Lecture Notes in Cosmology: Oliver F. Piattella

Lecture Notes in Cosmology
Oliver F. Piattella1
Núcleo Cosmo-UFES and Physics Department,

Federal University of Espı́rito Santo
Vitória, ES, Brazil
arXiv:1803.00070v1 [astro-ph.CO] 28 Feb 2018
1
E-mail: [email protected], [email protected]
Webpage: http://ofp.cosmo-ufes.org/
To Giuseppina, Lilli e Dorotea
Together we are gold

Contents
Preface 8
Notation 10
1 Cosmology 1
1.1 The expanding universe and its content . . . . . . . . . . . . . . . . 1
1.1.1 Olbers’s paradox . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 The accelerated expansion of the universe and Dark Energy 3
1.1.3 Dark Matter . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Cosmological observations . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 The cosmic microwave background . . . . . . . . . . . . . . 6
1.2.2 Redshift surveys . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Gravitational waves observatories . . . . . . . . . . . . . . . 8
1.2.4 Neutrino observation . . . . . . . . . . . . . . . . . . . . . . 8
1.2.5 Dark matter searches . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Redshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Open problems in cosmology . . . . . . . . . . . . . . . . . . . . . . 9
1.4.1 Cosmological constant and dark energy . . . . . . . . . . . . 10
1.4.2 Dark matter and small-scale anomalies . . . . . . . . . . . . 11
1.4.3 Other problems . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 The universe in expansion 13

2.1 Newtonian cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Relativistic cosmology . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Friedmann-Lemaı̂tre-Robertson-Walker metric . . . . . . . . 14
2.2.2 The conformal time . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.3 FLRW metric written with proper radius . . . . . . . . . . . 18
2.2.4 Light-cone structure of the FLRW space . . . . . . . . . . . 18
2.2.5 Christoffel symbols and geodesics . . . . . . . . . . . . . . . 20
2.3 Friedmann equations . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.1 The Hubble constant and the deceleration parameter . . . . 25
2.3.2 Critical density and density parameters . . . . . . . . . . . . 27
2.3.3 The energy conservation equation . . . . . . . . . . . . . . . 28
2.3.4 The ΛCDM model . . . . . . . . . . . . . . . . . . . . . . . 29
2.4 Solutions of the Friedmann equations . . . . . . . . . . . . . . . . . 32
2.4.1 The Einstein Static Universe . . . . . . . . . . . . . . . . . . 32
2.4.2 The de Sitter universe . . . . . . . . . . . . . . . . . . . . . 32
3
2.4.3 Radiation-dominated universe . . . . . . . . . . . . . . . . . 34
2.4.4 Cold matter-dominated universe . . . . . . . . . . . . . . . . 35
2.4.5 Radiation plus dust universe . . . . . . . . . . . . . . . . . . 36
2.5 Distances in cosmology . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5.1 Comoving distance and proper distance . . . . . . . . . . . . 38
2.5.2 The lookback time . . . . . . . . . . . . . . . . . . . . . . . 39
2.5.3 Distances and horizons . . . . . . . . . . . . . . . . . . . . . 40
2.5.4 The luminosity distance . . . . . . . . . . . . . . . . . . . . 41
2.5.5 Angular diameter distance . . . . . . . . . . . . . . . . . . . 43
3 Thermal history 45
3.1 Thermal equilibrium and Boltzmann equation . . . . . . . . . . . . 45
3.2 Short summary of thermal history . . . . . . . . . . . . . . . . . . . 47
3.3 The distribution function . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.1 Volume of the fundamental cell . . . . . . . . . . . . . . . . 50
3.3.2 Integrals of the distribution function . . . . . . . . . . . . . 51
3.4 The entropy density . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.5 Photons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.6 Neutrinos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.6.1 Temperature of the massless neutrino thermal bath . . . . . 59
3.6.2 Massive neutrinos . . . . . . . . . . . . . . . . . . . . . . . . 61
3.6.3 Matter-Radiation equality . . . . . . . . . . . . . . . . . . . 64
3.7 Boltzmann equation . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.7.1 Proof of Liouville theorem . . . . . . . . . . . . . . . . . . . 66
3.7.2 Example: The one-dimensional harmonic oscillator . . . . . 66
3.7.3 Boltzmann equation in General Relativity and Cosmology . 67
3.8 Boltzmann equation with a collisional term . . . . . . . . . . . . . . 70
3.8.1 Detailed calculation of the equilibrium number density . . . 72
3.8.2 Saha equation . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.9 Big-Bang Nucleosynthesis . . . . . . . . . . . . . . . . . . . . . . . 75
3.9.1 The baryon-to-photon ratio . . . . . . . . . . . . . . . . . . 75
3.9.2 The deuterium bottleneck . . . . . . . . . . . . . . . . . . . 76
3.9.3 Neutron abundance . . . . . . . . . . . . . . . . . . . . . . . 78
3.10 Recombination and decoupling . . . . . . . . . . . . . . . . . . . . . 84
3.10.1 Decoupling . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.11 Thermal relics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.11.1 The effective numbers of relativistic degrees of freedom . . . 93
3.11.2 Relic abundance of DM and the WIMP miracle . . . . . . . 96
4 Cosmological perturbations 98
4.1 From the perturbations of the FLRW metric to the linearised Ein-
stein tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.1.1 The perturbed Christoffel symbols . . . . . . . . . . . . . . . 101
4.1.2 The perturbed Ricci tensor and Einstein tensor . . . . . . . 103
4.2 Perturbation of the energy-momentum tensor . . . . . . . . . . . . 105
4.3 The problem of the gauge and gauge transformations . . . . . . . . 110
4.3.1 Coordinates and gauge transformations . . . . . . . . . . . . 111
4.3.2 The Scalar-Vector-Tensor decomposition . . . . . . . . . . . 113
4.3.3 Gauges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.4 Normal mode decomposition . . . . . . . . . . . . . . . . . . . . . . 119
4.5 Einstein equations for scalar perturbations . . . . . . . . . . . . . . 122
4.5.1 The relativistic Poisson equation . . . . . . . . . . . . . . . 123
4.5.2 The equation for the anisotropic stress . . . . . . . . . . . . 124
4.5.3 The equation for the velocity . . . . . . . . . . . . . . . . . 125
4.5.4 The equation for the pressure perturbation . . . . . . . . . . 125
4.6 Einstein equations for tensor perturbations . . . . . . . . . . . . . . 125
4.7 Einstein equations for vector perturbations . . . . . . . . . . . . . . 128
5 Perturbed Boltzmann equations 130

5.1 General form of the perturbed Boltzmann equation . . . . . . . . . 130
5.1.1 On the photon and neutrino perturbed distributions . . . . . 131
5.2 Force term . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.1 Scalar perturbations . . . . . . . . . . . . . . . . . . . . . . 133
5.2.2 Tensor perturbations . . . . . . . . . . . . . . . . . . . . . . 134
5.2.3 Vector perturbations . . . . . . . . . . . . . . . . . . . . . . 135
5.3 The perturbed Boltzmann equation for CDM . . . . . . . . . . . . . 136
5.4 The perturbed Boltzmann equation for massless neutrinos . . . . . 138
5.5 The perturbed Boltzmann equation for photons . . . . . . . . . . . 143
5.5.1 Computing the collisional term neglecting polarisation . . . 144
5.5.2 Full Boltzmann equation including polarisation . . . . . . . 147
5.5.4 Tensor perturbations . . . . . . . . . . . . . . . . . . . . . . 151
5.5.5 Vector perturbations . . . . . . . . . . . . . . . . . . . . . . 153
5.6 Boltzmann equation for baryons . . . . . . . . . . . . . . . . . . . . 155
6 Initial conditions 159

6.1 Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.2 Evolution equations in the kη 1 limit . . . . . . . . . . . . . . . 161
6.2.1 Multipoles in the kη 1 limit . . . . . . . . . . . . . . . . . 162
6.2.2 CDM and baryons velocity equations . . . . . . . . . . . . . 164
6.2.3 The kη 1 limit of the Einstein equations . . . . . . . . . . 164
6.3 The adiabatic primordial mode . . . . . . . . . . . . . . . . . . . . 166
6.3.1 Why “adiabatic”? . . . . . . . . . . . . . . . . . . . . . . . . 167
6.4 The neutrino density isocurvature primordial mode . . . . . . . . . 169
6.5 The CDM and baryons isocurvature primordial modes . . . . . . . . 169
6.6 The neutrino velocity isocurvature primordial mode . . . . . . . . . 171
6.7 Planck constraints on isocurvature modes . . . . . . . . . . . . . . . 172
7 Stochastic Properties of Cosmological Perturbations 174

7.1 Stochastic cosmological perturbations and power spectrum . . . . . 174
7.2 Random fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
7.3 Power spectrum and Gaussian random fields . . . . . . . . . . . . . 179
7.3.1 Definition of a Gaussian random field . . . . . . . . . . . . . 180
7.3.2 Estimator of the power spectrum and cosmic variance . . . . 181
7.4 Non-Gaussian perturbations . . . . . . . . . . . . . . . . . . . . . . 183
7.5 Matter power spectrum, transfer function and stochastic initial con-
ditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
7.6 CMB power spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
7.6.1 Cosmic Variance of angular power spectra . . . . . . . . . . 188
7.7 Power spectrum for tensor perturbations . . . . . . . . . . . . . . . 190
7.8 Ergodic theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
8 Inflation 194
8.1 The flatness problem . . . . . . . . . . . . . . . . . . . . . . . . . . 194
8.2 The horizon problem . . . . . . . . . . . . . . . . . . . . . . . . . . 196
8.3 Single scalar field slow-roll inflation . . . . . . . . . . . . . . . . . . 198
8.3.1 More slow-roll parameters . . . . . . . . . . . . . . . . . . . 201
8.3.2 Reheating . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
8.4 Production of gravitational waves during inflation . . . . . . . . . . 205
8.5 Production of scalar perturbations during inflation . . . . . . . . . . 212
8.6 Spectral indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
8.7 Observational results . . . . . . . . . . . . . . . . . . . . . . . . . . 219
8.8 Examples of models of inflation . . . . . . . . . . . . . . . . . . . . 220
8.8.1 General power law potential . . . . . . . . . . . . . . . . . . 221
8.8.2 The Starobinsky model . . . . . . . . . . . . . . . . . . . . . 222
9 Evolution of perturbations 225

9.1 Evolution on super-horizon scales . . . . . . . . . . . . . . . . . . . 226
9.1.1 Evolution through radiation-matter equality . . . . . . . . . 228
9.1.2 Evolution in the Λ-dominated epoch . . . . . . . . . . . . . 231
9.1.3 Evolution through matter-DE equality . . . . . . . . . . . . 232
9.2 The matter-dominated epoch . . . . . . . . . . . . . . . . . . . . . 235
9.2.1 Baryons falling into the CDM potential wells . . . . . . . . . 238
9.3 The radiation-dominated epoch . . . . . . . . . . . . . . . . . . . . 240
9.4 Deep inside the horizon . . . . . . . . . . . . . . . . . . . . . . . . . 243
9.5 Matching and CDM transfer function . . . . . . . . . . . . . . . . . 246
9.6 The transfer function for tensor perturbations . . . . . . . . . . . . 252
9.6.1 Radiation-dominated epoch . . . . . . . . . . . . . . . . . . 253
9.6.2 Matter-dominated epoch . . . . . . . . . . . . . . . . . . . . 253
9.6.3 Deep inside the horizon . . . . . . . . . . . . . . . . . . . . . 254
10 Anisotropies in the Cosmic Microwave Background 257

10.1 Free-streaming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
10.2 Anisotropies on large scales . . . . . . . . . . . . . . . . . . . . . . 262
10.3 Tight-coupling and acoustic oscillations . . . . . . . . . . . . . . . . 264
10.3.1 The acoustic peaks for R = 0 . . . . . . . . . . . . . . . . . 267
10.3.2 Baryon loading . . . . . . . . . . . . . . . . . . . . . . . . . 272
10.4 Diffusion damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
10.5 Line-of-sight integration . . . . . . . . . . . . . . . . . . . . . . . . 279
10.6 Finite thickness effect and reionization . . . . . . . . . . . . . . . . 283
10.7 Cosmological parameters determination . . . . . . . . . . . . . . . . 286
10.8 Tensor contribution to the CMB TT correlation . . . . . . . . . . . 292
10.9 Polarisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
10.9.1 Scalar perturbations contribution to polarisation . . . . . . . 298
10.9.2 Tensor perturbations contribution to polarisation . . . . . . 301
11 Miscellanea 306
11.1 Bayesian analysis using type Ia supernovae data . . . . . . . . . . . 306
11.2 Doing statistics in the sky . . . . . . . . . . . . . . . . . . . . . . . 310
11.2.1 Top Hat and Gaussian filters . . . . . . . . . . . . . . . . . . 312
11.2.2 Sampling and shot noise . . . . . . . . . . . . . . . . . . . . 313
11.2.3 Correlation function . . . . . . . . . . . . . . . . . . . . . . 315
11.2.4 Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
12 Appendices 318
12.1 Thermal distributions . . . . . . . . . . . . . . . . . . . . . . . . . . 318
12.1.1 Derivation of the Maxwell-Boltzmann distribution . . . . . . 318
12.1.2 Derivation of the Fermi-Dirac distribution . . . . . . . . . . 319
12.1.3 Derivation of the Bose-Einstein distribution . . . . . . . . . 320
12.2 Derivation of the Poisson distribution . . . . . . . . . . . . . . . . . 321
12.3 Helmholtz theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
12.4 Conservation of R on large scales and for adiabatic perturbations . 323
12.5 Spherical harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
12.5.1 Spin-weighted spherical harmonics . . . . . . . . . . . . . . . 330
12.6 Method of Green’s functions . . . . . . . . . . . . . . . . . . . . . . 333
12.7 Polarisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
12.7.1 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . . 334
12.7.2 Polarisation ellipse and Stokes parameters . . . . . . . . . . 336
12.8 Thomson scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Bibliography 347
Subject Index 361

Preface
Considerate la vostra semenza: fatti non foste a viver come bruti ma
per seguir virtute e canoscenza
(Consider well the seed that gave you birth: you were not made to
live as brutes, but to follow virtue and knowledge)
—Dante Alighieri, Divina Commedia, Canto XXVI
These lecture notes are based on the hand-written notes which I prepared for
the cosmology course taught to graduate students of PPGFis and PPGCosmo at
the Federal University of Espı́rito Santo (UFES), starting from 2014.
This course covers topics ranging from the evidence of the expanding universe
to Cosmic Microwave Background anisotropies. In particular, the first Chapter
commences with a bird’s eye view of cosmology, showing its tremendous evolu-
tion during the last fifty years and the open problems on which cosmologists are
working today. From Chapter 2 on starts the conventional content, hopefully ex-
posed in a not too conventional manner. The main topics are: the expansion of
the universe, relativistic cosmology and Friedmann equations (Chapter 2); ther-
mal history, Big Bang Nucleosynthesis, recombination and cold relic abundance
(Chapter 3); cosmological perturbation theory and perturbed Einstein equations
(Chapter 4); perturbed Boltzmann equations (Chapter 5); primordial modes of
perturbations (Chapter 6); random variables and stochastic character of cosmo-
logical perturbations (Chapter 7); inflationary paradigm (Chapter 8); evolution of
perturbations (Chapter 9); anisotropies in the Cosmic Microwave Background sky
(Chapter 10). A selection of extra topics is offered in Chapter 11 whereas Chapter
12 collects some important and well-known physical and mathematical results.
Scattered throughout the text are many exercises. I chose not to put them at
the end of a Chapter or at the end of the book because I want the reader to stop
and do some work in the moment in which this is needed. Most of the exercises
are not extensions of the material covered, but consist in developing calculations
necessary to the topics addressed. The idea here is, since these are lecture notes,
not only to provide a book for consultation but to put the reader-student to work.
When I prepared these lecture notes, I heavily relied on the following three
famous textbooks in cosmology:
• S. Dodelson, Modern Cosmology, [Dodelson, 2003],
• V. Mukhanov, Principles of physical Cosmology, [Mukhanov, 2005],
• S. Weinberg, Cosmology, [Weinberg, 2008],
but also on many other books and papers, which are cited throughout the text.
I tried not to simply develop the calculations there contained, but to provide an
original presentation. I hope to have succeeded, but of course the final word is up
to the reader.
I want now to make three recommendations to the reader-student.
First. Read the original papers in order to make contact with the original ideas
in their primeval forms and to appreciate the geniuses of their authors.
Second. The technological advance has provided us with sophisticated instru-
ments such as Inspire, http://inspirehep.net/, a powerful tool for researching
papers. One of the features I like most about it is the possibility to check which
papers have cited the one in which we are interested, thus allowing us to rapidly get
up-to-date on a given topic. I sometimes joke about this by rephrasing a sentence
which we all have many times read in papers and books when referring to a certain
work: see ... and references therein. Thanks to the above-mentioned citation tool
we can now add also the sentence see ... and references thereout. These notes will
be up-to-date when published, but cosmology is a very vivacious research field so
they will probably become outdated in a few years. My recommendation is thus to
take advantage of the references thereout tools and keep yourself always updated.
Third and final one. In preparing these notes I have intensively employed the
CLASS code. This is very user friendly, so it was not as hard as it may sound. My
point in doing so is that analytic calculations are very stimulating and very useful
in order to understand the physics ruling the cosmological phenomena, so they
have an enormous didactical value. On the other hand, observation needs precise
calculations, and these can be done only numerically. So, my recommendation here
is the following: do not be afraid of numerical codes and learn how to use them
proficiently.
I thank all my students and colleagues who have helped me, through questions
and suggestions, in the challenging but rewarding task of writing these notes. Extra
thanks to Rodrigo Von Marttens, whose help with CLASS has been crucial in order
for me to rapidly grasp how to run the code. Special thanks are due to my editor
Aldo Rampioni and his assistant Kirsten Kley-Theunissen for they kind help and
encouragement throughout the realisation of the project.
Oliver F. Piattella
Vitória, Brazil
February 2018
Notation
Humans are good, she knew, at discerning subtle patterns that are
really there, but equally so at imagining them when they are altogether
absent
—Carl Sagan, Contact
Latin indices (e.g. i, j, k) run over the three spatial coordinates and assume values
1, 2, 3. In a Cartesian coordinate system we shall use x1 ≡ x, x2 ≡ y and x3 ≡ z.
Greek indices (e.g. µ, ν, ρ) run over the four spacetime coordinates and assume
values 0, 1, 2, 3, with x0 being the time coordinate.
Repeated
P4 high and low indices are summed unless otherwise stated, i.e. xµ yµ ≡
µ i
P 3 i
µ=0 x yµ or x yi ≡ i=0 x yi . Repeated spatial indices are also summed, whether
one is high and the other is low or not, i.e. xi yi ≡ 3i=0 xi yi .
P
The signature employed for the metric is (−, +, +, +).
Spatial 3-vectors are indicate in boldface, e.g. v.
The unit vector corresponding to any 3-vector v is denoted with a hat, i.e. v̂ ≡
v/|v|, where |v| is the modulus of the 3-vector v defined as |v|2 = δij v i v j with δij
being the usual Kronecker delta.
A dot over any quantity denotes the derivation with respect to the cosmic time,
denoted with t, of that quantity, whereas a prime denotes derivation with respect
to the conformal time, denoted with η.
The operator ∇2 is the usual Laplacian operator in the Euclidian space, i.e. ∇2 ≡
δ ij ∂i ∂j , where ∂µ is used as a shorthand for the partial derivative with respect the
coordinate xµ .
Except on vector and tensor, a 0 subscript means that a time-dependent quantity

is evaluated today, i.e. at t = t0 or η = η0 where t0 and η0 represent the age of the
universe in the cosmic time or in the conformal time.
The subscripts b, c, γ and ν put on matter quantities such as density and pressure
refer to baryons, cold dark matter, photons and neutrinos, respectively. Subscripts
m and r refer to matter and radiation, in general.
With k is denoted the comoving wavenumber.

From Chapter 4 on natural units ~ = c = 1 shall be employed.
When employing the spherical coordinate system, the azimuthal coordinate is de-
noted with φ. A scalar field is denoted with ϕ.
Chapter 1
Cosmology
And that inverted Bowl we call the Sky,

Whereunder crawling coop’t we live and die,
Lift not thy hands to It for help—for It,
Rolls impotently on as Thou or I.
—Omar Khayyám, Rubáiyát
In this Chapter we present an overview of cosmology, addressing its most impor-

tant aspects and presenting some observational experiments and open problems.
1.1 The expanding universe and its content

The starting point of our study of cosmology is the extraordinary evidence that
we live in an expanding universe. This was a landmark discovery made in the
XX century, usually attributed to Edwin Hubble [Hubble, 1929], but certainly
resulting from the joint efforts of astronomers such as Vesto Slipher [Slipher, 1917]
and cosmologists such as George Lemaı̂tre [Lemaitre, 1927]. We do not enter here
the debate about who is deserving more credit for the discovery of the expansion of
the universe. The interested reader might want to read e.g. [Way and Nussbaumer,
2011] and [van den Bergh, 2011].
Hubble discovered that the farther a galaxy is the faster it recedes from us. See
Fig. 1.1. This is the famous Hubble’s law:
v = H0 r , (1.1)
where v is the recessional velocity, r is the distance and H0 (usually pronounced

“H-naught”) is a constant named after Hubble. The value of H0 determined by
Hubble himself was:
H0 = 500 km s−1 Mpc−1 , (1.2)
with huge error, as can be understood from Fig. 1.1 by observing how much the
data points are scattered. A more precise estimate was made by Sandage [Sandage,
1958] in 1958:
H0 = 75 km s−1 Mpc−1 . (1.3)
1
Figure 1.1: Figure 1 of Hubble’s original paper [Hubble, 1929].
A recent measurement done by the BOSS collaboration [Grieb et al., 2017] gives
−1
H0 = 67.6+0.7
−0.6 km s Mpc−1 . (1.4)
Roughly speaking, this number means that for each Mpc away a source recedes
67.6 km/s faster. At a certain radius, the receding velocity attains the velocity
of light and therefore we are unable to see farther objects. This radius is called
Hubble radius.
The Hubble constant can be measured also with fair precision by using the time-
delay among variable signals coming from lensed distant sources [Bonvin et al.,
2017] and via gravitational waves [Abbott et al., 2017b].
1.1.1 Olbers’s paradox

The expansion of the universe could have been predicted a century before Hubble
by solving Olbers’s paradox [Olbers, 1826]. See e.g. Dennis Sciama’s book [Sciama,
2012] for a very nice account of the paradox, which is also called the dark night
sky paradox and goes as follows: if the universe is static, infinitely large and
old and with an infinite number of stars distributed uniformly, then the night sky
should be bright.
Let us try to understand why. First of all, the stars are distributed uniformly,
which means that their number density say n is a constant. Consider a spherical
shell of thickness dR and radius R centred on the Earth. The number of stars
inside this spherical shell is:
dN = 4πnR2 dR . (1.5)
2
The total luminosity of this spherical shell is dN multiplied by the luminosity say
L of a single star, and we assume L to be the same for all the stars. Only a fraction
f of the radiation produced by the star reaches Earth, but this fraction is the same
for all the stars (because we assume them to be identical). Therefore, the total
luminosity of a spherical shell is dN f L. By the inverse-square law, the total flux
received on Earth is:
dN f L
dFtot = = nf LdR . (1.6)
4πR2
It does not depend on R and thus it diverges when integrated over R from zero
to infinity. This means not only that the night sky should not be dark but also
infinitely bright!
We can solve the problem of having an infinitely bright night sky by considering
the fact that stars are not points and do eclipse each other, so that we do not really
see all of them. Suppose that each star shows us a surface dA. Therefore, if a star
lies at a distance R, we receive from it the flux dF = dAL/(4πR2 ), where L is the
luminosity per unit area. But dA/R2 = dΩ is the solid angle spanned by the star
in the sky. Therefore:
L
dF = dΩ . (1.7)
4π
Once again, this does not depend on R! When we integrate it over the whole solid
angle, we obtain that Ftot = L, i.e. the whole sky is as luminous as a star! In other
words: it is true that the farther a star is the fainter it appears, but we can pack
more of them in the same patch of sky.
In order to solve Olbers’s paradox, we can drop one or more of the initial
assumptions. For example:
• The universe is not eternal so the light of some stars has not yet arrived to
us. This is plausible, but even so we could expect a bright night sky and also
to see some new star to pop out from time to time, without being a transient
phenomenon such as a supernova explosion. There is no record of this.
• Maybe there is not an infinite number of stars. But we have showed that
taking into account their dimension we do not need an infinite number and
yet the paradox still exists.
• Are stars distributed not uniformly? Even so, we would expect still a bright
night sky, even if not uniformly bright.
At the end, we must do something in order for the light of some stars not to reach
us. A possibility is to drop the staticity assumption. The farther a spherical shell
is, the faster it recedes from us (this is Hubble’s law). In this way, beyond a certain
distance (the Hubble radius) light from stars cannot reach us and the paradox is
solved.
1.1.2 The accelerated expansion of the universe and Dark

Energy
The discovery of the type Ia supernova 1997ff [Williams et al., 1996] marked the
beginning of a new era in cosmology and physics. The analysis of the emission
3
of this type of supernovae led to the discovery that our universe is in a state of
accelerated expansion. See e.g. [Perlmutter et al., 1999] and [Riess et al., 1998].
This is somewhat problematic because gravity as we know it should attract
matter, thereby causing the expansion to decelerate. So, what does cause the
acceleration in the expansion? A possibility is that there exists a new form of
matter, or rather energy, which acts as anti-gravity. This is widely known today
as Dark Energy (DE) and its nature is still a mystery to us. The most simple
and successful candidate for DE is the cosmological constant Λ.
1.1.3 Dark Matter

DE is not the only dark part of our universe. Many observations of different nature
and from different sources at different distance scales point out the existence of an-
other dark component, called Dark Matter (DM). In particular, these observations
are:
• The dynamics of galaxies in clusters. The pioneering applications of the
virial theorem to the Coma cluster by Zwicky [Zwicky, 1933] resulted in a virial
mass 500 times the observed one (which can be estimated by the light emission).
• Rotation curves of spiral galaxies. This is the famous problem of the
flattish velocity curves of stars in the outer parts of spiral galaxies. See e.g. [Sofue
et al., 1999] and [Sofue and Rubin, 2001].
The surprising fact of these flattish curves is that there is no visible
√ matter to
justify them and one would then expect a Keplerian fall V ∼ 1/ R, where R is
the distance from the galactic centre. In order to derive the Keplerian fall, simply
assume a circular orbit and use Newtonian gravity (which seems to be fine for
galaxies). Then, the centrifugal force is canceled by the gravitational attraction of
the galaxy as follows:
V2 GM (R)
= , (1.8)
R R2
where we assume also a spherical distribution of matter in the galaxy and therefore
use Gauss’ theorem. Inside the bulge of the galaxy, the visible mass goes as M ∝
R3 , and therefore V ∝ R, which represents the initial part of the velocity curve.
However, outside the √ bulge of the galaxy, the mass becomes a constant, thus the
Keplerian fall V ∝ 1/ R follows.
• The brightness in X-rays of galaxy clusters. This depends on the
gravitational potential well of the cluster, which is deduced to be much deeper
than the one that would be generated by visible matter only. See e.g. [Weinberg,
2008].
• The formation of structures in the universe. In other words, the fact
that the density contrast of the usual matter, called baryonic in cosmology, δb is
today highly non-linear, i.e. δb 1. As we shall see in Chapter 9, relativistic
cosmology predicts that δb grows by a factor of 103 between recombination and
today and observations of CMB show that the value of δb at recombination is
δb ∼ 10−5 . This means that today δb ∼ 10−2 which is in stark contrast with the
huge number of structures that we observe in our universe. So, there must be
something else which catalyses structure formation.
4
• The structure of the Cosmic Microwave Background peaks. The
temperature-temperature correlation spectrum in the CMB sky is characterised by
the so-called acoustic peaks. The absence of DM would not allow to reproduce
the structure shown in Fig. 1.2. We shall study this structure in detail in Chapter
10.
Figure 1.2: CMB TT spectrum. Figure taken from [Ade et al., 2016a]. The red
solid line is the best fit ΛCDM model.
• Weak Lensing. The bending of light is a method for measuring the mass
of the lens, and it is a classical test of General Relativity (GR) [Weinberg, 1972].
When the background source is distorted by the foreground lens, one has the so-
called weak lensing. Analysing the emission of the distorted source allows to map
the gravitational potential of the lens and therefore its matter distribution. See
e.g. [Dodelson, 2017] for a recent textbook reference on gravitational lensing. Weak
gravitational lensing is a powerful tool for the study of the geometry of the universe
and its observation is one of the primary targets of forthcoming surveys such as
the European Space Agency (ESA) satellite Euclid and the Large Synoptic Survey
Telescope (LSST).
A remarkable combination of X-ray and weak lensing observational techniques
made the Bullet Cluster famous [Clowe et al., 2006]. Indeed X-ray maps show
the result of a merging between the hot gases of two galaxy clusters which gravi-
tational lensing maps reveal to be lagging behind their respective centres of mass.
Therefore, most parts of the clusters simply went through one another, leaving
behind a smaller fraction of hot gas. This is considered a direct empirical proof of
the existence of DM forming a massive halo and a gravitational potential well in
which gas and galaxies lie.
5
Popular candidates to the role of DM are particles beyond the standard model
[Silk et al., 2010]. Among these, the most famous are the Sterile Neutrino [Dodel-
son and Widrow, 1994], the Axion, which is related to the process of violation of
CP symmetry [Peccei and Quinn, 1977], and Weakly Interacting Massive Particles
(WIMP) which count among them the lightest supersymmetric neutral stable par-
ticle, the Neutralino. See [Bertone and Hooper, 2016] for a historical account of
DM and [Profumo, 2017] for a textbook on particle DM.
The observational evidences of DM that we have seen earlier do not only point
to the existence of DM but also on the necessity of it being cold, i.e. with negligi-
ble pressure or, equivalently, with a small velocity (much less than that of light) of
its particles (admitting that DM is made of particles, which is the common under-
standing to which we adhere in these notes). Hence Cold Dark Matter (CDM)
shall be our DM paradigm.
As we shall see in Chapter 3 if DM was in thermal equilibrium with the rest of
the known particles in the primordial plasma, i.e. it was thermally produced,
then its being cold amounts to say that its particles have a sufficiently large mass,
e.g. about 100 GeV for WIMP’s. Decreasing the mass, we have DM candidates
characterised by increasing velocity dispersions and thus with different impacts on
the process of structure formation. Typically one refers to Warm Dark Matter
(WDM) as a thermally produced DM with mass of the order of some keV and
Hot Dark Matter (HDM) as thermally produced particles with small masses,
e.g. of the order of the eV, or even massless. In fact neutrinos can be considered
as a HDM candidate. Note that the axion has a mass of about 10−5 eV but
nonetheless is CDM because it was not thermally produced, i.e. it never was in
thermal equilibrium with the primordial plasma.
The combined observational successes of Λ and CDM form the so-called ΛCDM
model, which is the standard model of cosmology.
1.2 Cosmological observations

We dedicate this section to the most important cosmological observations which
are ongoing or ended recently, or are planned.
1.2.1 The cosmic microwave background

The cosmic microwave background (CMB) radiation provides a window onto the
early universe, revealing its composition and structure. It is a relic, thermal ra-
diation from a hot dense phase in the early evolution of our universe which has
now been cooled by the cosmic expansion to just three degrees above absolute
zero. Its existence had been predicted in the 1940s by Alpher and Gamow [Alpher
et al., 1948] and its discovery by Penzias and Wilson at Bell Labs in New Jersey,
announced in 1965 [Penzias and Wilson, 1965] was convincing evidence for most
astronomers that the cosmos we see today emerged from a Hot Big Bang more
than 10 billion years ago.
Since its discovery, many experiments have been performed to observe the CMB
radiation at different frequencies, directions and polarisations, mostly with ground-
6
and balloon-based detectors. These have established the remarkable uniformity of
the CMB radiation, at a temperature of 2.7 Kelvin in all directions, with a small
±3.3 mK dipole due to the Doppler shift from our local motion (at 1 million
kilometres per hour) with respect to this cosmic background.
However, the study of the CMB has been transformed over the last twenty
years by three pivotal satellite experiments. The first of these was the Cosmic
Background Explorer (CoBE, https://lambda.gsfc.nasa.gov/product/cobe/),
launched by NASA in 1990. In 1992 CoBE reported the detection of statistically
significant temperature anisotropies in the CMB, at the level of ±30 µK on 10
degree scales [Smoot et al., 1992] and it confirmed the black body spectrum with
an astonishing precision, with deviations less than 50 parts per million [Smoot
et al., 1992]. CoBE was succeeded by the Wilkinson Microwave Anisotropy Probe
(WMAP, https://map.gsfc.nasa.gov/) satellite, launched by NASA in 2001,
which produced full sky maps in five frequencies (from 23 to 94 GHz) mapping the
temperature anisotropies to sub-degree scales and determining the CMB polarisa-
tion on large angular scales for the first time.
The Planck satellite (http://sci.esa.int/planck/), launched by ESA in
2009, sets the current state of the art with nine separate frequency channels, mea-
suring temperature fluctuations to a millionth of a degree at an angular resolution
down to 5 arc-minutes.
Planck’s mission ended in 2013 and the full-mission data were released in 2015
in [Adam et al., 2016] and in many companion papers. A fourth generation of full-
sky, microwave-band satellite recently proposed to ESA within Cosmic Vision 2015-
2025 is the Cosmic Origins Explorer (COrE, http://www.core-mission.org/)
[Bouchet et al., 2011].
At the moment, a great effort is being devoted to the detection of the B-mode
of CMB polarization because it is the one related to the primordial gravitational
waves background, as we shall see in Chapter 10. Located near the South Pole,
BICEP3 (https://www.cfa.harvard.edu/CMB/bicep3/) and the Keck Array are
telescopes devoted to this purpose.
Among the non-satellite CMB experiments we must mention the Balloon Obser-
vations Of Millimetric Extragalactic Radiation ANd Geophysics (BOOMERanG)
which was a balloon-based mission which flew in 1998 and in 2003 and mea-
sured CMB anisotropies with great precision (higher than CoBE). From these
data the Boomerang collaboration first determined that the universe is spatially
flat [de Bernardis et al., 2000].
1.2.2 Redshift surveys

Redshift surveys are observations of certain patches of sky at certain wavelengths
with the aim of determining mainly the angular positions (declination and right
ascension) redshifts and spectra of galaxies.
The Sloan Digital Sky Survey (SDSS, http://www.sdss.org/) is a massive
spectroscopic redshift survey which is ongoing since the year 2000 and it is now in
its stage IV with 14 data releases available. It is ground-based and uses a telescope
located in New Mexico (USA). The SDSS-IV is formed by three sub-experiment:
7
• The Extended Baryon Oscillation Spectroscopic Survey (eBOSS), focusing
on redshifts 0.6 < z < 2.5 and on the Baryon Acoustic Oscillations (BAO)
phenomenon;1
• The Apache Point Observatory Galaxy Evolution Experiment (APOGEE-2)

is dedicated to the study of our Milky Way;
• The Mapping Nearby Galaxies at Apache Point Observatory (MaNGA) study

instead nearby galaxies by measuring their spectrum along their extension
and not only at the centre.
The V generation of the SDSS will start in 2020, consisting of three surveys: the
Milky Way Mapper, the Black Hole Mapper and the Local Volume Mapper. See
[Kollmeier et al., 2017].
The Dark Energy Survey (DES, https://www.darkenergysurvey.org/) mea-
sures redshifts photometrically using a telescope situated in Chile and looking for
Type Ia supernovae, BAO and weak lensing signals.
Planned surveys are the already mentioned satellite Euclid (http://sci.esa.
int/euclid/), whose launch is due possibly in 2021 and the telescope LSST, which
is being built in Chile and whose first light is due in 2019. We also cite the
NASA satellite Wide Field Infrared Survey Telescope (WFIRST, https://www.
nasa.gov/wfirst) and the Javalambre Physics of the accelerating universe As-
tronomical Survey (J-PAS, http://j-pas.org/). The main cosmological goals
of these experiments relay on the detection of weak lensing, BAO and type Ia
supernovae signals with high precision.
1.2.3 Gravitational waves observatories

The recent direct detection of gravitational waves (GW) by the LIGO-Virgo col-
laboration (https://www.ligo.org/) [Abbott et al., 2016b], [Abbott et al., 2016a]
has opened a new observational window on the universe. In particular, GW are
relevant in cosmology because they could be a relic from inflation containing in-
valuable informations on the very early universe. As already mentioned, they are
being searched via the detection of the B-mode polarisation of the CMB.2
There are now three functioning ground-based GW observatories: LIGO (Han-
ford and Livingstone, USA) and Virgo (near Pisa, Italy). KAGRA, in Japan, is
under construction and another one in India, INDIGO, is planned. The space-based
LISA GW observer is still in a preliminary phase (LISA pathfinder).
1.2.4 Neutrino observation

Neutrinos are relevant in cosmology, as we shall see throughout these notes, because
they should form a cosmological background as CMB photons do. The great prob-
1
We shall not address BAO extensively in these notes, but only mention them in Chapter
10. Together with weak lensing, BAO are another powerful observable upon which present and
future missions are planned.
2
The events detected by the LIGO-Virgo collaboration originated from merging of black holes
or neutron stars. Thus are not part of the primordial GW background.
8
lem is that it is incredibly difficult to detect them and even more if they have low
energy, as we expect to be the case for neutrinos in the cosmological background.
The most important neutrino observatory is IceCube (http://icecube.wisc.
edu/), operating since 2005 (its construction was completed in 2010) and located
near the South Pole. It detects neutrinos indirectly, via their emission of Cherenkov
light.
1.2.5 Dark matter searches

The search for DM particles counts on many observatories and the Large Hadron
Collider (LHC). See [Gaskins, 2016] and [Liu et al., 2017] for the status of indirect
and direct DM searches which, unfortunately, have not been successful until now.
1.3 Redshift
Redshift is a fundamental observable of cosmology. Its definition is the following:
λobs
z= −1 (1.9)
λem
It is always positive, i.e. observed radiation is redder than the emitted one, because
the universe is in expansion. For the closest sources, such as Andromeda, it is
negative, i.e. the observed radiation is bluer than the emitted one, because the
Hubble flow is overcome by the peculiar motion due to local gravitational effects.
For the moment we can think of the redshift as a Doppler effect due to the
relative motion of the sources. In Chapter 2 we will relate it to spacetime geometry
with GR.
Redshift is measured in two ways: spectroscopically or photometrically. For the
former one needs to do spectroscopy, i.e. detecting known emission or absorption
lines from a source and comparing their wavelengths with the ones measured in a
laboratory on Earth. Hence one uses Eq. (1.9) and thus calculate z.
Photometric redshifts are calculated by assuming certain spectral features for
the sources and measuring their relative brightness in certain wavebands, using
filters.
A simple example is the following. Sun’s spectrum is almost a blackbody one
with temperature of about 6000 K and then, by using Wien’s displacement law, it
has a peak emission at a wavelength of 500 nm. Therefore, if a star similar to the
Sun had a peak emission of say 600 nm, then using Eq. (1.9) one would calculate
z = 0.2.
The reason for using photometry instead of spectroscopy is that it is less time-
consuming and allows to obtain redshifts of very far sources, for which it is difficult
to do spectroscopy. On the other hand, photometric redshifts are less precise.
1.4 Open problems in cosmology

The fundamental issue in cosmology is to understand what are DM and DE. The
effort of answering this question makes cosmology, particle physics and quantum
9
field theory (QFT) to merge. The ways adopted in order to tackle these problems
are essentially the search for particles beyond the standard model and the inves-
tigation of new theories of gravity, which in most of the cases are extensions of
GR.
1.4.1 Cosmological constant and dark energy

Pure geometrical Λ and vacuum energy have the same dynamical behaviour in
GR. Estimating the latter via QFT calculations and comparing the result with
the observed value leads to the famous fine-tuning problem of the cosmological
constant. See e.g. [Weinberg, 1989]. This roughly goes as follows: the observed
value of ρΛ is about 10−47 GeV4 [Ade et al., 2016a]. The natural scale for the
vacuum energy density is the Planck scale, i.e. 1076 GeV4 . There are 123 orders of
magnitude of difference! Even postulating a false vacuum state after the electro-
weak phase transition at 108 GeV4 , the difference is 55 orders of magnitude. See
[Martin, 2012] for a comprehensive account of Λ and the issues related to it.
Another problem with Λ is the so-called cosmic coincidence [Zlatev et al.,
1999]. This problem stems from the fact that the density of matter decreases with
the inverse of the cube scale factor, whereas the energy density of the cosmological
constant is, as its name indicates, constant. However, these two densities are
approximately equal at the present time. This coincidence becomes all the more
intriguing when we consider that if the cosmological constant had dominated the
energy content of the universe earlier, galaxies would not have had time to form; on
the other hand, had the cosmological constant dominated later, then the universe
would still be in a decelerated phase of expansion or younger than some of its oldest
structures, such as clusters of stars [Velten et al., 2014].
The cosmic coincidence problem can also be seen as a fine-tuning problem in
the initial conditions of our universe. Indeed, consider the ratio ρΛ /ρm , of the
cosmological constant to the matter content. This ratio goes as a3 . Suppose that
we could extrapolate our classical theory (GR) up to the Planck scale, for which
a ≈ 10−32 . Then, at the Planck scale we have ρΛ /ρm ≈ 10−96 . This means that, at
trans-Planckian energies, possibly in the quantum universe, there must be a mech-
anism which establishes the ratio ρΛ /ρm with a precision of 96 significant digits!
Not a digit can be missed, otherwise we would have today 10 times more cosmo-
logical constant than matter, or vice-versa, thereby being in strong disagreement
with observation.
So, we find ourselves in a situation of impasse. On one hand, Λ is the simplest
and most successful DE candidate. On the other hand it suffers from the above-
mentioned issues. What do we do? Much of today research in cosmology addresses
this question. Answers are looked for mostly via investigation of new theories of
gravity, extensions or modifications of GR, of which DE would be a manifestation.
There are so many papers addressing extended theories of gravity that it is quite
difficult to choose representatives. Probably the best option is to start with a
textbook, e.g. [Amendola and Tsujikawa, 2010].
A different approach is to accept that Λ has the value it has by chance, and it
turns out to be just the right value for structures to form and for us to be here
doing cosmology. This also known as Anthropic Principle and exists in many
10
forms, some stronger than others. It is also possible that ours is one universe out of
an infinite number of realisations, called Multiverse, with different values of the
fundamental constants. Life as we know it then develops only in those universes
where the conditions are favourable. Again, it is difficult to cite papers on these
topics (which are more about metaphysics rather than physics, since there is no
possibility of performing experimental tests) but a nice reading is e.g. [Weinberg,
1992].
1.4.2 Dark matter and small-scale anomalies

On sub-galactic scales, of about 1 kpc, the CDM paradigm displays some difficulties
[Warren et al., 2006]. These are called CDM small-scales anomalies. See e.g.
[Bullock and Boylan-Kolchin, 2017] for a recent account. They are essentially three
and stem from the results of numerical simulations of the formation of structures:
1. The Core/Cusp problem [Moore, 1994]. The CDM distribution in the centre
of the halo has a cusp profile, whereas observation suggests a core one;
2. The Missing satellites problem [Klypin et al., 1999]. Numerical simulations

predict a large number of satellite structures, which are not observed;
3. The Too big to fail problem [Boylan-Kolchin et al., 2011]. The sub-structures
predicted by the simulations are too big not to be seen.
Possible solutions to these small-scale anomalies are the following:
Baryon feedback. The cross section for DM particles and the standard model
particles interaction must be very small, i.e. ∼ 10−39 cm2 , but in environments of
high concentration, such as in the centre of galaxies, such interactions may become
important and may provide an explanation for the anomalies. The problem is that
the models of baryon feedback are difficult to be simulated and they seem not to
be enough to resolve the anomalies [Kirby et al., 2014].
Warm dark matter. As anticipated, WDM are particles with mass around
the keV which decouple from the primordial plasma when relativistic. They are
subject to free streaming that greatly cut the power of fluctuations on small scales,
thereby possibly solving the anomalies of the CDM. The problem with WDM is
that different observations indicate mass limits which are inconsistent among them.
In particular:
• To solve the Core/Cusp problem is necessary a mass of ∼ 0.1 keV [Macciò

et al., 2012].
• To solve the Too big to fail problem is necessary a mass of ∼ 2 keV [Lovell
et al., 2012].
• Constraints from the Lyman-α observation require mWDM > 3.3 keV [Viel
et al., 2013].
11
These observational tensions disfavour WDM. In addition, it was shown in [Schnei-
der et al., 2014] that for mWDM > 3.3 keV WDM does not provide a real advantage
over CDM.
Interacting Dark Matter. CDM anomalies could perhaps be understood by

admitting the existence of self-interactions between dark matter particles [Spergel
and Steinhardt, 2000]. It has been shown that there are indications that interaction
models can alleviate the Core/Cusp and the Too big to fail problems [Vogelsberger
et al., 2014]. Recently, [Macciò et al., 2015] pointed out that a certain type of
interaction and mixing between CDM and WDM particles is very satisfactory from
the point of view of the resolution of the anomalies.
1.4.3 Other problems

Understanding the nature of DE and DM is the main open question of today
cosmology but here follows a small list of other open problems:
• The problem of the initial singularity, the so-called Big Bang. This issue is
related to a quantum formulation of gravity.
• There exists a couple of more technical, but nevertheless very important, is-
sues which are called tensions. These happen when observations of different
phenomena provide constraints on some parameters which are different up
to 68% or 95% confidence level. There is now tension between the determi-
nation of H0 via low-redshift probes and high-redshift ones (i.e. CMB). See
e.g. [Marra et al., 2013] and [Verde et al., 2013]. Moreover, there is also a
tension on the determination of σ8 [Battye et al., 2015], recently corroborated
by the analysis of the first year of the DES survey data collection [Abbott
et al., 2017d].
• Testing the cosmological principle and the copernican principle. See e.g.
[Valkenburg et al., 2014].
• The CMB anomalies [Schwarz et al., 2016]. These are unexpected (in the
sense of statistically relevant) features of the CMB sky.
• The Lithium problem [Coc, 2016]. The predicted Lithium abundance is much
larger than the observed one.
12
Chapter 2
The universe in expansion
oras ubicumque locaris extremas, quaeram: quid telo denique fiet?

(wherever you shall set the boundaries, I will ask: what will then
happen to the arrow?)
—Lucretius, De Rerum Natura
We introduce in this Chapter the geometric basis of cosmology and the ex-
pansion of the universe. A part from the technical treatment, historical, theolog-
ical and mythological introductions to cosmology can be found in [Ryden, 2003]
and [Bonometto, 2008].
2.1 Newtonian cosmology

In order to do cosmology we need a theory of gravity, because gravity is a long-range
interaction and the universe is pretty big. Electromagnetism is also a long-range
interaction, but considering the lack of evidence that the universe is charged or
made up of charges here and there, it seems reasonable that gravity is what we
need in order to describe the universe on large scales.
Which theory of gravity do we use for describing the universe? It turns out that
Newtonian physics works surprisingly well! It is also surprising that attempts of
doing cosmology with Newtonian gravity are well posterior to relativistic cosmology
itself.
In particular, the first work on Newtonian cosmology can be dated back to
Milne and McCrea in the 1930s [McCrea and Milne, 1934, Milne, 1934]. These
were models of pure dust, while pressure was introduced later by McCrea [McCrea,
1951] and Harrison [Harrison, 1965]. More recently the issue of pressure corrections
in Newtonian cosmology has been tackled again in [Lima et al., 1997], [Fabris and
Velten, 2012], [Hwang and Noh, 2013] and [Baqui et al., 2016].
Newtonian cosmology works as follows. Imagine a sphere of dust of radius r.
This radius is time-dependent because the configuration is not stable since there
is no pressure, thus r = r(t). We assume homogeneity of the sphere during the
evolution, i.e. its density depends only on the time:
3M
ρ(t) = , (2.1)
4πr(t)3
13
where M is the mass of the dust sphere and is constant. Now, imagine a small test
particle of mass m on the surface of the sphere. By Newton’s gravitation law and
Gauss’s theorem one has:
GM m 4πG
F =− ⇒ r̈ = − ρr , (2.2)
r(t)2 3
where we have used Eq. (2.1) and the dot denotes derivation with respect to the
time t. This is the same acceleration equation that we shall find later using GR,
cf. Eq. (2.51).
Exercise 2.1. Integrate Eq. (2.2) and show that:

ṙ2 8πG K
2
= ρ− 2 , (2.3)
r 3 r
where K is an integration constant.
We shall also see that Eq. (2.3) is the same as Friedmann equation in GR, cf.
(2.50). The integration constant K can be interpreted as the total energy of the
particle. Indeed, we can rewrite Eq. (2.3) as follows:
mK m GM m
E≡− = ṙ2 − , (2.4)
2 2 r
which is the expression of the total energy of a particle of mass m in the gravita-
tional field of the mass M .
2.2 Relativistic cosmology

In GR we have geometry and matter related by Einstein equations:
8πG
Gµν = Tµν , (2.5)
c4
where Gµν is the Einstein tensor, computed from the metric, and Tµν is the energy-
momentum or stress-energy tensor, and describes the matter content.
In cosmology, which is the metric which describes the universe and what is the
matter content? It turns out that both questions are very difficult to answer and,
indeed, there are no still clear answers, as we stressed in Chapter 1.
2.2.1 Friedmann-Lemaı̂tre-Robertson-Walker metric

The metric used to describe the universe on large scales is the Friedmann-Lemaı̂tre-
Robertson-Walker (FLRW) metric. This is based on the assumption of very high
symmetry for the universe, called the cosmological principle, which is minimally
stated as follows: the universe is isotropic and homogeneous, i.e. there is no
preferred direction or preferred position.
A more formal definition can be found in [Weinberg, 1972, pag. 412] and is
based on the following two requirements:
14
1. The hypersurfaces with constant cosmic standard time are maximally sym-
metric subspaces of the whole of the spacetime;
2. The global metric and all the cosmic tensors such as the stress-energy one
Tµν are form-invariant with respect to the isometries of those subspaces.
We shall come back in a moment to maximally symmetric spaces. Roughly speak-
ing, the second requirement above means that the matter quantities can depend
only on the time.
The cosmological principle seems to be compatible with observations at very
large scales. According to [Wu et al., 1999]: on a scale of about 100 h−1 Mpc the
rms density fluctuations are at the level of ∼10% and on scales larger than 300
h−1 Mpc the distribution of both mass and luminous sources safely satisfies the
cosmological principle of isotropy and homogeneity.
In a recent work [Sarkar and Pandey, 2016] find that the quasar distribution is
homogeneous on scales larger than 250 h−1 Mpc. Moreover, numerical relativity
seems to indicate that the average evolution of a generic metric on large scale is
compatible with that of FLRW metric [Giblin et al., 2016].
According to the cosmological principle, the constant-time spatial hypersurfaces
are maximally symmetric.1 A maximally symmetric space is completely charac-
terised by one number only, i.e. its scalar curvature, which is also a constant.
See [Weinberg, 1972, Chapter 13].
Let R be this constant scalar curvature. The Riemann tensor of a maximally
symmetric D-dimensional space is written as:
R
Rµνρσ = (gµρ gνσ − gµσ gνρ ) . (2.6)
D(D − 1)
Contracting with g µρ we get for the Ricci tensor:
R
Rνσ = gνσ , (2.7)
D
and then R is the scalar curvature, as we stated, since g νσ gνσ = D. Since any
given number can be negative, positive or zero, we have three possible maximally
symmetric spaces. Now, focusing on the 3-dimensional spatial case:
1. ds23 = |dx|2 ≡ δij dxi dxj , i.e. the Euclidean space. The scalar curvature is
zero, i.e. the space is flat. This metric is invariant under 3-translations and
3-rotations.
2. ds23 = |dx|2 + dz 2 , with the constraint z 2 + |x|2 = a2 . This is a 3-sphere of

radius a embedded in a 4-dimensional Euclidean space. It is invariant under
the six 4-dimensional rotations.
3. ds23 = |dx|2 − dz 2 , with the constraint z 2 − |x|2 = a2 . This is a 3-hypersphere,

or a hyperboloid, in a 4-dimensional pseudo-Euclidean space. It is invariant
under the six 4-dimensional pseudo-rotations (i.e. Lorentz transformations).
1
This means that they possess 6 Killing vectors, i.e. there are six transformations which leave
the spatial metric invariant [Weinberg, 1972].
15
Exercise 2.2. Why are there six independent 4-dimensional rotations in the 4-
dimensional Euclidean space? How many are there in a D-dimensional Euclidean
space?
Let us write in a compact form the above metrics as follows:
ds23 = |dx|2 ± dz 2 , z 2 ± |x|2 = a2 . (2.8)
Differentiating z 2 ± |x|2 = a2 , one gets:
zdz = ∓x · dx . (2.9)
Now put this back into ds23 :
(x · dx)2
ds23 = |dx|2 ± . (2.10)
a2 ∓ |x|2
In a more compact form:
(x · dx)2
ds23 = |dx|2 + K , (2.11)
a2 − K|x|2
with K = 0 for the Euclidean case, K = 1 for the spherical case and K = −1 for
the hyperbolic case. The components of the spatial metric in Eq. (2.11) can be
immediately read off and are:
(3) xi xj
gij = δij + K . (2.12)
a2 − K|x|2
Exercise 2.3. Write down metric (2.11) in spherical coordinates. Use the fact
that |dx|2 = dr2 + r2 dΩ2 , where
dΩ2 ≡ dθ2 + sin2 θdφ2 , (2.13)
and use:
1 1
x · dx = d|x|2 = d(r2 ) = rdr . (2.14)
2 2
Show that the result is:
a2 dr2
ds23 = + r2 dΩ2 (2.15)
a2 − Kr2
(3) (3)
Calculate the scalar curvature R(3) for metric (2.15). Show that Rij = 2Kgij /a2
and thus R(3) = 6K/a2 .
16
If we normalise r → r/a in metric (2.15), we can write:
dr2

2 2 2 2
ds3 = a + r dΩ , (2.16)
1 − Kr2
and letting a to be a function of time, we finally get the FLRW metric:
dr2

2 2 2 2
ds = −c dt + a (t) + r2 dΩ2 (2.17)
1 − Kr2
The time coordinate used here is called cosmic time, whereas the spatial co-
ordinates are called comoving coordinates. For each t the spatial slices are
maximally symmetric; a(t) is called scale factor, since it tells us how the distance
between two points scales with time.
The FLRW metric was first worked out by Friedmann in [Friedmann, 1922]
and [Friedmann, 1924] and then derived on the basis of isotropy and homogeneity
by Robertson and Walker in [Robertson, 1935], [Robertson, 1936] and [Walker,
1937]. Lemaı̂tre’s work [Lemaitre, 1931] had been also essential to develop it.2
A further comment concerning FLRW metric (2.17) is in order here. The di-
mension of distance is being carried by the scale factor a itself, since we rescaled
the radius r → r/a. Indeed, as we computed earlier, the spatial curvature is
R(3) = 6K/a(t)2 , also time-varying, and it is a real, dimensional number as it
should be.
2.2.2 The conformal time

A very useful form of rewriting FLRW metric (2.17) is via the conformal time η:
Z t
dt0
adη = dt ⇒ η − ηi = 0
. (2.18)
ti a(t )
As we shall see later, but as we already can guess from the above integration,
c(η − ηi ) represents the comoving distance travelled by a photon between the times
ηi and η, or ti and t. The conformal time allows to rewrite FLRW metric (2.17) as
follows:
dr2

2 2 2 2 2 2
ds = a(η) −c dη + + r dΩ (2.19)
1 − Kr2
i.e. the scale factor has become a conformal factor (hence the name for η). Re-
calling the earlier discussion about dimensionality, if a has dimensions then cη is
dimensionless. On the other hand, if a is dimensionless, then η is indeed a time.
Note also that metric (2.19) for K = 0 is Minkowski metric multiplied by a
conformal factor.
2
See also [Lemaı̂tre, 1997] for a recent republication and translation of Lemaı̂tre’s 1933 paper.
17
2.2.3 FLRW metric written with proper radius
A third useful way to write FLRW metric (2.17) is using the proper radius, which
is defined as follows:
D(t) ≡ a(t)r . (2.20)
We shall discuss in more detail the proper radius, or proper distance, in Sec. 2.5.
Exercise 2.4. Using D instead of r, show that the FLRW metric (2.17) becomes:
D2 /c2

2 2 2 2 2HDdtdD
ds = −c dt 1 − H 2 2
−
1 − KD /a 1 − KD2 /a2
dD2
+ + D2 dΩ2 , (2.21)
1 − KD2 /a2
where
ȧ
H≡ (2.22)
a
is the Hubble parameter. The dot denotes derivation with respect to the cosmic
time.
2.2.4 Light-cone structure of the FLRW space

Let us consider the K = 0 case, for simplicity. Moreover, consider also dΩ = 0. In
this case, the radial coordinate is also the distance. Then, putting ds2 = 0 in the
FLRW metric gives the following light-cone structures.
Cosmic time-comoving distance

From the FLRW metric (2.17), the condition ds2 = 0 gives us:
cdt
= ±a(t) . (2.23)
dr
We put our observer at r = 0 and t = t0 . The plus sign in the above equation
then describes an outgoing photon, i.e. the future light-cone, whereas the negative
sign describes an incoming photon, i.e. the past-light cone, which is much more
interesting to us. So, let us keep the negative sign and discuss the shape of the
light-cone.
Assume that a(0) = 0. Therefore, the slope of the past light-cone starts as
−a(t0 ), which we can normalise as −1, i.e. locally the past light-cone is identical
to the one in Minkowski space. However, a goes to zero, so the light-cone becomes
flat, encompassing more radii than it would for Minkowski space. See Fig. 2.1. We
can show this analytically by taking the second derivative of Eq. (2.23) with the
minus sign:
c2 d 2 t cdt
2
= −ȧ = aȧ . (2.24)
dr dr
18
Figure 2.1: Space-time diagram and light-cone structure for the FLRW metric
(2.17). Credit: Prof. Mark Whittle, University of Virginia.
Being a > 0 and ȧ > 0 (we consider just the case of an expanding universe), the
function t(r) is convex (i.e. it is “bent upwards”).
Conformal time-comoving distance

For the FLRW metric (2.19), the condition ds2 gives:
cdη
= ±1 . (2.25)
dr
The latter is exactly the same light-cone structure of Minkowski space. Indeed,
Friedmann metric written in conformal time and for K = 0 is Minkowski metric
multiplied by a conformal factor a(η). See Fig. 2.2.
Cosmic time-proper distance

In order to find the light-cone structure for the FLRW metric (2.21) with K = 0,
we need to solve the following equation:
c2 dt2 H 2 D2

2HD cdt
− 1 − − +1=0. (2.26)
dD2 c2 c dD
19
Exercise 2.5. Solve Eq. (2.26) algebraically for cdt/dD and show that:
−1
cdt HD
= ±1 . (2.27)
dD c
For t = t0 we have H(t0 ) > 0 and D = 0. Therefore, from Eq. (2.27) we

have that (cdt/dD)(t0 ) = ±1 and thus we must choose the minus sign in order to
describe the past light-cone. Going back in time, HD grows, until
HD
=1, (2.28)
c
for which cdt/dD diverges. This means that no signal can come from beyond this
distance D = c/H, which is the Hubble radius that we met in Chapter 1. See
Fig. 2.3. The lower part of this figure is explained as follows. First of all HD
becomes larger than 1, and this explains the change of sign of the slope of the
light cone. Then, H → ∞ for a → 0 (if we assume a model with Big Bang) and
therefore cdt/dD → 0. This is why the light-cone flattens close to t = 0 in Fig. 2.3.
2.2.5 Christoffel symbols and geodesics
20
21
Exercise 2.6. Assume K = 0 in metric (2.17), rewrite it in Cartesian coordinates
and calculate the Christoffel symbols. Show that:
aȧ H i
Γ000 = 0 , Γ00i = 0 , Γ0ij = δij , Γi0j = δj. (2.29)
c c
We now use these in the geodesic equation:

dP µ
+ Γµνρ P ν P ρ = 0 , (2.30)
dλ
where P µ ≡ dxµ /dλ is the four-momentum and λ is an affine parameter. For a
particle of mass m, one has λ = τ /m, where τ is the proper time. The norm of the
four-momentum is:
E2
P 2 ≡ gµν P µ P ν = − 2
+ p2 = −m2 c2 , (2.31)
c
where we have defined the energy and the physical momentum (or proper
momentum):
E2
≡ −g00 (P 0 )2 , p2 ≡ gij P i P j , (2.32)
c2
and the last equality of Eq. (2.31), which applies only to massive particles, comes
from:
ds2 m2 ds2
2
= 2
= −m2 c2 , (2.33)
dλ dτ
since, by definition, ds2 = −c2 dτ 2 . We have recovered above the well-known
dispersion relation of special relativity. The metric gµν used above is, in principle,
general. But, of course, we now specialise it to the FLRW one.
For a photon, m = 0 and E = pc. The time-component of the geodesic equation
is the following:
dP 0 aȧ
+ δij P i P j = 0 . (2.34)
dλ c
Introducing the proper momentum as defined in Eq. (2.32), one gets:
dp
+ Hp2 = 0 . (2.35)
dλ
Exercise 2.7. Solve Eq. (2.35) and show that p = E/c ∝ 1/a, i.e. the energy of
the photon is proportional to the inverse scale factor.
Therefore, we can write:

Eobs aem
= . (2.36)
Eem aobs
22
On the other hand the photon energy is E = hf , with f its frequency. Therefore:
aem Eobs fobs λem 1
= = = = . (2.37)
aobs Eem fem λobs 1+z
This is the relation between the redshift and the scale factor. We have connected
observation with theory. Usually, aobs = 1 and the above relation is simply written
as 1 + z = 1/a.
What does happen, on the other hand, to the energy of a massive particle? The
time-geodesic equation for massive particles is identical to the one for photons, but
the dispersion relation is different, i.e. E 2 = m2 c4 + p2 c2 . Therefore:
r
p 2 a2 c 2
E = m2 c4 + i 2i , (2.38)
a
where pi is some initial proper momentum, at the time ti and ai = a(ti ). For m = 0
we recover the result already obtained for photons. For massive particles the above
relation can be approximated as follows:
p2i a2i

2
E = mc 1 + 2 2 2 + . . . , (mc p) , (2.39)
2a m c
i.e. performing the expansion for small momenta which is usually done in special
relativity. The second contribution between parenthesis is the classical kinetic
energy of the particle, whose average is proportional to kB T . Therefore:
T ∝ a−1 , for relativistic particles, (2.40)

T ∝ a−2 , for non-relativistic particles. (2.41)
We shall recover the above result also using the Boltzmann equation.
Exercise 2.8. Show that p ∝ 1/a by using not the time-component geodesic
equation but the spatial one:
dP i
+ 2Γi0j P 0 P j = 0 . (2.42)
dλ
Why is there a factor two in this equation?
2.3 Friedmann equations

Given FLRW metric, Friedmann equations can be straightforwardly computed from
the Einstein equations:
1 8πG
Gµν + Λgµν = Rµν − gµν R + Λgµν = 4 Tµν , (2.43)
2 c
where Λ is the cosmological constant.
23
Exercise 2.9. Calculate from FLRW metric (2.17) the components of the Ricci
tensor. Show that:
Kc2

3 ä 1 2 ä
R00 = − 2 , R0i = 0 , Rij = 2 gij 2H + + 2 2 , (2.44)
c a c a a
and show that the scalar curvature is:
Kc2

6 ä 2
R= 2 +H + 2 . (2.45)
c a a
Finally, compute the Einstein equations:
Kc2 8πG Λc2

H2 += T00 + (2.46)
a2 3c2 3
ä Kc2

8πG
gij H 2 + 2 + 2 − Λc2 = − 2 Tij (2.47)
a a c
These are called Friedmann equations or Friedmann equation and acceler-

ation equation or Friedmann equation and Raychaudhuri equation.
Which stress-energy tensor Tµν do we use in Eqs. (2.46) and (2.47)? Having
fixed the metric to be the FLRW one, we have some strong constraints:
• First of all: G0i = 0 implies that T0i = 0, i.e. there cannot be a flux of energy
in any direction because it would violate isotropy;
• Second, since Gij ∝ gij , then Tij ∝ gij .
• Finally, since Gµν depends only on t, then it must be so also for Tµν .
Therefore, let us stipulate that
T00 = ρ(t)c2 = ε(t) , T0i = 0 , Tij = gij P (t) , (2.48)
where ρ(t) is the rest mass density, ε(t) is the energy density and P (t) is the
pressure. In tensorial notation we can write the following general form for the
stress-energy tensor:

P
Tµν = ρ + 2 uµ uν + P gµν (2.49)
c
where uµ is the four-velocity of the fluid element. In this form of Eq. (2.49), the
stress-energy tensor does not contain either viscosity or energy transport terms.
Matter described by (2.49) is known as perfect fluid. For more detail about the
latter see [Schutz, 1985] whereas for more detail about viscosity, heat fluxes and
the imperfect fluids see e.g. [Weinberg, 1972] and [Maartens, 1996].
Combine Eqs. (2.46), (2.47) and (2.48). The Friedmann equation becomes:
8πG Λc2 Kc2

H2 = ρ+ − 2 (2.50)
3 3 a
24
while the acceleration equation is the following:
Λc2

ä 4πG 3P
=− ρ+ 2 + (2.51)
a 3 c 3
Exercise 2.10. Write Eqs. (2.50) and (2.51) using the conformal time introduced
in Eq. (2.18). Show that the Friedmann equation becomes:
8πG 2 Λc2 a2
H2 = ρa + − Kc2 (2.52)
3 3
and that the acceleration equation becomes:
a00 2Λc2 a2

4πG 3P
= ρ − 2 a2 + − Kc2 , (2.53)
a 3 c 3
where the prime denotes derivation with respect to the conformal time η and
a0
H≡ (2.54)
a
is the conformal Hubble factor.
In the Friedmann and acceleration equations, ρ and P are the total density and
pressure. Hence, they can be written as sums of the contributions of the individual
components: X X
ρ≡ ρx , P ≡ Px . (2.55)
x x
The contribution from the cosmological constant can be considered either geomet-
rically or as a matter component with the following density and pressure:
Λc2
ρΛ ≡ , PΛ ≡ −ρΛ c2 . (2.56)
8πG
The scale factor a is, by definition, positive, but its derivative can be negative.
This would represent a contracting universe. Note that the left hand side of the
Friedmann equation (2.50) is non-negative. Therefore, ȧ can vanish only if K > 0,
i.e. for a spatially closed universe. This implies that, if K 6 0 and if there exists
an instant for which ȧ > 0, then the universe will expand forever.
2.3.1 The Hubble constant and the deceleration parameter

When the Hubble parameter H is evaluated at the present time t0 , it becomes
a number: the Hubble constant H0 which we already met in Chapter 1 in the
Hubble’s law (1.1). Its value is
H0 = 67.74 ± 0.46 km s−1 Mpc−1 , (2.57)
25
at the 68% confidence level, as reported by the Planck group [Ade et al., 2016a].
Usually H0 is conveniently written as
H0 = 100 h km s−1 Mpc−1 . (2.58)
The unit of measure of the Hubble constant is an inverse time:
H0 = 3.24 h × 10−18 s−1 , (2.59)
whose inverse gives the order of magnitude of the age of the universe:
1
= 3.09 h−1 × 1017 s = 9.78 h−1 Gyr , (2.60)
H0
and multiplied by c gives the order of magnitude of the size of the visible universe,
i.e. the Hubble radius that we have already seen in Eq. (2.28) but evaluated at the
present time t = t0 :
c
= 9.27 h−1 × 1025 m = 3.00 h−1 Gpc . (2.61)
H0
But what does “present time” t0 mean? Time flows, therefore t0 cannot be a
constant! That is true, but if we compare a time span of 100 years (the span
of some human lives) to the age of the universe (about 14 billion years), we see
that the ratio is about 10−8 . Since this is pretty small, we can consider t0 to be a
constant, also referred to as the age of the universe.3 We can calculate it as follows:
Z t0 Z 1 Z 1 Z ∞
da da dz
t0 = dt = = = . (2.62)
0 0 ȧ 0 H(a)a 0 H(z)(1 + z)
Exercise 2.11. Prove the last equality of Eq. (2.62).
The integration limits of Eq. (2.62) deserve some explanation. We assumed that
a(t = 0) = 0, i.e. the Big Bang. This condition is not always true, since there are
models of the universe, e.g. the de Sitter universe, for which a vanishes only when
t → −∞. The other assumption is that a(t0 ) = 1. This is a pure normalisation,
done for convenience, which is allowed by the fact that the dynamics is invariant
if we multiply the scale factor by a constant.
Recall that, in cosmology, when a quantity has subscript 0, it usually means
that it is evaluated at t = t0 .
3
Pretty much the same happens with the redshift. A certain source has redshift z which,
actually, is not a constant but varies slowly. This is called redshift drift and it was first
considered by Sandage and McVittie in [Sandage, 1962] and [McVittie, 1962]. Applications of
the redshift drift phenomenon to gravitational lensing are proposed in [Piattella and Giani, 2017].
26
The deceleration parameter
Let us focus now on Eq. (2.51). It contains ä, so it describes how the expansion of
the universe is accelerating. The key-point is that if the right hand side of Eq. (2.51)
is positive, i.e. ρ + 3P/c2 < 0, then ä > 0. There exists a parameter, named
deceleration parameter, with which to measure the entity of the acceleration.
It is defined as follows:
äa
q≡− 2 (2.63)
ȧ
In [Riess et al., 1998] and [Perlmutter et al., 1999] analysis based on type Ia
supernovae observation have shown that q0 < 0, i.e. the deceleration parameter
is negative and therefore the universe is in a state of accelerated expansion. We
perform a similar but simplified analysis in Sec. 11.1 in order to illustrate how data
in cosmology are analysed.
2.3.2 Critical density and density parameters

Let us now rewrite Eq. (2.50) incorporating Λ in the total density ρ:
8πGρ Kc2
H2 = − 2 . (2.64)
3 a
The value of the total ρ such that K = 0 is called critical energy density and
has the following form:
3H 2
ρcr ≡ (2.65)
8πG
Its present value [Ade et al., 2016a] is:
ρcr,0 = 1.878 h2 × 10−29 g cm−3 (2.66)
It turns out that ρ0 is very close to ρcr,0 , so that our universe is spatially flat.
Such an extreme fine-tuning in K is a really surprising coincidence, known as the
flatness problem. A possible solution is provided by the inflationary theory
which we shall see in detail in Chapter 8.
Instead of densities, it is very common and useful to employ the density pa-
rameter Ω, which is defined as
ρ 8πGρ
Ω≡ = (2.67)
ρcr 3H 2
i.e. the energy density normalised to the critical one. We can then rewrite Fried-
mann equation (2.50) as follows:
Kc2
1=Ω− . (2.68)
H 2 a2
Defining
Kc2
ΩK ≡ − , (2.69)
H 2 a2
27
i.e. associating the energy density
3Kc2
ρK ≡ − , (2.70)
8πGa2
to the spatial curvature, we can recast Eq. (2.68) in the following simple form:
1 = Ω + ΩK . (2.71)
Therefore, the sum of all the density parameters, the curvature one included, is
always equal to unity. In particular, if it turns out that Ω ' 1, this implies that
ΩK ' 0, i.e. the universe is spatially flat. From the latest Planck data [Ade et al.,
2016a] we know that:
ΩK0 = 0.0008+0.0040
−0.0039 (2.72)
at the 95% confidence level.
It is more widespread in the literature the normalisation of ρ to the present-time
critical density, i.e.
ρ 8πGρ
Ω≡ = (2.73)
ρcr,0 3H02
because it leaves more evident the dependence on a of each material component.
With this definition of Ω, Friedmann equation (2.50) is written as:
H2 X ΩK0
2
= Ωx0 fx (a) + 2 , (2.74)
H0 x
a
where fx (a) is a function which gives the a-dependence of the material component
x and fx (a0 = 1) = 1. Consistently:
X
Ωx0 + ΩK0 = 1 (2.75)
x
also known as closure relation. We shall use the definition Ωx ≡ ρx /ρcr,0 through-
out these notes.
2.3.3 The energy conservation equation

The energy conservation equation
∇ν T µν = 0 (2.76)
is encapsulated in GR through the Bianchi identities. Therefore, it is not indepen-
dent from the Friedmann equations (2.50) and (2.51). For the FLRW metric and
a perfect fluid, it has a particularly simple form:

P
ρ̇ + 3H ρ + 2 = 0 (2.77)
c
This is the µ = 0 component of ∇ν T µν = 0 and it is also known from fluid dynamics

as continuity equation.
28
Exercise 2.12. Derive the continuity equation (2.77) by combining Friedmann and
acceleration equations (2.50) and (2.51). Derive it in a second way by explicitly
calculating the four-divergence of the energy-momentum tensor.
The continuity equation can be analytically solved if we assume an equation of

state of the form P = wρc2 , with w constant. The general solution is:
ρ = ρ0 a−3(1+w) (w = constant) , (2.78)
where ρ0 ≡ ρ(a0 = 1).
Exercise 2.13. Prove the above result of Eq. (2.78).
There are three particular values of w which play a major role in cosmology:
Cold matter: w = 0, i.e. P = 0, for which ρ = ρ0 a−3 . As we have discussed

in Chapter 1, the adjective cold refers to the fact that particles making up this
kind of matter have a kinetic energy much smaller than the mass energy, i.e. they
are non-relativistic. If they are thermally produced, i.e. if they were in thermal
equilibrium with the primordial plasma, they have a mass much larger than the
temperature of the thermal bath. We shall see this characteristic in more detail in
Chapter 3.
Cold matter is also called dust and it encompasses all the non-relativistic known
elementary particles, which are overall dubbed baryons in the jargon of cosmology.
If they exist, unknown non-relativistic particles are called cold dark matter
(CDM).
Hot matter: w = 1/3, i.e. P = ρ/3, for which ρ = ρ0 a−4 . The adjective hot
refers to the fact that particles making up this kind of matter are relativistic.
For this reason they are known, in the jargon of cosmology, as radiation and
they encompass not only the relativistic known elementary particles, but possibly
the unknown ones (i.e. hot dark matter). The primordial neutrino background
belonged to this class, but since neutrino seems to have a mass of approximately
0.1 eV, it is now cold. We shall see why in Chapter 3.
Vacuum energy: w = −1, i.e. P = −ρ and ρ is a constant. It behaves as the

cosmological constant and provides the best (and the simplest) description that we
have for dark energy, though plagued by the serious issues that we have presented
in Chapter 1.
2.3.4 The ΛCDM model

The most successful cosmological model is called ΛCDM and is made up of Λ,
CDM, baryons and radiation (photons and massless neutrinos). The Friedmann
29
equation for the ΛCDM model is the following:
H2 Ωc0 Ωb0 Ωr0 ΩK0

2
= ΩΛ + 3 + 3 + 4 + 2 . (2.79)
H0 a a a a
We already saw in Eq. (2.72) the value of the spatial curvature contribution. From
Ref. [Ade et al., 2016a] here are the other ones:
ΩΛ = 0.6911 ± 0.0062 , Ωm0 = 0.3089 ± 0.0062 (2.80)
at 68% confidence level, where Ωm0 = Ωc0 + Ωb0 , i.e. it includes the contributions
from both CDM and baryons, since they have the same dynamics (i.e. they are
both cold). It is however possible to disentangle them and one observes:
Ωb0 h2 = 0.02230 ± 0.00014 , Ωc0 h2 = 0.1188 ± 0.0010 (2.81)
also at 68% confidence level. The radiation content, i.e. photons plus neutrinos,
can be easily calculated from the temperature of the CMB, as we shall see in
Chapter 3. It turns out that:
Ωγ0 h2 ≈ 2.47 × 10−5 , Ων0 h2 ≈ 1.68 × 10−5 (2.82)
Since h = 0.68, and recalling the closure relation of Eq. (2.75), we can conclude
that today 69% of our universe is made of cosmological constant, 26% of CDM
and 5% of baryons. Radiation and spatial curvature are negligible. That is, the
situation is pretty obscure, in all senses.
Let us now calculate the age of the universe for the ΛCDM model. Using
Eq. (2.62), we get:
Z 1
1 a
t0 = da √ . (2.83)
H0 0 ΩΛ a + Ωm0 a + Ωr0 + ΩK0 a2
4
Using the numbers shown insofar, we get upon numerical integration:
0.95
t0 = = 13.73 Gyr (2.84)
H0
The value reported by [Ade et al., 2016a] is 13.799 ± 0.021 at 68% confidence
level. Note how H0 t0 ≈ 1. This fact has been dubbed synchronicity problem
by [Avelino and Kirshner, 2016]. In Fig. 2.4 we plot the dimensionless age of the
universe H0 t0 for models with or without Λ and in Fig. 2.5 we plot the evolution
of H0 t as function of a in order to show indeed how H0 t0 ≈ 1 is quite a peculiar
instant of the history of the universe.
As one can see, in presence of Λ the dimensionless age of the universe reaches
values larger than unity. This, mathematically, is due to the a4 factor multiplying
ΩΛ in Eq. (2.83). Note that we can obtain the observed value H0 t0 ≈ 0.95 also
in absence of a cosmological constant and for a curvature-dominated universe, i.e.
ΩK0 ≈ 0.97.
30
��
��
��
��
��
Figure 2.4: Dimensionless age of the universe H0 t0 as function of ΩΛ (keeping

fixed the matter and radiation content) and as function of Ωm0 (keeping fixed the
radiation content and with no Λ).
�
� � � � � �
Figure 2.5: Dimensionless age of the universe H0 t as function of a for the ΛCDM
model (solid line) and in a model made only of CDM (dashed line).
31
2.4 Solutions of the Friedmann equations
The Friedmann equations can be solved exactly for many cases of interest.
2.4.1 The Einstein Static Universe

As the first application of his theory to cosmology, Einstein was looking for a static
universe, since at his time there was not yet compelling evidence of the contrary.
Therefore, we must set ȧ = ä = 0. Since ρ is positive, we must have K = 1,
therefore the Einstein Static Universe (ESU) is a closed universe. Its radius is:
s
2
8πG c 3c2
ρ= 2 ⇒ a= . (2.85)
3 a 8πGρ
From the acceleration equation we get that
ρ + 3P/c2 = 0 , (2.86)
therefore we cannot have simply ordinary matter because we need a negative pres-
sure. Here enters the cosmological constant Λ. We assume that ρ = ρm + ρΛ , so
that
ρ + 3P/c2 = 0 , ⇒ ρm + ρΛ − 3ρΛ = 0 , (2.87)
and therefore ρm = 2ρΛ . The radius can thus be written as
c 1
a= √ =√ . (2.88)
4πGρm Λ
Until here all seems to be fine. But it is not. The problem is indeed the condition
ρm = 2ρΛ , which makes the ESU unstable. In fact, if this condition is broken, say
ρm /ρΛ = 2 + , then the universe expands or collapses, depending on the sign of .
Exercise 2.14. Prove that the ESU is unstable. Hint: use ρm /ρΛ = 2 + in the
Friedmann and acceleration equations.
2.4.2 The de Sitter universe

For ρ = 0, the Friedmann equation (2.50) becomes:
Λc2 Kc2
H2 = − 2 . (2.89)
3 a
When spatial curvature is taken into account, it is more convenient to solve the
acceleration equation (2.51) rather than Friedmann equation. Indeed:
Λc2
ä = a, (2.90)
3
32
is straightforwardly integrated:
r ! r !
Λ Λ
a(t) = C1 exp ct + C2 exp − ct , (2.91)
3 3
where C1 and C2 are two integration constants. One of these is constrained by

Friedmann equation (2.89). Calculating ȧ2 and a2 from Eq. (2.91) we get:
" r ! r ! #
2
Λc Λ Λ
ȧ2 = C12 exp 2 ct + C22 exp −2 ct − 2C1 C2 , (2.92)
3 3 3
r ! r !
Λ Λ
a2 = C12 exp 2 ct + C22 exp −2 ct + 2C1 C2 . (2.93)
3 3
Combining them, one finds:
Λc2 2
ȧ2 = (a − 4C1 C2 ) . (2.94)
3
Using Eq. (2.89), we can put a constraint on the product of the two integration
constants:
4Λ
C1 C2 = K , (2.95)
3
so that we have freedom to fix just one of them. Assuming that C1 6= 0, we can
write the general solution as:
r ! r !
Λ 3K Λ
a(t) = C1 exp ct + exp − ct . (2.96)
3 4ΛC1 3
When K = ±1 we can set C1 such that we can write the solution as follows:
p p


 3/Λ sinh Λ/3ct , for K = −1 ,
 p
a(t) = a0 exp Λ/3ct , for K = 0 , (2.97)

 p p
 3/Λ cosh

Λ/3ct , for K = 1 .
where a0 is some initial t = 0 scale factor. Note that there is no a0 for the solutions
with spatial curvature because we fixed the integration constant in order to have
the hyperbolic sine and cosine.
Exercise 2.15. From the Einstein equations (2.43) show that R = 4Λ for the
de Sitter universe. Verify that the above solutions (2.96) and (2.97) satisfy this
relation by substituting them into the expression in Eq. (2.45) for the Ricci scalar.
The de Sitter universe [de Sitter, 1917], [de Sitter, 1918c], [de Sitter, 1918b],
[de Sitter, 1918a] is eternal with no Big-Bang (i.e. when a = 0) for K = 1. Here
we have rather a bounce at the minimum value a0 for the scale factor. For K = 0
33
there is a Big-Bang at t = −∞. For K = −1 we might have negative scale factors,
which we however neglect and consider the evolution as starting only at t = 0, for
which there is another Big-Bang. Note that the Big-Bang’s we are mentioning here
are not singularities. There is no physical singularity in the de Sitter space, since
it is maximally symmetric.
The deceleration parameter is the following
−1
Λc2 a2

äa 3K
q=− 2 =− =− 1− , (2.98)
ȧ 3 ȧ2 Λa2
i.e. always negative.
2.4.3 Radiation-dominated universe

For ρ = ρ0 a−4 , K = 0 and Λ = 0, the solution of Eq. (2.50) is:
r
t
a= , (2.99)
t0
The deceleration parameter is q0 = 1 and the age of the universe is:
1
t0 = . (2.100)
2H0
Exercise 2.16. Prove the results of Eqs. (2.99) and (2.100).
It is quite complicated to analytically solve Friedmann equation (2.50) for a

radiation-dominated universe when K 6= 0. On the other hand, solving the accel-
eration equation (2.53) is much easier. For ρc2 = 3P , Eq. (2.53) becomes:
a00 + Kc2 a = 0 , (2.101)
whose general solution is:


C1 exp(cη) + C2 exp(−cη) ,
 for K = −1 ,
a(η) = C3 + C4 η , for K = 0 , (2.102)

C5 sin(cη) + C6 cos(cη) , for K = 1 .

First of all, we can choose a(0) = 0. Thus, the general solution (2.102) becomes:

2C1 sinh(cη) ,
 for K = −1 ,
a(η) = C4 η , for K = 0 , (2.103)

C5 sin(cη) , for K = 1 .

Second, these solutions are subject to the constraint of Friedmann equation (2.52),
written of course in the radiation-dominated case:
8πG 4
a02 = ρa − Kc2 a2 , (2.104)
3
34
where notice that ρa4 = ρ0 , i.e. a constant. When a = 0, i.e. η = 0, then
8πG
a02 (η = 0) = ρ0 ≡ a2m c2 , (2.105)
3
and the solutions (2.103) become:

sinh(cη) ,
 for K = −1 ,
a(η) = am cη , for K = 0 , (2.106)

sin(cη) , for K = 1 .

If we want to recover the cosmic time from the above solutions, we need to solve
the following integration: Z η
a(η 0 )dη 0 = t . (2.107)
0
Using Eq. (2.106), one obtains:


cosh(cη) − 1 ,
 for K = −1 ,
2
ct = am (cη) /2 , (2.108)
for K = 0 ,

1 − cos(cη) , for K = 1 .

Inverting these relations allows you to find η = η(t), which once substituted in
Eq. (2.106) allows to find a = a(t).
Exercise 2.17. Using the solutions (2.108), find the explicit form of a(t). Show
that ct = am η 2 /2 leads to Eq. (2.99).
2.4.4 Cold matter-dominated universe

For ρ = ρ0 a−3 , K = 0 and Λ = 0, the solution of Friedmann equation (2.50) is
straightforwardly obtained:
2/3
t
a= . (2.109)
t0
The deceleration parameter is q0 = 1/2 and the age of the universe is:
2
t0 = = 6.52 h−1 Gyr . (2.110)
3H0
This model of universe is also known as the Einstein-de Sitter universe. A part
the fact that it does not predict any accelerated expansion, there are also problems
with the age of the universe given in Eq. (2.110): it is smaller than the one of some
globular clusters [Velten et al., 2014].
Exercise 2.18. Prove the results of Eqs. (2.109) and (2.110).
35
Worse than the radiation-dominated case, it is impossible to analytically solve
Friedmann equation (2.50) for a dust-dominated universe when K 6= 0. But, as in
the radiation-dominated case, it is possible to find an exact solution for a(η). Let’s
write Eq. (2.53) for the dust-dominated case:
4πG 3
a00 = ρa − Kc2 a . (2.111)
3
Note that ρa3 = ρ0 = constant. The general solution is therefore the general
solution of Eq. (2.101) plus a particular solution of Eq. (2.111), that is:

4πG
C1 sinh(cη) + C2 cosh(cη) − 3c2 ρ0 ,
 for K = −1 ,
a(η) = C3 + C4 η + 2πG 3
ρ0 η 2 , for K = 0 , (2.112)
 4πG
C5 sin(cη) + C6 cos(cη) + 3c2 ρ0 , for K = 1 .

Exercise 2.19. Using the condition a(0) = 0 and employing Friedmann equation
for a dust-dominated universe, i.e.
8πG 4
a02 + Kc2 a2 = ρa , (2.113)
3
as constraint, show that Eq. (2.112) can be cast as:

cosh(cη) − 1 , for K = −1 ,
4πG  2
a(η) = ρ0 (cη) /2 , for K = 0 , (2.114)
3c2 
1 − cos(cη) , for K = 1 .

Recovering the cosmic time from Eq. (2.114) one has:


sinh(cη) − cη , for K = −1 ,
4πG  3
ct = ρ0 (cη) /6 , for K = 0 , (2.115)
3c2 
cη − sin(cη) , for K = 1 .

Unfortunately, the above relations for K = ±1 cannot be explicitly inverted in

order to give η(t) and then a(t).
2.4.5 Radiation plus dust universe

The mixture of radiation plus matter is a cosmological model closer to reality and
with which we can describe the evolution of our universe on a larger timespan than
the single component-dominated cases. Consider the total density:
ρeq a3eq ρeq a4eq

ρ = ρm + ρr = + , (2.116)
2 a3 2 a4
where aeq is the equivalence scale factor, i.e. the scale factor evaluated at the time
at which dust and radiation densities were equal. At this time, we dub the total
36
density as ρeq . Now write down the acceleration equation (2.53) for the dust plus
radiation model:
4πG
a00 = ρm a3 − Kc2 a . (2.117)
3
It is identical to the dust-dominated case, viz. Eq. (2.111)! Indeed, the fact that
radiation is also present will enter when we set the constraint from Friedmann
equation, which is the following:
4πGρeq 3
a02 + Kc2 a2 = aeq a + a4eq .

(2.118)
3
Solving Eq. (2.117) with the condition a(0) = 0 leads to the following solutions:

C sinh(cη) + cosh(cη) − 1 , for K = −1 ,
2πGρeq a3eq  1 2
a(η) = C2 η + (cη) /2 , for K = 0 , (2.119)
3c2 
C3 sin(cη) + 1 − cos(cη) , for K = 1 .

Now use the constraint from Friedmann equation, i.e.

4πGρeq 4
a02 (η = 0) = aeq , (2.120)
3
and find that:
s
3
C1 = C3 = c ≡ cη̃ , C2 = c2 η̃ , (2.121)
πGρeq a2eq
so that:

cη̃ sinh(cη) + cosh(cη) − 1 , for K = −1 ,
2aeq  2 2
a(η) = 2 2 c η̃η + (cη) /2 , for K = 0 , (2.122)
c η̃ 
cη̃ sin(cη) + 1 − cos(cη) , for K = 1 .

In particular, the solution for K = 0 is:

η η2

a(η) = aeq 2 + 2 . (2.123)
η̃ η̃
Exercise 2.20. Show that the conformal time at equivalence ηeq and η̃ are related
by: √
ηeq = ( 2 − 1)η̃ . (2.124)
Exercise 2.21. Solve Friedmann equation for the ΛCDM model, neglecting radi-
ation and spatial curvature:
H2 Ωm0
2
= 3 + ΩΛ . (2.125)
H0 a
Show that: p 1/3
Ωm0 3
a(t) = sinh2 ΩΛ H0 t . (2.126)
ΩΛ 2
37
Exercise 2.22. Solve the Friedmann equation for the curvature-dominated uni-
verse:
Kc2
H2 = − 2 . (2.127)
a
This is the Milne model [Milne, 1935]. Clearly, only K = −1 is allowed.
Show then that a = ct. Substitute this solution into the expression for the Ricci
scalar (2.45). Show that R = 0.
Write down explicitly the FLRW metric with a = ct and show that it is
Minkowski metric written in a coordinate systems different from the usual.
The above last result is not completely surprising, since Milne model has no
matter (empty universe) and no cosmological constant. The spatial hypersurfaces
are already maximally symmetric because of the cosmological principle and the
absence of matter add even more symmetry to the spacetime.
2.5 Distances in cosmology

We present and discuss in this section the various notions of distance that are
employed in cosmology. See e.g. [Hogg, 1999] for a reference on the subject.
2.5.1 Comoving distance and proper distance

We have already encountered comoving coordinates in the FLRW metric (2.17) and
the proper radius D(t) ≡ a(t)r in the FLRW metric (2.21). We must be clearer
about the difference between the radial coordinate and the distance. They are
equal only when dΩ = 0. The comoving square infinitesimal distance is indeed,
from FLRW metric (2.17) the following:
dr2
dχ2 = + r2 dΩ2 , (2.128)
1 − Kr2
i.e. it has indeed a radial part, but also has a transversal part. So, if χ is the
comoving distance between two points, the proper distance at a certain time t is
d(χ, t) = a(t)χ.
The comoving distance is a notion of distance which does not include the ex-
pansion of the universe and thus does not depend on time.
The proper distance is the distance that would be measured instantaneously
by rulers. For example, imagine to extend a ruler between GN-z11 (the farthest
known galaxy, z = 11.09) and us. Our reading at the time t would be the proper
distance at that time.
Suppose that dΩ = 0. Then the comoving distance to an object with radial
coordinate r is the following:

Z r
dr 0 arcsin r , for K = 1 ,

χ= 02 = r , for K = 0 , (2.129)
0 1 − Kr 
arcsinh r , for K = −1 .

38
Deriving d with respect to the time one gets:
ȧ
d˙ = ȧχ = d = Hd , (2.130)
a
which recovers the Hubble’s law for t = t0 .
2.5.2 The lookback time

Imagine a photon emitted by a galaxy at a time tem and detected at the time t0
on Earth. A very basic notion of distance is c(t0 − tem ), i.e. it is the light-travel
distance, based on the fact that light always travels with speed c. The quantity
t0 − tem is called lookback time and suggestively reminds the fact that when we
observe some source in the sky we are actually looking into the past, because of
the finiteness of c.
From the FLRW metric, by putting ds2 = 0, we can relate the lookback time
with the comoving distance as follows:
cdt = a(t)dχ . (2.131)
This seems quite similar to the proper distance, but careful: the proper distance
is defined as aχ and evidently adχ 6= d(aχ). The lookback time is the photon
time of flight and thus it includes cumulatively the expansion of the universe.
On the other hand, the proper distance is the distance considered between two
simultaneous events and therefore the expansion of the universe is not taken into
account cumulatively.
Since we observe redshifts, is there a way to calculate the lookback time from
z? In principle yes: one solves Friedmann equation, finds a(t), inverts this function
in order to find t = t(a), uses 1 + z = 1/a and finally gets a relation t = t(z). For
example, for the flat Einstein-de Sitter universe, using Eqs. (2.109) and (2.110) one
gets:
2/3
2 2
1+z = ⇒ t= . (2.132)
3H0 t 3H0 (1 + z)3/2
This approach is model-dependent because in order to solve the Friedmann equa-
tion we must know it and this is possible only if we know, or model, the energy
content of the universe. Hence the model-dependence.
A model-independent way of relating lookback time and redshift is cosmog-
raphy, a word which means “measuring the universe”. In practice, cosmography
consists in a Taylor expansion of the scale factor about its today value:
da 1 d2 a
a(t) = a(t0 ) + (t − t0 ) + (t − t0 )2 + · · · (2.133)
dt t0 2 dt2 t0
where we stop at the second order, for simplicity. This can be written as

1 2 2
a(t) = a(t0 ) 1 + H0 (t − t0 ) − q0 H0 (t − t0 ) + · · · , (2.134)
2
i.e. the first coefficient of the expansion is the Hubble constant, whereas the second
one is proportional to the deceleration parameter. The third is usually called
39
jerk and the fourth snap. All these parameters are evaluated at t0 in the above
expansion.
Exercise 2.23. For a(t0 ) = 1 and introducing the redshift show that:
1
z ∼ H0 (t0 − t) + (q0 + 2)H02 (t0 − t)2 . (2.135)
2
Here is a direct, model-independent relation between the redshift and the look-
back time (t0 − t).
2.5.3 Distances and horizons

For a photon, not unexpectedly,
cdt
dχ = = cdη , (2.136)
a(t)
i.e. the comoving distance is equal to the conformal time, which we introduced in
Eq. (2.18). We might say that the comoving distance is a lookback conformal time.
By integrating cdt/a(t) from tem to t0 we get the comoving distance from the
source to us, or the conformal time spent by the photon travelling from the source
to us: Z t0 Z 1
cdt0 cda0
χ= 0
= 0 02
. (2.137)
tem a(t ) a H(a )a
For the dust-dominated case one has H = H0 /a3/2 and the comoving distance as
a function of the scale factor and of the redshift is:
Z 1 0
√

c da 2c 2c 1
χ(a) = √ = 1 − a , χ(z) = 1− √ . (2.138)
H0 a a0 H0 H0 1+z
When z → 0, χ ∼ cz/H0 . Comparing with Eq. (2.135) one sees that, at the first
order in the redshift, the lookback time distance is equivalent to the comoving one.
Exercise 2.24. Calculate the comoving distance as a function of the scale factor
and of the redshift for a radiation-dominated universe and for the de Sitter universe.
When the lower integration limit in Eq. (2.137) is a = 0, i.e. the Big Bang, one
defines the comoving horizon χp (also known as particle horizon or cosmo-
logical horizon). This is the conformal time spent from the Big Bang until the
cosmic time t or scale factor a. It is also the maximum comoving distance travelled
by a photon (hence the name particle horizon) since the Big Bang and so it is the
comoving size of the visible universe.
In the dust-dominated case, using Eq. (2.138), with a = 0 or z = ∞ one obtains:
2c
χp = cη0 = . (2.139)
H0
40
Note that this is not the age of the universe given in Eq. (2.110), but three times
its value.
When the upper integration limit of Eq. (2.137) is infinite, one defines the
event horizon: Z ∞ Z ∞
dt0 da0
χe (t) ≡ c = c , (2.140)
t a(t0 ) a H(a0 )a0 2
which of course makes sense only if the universe does not collapse. This represents
the maximum distance travelled by a photon from a time t. If it diverges, then no
event horizon exists and therefore eventually all the events in the universe will be
causally connected. This happens, for example, in the dust-dominated case:
Z ∞ 0
c da
χe = √ =∞. (2.141)
H0 a a0
But, in the de Sitter universe we have
Z ∞ 0
c da c
χe = 02 = . (2.142)
H0 a a H0 a
The proper event horizon for the de Sitter universe is a constant:
c
aχe = . (2.143)
H0
2.5.4 The luminosity distance

The luminosity distance is a very important notion of distance for observation. It is
based on the knowledge of the intrinsic luminosity L of a source, which is therefore
called standard candle. Type Ia supernovae are standard candles, for example.
Then, measuring the flux F of that source and dividing L by F , one obtains the
square luminosity distance:
L
d2L ∝ . (2.144)
F
Now, imagine a source at a certain redshift z with intrinsic luminosity L = dE/dt.
The observed flux is given by the following formula:
dE0
F = , (2.145)
dt0 A0
where A0 is the area of the surface on which the radiation is spread:
A0 = 4πa20 χ2 , (2.146)
i.e. over a sphere with the proper distance as the radius. We must use the proper
distance, because this is the instantaneous distance between source and observer
at the time of detection. Note that χ is the comoving distance between the source
and us.
We do not observe the same photon energy as the one emitted, because photons
suffer from the cosmological redshift, thus:
dE a0
= . (2.147)
dE0 a
41
Finally, the time interval used at the source is also different from the one used at
the observer location:
dt a
= . (2.148)
dt0 a0
We can easily show this by using FLRW metric with ds2 = 0, i.e. cdt = a(t)dχ.
Consider the same dχ at the source and at the observer’s location. Thus, cdt =
a(t)dχ and cdt0 = a(t0 )dχ and the above result follows.
Putting all the contributions together, we get
dE0 a2 dE dE
F = = 2 2 2
= 2 2
. (2.149)
dt0 A0 a0 dt4πa0 χ dt4πa0 χ (1 + z)2
Hence, the luminosity distance is defined as:
dL ≡ a0 (1 + z)χ (2.150)
From this formula and the observed redshifts of type Ia supernovae we can deter-
mine if the universe is in an accelerated expansion, in a model-independent way.
In order to do this, we first need to know how to expand χ in series of powers of
the redshift.
Using the definition (2.137) and the expansion (2.135), we get:
Z t0
cdt0 c(t0 − t) cH0
χ= 0
= + (t0 − t)2 + · · · (2.151)
t a(t ) a 0 2a 0
where we stop at the second order only. This is the expansion of the comoving
distance with respect to the lookback time. We must invert the power series of
Eq. (2.135) in order to find the expansion of the lookback time with respect to the
redshift. This can be done, for example, by assuming the following ansatz:
H0 (t0 − t) = α + βz + γz 2 + · · · (2.152)
and substitute it into Eq. (2.135), keeping at most terms O(z 2 ).
Exercise 2.25. Show that:

1
α=0, β=1, γ = − (q0 + 2) , (2.153)
2
and thus
1
H0 (t0 − t) = z − (q0 + 2)z 2 + · · · (2.154)
2
Substituting the expansion of Eq. (2.154) back into Eq. (2.150), one gets

c 1 2
dL = z + (1 − q0 )z + · · · , (2.155)
H0 2
where note again that at the lowest order the luminosity distance is cz/H0 , identical
to the comoving distance and to the lookback time distance.
42
Since dL and z are measured, one can fit the data with this quadratic function
and determine q0 , thereby establishing if the universe expansion is accelerated or
not. Note that H0 is an overall multiplicative factor, thus does not determine the
shape of the function dL (z).
In the case of a dust-dominated universe, using Eq. (2.138), the luminosity
distance has the following expression:
2c √
dL = 1+z− 1+z . (2.156)
H0
For small z, this distance can be expanded in powers of the redshift as:

c 1 2
dL = z + z + ··· , (2.157)
H0 4
which, when compared with Eq. (2.150), provides q0 = 1/2, as expected.
2.5.5 Angular diameter distance

The angular diameter distance is based on the knowledge of proper sizes. Objects
with a known proper size are called standard rulers. Suppose a standard ruler
of transversal proper size ds (small) to be at a redshift z and comoving distance
χ. Moreover, this object has an angular dimension dφ, also small. See Fig. 2.6 for
reference.
O ds S
dφ χ
Figure 2.6: Defining the angular diameter distance.
At a fixed time t, we can write the FLRW metric as:
ds2 = a(t)2 dχ2 . (2.158)
Since the object is small and we are at the origin of the reference frame, the
comoving distance χ is also the radial distance. Therefore, the transversal distance
is:
ds = a(t)χdφ . (2.159)
Dividing the proper dimension of the object by its angular size provides us with
the angular diameter distance:
dA = a(t)χ . (2.160)
For the case of a dust-dominated universe, one has:

2c 1 1
dA = − . (2.161)
H0 1 + z (1 + z)3/2
43
In the limit of small z, we find dA ∼ cz/H0 . All the distances that we defined
insofar coincide at the first order expansion in z.
Note the relation:
dL = (1 + z)2 dA , (2.162)
known as Etherington’s distance duality [Etherington, 1933].
In gravitational lensing applications it is often necessary to know the angular-
diameter distance between two sources at different redshifts (i.e. the angular-
diameter distance between the lens and the background source). In order to com-
pute this, let us refer to Fig. 2.7.
O ds S
dφS L dφL
χL χLS
χS
Figure 2.7: The angular diameter distance between two different redshifts.
The problem is to determine the angular-diameter distance between L and S,

say dA (LS). Is this the difference between the angular-diameter distances dA (S) −
dA (L)? We now show that this is not the case. Simple trigonometry is sufficient
to establish that:
ds = a(tS )χS dφS = a(tS )χLS dφL , (2.163)
And for comoving distances we do have that χLS = χS − χL . Therefore, we have
dA (LS) = a(tS )χLS = a(tS )(χS − χL ) (2.164)
which is the relation we were looking for, and it is different from the difference
between the angular diameter distances:
dA (S) − dA (L) = a(tS )χS − a(tL )χL . (2.165)
44
Chapter 3
Thermal history
L’umanità non sopporta il pensiero che il mondo sia nato per caso,
per sbaglio. Solo perché quattro atomi scriteriati si sono tamponati
sull’autostrada bagnata
(Humanity cannot bear the thought that the world was born by acci-
dent, by mistake. Just because four mindless atoms crashed on the
wet highway)
—Umberto Eco, Il Pendolo di Foucault
In this Chapter we discuss the application of Boltzmann equation in cosmol-

ogy. In particular, we address Big Bang Nucleosynthesis (BBN), recombination of
protons and electrons in neutral hydrogen atoms and the relic abundance of CDM.
Our main references are [Dodelson, 2003], [Kolb and Turner, 1990] and Daniel
Baumann’s lecture notes (chapter 3).1 See also [Bernstein, 1988].
3.1 Thermal equilibrium and Boltzmann equation

We have encountered in the previous Chapter the continuity equation (2.77):
ε̇ + 3H(ε + P ) = 0 , (3.1)
where recall that the dot represents derivation with respect to the cosmic time.
Surprisingly, it is possible to obtain the continuity equation by using the first and
second law of thermodynamics, i.e.
T dS = P dV + dU , (3.2)
where T is the temperature, S is the entropy, V is the volume and U is the internal
energy of the cosmic fluid. Now, assuming adiabaticity, i.e. dS = 0, and writing
U = εV , one gets
P dV + d(εV ) = 0 ⇒ V dε + (ε + P )dV = 0 . (3.3)
1
http://www.damtp.cam.ac.uk/user/db275/Cosmology/Lectures.pdf
45
Exercise 3.1. Since the volume is proportional to the cube of the scale factor, i.e.
V ∝ a3 , show that Eq. (3.3) leads to the the continuity equation (3.1).
In using Eq. (3.2), we have made a very strong assumption: the evolution of the
universe is an adiabatic reversible transformation, i.e. at each instant the universe
is in an equilibrium state.
In some instances we can trust this assumption and it gives the correct con-
tinuity equation. In particular, we will see that Eq. (3.1) can be obtained from
Boltzmann equation assuming no interactions among particles or assuming a very
high rate of interactions so that thermal equilibrium is reached. The latter instance
can be mathematically represented as follows:
ΓH (3.4)
i.e. the interaction rate is much larger than the Hubble rate, where the
interaction rate is defined as follows:
Γ ≡ nσvrel (3.5)
where n is the particle number density of projectiles, vrel is the relative velocity
between projectile and targets and σ is the cross section.
Equation (3.4) can also be rephrased as the fact that the mean-free-path is much
smaller than the Hubble radius. In this situation, particles interact so frequently
that they do not even care about the cosmological expansion and any fluctuation in
their energy density is rapidly smoothed out, thus recovering thermal equilibrium.
It is important to make distinction between kinetic equilibrium and chem-
ical equilibrium. When Γ H refers to a process of the type:
1+2↔3+4, (3.6)
i.e. we have four different particle species which transform into each other in a
balanced way, then we have chemical equilibrium. This can be also reformulated
as:
µ1 + µ2 = µ3 + µ4 (3.7)
where the µ’s are the chemical potentials.
On the other hand, when Γ H refers to a reaction such as a scattering:
1+2↔1+2, (3.8)
then we have kinetic equilibrium. Kinetic or chemical equilibrium (or both)

imply thermal equilibrium. In general, it is possible for a species to break chemical
equilibrium and still remain in kinetic, therefore thermal, equilibrium with the rest
of the cosmic plasma through scattering processes.2
Note that Γ is different for different fundamental interactions and for different
particle species masses. Therefore, the above condition (3.4) is valid for all the
2
This occurs in some DM particle models. For example a m = 100 GeV WIMP chemically
decouples at 5 GeV and kinetically decouples at 25 MeV. See e.g. [Profumo et al., 2006].
46
known (and perhaps unknown) particles in the very early universe but is broken
at different times for different species. This is the essence of the thermal history
of the universe.
So, at the very beginning (we are talking about tiny fractions of seconds after
the Big Bang) all the particles were in thermal equilibrium in a primordial soup,
the primordial plasma. When for a species, the condition Γ ∼ H is reached, it
decouples from the primordial plasma. If it does this by breaking the chemical
equilibrium, then it is said to freeze out and attains some fixed abundance.3
When we want to explicitly calculate the residual abundance of some species,
we have to track its evolution until Γ ∼ H. In this instance, equilibrium thermo-
dynamics fails and we are compelled to use Boltzmann equation. For example, we
shall use Boltzmann equation when analysing the formation of light elements dur-
ing BBN, the recombination of protons and electrons in neutral hydrogen atoms
and the relic abundance of CDM.
The fundamental interactions which characterise the above-mentioned processes
compel some particles to react and transform into others and vice-versa, such as
in Eq. (3.6). When Γ H these reactions take place with equal probability in
both directions, hence the ↔ symbol, but when Γ ∼ H eventually one direction is
preferred over the other. This is the characteristic of irreversibility which demands
the use of Boltzmann equation.
3.2 Short summary of thermal history

We present in this section a brief scheme of the main events characterising the
thermal history of the universe. A minimal knowledge of the standard model of
particle physics is required.
Planck scale, inflation and Grand Unified Theory. Planck scale is usually
considered as an upper threshold in the energy, being the Planck mass is MPl c2 =
1019 GeV, or as a lower threshold in the time, 10−43 seconds, to which we can extend
our classical theory of gravity. Beyond that threshold, the common understanding
is that we should incorporate quantum effects.
Inflation is a very important piece of the current description of the primordial
universe. We shall dedicate to it Chapter 8 so, for now, it is enough to say that it
occurs at an energy scale of the order 1016 GeV, which is the same of the Grand
Unified Theory (GUT), i.e. a model in which electromagnetic, weak and strong
interactions are unified.
Baryogenesis and Leptogenesis. Baryogenesis is the creation, via some still

unclear mechanism, of a positive baryon number. In other words, the creation of
a primordial quark-antiquark asymmetry by virtue of which protons and neutrons
are much more common than anti-protons and anti-neutrons. In order to maintain
the neutrality of the universe, we need thus also Leptogenesis, i.e. a mechanism
which produces a non-vanishing lepton number in the form of an excess of electrons
3
It attains a fixed abundance if it is a stable particle, of course. If not it disappears.
47
over positrons. There is no evidence for a similar asymmetry in neutrinos and anti-
neutrinos,4 therefore in the universe energy budget, after the annihilation epochs,
we shall take into account both of them while neglecting anti-protons, anti-neutrons
and positrons.
We know that antimatter exists but we also know that matter is the most
abundant in the universe. If they were produced exactly in the same quantity, they
should have annihilated almost completely when in equilibrium in the primordial
plasma, leaving just photons. Instead, we have a small but nonvanishing baryon-
to-photon ratio ηb = 5.5 × 10−10 .
Decoupling of the Top quark. The Top quark is the most massive of all
observed fundamental particles, mTop c2 = 173 GeV, so it is probably the first to
decouple from the primordial plasma. For this reason it is considered here in this
list.
Electroweak phase transition. At a thermal energy of about 100 GeV, which

corresponds to 10−12 seconds after the Big Bang, the electromagnetic and weak
forces start to behave distinctly. This happens because the vector bosons W ± and
Z 0 gain their masses, of roughly 80 and 90 GeV respectively, through the Higgs
mechanism and the weak interaction “weakens”, since it is now ruled by the Fermi
constant:
GF
= 1.17 × 10−5 GeV−2 . (3.9)
(~c)3
QCD phase transition. Below 150 MeV quarks pass from their asymptotic
freedom to bound states form by two (mesons) or three (baryons) of them. The
above energy corresponds roughly to 20 µs after the Big Bang.
DM freeze-out. We do not know if DM is made up of particles and, if so, which

ones. For the neutralino case, the freeze-out takes place at about 25 MeV.
Neutrino decoupling. Neutrinos maintain thermal equilibrium with the pri-

mordial plasma through interactions such as
p + e− ↔ n + ν , p + ν̄ ↔ n + e+ , n ↔ p + e− + ν̄ , (3.10)
down to a thermal energy of 1 MeV, which corresponds roughly to 1 second after

the Big Bang. Below this energy threshold they decouple. We can calculate roughly
the 1 MeV scale of decoupling in the following way. Using Eq. (3.5) with vrel = c,
since neutrinos are relativistic, one gets:
3
G2F

Γ nν σc 2 3 kB T
= ≈ MPl c (kB T ) ≈ , (3.11)
H H (~c)6 1 MeV
4
It is still unclear whether neutrino is a Majorana fermion, i.e. a fermion which is its own
anti-particle, or a Dirac fermion, i.e. a fermion which is distinct from its anti-particle.
48
where we have used some results that we shall prove later, such as that the particle
number density nν goes as T 3 and H ∝ T 2 /MPl . When kB T ∼ 1 MeV then
decoupling occurs.
We have just used the effective field theory of the weak interaction in as-
suming σ ∝ G2F T 2 . An effective field theory is a low energy approximation of the
full theory. Indeed, the cross section σ ∝ G2F T 2 diverges at high energies and this
is unphysical. In the case of the weak interaction, we have chosen to work at an
energy scale much smaller than 80 GeV, which is the mass of the boson vector W ±
(which mediates the weak interaction). In this approximation, it is as if the boson
vectors W ± and Z 0 had infinite masses and therefore the range of the weak inter-
action is zero. In other words, we have considered the interactions of Eq. (3.10)
as if they occurred in a point. The result 1 MeV 80 GeV is consistent with the
effective field theory approximation and thus is reliable.
Another approximation that we have used is the neutrino masslessness. Even
considering a small mass mν c2 . 0.1 eV, the above calculation is solid, since
mν c2 kB T ∼ 1 MeV. The condition mc2 kB T for a generic particle of mass
m in thermal equilibrium guarantees that such particle is relativistic, as we shall
demonstrate later.
Other kinds of interactions involving neutrinos are those of annihilation, such
as:
e− + e+ ↔ νe + ν̄e . (3.12)
and these also are no more efficient on energies scales below 1 MeV.
Electron-positron annihilation. The interaction
e− + e+ ↔ γ + γ , (3.13)
is balanced for energies higher than the mass of electrons and positrons me c2 =
511 keV. When the temperature of the thermal bath drops below this value pair
production is no longer possible and annihilation takes over. Just for completeness,
the cross section for pair production is [Berestetskii et al., 1982]:

3 2 4 1+v 2
σγγ = σT (1 − v ) (3 − v ) ln − 2v(2 − v ) , (3.14)
16 1−v
where the parameter v is defined as follows:
p
v ≡ 1 − (me c2 )2 /(~2 ω1 ω2 ) ; (3.15)
the Thomson cross section is

2
8π α~c
σT = ≈ 66.52 fm2 , (3.16)
3 m e c2
with α = 1/137 being the fine structure constant. Finally, ω1 and ω2 are the
frequencies of the two photons. Being in a thermal bath, we have that ~ω1 ≈
~ω2 ≈ kB T . From Eq. (3.15) it is then clear that the condition
~2 ω1 ω2 ≈ (kB T )2 > (me c2 )2 , (3.17)
49
must be satisfied in order to produce pairs. Therefore, when the temperature of
the thermal bath drops below the value of the electron mass, annihilation becomes
the only relevant process. Thanks to leptogenesis, positrons disappear and only
electrons are left.
Big Bang Nucleosynthesis (BBN). We shall discuss BBN in great detail in

Sec. 3.9. It occurs at about 0.1 MeV (some three minutes after the Big Bang) as
deuterium and Helium form.
Recombination and photon decoupling. Proton and electrons form neutral

hydrogen at about 0.3 eV. Having no more free electrons with which to scatter,
photons decouple and can be seen today as CMB. We shall study this process in
detail in Sec. 3.10.
3.3 The distribution function

Before applying Boltzmann equation to cosmology, we have to introduce the main
character of our story: the distribution function f . This is a function f =
f (x, p, t) of the position, of the proper momentum and of the time, i.e. it is a
function which takes its values in the phase space. It can be thought of as a
probability density, i.e.
d3 xd3 p
f (t, x, p) , (3.18)
V
is the probability of finding a particle at the time t in a small volume d3 xd3 p of
the phase space centred in (x, p), and V is some suitable normalisation.
Because of Heisenberg uncertainty principle of quantum mechanics no particle
can be localised in the phase space in a point (x, p), but at most in a small volume
V = h3 about that point, where h is Planck constant. Therefore, the probability
density is the following:
d3 xd3 p d3 xd3 p
dP(t, x, p) = f (t, x, p) = f (t, x, p) . (3.19)
h3 (2π~)3
Note the dimensions: h has dimensions of energy times time, or momentum times
space. Hence f is dimensionless.
3.3.1 Volume of the fundamental cell

Why the fundamental cell has volume precisely equal to h3 ? We will see that
this value is very important in order to calculate correctly the abundances of the
universe components. That is, if it were 2h3 instead of h3 we would estimate half
of the present abundance of photons.5
Consider a quantum particle confined in a cube of side L. The eigenfunctions
up (x) of the momentum operator p̂ are determined by the equation:
p̂up (x) = pup (x) ⇒ −i~∇up (x) = pup (x) . (3.20)
5
Thanks to Dr. Luciano Casarini for raising this question.
50
Assuming variable separation,
up (x) = ux (x)uy (y)uz (z) , (3.21)
Eq. (3.20) is easily solved:

1 ip · x
up (x) = 3/2 exp , (3.22)
L ~
where the factor 1/L3/2 comes from the normalisation of the eigenfunction, which
has integrated square modulus equal to 1 in the box.
Exercise 3.2. Prove the result (3.22).
Since we have used variable separation, from now on focus only on the x di-
mension, for simplicity. Since the particle is restricted to be in the box, we must
impose periodic boundary conditions in Eq. (3.22):
ux (0) = ux (L) . (3.23)
These conditions imply that the momentum is quantised, i.e.
px L
= 2πnx , (3.24)
~
where nx ∈ Z.
Exercise 3.3. Prove the result (3.24).
The phase space occupied by a single state is thus:

∆(px L) = 2π~ = h , (3.25)
and recovering the three dimensions we get the expected result
V = ∆(px L)∆(py L)∆(pz L) = h3 . (3.26)
3.3.2 Integrals of the distribution function

Integrating the distribution function with respect to the momentum, one gets the
particle number density:
d3 p
Z
n(t, x) ≡ gs f (t, x, p) (3.27)
(2π~)3
where gs is the degeneracyp of the species, e.g. the number of spin states. If we
weigh the energy E(p) = p2 c2 + m2 c4 with the distribution function, we get the
energy density
d3 p
Z p
ε(t, x) = gs f (t, x, p) p2 c2 + m2 c4 (3.28)
(2π~)3
51
where note that we are identifying p ≡ |p|. The pressure is defined as follows:
d3 p p 2 c2
Z
P (t, x) = gs f (t, x, p) (3.29)
(2π~)3 3E(p)
For photons one has E(p) = pc and therefore combining Eqs. (3.28) and (3.29) one
gets the familiar result P = ε/3.
In general, the energy-momentum tensor written in terms of the distribution
function has the following form:
dP1 dP2 dP3 1 cP µ Pν

Z
µ
T ν (t, x) = gs √ f (t, x, p) (3.30)
(2π~)3 −g P 0
where Pµ ≡ dxµ /dλ is the comoving momentum and g is the determinant of gµν .
√
Note that the combination dP1 dP2 dP3 /( −gP 0 ) which appears in the integrand is
actually covariant. We shall prove this later in Sec 3.8, but we give now a simple
proof within special relativity. Indeed, for g = −1, the volume element reduces
to d3 P/E, which is invariant under Lorentz transformations. Let us show this,
focussing on a single spatial dimension, i.e. the one along which the boost takes
place and along which the particle travels. Lorentz transformations between two
inertial reference frames are:
E 0 = γ(E − βpc) , p0 c = γ(pc − βE) , (3.31)
with β ≡ V /c, being V the boost velocity. Combining the differential form of the
second, with the first, we get:
dp0 c dpc − βdE
= . (3.32)
E0 E − βpc
Exercise 3.4. Differentiate the dispersion relation E 2 = p2 c2 + m2 c4 , use it into

(3.32) and show that
dp0 dp
0
= , (3.33)
E E
i.e. this combination is invariant.
We can rewrite Eq. (3.30) in terms of the proper momentum instead of the
comoving one. Using the FLRW metric with K = 0, for simplicity, the dispersion
relation (2.31) and Eq. (2.32) we can write
√
g00 (P 0 )2 = −E 2 /c2 ⇒ −g00 P 0 = E/c , (3.34)
so that Eq. (3.30) becomes:
dP1 dP2 dP3 1 c2 P µ Pν

Z
µ
T ν (t, x) = gs f (t, x, p) (3.35)
(2π~)3 a3 E
52
Exercise 3.5. Using the definition (2.32) of the proper momentum, i.e.
1 ij
p2 = gij P i P j = g ij Pi Pj = δ Pi P j , (3.36)
a2
show that:
pi pp̂i
Pi = api = app̂i Pi = = (3.37)
a a
where p̂i is the direction of the proper or comoving momentum, satisfying:
δ ij p̂i p̂j = 1 (3.38)
Note the following:
pj
api = Pi = gij P j = a2 δij = aδij pj . (3.39)
a
This means that
pi = δij pj (3.40)
i.e. it is as if the proper momentum were a 3-vector in the Euclidean space.
Therefore, Eq. (3.35) becomes:
d3 p c2 P µ Pν
Z
µ
T ν (t, x) = gs f (t, x, p) (3.41)
(2π~)3 E
For µ = ν = 0:
d3 p d3 p
Z Z
0
T 0 (t, x) = gs cP0 f (t, x, p) = gs E(p)f (t, x, p) . (3.42)
(2π~)3 (2π~)3
For µ = 0 and ν = i:
d3 p d3 p
Z Z
0
T i (t, x) = gs cPi f (t, x, p) = gs apcp̂i f (t, x, p) . (3.43)
(2π~)3 (2π~)3
Finally, for µ = i and ν = j:
d3 p c2 p2 p̂i p̂j
Z
i
T j (t, x) = gs f (t, x, p) . (3.44)
(2π~)3 E
Taking the spatial trace and dividing by 3 we get the pressure, consistently with
Eq. (3.29).
53
3.4 The entropy density
For many of the forthcoming purposes the hypothesis of thermal equilibrium is
suitable and very useful. As we have stated at the beginning of this chapter,
it is justified in those instances in which the interaction rate among particles is
much higher than the expansion rate. In these cases, one can use equilibrium
thermodynamics and a very useful quantity is the entropy density:
S
s≡ , (3.45)
V
because, as we will show in a moment, sa3 is conserved.
In thermal equilibrium, we can cast the thermodynamical relation (3.2) in the
following form:
dε
T dS = V dε + (ε + P )dV = V dT + (ε + P )dV , (3.46)
dT
because the energy density (and also the pressure) only depends on the temperature
T . The integrability condition applied to Eq. (3.46) yields to:
∂ 2S ∂ 2S dP
= ⇒ T =ε+P . (3.47)
∂T ∂V ∂V ∂T dT
Bosons and fermions in thermal equilibrium are distributed according to the Bose-
Einstein and Fermi-Dirac distributions:
1 1
fBE = , fFD = (3.48)
E−µ E−µ
exp kB T
−1 exp kB T
+1
We derive these distributions in Chapter 12.

We now prove Eq. (3.47) in another way, assuming a distribution function of
the type f = f (E/T ). We first must know how to calculate dP/dT .
Call E/T = x and f 0 ≡ df /dx. Then:
f0 f 0E
df = f 0 dx = dE − 2 dT . (3.49)
T T
Comparing this with
∂f ∂f
df = dE + dT , (3.50)
∂E ∂T
we can establish that
∂f E ∂f
=− . (3.51)
∂T T ∂E
Now we use this result into:
d3 p ∂f p2 c2
Z
dP
= gs , (3.52)
dT (2π~)3 ∂T 3E
and obtain
d3 p E ∂f p2 c2
Z
dP
= −gs . (3.53)
dT (2π~)3 T ∂E 3E
54
Now we introduce spherical coordinates in the proper momentum space:
d3 p = p2 dpd2 p̂ , (3.54)
and use the dispersion relation E 2 = p2 c2 + m2 c4 in order to write
∂f ∂f dp ∂f E
= = . (3.55)
∂E ∂p dE ∂p pc2
Equation (3.53) thus becomes:
p2 dpd2 p̂ E ∂f p
Z
dP
= −gs . (3.56)
dT (2π~)3 T ∂p 3
Integrating by parts:
∞
4π Ep3 dpd2 p̂ 1 2
Z
dP 2 3 pc
= −gs f + gs f 3p E + p , (3.57)
dT (2π~)3 3T 0 (2π~)3 3T E
where we have used again dE/dp = pc2 /E. The first contribution vanishes for p →
∞, since f → 0, i.e. there are no particles with infinite momentum. Recovering
the momentum volume d3 p, we get:
d3 p 1 p 2 c2
Z
dP dP ε+P
= gs 3
f E+ ⇒ = (3.58)
dT (2π~) T 3E dT T
as we wanted to show.
We now prove that the temperature derivative of the pressure is the entropy
density. Substituting Eq. (3.47) into Eq. (3.46), i.e.
T dS = d(εV ) + P dV = d[(ε + P )V ] − V dP , (3.59)
one gets
1 ε+P (ε + P )V
dS = d[(ε + P )V ] − V dT = d , (3.60)
T T2 T
i.e. up to an additive constant:
S ε+P
s≡ = . (3.61)
V T
Taking into account the chemical potential µ, the thermodynamical relation (3.2)
becomes
T dS = d(εV ) + P dV − µd(nV ) , (3.62)
and the entropy density is redefined as
S ε + P − µn
s≡ = (3.63)
V T
Exercise 3.6. Using the continuity equation and/or Eq. (3.2) show that sa3 is a
constant:
d
(sa3 ) = 0 (3.64)
dt
i.e. the entropy density is proportional to 1/a3 .
55
3.5 Photons
In these notes we use “photons” as synonym of CMB, though this is not correct
since there exist photons whose origin is not cosmological, e.g. those produced in
our Sun as well as in other stars or emitted by hot interstellar gas. These say non-
cosmological photons contribute at least one order of magnitude less than CMB
photons [Camarena and Marra, 2016] so our sloppiness is partially justified.
Assuming a vanishing chemical potential since µ/(kB T ) < 9 × 10−5 , as reported
by [Fixsen et al., 1996], the Bose-Einstein distribution for photons becomes:
1 1
fγ = = , (3.65)
E pc
exp kB T
−1 exp kB T
−1
where we have used the dispersion relation E = pc. Taking into account the
chemical potential is important in order to study distortions in the CMB spectrum,
which is a very recent and promising research field [Chluba and Sunyaev, 2012],
[Chluba, 2014].
Let us calculate the photon energy density:
d3 p
Z
pc
εγ = 2 3
, (3.66)
(2π~) exp pc − 1
kB T
where the factor 2 represents the two states of polarisation of the photon. The
angular part can be readily integrated out, giving a factor 4π. We are left then
with: Z ∞
c p3
εγ = 2 3 dp . (3.67)
π ~ 0 exp pc − 1 kB T
Let us do the substitution x ≡ pc/(kB T ). We obtain:

4 Z ∞
x3

c kB T
εγ = 2 3 dx x . (3.68)
π ~ c 0 e −1
The integration is proportional to the Riemann ζ function, which has the following
integral representation:
Z ∞ Z ∞
1 xs−1 1 xs−1
ζ(s) = dx x = dx , (3.69)
Γ(s) 0 e −1 (1 − 21−s )Γ(s) 0 ex + 1
where Γ(s) is Euler gamma function.

In alternative, a possible way to perform the integration is the following. Let
I− be: Z ∞ Z ∞
x3 e−x x3
I− ≡ dx x = dx . (3.70)
0 e −1 0 1 − e−x
Now use the geometric series in order to write
∞
1 X
−x
= e−nx , (3.71)
1−e n=0
56
and substitute this in the integral I− :
∞ Z
X ∞
I− = dx e−(n+1)x x3 . (3.72)
n=0 0
Exercise 3.7. Integrate the above equation three times by part and show that:
∞ Z ∞
X 1
I− = 6 3
dx e−nx . (3.73)
n=1
n 0
Integrate again and find:

∞
X 1 π4
I− = 6 ≡ 6ζ(4) = , (3.74)
n=1
n4 15
in agreement with Eq. (3.69).
Therefore, the photon energy density is the following:
π2
εγ = (kB T )4 . (3.75)
15~3 c3
This is the Stefan-Boltzmann law, of the black-body radiation. From the continuity
equation we know that εγ = εγ0 /a4 so that we can infer that
T0
T = , T0 = 2.725 K (3.76)
a
i.e. the temperature of the photons decreases with the inverse scale factor. This is a
result that we will prove also using the Boltzmann equation (indeed the continuity
equation that provides εγ = εγ0 /a4 is a way of writing the Boltzmann equation).
In the above equation (3.76), the value of T0 is the measured one of the CMB.
Knowing this value, we can estimate the photon energy content today (and thus
at any times):
εγ0 8π 3 G
Ωγ0 = = (kB T0 )4 . (3.77)
εcr0 45~3 c3 H02 c2
Exercise 3.8. Using H0 = 100 h km s−1 Mpc−1 and the known constants of nature
show that:
Ωγ0 h2 = 2.47 × 10−5 . (3.78)
The photon number density is calculated as follows:

Z ∞
1 p2
nγ = 2 3 dp . (3.79)
π ~ 0 exp kpc − 1
BT
57
Making the usual substitution x ≡ pc/(kB T ) and using Eq. (3.69) we get:
(kB T )3 ∞ x2
Z
2ζ(3)
nγ = 2 3 3 dx x ⇒ nγ = 2 3 3 (kB T )3 (3.80)
π ~c 0 e −1 π ~c
Exercise 3.9. Calculate the photon number density today. Show that it is nγ0 =
411 cm−3 .
3.6 Neutrinos
The same comment made ate the beginning of the previous section also applies
here: with “neutrinos” we mean the cosmological, or primordial, ones and not
those produced e.g. in supernovae explosions.
The massless neutrino energy density also scales as εν = εν0 /a4 , as the photons
energy density, but since neutrinos are fermions we need now to employ the Fermi-
Dirac distribution.
Exercise 3.10. Show that the energy-density of a massless fermion species is given
by the following integral:
4 Z ∞
x3

c kB T
ε= 2 3 gs dx x . (3.81)
2π ~ c 0 e +1
Since neutrinos are spin 1/2 fermions, then gν = 2, where gν is the neutrino gs .
On the other hand, neutrinos are particles which interact only via weak interaction
and this violates parity. In other words, only left-handed neutrinos can be detected.
Right-handed neutrinos, if they exist, would interact only via gravity and via
the seesaw mechanism with the left-handed neutrino [Gell-Mann et al., 1979].
Right-handed neutrinos are also called sterile neutrinos and are advocated as
possible candidates for DM [Dodelson and Widrow, 1994].
Let I+ be the integral in Eq. (3.81):
Z ∞
x3
I+ ≡ dx x . (3.82)
0 e +1
Using Eq. (3.69), we have:
7
I+ = (1 − 2−3 )I− = I− , (3.83)
8
i.e. the difference between bosons and fermions energy densities per spin state is
simply a factor 7/8. Taking into account that I− = π 4 /15, the neutrino energy
density (3.81) is:
7 π2
εν = Nν gν (kB Tν )4 (3.84)
8 30~3 c3
where Nν is the number of neutrino families. Of course, an equivalent expression
holds true for antineutrinos.
58
3.6.1 Temperature of the massless neutrino thermal bath
As we have seen in Eq. (3.64), for a species in thermal equilibrium its entropy
density s is proportional to 1/a3 . Moreover, if that species is relativistic then
its temperature T scales as 1/a. Therefore, for a relativistic species in thermal
equilibrium s ∝ T 3 .
Since photons and neutrinos do not interact, it is reasonable to ask whether
the temperatures of the photon thermal bath and of the neutrino thermal bath are
the same. We cannot observe the neutrino thermal bath today, but we can indeed
predict different temperatures.
When the temperature of the photon thermal bath was sufficiently high, positron-
electron annihilation and positron-electron pair production were balanced reac-
tions:
e+ + e− ↔ γ + γ . (3.85)
In order to produce e+ -e− pairs, the photons must have temperature of the order
of 1 MeV at least.
Exercise 3.11. Knowing that today, i.e. for z = 0, the photon thermal bath has
temperature T0 = 2.725 K, estimate the redshift at which kB T = 1 MeV.
Therefore, when the temperature of the photon thermal bath drops below that
value, the above reactions are unbalanced and more photons are thus injected in
the thermal bath. For this reason we expect the photon temperature to drop more
slowly than 1/a and then to be different from the neutrino temperature. We now
quantify this difference using the conservation of sa3 .
For a relativistic bosonic species (such as the photon), the entropy density is:
ε+P 4ε 2π 2 kB4 3
sboson = = = gs T (3.86)
T 3T 45~3 c3
whereas for a relativistic fermion species (such as the neutrino), the entropy density
is:
7 2π 2 kB4 3
sfermion = gs T (3.87)
8 45~3 c3
Therefore, at a certain scale factor a1 earlier than e− -e+ annihilation, the entropy
density is:
2π 2 kB4 3

7 7
s(a1 ) = T 2 + (2 + 2) + Nν (gν + gν ) , (3.88)
45~3 c3 1 8 8
where we have left explicit all the degrees of freedom, i.e. 2 for the photons, 2 for
the electrons, 2 for the positrons, gν for the neutrinos and gν for the antineutrinos.
Moreover, we have assumed the same temperature T1 for photons and neutrinos
because they came from the original thermal bath (the Big Bang).
At a certain scale factor a2 after the annihilation the entropy density is:
2π 2 kB4

3 7 3
s(a2 ) = 2Tγ + Nν gν Tν , (3.89)
45~3 c3 4
59
where we have now made distinction between the two temperatures. Equating
s(a1 )a31 = s(a2 )a32 , (3.90)
we obtain
" #
3
7 Tγ 7
(a1 T1 )3 2 + (Nν gν + 2) = (a2 Tν )3 2 + Nν gν . (3.91)
4 Tν 4
Since (a1 T1 )3 = (a2 Tν )3 , because the neutrino temperature did not change its ∝ 1/a
behaviour, we are left with:
1/3
Tν 4
= ≈ 0.714 (3.92)
Tγ 11
Therefore, since the CMB temperature today is of Tγ0 = 2.725 K, we expect

a thermal neutrino background of temperature Tν0 ≈ 1.945 K. Remarkably, the
result of Eq. (3.92) does not depend on the neutrino gν .
Let us open a brief parenthesis in order to justify the procedure that we have
used in order to determine the result in Eq. (3.92).
We saw that for a single species in thermal equilibrium sa3 is conserved because
of the continuity equation. However, during electron-positron annihilation the
continuity equation does not hold true neither for photons nor for electron and
positron, i.e.
ε̇γ + 4Hεγ = +Γann , (3.93)

ε̇e + 3H(εe + Pe ) = −Γann , (3.94)
where Γann is the e− -e+ annihilation rate. Therefore, we cannot use the constancy
of sa3 for each of these species separately. However, we can and did use it for the
total, since the sum of the two above equations gives:
ε̇tot + 3H(εtot + Ptot ) = 0 . (3.95)
In other words, the continuity equation always applies if one suitably extends the
set of species, ultimately because of Bianchi identities.
Using Eq. (3.92), the neutrino and antineutrino energy densities can be thus
related to the photon energy density as follows:
4/3
7 Nν gν 4
εν = εν̄ = εγ (3.96)
8 2 11
where we have left unspecified the value of the neutrino degeneracy gν and the
number of neutrino families Nν .
The total radiation energy content can thus be written as
" 4/3 #
7 4
εr ≡ εγ + εν + εν̄ = εγ 1 + Nν gν . (3.97)
8 11
60
The Planck collaboration [Ade et al., 2016a] has put the constraint
Neff ≡ Nν gν = 3.04 ± 0.33 (3.98)
at 95% CL. Therefore, three neutrino families (Nν = 3) and one spin state for each
neutrino (gν = 1) are values which work fine.
Calculating the neutrino + antineutrino number density today is straightfor-
ward:
Nν gν ∞ p2
Z
nν = 2 3 dp , (3.99)
π ~ 0 exp kBpcTν + 1
where now the subscript ν indicates both neutrinos and antineutrinos. Making the
usual substitution x ≡ pc/(kB Tν ), we get:
Nν gν (kB Tν )3 ∞ x2 Nν gν (kB Tν )3 3ζ(3)Γ(3) 3ζ(3)Nν gν (kB Tν )3

Z
nν = dx = = .
π 2 ~3 c3 0 ex + 1 π 2 ~3 c3 2 2π 2 ~3 c3
(3.100)
Exercise 3.12. Taking into account (3.92), show that
3
nν = Nν gν nγ (3.101)
11
3.6.2 Massive neutrinos

The 2015 Nobel Prize in Physics has been awarded to Takaaki Kajita and Arthur
B. McDonald for the discovery of neutrino oscillations, which shows that neutri-
nos have mass (quoting from the Nobel Prize website). Indeed, neutrino flavours
oscillate among the leptonic families (electron, muon and tau), e.g. an electronic
neutrino can turn into a muonic one and a tau one, depending on its energy and
on how far it travels.
The most stringent constraints
P on neutrino mass do not come from particle
accelerators but cosmology: mν < 0.194 eV at 95% CL from the Planck collab-
oration [Ade et al., 2016a]. The neutrino mass is thus very small and this does not
change relevantly the early history of the universe whereas it has some impact at
late-times for structure formation [Lesgourgues and Pastor, 2006].
Using Eq. (3.28) and the FD distribution, the massive neutrino energy density
can be calculated as follows:
p
d3 p p2 c2 + m2ν c4
Z
εν = h i , (3.102)
(2π~)3 exp p
p2 c2 + m2ν c4 /(kB T ) + 1
where we are assuming gs = 1.

Actually, Eq. (3.102) is valid for any fermion (or boson, if we have the −1 at the
denominator), provided we can neglect the chemical potential. For this reason, we
61
drop the subscript ν and consider a generic species of mass m, fermion or boson,
in order to make more general statements.
Rewrite Eq. (3.102) as follows:
Z ∞ p
mc2 2 p2 /(m2 c2 ) + 1 mc2
ε= 2 3 dp p h p i , A≡ . (3.103)
2π ~ 0 exp A p2 /(m2 c2 ) + 1 ± 1 kB T
p
Exercise 3.13. Calling x ≡ A p2 /(m2 c2 ) + 1, show that the above integration
(3.103) becomes:
Z ∞ √ √
m4 c5 x2 x2 − A2 (kB T )4 ∞ x2 x2 − A2
Z
ε= 2 3 4 dx = 2 3 3 dx . (3.104)
2π ~ A A ex ± 1 2π ~ c A ex ± 1
Note the lower integration limit.
Relativistic and non-relativistic regimes of particles in thermal equilib-

rium
Unfortunately, the above integral in Eq. (3.104) cannot be solved analytically.
When A 1, i.e. the thermal energy is much larger than the mass energy, the
integral in Eq. (3.104) can be expanded as follows:
Z ∞ √
x2 x2 − A2 π 4 π 2 A2
dx x−1
= − + O(A3 ) , (3.105)
A e 15 12
Z ∞ √
x2 x2 − A2 7π 4 π 2 A2
dx x+1
= − + O(A3 ) . (3.106)
A e 120 24
The zero-order terms recovers the result of Eq. (3.75), for photons, and of Eq. (3.84),
obtained for massless neutrinos. In general, we can state that particles with
mass m in thermal equilibrium at temperature T behave as relativistic
particles when mc2 kB T .
The opposite limit mc2 kB T is much trickier to investigate analytically, so
we shall do it numerically.
p
Exercise 3.14. Calling x ≡ A p2 /(m2 c2 ) + 1, show that the number density can
be written as: Z ∞ √
m3 c3 x x2 − A2
n= 2 3 3 dx . (3.107)
2π ~ A A ex ± 1
The ratio ε/(nmc2 ) can be written as:

R∞ √
x2 x2 −A2
ε 1 A dx ex ±1
γ≡ = R∞ √ , (3.108)
(nmc2 ) A 2
dx x xx −A
2
A e ±1
62
��
��
��
��
��
�
�
��
Figure 3.1: Plot of γ as function of A. The solid line is for fermions whereas the
dashed one for bosons.
where γ is indeed some sort of averaged Lorentz factor since it is the ratio of the
energy density to the mass energy density. The behaviour of γ as function of A is
shown in Fig. 3.1.
From Fig. 3.1 we can infer that ε/(nmc2 ) ≈ 1 when mc2 kB T and there-
fore the particle becomes non-relativistic since all its energy is mass energy. In
general, we can state that particles with mass m in thermal equilibrium at
temperature T behave as non-relativistic particles when mc2 kB T . The
transition relativistic → non-relativistic takes place for kB T ≈ 10mc2 .
Now, suppose that a single family of neutrinos and antineutrinos became non-
relativistic only recently, what would be their energy density and density parameter
today? Being non-relativistic, we can write their energy density as follows:
εmν = ρν c2 = nν mν c2 , (3.109)
and thus the density parameter today is:

8πGnν0 mν
Ωmν 0 = . (3.110)
3H02
Exercise 3.15. Using Eq. (3.101) and the result for nγ0 , prove that Eq. (3.110)
can be written as:
1 mν c2
Ωmν 0 = gν (3.111)
94h2 eV
63
3.6.3 Matter-Radiation equality
The epoch, or instant, at which the energy density of matter (i.e. baryons plus
CDM) equals the energy density of radiation (i.e. photons plus neutrinos) is par-
ticularly important from the point of view of the evolution of perturbations, as we
shall see in Chapter 9. From Eq. (3.97) we have that:
" 4/3 #
7 4
Ωr0 = Ωγ0 1 + Neff (3.112)
8 11
In order to calculate the scale factor aeq of the equivalence we only need to solve
the following equation:
Ωr0 Ωm0
4
= 3 , (3.113)
aeq aeq
which gives " 4/3 #
Ωr0 Ωγ0 7 4
aeq = = 1 + Neff . (3.114)
Ωm0 Ωm0 8 11
Using Eq. (3.78) and Neff = 3 one obtains:
4.15 × 10−5
aeq = ⇒ 1 + zeq = 2.4 × 104 Ωm0 h2 (3.115)
Ωm0 h2
What does happen to the equivalence redshift zeq if one of the neutrino species has
mass mν 6= 0?
If that neutrino species becomes non-relativistic after the equivalence epoch,
then the above calculation still holds true. Therefore, let us assume that the
neutrino species becomes non-relativistic, thereby counting as matter, before the
equivalence. Since there is more matter and less radiation we expect the equivalence
to take place earlier, i.e.
3.59 × 10−5
aeq = 1 + zeq = 2.79 × 104 (Ωm0 + Ωmν 0 )h2 . (3.116)
(Ωm0 + Ωmν 0 )h2
Since T = T0 (1 + z), the photon temperature at equivalence is:
Tγ,eq = 2.79 × T0 · 104 (Ωm0 + Ωmν 0 )h2 = 7.60 × 104 (Ωm0 + Ωmν 0 )h2 K . (3.117)
Using Eq. (3.92), the temperature of neutrinos at equivalence is:
Tν,eq = 5.43 × 104 (Ωm0 + Ωmν 0 )h2 K . (3.118)
The neutrino mass energy mν c2 has to be larger than kB Tν,eq in order for the above
calculation to be consistent. This yields:
mν c2 > 5.43 kB × 104 (Ωm0 + Ωmν 0 )h2 K = 4.68 (Ωm0 + Ωmν 0 )h2 eV . (3.119)
Using Ωm0 h2 = 0.14 P

and Eq. (3.111) we get mν c2 > 0.69 eV, which is incompatible
with the constraint mν < 0.194 found by the Planck collaboration.
64
3.7 Boltzmann equation
Here is the main character of this Chapter: Boltzmann equation. It is very simple
to write:
df
= C[f ] , (3.120)
dt
but nonetheless very meaningful, as we shall appreciate. Here f is the one-particle
distribution function and C[f ] is the collisional term, i.e. a functional of f describ-
ing the interactions among the particles constituting the system under investiga-
tion. The one-particle distribution is a function of time t, of the particle position
x and of the particle momentum p. In turn, also x and p are functions of time,
because of the particle motion. Therefore, the total time derivative can be written
as:
df ∂f dx dp ∂f
= + · ∇x f + · ∇p f = + v · ∇x f + F · ∇p f ≡ L̂(f ) , (3.121)
dt ∂t dt dt ∂t
where v is the particle velocity and F is the force acting on the particle. The
operator L̂ acting on f is similar to the convective derivative used in fluid dynamics
and is also called Liouville operator.
If interactions are absent, then
df
=0, (3.122)
dt
is the collisionless Boltzmann equation, or Vlasov equation. It represents math-
ematically the fact that the number of particles in a phase space volume element
does not change with the time. Note that we are starting here with the non rel-
ativistic version of the Boltzmann equation. We shall see it later in the general
relativistic case and cosmology.
The collisionless Boltzmann equation is a direct consequence of Liouville theo-
rem:
dρ(t, xi , pi )
=0, i = 1, · · · , N (3.123)
dt
where ρ(xi , pi , t) is the N -particle distribution function, i.e.
ρ(t, xi , pi )dN xdN p , (3.124)
is the probability of finding our system of N particles in a small volume dN xdN p

of the phase space centred in (xi , pi ).
If the particles are not interacting, then the probability of finding N particles in
some configuration is the product of the single probabilities. That is, the positions
in the phase space of the individual particles are independent events. Therefore:
dρ df
ρ ∝ fN , = N f N −1 , (3.125)
dt dt
and using Liouville theorem (3.123) one obtains the collisionless Boltzmann equa-
tion (3.122).
65
3.7.1 Proof of Liouville theorem
In order to complete this brief introduction to the Boltzmann equation, we prove
Liouville theorem. As first step, expand the time derivative of ρ:
N
dρ(t, xi , pi ) ∂ρ X dxi dp
= + ∇xi ρ + i ∇pi ρ . (3.126)
dt ∂t i=1
dt dt
This can be written as

N
dρ(t, xi , pi ) ∂ρ X
= + ∇xi (ρẋi ) + ∇pi (ρṗi ) , (3.127)
dt ∂t i=1
since the extra term
∇xi (ẋi ) + ∇pi (ṗi ) = ∇pi ∇xi (H) − ∇xi ∇pi (H) = 0 , (3.128)
is vanishing because of Hamilton equations of motion. Moreover, Eq. (3.127) can

be cast as:
dρ(t, xi , pi ) ∂ρ
= + ∇y · (ρẏ) , (3.129)
dt ∂t
where we have indicated as y the generic variable of the phase space, i.e. y =
{xi , pi } for i = 1, · · · , N . Equation (3.129) is a continuity equation. Therefore, if
the particle number N is conserved, then Eq. (3.123) must hold.
3.7.2 Example: The one-dimensional harmonic oscillator

Consider the one-dimensional harmonic oscillator:
p2 kx2
H= + . (3.130)
2m 2
The Boltzmann equation is the following:
∂f ∂f dx ∂f dp
+ + =0, (3.131)
∂t ∂x dt ∂p dt
where
dx ∂H p dp ∂H
= = , =− = −kx . (3.132)
dt ∂p m dt ∂x
The general strategy for solving Boltzmann equation is to use the method of the
characteristics. A characteristic is a curve in the space formed by the phase space
plus the time axis and it is parametrised by its arc length s such that
df (s)
=0, (3.133)
ds
i.e. f is constant along the characteristic. So, given an initial value s0 , one has
f = f (s0 ).
66
In our example of the harmonic oscillator, the characteristic is a curve in the
3-dimensional space (t, x, p), described by t(s), x(s) and p(s). Using the chain rule,
we get:
df ∂f dt ∂f dx ∂f dp
= + + =0. (3.134)
ds ∂t ds ∂x ds ∂p ds
Comparing with (3.131), we get the following system:
 
dt
 ds = 1
 t = s − s 0

dx p
ds
=m ⇒ p = m dx
ds
. (3.135)
 dp
  d2 x
 k
ds
= −kx ds2
+ m
x=0
We solve the last equation supposing that x(s0 ) = x0 . Therefore:

"r #
k
x(s) = x0 cos (s − s0 ) , (3.136)
m
and
"r # "r #
dx(s) √ k k
p(s) = m = −x0 km sin (s − s0 ) ≡ p0 sin (s − s0 ) . (3.137)
ds m m
The distribution function is thus a function of:
f = f (s0 ) = f [t(s0 ), x(s0 ), p(s0 )] , (3.138)
that is
f = f (s0 ) = f [x(s0 )] = f (x0 ) , (3.139)
i.e. it is simply a function of the initial position x0 . The latter can be related to
the position and momentum at any time in the following way:
x2 p 2
+ =1, (3.140)
x20 p20
which turns out to be:

x2 p2 2E
+ =1 ⇒ x20 = , (3.141)
x20 x20 km k
where E is the energy of the harmonic oscillator. Therefore, f = f (E), i.e. the dis-
tribution function is a function of the energy. The precise functional form depends
on the initial condition that we give on f .
3.7.3 Boltzmann equation in General Relativity and Cos-

mology
In GR the distribution function must be expressed covariantly as f = f (xµ , P µ ),
and the total derivative of f cannot be taken with respect to the time because
67
this would violate the general covariance of the theory. The total derivative of f
is taken with respect to an affine parameter λ, as follows:
df ∂f dxµ ∂f dP µ
= + . (3.142)
dλ ∂xµ dλ ∂P µ dλ
The geometry enters through the derivative of the four-momentum, which can be
expressed via the geodesic equation:
dP µ
+ Γµνρ P ν P ρ = 0 , (3.143)
dλ
so that
df ∂f ∂f
= P µ µ − Γµνρ P ν P ρ µ ≡ L̂rel (f ) , (3.144)
dλ ∂x ∂P
where we have defined the relativistic Liouville operator L̂rel . It might seem that in
the relativistic case we have gained one variable, i.e. P 0 , but this is not so because
P 0 is related to the spatial momentum P i via the relation gµν P µ P ν = −m2 c2 . For
this reason, we can reformulate the Liouville operator as follows:
df ∂f ∂f
= P µ µ − Γiνρ P ν P ρ i , (3.145)
dλ ∂x ∂P
i.e. by considering f = f (xµ , P i ).
Collisionless Boltzmann equation in relativistic cosmology

When we couple Eq. (3.145) with FLRW metric, we must take into account that
f cannot depend on the position xi , because of homogeneity and isotropy. The
collisionless Boltzmann equation thus becomes:
∂f ∂f
P0 − Γiνρ P ν P ρ i = 0 . (3.146)
∂t ∂P
Exercise 3.16. Considering the spatially flat case K = 0, show that the above
equation can be cast as follows:
∂f ∂f
− 2HP i i = 0 . (3.147)
∂t ∂P
Again, because of isotropy, f cannot depend on the direction of P i , but only

on its modulus P 2 = δij P i P j .
Exercise 3.17. Show for a generic function f = f (x2 ) that:

∂f ∂f
xi = x , (3.148)
∂xi ∂x
with x2 ≡ δij xi xj .
68
Therefore, we can write
∂f ∂f
− 2HP =0. (3.149)
∂t ∂P
Exercise 3.18. Show that the solution of the above equation is a generic function:
f = f (a2 P ) = f (ap) , (3.150)
where in the second equality we have used the definition of the proper momentum.
Show that the Boltzmann equation written using the proper momentum has
the following form:
∂f ∂f
− Hp =0 (3.151)
∂t ∂p
Moments of the collisionless Boltzmann equation

Taking moments of the Boltzmann equation means to integrate it in the momentum
space, weighed with powers of the proper momentum. This method is due to
Grad [Grad, 1958]. For example, the moment zero of Eq. (3.151) is the following:
d3 p
Z
∂f ∂f
− Hp =0 (Moment zero) . (3.152)
(2π~)3 ∂t ∂p
Exercise 3.19. Using the definition of the particle number density (3.27), show
that Eq. (3.152) becomes:
1 d(na3 )
ṅ + 3Hn = 0 ⇒ =0 (3.153)
a3 dt
The particle number na3 is conserved. This is an expected result since we have
considered a collisionless Boltzmann equation, i.e. absence of interactions and thus
no source of creation or destruction of particles.
Weighing Eq. (3.151) with the energy and integrating in the momentum space
we get:
d3 p
Z
∂f
ε̇ − H 3
pE(p) =0. (3.154)
(2π~) ∂p
Exercise 3.20. Integrate by parts the above equation and show that it becomes:
ε̇ + 3H (ε + P ) = 0 , (3.155)
i.e. the continuity equation.
69
Weighing Boltzmann equation (3.151) with p̂i will always result in an identity,
because of isotropy. We can show this as follows:
d3 p i ∂f dp p2 ∂f
Z Z Z
∂f 2 i ∂f
p̂ − Hp = d p̂ p̂ − Hp = 0 . (3.156)
(2π~)3 ∂t ∂p (2π~)3 ∂t ∂p
Since f = f (ap) then only the integration in p contains f and the angular integra-
tion is just a multiplicative factor.
Exercise 3.21. Show that: Z

d2 p̂ p̂i = 0 , (3.157)
and thus Eq. (3.156) is an identity.
3.8 Boltzmann equation with a collisional term

For the forthcoming applications we shall need to investigate interactions of the
following type:
1+2↔3+4, (3.158)
among generic species that we dub 1, 2, 3 and 4. This reaction can describe
scattering or annihilation and will be suitable for discussing BBN, recombination
and calculating the expected relic abundance of CDM.
Let us take the particle 1 as reference and let us focus on its Boltzmann equa-
tion:
df1
= C[f ] . (3.159)
dλ
The collisional term is the same for all the particles, and has dimension of an
inverse time, i.e. it is a scattering rate. Using the Liouville operator of Eq. (3.146),
we can cast the above equation as
∂f1 ∂f1 1
− Hp1 = 0 C[f ] , (3.160)
∂t ∂p1 P1
where we have passed from the affine parameter λ to the time t, and for this reason
P10 appears. Taking the moment zero of the above equation, and using the result
of Eq. (3.153), we get:
1 d(n1 a3 ) d3 p 1
Z
= C[f ] . (3.161)
a3 dt (2π~)3 P10
Introducing the general expression of the right hand side [Kolb and Turner, 1990],
the above equation becomes:
1 d(n1 a3 ) d3 p 1 d3 p 2 d3 p 3 d3 p4
Z Z Z Z
=
a3 dt (2π~)3 2E1 (2π~)3 2E2 (2π~)3 2E3 (2π~)3 2E4
(2π)4 δ (3) (p1 + p2 − p3 − p4 )δ(E1 + E2 − E3 − E4 )|M|2
[f3 f4 (1 ± f1 )(1 ± f2 ) − f1 f2 (1 ± f3 )(1 ± f4 )] , (3.162)
70
p
where Ei = p2i + m2i . We have to provide several comments.
• Since we took the zero moment of Eq. (3.159), the left hand side is the time
derivative of the number of particles of the species 1. See also Eq. (3.153).
• We have incorporated the particles degeneracies gs in the distribution func-
tions.
• The integrals are over the particles momenta. On the other hand, the total
four-momentum must be conserved, hence the Dirac deltas in the second line. Note
the E1 contribution in the first integration. You may think that it comes from the
P10 factor in Eq. (3.161), but this would not explain the factor 2. Indeed, one must
rather look at the right hand side of Eq. (3.162) as a definition of the momentum
integrated C[f ].
• The fundamental physics of the interaction is represented by the amplitude
|M|2 . We have also assumed symmetric interaction, i.e. for fixed particles four-
momenta the amplitude probability for 1+2 → 3+4 is the same as for 1+2 ← 3+4.
• In the last line, we have a balance: the more particle 3 and 4 we have, the
more they react and produce particles 1 and 2. And vice-versa. Since our reference
particle is the 1, we have the combination f3 f4 − f1 f2 .
• Finally, the contributions of the type 1 + f and 1 − f are called Bose enhance-
ment and Pauli blocking, respectively. They represent the fact that it is easier to
produce a boson rather than a fermion because, due to Pauli exclusion principle,
there are more states available to the former than to the latter.
• The volume element d3 p/2E is covariant. It comes from the fact that we
have enforced E 2 = p2 c2 + m2 c4 for each particle. We now prove it. When we want
to enforce the dispersion relation E 2 = p2 c2 + m2 c4 we use a Dirac delta in the
four-momentum space:
Z Z ∞ Z Z ∞
3 2 2 2 3
dp dE δ(P + m c ) = d p dE δ(E 2 − p2 c2 − m2 c4 ) θ(E) , (3.163)
0 −∞
where the square modulus of the four-momentum is P 2 = −E 2 /c2 + p2 . The

Heaviside function θ(E) serves to choose only the positive values of the energy.
The above equation is covariant. Using the known relation:
X δ(x − xi )
δ[F (x)] = , (3.164)
i
|F 0 (xi )|
where the xi ’s are the roots of the generic function F (x), one gets:
p p
δ(E − p 2 c2 + m2 c4 ) δ(E + p2 c2 + m2 c4 )
δ(E 2 − p2 c2 − m2 c4 ) = + , (3.165)
2E+ 2E−
p
where E± = ± p2 c2 + m2 c4 .
The second term of Eq. (3.165) integrated with θ(E) in Eq. (3.163) vanishes
and we are left with:
Z Z ∞ Z 3
3 2 2 2 2 4 dp
dp dE δ(E − p c − m c ) θ(E) = , (3.166)
−∞ 2E+
which is what we wanted to prove.
71
We now focus our attention to Eq. (3.162) and make some assumptions in
order to simplify it. In particular, we shall always assume thermal equilibrium and
sufficiently small temperatures such that:
E − µ kB T (3.167)
in order to use the FD and BE distributions simplified as follows:
f ≈ e−E/(kB T ) eµ/(kB T ) , (3.168)
i.e. forgetting the ±1 (using thus the Maxwell-Boltzmann distribution).
Exercise 3.22. Show that using Eq. (3.168), the last line of Eq. (3.162) can be
simplified as:
f3 f4 (1 ± f1 )(1 ± f2 ) − f1 f2 (1 ± f3 )(1 ± f4 ) ≈
e−(E1 +E2 )/(kB T ) e(µ3 +µ4 )/(kB T ) − e(µ1 +µ2 )/(kB T ) .

(3.169)
In particular, we can neglect the Bose enhancement and Pauli blocking terms. One
has to use energy conservation E3 + E4 = E1 + E2 in order to obtain the above
equation.
With the approximation given in Eq. (3.168), the particle number density can
be simplified as follows:
d3 p d3 p −E/(kB T )
Z Z
µ/(kB T )
n = gs f ≈ gs e e , (3.170)
(2π~)3 (2π~)3
where we could extract the chemical potential from the integral since it does not
depend on the particle momentum but rather on the temperature of the thermal
bath. For µ = 0 we define the equilibrium number density as:
d3 p −E/(kB T )
Z
(0)
n = gs e (3.171)
(2π~)3
3.8.1 Detailed calculation of the equilibrium number density

Now we show how to calculate the equilibrium number density in the regimes
mc2 kB T and mc2 kB T , which we proved earlier to be regimes in which the
particles are non-relativistic and relativistic, respectively. At the same time, we
justify the approximation of Eq. (3.168). Let us start from:
Z ∞
d3 p p2
Z
(0) 1 gs
n = gs = dp √ , (3.172)
(2π~)3 eE/(kB T ) ± 1 2π 2 ~3 0 2 2 2 4
e p c +m c /(kB T ) ± 1
and split the integration, let us dub it I, into two contributions, one from zero up
to mc and the other from mc to infinity, i.e.
Z mc Z ∞
p2 p2
I= dp √ 2 2 2 4 + dp √ 2 2 2 4 . (3.173)
0 e p c +m c /(kB T ) ± 1 mc e p c +m c /(kB T ) ± 1
72
The non-relativistic case mc2 kB T
Now let us start considering the non-relativistic case mc2 kB T . We shall use
different approximations within the two integrals in which we split I:
Z mc Z ∞
p2 p2
I= dp 2 √ 2 2 2 + dp √ 2 2 2
. (3.174)
0 emc p /(m c )+1/(kB T ) ± 1 mc epc 1+m c /p /(kB T ) ± 1
Now we use the mc2 kB T condition. In the first integral p < mc because of
the integration limits, so we can expand the square root with respect to p/(mc).
The second integral is negligible with respect to the first one because, since pc >
mc2 kB T , the integrand is always exponentially small. Thus we are left with:
Z mc
−mc2 /(kB T ) 2
I=e dp p2 e−p /(2mkB T ) , (3.175)
0
where we kept just the dominant term for the first integral. This is the same
integral that we would have obtained had we started from Eq. (3.171).
After changing variable, we get:
Z mc/√2mkB T
−mc2 /(kB T ) 2
I=e (2mkB T ) 3/2
dx x2 e−x , (3.176)
0
and the result is:

" s
2
√ s
2
!#
2 mc 2 π mc
I = e−mc /(kB T ) (2mkB T )3/2 − e−mc /(2kB T ) + Erf .
8mkB T 4 2mkB T
(3.177)
The dominant contribution of the above expression comes from the Erf function,
which can be approximated to 1 in the mc2 kB T limit, and thus:
√
−mc2 /(kB T ) 3/2 π
I=e (2mkB T ) . (3.178)
4
Finally, we can express the equilibrium number density for any species in thermal
equilibrium and mc2 kB T as:
3/2
mkB T 2
(0)
n = gs 2
e−mc /(kB T ) , for mc2 kB T . (3.179)
2π~
The relativistic case mc2 kB T

Now we adopt the same technique in the other relevant limit, the one of relativistic
particles, for which mc2 kB T . We split the integral I in the same way as before:
Z mc Z ∞
p2 p2
I= dp 2 √ 2 2 2 + dp √ , (3.180)
mc p /(m c )+1/(kB T ) pc 1+m2 c2 /p2 /(kB T )
0 e ±1 mc e ±1
but now the mc2 kB T condition allows us to make the following approximations:
Z mc Z ∞
p2 p2
I= dp 2
p + dp pc(1+m2 c2 /2p2 )/(k T ) .
0 1 + kmc p2 /(m2 c2 ) + 1 ± 1 mc e B ±1
BT
(3.181)
73
Now it is the first integral that is subdominant with respect to the second one, and
I can be written as:
Z ∞
(kB T )3 ∞
Z
2 −pc/(kB T )
I= dp p e = dx x2 e−x . (3.182)
mc c3 mc2 /(kB T )
The solution of the integration is:
(kB T )3 −mc2 /(kB T ) mc2 mc2

I= e 2+ 2+ . (3.183)
c3 kB T kB T
In the mc2 kB T regime, this becomes:
2(kB T )3
I= . (3.184)
c3
Finally, we can express the equilibrium number density for any species in thermal
equilibrium and mc2 kB T as:
(kB T )3
n(0) = gs , for mc2 kB T . (3.185)
π 2 ~3 c3
3.8.2 Saha equation

(0)
Expressing the contributions eµi /(kB T ) as ratios ni /ni , we can recast the collisional
Boltzmann’s equation (3.162) as follows:
!
3
1 d(n1 a ) (0) (0) n3 n4 n1 n2
3
= n1 n2 hσvi (0) (0)
− (0) (0)
(3.186)
a dt n3 n4 n1 n2
where we have defined the thermally averaged cross section as follows:
d3 p 1 d3 p 2 d3 p 3 d3 p 4
Z Z Z Z
1
hσvi ≡ (0) (0)
n1 n2 (2π~)3 2E1 (2π~)3 2E2 (2π~)3 2E3 (2π~)3 2E4
(2π)4 δ (3) (p1 + p2 − p3 − p4 )|M|2 e−(E1 +E2 )/(kB T ) . (3.187)
Observe the following very important point about Eq. (3.186). When there are no
interactions, i.e. when hσvi = 0, we recover the collisionless Boltzmann equation
that we have discussed earlier. On the other hand, when the interaction rate is
extremely high, i.e. it is much larger than than the Hubble rate:
(0) (0) 1 d(n1 a3 )

n1 n2 hσvi ∼ n1 H , (3.188)
a3 dt
then in order for Eq. (3.186) to hold true, we must have:
n3 n4 n1 n2
(0) (0)
= (0) (0)
(3.189)
n3 n4 n1 n2
74
This equation can be written as Eq. (3.7) and thus represents chemical equilib-
rium, as we saw at the beginning of this chapter. Equation (3.189) is also known
as Saha equation.
So we see that when Γ H then Saha equation (3.189) must hold and Boltz-
mann equation (3.186) becomes:
1 d(n1 a3 )
=0, (3.190)
a3 dt
which is the same Boltzmann equation that we would have if interactions were
absent! This similarity between high interaction rate and no interaction at all is
intriguing and responsible of the very low degree of CMB polarisation, as we shall
see in Sec. 10.
3.9 Big-Bang Nucleosynthesis

The BBN is the formation of the primordial light elements, mainly helium. It took
place at a temperature (photon temperature) of about 0.1 MeV, which corresponds
to a redshift z ≈ 109 .
In order to investigate the BBN, we need to know the characters of the story.
At temperatures say larger than 1 MeV the primordial plasma was formed by pho-
tons, electrons, positrons, neutrinos, antineutrinos, protons and neutrons. We have
already seen how photons, electrons, positrons, neutrinos, antineutrinos interact
among each other, so now we focus on protons and neutrons. Their interactions
relevant to BBN are:
n ↔ p + e− + ν̄e (beta decay) , (3.191)

−
p + e ↔ νe + n (electron capture) , (3.192)
+
p + ν̄e ↔ e + n (inverse beta decay) . (3.193)
As we have seen, at a temperature of about 1 MeV neutrinos decouple. Therefore,

the β-decay reaction in Eq. (3.191) (from left to right) takes over and the number
of neutrons starts to diminish. On the other hand, they can also be captured
by protons and form deuterium nuclei. The BBN is essentially a competition in
capturing neutrons before they decay.
3.9.1 The baryon-to-photon ratio

A very important number for BBN and cosmology is the baryon-to-photon ratio
ηb , which we have already encountered at the beginning of this chapter. It is defined
as follows:
Ωb0 h2

nb −10
ηb ≡ = 5.5 × 10 (3.194)
nγ 0.020
i.e. as the ratio between the number of baryons and the number of photons.
We have defined it via number densities, which individually are time-dependent
75
quantities but whose ratio is fixed since both scales as 1/a3 . The above numbers
in Eq. (3.194) can be found as follows:
nb εb0
ηb = = , (3.195)
nγ mb c2 nγ0
where we have assumed the baryons to be nonrelativistic, which is indeed the case
at the temperatures we are dealing with (kB T ∼ MeV) since the proton mass is of
the order of 1 GeV.
Using Eq. (3.75) and Eq. (3.80), we can write
nγ0 30ζ(3)
= 2 . (3.196)
εγ0 π kB T0
Therefore,
π 4 kB T0 εb0 π 4 kB T0 Ωb0
ηb = = , (3.197)
30ζ(3)mb c2 εγ0 30ζ(3)mb c2 Ωγ0
where in the last equality we have multiplied and divided by the present critical
energy density in order for the density parameters to appear.
Exercise 3.23. Show that Eq. (3.197) leads to Eq. (3.194). Use the mass of the
proton as mb since the baryon energy density is indeed dominated by the mass
energy density of the protons.
The fact that there is a billion photon for each proton and electron is very
important for the following reason. Even if the temperature of the thermal bath
is lower than the binding energy of deuterium, i.e. 2.2 MeV, there are still many
photons with energy higher than 2.2 MeV which are able to break newly formed
deuterium nuclei. This is also known as deuterium bottleneck.
As Kolb and Turner comment in their book [Kolb and Turner, 1990, page 92],
it is not deuterium’s fault if BBN takes place at temperature much smaller than
2.2 MeV. Rather, the very high entropy of the universe, i.e. the smallness of ηb , is
the culprit.
3.9.2 The deuterium bottleneck

The deuterium bottleneck is the situation in which newly formed deuterium nuclei
are destroyed by photons. With no deuterium available, BBN cannot take place.
Let us start considering the reaction:
p+n↔D+γ , (3.198)
at chemical equilibrium. Using Saha equation (3.189), we have

nD nγ np nn
(0) (0)
= (0) (0)
. (3.199)
nD nγ np nn
76
(0)
Neglecting the photon chemical potential, i.e. nγ = nγ , and using Eq. (3.179) we
obtain:
(0) 3/2
2π~2 mD

nD nD gD 2
= (0) (0) = e−(mD −mp −mn )c /(kB T ) . (3.200)
np nn np nn g g
p n m m k
p n B T
The deuterium has spin 1, whereas protons and neutrons have spin 1/2. Therefore,
3/2
2π~2 mD

nD 3
= eBD /(kB T ) , (3.201)
np nn 4 mp mn kB T
where BD = 2.22 MeV is the deuterium binding energy. We can write the masses
ratio as follows:
mD mp + mn − BD /c2 1 1 BD
= = + − . (3.202)
mp mn mp mn mn mp mp mn c2
Introducing the neutron-proton mass difference [Wilczek, 2015]:
Q ≡ (mn − mp )c2 = 1.239 MeV , (3.203)
we can write
mD 1 1 BD
= + − 2 . (3.204)
mp mn mp (1 + Q/mp c ) mp mp (1 + Q/mp c2 )c2
2
Since Q/mp c2 ∼ BD /mp c2 ∼ 10−3 , we approximate:

mD 2
≈ . (3.205)
mp mn mp
Moreover, being nn = np = nb , we can write:

3/2
4π~2

nD 3
= nb eBD /(kB T ) , (3.206)
nb 4 mp kB T
i.e. we obtain a deuterium-to-baryon ratio which defines how many baryons exist
that are deuterium nuclei. Dealing with the nb on the right hand side as follows:
(kB T )3
nb = ηb nγ = ηb n(0)
γ = 2ηb 2 3 3 , (3.207)
π ~c
where we have used Eq. (3.185) for the photon number density, we get
3/2
nD 12 kB T
= √ ηb eBD /(kB T ) . (3.208)
nb π mp c2
As long as kB T BD the relative deuterium abundance is completely negligible

since it is exponentially suppressed. However, even when kB T ∼ BD the rela-
tive abundance is very small, because of the prefactor ηb . This is the deuterium
bottleneck.
77
Typically, the temperature TBBN at which the BBN starts is the one at which
the bottleneck is overcome. This is because, as numerical calculations show, once
deuterium is formed it rapidly combines into Helium.
So, we define TBBN as the one for which nD = nb :

12 3 kB TBBN BD
log √ ηb + log 2
=− . (3.209)
π 2 mp c kB TBBN
Numerically solving this equation, one finds
kB TBBN ≈ 0.07 MeV (3.210)
Note that we can use Saha equation only until chemical equilibrium holds true.
When the reaction p + n ↔ D + γ unbalances and deuterium is formed, we must
use the full Boltzmann equation (3.186). However, it can be shown numerically
that the two equations provide compatible results up to the moment in which the
equilibrium is broken. Therefore, Saha equation is an useful tool for estimating
when the equilibrium is broken but, of course, if one needs precise results one
should solve the full Boltzmann equation.
3.9.3 Neutron abundance

After the deuterium bottleneck is overcome, BBN takes place in the following chain
reactions:
p+n→D+γ , (3.211)
D + D → 3 He + n , (3.212)
D + 3 He → 4 He + p . (3.213)
In the following, we shall assume that the three above reactions take place in-
stantaneously and the neutrons are captured in Helium nuclei. This is not what
occurred, of course, but it turns out to be a good approximation which allows us
to perform easy calculations.
Note that Lithium 3 Li is also produced, but in tiny fraction (one billionth of
the hydrogen abundance). However, measuring its abundance in the universe is a
very important independent measure of Ωb0 .
The main prediction of BBN is on the abundance of 4 He because this is the
element which is mostly formed, due to both its high binding energy per nucleon,
which is about 7 MeV, see Fig. 3.2, but also to the fact that there is not much
time for forming heavier nuclei since the thermal bath is rapidly cooling and ηb is
so small.
Our objective is thus to determine the neutron abundance at TBBN . This is
done considering two reactions. The electron capture:
p + e− ↔ n + ν , (3.214)
and the β-decay:

n ↔ p + e− + ν̄ . (3.215)
78
Figure 3.2: Binding energy per nucleon. Figure taken from https://en.
wikipedia.org/wiki/Helium-4.
The β-decay will provide just an exponential suppression on the abundance pre-
dicted by the the electron capture. Therefore, we focus on the latter.
For temperature kB T 1 MeV, protons and neutrons are in chemical equilib-
rium:
(0) 3/2
np np mp 2
= (0) = e(mn −mp )c /(kB T ) ∼ eQ/(kB T ) , (3.216)
nn nn mn
where Q = 1.239 MeV, see Eq. (3.203). When kB T Q, the mass difference
between protons and neutrons is irrelevant, and therefore they are in chemical
equilibrium. When kB T drops below Q nature starts to favor protons because they
are energetically more “economic” and neutrons start to disappear. However, as
we mentioned earlier, out of equilibrium we cannot use Saha equation but have to
solve the full Boltzmann equation:
!
1 d(nn a3 ) (0) n n
p l n n
n l
= n(0)
n nl hσvi (0) (0)
− (0) (0) , (3.217)
a3 dt np nl nn nl
where we have denoted with the subscript l the leptons, either electron or neutrino,
involved in the electron capture process. We assume their chemical potentials to
be zero and simplify Eq. (3.217) as follows:
!
3 (0)
1 d(nn a ) (0) np nn
3
= nl hσvi (0)
− nn . (3.218)
a dt np
79
Let us define the neutron abundance and the scattering rate as follows:
nn (0)
Xn ≡ , nl hσvi ≡ λnp . (3.219)
nn + np
Exercise 3.24. Show that Eq. (3.218) can be cast, using Eq. (3.219), as follows:
dXn
= λnp (1 − Xn )e−Q/(kB T ) − Xn .

(3.220)
dt
In order to solve (numerically) Eq. (3.220), we introduce the variable:
Q
x≡ . (3.221)
kB T
Exercise 3.25. Show that the time derivative of x ≡ Q/(kB T ) can be cast as
follows: r
dx 8πGε
= Hx = x . (3.222)
dt 3c2
Since we are deep in the radiation-dominated era, we can write the energy
density of Eq. (3.222) as follows:
π 2 (kB T )4
ε= g∗ (3.223)
30(~c)3
where the effective number of relativistic degrees of freedom is

X 7 X
g∗ ≡ gi + gi , (3.224)
i=bosons
8 i=fermions
where recall the 7/8 factor coming from Eq. (3.83). The effective number of rel-
ativistic degrees of freedom is actually a function of the temperature, since it
decreases when a certain species becomes non-relativistic. For temperature larger
than 1 MeV g∗ is roughly a constant and its value is:
7
g∗ = 2 + (3 + 3 + 2 + 2) = 10.75 , (3.225)
8
where we have considered two degrees of freedom coming from the photons, 3
+ 3 coming from neutrinos and anti-neutrinos and 2 + 2 coming from electrons
and positrons. We have considered just a single spin state for each neutrino and
anti-neutrino.
80
Exercise 3.26. Show that the Hubble parameter can thus be cast in the following
form:
8πG π 2 (kB T )4 4π 3 Gg∗ Q4 1
H2 = g ∗ = . (3.226)
3c2 30(~c)3 45c2 (~c)3 x4
Calculate:
H(x = 1) = 1.13 s−1 . (3.227)
Show that Eq. (3.220) can be cast as:
dXn xλnp −x
e − Xn (1 + e−x ) .

= (3.228)
dx H(x = 1)
We only need a last piece of information, i.e. the interaction rate λnp :
255
λnp = 5
(12 + 6x + x2 ) , (3.229)
τn x
where
τn = 886.7 s , (3.230)
is the neutron lifetime. The above scattering rate can be found in [Bernstein, 1988].
We can now solve numerically Eq. (3.228) together with Eq. (3.229). The
initial condition on Xn is of course Xn (x → 0) = 1/2, as we can see from the Saha
equation (3.216). We plot the evolution of Xn in Fig. 3.3.
��
��
��
��
��
��
��
��
��
Figure 3.3: Evolution of Xn from Eq. (3.228).
As we can appreciate from Fig. 3.3, Xn → 0.15 as x → ∞. This number

is not a very good approximation of the residual abundance of neutrons at TBBN
81
because there are other relevant processes which contribute to deplete or enhance
the number of neutrons. Namely:
n → p + e− + ν̄ , (3.231)
n+p→D+γ , (3.232)
i.e. the β-decay and the neutron-proton capture. These processes lower the number
of free neutrons. The Helium-3 formation:
D + D → 3 He + n , (3.233)
puts back into play another neutron and the Helium-4 formation:
3
He + D → 4 He + p , (3.234)
reinserts into play another proton, which helps in capturing neutrons, thus lowering
their number.
The right way to calculate the abundances of the light elements produced dur-
ing BBN is to consider all the coupled Boltzmann equations for all the relevant
reactions taking place. This is, of course, done numerically and the standard code
is Wagoner’s one [Wagoner, 1973] (there has been refinements since then).
We now show that correcting Xn = 0.15 by taking into account only the β-
decay gives a result which is in surprising agreement with the more reliable one
which takes into account all the reactions.
What we have to do is to weigh Xn = 0.15 with exp(−tBBN /τn ),6 where tBBN
is the time corresponding to kB TBBN = 0.07 MeV, i.e. the time at which BBN
starts. Moreover, we suppose that at this time all the free neutrons are immediately
captured and produce Helium-4. Thereby, estimating Xn gives a direct estimation
of X4 He .
At kB TBBN = 0.07 MeV, electrons and positrons have already annihilated.
Therefore, the effective relativistic degrees of freedom are:
4/3
7 4
g∗ = 2 + 6 ≈ 3.36 , (3.235)
8 11
where we have taken into account the temperature difference between photons and
neutrinos. √
Since T ∝ 1/a and in the radiation-dominated epoch a ∝ t, we can relate
time and temperature as follows:
1 2 4π 3 G(kB T )4
= H = g∗ . (3.236)
4t2 45c2 (~c)3
Exercise 3.27. Show from Eq. (3.236) that:

2
0.07 MeV
t = 271 s. (3.237)
kB T
6
This exponential weight comes from Poisson distribution, which governs stochastic processes
such as the β-decay. We derive it in Chapter 12.
82
Therefore, the expected abundance of neutrons at kB TBBN = 0.07 MeV is:
Xn (TBBN ) = 0.15 · e−271/886.7 ≈ 0.11 . (3.238)
Assuming that all the neutrons end up in Helium-4 nuclei, we have the prediction:
4n4 He
YP ≡ 4X4 He ≡ = 2Xn (TBBN ) = 0.22 . (3.239)
nb
The factor 4 comes from the fact that YP is a mass fraction and each Helium-4
nucleus contains 4 baryons. We have also assumed mb = mp = mn .
The more accurate numerical result is [Kolb and Turner, 1990]:
η
b
YP = 0.2262 + 0.0135 log −10
, (3.240)
10
which is in very good agreement with our “back-of-the-envelope” calculation. In
Fig. 3.4 the time-evolution plots of the mass fractions of various elements are
displayed.
η = 6.23x10-10 Nν= 3.0 H0= 70.50 km/s/Mpc Tcmb= 2.725 K

1
10
p p
-1
10 n he4
he4 n
-3 d
10
t d
-5 t
10 he3
Mass Fraction
he3
t
-7
10
be7
-9
li7
10 n
d t
li7
-11
10 he4
-13
10
li7 li6 be7 li6
-15
he3 b11
10
1 2 3 4 5
10 10 10 10 10
Time (s)
Figure 3.4: Time-evolution of the mass fraction of various elements. Figure

taken from http://cococubed.asu.edu/images/net_bigbang/bigbang_time_
2010.pdf.
83
3.10 Recombination and decoupling
Recombination is the process by which neutral hydrogen is formed via combi-
nation of protons and electrons. Decoupling is generally refered to be the epoch
when photons stop to interact with free electrons and their mean free path becomes
larger than the Hubble radius and we are able to detect them as CMB coming from
the last scattering surface. For the two events, the relevant interactions are:
p + e− ↔ H + γ , (3.241)
e− + γ ↔ e− + γ (Compton/Thomson scattering) . (3.242)
Recombination and decoupling temporally occur close to each other for the follow-
ing reason. At sufficiently low temperatures, which we will calculate, photons are
no more able to break hydrogen atoms and so these start to form in larger number
(recombination). Being captured in hydrogen atoms, the number of free electrons
dramatically drops and the Thomson scattering rate goes to zero (decoupling).
The seminal paper on recombination is [Peebles, 1968].
In order to determine the epoch of recombination, let us use again Saha equa-
tion:
(0) (0)
ne np ne np
= (0)
. (3.243)
nH nH
Let us assume neutrality of the universe, i.e. ne = np and define the free electron
fraction:
ne np
Xe ≡ = , (3.244)
ne + nH np + nH
Exercise 3.28. Considering that the degeneracy of the hydrogen atom, in the
state 1s, is g1s = 4 (it has two hyperfine states, one of spin 0 and the other of spin
1), show that Saha equation can be written, using Eq. (3.179), as:
3/2
Xe2

1 me mp kB T 2 /(k
= e−(me +mp −mH )c BT )
. (3.245)
1 − Xe ne + nH 2mH π~2
Consider the contribution at the denominator of the right hand side as:
3
2ζ(3) 3 −9 (kB T )
ne + nH = nb = ηb nγ = ηb (kB T ) ≈ 10 . (3.246)
π 2 ~3 c3 ~3 c3
Look at the first equality as follows: the total electron number is made up of
those which are free plus those which have already been captured. Moreover, the
total electron number density is the same as the baryon number density because
electrons are baryons (in the jargon of cosmology).
We again neglect the mass difference elsewhere than at the exponential, and
write: 3/2
Xe2 me c2

9 13.6 eV
≈ 10 exp − , (3.247)
1 − Xe 2πkB T kB T
84
where we used
ε0 ≡ (me + mp − mH )c2 = 13.6 eV , (3.248)
i.e. the ionisation energy of the hydrogen atom.
The high photon-to-baryon number delays recombination as well as it delayed
BBN. Indeed, when kB T = 13.6 eV, we get from Eq. (3.247):
Xe2
≈ 1015 , (3.249)
1 − Xe
From which one gets that Xe ≈ 1. This means that even when the energy of
the thermal bath drops below the ionisation energy of the hydrogen atom, still no
hydrogen is formed and the electrons remain free. This, again, happens because
there are still many photons with energy much higher than 13.6 eV.
As we already mentioned, Saha equation works until chemical equilibrium is
maintained. In Tab. 3.1 we show numerical calculations of Xe from Eq. (3.247) in
order to have a hint about the time of recombination.
kB T [eV] Xe
0.5 1
0.38 0.995
0.36 0.970
0.34 0.819
0.32 0.434
0.30 0.137
0.29 0.067
0.25 0.001
Table 3.1: Free electron fraction at different photon temperatures.
From Tab. 3.1 we see that the free electron fraction falls abruptly at about
kB T ≈ 0.30 eV.
Exercise 3.29. Calculate at which redshift corresponds the energy kB T = 0.3 eV

of the photon thermal bath.
In Fig. 3.5 we numerically solve Saha equation (3.247) and use both kB T and
the redshift as variables.
In order to accurately calculate Xe , we need to use the full Boltzmann equation
(3.186), which for recombination becomes:
!
1 d(ne a3 ) n n
H γ n n
e p
= n(0) (0)
e np hσvi (0) (0)
− (0) (0) . (3.250)
a3 dt nH nγ ne np
(0)
We assume nγ = nγ and ne = np again. Therefore,
" 3/2 #
3
1 d(ne a ) me kB T
3
= hσvi nH 2
e−ε0 /(kB T ) − n2e . (3.251)
a dt 2π~
85
��
��
��
��
��
��
��
��
Figure 3.5: Numerical solution of the Saha equation (3.247).
Introducing now the free electron fraction Xe defined in Eq. (3.244), we get:
" 3/2 #
dXe me kB T
= hσvi (1 − Xe ) e−ε0 /(kB T ) − Xe2 nb . (3.252)
dt 2π~2
As we did for BBN in Eq. (3.207), we can replace nb with:

(kB T )3
nb = 2ηb . (3.253)
π 2 ~ 3 c3
Now we need the fundamental physics of the capture process. It is given by:
2
1/2
(2) 2 ~ ε0 ε0
hσvi ≡ α = 9.78 α 2 log , (3.254)
me c kB T kB T
where α = 1/137 is the fine structure constant. The superscript (2) serves to
indicate that the best way to form hydrogen is not via the capture of an electron
in the 1s state, because this generates a 13.6 eV photon which ionises another
newly formed H.
The efficient way to form hydrogen is to form it in a excited state. When it
relaxes to the ground state, the photons emitted have not enough energy to ionise
other hydrogen atoms.
For example, an electron captured in the n = 2 state generates a 3.4 eV photon.
Subsequently, when the electron falls in the ground state, the hydrogen releases
86
another 10.2 eV photon. Neither of the two photons has sufficient energy for
ionising another hydrogen atom in the ground state.
We now solve numerically Eq. (3.252) together with Eq. (3.207) and Eq. (3.254).
We use the redshift as independent variable:
dXe dXe dz dXe
= =− H(1 + z) , (3.255)
dt dz dt dz
and the following Hubble parameter:
H2
2
= Ωm0 (1 + z)3 + Ωr0 (1 + z)4 + ΩΛ , (3.256)
H0
where for z ≈ 1100 is the matter contribution the dominant one.
Therefore, we rewrite Eq. (3.252) as follows:
" 3/2 #
dXe hσvi me kB T
=− (1 − Xe ) 2
e−ε0 /(kB T ) − Xe2 nb , (3.257)
dz H(1 + z) 2π~
also taking into account that the photon temperature T scales as:
T = T0 (1 + z) , (3.258)
with T0 = 2.725 K.
In Fig. 3.6 we show the numerical solution of the Boltzmann equation (3.257),
compared with the solution of the Saha equation (3.247). Note how the two solu-
tions are compatible for high redshifts, but that of the Boltzmann equation predicts
a residual free electron fraction of about Xe ≈ 10−3 .
At the same time of recombination the decoupling of photons from electron
takes place. As we anticipated, this happens because very few free electrons remain
after hydrogen formation and therefore photons are free to propagate undisturbed
and seen by us as the CMB.
3.10.1 Decoupling
As we have already mentioned at the beginning of this chapter, roughly speaking
in the expanding universe any kind of reaction stops occurring when its interaction
rate Γ becomes of the order of H. In the case of photons and electrons, the relevant
process is Thomson scattering, for which:
ΓT = ne σT c = Xe nb σT c , (3.259)
where ne is the free-electron number density, which we have written as Xe nb be-

cause we have neglected the Helium abundance. The baryon number density can
be expressed as
ρb 3H02 Ωb0
nb = = , (3.260)
mb 8πGmp a3
where we have identified mb = mp since it is the proton mass that dominates the
baryon energy density.
87
��
��
��
��
��-�
��
Figure 3.6: Solid line: numerical solution of the Boltzmann equation (3.257).
Dashed line: solution of the Saha equation (3.247).

ΓT ne σT c Xe Ωb0 H0
= = 0.0692 h . (3.261)
H H Ha3
As for H, we consider a matter plus radiation universe, for which:

Ωm0 aeq
H 2 = H02 3 1 + . (3.262)
a a
Exercise 3.31. Using the above Hubble parameter show that:

1/2 3/2 −1/2
Ωb0 h2

ΓT 0.15 1+z 1 + z 0.15
= 113Xe 1+ .
H 0.02 Ωm0 h2 1000 3600 Ωm0 h2
(3.263)
The decoupling redshift zdec is defined to be that for which ΓT = H. Note that
in the above equation Xe is a function of the redshift, which we have plotted in
Fig. 3.6. On the other hand, Xe drops abruptly during recombination, so that the
factor 113 is easily overcome. For this reason recombination and decoupling take
place at roughly the same time.
88
��
��
��
��
��
��
��
Figure 3.7: Solid line: numerical solution of the ratio ΓT /H of Eq. (3.263). Dashed
line: numerical solution of Eq. (3.257) for Xe .
Now imagine that no recombination takes place, i.e. Xe = 1. The above

equation then gives the decoupling redshift:
2/3 1/3
Ωm0 h2

0.02
1 + zdec = 43 . (3.264)
Ωb0 h2 0.15
This is the freeze-out redshift of the electrons, i.e. eventually photons and electrons
do not interact anymore because they are too diluted by the cosmological expan-
sion. This would happen for a redshift 42. This number is important because of the
following. Well after decoupling, ultraviolet light emitted by stars and gas is able
to ionise again hydrogen atoms. This phase is called reionisation. If the latter
ocurred for redshifts smaller than 42, the newly freed electron would not interact
with photons because they are too much diluted by the cosmological expansion.
Indeed, reionization takes place for zreion ≈ 10, so that the CMB spectrum is poorly
affected, as we shall see in Chapter 10.
3.11 Thermal relics

In this section we investigate in some detail the freeze-out and relic abundance
of CDM. In general, with thermal relic one refers to the abundance of a certain
species left over from the annihilation suffered in thermal bath and, after its decou-
pling, from the dilution caused by the expansion of the universe. To this purpose,
89
consider the following process:
X + X̄ ↔ l + ¯l , (3.265)
where X represents the DM particle and l a lepton. Due to the expansion of the
universe, at a certain point the annihilation of DM is no more efficient and thus
its abundance is fixed. This is precisely what we want to calculate. We could then
put constraints on the cross-section of the above process and on the mass of the
DM particle by measuring the abundance of DM necessary today in order to be in
agreement with the cosmological observations.
We are assuming here that DM is made up of massive particles which were in
thermal equilibrium with the rest of the standard model particles in the primordial
universe, hence the name thermal relics. Moreover, since we are considering
CDM, or more in general particle which decouple from the primordial plasma
when nonrelativistic, ours is a calculation of cold relics abundance.
The Boltzmann equation for the above process has the following form:
1 d(nX a3 )
(0)2 2

= hσvi nX − nX , (3.266)
a3 dt
(0)
where we have assumed nl = nl , because we are in thermal bath with very high
temperature (e.g. larger than 100 GeV for WIMPs).
Now we take advantage of the scaling T ∝ 1/a and define the following dimen-
sionless quantities:
(0)
nX ~3 n ~3 mc2 (mc2 )3 hσvi
Y ≡ , Yeq ≡ X 3 , x≡ , λ≡ , (3.267)
(kB T )3 (kB T ) kB T H(m)~3
where H(m) = H(x = 1) is the Hubble parameter corresponding to a thermal
energy kB T = mc2 , with m being the mass of the DM particle. Note that, being
deep in the radiation-dominated era, the Hubble parameter scales as follows:
H(x = 1)
H= . (3.268)
x2
The Boltzmann equation thus becomes:
(kB T )3 dY H(x = 1) (kB T )6 2 2

= hσvi Y eq − Y , (3.269)
~3 dx x ~6
and finally
dY λ
= − 2 Y 2 − Yeq2

(3.270)
dx x
In this case Saha equation is simply Y = Yeq and from Eq. (3.107) we know that
Yeq → 0 for x → ∞, because of the dilution due to the cosmological expansion.
On the other hand, we also expect, as we saw for recombination, that Y attains an
asymptotic value which we call Y∞ and with which we shall calculate the present
abundance of DM. The departure between the Saha equation solution and the
Boltzmann equation solution is the freeze-out and, as we know, it takes place
approximately when Γ ∼ H.
90
Using Eq. (3.107) in order to express Yeq , we can numerically solve Eq. (3.270).
In order to do this, we set a small initial value x = xi such that Y (xi ) = Yeq (xi ) and
fix λ to be a constant. In Figs. 3.8 and 3.9 we choose xi = 0.01 and λ = 1, 10, 100
for a fermionic DM species.
��
��
��
��
��
��
Figure 3.8: Numerical solution of Eq. (3.270) for the case of fermionic DM.
From Fig. 3.8 one can appreciate that Y attains a residual abundance and that,
as expected, the latter is smaller the larger λ is. This happens because for larger
values of λ the interaction is more efficient.
In Fig. 3.9 we show the behaviour of the relative difference Y /Yeq −1 in order to
understand when the freeze-out approximately takes place. Of course, the freeze-
out is not a specific instant, but depends on a criterion that we choose. For example,
from inspection of Fig. 3.9, we see that Y /Yeq − 1 at x ≈ 10 starts to increase with
more steepness and therefore we might establish that xf ≈ 10. Moreover, this value
is very weakly dependent on λ.
Now, what is the difference between fermionic and bosonic DM? In Fig. 3.10
we plot Y in the bosonic DM (solid lines) and fermionic DM (dashed lines) for
λ = 10, 100. As one can see, there are two differences: i) When x is small the
baryonic abundance is larger than the fermionic one (because of the ±1 at the
denominator of the distribution function) of a factor 4/3; ii) for large x, the larger
λ is, the smaller becomes the difference between the fermionic and bosonic relic
abundances.
We now relate the relic abundance to the DM annihilation cross-section. For
x 1 we know that Yeq is vanishing, and the Boltzmann equation can be written
as:
dY λ
≈ − 2Y 2 , (3.271)
dx x
whose solution is:
1 1 λ
− ≈ . (3.272)
Y∞ Yf xf
91
��
��
��
��
��-�
��
Figure 3.9: Relative difference Y /Yeq − 1.
��
��
��
��
Figure 3.10: Comparison of the numerical solutions of Eq. (3.270) for the cases of
bosonic DM (solid lines) and fermionic DM (dashed lines) and λ = 10 (top two
lines) and λ = 100 (bottom two lines, almost superposed).
92
We have again considered here a constant λ, for simplicity. As we can see from
Fig. 3.8, Yf − Y∞ > 0 and the difference gets larger the larger λ is. Moreover,
xf ≈ 10, so that we can simplify
10
Y∞ ≈ . (3.273)
λ
When Y has attained Y∞ , the abundance of particle is fixed and their number
density starts to be diluted as nX ∝ a−3 . Therefore, we can can write down the
present-time energy density as follows:
a31 (kB T1 )3 a31 10m (kB T0 )3 a31 T13

ρX0 = n1 m 3 = mY∞ = . (3.274)
a0 ~3 a30 λ ~3 a30 T03
We have introduced here the photon temperature (the only one we can measure).
The ratio between parenthesis is not just equal to 1. We have seen an example
of this discrepancy when we calculated the photon-neutrinos temperatures ratio.
The reason is that not all along the cosmological evolution T decays as the inverse
scale factor. When there are processes such as electron-positron annihilation, more
photons are injected in the thermal bath and the temperature scales in a milder
way than 1/a.
It is now time to tackle more seriously the issue of the effective numbers of
relativistic degrees of freedom.
3.11.1 The effective numbers of relativistic degrees of free-

dom
Let T be the photon temperature, which we always use as reference since it is the
only one we can measure, from CMB.
In Eq. (3.223), we wrote the energy density of all the relativistic species in the
following way:
π 2 (kB T )4
ε= g∗ , (3.275)
30(~c)3
where, actually g∗ = g∗ (T ) because when kB T drops below the mc2 of a species,
this becomes non-relativistic and is removed from the above equation. Thus g∗
varies, but very rapidly close to the thresholds of the mass energies. Far from
those, g∗ it is practically constant.
In Fig. 3.11 we plot g∗ calculated from Eq. (3.103) and Eq. (3.223) for the species
of Table 3.2. The residual non-vanishing value of g∗ (T ) for low temperatures is due
to the massless species, i.e. photons and neutrinos.
Note that this plot is just an illustrative example, because it employs Eq. (3.103)
during the entire evolution, i.e. it assumes thermal equilibrium all the time, and,
moreover, the QCD phase transition at 200 MeV and the difference between photon
and neutrino temperature after electron-positron annihilation at 0.5 MeV have
been put “by hand”. The correct calculation should employ the energy density
obtained from the solutions of the Boltzmann equations of the various species,
which correctly track the evolutions when Γ ∼ H, i.e. out of equilibrium.
93
��
��
��
��
Figure 3.11: Evolution of g∗ calculated from Eq. (3.103) and Eq. (3.223) for the
species of Table 3.2. Note the QCD phase transition at 200 MeV and the difference
between photon and neutrino temperature after electron-positron annihilation at
0.5 MeV.
The effective numbers of relativistic degrees of freedom gets two contributions.

One from the relativistic particles that are in thermal equilibrium with the photons:
X 7 X
g∗therm ≡ gi + gi , (3.276)
i=bosons
8 i=fermions
and another from the relativistic particles that are no more in thermal equilibrium
with the photons:
X Ti4 7 X Ti4
g∗dec ≡ gi + gi . (3.277)
i=bosons
T 4 8 i=fermions T 4
The latter are basically neutrinos only.

Why do we insist on relativistic species? Because, as we showed, these are the
only one which contribute to the entropy density s:
2π 2 kB4 T 3
s= g∗S (3.278)
45(~c)3
where g∗S is the effective number of degrees of freedom for the entropy. Now, also
to g∗S contribute species which are in thermal equilibrium with the photons and
for which:
therm
X 7 X
g∗S = g∗therm = gi + gi , (3.279)
i=bosons
8 i=fermions
94
and species that are no more in thermal equilibrium with the photons and for
which:
X T3 7 X T3
dec
g∗S = gi i3 + gi i3 6= g∗dec . (3.280)
i=bosons
T 8 i=fermions T
Because of the last inequality, due to the different scaling of s and ε with the
temperature, we expect g∗ 6= g∗S .
Now, sa3 is a very useful quantity since it is conserved, as we showed earlier.
When a species becomes non-relativistic, its contribution to s is exponentially
suppressed as exp(−mc2 /kB T ). Therefore, the non-relativistic species passes its
entropy to rest of the thermal bath such that sa3 does not change.
What is the value of g∗ and of g∗S ? We need to recover our knowledge of particle
physics in Table 3.2.
Particle mass spin gs

1
Quarks t, t̄ 173 GeV 2
2 · 2 · 3 = 12
b, b̄ 4 GeV
c, c̄ 1 GeV
s, s̄ 100 MeV
d, d¯ 5 MeV
u, ū 2 MeV
Gluons gi 0 GeV 1 8 · 2 = 16
Leptons τ± 1777 MeV 1
2
2·2=4
µ± 106 MeV
e± 511 keV
1
ντ , ν̄τ < 0.6 eV 2
2·1=2
νµ , ν¯µ < 0.6 eV
νe , ν̄e < 0.6 eV
Gauge Bosons W+ 80 GeV 1 3
W− 80 GeV
Z0 91 GeV
γ 0 2
Higgs Bosons H0 125 GeV 0 1
Table 3.2: The standard model particles, with their mass, spin and degeneracies
gs .
Summing up all the contributions we have from bosons and fermions:
gbosons = 28 , gfermions = 90 , (3.281)
so that
7
g∗ = 28 + · 90 = 106.75 . (3.282)
8
When the temperature drops below one species mass, this becomes non-relativistic
and then its gs does not contribute anymore to the above sum. Note that before
neutrino decoupling g∗ = g∗S .
95
3.11.2 Relic abundance of DM and the WIMP miracle
On the basis of the discussion of the previous subsection, we can therefore write
Eq. (3.274) as follows:
ρX0 10m (kB T0 )3 g∗S (T0 )
ΩX0 = = , (3.283)
ρcr,0 λρcr,0 ~3 g∗S (m)
where g∗S (m) is the number of effective degrees of freedom in entropy for a thermal
energy equal to the DM particle mass. This must be certainly larger than 1 MeV,
roughly when neutrino decoupling takes place, therefore g∗ = g∗S . Moreover,
3
7 Tν 7 4
g∗S (T0 ) = 2 + · 6 · =2+ ·6· = 3.91 , (3.284)
8 T0 8 11
and so we can rewrite Eq. (3.283) as follows:
8πGH(m)xf (kB T0 )3 g∗S (T0 )
ΩX0 = , (3.285)
3H02 m2 c6 hσvi g∗ (m)
where we have also made λ explicit. The Hubble parameter H(m) can be written
as:
8πG π 2 (mc2 )4
H 2 (m) = g∗ (m) , (3.286)
3c2 30 (~c)3
so that we have finally:
2 10−37 cm2
xf xf 2.57 × 10−10 GeV−2
ΩX0 h ≈ 0.331 p ≈ 0.331 p . (3.287)
g∗ (m) hσv/ci g∗ (m) hσv/ci
As we saw, reasonable values are xf ≈ 10 and g∗ ≈ 100. A cross-section of the order

of G2F ∼ 10−10 GeV−2 gives the right order of magnitude of the present abundance
of CDM. This coincidence is known as WIMP miracle.7
Relic abundance of baryons

The very same result of Eq. (3.287) can be used for the annihilation of baryons:
b + b̄ ↔ γ + γ . (3.288)
Let us consider only nucleons, i.e. protons and neutrons, since their mass is the
dominant one for the baryonic energy density. The annihilation cross-section is of
the order of:
(~c)2
hσv/ci ≈ , (3.289)
(mπ c2 )2
where mπ c2 ≈ 140 MeV is the mass of the meson π, which can be thought as the
mediator of the strong interaction among nucleons. Substituting into Eq. (3.287)
we get:
Ωb0 h2 ≈ 10−11 , (3.290)
7
One can think of different cross sections and thus find CDM candidates lighter than WIMPs.
See e.g. [Profumo, 2017]. This possibility is also called WIMPless miracle.
96
i.e. a value many orders of magnitude below the observed one and that thus
constitutes a compelling argument for the necessity of baryogenesis.
Focusing on electrons and positrons, the annihilation cross section is of the
order
α2 (~c)2
hσv/ci ≈ ≈ 204 GeV−2 , (3.291)
(me c2 )2
and from Eq. (3.287) we get:
Ωe0 h2 ≈ 10−12 . (3.292)
97
Chapter 4
Cosmological perturbations
Без труда не вынешь рыбку из пруда

(Without effort, you can’t pull a fish out of the pond)
—Russian proverb
As we have seen in the previous Chapters, the assumption of homogeneous

and isotropic universe is very useful and productive, but it is reliable only on very
large scales (above 200 Mpc). Its shortcomings become evident when we start to
investigate how structures, such as galaxies and their clusters, form, since these
are huge deviations from the cosmological principle.
In this Chapter we address small deviations from the cosmological principle,
considering perturbations in the FLRW metric. This is the starting point of the
incredibly difficult task of understanding how structures form in an expanding
universe, which ultimately needs powerful machines and numerical simulations.
The material for this Chapter is mainly drawn from the textbooks [Dodelson,
2003], [Mukhanov, 2005], [Weinberg, 2008] and from the papers/reviews [Bardeen,
1980], [Kodama and Sasaki, 1984], [Mukhanov et al., 1992], [Ma and Bertschinger,
1995].
We assume hereafter K = 0, both for simplicity and because we have seen
that there is strong observational evidence of a spatially flat universe. From this
Chapter on, we also start to adopt natural ~ = c = 1 units.
4.1 From the perturbations of the FLRW metric to

the linearised Einstein tensor
Let ḡµν be the FLRW metric, and write it using the conformal time:
ḡµν = a2 (η)(−dη 2 + δij dxi dxj ) . (4.1)
This metric describes the background spacetime, or manifold. However, the
background spacetime is fictitious, in the sense that we are now considering devia-
tions from homogeneity and isotropy and therefore the actual physical spacetime
is a different manifold described by the metric gµν . Defining the difference:
δgµν (x) = gµν (x) − ḡµν (x) , (4.2)
98
at a certain spacetime coordinate x is an ill-posed statement because gµν and ḡµν
are tensors defined on different manifolds and x is a coordinate defined through
different charts. Even if we embed the two manifolds in a single one, still the differ-
ence between two tensors evaluated at different points is an ill-defined operation.
Therefore, in order to make Eq. (4.2) meaningful, we need an extra ingredient: a
map which identifies points of the background manifold with those of the physi-
cal manifold. This map is called gauge, is arbitrary and allows us to use a fixed
coordinate system (a chart) in the background manifold also for the points in the
physical manifold. In other words, we shall still use conformal or cosmic time plus
comoving spatial coordinates even when describing perturbative quantities. This
property leads to the so-called problem of the gauge, which we will briefly dis-
cuss later. For more details on these topics, see [Stewart, 1990] and [Malik and
Matravers, 2013].
Metric gµν has in general 10 independent components that, in a generic gauge,
we write down in the following form:
 
 −[1 + 2ψ(η, x)] wi (η, x) 
2
gµν = a (η) , δ ij χij = 0 ,
wi (η, x) δij [1 + 2φ(η, x)] + χij (η, x)
 
(4.3)
where ψ, φ, wi , χij (i, j = 1, 2, 3) are functions of the background spacetime
coordinates xµ , in our case conformal time and comoving spatial coordinates.1
From now on we omit their explicit functional dependence wherever possible, in
order to keep a lighter notation.
As we mentioned above, the liberty of choosing a gauge allows us to fix a
coordinate system in the background manifold. The latter shall be one in which
homogeneity and isotropy are manifest, of course. Therefore, we could use any
time parametrisation, though we will employ conformal time the most because it
has the important physical meaning of the comoving particle horizon, and as for
the spatial coordinates we shall always choose the comoving ones for which at any
given and fixed time the background spatial metric is Euclidean, i.e. δij dxi dxj .
For this reason, the perturbations in Eq. (4.3) can be regarded as usual 3-vectors,
defined from the rotation group SO(3).
Let us elaborate more on this point. Consider the coordinate transformation:
 
µ 1 0
∂x
=  , (4.4)
∂xν 0 i
0 Rj
where Ri j is a rotation. By definition, a rotation is characterised by RT R = I,
i.e. the transposed matrix is also the inverse, hence δkl Rk i Rl j = δij . Applying this
transformation to metric (4.3) we get:
0 0
g00 = g00 , g0i = Rk i g0k , gij0 = Rk i Rl j gkl , (4.5)
and hence, recalling that Rk i Rl j δkl = δij , we finally find:
ψ0 = ψ , wi0 = Rk i wk , φ0 = φ , χ0ij = Rk i Rl j χkl . (4.6)
1
The reason why ψ and φ are multiplied by 2 in the perturbed metric is just for pure future
convenience of calculation.
99
Therefore, ψ and φ are two 3-scalars, wi (i = 1, 2, 3) is a 3-vector, χij is a 3-tensor
(i, j = 1, 2, 3) and the indices of wi and χij are raised and lowered by δij . Note that
the 6 components of χij are not independent because δ ij χij = 0. In other words,
χij is traceless and we have already put in evidence the spatial trace of the metric
through φ. One does this also because the spatial intrinsic curvature depends only
on φ, not on χij .2
Now, if the gauge chosen in Eq. (4.3) is such that:
|ḡµν | |δgµν | , (4.7)
then we are dealing with perturbations. They are considered small, or linear, or
at first-order if we neglect powers with exponent larger than one in the quantities
themselves and in their derivatives. Not only, also combinations among different
perturbations are neglected. For example, ψ 2 , φwi , wi χij , φ0 φ and so on are all sec-
ond order perturbations, and therefore negligible. Let us see how this works when
computing the Christoffel symbols for metric (4.3). Substituting the decomposition
(4.2) into the definition of Christoffel symbol we have:
1 1
Γµνρ = g µσ (gσν,ρ + gσρ,ν − gνρ,σ ) = ḡ µσ (ḡσν,ρ + ḡσρ,ν − ḡνρ,σ )
2 2
1 µσ 1 µσ
ḡ (δgσν,ρ + δgσρ,ν − δgνρ,σ ) + δg (ḡσν,ρ + ḡσρ,ν − ḡνρ,σ ) , (4.8)
2 2
where the comma denotes the usual partial derivative. Note that we have assumed
the same decomposition of Eq. (4.2) also for the covariant components of the metric
and neglected terms such as δg µσ δgσν,ρ , i.e. perturbative quantities multiplied by
their derivatives. It is important to realise that δg µσ is not simply δgµσ with indices
raised by ḡ µσ .
Exercise 4.1. Since

g µρ gρν = δ µ ν , ḡ µρ ḡρν = δ µ ν , (4.9)
because both are metrics, using Eq. (4.2) show that
δg µν = −ḡ µρ δgρσ ḡ νσ (4.10)
In particular, in our scenario of cosmological perturbations, we shall write the total

metric, using the conformal time, as follows:
gµν = a2 (ηµν + hµν ) . (4.11)
Hence, the perturbed contravariant metric is the following:
1 1 il 1 1 il 1
δg 00 = − h00 δg 0i = δ h0l = 2 h0i , δg ij = −
δ hlm δ mj = − 2 hij ,
a2 a 2 a a 2 a
(4.12)
ij
where we have used our hypothesis that the indices of hij are raised by δ and the
property hij = hij .
2
One can guess this by simply noting that the spatial curvature is a 3-scalar and one cannot
form any 3-scalar at first-order from χij since it is traceless.
100
Do not be confused by the fact that δgµν is the perturbed covariant metric but
µν
δg is not the contravariant perturbed metric.
One does raise the indices of the contravariant perturbed metric with the back-
ground one, but a minus sign must be taken into account. This fact is not dissimilar
from considering the Taylor expansion
1
= 1 − x + O(x2 ) . (4.13)
1+x
4.1.1 The perturbed Christoffel symbols

It is clear from Eq. (4.8) that we can decompose the affine connection as follows:
Γµνρ = Γ̄µνρ + δΓµνρ , (4.14)
where the barred one is computed from the background metric only.
Exercise 4.2. Show that
1
δΓµνρ = ḡ µσ δgσν,ρ + δgσρ,ν − δgνρ,σ − 2δgσα Γ̄ανρ

(4.15)
2
The background Christoffel symbols were calculated already in Chapter 2 but

for the FLRW metric written in the cosmic time.
Exercise 4.3. Show that the only non-vanishing background Christoffel symbols
are:
a0 a0 a0
Γ̄000 = , Γ̄0ij = δij , Γ̄i0j = δ i j , (4.16)
a a a
where the prime denotes derivation with respect to the conformal time. One can
make the calculation directly from the FLRW metric written in the conformal time
or use the results already found in Eq. (2.29) for the FLRW metric written in the
cosmic time and use the transformation relation for the Christoffel symbol:
∂xβ ∂xγ ∂x0µ ∂x0µ ∂ 2 xσ

Γ̄0µ α
νρ = Γ̄βγ + , (4.17)
∂x0ν ∂x0ρ ∂xα ∂xσ ∂x0ν ∂x0ρ
where the primed coordinates are in the conformal time and hence:
∂x00 1 ∂x0l
= , = δl m , (4.18)
∂x0 a ∂x m
being the other cases vanishing. It is a good and reassuring exercise to do in both
ways and check that the result is the same.
101
A moment of reflection. First of all, we shall use mostly the conformal time
throughout these notes since, as we saw earlier, it represents the comoving particle
horizon and it will allow us to clearly distinguish the evolution of super-horizon
(hence causally disconnected) scales from sub-horizon ones.
On the other hand, the most economic way to compute the linearised Einstein
equations is using the cosmic time and stopping to the calculation of the Ricci
tensor by considering the Einstein equations in the form

1
Rµν = 8πG Tµν − gµν T , (4.19)
2
as done e.g. in [Weinberg, 2008]. It is the most economic way because in the
cosmic time we have only 2 non-vanishing Christoffel symbols, whereas in the
conformal time we have three, and because we are spared to compute the perturbed
Ricci scalar. Once done this calculation, it is straightforward to come back to the
conformal time.
In the following we shall anyway go on using the conformal time, because
it is a good workout. Compare the results found here with those in [Weinberg,
2008, Chapter 5] using the tensorial properties:
∂ x̃ρ ∂ x̃σ ∂ x̃ρ ∂ x̃σ
Rµν = R̃ρσ , hµν = h̃ρσ , (4.20)
∂xµ ∂xν ∂xµ ∂xν
where the quantities with tilde are in the cosmic time. Since the change from
cosmic to conformal time does not affect the spatial coordinates, we have that:
R00 = R̃00 a2 , R0i = R̃0i a , Rij = R̃ij , (4.21)
and similarly for hµν . With this map, we can check the results that we are going to
find here with those in [Weinberg, 2008]. Mind that in [Weinberg, 2008] the Ricci
tensor is defined with the opposite sign with respect to ours here.
Exercise 4.4. We are now in the position of writing the perturbed Christoffel
symbols. We do so without leaving hµν explicit from Eq. (4.3). Find that:
1 1
δΓ000 = − h000 , δΓ0i0 = − (h00,i − 2Hh0i ) , (4.22)
2 2
1
δΓi00 = h0i0 + Hhi0 − h00,i , (4.23)
2
1
δΓ0ij = − + h0j,i − h0ij − 2Hhij − 2Hδij h00 ,

h0i,j (4.24)
2
1 1
δΓij0 = h0ij + (hi0,j − h0j,i ) , (4.25)
2 2
1
δΓijk = (hij,k + hik,j − hjk,i − 2Hδjk hi0 ) . (4.26)
2
The prime denotes derivative with respect to the conformal time. The indices
might seem unbalanced, but we have used the fact that hi0 and hij are 3-tensors
with respect to the metric δij and hence, for example, hi 0 = hi0 . Recall that
a0
H≡ , (4.27)
a
102
i.e. H is the Hubble factor written in conformal time.
4.1.2 The perturbed Ricci tensor and Einstein tensor

With this result we are going to compute the components of the perturbed Ricci
tensor. Recall that this is defined as:
Rµν = Γρµν,ρ − Γρµρ,ν + Γρµν Γσρσ − Γρµσ Γσνρ , (4.28)
and hence, when substituting Eq. (4.14), we get
Rµν = Γ̄ρµν,ρ − Γ̄ρµρ,ν + Γ̄ρµν Γ̄σρσ − Γ̄ρµσ Γ̄σνρ

+δΓρµν,ρ − δΓρµρ,ν + Γ̄ρµν δΓσρσ + δΓρµν Γ̄σρσ − Γ̄ρµσ δΓσνρ − δΓρµσ Γ̄σνρ , (4.29)
by neglecting second order terms in the connection. It is clear that we can expand
also the Ricci tensor as:
Rµν = R̄µν + δRµν , (4.30)
and we compute now its perturbed components.
Exercise 4.5. After opening up all the sums, show that:
δR00 = δΓl00,l − δΓl0l,0 − HδΓl0l + 3HδΓ000 , (4.31)
and substituting the previous results, one gets:

1 3 1
δR00 = − ∇2 h00 − Hh000 + h0k0,k + Hhk0,k − (h00kk + Hh0kk ) . (4.32)
2 2 2
Here we have defined δ ij ∂i ∂j ≡ ∇2 as the Laplacian in comoving coordinates. Note
that hkk = δ lm hlm , i.e. in the present instance repeated indices are summed even
if both covariant or contravariant.
Exercise 4.6. Of course, repeat the previous exercise also for the other compo-
nents. Find that:
00
1 a 1 0
2 2
hkk,i − h0ki,k ,(4.33)

δR0i = −Hh00,i − ∇ h0i − hk0,ik + + H h0i −
2 a 2
and
a00

1 H 0 2
δRij = h00,ij + h00 δij + H + h00 δij
2 2 a
a00

1 2
1 00 0 2
− ∇ hij − hki,kj − hkj,ki + hkk,ij + hij + Hhij + H + hij
2 2 a
H 1
+ h0kk δij − Hhk0,k δij − (h00i,j + h00j,i ) − H(h0i,j + h0j,i ) . (4.34)
2 2
103
In the same way we decomposed metric (4.2), we decompose the Einstein tensor.
We shall work with mixed indices:
1 1 1
Gµ ν = g µρ Rρν − δ µ ν R = ḡ µρ R̄ρν − δ µ ν R̄ + ḡ µρ δRρν + δg µρ R̄ρν − δ µ ν δR , (4.35)
2 2 2
where Ḡµ ν = R̄µ ν − 12 δ µ ν R̄ is the background Einstein tensor and depends purely
from the background metric ḡµν whereas
1
δGµ ν = ḡ µρ δRρν + δg µρ R̄ρν − δ µ ν δR , (4.36)
2
is the linearly perturbed Einstein tensor, which depends from both ḡµν and hµν .
Exercise 4.7. Compute the perturbed Ricci scalar:
δR = g µν Rµν = ḡ µν δRµν + δg µν R̄µν . (4.37)
Expand the above expression and use formula (4.10) in order to find:
1 1
δR = − 2
δR00 + 2 δ ij δRij − a2 hρσ ḡ ρµ ḡ σν R̄µν , (4.38)
a a
and then, recalling that the background Ricci tensor is:
a00 a00

2 2
R̄00 = 3 H − , R̄ij = δij H + , (4.39)
a a
one can write:
a00 a00

1 3 2 1 ij 1 2
δR = − 2 δR00 − 2 h00 H − + 2 δ δRij − 2 hkk H + , (4.40)
a a a a a a
Substituting the formulae for the perturbed Ricci tensor, one finds:
a00
a2 δR = ∇2 h00 + 3Hh000 + 6 h00 − 2h0k0,k − 6Hhk0,k
a
+h00kk + 3Hh0kk − ∇2 hkk + hkl,kl . (4.41)
The second line represents a2 δR(3) , i.e. the intrinsic spatial perturbed curvature
scalar.
Now, let us calculate the mixed components of the perturbed Einstein tensor.
2a2 δG0 0 = −6H2 h00 + 4Hhk0,k − 2Hh0kk + ∇2 hkk − hkl,kl (4.42)
and
2a2 δG0 i = 2Hh00,i + ∇2 h0i − hk0,ki + h0kk,i − h0ki,k (4.43)
104
and
a00

2a δG j = −4 h00 − 2Hh000 − ∇2 h00 + 2H2 h00 − 2Hh0kk
2 i
a
+∇ hkk − hkl,kl + 2h0k0,k + 4Hhk0,k − h00kk δ i j + h00,ij − ∇2 hij
2

+hki,kj + hkj,ki − hkk,ij + h00ij + 2Hh0ij − (h00i,j + h00j,i ) − 2H(h0i,j + h0j,i ) . (4.44)
Now we turn to the right hand side of the Einstein equations, i.e. the energy-
momentum tensor.
4.2 Perturbation of the energy-momentum tensor

In the following we shall use mostly the energy-momentum tensor defined through
the distribution function and its perturbation. However, let us see how to perturb
the background, perfect fluid energy-momentum tensor. This was introduced as:
T̄µν = (ρ̄ + P̄ )ūµ ūν + P̄ ḡµν , (4.45)
as the tensor describing a fluid with no dissipation, i.e. constant entropy along the
flow.
Exercise 4.9. Compute ūν ∇ν ūµ , demand its vanishing and check via the second
law of thermodynamics that this corresponds to a constant entropy.
Let us rewrite T̄µν as follows:
T̄µν = ρ̄ūµ ūν + P̄ θ̄µν , θ̄µν ≡ ḡµν + ūµ ūν . (4.46)
The tensor θ̄µν acts as a projector on the hypersurface orthogonal to the four-
velocity.
Exercise 4.10. Show that θ̄µν ūµ = 0.
This is an example of 3+1 decomposition, which is particularly useful when we

study a fluid flow.
ρ̄ = T̄µν ūµ ūν , 3P̄ = T̄µν θ̄µν , T ≡ ḡ µν T̄µν = −ρ + 3P , (4.47)
i.e. the fluid density is the projection of the energy-momentum tensor along the
4-velocity of the fluid element and the pressure is the projection of the energy-
momentum tensor on the 3-hypersurface orthogonal to the four-velocity.
105
The most general energy-momentum tensor, which also includes the possibility
of dissipation, can be then written by straightforwardly generalising Eq. (4.46), i.e.
Tµν = ρuµ uν + qµ uν + qν uµ + (P + π)θµν + πµν (4.48)
where qµ is the heat transfer contribution, satisfying qµ uµ = 0 and thus contribut-
ing with 3 independent components; πµν is the anisotropic stress, it is traceless
and satisfies πµν uµ = 0, hence providing 5 independent components. The trace of
πµν is π, it is called bulk viscosity and has been put in evidence together with the
pressure. For more detail about dissipative processes in cosmology and the above
decomposition of the energy-momentum tensor, see the review [Maartens, 1996].
The anisotropic stress πµν is not necessarily related to viscosity, but can exist for
relativistic species such as photons and neutrinos because of the quadrupole mo-
ments of their distributions, as we shall see later. On the other hand, heat fluxes
and bulk viscosity are related to dissipative processes, and we neglect them in these
notes starting from the next section.
Now we consider the energy-momentum tensor of Eq. (4.48) as made up of a
background contribution plus a linear perturbation, i.e. we expand
ρ = ρ̄ + δρ(η, x) , P = P̄ + δP (η, x) , uµ = ūµ + δuµ (η, x) , (4.49)
which are the physical density, pressure and four-velocity, which, remember, de-
pend on the background quantities because of our choice of a gauge. The barred
quantities depend only on η, since they are defined on the FLRW background.
Heat fluxes, bulk viscosity and anisotropic stresses are purely perturbed quanti-
ties.3 Therefore, we have:
Tµν = T̄µν + δρūµ ūν + ρ̄δuµ ūν + ρ̄ūµ δuν + qµ ūν + qν ūµ + θ̄µν (δP + π) + P̄ δθµν + πµν ,
(4.50)
where one can straightforwardly identify the perturbed energy-momentum tensor
and where
δθµν = hµν + δuµ ūν + ūµ δuν . (4.51)
Now, recall that the background four-velocity satisfies the normalisation:
ḡµν ūµ ūν = −1 . (4.52)
Let us choose a frame in which to make explicit the components of the energy-
momentum tensor. Of course, we use comoving coordinates, for which one has
ūi = 0. From Eq. (4.52) we have then
a2 (ū0 )2 = 1 , (4.53)
and we choose the positive solution ū0 = 1/a, which implies u0 = −a.4 When we
choose these coordinates, the relations qµ uµ = πµν uµ = 0 imply that q0 = πµ0 = 0,
i.e. the heat flux and the anisotropic stress have only spatial components.
3
Bulk viscosity is compatible with the cosmological principle and can be contemplated also at
background level. It plays a central role in the so-called bulk viscous cosmology, see [Zimdahl,
1996].
4
It amounts to choose that the conformal time and the fluid element proper time flow in the
same direction.
106
Exercise 4.12. Calculate the components of the energy-momentum tensor (4.50).
Show that:
T00 = ρ̄(1 + δ)a2 − 2a(ρ̄ + P̄ )δu0 + P̄ h00 , (4.54)

T0i = −a(ρ̄ + P̄ )δui − aqi + P̄ h0i , (4.55)
Tij = (P̄ + δP + π)a2 δij + P̄ hij + πij , (4.56)
where we have introduced one of the main characters of these notes, the density
contrast:
δρ
δ≡ (4.57)
ρ̄
The density contrast is very important because describes how structure formation
begins. Note how the presence of perturbations in the four-velocity gives rise to
mixed time-space components in the energy-momentum tensor, i.e. the breaking
of homogeneity and isotropy allows for extra fluxes beyond the Hubble one.
Note that the total four-velocity also satisfies a normalisation relation, with
respect to the total metric, i.e.
gµν uµ uν = −1 . (4.58)
If we expand this relation up to the first order we find:
ḡµν ūµ ūν + hµν ūµ ūν + 2ḡµν δuµ ūν = −1 . (4.59)
Using Eq. (4.52) and ūi = 0, we find that:
h00 + 2ḡ00 aδu0 = 0 , (4.60)
and thus we can relate the metric perturbation h00 to δu0 as follows:
h00
δu0 = (4.61)
2a3
Care is needed when we want to compute the covariant components of the per-
turbed four-velocity δuµ . These are not simply δuµ = ḡµν δuν . It is the same care
we had to apply when considering the relation between δg µν and hµν . So, let us
define:
δui = avi (4.62)
with vi components of a 3-vector, i.e. its index is raised by δ ij so that v i = vi . Let
us compute now the components δui . We must start from the covariant expression
for the total four velocity, i.e.
ūµ + δuµ = uµ = gµν uν = gµν (ūν + δuν ) , (4.63)
and expanding up to first order, we get
ūµ + δuµ = ḡµν ūν + ḡµν δuν + hµν ūν . (4.64)
107
Equating order by order we obtain ūµ = ḡµν ūν , as expected, and
δuµ = ḡµν δuν + hµν ūν (4.65)
so that
δuµ = ḡ µν δuν − ḡ µρ hρν ūν (4.66)
So, there is an extra term hµν ūν . This comes from the fact that ḡµν raises or lowers
indices for the background quantities only whereas gµν raises or lowers indexes for
the full quantities only, and δuµ is neither. From Eq. (4.65) and (4.66) we have
that
h00 1
δu0 = , aδui = vi − 2 hi0 (4.67)
2a a
Exercise 4.13. Rewrite the components of the energy-momentum tensor as:
T00 = ρ̄(1 + δ)a2 − ρ̄h00 , (4.68)

T0i = −a2 (ρ̄ + P̄ )vi − aqi + P̄ h0i , (4.69)
Tij = (P̄ + δP + π)a2 δij + P̄ hij + πij , (4.70)
In order to calculate the mixed components we use the standard relation:
T µ ν = g µρ Tρν = ḡ µρ T̄ρν + ḡ µρ δTρν + δg µρ T̄ρν . (4.71)
Exercise 4.14. Compute the mixed components of the energy-momentum tensor.

Show that:
T 0 0 = −ρ̄(1 + δ) , (4.72)
T 0 i = ρ̄ + P̄ vi + a−1 qi ,

(4.73)
T i0 = − ρ̄ + P̄ (vi − h0i a−2 ) − a−1 qi ,

(4.74)
T i j = δ i j (P̄ + δP + π) + π i j , (4.75)
where we have stipulated that δ il πlj a−2 = π i j .
In the following we shall mainly use, especially for photons and neutrinos, the
energy-momentum tensor computed from kinetic theory, i.e. Eq. (3.41):
d3 p pµ pν
Z
µ
T ν= f, (4.76)
(2π)3 p0
in which the distribution function is also perturbed:
f = f¯ + F (4.77)
108
thus allowing to define a perturbed energy-momentum tensor as follows:
d3 p p µ pν
Z
µ
δT ν = F, (4.78)
(2π)3 p0
The momentum used here is the proper one which, recall, has the metric embedded
in its definition so that:
pi = p i , p2 = δij pi pj , E = p0 , p0 = −E . (4.79)
These relations provide directly the mass-shell one E 2 = p2 + m2 from gµν P µ P ν =
−m2 . Note that we are assuming g0i = 0, also at perturbative level thanks to
gauge freedom, otherwise the relations above would be incorrect since the spatial
metric would not be gij but gij − g0i g0j /g00 . See [Landau and Lifschits, 1975] for a
nice explanation of this fact.
Then, from the above definition of perturbed energy-momentum, we have:
d3 p
Z
0
δT 0 = − E(p)F = −δρ , (4.80)
(2π)3
which makes sense, and is consistent with Eq. (4.72), because we are weighting the
particle energy with the perturbed distribution function. The mixed components
are:
d3 p d3 p i
Z Z
0 i
δT i = p i F = (ρ̄ + P̄ )vi , δT 0 = − p F = −(ρ̄ + P̄ )vi , (4.81)
(2π)3 (2π)3
compatible with Eqs. (4.73) and (4.74) since we are assuming h0i = 0 (and ne-
glecting already the heat fluxes). Note the multiplication by ρ̄ + P̄ . Physically the
integral gives the perturbed spatial momentum density, which can be decomposed
in the velocity flow times the inertial mass density, which is ρ̄ + P̄ in GR.
Finally:
d3 p pi pj
Z
i
δT j = 3
F = δ i j δP + π i j , (4.82)
(2π) E
from which we can define the perturbed pressure as the trace part:
d3 p p2
Z
1 ij
δP = δ δTij = F, (4.83)
3 (2π)3 3E
and the anisotropic stress as the traceless part:
d3 p 1
Z
i i 1 i lm i 1 i 2
π j = δT j − δ j δ δTlm = p pj − δ j p F . (4.84)
3 (2π)3 E 3
The evolution equations for the perturbations are given by the linearised Ein-
stein equations:
δGµ ν = 8πGδT µ ν . (4.85)
Unfortunately, Eq. (4.85) is not sufficient to completely describe the behaviour of
both matter and metric quantities if the fluid components are more than one. See
e.g. [Gorini et al., 2008]. We shall make use of the perturbed Boltzmann equations
for each component of our cosmological model and we shall derive them in Chapter
5.
109
4.3 The problem of the gauge and gauge transfor-
mations
As we have mentioned in the previous section, a gauge is a map between the points
of the physical manifold and those of the background one which allows us to define
the difference between tensors defined on the two manifolds. Suppose we change
gauge from a G to a Ĝ. Metric (4.3) then becomes:
 
 −[1 + 2ψ̂(η, x)] ŵi (η, x) 
2
gµν = a (η) , δ ij χ̂ij = 0 .
 
ŵi (η, x) δij [1 + 2φ̂(η, x)] + χ̂ij (η, x)
(4.86)
We have new hatted functions representing perturbations depending again on the
background coordinates, which we have fixed.
A similar map can be defined also on the single background manifold, so in
absence of perturbations, and it can be deceiving, in the sense that quantities
similar to perturbations might appear even if we are in the background manifold.
Let us rephrase this. The problem when considering fluctuations in GR is that we
cannot be sure, by only looking at a metric, that there are real fluctuations about
a known background or it is the metric which is written in a not very appropriate
coordinate system. For example, consider the following time transformation for
the FLRW metric:
dη = g(t, x)dt . (4.87)
Then, the FLRW metric becomes:
ds2 = −a(t, x)2 g(t, x)2 dt2 + a(t, x)2 δij dxi dxj , (4.88)
and recalling back t → η we get:
ds2 = −a(η, x)2 g(η, x)2 dη 2 + a(η, x)2 δij dxi dxj . (4.89)
Now the metric coefficients depend on the position so we may think that homo-
geneity and isotropy are lost, but it was just a coordinate transformation. This
is not the problem of the gauge, but is the general covariance typical of GR. As
we show above, it is not even necessarily related to perturbations but it is just a
coordinate transformation masking the metric in the form that we are used to see
it. Computing relativistic invariants or Killing vectors allows to determine if the
above is the FLRW metric or not.
The problem of the gauge is the dependence of the perturbations on the
gauge. In the following we will see how a gauge transformation manifests itself
through a change of coordinates and deal with the problem of the gauge by intro-
ducing gauge-invariant variables. Loosely speaking, we will see that the gauge
is the functional dependence of the perturbative quantities on the coordinates,
whereas the background quantities maintain their functional form.
For a more complete treatment of the gauge problem, see for example [Stewart,
1990], [Mukhanov et al., 1992] and [Malik and Matravers, 2013].
110
4.3.1 Coordinates and gauge transformations
In order to understand how a gauge transformation changes the functional depen-
dence of the perturbative quantities we express the change from a gauge G to a
gauge Ĝ as the following infinitesimal coordinate transformation:
xµ → x̂µ = xµ + ξ µ (x) , (4.90)
where xµ are the background coordinates and ξ µ is a generic vector field, the
gauge generator, which must be |ξ µ | 1 in order to preserve the smallness of
the perturbation. From a geometric point of view, we fix a point on the background
manifold and by changing gauge we change the corresponding point on the physical
manifold. This point has coordinates different from the first one and given by
Eq. (4.90). The gauge generator can be seen then as a vector field on the physical
manifold and the Lie derivative of the metric along ξ tells us how the perturbative
quantities change their functional form.
Under a coordinate transformation, the metric tensor gµν , as well as any other
tensor of the same rank, transforms in the following way:
∂ x̂ρ ∂ x̂σ
gµν (x) = ĝρσ (x̂) , (4.91)
∂xµ ∂xµ
which, using Eq. (4.90), can be cast as follows:
gµν (x) = (δ ρ µ + ∂µ ξ ρ ) (δ σ ν + ∂ν ξ σ ) ĝρσ (x̂) . (4.92)
Writing down Eq. (4.92) up to first order, one obtains:
gµν (x) = ĝµν (x) + ∂α ĝµν (x)ξ α + ∂µ ξ ρ ĝρν (x) + ∂ν ξ ρ ĝρµ (x) , (4.93)
where we have also expanded ĝρσ (x̂) about x using Eq. (4.90).
Exercise 4.15. Show that Eq. (4.93) can be cast in the following form:
gµν (x) = ĝµν (x) + ∂α gµν (x)ξ α + ∂µ ξ ρ gρν (x) + ∂ν ξ ρ gρµ (x) , (4.94)
i.e. prove that we can remove the hat from the metric when it is multiplied by ξ.
Then, cast the above equation as follows:
gµν (x) = ĝµν (x) + ∇ν ξµ + ∇µ ξν . (4.95)
This equation shows how the functional form of the metric components, i.e. the
gauge, changes upon a coordinate transformation.
If Eq. (4.90) is an isometry, i.e. gµν (x) = ĝµν (x), one gets the Killing equation:
∇ν ξµ + ∇µ ξν = 0 . (4.96)
Now use the perturbed FLRW metric (4.3) in Eq. (4.93).
111
Exercise 4.16. Show that at the order zero:
â(η) = a(η) (4.97)
i.e. the scale factor maintains its functional form, confirming the property of the
gauge, which maintains the functional form of the background quantities. Then,
show that at first-order the following transformation relations hold:
0
ψ̂ = ψ − Hξ 0 − ξ 0 ŵi = wi − ζi0 + ∂i ξ 0 (4.98)
1 2
φ̂ = φ − Hξ 0 − ∂l ξ l χ̂ij = χij − ∂j ζi − ∂i ζj + δij ∂l ξ l (4.99)
3 3
where ζi ≡ δil ξ l . We have introduced ζi in order not to make confusion with the
spatial part of ξµ , which is ξi = a2 δil ξ l = a2 ζi .
In the very same fashion we adopted for the metric, we can also find the trans-
formation rules for the components of the energy-momentum tensor. That is,
through the same steps that we have just used for the metric, we can write:
T̂µν (x) = Tµν (x) − ∂α Tµν (x)ξ α − ∂µ ξ ρ Tρν (x) − ∂ν ξ ρ Tρµ (x) . (4.100)
Exercise 4.17. Use the above transformation with Tµν given by Eq. (4.68)-(4.70).
Find that at zeroth order one has:
ρ̄ˆ(η) = ρ̄(η) P̄ˆ (η) = P̄ (η) (4.101)

i.e. the background density and pressure maintain their functional forms. Then,
show that at first order the following transformation relations hold:
ˆ = δρ − ρ̄0 ξ 0
δρ v̂i = vi + ∂i ξ 0 q̂i = qi (4.102)
π̂ = π ˆ = δP − P̄ 0 ξ 0
δP π̂ij = πij (4.103)
Hint. In order to find these relations you have to use those for the metric quantities.
Moreover, one obtains q̂i = qi noticing that ρ̄ + P̄ is arbitrary. One the other hand,
we do not have a mathematical way to separate the transformations for δP and π.
We do that by giving the physical argument by which π is related to dissipative
processes whereas δP is not.
In general, perturbations of quantities which are vanishing or constant in the

background are automatically gauge-invariant and one can see this explicitly above
for the heat flux, the bulk viscosity and the anisotropic stress. This property
is formalised in the Stewart-Walker lemma [Stewart and Walker, 1974]. It
stimulates for example the use of the perturbed Weyl tensor (which is vanishing
in FLRW metric) and the quasi-Maxwellian equations [Hawking, 1966], [Jordan
et al., 2009].
112
4.3.2 The Scalar-Vector-Tensor decomposition
The scalar-vector-tensor (SVT) decomposition was introduced by Lifshitz in 1946
[Lifshitz, 1946], who was the first to address cosmological perturbations. See also
[Lifshitz and Khalatnikov, 1963] and [Ma and Bertschinger, 1995], for a particularly
detailed account. It consists of the following procedure. We have already seen that
the perturbed metric can be written in terms of two scalars ψ and φ, a 3-vector wi
and a 3-tensor χij . However, we can “squeeze out” two more scalars from wi and
χij and one more vector from χij .
Helmholtz theorem (see Sec. 12.3 for a brief reminder) states that, under certain
conditions of regularity, any spatial vector wi can be uniquely decomposed in its
longitudinal part plus its orthogonal contribution:
k
wi = wi + wi⊥ , (4.104)
which are respectively irrotational and solenoidal (divergenceless), namely:
k
ijk ∂j wk = 0 , ∂ k wk⊥ = 0 , (4.105)
where ijk is the Levi-Civita symbol.5 By Stokes theorem, the irrotational part can
be written as the gradient of a scalar say w so that, finally, we can write wi as
follows:
wi = ∂i w + Si (4.106)
where we have defined Si ≡ wi⊥ , because it is simpler to write. Therefore, w is the
scalar part of wi and Si is the vector part of wi . Usually, when in cosmology one
talks about a vector perturbation one is referring to Si , i.e. to a vector which
cannot be written as a gradient of a scalar.
Similarly to the vector case, any spatial rank-2 tensor say χij can be decom-
k
posed in its longitudinal part χij plus its orthogonal part χ⊥ij plus the transverse
T
contribution χij :
k
χij = χij + χ⊥ T
ij + χij , (4.107)
defined as follows:
k
ijk ∂ l ∂j χlk = 0 , ∂ i ∂ j χ⊥
ij = 0 , ∂ j χTij = 0 . (4.108)
Basically, one builds a vector by taking the divergence of χij and then applies
Helmholtz theorem to it. This implies that the longitudinal and the orthogonal
parts can be further decomposed in the same spirit of Eq. (4.105) in the following
way:

k 1 2
χij = ∂i ∂j − δij ∇ 2µ , χ⊥
ij = ∂j Ai + ∂i Aj , ∂ i Ai = 0 , (4.109)
3
where µ is a scalar, Ai is a divergenceless vector and recall that ∇2 ≡ δ lm ∂l ∂m . We
can thus write χij in the following form:

1
χij = ∂i ∂j − δij ∇ 2µ + ∂j Ai + ∂i Aj + χTij
2
(4.110)
3
5
Reminder: 123 = 1 and it changes sign upon any odd permutation of its indices. It follows
that ijk = 0 if two or more indices are equal.
113
The transverse part χTij cannot be decomposed in any scalar or divergenceless
vector. Therefore, it constitutes a tensor perturbation.
The SVT decomposition is a fundamental tool for the investigation of first order
perturbations because the three classes do not mix up and therefore they can be
independently analysed. The absence of mixing is due to the fact that any kind of
interaction term among the three categories would be of second order and therefore
negligible.
Let us see how each class of perturbations transforms. Apply Helmholtz theo-
rem also to the spatial part of ξ µ as follows:
ξ0 ≡ α , ζi = ∂i β + i , (∂ l l = 0) , (4.111)
where α and β are scalars and i is a divergenceless vector. Now let us write the
transformations found in Eqs. (4.98) and (4.99) using the SVT decomposition:
ψ̂ = ψ − Hα − α0 , (4.112)
∂i ŵ + Ŝi = ∂i w + Si − ∂i β 0 − 0i + ∂i α , (4.113)
1
φ̂ = φ − Hα − ∇2 β , (4.114)
3
1 1
∂i ∂j − δij ∇2 2µ̂ + ∂j Âi + ∂i Âj + χ̂Tij = ∂i ∂j − δij ∇2 2µ
3 3
2
+∂j Ai + ∂i Aj + χTij − 2∂j ∂i β − ∂j i − ∂i j + δij ∇2 β . (4.115)
3
We are now in the position of writing explicitly the transformation rules for each
class of perturbation.
Scalar perturbations and their gauge-invariant combinations

By taking the divergence ∂ i of Eq. (4.113) and twice the divergence ∂ i ∂ j of Eq. (4.115),
we eliminate all the vector and tensor contributions and are left with the transfor-
mation equations for the scalar perturbations only:
ψ̂ = ψ − Hα − α0 , (4.116)
∇2 ŵ = ∇2 (w − β 0 + α) , (4.117)
1
φ̂ = φ − Hα − ∇2 β , (4.118)
3
2 2 2 2
∇ ∇ µ̂ = ∇ ∇ (µ − β) . (4.119)
From the above transformations we obtain:
ψ̂ = ψ − Hα − α0 (4.120)
ŵ = w − β 0 + α (4.121)
1
φ̂ = φ − Hα − ∇2 β (4.122)
3
µ̂ = µ − β (4.123)
114
In principle, we should have extra functions in the second and fourth equations,
coming from the integration of ∇2 and ∇2 ∇2 , but these are spurious gauge
modes which can be set to zero without losing of generality.
The following combinations of scalar perturbations are gauge-invariant:
1 0 1
Ψ=ψ+ [(w − µ0 ) a] Φ = φ + H (w − µ0 ) − ∇2 µ (4.124)
a 3
They are the famous Bardeen’s potentials [Bardeen, 1980].
Exercise 4.18. Prove that the Bardeen potentials are gauge-invariant. Why are
there only two of them?
The same technique that we have just used for the metric perturbations can be
applied to the matter quantities in Eqs. (4.102) and (4.103).
Exercise 4.19. Applying the SVT decomposition to vi :
vi = ∂i v + Ui , (∂ l Ul = 0) , (4.125)
show that, for scalar perturbations, one gets:
ˆ = δρ − ρ̄0 α
δρ ⇒ δ̂ = δ + 3H(1 + P̄ /ρ̄)α (4.126)
v̂ = v + α (4.127)
ˆ = δP − P̄ 0 α
δP (4.128)
Note how a cosmological constant has gauge invariant perturbations.
As we did earlier for the geometric quantities, we can combine the above trans-
formations for matter in order to obtain gauge-invariant variables. The strategy
is, in general, to combine the transformations in order to eliminate α and β. The
result is then manifestly gauge-invariant. Using matter quantities only, we can
eliminate α from Eqs. (4.126) and (4.128), thus obtaining the following gauge-
invariant perturbation:
P̄ 0
Γ ≡ δP − 0 δρ (4.129)
ρ̄
This is the entropy perturbation. The ratio δP/δρ is called effective speed of
sound, whereas the ratio P̄ 0 /ρ̄0 is called adiabatic speed of sound. When the
two are equal, i.e. Γ = 0 one finds dS = 0, i.e. one has adiabaticity.
The other gauge-invariant combinations are:
0
δρ(gi)
m ≡ ρ̄∆ ≡ δρ + ρ̄ v δPm(gi) ≡ δP + P̄ 0 v (4.130)
115
i.e. the gauge-invariant density, also called comoving-gauge density pertur-
bation, and pressure perturbations. The subscript m refers to “matter”, since it
is also possible to build gauge-invariant perturbations of the density and pressure
using metric quantities. We are borrowing this notation from [Bardeen, 1980].
We can now think of combining the geometric and matter transformations, a
total of 7 relations, trying to eliminate α and β in order to create new gauge-
invariant variables.
Indeed, we can form the so-called comoving curvature perturbation
1
R ≡ φ + Hv − ∇2 µ (4.131)
3
also known as Lukash variable [Lukash, 1980], and we can form the quantity:
δρ 1
ζ ≡φ+ − ∇2 µ (4.132)
3(ρ̄ + P̄ ) 3
which was introduced first in [Bardeen et al., 1983] but started to be exploited
in [Wands et al., 2000]. These R and ζ are especially important in the framework of
inflation because they are conserved on large scales and for adiabatic perturbations,
as noticed in [Bardeen, 1980] (at least for R), and as we shall prove in Sec. 12.4.
Again, we can form gauge-invariant density, velocity and pressure perturba-
tions:
δρ(gi)
g ≡ δρ + ρ̄0 (w − µ0 ) v (gi) ≡ v − (w − µ0 )
δPg(gi) ≡ δP + P̄ 0 (w − µ0 )
(4.133)
In general, having 2 gauge variables α and β and 7 transformations, we can build
5 independent scalar gauge-invariant perturbations.
Vector perturbations and their gauge-invariant combinations

We can now eliminate the scalar contribution from Eq. (4.113) and consider the
divergence ∂ j of Eq. (4.115). In this way we shall find the transformations for
vector perturbations:
Ŝi = Si − 0i (4.134)

∇2 Âi = ∇2 (Ai − i ) . (4.135)
From the second equation, we can find the following transformation:
Âi = Ai − i (4.136)
It is possible to define a new gauge-invariant vector potential, which has the fol-
lowing form:
Wi ≡ Si − A0i (4.137)
116
Exercise 4.20. Prove that Wi is gauge-invariant. Why is there only one gauge-
invariant vector perturbation?
Using the SVT decomposition of Eq. (4.125), show that from the matter sector
we just have:
Ûl = Ul , (4.138)
i.e. the vector contribution of vi is already gauge-invariant.
Tensor perturbations
Since an infinitesimal gauge transformation, cf. Eq. (4.90), cannot be realised by
any rank-2 tensor, the following result is not unexpected:
χ̂Tij = χTij (4.139)
i.e. that the transverse part of χij is already gauge-invariant.
Summary
We have thus seen that a generic perturbation of the metric can be split in:
• 4 scalar functions;
• 2 divergenceless 3-vectors, for a total of 4 independent components (2 each);
• A transverse, traceless spatial tensor of rank 2, χTij . It has 2 independent

components.
The total number of independent components sums up to 10, as it should be.
Exercise 4.21. Why does χTij have two independent components only?
The above decomposition holds true not only for the metric but for any rank-2
symmetric tensor.
4.3.3 Gauges
Thanks to gauge freedom we can set any 4 components of the metric to zero. There
are two particularly useful choices: the synchronous gauge and the Newtonian
gauge.
117
Synchronous gauge. This is realised by the choice:
ψ̂ = 0 , ŵi = 0 . (4.140)
Note that ŵi = 0 means that both the scalar and the vector part of wi are being
set to zero. Using the transformations found in the previous subsection, we find:


 ψ − Hα − α0 = 0



w − β0 + α = 0 . (4.141)




Si − 0i = 0

This must be interpreted as a system of equations for the unknowns α, β and i , i.e.
from a generic gauge we want to know which transformations we have to perform
in order to go to the synchronous gauge. We have 4 equation for 4 unknowns,
so we expect to determine a single solution. However, the above equations are
differential and this implies the following:

α = a1 dη aψ + f (x)
R




 R
β = dη (w + α) + g(x) . (4.142)




 0 R
i = dη Si + h(x)
Since we have only time derivatives, the integrations give rise to purely space-
dependent functions, which we have called f , g and h here, and which are spurious
gauge modes.
Newtonian gauge. This is realised by the choice:
ŵ = 0 , µ̂ = 0 , χ̂⊥
ij = 0 . (4.143)
Using the transformations found in the previous subsection, we find:



 w − β0 + α = 0



µ−β =0 . (4.144)




Ai − i = 0

It easy to see that the second and the third equation are algebraic and determine
β and i . Substituting the solution for β into the first equation, we then find α.
There is no integration to perform, therefore no spurious gauge mode appears.
An important fact that makes the conformal Newtonian gauge somewhat special
is that ψ̂ = Ψ and φ̂ = Φ, i.e. the metric perturbations become identical to the
Bardeen potentials. These lecture notes are based on the use of the Newtonian
gauge.
118
Transformations between the two gauges It is useful to provide the trans-
formation rules among the metric perturbations in the two gauges, for those readers
who might want to translate into the synchronous gauge the results of these notes
and comparing them with the huge literature in which this gauge is employed. For
the scalar case, using Eqs. (4.120)-(4.123) and assuming that the hatted quantities
are the synchronous ones whereas the non-hatted perturbations are the conformal
Newtonian ones, we get:
0 = ψN − Hα − α0 , (4.145)
0 = 0 − β0 + α , (4.146)
1
φS = φN − Hα − ∇2 β , (4.147)
3
µS = −β . (4.148)
The second and the fourth equation completely specify the transformation in terms
of µS , i.e. we have:
α = β 0 = −µ0S (4.149)
and the metric potentials are related by:
1
ψN = −Hµ0S − µ00S φN = φS − Hµ0S − ∇2 µS (4.150)
3
In the literature, see e.g. [Ma and Bertschinger, 1995], the Fourier transforms of
φS and µS are usually named h and 6η, respectively.
For vector perturbations we have:
i 0
0 = SN − i , (4.151)
AiS = 0 − i , (4.152)
from which it is straightforward to obtain:

i 0
SN = −AiS . (4.153)
Of course, tensor perturbations are naturally gauge-invariant and thus have the
same functional form in the two gauges. This is also true for any tensor of rank
equal or higher than 2.
4.4 Normal mode decomposition

We are now in the position of writing down explicitly the Einstein equations, fix-
ing a gauge of our choice, which will be the Newtonian one. We expect these
equations to be linear second order partial differential equations, given the per-
turbation scheme employed. Because of this, it is very convenient to express the
perturbations as superpositions of normal modes, i.e. the eigenmodes Q(k, x) of
the Laplacian operator, defined via the Helmholtz equation:
∇2 Q(k, x) = −k 2 Q(k, x) . (4.154)
119
For flat spatial slicing, which is considered here, this normal mode decomposition
of course amounts to a Fourier transform, i.e.
Q(k, x) = eik·x , (4.155)
and a given quantity X(η, x) is expressed as:

d3 k
Z Z
3 −ik·x
X̃(η, k) = d x X(η, x)e , X(η, x) = 3
X̃(η, k)eik·x . (4.156)
(2π)
Consider for example the metric perturbations ψ, φ, wi and χij . For ψ and φ we
simply have that:
d3 k d3 k
Z Z
ik·x
ψ(η, x) = 3
ψ̃(η, k)e , φ(η, x) = 3
φ̃(η, k)eik·x , (4.157)
(2π) (2π)
and similar expressions also apply for δρ and δP .
Using its SVT decomposition in Eq. (4.106), we have for wi the following Fourier
transformation:
d3 k d3 k
Z Z
wi (η, x) = 3
w̃ i (η, k)e ik·x
= 3
[∂i˜w(η, k) + S̃i (η, k)]eik·x . (4.158)
(2π) (2π)
The Fourier transform of a partial spatial derivative is
∂i˜w(η, k) = iki w̃(η, k) , (4.159)
and therefore we have:

d3 k d3 k
Z Z
ik·x
wi (η, x) = i ki w̃(η, k)e + S̃i (η, k)eik·x . (4.160)
(2π)3 (2π)3
So we see that the Fourier transforms of ψ or φ and w are not treated on an equal
footing, because w̃ is multiplied by a ki . A similar argument goes also for vi .
Finally, doing the same for χij in Eq. (4.110), one gets:
d3 k
Z
1
χij (η, x) = −ki kj + δij k 2µ̃(η, k)eik·x
2
(2π)3 3
3
d3 k T
Z Z
dk ik·x
+i [kj Ãi (η, k) + ki Ãj (η, k)]e + χ̃ (η, k)eik·x , (4.161)
(2π)3 (2π)3 ij
with a similar expansion holding true for π i j . We see that µ̃ is multiplied by
a factor k 2 and Ãi is multiplied by a factor k. Therefore, in order to properly
compare perturbations, par condicio is restored by “correcting” as follows:
1 1 1 1
w̃ ≡ − B̃ , ṽ ≡ − Ṽ , µ̃ ≡ Ẽ , Ãi ≡ − F̃i . (4.162)
k k k2 2k
In this way, all scalar quantities are treated on the same footing. Let us see what
happens to the comoving gauge density perturbation:
˜ = δ̃ + 3(1 + P̄ /ρ̄) H Ṽ
∆ (4.163)
k
120
��
��
��
��-� ��-� ��
Figure 4.1: Plot of the modulus of the CDM density contrast at z = 0 (today)
as function of k, using CLASS. The solid line is the result obtained using the
synchronous gauge whereas the dashed one is obtained using the Newtonian one.
The initial conditions are adiabatic and normalised in order to have R = 1. All
the cosmological parameters have been set, as default, corresponding to the Planck
best fit of the ΛCDM model [Ade et al., 2016a].
Since H ∝ 1/η, on sub-horizon scales, i.e. for kη 1, the density contrast becomes
gauge-invariant. We present this fact in Fig. 4.1, where we plot the evolution of the
modulus of the CDM density contrast δc as function of k and for z = 0, computed
with CLASS [Lesgourgues, 2011] in the synchronous (solid line) and Newtonian
(dashed line) gauges.
The plots in Fig. 4.1 are drawn for z = 0, hence the value of the Hubble
parameter is the Hubble constant, H0 = 3 × 10−4 h Mpc−1 . Indeed, when k > H0
the two evolutions coincide.
Now we have to understand how to extract the scalar and vector part from a
full 3-vector quantity. Let us reformulate the FT of wi as follows:
w̃i (η, k) = −ik̂i B̃(η, k) + S̃i (η, k) . (4.164)
We see that we can isolate the scalar part of the FT transform of a 3-vector
perturbation by contracting it with ik̂ i . Indeed:
ik̂ i w̃i (η, k) = B̃(η, k) , (4.165)
because k̂ i S̃i = 0, since ∂ i Si = 0. The vector part is therefore obtained by sub-

tracting the scalar part:
S̃i = δ j i − k̂ j k̂i w̃j . (4.166)
121
We can easily check that this formula satisfies k̂ i S̃i = 0.
What about a tensorial quantity? For the traceless spatial metric perturbation
in Eq. (4.110), we can write:

1 i i
χ̃ij = − k̂i k̂j − δij 2Ẽ − k̂j F̃i − k̂i F̃j + χ̃Tij . (4.167)
3 2 2
Here the scalar contribution can be isolated by contracting with −3k̂ i k̂ j /2. In fact:
3
− k̂ i k̂ j χ̃ij = 2Ẽ . (4.168)
2
Exercise 4.22. Show that the vector contribution is obtained contracting once
with 2ik̂ i and using Eq. (4.168), i.e.

F̃i = 2i δ j i − k̂ j k̂i k̂ l χ̃jl . (4.169)
For the tensor part, show that:
1
χ̃ij = δ i − k̂ k̂i δ j − k̂ k̂j χ̃lm + (δij − k̂i k̂j )k̂ l k̂ m χ̃lm .
T l l m m
(4.170)
2
Verify that k̂ i χ̃Tij = k̂ j χ̃Tij = 0 and δ ij χ̃Tij = 0. Recall that χ̃ij is already traceless.
4.5 Einstein equations for scalar perturbations

In this section we focus on scalar perturbations and employ the Newtonian gauge.
Our perturbed metric is then:
g00 = −a(η)2 [1 + 2Ψ(η, x)] , g0i = 0 , gij = a(η)2 δij [1 + 2Φ(η, x)] , (4.171)
where we are employing the Bardeen potentials, exploiting the fact that w = µ = 0.
In metric (4.171) Ψ plays the role of the Newtonian potential and Φ is the spatial
curvature perturbation.
Exercise 4.23. Calculate from scratch the perturbed Einstein tensor δGµ ν from
metric (4.171), AND do it again using also Eqs. (4.42)-(4.44). Show that:
a2 δG0 0 = −6HΦ0 + 6H2 Ψ + 2∇2 Φ , (4.172)
a2 δG0 i = 2∂i (Φ0 − HΨ) , (4.173)
a00

00 0 0
a δG j = −2Φ − 4HΦ + 2HΨ + 4 Ψ − 2H Ψ + ∇ (Φ + Ψ) δ i j
2 i 2 2
a
−∂ i ∂j (Φ + Ψ) . (4.174)
Note again that ∂ i = ∂i since it is the partial derivative with respect to comoving
coordinates and ∇2 = δ lm ∂l ∂m is the comoving Laplacian. Being comoving, it
is always accompanied by a factor 1/a2 , since together they form the physical
Laplacian.
122
4.5.1 The relativistic Poisson equation
The 0 − 0 Einstein equation is the following:
− 3HΦ0 + 3H2 Ψ + ∇2 Φ = 4πGa2 δT 0 0 . (4.175)
Using the perturbed energy momentum tensor component δT 0 0 , as we can read

from Eq. (4.72), we have:
3HΦ0 − 3H2 Ψ − ∇2 Φ = 4πGa2 δρtot (4.176)
where the total perturbed density is

X X
δρtot = δρi = ρ i δi , (4.177)
i i
i.e. it is the sum of the perturbed densities of all the material components that
constitute our cosmological model, in the same way that we did for the background.
We start here to eliminate the bar over the background quantities. We shall con-
sider the ΛCDM as our standard cosmological model. Thus, we have to deal with
4 contributions:
1. Photons;
2. Neutrinos;
3. CDM;
4. Baryons.
Each of these has its own energy-momentum tensor and the total one, which enters
the right hand side of the Einstein equations, is their sum. The cosmological
constant only contributes at background level.
We have found above the relativistic Poisson equation. Indeed, if we con-
sider a = 1, and hence H = 0, we recover the usual Newtonian Poisson equation
(or almost, since we have two potentials in GR). Written in terms of the density
contrast, using Eq. (4.72), the relativistic Poisson equation is the following:
3HΦ0 − 3H2 Ψ − ∇2 Φ = 4πGa2 (ρc δc + ρb δb + ργ δγ + ρν δν ) (4.178)
As one can see, Eq. (4.178) is a second order partial differential equation (PDE)
which is linear, because we are doing first-order perturbation theory and thus all the
perturbative variables appear with power 1. Because of this, as we anticipated, it is
very convenient to introduce the Fourier transform of the latter, thus transforming
Eq. (4.178) in a linear ordinary differential equation (ODE).
Hereafter, we shall constantly employ the Fourier transform, but drop the tilde
above the transformed quantities as it customary in cosmology because almost
always one deals directly with the Fourier modes rather than with the configuration
space. Therefore, Eq. (4.178) Fourier-transformed and written in the conformal
time is the following:
3HΦ0 − 3H2 Ψ + k 2 Φ = 4πGa2 (ρc δc + ρb δb + ργ δγ + ρν δν ) (4.179)
123
4.5.2 The equation for the anisotropic stress
The next Einstein equation that we present is the traceless part of δGi j , which we
know from Eq. (4.75) to be related to the anisotropic stress πij .
Exercise 4.24. From Eq. (4.174), calculate the trace and the traceless part of
δGi j . Show that:
a00
a2 δGl l = −6Φ00 − 12HΦ0 + 6HΨ0 + 12 Ψ − 6H2 Ψ + 2∇2 (Φ + Ψ) , (4.180)
a
and then
2 i 1 i 2 l i 1 i 2
a δG j − δ j a δG l = − ∂ ∂j − δ j ∇ (Φ + Ψ) . (4.181)
3 3
The Fourier transform of the latter can be written in the following form:

2 i 1 i 2 l 2 i 1 i
a δG j − δ j a δG l = k k̂ k̂j − δ j (Φ + Ψ) , (4.182)
3 3
where we have used k i = k k̂ i and k̂ i is the unit vector denoting the direction of k.
The spatial traceless Einstein equation can thus be written as:

1 i
2 i
k k̂ k̂j − δ j (Φ + Ψ) = 8πGa2 π i j (4.183)
3
since π i j is the spatial traceless part of the energy-momentum tensor, as we know

from Eq. (4.75). On the left hand side, we notice the same operator multiplying the
scalar contribution of χij in Eq. (4.167). Hence, only the scalar contribution of π i j
would contribute on the right hand side. Indeed, contracting the above equation
with k̂i k̂ j , as in Eq. (4.168), we obtain:
k 2 (Φ + Ψ) = 12πGa2 k̂i k̂ j π i j (4.184)
Our k̂i k̂ j π i j corresponds to the −(ρ + P )σ used in [Ma and Bertschinger, 1995].
We leave this equation as it is for the moment. We shall see that k̂i k̂ j π i j is sourced
by the quadrupole moments of the photon and neutrino distributions.
This equation tells us that Φ = −Ψ, i.e. there exists only one gravitational
potential, unless a quadrupole moment of the energy content distribution is present.
For example, when CDM dominated the universe then Φ = −Ψ but this is not the
case in the early universe, because of neutrinos. Even when CDM or DE dominates
but the underlying theory of gravity is not GR one might have Φ 6= −Ψ.6
6
One can probe the value of Φ + Ψ via weak lensing, which we do not address in these notes.
See e.g. [Dodelson, 2017] for a treatise on gravitational lensing.
124
4.5.3 The equation for the velocity
The 0 − i Einstein equation can be written using Eqs. (4.173) and (4.73) as follows:
∂i (Φ0 − HΨ) = 4πGa2 (ρ + P ) vi . (4.185)
Upon FT we get:
iki (Φ0 − HΨ) = 4πGa2 (ρ + P ) vi , (4.186)
and now we must get the scalar part of this vectorial equation, contracting by ik̂ i ,
as we showed in Eq. (4.165). Hence, we get:
k(−Φ0 + HΨ) = 4πGa2 (ρ + P ) V (4.187)
Comparing with the notation employed in [Ma and Bertschinger, 1995], one has
θ = kV . Note that the (ρ + P ) V in the above equation is the total one, hence
making explicit the various contributions one has:

0 2 4 4
k(−Φ + HΨ) = 4πGa ρc Vc + ρb Vb + ργ Vγ + ρν Vν (4.188)
3 3
where we have considered the usual equations of state for the various components,
i.e. Pc = Pb = 0 and Pγ = ργ /3 and Pν = ρν /3.
4.5.4 The equation for the pressure perturbation

Using Eq. (4.180), we can immediately write down the last Einstein equation for
scalar perturbations:
k2
Φ00 + 2HΦ0 − HΨ0 − (2H0 + H2 )Ψ + (Φ + Ψ) = −4πGa2 δP (4.189)
3
Exercise 4.25. Use the previously derived transformations (4.150) from the New-
tonian to the synchronous gauge and write there the Einstein equations.
4.6 Einstein equations for tensor perturbations

Our perturbed FLRW metric, with tensor perturbations only, can be cast as follows:
g00 = −a2 , g0i = 0 , gij = a2 (δij + hTij ) , (4.190)
where hTij is divergenceless and traceless.
125
Exercise 4.26. Start from metric (4.190) and calculate the perturbed Einstein
tensor δGµ ν . Verify the calculations also using Eqs. (4.42)-(4.44). Show that the
only non vanishing components are:
00 0
2a2 δGi j = hTij + 2HhTij − ∇2 hTij (4.191)
Notice that the wave operator has appeared.
The calculation of the above exercise is pretty straightforward because the

tensor nature of the perturbation hTij already suggests that it cannot contribute to
R00 , R0i and the Ricci scalar R (at first-order).
The tensor part of the Einstein equations is thus:
00 0
hTij + 2HhTij + k 2 hTij = 16πGa2 πijT (4.192)
where πijT is the tensorial part of the anisotropic stress, which can be computed
from the total as in Eq. (4.170):
1
πijT = δ l i − k̂ l k̂i δ m j − k̂ m k̂j πlm + k̂ l k̂ m πlm δij − k̂i k̂j . (4.193)
2
In Fourier space, the divergenceless condition can be written down as:
k̂ i hTij = 0 . (4.194)
Therefore, hTij can be expanded with respect to a 2-dimensional basis {ê1 , ê2 } de-
fined on the 2-dimensional subspace orthogonal to k̂. The basis satisfies thus the
condition:
γ ij ea,i k̂j = 0 , γ ij ea,i eb,j = δab , (4.195)
where γij is the metric on the 2-dimensional subspace and a, b ∈ {1, 2}. We can
write then hTij as follows:
hTij (k) = (e1,i e1,j − e2,i e2,j )(k̂)h+ (k) + (e1,i e2,j + e2,i e1,j )(k̂)h× (k) . (4.196)
Note the dependence on k̂ of the combinations of the 2-dimensional basis vectors.

In fact these depend on the orientation of k̂.
Substituting the above expansion into Eq. (4.192) and in absence of quadrupole
moments, h+,× satisfy then the equation:
h00+,× + 2Hh0+,× + k 2 h+,× = 0 (4.197)
which we will employ in order to investigate the GW production during inflation

in Sec. 8.4.
If we choose a Cartesian reference frame and k̂ = ẑ, i.e. a propagation direction
of a gravitational wave along ẑ, then a natural choice is ê1 = x̂ and ê2 = ŷ and the
perturbed metric can be written as:
 
h+ h× 0
hTij (kẑ) =  h× −h+ 0  . (4.198)
0 0 0
126
Though this a convenient way of expressing hTij (kẑ), one usually prefers to use the
combinations:
h+ ∓ ih× , (4.199)
since these have helicity ±2, see e.g. [Weinberg, 1972]. In order to see this, we
apply a rotation about ẑ and calculate how hTij (kẑ) transforms.
Exercise 4.27. Apply the rotation:

 
cos θ − sin θ 0
Ri j (θ) =  sin θ cos θ 0  , (4.200)
0 0 1
about the ẑ axis to hTij , i.e. compute the components:
h̄Tlm = Ri l Rj m hTij , (4.201)
and show that:
h̄+ = cos2 θh+ + 2 sin θ cos θh× − sin2 θh+ = cos 2θh+ + sin 2θh× , (4.202)
h̄× = cos2 θh× − 2 sin θ cos θh+ − sin2 θh× = cos 2θh× − sin 2θh+ . (4.203)
Hence, the aforementioned combinations h+ ± ih× transform as:
h̄+ ± ih̄× = e∓2iθ (h+ ± ih× ) (4.204)
and have thus helicity ∓2. Sometimes the sign could be a bit confusing depending
in which sense the rotation is performed. By convention, θ > 0 denotes an anti-
clockwise rotation so a rotation of θ about the ẑ axis corresponds to a −θ rotation
about the −ẑ axis, which is the line of sight and therefore the relevant direction
for the observer. So, the observed helicities have opposite sign with respect to the
propagating ones.
So, we write the total tensor perturbation as a sum over the helicities:
X
hTij (η, kẑ) = eij (ẑ, λ)h(η, kẑ, λ) , (4.205)
λ=±2
where:
1
e11 (ẑ, ±2) = −e22 (ẑ, ±2) = ∓ie12 (ẑ, ±2) = ∓ie21 (ẑ, ±2) = √ , (4.206)
2
and of course e3i = ei3 = 0. Therefore, from Eq. (4.205) we get:
1 1
h+ (η, kẑ) = √ h(η, kẑ, +2) + √ h(η, kẑ, −2) , (4.207)
2 2
i i
h× (η, kẑ) = √ h(η, kẑ, +2) − √ h(η, kẑ, −2) , (4.208)
2 2
127
and inverting:
√
2h(η, kẑ, +2) = h+ (η, kẑ) − ih× (η, kẑ) , (4.209)
√
2h(η, kẑ, −2) = h+ (η, kẑ) + ih× (η, kẑ) . (4.210)
For k in a generic direction, we have that:
X
hTij (η, k) = eij (k̂, λ)h(η, k, λ) , (4.211)
λ=±2
where the polarisation tensor is defined as:

√
eij (k̂, ±2) = 2e±,i e±,j , (4.212)
where the polarisation vectors are:
(e1 ± ie2 )i
e±,i (k̂) ≡ √ . (4.213)
2
Of course eij (k̂, λ) has the same symmetry of hTij (η, k) and thus is traceless and
transverse, i.e.
k̂ l elm (k̂, λ) = 0 , (4.214)
Again, the two h(η, k, ±2) satisfy the same Eq. (4.197) as for h+,× .
When we will discuss about the effect of GW on photon propagation and CMB
we shall have then to deal with two directions: one is the GW direction k̂ and
the other is the photon direction of propagation, p̂. In general a frame k̂ = ẑ
is chosen in order to simplify the calculations. But then, before taking the anti-
Fourier transform and obtaining the physical quantities in the real space one has
to remember of performing a rotation which brings k̂ back in a generic direction.
We shall see this in detail in Chapter 10.
4.7 Einstein equations for vector perturbations

Finally, we address vector perturbations. The vector-perturbed FLRW metric, in
the Newtonian gauge, has the following form:
g00 = −a2 , g0i = 0 , gij = a2 (δij + hVij ) , (4.215)
where
hVij = ∂i Aj + ∂j Ai , (4.216)
and Ai is a divergenceless vector, i.e. ∂i Ai = 0.
Exercise 4.28. Repeat the very same calculations performed in the tensor case
but now for hVij . Notice a very important difference: hVij is traceless but NOT
divergenceless. Compare the results with those found using Eqs. (4.42)-(4.44).
Show that the non-vanishing components of the perturbed Einstein tensor are:
0
2a2 δG0 i = −∂l hVli = −∇2 A0i , (4.217)
00 0
2a2 δGi j = hVij + 2HhVij = (∂i Aj + ∂j Ai )00 + 2H(∂i Aj + ∂j Ai )0 . (4.218)
128
With the Laplacian missing, the last equation has no more the wave behaviour
that the corresponding tensor equation has. With no vector sources, show then
that in the early, radiation-dominated universe, for which H = 1/η, one has:
hVij ∝ 1/η 2 , (4.219)
and hence vector perturbations vanish, if not sourced.
Einstein equations are thus:

kFi0 = −32πGa2 (ρ + P )Ui , (4.220)
(ik̂i Fj + ik̂j Fi )00 + 2H(ik̂i Fj + ik̂j Fi )0 = −32πGa2 πijV , (4.221)
where Ui is the vector part of vi , defined in Eq. (4.125), Fi is defined in Eq. (4.162)
and πijV is the vector part of the anisotropic stress, defined as from Eq. (4.169):

πijV = 2i δ m i − k̂ m k̂i k̂ l πlm k̂j + 2i δ m j − k̂ m k̂j k̂ l πlm k̂i , (4.222)
Of course, we have that:
k̂ i Fi = 0 , k̂ i Ui = 0 , k̂ i k̂ j πijV = 0 . (4.223)
Contracting the second Einstein equation with ik̂ j leaves us with:
Fi00 + 2HFi0 = 32πGa2 ik̂ j πijV (4.224)
We choose, as in the tensor case, to align k̂ to ẑ. Therefore, the divergenceless of

Ai implies that
ik i Ai = 0 ⇒ A3 = 0 , (4.225)
and from Eq. (4.216) we have:
i i
hVij = iki Aj + ikj Ai = − k̂i Fj − k̂j Fi . (4.226)
2 2
V
So, hij can be written as follows:
 
0 0 −iF1 /2
hVij =  0 0 −iF2 /2  . (4.227)
−iF1 /2 −iF2 /2 0
Exercise 4.29. Applying the same rotation about ẑ as we did for tensor pertur-
bations. Show that:
F̄1 = F1 cos θ − F2 sin θ , (4.228)
F̄2 = F1 sin θ + F2 cos θ . (4.229)
Hence we have:
F̄1 + iF̄2 = (F1 + iF2 )eiθ , (4.230)
F̄1 − iF̄2 = (F1 − iF2 )e−iθ , (4.231)
and therefore the quantities:
F± ≡ F1 ± iF2 , (4.232)
are fields with helicity ±1.
129
Chapter 5
Perturbed Boltzmann equations
the Boltzmann equation plays a similar role for physicists and as-
tronomers: no one ever talks about it, but everyone is always thinking
about it
—Scott Dodelson, Modern Cosmology
We derive in this Chapter the perturbed Boltzmann equations for photons,

massless neutrinos, CDM and baryons. We shall use these in order to track the
evolution of small fluctuations in these components, and couple them to Einstein
equations. For deriving the hierarchy of temperature and polarisation for photons
we follow mainly [Ma and Bertschinger, 1995], [Hu and White, 1997] and [Tram
and Lesgourgues, 2013]. We shall focus most of the time on the photon perturbed
Boltzmann equation which is much more laborious than the others. This is because
photons are massless and interact, therefore we cannot truncate the hierarchy and
a collisional term must be taken into account. The latter comes from Thomson
scattering, whose cross-section depends also on the polarisation, thus further com-
plicating the treatment of photons fluctuations, which we nonetheless will bravely
face.
5.1 General form of the perturbed Boltzmann equa-

tion
Let us build in a general fashion the perturbed Boltzmann equation, in order to
understand first and to tackle separately afterwards the terms which constitute it.
As we did in Eq. (4.77), the distribution function f can be split in a background
contribution plus a perturbation:
f (η, x, p) = f¯(η, p) + F(η, x, p) , (5.1)
where we have left explicit the functional dependence in order to stress that, after
breaking homogeneity and isotropy, the perturbed distribution function depends on
7 variables. Covariance demands the total distribution function f to be a function
of the 4-position and of the 4-momentum:
dxµ
f = f (xµ , P µ ) , Pµ = , (5.2)
dλ
130
where λ is an affine parameter. However, the total number of independent variables
is not 8 but 7, due to the mass-shell relation gµν P µ P ν = −m2 .
We choose these seven variables to be xµ , the proper momentum modulus p and
the proper momentum direction p̂i . Now, we are ready to write down the Liouville
operator with the above choice of coordinates in the phase space:
df ∂f ∂f ∂f dp ∂f dp̂i
= P0 + P i
+ + , (5.3)
dλ ∂η ∂xi
|{z} ∂p dλ ∂ p̂i dλ
| {z }
1st order 2nd order
where we have already put in evidence the fact that ∂f /∂xi , ∂f /∂ p̂i and dp̂i /dλ
are pure first-order quantities, thereby making the last term of second order and
thus negligible. Let us see in some more detail why.
The term ∂f /∂xi is of first order because it breaks homogeneity and the latter
can be broken only at first order. The same reasoning goes for ∂f /∂ p̂i , which
breaks isotropy. Finally, dp̂i /dλ is identically zero in the background cosmology,
because there is nothing that could deviate the path of a photon in a homogeneous
and isotropic space. Therefore, dp̂i /dλ is a first-order quantity.
Dividing by P 0 , we can write the first-order perturbed Boltzmann equation as
follows:
df ∂f ∂f P i ∂f dp 1
= + i 0+ = 0 C[f ] (5.4)
dη ∂η ∂x P ∂p dη P
So, we have to deal with 3 terms:
1. The velocity term P i /P 0 .
2. The force term dp/dη.
3. The collisional term C[f ].
The velocity term is the easiest one to compute because it must be at zeroth order,
being ∂f /∂xi a first-order quantity, as we have just discussed. Therefore, as we
already showed in Chapter 3:
Pi p
0
= p̂i , (5.5)
P E
which for photons and neutrinos simply becomes p̂i .
The force term is made explicit through the geodesic equation and hence de-
pends on the metric perturbations. The only collisional term that we shall consider
explicitly is the one related to Thomson scattering and will be of interest just for
photons and free electrons.
5.1.1 On the photon and neutrino perturbed distributions

As we know, in the Hot Big Bang model the background distribution function for
photons and neutrinos are the BD and the FD ones:
1 1
f¯γ = , f¯ν = , (5.6)
epγ /Tγ −1 epν /Tν +1
131
with Tγ (η) ∝ 1/a(η) (far from electron-positron annihilation) and Tν (η) ∝ 1/a(η),
because of the thermal equilibrium and the cosmological principle. Their perturbed
distribution functions can be written in the same form, but with a perturbed
temperature:
Tγ (η, x, p) = Tγ (η) + δTγ (η, x, p) , Tν (η, x, p) = Tν (η) + δTν (η, x, p) . (5.7)
So, let us write the distribution functions as follows:

−1 p
f = exp ∓1. (5.8)
T + δT
Exercise 5.1. For δT T , show that:
∂ f¯ δT
f = f¯ − p (5.9)
∂p T
Hence:
∂ f¯ δT
F = −p , (5.10)
∂p T
the perturbed distribution function for neutrinos and photons can be physically
interpreted as their relative temperature fluctuations.
5.2 Force term

We now deal with the force term, which originates from the geodesic equation. We
shall calculate it first for a perturbed metric a2 hµν with h0i = 0 and then specify
the result obtained for the various perturbation types: scalar, tensor and vector.
Let us write the mass-shell relation gµν P µ P ν as follows:
− a2 (1 − h00 )(P 0 )2 + a2 (δij + hij )P i P j = −m2 , (5.11)
and hence define the energy and proper momentum modulus as:
E 2 = a2 (1 − h00 )(P 0 )2 , p2 = a2 (δij + hij )P i P j . (5.12)
Exercise 5.2. Determine dE/dη using the geodesic equation:
dP 0
= −Γ0αβ P α P β . (5.13)
dλ
First show that:
dE d(h00 /2) P αP β 2
= HE − E − Γ0αβ a (1 − h00 ) . (5.14)
dη dη E
132
Working out the Christoffel symbol term:
−2Γ0αβ P α P β a2 (1 − h00 ) = −2aa0 (1 − h00 )(P 0 )2 + a2 h000 (P 0 )2 + 2a2 ∂l h00 P 0 P l

−2aa0 (δij + hij )P i P j − a2 h0ij P i P j . (5.15)
Hence, using again the definitions of P 0 and of the proper momentum, get:
P α
Pβ 2 h000 p2 1 0 i j
− Γ0αβ a (1 − h00 ) = −HE + E l
+ pp̂ ∂l h00 − H − h p p . (5.16)
E 2 E 2E ij
Collecting all the above found contributions, we have
dE p2 h00 1 0 i j
= −H + pp̂l ∂l − h pp , (5.17)
dη E 2 2E ij
and using the differentiated mass-shell relation EdE = pdp, we can write:
dp h00 p 0 i j
= −Hp + E p̂l ∂l − hij p̂ p̂ (5.18)
dη 2 2
As we can see, if we choose the synchronous gauge then h00 = 0 and all the per-
turbation types are elegantly taken into account in the last term. The Boltzmann
equation thus becomes:
pp̂i ∂f

0 E l h00 1 0 i j ∂f a
f + i
+ p −H + p̂ ∂l − hij p̂ p̂ = C[f ] , (5.19)
E ∂x p 2 2 ∂p E
where we have already used the fact that the collisional term is a first-order quantity
and thus neglected its multiplication by h00 . Separating the distribution function
in its background plus perturbed part f = f¯ + F, the perturbed part of the above
Boltzmann equation can be written as:
h00 p 0 i j ∂ f¯

0ikµp ∂F a
F + F − Hp + ikµE − hij p̂ p̂ = C[F] (5.20)
E ∂p 2 2 ∂p E
where we have defined

µ ≡ k̂ · p̂ (5.21)
This quantity plays a major role in the rest of these notes, so keep it in mind.
Note that the left hand side of the perturbed Boltzmann equation has an azimuthal
dependence possibly coming from h0ij p̂i p̂j . We shall see that for scalar perturbations
no azimuthal dependence appears.
5.2.1 Scalar perturbations

Using metric (4.171), we can identify h00 = −2Ψ and hij = 2Φδij . the proper
momentum modulus is defined as:
p2 = a2 (1 + 2Φ)δij P i P j , pi = a(1 + Φ)P i , (5.22)
133
and
E
P0 = (1 − Ψ) . (5.23)
a
The velocity term up to first-order is thus the following:
dxi Pi p(1 − Φ)p̂i p

= 0 = = p̂i (1 − Φ + Ψ) , (5.24)
dη P E(1 − Ψ) E
but of course we have to neglect Ψ − Φ in the Liouville operator since ∂f /∂xi is of

first order.
From Eq. (5.18), we have:
dE p2 p2
= −H − Φ0 − pp̂i ∂i Ψ , (5.25)
dη E E
and then:
dp
= −Hp − pΦ0 − E p̂i ∂i Ψ (5.26)
dη
The first term on the right hand side of the above equation is of zeroth order and
is responsible for the cosmological redshift. The second takes into account the time
variation of the spatial curvature and the third is the gradient of the gravitational
potential along the direction of the photon.
Therefore, the perturbed Boltzmann equation with scalar perturbations has the
following general form:
p i ∂F ∂ f¯ a
F0 + p̂ ∂i F − Hp − pΦ0 + E p̂i ∂i Ψ = C[F] (5.27)
E ∂p ∂p E
The Fourier transform of the above Boltzmann equation can be cast as follows:
p ∂F ∂ f¯ a
F 0 + ikµ F − Hp − (pΦ0 + ikµEΨ) = C[F] (5.28)
E ∂p ∂p E
5.2.2 Tensor perturbations

From Eq. (4.190) we have that h00 = 0 and hij = hTij , transverse and traceless.
From Eq. (5.18), we have:
dE p2 p2 T 0 i j dp p 0
= −H − h p̂ p̂ , = −Hp − hTij p̂i p̂j , (5.29)
dη E 2E ij dη 2
and the Fourier transform of the Boltzmann equation can be cast as follows:
p ∂F p 0 ∂ f¯ a
F 0 + ikµ F − Hp − hTij p̂i p̂j = C[F] (5.30)
E ∂p 2 ∂p E
If we choose k̂ = ẑ, from Eq. (4.198) we have hT11 = −hT22 = h+ and hT12 =
hT21 = h× , hence the contributions p̂1 and p̂2 shall be selected, producing thus
134
an azimuthal dependence. The particle direction unit vector can be written in
spherical coordinates as follows:
p p
p̂ = p̂ = ( 1 − µ2 cos φ, 1 − µ2 sin φ, µ) (5.31)
again, because we have fixed beforehand k̂ = ẑ.

0
hTij p̂i p̂j = h0+ (1 − µ2 ) cos 2φ + h0× (1 − µ2 ) sin 2φ =
r
2π 2
Y2 (µ, φ)(h0+ − ih0× ) + Y2−2 (µ, φ)(h0+ + ih0× ) =

2
15
r
π 2
Y2 (µ, φ)h0 (λ = +2) + Y2−2 (µ, φ)h0 (λ = −2) .

4 (5.32)
15
So the metric contribution in the case of tensor perturbations carries a Y2±2 (µ, φ)
proportionality, each coupled to the respective helicity of the GW.
5.2.3 Vector perturbations

From Eq. (4.215) we have that h00 = 0 and hij = hVij , traceless and with vanishing
double divergence. From Eq. (5.18), we have:
dE p2 p2 V 0 i j dp p 0
= −H − h p̂ p̂ , = −Hp − hVij p̂i p̂j , (5.33)
dη E 2E ij dη 2
and the Fourier transform of the Boltzmann equation can be cast in a form identical
to the tensor:
p ∂F p 0 ∂ f¯ a
F 0 + ikµ F − Hp − hVij p̂i p̂j = C[F] (5.34)
E ∂p 2 ∂p E
In Eq. (4.227) we have defined hV13 = −iF1 /2 and hV23 = −iF2 /2, hence again an
azimuthal dependence shall appear. In particular, using Eq. (5.31):
0
p p
hVij p̂i p̂j = −iF10 1 − µ2 µ cos φ − iF20 1 − µ2 µ sin φ =
r
2π 1
Y2 (µ, φ)(iF10 + F20 ) + Y2−1 (µ, φ)(iF10 − F20 ) =

−
15
r
2π 1
Y2 (µ, φ)F−0 + Y2−1 (µ, φ)F+0 .

−i (5.35)
15
So the metric contribution in the case of tensor perturbations carries a Y2±1 (µ, φ)
proportionality. Again, this result holds true only for k̂ = ẑ.
One might ask about a possible Y1±1 (µ, φ) contribution. This is absent because
of our choice h0i = 0.
135
5.3 The perturbed Boltzmann equation for CDM
We are going to consider only scalar perturbations sourcing the evolution of CDM
since, as we are going to see, we shall need only two equations: one for the density
contrast and the other for the fluid velocity. The former only has a scalar contri-
bution whereas the latter does have a vector contribution, but negligible, as vector
perturbations usually are.
Hence, the perturbed Boltzmann equation for CDM is Eq. (5.27) with the
collisional term vanishing. We have seen that the latter may be important, at least
for WIMPs, at energies of the order of tens of MeV, before kinetic decoupling.
However, we are not interested in epochs so primordial.
We are going to take moments of Eq. (5.27) and show that we can neglect
all of them except the first two, because CDM particles must be very massive, if
thermally produced. That is, we can treat CDM in the fluid approximation.
Recalling the definitions Eqs. (4.80)-(4.84), multiply Eq. (5.27) with no C[F]
by E and integrate in the momentum space:
d3 p ∂F d3 p ∂ f¯
Z Z
0 i 0
δρ + (ρ + P )∂i v − H pE − Φ pE = 0 . (5.36)
(2π)3 ∂p (2π)3 ∂p
The integral multiplying ∂i Ψ is vanishing because f¯ has no angular dependence
and thus when integrated with p̂i the result is zero.
Exercise 5.4. Show, integrating by parts, that:

d3 p ∂F d3 p p2
Z Z
−H pE = 3H F E+ = 3H (δρ + δP ) , (5.37)
(2π)3 ∂p (2π)3 3E
and that
d3 p ∂ f¯ d3 p ¯ p2
Z Z
0 0
−Φ 3
pE = 3Φ 3
f E+ = 3Φ0 (ρ + P ) . (5.38)
(2π) ∂p (2π) 3E
Thus the integrated perturbed Boltzmann equation becomes:

δρ0 + (ρ + P )∂i v i + 3H(δρ + δP ) + 3(ρ + P )Φ0 = 0 . (5.39)
Exercise 5.5. Show, using δρ = ρδ, that:

0 δP
δ + (1 + w)kV + 3H − w δ + 3(1 + w)Φ0 = 0 (5.40)
δρ
where we have introduced the background equation of state

P
w≡ . (5.41)
ρ
It is necessary to use the background conservation equation ρ0 = −3H(ρ + P ) at
some point during the calculation.
136
Equation (5.40) is valid for any kind of fluid with pressure, though for CDM
we shall consider w = 0 and δP = 0.
Equation (5.40) is not enough for describing the behaviour of CDM because we
do not know how V evolves. In order to find an evolution equation for V we take
the first moment of Eq. (5.28). That is, we multiply it by pp̂i and then integrate.
There are four contributions which we address separately. We already consider
contraction with ik̂i in order to single out the scalar contribution.
1. The first term is very straightforward:

d3 p ∂F i
Z
∂
ik̂i 3
pp̂ = [(ρ + P )V ] = ρ(1 + w)V 0 − 3Hρ(1 + w)2 V + ρw0 V . (5.42)
(2π) ∂η ∂η
For CDM, we shall take w = 0.
2. The second one is:

d3 p pp̂l i
Z
− k̂i kl F pp̂ = −k k̂i k̂l δT il = −kδP − k k̂i k̂l π il , (5.43)
(2π)3 E
where we have already met k̂i k̂l π il in Eq. (4.184). It corresponds to −(ρ + P )σ
of [Ma and Bertschinger, 1995]. For CDM we shall neglect these contributions
because they are of order p2 /E 2 and hence negligible, being CDM cold.
3. The third integration is the following:

d3 p ∂F 2 i
Z
− ik̂i H p p̂ . (5.44)
(2π)3 ∂p
Exercise 5.6. Integrate by parts and show that:

d3 p ∂F 2 i d3 p
Z Z
− ik̂i H p p̂ = 4iki H Fpp̂i = 4Hρ(1 + w)V . (5.45)
(2π)3 ∂p (2π)3
The similar integration performed with f¯pp̂i in the Φ0 term is vanishing.
4. Finally, the fourth and last integration to be performed is the following:
d3 p ∂ f¯
Z
k k̂i k̂l Ψ 3
Epp̂l p̂i . (5.46)
(2π) ∂p
Since f¯ does not depend on p̂i , the angular integration yields a δ il /3. Hence,
integrating by parts, we arrive at the following result:
d3 p ∂ f¯ d3 p ¯ p2
Z Z
l i
k k̂i k̂l Ψ Epp̂ p̂ = −kΨ f E+ = −kΨρ(1 + w) . (5.47)
(2π)3 ∂p (2π)3 3E
137
We can finally put together the four contributions and the first moment of the
perturbed Boltzmann equation for CDM is the following:
w0 δP/δρ k k̂i k̂l π il

V 0 + H(1 − 3w)V + V − kδ − − kΨ = 0 (5.48)
1+w 1+w ρ(1 + w)
This equation corresponds to the Euler equation.
Exercise 5.7. Find the above two equations starting now from:
∇µ δT µν = 0 , (5.49)
i.e. using the fluid approximation directly, without passing through kinetic theory.
As we can appreciate, at each moment we take new variables appear, such as

V , δP and k̂i k̂l π il , and this demands to take further moments unless we decide
to truncate this procedure. This is possible if the particle velocity is very small,
which amounts to say a very large mass if the particle is in a thermal bath, as
we showed in Chapter 3.1 This is the case for CDM, for which we shall use the
truncated equations:
δc0 + kVc + 3Φ0 = 0 (5.50)
and
Vc0 + HVc − kΨ = 0 (5.51)
We shall employ the same technique of taking momenta of the Boltzmann equation
also for baryons (free electrons). We shall find the same results there as the ones
found here but with a source term in the equation for V , coming from the Thomson
collisional term which couples photons with free electrons.
Exercise 5.8. Find Eqs. (5.50) and (5.51) in the synchronous gauge. Show that
one can exploit the residual gauge freedom in order to choose Vcsyn = 0. Therefore,
in the synchronous gauge CDM is described by a single equation only.
5.4 The perturbed Boltzmann equation for mass-

less neutrinos
We do not discuss massive neutrinos here, though at least one of their families
does have a mass. On the other hand, for sufficiently early times, e.g. before
recombination, they certainly behave as relativistic particles. For a treatment of
the Boltzmann equation for massive neutrinos see [Ma and Bertschinger, 1995].
1
In [Piattella et al., 2013] and [Piattella et al., 2016] the CDM velocity dispersion is taken
into account and the second moment of the Boltzmann equation is computed.
138
Specializing Eq. (5.20) for massless neutrinos, i.e. setting p = E and neglecting
the collisional term, we have:
h00 1 0 i j ∂ f¯ν

0 ∂Fν
Fν + ikµFν − Hp + p ikµ − hij p̂ p̂ =0. (5.52)
∂p 2 2 ∂p
In order to eliminate the partial derivative with respect to p, we multiply the

above equation by p and integrate it in the proper momentum modulus, using the
definition:
dp p2
Z
pFν ≡ 4ρν (η)N (η, k, p̂) . (5.53)
2π 2
The scalar part of N (η, k, p̂) corresponds to the Fν /4 defined in [Ma and Bertschinger,
1995].
We could have started tackling Eq. (5.52) by taking its moments as we did in
the previous section for CDM. However, since massless neutrinos are relativistic,
we are not able to truncate the procedure at some point because p/E = 1, i.e.
p/E never is a small parameter. For this reason it is more convenient to build a
hierarchy from N , as we shall see in this section. For massive neutrinos we also
need a hierarchy because their mass is very small, see [Ma and Bertschinger, 1995].
Performing the dp integration and using the definition of Eq. (5.53) in the
Boltzmann equation (5.52), we get:

0 h00 1 0 i j
(4ρν N ) + 4ikµρν N + 16Hρν N − 4ρν ikµ − hij p̂ p̂ = 0 . (5.54)
2 2
Using the background evolution of the density:
ρ0ν + 4Hρν = 0 , (5.55)
we can finally cast the neutrino Boltzmann equation as follows:

0 h00 1 0 i j
N + ikµN − ikµ − hij p̂ p̂ =0 (5.56)
2 2
Any calculations we would do here for neutrinos shall be repeated almost step-by-
step later for photons. Indeed, the two Boltzmann equations are identical being
the only (huge) difference that for neutrinos we do not have a collisional term or
polarisation.

For scalar perturbations the neutrino Boltzmann equation becomes:
0
N (S) + ikµN (S) + Φ0 + ikµΨ = 0 (5.57)
Inspecting Eq. (5.57) we note that it contains a differential operator which depends
just on µ. Hence, if the initial condition on N (S) is axisymmetric, i.e. also depends
only on µ, at any time we have N (S) (η, k, µ). This shall be our hypothesis.
139
We then expand N (S) (η, k, µ) in Legendre polynomials, or partial waves, fol-
lowing the convention of [Ma and Bertschinger, 1995]:
∞
X
(S) (S)
N (η, k, µ) = (−i)` (2` + 1)N` (η, k)P` (µ) , (5.58)
`=0
Z 1
(S) 1 dµ
N` (η, k) = P` (µ)N (S) (η, k, µ) . (5.59)
(−i)` −1 2
So, applying Eq. (5.59) to Boltzmann equation (5.57) we obtain the following:
Z 1 Z 1 Z 1
(S)0 ik dµ (S) Φ0 dµ ikΨ dµ
N` + µP ` N + P` + µP` = 0 . (5.60)
(−i)` −1 2 (−i)` −1 2 (−i)` −1 2
Using the orthogonality relation of the Legendre polynomials:

Z 1
dx δ``0
P` (x)P`0 (x) = , (5.61)
−1 2 2` + 1
the integrals of the above equation are easily computed as follows:

Z 1 Z 1
dµ δ`0 dµ δ`1
P` = , µP` = . (5.62)
−1 2 2` + 1 −1 2 2` + 1
Therefore, we must distinguish 3 cases: when only one of the above integrals
contributes or none of them does (for ` ≥ 2). Moreover, we employ the following
recurrence relation:
(` + 1)P`+1 (µ) = (2` + 1)µP` (µ) − `P`−1 (µ) , (5.63)
which allows us to write:

Z 1
ik dµ k(l + 1) (S) kl (S)
l
µPl N (S) = Nl+1 − Nl−1 . (5.64)
(−i) −1 2 2l + 1 2l + 1
Exercise 5.9. Show that the hierarchy of Boltzmann equations for neutrinos is
the following:
h i
(S)0 (S) (S)
(2` + 1)N` + k (` + 1)N`+1 − `N`−1 = 0 (` ≥ 2) , (5.65)
(S)0 (S) (S)

3N1 + 2kN2 − kN0 = kΨ (` = 1) , (5.66)
(S)0 (S)
N0 + kN1 = −Φ0 (` = 0) . (5.67)
This infinite set of equations is called hierarchy because the ` equation is sourced
by the ` + 1 and ` − 1 multipoles. Of course, it is not possible to solve numerically
an infinite number of equations so some truncation it is necessary at a certain `max .
140
(S)
Using the definitions (5.53) and (5.59), the monopole N0 can be written as:
dp p2 d3 p
Z Z Z
(S) 1 dµ 1 δρν
4N0 = 2
pF ν = 3
pFν = = δν . (5.68)
ρν 2 2π ρν (2π) ρν
(S)
Hence 4N0 = δν , i.e. the monopole is proportional to the density contrast.2
In Chapter 6 we shall see that indeed the primordial mode excited is just the
monopole, making thus reliable our assumption of axial symmetry. For the dipole
(S)
N1 :
dp p2 d3 p
ik̂ l
Z Z Z
(S) i dµ (ρν + Pν ) 4
4N1 = µ pF ν = pp̂l Fν = Vν = Vν .
ρν 2 2π 2 (2π) 3
ρν ρν 3
(5.69)
(S)
Hence, 3N1 = Vν . Finally, for the quadrupole:
dp p2 d3 p
Z Z Z
(S) 1 dµ 3
4N2 = − P2 (µ) pFν = − p(k̂l p̂l k̂m p̂m − 1/3)Fν
ρν 2 2π 2 2ρν (2π)3

3 lm 1 lm i 3
=− k̂l k̂m δTν − δ δTν i = − k̂l k̂m πνlm . (5.70)
2ρν 3 2ρν
(S)
Therefore, k̂l k̂m πνlm = −4ρν N2 /3. Similar expressions hold true for photons.
The above equations can be compared with those in [Ma and Bertschinger,
(S) (S) (S)
1995] by making the identification 4N0 = δν , 3kN1 = θν and 2N2 = σν .
Exercise 5.10. Show that Eq. (5.57) can be formally integrated as follows:
Z η0
ikµ(ηi −η0 )
(S) (S)
N (η0 , k, µ) = N (ηi , k, µ)e − dη(Φ0 + ikµΨ)eikµ(η−η0 ) , (5.71)
ηi
where ηi is some initial conformal time.
This result is called line-of-sight integral [Seljak and Zaldarriaga, 1996]. It is

a simple formal integration but it is very effective for the numerical calculation of
the evolution of CMB anisotropies. We shall see this in Chapter 10. Note that the
(S)
gravitational potentials Φ and Ψ couple only with N0,1,2 via the Einstein equations.
Hence, a way of avoiding the truncation of the hierarchy of Boltzmann equation
(S)
is to solve Eq. (5.71) together with its integrals which give N0,1,2 . See [Weinberg,
2006].
We now introduce the expansion of a plane wave into spherical harmonics:
∞ X
X `
eik·r = 4π i` Y`m∗ (k̂)Y`m (r̂)j` (kr) , (5.72)
`=0 m=−`
where j` is a spherical Bessel function and Y`m is a spherical harmonic. The latter
is very often used in cosmology and especially in CMB physics because it is the
2 (S)
It is redundant to write N0 since the monopole has only the scalar contribution.
141
natural basis over which to expand quantities defined on the sphere (the celestial
sphere, in astronomy). Since they are extremely important for our work to come,
Sec. 12.5 is dedicated to them as a reminder.
Recalling that µ = k̂ · p̂, we introduce in Eq. (5.71) the following plane wave
expansion:
∞ ` 0
0 0 0
X X
−ikk̂·p̂(η0 −η) ∗
e = 4π (−i)` Y`m
0 (k̂)Y`m
0 (p̂)j`0 (kr) (5.73)
`0 =0 m0 =−`0
where we have defined:
r(η) ≡ η0 − η (5.74)
Using the addition theorem for spherical harmonics, the above result can be written
as: X
eikµ(η−η0 ) = (−i)` (2` + 1)P` (µ)j` (kr) , (5.75)
`
where r ≡ η0 − η and j` (kr) is a spherical Bessel function. Now, note that:
d ikµ(η−η0 )
− ikµΨeikµ(η−η0 ) = Ψ e . (5.76)
dη0
Therefore, Eq. (5.71) can be written as follows:

X
N (S) (η0 , k, µ) = (−i)` (2` + 1)P` (µ)
`
Z η0
(S) d 0
N (ηi , k, µ)j` (kri ) − dη Φ − Ψ j` (kr) . (5.77)
ηi dη0
We shall see in Chapter 6 that at early times (ηi → 0) the monopole contribution is
(S)
dominant, hence we may approximate N (S) (ηi , k, µ) ≈ N0 (ηi , k) and thus write:
Z η0
(S) (S)
N` (η0 , k) = N0 (ηi , k)j` (kη0 ) − dη [Φ0 j` (kr) − Ψj`0 (kr)] . (5.78)
0
Of course we cannot really set ηi = 0 since this is the cosmological singularity. Such
equation should be understood as ηi → 0. Knowing the gravitational potentials
(S)
allows us to determine N` without solving the hierarchy of Boltzmann equations.
The advantage is that Φ and Ψ are determined from the Einstein equations, which
(S)
couple only with N0,1,2 .
Exercise 5.11. Write the Boltzmann equation for neutrinos in the cases of tensor
and vector perturbations. It might be useful to first study those for photons in the
next section.
142
5.5 The perturbed Boltzmann equation for pho-
tons
The Boltzmann equation for photons has a collisional term, coming from Thomson
scattering among photons and free electrons. Thomson scattering is an excellent
approximation as long as the energy of the photon is much smaller than the elec-
tron mass, i.e. hν 511 keV which corresponds to a redshift ≈ 1010 (the same
order as the BBN one). So, as long as we deal with much smaller redshifts, our
approximation is fine. If not, then the Klein-Nishina cross-section should be taken
into account.
The scales of cosmological interest today were well outside the particle horizon
for those large redshifts and we shall see that the evolution of perturbations on
these super-horizon regimes is somewhat peculiar. In order to be convinced of
this, consider a scale k today which, in order to be observationally interesting,
must be much smaller than the particle horizon, which is proportional to H0 .
Thus k H0 . At some time in the past, this condition becomes k H, since
H diverges for η → 0. To be more quantitive, using Friedmann equation in the
radiation-dominated epoch one has:
k ka 102 k
≈ √ ≈ . (5.79)
H H0 Ωr0 (1 + z) H0
So, even a scale k = 100H0 , which is today of order of 40 Mpc (already in the non-
linear regime of evolution) for redshifts larger than 104 was outside the horizon.
An important feature of Thomson scattering is that the photon energy re-
mains unchanged. This means that C[Fγ ] does not depend on the photon energy
p. Hence, as we did in the massless neutrino case, we can integrate the whole
Boltzmann equation with respect to p, and define:
dp p2
Z
pFγ ≡ 4ργ (η)Θ(η, k, p̂) (5.80)
2π 2
where Θ corresponds to Fγ /4 of [Ma and Bertschinger, 1995]. The treatment of the

Boltzmann equation for photons is more complicated with respect to the neutrino
case because of the presence of a collisional term which depends on polarisation.
We have therefore to develop three Boltzmann equations: One for each of Θ, Q
and U . The latter are the two Stokes parameters related to linear polarisation. Θ
is also a Stokes parameter, related to the intensity of light, i.e. its energy density,
and for this reason it is the only parameter which couples to the metric. Thomson
scattering does not produces circular polarisation and thus V = 0. If the reader
is not familiar with polarisation of light and Stokes parameters, Sec. 12.7 offers a
brief reminder.
In the next subsection we explicitly compute the collisional term for the Boltz-
mann equation for photons without considering polarisation.
143
5.5.1 Computing the collisional term neglecting polarisation
Consider scattering among photons and free electrons:
e− (q) + γ(p) ↔ e− (q0 ) + γ(p0 ) , (5.81)
where the the photon energy is unchanged, i.e. p = p0 , but its direction has been
modified. The right hand side of Eq. (5.28) can be written as follows:
d3 q d3 q 0 d3 p0
Z Z Z
a a
Ceγ ≡ C[Fγ (p)] = |M|2
p p (2π)3 2Ee (q) (2π)3 2Ee (q 0 ) (2π)3 2p0
(2π)4 δ (3) (p + q − p0 − q0 )δ[p + Ee (q) − p0 − Ee (q 0 )]
f¯e (q 0 )Fγ (p0 ) + Fe (q0 )f¯γ (p0 ) − f¯e (q)Fγ (p) − Fe (q)f¯γ (p) , (5.82)

Evidently, the zeroth order term
f¯e (q 0 )f¯γ (p0 ) − f¯e (q)f¯γ (p) , (5.83)
yields a vanishing integration because the energies do not change. Hence this
collisional term is a perturbative quantity.
At recombination Θ is of order 10−5 , so the linear approximation is reasonable.
However, there is another term which we must take into account: the electron
velocity q/me , which causes a Doppler shift in the CMB photons. Therefore,
we should keep track also of that when working on Eq. (5.82). In particular,
being thepelectron non-relativistic and in the thermal bath with photons, then
q/me ∼ T /me . For thermal energies of ∼ 1 eV then q/me ∼ 10−3 , so it is a
contribution that we definitely want to take into account.
Let us start our task. Expand the electron energy as usual:
p q2
Ee (q) = q 2 + m2e = me + + ... , (5.84)
2me
and write Eq. (5.82) in the following way:
d3 q dp0 p0 d2 p̂0 q2 (p + q − p0 )2
Z Z
aπ 0 2 0
Ceγ = |M(p̂, p̂ )| δ p + −p −
4m2e p (2π)3 (2π)3 2me 2me
f¯e (q)Fγ (p ) + Fe (q)f¯γ (p ) − f¯e (q)Fγ (p) − Fe (q)f¯γ (p) . (5.85)
0 0

Here, we have used the 3-momentum Dirac delta in order to eliminate the integra-
tion with respect to q0 and at the denominator of the volume elements we have
used just Ee = me , because we need to keep track only of terms of the first order
in q/me . Indeed:
1 1 1 − q 2 /(2m2e )
= 2 ≈ , (5.86)
Ee me + q 2me
me
and the q 2 /m2e correction to the collisional term is totally negligible (it is some-
thing of order 10−12 ). We have also put in evidence the (p̂, p̂0 ) dependence of the
probability amplitude and used
fe (p + q − p0 ) ≈ fe (q) (5.87)
144
Physically, this happens because the electron mass-energy is so much larger than
that of the average photon (something like 6 orders of magnitude) that the latter
is unable to deviate the former from its path. It is like deviating a truck with a
tennis ball.
More quantitatively, we can prove the goodness of the above approximation by
using the 4-momentum conservation:
0 0
pµγ + qeµ = pγµ + qeµ . (5.88)
The zero component gives us the energy conservation. Since the photon energy E
does not change, so does not the electron energy Ee .
Exercise 5.12. In Eq. (5.88), put the photon 4-momenta on the same side and
the electron ones on the other side and square. Show that:
E 2 (1 − cos αγ ) = q 2 (1 − cos αe ) , (5.89)
where αγ is the angle between the photon directions and αe is the angle between
the electron directions.
Being in a thermal bath, the photon

√ energy is of order E ∼ T , whereas the
electron momentum is of order q ∼ me T . Therefore:
T
(1 − cos αγ ) ∼ 1 − cos αe . (5.90)
me
Since
p T me (because the electron is non-relativistic) one can see that αe ∼
T /me , i.e. the electron is only slightly deviated upon the scattering.
The electron final energy can be expanded as follows:
(p + q − p0 )2 q2 q · (p − p0 ) (p − p0 )2
= + + , (5.91)
2me 2me me 2me
and using this expansion in the energy Dirac delta in Eq. (5.85) we obtain:
q · (p0 − p)

0
δ p−p + , (5.92)
me
where we have neglected the (p − p0 )2 /me contribution, because |p − p0 | is at most
2p and the typical energy of the photon is that of the thermal bath, i.e. T me .
The Dirac delta can be formally Taylor-expanded as follows:
q · (p0 − p) ∂δ(p − p0 ) q

0
δ p−p + = δ(p − p0 ) + · (p − p0 ) . (5.93)
me ∂p0 me
and the collisional term (5.85) is cast as follows:
d3 q dp0 p0 d2 p̂0
Z Z
aπ
Ceγ = |M(p̂, p̂0 )|2 ×
4m2e p (2π)3 (2π)3
∂δ(p − p0 ) q

0 0
× δ(p − p ) + · (p − p ) ×
∂p0 me
× f¯e (q)Fγ (p0 ) + Fe (q)f¯γ (p0 ) − f¯e (q)Fγ (p) − Fe (q)f¯γ (p) .

(5.94)
145
Exercise 5.13. Show that, exploiting the two Dirac deltas, we obtain:
Z
ane
Ceγ = d2 p̂0 |M(p̂, p̂0 )|2 [Fγ (p0 ) − Fγ (p)]
32π 2 m2e
0
0 2 ∂δ(p − p )
Z
ane vb 0 0 2 0 0

¯γ (p0 ) − f¯γ (p) , (5.95)

+ · p dp d p̂ |M(p̂, p̂ )| (p − p ) f
32π 2 m2e p ∂p0
where
d3 q ¯ d3 q
Z Z
ne ≡ f e (q) , n e m e v b ≡ Fe (q)q . (5.96)
(2π)3 (2π)3
Note that we have assumed that ve = vp = vb , i.e. the velocity of the electron
fluid to be equal to that of the proton one and then we have called it baryon
fluid velocity. This is justified by the fact that Coulomb scattering tightly couples
electrons and protons.
For Thomson scattering we have that (see Sec. 12.8):
Z Z
2 0 0 2 2 2
d p̂ |M(p̂, p̂ )| = 32π me σT , d2 p̂0 |M(p̂, p̂0 )|2 p̂0 = 0 , (5.97)
where the latter equation simply establishes the fact that Thomson scattering has
not a a priori preferred direction. Therefore, integrating by parts the Dirac delta
derivative, we get:
∂ f¯γ 0
Z
0 ane 2 0 0 2 0
Ceγ = τ Fγ (pp̂) + d p̂ |M(p̂, p̂ )| F γ (pp̂ ) + p τ vb · p̂ , (5.98)
32π 2 m2e ∂p
where we have introduced the optical depth:
Z η0
τ (η) ≡ dη 0 ne σT a , τ 0 = −ne σT a (5.99)
η
As we did for neutrinos, since |M|2 does not depend on p, we can multiply the full
Boltzmann equation by p and integrate in the modulus of the proper momentum.
The Boltzmann equation for photons thus becomes:
dp p2
Z
0 h00 1 0 i j
4ργ Θ + ikµΘ − ikµ + hij p̂ p̂ = p Ceγ , (5.100)
2 2 2π 2
and substituting the collisional term of Eq. (5.98) we find:
Z
0 h00 1 0 i j 0 ane
Θ + ikµΘ − ikµ + hij p̂ p̂ = τ Θ + d2 p̂0 |M(p̂, p̂0 )|2 Θ(p̂0 ) − τ 0 p̂ · vb .
2 2 32π 2 m2e
(5.101)
The term containing the Thomson scattering amplitude can be split as follows:
Z Z 2 0 Z
ane 2 0 0 0 d p̂ 0 ane
2
d p̂ |M| Θ(p̂ ) = −τ Θ(p̂ ) + d2 p̂0 |M0 |2 Θ(p̂0 ) .
32π 2 m2e 4π 32π 2 m2e
(5.102)
The last contribution of the above equation has the following form:
2
τ0 X m
Z Z
ane 2 0 0 2 0 0
2 2
d p̂ |M | Θ(p̂ ) = − Y2 (p̂) d2 p̂0 Y2m ∗ (p̂0 )Θ(p̂0 ) . (5.103)
32π me 10 m=−2
We now introduce the contribution from polarisation. We work out the collisional
term in Sec. 12.8.
146
5.5.2 Full Boltzmann equation including polarisation
The original derivation of the Thomson scattering matrix can be found in [Chan-
drasekhar, 1960]. The theoretical framework for the CMB polarisation can be
found e.g. in [Kosowsky, 1996], but also in [Weinberg, 2008]. Here we write already
the final equation that we are going to analyse, following [Hu and White, 1997]
and [Tram and Lesgourgues, 2013], and derive the collisional term in Sec. 12.8.
Following [Tram and Lesgourgues, 2013], the three combined Boltzmann equa-
tions for Θ, Q and U can be written as:
   R d2 p̂0 0

Θ Θ − Θ(p̂ ) − p̂ · v b
∂ 4π
+ ikµ  Q  − τ 0  Q 
∂η
iU iU
−ikµ h200 + 12 h0ij p̂i p̂j
 
+ 0 =
0
 q q 
3 m∗ 0 3 m∗
 m  m∗
2 Y (p̂) Y Θ − E Q − B iU
τ 0 X  1 2m 2
Z
 d2 p̂0  √ m∗ 2 2  0
− 2
E (p̂) − 6Y Θ + 3E m∗
Q + 3B m∗
iU  (p̂ ) , (5.104)
10 m=−2 1 m √ m∗
 2
2
B (p̂) m∗
− 6Y2 Θ + 3E Q + 3B iU m∗
where we have used the definition:
E m ≡ 2 Y2m + −2 Y2m , B m ≡ 2 Y2m − −2 Y2m , (5.105)
where ±2 Y2m are the spin-2 weighted spherical harmonics. Their explicit form is
given in Tabs. 5.1 and 5.2.
Inspecting the above trio of Boltzmann equations in Eq. (5.104), one can see
that iU and B m Q/E m satisfy the same Boltzmann equation. Hence, since they
have the same initial condition (which is zero, as we shall see in Chapter 6), we
just need a single polarisation hierarchy. This is the principal result of [Tram and
Lesgourgues, 2013], which contributes to make the CLASS code faster. Choosing
the Q one, we have thus for the temperature fluctuation:
Z 2 0
∂ 0 d p̂ 0 h00 1 0 i j
+ ikµ Θ − τ Θ − Θ(p̂ ) − p̂ · vb − ikµ + hij p̂ p̂ =
∂η 4π 2 2
2
" r r #
τ0 X m m∗ 2
Z
3 3 (B )
− Y (p̂) d2 p̂0 Y2m∗ Θ − E m∗ Q − Q (p̂0 ) ,(5.106)
10 m=−2 2 2 2 E m∗
and for polarisation:

∂
+ ikµ Q − τ 0 Q =
∂η
2 r " r r #
τ0 X m∗ 2
Z
3 m 3 3 (B )
E (p̂) d2 p̂0 Y2m∗ Θ − E m∗ Q − Q (p̂0 ) . (5.107)
10 m=−2 2 2 2 E m∗
Note that on the left hand side of Eq. (5.106) the dependence on the photon
direction p̂ enters through µ = k̂ · p̂ and through h0ij p̂i p̂j , which also depends on µ
147
m q Y2m q2 Y2
m
0 1
4
5
π
(3 cos2 θ − 1) 3
4
5
6π
sin2 θ
q q
±iφ
±1 1
2
15
2π
sin θ cos θe 1
4
5
π
sin θ(1 ∓ cos θ)e±iφ
q q
±2 1
4
15
2π
sin2 θe±2iφ 1
8
5
π
(1 ∓ cos θ)2 e±2iφ
Table 5.1: Explicit functional form of the spin-0 and spin-2 spherical harmonics.
Note that the spin-−2 spherical harmonics can be obtained by the spin-2 by spatial
inversion, i.e. −2 Y2m (p̂) = 2 Y2m (−p̂). In these notes, we omit the Condon-Shortley
phase.
m
m q E Bm
0 15
8π
sin2 θ 0
q q
±1 − 12 π5 sin θ cos θe±iφ 1
2
5
sin θe±iφ
q qπ
±2 1
4
5
π
(1 + cos2 θ)e±2iφ − 12 5
π
cos θe±2iφ
Table 5.2: Explicit functional form of E m and B m .
for scalar perturbations, cf. Eq. (5.28), and on µ and and the azimuthal angle φ
for tensor and vector perturbations if k̂ = ẑ, cf. Eqs. (5.32) and (5.35). Instead,
on the right hand side of Eq. (5.106) the dependence on the photon direction is
Y2m (p̂).
In order to match the angular dependences on the two sides, making the equa-
tion easier to manipulate, and also in order to easily express the metric contribution
for tensor and vector perturbations, it is convenient to set k̂ = ẑ. In order to recall
this, we define:
ΘP (kẑ) ≡ Q(kẑ) . (5.108)
As we shall see in Chapter 10, before performing the anti-Fourier transform in
order to recover the physical quantities in the real space, we shall have to apply
first a spatial rotation in order to recover a generic direction for k̂. This rotation
compensates if one computes straightaway the angular power spectra, since they
are rotational-invariant quantities. In this case one can use at once the results of
the equations of this section.
Choosing then a frame in which k̂ = ẑ, comparing Y2m (µ, φ) with Eqs. (5.32)
and (5.35) allows us to see that the m = 0 contribution of the sum on the right hand
side of Eq. (5.106) couples to scalar perturbations only, the m = ±2 contributions
couple to tensor perturbations and the m = ±1 to vector perturbations.
Note that for scalar perturbations one has that U = 0 in the reference frame
k̂ = ẑ, since B 0 = 0.
Unless differently stated, in the following the functional dependences of Θ and
ΘP are on η, k = kẑ and µ and φ. The metric quantities do not depend on µ and
φ.
148
For scalar perturbations we have that:
h00 1 0 i j
−ikµ + hij p̂ p̂ = Φ0 + ikµΨ , (5.109)
2 2
(p̂ · vb )(S) = −iµVb , (5.110)
(S)
since recall that the scalar part of vb is vb = −ik̂Vb . Note that, as in the case of
neutrinos, since no azimuthal dependence appears in the differential equation, it is
natural to assume axial symmetry in Θ(S) , i.e. Θ(S) (η, kẑ, µ).
As we shall see in Chapter 10, the scalar contribution Θ(S) is the one which
dominates the CMB temperature fluctuations because it is sourced by scalar per-
turbations in the metric, which are the strongest since they can grow (they are
compressional modes).
(S)
The same goes for ΘP (η, kẑ, p̂), which is not sourced by metric fluctuations
and therefore its angular dependence is less constrained. Nonetheless, we assume
that the scalar part also depends only on µ. Note the dependence on k = kẑ as a
reminder that k̂ = ẑ.
Exercise 5.14. Perform the integration in dφ0 for the m = 0 contribution of

Eqs. (5.106) and (5.107) and using the results of Tabs. 5.1 and 5.2 show that:
dµ0 (S) 0
Z
∂ 0
(S)
+ ikµ Θ − τ Θ − (S)
Θ (µ ) + iµVb + Φ0 + ikµΨ =
∂η 2
τ0 dµ0 h
Z i
(S)
− P2 (µ) P2 Θ(S) − (1 − P2 )ΘP (µ0 ) , (5.111)
2 2
and

∂ (S) (S)
+ ikµ ΘP − τ 0 ΘP =
∂η
0 Z 0 h
τ dµ (S)
i
− [1 − P2 (µ)] −P2 Θ + (1 − P2 )ΘP (µ0 ) .
(S)
(5.112)
2 2
Recall now the partial wave expansions already used for the neutrino distribu-
tion:
Z 1 Z 1
(S) 1 dµ (S) (S) 1 dµ (S)
Θ` = `
P` (µ)Θ (µ) , ΘP ` = `
P` (µ)ΘP (µ) , (5.113)
(−i) −1 2 (−i) −1 2
Exercise 5.15. Show that the Boltzmann equations for the scalar contribution to
the photon temperature and polarisation are:

(S)0 (S) 0 0 (S) (S) 1
Θ + ikµΘ + Φ + ikµΨ = −τ Θ0 − Θ − iµVb − P2 (µ)Π (5.114)
2

(S)0 (S) 0 (S) 1
ΘP + ikµΘP = −τ −ΘP + [1 − P2 (µ)]Π (5.115)
2
149
where Π is defined as follows:
(S) (S) (S)
Π ≡ Θ2 + ΘP 2 + ΘP 0 (5.116)
The anisotropic nature of Thomson scattering and polarisation were neglected

by [Peebles and Yu, 1970] whereas the former only was included by [Wilson and
Silk, 1981]. Polarisation was considered in [Bond and Efstathiou, 1984].
The above two equations are usually expanded in Legendre polynomials, as we
did for the neutrino equation. Using then Eq. (5.113) applied to Eq. (5.114), we
obtain the following equation:
Z 1 Z 1 Z 1
(S)0 ik dµ (S) Φ0 dµ ikΨ dµ
Θ` + µP ` Θ = − P` + µP`
(−i)` −1 2 (−i)` −1 2 (−i)` −1 2
" #
(S) Z 1 Z 1 Z 1
0 Θ0 dµ (S) iVb dµ Π dµ
−τ P` − Θl − µP` − P2 P` .(5.117)
(−i)` −1 2 (−i)` −1 2 2(−i)` −1 2
Using the orthogonality relation of the Legendre polynomials, cf. Eq. (5.61), the
integrals of the above equation are easily computed as follows:
Z 1 Z 1 Z 1
dµ δ`0 dµ δ`1 dµ δ`2
P` = , µP` = , P2 P` = . (5.118)
−1 2 2` + 1 −1 2 2` + 1 −1 2 2` + 1
Therefore, we must distinguish among 4 cases, i.e. when one of the above integrals
contributes or none of them does (for ` > 2). We shall make use again of the
recurrence relation of Eq. (5.63), which allows us to write:
h i
(S)0 (S) (S) (S)
(2` + 1)Θ` + k (` + 1)Θ`+1 − `Θ`−1 = τ 0 (2` + 1)Θ` , `>2 (5.119)
The equation for the quadrupole ` = 2:

(S)0 (S) (S) (S)
10Θ2 + 2k 3Θ3 − 2Θ1 = 10τ 0 Θ2 − τ 0 Π (5.120)
The equation for the dipole ` = 1:

(S)0 (S) (S) (S)
3Θ1 + k 2Θ2 − Θ0 = kΨ + τ 0 3Θ1 − Vb (5.121)
Finally, the equation for the monopole ` = 0:

(S)0 (S)
Θ0 + kΘ1 = −Φ0 (5.122)
For the polarisation equation the steps to be performed are the same. Therefore:
(S)0 k(` + 1) (S) k` (S)

ΘP ` + ΘP (`+1) − ΘP (`−1) =
2` + 1 2` + 1
Z 1 Z 1
0 (S) Π dµ Π dµ
−τ −ΘP ` + P` − P` P2 (µ) . (5.123)
2(−i)` −1 2 2(−i)` −1 2
150
So, the equation for ` > 2 is the following:
h i
(S)0 (S) (S) (S)
(2` + 1)ΘP ` + k (` + 1)ΘP (`+1) − `ΘP (`−1) = τ 0 (2` + 1)ΘP ` , `>2
(5.124)
The equation for the quadrupole ` = 2:

(S)0 (S) (S) (S)
10ΘP 2 + 2k 3ΘP 3 − 2ΘP 1 = 10τ 0 ΘP 2 − τ 0 Π (5.125)
The equation for the dipole ` = 1:

(S)0 (S) (S) (S)
3ΘP 1 + k 2ΘP 2 − ΘP 0 = 3τ 0 ΘP 1 (5.126)
Finally, the equation for the monopole ` = 0:
(S)0 (S) (S)

2ΘP 0 + 2kΘP 1 = 2τ 0 ΘP 0 − τ 0 Π (5.127)
In order to compare with [Ma and Bertschinger, 1995], one has to make the iden-
(S) (S) (S)
tification 4Θ0 = δγ , 3kΘ1 = θγ , 2Θ2 = σγ and kVb = θb .
Note that the monopole and the dipole are thus related to the density contrast
and to the photon fluid velocity, hence they are gauge-dependent. In particular, the
monopole can be reabsorbed into the determination of the background temperature
whereas the dipole by a suitable boost. On the other hand, the quadrupole is
related to the anisotropic stress, so it is gauge-invariant as well as the higher-order
multipoles.
5.5.4 Tensor perturbations

We derive in this section the hierarchy of equations describing the evolution of
fluctuations in the photon distribution caused by tensor perturbations, i.e. by
GW. Consider the tensor contribution of Eq. (5.106). Using Eq. (5.32) we have:
h0+ h0×

∂ (T ) 0 (T ) 2
+ ikµ Θ − τ Θ + (1 − µ ) cos 2φ + (1 − µ2 ) sin 2φ =
∂η 2 2
2
" r r #
τ0 X m 3 (B m∗ )2 (T )
Z
2 0 3 m∗ (T )
− m∗ (T )
Y2 (p̂) d p̂ Y2 Θ − E ΘP − m∗
ΘP (p̂0 ) ,(5.128)
10 m=−2 2 2 E
whereas for the polarisation part:

∂ (T ) (T )
+ ikµ ΘP − τ 0 ΘP =
∂η
2 r " r r #
τ0 X m∗ 2
Z
3 m 3 (T ) 3 (B ) (T )
E d2 p̂0 Y2m∗ Θ(T ) − E m∗ ΘP − ΘP (p̂0 ) . (5.129)
10 m=−2 2 2 2 E m∗
R
Note that in Eq. (5.106) the monopole term dp̂ Θ/4π is a pure scalar and therefore
does not provide any tensor contribution. The scalar product vb · p̂ has a scalar
151
contribution, used in the previous subsection, which is iµVb , and it has also a vector
contribution which can be written as:
(vb · p̂)(V ) = Ub1 sin θ cos φ + Ub2 sin θ sin φ , (5.130)
where we have used Eq. (5.31) and the fact that ki Ubi = 0 because of the vector
nature of Ubi and the choice of having k 3 = k, i.e. k̂ = ẑ. So, vb · p̂ does have an
azimuthal dependence, but not in the form of ei2φ and thus cannot contribute to
the tensor Boltzmann equation (it contributes to the vector one).
Mimicking the azimuthal dependence produced by tensor perturbations in the
metric, we can split the tensor contribution to the temperature anisotropy as fol-
lows, following [Polnarev, 1985] and [Crittenden et al., 1993]:
(T ) (T )
Θ(T ) (µ, φ) = Θ+ (µ)(1 − µ2 ) cos 2φ + Θ× (µ)(1 − µ2 ) sin 2φ
r
π X (T )
=4 Θ (µ)Y2λ (µ, φ) (5.131)
15 λ=±2 λ
where we have reproduced the sum over the helicities of Eq. (5.32), and similarly
for the polarisation field:
(T ) (T ) (T )
ΘP (µ, φ) = ΘP + (µ)(1 + µ2 ) cos 2φ + ΘP × (µ)(1 + µ2 ) sin 2φ
r r
π 3 X (T )
=4 Θ (µ)E λ (µ, φ) (5.132)
15 2 λ=±2 P λ
So we have to select m = ±2 in sum on the right hand side of Eq. (5.106), and we
get:
h0

∂ 0 (T )
+ ikµ − τ Θλ + λ =
∂η 2
0 Z λ∗ 2

τ (T ) 3 (T ) 3 (B ) (T ) λ
− d2 p̂0 Y2λ∗ Θλ Y2λ − E λ∗ ΘP λ E λ∗ − ΘP λ E (p̂0 ) , (5.133)
10 2 2 E λ∗
whereas for the polarisation part:

∂ 0 (T )
+ ikµ − τ ΘP λ =
∂η
τ0 3 λ∗ (T ) λ 3 (B λ∗ )2 (T ) λ
Z
2 0 λ∗ (T ) λ
d p̂ Y2 Θλ Y2 − E ΘP λ E − λ∗
ΘP λ E (p̂0 ) , (5.134)
10 2 2 E
where λ = ±2 and the equations are identical for the two choices. Let us work out
the right hand sides.
Exercise 5.16. With the help of Tabs. 5.1 and 5.2 show that the integral on the
right hand sides becomes:
τ0
Z
15 h 2 2 (T ) 2 4 (T )
i
− dµdφ (1 − µ ) Θλ − (1 + 6µ + µ )ΘP λ . (5.135)
10 32π
152
(T ) (T )
Let us introduce an expansion in Legendre polynomials for Θλ (µ) and ΘP,λ (µ)
similar to that in Eq. (5.113):
Z 1 Z 1
(T ) 1 dµ (T ) (T ) 1 dµ (T )
Θλ,` = `
P` (µ)Θλ (µ) , ΘP λ,` = `
P` (µ)ΘP λ (µ) .
(−i) −1 2 (−i) −1 2
(5.136)
Exercise 5.17. Rewrite the contributions (1 − µ2 )2 and (1 + 6µ2 + µ4 ) in terms

of Legendre polynomials, and show that:
3τ 0
Z
dµ 8 80 8 (T )
− P4 (µ) − P2 (µ) + P0 (µ) Θλ
16 2 35 105 15
3τ 0
Z
dµ 8 32 16 (T )
+ P4 (µ) + P2 (µ) + P0 (µ) ΘP λ . (5.137)
16 2 35 7 5
Hence, using Eq. (5.136), we can write the tensor Boltzmann equation for pho-
tons as follows [Crittenden et al., 1993]:

∂ (T ) 1
+ ikµ − τ Θλ + h0λ =
0
∂η 2

0 3 (T ) 1 (T ) 1 (T ) 3 (T ) 6 (T ) 3 (T )
−τ Θ + Θ + Θ − Θ + Θ − Θ , (5.138)
70 λ,4 7 λ,2 10 λ,0 70 P λ,4 7 P λ,2 5 P λ,0

∂ 0 (T )
∂η

0 3 (T ) 1 (T ) 1 (T ) 3 (T ) 6 (T ) 3 (T )
τ Θ + Θ + Θ − Θ + Θ − Θ . (5.139)
70 λ,4 7 λ,2 10 λ,0 70 P λ,4 7 P λ,2 5 P λ,0
(T ) (T )
The same equations hold true also for Θ+,× and ΘP +,× . The combination of terms
between square brackets in the above equations is sometimes dubbed as Ψ in the
literature. Since for us Ψ is already used as one of the Bardeen potentials, in
Chapter 10 we shall use another notation.
5.5.5 Vector perturbations

For completeness, we present here the Boltzmann equation for photons sourced by
vector perturbations in the metric, though we shall not use it. Using Eqs. (5.35)
and (5.130), we can write:
r
0

∂ τ 2π 1
+ ikµ − τ 0 Θ(V ) − (Y Ub,− + Y2−1 Ub,+ )
∂η cos θ 15 2
r
i 2π 1 0
− (Y2 F− + Y2−1 F+0 ) =
2 15
2
" r r #
0 X Z m∗ 2
τ 3 (V ) 3 (B ) (V )
− Y m (p̂) d2 p̂0 Y2m∗ Θ(V ) − E m∗ ΘP − ΘP (p̂0 ) ,(5.140)
10 m=−2 2 2 2 E m∗
153
where
Ub,± ≡ Ub1 ± iUb2 , (5.141)
and for the polarisation:

∂ 0 (V )
+ ikµ − τ ΘP =
∂η
2 r " r r #
τ0 X 3 (B m∗ )2 (V )
Z
3 m 2 0 3 m∗ (V )
E m∗ (V )
d p̂ Y2 Θ − E ΘP − m∗
ΘP (p̂0 ) . (5.142)
10 m=−2 2 2 2 E
Introduce the vector contribution to temperature anisotropy as follows:

r
i 2π X (V )
Θ (V )
(µ, φ) = Θλ (µ)Y2−λ (µ, φ) (5.143)
cos θ 15 λ=±1
where the factor 1/ cos θ is due to the fact that in Eq. (5.130) only a sin θ appears
and thus it is not proportional to Y2±1 . Similarly, for the polarisation field:
r
(V ) π X (V )
ΘP (µ, φ) =− Θ (µ)E −λ (µ, φ) (5.144)
5 λ=±1 P λ
The vector Boltzmann equation for the temperature can then be written as:

∂ (V ) 1 0
+ ikµ − τ Θλ + iτ 0 Ubλ − Fλ µ =
0
∂η 2
iτ 0 3 −λ∗ (V ) −λ 3 (B −λ∗ )2 (V ) −λ
Z
2 0 −λ∗ (V ) i −λ
µ d p̂ Y2 Θλ 0 Y2 + E ΘP λ E + ΘP λ E , (5.145)
10 µ 2 2 E −λ∗

∂ 0 (V )
∂η
0 Z
3 −λ∗ (V ) −λ 3 (B −λ∗ )2 (V ) −λ

τ 2 0 −λ∗ (V ) i −λ
− d p̂ Y2 Θλ 0 Y2 + E ΘP λ E + ΘP λ E , (5.146)
10 µ 2 2 E −λ∗
With the help of Tab. 5.1 we have then to work out the integral:
τ0
Z
15 h (V ) (V )
i
− dµdφ µ(1 − µ2 )iΘλ + (1 − µ4 )ΘP λ , (5.147)
10 8π
which, written in Legendre polynomial, becomes:
3τ 0
Z
dµ 2 2 (V )
− P1 (µ) − P3 (µ) iΘλ
4 2 5 5
0 Z

3τ dµ 4 4 8 (V )
− P0 (µ) − P2 (µ) − P4 (µ) ΘP λ . (5.148)
4 2 5 7 35
154
Hence, using the usual Legendre expansion, we can write the vector Boltzmann
equation for photons as follows:

∂ (V ) 1 0
+ ikµ − τ µ Θλ + iτ 0 Ub,λ − Fλ =
0
∂η 2

0 3 (V ) 3 (V ) 6 (V ) 3 (V ) 3 (V )
iτ P1 (µ) Θ + Θλ,3 − ΘP λ,4 + ΘP λ,2 + ΘP λ,0 . (5.149)
10 λ,1 10 35 7 5

∂ 0 (V )
+ ikµ − τ µ ΘP λ =
∂η

0 3 (V ) 3 (V ) 6 (V ) 3 (V ) 3 (V )
−τ Θ + Θλ,3 − ΘP λ,4 + ΘP λ,2 + ΘP λ,0 . (5.150)
10 λ,1 10 35 7 5
We end here our treatment of vector modes, dealing just with the scalar and tensor
ones from now on.
5.6 Boltzmann equation for baryons

As baryons, we refer here generically to electrons and protons, neglecting helium
nuclei. The latter can be straightforwardly included, as we are going to comment
throughout the derivation. Electrons and protons interact via Coulomb scattering:
e+p↔e+p, (5.151)
and in turn recall that electrons are also coupled to photons via Thomson scat-
tering. Electrons also interact with Helium nuclei via Coulomb scattering. We
assume the interactions to be so efficient that:
δe = δp = δb , ve = vp = vb . (5.152)
In other words, we do not expect to have more electrons here and more protons
there, thereby generating cosmic dipoles. So, baryons together form the so-called
baryonic plasma. Baryons are heavy and non-relativistic in the epochs of interest
and thus there is no exchange of energy via Thomson scattering with photons. For
this reason we expect that their density contrast is governed by an equation similar
to Eq. (5.50).
We can check this by considering the Boltzmann equation for electrons:
dFe (η, x, q)
= hcep iQQ0 q0 + hceγ ipp0 q0 , (5.153)
dη
where hcep iQQ0 q0 is the collisional term relative to Coulomb scattering:
e(q) + p(Q) ↔ e(q0 ) + p(Q0 ) , (5.154)
whereas hceγ ipp0 q0 is the collisional term relative to Thomson scattering:
e(q) + γ(p) ↔ e(q0 ) + γ(p0 ) . (5.155)
155
For protons, the Boltzmann equation is similar:
dFp (η, x, Q)
= hcep iqq0 Q0 , (5.156)
dη
the only difference being that we neglect their interaction with photons, since the
Thomson cross-section goes as ∝ 1/m2 and therefore it is 106 less important for
protons rather than for electrons. A Boltzmann equation for Helium nuclei is
similar to the one for protons. The brackets mean integration in the phase space:
Z 3 0 Z 3 0
d3 q
Z
dq dQ
h(· · · )iqq0 Q0 ≡ 3 3
(· · · ) , (5.157)
(2π) (2π) (2π)3
whereas the integrands are similar to those in Eq. (5.82).
a (2π)4 δ 4 (q + p − q 0 − p0 )|M|2 [fe (q 0 )fγ (p0 ) − fe (q)fγ (p)]

ceγ ≡ , (5.158)
E(q) 8E(p0 )Ee (p)Ee (q 0 )
with a similar expression, but with different |M|2 , for cep . Note the a/E(q) factor
in the above definition of ceγ that we have left in evidence in order to recall that it
comes comes from changing the affine parameter derivative for the conformal time
derivative in the Boltzmann equation, cf Eq. (5.27).
An important approximation that we are making is the following: we are con-
sidering all the electron ionised (hence the name baryonic plasma). This is fine
before recombination, of course, but it does not work after. After recombination,
electrons and protons (and also helium) can be considered altogether as collision-
less baryons. Therefore, their evolution is governed by equations identical to the
ones we have developed earlier for CDM.
We are going to exploit again the fact that electrons and protons are non-
relativistic, as we did for CDM, and take moments of the perturbed Boltzmann
equations.
The calculations of the integrals of Eqs. (5.153) and (5.156) multiplied by the
respective energies follow the same steps that we have used for Eqs. (5.40) and
(5.48):
d3 q dFe
Z
0 δPe
= δe + (1 + we )kVe + 3H − we δe + 3(1 + we )Φ0 , (5.159)
(2π)3 dη δρe
d3 q dFp
Z
0 δPp
3
= δp + (1 + wp )kVp + 3H − wp δp + 3(1 + wp )Φ0 . (5.160)
(2π) dη δρp
These integrations are equal to the corresponding ones of the collisional terms,
which are very simple and are the following:
d3 q
Z Z 3
dQ
[hc ep iQQ 0 q 0 + hceγ ipp0 q 0 ] = 0 , hcep iQ0 q0 q = 0 , (5.161)
(2π)3 (2π)3
i.e. vanishing because the scattering processes do not change the number of baryons
but only reshuffle their momenta. Mathematically, this can be proved by noticing
that |M|2 is symmetric with respect to the change of initial and final momenta.
156
Therefore, using the conditions stated earlier in Eq. (5.152), the zero-moment
equations that we find for electrons and protons are identical if we set we = wp = 0
and no effective speeds of sound. Neglecting these, we obtain an equation similar
to Eq. (5.50), which we found for CDM:
δ 0 + kVb + 3Φ0 = 0 (5.162)
where, again, kVb = θb in the notation of [Ma and Bertschinger, 1995].

Taking care of the first moments of the perturbed Boltzmann equations for
electrons and protons, as we did for Eq. (5.48), leaves us with:
d3 q i dFe
Z
ik̂i q q̂ = ρe (Ve0 + HVe − kΨ) , (5.163)
(2π)3 dη
Z 3
dQ i dFp
ik̂i 3
Q Q̂ = ρp (Vp0 + HVp − kΨ) , (5.164)
(2π) dη
where we have already neglected pressure and effective speed of sound. The first
moments of the Boltzmann equations are thus:
d3 q q q̂ i dfe (η, x, q) d3 q q q̂ i
Z Z
= [hcep iQQ0 q0 + hceγ ipp0 q0 ] , (5.165)
(2π)3 Ee dη (2π)3 Ee
Z 3
d Q QQ̂i dfp (η, x, Q)
Z 3
d Q QQ̂i
= hcep iqq0 Q0 . (5.166)
(2π)3 Ep dη (2π)3 Ep
Consider the sum of the two above Liouville operator, and using Ve = Vp = Vb , we
can write:
(ρe + ρp )(Vb0 + HVb − kΨ) =

d3 q i
Z Z 3
dQ i
ik̂i q [hcep iQQ0 q0 + hceγ ipp0 q0 ] + ik̂i Q hcep iqq0 Q0 . (5.167)
(2π)3 (2π)3
We write the sum of the background energies as
ρe + ρp = ne me + np mp = nb me + nb mp ≈ nb mp ≡ ρb , (5.168)
since the proton mass is much larger than the electron one, and hence
ik̂i
Vb0 + HVb − kΨ = [hcep (q i + Qi )iQQ0 q0 q + hceγ q i ipp0 qq0 ] . (5.169)
ρb
Now we have to calculate the terms on the right hand side. The first one is:
Z 3 0 Z 3 Z 3 0
d3 q
Z
i i dq dQ dQ i
hcep (q + Q )iQQ0 q0 q = a (q + Qi )
(2π)3 (2π)3 (2π)3 (2π)3
(2π)4 δ 4 (q + Q − q 0 − Q0 )|M|2 [fe (q 0 )fp (Q0 ) − fe (q)fp (Q)]
, (5.170)
8m2e m2p
which is vanishing because q i + Qi is the total initial 3-momentum, so it is equal to

the final one q 0i +Q0i because it is conserved. Another way to see this is noticing that
157
also (q i + Qi )|M|2 is symmetric with respect to the change of the initial momenta
with the final ones and this implies that the integration is zero. Physically, being
conserved the total 3-momentum is unaffected by the scattering.
The only survivor is the second term on the right hand side, which comes from
Thomson scattering. By 3-momentum conservation we can write:
hceγ (q i + pi )ipp0 qq0 = 0 ⇒ hceγ q i ipp0 qq0 = −hceγ pi ipp0 qq0 . (5.171)
That is, we have changed the average with respect to the photon momentum. We
have then:
Z 1
dp p2
Z
0 i i dµ
Vb + HVb − kΨ = − hceγ µpipp0 qq0 = − µ p Ceγ . (5.172)
ρb ρb −1 2 2π 2
Part of the right hand side has already been calculated for photons, when we
integrated over the modulus p and defined Θ(η, k, µ). This integration gives a 4ργ
contribution and leaves us with the same collisional term of Eq. (5.114) integrated
over dµ/2 and without the polarisation contribution:
4iτ 0 ργ 1 dµ
Z
0 (S) (S) 1 (S)
Vb + HVb − kΨ = µ Θ0 − Θ − iµVb − P2 (µ)Θ2 . (5.173)
ρb −1 2 2
(S)
Performing the dµ integration and introducing the dipole Θ1 , we obtain:
4τ 0 ργ (S)

Vb0 + HVb − kΨ = Vb − 3Θ1 (5.174)
3ρb
This equation can be also obtained exploiting momentum conservation for the
two-fluid system baryonic plasma plus photons and using Eq. (5.48). In fact,
momentum is not conserved individually for baryons and photon, but of course it
is for the photon-baryon fluid. Let us write then Eq. (5.48) for the latter:
4 k 4k
ρb (Vb0 + HVb − kΨ) + ργ Vγ0 − ργ δγ − k k̂l k̂m πγlm − ργ Ψ = 0 , (5.175)
3 3 3
where we have neglected pressure and anisotropic stress for baryons, being them
non-relativistic. Now, Eqs. (5.68), (5.69) and (5.70) hold true also for photons. So
we have:
(S) (S) 4 (S) 3
4Θ0 = δγ , 4Θ1 = Vγ , 4Θ2 = − k̂l k̂m πγlm . (5.176)
3 2ργ
Exercise 5.18. Using the above definitions and Eq. (5.121) obtain Eq. (5.174).
We have thus obtained all the relevant equations which describe small fluctua-
tions in the components of the universe. In Chapter 6 we shall tackle the issue of
which initial conditions employ to our set of differential equations.
If we compare the above Eq. (5.174) with the corresponding one in [Ma and
Bertschinger, 1995] we shall notice in this reference an extra term k 2 c2s δb , com-
ing from the fact that the authors did not neglect the effective speed of sound
contribution for baryons, cf. Eq. (5.48).
158
Chapter 6
Initial conditions
Finirai per trovarla la Via... se prima hai il coraggio di perderti

(You will eventually find your Way... if you have first the courage of
lose yourself )
—Tiziano Terzani, Un altro giro di giostra
In Chapters 4 and 5 we have derived the evolution equations for small fluctu-
ations about the homogeneous and isotropic FLRW background. These equations
are differential and so, in order to have a well-posed Cauchy problem, we need
to know which initial conditions to use. This is the topic of this chapter, where
we shall be concerned with scalar perturbations only. Many papers in the litera-
ture are concerned about initial conditions in cosmology, but we shall mainly refer
to [Ma and Bertschinger, 1995] and [Bucher et al., 2000].
We shall discuss primordial modes for scalar perturbations only, leaving the
tensor one in Chapter 8, when we shall discuss of inflation. For this reason in this
Chapter we drop the superscript (S).
6.1 Initial conditions

We start by summarizing here the relevant evolution equations that we have found
in Chapters 4 and 5. For the photon temperature fluctuations, Eqs. (5.119)-(5.122):
(2` + 1)Θ0` + k [(` + 1)Θ`+1 − `Θ`−1 ] = τ 0 (2` + 1)Θ` , (` > 2) , (6.1)

2
10Θ02 + 2k 3Θ3 − Vγ = 10τ 0 Θ2 − τ 0 Π , (6.2)
3

δγ
Vγ0 + k 2Θ2 − = kΨ + τ 0 (Vγ − Vb ) , (6.3)
4
4
δγ0 + kVγ = −4Φ0 , (6.4)
3
159
where we have used 4Θ0 = δγ and 3Θ1 = Vγ . For the polarisation field, Eqs. (5.124)-
(5.127):
(2` + 1)Θ0P ` + k (` + 1)ΘP (`+1) − `ΘP (`−1) = τ 0 (2` + 1)ΘP ` ,

(` > 2) , (6.5)
10Θ0P 2 + 2k (3ΘP 3 − 2ΘP 1 ) = 10τ 0 ΘP 2 − τ 0 Π , (6.6)
3Θ0P 1 + k (2ΘP 2 − ΘP 0 ) = 3τ 0 ΘP 1 , (6.7)
2Θ0P 0 + 2kΘP 1 = 2τ 0 ΘP 0 − τ 0 Π , (6.8)
with
Π = Θ2 + ΘP 2 + ΘP 0 . (6.9)
For neutrinos, Eqs. (5.65)-(5.67):
(2` + 1)N`0 + k [(` + 1)N`+1 − `N`−1 ] = 0 , (` > 1) , (6.10)
δν
Vν0 + 2kN2 − k = kΨ , (6.11)
4
4
δν0 + kVν = −4Φ0 , (6.12)
3
where we have used 4N0 = δν and 3N1 = Vν .
For CDM, Eqs. (5.50) and (5.51):
δc0 + kVc + 3Φ0 = 0 , Vc0 + HVc − kΨ = 0 . (6.13)
For baryons, Eqs. (5.162) and (5.174):
4τ 0 ργ
δb0 + kVb + 3Φ0 = 0 , Vb0 + HVb − kΨ = (Vb − Vγ ) , (6.14)
3ρb
Finally, we consider the Einstein equations (4.179) and (4.184):
3H (Φ0 − HΨ) + k 2 Φ = 4πGa2 (ρδ + ρb δb + ργ δγ + ρν δν ) , (6.15)
k 2 (Φ + Ψ) = −32πGa2 (ργ Θ2 + ρν N2 ) , (6.16)
where we have expressed the photons and neutrinos anisotropic stresses in terms
of their quadrupole moments.
Now we have to understand when to set our initial conditions. These should be
values for the above quantities Θ` , N` , δc , δb , Vc , Vb , Φ and Ψ at a certain initial
instant. This instant cannot be η = 0, i.e. the Big Bang, evidently, because it is a
singularity, and so it should be some small ηi > 0.
Now, any scale k, being η growing, shall pass from a η 1/k regime called
super-horizon evolution, to a η ∼ 1/k regime called horizon crossing, finally
to a η 1/k called sub-horizon evolution. The mentioned horizon is the
particle one, since recall from Eq. (2.137) that:
Z t Z η
dt0
χp (η) = 0)
= dη 0 = η , (6.17)
0 a(t 0
i.e. the conformal time represents the comoving distance travelled by a photon
since the Big Bang.
Therefore, the initial values for the perturbative quantities are set when kη 1
and are also called primordial modes for the scales of observational interest
today because are deep into the radiation-dominated epoch. For this reason, we
shall frequently use the result a ∝ η.
160
6.2 Evolution equations in the kη 1 limit
All the equations that we shall use here are valid only at early times, in the
radiation-dominated epoch, and on very large scales, i.e. kη 1.
What one does is basically an expansion with respect to kη of the perturbative
variables. Let us see what happens for the case of the photon monopole Eq. (6.4),
just to fix the ideas. Suppose that we have the following expansions:
∞
X ∞
X ∞
X
δγ = δγ(n) (kη)n , Vγ = Vγ(n) (kη)n , Φ= Φ(n) (kη)n . (6.18)
n=0 n=0 n=0
Insert them into Eq. (6.4), and find:

∞ ∞ ∞
X 1 4 X (n) X 1
δγ(n) n(kη)n + k Vγ (kη)n = −4 Φ(n) n(kη)n . (6.19)
n=0
η 3 n=0 n=0
η
This equation can be cast as follows:

∞ ∞ ∞
X 4 X (n) X
δγ(n) n(kη)n + V (kη)n+1
= −4 Φ(n) n(kη)n . (6.20)
n=0
3 n=0 γ n=0
(0) (1)
So, we see that Vγ couples with δγ and Φ(1) in the expansion. In other words,
Vγ is of order (kη)δγ and (kη)Φ, hence is subdominant with respect to δγ and Φ.
This means that, at the lowest order, Eq. (6.4) becomes:
δγ0 = −4Φ0 (6.21)
This procedure is equivalent to neglecting a contribution proportional to k with

respect to a time derivative.
Exercise 6.1. Apply the same reasoning which brought us to Eq. (6.21) to the
neutrinos monopole and to CDM and baryons density contrasts, showing that at
the lowest order of approximation we have:
δν0 = −4Φ0 (6.22)

δc0 = −3Φ0 (6.23)
δb0 = −3Φ0 (6.24)
Integrating the above equations for the monopoles, we get:
δγ = −4Φ + 4Cγ , (6.25)

δν = −4Φ + Cν , (6.26)
δc = −3Φ + Cc , (6.27)
δb = −3Φ + Cb , (6.28)
161
where we have introduced 4 integration constants. Cast the above equations as
follows:
δγ = −4Φ + 4Cγ , (6.29)

δν = δγ + Sν , (6.30)
3
δc = δγ + Sc , (6.31)
4
3
δb = δγ + Sb , (6.32)
4
for a reason that will be clearer later. We have defined:
Sν ≡ Cν − 4Cγ , Sc ≡ Cc − 3Cγ , Sb ≡ Cb − 3Cγ , (6.33)
and these are called density isocurvature modes or entropy modes. Be care-
ful that the C’s and thus the S’s are not actually constants, but functions of k.
Hereafter we will call inappropriately as “constants” those quantities which are
time-independent.
6.2.1 Multipoles in the kη 1 limit

From the hierarchy of the equations for photons we can see that each multipole
Θ` enters the differential equation for Θ`+1 multiplied by k. This means that
Θ`+1 ∼ (kη)Θ` , i.e. each multipole is subdominant with respect the previous one.
This is also true for ΘP ` and for N` and it is an important fact that we shall employ
afterwards.
Applying the limit kη 1 to the photon equations we get, at the dominant
order:
(2` + 1)Θ0` − k`Θ`−1 = τ 0 (2` + 1)Θ` , `>2, (6.34)

4
10Θ02 − kVγ = τ 0 (9Θ2 − ΘP 0 − ΘP 2 ) , (6.35)
3
δγ
Vγ0 − k = kΨ + τ 0 (Vγ − Vb ) . (6.36)
4
Similarly, for the polarization field:
(2` + 1)Θ0P ` − k`ΘP (`−1) = τ 0 (2` + 1)ΘP ` , `>2, (6.37)

0 0
10ΘP 2 − 4kΘP 1 = τ (9ΘP 2 − ΘP 0 − Θ2 ) , (6.38)
3Θ0P 1 − kΘP 0 = 3τ 0 ΘP 1 , (6.39)
2Θ0P 0 = τ 0 (ΘP 0 − ΘP 2 − Θ2 ) . (6.40)
Here comes a very useful approximation, based on the fact that:
− τ 0 = ne σT a ∝ a−2 ∝ η −2 . (6.41)
That is, the optical depth time derivative, which is the photon-electron interaction
rate, diverges for η → 0 as 1/η 2 . This is logical, since the universe becomes denser
the more we go backwards in time and so more interactions take place. This
162
is also called tight-coupling and we will use it as an approximation closer to
recombination in Chapter 10.
But then, in order to prevent the perturbations from diverging, the quantities
multiplying τ 0 must vanish for η → 0, i.e.:
Θ` = ΘP ` = 0 , `>2, (6.42)
9Θ2 − ΘP 0 − ΘP 2 = 0 , (6.43)
9ΘP 2 − ΘP 0 − Θ2 = 0 , (6.44)
Vγ = Vb , ΘP 1 = 0 , (6.45)
ΘP 0 = ΘP 2 + Θ2 . (6.46)
Exercise 6.2. Combine the above equations and show that:
Θ2 = ΘP 2 = ΘP 0 = 0 (6.47)
Hence, the coupling between electrons and photons is so efficient that it washes
out the quadrupole and higher moments of the temperature fluctuations and all
the polarisation moments. Moreover, it tightly couples photons and baryons in a
single fluid with velocity Vγ = Vb .
The kη 1 neutrino equations are the same as the photon ones, but with no
interaction terms:
(2` + 1)N`0 − k`N`−1 = 0 , `≥2, (6.48)

δν
Vν0 − k = kΨ . (6.49)
4
Again, we can see that Vν ∼ (kη)δν , N2 ∼ (kη)2 δν , and so on. Nothing can make
the multipoles of the neutrino temperature to vanish, as Thomson scattering for
photons, therefore we need initial conditions for all of them.
Exercise 6.3. Suppose that:

N` = c` (kη)` , (6.50)
at the lowest order, where c` is some constant (in this instance a true constant, a
number). From Eq. (6.48) show that:
1
c` = c`−1 , (6.51)
2` + 1
and thus:
kη
N` = N`−1 , `≥2. (6.52)
2` + 1
Therefore, once we know the initial condition on N1 , we can determine the

initial conditions for all the subsequent multipoles from the above equation. But,
what about the initial condition on Vν ?
163
Exercise 6.4. Combining Eqs. (6.21), (6.22), (6.36) and (6.49) show that:
Vγ00 = Vν00 . (6.53)
This does not necessarily means that Vγ = Vν , but in general:
Vγ = Vν + qν , (6.54)
where qν is at most a linear function of η and is the neutrino velocity isocur-

vature mode or relative neutrino heat flux.
6.2.2 CDM and baryons velocity equations

Consider now the velocity equations for CDM and baryons. At the lowest order in
kη they become:
ηVc0 + Vc = (kη)Ψ , (6.55)

0 4ητ 0 ργ
ηVb + Vb = (kη)Ψ + (Vb − Vγ ) . (6.56)
3ρb
First of all, ητ 0 ργ /ρb ∝ 1/a2 ∝ 1/η 2 . Thus, in order for Vb not to diverge one has
to have Vγ = Vb , consistently with what we have found earlier for photons. With
this condition holding true, the equation for Vb is similar to the one for Vc :
ηVc0 + Vc = (kη)Ψ , (6.57)

ηVb0 + Vb = (kη)Ψ . (6.58)
From these equations we can conclude that Vc and Vb are subdominant with respect
to Ψ.
6.2.3 The kη 1 limit of the Einstein equations

Noting that:
3H2 3
4πGa2 = = , (6.59)
2ρtot 2ρtot η 2
where ρtot is the total energy density, the generalised Poisson equation can be
written as:
ργ ρν ρc ρb
2ηΦ0 − 2Ψ + 2(kη)2 Φ = δγ + δν + δc + δb . (6.60)
ρtot ρtot ρtot ρtot
Neglect now (kη)2 and define the density fraction Ri ≡ ρi /ρtot for each component,
such that:
Rγ + Rν + Rc + Rb = 1 . (6.61)
Using the solutions found for the monopoles, we arrive at:

0 3 3
2ηΦ − 2Ψ = −4Φ Rγ + Rν + Rc + Rb + 4Rγ Cγ + Rν Cν + Rc Cc + Rb Cb .
4 4
(6.62)
164
Note that
3 3
4 Rγ + Rν + Rc + Rb = 3 + Rγ + Rν , (6.63)
4 4
and since we are deep into the radiation dominated epoch, photons and neutrinos
dominate and thus:
Rγ + Rν ≈ 1 . (6.64)
Exercise 6.5. Show that Rν is constant at early times and its value is:
ρν Neff (7/8)(4/11)4/3
Rν = = = 0.4089 , (6.65)
ρν + ργ 1 + Neff (7/8)(4/11)4/3
the last number coming from choosing Neff = 3.046.
Equation (6.62) can be thus written as follows:
2ηΦ0 + (3 + Rγ + Rν )Φ − 2Ψ = (3 + Rγ + Rν )Cγ + Rν Sν + Rc Sc + Rb Sb
(6.66)
where we have also introduced the density isocurvature modes.
Since Θ2 = 0 for kη 1, the Einstein equation for the anisotropic stress
becomes:
k 2 η 2 (Φ + Ψ) = −12Rν N2 (6.67)
This equation tells us that, consistently, N2 ∼ (kη)2 δν , since Φ ∼ δν . However,
because of that factor (kη)2 , we have to keep track of the time-evolution of N2 ,
which is of the same order. From the neutrino equations that we have derived
earlier it is not difficult to obtain that, upon differentiating Eq. (6.67):
8
2k 2 η(Φ + Ψ) + k 2 η 2 (Φ0 + Ψ0 ) = −12Rν N20 = − Rν kVν . (6.68)
5
Differentiate again, and obtain:
8
2k 2 (Φ + Ψ) + 4k 2 η(Φ0 + Ψ0 ) + k 2 η 2 (Φ00 + Ψ00 ) = − Rν kVν0 =
5
8 2 δν
− Rν k Ψ + . (6.69)
5 4
Now the k 2 can be simplified and differentiating again, one obtains:
8
6(Φ0 + Ψ0 ) + 6η(Φ00 + Ψ00 ) + η 2 (Φ000 + Ψ000 ) = − Rν (Ψ0 − Φ0 ) (6.70)
5
where we used δν0 = −4Φ0 and the fact that Rν is constant when radiation domi-
nates.
We have obtained a closed equation for Φ and Ψ, despite of the fact that it is
of third order. Combining it with the generalised Poisson equation, one can obtain
an equation of fourth order for Φ or Ψ only.
Now we solve the equations found in this section in 5 cases, i.e. when only one
of the 5 constants is different from zero.
165
6.3 The adiabatic primordial mode
The adiabatic mode is defined by Sν = Sc = Sb = qν = 0. Only Cγ 6= 0. With this
condition, we have for the density contrasts:
1 1 1 1
δc = δb = δγ = δν = −Φ + Cγ (6.71)
3 3 4 4
As for the gravitational potentials, Eq. (6.66) becomes:
ηΦ0 + 2Φ − Ψ = 2Cγ , (6.72)
and we have to combine this together with Eq. (6.70) in order to obtain a fourth
order equation for Φ or Ψ. The result is, for Φ:

2 2
η Φ + 12η Φ + 4 9 + Rν ηΦ + 8 3 + Rν Φ0 = 0 .
3 0000 2 000 00
(6.73)
5 5
This equation can be solved exactly, recalling that Rν is constant. The fact that
only derivatives of Φ appear already tells us that Φ = constant is a solution and in-
deed it is the growing mode that we are going to consider. The other 3 independent
solutions of the above equation are:
5 1
√ 32Rν 1
Φ ∝ η − 2 ± 2 1− 5 , Φ∝ . (6.74)
η
Hence, these are decaying modes, the first two even complex if Rν > 5/32, and we
neglect them.
Since Cγ and Φ are constant, then from Eq. (6.72) we deduce that Ψ is also
constant. In order to determine their values in terms of Cγ we have to solve the
system of Eqs. (6.72) and (6.69):
2Φ − Ψ = 2Cγ , (6.75)
4
Φ + Ψ = − Rν (Ψ − Φ + Cγ ) . (6.76)
5
which allows us to establish:

2Cγ (5 + 2Rν ) 10Cγ 2
Φ= , Ψ=− , Φ = −Ψ 1 + Rν (6.77)
15 + 4Rν 15 + 4Rν 5
We shall later show in Chapter 8 how the inflationary mechanism is able to establish
a value for Cγ . Note how the presence of neutrinos via Rν prevents Φ = −Ψ.
The Cγ that we are using here corresponds to the −2C of [Ma and Bertschinger,
1995] and in order to obtain the result of [Bucher et al., 2000] one has to set
Cγ = −1.1
1
There must be a typo in Eq. 28 of [Bucher et al., 2000], since the gravitational potentials
appear to be equal but they cannot be because of the presence of neutrinos and the quadrupole
moment of their distribution.
166
From Eq. (6.71) we can write:
20Cγ
δγ = = −2Ψ . (6.78)
15 + 4Rν
Using Eq. (6.67), the primordial mode of the neutrino quadrupole moment is:
k 2 η 2 (Φ + Ψ) k2η2
N2 = − = Ψ (6.79)
12Rν 30
What about the velocities? We have just seen that Ψ ∼ (kη)0 so if also Vc ∼ Vb ∼
(kη)0 , then Eqs. (6.57) and (6.58) become, at the lowest order:
ηVc0 + Vc = 0 , (6.80)
ηVb0 + Vb = 0 . (6.81)
The solutions are Vc ∝ Vb ∝ 1/η. These solutions are not admissible because
they diverge for η → 0. This means that Vc ∝ Vb ∼ (kη), at the lowest order.
Substituting this ansatz again in Eqs. (6.57) and (6.58) it is easy to find:
kηΨ
Vc = Vb = . (6.82)
2
Since qν = 0, then Vν = Vγ and recalling that Vγ = Vb , we have for the velocities
primordial modes:
kηΨ
Vν = Vγ = Vb = Vc = (6.83)
2
All the above modes depend on the same unique constant Cγ , which is the primor-
dial perturbation, in the sense that its non-vanishing generates the fluctuations for
all the matter components.
Exercise 6.6. Compare the solutions that we have found for the adiabatic mode
with those of Refs. [Ma and Bertschinger, 1995] and [Bucher et al., 2000].
6.3.1 Why “adiabatic”?

Now we are going to understand the reason for the adjective “adiabatic”. From the
thermodynamical relation:
T dS = dU + pdV , (6.84)
we can write:
T dS ρtot + Ptot
= dρtot − dntot . (6.85)
V ntot
Recast this as follows:
T dS ρtot + Ptot
= ρc δc + ργ δγ + ρb δb + ρν δν − (δnc + δnγ + δnb + δnν ) . (6.86)
V ntot
167
Using:
δnc δnb 4 δnγ 4 δnν
δc = , δb = , δγ = , δν = , (6.87)
nc nb 3 nγ 3 nν
we finally have:

T dS ρtot + Ptot 3 ρtot + Ptot 3
= ρc − nc δc − δγ + ρ b − n b δb − δγ
V ntot 4 ntot 4

3 ρtot + Ptot
ρ ν − nν (δν − δγ ) . (6.88)
4 ntot
Therefore, δc /3 = δγ /4 = δν /4 = δb /3, i.e. Sc = Sν = Sb = 0, implies dS = 0
and hence adiabaticity. When one of Sν , S, Sb and qν is different from zero we
have isocurvature modes, respectively divided into isocurvature neutrino den-
sity, isocurvature CDM, isocurvature baryons and isocurvature neutrino
velocity. However, it might be more appropriate to call them entropy pertur-
bations, on the basis of Eq. (6.88).
The name “isocurvature” comes from the fact that when these modes are present
it is possible to have R = 0 or ζ = 0, from Eqs. (4.131) and (4.132). In the
Newtonian gauge, we have:
Rγ δγ + Rν δν + Rc δc + Rb δb
R = Φ + Hvtot , ζ =Φ+ . (6.89)
3 + Rγ + Rν
Some care must be used when computing vtot in the above formula for R since
it is the total one but it is not the sum of the single components’ v’s. This hap-
pens because the SVT decomposition is made not directly on vi but rather on the
perturbed energy-momentum tensor δT 0 i and the latter is, from Eq. (4.73):
δT 0 i (tot) = (ρtot + Ptot )vi (tot) . (6.90)
Hence, vtot is defined through a weighted average with weight ρ + P , i.e.
P
(ρs + Ps )vs
vtot = s , (6.91)
ρtot + Ptot
which, in our 4-components model, gives:
4Rγ vγ + 4Rν vν + 3Rc vc + 3Rb vb
vtot = . (6.92)
3 + Rγ + Rν
Now, using the relation v = −V /k and the condition found in Eq. (6.83), one can
finally write:
4Rγ Vγ + 4Rν Vν + 3Rc Vc + 3Rb Vb Ψ
R=Φ− =Φ− . (6.93)
kη(3 + Rγ + Rν ) 2
One finds the same result for ζ, when substituting Eq. (6.66) in Eq. (6.89) (and
considering the constant mode). Hence, we have that:
Ψ
R=ζ =Φ− = Cγ (6.94)
2
Hence, the Cγ is related to the curvature perturbations and we have also proved
that these are equal and constant on large scales for adiabatic perturbations. We
shall prove this important result again in Sec. 12.4.
168
6.4 The neutrino density isocurvature primordial
mode
Neglect CDM and baryons in Eq. (6.89).
Exercise 6.7. Since Rb,c → 0 at early times, use Eq. (6.30) in order to find:
Rν Sν
ζ = Cγ + . (6.95)
4
If Sν = 0 we recover the adiabatic case. In order to make ζ = 0, we need

4Cγ = −Rν Sν . Therefore, the density contrasts are related as follows:
1 1 1 1 Sν Rν Sν
δ = δb = δγ = δν − = −Φ − (6.96)
3 3 4 4 4 4
In order to determine the gravitational potentials, from Eq. (6.66) we have:
ηΦ0 + 2Φ − Ψ = 0 , (6.97)
and combining it with Eq. (6.70) one finds the very same Eq. (6.73) for Φ that we
obtained in the adiabatic case. Hence, we choose the constant Φ solution, which
allows us to establish:
Rν2 Sν
Ψ = 2Φ , Φ= (6.98)
15 + 4Rν
In order to obtain the same results of [Bucher et al., 2000] one has to set Rν Sν = −1.
Using Eq. (6.67), the initial condition on the neutrino quadrupole moment is:
k 2 η 2 (Φ + Ψ) k2η2Φ
N2 = − =− (6.99)
12Rν 4Rν
Since qν = 0, then Vν = Vγ and the initial conditions on the velocities are again:
kηΨ
Vν = Vγ = Vb = V = = kηΦ (6.100)
2
6.5 The CDM and baryons isocurvature primordial

modes
The treatment for CDM or baryons isocurvature modes is the same, therefore we
present explicitly the CDM case only. Suppose that only Sc 6= 0, then we have for
ζ:
Rc Sc
ζ= . (6.101)
4
169
Since we are deep into the radiation-dominated epoch:
ρc ρc0
Rc ≡ ∼ a = Ωc0 a . (6.102)
ρtot ρtot,0
Then Rc goes to zero and so thus ζ, denoting thus correctly an isocurvature mode.
The density contrasts are related as follows:
δγ δν δb δc − S c
= = = = −Φ (6.103)
4 4 3 3
and all of them vanishes for η → 0, except δc = Sc , which sources all the pertur-
bations. Equation (6.66) tells us that:
2ηΦ0 − 2Ψ + 4Φ = Rc Sc , (6.104)
and using Eq. (6.102), we write:
ηΦ0 − Ψ + 2Φ = Sc Ωc0 η , (6.105)
where we have incorporated the factor 1/2 and the proportionality constant of
a ∝ η into S. This can be done, of course, without losing generality since S is itself
a constant. Now, since the right hand side of the above equation depends linearly
on η, and Φ → 0 for η → 0, the only possibility is that the gravitational potentials
are linearly dependent on the time. Using the ansatz:
Φ = Φ̄η , Ψ = Ψ̄η , (6.106)
Eqs. (6.105) and (6.69) become:
3Φ̄ − Ψ̄ = Sc Ωc0 (6.107)

4
3(Φ̄ + Ψ̄) = − Rν Ψ̄ − Φ̄ . (6.108)
5
Solving this system of two equations, we find the CDM isocurvature modes:
(15 + 4Rν )Sc Ωc0 (4Rν − 15)Sc Ωc0 15 + 4Rν

Φ= η, Ψ= η , Φ = −Ψ (6.109)
4(15 + 2Rν ) 4(15 + 2Rν ) 15 − 4Rν
Again as we expected, the absence of neutrinos causes Φ = −Ψ since there is no

other source of anisotropic stress. Using Eq. (6.67), the primordial mode of the
neutrino quadrupole moment is:
k 2 η 2 (Φ + Ψ) k2η3
N2 = − =− Sc Ωc0 (6.110)
12Rν 6(15 + 2Rν )
Again for the velocities:
kΨ
Vν = Vγ = Vb = Vc = (6.111)
2
The isocurvature modes for baryons have exactly the same form, just change Ωc0
with Ωb0 .
170
6.6 The neutrino velocity isocurvature primordial
mode
These modes are problematic in the conformal Newtonian gauge because they
diverge for η → 0. On the other hand, they are well-defined in the synchronous
gauge [Bucher et al., 2000]. Let us see what is the problem. When we set Cγ =
Sν = Sc = Sb = 0, we can straightforwardly find:
δγ δν δc δb
ζ=0, = = = = −Φ (6.112)
4 4 3 3
and Eq. (6.66) becomes:
ηΦ0 + 2Φ − Ψ = 0 . (6.113)
Here a constant mode seems to be viable, just as in the case of the neutrino density
isocurvature mode. However, we need here also Eq. (6.69), which has the following
form, taking into account δν = −4Φ:
8
2(Φ + Ψ) + 4η(Φ0 + Ψ0 ) + η 2 (Φ00 + Ψ00 ) = − Rν (Ψ − Φ) . (6.114)
5
If we choose the constant mode, we get:
2Φ = Ψ , (6.115)
4
Φ + Ψ = − Rν (Ψ − Φ) . (6.116)
5
From this system we clearly see that the only solution is the trivial one. Therefore,
from Eq. (6.113) one sees that we are left with the only following possibility:
Φ̄ Ψ̄
Φ= Ψ= (6.117)
η η
which is problematic in the limit η → 0.

Substituting this ansatz into the generalised Poisson equation we can easily
determine that:
Φ̄ = Ψ̄ , (6.118)
i.e. the two gravitational potentials are equal. Substituting this result into Eq. (6.68),
one obtains:
4
k Φ̄ = − Rν Vν , (6.119)
5
From the dipole neutrino and the dipole photon equations we can readily establish
that:
Vγ0 = Vν0 = 0 , (6.120)
i.e. Vγ and Vν are constant and their difference is indeed our neutrino velocity
isocurvature mode qν . Thus:
4
k Φ̄ = Rν (Vγ − qν ) . (6.121)
5
171
We can determine Vγ in terms of Φ considering the tight coupling with baryons,
which implies Vγ = Vb and the baryon velocity equation (6.58), which now is:
ηVb0 + Vb = kηΨ = k Φ̄ . (6.122)
The right hand side is a constant and so Vb must be. In particular:
Vb = k Φ̄ = Vγ . (6.123)
This relation, inserted into Eq. (6.121), gives:
4
k Φ̄ = Rν k Φ̄ − qν . (6.124)
5
Finally, we have for the potentials:
4Rν qν 1
Φ=Ψ=− (6.125)
5 − 4Rν kη
and for the velocities:

Rν qν
V = Vb = Vγ = − (6.126)
5 − 4Rν
The neutrino velocity and quadrupole moment are:
k 2 η 2 (Φ + Ψ) 2kηqν
Vν = qν − Vγ N2 = − = (6.127)
12Rν 3(5 − 4Rν )
6.7 Planck constraints on isocurvature modes

In [Ade et al., 2016a] and [Ade et al., 2016c] the isocurvature mode are constrained
through the Planck data. These tests are related also to inflation, since this is
a mechanism of production of primordial fluctuations, as we shall see later in
Chapter 8. In particular, the most successful and simple inflationary models are
those based on a single scalar field and they predict adiabatic initial perturbations.
If other fields are present, then isocurvature modes can be produced. Depending on
the mechanism of production, adiabatic and isocurvature modes can be correlated.
In [Ade et al., 2016a] for example the totally correlated case:
s
|α|
Sm = sgn(α) ζ, (6.128)
1 − |α|
is considered, where
Rb
Sm = Sc + Sb , (6.129)
Rc
is the total matter isocurvature mode. The fact that Sm is proportional to ζ
means that the totally-correlated case is being considered (because of this the
modes are described by the same parameters and hence put in the above form of
proportionality). The constraint found on α is:
α = 0.0003+0.0016
−0.0012 , (6.130)
172
at 95% CL, using also polarisation data. A much more detailed analysis, including
possible correlations, is performed in [Ade et al., 2016c] with the result
|αnon−adi | < 1.9%, 4.0%, 2.9% , (6.131)
for CDM, neutrino density and neutrino velocity isocurvature modes respectively.
In the above equation, αnon−adi is the non-adiabatic contribution to the CMB tem-
perature power spectrum. How the different primordial modes affect the latter will
be shown in Chapter 10.
173
Chapter 7
Stochastic Properties of
Cosmological Perturbations
The more the universe seems comprehensible, the more it also seems
pointless
—Steven Weinberg, The first three minutes
The attentive reader must have noticed that, in fact, we have given no actual
initial conditions on cosmological perturbations in Chapter 6. What we have done
is to give the form of the primordial modes and show how, in the 5 different
cases that we investigated, all scalar perturbations are sourced by a single scalar
potential. For example, in the adiabatic case we showed in Eq. (6.77) how Φ(k) is
related to Cγ (k), which we proved to be equal to ζ(k) and R(k).
Note the following important point: all the evolution equations that we derived
for the perturbations, i.e. all the evolution equations for photons, neutrinos, CDM
and baryons, plus Einstein equations, depend only on the modulus k and not on
k. This means that only the initial conditions depend on the latter. We could
solve the evolution equation for some initial condition in which R(ηi , k) = 1, i.e.
normalised to unity, and then the result will depend only on k: R(η, k). This
is usually called transfer function. Multiplying it by the true initial condition
R(ηi , k) we obtain the correct result. Of course, this reasoning applies to any
perturbation, not only R.
In this chapter we discuss a very important point of cosmology, both from
the viewpoint of theory and especially from the observational one: the stochastic
nature of cosmological perturbations, embedded in the stochastic character of the
initial conditions. This means that R(ηi , k) shall be regarded as a random variable.
This chapter mainly follows [Weinberg, 2008] and [Lyth and Liddle, 2009].
7.1 Stochastic cosmological perturbations and power

spectrum
First of all, what do cosmological perturbations have to do with statistics? After all,
we have derived differential equations describing the evolutions of the perturbative
174
variables and, provided that we are able to solve them, we should determine such
variables exactly with no room for randomness.
Consider the following point: the set of differential equations that we have
found are ODE which only describe the time evolution of the perturbative vari-
ables, with their spatial (or scale) dependences to be provided as initial conditions.
For example, focus on the CDM density contrast δc (η, x), for which the evolution
equations Eqs. (5.50) and (5.51) are the simplest:
δc0 + kVc + 3Φ0 = 0 , Vc0 + HVc − kΨ = 0 . (7.1)
These equations depend only on k, therefore the k-dependence comes from the
initial condition, which indeed we know in the adiabatic case from Eq. (6.78) to
be:
15
δc (k) = ζ(k) . (7.2)
15 + 4Rν
Do we have such initial conditions, i.e. do we know the scale dependence of the
primordial curvature perturbation? Not exactly. According to our current under-
standing, the universe had a quantum origin which we are not yet able to describe
in detail because of the lack of a quantum theory of gravity and also because we
have no experiments penetrating the trans-Planckian scales, allowing us at least
to see what is happening there.
The evolution equations which we had found start to be valid well after the
Planck scale, when gravity is safely classical. However, the initial conditions that
we use are coming from a quantum phase and as such they carry a probabilistic
feature. For example, we shall investigate the inflationary paradigm and we shall
see that it provides a prediction not on ζ(k) itself but rather on its probability
distribution. In particular, if it is a Gaussian one hence all the information is
contained in its variance, which is called power spectrum.
There is also another important point which stresses how useful is to treat
cosmological perturbations as random variables. Observationally, it is impossible
to determine δ(η, x) for any given time because we receive signals from our past
light-cone only. It is impossible thus to recover the initial ζ(x) from what we
observe. Even if we would be able to concoct a theory predicting the initial ζ(x),
we would not be able to fully test it since we cannot determine δ(η, x) completely
from observation.
Moreover, determining exactly δ(η, x) is an awful lot of information. For exam-
ple, it would mean that we are able to predict the position of a certain galaxy at a
certain time. Though this might be interesting to some extent, we are rather more
concerned with averaged quantities, such as the average distance among galaxies,
because these contain information on gravity and the expanding universe.
Hence, promoting (or rather demoting) δ(η, x) to a random variable, we are
then able to infer informations about its expectation value, variance or other high-
order correlators from the observation limited to our past light-cone. This allows
us to make contact with the descriptive statistics that we are able to make on
the distribution of structures in the sky.
175
7.2 Random fields
A function G(x) is a random field if its values are random variables for any x and
distribution functions
F1,2,...,n (g1 , g2 , . . . , gn ) = F [G(x1 ) < g1 , G(x2 ) < g2 , . . . , G(xn ) < gn ] , (7.3)
exist for any n. We denote with capital G the random field, so it has x dependence,
and with g a certain value that it can assume at a given x among all the possible
ones which form the ensemble. The subscripts in g1 , g2 , . . . , gn refer to the different
points x1 , x2 , . . . , xn in which the random field is evaluated.
In particular, at a given point x1 some probability density functional exists:
p1 (g1 )dg1 , (7.4)
describing how probable is for G to assume a certain value g1 in x1 . In terms of

the distribution function, p(g1 ) is defined as:
dF1 (g1 )
p1 (g1 ) = . (7.5)
dg1
Since F is a cumulative probability then F1 (−∞) = 0 and F1 (∞) = 1. In cosmol-
ogy, G(x) is a perturbative quantity, such as the density contrast δ(x).
As usual, an expectation value of the random field is defined via an ensemble
average:
Z
hG(x1 )i ≡ g1 p1 (g1 )dg1 (7.6)
Ω
where Ω denotes the ensemble.

In general, one has:
p1 (g1 ) 6= p2 (g2 ) , (7.7)
i.e. the probability distribution of the values which G may assume in x1 is in
general different from the one of the values which G may assume in x2 . When this
does not happen, i.e. the probability of the realisation is translationally invariant,
the random field is said to be statistically homogeneous. For a statistically
homogeneous random field then Eq. (7.6) is independent of x
Z
hGi ≡ gp(g)dg (7.8)
Ω
Now we can think of more complicated and richer configurations of the random
field G asking for example what is the probability of G(x1 ) and G(x2 ) being g1
and g2 , respectively. This is given by
p12 (g1 , g2 )dg1 dg2 , (7.9)
which can be written again as a derivative of the distribution function F12 . In

general:
p12 (g1 , g2 ) 6= p1 (g1 )p2 (g2 ) , (7.10)
176
unless the realisations are independent, in which case the random process is said
to be Poissonian. The 2-dimensional probability density allows us to define the
2-point correlation function as follows:
Z
ξ(x1 , x2 ) ≡ hG(x1 )G(x2 )i ≡ g1 g2 p12 (g1 , g2 )dg1 dg2 (7.11)
Ω
In a similar fashion, one can define N -point correlation functions as
ξ (N ) (x1 , x2 , . . . , xN ) ≡ hG(x1 )G(x2 ) . . . G(xN )i

Z
≡ g1 g2 . . . gN p12...N (g1 , g2 , . . . , gN )dg1 dg2 . . . gN . (7.12)
Ω
The order of the points matters, i.e. in general p12...N 6= p21...N for example, and
we are using the same ensemble Ω for each point.
Exercise 7.1. If the random field is statistically homogeneous, show that the
2-point correlation function has the following property:
ξ(x1 , x2 ) = ξ(x1 − x2 ) . (7.13)
Given a rotation matrix R, define xR1 = Rx1 . The random field is statistically
isotropic if
p1 (g1 ) = PR1 (gR1 ) ∀R , (7.14)
i.e. the probability of the realisation is rotationally invariant.
Exercise 7.2. If the random field is statistically homogeneous and statistically

isotropic, show that the 2-point correlation function has the following property:
ξ(x1 , x2 ) = ξ(x1 − x2 ) = ξ(r12 ) , (7.15)
where r12 = |x1 − x2 |. That is, the 2-point correlation function depends only on
the distance between the two points.
Following the standard definition that we learn in the first course of statistics,
the ensemble variance of the random field is defined as:
σ 2 (x1 , x2 ) ≡ hG(x1 )G(x2 )i − hG(x1 )ihG(x2 )i (7.16)
Therefore, if the random field is statistically homogeneous and isotropic, we get:
σ 2 (r12 ) = ξ(r12 ) − hGi2 , (7.17)
where recall that hGi2 does not depend on the position. Note that if random
process is Poissonian, i.e. p12 (g1 , g2 ) = p1 (g1 )p2 (g2 ), then the variance is zero,
since:
ξ(r12 ) = hGi2 . (7.18)
177
If G is a variable representing the distribution of galaxies, we do not expect it to be
a Poissonian random variable because gravity turns the odds for the galaxies to be
closer to each other rather than farther. Hence, we attribute deviations from the
Poissonian behaviour to gravity and for this reason it is very important to study
the 2-point correlation function.
In observational cosmology we are able to compute just a spatial average of the
field G(x), since we have just a single realisation, i.e. our universe. So, consider
the spatial average on a given volume V :
Z
1
Ḡ ≡ d3 x G(x) . (7.19)
V V
What can we say about X ≡ Ḡ − hGi? We use this as an estimator of the error
which we are making by exchanging the ensemble average with the spatial one.

Z
1
hXi = d3 x hG(x)i − hGi = 0 , (7.20)
V V
and the variance is:

Z Z
2 1 3
hX i = 2 d x1 d3 x2 hG(x1 )G(x2 )i − hGi2 . (7.21)
V V V
Assuming statistical homogeneity we can then obtain

Z Z
2 1 3
hX i = 2 d x1 d3 x2 ξ(x1 − x2 ) − hGi2 . (7.22)
V V V
We see that for a Poissonian process this variance is vanishing. In fact, if the
galaxies are randomly distributed, any volume is a good sample of the realisation.
The above integral can be written as
Z
2 1
hX i = d3 r ξ(r) − hGi2 . (7.23)
V V
In Sec. 7.8 we show that this variance goes to zero if V → ∞, a fact known as
ergodic theorem and based on a couple of reasonable assumptions. On the other
hand, in practice the volume is finite and thus hX 2 i is in general different from
zero. In cosmology, it is called cosmic variance. Assuming statistical isotropy in
the above integral and using a spherical volume for convenience, we can write:
Z R
2 3
hX i = 3 dr r2 ξ(r) − hGi2 . (7.24)
R 0
This is almost all we need to know about random fields in order to apply them to
cosmology, at least from the theoretical point of view. How to compute correlators
from the observational data is another issue which is not addressed in these notes.
178
Note that in the above description of a random field G we have used the config-
uration space whereas we have mostly used Fourier-transformed quantities repre-
senting our cosmological perturbations. We assume that the FT of a random field
is also a random field, and all the above described properties apply.
Observationally, we are able to probe a realisation in a certain finite volume,
say a box of volume L3 . Hence, the FT is defined here as a Fourier series:
1 X
G(x) = Gn eikn ·x (7.25)
L3 n
where the coefficients are given by:

Z
Gn = d3 x G(x)e−ikn ·x , (7.26)
and the wavenumbers are quantised, because of the periodic boundary conditions
that we must impose, which are equivalent to ask that G vanishes on the sides of
the box:
2π
kn = n, (7.27)
L
where n is a generic vector whose components are integers.
If the side of the box goes to infinity, we recover the usual FT, i.e.
d3 k
Z Z
G(x) = G̃(k)eik·x
, G̃(k) = d3 x G(x)e−ik·x . (7.28)
(2π)3
Note that if G(x) is a real field, then G̃(−k) = G̃∗ (k), because of the following:
Exercise 7.4. Compare:

d3 k d3 k
Z Z
∗
G(x) = 3
G̃(k)eik·x , G (x) = 3
G̃∗ (k)e−ik·x . (7.29)
(2π) (2π)
The two expressions are equal if G(x) is real. Therefore, show that G̃(−k) = G̃∗ (k).
The relation G̃(−k) = G̃∗ (k), is called reality condition and we shall exploit
it in the following Chapters for the predictions on the observed spectra.
7.3 Power spectrum and Gaussian random fields

Consider the 2-point correlation function for the Fourier transform of G(x):
Z Z
0 0
hG̃(k)G̃ (k )i = d x d3 x0 hG(x)G(x0 )ie−ik·x eik ·x .
∗ 0 3
(7.30)
Assume statistical homogeneity. Then:

Z Z
0 0
hG̃(k)G̃ (k )i = d x d3 x0 ξG (x0 − x)e−ik·x eik ·x ,
∗ 0 3
(7.31)
179
and changing variable from x0 to z = x0 − x we get:
Z Z
∗ 0 −i(k−k0 )·x 0
3
hG̃(k)G̃ (k )i = d x e d3 z ξG (z)eik ·z . (7.32)
Now, using the representation of the Dirac delta we get
hG̃(k)G̃∗ (k0 )i = (2π)3 δ (3) (k − k0 )PG (k) (7.33)

with the power spectrum defined as:
Z
PG (k) ≡ d3 x ξG (x)e−ik·x (7.34)
i.e. as the FT of the 2-point correlation function. If also statistical isotropy is

assumed, one has:
Z ∞ Z 1
PG (k) = 2π 2
dr r ξG (r) du e−ikru , (7.35)
0 −1
where u is the cosine of the angle between k and x, which we have used as variable
in the integration. Because of statistical isotropy then the power spectrum depends
only on the modulus of k. Performing the u integration, one gets:
Z ∞
sin(kr)
PG (k) = 4π dr r2 ξG (r) . (7.36)
0 kr
7.3.1 Definition of a Gaussian random field

Now we define a Gaussian random field, from the point of view of its Fourier
components first. FT Gaussian random fields are defined by uncorrelated modes,
i.e. precisely by Eq. (7.33). So, we have proved that Gaussianity implies statistical
homogeneity. Moreover, Gaussian random field are characterised by the fact that
the expectation value and all the correlators of odd order are vanishing, i.e.
hG̃(k)i = hG̃(k1 )G̃(k2 )G̃(k3 )i = · · · = 0 . (7.37)
All the correlators of even order can be written in terms of the second-order corre-
lators, i.e. the power spectrum, which thus contains all the information about the
random field. For example, the 4-point correlator is:
hG̃(k1 )G̃(k2 )G̃(k3 )G̃(k4 )i = hG̃(k1 )G̃(k2 )ihG̃(k3 )G̃(k4 )i
+hG̃(k1 )G̃(k3 )ihG̃(k2 )G̃(k4 )i + hG̃(k1 )G̃(k4 )ihG̃(k2 )G̃(k3 )i . (7.38)
Since all the Fourier modes are uncorrelated, their superposition is Gaussian-
distributed by virtue of the central limit theorem and hence the reason why they
are called Gaussian perturbations. So the g’s are distributed as a Gaussian:
1 2 2
p(g) = √ e−g /2σg , (7.39)
2πσg
where we are assuming a vanishing expectation value and the variance has been
defined in Eq. (7.16). We have used g here instead of G(x) because of statistical
homogeneity.
180
7.3.2 Estimator of the power spectrum and cosmic variance
Theoretically, we shall provide predictions on PG (k) whereas observationally we
determine ξG (r), so it is more appropriate to invert the above FT. So, starting
from Eq. (7.11), we have:
Z 3 0
d3 k
Z
dk 0
ξG (r) = hG(x)G(x + r)i = 3 3
hG̃(k)G̃∗ (k0 )eik·x−ik ·(x+r) i . (7.40)
(2π) (2π)
We have included the Fourier modes in the average in order to investigate the
difference between doing the spatial average or the ensemble one.
Doing the ensemble average. Using Eq. (7.33) in Eq. (7.40), the average must
be carried on the G’s:
d3 k
Z Z
0
ξG (r) = 3
d3 k0 PG (k)δ (3) (k − k0 )eik·x−ik ·(x+r) , (7.41)
(2π)
and gives us:
d3 k
Z
ξG (r) = PG (k)e−ik·r . (7.42)
(2π)3
Note that, since ξ is dimensionless, then PG (k) has dimension of a volume. Working
out the angular integration, we get:
Z 1
dk k 2
Z
ξG (r) = PG (k) du e−ikru , (7.43)
(2π)2 −1
where again u is the cosine of the angle between k and r. Integrating for the last
time, we get:
Z ∞
k 2 PG (k) sin(kr)
ξG (r) = dk (7.44)
0 2π 2 kr
It is customary then to define a dimensionless power spectrum as
k 3 PG (k)
∆2G (k) ≡ (7.45)
2π 2
so that Eq. (7.44) becomes:
Z ∞
dk 2 sin(kr)
ξG (r) = ∆G (k) (7.46)
0 k kr
Doing the spatial average. Suppose that none of the quantities in the inte-
grand of Eq. (7.40) is a stochastic variable and that h. . . i is a spatial average, which
is the one relevant from the point of view of observation. We must rewrite ξG (r)
as follows:
Z
ˆ 1 X 1 2π
ξG (r) = d3 x 2
Gn G∗m eikn ·x−ikm ·(x+r) , Gn = G̃(kn ) , kn = n,
V V n,m
V L
(7.47)
181
where we have used the Fourier series expansion because we are in a finite volume,
that we have chosen to be a cube of side L and hence V = L3 . This volume can be
thought as the survey volume. Note that the summation over n and m is intended
as a sum over the three components of n and m since we cannot use statistical
isotropy now. The spatial integration gives:
Z
1
d3 x ei(kn −km )·x = δnm . (7.48)
L3 V
Exercise 7.5. Show in one dimension, for simplicity, that:

1 L/2 2 sin[(kn − km )L/2]
Z
dx ei(kn −km )x = . (7.49)
L −L/2 (kn − km )L
Since k is quantised, the difference kn −km is always a multiple of 2π/L, thereby
making the sine to vanish. The only exception is when the difference is zero, i.e.
for n = m.
So, we have for the correlation function:

X 1
ξˆG (r) = 2
|Gn |2 e−ikn ·r , (7.50)
n
V
from which we infer the power spectrum as:

|Gn |2
Pn ≡ P (kn ) = . (7.51)
V
Since the survey volume is finite we saw that the smallest wavenumber that we
can build is 2π/L and the smallest cell in the wavenumber space has thus volume
(2π/L)3 .
Exercise 7.6. Prove that the number of independent modes between k and k + dk
is:1 3
2 L 1 dk
Nk = 4πk dk = 2 (kL)3 . (7.52)
2π 2π k
From this number of independent modes, we then know that the cosmic variance
of the power spectrum is:
σP (k) 1 1
'√ ' 1/2 , (7.53)
P (k) Nk rk (kL)3/2
where we have defined rk ≡ dk/k as the resolution of the survey. If we assume
infinite resolution then we have to derive again the above result since the dimension
is reduced by one.
1
This calculation is essentially identical to the one which leads to the definition of the Fermi
momentum of a gas of fermions, which might be more familiar to the reader. See e.g. [Huang,
1987]
182
Exercise 7.7. Show that the number of independent modes on the sphere of radius
k in the wavenumber space are:
2
2 L (kL)2
Nk = 4πk = . (7.54)
2π π
In this case, one gets:

σP (k) 1 1
'√ ' , (7.55)
P (k) Nk kL
for the cosmic variance, which is the result obtained in [Weinberg, 2008].
These formulae for the cosmic variance tell us that the latter is negligible if the
scale probed is much smaller that the dimension L of the survey, i.e. kL 1.
7.4 Non-Gaussian perturbations

For non-Gaussian perturbations the odd-order correlators are non-vanishing. For
example:
hG̃(k1 )G̃(k2 )G̃(k3 )i = (2π)3 δ (3) (k1 + k2 + k3 )BG (k1 , k2 , k3 ) , (7.56)
where BG (k1 , k2 , k3 ) is called bispectrum and can be rewritten as:
BG (k1 , k2 , k3 ) = BG (k1 , k2 , k3 ) [PG (k1 , k2 ) + PG (k2 , k3 ) + PG (k1 , k3 )] , (7.57)
i.e. in terms of a purely FT of a 3-point correlation function, called reduced
bispectrum, multiplied by all the possible combinations of the FT transforms of
the 2-point correlation functions (i.e. the PS). This formula can be obtained using
the following expansion:
G(x) = GG (x) + fN L G2G (x) − hG2G (x)i + . . . ,

(7.58)
where hG2G (x)i ≡ σG2
. This expansion is called local type non-Gaussianity
and is based on the fact that the square of a Gaussian random field GG (x) is not
Gaussian. The amount of non-Gaussianity is indicated by the free parameter fN L ,
which Planck has constrained to be [Ade et al., 2016b]:
fN L = 2.5 ± 5.7 (68% CL, statistical) (7.59)
The huge relative error shows how difficult is to extract this kind of information
and at the same time how Gaussianity (fN L = 0) is fully consistent with data. Note
that we are talking here of primordial non-Gaussianity. In the structure formation
process, non-Gaussianity naturally arises in the non-linear regime of evolution.
7.5 Matter power spectrum, transfer function and

stochastic initial conditions
All the formula derived in the previous section can be in principle directly adapted
to the matter density contrast field δ(η, k),2 but how do we deal with the time
2
We drop here the subscript of δ in order to keep a light notation. The treatment can be in
principle applied to any δ, but the most interested case is that for matter, i.e. CDM plus baryons.
183
dependence of the latter? In principle, one just needs to substitute G(x) for δ(η, x),
where we are leaving on purpose the time-dependence. We then have:
Z ∞
dk 2 sin kr k 3 Pδ (η, k)
ξδ (η, r) = ∆δ (η, r) , ∆2δ (η, r) = , (7.60)
0 k kr 2π 2
and, from Eq. (7.33) we have
hδ(η, k)δ ∗ (η, k0 )i = (2π)3 δ (3) (k − k0 )Pδ (η, k) . (7.61)
In the above equation we have assumed Gaussianity in the initial conditions. In

the linear case the modes evolve independently and hence Gaussianity is preserved,
for this reason the above equation is valid even if it is considered at any time η.
As we have anticipated, the stochastic character of δ(η, k) is due just to its
initial value δ(k). Hence, it is customary to write:
Pδ (η, k) = T 2 (η, k)Pδ (k) (7.62)
where T (η, k) is the transfer function and Pδ (k) is the primordial power spec-
trum, i.e.
hδ(k)δ ∗ (k0 )i = (2π)3 δ (3) (k − k0 )Pδ (k) , (7.63)
where δ(k) is the initial condition of δ. The primordial power spectrum is a predic-
tion of inflation that we shall compute in Chapter 8 whereas the transfer function
is the solution of the evolution equations that we have found in Chapters 4 and 5,
with the primordial modes found in Chapter 6 as initial conditions. These are char-
acterised by constants multiplied by a function of k, which becomes our stochastic
initial condition. We call this α(k) for scalar modes and β(k, λ) for tensor modes,
borrowing the notation used in [Weinberg, 2008]. Also, in the rest of these notes
we shall drop the explicit use of the transfer function T (η, k), by adopting the
renormalisation:
δ(η, k) → α(k)δ(η, k) , Θ(η, k, p̂) → α(k)Θ(η, k, p̂) , (7.64)
e.g. for the density contrast and for Θ, when it shall be necessary.
Let us see a practical example of how to calculate T 2 (η, k) and its normalised
initial conditions. Consider adiabaticity. Then, all perturbations are sourced by
α(k) = ζ(k) = R(k) = Cγ (k). If we normalise the initial condition Cγ (k) = 1,
then we have from Eq. (6.71) and the subsequent ones:

1 1 1 1 10 2
δc = δb = δγ = δν = −Φ + 1 , Ψ = − , Φ = −Ψ 1 + Rν .
3 3 4 4 15 + 4Rν 5
(7.65)
With this collection of initial conditions we can compute the transfer functions for
each variable, square them and propagate the primordial power spectrum forward
to any time. Hence, for matter, we can write:
2
2 2 15
Pδ (η, k) = T (η, k)Pδ (k) = T (η, k) Pζ (k) . (7.66)
15 + 4Rν
184
Finally, a comment on the following. Note that in Eq. (7.63) we have the product
of two δ’s in hδ(k)δ ∗ (k0 )i. Is not this of second order and thus negligible? Locally
it is, but the ensemble average can be seen as a spatial average, cf. Eq. (7.47), and
thus we have a vanishing quantity integrated over an infinite volume giving a finite
result.
7.6 CMB power spectra

CMB photons do not come from a localised source but from the whole celestial
sphere and from a finite distance, the last scattering surface, which corresponds to
a redshift of z ∼ 1100. Hence, one usually approaches the CMB power spectrum
differently from the matter one (this coming from the observation of galaxies).
The relative fluctuation in the temperature δT /T is the same Θ that we defined
in Eq. (5.80) if we suppose that the perturbed distribution function Fγ does not
depend on the photon energy p. Since this assumption is justified by the fact that
Thomson scattering with electrons is the relevant physical process generating Θ
and in Thomson scattering the energy of the photon is unchanged, we shall use
δT /T = Θ.
Note that Θ = Θ(η, x, p̂), but from the point of view of observation, we are just
interested in temperature fluctuations here and today, i.e. for η = η0 and x = x0 ,
where the latter is the position of our laboratory (i.e. Earth). Moreover, photons
travel on the light-cone (they come from our past light-cone) and from the fixed
distance to the last-scattering surface. Hence:
x∗ = −(η0 − η∗ )p̂ = −r∗ p̂ , (7.67)
where η0 is the present conformal time, η∗ the recombination one and r∗ ≡ η0 −η∗ is
the comoving distance to recombination. Only photons satisfying this relation can
be detected because their momentum is in the “correct” direction (i.e. towards us)
and they free stream only after recombination. Therefore, of the 6 variables upon
which Θ(η, x, p̂) in principle depends only 2 are observationally relevant, since we
observe
Θ(η0 , x0 , n̂) , (7.68)
where n̂ = −p̂ are indeed coordinates on the celestial sphere. The direction to the
photon towards us is p̂, so our line of sight unit vector has the opposite sign.
Now we omit the (η0 , x0 ) dependence of Θ. It is customary to expand a func-
tion on the sphere in spherical harmonics Y`m (θ, φ). Hence, for our temperature
fluctuation we have:3
∞ X
X `
Θ(n̂) = aT,`m Y`m (θ, φ) (7.69)
`=0 m=−`
3
The expansion is usually considered for ∆T , the temperature fluctuation. In this case, the
aT,`m ’s carry dimensions of temperature. The only difference between the coefficients of the two
expansions is a factor T0 , i.e. the temperature of the CMB.
185
where we have written n̂ = (sin θ cos φ, sin θ sin φ, cos θ) (i.e. employed spheri-
cal coordinates) and Y`m (θ, φ) are the spherical harmonics. Recall that these are
revised in some detail in Sec. 12.5.
In these notes we adopt a normalisation which guarantees orthormality, i.e.
Z
0∗
d2 n̂ Y`m (n̂)Y`m
0 (n̂) = δ``0 δmm0 . (7.70)
Note that the spherical harmonics Y`m are in general complex4 and so the a`m must
also be such, in order for Θ(θ, φ) to be real.
Exercise 7.8. According to our conventions established in Section 12.5 show that
Y`m∗ (n̂) = Y`−m (n̂) , (7.71)
and hence, in order for Θ(n̂) to be real, show that:
a∗T,`m = aT,`,−m , (7.72)
which is a reality condition for the coefficients aT,`m .
Note that the expansion (7.69) can be done at each point of spacetime, but
then the angular dependence changes because the (θ, φ) measured from one point
in space correspond to different celestial coordinates as seen from another spot in
space.
The expansion of Eq. (7.69) can be inverted as follows:
Z
aT,`m = d2 n̂ Y`∗m (n̂)Θ(n̂) , (7.73)
and the a`m ’s are promoted to stochastic variable, just as Θ is. Again, it is the initial
conditions for cosmological perturbations which are actual stochastic variables for
which inflation predicts a power spectrum, so we shall introduce a transfer function.
For Gaussian perturbations, the expectation value and variance of the aT,`m ’s
are:
haT,`m i = 0 haT,`m a∗T,`0 m0 i = δ``0 δmm0 CT T,` (7.74)
and the CT T,` = h|aT,`m |2 i form the CMB power spectrum. Introducing the Fourier
transform, we have:
Z 3 0
d3 k
Z Z Z
d k i(k−k0 )·x0
CT T,` = h e d n̂ Y` (n̂)Θ(k, n̂) d2 n̂0 Y`m (n̂0 )Θ∗ (k0 , n̂0 )i .
2 ∗m
(2π)3 (2π)3
(7.75)
The ensemble average acts on the temperature fluctuations, which we renormalise
to some primordial mode α(k):
hΘ(k, n̂)Θ∗ (k0 , n̂0 )i = hα(k)α∗ (k0 )iΘ(k, n̂)Θ∗ (k 0 , n̂0 ) . (7.76)
4
The notation Y`m usually denotes spherical harmonics in the real form, obtained by combining
Y`m in a suitable way in order to trade the imaginary exponential exp(imφ) for a sine or cosine
function of mφ.
186
We should have perhaps written TΘ (k, n̂), stressing the introduction of a transfer
function, but we have decided to keep a simple notation.
For scalar perturbations we saw that, assuming axially symmetric initial condi-
tions, the dependence is only on k̂ · p̂ = µ = −k̂ · n̂ and we have used the multipole
expansion in Eq. (5.113), which can be easily inverted as follows:
X (S)
Θ(S) (k, µ) = (−i)` (2` + 1)P` (µ)Θ` (k) . (7.77)
`
(S)
The evolution of the multipoles Θ` (η, k) until η0 (today) is given by the hierarchy
of eqs. (5.119)-(5.122) which in fact depend only on the modulus k.
Assuming adiabatic Gaussian perturbations, i.e. α(k) = R(k) and
hR(k)R∗ (k0 )i = (2π)3 δ (3) (k − k0 )PR (k) , (7.78)
we have from Eq. (7.75):
d3 k
Z Z Z
(S)
S
CT T,` = 3
|Θ` (η0 , k)| PR (k) d n̂ Y` (n̂) d2 n̂0 Y`m (n̂0 )
2 2 m∗
(2π)
0
X X 00
(−i)` (2`0 + 1)P`0 (µ) i` (2`00 + 1)P`00 (µ) . (7.79)
`0 `00
Note that a similar result can be obtained from Eq. (7.75) if we perform as spatial
average, i.e. an average over x0 , as we saw in the case of the three-dimensional
power spectrum. In this case the square modulus |Θ(k, n̂)|2 appears, as our esti-
mator of the angular power spectrum.
Recall that the Legendre polynomial is proportional to Y`0 and thus from the
orthogonality of the spherical harmonics and the addition theorem we get:
Z Z
4π
2
d n̂ P` (µ)Y` (n̂) = d2 n̂ P`0 (−k̂ · n̂)Y`m∗ (n̂) =
0
m∗
Y`m∗ (k̂)δ``0 . (7.80)
2` + 1
Hence we have:
Z Z
S 2 2 (S) 2 dk (S)
CT T,` = dk k |Θ` (η0 , k)| PR (k) = 4π |Θ` (η0 , k)|2 ∆2R (k) (7.81)
π k
where we have used: Z

d2 k̂ |Y`m (k̂)|2 = 1 . (7.82)
We shall discuss the tensor contribution to the CMB TT spectrum in Chapter 10.
Not only temperature anisotropies are measurable in the CMB sky, but also
polarisation ones. Hence, we have more correlation functions and spectra than
the temperature-temperature (TT) one. As we discussed in Chapter 5, Thomson
scattering provides only linear polarisation, which is described by the two Stokes
parameters Q and U . It is customary to use the combinations Q ± iU since these
have helicity 2, i.e. under a rotation of an angle θ in the polarisation plane they
transform as:
(Q ± iU ) → e±2iθ (Q ± iU ) . (7.83)
187
Similar quantities were defined for gravitational waves. Now, Q±iU represent fields
of helicity 2 on the sphere and as such they can be expanded in spin 2-weighted
spherical harmonics:
∞ X
X `
(Q + iU )(n̂) = aP,`m 2 Y`m (n̂) , (7.84)
`=2 m=−`
∞ X
X `
(Q − iU )(n̂) = a∗P,`m 2 Y`m∗ (n̂) . (7.85)
`=2 m=−`
The spin-weighted spherical harmonics do not satisfy simple reality conditions as

the spin-0 ones (the usual spherical harmonics) do, i.e. Y`m∗ = Y`−m , and therefore
it is convenient to use the following combinations of aP,`m :
aE,`m = −(aP,`m + a∗P,`−m )/2 , aB,`m = i(aP,`m − a∗P,`−m )/2 . (7.86)
These satisfy reality conditions and have the following properties under spatial
inversion: aE,`m gains a factor (−1)` , as the a`m of the temperature-temperature
correlation, whereas aB,`m gain an extra −1 factor.
If we assume thus stochastic initial conditions which are invariant under spatial
inversion, the only four spectra that we can build from CMB temperature and
polarisation measurement are thus:
haT,`m a∗T,`0 m0 i = δ``0 δmm0 CT T,` ha∗T,`m aE,`0 m0 i = δ``0 δmm0 CT E,` (7.87)
haE,`m a∗E,`0 m0 i = δ``0 δmm0 CEE,` haB,`m a∗B,`0 m0 i = δ``0 δmm0 CBB,` (7.88)
Note that there is no dependence on m in the power spectra. This is again due
to statistical isotropy. Violation of the latter is related to the so-called CMB
anomalies. These are unexpected features in the CMB sky, such as the axis of
evil and the large cold spot in the southern hemisphere [Schwarz et al., 2016]. The
explanation of these anomalies is an open issue of modern cosmology.
Statistical isotropy is motivated by the Copernican principle and predicted by
many inflationary theories.5
7.6.1 Cosmic Variance of angular power spectra

In this section we compute the cosmic variance of the angular power spectra. The
result can be applied to any field defined on the sphere and for Gaussian pertur-
bations, which are characterised by a correlation function which depends only on
the angular separation and thus the power spectrum is C` , depending only on `.
In order to compute the cosmic variance, let us do first an estimate. Our objec-
tive is to determine C` observationally, in order to compare it with our theoretical
prediction. In order to do that, we probe ha`m a∗`m i for different values of m, which
5
According to the Copernican principle we do not occupy a special position in the universe.
Therefore, its average properties that we are able to determine should be the same as those
determined by any other observer.
188
are 2` + 1. Hence, we have 2` + 1 possible sampling of C` for any given ` and a
sampling error √
∆C` ∝ 2` + 1 . (7.89)
The relative error associated to the sampling, i.e. the cosmic variance, is thus:
∆C` 1
σC` = ∝√ . (7.90)
2` + 1 2` + 1
Now, let us make a more precise calculation following [Weinberg, 2008]. Consider
the temperature-temperature correlation function, as an example:
0
X X
hΘ(n̂)Θ(n̂0 )i = ha`m a`0 m0 iY`m (n̂)Y`m 0
0 (n̂ ) = C` Y`m (n̂)Y`−m (n̂0 ) , (7.91)
`m`0 m0 `m
where cos θ ≡ n̂·n̂0 and we have performed the ensemble average assuming Gaussian
perturbations. The −m of the second spherical harmonics comes from the reality
condition by which a∗`m = a`,−m .
Using the addition theorem of the spherical harmonics, we can do the sum over
m in the above formula and obtain:
X 2` + 1 X 2` + 1
C(θ) ≡ hΘ(n̂)Θ(n̂0 )i = C` P` (n̂ · n̂0 ) = C` P` (cos θ) . (7.92)
`
4π `
4π
So we have explicitly proved that for Gaussian perturbations the correlation func-
tion depends only on the angle between the two directions. Inverting this relation
using the orthonormality of the Legendre polynomials, we get:
Z
1
C` = d2 n̂ d2 n̂0 P` (n̂ · n̂0 )hΘ(n̂)Θ(n̂0 )i , (7.93)
4π
which is Eq. (7.75), without introducing the FT, which we do not need here.
The integral is of course on the whole sky. These are the theoretical C` ’s, and
the average is the ensemble one. Observationally, the only average that we can do
is the angular one, i.e.
Z
1
obs
C` = d2 n̂ d2 n̂0 P` (n̂ · n̂0 )Θ(n̂)Θ(n̂0 ) . (7.94)
4π
Exercise 7.9. Show that, substituting the spherical harmonics expansions, we

have:
Z
1 X 0
obs
C` = d2 n̂ d2 n̂0 P` (n̂ · n̂0 )aLM YLM (n̂)aL0 M 0 YLM0 (n̂0 )
4π LM L0 M 0
1 X
= a`m a`,−m . (7.95)
2` + 1 m
189
Here it appears more clearly that for each value of √ ` we have 2` + 1 possible
realisations and thus we expect that the counting error is 2` + 1. We can calculate
this exactly, and the cosmic variance is the following ensemble average:
* +
obs 2
C ` − C ` hC obs i 1 2
σC2 ` = =1−2 ` + 2 hCòbs i . (7.96)
C` C` C`
Of course, the ensemble average of hCòbs i is C` , just look at the integral which
defines it, and the ensemble average of C` is C` , since it is already averaged.
Therefore, let us focus on
2 1 X
hCòbs i = ha`m a`,−m a`m0 a`,−m0 i . (7.97)
(2` + 1)2 mm0
Since the perturbations are Gaussian, this 4-point correlator can be split into the
following sum, by Eq. (7.38):
ha`m a`,−m a`m0 a`,−m0 i = ha`m a`,−m iha`m0 a`,−m0 i + ha`m a`m0 iha`,−m a`,−m0 i
+ha`m a`,−m0 iha`m0 a`,−m i . (7.98)
Using Eq. (7.74), we finally get:
2
σC2 ` = (7.99)
2` + 1
as expected.
7.7 Power spectrum for tensor perturbations

When we compute the power spectrum for GW, using the decomposition of helic-
ities of Eq. (4.211), we get:
X
hhTij (η, k)hTlm∗ (η, k0 )i = eij (k̂, λ)e∗lm (k̂ 0 , λ0 )hh(η, k, λ)h∗ (η, k0 , λ0 )i , (7.100)
λ,λ0 =±2
the stochastic behaviour being carried by the initial condition on h(η, k, λ), which
we then renormalise as:
h(η, k, λ) = β(k, λ)h(η, k) (7.101)
Since the evolution equation for tensor perturbations depends only on k and does
not depend on the helicity λ, we incorporate the latter in the stochastic initial
condition.
Asking translational invariance (Gaussian perturbations) and rotational invari-
ance, we have:
hβ(k, λ)β ∗ (k0 , λ0 )i = (2π)3 Ph (k)δλλ0 δ (3) (k − k0 ) . (7.102)
190
Therefore, in the ensemble average of hhTij (η, k)hTlm∗ (η, k0 )i it appears the sum over
the helicities:
X
Πij,lm (k̂) ≡ eij (k̂, λ)e∗lm (k̂, λ) . (7.103)
λ=±2
We can make this sum explicit as follows. Consider the fact that Πij,lm (k̂) is a
tensor which depends on k̂. In order to have the correct combination ij, lm of
indices it must be a combination of the only independent tensors that are around,
i.e. δij and k̂m . In order to have the correct combination ij, lm of indices, we
can combine two δij , one δij and two k̂m or four k̂m . So we can put forward the
following ansatz:
Πij,lm (k̂) = A(δil δjm + δim δjl ) + Bδij δlm

+Cδij k̂l k̂m + Dδlm k̂i k̂j + E(δil k̂j k̂m + δjm k̂i k̂l + δim k̂j k̂l + δjl k̂i k̂m )
+F k̂i k̂j k̂l k̂m , (7.104)
where we have grouped together all the terms which must be symmetrised in order
to respect the symmetry of Πij,lm (k̂), which is symmetric in ij and in lm separately.
The coefficient introduced above are, in principle, complex.
Exercise 7.10. Contract the above ansatz for Πij,lm (k̂), with k̂ i , with k̂ l and with
δ ij . Since the result must be zero, show that have the following conditions on the
coefficients:
A = −B = C = D = −E = F . (7.105)
Thus, we can write:
Πij,lm (k̂) = A(δil δjm + δim δjl − δij δlm

+δij k̂l k̂m + δlm k̂i k̂j − δil k̂j k̂m − δjm k̂i k̂l − δim k̂j k̂l − δjl k̂i k̂m
+k̂i k̂j k̂l k̂m ) . (7.106)
Now, since Π∗ij,lm (k̂) = Πlm,ij (k̂), i.e. complex conjugation amounts to exchange
the couple ij with lm, it is not difficult to conclude that A is a real number. In
order to determine which number, consider the normalisation used in Eq. (4.206).
If we set k̂ = ẑ in our improved ansatz above and choose i = j = l = m = 1, we
get: X
Π11,11 (k̂) = e11 (k̂, λ)e∗11 (k̂, λ) = 1 = A . (7.107)
λ=±2
Therefore, we can conclude that:
Πij,lm (k̂) = δil δjm + δim δjl − δij δlm

+δij k̂l k̂m + δlm k̂i k̂j − δil k̂j k̂m − δjm k̂i k̂l − δim k̂j k̂l − δjl k̂i k̂m
+k̂i k̂j k̂l k̂m . (7.108)
191
7.8 Ergodic theorem
Following [Weinberg, 2008], here we briefly presents the ergodic theorem, which
allows us to exchange ensemble and spatial average under certain conditions.
Consider a random variable G(x) in a D-dimensional Euclidean space and as-
sume statistical homogeneity, i.e.
hG(x1 )G(x2 ) . . . G(xn )i = hG(x1 + z)G(x2 + z) . . . G(xn + z)i , (7.109)
i.e. when we calculate the correlator, the result does not change upon translation
of the fields.
Also, let us make the following reasonable assumption. When we calculate a
correlator where a certain amount of points are very far from another set, then the
correlator breaks. That is:
hG(x1 + u)G(x2 + u) . . . G(y1 − u)G(y2 − u) . . . i

→|u|→∞ hG(x1 + u)G(x2 + u) . . . ihG(y1 − u)G(y2 − u) . . . i
= hG(x1 )G(x2 ) . . . ihG(y1 )G(y2 ) . . . i ,
i.e., if we think of a simple 2-point correlation function, the points become uncor-
related on a large distance and so it is like having independent ensembles, which
is the key in order to exchange the spatial average with the ensemble one. In the
last line of the above equation we have used statistical homogeneity.
Now define what in some sense is closely related to the cosmic variance:
σR2 (x1 , x2 , . . . ) ≡
*Z 2 +
dD zNR (z)G(x1 + z)G(x2 + z) . . . − hG(x1 )G(x2 ) . . . i , (7.110)
where the integral is a spatial average about a point z0 , recall that the dimension
of the space is D, and the window function or filter is for example:
1 2 2
NR (z) ≡ √ D
e−|z−z0 | /R , (7.111)
( πR)
R D
which, as it can be checked, is normalised to unity, i.e. d z NR (z) = 1, the
integration being over all the space. This function is basically a filter and its form
is not important as long it is almost constant for |z − z0 |2 < R2 and decays rapidly
for |z − z0 |2 > R2 . It introduces a scale R which is the limiting one until which we
are able to do a spatial average and so we want to check how this variance depends
on R. We can think of R as the size of a survey, for example.
The ergodic theorem states that
σR2 (x1 , x2 , . . . ) ∼ R−D for R → ∞ . (7.112)
In order to prove it, expand the square in the variance:

Z Z
2
σR = d zNR (z) dD wNR (w)(G(x1 + z)G(x2 + z) · · · − hG(x1 )G(x2 ) . . . i)
D
× (G(x1 + w)G(x2 + w) · · · − hG(x1 )G(x2 ) . . . i)i . (7.113)
192
R
We can incorporate the ensemble average into the integral because of dD z NR (z) =
1. Now apply the average and use statistic homogeneity:
Z Z
2
σR = d zNR (z) dD wNR (w)(hG(x1 + z)G(x2 + z) . . . G(x1 + w)G(x2 + w) . . . i
D
− hG(x1 )G(x2 ) . . . i2 ) . (7.114)
Now change integration variables:
u = (z − w)/2 , v = (z + w)/2 , (7.115)
and, remembering to take into account the Jacobian of the transformation, which
is 2D , and the expression we have adopted for the window function:
D Z Z
2 2 2 2 2
σR2
= 2
d v dD u e−|u+v−z0 | /R −|v−u−z0 | /R
D
πR
(hG(x1 + u + v)G(x2 + u + v) . . . G(x1 + v − u)G(x2 + v − u) . . . i −
hG(x1 )G(x2 ) . . . i2 ) . (7.116)
Now, the v in the fields is always summed to the coordinate so we can use again
statistic homogeneity and eliminate v from the average, getting:
D Z Z
2 2 2 2 2
σR2
= 2
d v dD u e−2|u| /R e−|v−z0 | /R
D
πR
(hG(x1 + u)G(x2 + u) . . . G(x1 − u)G(x2 − u) . . . i − hG(x1 )G(x2 ) . . . i2 ) (7.117)
.
The integration in v can be immediately performed, being a Gaussian integral

equal to (π/2)D/2 RD :
D/2 Z
2 2 2
= σR2 2
dD u e−2|u| /R
πR
(hG(x1 + u)G(x2 + u) . . . G(x1 − u)G(x2 − u) . . . i − hG(x1 )G(x2 ) . . . i2 ) . (7.118)
Note that z0 is no more present, as it should be because of statistical homogene-

ity. The term between parenthesis goes to zero for large |u|, on the basis of the
hypothesis made at the beginning of our discussion, therefore the |u| integral is
finite and above all, its value does not depend on R because the difference among
the ensemble averages does not depend on our choice of R. Then, we can conclude
that:
σR = O(R−D/2 ) (7.119)
193
Chapter 8
Inflation
And if inflation is wrong, then God missed a good trick. But, of

course, we’ve come across a lot of other good tricks that nature has
decided not to use
—Jim Peebles, interview at Princeton (1994)
We dedicate this chapter to inflation, a model of the primordial universe in

which an almost constant H provides a scale factor a growing exponentially with
the cosmic time. Inflation is able to solve some puzzles related to background cos-
mology and also to provide a testable prediction of the power spectrum of primor-
dial fluctuations. Among the first pioneering works on inflation there are [Starobin-
sky, 1979], [Guth, 1981], [Linde, 1982] and [Albrecht and Steinhardt, 1982].
An alternative to inflation which has risen some interest is the so-called bounc-
ing cosmology in which the universe is eternal and whose evolution is charac-
terised by a contracting phase followed by an expanding one, through a bouncing
phase whose physical details are governed by quantum cosmology. See [Novello
and Bergliaffa, 2008] and [Brandenberger and Peter, 2017] for recent reviews on
the subject which, though interesting, we shall not tackle here.
8.1 The flatness problem

Recall the definition of the curvature density parameter that we gave in Eq. (2.69):
K
ΩK ≡ − , (8.1)
H 2 a2
and of its observed value coming from the latest Planck data [Ade et al., 2016a]:
ΩK0 = 0.0008+0.0040
−0.0039 . (8.2)
at the 95% confidence level. It is with a lot of confidence, much larger that 95%,
that we can conclude that:
|ΩK0 | < 1 . (8.3)
But then notice the following:
K K H02 a20 H02 H02
|ΩK | = − = − = |ΩK0 | < , (8.4)
H 2 a2 H 2 a2 H 2 a2 H 2 a2 H 2 a2
194
where we have used a0 = 1. Now, how does the function on the right hand side
scale? When a → 0, since radiation dominates, we have that:
H02
∼ a2 → 0 . (8.5)
H 2 a2
That is, the more in the past we go the closer to zero ΩK gets. And closer means
really closer. Indeed:
Exercise 8.1. Show that at the radiation-matter equivalence, i.e. at z = 104 ,

ΩK < 10−4 . Show that at BBN, i.e. at z = 1010 , ΩK < 10−16 . Then show that the
Planck time corresponds roughly to z = 1032 , and then there we have ΩK < 10−60 .
The Planck time is the farthest we can extrapolate our classical theory and
10−60 is something really close to zero. What is the problem here? The problem
is that if by some reason ΩK ∼ 10−59 at the Planck era, then today ΩK0 would
be ten times larger and in complete disagreement with observation. In order to
match what we observe today, ΩK has to be determined at the Planck scale with
a precision of 60 significative digits! This is an example of fine-tuning.
Fine tunings might not be actual problems for theories, in the same sense
that falsified predictions rule out wrong theories, but denote something ad hoc,
unnatural and therefore something that we would like to explain in another way if
possible.
As for the flatness problem the very simple idea put forward in [Guth, 1981]
is the following. Regard Eq. (8.1). If H is constant, then ΩK ∝ 1/a2 . Thus, if
before the radiation-dominated era the very early universe experienced a phase in
which H is constant, this could explain why ΩK is so small to begin with, without
recurring to any fine-tuning. This primordial phase with H constant, or almost
constant, is inflation.
Inflation must take place in a primordial phase of evolution, prior to radiation
domination, in order not to spoil the successful predictions of the Hot Big Bang
model.
Exercise 8.2. Show that a constant H implies that:
H = constant ⇒ a ∝ eHt , (8.6)
i.e. an exponential growth with time of the scale factor.
Let us see quantitatively how inflation can solve the flatness problem. Suppose
that
|K|
= O(1) , (8.7)
a2i Hi2
at the beginning of the inflationary phase, to which we refer with a subscript i.
That is, at the beginning of inflation the spatial curvature might be a relevant
fraction of the total energy density content of the universe.
195
Now, at the end of inflation, which we denote with a subscript I, the scale
factor a has grown of a factor eN . The number N is called e-folds number. Then
we have:
|K| |K|
2 2
= 2 2 e−2N ≈ e−2N . (8.8)
aI HI ai Hi
Therefore, today we have that:
2 2
|K| |K| aI HI −2N aI HI
|ΩK0 | = 2 = 2 2 ≈e . (8.9)
H0 aI HI H0 H0
In order to have |ΩK0 | < 1, we need
aI HI
< eN (8.10)
H0
In order to estimate the ratio on the left hand side, let us suppose that inflation
ends into the radiation-dominated epoch, so that:
HI2 ≈ H02 Ωr0 /a4I , (8.11)
and thus: r
1/4 H0
aI ≈ Ωr0 . (8.12)
HI
We have then:
r 1/4 1/4
N 1/4 HI 1/4 ρI ρI
e > Ωr0 = Ωr0 = , (8.13)
H0 ρ0 0.037 h eV
where ρI is the energy scale at which inflation ends, but since H is constant during
the inflationary phase, then ρI is energy density scale of inflation tout court.
Certainly we do not want to spoil BBN, therefore ρI must be larger than 1
MeV4 . With this constraint we get N > 17. Choosing the Planck scale, one gets
N > 68 and choosing the GUT energy scale one gets N > 62.
8.2 The horizon problem

The horizon problem is an issue that appears when we calculate the angular size
of the particle horizon at recombination and notice that it is only a small portion
of the CMB sky. Then, how is it possible that the latter is so isotropic if no causal
process could have provided the conditions to be so?
Let us again consider this issue more quantitatively. The proper particle-
horizon distance is the following:
Z t Z a
dt0 da0
dH = a(t) 0
= a 0 02
. (8.14)
0 a(t ) 0 H(a )a
In an universe dominated by matter and radiation the above expression becomes:

Z a
da0 2a p p
dH = a √ 0
= Ωm0 a + Ωr0 − Ωr0 . (8.15)
0 H0 Ωm0 a + Ωr0 H0 Ωm0
196
Now, recall that the angular diameter distance has the following form:
Z t0 Z 1
dt0 da0
dA = a(t) = a 0 02
. (8.16)
t a(t0 ) a H(a )a
Again, in an universe dominated by matter and radiation the above expression

becomes:
Z 1
da0 2a p p
dA = a √ 0
= Ω m0 + Ω r0 − Ωm0 a + Ω r0 . (8.17)
a H0 Ωm0 a + Ωr0 H0 Ωm0
The ratio dH /dA represents the angular radius of the particle horizon at a given
scale factor. From the above calculations we obtain:
√ √
dH Ωm0 a + Ωr0 − Ωr0
=√ √ . (8.18)
dA Ωm0 + Ωr0 − Ωm0 a + Ωr0
This ratio tends to zero for a → 0 and at recombination it is equal to:
dH
(arec = 10−3 ) = 0.018 , (8.19)
dA
which corresponds to about 1◦ in the CMB sky. Therefore, we have roughly
4π/(0.018)2 ≈ 104 causally disconnected regions in the sky.
Another way to see this problem is to compare dH with the radius of the uni-
verse, say R, which is proportional to the scale factor a, and calculating thus the
number of causally disconnected regions, i.e.:
1/3 R Ωm0 H0 Ri
Ncdr ≡ = √ √ , (8.20)
dH 2ai Ωm0 a + Ωr0 − Ωr0
where Ri is some initial radius of the universe corresponding to an initial scale

factor ai .
Exercise 8.3. Typically one assumes that Ncdr ∼ 1090 at the Planck scale, i.e.
of the order or larger than the observed number of particles in the universe. See
e.g. [Linde, 2017] for a recent discussion on the issue of the initial conditions of the
universe.
Given Ncdr ∼ 1090 at a = 10−32 determine then H0 Ri /ai in Eq. (8.20) and
deduce the value of Ncdr at recombination. Show that, again, Ncdr ≈ 104 .
So, the particle horizon at recombination is a very small fraction of the CMB
sky. If no causal process could have taken place beyond 1 degree then what did
cause the high isotropy of the CMB temperature?
This issue can be solved in pretty much the same way as we did for the flatness
problem. Assume an initial inflationary phase such that:
a(t) = ai eHI (t−ti ) = aI e−HI (tI −t) . (8.21)
197
Then, the proper particle-horizon distance that we have calculated in Eq. (8.15)
acquires the following contribution for very small times:
Z tI
eHI (tI −t)
dH ≈ a dt , (8.22)
ti aI
which we actually put equal to the same dH because we know that it has to dominate
it in order to solve the horizon problem. Therefore:
Z tI
eHI (tI −t) a
dH ≈ a dt = (eN − 1) . (8.23)
ti a I a I HI
Since dA ≈ a/H0 for small scale factors, we can conclude that:
dH H0 N
≈ e , (8.24)
dA aI HI
and so, in order to have dH > dA , we obtain the condition:
aI HI
< eN (8.25)
H0
which is the same as in Eq. (8.10). The same solution for two different problems
seems to be a good sign and a good point in favour of inflation.
Usually a third problem concerning standard cosmology goes along with the
above two and it is the one related to the abundance of unwanted relics, such as
magnetic monopoles, which are produced via some symmetry breaking mechanism
in the very early universe and are not observed today. We do not treat this issue
here, see [Weinberg, 2013] for more details, but it is clear that inflation provides a
mechanism for diluting these unwanted relics beyond the possibility of observation.
We now discuss how inflation can be realised.
8.3 Single scalar field slow-roll inflation

We present here the possibility of implementing inflation via a single canonical
scalar field ϕ, called inflaton, which is subject to some potential V (ϕ) with the
property that the scalar field initially rolls slowly down it attaining its minimum
after the end of inflation. This kind of behaviour is called slow-roll inflation and
it proves to be very successful. We now see in some detail how the inflationary
phase occurs.
The Lagrangian of a canonical scalar field is:
1
L = g µν ∂µ ϕ∂ν ϕ + V (ϕ) , (8.26)
2
where the plus sign might be deceiving (we are used in Lagrangian mechanics to a
difference between the kinetic term and the potential one) if we do not recall that
the signature in use is (−, +, +, +).
198
The energy-momentum tensor of a canonical scalar field has the following form:

α αν α 1 µν
T β = g ∂ν ϕ∂β ϕ − δ β g ∂µ ϕ∂ν ϕ + V (ϕ) , (8.27)
2
where V (ϕ) is some potential. In the background FLRW metric ds2 = −dt2 +
a(t)2 δij dxi dxj , the scalar field has to depend only on t, and thus the energy density
and pressure are:
1 1 1
ρϕ = −T 0 0 = ϕ̇2 + V (ϕ) , Pϕ = δ i j T j i = ϕ̇2 − V (ϕ) . (8.28)
2 3 2
Exercise 8.4. Derive the Klein-Gordon equation for ϕ using the continuity equa-
tion. Show that:
ϕ̈ + 3H ϕ̇ + V,ϕ = 0 , (8.29)
where V,ϕ ≡ dV /dϕ. Show that in the conformal time
ϕ00 + 2Hϕ0 + a2 V,ϕ = 0 . (8.30)
Now, Friedmann and the acceleration equations for this scalar field as the unique
content of the universe are written as:

2 8πG 1 2 ä 8πG 2
H = ϕ̇ + V (ϕ) , =− ϕ̇ − V (ϕ) . (8.31)
3 2 a 3
Exercise 8.5. Show that the acceleration equation can be written in the following
form:
Ḣ = −4πGϕ̇2 . (8.32)
In order to have an accelerated phase of expansion we need that H varies slowly,

i.e. Ḣ ∼ 0, but how much slowly? Since the only time scale present in the problem
is 1/H itself, we need that
|Ḣ|
1, (8.33)
H2
because this is the only dimensionless combination possible with Ḣ. It means
that, during an expansion time 1/H, the relative variation of H is much less than
unity. Using Friedmann equation and the expression for Ḣ we can write the above
condition as:
ϕ̇2 V (ϕ) (8.34)
which is our first slow-roll condition. When the kinetic term of the scalar field
is negligible with respect to the potential one, one has from Eq. (8.28):
Pϕ ≈ −ρϕ ≈ −V (ϕ) ≈ constant . (8.35)
199
That is, the scalar field potential, when it dominates over the kinetic term, behaves
as a cosmological constant.
Note that condition (8.34) does not really demand the potential V (ϕ) to be flat,
as for example the Coleman-Weinberg (Erick, not Steven) potential [Coleman and
Weinberg, 1973] used in the “new” inflationary scenario [Albrecht and Steinhardt,
1982]. What counts is the kinetic term to be very small compared with the potential
one and this can be achieved even for V (ϕ) ∝ ϕ2 or ϕ4 potentials. We shall see
this in more detail later.
The condition given in Eq. (8.33) is usually reformulated in term of a parameter
, called slow-roll parameter and defined as the ratio in Eq. (8.33), i.e.

Ḣ d 1
≡− 2 = (8.36)
H dt H
and it is the derivative of the Hubble radius. For an exactly exponential expansion
H is constant (de Sitter space), and thus = 0. The slow-roll condition is thus√
realised by demanding that || 1. During the radiation-dominated epoch, a ∝ t
and thus 1/H = 2t and = 2.
Exercise 8.6. The minimal requirement for inflation is that it must produce an
acceleration, i.e. ä > 0. For the limiting case in which we have ä = 0, show that
= 1.
In Fig. 8.1 the dashed lines represents generic physical scales λphys ∝ a/k grow-
ing proportionally to the scale factor and exiting the Hubble radius during inflation,
since 1/H stays almost constant (it is really constant in the figure), and re-entering
after inflation ends, during the radiation-dominated epoch. Of course, the last scale
to exit is also the first one to re-enter, whereas the observable scale which first ex-
its the horizon, is the one which re-enters the horizon today. According to the
normalisation used in Fig 8.1, this scale is a and exited the horizon when:
H0
aexit = . (8.37)
HI
Using Eq. (8.13), we can thus write down the maximum number of e-folds to which
we could have access observationally as:
!
1/4
aI aI HI ρI
Nmax = ln = ln = ln . (8.38)
aexit H0 0.037 h eV
For the energy scale of GUT, i.e. 1016 GeV, one gets Nmax ≈ 61. Note that this is
not the duration of inflation. It can last much longer, e.g. N ≈ 145 as computed
in [Bolliet et al., 2017], but we can observationally probe only the last 61 e-folds
(depending on the energy scale of inflation).
We shall see later that when a perturbative scale crosses the Hubble horizon
during inflation, it gets a specific value which remains constant during it super-
horizon evolution and then serves as initial condition when re-entering the horizon
during the radiation-dominated epoch.
200
��-��
��-��
��-��
��-��
��-��
��-��
��-�� -�� -�� -�� -��
Figure 8.1: The solid line represents the evolution of the Hubble radius 1/H, nor-
malised to the present one 1/H0 for the ΛCDM model. The dashed lines represents
different scales, which grow as a, exiting the Hubble radius and then re-entering
again during the radiation-dominated epoch. The dotted line represents the Planck
scale. We have chosen inflation to end at a = 10−29 , corresponding to the GUT
scale 1016 GeV.
In Fig. 8.1 there is also a horizontal dotted line. This represents the Planck
scale. Note that scales which re-enter the horizon after the end of inflation, and are
thus observable by us for example in the CMB, were smaller than the Planck scale.
Since we do not know how gravitation behaves on such smaller scales, the above is
known as Trans-Planckian problem. See e.g. [Brandenberger and Martin, 2013]
for a recent review on the subject.
8.3.1 More slow-roll parameters

It is not only important that the inflaton slow-rolls, but also that it does so for
a sufficiently long time in order to provide at least N = 60 e-folds. In order to
make this claim more quantitative, let us investigate how varies by computing
its time-derivative. Using the definition (8.36) we get:
2Ḣ 2 Ḧ
˙ = − , (8.39)
H3 H2
and from Eq. (8.32) we get:
8πGϕ̇ Ḣ ϕ̈ ϕ̈
˙ = 2H2 + 2
ϕ̈ = 2H2 − 2 2 = 2H2 + 2 . (8.40)
H H ϕ̇ ϕ̇
201
Let us write this expression as:
˙ = 2H( − η) , (8.41)
where
1 ϕ̈
η≡− (8.42)
H ϕ̇
is our second slow-roll parameter, not to be confused with the conformal time. The
smallness of η then guarantees that varies slowly. Indeed, η 1 gives us, from
the KG equation (8.29) the condition
3H ϕ̇ ≈ −V,ϕ (8.43)
Considering , η 1 as first-order quantities, the time-derivative ˙ is thus a second-
order quantity, cf. Eq. (8.41).
Nothing prevents us of considering now the time-derivative of η and then to
define a third slow-roll parameter α, of which we could consider again the time-
derivative, defining a fourth slow-roll parameter β and so on, constructing a hier-
archy of slow-roll parameters. This was done in [Liddle et al., 1994], but we limit
ourselves here to and η, upon which the predictions on the scalar and tensor spec-
tral indices depend, as we shall see. It must be stressed, on the other hand, that
future observation might be able to constrain with great precision the runnings
of the spectral indices, which depend on the higher-order slow-roll parameters,
see [Muñoz et al., 2017].
Since it is the scalar field that triggers the inflationary phase, it is useful to
express and η in terms of quantities related to the scalar field itself, i.e. the
potential V (ϕ) and its derivatives. This can be done as follows. Combining the
definition of in Eq. (8.36) with Friedmann equation and Eq. (8.32), one gets:
" #
2
3ϕ̇2 3ϕ̇2 ϕ̇2
= = +O , (8.44)
2V + ϕ̇2 2V 2V
where we are implementing the slow-roll condition (8.34). Using now Eq. (8.43),
we find:
2
1 V,ϕ
≈ ≡ V (8.45)
16πG V
at the lowest order in ϕ̇2 /2V . In this equation we have defined V as a quantity
describing the steepness of the inflaton potential.
Since η depends on the second time derivative of the scalar field, we expect it
to depend on the second derivative of the potential, i.e. V,ϕϕ , in the slow-roll limit.
Using Eq. (8.43) and the above definition (8.42), we can write:
V,ϕϕ + 3Ḣ
η≈ . (8.46)
3H 2
Now, recalling the definition of and that, at the lowest order in the slow-roll
approximation, 3H 2 ≈ 8πGV , we can conclude that:
1 V,ϕϕ
η+≈ ≡ ηV (8.47)
8πG V
202
It exists thus a hierarchy of slow-roll parameters based on the derivative of the
Hubble factor and another one based on those of the potential. They of course can
be put in relation, as we did above for the first two slow-roll parameters.
So, in order to have V , ηV 1 and thus trigger inflation not necessarily V has
to be constant but rather its first and second derivatives must be much smaller
than the value of V itself.
We can relate the number of e-folds to the slow-roll parameter as follows.
Recalling that the number of e-folds is N = ln a, we can immediately write that:
Z t2
Ṅ = H ⇒ ∆N12 = Hdt , (8.48)
t1
where t1 < t2 are two generic instants during the inflationary phase. Changing
variable in favor of the inflaton field, we can write:
Z ϕ2
H
∆N12 = dϕ , (8.49)
ϕ1 ϕ̇
where ϕ1 ≡ ϕ(t1 ) and ϕ2 ≡ ϕ(t2 ). Using the slow-roll conditions presented earlier
one has: Z ϕ1
V
∆N12 = 8πG dϕ . (8.50)
ϕ2 V ϕ
Since V /Vϕ is almost constant and very large during inflation (its square is pro-
portional to 1/), we can pull it out of the integral and approximate the above
equation as:
8πGV V ϕ1 − ϕ2
∆N12 ≈ (ϕ1 − ϕ2 ) = 2
, (8.51)
Vϕ Vϕ MPl
√
where we have introduced the Planck mass MPl instead of 1/ 8πG. In order to
produce a large N it might be possible that |ϕ1 −ϕ2 | > MPl , but this not necessarily
leads to a trans-Planckian problem because the energy scale of inflation is V (ϕ),
not the field itself, and so it is the potential which has to be smaller than the
Planck scale in order for a classical treatment to be valid. This can be achieved,
for example, if there is a sufficiently small coupling constant.
Using Eq. (8.45), we can write in general that:
1 ∆ϕ
∆N = √ . (8.52)
2 MPl
Suppose that ϕ1 and ϕ2 are the value of the inflaton field for which the wavenumbers
corresponding to the CMB multipole ` = 1 and ` = 100 exit the horizon. Since, as
we are going to see in Chapter 10, ` ∝ k, then:
∆N = ln 100 = 4.6 . (8.53)
The above formula then gives a bound between the variation of the inflaton field
and , based on the observable scales of the CMB. Such bound is known as Lyth
bound [Lyth, 1997]. See also [Di Marco, 2017] for a recent perspective on the
Lyth bound.
203
8.3.2 Reheating
When V and ηV cease to be very small and attain values of order unity, inflation
ends. Close to the minimum say ϕ0 the inflaton potential can be expanded in the
following way:
1 1
V (ϕ) = V0 + V,ϕϕ |ϕ=ϕ0 (ϕ − ϕ0 )2 + · · · ≈ m2ϕ (ϕ − ϕ0 )2 , (8.54)
2 2
where V0 ≡ V (ϕ0 ) is the minimum of the potential which we assume to be very
small, in fact negligible, in order not to generate an important vacuum energy
contribution (which dominates only at late times). We have also introduced the
inflaton mass as it is customary, i.e. as the second derivative of the potential
evaluated at ϕ0 .
The above approximated potential is the one of a harmonic oscillator with
proper frequency mϕ and so we expect the inflaton to perform oscillations about
the minimum, damped by the Hubble flow H but not only. Since inflation needs
to end and there the radiation-dominated epoch must start, we need to couple the
inflaton to other matter fields in order for the inflaton to lose energy in favour of
the latter. Phenomenologically, one can write:
ρ̇ϕ + 3Hρϕ (1 + wϕ ) = −Γρϕ , (8.55)

ρ̇M + 3HρM (1 + wM ) = Γρϕ , (8.56)
where Γ is some scattering rate governing the decay of the inflaton and with M
we refer to matter in general which has to be relativistic in order to gives rise to a
radiation-dominated epoch and thus wM ≈ 1/3.
Therefore, the inflaton oscillations are also damped by the presence of Γ. This
final phase of inflation, transiting to the radiation-dominated epoch is called re-
heating.
Exercise 8.7. Assuming wϕ constant, show that:

3(1+wϕ ) Z t
a(tI ) 0 0
ρϕ (t) = ρϕ (tI ) exp − dt Γ(t ) , (8.57)
a(t) tI
where tI is the time at which inflation ends.
With this formal solution, we can find another formal one for the matter part.
Exercise 8.8. Assuming wM = 1/3, show that:

!
t t0
ρϕ (tI )a(tI )3(1+wϕ )
Z Z
0 0 0 1−3wϕ 00 00
ρM (t) = dt Γ(t )a(t ) exp − dt Γ(t ) . (8.58)
a(t)4 tI tI
From the above equation one sees that the energy density of the matter fields
is zero at the end of inflation, then it rises more or less abruptly (depending on
204
Γ) and then finally decreases again as 1/a4 after the inflation has given up all its
energy. Now, assume Γ to be constant and wϕ = 0, for simplicity. We get for the
matter density:
ρϕ (tI )Γa(tI )3 t 0 0
Z
ρM (t) = dt a(t ) exp [−Γ(t0 − tI )] . (8.59)
a(t)4 tI
Integrating once by parts and considering the limit Γ(t − tI ) 1, i.e. a very large
decay rate, we get:
ρϕ (tI )a(tI )4
ρM (t) = , (8.60)
a(t)4
i.e. the decay takes place so rapidly that all the energy of the inflaton is passed to
the matter.
If, on the other hand, Γ(t − tI ) 1, i.e. a very small decay rate, we can
approximate the matter density as:
ρϕ (tI )Γa(tI )3 t 0 0
Z
ρM (t) ≈ dt a(t ) , (8.61)
a(t)4 tI
and since Γ is so small, the inflaton is still dominating the dynamics giving thus
for the scale factor
a(t) = a(tI )(t/tI )2/3 , (8.62)
because we have chosen wϕ = 0. Integrating, we have thus:
3 ρϕ (tI )ΓtI
(t/tI )5/3 − 1 ,

ρM (t) ≈ 8/3
(8.63)
5 (t/tI )
and the maximum is attained at tmax /tI = (8/3)3/5 and its value is:
Γ
ρM,max (t) ≈ 0.139 ρϕ (tI ) . (8.64)
H(tI )
Hence, in this case the energy density of matter is much smaller than that of the
inflaton. Most of the latter is still spent driving the expansion of the universe
instead of generating matter, because of the small Γ.
We have seen how to produce an inflationary phase and how to quantify it,
through the slow-roll parameters. Now we are going to see what inflation has to
say about quantum fluctuations. We hypothesise that before inflation the universe
was quantum and quantum fluctuations were turned into classical ones by inflation
itself, though we do not address the details of the quantum-to-classical transition
of the primordial fluctuations. About the latter topic, see e.g. [Kiefer and Polarski,
2009].
8.4 Production of gravitational waves during infla-

tion
We are going to quantize the equation that we found for the evolution of gravita-
tional waves, i.e. from Eq. (4.191):
00 0
hTij + 2HhTij − ∇2 hTij = 0 , (8.65)
205
where recall that hTij is traceless and transverse, i.e.
hTii = 0 , ∂ j hTij = 0 , (8.66)
and no source term πijT appears in the GW equation since this is vanishing for a
scalar field. This can be understood intuitively, since the inflaton ϕ is a scalar
and thus unable of producing a tensor perturbation. Mathematically, just look
Eq. (8.117) and realise that no anisotropic stresses appear in the δT i j of a scalar
field, since this is diagonal.
The above Eq. (8.65) has no source term, is gauge-invariant and paves the way
for a similar treatment that we shall do for scalar perturbations. For this reason
we tackle the tensor ones first.
The Fourier transform of the above equation is Eq. (4.197), which we write here
in the following compact form:
h00 + 2Hh0 + k 2 h = 0 , (8.67)
where h(η, k) represents either h+,× or h(λ = ±2). Since the above equation only
contains the modulus k, only the initial condition on h shall have a dependence on
k, cf. Eq. (7.101).
It shall be useful to cast Eq. (8.67) in the form of a harmonic oscillator one
with no damping term by defining:
ah MPl
g≡√ = ah (8.68)
32πG 2
The normalisation √ comes in order to give g dimensions of a mass. Indeed, h is

dimensionless and 8πG = 1/MPl . This is necessary in order to give the correct
dimensionality, a cube length or inverse
√ cube mass, to the power spectrum.
A very direct way to see that 32πG is the correct normalisation is e.g. to
look to Mukhanov’s calculations in [Mukhanov et al., 1992]. He finds the action
which gives Eq. (8.65) by perturbing up to the second order a f (R) action about
a generic FLRW background. In the Einstein-Hilbert case f (R) = R and for a
spatially flat FLRW background one has:
Z
1 0
T k0

δ2 S = a h k h i − ∂ l h k ∂ h i d4 x .
2 Ti Ti l Tk
(8.69)
64πG
In [Mukhanov et al., 1992] the corresponding second order action for the scalar
field is also found and the factor outside the integral is 1/2. Hence, in order to
properly compare tensor and scalar fluctuations we must rescale the former as:
hT i j
hT i j → √ . (8.70)
32πG
Exercise 8.9. Use Eq. (8.68) into Eq. (8.67) in order to find:
a00

00 2
g + k − g=0 (8.71)
a
206
The GW wave can be written as:
X Z d3 k h
T
hij (η, x) = 3
h(η, k)eik·x a(k, λ)eij (k̂, λ)
λ=±2
(2π)
i
h∗ (η, k)e−ik·x a∗ (k, λ)e∗ij (k̂, λ) , (8.72)
where the sum is over the helicity of the GW and eij (k̂, λ) is the polarisation tensor
defined in Sec. 4.6. We have put the initial dependence on k of the GW in a(k, λ)
and its complex conjugate, which is introduced in order to guarantee that hTij (η, x)
is real.
Now, we assume the initial state of the GW field to be a quantum one on very
small scales k aH. Thus, we promote hTij and a to operators and impose the
canonical commutation relations:
[a(k, λ), a(k0 , λ0 )] = 0 , a(k, λ), a† (k0 , λ0 ) = (2π)3 δ (3) (k − k0 )δλλ0 , (8.73)

which tell us that a† (k, λ) creates a graviton of momentum-energy k and helicity

λ, whereas a(k, λ) destroys it.
The quantum state of the universe during inflation is the vacuum |0i, which by
definition is annihilated by a(k, λ). The expectation value on the vacuum:
h0|hTij (η, x)hTlm (η, x0 )|0i , (8.74)
is what shall define our primordial spectrum. Two comments are in order here.
First of all, we could choose another quantum state for the universe which is not
necessarily the vacuum one. Second, the above is a vacuum expectation value
which we expect to become an ensemble average over classical random fields well
outside the horizon.
Exercise 8.10. Compute the vacuum expectation value in Eq. (8.74) using the
plane-wave expansion (8.72) and the commutation relations (8.73). Show that:
d3 k
Z
T T 0 2 ik·(x−x0 )
h0|hij (η, x)hlm (η, x )|0i = |h(η, k)| e Πij,lm (k̂) , (8.75)
(2π)3
where X
Πij,lm (k̂) ≡ eij (k̂, λ)e∗lm (k̂, λ) , (8.76)
λ=±2
is the sum over the helicities, defined in Eq. (7.103).
Comparing the result found in the exercise with Eq. (7.42) it is quite straight-
forward to see that the tensor quantum perturbations are Gaussian with power
spectrum:
Ph (η, k) ∝ |h(η, k)|2 , (8.77)
207
hence we need to solve Eq. (8.67), or rather Eq. (8.71), in order to determine
|h(η, k)|2 .
Inspect Eq. (8.71). There are two regimes of interest. The first is for very short
wavelength, i.e. k 2 a00 /a, for which the equation becomes:
g 00 + k 2 g = 0 , (k 2 a00 /a) , (8.78)
which is the usual harmonic oscillator equation. This means that the details of
the inflationary evolution, encoded in a(η), are not important on very small scales.
This was not unexpected since on very small scales the expansion of the universe
is irrelevant. Moreover, from the QFT point of view, we can then look at the
quantisation as that of a free field in Minkowski space and thus determine:
1 1
g(η, k) = √ e−ikη , g ∗ (η, k) = √ eikη , (k 2 a00 /a) , (8.79)
2k 2k
√
where the normalisation 1/ 2k comes from the quantisation procedure in Minkowski
space. We can take this solution as the “initial condition” in solving Eq. (8.71)
therefore addressing only those modes which satisfy the condition k 2 a00 /a dur-
ing the inflationary epoch.
The second regime of evolution is given by the condition k 2 |a00 /a|, i.e. for
very long wavelength, for which Eq. (8.71) becomes:
a00
g 00 − g=0, (k 2 |a00 /a|) . (8.80)
a
This equation has the following formal solution:
Z η
dη̄
g(η, k) = C1 (k)a + C2 (k)a , (k 2 |a00 /a|) , (8.81)
a(η̄)2
which contains the solution
g h
=√ = constant , (8.82)
a 32πG
representing thus the constant value that h reaches outside the horizon and that
will eventually become the initial condition at the re-entering, during the radiation-
dominated epoch.
Let us try to be more quantitative.
Exercise 8.11. Assume that during inflation H = a0 /a2 is a constant say HΛ , i.e.
consider a de Sitter space. Show that:
1
a(η) = − . (8.83)
HΛ η
Since the scale factor and HΛ are positive, then η is negative during inflation.
With the solution of Eq. (8.83), which we could take valid also in the case of
slowly varying H, show that the condition used in Eq. (8.80) can be written as:
k 2 |a00 /a| = 2HΛ2 a2 . (8.84)
208
This is a condition for the physical wavenumber k/a to be much smaller than
the Hubble factor, i.e. it is a condition for super-Hubble scales. In general
|a00 /a| ∝ 1/η 2 , for dimensional reasons, hence the condition in Eq. (8.80) can
also be understood as:
kη 1 , (8.85)
which means super-horizon scales, since η is the comoving particle horizon.
In this approximation then Eq. (8.71) becomes:

00 2 2
g + k − 2 g=0. (8.86)
η
Exercise 8.12. Show that the solution of Eq. (8.86) can be written as:
C1 (k) C2
g(η, k) = [kη cos(kη) − sin(kη)] + [kη sin(kη) + cos(kη)] . (8.87)
η η
Now, let us select the exp(−ikη) mode. Show that in this case the solution becomes:

−ikη i
g(η, k) = C1 (k)ke 1− . (8.88)
kη
Now suppose an initial condition in η = ηi , such that:

a00 2
k2 = 2 , ⇒ k|ηi | 1 , (8.89)
a ηi
so that we can do the matching with Eq. (8.79) and obtain:

1 1
C1 (k)ke−ikηi = √ e−ikηi , ⇒ C1 (k)k = √ , (k|ηi | 1) . (8.90)
2k 2k
and so we can write:
e−ikη

i
g(η, k) = √ 1− , (k|ηi | 1) . (8.91)
2k kη
Using now Eq. (8.68), we get for the power spectrum:

32πG 16πG 1 16πG 2 2 1
Ph (η, k) = Pg (η, k) = 1+ 2 2 = HΛ η 1 + 2 2 ,
a(η)2 a(η)2 k k η k k η
(8.92)
and the dimensionless one, according to the definition of Eq. (7.45), is:
k 3 Ph (η, k) 8πG 2 HΛ2

∆2h (η, k) 2 2 2 2

= = HΛ 1 + (k η ) = 2
1 + (k η ) , (8.93)
2π 2 π2 π 2 MPl
and recall that this result is valid only for those scales for which k|ηi | 1.
209
This spectrum depends on time, so which η should we choose? We have seen,
through the discussion culminating in Eq. (8.84), that when a scale k exits the
horizon, its corresponding h(η, k) becomes constant. Its value will be the initial
condition during the radiation-dominated epoch, when the scale re-enters the hori-
zon. Therefore, for kη → 0 we have:
HΛ2
∆2h (η, k) = 2
, (8.94)
π 2 MPl
which does not depend on the scale k anymore, i.e. it is a scale-invariant power
spectrum. The only caveat is that this spectrum is valid only for those scales for
which k|ηi | 1, i.e. scales which exit the horizon:
a2exit HΛ2 ηi2 1 , ⇒ ηi2 ηexit

2
, (8.95)
well after ηi . Recall that η is negative during inflation, so the above condition
makes sense.
Note that for a constant HΛ the slow-roll parameter is vanishing. In order to
describe the inflationary phase more realistically, we should consider a small but
non-vanishing . In this case H varies and a k-dependence is gained by the power
spectrum.
Let us write the definition of in Eq. (8.36), but using the conformal time:
a 0 1 H0
= =1− . (8.96)
H a H
Exercise 8.13. Solve the above differential equation for H assuming a constant
. Show that:
1
H=− . (8.97)
(1 − )η
From this solution, it is not difficult to show that:

a00 1 1 2 + 3
= H0 + H2 = 2
+ 2 2
≈ , (8.98)
a (1 − )η (1 − ) η η2
where in the last approximation we have considered 1, as it should be during
inflation, and kept to first order.
Substituting into Eq. (8.71), we get then:

00 2 2 + 3
g + k − g=0, (8.99)
η2
which can be solved exactly, giving as general solution a combination of Hankel

functions:
√
√ (1)
√ (2) 3√
g(η, k) = C1 (k) −ηHν (−kη) + C2 (k) −ηHν (−kη) , ν = 3 + 4 ,
2
(8.100)
210
where the minus sign must be introduced in order to account for the fact that
η < 0. We have chosen to express the solution in terms of the Hankel functions
since it is easier to see which one of them matches the initial condition of Eq. (8.79).
Indeed, consider the asymptotic expansion, cf. [Abramowitz and Stegun, 1972], of
(1)
Hν (−kη):
r
2
(1)
Hν (−kη) ∼ e−ikη−iνπ/2−iπ/4 , (k|η| 1) . (8.101)
−πkη
Hence, this is the right behavior at very small scales for g and the integration
constant must be: √
π iνπ/2+iπ/4
C1 (k) = e , (8.102)
2
i.e. indeed a (complex) constant, since it bears no dependence from k. The solution
we look for is then:
√ √
π iνπ/2+iπ/4 √ (1) 3√
g(η, k) = e −ηHν (−kη) , ν = 3 + 4 , (8.103)
2 2
and the power spectrum is thus:
4 π|η|
Ph (η, k) = 2
|g(η, k)|2 = 2
|Hν(1) (−kη)|2 . (8.104)
MPl a(η)2 MPl a(η)2
For kη → 0, i.e. on very large scales, one has that
−ν
(1) Γ(ν) k|η|
Hν (−kη) ∼ −i , (k|η| → 0) , (8.105)
π 2
and hence the dimensionless power spectrum becomes:
−2ν
k 3 |η|Γ(ν)2 k|η|

2
∆h (η, k) = 2 2
, (k|η| → 0) . (8.106)
2π 3 MPl a 2
Considering at first order in the exponent of k|η| and negligible elsewhere, we
can write:
−3−2
k 3 |η|Γ(ν)2 k|η|

1
2
∆h (η, k) = 3 2 2
= 2 2 2 2 (k|η|)−2 . (8.107)
2π MPl a 2 π MPl a |η|
Now, the power spectrum is no more scale-invariant but gained a small k-dependence:
a power law with exponent −2. The latter is usually named as tensor spectral
index and denoted as nT . We shall see a little more detail about it later.
On large scales, h is time-independent and so its power spectrum is. Hence, we
can choose any convenient value of η when to evaluate ∆2h . It is customary to use
the time ηk at which a given scale k crosses the horizon, i.e.
k = H(ηk ) , (8.108)
because the details of the k-dependence of the power spectrum depend on the in-
flationary model but, as we saw in Eqs. (8.78) and (8.80), they manifest themselves
only at horizon-crossing.
211
Using Eq. (8.97), the horizon crossing condition gives the following relation
between |η| and k:
1
k = H(ηk ) = , (8.109)
(1 − )|ηk |
which at fist order in gives:
k|ηk | = 1 + . (8.110)
Using again Eq. (8.97), we can write the power spectrum evaluated at horizon
crossing as:
1 H(ηk )2
∆2h (k) = 2 2 2 , (8.111)
π MPl a (ηk )
which is usually put as:
H2
∆2h (k) = 2 2 , (8.112)
π MPl k=aH
i.e. it has the same form as the one in the de Sitter case, computed in Eq. (8.94),
but allows for a time-dependent Hubble factor.
It is very interesting to notice that if we could directly measure the gravitational
waves background, we would be able to determine the energy scale of inflation, i.e.
H.
8.5 Production of scalar perturbations during in-

flation
Consider our FLRW metric perturbed with scalar perturbations only and written
in the conformal Newtonian gauge, i.e.
ds2 = a(η)2 − (1 + 2Ψ) dη 2 + (1 + 2Φ) δij dxi dxj .

(8.113)
Accordingly, consider also perturbations of our inflaton field:
ϕ = ϕ̄(t) + δϕ(η, x) (8.114)
Exercise 8.14. Using Eqs. (8.27) and (8.114), write down the perturbed energy-
momentum tensor. Show that:
1 0 1 1 0
T 0 0 = − 2 ϕ̄ 2 − V (ϕ) − 2 ϕ̄0 δϕ0 + 2 Ψϕ̄ 2 − V,ϕ δϕ , (8.115)
2a a a
1
T 0 i = − 2 ϕ̄0 ∂i δϕ , (8.116)
a
i i 1 0 2 i 1 0 0 1 0 2
T j=δj ϕ̄ − V (ϕ) + δ j ϕ̄ δϕ − 2 Ψϕ̄ − V,ϕ δϕ . (8.117)
2a2 a2 a
The perturbed contribution in the above expressions is grouped to the right.

Note the following important fact: T i j is diagonal and thus no anisotropic stresses
can be sourced by a single scalar field. We already know from Eq. (4.184) that this
fact implies Ψ = −Φ.
212
Exercise 8.15. Obtain the perturbed Klein-Gordon equation. Start from:
∇µ T µ ν = 0 , (8.118)
and then put ν = 0. Show that:
δϕ00 + 2Hδϕ0 + k 2 + V,ϕϕ a2 δϕ = 2ΦV,ϕ a2 − 4Φ0 ϕ̄0

(8.119)
where we have already used the conformal time and Ψ = −Φ. Compare this equa-
tion with the corresponding one found by Mukhanov and co-authors in [Mukhanov
et al., 1992]. It is also useful to consider that found in [Weinberg, 2008] and recover
the above one by transforming from the cosmic time to the conformal one.
Unfortunately, the contributions V,ϕϕ a2 , 2ΦV,ϕ a2 and 4Φ0 ϕ̄0 make life more dif-
ficult. Indeed, if they were absent, we would get the following equation:
δϕ00 + 2Hδϕ0 + k 2 δϕ = 0 , (8.120)
which is formally identical to Eq. (8.67), and therefore all the calculations per-
formed for the GW case would follow in the same fashion.
Even if we could neglect V,ϕϕ in Eq. (8.119) during the slow-roll phase, there
would be no a priori reason to neglect the whole right hand side of Eq. (8.119).
Moreover, we do not really want to follow the evolution of δϕ, for at least two rea-
sons. First, the inflaton supposedly decays into ordinary matter during reheating,
and we do not want to address this problem. Second, as we saw in Eq. (6.94),
what we really need is to know R or ζ on very large scales when radiation starts
to dominate.
Let us work with R. It turns out that R is conserved (i.e. is a constant) on large
scales and for adiabatic perturbations. We have already seen this in Chapter 6 and
we show this in detail again in Chapter 12.
It turns out that, as it happened for h, when inflation brings a scale k outside
the horizon, its corresponding R(η, k) attains a constant value which shall serve
as initial condition during the radiation-dominated epoch. There shall be a crucial
difference with respect to the case of GW, which we shall meet shortly.
Exercise 8.16. Express v in Eq. (4.131) in terms of the inflaton field velocity
potential. From Eq. (4.73) we know that:
(ϕ) ϕ̄02 (ϕ)
T 0(ϕ) i = ρ̄ϕ + P̄ϕ vi = 2 vi . (8.121)
a
Using Eq. (8.116), show that:
(ϕ) ∂i δϕ
vi =− . (8.122)
ϕ̄0
Thus, writing
(ϕ)
vi = ∂i v (ϕ) , (8.123)
we can identify the velocity potential of the inflation field (we drop the superscript
ϕ now) as:
δϕ
v=− 0 . (8.124)
ϕ̄
213
Hence, using the result of the exercise, we can write:
H
R=Φ− δϕ . (8.125)
ϕ̄0
We need to find an evolution equation for R. In principle, combining the above
expression with the Klein-Gordon equation (8.119) we can eliminate δϕ. Then, Φ
can be dealt with via the use of the Einstein equations, which are:
0

3H(Φ0 + HΦ) + k 2 Φ = 4πG ϕ̄0 δϕ0 + Φϕ̄ 2 + V,ϕ a2 δϕ , (8.126)
Φ0 + HΦ = −4πGϕ̄0 δϕ , (8.127)
0

Φ00 + 3HΦ0 + (2H0 + H2 )Φ = −4πG ϕ̄0 δϕ0 + Φϕ̄ 2 − V,ϕ a2 δϕ . (8.128)
obtained combining Eq. (4.179) with Eq. (8.115), Eq. (4.185) with Eqs. (8.123)
and Eq. (8.124), Eq. (4.189) with Eq. (8.117). All with Φ = −Ψ, of course.
Though in principle possible, it is a mountain of calculations which we can at
least limit by choosing a different gauge. In particular, following [Weinberg, 2008],
let us consider the following gauge:
ds2 = −a2 (1 + E)dη 2 + 2a2 F,i dηdxi + a2 (1 + A)δij dxi dxj , δ ϕ̂ = 0 , (8.129)
i.e. a gauge in which the perturbed scalar field (which we denote with a hat, in
the new gauge) is zero. Thus in this gauge
A
R= , (8.130)
2
and so our objective is to find a closed equation for A. The price to pay in order
to put the perturbed scalar field equal to zero is to introduce one more scalar
perturbation in the metric. Now, let us calculate the energy-momentum tensor in
the new gauge.
Exercise 8.17. Using Eqs. (8.27) and (8.129), write down the perturbed energy-
momentum tensor. Show that:
1 0
T̂ 0 0 = − 2 (1 − E)ϕ̄ 2 − V (ϕ) , (8.131)
2a
1 0
T̂ 0 i = 0 , T̂ i 0 = 2 F,i ϕ̄ 2 , (8.132)
a
i i 1 02
T̂ j = δ j (1 − E)ϕ̄ − V (ϕ) . (8.133)
2a2
The above is a pretty simple energy-momentum tensor and indeed we now are
going to exploit its simplicity. First of all, since δ T̂ 0 i = 0, we have from Eq. (4.43)
that:
− HE + A0 = 0 , (8.134)
i.e. a simple algebraic relation between two of the three scalar potentials.
The second relation that we are going to us is the continuity equation.
214
Exercise 8.18. Compute the continuity equation ∇ν T ν 0 = 0 using the energy-
momentum tensor given in Eqs. (8.131)-(8.133). Compute from the metric in
Eq. (8.129) the only necessary Christoffel symbol:
A0 i
Γi0j = Hδ i j + δj. (8.135)
2
Then show that:
0
a2 E 2

0 2 0 2 3 0
(H − H ) + (H − H ) ∇ F + 3HE − A = 0 , (8.136)
2 a2 2
0
where we have used the relation H2 − H0 = 4πGϕ̄ 2 .
As last relation, we shall use δRij computed from the metric in Eq. (8.129). So,
using Eq. (4.34), we have:
1 H 1
δRij = − E,ij − E 0 δij − (2H2 + H0 )Eδij − (∇2 Aδij + A,ij )
2 2 2
1 00 5
+ A δij + HA0 δij + (2H2 + H0 )Aδij
2 2
−H∇2 F δij − Fij0 − 2HF,ij . (8.137)
Through the Einstein equations, we know that:

2 T̄ 2 δT
δRij = 8πG δTij − a Aδij − a δij , (8.138)
2 2
where T̄ and δT are the background and perturbed trace, respectively, of the
energy-momentum tensor.
Exercise 8.19. Calculate the right hand side of the above equation. Show that:
δRij = 8πGa2 Aδij V = (2H2 + H0 )Aδij . (8.139)
Extracting thus only the part of δRij which is proportional to δij , we finally
get:
H 0 1 1 5
− E − (2H2 + H0 )E − ∇2 A + A00 + HA0 − H∇2 F = 0 . (8.140)
2 2 2 2
Exercise 8.20. Combine Eqs. (8.134), (8.136) and (8.140), eliminating E and F
and thus finding the following equation for R:
H (H2 − H0 )0

00 2 0 2 0
R + R H −H + + k2R = 0 . (8.141)
H 2 H2 − H0
215
Show that this equation can be written in a more compact form as:
z0
R00 + 2 R0 + k 2 R = 0 (8.142)
z
with
aϕ̄0
z≡ . (8.143)
H
Equation (8.142) is called Mukhanov-Sasaki equation, cf. [Mukhanov, 1985]

and [Sasaki, 1986]. Notice that this equation has the very same structure as
Eq. (8.67) if one makes the change a → z. Therefore, a similar analysis applies
and a constant solution for R is allowed when:
k 2 |z 00 /z| , (8.144)
which is a similar, but not identical, condition as k 2 |a00 /a|. In fact, note that:
2
ϕ̄0 3ϕ̄02

2 2
z =a = a2 02 2
= a2 , (8.145)
H 8πG(ϕ̄ /2 + V a ) 4πG
where we have used Eq. (8.44), written in the conformal time. The above relation
between z and a is exact and it calls into play the slow roll parameter .
Now, the procedure that we need in order to obtain the power spectrum for
R is the same as the one used for h. We promote R to a quantum field and
compute its vacuum expectation value, which will become the power spectrum
itself. The vacuum expectation value is computed first by solving the Mukhanov-
Sasaki equation, choosing the correct Minkowski behaviour at large k.
Finally, we can use the same result of Eq. (8.112), remembering to divide by
the 32πG factor that we have introduced in order to give dimensionality to h, and
multiplying by a2 /z 2 = 4πG/, in order to recover the 2z 0 /z factor in front of R0 ,
instead of 2H.
Thus, we can finally write down the scalar power spectrum as:
H2 H2
PR = 2
∆2R = 2
(8.146)
4MPl k 3 k=aH 8π 2 MPl k=aH
This is the most important result of this chapter, since it provides a prediction
which can be (and indeed is) tested observationally. Using Eq. (6.94), one can
relate this power spectrum to those of the other quantities which become relevant
at horizon re-entering during the radiation-dominated epoch. For example, using
Eq. (6.77), we can write that:
4(5 + 2Rν )2 2
∆2Φ = ∆ . (8.147)
(15 + 4Rν )2 R
In general, Φ is not constant on large scales during inflation, but it is indeed

constant on large scales during radiation domination, as we saw in Chapter 6. The
216
above thus constitutes the initial condition, on the power spectrum, at horizon
re-entering.
Now, let us perform a calculation similar to that in the previous section, cul-
minated in Eq. (8.112).
Exercise 8.21. Employing Eq. (8.41) written in the conformal time, show that:
z0
= H(1 + − η) . (8.148)
z
Afterwards, assume and η to be constant. Then, use Eq. (8.97) and recast the
Mukhanov-Sasaki equation as follows:
2(1 + 2 − η) 0
R00 − R + k2R = 0 , (8.149)
τ
where we are now employing τ as conformal time, in order to avoid confusion with
the second slow-roll parameter η. Finally, show that the above equation can be
written as:
2 00 2 2 + 6 − 3η
(z R) + k − (z 2 R) = 0 . (8.150)
τ2
This equation has the same form as Eq. (8.99), and hence a similar analysis
applies. In particular, its solution is of the form:
√ √
z 2 (τ )R(τ, k) = C1 (k) −τ Hν(1) (−kτ ) + C2 (k) −τ Hν(2) (−kτ ) , (8.151)
with order √
3p
ν= 3 + 8 − 4η . (8.152)
2
Exercise 8.22. Following the same steps that we saw in the tensor case, show
that:
1 1
∆2R (τ, k) = 2 2 2 2 (k|τ |)−4+2η . (8.153)
8π MPl a τ
For the scalar case a different k-dependence of the spectrum appears, involving
the second slow-roll parameter. The combination −4 + 2η is the first order ex-
pression of the scalar spectral index. Evaluating the above spectrum at horizon
crossing, we get the result already shown in Eq. (8.146).
8.6 Spectral indices

It is customary to cast the dimensionless scalar and tensor power spectra as follows:
nS (k)−1
2 2 k 3 PR (k) H2 k
∆S ≡ ∆R ≡ 2
= 2 2
≡ AS , (8.154)
2π 8π MPl k=aH k∗
nT (k)
2 2 k 3 Ph (k) 2H 2 k
∆T ≡ 2∆h ≡ = 2
≡ AT , (8.155)
π2 π 2 MPl k=aH k∗
217
where the general k-dependence (given by the specific model of inflation) is embed-
ded in nS (k) and nT (k), which are known as scalar spectral index and tensor
spectral index.
We have introduced here the pivot scale k∗ , which is usually taken to be 0.002
Mpc−1 or 0.05 Mpc−1 [Ade et al., 2016c] and the factor 2 in ∆2T ≡ 2∆2h takes into
account the two polarisations of the GW. The spectral amplitudes AS and AT ,
of which only the first is constrained by observation since we do not detect yet the
primordial GW background, are related to the energy scale of inflation.
Since the spectral indices are not necessarily constant, the spectra can be writ-
ten in the following more general form:
" 2 #
∆2S 1 d2 nS

1 dnS k k k
ln = nS − 1 + ln + ln + · · · ln , (8.156)
AS 2 d ln k k∗ 6 d(ln k)2 k∗ k∗
∆2T

1 dnT k k
ln = nT + ln + · · · ln . (8.157)
AT 2 d ln k k∗ k∗
The derivative of the spectral index with respect to ln k is called running of the
spectral index. We do not consider runnings in detail here, for simplicity, and
Planck has not constrained them very tightly, as we shall see. However, they (at
least the first one) will probably become of great interest in future CMB experi-
ments [Muñoz et al., 2017].
From the above Eq. (8.156) at first order, it is straightforward to write:
d ln ∆2S d ln ∆2T
nS − 1 = , nT = . (8.158)
d ln k d ln k
For the tensor case:
k dH
nT = 2 . (8.159)
H dk aH=k
Recall that H does not depend on the scale in general, being it a background
quantity, but it is the evaluation at aH = k that makes H depending on k. This
is because different scales cross the horizon at different times during inflation. In
particular, recall from Eq. (8.83) that:
Z a
1 a da
Z
da 1
η= 2
≈ 2
=− , (8.160)
0 Ha H 0 a Ha
since H is almost constant during inflation. Now, when we evaluate aH = k, we
have that η = −1/k. Therefore, we obtain for the tensor spectral index:
k dH dη 1 dH
nT = 2 =2 . (8.161)
H dη dk aH=k kH dη aH=k
By definition of , cf. Eq. (8.36), we know that:

dH
= −aH 2 , (8.162)
dη
so that we can finally determine:
nT = −2 (8.163)
218
which is the result already derived in Eq. (8.107).
For the scalar spectral index is pretty much the same calculation:
d ln(H 2 /) d ln
nS − 1 = = −2 − . (8.164)
d ln k aH=k d ln k aH=k
We can deal with the last term as follows:

d ln k 0 dη 1 0
= = . (8.165)
d ln k aH=k dk aH=k k aH=k
Using Eq. (8.41) one has:

0 = 2aH( − η) , (8.166)
and the scalar spectral index is thus written as:
nS − 1 = −4 + 2η = −6V + 2ηV (8.167)
again in agreement with the result found in Eq. (8.153).

Finally, define the tensor-to-scalar ratio:
∆2T (k∗ ) AT
r∗ ≡ 2
= = 16 = −8nT (8.168)
∆S (k∗ ) AS
It is customary [Ade et al., 2016c] to use r in order to express the energy scale of
Inflation at the time when the pivot scale exits the Hubble radius. From Eq. (8.155)
we have:
π 2 MPl
2
H∗2 = ∆2T (k∗ ) . (8.169)
2
Using the definition of r∗ and Eq. (8.154), one gets:
π 2 MPl
2
π 2 MPl
2
H∗2 = r∗ ∆2S (k∗ ) = r∗ AS , (8.170)
2 2
from which:
3π 2 MPl
4
V∗ = r∗ AS (8.171)
2
8.7 Observational results

In Ref. [Ade et al., 2016c] one can find recent constraints on the inflationary pa-
rameters. They slightly change with respect to the different datasets considered.
We report here the spectral index and its running and running of the running at
68% CL:
nS = 0.9586 ± 0.0056 , (8.172)

dnS
= 0.009 ± 0.010 , (8.173)
d ln k
d2 nS
= 0.025 ± 0.013 , (8.174)
d(ln k)2
219
using the pivot scale k∗ = 0.05 Mpc−1 . For the scalar amplitude at 68% CL:
ln(1010 AS ) = 3.094 ± 0.034 . (8.175)
For the tensor-to-scalar ratio:

r0.002 < 0.10 , (8.176)
at 95% confidence level. From these numbers, we can write the energy scale of
Inflation at the time when the pivot scale exits the Hubble radius as follows:
4 r
V∗ = 1.88 × 1016 GeV , (8.177)
0.10
and therefore obtain an upper bound, which corresponds to GUT energy scale.
Figure 8.2: This is Fig. 12 of [Ade et al., 2016c] and shows the marginalised
joint 68% and 95% CL regions for ns and r at k = 0.002 Mpc−1 from Planck
compared to the theoretical predictions of selected inflationary models. Note that
the marginalised joint 68% and 95% CL regions have been obtained by assuming
dns /d ln k = 0.
In Fig. 8.2 we borrow (with permission, of course) Fig. 12 of [Ade et al., 2016c]
showing the constraints on some inflationary models in the parameter space ns vs.
r. Note here that N∗ is the number of e-folds to the end of inflation, which is
limited in the range 50 − 60 because the observed scales, in particular the pivot
one k = 0.002 Mpc−1 must have had time to leave the horizon and then to re-enter.
8.8 Examples of models of inflation

We conclude this Chapter presenting some important models of inflations, some
of those in Fig. 8.2, limiting ourselves to the class of single field models. There
is a huge number of inflationary models, in [Martin et al., 2014] 193 of them are
220
analysed, and it seems that data favour the simplest category of single field slow-
roll inflation that we have presented in this chapter. See also [Tsujikawa, 2014] for
a very nice review on many inflationary models.
8.8.1 General power law potential

A very simple model of inflation consists in an inflaton field described by the
potential:
ϕn
V (ϕ) = λn , (8.178)
n
from which one can easily determine the slow-roll parameters from Eqs. (8.45) and
(8.47):
n2 n2 MPl
2
n(n − 1) 2
n(n − 1)MPl
V = = , ηV = = . (8.179)
16πGϕ2 16πϕ2 8πGϕ2 8πϕ2
In order for these parameters to be very small and thus trigger inflation, one has
to have ϕ MPl and note that this condition does not depend on the coupling.
We can put the predictions on the spectral indices as functions of the number of
e-folds to the end of inflation by defining the value of the field at which inflation
ends as V (ϕf ) = 1.
For the general power law potential, this condition becomes:
n2 n
=1 ⇒ ϕf = √ . (8.180)
16πGϕ2f 16πG
The number of e-folds to the end of inflation can be obtained exactly from Eq. (8.50):
8πG ϕi n2
Z
4πG 2
N= ϕ dϕ = ϕi − , (8.181)
n ϕf n 16πG
from which the initial value of the field is:
2 n n
ϕi = N+ . (8.182)
4πG 4
From this, the slow-roll parameters can be written as:
n 2(n − 1)
V = , ηV = , (8.183)
4N + n 4N + n
and the scalar spectral index and the tensor-to-scalar ratio are the following:
2(n + 2) 16n 8n
nS − 1 = −6V + 2ηV = − , r = 16V = = (1 − nS ) .
4N + n 4N + n n+2
(8.184)
If we substitute Eq. (8.182) into Eq. (8.178), we get:
V (ϕ) ≈ λn nn/2−1 N n/2 MPl
n
. (8.185)
4
In order for the classical treatment to be valid, we need V (ϕ) MPl and hence:
4−n
MPl
λn . (8.186)
nn/2−1 N n/2
For n = 4 and N = 60 we have λ4 10−5 .
221
8.8.2 The Starobinsky model
The Starobinsky model is actually a f (R) model, hence a modification of GR,
whose action is
2 Z √
MPl
S= d4 x −gf (R) , (8.187)
2
where f (R) is a generic function of the Ricci scalar. When f (R) = R we recover
GR. The above action can be recast as follows:
2
√
Z
4 MPl
S = d x −g ϕR − V (ϕ) , (8.188)
2
i.e. as a non-minimally coupled scalar-tensor theory, where the scalar field

ϕ and its potential are defined as:
M2Pl

df df
ϕ ≡ MPl , V (ϕ) ≡ R −f . (8.189)
dR 2 dR
This kind of theory has a non-minimal coupling because of the term ϕR (the
√
minimal coupling to geometry occurs just with −g). The action written as in
Eq. (8.188) is also said to be in the Jordan frame and corresponds to the Brans-
Dicke theory [Brans and Dicke, 1961] with a special choice for its free parameter
(i.e. ω = 0).
Exercise 8.23. Performing the conformal transformation

df ϕ
ĝµν = gµν = gµν , (8.190)
dR MPl
show that: 2
√
Z
4 MPl 1 µν
Ŝ = d x −g R̂ − ĝ ∂µ χ∂ν χ − U (χ) , (8.191)
2 2
where R̂ is the Ricci scalar computed from ĝµν and:
r
2 2

V MPl MPl df 3 ϕ
U≡ = R −f , χ≡ MPl ln . (8.192)
ϕ2 2(df /dR)2 dR 2 MPl
Note that, in order for the above definition of χ to make sense, df /dR > 0.
The action Ŝ is said to be in the Einstein frame. Thus, any f (R) theory can be
reformulated as a scalar-tensor theory, the scalar field being ϕ ≡ MPl df /dR. The
f (R) theories have been intensively studied in the past decades, see the compre-
hensive reviews [Sotiriou and Faraoni, 2010] and [De Felice and Tsujikawa, 2010].
Since any f (R) can be conformally mapped in GR plus a canonical scalar
field, it seems natural to investigate how this kind of modified gravity accounts for
inflation. Indeed, probably the most successful application of a f (R) theory is in
the inflationary paradigm, as we are going to show in a moment.
222
Now it is χ in action (8.191) which plays the role of the inflaton. The Starobin-
sky model [Starobinsky, 1979] is given by a quadratic R2 correction to the usual
Einstein-Hilbert term in the action for gravity:
R2
f (R) = R + . (8.193)
6M 2
Then, one can easily calculate from Eq. (8.192):
r
2
3MPl M 2 R2 3 R + 3M 2
U= , χ = MPl ln . (8.194)
4(R + 3M 2 )2 2 3M 2
Exercise 8.24. Invert the above relation for χ(R), obtaining thus a R(χ), and
substitute it into the expression for U (χ), obtaining thus the expression for the
inflaton potential in the Starobinsky model:
3 2 2 √ 2
U (χ) = MPl M 1 − e− 2/3χ/MPl (8.195)
4
��
��
��
��
��
��
��
� � � � � ��
Figure 8.3: Plot of the Starobinsky potential, in Eq. (8.195).
In Fig. 8.3 we plot the Starobinsky potential U (χ) as function of the scalar field
χ. The most important feature is the plateau at large values of the field which is
the ideal feature for a slow-roll evolution of the inflaton field.
223
The slow-roll parameters are easily calculated from Eq. (8.195) and Eqs. (8.45)
and (8.47):
√ √ √
−2 2/3χ/MPl −2 2/3χ/MPl − 2/3χ/MPl
4 e 4 2e −e
U = √ 2 , ηU = √ 2 . (8.196)
3 3

1 − e− 2/3χ/MPl 1 − e− 2/3χ/MPl
The number of e-folds to the end of inflation is obtained by solving the following
integral:
dχ √2/3χ/MPl 3 √
r Z χ
3
N= e − 1 ≈ e 2/3χ/MPl , (8.197)
8 χf MPl 4
so that we can write the slow-roll parameters as:
12 3 9 − 4N 1
U = 2
≈ , ηU = 4 ≈− , (8.198)
(4N − 3) 4N 2 (4N − 3)2 N
where we have kept the dominant contributions for large N . The scalar spectral
index and the tensor-to-scalar ratio are thus written, using Eqs. (8.167) and (8.168),
as follows:
2 12
nS = 1 − , r= 2 . (8.199)
N N
Substituting N = 50 and 60, the predictions obtained are in excellent agreement
with the Planck constraints shown in Fig. 8.2, making thus the Starobinsky model
quite successful.
224
Chapter 9
Evolution of perturbations
On peut braver les lois humaines, mais non résister aux lois na-
turelles
(One can challenge human laws, but not the natural ones)
—Jules Verne, Vingt mille lieues sous le mers
In this Chapter we solve exactly some of the equations that we found in Chap-
ter 4, using some approximations. In particular, we can distinguish 4 cases of
evolution:
1. On super-horizon scales,
2. In the matter-dominated epoch,
3. In the radiation-dominated epoch,
4. Deep inside the horizon,
for which it is possible to perform analytic calculations and thus gain a clearer
physical insight.
Our scope is to understand the shape of the matter power spectrum, plotted in
Fig. 9.1 from the analysis of the SDSS DR5 (Data Release 5) performed in [Percival
et al., 2007]. Since we already know the form of the primordial power spectrum, cf.
Eq. (8.154), the above task amounts to determine the matter, CDM plus baryons,
transfer function.
It must be noted that the data points in Fig. 9.1 are derived from the observation
of the distribution of galaxies in the sky and hence provide information on the
galaxy density contrast δg , which is in general a biased tracer of the underlying
distribution of matter in the sense that the galaxy correlation function is not equal
to the total matter correlation function [Kaiser, 1984]. At low redshift this bias is
usually considered as a constant
δg = bδm , (9.1)
with δm being the matter density contrast (baryonic plus CDM), but for larger
redshift it might be a function of redshift and of the wavenumber, i.e. b = b(k, z).
225
Beyond the cosmic variance affecting in a relevant way any large-scale obser-
vation, the determination of the power spectrum is also afflicted by another noise,
which is called shot noise and is due to the fact that δg is given by a discrete
distribution (that of galaxies) tracing a continuous one (that of the underlying
matter).
Finally, spectra such as that in Fig. 9.1 are 3-d, in the sense that they are
computed from the spatial distribution of galaxies. While determining the angular
positions on the celestial sphere is not complicated, the only direct measure of dis-
tance that we have is the redshift. It is possible, of course, to transform the redshift
into an actual distance (a proper distance, for example) through the cosmological
model that we want to test, but determining redshift is time-consuming, especially
if it is done via spectroscopy, and introduces extra errors due to peculiar motions
and to photometry (if z is determined photometrically). Hence, it is perhaps more
convenient to work with a 2-d power spectrum, the angular one w(θ), since angular
positions on the celestial sphere are easily, precisely and rapidly determined.
9.1 Evolution on super-horizon scales

A given comoving wavenumber k is super-horizon at a certain conformal time η, if
kη 1 (9.2)
Since usually H ∝ 1/η, the above condition amounts to:
kH (9.3)
which can be rewritten for the physical scale as follows:
k
H (9.4)
a
The super-horizon regime is the same one that we used in Chapter 6 when we
investigated the primordial modes. The main difference with what we are going
to see here is that we do not limit ourselves to the radiation-dominated epoch,
but investigate what happens through radiation-matter equality and also through
matter-DE equality.
Thus, the above conditions can be written as follows in the epochs of interest:
1
kH= ⇒ kη 1 , (9.5)
η
during the radiation-dominated epoch (for which a ∝ η),
2
kH= ⇒ kη 2 , (9.6)
η
during the matter-dominated epoch (for which a ∝ η 2 ) and
HΛ
kH= , (9.7)
1 + HΛ (η0 − η)
226
Figure 9.1: Matter power spectrum, from [Percival et al., 2007].
for the Λ-dominated era. This is similar to the inflationary phase, but with η > 0,
hence the above expression for the conformal Hubble factor.
We use Eqs. (6.21)-(6.24):
δγ0 = −4Φ0 , δν0 = −4Φ0 , δc0 = −3Φ0 , δb0 = −3Φ0 , (9.8)
227
the Poisson equation, written using Eq. (6.59) as follows:
3 k2 3
(Φ0 − HΨ) + 2 Φ = (ρc δc + ρb δb + ργ δγ + ρν δν ) , (9.9)
H H 2ρtot
where we put in evidence the k 2 /H2 factor that we are going to neglect, and the
anisotropic stress equation (4.184):
k 2 (Φ + Ψ) = −32πGa2 ρν N2 . (9.10)
We could use again Eq. (6.59) in order to put in evidence the k 2 /H2 term on the
left, but it is not necessary because the neutrino quadrupole is of the same order
and thus we need to perform differentiations as we did in Chapter 6 in order to get
a closed equation for the potentials.
Note that we have neglected the photon quadrupole contribution. It is an
approximation motivated by the fact that before recombination the tight coupling
with electrons washes out Θ2 , whereas after radiation-matter equality Rγ becomes
rapidly (∝ 1/a ∝ 1/η 2 ) negligible.
When matter dominates, also Rν is negligible, so we expect the potentials to
become equal. Since our objective here is to perform an analytic calculation, we
assume already Φ = −Ψ. This is incorrect, strictly speaking, when considering the
radiation-matter domination transition but it is fine when considering the matter-
DE one. Therefore, let us rewrite Eq. (9.9) as follows:
3H2
3H (Φ0 + HΦ) = (ρc δc + ρb δb + ργ δγ + ρν δν ) . (9.11)
2ρtot
This equation holds true also in presence of DE, provided that the latter does not
cluster (e.g. it is Λ).
9.1.1 Evolution through radiation-matter equality

In presence of radiation and matter only, Friedmann equation can be written as
follows:
8πGa2

2 8πG 1
H = (ρm + ρr ) = ρm 1 + , (9.12)
3 3 y
where we have grouped together the species which evolve in the same way and
indicated them with a subscript r, i.e. radiation (photons and neutrinos) and
with a subscript m, i.e. matter (CDM and baryons). We have also employed the
definition:
ρm a
y≡ = , (9.13)
ρr aeq
where aeq is the equivalence scale factor.
Exercise 9.1. Assume adiabaticity, i.e.

3 3
δc = δb = δγ = δν ≡ δm . (9.14)
4 4
228
Rewrite Eq. (9.11) using y as independent variable. Show that:
dΦ 4 + 3y
y +Φ= δm . (9.15)
dy 6(y + 1)
Exercise 9.2. We are looking now for a closed equation for Φ. Therefore, differ-
0
entiate Eq. (9.15) with respect to y and use δm = −3Φ0 in order to find:
d2 Φ (7y + 8)(3y + 4) + 2y dΦ 1
2
+ + Φ=0 (9.16)
dy 2y(y + 1)(3y + 4) dy y(y + 1)(3y + 4)
Exercise 9.3. Quite unexpectedly, the above equation (9.16) can be solve exactly.
Indeed, use the following transformation [Kodama and Sasaki, 1984]:
y3Φ
u≡ √ , (9.17)
1+y
and show that
d2 u

2 3 3 du
+ − + − =0. (9.18)
dy 2 y 2(y + 1) 3y + 4 dy
Exercise 9.4. Integrate once Eq. (9.18) and show that:

du 3
ln = C1 + 2 ln y − ln(y + 1) + ln(3y + 4) , (9.19)
dy 2
where C1 is an integration constant. By exponentiating and integrating again show
that: Z y
y3Φ y 2 (3y + 4)
√ =A dy , (9.20)
1+y 0 (1 + y)3/2
where A is an integration constant related to C1 and assume that y 3 Φ → 0 for
y → 0.
Exercise 9.5. Solve the above integration and show that:

ΦP p 3 2

Φ(y) = 16 1 + y + 9y + 2y − 8y − 16 , (9.21)
10y 3
where ΦP is the primordial gravitational potential, which we introduced in Eq. (6.77)
in place of Cγ = R = ζ. Show that Φ(y → 0) = ΦP .
From the above solution we see that for y → ∞ (which here means deep into
the matter-dominated epoch) the gravitational potential drops of 10%, i.e.
9
Φ→ ΦP , for y → ∞ . (9.22)
10
229
��
��
��
��
��
��
��
Figure 9.2: Evolution of the gravitational potential Φ on super-horizon scales (k =

0) through radiation-matter equality. From Eq. (9.22).
In Fig. 9.2 we display the evolution of Φ/ΦP as function of y, as given by Eq. (9.22).
It can be shown that Φ is constant on super-horizon scales for any background
evolution with w 6= −1 constant and for adiabatic perturbations [Mukhanov et al.,
1992].
Exercise 9.6. Setting Φ = −Ψ, combine Eq. (4.179) with Eq. (4.189) and use
Eq. (4.129). Show that:
Φ00 + 3H 1 + c2ad Φ0 + 2H0 + H2 1 + 3c2ad Φ + k 2 c2ad Φ = −4πGa2 Γ . (9.23)

where we have defined c2ad ≡ P 0 /ρ0 as the adiabatic speed of sound. Show that
this equation can be cast as follows:
θ00

00 2 2
u + k ca − u = −a2 (ρ + P )−1/2 Γ (9.24)
θ
where 1/2
Φ 1 ρ
u≡ , θ≡ . (9.25)
4πG(ρ + P )1/2 a ρ+P
It is clear that the above transformation cannot treat the cosmological constant,
for which P = −ρ. Nevertheless, it is interesting to see that Eq. (9.24) has a form
similar to Eq. (8.71) and hence, for k = 0 and Γ = 0 (adiabatic perturbations) its
general solution is: Z
dη
u = C1 θ + C2 θ . (9.26)
θ2
230
Exercise 9.7. Consider a single fluid model P = wρ, with w constant and different
from −1. Show, from solving the Friedmann equation, that:
a = (η/η0 )2/(1+3w) , (9.27)
where w 6= −1/3 (in this case the solution grows exponentially with the conformal
time). Show then that: Z
dη a
θ 2
∝ , (9.28)
θ H
and thus show, using Eq. (9.25), that Φ is constant.
Hence the 9/10 drop of Φ between the radiation-dominated era and the matter-
dominated one can be extended to any kind of adiabatic fluid with w 6= −1 con-
stant, using the constancy of R on large scales. See e.g. [Mukhanov, 2005]. Indeed,
from (12.33) we have
Φ0 − HΨ 2 H−1 Φ0 + Φ
R=Φ+H = Φ + . (9.29)
4πGa2 (ρ + P ) 3 1+w
Now, assume that w changes from a constant value wi to another constant value
wf . For each of the two cases Φ is a constant, Φi and Φf , respectively. Then,
taking advantage of the constancy of R, we can say that:
2 Φi 2 Φf
Φi + = Φf + , (9.30)
3 1 + wi 3 1 + wf
i.e
5 + 3wi 1 + wf
Φf = Φi , (9.31)
5 + 3wf 1 + wi
with which we can easily check the result of Eq. (9.22).
9.1.2 Evolution in the Λ-dominated epoch

As already mentioned, we cannot use the above formulae for the case of greatest
interest, which is for w = −1, i.e. the cosmological constant. In this case, we have
to start directly from Eq. (9.23). Since P = −ρ and constant, then P 0 = 0 and
thus cad = 0.
Exercise 9.8. Using Eq. (9.7) into Eq. (9.23) with cad = 0 and Γ = 0, show that:
3HΛ 3HΛ2
Φ00 + Φ0 + Φ=0. (9.32)
1 + HΛ (η0 − η) [1 + HΛ (η0 − η)]2
Note that η does not go to infinity, but to a maximum value η∞ = η0 −1/HΛ for
which the scale factor diverges. Moreover, note that this equation, and its solution,
are valid also for small scales because for cad = 0 the k-dependence is suppressed.
231
Exercise 9.9. Find the relation between the cosmic time t and the conformal time
using Eq. (9.7) and show that η∞ = η0 − 1/HΛ corresponds to an infinite t.
Exercise 9.10. Change variable to the scale factor in Eq. (9.32), and show that:
d2 Φ 5 dΦ 3
2
+ + 2Φ = 0 . (9.33)
da a da a
Solve this equation, using a power-law ansatz, and show that:
Φ = C1 a−1 + C2 a−3 . (9.34)
Hence, when Λ dominates, Φ is not constant on super-horizon scales but van-

ishes rapidly, as Φ ∝ 1/a or Φ ∝ 1 + HΛ (η0 − η).
9.1.3 Evolution through matter-DE equality

In order to study the transition between the matter-dominated epoch and the
DE-dominated one, we consider Eq. (9.11) neglecting radiation:
3H2 ρm
3H (Φ0 + HΦ) = δm , (9.35)
2ρtot
where we have already considered adiabatic perturbations. Now, let us introduce
the following variable:
−3wx
ρx ρ̃(a/ax )−3(1+wx ) a
x≡ = −3
= , (9.36)
ρm ρ̃(a/ax ) ax
where ρx is a DE component with equation of state wx , which we assume constant,
and ax is the scale factor at matter-DE equivalence, for which both the densities
are equal to ρ̃. Note that wx < −1/3, in order to have a useful DE (it has to
produce an accelerated expansion) and recall that, in order for Eq. (9.35) to be
valid, DE must not cluster.
Exercise 9.11. Following the same steps which brought us to Eq. (9.16), find a
closed equation for Φ, using x as independent variable. Show that:
d2 Φ 6wx (x + 1) − 2x − 5 dΦ 1
+ − Φ=0. (9.37)
dx2 6wx x(x + 1) dx 3wx x(x + 1)
This equation can be cast as a hypergeometric equation and thus solved exactly
[Piattella et al., 2014]. Only one of the two independent solutions is well-behaved
for x → 0, i.e. is a constant:

4wx (1 + x) 1 5 6wx
Φ ∝ −2wx + 2 F1 1, 1 − ,1 − , −x →x→0 − . (9.38)
5 3wx 6wx 5
232
��
��
��
��
��
��
Figure 9.3: Evolution of the gravitational potential Φ on super-horizon scales (k =

0) through matter-DE equality, from Eq. (9.38). The solid line represents the
cosmological constant case wx = −1, the dashed line wx = −1.5 and the dash-
dotted one wx = −0.5.
The integration constant has to be picked in order to match with 9ΦP /10.
In Fig. 9.3 we display the evolution of Φ, computed for wx = −1.5, −1, −0.5
from Eq. (9.38).
In Fig. 9.4 we display the evolution of the potentials Φ (solid line) and −Ψ
(dashed-line) for the ΛCDM model with adiabatic initial condition using CLASS.
The wavenumber chosen here is k = 10−4 Mpc−1 , which corresponds to a scale
larger than the horizon today and hence which spent the whole evolution outside
the horizon. Note how the two potentials display a difference at early times, due
to the presence of neutrinos, which are of course taken into account in CLASS.
The investigation of the evolution of scales larger than the horizon today is not
very useful because they are not observable. However, the scales that we do observe
today were outside the horizon in the past. We shall see in the next section that in
the matter-dominated regime the gravitational potential Φ is constant at all scales,
even through horizon-crossing, and thus it is interesting to know the behaviour of
super-horizon modes (the same behaviour is not shared by the density contrast).
The same does not happen when radiation dominates, but instead the gravitational
potential decays and oscillates rapidly for those scales which enter the horizon.
Since the behaviour in the radiation-dominated and in the matter-dominated
epoch is so dramatically different, it is useful to introduce the so-called equiva-
lence wave number, i.e. the wavenumber corresponding to a scale which enters
233
��
��
��
��
��
��
Figure 9.4: Evolution of the potentials Φ (solid line) and −Ψ (dashed-line) for
the ΛCDM model with adiabatic initial condition using CLASS. The wavenumber
chosen here is k = 10−4 Mpc−1 .
the horizon at the equivalence epoch, and thus defined as:
keq ≡ Heq . (9.39)
Neglecting DE, Friedmann equation in presence of radiation and matter is written

as follows:
8πG
H2 = (ρm + ρr )a2 = H02 (Ωm0 a−1 + Ωr0 a−2 ) , (9.40)
3
and from this we can establish that, at equivalence, the conformal Hubble param-
eter has the following expression:
2 16πG ρR0
Heq = = 2H02 Ωr0 (1 + zeq )2 = keq
2
, (9.41)
3 a2eq
or, using the matter density parameter:
2 16πG ρM0
Heq = = 2H02 Ωm0 (1 + zeq ) = keq
2
, (9.42)
3 aeq
from which it is clear that:

Ωm0
1 + zeq =. (9.43)
Ωr0
Using the observed values for the density parameters, we have that:
√
2H0 Ωm0
keq = √ ≈ 0.014 h Mpc−1 (9.44)
Ωr0
234
The behaviour of super-horizon modes is especially important in CMB physics. In-
deed, in Chapter 10 we shall see that for a given multipole ` the CMB temperature
correlation spectrum C` is determined mostly by those wavenumbers which satisfy:
` ≈ k(η0 − η∗ ) = kr∗ ≈ kη0 , (9.45)
where η0 is the present conformal time, η∗ is the one corresponding to recombination
and r∗ ≡ η0 −η∗ is the comoving distance to recombination. The last approximation
is motivated by the fact that, using CLASS and the ΛCDM model, η∗ ≈ 3 × 102
Mpc whereas η0 ≈ 104 Mpc.
The above expression can be manipulated as follows:
η0 1
` ≈ kη ≈ kη √ , (9.46)
η a
where η is a generic past conformal time in the matter-dominated epoch and,
for this reason, we have used a ∝ η 2 . So, e.g. at radiation-matter equality, i.e.
a ≈ 10−4 , the super-horizon scales kη < 1 contribute to the monopoles ` / 100
and at recombination a∗ ≈ 10−3 , ` / 30.
9.2 The matter-dominated epoch

Let us now neglect completely radiation. Being no photon and neutrino quadrupoles,
then Φ = −Ψ and since matter dominates, δPtot = 0.
Equation (4.189) then provides us straightaway with a closed equation for Φ:
Φ00 + 3HΦ0 + 2H0 Φ + H2 Φ = −4πGa2 δPtot = 0 , (9.47)
which can be solved exactly since, being a ∝ η 2 , we have:
6
Φ00 + Φ0 = 0 . (9.48)
η
The general solution is:
Φ(k, η) = A(k) + B(k)(kη)−5 = A(k) + B̂(k)a−5/2 (9.49)
where A(k), B(k) and B̂(k) are functions of k, being the latter equal to B times the
proportionality factor between the conformal time and the scale factor, which are
related by a ∝ η 2 . We have kept a kη dependence above because it is dimensionless,
as Φ, A and B are.
Now, let us neglect the decaying mode (kη)−5 since it disappears very fast. The
important result here is that the gravitational potential is constant at all scales,
through horizon crossing, during matter domination. We shall see a very different
behaviour when radiation dominates.
Hence, provided that we are deep into the matter-dominated epoch, at large
scales, when kη < 1, we can match the results for Φ of this section and the previous
one and see that:
9
A(k < 1/η) = ΦP (k) , (η > ηeq ) . (9.50)
10
235
So, the gravitational potential Φ, for those scales which enter the horizon during
the matter dominated epoch, is a constant with value:
9
Φ(k) = ΦP (k) , (kηeq < 1) (9.51)
10
The condition kηeq < 1, which corresponds to k < keq guarantees that the mode
was outside the horizon at equivalence and hence that it entered during matter
domination.
Considering the generalised Poisson equation with Φ constant, we have:
3H2
3H2 Φ + k 2 Φ = δm , (9.52)
2
where δm is the density contrast of matter, which we define here as:
δm ≡ (1 − Ωb0 )δc + Ωb0 δb , (9.53)
which comes from the fact that, being in the matter-dominated epoch, ρtot ∝ a−3
and Ωc0 + Ωb0 = 1 (of course we keep on considering a spatially flat universe).
We must be careful that even in the matter-dominated epoch, but before re-
combination, for those modes inside the horizon we can have that δc δb because
baryons are tightly coupled to photons and thus δb cannot grow, whereas CDM
fluctuations can. For these modes, soon after recombination, δb becomes equal
to δc or, in other words, baryons fall into the potential wells of CDM. We shall
see this in more detail later. For large scales, those which enter the horizon after
recombination, we have δc = δb = δm (assuming, as usual, adiabaticity).
For the moment, let us focus on the evolution for δm . It follows immediately
that:
k2 k2η2

9ΦP (k)
δm (k, η) = 2A(k) 1 + = 1+ , (k < keq ) . (9.54)
3H2 5 12
Therefore, δm is constant on super-horizon scales, but when a scale crosses the
horizon it starts to grow as δ ∝ η 2 ∝ a, as the plot in Fig. (9.5) shows. Note that
the first equality holds true at any scales, but the second one only for k < keq ,
because only in this regime we are allowed to use Eq. (9.50).
The power spectrum at any time during the matter dominated epoch can thus
be put immediately in relation with the primordial one:
2
k2η2

81
Pδ (k, η) = 1+ PΦ (k) , (k < keq ) . (9.55)
25 12
For nS = 1 the primordial power spectrum goes as PΦ ∝ 1/k 3 , cf. Eq. (8.154), and
thus the above matter one grows linearly with k (when kη > 1). From the above
equation we can read off the matter transfer function, i.e.:
9 k2η2
Tδ (k, η) = , (1/η < k < keq ) . (9.56)
5 12
From this solution we can infer that the larger k is, i.e. the smaller the scale under
consideration is, the more it grows. This scenario is called bottom-up because
236
��
��
��
��
��
�
�
��
Figure 9.5: Evolution of the matter density contrast δm normalised to the primor-
dial potential, in the matter-dominated era. From Eq. (9.54).
smaller scales becomes non-linear before the larger ones. In other words, first small
structures form and then these can merge in order to form larger structures. The
bottom-up scenario is the contrary of the top-down scenario [Zeldovich, 1984] by
which first the largest structures form and then fragmentise in order to form the
smaller ones.
In Eq. (9.56) we can appreciate that the k-dependence and the η one are sep-
arate. The latter is proportional to η 2 ∝ a and is usually called growth factor.
The separation of the k and η dependences takes place because matter has van-
ishing adiabatic speed of sound and the k-dependence of the equation governing
δm comes in multiplied by the gravitational potential, which is a constant (during
matter domination).
We can see this in detail, by combining Eqs. (5.40) and (5.48).
Exercise 9.12. Combine Eqs. (5.40) and (5.48) after putting to zero w, δP and
the anisotropic stresses, which are indeed vanishing for CDM and also for baryons,
after recombination. Show that:
00 0
δm + Hδm = −k 2 Ψ − 3Φ00 − 3HΦ0 (9.57)
In the above equation, the k-dependence enters only through k 2 Ψ and then, in
the matter-dominated epoch we have that:
00 0
δm + Hδm = k2Φ , (9.58)
237
with Φ constant (if we neglect already the decaying mode).
Exercise 9.13. Check that the solution of the above equation is the same as
Eq. (9.54).
The transfer function that we have determined in this section is valid for small
values of k, i.e. k < keq ≈ 0.014 h Mpc−1 , which correspond to very large scales
which we do not actually observe or for which the errors and the cosmic variance
are too large. Indeed, in Fig. 9.1 there are data points only for scales larger than
keq . It is necessary therefore to understand how matter fluctuations behave during
radiation-domination.
9.2.1 Baryons falling into the CDM potential wells

We offer here a simple calculation which should convey the idea of how important
CDM is for structure formation. This is often stated otherwise as the fact that
after recombination, baryons fall into the gravitational potential wells of CDM.
As we anticipated earlier, before recombination baryons were tight-coupled to
photons in the early-times plasma and, when they decouple and their over-densities
are free to grow, in general we have δb δc for those modes which were well inside
the horizon during recombination.
Let us see this more quantitatively. In the same fashion by which we obtained
Eq. (9.58), we can write the following coupled equations for CDM and baryons:
δc00 + Hδc0 = k 2 Φ , (9.59)

δb00 + Hδb0 = k 2 Φ , (9.60)
of which the baryonic one is valid only after recombination. These equations are
coupled since Φ is determined by both the components. Indeed, from the Poisson
equation we have that:
3H2 3H2
(3H2 + k 2 )Φ = (ρc δc + ρb δb ) ≡ δm , (9.61)
2ρtot 2
Now, the two Eqs. (9.59) and (9.60) have solutions (neglecting the decaying mode):
k2η2 k2η2
δc (k, η) = C1 (k) + A(k) , δb (k, η) = C2 (k) + A(k) , (9.62)
6 6
where we have used the constant potential solution for the potential, in Eq. (9.49)
(also here neglecting the decaying mode). Using this solution in Eq. (9.53) and
comparing with Eq. (9.54), we can conclude that:
C1 (k) + Ωb0 [C2 (k) − C1 (k)] = 2A(k) . (9.63)
For large scales kη 1 we already know that δc = δb = δm , because of adiabaticity,

and hence C1 = C2 = 2A.
238
For small scales, at recombination δc δb because baryons were tightly coupled
to photons and thus δb could not grow. This, combined with the crucial fact that
Ωb0 = 0.04 is small, makes δm (and the gravitational potential) dominated by δc .
Hence, again δc = δb soon after recombination, the detailed transient coming from
the decaying modes that we have neglected.
In other words, baryons fall in the potential wells already created by CDM.
Without CDM, δb would grow proportionally to a, by a factor 103 by today, being
of order 10−2 . This is way too small in order for account of the structures that we
observe.
In Fig. 9.6 we plot the evolution of δc (solid line) and δb (dashed line) for k = 1
Mpc−1 computed with CLASS. Note the oscillations in δb , which are related to
the baryon acoustic oscillations (BAO) of the matter power spectrum, cf. the
small box inside Fig. 9.1, but are not the same thing. The oscillations in the plot
of Fig. 9.6 are function of the time (scale factor) evolution, whereas those in the
matter power spectrum of Fig. 9.1 are function of the wavenumber k.
��
��
��
��-�
��-� ��-� ��
Figure 9.6: Evolution of δc (solid line) and δb (dashed line) for k = 1 Mpc−1
computed with CLASS.
The oscillations of Fig. 9.6 are caused by the tight coupling of baryons with
photons before recombination, when structure formation is impossible. On the
other hand, CDM can grow even when radiation dominates and thus, for the chosen
scale k = 1 Mpc−1 , the ratio δc /δb is of about 3 orders of magnitude.
In Fig. 9.7 we plot again the evolution of δc (solid line) and δb (dashed line) for
k = 1 Mpc−1 computed with CLASS, but this time with a negligible amount of
CDM (Ωc0 h2 = 10−6 ). Note how δb grows six orders of magnitude less today than
in the standard case.
239
��
��
��
��-�
��-� ��-� ��
Figure 9.7: Evolution of δc (solid line) and δb (dashed line) for k = 1 Mpc−1
computed with CLASS with a negligible amount of CDM (Ωc0 h2 = 10−6 ).
9.3 The radiation-dominated epoch

Consider now the case of full radiation dominance and neglect δc and δb as source
of the gravitational potentials. Moreover, assume adiabaticity, so that δγ = δν = δr
and neglect the neutrino anisotropic stress, so that Φ = −Ψ.
Since we are deep into the radiation-dominated epoch, then w = c2ad = 1/3 and
thus from Eq. (9.23) we can immediately write:
4 k2
Φ00 + Φ0 + Φ = 0 , (9.64)
η 3
where we have used a ∝ η. Defining
u ≡ Φη , (9.65)
the above equation becomes
k2

2 2
u + u0 +
00
− 2 u=0. (9.66)
η 3 η
√ √
This is a Bessel equation with solutions j1 (kη/ 3) and n1 (kη/ 3), i.e. the spher-
ical Bessel functions of order 1. Since n1 diverges for kη → 0, we discard it as
unphysical.
Recovering the gravitational potential Φ and using the fact that [Abramowitz
and Stegun, 1972]
sin x cos x
j1 (x) = 2 − , (9.67)
x x
and
sin x − x cos x 1
lim 3
= , (9.68)
x→0 x 3
240
we can write √ √ √
sin(kη/ 3) − (kη/ 3) cos(kη/ 3)
Φ = 3ΦP √ . (9.69)
(kη/ 3)3
This solution shows that as soon as a mode k of the gravitational potential enters
the horizon, it rapidly decays as 1/η 3 or 1/a3 while oscillating.
��
��
��
��
��
��
��
Figure 9.8: Evolution of the gravitational potential Φ deep into the radiation-
dominated era. From Eq. (9.69).
In Fig. 9.69 we plot the evolution of the gravitational potential, according to

Eq. (9.69). Note how Φ starts to decay right after kη > 1, i.e. after horizon
crossing.
The goodness of the solution (9.69) plotted in Fig. 9.8 can be appreciated in
Fig. 9.9 where both Φ (solid line) and −Ψ (dashed line) are plotted. Outside
the horizon the two potentials are constant with a difference due to the neutrino
fraction Rν . As soon as they enter the horizon they rapidly decay to zero.
Now, recall Eq. (9.57) that we derived for the matter density contrast. It can
be used in the radiation-dominated epoch, but only for CDM:
1 3
δc00 + δc0 = k 2 Φ − 3Φ00 − Φ0 . (9.70)
η η
Using Eq. (9.64), we can write
1 9
δc00 + δc0 = 2k 2 Φ + Φ0 ≡ S(k, η) . (9.71)
η η
As stated at the beginning of the section, we are so deep into the radiation-
dominated epoch that the matter density contrast does not contribute to the grav-
itational potential but only feels it.
241
��
��
��
��
��
��
��
Figure 9.9: Evolution of the gravitational potentials Φ (solid line) and −Ψ (dashed
line) deep into the radiation-dominated era computed with CLASS for the ΛCDM
model and k = 10 Mpc−1 . Adiabatic perturbations have been used and the initial
values have been normalised to that of Φ.
Using Eq. (9.69) with x ≡ kη as new independent variable, the function S(x)
has the following form:
√ √ √
S(x) 9 (27x − 2x3 ) cos x/ 3 + 3 (5x2 − 27) sin x/ 3
= , (9.72)
k 2 ΦP x5
and the equation for δc becomes:
√ √ √
(27x − 2x3 ) cos x/ 3 + 3 (5x2 − 27) sin x/ 3

d dδ
x = 9ΦP . (9.73)
dx dx x4
The homogeneous part of this equation has a simple solution:
δhom = C1 + C2 ln x , (9.74)
i.e. a constant C1 times a logarithmic contribution. A particular solution is ob-

tained by integrating twice the right-hand side, thus obtaining:
√ √ √ √
9ΦP −x3 Ci x/ 3 + 3 (x2 − 3) sin x/ 3 + 3x cos x/ 3
δpart = , (9.75)
x3
where Ci(z) is the cosine integral function, defined as
Z ∞
cos t
Ci(z) ≡ − dt . (9.76)
z t
242
The above is a particular solution, therefore the integration constants which stem
from the indefinite integration can be incorporated in C1 . The general solution for
δc is then:
√ √ √ √
9ΦP −x3 Ci x/ 3 + 3 (x2 − 3) sin x/ 3 + 3x cos x/ 3
δc = C1 +C2 ln x+ .
x3
(9.77)
For x → 0, we can expand the above solution as follows:

9 ln(3)
+ O x2 , (9.78)

δ(x → 0) = C1 + C2 ln(x) + ΦP −9 ln(x) − 9γ + 6 +
2
where γ is the Euler constant.

Since ln x is divergent for x → 0 and we do not want δc to diverge, we have to
ask:
C2 = 9ΦP . (9.79)
Moreover, we know that δc (x → 0) = 3ΦP /2 when we choose adiabatic initial
conditions (and neglect neutrinos), cf. Eqs. (6.71) and (6.77), thus:
9
C1 = − ΦP [−2γ + 1 + ln(3)] . (9.80)
2
We plot the evolution of δc through horizon-crossing in Fig. 9.10.
For x 1, deep inside the horizon, we can neglect the contribution δpart to the
solution for δc since it decays rapidly. Thus, we can write the density contrast as
follows:
9
δc = − ΦP [−2γ + 1 + ln(3)] + 9ΦP ln x = AΦP ln(Bkη) , (9.81)
2
where
1 ln(3)
A=9, B = exp γ − − ≈ 0.62 . (9.82)
2 2
9.4 Deep inside the horizon

The last domain in which it is possible to analytically solve the equations for the
perturbations is when k H, i.e. deep inside the horizon. We shall neglect
baryons in this calculation (this is imprecise and we shall see why numerically)
and assume Φ = −Ψ (which is a good approximation, on the basis of the results
of the previous section).
The relevant equations are thus the following ones:
δc0 + kVc = −3Φ0 , (9.83)

Vc0 + HVc = −kΦ , (9.84)
k 2 Φ = 4πGa2 ρc δc . (9.85)
In the latter equation we have neglected all the potential terms except for the one
accompanied by k 2 and we have also neglected radiation perturbations. It is not
evident why should we neglect ρr δr with respect ρc δc even when ρr ρc deep into
243
��
��
��
��
�
��
Figure 9.10: Evolution of δc deep into the radiation-dominated era computed from
Eqs. (9.77), (9.79) and (9.80) (solid line) compared with the numerical calculation
performed with CLASS for k = 10 Mpc−1 (dashed line). Note the semi-logarithmic
scale employed.
the radiation-dominated epoch. Neglecting δb , at least, is justified by the fact that

before recombination it behaves as the fluctuation in radiation and afterwards as
that in CDM, while being Ωb always subdominant.
An explanation of why we can neglect ρr δr is offered by Weinberg who shows
that new modes appear (dubbed fast) which rapidly decay and oscillate [Weinberg,
2002]. He also takes into account baryons at first order in Ωb0 .
Exercise 9.14. Use again the variable y ≡ a/aeq and manipulate the three equa-
tions above in order to obtain a single second-order equation for δc :
d 2 δc 2 + 3y dδc 3
+ − δc = 0 . (9.86)
dy 2 2y(y + 1) dy 2y(y + 1)
This equation is known as Mészáros equation [Meszaros, 1974]. A solution

can be found at once, multiplying by 2y(y + 1):
d 2 δc dδc dδc
2y(y + 1) 2
+2 + 3y − 3δc = 0 . (9.87)
dy dy dy
A linear ansatz δc ∝ y kills the second derivative and the last two terms on the left
244
hand side. Therefore, the simple solution we looked for is:
2
D1 (y) ≡ y + (9.88)
3
This is also the growing mode. For y 1, i.e. before matter-radiation equality, δ
is practically constant, whereas for y 1 we have the known growth linear with
respect to the scale factor.
In order to find the other independent solution say D2 , we can use the Wron-
skian:
dD2 dD1
W (y) = D1 − D2 . (9.89)
dy dy
Exercise 9.15. Show that the Wronskian satisfies the simple first-order differential
equation:
dW 2 + 3y
=− W , (9.90)
dy 2y(y + 1)
from which one gets
1
W = √ . (9.91)
y 1+y
Exercise 9.16. From the very definition of the Wronskian in Eq. (9.89), write a
first-order equation also for D2 , which is the following:

2 d D2 1
(y + 2/3) = √ . (9.92)
dy y + 2/3 y 1+y
Integrate it and show that the result is:

√
9p 9 y+1+1
D2 = 1 + y − (y + 2/3) ln √ (9.93)
2 4 y+1−1
This mode grows logarithmically when y 1, recovering the logarithmic solu-

tion of the previous section. It decays as 1/y 3/2 for y 1.
The complete solution for δc on small scales k H and through radiation-
matter equality is then:
δc (k, a) = C1 (k)D1 (a) + C2 (k)D2 (a) . (9.94)
The dependence of C1 and C2 from k can be established by matching this solution

with the one of the previous section in Eq. (9.81).
245
9.5 Matching and CDM transfer function
As we discussed earlier and as it is clear from Fig. 9.1 the today observed scales in
the matter power spectrum are those for which k > keq , i.e. those which entered
the horizon before matter equality.
In the last two section we have obtained exact solution for δc deep into the
radiation-dominated epoch for all scales and through radiation-matter equality at
very small scales. There is then the possibility of matching the two solutions on
very small scales and thus obtain in this regime the CDM transfer function until
today.
Consider the following two solutions found in the previous sections, i.e.
δc (k, η) = AΦP (k) ln(Bkη) , (9.95)

δc (k, a) = C1 (k)D1 (a) + C2 (k)D2 (a) , (9.96)
which are valid on very small scales, i.e. kη 1. The purpose is to find the
functional forms of C1 (k) and C2 (k).
Using Eqs. (9.40) and (9.41), one can approximate the Hubble parameter deep
into the radiation era as follows:
Heq aeq
H≈ √ , (9.97)
2a
and solving using the conformal time, one has:
Heq aeq
a= √ η. (9.98)
2
The proportionality constant is the correct one which gives H = 1/η when substi-
tuted in the approximated formula for H.
We can thus write the logarithmic solution for δc , deep into the radiation-
dominated era, as follows:
√ !
2a
δc (k, a) = AΦP (k) ln Bk . (9.99)
Heq aeq
Introducing y ≡ a/aeq , the equivalence wavenumber keq = Heq and the rescaled
wavenumber: √ √
2k k Ωr0 k
κ≡ = = (9.100)
keq H0 Ωm0 0.052 Ωm0 h2 Mpc−1
one has:
δc (k, a) = AΦP (k) ln (Bκy) . (9.101)
Recall that this solution is valid deep into the radiation-dominated epoch, thus
y 1. At the same time, it holds true only on very small scales, i.e. kη 1.
Using the above formulae, this means:
√ √
2a 2a
kη = k =k = κy 1 . (9.102)
Heq aeq keq aeq
246
Therefore, if we want to match radiation-domination solution with the solution of
the Mészáros equation, we need to choose a suitable ym at which performing the
junction of the two solutions such that 1/κ ym 1.
Asking the equality of the two solutions and their derivatives at ym implies to
solve the following system:
AΦP ln(Bκym ) = C1 (ym + 2/3) + C2 D2 (ym ) , (9.103)

AΦP dD2
= C 1 + C2 . (9.104)
ym dy y=ym
Exercise 9.17. Solve the above system and take the dominant contribution for
ym → 0 (because the junction condition has to be imposed for ym 1). Show
that:
3 2
C1 = AΦP ln 4Bκe−3 ,

C2 = AΦP . (9.105)
2 3
Therefore, the solution for δc valid at all time and at small scales is the following:
3 √ 2
−3
δc (k, y) = AΦP ln 4 2Bκe (y + 2/3) + AΦP D2 (y) , (κ 1) . (9.106)
2 3
We can project this solution at late times, when matter dominates and thus neglect
the decaying mode D2 and write:
√ !
3 4 2Be−3 k a
δc (k, a) = AΦP ln , (a aeq , k keq ) (9.107)
2 keq aeq
Note again the separated dependence from k and from the scale factor, being the
growth function D(a) = a. This allows us to write the transfer function for CDM
as follows, using Eq. (9.42) in order to eliminate aeq :
2
√ !
Akeq 4 2Be−3 k
Tδ (k) = ln D(a) , (a aeq , k keq ) (9.108)
2H02 Ωm0 keq
Recall that the factor 3ΦP /2 is the adiabatic initial condition on δc and thus enters
the primordial power spectrum. The above transfer function can be generalised
including DE, if the latter does not cluster. If it is the case, as for the cosmological
constant, its effect enters only the growth factor and in keq since, being another
component and having the fixed total Ωr0 + Ωm0 + ΩΛ = 1, the relative amount of
radiation and matter has to change (usually the amount of radiation is fixed) and
thus keq changes.
We can also determine the transfer function for the gravitational potential Φ
at late times. From Eq. (9.85) we have that:
3H02 Ωm0
k 2 Φ = 4πGa2 ρm δm = δm , (9.109)
2a
247
where we have included baryons, since we are considering late times and we have
seen that after recombination δb = δc . We are neglecting any DE contribution
in the expansion of the universe and considering a radiation plus matter model.
Hence, Ω0 ≈ 1.
We thus have for the gravitational potential, combining Eqs. (9.107) and (9.109):
2
√ !
9Akeq 4 2Be−3 k
Φ(k) = ΦP (k) ln , k keq (9.110)
8k 2 keq
The transfer function for the gravitational potential is usually normalised to 9ΦP /10
and therefore:
2
√ !
5Akeq 4 2Be−3 k
TΦ (k) = ln , k keq (9.111)
4k 2 keq
Exercise 9.18. Using Ωr0 h2 = 4.15 × 10−5 in Eq. (9.44) and defining
Mpc
q≡k× , (9.112)
Ωm0 h2
show that we can cast the above transfer function in the following form:
ln (2.40q)
TΦ (k) = (9.113)
(4.07q)2
See also [Weinberg, 2008, pag. 310].
We have written the transfer function as in Eq. (9.113) because it is simpler to

compare it with the numerical fit of Bardeen, Bond, Kaiser and Szalay (BBKS) of
the exact transfer function [Bardeen et al., 1986], which is the following:
ln (1 + 2.34q) −1/4
TBBKS (k) = 1 + 3.89q + (16.2q)2 + (5.47q)3 + (6.71q)4 ,
2.34q
(9.114)
and see that for large q it goes as ln(2.34q)/(3.96q)2 , which is in good agreement
with our analytic estimate (9.113).
Given the transfer function TΦ , δm can be written, from Eq. (9.109), as:
2a 2 3k 2 3k 2
δm (k, a) = k Φ= ΦP (k)TΦ (k)a = ΦP (k)TΦ (k)D(a) .
3H02 Ωm0 5H02 Ωm0 5H02 Ωm0
(9.115)
In the last equality we have recovered the growth factor D(a) in order to provide
a more general formula. From this solution we can obtain the power spectrum for
δm starting from the primordial one for Φ, i.e.
9k 4
Pδ (k, a) = 4 2
PΦ (k)TΦ2 (k)D2 (a) . (9.116)
25H0 Ωm0
248
Using Eq. (6.94), with Φ = −Ψ since we are neglecting neutrinos, ΦP = 2R/3 and
thus the primordial power spectrum of Φ can be traded with the R one, and we
get:
4k 4
Pδ (k, a) = PR (k)TΦ2 (k)D2 (a) . (9.117)
25H04 Ω2m0
From Eq. (8.154), we can write the explicit dependence of the primordial power
spectrum of k as follows:
nS −1
8π 2 k

k
Pδ (k, a) = AS TΦ2 (k)D2 (a) (9.118)
25H04 Ω2m0 k∗
The power spectrum Pδ (k, a) can be determined through the observation of the
distribution of galaxies in the sky. Therefore, through the above formula we can
probe many quantities of great interest such as the primordial tilt of the power
spectrum.
Now, let us make a plot of the power spectrum today (z = 0) using CLASS and
fixing all the parameters to the ΛCDM best fit values except for Ωm0 , which we let
free. We show in Fig. 9.11 the shape of the matter power spectrum for Ωm0 = 0.1
(blue) and Ωm0 = 0.3 (yellow), Ωm0 = 0.7 (green) and Ωm0 = 0.99 (red).
��
��
��
��
��-� ��-� ��
Figure 9.11: Matter power spectra for (left to right, in case no colours are available)
Ωm0 = 0.1 (blue), Ωm0 = 0.3 (yellow), Ωm0 = 0.7 (green) and Ωm0 = 0.99 (red).
The first interesting feature of the power spectrum is that it has a maximum.
This maximum takes place roughly at keq for the following reason: entering the
horizon at equivalence is the best time to do that in order for a matter fluctuation
to grow more. In fact, as we have seen, scales that entered the horizon earlier
(i.e. k > keq ) are suppressed because radiation is dominating and that δ grows
logarithmically during this epoch, cf. Eq. (9.77). These are the scales of great
observational interest, as can be appreciated from Fig. 9.1.
249
On the other hand, the scales that entered after the equivalence grow propor-
tionally to a, cf. Eq. (9.54). Evidently, the scale which entered at equivalence has
had more time to grow than all the others, hence the the maximum or turnover
in the power spectrum.
The second interesting feature of the power spectrum is that its maximum is
shifted to the left when we reduce Ωm0 . This means that the equivalence wavenum-
ber keq is smaller when Ωm0 is smaller and this can be immediately seen from
Eq. (9.44) if we fix Ωr0 .
Remarkably, observation favours the line for Ωm0 = 0.3 of Fig. 9.11, as can
also be seen from Fig. 9.1. Since the radiation content is very well established from
CMB observation (and our knowledge of neutrinos) as well as the spatial flatness of
the universe (from the CMB, we shall see this in Chapter 10), and its age (at least a
lower bound) and H0 are also well-determined, the only way to make the total is to
add a further component, which clearly is DE. The important point here is that the
necessity for DE can already be seen by analysing the large scale structure of the
universe (together with CMB) and this was realised well before type Ia supernovae
started to be used as standard candles [Maddox et al., 1990], [Efstathiou et al.,
1990].
For completeness, in Fig. 9.12 we plot with CLASS the matter power spectrum
at z = 0 for the ΛCDM model, with different choices of the initial conditions.
��
��
��-�
��-�
��-� ��-� ��
Figure 9.12: Matter power spectra for different initial conditions: Adiabatic (blue),
Baryon Isocurvature (yellow), CDM Isocurvature (green), Neutrino density Isocur-
vature (red), and Neutrino velocity Isocurvature (purple).
The transfer function thus tells us how the shape of the primordial power spec-
trum is changed through the cosmological evolution and through the analytic es-
timates that we have done in this Chapter we understood that the k-dependence
is set up during radiation-domination.
However, we have made two important assumptions in our calculations: we
have neglected baryons and neutrino anisotropic stress (we have taken into account
neutrinos from the point of view of the background expansion). If we aim to
precise predictions, and we have to in order to keep the pace with the increasing
250
sophistication of the observational techniques, we must take into account them. A
more precise fitting formula (to numerical calculations performed with CMBFAST)
taking into account baryons and neutrino anisotropic stress is given by [Eisenstein
and Hu, 1998].
��
��
��
��
��
��
��
��
��
��
��
��
��-� �� -� ��
Figure 9.13: Left Panel. Evolution of BBKS transfer function of Eq. (9.114) (solid
line) compared with the numerical computation of CLASS, using Ωm0 = 0.95
(which means negligible DE). Right Panel. Ratio between the numerical result
and the BBKS transfer function.
In Fig. 9.13 we compare the BBKS transfer function with the numerical calcu-
lation of CLASS, adopting Ωm0 = 0.95 while leaving all the remaining cosmological
parameters as in the ΛCDM model, except for ΩΛ which is adjusted to the value
ΩΛ = 1.632908 × 10−3 in order to match the correct budget of energy density. So,
in practice, we are neglecting DE.
As can be appreciated from the plots, the BBKS transfer function overestimates
of about 5% the correct transfer function for scales k & 0.1 h Mpc−1 , see also
[Dodelson, 2003, pag. 208]. The reason is that baryons behave like radiation before
decoupling, because of their tight coupling due to Thomson scattering. Therefore,
they contribute further to thwart scales entering the horizon before the equivalence.
��
��
��
��
��
��
��
��
��
��-� ��-� �� -� ��-� ��
Figure 9.14: Left Panel. Ratio between the numerical result computed with
CLASS assuming the ΛCDM model, and the BBKS transfer function with Ωm0 h2 =
0.12038. Right Panel. Ratio between the numerical result computed with CLASS
assuming the ΛCDM model, and the BBKS transfer function with Ωm0 = 1.
In Fig. 9.14 we display the ratio between the numerical transfer function com-
251
puted with CLASS for the ΛCDM model and the BBKS transfer function in two
cases. In the left panel, the same matter density parameter of the ΛCDM model is
used for both the transfer functions. In the right panel, we used the BBKS transfer
function with Ωm0 = 1.
The latter choice was made in order to reproduce the plot of [Dodelson, 2003,
pag. 208] and thus to show the very large correction due to the cosmological
constant which, if not taken into account, leads to a 80% error at the scale k = 0.1
h Mpc−1 .
When Λ is taken into account properly, the BBKS transfer function still over-
estimates the correct transfer function of at least a 10% on small scales, which
implies an imprecision of 1% in the power spectrum (since this depends on the
squared transfer function). Moreover, as it is obvious since it does not include
baryons, the BBKS cannot describe the BAO, which appear as oscillations in the
transfer function at about k = 0.1 h Mpc−1 in Fig. 9.14.
9.6 The transfer function for tensor perturbations

Treating the evolution of tensor perturbations, which we have done in the context
of inflation in Chapter 8 is much easier than for scalar modes because of two
reasons. The first is that CDM and baryons are very non-relativistic and so do
not possess anisotropic stress, which would source GW. The second reason is that,
though photons and neutrinos do have anisotropic stresses this are always very
small so that it is a good approximation to neglect them.
Therefore, what we have to do is to recover the calculations of Sec. 8.4 and
solve Eq. (8.67) with a H given in the radiation- and matter-dominated epoch. As
we already know, on very large scales, meaning k 2 a00 /a, h is a constant (we
shall call it hP ) and this holds true for whatever background expansion we choose.
We used this fact in order to determine the primordial power spectrum and in the
present section we use it again to determine the initial value for solving Eq. (8.67).
Using Eqs. (9.40) and (9.41), we can write the conformal Hubble parameter as
follows: √
Heq 1 + y
H= √ , (9.119)
2y
where we have introduced again y ≡ ρm /ρr as new independent variable.
Exercise 9.19. Cast Eq. (8.67) using y as independent variable and the above
form of the conformal Hubble factor. Show that:
d2 h

1 2(1 + y) dh
(1 + y) 2 + + + κ2 h = 0 (9.120)
dy 2 y dy
using also Eq. (9.100).
The above equation cannot be solved analytically, but we can find an exact
solution in the same four instances that we worked out for scalar perturbations. On
super-horizon scales we already know that h = hP irrespective of the background
evolution (this is relevant only for the decaying mode), so we skip this case.
252
9.6.1 Radiation-dominated epoch
In this case, we consider y 1 in Eq. (9.120):
d2 h 2 dh
2
+ + κ2 h = 0 , (9.121)
dy y dy
which can be rewritten as:
d2 (κhy)
+ κhy = 0 , (9.122)
d(κy)2
and hence is a harmonic oscillator equation for the quantity κhy with respect to
the variable κy, with unitary frequency. Hence, the solution for h is:
sin(κy) cos(κy)
h(k, y) = C1 (k) + C2 (k) , (9.123)
κy κy
where as usual C1 and C2 are generic k-dependent functions. For y → 0, or
equivalently on very large scales κy → 0 only the sine tends to a finite result and
hence we must put C2 to zero:
sin(κy)
h(k, y) = hP (k) = hP (k)j0 (κy) . (9.124)
κy
Equivalently, putting H = 1/η into the original equation (8.67) one finds:
2
h00 + h0 + k 2 h = 0 , (9.125)
η
and so:
sin(kη)
h(k, η) = hP (k) = hP (k)j0 (kη) , (9.126)
kη
since, indeed, κy = kη when radiation dominates, cf. Eq. (9.102).
9.6.2 Matter-dominated epoch

In this case, we consider y 1 in Eq. (9.120):
d2 h 5 dh
y + + κ2 h = 0 , (9.127)
dy 2 2 dy
or, using Eq. (8.67) with H = 2/η:
4
h00 + h0 + k 2 h = 0 . (9.128)
η
This equation can be cast in the form of a Bessel equation.
Exercise 9.20. Introduce the new function:

h ≡ g(kη)α , (9.129)
where α is a number to be determined. Then, show that:
η 2 g 00 + (2α + 4)ηg 0 + g[k 2 η 2 + α2 + 3α] = 0 . (9.130)
253
Then, we recover the form of the Bessel function for α = −3/2, for which the
order is 3/2 and hence:
r
2kη
g ∝ J3/2 (kη) = j1 (kη) , (9.131)
π
where we have introduced the spherical Bessel function and neglected the one of
second kind, since it diverges for kη → 0. The GW amplitude thus evolves as:

j1 (kη) sin(kη) cos(kη)
h(k, η) = C1 (k) = C1 (k) − . (9.132)
kη (kη)3 (kη)2
In the limit kη → 0:
sin(kη) cos(kη) 1
− → , (9.133)
(kη)3 (kη)2 3
hence we can write:

sin(kη) cos(kη)
h(k, η) = 3hP (k) − , (9.134)
(kη)3 (kη)2
but this is valid only for those modes which enters the horizon well deep into the
matter-dominated epoch. Using the definition of κ in Eq. (9.100) and the fact that,
deep in the matter-dominated epoch we have that:
2 Heq
H= =√ , (9.135)
η 2y
we can conclude that:

√
kη = 2κ y , (9.136)
and thus the solution for h(k, y) is:
√ √
3hP (k) sin(2κ y) cos(2κ y)
h(k, y) = √ − √ , (9.137)
4 2(κ y)3 (κ y)2
√
which again is valid only for those modes κ y 1 and hence, since y 1, then
κ 1, i.e. modes which entered the horizon well deep into the matter-dominated
epoch.
9.6.3 Deep inside the horizon

In the case k H we work directly on Eq. (8.67) and, following [Weinberg, 2008],
we introduce a new variable Z
dη
x≡ , (9.138)
a2
with the purpose of eliminating the first-order derivative, and obtain thus the
following equation:
d2 h
+ k 2 a4 h = 0 . (9.139)
dx2
254
The latter is not analytically solvable for a generic a(x) but it is possible to find
an approximated, WKB solution. Indeed, if a(x) is approximately constant, the
above equation becomes that of a harmonic oscillator and the solution is simply:
2
h(k, x) ∝ e±ika . (9.140)
Taking into account the x dependence of a, we can use the following ansatz:
Z
2
h(k, x) = A(x) exp ±ik a dx , (9.141)
and substitute it back into Eq. (9.139), obtaining:

d2 A

dA 2 da
2
+ 2 (±ika ) + A ±2ika + k 2 a4 A = 0 . (9.142)
dx dx dx
Equating the imaginary part to zero, we have then:
dA 2 da
a + Aa =0, (9.143)
dx dx
for which the solution is:
A(x) ∝ 1/a(x) . (9.144)
Hence, the approximated WKB solution for the GW is:
Z
1 2 1
h(k, η) ∝ exp ±ik a dx = exp (±ikη) . (9.145)
a a
Note that this solution is valid for any background expansion, since we have made
no assumptions on the latter. Moreover, note the oscillatory behaviour exp (±ikη)
with amplitude damped with a factor 1/a. Recall from [Weinberg, 1972] that the
energy-momentum tensor of a GW is:
pµ pν
|h+ |2 + |h× |2 ,

htµν i = (9.146)
16πG
where tµν is the gravitational pseudo-tensor, the average on it is on a sufficiently
large region of spacetime such that the oscillatory terms gives a constant result and
pµ is the physical momentum pµ = dxµ /dλ of the GW (of the graviton). We now
that in a FLRW metric for a massless particle p0 = p ∝ 1/a and hence ht00 i ∝ 1/a4
as we expected for the energy density of a relativistic species.
Now, we can match the above WKB solution (for y 1) with the one we
found in Eq. (9.126) (for large κy), where we use y instead of η because it is
straightforward then to project the matching at late times. It is the same procedure
we employed in order to find the transfer function for CDM. So, we have:
C1 (k) sin(κy)
sin(κy) = hP (k) , (9.147)
yaeq κy
and thus
√
aeq aeq H0 Ωm0 H0 Ωr0
C1 (k) = hP (k) = hP (k) √ = hP (k) . (9.148)
κ k Ωr0 k
255
So, the solution for small scales is:
√
H0 Ωr0
h(k, η) = hP (k) sin(kη) (9.149)
ka(η)
Note that this solution is valid for any content of the universe, since there is no
more a y dependence appearing (any content after radiation-domination, since we
have matched there the solutions). Close to the present time η0 we can set as usual
a(η0 ) = 1 and then the gravitational wave solution for small scales is then:
√
H0 Ωr0
h(k, t) = hP (k) sin [kη0 + k(t − t0 )] (9.150)
k
for t close to t0 and where we have used dη = dt/a(t) close to present time. This
GW profile is very different from those detected by LIGO for the merging of black
holes, the first in [Abbott et al., 2016b], and from the spectacular event GW170817
of the merging of two neutron stars detected both by LIGO and VIRGO [Abbott
et al., 2017a] and whose electromagnetic counterpart was also seen as a short
gamma-ray burst (GRB) [Abbott et al., 2017c]. The main characteristics of these
profiles is their growth in frequency and amplitude for times close to the merging
one. The cosmological GW profile is a simple sine function (for very small scales),
with an amplitude containing cosmological information of great relevance such as
the primordial amplitude, which we have seen to be related to the energy scale of
inflation.
256
Chapter 10
Anisotropies in the Cosmic

Microwave Background
Long the realm of armchair philosophers, the study of the origins and
evolution of the universe became a physical science with falsifiable
theories
—Wayne Hu, PhD Thesis
In this Chapter we attack the hierarchy of Boltzmann equations that we have

found for photons and present an approximate, semi-analytic solution which will
allow us to understand the temperature correlation in the CMB sky and its relation
with the cosmological parameters. Our scope is to understand the features of the
angular, temperature-temperature power spectrum in Fig. 10.1.
Note that in this plot the definition
`(` + 1)CT T,`

D`T T ≡ , (10.1)
2π
is used. We shall see the reason for the `(` + 1) normalisation, whereas the CT T,` ’s
are given in Eq. (7.81) as functions of the multipole moments of the temperature
distribution and the primordial power spectrum for scalar perturbations.
In Fig. 10.1 we can also see data points up to ` ≈ 2500. What can we say from
this number about the angular sensitivity of Planck ? It can be roughly computed
as follows. For a given `max how many realisations of a`m do we have?
Exercise 10.1. For each ` we have 2` + 1 possible values of m, thus show that:
`X
max
N`max = (2` + 1) = (`max + 1)2 . (10.2)

`=0
The full sky has:

4
4π rad2 = (180 deg)2 ≈ 41000 deg2 . (10.3)
π
257
Figure 10.1: CMB TT spectrum. Figure taken from [Ade et al., 2016a]. The red
solid line is the best fit ΛCDM model.
If an experiment has sensitivity of 7 deg, then we can have at most

4
(180/7)2 ≈ 842 , (10.4)
π
pieces of independent information and therefore we can determine as many a`m .
This gives `max ≈ 28 and it was the sensitivity of CoBE. For Planck, the angular
sensitivity was of 5 arcmin, which corresponds to
4
(180 × 60/5)2 ≈ 106 , (10.5)
π
pieces of independent information and then to `max = 2436.
In this Chapter we omit the superscript S referring to the scalar perturbations
contribution to Θ, since most of the time we shall discuss of it. We shall only use
T in order to distinguish the tensor contribution.
10.1 Free-streaming
It is convenient to start neglecting the collisional term in the Boltzmann equation
and considering thus the phase of photon free-streaming. The following discus-
sion is similar to the one in Sec. 5.4. Consider Eq. (5.27) for photons and with no
collisional term. Using the definition of Θ in Eq. (5.80), we can write for scalar
perturbations:
dxi ∂

∂
+ i
(Θ + Ψ) = Ψ0 − Φ0 . (10.6)
∂η dη ∂x
258
As we know from Boltzmann equation, the differential operator on the left hand
side is a convective derivative, i.e. a derivative along the photon path:
d
(Θ + Ψ) = Ψ0 − Φ0 , (10.7)
dη
whose inversion is the basis of the line-of-sight integration approach to CMB
anisotropies [Seljak and Zaldarriaga, 1996], which is an alternative to attacking
the hierarchy of coupled Boltzmann equations (which still must be attacked but
can be truncated at much lower `’s) as it was done e.g. in [Ma and Bertschinger,
1995]. We shall see this technique in some detail in Sec. 10.5.
For time-independent potentials, as they are in the matter-dominated epoch,
the collisionless Boltzmann equation for photons tells us that Θ + Ψ is constant
along the photons paths, i.e. along our past light-cone, since recombination.
Recall that the scalar-perturbed metric that we are using is given in Eq. (4.171):
ds2 = −a2 (η)(1 + 2Ψ)dη 2 + a2 (η)(1 + 2Φ)δij dxi dxj . (10.8)
Inside a potential well, Ψ is negative. In order to be convinced of this one has

just to think about the Newtonian limit and realise that 2Ψ is the Newtonian
gravitational potential, hence negative. So, since Θ + Ψ stays constant, we have
that:
Θ(η∗ , x∗ , p̂) + Ψ(η∗ , x∗ ) = Θ(η0 , x0 , p̂) + Ψ(η0 , x0 ) . (10.9)
where on the left hand side we have chosen the quantities at recombination whereas
on the right hand side we have chosen the present time. Note that x0 , is where our
laboratory (the CMB experiment) is, i.e. Earth, and as such is fixed. Therefore,
since we can only detect photons on our past light-cone, and those from CMB
comes from a fixed comoving distance r∗ = η0 − η∗ , we have that:
x∗ = x0 − r∗ p̂ . (10.10)
Note that p̂ is the photon direction and so it is opposite to the direction of the
line of sight n̂ = −p̂. So, the only independent variables are 2, the components of
p̂. They become just a single one, µ, because of the way in which we factorise the
azimuthal dependence (and assuming axial symmetry).
The potential Ψ(η0 , x0 ) is usually neglected, or incorporated in the potential
at recombination, since it is not detectable. As it is well known, classically only
potential differences are physically meaningful. The above equation then tells us
that:
Θ(η∗ , x0 − r∗ p̂, p̂) + Ψ(η∗ , x0 − r∗ p̂) = Θ(η0 , x0 , p̂) , (10.11)
i.e. the observed temperature fluctuation (on the right hand side) accounts for the
energy loss due to climbing out the potential well or falling down a potential hill.
This is the so-called Sachs-Wolfe effect [Sachs and Wolfe, 1967]. Writing the
above equation in Fourier modes, we have:
d3 k d3 k
Z Z
3
Θ(η0 , k, p̂)eik·x0
= 3
[Θ(η∗ , k, p̂) + Ψ(η∗ , k)] eik·(x0 −r∗ p̂) . (10.12)
(2π) (2π)
259
We can set now x0 = 0, without losing of generality, and manifest the depen-
dences as k and µ, the former since we normalise to the scalar primordial mode
α(k), cf. Eq. (7.64), and the latter since we are considering axisymmetric scalar
perturbations. Hence we have for the Fourier modes:
Θ(η0 , k, µ) = [Θ(η∗ , k, µ) + Ψ(η∗ , k)] e−ikµr∗ . (10.13)
Using the partial wave expansion, we get:
Z 1
1 dµ
Θ` (η0 , k) = `
P` (µ) [Θ(η∗ , k, µ) + Ψ(η∗ , k)] e−ikµr∗ , (10.14)
(−i) −1 2
and using the relation:
Z 1
dµ
P` (µ)e−ikµr∗ = (−i)` j` (kr∗ ) , (10.15)
−1 2
which can be obtained by inverting the expansion of Eq. (5.75), we can write:
Z 1
1 dµ
Θ` (η0 , k) = Ψ(η∗ , k)j` (kr∗ ) + `
P` (µ)Θ(η∗ , k, µ)e−ikµr∗ . (10.16)
(−i) −1 2
Using again the partial wave expansion, we can write the above formula as:
Θ` (η0 , k) = Ψ(η∗ , k)j` (kr∗ )
Z 1
1 X
`0 0 dµ
+ `
(−i) (2` + 1)Θ`0 (η∗ , k) P` (µ)P`0 (µ)e−ikµr∗ . (10.17)
(−i) `0 −1 2
We shall see later that, because of tight-coupling, the monopole and the dipole con-
tribute the most at recombination. Hence, we can write, truncating the summation
at `0 = 1:
3Θ1 (η∗ , k) 1 dµ
Z
Θ` (η0 , k) = (Θ0 + Ψ) (η∗ , k)j` (kr∗ ) + `−1
P` (µ)µe−ikµr∗ . (10.18)
(−i) −1 2
The integral can be performed as follows:

Z 1 Z 1
dµ −ikµr∗ d dµ 1 d
P` (µ)µe =i P` (µ)e−ikµr∗ = `−1 j` (kr∗ ) .
−1 2 d(kr∗ ) −1 2 i d(kr∗ )
(10.19)
The same technique can be used, in principle, to calculate the integral for any `0 :
for each power of µ one gains a derivative of the spherical Bessel function. Recalling
the formula [Abramowitz and Stegun, 1972]:1
dj` (x) `+1
= j`−1 (x) − j` (x) , (10.20)
dx x
we can write:
Θ` (η0 , k) = (Θ0 + Ψ) (η∗ , k)j` (kr∗ )

`+1
+3Θ1 (η∗ , k) j`−1 (kr∗ ) − j` (kr∗ ) . (10.21)
kr∗
1
The website http://functions.wolfram.com is also very useful.
260
So, the spherical Bessel functions that we have mentioned in Chapter 9 start to
appear. We have obtained the above free-streaming solution neglecting the po-
tentials derivatives in Eq. (10.7). Taking them into account is not difficult, since
an additional piece containing the integration of the potential derivatives would
appear in Eq. (10.13):
Z η0
−ikµr∗
Θ(η0 , k, µ) = [Θ(η∗ , k, µ) + Ψ(η∗ , k)] e + dη (Ψ0 − Φ0 )(η, k)e−ikµ(η0 −η) .
η∗
(10.22)
The exponential factor in the integral comes from the Fourier transform of the
potentials and from considering:
x = x0 − (η0 − η)p̂ , (10.23)
at any given time η along the photon trajectory (this is the “line of sight”, in
practice). Performing again the expansion in partial waves, we get:
Θ` (η0 , k) = (Θ0 + Ψ) (η∗ , k)j` (kr∗ )

`+1
+3Θ1 (η∗ , k) j`−1 (kr∗ ) − j` (kr∗ )
kr∗
Z η0
+ dη [Ψ0 (η, k) − Φ0 (η, k)]j` (kr) , (10.24)
η∗
where
r ≡ η0 − η . (10.25)
As we are going to see, the first two terms of the above formula contains the
primary anisotropies of the CMB, which are the acoustic oscillations and
the Doppler effect. The Ψ(η∗ , k) contribution in the first term is, as we have
already anticipated, the Sachs-Wolfe effect. The last term is the Integrated
Sachs-Wolfe (ISW) effect [Sachs and Wolfe, 1967] and contributes only when
the gravitational potentials are time-varying. This happens, as we have seen in
Chapter 9, when radiation and DE are relevant. For this reason the ISW effect is
usually separated in the early-times one, due to a small presence of radiation still
at decoupling, and in the late-times one, due to DE.
Once we know all the contributions of the above formula, we can use Eq. (7.81)
and provide the prediction on the CTST,` spectrum.
The presence of the spherical Bessel function is interesting for two reasons,
which we display in Fig. 10.2 for the arbitrary choice ` = 10.
We have chosen to plot the squared spherical Bessel function because it is the
relevant window function when computing the C` ’s, as we shall see briefly. First,
the maximum value is attained roughly when x ≈ ` and for x < ` the spherical
Bessel function is practically vanishing. Therefore, for a given multipole ` the scale
which contribute most for the observed anisotropy is:
`
k≈ . (10.26)
η0 − η∗
We have anticipated this already in Chapter 9. The second reason of interest is
that the spherical Bessel function goes to zero for large x. This means that scales
261
��
��
��
��
��
��
��
� � ��
2
Figure 10.2: Evolution of the spherical Bessel function j10 (x).
such that kr∗ 1 do not contribute to the observed anisotropy. Physically, this is
an effect due to the free-streaming phase for which, on very small scales, hot and
cold photons mix up destroying thus the anisotropy.
We have thus seen that the predicted anisotropy today is given by formula
Eq. (10.24). We have now to justify the fact of considering only the monopole and
the dipole at recombination. We shall commence in the next section discussing
very large scales.
In principle, Eq. (10.24) has the very same form for neutrinos, but with an initial
conformal time ηi which is well anterior to η∗ , since neutrinos do not interact and
therefore they only free-stream (at least for temperatures of the primordial plasma
below 1 MeV).
10.2 Anisotropies on large scales

On large scales, i.e. kη 1, the relevant equations are those of Chapter 6, which
we report here:
δγ0 = −4Φ0 , δν0 = −4Φ0 , δc0 = −3Φ0 , δb0 = −3Φ0 , (10.27)
i.e. only the monopoles are relevant. Since we want to describe CMB, let us focus
on the photon density contrast, which can be written as:
δγ (k, η) = 4Θ0 (k, η) , (10.28)
introducing the monopole of the temperature fluctuation. The equation Θ00 = −Φ0
can be immediately integrated, obtaining:
Θ0 (k, η) = −Φ(k, η) + Cγ (k) . (10.29)
262
For the adiabatic primordial mode, the only which we are going to consider, we
know from Eq. (6.94) that Cγ (k) = ΦP (k) − ΨP (k)/2 and thus:
1
Θ0 (k, η) = −Φ(k, η) + ΦP (k) − ΨP (k) . (10.30)
2
As we know from Chapter 9, we can consider the gravitational potentials to be
equal in modulus and on large scales Φ(k, η) is independent of time and since
recombination η∗ ηeq takes place well after radiation-matter equality, we know
that Φ(k, η∗ ) = 9ΦP (k)/10, i.e. the value of the gravitational potential drops of
10% in passing through radiation-matter domination. Therefore:
3 2 2
Θ0 (k, η∗ ) = ΦP (k) = Φ(k, η∗ ) = − Ψ(k, η∗ ) . (10.31)
5 3 3
As we saw earlier in Eq. (10.24), the observed anisotropy is not Θ0 (k, η∗ ) but
Θ0 (k, η∗ ) + Ψ(k, η∗ ), because of the gravitational redshift. Again, this is the Sachs-
Wolfe effect, amounting to a shift in the photons frequency when they decouple
from the baryonic plasma depending whether they are in a well or hill of the
gravitational potential. So, we have from Eq. (10.31) that:
1
(Θ0 + Ψ)(k, η∗ ) = Ψ(k, η∗ ) . (10.32)
3
On the other hand, for δc we know that
9ΦP (k)
δc (k, η) = −3Φ(k, η) + , (10.33)
2
again assuming adiabatic primordial modes. Using again Φ(k, η∗ ) = 9ΦP (k)/10,
we get:
δc (k, η∗ ) = 2Φ(k, η∗ ) = −2Ψ(k, η∗ ) . (10.34)
The fluctuations in CDM contribute more in generating the potential wells than
photons, a factor 2 against a factor −2/3. Combining the two equations:
δc (k, η∗ )
(Θ0 + Ψ)(k, η∗ ) = − (10.35)
6
This result tells us that on large scales colder spots represent larger overdensities,
a counter-intuitive result. One expects hotter photons the deeper the well is and
in fact this is the case with just Θ0 (k, η∗ ), since we have:
2 δc (k, η∗ )
Θ0 (k, η∗ ) = − Ψ(k, η∗ ) = , (10.36)
3 3
i.e. the larger the CDM overdensity, the larger the well and Θ0 (k, η∗ ) are. However,
photons’ response to the gravitational potential is only a factor −2/3 whereas the
gravitational redshift adds a Ψ contribution, changing thus the sign of the observed
anisotropy. In the limit of δc → −1, one gets (Θ0 + Ψ)(k, η∗ ) → 1/6, so cosmic
voids correspond to hot spots!
The results found here are valid only on large scales, i.e. for kη∗ 1, scales
much larger than the horizon at recombination, which has an angular size of approx-
imatively 1 degree. Moreover, they also depend on the choice of initial conditions.
We have opted for the adiabatic ones, as usual.
263
Exercise 10.2. Reproduce the above argument for the other primordial modes.
Let us use the theoretical prediction on the CTST,` given in Eq. (7.81) together
with the first contribution only from Eq. (10.24). The latter approximation is jus-
tified by the fact that we are considering large scales, hence the dipole contribution
is negligible and the ISW effect is vanishing because the potentials are constant.
Since:
1 1 3 1
(Θ0 + Ψ)(k, η∗ ) = Ψ(k, η∗ ) = − Φ(k, η∗ ) = − ΦP (k) = − R(k) , (10.37)
3 3 10 5
the transfer function is just the constant −1/5 (recall that we are neglecting the
neutrino fraction Rν ) and thus the angular power spectrum is:
4π ∞ dk 2
Z
S
CT T,` (SW) = ∆ (k)j`2 (kη0 ) , (10.38)
25 0 k R
since η∗ η0 . Note that kη∗ kη0 and we have seen in Fig. 10.2 that the spherical
Bessel function contributes the most about kη0 ≈ `. Thus, for small `, i.e. large
angular scales, kη0 is small and kη∗ is very small, where in fact | (Θ0 + Ψ) (k, η∗ )|2
is constant. In other words, the above approximation is valid for small `, typically
` . 30.
In the above integral we can look at j`2 (kη0 ) as a very peaked window function
and approximate it as:
Z ∞
S 4π 2 dk 2
CT T,` (SW) ≈ ∆R (`/η0 ) j (kη0 ) . (10.39)
25 0 k `
Using the result: Z ∞
dx 2 1
j` (x) = , (10.40)
0 x 2`(` + 1)
we have then:
`(` + 1)CTST,` (SW) 1
≈ ∆2R (`/η0 ) . (10.41)
2π 25
Hence, for a scale-invariant spectrum nS = 1 the combination `(` + 1)CTST,` (SW)
is constant and it is called Sachs-Wolfe plateau. This also explains why CMB
power spectra are usually presented with the `(` + 1) normalisation, as in Fig. 10.1.
If nS 6= 1, then `(` + 1)CTST,` (SW) is proportional to `nS −1 , i.e. the primordial
tilt in the power spectrum leaves its mark in a tilted plateau for small `.
10.3 Tight-coupling and acoustic oscillations

We have seen that in order to determine the prediction on the present time CT T,` ’s
we need to know what happens at recombination. We devote this section to such
purpose, showing that the monopole and the dipole contribute the most.
264
Let us recover here the hierarchy of Boltzmann equations for the Θ` ’s (not
taking into account polarisation) that we have derived in Chapter 5:
(2` + 1)Θ0` + k[(` + 1)Θ`+1 − `Θ`−1 ] = (2` + 1)τ 0 Θ` , (` > 2) , (10.42)

10Θ2 + 2k (3Θ3 − 2Θ1 ) = 10τ Θ2 − τ 0 Π ,
0 0
(10.43)
3Θ01 + k(2Θ2 − Θ0 ) = kΨ + τ 0 (3Θ1 − Vb ) , (10.44)
Θ00 + kΘ1 = −Φ0 , (10.45)
where recall that δγ = 4Θ0 and 3Θ1 = Vγ . The best way to deal with these
equations is to solve them numerically by using Boltzmann codes such as CAMB
or CLASS, but in this way the physics behind the CT T,` ’s remains hidden or un-
clear. For this reason we attack these equations in an approximate fashion, but
analytically.
We take the limit −τ 0 H, which is called tight-coupling (TC) approx-
imation. This limit physically means that the Thomson scattering rate between
photons and electrons is much larger than the Hubble rate until recombination and
then drops abruptly since the free electron fraction Xe goes to zero very rapidly, as
we have seen when studying thermal history in Chapter 3. We shall first consider
the case of sudden recombination, where all the photons last scatter at the
same time. It is a fair approximation, though unrealistic.
Exercise 10.3. From the definition of the optical depth:

Z η0
τ≡ dη 0 ne σT a , (10.46)
η
show that τ ∝ 1/η 3 when matter dominates and τ ∝ 1/η when radiation dominates.
We can be more quantitative and write:

ρb
− τ 0 = ne σT a = nb σT a = σT a , (10.47)
mb
where we have used the definition of τ and assumed to be in an epoch before
recombination, so that we can approximate ne with nb , since all the electrons are
free.
Exercise 10.4. Introducing the baryon density parameter and using mb = 1 GeV,
the mass of the proton, show that:
Ωb0 h2 −1
− τ 0 ≈ 1.46 × 10−19 s . (10.48)
a2
Now we need to compare this scattering rate with the Hubble rate, in order
to check the goodness of the TC approximation. Assuming matter-domination
265
and using the conformal time Friedmann equation (this because τ 0 is derived with
respect to the conformal time), we have:
p p
H = H0 Ωm0 a−1/2 ≈ 3.33 × 10−18 h Ωm0 a−1/2 s−1 . (10.49)
Therefore, the ratio:
−τ 0 Ωb0 h2 −3/2
= 0.044 √ a , (10.50)
H Ωm0 h2
diverges for a → 0 as expected (though the formula should be generalised to the
case of radiation-domination), so if it is sufficiently big at recombination then the
TC approximation would be reliable. Substituting the Planck values Ωb0 h2 = 0.022
and Ωm0 h2 = 0.12 one gets at recombination, i.e. for a = 10−3 :
−τ 0
≈ 102 . (10.51)
H
This means that the scattering rate is much larger than the Hubble rate even at
recombination as long as there are free electrons around and thus we are going to
use the tight-coupling approximation with reliability.
Let us see in detail how the TC limit works. Let us compare in the hierarchy
for ` ≥ 2 the terms Θ0` and kΘ` with τ 0 Θ` , which have all the same dimensions of
inverse time. There are two physical time scales in our problem, one is given by
the expansion rate and the other by the scattering rate, hence
Θ0` ∝ HΘ` , τ 0 Θ` , (10.52)
from a dimensional analysis. However, the mode for which Θ0` ∝ τ 0 Θ` implies that
Θ` ∝ exp τ and hence diverges at early times, which is unacceptable for a small
fluctuation. We then dismiss this mode as unphysical and take into account just
that for which Θ0` ∝ HΘ` , which is small compared to τ 0 Θ` .
Now, let us inspect the ratio
−τ 0
. (10.53)
k
This is the number of collisions which take place on a scale 1/k. Hence, this number
is very large, provided that we consider sufficiently large scales, i.e. small k. If the
scale is too small, i.e. large k, then the TC approximation does not work well and
we must take into account the multipole moments for ` ≥ 2. We will see this when
investigating the diffusion damping or Silk damping effect.
From the above analysis, for sufficiently large scales we can conclude then that
Θ` ≈ 0 for ` ≥ 2. Sufficiently large means much larger than the mean free path
−1/τ 0 which is approximately of the order of 10 Mpc at recombination. This
number can be computed from Eq. (10.48) and is a comoving scale; the physical
one is divided by a factor a thousand and so it is 10 kpc.
Finally, note that Θ` ∼ τ 0 /kΘ`−1 . Therefore, considering smaller and smaller
scales makes necessary to include higher and higher order multipoles.
Eliminating all the multipoles ` ≥ 2, the relevant equations are just the follow-
ing two:
Θ00 + kΘ1 = −Φ0 , (10.54)
3Θ01 − kΘ0 = kΨ + τ 0 (3Θ1 − Vb ) , (10.55)
266
i.e. the TC approximation allows us to treat photons as a fluid until recombination.
Note the coupling to baryons via the baryon velocity Vb . Thus, we need also the
equations for baryons:
δb0 + kVb = −3Φ0 , (10.56)

τ0
Vb0 + HVb = kΨ + (Vb − 3Θ1 ) , (10.57)
R
where we have introduced R ≡ 3ρb /4ργ , i.e. the baryon density to photon density
ratio. This number can be cast as:
3Ωb0
R= a ≈ 600a , (10.58)
4Ωγ0
using the usual values and it grows from zero at early times to R∗ ≈ 0.6 at re-
combination. So it is small, but not that negligible. Let us rewrite the velocity
equation for baryons in the following way:
R 0
Vb = 3Θ1 + (V + HVb − kΨ) . (10.59)
τ0 b
We can solve this equation via successive approximation, exploiting the fact that
R < 1 before recombination. That is, assume the expansion:
(0) (1) (2)
Vb = Vb + RVb + R2 Vb + . . . . (10.60)
(0)
The solution for R = 0 simply gives Vb = 3Θ1 , which we have used in Chapter 6
in order to investigate the primordial modes. This solution is reliable well before
recombination, say at a = 10−7 for example, because R ≈ 6 × 10−5 there, but it is
not satisfactory at recombination and we shall take into account the first order in
R in the above expansion.
10.3.1 The acoustic peaks for R = 0

Let us start with the simple case of R = 0, which amounts to neglect baryons.
Exercise 10.5. Combine the photon Eqs. (10.54)-(10.55) with the zeroth-order
TC condition Vb = 3Θ1 and find the following second-order equation for Θ0 :
k2 k2Ψ
Θ000 + Θ0 = − − Φ00 . (10.61)
3 3
We have here already the first fundamental piece of physics of the CMB. This is
the equation of motion of a driven harmonic oscillator where instead of the position
we have the monopole of the temperature fluctuation and the driving force is given
by the gravitational potential. This equation describe acoustic oscillations of
the baryon-photon fluid until recombination. After recombination we expect to
observe these fluctuations in the CT T,` ’s, using the free-streaming formula (10.24),
and in fact we do, cf. Fig. 10.1.
267
Note that these oscillations are in the baryon-photon fluid and therefore affect
also baryons. We therefore expect to see oscillations in the baryon distribution
after recombination, called baryon acoustic oscillations (BAO), and detected
by Eisenstein and collaborators in 2005 [Eisenstein et al., 2005]. The BAO are
the manifestation of a special length, the sound horizon at recombination, in the
correlation function of galaxies which appears as a bump, i.e. an excess probability.
In the Fourier space, i.e. for the power spectrum, a given scale is represented with
various oscillations. We have already encountered BAO in Chapter 9. BAO and
weak gravitational lensing are among the main observables on which current and
future experiments (such as Euclid and LSST ) are based.
Exercise 10.6. Combine Eqs. (10.56) and (10.57) and the TC condition Vb = 3Θ1
and find the following equation for δb :
δb0 = 3Θ00 . (10.62)
Hence, the same oscillatory solution of Θ0 holds true for δb .
Now, consider the fact that close to recombination CDM is already dominating
and thus the potentials are equal and constant at all scales. We get:
k2 k2Ψ
Θ000 + Θ0 = − . (10.63)
3 3
This equation can be put in the following form:
k2
(Θ0 + Ψ)00 + (Θ0 + Ψ) = 0 , (10.64)
3
where we have used the constancy of Ψ. Note how the observed temperature
fluctuation, used in Eq. (10.24), has appeared. The solution is:

kη kη
(Θ0 + Ψ)(η, k) = A(k) sin √ + B(k) cos √ , (10.65)
3 3
with the driving potential, i.e. CDM, providing just an offset for the oscillations.
At recombination we have

kη∗ kη∗
(Θ0 + Ψ)(η∗ , k) = A(k) sin √ + B(k) cos √ , (10.66)
3 3
with peaks and valleys in the temperature fluctuations given by this combination
of sine and cosine, therefore dependent on the functions A(k) and B(k). Inserting
these formula into Eq. (10.24) in order to compute the Θ` (η0 , k) (the anisotropies
today) and then into Eq. (7.81) in order to compute the CT T,` ’s, we are able to
explain the acoustic oscillations feature of the CMB TT spectrum, of Fig. 10.1.
The functions A(k) and B(k) are determined by the initial condition, i.e. for
kη∗ 1:
kη∗
(Θ0 + Ψ)(kη∗ 1) ∼ A(k) √ + B(k) . (10.67)
3
268
Hence, if we choose adiabatic modes, we must put A(k) = 0. So, considering dif-
ferent initial conditions changes the position of the acoustic peaks and observation
allows to test the choice made. As we saw in Chapter 6, Planck limits the pres-
ence of isocurvature modes to a few percent. With A(k) = 0, i.e. for adiabatic
perturbations, using the large-scale solution that we found in Eq. (10.37), we have:

1 kη∗
(Θ0 + Ψ)(η∗ , k) = − R(k)T (k) cos √ , (10.68)
5 3
where T (k) is the transfer function of Θ0 + Ψ. We did not calculate it in Chap-
ter 9, but it can be shown that it is limited to a range 0.4-2, approximately.
See [Mukhanov, 2005].
The extrema of the effective temperature fluctuations are thus given by:
kη∗
√ = nπ , (n = 1, 2, . . . ) , (10.69)
3
where the odd values provide peaks, corresponding to the highest temperature
fluctuations and thus to scales at which photons are maximally compressed and hot,
whereas the even values provide throats, corresponding to the lowest temperature
fluctuations and thus to scales at which photons are maximally rarefied and cold.
In the spectrum, cf. Fig. 10.1, only peaks appear because of the quadratic nature
of the CT T,` ’s as functions of the Θ` ’s, but it should be clear that the first and the
third peaks are compressional.
From Eqs. (10.54) and (10.68), we can determine easily the dipole contribution:
Θ00 (η∗ , k)

1 kη∗
Θ1 (η∗ , k) = − = − √ R(k)T (k) sin √ , (10.70)
k 5 3 3
where we are still continuing in keeping the potentials constant. Substituting this
equation and Eq. (10.68) into Eq. (10.24) and then into Eq. (7.81) in order to
compute the angular power spectrum, we get:
2
4π ∞ dk 2 √
Z
kη∗ kη∗ dj` (kη0 )
CT T,` = ∆ (k) cos √ j` (kη0 ) + 3 sin √ ,
25 0 k R 3 3 d(kη0 )
(10.71)
where the derivative of the spherical Bessel function is given in Eq. (10.20). We
have put T (k) = 1 here for simplicity.
We can manipulate analytically this integral following the technique used in
[Mukhanov, 2004] and [Mukhanov, 2005]. In these references, baryon loading and
diffusion damping are taken into account but here we just tackle a simpler case.
The idea is to avoid the oscillatory nature of the Bessel function and of the
trigonometric ones (which are also problematic from a numerical perspective) by
approximating j` (x) as follows, for large `:
(
0, (x < `) ,
j` (x) ≈ 1
√
2 − `2 − ` arccos(`/x) − π/4 ,
√ 2 2
x(x −` ) 1/4 cos x (x > `) .
(10.72)
269
This approximation is identical either for j` (x) and for j`−1 (x), since we are as-
suming ` to be large. Hence, when we deal with the derivative of the spherical
Bessel function in Eq. (10.71), we can factorise a j`2 (x) and we can approximate
the squared cosine coming from the above approximation with its average, i.e. a
factor 1/2. We thus have the following integration:
2
2π∆2R ∞ √
Z
dk kη∗ ` kη∗
CT T,` = p cos √ + 3 1− sin √ ,
25 2
`/η0 k η0 (kη0 )2 − `2 3 kη0 3
(10.73)
where we have already assumed a scale-invariant spectrum, for simplicity. Using
now the variable
kη0
x≡ , (10.74)
`
we can write:
2
2π∆2R ∞ √ x−1
Z
2 dx
` CT T,` = √ cos (`%x) + 3 sin (`%x) , (10.75)
25 1 x2 x2 − 1 x
where note the appearance of the factor `2 on the left hand side and we have defined
the quantity:
η∗
%≡ √ . (10.76)
3η0
Now, developing the square and using the trigonometric formulae:
1 + cos 2α 1 − cos 2α
cos2 α = , sin2 α = , 2 sin α cos α = sin 2α ,
2 2
(10.77)
we can write:
2π∆2R (k) ∞
Z
2 dx
` CT T,` = √
25 1 x2 x2 − 1
" √ #
x2 + 3(x − 1)2 x2 − 3(x − 1)2 3(x − 1)
+ cos (2`%x) + sin (2`%x) . (10.78)
2x2 2x2 x
Now, let us treat separately the three integrands. The first, non-oscillatory one is
simplest one:
Z ∞
dx x2 + 3(x − 1)2 π
N≡ √ =3 1− , (10.79)
1 x2 x2 − 1 2x2 4
but also the less interesting. The oscillatory ones can be dealt with following
[Mukhanov, 2005]. Define:
Z ∞
dx x2 − 3(x − 1)2
O1 ≡ √ √ cos (2`%x) , (10.80)
1 x − 1 2x4 x + 1
then solving the problem in [Mukhanov, 2005, page 383], we can use the formula:
Z ∞ r
dx π
√ f (x) cos(bx) ≈ f (1) cos(b + π/4) , (10.81)
1 x−1 b
270
for large values of b and a slowly varying f (x). A similar result holds true also for
the sine function. Using this formula we have then:
r
1 π
O1 = √ cos(2`% + π/4) , (10.82)
2 2 2`%
whereas for the integral containing the sine:
Z ∞ √
dx 3(x − 1)
O2 ≡ √ √ sin (2`%x) ≈ 0 , (10.83)
1 x−1 x x+1
3
since f (1) = 0 here. The contribution O2 comes from the cross product between
the monopole and the dipole terms and it is usually neglected in the calculations.
We have explicitly shown why here. Gathering the N and O1 contributions, we
plot the sum N + O1 in Fig. 10.3.
��
��
��
��
��
��
Figure 10.3: Sum of the N and O1 contributions.
In order to make this plot, we have used a ∝ η 2 , since we are in the matter-
dominated epoch, and thus we have evaluated % as follows:
η∗ 1 1
%= √ =p =√ ≈ 0.0183 . (10.84)
3η0 3(1 + z∗ ) 3000
The agreement between the plots of Fig. 10.1 and Fig. 10.3 is poor but at least we
have understood how the acoustic oscillations free-stream until today and are seen
in the CMB TT power spectrum. There are several feature missing in Fig. 10.3:
there are too many peaks, their relative height diminishes too slowly and the overall
trend does not decay as in Fig. 10.1. The reason is that we have neglected baryons
and diffusion damping, which we are going to tackle in the next sections.
271
10.3.2 Baryon loading
√
The oscillations √
in Eq. (10.64) take place with frequency k/ 3, i.e. as if the speed
of sound was 1/ 3, i.e. the speed of sound of a pure photon fluid. We have been
too radical in assuming Vb = 3Θ1 in the equation for baryons. In fact we saw
that this assumption is equivalent to say that R = 0, i.e. the baryon density is
negligible with respect to the photon one. That is why photons do not feel baryons
at all and baryons fluctuations oscillate in the same way as photons do.
(0)
We now take into account R up to first-order. If we consider Vb = 3Θ1
substituted in Eq. (10.59) we get up to order R:
R
Vb = 3Θ1 + (3Θ01 + 3HΘ1 − kΨ) . (10.85)
τ0
Exercise 10.7. Combine the above equation and Eqs. (10.54)-(10.55) in order to
find the following second-order equation for Θ0 :
R k2 k2Ψ R
Θ000 + H Θ00 + Θ0 = − − Φ00 − H Φ0 . (10.86)
1+R 3(1 + R) 3 1+R
Now the speed of sound, i.e. the quantity multiplying k 2 , has been reduced:
1
c2s = . (10.87)
3(1 + R)
The extrema of the temperature fluctuation at recombination are now expected to

be slightly changed, since:
kη∗
p = nπ , (n = 1, 2, . . . ) , (10.88)
3(1 + R)
is now the condition defining them. Moreover, baryons are also responsible for
the damping term HRΘ00 /(1 + R), hence we also expect the extrema to have less
and less amplitude. These features translate, once free-streamed,2 in a relative
suppression of the second peak with respect to the first one, as seen in Fig. 10.1.
This effect is due to the baryon loading and it is also called baryon drag.
Physically, baryons are heavy and prevent the oscillations in Θ0 to be symmetric,
favouring compression over rarefaction. Since R ∝ a, then we have that:
R0 = HR . (10.89)
Let us write Eq. (10.86) in the following form:
R0 d
2
k2

d 2 2 Φ
+ + k cs (Θ0 + Φ) = −Ψ . (10.90)
dη 2 1 + R dη 3 1+R
2
To “free-stream” means to calculate the CT T,` ’s weighting the solution at recombination with
the spherical Bessel function of Eq. (10.24).
272
The above equation cannot be solved analytically, but we can use a semi-analytic
approximation, provided by [Hu and Sugiyama, 1996]. Let us employ the WKB
method and use the following ansatz:
(Θ0 + Φ)(η, k) = A(η)eiB(η,k) , (10.91)
where A(η) and B(η, k) are functions to be determined via Eq. (10.90).
Exercise 10.8. Substitute this ansatz into the homogenous part of Eq. (10.90)
and find the following couple of equations, by separately equating the real and
imaginary parts to zero:
R0
−A(B 0 )2 + A00 + A0 + k 2 c2s A = 0 , (10.92)
1+R
R0
2B 0 A0 + AB 00 + AB 0 = 0 . (10.93)
1+R
In the first equation, let us neglect the second and the third term with respect
to the first one. That is, the oscillations provide almost at any time (except at the
extrema) a much larger derivative than that of the amplitude or R. Then, the first
equation is readily solved as:
Z η
B(η, k) = k cs (η 0 )dη 0 ≡ krs (η) (10.94)
0
where in the last step we have defined the sound horizon, i.e. the conformal
distance travelled by a sound wave propagating in the baryon photon fluid. When
evaluated at recombination, rs (η∗ ) = 150 Mpc and this scale is fundamental for
BAO, making them standard rulers.
Exercise 10.9. Determine now A(η). Show that the above equations, together
with the found solution for B(η, k), can be cast as:
A0 1 R0
=− , (10.95)
A 41+R
which gives:
A(η) = (1 + R)−1/4 . (10.96)
The general, approximate, solution of the homogeneous equation is then:

1
(Θ0 + Φ)(η, k) = [C(k) sin(krs ) + D(k) cos(krs )] (10.97)
(1 + R)1/4
The condition |A0 |, R0 |B 0 |, which was employed in order to find the above
solution, can be checked as follows:
R0 k
5/4
, R0 p , (10.98)
4(1 + R) 3(1 + R)
273
which essentially amounts to say that:
k R0 , (10.99)
i.e. the solution found is good on sufficiently small scales. Since R0 = HR ∼ R/η,
we must have that kη R. Since R is pretty small, being at most R∗ ≈ 0.6
at recombination, this condition means any scale at early times, but sub-horizon
scales at recombination.
Equation (10.97) gives us the general solution of the homogeneous part of
Eq. (10.90). In order to find the general solution of the full equation we need
to find a particular solution of Eq. (10.90). This can be obtained via Green’s func-
tions method, which we recall in Chapter 12. Let us define, in order to keep a more
compact notation, the independent solutions of the homogeneous equation that we
have just found in Eq. (10.97) as follows:
1 1
S1 (η, k) ≡ sin(krs ) , S2 (η, k) ≡ cos(krs ) . (10.100)
(1 + R)1/4 (1 + R)1/4
Taking into account the non-homogeneous term, the general solution of Eq. (10.90)
is:
k 2 η 0 Φ(η 0 )
Z
(Θ0 +Φ)(η, k) = C(k)S1 +D(k)S2 + dη − Ψ(η ) G(η, η 0 ) , (10.101)
0
3 0 1+R
where G(η, η 0 ) is the Green’s function.
Exercise 10.10. As done in Chapter 12, cf. Eq. (12.121), determine the Green’s
function:
S1 (η 0 )S2 (η) − S1 (η)S2 (η 0 )
G(η, η 0 ) = , (10.102)
W (η 0 )
using the homogeneous solution. Show that:
1 sin[krs (η 0 ) − krs (η)]
G(η, η 0 ) = p , (10.103)
1 + R) W (η 0 )
and
1
W (η 0 ) = − √ . (10.104)
3(1 + R)
We are omitting the k-dependence for simplicity.
With the above results we can write:
(Θ + Φ)(η, k) = C(k)S1 + D(k)S2

Z η 0
0
k Φ(η ) p
+√ dη 0 − Ψ(η 0 ) 1 + R(η 0 ) sin[krs (η) − krs (η 0 )] . (10.105)
3 0 1 + R(η 0 )
For the primordial modes, in the limit kη → 0, one gets at the dominant order:
(Θ0 + Φ)(0, k) = D(k) . (10.106)
274
Hence, it is the adiabatic mode which multiplies the cosine. Since sine and cosine
have a π/2 phase difference, the effect of different initial conditions is to change the
scales for which the effective temperature fluctuations is maximum or minimum
and hence the positions of the peaks in the CT T,` ’s.
In the adiabatic case, we have:
cos[krs (η)]
(Θ0 + Φ)(η, k) = (Θ0 + Φ)(0, k)
(1 + R)1/4
η
Φ(η 0 ) √
Z
k 0 0
+√ dη − Ψ(η ) 1 + R sin[krs (η) − krs (η 0 )] (10.107)
3 0 1+R
This is the semi-analytic (semi because the integral has to be performed numeri-
cally) formula of [Hu and Sugiyama, 1996].
The above solution (10.107) can also be used for baryons. Indeed, combining
Eq. (10.56) with Eq. (10.85) and then with Eq. (10.54) we get:
k2Ψ

0 0 3R 00 00 0 0
δb = 3Θ0 + 0 Θ0 + Φ + H(Θ0 + Φ ) + . (10.108)
τ 3
Eliminating the second derivative by means of the differential equation (10.90), we

have:
R
δb0 = 3Θ00 + 0 −k Θ0 + 3H(Θ00 + Φ0 ) .
2
(10.109)
τ (1 + R)
Just to make a rough estimative, let us neglect the second contribution (which is
divided by τ 0 anyway which is much larger than H and also than k, for suitable
scales) and use the homogeneous part of Eq. (10.107). It is straightforward then
to integrate δb0 and obtain at recombination:

rs (η∗ )
δb (η∗ , k) ∝ cos[krs (η∗ )] = cos 2π . (10.110)
λ
So the scale rs (η∗ ) ≈ 150 Mpc is relevant for baryons, too. Indeed, at about this
scale the matter power spectrum display the BAO feature, as we saw in Chapter 9.
10.4 Diffusion damping

In order to understand what happens to the CT T,` ’s when ` grows larger and larger
we need to take into account smaller and smaller scales, because of the relation
` ≈ kη0 . As discussed earlier, for larger and larger k the ratio −τ 0 /k becomes
smaller and smaller and so the TC approximation must be relaxed.
In this section then we investigate what happens to the temperature fluctua-
tions when the quadrupole moment Θ2 is taken into account. Since this analysis
accounts for the behavior of very small scales which entered the horizon deep into
the radiation-dominated epoch, we can neglect the gravitational potentials since
these, as we saw in Chapter 9, rapidly decay.
275
Moreover, being deep into the radiation-dominated epoch, we can also neglect
R and thus take 3Θ1 = Vb . Neglecting also polarisation, we have the following set
of three equations for Θ0 , Θ1 and Θ2 :
Θ00 + kΘ1 = 0 , (10.111)

3Θ01 + 2kΘ2 − kΘ0 = 0 , (10.112)
10Θ02 − 4kΘ1 = 9τ 0 Θ2 . (10.113)
In the last equation we can neglect Θ02 with respect τ 0 Θ2 , as we already did earlier,
and then find:
4k
Θ2 = − 0 Θ1 . (10.114)
9τ
The minus sign might ring some alarm, but recall that τ 0 is always negative by
definition.
Exercise 10.11. Combine the above condition with the remaining equations in
order to find a closed equation for Θ0 :
8k 2 k2

Θ000 + − Θ0
0 + Θ0 = 0 (10.115)
27τ 0 3
This is the equation for an harmonic oscillator that we have already found earlier
in Eq. (10.64), only that now there appears a damping term which is relevant
on small scales, i.e. when k ∼ −τ 0 . Baryons also provide a damping term, cf.
Eq. (10.86), but they are irrelevant in the present case since we set R = 0.
This damping term here depends on Θ2 and is time-dependent. Let us consider
it constant and assume a solution of the type Θ0 ∝ exp(iωη). Substituting this
ansatz in the equation, we find:
8k 2 k2

2
−ω + − iω + =0. (10.116)
27τ 0 3
The frequency must have an imaginary part, which accounts for the damping, thus
let us stipulate:
ω = ωR + iωI , (10.117)
Exercise 10.12. Substitute this ansatz in the equation and find:
k 4k 2
ωR = √ , ωI = − . (10.118)
3 27τ 0
Hence, we can write the general solution for Θ0 as:

√ 2 /k 2
Θ0 ∝ eikη/ 3 e−k Silk , (10.119)
276
where we have introduced the comoving diffusion length, the Silk length, as
1 4η
λ2Silk = 2
≡− . (10.120)
kSilk 27τ 0
What does the diffusion length physically represent? It is the comoving distance
travelled by a photon in a time η, but taking into account the collisions which it
is suffering, i.e. its diffusion. Let us see this in some more detail.
Since −τ 0 is the scattering rate, i.e. how many collisions take place per unit
conformal time, then −1/τ 0 is the average conformal time between 2 consecutive
collisions, which for a photon is also the average comoving distance between two
collision, i.e. the mean free path.
Now, we have:
η
λ2Silk ∝ − 0 ∝ λMFP η , (10.121)
τ
where we have used the comoving mean free path, λMFP . Now, multiply and divide
by λMFP and take the square root:
r
η
λSilk ∝ λMFP , (10.122)
λMFP
Under the square root we have the comoving distance η divided by the photon
comoving mean free path. This gives us the average number of collision N which
the photons experience up to the time η and hence:
√
λSilk ∝ N λMFP , (10.123)
which is the typical relation for diffusion. Below this scales λSilk all fluctuations
are suppressed because photons cannot agglomerate since they escape away. This
effect is known as Silk damping [Silk, 1967]. Therefore, the behaviour of the Cl ’s
for large l’s, as seen in Fig. 10.1, is also decaying, though not exactly as in the
above solution since this has to be free-streamed first.
We can do a more detailed calculation of the damping scale as follows. Let
us neglect the gravitational potential and the ` ≥ 3 multipoles as before, but
let us deal with more care of baryons and take into account polarisation. From
Eq. (10.59) we have:
R
Vb = 3Θ1 + 0 (Vb0 + HVb ) , (10.124)
τ
and the six equations for the monopole, dipole and quadrupole of the temperature
fluctuations and polarisation:
Θ00 + kΘ1 = 0 , (10.125)

3Θ01 + 2kΘ2 − kΘ0 = τ 0 (3Θ1 − Vb ) , (10.126)
10Θ02 − 4kΘ1 = 9τ 0 Θ2 − τ 0 ΘP 0 − τ 0 ΘP 2 , (10.127)
2Θ0P 0 + 2kΘP 1 = τ 0 ΘP 0 − τ 0 ΘP 2 − τ 0 Θ2 , (10.128)
3Θ0P 1 + 2kΘP 2 − kΘP 0 = 3τ 0 ΘP 1 , (10.129)
10Θ0P 2 − 4kΘP 1 = 9τ 0 ΘP 2 − τ 0 ΘP 0 − τ 0 Θ2 . (10.130)
277
R
Now, assuming a solution of the type exp(i ωdη) for all the above 7 variables and
also assuming that ω H, we have:
3Θ1
Vb = , (10.131)
1 + Riωηc
where we have defined ηc ≡ −1/τ 0 as the the average conformal time between 2
consecutive collisions. We have thus a closed system for Θ0 , Θ1 , Θ2 , ΘP 0 , ΘP 1 and
ΘP 2 :
iωΘ0 + kΘ1 = 0 , (10.132)

R
−kΘ0 + 3iωΘ1 1 + + 2kΘ2 = 0 , (10.133)
1 + Riωηc
−4kηc Θ1 + (10iωηc + 9)Θ2 − ΘP 0 − ΘP 2 = 0 , (10.134)
−Θ2 + (2iωηc + 1)ΘP 0 + 2kηc ΘP 1 − ΘP 2 = 0 , (10.135)
−kΘP 0 + 3(iωηc + 1)ΘP 1 + 2kηc ΘP 2 = 0 , (10.136)
−Θ2 − ΘP 0 − 4kηc ΘP 1 + (10iωηc + 9)ΘP 2 = 0 . (10.137)
We have already arranged the variables in order for the system matrix to appear
clearly. The determinant of this matrix, in order to have a non trivial solution,
must be zero. Considering the limit ωηc 1, and keeping the first-order only in
ωηc we get:
k2 2i
− ω 2 (1 + R) + ωηc 37k 2 − 285(1 + R)ω 2 + 15ω 2 R2 = 0 .

(10.138)
3 30
In order to solve for ω, let us again employ the smallness of ωηc and stipulate that:
ω = ω0 + δω , (10.139)
where δω is a small correction. From the above equation is then straightforward

to obtain:
k2
− ω02 (1 + R) = 0 , (10.140)
3
2i
−2ω0 δω(1 + R) + ω0 ηc 37k 2 − 285(1 + R)ω02 + 15ω02 R2 = 0 .

(10.141)
30
The first equation gives the result that we have already encountered:
k2
ω02 = = k 2 c2s (10.142)
3(1 + R)
which, substituted in the second equation, gives us:
iηc k 2 R2

16
δω = + (10.143)
6(1 + R) 15 1 + R
This result was obtained for the first time by [Kaiser, 1983]. See also the derivation
of [Weinberg, 2008].
278
Therefore, the evolution of the multipoles is proportional to the following factor:
Z
2 2
exp i ωdη = eikrs (η) e−k /kSilk , (10.144)
where
η
R2
Z
1 0 1 16
2
≡− dη 0 + (10.145)
kSilk 0 6τ (1 + R) 15 1 + R
From the best fit values of the parameter of the ΛCDM model we have:
dSilk = 0.0066 Mpc (10.146)
10.5 Line-of-sight integration

The approximate solutions found earlier are based on the TC limit, which allows us
to take into account just the monopole and the dipole until recombination and then
to better understand the physics behind the CMB anisotropies. On the other hand,
observation demands more precise calculations to be compared with and therefore,
at the end, numerical computation and codes such as CLASS are needed. Even
so, there is a more efficient way of computing predictions on the CMB anisotropies
than dealing directly with the hierarchy of Boltzmann equation and that is to
formally integrate along the photon past light-cone according to a semi-analytic
technique called line-of-sight integration, due to Seljak and Zaldarriaga [Seljak
and Zaldarriaga, 1996], and which was the basis for the CMBFAST code.3
Recall the photon Boltzmann equations (5.114) and (5.115):

0 0 0 1
Θ + ikµΘ = −Φ − ikµΨ − τ Θ0 − Θ − iµVb − P2 (µ)Π , (10.147)
2

0 0 1
ΘP + ikµΘP = −τ −ΘP + [1 − P2 (µ)]Π , (10.148)
2
where Π = Θ2 + ΘP 2 + ΘP 0 . Let us rewrite them as follows:

0 0 0 0 1
Θ + (ikµ − τ )Θ = −Φ − ikµΨ − τ Θ0 − iµVb − P2 (µ)Π ≡ S(η, k, µ) , (10.149)
2
0
τ
Θ0P + (ikµ − τ 0 )ΘP = − [1 − P2 (µ)]Π ≡ SP (η, k, µ) , (10.150)
2
where we have introduced two source functions on the right hand sides. Note
that the dependence in on k and not on k = kẑ because we are considering the
equations for the transfer functions. Afterwards, before performing the anti-Fourier
transform, we must rotate back k̂ in a generic direction.
Let us write the left hand sides as follows:
d
Θ0 + (ikµ − τ 0 )Θ = e−ikµη+τ Θ eikµη−τ ,

(10.151)
dη
3
https://lambda.gsfc.nasa.gov/toolbox/tb_cmbfast_ov.cfm
279
with a similar expression for ΘP . Substituting these into the Boltzmann equations
and integrating formally from a certain initial ηi → 0 to today η0 , we get:
Z η0
−τ (η0 ) ikµ(ηi −η0 )−τ (ηi )
Θ(η0 )e = Θ(ηi )e + dη eikµ(η−η0 )−τ (η) S(η, k, µ) , (10.152)
Z ηηi0
ΘP (η0 )e−τ (η0 ) = ΘP (ηi )eikµ(ηi −η0 )−τ (ηi ) + dη eikµ(η−η0 )−τ (η) SP (η, k, µ) , (10.153)
ηi
Now recall the definition of the optical depth:

Z η0
τ≡ dη 0 ne σT a . (10.154)
η
It is clear then that τ (η0 ) = 0 and, since ηi → 0 is deep into the radiation-
dominated epoch, then τ ∝ 1/η is very large and we can neglect exp[−τ (ηi )].
Therefore, we are left with
Z η0
Θ(η0 , k, µ) = dη eikµ(η−η0 )−τ (η) S(η, k, µ) , (10.155)
Z 0η0
ΘP (η0 , k, µ) = dη eikµ(η−η0 )−τ (η) SP (η, k, µ) . (10.156)
0
where we have already implemented the limit ηi → 0. Now we calculate the Θ` ’s

inverting the Legendre expansion as done in Eq. (5.113) and obtain:
Z 1 Z η0
1 dµ
Θ` (η0 , k) = P` (µ) dη eikµ(η−η0 )−τ (η) S(k, η, µ) , (10.157)
(−i)` −1 2 0
Z 1 Z η0
1 dµ
ΘP ` (η0 , k) = `
P` (µ) dη eikµ(η−η0 )−τ (η) SP (k, η, µ) , (10.158)
(−i) −1 2 0
The source terms have µ-dependent contributions (up to µ2 ) that we can handle
integrating by parts. Take for example the −ikµΨ contribution of S(k, η, µ). Let
IΨ be its integral, which can be rewritten as follows:
Z η0 Z η0
d ikµ(η−η0 )
IΨ ≡ − dη ikµΨe ikµ(η−η0 )−τ (η)
=− dη Ψe−τ (η) e , (10.159)
0 0 dη
and now it is easy to integrate by parts and obtain:
Z η0
−τ (η) ikµ(η−η0 ) η0 d −τ (η)
IΨ = − Ψe e 0
+ dη eikµ(η−η0 ) Ψe . (10.160)
0 dη
The first contribution gives −Ψ(η0 ), i.e. the gravitational potential evaluated at
present time. This is just an undetectable offset that we incorporate into the
definition of Θ` (η0 , k), as the observed anisotropy, like we did at the beginning of
this Chapter when dealing with the free-streaming solution.
Exercise 10.13. Take care of the term containing µ2 , in P2 (µ). Show that:
Z η0
1 η0 d2
Z
0 2 ikµ(η−η0 )−τ (η)
dη eikµ(η−η0 ) 2 τ 0 Πe−τ (η) . (10.161)

dη τ µ Πe =− 2
0 k 0 dη
280
Combining all the terms treated with integration by parts, we get:
Z 1 Z η0
1 dµ
Θ` (k, η0 ) = P` (µ) dη eikµ(η−η0 )
(−i)` −1 2 0
0 0 −τ 0

0 0 τ Π −τ −τ τ Vb e 3 0 −τ 00

− Φ + τ Θ0 + e + Ψe − − 2 τ Πe , (10.162)
4 k 4k
Z 1 Z η0
3 dµ ikµ(η−η0 ) 0 −τ 1 0 −τ 00
ΘP ` (k, η0 ) = − P` (µ) dη e τ Πe + 2 τ Πe . (10.163)
4(−i)` −1 2 0 k
Using now the relation of Eq. (10.15), we can cast the above equations as:
Z η0
Θ` (η0 , k) = dη S(η, k)j` [k(η0 − η)] , (10.164)
0
Z η0
ΘP ` (η0 , k) = dη SP (η, k)j` [k(η0 − η)] . (10.165)
0
with

Π 1 3
0
S(η, k) ≡ (Ψ − Φ )e0 −τ
+ g Θ0 + + Ψ + (gVb )0 + 2 (gΠ)00 , (10.166)
4 k 4k
3 3
SP (η, k) ≡ gΠ + 2 (gΠ)00 , (10.167)
4 4k
where we have introduced the visibility function:
g(η) ≡ −τ 0 e−τ (10.168)
Exercise 10.14. Show that the visibility function is normalised to unity, i.e.
Z η0
dη g(η) = 1 . (10.169)
0
The visibility function represents the Poissonian probability that a photon is

last scattered at a time η. It is very peaked at a time that we define as the one of
recombination, i.e. at η = η∗ , because for η > η∗ it is basically zero, since τ 0 = 0.
Before recombination, in the radiation-dominated epoch, we saw that −τ 0 ∝ 1/η 2
and thus τ ∝ 1/η and g ∝ exp(−1/η)/η 2 , i.e. it goes to zero exponentially fast.
In Fig. 10.4 we plot the numerical calculation of the visibility function per-
formed with CLASS for the standard model. Note the peak at about z = 1000,
which has always been our reference for the recombination redshift. Note also an-
other peak at about z = 10, representing the epoch of reionisation. Until now
we have used the peakedness of the visibility function as if it were a Dirac delta
δ(η − η∗ ), i.e. we have made the sudden recombination approximation. From
Fig. 10.4 we can appreciate that it is a good approximation (mind the logarithmic
scale there). As usual, in cosmology but not only, the calculations get more and
more complicated and impossible to do analytically the more precision we demand.
281
��-�
��-�
��-��
��
Figure 10.4: Visibility function g as function of the redshift from the numerical
calculation performed with CLASS for the standard model.
Inserting the source terms (10.166) and (10.167) in the expressions for Θ` (η0 , k)
and ΘP ` (η0 , k) in the expression of Θ` and integrating by parts, we get:
Z η0
Π
Θ` (k, η0 ) = dη g Θ0 + Ψ + j` [k(η0 − η)]
0 4
Z η0 Z η0
gVb d 3gΠ d2
− dη j` [k(η0 − η)] + dη 2 2 j` [k(η0 − η)]
0 k dη 0 4k dη
Z η0
+ dη e−τ (Ψ0 − Φ0 )j` [k(η0 − η)] ,(10.170)
0
Z η0 Z η0
3gΠ 3gΠ d2
ΘP ` (k, η0 ) = dη j` [k(η0 − η)] + dη 2 2 j` [k(η0 − η)] .(10.171)
0 4 0 4k dη
Assuming the visibility function to be a Dirac delta δ(η − η∗ ), i.e. the sudden re-
combination mentioned earlier, and neglecting Π, we recover formula (10.24). Note
that neglecting Π no polarisation is present. Indeed, from the above equation we
see that a non-zero quadrupole moment of the photon distribution at recombination
is essential in order to have polarisation.
The above equations still need the Boltzmann hierarchy in order to be inte-
grated, but just up to ` = 4 (because Θ2 and Θ4 moments are contained in the
equation for Θ03 ) and hence are much more convenient from the computational
point of view.
The partial wave expansion of Θ given in Eq. (7.77):
X
Θ(k, µ) = (−i)` (2` + 1)P` (µ)Θ` (k) , (10.172)
`
282
and that we have used in the above calculations is valid as long as k̂ = ẑ. Now we
have to rotate it in a general direction before performing the Fourier anti-transform.
The task is simple because the temperature fluctuation is a scalar. Therefore:
X
Θ(k, k̂ · p̂) = (−i)` (2` + 1)P` (k̂ · p̂)Θ` (k) . (10.173)
`
The same is not true for ΘP , since the Stokes parameters are not scalars.
Using the definition of aT,`m given in Eq. (7.73) we can then write:
d3 k
Z X Z
S 2 m∗ `
aT,`m = d n̂ Y` (n̂) (−i) (2` + 1) P (k̂ · p̂)α(k)Θ` (k) . (10.174)
3 `
l
(2π)
The integration is over d2 n̂, hence we must change p̂ → n̂ = −p̂ in the Legendre
polynomial. This gives an extra (−1)` factor, due to the parity of the Legendre
polynomials, and then using the addition theorem we obtain:
d3 k
Z Z
S 2 m∗
X
` 4π
aT,`m = d n̂ Y` (n̂) i (2` + 1) 3
α(k) 0
l
(2π) 2` + 1
` 0
0 0
X
Y`∗m
0 (k̂)Y`m
0 (n̂)Θ` (k) . (10.175)
m0 =−`0
Now the integration over the whole solid angle can be performed and the orthonor-
mality of the spherical harmonics can be employed, obtaining thus:
d3 k m∗
Z
aST,`m `
= 4πi Y (k̂)α(k)Θ` (k) (10.176)
(2π)3 `
This formula, together with Eq. (10.170) allows us to explicitly calculate the scalar
contribution to the aT,`m ’s. Earlier, we have focused on the CT T,` ’s only, for which
the calculations are simpler because there is no need of performing a spatial rota-
tion, but we need to know the explicit form of the aT,`m ’s in order to compute the
TE correlation spectrum of Eq. (7.87).
10.6 Finite thickness effect and reionization

In this section we discuss two more effects that influence the CMB spectrum,
namely the finite thickness effect and the reionisation. The first one is related to
the fact that the visibility function g in Fig. 10.4 is very peaked but it is not a
Dirac delta. In other words, CMB photons do not last scatter all at once at η∗
but during a finite amount of time say ∆η∗ . This is the finite thickness effect.
Physically, on scales smaller than the thickness ∆η∗ we expect fluctuations to be
washed out because they are averaged over a finite amount of time. This is similar
to what is called Landau damping (although the latter arises from a spread in
frequency and not in time). It may seem that Landau damping is a small effect,
but actually is of the same order of Silk damping and therefore it must be taken
into account.
283
Let us take advantage of this investigation and derive the form of the visibility
function here. The question is: what is the probability that a photon last scatter
during some sufficiently small interval between the instants η and η + ∆η? The
time interval ∆η is sufficiently small so that only one collision can take place into
it. The attentive reader has noticed that this is the same requirement we make
when we derive the Poisson distribution, cf. Chapter 12, which in fact rules the
statistics of e.g. scattering process.
So, we divide the time interval η0 − η in many, i.e. N ≡ (η0 − η)/∆η, intervals
and write the probability as:

∆η ∆η ∆η ∆η
∆P = 1− 1− ... 1 − , (10.177)
ηc (η) ηc (η1 ) ηc (η2 ) ηc (ηN )
where recall that ηc ≡ −1/τ 0 is the average time between two consecutive collisions
and it is time-dependent. We have chosen the time interval ∆η small enough in
order for ∆η/ηc to be the probability of having one scattering during its duration
and hence 1 − ∆η/ηc being that of having no scattering. Now, in the limit ∆η → 0
we can write:
Z η0
dη dη
dP = exp − = −τ 0 exp [−τ (η)] dη = g(η)dη , (10.178)
ηc η ηc
i.e. the visibility function defined in Eq. (10.168) appears. As we have anticipated,
the maximum of the visibility function occurs in a time that we dub η∗ , i.e. the
recombination time if we make the assumption of sudden recombination. From the
condition for the extrema of a function:
g 0 (η) = −τ 00 e−τ + (τ 0 )2 e−τ = 0 , (10.179)
we get:
− τ 00 = (τ 0 )2 , (10.180)
as the condition which defines η∗ . Therefore, employing the definition of the optical
depth, we get:
(ne σT a)0∗ = −(ne σT a)2∗ , (10.181)
where the derivative is evaluated at η∗ , as well as the function on the right hand side.
Now, let us write the free-electron number density as ne = Xe nb , i.e. introducing
the free-electron fraction and the baryon number density and recall from Boltzmann
equation, cf. Chapter 3, that:
1.44 × 104
Xe0 = − HXe . (10.182)
z
Evaluating the above equation at η∗ and combining it with the extrema condition
for the visibility function, we get:
1.44 × 104 K
Xe (η∗ ) ≈ H(η∗ ) ≡ H(η∗ ) , (10.183)
z∗ (nb σT a)∗ (nb σT a)∗
and from this we get the recombination redshift z∗ ≈ 1050.
284
Let us approximate the visibility function by expanding it about its maximum:

0 1 0 00 2
g(η) = exp[ln(−τ ) − τ ] ≈ exp − [τ − ln(−τ )]∗ (η − η∗ ) , (10.184)
2
i.e. we have a Gaussian function. Now we determine the second derivative, and
hence the variance of the distribution by using Boltzmann equation and employing
the following approximation: we consider only the first derivative of Xe to be
different from zero. All the other quantities are approximately constant indeed
since recombination takes place quite rapidly. So, we can write:
0 0
τ 00 nb Xe0 σT a

0
τ − 0 = −nb Xe σT a − . (10.185)
τ ∗ nb Xe σ T a ∗
Now, within our approximation Xe0 /Xe is constant, and thus:
0
τ 00

0
τ − 0 = −(nb Xe0 σT a)∗ = (nb Xe σT a)2∗ = K 2 H(η∗ )2 . (10.186)
τ ∗
Hence the visibility function can be approximated as a Gaussian function:

KH∗ 1 2 2
g(η) ≈ √ − (KH∗ ) (η − η∗ ) , (10.187)
2π 2
with variance 1/(KH∗ ).
When we substitute this approximation of the visibility function in the line-of-
sight integrals of Eqs. (10.170) and (10.171) we can extract the spherical Bessel
functions as jl [k(η0 − η∗ )], since η0 is much larger than any conformal time about
recombination, and the other integrals are oscillating functions of krs (η), as we
saw in Eq. (10.107). We can approximate this about η∗ , as follows:
Z η
1 k
krs (η) = k dη 0 p ≈ krs (η∗ ) + p (η − η∗ ) . (10.188)
0 3(1 + R) 3(1 + R∗ )
Hence, we have finally a conformal time Gaussian integral of the following type:
Z ∞ " #
1 ik
dη exp − (KH∗ )2 (η − η∗ )2 + p (η − η∗ ) . (10.189)
−∞ 2 3(1 + R∗ )
Now, using the formula for the Gaussian integral:
Z ∞ r
−ax2 +bx π b2 /4a
e dx = e , (10.190)
−∞ a
we can conclude that the Θ` ’s get an additional damping factor of the following
form:
k2

exp − , (10.191)
6(1 + R∗ )(KH∗ )2
so a new damping scale appears:
1 1
d2Landau = 2
≡ (10.192)
kLandau 6(1 + R∗ )(KH∗ )2
285
which, following [Weinberg, 2008], we call Landau damping scale. For the
ΛCDM model best fit parameters we have:
dLandau ≈ 0.0048 Mpc (10.193)
Now let us turn to reionisation. At a redshift of about 10 hydrogen gets ionised

again by the ultraviolet radiation of the first structures. Hence, the free-electron
fraction grows again increasing the probability of a CMB photon to be scattered
again, cf. the extra bump in the visibility function in Fig. 10.4. Following the
calculation done above in order to obtain the visibility function, we know that the
probability for a photon not to be scattered from reionisation until today is:
exp[−τ (ηreion )] , (10.194)
and of course the one for being scattered is 1 minus the above quantity, which we
compute now:
Z η0 Z zreion
dz
τ (ηreion ) = dη ne σT a = σT ne . (10.195)
ηreion 0 (1 + z)2 H
For the free-electron number density we can write:

3H02
ne = 0.88nb Xe = 0.88 Ωb0 (1 + z)3 , (10.196)
8πmb
where the factor 0.88 is due to the fact that not all the baryons are electrons or pro-
tons, but there are also neutrons in Helium nuclei. Assuming matter-domination
and instantaneous reionisation, i.e.
H2 = H02 Ωm0 (1 + z)3 , (10.197)
and Xe = 1 for z < zreion , we get:

Ωb0 h2 3/2
τ (zreion ) ≈ 0.04 √ z . (10.198)
Ωm0 h2 reion
Hence, the probability for a CMB photon not to be scattered for reionisation taking
place at redshift zreion = 10 is about 0.99, i.e. very high. Those photons which are
scattered are mixed up, hence the correlation in their temperature is destroyed.
So, the effect of reionisation on the CT T,` ’s is simply to weigh them by a factor
exp(−2τreion ), the factor 2 appearing because the spectrum is a quadratic function
of the temperature fluctuations.
10.7 Cosmological parameters determination

In this section we discuss how the CMB TT spectrum, i.e. the CT T,` ’s, are sensitive
to the cosmological parameters. We have learned in this Chapter about many
quantities which are of relevance in forming the shape of the spectrum but we
have not actually derived an analytic, approximated formula in order to see this
explicitly. These can be found in [Mukhanov, 2005] and [Weinberg, 2008]. Here
286
instead we plot with CLASS various spectra for varying parameters and discuss
the physics behind the changes.
Note that, for the standard Λ model, 6 of the overall parameters are usually
left free and constrained by observation:
1. The amplitude of the primordial power spectrum: AS ;
2. The primordial tilt: nS ;
3. The baryonic abundance: Ωb0 h2 ;
4. The CDM abundance: Ωc0 h2 ;
5. The reionization epoch: zreion ;
6. The sound horizon at recombination: rs (η∗ ), which is related to the Hubble

constant value H0 .
The other parameters can be derived by these ones. In particular, the amount of
radiation is already well known by measuring the CMB temperature and so the
amount of Λ and curvature is determined via the positions of the peaks, which
depend on rs (η∗ ), which in turn depends on the baryon content.
In Figs. 10.5 and 10.6 we start to show the numerical calculation of CMB TT
power spectrum decomposed in the contributions discussed in this Chapter. See
also [Wands et al., 2016]. We consider the ΛCDM as fiducial model.
��
��
��
��
��
�
� ��
Figure 10.5: Total CMB TT power spectrum (blue line) computed with CLASS and
decomposed in the physically different contributions: Sachs-Wolfe effect (yellow
line), early-times ISW effect (green line), late-times ISW effect (red line), Doppler
effect (purple line), and polarisation (brown line).
In Fig. 10.7 we show what happens to CMB TT the spectrum for Ωb0 h2 =
0.010, 0.014, 0.018, 0.022, 0.026, 0.030, 0.034. Taking the first peak height as refer-
ence, the larger the value of Ωb0 h2 is, the higher the peak is. When we vary one
287
��
��
��
��
� ��
Figure 10.6: Same as Fig. 10.5 but in logarithmic scale, in order to better distin-
guish the weakest contributions.
��
��
��
��
��
��
��
�
� ��
Figure 10.7: CMB TT power spectrum computed with CLASS and vary-
ing Ωb0 h2 . From the lowest first peak to the highest: Ωb0 h2 =
0.010, 0.014, 0.018, 0.022, 0.026, 0.030, 0.034.
288
of the density parameters, since their sum must be equal to one that means that
also something else must vary. In this case we have chosen to vary ΩΛ .
Why so? We have seen that baryons loading makes compression favoured over
rarefaction and hence the first and the third peaks are higher for higher values of
Ωb0 h2 , but the second one is lower. In other words, the peaks relative height is
very sensitive to the baryon content. The position of the first peak does not change
much because it is most sensitive to the spatial curvature and this has been fixed
to zero. Finally, the curves for larger Ωb0 h2 , as we commented, have less ΩΛ and
therefore less ISW effect. For these reason they are slightly lower for small `.
��
��
��
��
��
��
�
� ��
Figure 10.8: CMB TT power spectrum computed with CLASS and vary-
ing Ωc0 h2 . From the highest first peak to the lowest: Ωc0 h2 =
0.09, 0.10, 0.11, 0.12, 0.13, 0.14, 0.15.
In Fig. 10.8 we show what happens to CMB TT the spectrum for Ωc0 h2 =
0.09, 0.10, 0.11, 0.12, 0.13, 0.14, 0.15. Taking the first peak height as reference, the
larger the value of Ωc0 h2 is, the lower the peak is. This behaviour is the opposite of
the one that we found by varying Ωb0 h2 . Mostly CDM intervenes through the SW
effect since it dominates the gravitational potential Ψ at recombination. The first
peak is affected more because it corresponds to large scales, basically the horizon
at recombination, and there the transfer function is approximately unit, meaning
that −Ψ is as large as possible. The subsequent peaks correspond to scales which
entered the horizon much earlier and therefore the CDM influence there is weak.
In this case also we have chosen to vary ΩΛ in order to keep the total density
budget. Indeed, the more CDM, the less Λ and the less ISW effect, as expected.
In Fig. 10.9 we show what happens to CMB TT the spectrum for τreion =
0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7. As we have commented in the previous section, the
overall effect of reionisation is simple because it happens very lately: a damping of
the order exp(−2τreion ) for multipoles larger than a certain `reion which we infer to
be about 10 from the plots in Fig. 10.9.
From Fig. 10.10 we can appreciate how the CMB TT power spectrum is af-
fected by the spatial geometry of the universe. From the leftmost spectrum to the
289
��
��
��
��
��
�
� ��
Figure 10.9: CMB TT power spectrum computed with CLASS and varying τreion .
From the highest first peak to the lowest: τreion = 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7.
��
��
��
��
��
�
� ��
Figure 10.10: CMB TT power spectrum computed with CLASS and varying ΩK0 .
From the left to the right: ΩK0 = −0.2, −0.1, 0, 0.1, 0.2.
290
rightmost one ΩK0 = −0.2, −0.1, 0, 0.1, 0.2. Hence, the position of the first peak is
of great importance in order to determine whether our universe is closed or open.
Note that the flat case is a limiting value which we cannot determine observation-
ally, because of the experimental error; we can only conclude that observation is
consistent with ΩK0 = 0, i.e. this value is not ruled out.
As we saw in Eq. (10.107), the length scale associated to the acoustic peaks is
the sound horizon at recombination:
Z η∗
rs (η∗ ) = cs dη , (10.199)
0
where the speed of sound of the baryon-photon plasma is given by Eq. (10.87):
1 4Ωγ0
c2s = = . (10.200)
3(1 + R) 3(4Ωγ0 + 3Ωb0 a)
The physical sound horizon is given by:
Z t∗ Z ∞
phys cs (z)
rs (z∗ ) = cs (t)dt = dz , (10.201)
0 z∗ H(z)(1 + z)
i.e. integrating the lookback time. We need the physical quantity in order to relate
it with the angular-diameter distance to recombination:
Z z∗
1 dz
dA (z∗ ) = , (10.202)
(1 + z∗ ) 0 H(z)
and thus estimate the multipole corresponding to the first peak:
1 dA (z∗ )
`1st ≈ = . (10.203)
θ1st rsphys (z∗ )
Let us approximate the physical sound horizon by assuming cs constant and a
matter-dominated universe. We have thus:
Z ∞
phys cs dz 2cs 1
rs (z∗ ) ≈ √ 5/2
= √ , (10.204)
H0 Ωm0 z∗ (1 + z) 3H0 Ωm0 (1 + z∗ )3/2
and for the angular-diameter distance we also assume a matter plus Λ universe:
Z z∗
1 dz
dA (z∗ ) = p . (10.205)
H0 (1 + z∗ ) 0 Ωm0 (1 + z)3 + (1 − Ωm0 )
Exercise 10.15. Show that dA (z∗ ) can be approximated as:

2
dA (z∗ ) ≈ √ (9 − 2Ω3m0 ) . (10.206)
7H0 (1 + z∗ ) Ωm0
Hence, we have:
√
`1st ≈ 0.74 1 + z∗ (9 − 2Ω3m0 ) ≈ 220 , (10.207)
which clearly shows how the position of the first peak changes as function of the
total matter content.
In fig. 10.11 we show how the initial conditions dramatically affect the CMB
TT power spectrum and how the adiabatic ones are favoured by observation (when
comparing with the data points of Fig. 10.1).
291
��
��
��
��
��
�
� ��
Figure 10.11: CMB TT power spectrum computed with CLASS and varying initial
conditions: adiabatic (blue line), baryon isocurvature (yellow line), CDM isocurva-
ture (green line), neutrino density isocurvature (red line), neutrino velocity isocur-
vature (purple line).
10.8 Tensor contribution to the CMB TT correla-

tion
Tensor perturbations also contribute to generate temperature anisotropies, as we
see in Eq. (5.138), which we report here after renormalising to the primordial mode
β(k, λ), cf. Eq. (7.101):
0
hT

∂ 0
+ ikµ − τ Θ(T ) (η, k, µ) + =
∂η 2

0 3 (T ) 1 (T ) 1 (T ) 3 (T ) 6 (T ) 3 (T )
−τ Θ + Θ2 + Θ0 − ΘP 4 + ΘP 2 − ΘP 0
70 4 7 10 70 7 5
0 T
≡ −τ S (η, k) , (10.208)

∂ (T )
+ ikµ − τ 0 ΘP (η, k, µ) = τ 0 S T (η, k) , (10.209)
∂η
The label λ representing the two possible states of helicity is absent because of the
renormalisation with β(k, λ). It represents the fact that the evolution of the two
helicities is the same.
The line-of-sight solutions of the above equations are the following:
Z η0
(T )
dη eikµ(η−η0 )−τ −h0T /2 − τ 0 S T (η, k) ,

Θ (η0 , k, µ) = (10.210)
0
Z η0
(T )
ΘP (η0 , k, µ) = dη eikµ(η−η0 )−τ τ 0 S T (η, k) . (10.211)
0
We now focus on Θ(T ) (η0 , k, µ). Defining:
S T (η, k) ≡ e−τ −h0T /2 − τ 0 S T (η, k) ,

(10.212)
292
and using Eq. (5.131) and Eq. (5.32), the tensor contribution to the temperature
fluctuation is made up of the sum of the following two contributions:
r Z η0
π λ
fλ (kẑ, p̂) ≡ 4 Y2 (p̂) dη S T (η, k)e−iµkr(η) , (10.213)
15 0
where r(η) ≡ η0 − η and where we stress that the result holds true for k̂ = ẑ
since this was the condition under which we derived the Boltzmann equation for
photons. We cannot yet sum over λ because we have to include β(k, λ) first. For
this reason, we shall work on fλ (kẑ, p̂).
In order to investigate temperature fluctuations in the sky, we need to anti-
transform Θ(T ) (k, p̂) in order to employ the usual expansion:
X Z
(T )
Θ (n̂) = aT,`m Y` (n̂) , aT,`m = d2 n̂ Y`m∗ (n̂)Θ(T ) (n̂) .
T m T
(10.214)
`m
So, let us proceed as follows. We trade p̂ for the line-of-sight n̂ = −p̂ and use the
expansion of a plane wave in spherical harmonics:
X
eik̂·n̂kr = iL YLM ∗ (k̂)YLM (n̂)jL (kr) , (10.215)
LM
in Eq. (10.213).
Exercise 10.16. First of all, since k̂ = ẑ then show that:

r
2L + 1
YLM ∗ (k̂) = YLM ∗ (ẑ) = δM 0 , (10.216)
4π
i.e. for θ = 0 (representing the ẑ direction) the spherical harmonics are non-
vanishing only if M = 0.
Therefore, we can write:

2 X √ Z η0
fλ (kẑ, n̂) = √ Y2λ (n̂) iL 2L + 1YL0 (n̂) dη S T (η, k)jL (kr) . (10.217)
15 L 0
The idea is now to perform a rotation in order to put k̂ in a generic direction.

But then also n̂ rotates and therefore we need to know how a spherical harmonics
behaves under rotations. In order to deal with just one spherical harmonic we take
advantage of the following decomposition:
r
5(2L + 1) X √ 0
Y2±2 (n̂)YL0 (n̂) = 2L + 1
4π L0
L 2 L0 L 2 L0

YL±2
0 (n̂) , (10.218)
0 ±2 ∓2 0 0 0
293
where we have employed the Wigner 3j-symbols, which are coefficients appear-
ing in the quantum theory of angular momentum, when we combine two angular
momenta and we want to write the state of total angular momentum as a lin-
ear combination on the basis of the tensor product of the two combined angular
momenta. They are an alternative to the (perhaps more commonly used) Clebsch-
Gordan coefficient See e.g. [Landau and Lifshits, 1991] and [Weinberg, 2015].
This expansion allows us to deal with just one spherical harmonics. Now we
take advantage of the properties of the spherical harmonics under spatial rotation,
i.e.
`
0
X (`)
m
Y` (Rn̂) = Dm0 m (R−1 )Y`m (n̂) (10.219)
m0 =−`
(`)
where the Dm0 m are the elements of the Wigner D-matrix. See [Landau and
Lifshits, 1991] for more detail. The above R is a generic rotation. Of course, we
are interested in a R(k̂) rotation which brings k̂ in a generic direction. Hence, we
can write:
X 2L + 1 X √
L 2 L0

L 0
fλ (k, n̂) = i √ 2L + 1
3π L0 0 λ −λ
L
η0
L 2 L0 X (L0 )
Z
0
Dm0 λ [R(k̂)]YLm0 (n̂) dη S T (η, k)jL (kr) . (10.220)
0 0 0 0 0
m
Here we have dubbed Rn̂ the original line of sight and n̂ the resulting one after
the rotation.
Now we can perform the Fourier anti-transform. Let us multiply fλ (k, n̂) by
β(k, λ) and Y`m∗ (n̂) and integrate over d2 n̂ in order to obtain the aTT,`m ’s. We
obtain:
X 2L + 1 √
L 2 `

T L
a`m,±2 = i √ 2` + 1
3π 0 ±2 ∓2
L
η0
d3 k (`)
Z Z
L 2 `
D m±2 [R( k̂)]β(k, ±2) dη S T (η, k)jL (kr) . (10.221)
0 0 0 (2π) 3
0
We have used here the orthonormality relation of the spherical harmonics and
distinguished the contributions from different helicities. Of course a`m = a`m,+2 +
a`m,−2 .
It is now time to compute the 3j symbols and to perform the summation over
L. A general formula for those was obtained in [Racah, 1942], but we can read
their expression from [Landau and Lifshits, 1991]. We then have the only following
non-vanishing occurrences:
s
` 2 ` `(` + 1)
= (−1)`+1 , (10.222)
0 0 0 (2` − 1)(2` + 1)(2` + 3)
s
`+2 2 ` 3(` + 1)(` + 2)
= (−1)` , (10.223)
0 0 0 2(2` + 1)(2` + 3)(2` + 5)
s
`−2 2 ` 3`(` − 1)
= (−1)` . (10.224)
0 0 0 2(2` − 3)(2` − 1)(2` + 1)
294
In particular, there is no contribution coming from L = ` ± 1. The other three
relevant (i.e. not considering those for L = ` ± 1 which are non-vanishing in this
case) symbols are:
s
` 2 ` 3(` − 1)(` + 2)
= (−1)` , (10.225)
0 ±2 ∓2 2(2` − 1)(2` + 1)(2` + 3)
s
`+2 2 ` 1 (` − 1)`
= (−1)` , (10.226)
0 ±2 ∓2 2 (2` + 1)(2` + 3)(2` + 5)
s
`−2 2 ` 1 (` + 1)(` + 2)
= (−1)` , (10.227)
0 ±2 ∓2 2 (2` − 3)(2` − 1)(2` + 1)
Exercise 10.17. Derive the above expressions for the relevant Wigner 3j symbols
given in [Landau and Lifshits, 1991] and put them in Eq. (10.221). Show that:
s Z η0
d3 k (`)
Z
T ` (2` + 1)(` + 2)!
aT,`m,±2 = −i 3
Dm±2 [R(k̂)]β(k, ±2) dη S T (η, k)
8π(` − 2)! (2π) 0

j`−2 (kr) 2j` (kr) j`+2 (kr)
+ + (10.228)
.
(2` − 1)(2` + 1) (2` − 1)(2` + 3) (2` + 1)(2` + 3)
Recall that r = r(η) ≡ η0 − η.
Exercise 10.18. Show that, using the recurrence relation [Abramowitz and Ste-
gun, 1972]:
j` (x) j`−1 (x) + j`+1 (x)
= , (10.229)
x 2` + 1
we can write:
s
d3 k (`)
Z
T ` (2` + 1)(` + 2)! X
aT,`m = −i D [R(k̂)]β(k, λ)
8π(` − 2)! λ=±2 (2π)3 m,λ
Z η0
j` (kr)
dη S T (η, k) . (10.230)
0 (kr)2
The Wigner D-matrix can be related to the spin-weighted spherical harmonics

as follows:
r r
(`) 4π −m 4π m∗
Dm,±2 (k̂) = ±2 Y` (k̂) = ∓2 Y` (k̂) , (10.231)
2` + 1 2` + 1
so we have:
s Z η0
d3 k
Z
(` + 2)! X j` (kr)
aTT,`m `
= −i λ Y `
m∗
(k̂)β(k, λ) dη S T
(η, k)
2(` − 2)! λ=±2 (2π)3 0 (kr)2
(10.232)
295
This is our main result of this section. It is not surprising that Y`m∗ (k̂) eventu-
ally appeared, being GW a spin-2 field.
In order to compute the tensor contribution to the CT T,` ’s, we perform the
ensemble average:
haTT,`m aTT,`∗ 0 m0 i = CTTT,` δ``0 δmm0 . (10.233)
Exercise 10.19. Assuming Gaussian perturbations, using Eq. (7.102) and the
orthogonality property of the Wigner D-matrices or the spin-weighted spherical
harmonics:
Z
(`) (`0 )∗ 4π
d2 k̂ Dm,±2 [R(k̂)]Dm0 ,±2 [R(k̂)] = δ``0 δmm0 , (10.234)
2` + 1
show that:
Z ∞ Z η0 2
(` + 2)! dk 2 j` (kr)
CTTT,` = ∆ (k) dη S T (η, k) (10.235)
4π(` − 2)! 0 k h 0 (kr)2
Note that a factor 2 arises because of the two polarisation states.

The above result was originally obtained in [Abbott and Wise, 1984] (though
not exactly in the same way and final form).
The main difficulty we faced in computing the aTT,`m was the spatial rotation
which brought k̂ in a generic direction. This can be avoided if we calculate straight-
away CTTT,` because it is rotationally invariant. Note that no correlation exists
between scalar and tensor modes. In fact if we compute:
haTT,`m aS∗
T,`0 m0 i , (10.236)
we would get zero, mathematically because of the integral:

Z
0
d2 k̂ 2 Y`m (k̂)Y`m
0 (k̂) = 0 , (10.237)
between a spin-2 spherical harmonic and a spin-0 one. Physically, because we know
that at the linear order scalar and tensor perturbations do not couple.
We can again approximate this angular power spectrum for large values of `
as follows. First, S T (η, k) contains the derivative of h, which is maximum when
a mode enters the horizon, for kη ≈ 1, being almost zero elsewhere. Therefore,
assuming instantaneous recombination, we can write:
(` − 1)`(` + 1)(` + 2) ∞ dk 2 j`2 (kη0 )
Z
T
CT T,` = ∆ (k) . (10.238)
4π 0 k h (kη0 )4
Defining the new variable x ≡ kη0 and introducing the primordial tensor power
spectrum we get:
(` − 1)`(` + 1)(` + 2) ∞
Z
T
CT T,` ∝ dx xnT −5 j`2 (x) . (10.239)
4π 0
296
The integral can be performed exactly:
Z ∞ √
nT −5 π Γ[1 − (nT − 4)/2]Γ[(nT − 4)/2 + `]
dx x = ,
0 2 (4 − nT )Γ[1/2 − (nT − 4)/2]Γ[` + 2 − (nT − 4)/2]
(10.240)
but in the case of nT = 0, a scale-invariant primordial tensor spectrum, we get:
`(` + 1)CTTT,` `(` + 1)

∝ . (10.241)
2π (` − 2)(` + 3)
The behaviour of the tensor contribution to the TT power spectrum is thus very
different from the one coming from scalar perturbations. In Fig. 10.12 and 10.13
we display the numerical calculations done with CLASS of the total (solid line),
scalar (dashed line) and tensor (dotted line) angular power spectra.
��
��
��
��
��
�
� ��
Figure 10.12: Numerical calculations done with CLASS of the total (solid line),
scalar (dashed line) and tensor (dotted line) angular power spectra
The tensor contribution is practically irrelevant on very small angular scale

(i.e. large `) and on large angular scales they can be as large as 10% of the total.
Typically then one can give upper limits on CTTT,` /CTST,` for small multipoles (` = 2
or ` = 10) and this ratio is proportional to AT /AS and therefore on the parameter
r, the tensor-to-scalar ratio. From Eq. (8.176) we saw that r < 0.1. In order to
determine this constraint one has also to use polarisation data since with these
we are able to disentangle the AS exp(−2τreion ) dependence coming from the scalar
contribution only to the temperature power spectrum.
10.9 Polarisation
In this section we address CMB polarisation. Recall that before recombination
polarisation is also erased because of tight-coupling. Polarisation is generated
thanks to the fact that recombination does not take place instantaneously, so the
297
��
��
��
� ��
Figure 10.13: Same as Fig. 10.12 but using a logarithmic scale.
finite-thickness effect is indeed important. Moreover, since Thomson scattering is

axially-symmetric, circular polarisation is not produced.
In Sec. 12.7 we recall the main terminology regarding polarisation and in par-
ticular the Stokes parameters.
10.9.1 Scalar perturbations contribution to polarisation

Now, let us focus on scalar perturbations only and write down from Eq. (5.115) the
line of sight solution for the combination Q + iU . Since we have chosen a reference
frame in which k̂ = ẑ, there is no U polarisation. This can be also seen from the
fact that B 0 = 0. Hence, we shall again perform a rotation in order to compute
the aP,`m .
We have called ΘP the Stokes parameter Q in the k̂ = ẑ frame. So, let us work
on its line-of-sight solution.

r Z η0
3 8π 0
ΘP (kẑ, n̂) = 2 Y (n̂) dη e−iµkr SPS (η, k) , (10.242)
2 15 2 0
where we have defined a new source term:

SPS (η, k) ≡ g(η)Π(η, k) . (10.243)
We could have written −2 Y20 (n̂) instead of 2 Y20 (n̂), since they are equal. How-
ever, we are going to deal with Q + iU first. The above equation can written
as:
r
9 X √ Z η0
0 L 0
ΘP (kẑ, n̂) = 2 Y2 (p̂) i 2L + 1YL (n̂) dη SPS (η, k)jL (kr) , (10.244)
30 L 0
298
where again r ≡ η0 − η and we have used the well-known-by-now expansion of a
plane wave into spherical harmonics plus the fact that k̂ = ẑ.
Now, as in Eq. (10.218) we can write the product of spherical harmonics as
follows:
r
0 0 5(2L + 1) X √ 0
2 Y2 (n̂)YL (n̂) = 2L + 1
4π L0
L 2 L0 L 2 L0

0
2 YL0 (n̂) , (10.245)
0 −2 +2 0 0 0
and thus obtain:

r
3 X L X√
L 2 L 0

S
(Q + iU ) (n̂) = i (2L + 1) 2L0 + 1
8π L 0 −2 2
L0
Z η0
L 2 L0 X m0 d3 k (L0 )
Z
2 YL0 (n̂) D (k̂)α(k)
3 m0 0
dη SPS (η, k)jL (kr) , (10.246)
0 0 0 (2π) 0
m0
where we have already considered the rotation which brings k̂ in a generic direction.
Now, from the expansion:
X
(Q + iU )S (n̂) = aSP,`m 2 Y`m (n̂) , (10.247)
`m
we are able to calculate the coefficients aSP,`m by taking advantage of the orthonor-
mality of the spin-2 spherical harmonics. We can therefore write:
√
r
S 3 X L L 2 `
aP,`m = i (2L + 1) 2` + 1
8π L 0 −2 2
Z η0
d3 k (`)
Z
L 2 `
Dm0 (k̂)α(k) dη SPS (η, k)jL (kr) . (10.248)
0 0 0 (2π) 3
0
Remarkably, the sum over L can be performed in the very same way we did for the
aTT,`m , since the 3j symbols are the same. Therefore, we have:
s
`
d3 k (`)
Z
S 3i (2` + 1)(` + 2)!
aP,`m = − D (k̂)α(k)
8 π(` − 2)! (2π)3 m0
Z η0
j` (kr)
dη SPS (η, k) , (10.249)
0 (kr)2
and using r r
(`) 4π 4π
Dm0 (k̂) = Y −m (k̂) = Y m∗ (k̂) , (10.250)
2` + 1 ` 2` + 1 `
we can write:
s
η0
3i` d3 k m∗
Z Z
(` + 2)! j` (kr)
aSP,`m =− Y (k̂)α(k) dη SPS (η, k) (10.251)
4 (` − 2)! (2π)3 ` 0 (kr)2
299
The expansion for (Q − iU )S (n̂) can be obtained by complex conjugation, i.e.
X X
(Q − iU )(n̂) = a∗P,`m 2 Y`m∗ (n̂) = a∗P,`m −2 Y`−m (n̂)
`m `m
X
= a∗P,`,−m m
−2 Y` (n̂) . (10.252)
`m
There is no reality condition here holding true for the aP,`m as the one holding true
for the aT,`m , because Q + iU is not real and is not a scalar. It is thus convenient
to define the following combinations:
aE,`m ≡ −(aP,`m + a∗P,`,−m )/2 , aB,`m ≡ i(aP,`m − a∗P,`,−m )/2 , (10.253)
because the first has parity (−1)` whereas the second (−1)`+1 . Thus Q ± iU can
be expanded as:
X
(Q ± iU )(n̂) = (−aE,`m ∓ iaB,`m ) 2 Y`m (n̂) . (10.254)
`m
Now, if we compute aS∗

P,`m , we obtain:
s
`
d3 k −m∗
Z
S∗ 3(−i) (` + 2)!
aP,`m = − Y (k̂)α(−k)
4 (` − 2)! (2π)3 `
Z η0
j` (kr)
dη SPS (η, k) , (10.255)
0 (kr)2
since α(k)∗ = α(−k) because of the reality condition of the power spectrum.
Changing the integration variable to k and using the parity of the spherical har-
monic:
Y`−m∗ (−k̂) = (−1)` Y`−m∗ (k̂) , (10.256)
we can finally conclude that:
aSP,`m = aS∗
P,`,−m , (10.257)
and therefore scalar perturbations only affect the E-mode, i.e.
aSE,`m = −aSP,`m , aSB,`m = 0 . (10.258)
This means that, if the B-mode was detected, it would be a clear indication of the
existence of primordial gravitational waves.
From Eq. (10.251) we can then obtain the scalar contribution to the EE spec-
trum. Assuming adiabatic Gaussian perturbations:
Z Z η0 2
S 9 (` + 2)! dk 2 j` (kr)
CEE,` = ∆ dη SPS (η, k) (10.259)
64π (` − 2)! k R 0 (kr)2
Using instead Eq. (10.176) we can compute the cross-correlation TE multipole

coefficients:
s Z Z η0
S 3 (` + 2)! dk 2 j` (kr)
CT E,` = − ∆R Θ` (k) dη SPS (η, k) (10.260)
4 (` − 2)! k 0 (kr)2
300
10.9.2 Tensor perturbations contribution to polarisation
Let us now calculate the contribution to CMB polarisation coming from tensor
perturbations. From Eq. (10.211) we have
Z η0
(T )
ΘP (η0 , kẑ, µ) = dη eikµ(η−η0 ) SPT (η, k) , (10.261)
0
with
SPT (η, k) ≡ g(η)S T (η, k) . (10.262)
and then use Eq. (5.132) in order to write part of the tensor contribution to po-
larization:
r Z η0
(T ) 8π λ
Qλ (kẑ, p̂) ≡ E (p̂) dη e−iµkr(η) SPT (η, k) , (10.263)
5 0
where r(η) ≡ η0 − η and where we stress that the result holds true for k̂ = ẑ
since this was the condition under which we derived the Boltzmann equation for
photons.
In the scalar case no U contribution to polarisation is produced, so the above
expression already furnishes the quantity Q + iU . However, the same is not true
for tensor perturbation. We have thus to add the iU contribution. As we saw in
chapter 5 this is equal to B m Q/E m and for this reason we have just one polarisation
hierarchy. Hence, summing up we get:
r Z η0
(T ) 32π λ
(Qλ + iUλ ) (kẑ, n̂) = 2 Y2 (n̂) dη eik̂·n̂kr(η) SPT (η, k) . (10.264)
5 0
Using the usual plane-wave expansion and recalling that k̂ = ẑ we get:

r
8 λ X √
(T )
(Qλ + iUλ ) (kẑ, n̂) = 2 Y2 (n̂) iL 2L + 1YL0 (n̂)
5
Z η0 L
dη SPT (η, k)jL (kr) . (10.265)
0
The product of the two spherical harmonics can be written via the Wigner 3j-
symbols as follows:
r
±2 0 5(2L + 1) X √ 0
2 Y2 (n̂)YL (n̂) = 2L + 1
4π L0
0 0

L 2 L L 2 L ±2
2 YL0 (n̂) , (10.266)
0 ±2 ∓2 0 −2 2
and we rotate in a generic k̂ direction the only spherical harmonic left, i.e.
L 0
(L0 ) 0
X
±2
2 YL0 (Rn̂) = Dm0 ,±2 [R−1 (k̂)]2 YLm0 (n̂) . (10.267)
m0 =−L0
301
Hence, we can write:
r
2X L X√
L 2 L0

(T ) 0
(Qλ + iUλ ) (k, n̂) = i (2L + 1) 2L + 1
π L 0
0 λ −λ
L
η0
L 2 L0 X (L0 )
Z
m0
Dm0 λ [R(k̂)]YL0 (n̂) dη SPT (η, k)jL (kr) . (10.268)
0 −2 2 0 0
m
Now we can perform the Fourier anti-transform. Multiply by β(k, λ) and 2 Y`m∗ (n̂)
and integrate over d2 n̂ in order to obtain the aTP,`m ’s. We obtain:
√
r
2X L L 2 `
aTP,`m,±2
= i (2L + 1) 2` + 1
π L 0 ±2 ∓2
η0
d3 k (`)
Z Z
L 2 `
D [R( k̂)]β(k, λ) dη SPT (η, k)jL (kr) . (10.269)
0 −2 2 (2π)3 m±2 0
We have used here the orthonormality relation of the spin-2 spherical harmon-
ics and distinguished the contributions of different helicity. Of course aTP,`m =
aTP,`m,+2 + aTP,`m,−2 .
We have already computed some of the 3j symbols earlier, for the tensor case
but now two more enter: those for L = ` ± 1. We shall see that these contributions
will characterise the B-mode of polarisation. They are:
s
`+1 ` 2 (` − 1)
= (−1)`+1 , (10.270)
0 −2 +2 2(2` + 1)(2` + 3)
s
` `−1 2 (` + 2)
= (−1)` . (10.271)
−2 0 +2 2(2` − 1)(2` + 1)
Extra care has to be used when manipulating these terms. The reason is that the
3j symbols gain an overall phase factor
(−1)j1 +j2 +j3 , (10.272)
where j1,2,3 are the momenta which are being combined, each time we swap two
columns or change simultaneously all the signs of the bottom row.4 Therefore,
as long as j1 + j2 + j3 is even, no matter how many times we perform the above
operations. This is the case for L = ` ± 2 or L = `. However, for L = ` ± 1 we
have that
L + ` + 2 = 2(` + 1) ± 1 , (10.273)
which is odd and thus we have to keep track of the correct sign.
4
These signs can be changed only simultaneously since the sums m1 + m2 + m3 = 0 always.
This is a selection rule.
302
Exercise 10.21. Using the formulas for the 3j symbols, show that:
√ Z η0
−i` 2` + 1 d3 k (`)
Z
T
aP,`m,±2 = √ D
3 m±2
[R(k̂)]β(k, λ) dη SPT (η, k)
8π (2π) 0

(` + 1)(` + 2) 6(` − 1)(` + 2) (` − 1)`
j`−2 (kr) − j` (kr) + j`+2 (kr)
(2` − 1)(2` + 1) (2` − 1)(2` + 3) (2` + 1)(2` + 3)

`−1 `+2
±2i j`+1 (kr) ∓ 2i j`−1 (kr) . (10.274)
2` + 1 2` + 1
Recall that r = r(η) ≡ η0 − η.
Exercise 10.22. Show that, using the recurrence relation of Eq. (10.229) and the
following ones for the derivatives:
`+1 `
j`0 (x) = j`−1 (x) − j` (x) , j`0 (x) = j` (x) − j`+1 (x) , (10.275)
x x
we can write:
√ Z η0
−i` 2` + 1 X d3 k (`)
Z
aTP,`m = √ D [R(k̂)]β(k, λ)
3 m,λ
dη SPT (η, k)
8π λ=±2
(2π) 0

2 0 2 + `(` + 1) 0 2
j − 2j` + j` − iλ j` + j` . (10.276)
kr ` (kr)2 kr
Show that the term between square brackets is equal to the corresponding one
in [Weinberg, 2008, pag. 389]. In order to make the second derivative of the
spherical Bessel function to appear one must use Bessel differential equation:
2 1 − `(` + 1)
j`00 + j` + j` = 0 . (10.277)
x x2
Now we are ready to investigate the reality property of aTP,`m and discern from
it the E-mode and B-mode contributions. Recall that Wigner D-matrix can be
related to the spin-weighted spherical harmonics as follows:
r r
(`) 4π −m 4π m∗
Dm,±2 (k̂) = ±2 Y` (k̂) = ∓2 Y` (k̂) . (10.278)
2` + 1 2` + 1
So taking the complex conjugate we find:
Z η0
(−i)` X d3 k
Z
T∗ m
aP,`m = − √ ∓2 Y` (k̂)β(−k, λ) dη SPT (η, k)
2 λ=±2 (2π)3 0

2 0 2 + `(` + 1) 0 2
j − 2j` + j` + iλ j` + j` . (10.279)
kr ` (kr)2 kr
Note that β(k, λ)∗ = β(−k, λ) because of the reality condition and beware that the
sign of the imaginary unit inside the square brackets has changed. Now, changing
integration variable
k → −k , (10.280)
303
and taking advantage of the parity property and the complex conjugation property:
m
∓2 Y` (−k̂) = (−1)` ±2 Y`m (k̂) = (−1)` ∓2 Y`−m∗ (k̂) , (10.281)
we get:
Z η0
i` X d3 k
Z
∗ −m∗
aTP,`m = −√ 3 ∓2 Y` (k̂)β(k, λ) dη SPT (η, k)
2 λ=±2 (2π) 0

2 0 2 + `(` + 1) 0 2
j − 2j` + j` + iλ j` + j` . (10.282)
kr ` (kr)2 kr
This time we have not the same situation as in Eq. (10.257) because of the iλ
contribution. Hence, we can compute the E-mode:
Z η0
i` X d3 k
Z
T m∗
aE,`m = √ Y (k̂)β(k, λ)
3 ∓2 `
dη SPT (η, k)
2 λ=±2 (2π) 0

2 0 2 + `(` + 1)
j − 2j` + j` . (10.283)
kr ` (kr)2
and the B-mode is also present:
Z η0
i` X d3 k
Z
T m∗ T 0 2
aB,`m = − √ λ ∓2 Y` (k̂)β(k, λ) dη SP (η, k) j` + j` (10.284)
2 λ=±2 (2π)3 0 kr
Now we are in position of giving the formulas for the angular power spectra. As-
suming Gaussian perturbations we have:
Z Z η0 2
T dk 2 T 2 0 2 + `(` + 1)
CEE,` = ∆ (k) dη SP (η, k) j − 2j` + j` , (10.285)
4πk h 0 kr ` (kr)2
and
Z Z η0 2
T dk 2 0 4
CBB,` = ∆ dη SPT (η, k) 2j` + j` . (10.286)
4πk h 0 kr
The cross-correlation CTTE,` , using Eq. (10.232) gives:
s Z Z η0
T (` + 2)! dk 2 j`
CT E,` = − ∆h (k) dη S T (η, k)
(` − 2)! 8πk 0 (kr)2
Z η0
2 0 2 + `(` + 1)
dη 0 SPT (η 0 , k) j` − 2j` + j` . (10.287)
0 kr (kr)2
If we try to compute the cross correlations CTTB,` and CEB,` T
we obtain a vanishing
result, as expected, because of the term λ in the sum of Eq. (10.284). In fact we
T
get, considering for example CEB,` :
X λ Z dk Z η0
T 2 T 2 0 2 + `(` + 1)
CEB,` = − ∆h dη SP (η, k) j` − 2j` + 2
j`
λ=±2
2 4πk 0 kr (kr)
Z η0
T 0 4
dη̄ SP (η̄, k) 2j` + j` . (10.288)
0 kr
304
��
��
��
��
��
�� -��
�� -��
� -��
� ��
��
��
��
��
��
��
��
��
�
� ��
Figure 10.14: The four angular power spectra characterising the CMB, computed
with CLASS for the standard model. Top left: TT. Top right: TE. Bottom left:
EE. Bottom right: BB. As for the TT spectrum, D ≡ `(` + 1)C` /(2π).
Now, the sum over λ is equivalent to a difference and since nothing else depends
on λ the result is zero. The same happens with the correlation CTTB,` .
In Fig. 10.14 we display the 4 angular CMB power spectra which constitute a
wealth of cosmological information. The only possible cross-correlation is the one
between temperature and the E-mode of polarisation. The largest signal is the TT
one and then, in order of decreasing power, the TE, EE and BB one. The latter is
5 orders of magnitude smaller than the TT one and it has not yet been detected.
The bump it displays for small `’s is due to reionisation.
From Fig. 10.14 we can appreciate how small the polarisation spectra are with
respect to the TT one. This is due to the fact that a quadrupole moment in
the distribution of photons is needed in order to have production of polarisation.
Before recombination, Thomson scattering rate is so high that photons are in nearly
perfect thermal equilibrium and any moment from the quadrupole up is washed
out. After recombination, photons free stream and thus have no more chance of
being polarised by Thomson scattering.
305
Chapter 11
Miscellanea
E desse modo ele ia levando a vida, metade na repartição, sem ser

compreendido, a outra metade em casa, também sem ser compreen-
dido
(And in this way he took life, half in the division, without being un-
derstood, the other half at home, also without being understood)
—Lima Barreto, Triste fim de Policarpo Quaresma
In this Chapter we collect some extra material which extends the topics treated
so far.
11.1 Bayesian analysis using type Ia supernovae

data
We discuss here how we can use type Ia supernovae in order to infer cosmological
information. We do not address important issues such as the calibration of the light
curves, how the distance moduli are calculated and the treatment of the systematic
errors. Moreover, some knowledge of statistics and Bayesian analysis is required.
The latter can be helped by e.g. [Trotta, 2017].
First of all, the measured quantities of interest are fluxes, which for historical
reasons are given in a logarithmic scale called apparent magnitude m, defined
as follows:
5 F
m ≡ − log10 , (11.1)
2 F0
where F is the observed flux and F0 is some reference flux for which m = 0. One
should take into account that observations are made via filters which select a certain
portion of the emitted spectrum. Therefore, the measured apparent magnitude m
depends on the filter which has been used. Moreover, one should also correct m
by the effect of interstellar absorption in such a way that its value really depends
only on how far the source is and not on the matter existing along the line of sight.
We do not take into account these important issues here and simply suppose
to be given with the bolometric magnitude, i.e. the magnitude corresponding
to the emission in the whole spectrum.
306
zb µb zb µb zb µb
0.010 32.9538 0.051 36.6511 0.257 40.5649
0.012 33.8790 0.060 37.1580 0.302 40.9052
0.014 33.8421 0.070 37.4301 0.355 41.4214
0.016 34.1185 0.082 37.9566 0.418 41.7909
0.019 34.5934 0.097 38.2532 0.491 42.2314
0.023 34.9390 0.114 38.6128 0.578 42.6170
0.026 35.2520 0.134 39.0678 0.679 43.0527
0.031 35.7485 0.158 39.3414 0.799 43.5041
0.037 36.0697 0.186 39.7921 0.940 43.9725
0.043 36.4345 0.218 40.1565 1.105 44.5140
1.300 44.8218
Table 11.1: Binned type Ia supernovae data. From [Betoule et al., 2014].
The absolute magnitude is the hypothetical apparent magnitude of an object

as if it were at a distance of 10 pc. Since F ∝ 1/d2L , using Eq. (11.1) the absolute
magnitude can be written as:
" 2 #
5 F dL dL
M ≡ − log10 = m − 5 log10 . (11.2)
2 F0 10 pc 10 pc
The distance modulus is defined as:

dL dL
µ ≡ m − M = 5 log10 = 5 log10 + 25 , (11.3)
10 pc 1 Mpc
where in the last equality we have introduced the Megaparsec (Mpc) as a more
appropriate distance scale for cosmology.
Almost a thousand type Ia supernovae are known today and many more are
expected to be detected by future experiments such as the LSST. For the sake of
simplicity, we use now the 31 binned data of [Betoule et al., 2014], which we report
in Tab. 11.1, and the relative binned 31×31 covariant matrix Cb , which we do not
report here.
Combining Eq. (11.3) with Eq. (2.155) allows us to find the theoretical predic-
tion on µ(z), which we can adjust to the observed binned values of Tab. 11.1 and
thus find the best-fit values for H0 and q0 .
Note that there are binned redshifts larger than 1 in Tab. 11.1 and for these
certainly the expansion of Eq. (2.155) is inaccurate. However, let us not worry
about this now, but focus only on how the analysis is performed.
The χ2 is calculated as follows:
χ2 (h, q0 ) = rT · C−1
b ·r , (11.4)
where r is the vector of the differences between the observed µb and the predicted
one, i.e.
3000 1 2
ri ≡ µbi − 5 log10 zi + (1 − q0 )zi − 25 , (11.5)
h 2
307
for i = 1, · · · , 31. The µbi and zi are those of Tab. 11.1. Note that we have
expressed H0 as 100 h km s−1 Mpc−1 and c = 3 × 105 km s−1 .
Through the χ2 we define the likelihood:
2 (h,q )/2
L(h, q0 ) = N e−χ 0
, (11.6)
where N is a normalisation constant. The likelihood represents the probability of

having a dataset given a cosmological model. We are interested in the contrary, i.e.
in the probability of having a certain cosmological model given a dataset. This is
called posterior probability. If there is no a priori reason for which some values
of the parameters are preferred with respect to others, then the likelihood and the
posterior probability are equal.
Let us see what happens in our case. What we do is the following: we set
up a 100×100 grid in the parameter space (h, q0 ) with 0.65 < h < 0.75 and
−0.7 < q0 < 0.2. We choose these values because we already know the results, but
in general one has to explore the parameter space. Also the 100×100 grid is purely
arbitrary as one may choose a finer one in order to improve precision.
For each point of the grid we compute the χ2 of Eq. (11.4) and thus construct
the likelihood as a function in the parameter space.1 Its maximum corresponds to
the minimum χ2 which in turn corresponds to the best fit values for the parameters.
In our case, we find:
(bf)
χ2min = 37.94 , h(bf) = 0.69 , q0 = −0.18 , (11.7)
where the superscript bf means “best fit”, of course. Note the negative best fit
value of q0 , which corresponds to an accelerated expansion.
We could in principle plot the likelihood as function of the parameters as a 3D
plot, but it is more convenient and visually clearer to plot contours, i.e. horizontal
slices of the likelihood, where horizontal means parallel to the (h, q0 ) plane. This
is what we have in Fig. 11.1.
In Fig. 11.1 there are three contour plots, which correspond to 68%, 95% and
99% confidence level. What are these? Starting from the maximum value of
the likelihood (represented by the big dot in Fig. 11.1) we slice its graph with
a horizontal plane such that the volume enclosed is 68%, 95% and 99%. If the
contours are very close one to each other then it means that the likelihood is very
peaked and thus the measure has been very precise.
Therefore, from Fig. 11.1 we can say that q0 is negative with almost 99% of
confidence and this is remarkable.
What happens if there are more than two parameters? In this case the likelihood
cannot be plotted, nor its 2-dimensional contours. Then, one performs marginal-
isation, i.e. one integrates the likelihood with respect to all the parameters but
two. In our case we did not need to do this because we had two parameters since
the beginning, but we can marginalise over one and get the probability distribution
function (PDF) for the other. This is done in Fig. 11.2.
Now we can do something similar to the calculation of the contour plots based
on the confidence levels. For each one of the PDF in Fig. 11.2, starting from the
1
Mathematica codes which perform this task are available on my personal webpage http:
//ofp.cosmo-ufes.org.
308
��
-��
-��
-��
-��
��
Figure 11.1: Contour plots at 68%, 95% and 99% confidence level (from the inner
to the outer contour). The big dot represent the best fit values of Eq. (11.7).
��
��
��
��
��
��
��
��
��
��
��
�� -�� -�� -�� -��
Figure 11.2: Normalised probability distribution functions for h and q0 .
maximum and going down, we calculate the values of the parameters for which the
area encompassed is 68%, 95% and 99%. these values represent the uncertainty
309
over the best fit values of our parameters. In our case we have:
h = 0.69+0.015
−0.015 , q0 = −0.18+0.12
−0.13 , (11.8)
at 95% confidence level.

Both the PDF’s for h and q0 are beautifully symmetric. This happens because
of the special way h and q0 enter in the χ2 , i.e. the latter is quadratic in h and
q0 and therefore the PDF for each one of these parameters is a Gaussian function.
This can also be seen in the contour plots of Fig. 11.1, which are ellipses. On the
other hand, if a parameter enters in a more complicated way into the χ2 , then the
contour plots might be very asymmetric (typically they are “banana” shaped).
Exercise 11.1. Reproduce the analysis of this section developing a suitable nu-
merical code.
11.2 Doing statistics in the sky

In Chapter 7 we have tackled the stochastic character of cosmological perturbations
from a theoretical perspective. Here we offer a simplistic approach to how statistical
methods are applied observationally. More detailed and comprehensive treatments
can be found e.g. in [Peebles, 1980] and [Bonometto, 2008].
As we learn in the very first year of our Physics course, determining the value
of some physical quantity is no trivial task. The true value that we hope to find
is only an abstraction, being our measurements imperfect i.e. characterised by
an error that we try to harness with some mathematical tools. Statistics is one
of them and tells us for example that the more we repeat our measurements the
closer to a certain value the average goes. This certain value might be the true
value, or not if systematic errors are present. In other words, we might have a
very precise but inaccurate experiment. The approach of repeating experiments
and measurements in order to extract informations about an underlying pattern
is called frequentist. The question is: how do we apply that machinery to the
universe, being this only one?
We want to learn something about gravity by observing how structures (galax-
ies) are distributed in the universe. We need statistics in order for our results to be
meaningful and thankfully there are many galaxies. Imagine a big volume V which
contains N galaxies. The galaxy number density is n = N/V . This, for example,
represents the volume of a certain survey. Getting information on the volume V in
itself is not meaningful, because we have just a single realisation of it. Therefore,
let us consider small spheres of radius R and volume
4π 3
VR = R , (11.9)
3
with, of course, VR V . Now, inside these spheres we would expect a number
N̄R ≡ nVR of galaxies if these were randomly distributed, i.e. distributed according
310
to a Poisson distribution. But actually this is not the case, because of gravity. The
latter is attractive and therefore it is more probable to find a galaxy closer to a big
cluster rather than to a smaller one. Therefore, studying the large scale structure
of the universe amounts to study the properties of gravity on large scales.
Now, let us count the galaxies in each one of the sphere that we have con-
structed. We have different numbers depending on the sphere considered, say
NR (x) because the sphere has centre in x. The number density thus becomes
NR (x)
nR (x) = . (11.10)
VR (x)
It is not really a function of a continuous variable, because in practice x gets only a

countable amount of values, as many as the centres of the spheres chosen. Formally,
we have: ν
1X
hnR (x)i = lim nR (xi ) = n . (11.11)
ν→∞ ν
i=1
For x being a continuous variable, we would have an integral:

Z
1
hnR (x)i = d3 x nR (x) = n . (11.12)
V V
So, we have divided the initial big volume V into small spherical regions VR in
order to gain statistics, i.e. in order to define averages. In particular, the mass
variance
h(NR (x) − N̄R )2 i
σR2 = (11.13)
N̄R2
is a very important indicator, as we shall see in a moment.
Exercise 11.2. Assume that all galaxies have the same mass mg . Show that from
Eq. (11.13) we get
* 2 +
2
h(ρR (x) − ρ) i δρR (x)
σR2 = 2
= = hδR (x)2 i , (11.14)
ρ ρ
where ρR (x) ≡ nR (x)mg is the mass-density inside a sphere of radius R, centered

in x and ρ = nmg , i.e. it is the background density, averaged over all V .
The important point shown by the above exercise is that the mass variance
is related to the averaged smoothed squared density contrast. From the theory
developed in these lecture notes we know how to get δ(x). The point now is to
learn how to smooth it and for this purpose we need to use filters.
311
11.2.1 Top Hat and Gaussian filters
The density contrast field smoothed over a sphere of radius R can be formally
written as: Z
δR (x) = d3 u δ(u) . (11.15)
VR (x)
Note that the x dependence of δR (x) comes from where we have centred the sphere.
We can transform the above integral into an equivalent one over the whole space,
which we need in order to use the Fourier transform:
Z Z
δR (x) = d u δ(u) = d3 u WR (|x − u|)δ(u) ,
3
(11.16)
VR (x)
where (
1/VR for y < R ,
WR (y) = (11.17)
0 for y > R ,
is the Top Hat filter, or window function (hence the symbol W for denoting it).
It is a simple trick to exclude all the information content beyond a certain scale R.
There is no reason for this to depend from the direction, therefore the argument
of WR is the modulus |x − u|. A filter is a function that must not add or subtract
any information, therefore it must be normalised to unity:
Z
d3 y WR (y) = 1 . (11.18)
In general, the functional form of a filter is:

1
WR (y) = w(y/R) , (11.19)
VR
where w(y/R) is a generic function which goes to zero rapidly as y/R grows. In
the case of the Top Hat filter, for example, w is a Heaviside function. One can also
have a Gaussian filter:
2 /(2R2 )
WR (y) = (2πR2 )−3/2 e−y . (11.20)
Equation (11.16) is a convolution between the density contrast and the filter, hence
its Fourier transform is the product between the two Fourier transforms:
δR (k) = δ(k)W̃ (kR) , (11.21)
where we have already guessed the dependence kR for the Fourier transformed
filter since WR is a function of y/R.
Exercise 11.3. Calculate the Fourier transform of the Top Hat filter. Show that:
3
W̃ (kR) = [sin(kR) − (kR) cos(kR)] . (11.22)
(kR)3
312
Of course, the Fourier transform of the Gaussian filter is still a Gaussian. The
Top Hat filter is very effective in the configuration space since it cuts out all the
scales above a given one, i.e. WR = 0 if y/R > 1 or kR < 2π. However, its
Fourier transform scales as 1/(kR)3 , so it is not cut out drastically. This gives rise
to spurious effects and therefore sometimes it might be safer to use the Gaussian
filter, for which one has the same cutoff also for the Fourier transform.
So, in Eq. (11.21) the left hand side is the smoothed density contrast, which
on should compare with observation. On the right hand side we have the full
density contrast and the window function. Actually, there is another observational
limitation. We have integrated over the whole space, together with a window
function, in order to smooth out but often a survey does not cover the full sky.
Certainly not in depth (we cannot observe up to infinite redshift) and also not
the full celestial sphere. There is a mask say f (x) by which the effective density
contrast that we are going to use is
δeff (x) = f (x)δ(x) . (11.23)
Hence, its Fourier transform is the convolution of the two Fourier transforms:
Z
δeff (k) = (f ∗ δ)(k) = d3 k0 f˜(k0 )δ(k − k0 ) .
˜ (11.24)
This convolution has important effects on the final predictions of a model. In

fact, suppose that we have a theory providing a density contrast which oscillates
and then we expect of course an oscillating power spectrum. This would be ruled
out if compared directly with the data, but since the result of the convolution
smooths out oscillations, it might happen that the model still remains viable. See
e.g. [Moffat and Toth, 2011] for a discussion on this point.
11.2.2 Sampling and shot noise

We can formally define the following quantity:
n(x) ≡ lim nR (x) , (11.25)

R→0
as the number density field. On the other hand, we do not observe such field, but
only a finite amount of galaxies. Hence, observationally n(x) is realised by
N
X
n(x) = δ (3) (x − xi ) , (11.26)
i=1
i.e. abusing (since it is a distribution) of the Dirac delta of the positions of the
galaxies xi . Avoiding any abuse we could write:
N
X 1 2 /a2
n(x) = e−|x−xi | , (11.27)
i=1
a3 π 3/2
for a → 0, i.e. using a representation of the Dirac delta.
313
Technically, when integrating Eq. (11.26) over a certain volume V we should
obtain the number of galaxies there contained. This is formally achieved by using
the Top Hat filter:
Z N Z
X N
X
(3)
N= dV n(x) = dV w(x)δ (x − xi ) = w(xi ) , (11.28)
V i=1 i=1
where w(x) = V W (x) and W (x) is equal to 1/V when x is inside the volume V
and zero otherwise (this ensures the normalisation to unity). In the last sum above
w(xi ) = 1 only when xi is inside the volume V and hence the equation holds true.
Now let us keep our sampling volume V , imagining that is the volume probed
by a certain survey. As we commented earlier, the effective density contrast field
is then given by Eq. (11.23). The difference now is that we are constructing the
observed density contrast field whereas in that equation it was our theoretical
prediction. The density field is:
N
X
ρ(x) = mi δ (3) (x − xi ) , (11.29)
i=1
where mi is the mass of the i-th galaxy. The effective, or sampled, density contrast
is
PN !
(3)

ρ(x) i=1 m i δ (x − x i )
δs (x) = − 1 w(x) = PN − 1 w(x) . (11.30)
ρ i=1 mi /V
where ρ is the background density, which is equal to N

P
i=1 mi /V in the sampled
volume. Assuming the same mass for each galaxy, one obtains:
N
!
V X (3)
δs (x) = δ (x − xi ) − 1 w(x) . (11.31)
N i=1
Exercise 11.4. Show that the Fourier transform of the density contrast is then:
N
1 X
δs (k) = w(xi )e−ik·xi − W̃ (k) , (11.32)
N i=1
where W̃ (k) is the Fourier transform of the Top Hat filter, computed in Eq. (11.22).
When we compute the power spectrum from the above realisation, and we
suppose a Gaussian distribution, we get for the mode k:
N N
1 X W̃ (k) X
hδs (k)δs∗ (k)i = 2 hw(xi )w(xj )ie −ik·(xi −xj )
− hw(xi )ie−ik·xi
N i,j=1 N i=1
N
W̃ (k) X
− hw(xi )ieik·xi + W̃ (k)2 ,(11.33)
N i=1
314
where the average is an ensemble average. Since by definition:
hδs (k)i = 0 , (11.34)
we have that:
N
1 X
hw(xi )ie−ik·xi = W̃ (k) . (11.35)
N i=1
Therefore:
N
1 X
hδs (k)δs∗ (k)i = 2 hw(xi )w(xj )ie−ik·(xi −xj ) − W̃ (k)2 . (11.36)
N i,j=1
When i = j there is a contribution above of the form:

N
1 X 2 1
hw(xi ) i = . (11.37)
N 2 i=1 N
This is an error (a variance) on the Fourier mode k which does not depend on
the wavenumber and is called shot noise. It comes from the fact that we are
mapping a continuous density field with a discrete distribution, or sampling. It is
the Poissonian part of the distribution of galaxies, which must be subtracted in
order to obtain the true spectrum which is given by the correlations hw(xi )w(xj )i.
Multiplying by V for dimensional reasons, we have thus the true power spec-
trum:
N
V X
P (k) = 2 hw(xi )w(xj )ie−ik·(xi −xj ) − V W̃ (k)2 , (11.38)
N i6=j
and the noise spectrum:

V
Pn = . (11.39)
N
11.2.3 Correlation function

How do we measure P (k) and put the data on a plot? All we observe are galaxies in
the sky. Consider again the very large volume V upon which we have worked until
now, containing N galaxies. So, the mean number density is n = N/V . Assuming
the same mass mg for every galaxy, ρ = N mg /V is the mean density. Consider two
small volumes δV1 and δV2 , centered in x1 and x2 respectively and much smaller
than V . If the distribution of galaxies was random, i.e. Poissonian, one would
expect δN1 = nδV1 . Deviations from this randomness are due to gravity and are
encoded in the correlation function ξ(x1 , x2 ):
hδN1 (x1 )δN2 (x2 )i = n2 δV1 δV2 [1 + ξ(x1 , x2 )] . (11.40)
We have already met the correlation function in Chapter 7, in Eq. (7.11). There it
was defined through the ensemble average, whereas here the average is made over
all the couples of volumes and clearly these must be chosen large enough in order
315
to contain many galaxies but sufficiently small in order to identify many of them
in V . Usually one assume that
ξ(x1 , x2 ) = ξ(r) , (11.41)
where r = |x1 − x2 |, i.e. the correlation function depends only on the distance
of the volumes, not on their positions or along which direction they are aligned.
This is again statistical homogeneity and isotropy. Dividing the above equation by
δV1 δV2 , one gets
hδn1 (x)δn2 (x + r)i = n2 [1 + ξ(r)] , (11.42)
where now the average can be interpreted as the integration over x.
Exercise 11.5. Show that the above equation can be written as:
hδ(x)δ(x + r)i = ξ(r) , (11.43)
and this gives a direct relation between the density contrast and the correlation
function. Compare with Eq. (7.40).
The correlation function measures the galaxy clustering. From observation we

have the following empirical formula:
r γ
0
ξ(r) = , (11.44)
r
where r0 = 5.5 h−1 Mpc and γ ≈ 1.77 [Peebles, 1980]. The mass variance can be
related to the correlation function as follows:
σR2 = G(γ)ξ(R) , (11.45)
where G(γ) ≈ 2. We can argue that the non-linear regime of cosmological becomes
dominant when σR2 = 1 and from the above formulae one can find that this happens
at R = 8 h−1 Mpc. Hence, this is why σ8 is so often used in cosmology.
The mass variance can be written as:
Z 3 0
d3 k
Z
dk
2 2
σR = hδR (x)i = 3 3
hδ(k)W̃ (kR)δ(k0 )W̃ (k 0 R)i , (11.46)
(2π) (2π)
where we have used the Fourier transform of the smoothed density contrast, which
is the product of the density contrast times the window function (or filter). Using
the ensemble average we get:
Z ∞
d3 k
Z
2 2 dk 2
σR = 3
Pδ (k)W̃ (kR) = ∆δ (k)W̃ (kR)2 (11.47)
(2π) 0 k
The above expression for the mass variance and that for the correlation function
in Eq. (7.46) are very similar being the difference in the function which weighs the
dimensionless power spectrum ∆2 . For the correlation function it is sin(kr)/(kr)
whereas for the mass-variance is the squared Fourier transform of the filter. If we
take the Top Hat filter then from Eq. (11.22) we see that W̃ (kR)2 decays as (kR)6 .
316
11.2.4 Bias
The bias is the deviation of the clustering behaviour of ordinary matter from the
one of CDM. In general, we might suppose that structures of scale R are formed
in those sites where
δR (x) > νσR , (11.48)
where ν is some parameter. In this way we are stating that galaxies or clusters
have not exactly the same fluctuation pattern as CDM. Define the biased density
field:
ρR,ν = θ[δR (x) − νσR ] , (11.49)
where θ is the step function. For a Gaussian process, it can be shown that
[Bonometto, 2008]
1 1 −ν 2 /2
hρR,ν i ≈ √ e , (11.50)
2π ν
which gives
3 2 /2
hnR,ν i ≈ e−ν . (11.51)
(2π)3/2 R3 ν
The correlation function can be written then as
2 ξ R (r)/σ 2
ξ (ν,R) (r) = eν R −1, (11.52)
which, for a small exponential, can be cast as
ν2 R
ξ (ν,R) (r) = ξ (r) ≡ b2 ξ R (r) , (11.53)
σR2
and in general the relation between galactic and CDM density contrast is written
as:
δg = bδ , (11.54)
i.e. the same Eq. (9.1) that we presented in Chapter 9. As we mentioned there,
typically b is treated as a parameter, though it may depend on the scale and on
time.
317
Chapter 12
Appendices
Like all people who try to exhaust a subject, he exhausted his listeners
—Oscar Wilde, The picture of Dorian Gray
Here are collected miscellaneous topics which are helpful in order to understand
the former Chapters and to make these lecture notes as self-contained as possible.
12.1 Thermal distributions

In this section we derive the functional form of the Maxwell-Boltzmann, Fermi-
Dirac and Bose-Einstein thermal distributions in the grand-canonical ensemble.
See e.g. [Huang, 1987] for a textbook reference.
12.1.1 Derivation of the Maxwell-Boltzmann distribution

Suppose that we have a system of N classical particles distributed in K states of
different energies. Let us assume N > K. For each state i there are ni particles
with energy ei and the total energy of the system is fixed and equal to E. Therefore,
we have two constraints:
K
X K
X
N= ni , E= ni ei . (12.1)
i=1 i=1
The number of microstates Ω which correspond to the macroscopical configuration

of energy E is the following:
N!
Ω ({ni }) = . (12.2)
n1 !n2 ! · · · nK !
The numerator is N ! because classical particles are distinguishable, so each of their
permutations counts as a microstate. However, permutations done within the same
state i do not count, hence we eliminate these possibilities dividing by n1 !n2 ! · · · nK !.
Now take the logarithm of Ω and use Stirling’s approximation:
K
X K
X
log Ω = log(N !) − log(ni !) ≈ N log N − N − (ni log ni − ni ) . (12.3)
i=1 i=1
318
We have to find the maximum for this expression, but taking into account the
constraints (12.1). We then introduce Lagrange multipliers α and β and calculate
the differential:
" K K K
#
X X X
d N log N − N − (ni log ni − ni ) − α ni − β ni ei = 0 . (12.4)
i=1 i=1 i=1
The variables are the ni ’s, thus:

K
X
(− log ni − α − βei ) dni = 0 . (12.5)
i=1
From this equation we obtain:

ei − µ
ni = exp(−α − βei ) ≡ exp − , (12.6)
kB T
where in the last equality we have associated the Lagrange multiplier α as the
chemical potential and 1/β as the thermal energy of the bath.
12.1.2 Derivation of the Fermi-Dirac distribution

When calculating the Fermi-Dirac distribution two main differences with respect
to the Maxwell-Boltzmann case appear: the first is that we have now quantum
particles, which are indistinguishable, and the second is that fermions obey Pauli’s
exclusion principle, so there could be at most one fermion per quantum state.
Let us assume again N quantum particles and K energy states. The only
difference now is that for each energy state i there are gi sub-states, i.e. the energy
states are degenerate. Let us focus on the energy state i. Here we must place ni
fermions. Because of Pauli’s exclusion principle, gi ≥ ni , otherwise we would have
more than one fermion per state.
Exercise 12.1. In how many ways can we fit ni indistinguishable fermions in gi

energy “slots”? Prove that the answer is:
gi !
Ωi = . (12.7)
ni !(gi − ni )!
Since combinatorics might be sometimes irksome, and after all these are lecture
notes in cosmology, we offer a proof in the footnote. Try however not jumping to
it right away and to work out the solution yourself.1
1
We can choose gi slots for the first fermion, gi − 1 for the second and so on until finally we
can choose gi − ni + 1 slots for the ni -th fermion. This gives gi !/(gi − ni )! and we only used
Pauli’s exclusion principle. Until here, then, the order of the chosen particles matter, e.g. having
fermion 1 in the first slot is different from having fermion 2 in the first slot. But fermions are
indistinguishable, therefore we must divide by ni ! and hence the result (12.7).
319
The total number of micro-states is then:
K K
Y Y gi !
Ω ({ni }) = Ωi = . (12.8)
i=1 i=1
ni !(gi − ni )!
Now we proceed as before, taking the logarithm of Ω:

K
X
log Ω = [log(gi !) − log(ni !) − log((gi − ni )!)] , (12.9)
i=1
using Stirling’s approximation and calculating the constrained maximum:

K
X
[− log ni + log(gi − ni ) − α − βei ] dni = 0 , (12.10)
i=1
one obtains:
gi − ni
log = α + βei , (12.11)
ni
and finally:
1 1
ni = gi ≡ gi , (12.12)
1 + exp(α + βei ) 1 + exp ei −µ
kB T
with the same physical meanings for α and β as those stated for the Maxwell-
Boltzmann distribution.
12.1.3 Derivation of the Bose-Einstein distribution

The setup for deriving the Bose-Einstein distribution is the same as the one used
in the previous subsection for the Fermi-Dirac distribution except for the fact that
now Pauli’s exclusion principle does not apply and so the constraint ni ≤ gi does
not hold true anymore. This changes the way in which we calculate Ωi .

ni + gi − 1 (ni + gi − 1)!
Ωi = = . (12.13)
gi − 1 ni !(gi − 1)!
This is the same calculation of how many ni -th partial derivatives of a function of
gi variables there are. This is of course related to the fact that the wave-function
of an ensemble of bosons is symmetric, just as partial derivatives are. Again, a
proof is given in the footnote.2
2
Imagine ni particles and gi slots where to fit them. These slots are separated by gi − 1
walls. So, compute all the permutations among these objects, which are (ni + gi − 1)!, but do
not consider the permutations among the walls (gi − 1)! and the particles, ni !, because they are
indistinguishable. So, we find Eq. (12.13).
320
Proceeding as we did in the previous two subsections, we find:
K
X
[log(ni + gi − 1) − log ni − α − βei ] dni = 0 , (12.14)
i=1
and finally:
1 1
ni = gi ≡ gi . (12.15)
exp(α + βei ) − 1 exp ei −µ − 1
kB T
12.2 Derivation of the Poisson distribution

Poisson distribution describes stochastic independent events happening randomly
in time but with a certain average rate say λ. It is important for the description
particle scattering or decaying processes and we have used it in Chapter 3 in order
to take into account the neutron decay during BBN and also in Chapter 10 when
we have discussed the visibility function.
Let P (n; λ, t) be the probability that n events occur in a time interval t, given
the rate λ. Consider a sufficiently small time interval δt, for which:
P (1; λ, δt) = λδt , P (0; λ, δt) = 1 − λδt . (12.16)
Indeed, we can regard the first formula as the definition of λ. Therefore, calculating
P (0; λ, t + δt) one gets:
P (0; λ, t + δt) = P (0; λ, t)(1 − λδt) , (12.17)
where we have used the independence of the randomly occurring events.
Exercise 12.3. From Eq. (12.17) show that:
P (0; λ, t) = e−λt . (12.18)
Now, exploiting the independence of the events, we can find a recurrence rela-
tion for P (n; λ, t). Consider the following:
P (n; λ, t + δt) = P (n; λ, t)(1 − λδt) + P (n − 1; λ, t)λδt , (12.19)
i.e. n events in a time interval t + δt can either occur as n in the time interval
t and none in the subsequent δt or n − 1 in the time interval t and just one in
the subsequent δt. This is because we have chosen δt sufficiently small in order to
accommodate at most one event.
From the above equation we have then:
dP (n; λ, t)
+ λP (n; λ, t) = λP (n − 1; λ, t) . (12.20)
dt
321
Exercise 12.4. From Eq. (12.20) show that:
P (1; λ, t) = λte−λt . (12.21)
By induction show that:

(λt)n −λt
P (n; λ, t) = e . (12.22)
n!
This is Poisson distribution. Another way to find it is from the binomial dis-
tribution:
N n
P (N ; n, p) = p (1 − p)N −n , (12.23)
n
where N is the number of trials, n are the successful ones and p is the probability
of success for a single trial.
Exercise 12.5. Assume N → ∞ and p → 0 such that N p stays constant. Per-

forming these limits in Eq. (12.23) and using Stirling’s approximation show that:
(N p)n −N p
P (n, N p) = e . (12.24)
n!
Defining N p ≡ λt, we recover again Poisson distribution (12.22).
12.3 Helmholtz theorem

We follow here Appendix B of [Griffiths, 2017]. Let F(r) be a vector field. Let its
divergence and curl be the following:
∇ · F(r) ≡ D(r) , ∇ × F(r) ≡ C(r) . (12.25)
By construction, the divergence of the curl is vanishing:
∇ · [∇ × F(r)] = ∇ · C(r) = 0 , (12.26)
hence C(r) is a solenoidal, i.e. divergenceless, vector field.

Helmholtz theorem states that F(r) can be decomposed as:
F(r) = −∇U (r) + ∇ × W(r) , (12.27)
where:
D(r0 ) C(r0 )
Z Z
1 3 0 1
U (r) ≡ dr , W(r) = d3 r0 . (12.28)
4π V |r − r0 | 4π V |r − r0 |
Since the integration is over the whole space, in order for it to be well-defined D(r)
and C(r) must tend to zero more rapidly than 1/r2 for r → ∞. This can be seen
from the fact that:
d3 r0
0
∼ dr0 r0 , for r0 → ∞ , (12.29)
|r − r |
322
0
and then if the two functions D and C scaled as 1/r 2 we would have a logarithmic
divergence.
By taking the divergent of Eq. (12.27), remembering that the divergent of a
curl is vanishing, and using Eq. (12.25) we can easily check that:
D(r) = ∇ · F(r) = −∇2 U (r) , (12.30)
which is a Poisson equation and its solution is the first equation of Eq. (12.28).
On the other hand, by taking the curl of (12.27), remembering that the curl of
the gradient is vanishing, we can check that:
C(r) = ∇ × [∇ × W(r)] = ∇ [∇ · W(r)] − ∇2 W . (12.31)
The Laplacian term only gives the second equation of Eq. (12.28), but what about
the ∇ [∇ · W(r)] term? Does it vanishes? Let us check that this is the case:
Z
1 3 0 0 1
∇ · W(r) = d r C(r ) · ∇
4π V |r − r0 |
Z
1 3 0 0 0 1
=− d r C(r ) · ∇ =
4π V |r − r0 |
Z Z
1 0 0 1 1 1
− dS · C(r ) 0
+ d3 r0 ∇0 · C(r0 ) . (12.32)
4π ∂V |r − r | 4π V |r − r0 |
In the last line, both contributions vanish. The first is a surface integral at infinity
and in the second the divergence of C(r) is zero by construction.
In principle, the decomposition of Eq. (12.27) might not be unique because we
could add to F(r) a vector field say G(r) which has both vanishing divergence and
curl. On the other hand, it can be proved that there is no such G(r), with both
vanishing divergence and curl and which goes to zero at infinity. Therefore, if F(r)
goes to zero sufficiently fast at infinity then the decomposition in Eq. (12.27) is
indeed unique.
12.4 Conservation of R on large scales and for adi-

abatic perturbations
In order to show the constancy of R on large scales and for adiabatic perturbations,
let us start from Eq. (4.185) and combine its scalar part with the definition of R
given in Eq. (4.131):
Φ0 − HΨ
R=Φ+H . (12.33)
4πGa2 (ρ + P )
Exercise 12.6. Differentiate Eq. (4.131) with respect to the conformal time and
show that:
4πGa2 (ρ + P )R0 = 4πGa2 (ρ + P )Φ0 + H0 (Φ0 − HΨ)
P0
+H (Φ00 − HΨ0 − H0 Ψ) + H2 (Φ0 − HΨ) + 3H2 0 (Φ0 − HΨ) , (12.34)
ρ
where ρ0 = −3H(ρ + P ) has been used.
323
Exercise 12.7. Use the generalised Poisson equation (4.179), written as
3H (Φ0 − HΨ) + k 2 Φ = 4πGa2 δρ , (12.35)
and the background relation:
4πGa2 (ρ + P ) = H2 − H0 , (12.36)
in order to cast Eq. (12.34) as follows:
4πGa2 (ρ + P )R0 = H Φ00 + 2HΦ0 − HΨ0 − 2H0 Ψ − H2 Ψ

P0
+H 0 4πGa2 δρ − k 2 Φ .

(12.37)
ρ
The first term on the right hand side can be simplified using the Einstein
equation (4.189), so that we have:
k2H P0
4πGa2 (ρ+P )R0 = −4πGa2 HδP − (Φ+Ψ)+H 0 4πGa2 δρ − k 2 Φ . (12.38)

3 ρ
Recalling the gauge-invariant entropy perturbation that we introduced in Eq. (4.129):
P0
Γ ≡ δP − δρ , (12.39)
ρ0
we can finally write:
Γ P 0 k2Φ k 2 (Φ + Ψ)
R0 = −H −H 0 2 − H (12.40)
ρ+P ρ H − H0 3(H2 − H0 )
The first term on the right-hand side vanishes if the perturbations are adiabatic,
whereas the second and third terms on the right-hand side vanish on large scales,
leaving thus R constant.
The gauge-invariant variables R and ζ can be related in the following way.
First, using their definitions (4.131) and (4.132) in the Newtonian gauge we can
write:
δρ
ζ = R − Hv + . (12.41)
3(ρ + P )
Second, write down the relativistic Poisson equation (4.179) and the velocity equa-
tion (4.185) in the following form:
3HΦ0 − 3H2 Ψ + k 2 Φ = 4πGa2 δρ , (12.42)

Φ0 − HΨ = 4πGa2 (ρ + P ) v , (12.43)
and combine them in order to give the constraint:
12πGHa2 (ρ + P )v + k 2 Φ = 4πGδρ . (12.44)
324
Using this constraint, the relation in Eq. (12.41) becomes:
k2Φ
ζ =R+ (12.45)
12πGa2 (ρ + P )
Since (ρ + P )a2 is of order H2 , then ζ − R is of order k 2 /H2 , which vanishes for

large scales. This means that if R is conserved then ζ also is.
The result found here confirms our previous calculation ended with Eq. (6.94),
in Chapter 6, when we investigated the adiabatic primordial modes.
12.5 Spherical harmonics

In this section we briefly review spherical harmonics. These are the eigenfunctions
of the Laplacian operator on the sphere or, equivalently, the eigenfunctions of the
square of the angular momentum operator. They form a complete orthonormal
system and therefore any function on the sphere can be expanded in a linear
combination of spherical harmonics. That is why they are extensively used in
cosmology to analyse CMB anisotropies in temperature and polarisation: the latter
are fields (of spin zero and spin 2, respectively) on the celestial sphere. In this
section we shall follow principally [Butkov, 1968]. However, many references exist
dealing with the theory of the angular momentum and spherical harmonics, so the
reader is encouraged to find the treatment which most suits them.
Exercise 12.8. Find the expression of the Laplacian operator on the sphere. The
line element on the unit sphere is:
ds2S = gab dxa dxb = dθ2 + sin2 θdφ2 , (12.46)
with a, b = θ, φ, and naturally we have chosen spherical coordinates. Use the

formula:
2 1 ∂ ab √ ∂
∇ =√ g g b , (12.47)
g ∂xa ∂x
√
where g is the determinant of the metric and find:
1 ∂2

2 1 ∂ ∂
∇ = sin θ + . (12.48)
sin θ ∂θ ∂θ sin2 θ ∂φ2
Let us then set up the eigenvalue equation:

1 ∂ 2 Y (θ, φ)

1 ∂ ∂Y (θ, φ)
sin θ + = λY (θ, φ) . (12.49)
sin θ ∂θ ∂θ sin2 θ ∂φ2
Assume that the θ and φ dependences can be factorised, i.e. let us write:3
Y (θ, φ) = Θ(θ)Φ(φ) , (12.50)

3
The function Θ(θ) here is not the relative temperature fluctuation and Φ(φ) is not the Bardeen
potential.
325
so that:
1 d2 Φ

sin θ d dΘ
sin θ − λ sin2 θ = − . (12.51)
Θ dθ dθ Φ dφ2
Now, the left hand side depends only on θ whereas the right hand side only on φ.
Therefore, the only possible way for them to be equal is to be equal to the same
constant, which we call m2 . In this way we have now two equations:
d2 Φ

2 d dΘ
+m Φ=0, sin θ sin θ − λ sin2 θΘ = m2 Θ . (12.52)
dφ dθ dθ
The first one is very simple to solve and gives us:
Φ = e±imφ , (12.53)
with some normalisation which we will determine afterwards for the full Y (θ, φ).
On the other hand, Φ must be periodic since φ or φ + 2π denote the same angular
position. Hence:
eimφ = eim(φ+2π) , (12.54)
which implies:
ei2mπ = 1 , (12.55)
and therefore m must be an integer, positive or negative, or zero. The equation
for Θ can be treated as follows.
Exercise 12.9. Consider the new variable:
x ≡ cos θ . (12.56)
Show that the derivative with respect to θ satisfies:

d d
sin θ = −(1 − x2 ) , (12.57)
dθ dx
and hence the equation for Θ(x) can be written as follows:
m2

d 2 dΘ(x)
(1 − x ) − λΘ(x) − Θ(x) = 0 . (12.58)
dx dx 1 − x2
Now, we need Θ(x) to be regular at x = ±1, i.e. for θ = 0 or θ = π. A way to

investigate this is to change variable:
z =1∓x, (12.59)
in order to trade the neighbourhood of x = ±1 for that of z = 0.
Exercise 12.10. Obtain the new equation for Θ(z) and z = 1 − x:
d2 Θ dΘ m2
z(2 − z) 2 + 2(1 − z) − λΘ − Θ=0. (12.60)
dz dz z(2 − z)
326
Now, using Frobenius method we look for a solution of the form:
∞
X
s
Θ(z) = z an z n . (12.61)
n=0
where s ≥ 0 in order for Θ(z) to be regular for z → 0.
Exercise 12.11. Substitute the above ansatz in Eq. (12.60) and equate to zero
power by power. Show that we get from the lowest power (which is z s ) that:
m
s=± . (12.62)
2
Let us stipulate that m ≥ 0. Then we must choose the positive sign, i.e.
s = m/2, since z −m/2 → ∞ for z → 0.
Exercise 12.12. Carry on a similar analysis for x = −1, by transforming variable

to z = 1 + x. Show that again s = m/2.
Combining the results from the above exercises we can conclude that Θ(x) must
have the following form:
Θ(x) = (1 − x2 )m/2 f (x) , (12.63)
where f (x) is an analytic function, non-vanishing for x = ±1. We have also learnt
that Θ(x) does vanish for x ± 1, except when m = 0. Now, let us find an equation
for f .
Exercise 12.13. Substituting Eq. (12.63) into Eq. (12.58) show that:
d2 f df
(1 − x2 ) 2
− 2x(m + 1) − (λ + m + m2 )f = 0 . (12.64)
dx dx
We now prove that λ = −`(` + 1), with ` integer such that ` ≥ m. This result
will stem out of the requirement of regularity of f . Adopting again Frobenius
method, let us stipulate that:
∞
X
s
f (x) = x an x n . (12.65)
n=0
Exercise 12.14. Substitute the above ansatz into Eq. (12.64) and find the follow-
ing relation:
∞
X
an (n + s)(n + s − 1)xn+s−2 =
n=0
∞
X
an [m(m + 1) + λ + 2(m + 1)(n + s) + (n + s)(n + s − 1)] xn+s . (12.66)
n=0
327
The two series can be combined starting from n = 2.

∞
X
s−2 s−1
a0 s(s − 1)x + a1 s(s + 1)x + an (n + s)(n + s − 1)xn+s−2 =
n=2
∞
X
an−2 [m(m + 1) + λ + 2(m + 1)(n + s − 2) + (n + s − 2)(n + s − 3)] xn+s−2 .
n=2
(12.67)
Now the two series can be merged in a single one and equating to zero each
power we have that either s = 0 or s = 1. These conditions lead to the following
recursion relations for the series coefficients:
n(n − 1) + λ + m(m + 1) + 2n(m + 1)
an+2 = an , (s = 0) , (12.68)
(n + 2)(n + 1)
n(n + 1) + λ + m(m + 1) + 2(n + 1)(m + 1)
an+2 = an , (s = 1) . (12.69)
(n + 3)(n + 2)
The integral test of convergence fails because an+2 /an → 1 for n → ∞, hence the
series solution would diverge unless:
n(n − 1) + λ + m(m + 1) + 2n(m + 1) = 0 , (s = 0) , (12.70)

n(n + 1) + λ + m(m + 1) + 2(n + 1)(m + 1) = 0 , (s = 1) . (12.71)
These are constraints on λ, which has then to assume the following form:
λ = −(m + n)(m + n + 1) , (s = 0) , (12.72)

λ = −(m + n + 1)(m + n + 2) , (s = 1) , (12.73)
or, in general:
λ ≡ −`(` + 1) (12.74)
with ` integer and ` ≥ m, which is what we wanted to prove. This result can be
obtained also in quantum mechanics, by exploiting the commutation relations of
the components of the angular momentum operator, see e.g. [Weinberg, 2015]. We
have adopted here an approach based on calculus.
The above condition on λ implies that the series solution for f terminates
after the (` − m)-th term and thus Θ(x) is a polynomial. The polynomials thus
obtained for all the possible choices of (`, m) are known as associated Legendre
polynomials and denoted as P`m (x). Equation (12.58) can be thus written as:
m
m2

d 2 dP` (x)
(1 − x ) + `(` + 1) − 2
P`m (x) = 0 (12.75)
dx dx 1−x
328
and it is known as the general Legendre equation. Via some manipulation it
is possible to obtain it from the Legendre equation:

d 2 dP` (x)
(1 − x ) + `(` + 1)P` (x) = 0 (12.76)
dx dx
and therefore write the associated Legendre polynomials as:

dm
P`m (x) = (1 − x2 )m/2 [P` (x)] . (12.77)
dxm
Using Rodrigues’ formula:
1 d`
P` (x) = [(x2 − 1)` ] , (12.78)
2` `! dx`
we have then
(1 − x2 )m/2 d`+m
P`m (x) = [(x2 − 1)` ] , (12.79)
2` `! dx`+m
and this formula allows us to extend the definition of P`m (x) also for negative values
of m, such that ` ≥ |m| or −` ≤ m ≤ `, as we are used from the theory of the
angular momentum in quantum mechanics. In these notes we have adopted the
following convention:
(` − m)! m
P`−m = P . (12.80)
(` + m)! `
Another, perhaps more standard convention includes a (−1)m factor, but we choose
to omit it in order to have the following property for the spherical harmonics under
complex conjugation:
Y`m∗ (θ, φ) = Y`−m (θ, φ) (12.81)
We can now finally write down the explicit formula for the spherical harmonics:
s
(2` + 1)(` − m)! m
Y`m (θ, φ) = P` (cos θ)eimφ (12.82)
4π(` + m)!
where the normalisation has been chosen in order to have

Z π Z 2π
0∗
dθ sin θ dφ Y`m (θ, φ)Y`m
0 (θ, φ) = δ``0 δmm0 (12.83)
0 0
which is probably the single most important relation about spherical harmonics
because it tells us that they form an orthonormal system. This is also complete,
i.e. the spherical harmonics satisfy the following completeness relation:
∞ X
`
X 1
Y`m (θ, φ)Y`m∗ (θ0 , φ0 ) = δ(θ − θ0 )δ(φ − φ0 ) , (12.84)
`=0 m=−`
sin θ
Therefore, any square-integrable function f (θ, φ) can be expanded as:

∞ X
X `
f (θ, φ) = a`m Y`m (θ, φ) , (12.85)
`=0 m=−`
329
in an unique way (meaning that the a`m ’s are unique for that function). The δmm0
in the orthonormality can be easily understood from the dφ integration, since:
Z 2π 0 2π
i(m−m0 )φ ei(m−m )φ
dφ e = =0, (12.86)
0 i(m − m0 ) 0
for m 6= m0 , leaving only the m = m0 possibility and the 2π value of the integral.
The δ``0 part of the orthonormality relation depends of course on the properties of
the associated Legendre polynomials, but we do not give further detail here.
It is important to know the parity of the spherical harmonics, i.e. how Y`m (n̂)
changes under n̂ → −n̂. In terms of the angles:
n̂ → −n̂ ⇒ (θ, φ) → (π − θ, π + φ) . (12.87)
Since cos(π − θ) = − cos θ, spatial inversion amount to x → −x for the associated

Legendre polynomial and therefore a (−1)`+m factor coming from the derivative in
the formula of Eq. (12.79). The exponential exp(imφ) instead becomes:
eimφ → eim(π+φ) = eimπ eimφ = (−1)m eimφ . (12.88)
Therefore, we finally have:
Y`m (π − θ, π + φ) = (−1)` Y`m (θ, φ) (12.89)
A formula that we have intensively employed in these notes is the expansion of a

plane wave into spherical harmonics:
∞ X
X `
eik·r = 4π i` Y`m∗ (k̂)Y`m (r̂)j` (kr) (12.90)
`=0 m=−`
where the complex conjugation can be switched from one spherical harmonic to the
other, since k · r = r · k. Also, we have made great use of the addition theorem:
`
4π X m∗
P` (x · y) = Y (y)Y`m (x) (12.91)
2` + 1 m=−` `
We can then combine the plane wave expansion with the addition theorem and
get:
∞
X
eik·r = i` (2` + 1)P` (k̂ · r̂)j` (kr) . (12.92)
`=0
We now turn to the spin-weighted spherical harmonics.
12.5.1 Spin-weighted spherical harmonics

The usual spherical harmonics allow to expand a scalar quantity such as the tem-
perature on the sphere. Consider a rotation R such that
n̂ → n̂0 = Rn̂ . (12.93)
330
Since the relative temperature fluctuation is a scalar, we have that:
Θ0 (n̂0 ) = Θ(n̂) . (12.94)
Using the expansion in spherical harmonics we can write:

X X
a0`m Y`m (n̂0 ) = a`m Y`m (n̂) . (12.95)
`m `m
Now we take advantage of the properties of the spherical harmonics under spatial
rotation, i.e.
`
0
X (`)
m
Y` (Rn̂) = Dm0 m (R−1 )Y`m (n̂) (12.96)
m0 =−`
(`)
where the Dm0 m are the elements of the Wigner D-matrix. See [Landau and
Lifshits, 1991] for more detail. Hence, we can write:
0
X X (`)
a0`m Y`m (n̂0 ) = a`m Dm0 m (R)Y`m (n̂0 ) , (12.97)
`m `mm0
and readjusting the indices we finally have:

X (`)
a0`m = Dm0 m (R)a`m0 (12.98)
m0
since the spherical harmonics form an orthonormal basis. This is an important

relation because it tells us that the angular power spectrum is independent from
the direction, i.e. it is rotationally invariant:
0∗ 0
X (`) (`)∗ X (`) (`)∗
C`0 = ha0`m a`m i= DmM DmM 0 ha`M a∗`M i = C` DmM DmM = C` , (12.99)
MM0 M
the product of Wigner D-matrices being unity since it represents a product of

one rotation with its inverse. Hence we have found that the power spectrum is a
rotational invariant. This is a consequence of the fact that Θ is a scalar, but what
about another function which is not? In Sec. 12.7 we shall see that indeed the
Stokes parameters are not scalars, i.e. upon a rotation n̂ → n̂0 = Rn̂ we have in
general that:
Q0 (n̂0 ) 6= Q(n̂) , U 0 (n̂0 ) 6= U (n̂) . (12.100)
Therefore, if we expand them on a spherical harmonics basis we would get power
spectra depending on the orientation of our coordinate frame.
In order to avoid this, we must employ the spin-weighted spherical har-
monics. These were introduced in [Newman and Penrose, 1966] in the context
of gravitational waves, which also have polarisation. Using their notation, η is a
spin-s quantity if it transforms as:
η(n̂) → η(n̂)0 = esiψ η(n̂) , (12.101)
331
under a rotation of an angle ψ about the line of sight n̂. Then, one defines the
differential operators ð and ð̄ as:

∂ i ∂
s
(sin θ)−s η ,

ðη = −(sin θ) + (12.102)
∂θ sin θ ∂φ

−s ∂ i ∂
ð̄η = −(sin θ) + [(sin θ)s η] , (12.103)
∂θ sin θ ∂φ
in spherical coordinates and proves that ðη is a spin-(s + 1) quantity whereas ð̄η
is a spin-(s − 1) quantity. These operators are essentially covariant derivative on
the sphere, as we shall see in Sec. 12.7. The spin-s spherical harmonics are thus
defined as:
s
m (` − s)! s m
s Y` = ð Y` (0 ≤ s ≤ `) , (12.104)
(` + s)!
s
m s (` + s)! −s m
s Y` = (−1) ð̄ Y` (−` ≤ s ≤ 0) . (12.105)
(` − s)!
The spin-weighted spherical harmonics have fundamental properties similar to
those characterising the usual ones. First of all:
m
0 Y` (θ, φ) = Y`m (θ, φ) , (12.106)
i.e. the usual spherical harmonics can be regarded as spin-0. Then, the s Y`m form
a complete orthonormal system, i.e. they are orthonormal
Z π Z 2π
0∗
dθ sin θ dφ s Y`m (θ, φ)s Y`m
0 (θ, φ) = δ``0 δmm0 (12.107)
0 0
and complete:
∞ X
`
X
m∗ 0 0 1
m
s Y` (θ, φ)s Y` (θ , φ ) = δ(θ − θ0 )δ(φ − φ0 ) (12.108)
`=0 m=−`
sin θ
and therefore any spin-s quantity η can be expanded as:

∞ X
X `
η(θ, φ) = η`m s Y`m (θ, φ) (12.109)
`=0 m=−`
Under complex conjugation and spatial inversion the spin-weighted spherical har-
monics transform as:
s Y`
m∗
= (−1)s −s Y`−m , m
s Y` (−n̂) = (−1)` −s Y`m (n̂) , (12.110)
and they can be related to the Wigner D-matrix as follows:
r
(`) 4π −m
Dms (φ, θ, ψ) = s Y` (θ, φ)eisψ , (12.111)
2` + 1
where (φ, θ, ψ) are the Euler angles.
332
12.6 Method of Green’s functions
Suppose that we have a linear differential equation, which we write as follows:
Ly(x) = f (x) , (12.112)
where L is a second-order linear differential operator. Suppose also that we have
been able to solve the homogeneous equation Ly(x) = 0 and found the solution:
yhom (x) = C1 y1 (x) + C2 y2 (x) . (12.113)
How, knowing this, do we determine the solution to the complete, non-homogeneous
equation? Recall that Green’s function is defined by:
LG(x, x0 ) = δ(x − x0 ) , (12.114)
and from G(x, x0 ) the general solution to the complete equation is readily found
via convolution with the non-homogeneous term:
Z x
y(x) = dx0 G(x, x0 )f (x0 ) , (12.115)
xi
where xi is our initial value.
Exercise 12.16. Check that the above convolution Eq. (12.115) is solution of
Eq. (12.112).
So, we must determine G(x, x0 ). We can do this by noticing that it is solution

of the homogeneous equation when x 6= x0 and thus we can write:
(
0 A1 y1 (x) + A2 y2 (x) for x < x0
G(x, x ) = , (12.116)
B1 y1 (x) + B2 y2 (x) for x > x0
where the A’s and B’s are integration constants to be determined. Since L is a
second order linear differential equation, we want that:
• Since LG(x, x0 ) = δ(x − x0 ), the derivative of G(x, x0 ) must be discontinuous
in x = x0 , i.e. it must be a θ function.
• On the other hand, G(x, x0 ) is continuous in x = x0 .
So, with these two conditions, we impose the following matching:
(
A1 y1 (x0 ) + A2 y2 (x0 ) = B1 y1 (x0 ) + B2 y2 (x0 )
. (12.117)
B1 y10 (x0 ) + B2 y20 (x0 ) − A1 y10 (x0 ) − A2 y20 (x0 ) = 1
Note that whatever the gap of the discontinuity might be, we can always normalise
it to unity by renormalising the integration constants. The above system can be
cast in the following form:
(
(B1 − A1 )y1 (x0 ) + (B2 − A2 )y2 (x0 ) = 0
. (12.118)
(B1 − A1 )y10 (x0 ) + (B2 − A2 )y20 (x0 ) = 1
333
The unknowns of this system are the differences among the integration constants
and the determinant of the system is:
W (x0 ) = (y1 y20 − y10 y2 )(x0 ) , (12.119)
i.e. the Wronskian, which is certainly different from zero because y1 and y2 are
independent solutions. Therefore, we use Cramer’s rule and find:
−y2 (x0 ) y1 (x0 )
B1 − A1 = , B2 − A2 = . (12.120)
W (x0 ) W (x0 )
We have freedom of choosing two of the four constants. Therefore, choosing A1 =
A2 = 0:
(
0 for x < x0
G(x, x0 ) = −y2 (x0 )y1 (x)+y1 (x0 )y2 (x) 0
, (12.121)
0
W (x )
for x > x
and so this is our Green function which, in convolution with f (x0 ) gives us the
complete solution to our non-homogenous differential equation:
x
−y2 (x0 )y1 (x) + y1 (x0 )y2 (x)
Z
y(x) = dx0 0
f (x0 ) (12.122)
xi W (x )
12.7 Polarisation
In this section we recall the terminology regarding the polarisation of an electro-
magnetic wave. The standard reference is [Jackson, 1998], but see also [Griffiths,
2017].
12.7.1 Electromagnetic waves

Let us start from Maxwell equations in vacuum and in Minkowski space:
∂B 1 ∂E
∇·E=0, ∇×E=− , ∇·B=0, ∇×B= . (12.123)
∂t c2 ∂t
Exercise 12.17. Combine Maxwell equations in order to find the wave equations:
1 ∂ 2E 1 ∂ 2B
E = − ∇2 E = 0 , B = − ∇2 B = 0 . (12.124)
c2 ∂t2 c2 ∂t2
The general solutions of the wave equations are thus:
E(t, x) = E0 f (p̂ · x − ct) , B(t, x) = B0 g(p̂0 · x − ct) , (12.125)
where p̂ and p̂0 are in principle distinct directions of propagation, f and g two
distinct functions and E0 and B0 are the amplitudes of the wave describing the
electric field and of the one describing the magnetic field.
334
Now we can constrain all this freedom using indeed the very Maxwell equations.
We have:
∇·E=0 ⇒ E0 · p̂f 0 = 0 ⇒ E0 · p̂ = 0 , (12.126)

∂B ∂B
∇×E=− ⇒ p̂ × E0 f 0 = − ⇒ p̂ × E0 f 0 = B0 g 0 c , (12.127)
∂t ∂t
where f 0 and g 0 are the derivatives of f and g with respect to their arguments.
From the first equation we learn that the electric field oscillates perpendicularly to
the direction of its propagation.
Exercise 12.18. Show from the other couple of Maxwell equations that:
∇·B=0 ⇒ B0 · p̂0 = 0 , (12.128)

1 ∂E 1
∇×B= ⇒ p̂0 × B0 g 0 = − E0 f 0 , (12.129)
c2 ∂t c
from which we learn that also the magnetic field oscillates perpendicularly to the
direction of its propagation. For now this is distinct from the one of the electric
field.
Now, since p̂ · (p̂ × E0 ) = 0, we can conclude that p̂ × B0 = 0, and therefore

that the magnetic field is perpendicular to both the directions of propagation. On
the other hand:
p̂0 × (p̂ × E0 )f 0 = p̂0 × B0 g 0 c = −E0 f 0 . (12.130)
Hence, being f 0 not identically vanishing otherwise we would not have a wave, we
can conclude that:
p̂0 × (p̂ × E0 ) = −E0 . (12.131)
Exercise 12.19. Use the rule for the triple cross product and show that:
p̂(p̂ · E0 ) − E0 (p̂ · p̂0 ) = −E0 . (12.132)
Therefore, being p̂ · E0 = 0 and E0 non-vanishing, we must conclude that:
p̂ · p̂0 = 1 ⇒ p̂ = p̂0 , (12.133)
since both are unit vectors and therefore the scalar product equal to unity means
that the angle between them must be vanishing. Thus, we have proven that electric
and magnetic field propagate in the same direction. Now we prove that f = g:
|E0 | g0
= 0 . (12.134)
|B0 |c f
335
In principle we should ask f 0 =
6 0 in order to make meaningful the above expression,
but since the left hand side is time-independent, so the right hand side must be.
Therefore:
|E0 |
g= f +K , (12.135)
|B0 |c
where K is an integration constant. Hence, we can write:

|E0 |
E = E0 f , B = B0 f +K . (12.136)
|B0 |c
Without losing any generality we can take |E0 | = |B0 | and K = 0, and still we
have a solution to Maxwell equations. Hence:
1
B = p̂ × E (12.137)
c
For this reason, the polarisation properties of an electromagnetic wave can be

analysed with respect to the electric field only.
12.7.2 Polarisation ellipse and Stokes parameters

Now we focus on a monochromatic electromagnetic plane wave, propagating in the
direction p̂ = ẑ. In complex notation the electric field can be written as:
 
Ex
E(t, x) = E0 ei(z−ct) =  Ey  ei(z−ct) , (12.138)
0
where the two-dimensional vector is called the Jones vector. It is often useful to
work with complex fields in electrodynamics, but here it will not be necessary so
let us instead work with:
   
Ex (t, x) Ax cos(z − ct + φx )
E(t, x) =  Ey (t, x)  =  Ay cos(z − ct + φy )  . (12.139)
0 0
The wave equation that we have solved in the previous subsection is for a vector
quantity, the electric field. Hence, there are actually three wave equations, one
for each component. However, we have seen that the electromagnetic wave is
transversal, therefore one of the component can always be put to zero, as we
have done above by choosing p̂ = ẑ. We have then two second-order differential
equations and thus we expect 4 integration constants, which we have introduced
above as the amplitudes Ax and Ay and the phases φx and φy . From these 4
quantities depends the polarisation state of the wave.
We can always redefine our clock such that:
   
Ex (t, x) Ax cos(z − ct)
E =  Ey (t, x)  =  Ay cos(z − ct + β)  , (12.140)
0 0
336
where β ≡ φy − φx is the relative phase, which is indeed the relevant physical quan-
tity. So we have actually 3 independent parameters which describe polarisation.
When β = 0, the polarisation is purely linear, whereas when β = π/2 it is purely
circular. This can be seen in general by studying the time-evolution of the field in
the Ey − Ex plane. In fact, we have:
Ex2 Ey2 2Ex Ey

+ − cos β = sin2 β (12.141)
A2x A2y Ax Ay
This is the equation of an ellipse, the polarisation ellipse, rotated by an angle α

in the Ey − Ex plane.
Exercise 12.20. Defining:

Ax = A cos θ , Ay = A sin θ , (12.142)
show that:
tan 2α = tan 2θ cos β , (12.143)
and that the semi-major and semi-minor axis of the polarisation ellipse are given
by:
A2
q
2 2 2
a = 1 + 1 − sin 2θ sin β , (12.144)
2
A2
q
2 2 2
b = 1 − 1 − sin 2θ sin β . (12.145)
2
When β = 0 we have b = 0, i.e. the ellipse degenerates in a straight line, titled

by an angle α with respect to the Ex axis. In this case the wave is purely linearly
polarised. On the other hand, when β = π/2 the ellipse degenerates in a circle,
and thus we have a purely circularly polarised wave.
The Stokes parameters are defined as follows in terms of the polarisation
ellipse parameters:
I ≡ a2 + b2 = A2 , (12.146)
q
Q ≡ (a2 − b2 ) cos 2α = A2 1 − sin2 2θ sin2 β cos 2α = A2 cos 2θ , (12.147)
q
U ≡ (a − b ) sin 2α = A 1 − sin2 2θ sin2 β sin 2α = A2 sin 2θ cos β , (12.148)
2 2 2
V ≡ 2abh = (A2 sin 2θ sin β)h , (12.149)

where h = ±1 simply establishes the direction of rotation, clockwise (h = −1) or
anti-clockwise (h = 1). Note that the Stokes parameter I is the intensity of the
wave.
Pure Q polarisation. If U = V = 0, then β = 0 and therefore we have pure

linear polarisation. Since sin 2α = 0 then either α = 0 or α = π/2. In the former
case the electric field oscillates along the x-axis and Q = A2 = I > 0. If α = π/2
then the electric field oscillates along the y-axis and Q = −A2 = −I < 0.
337
Pure U polarisation. If Q = V = 0, then β = 0 and therefore we have again
pure linear polarisation. This time cos 2α = 0 then either α = π/4 or α = 3π/4.
In the former case the electric field oscillates along the y = x line and Q = A2 =
I > 0. If α = 3π/4 then the electric field oscillates along the y = −x line and
Q = −A2 = −I < 0.
Pure V polarisation. If Q = U = 0, then θ = π/4 and β = π/2 and therefore

we have pure circular polarisation. We have V = A2 h = ±A2 = ±I, with the sign
determined by the direction of rotation.
As we have seen, the polarisation is fully determined by 3 parameters. These
means that the 4 Stokes parameters are not independent.
Exercise 12.21. Show that the Stokes parameters satisfy the following constraint:
Q2 + U 2 + V 2 = I 2 (12.150)
Other definitions of the Stokes parameters are the following. Using the complex
electric field:  
Ex
E(t, x) =  Ey  ei(z−ct) , (12.151)
0
they are defined as:
I ≡ |Ex |2 + |Ey |2 , Q ≡ |Ex |2 − |Ey |2 , (12.152)

U ≡ 2Re(Ex Ey∗ ) , V ≡ −2Im(Ex Ey∗ ) . (12.153)
We have considered so far a monochromatic wave, possibly made up of many waves

but all coherent. On the other hand, if we consider the superposition of incoherent
waves then we can write a generic time-dependence for the electric field:
 
Ex (t, x)
E(t, x) =  Ey (t, x)  , (12.154)
0
and therefore the Stokes parameters are defined as expectation values:
I ≡ hEx2 i + hEy2 i , Q ≡ hEx2 i − hEy2 i , (12.155)

U ≡ 2hEx Ey i cos β , V ≡ 2hEx Ey i sin β , (12.156)
as for example time-averages:

Z T
1
hXi = dt X(t) . (12.157)
T 0
In this case the electric field is a random variable and we can just define polarisation
through averages. Moreover, in this case one has:
Q2 + U 2 + V 2 ≤ I 2 (12.158)
338
since each Stokes parameter is defined as the sum of the corresponding Stokes
parameters of all the waves composing the total one. See e.g. [Chandrasekhar,
1960].
Exercise 12.22. Show that the three definitions of Stokes parameters given above
are in fact equivalent for a single monochromatic sinusoidal wave.
Now, suppose to make successive anti-clockwise rotations of π/4 in the polari-

sation plane. What happens is the following:
Q → U → −Q → −U → Q . (12.159)
That is, after a half rotation about the propagation direction we recover the initial
polarisation state, as if we applied an identity operator. This suggests us that we
are dealing with a spin-2 field. Indeed, let us introduce the intensity matrix for an
electromagnetic wave propagating along −ẑ:
 
I(ẑ) + Q(ẑ) U (ẑ) − iV (ẑ) 0
1
Jij (−ẑ) =  U (ẑ) + iV (ẑ) I(ẑ) − Q(ẑ) 0  , (12.160)
2
0 0 0
where the factor 1/2 serves in order to reproduce the correct result that the trace
Jii = I. This intensity matrix reminds us of hTij , the tensor perturbation of the
metric, defined in Eq. (4.198).
Exercise 12.23. Show that applying the a rotation of an angle θ about the axis
ẑ we have that:
I→I, (12.161)
±2iθ
Q ± iU → e (Q ± iU ) , (12.162)
V →V . (12.163)
The different sign with respect to the GW case is due to the fact that the
rotation is made about the propagation direction −ẑ. Hence, a rotation by θ about
−ẑ corresponds to a rotation by −θ about ẑ, which is the line-of-sight direction.
The matrix defined on the two-dimensional subspace orthogonal to the propa-
gation direction:

1 I(ẑ) + Q(ẑ) U (ẑ) − iV (ẑ)
ρij (−ẑ) = , (12.164)
2 U (ẑ) + iV (ẑ) I(ẑ) − Q(ẑ)
is called photon density matrix and can be suggestively put in the following
form:
1
ρ = (I1 + Qσ3 + U σ1 + V σ2 ) , (12.165)
2
339
where 1 is the 2 × 2 identity matrix and the σ1,2,3 are the Pauli matrices. It
might seem strange that the latter could be associated to a photon, since they are
usually employed for the description of spin-1/2 particles whereas the photon is a
spin-1 boson. However, since the photon is massless it is characterised by only two
spin states ±1 (or helicities), just as any spin-1/2 particle, massive or massless, is
characterised by two spin states ±1/2.
The polarisation vectors introduced in Eq. (4.213) can be defined equivalently
here. Along the ẑ axis they are simply:
√
e± (ẑ) = (1, ±i, 0)/ 2 , (12.166)
and the Stokes parameters are defined as:
Q(ẑ) ± iU (ẑ) = 2e±,i (ẑ)e±,j (ẑ)Jij (−ẑ) . (12.167)
For a generic direction of propagation, say −n̂, we define:
Q(n̂) ± iU (n̂) = 2e±,i (n̂)e±,j (n̂)Jij (−n̂) , (12.168)
where in spherical coordinates:
n̂ = (sin θ cos φ, sin θ sin φ, cos θ) , (12.169)
and the basis vectors which form the polarisation vectors can be chosen as:
θ̂ = (cos θ cos φ, cos θ sin φ, − sin θ) , φ̂ = (− sin φ, cos φ, 0) . (12.170)
Hence:
θ̂ ± iφ̂
e± (n̂) = √ . (12.171)
2
As we have already discussed, Q ± iU are spin-2 fields and thus are expanded as:
X
(Q + iU )(n̂) = aP,`m 2 Y`m (n̂) . (12.172)
`m
Through the polarisation vectors we can characterise the operator ð as
˜ i1 · · · ∇
ðs = (−1)s 2s/2 e+,i1 (n̂) · · · e+,is (n̂)∇ ˜ is (12.173)
where
˜ = θ̂ ∂ + φ̂ ∂ ,
∇ (12.174)
∂θ sin θ ∂φ
is the covariant derivative on the sphere.
Exercise 12.24. Show that the definition of Eq. (12.102) and formula (12.173)
are identical for s = 1 and s = 2.
340
12.8 Thomson scattering
In this section we work out the collisional term of Eq. (5.104). We follow the
calculations of [Chandrasekhar, 1960], but take advantage of the properties of
spherical harmonics, as done in [Hu and White, 1997]. See also [Kosowsky, 1996].
Consider an incoming wave propagating in the direction p̂0 and scattered in the
direction p̂. The plane in which p̂0 and p̂ lie is called scattering plane.
p̂
x̂
x̂0 ŷ
β
p̂0 = ẑ 0
ŷ 0 e−
Figure 12.1: Scattering plane. The unit vectors ŷ 0 and ŷ are equal and going out
from the page orthogonally.
Choosing the simple geometry of Fig. 12.1, the incoming electric field can be
decomposed as follows:
E0 = Ex0 x̂0 + Ey0 ŷ 0 , (12.175)
i.e. with no ẑ 0 component, since this is the direction of propagation. When the
electric field interacts with the electron, this starts to oscillate and to emit elec-
tromagnetic waves in almost all directions. The “almost” is make quantitative by
Larmor formula, for which the irradiated power per solid angle is:
dP e2
= 3
|p̂ × (p̂ × a)|2 , (12.176)
dΩ 4πc
where a is the electron acceleration. See e.g. [Jackson, 1998] and note that here we
are using Gaussian units for which [F ] = [e2 /r2 ], i.e. the force can be dimensionally
expressed as squared Coulomb divided by a squared length. For a particular out-
going polarisation say ˆ, we make the scalar product with ˆ in the square modulus
of the above formula and obtain:
dP e2
= 3
|ˆ · a|2 . (12.177)
dΩ 4πc
The acceleration of the electron is produced by the incident electric field E0 :
e 0
a= E , (12.178)
m
so that:
dP e4 2
= 2 3
|ˆ · E0 | . (12.179)
dΩ 4πm c
0 0 0
Writing now E = ˆ |E |, we get:
2
c 0 2 e2

dP 2
= |E | 2
|ˆ · ˆ0 | . (12.180)
dΩ 4π mc
341
The contribution c|E0 |2 /(4π) is the Poynting vector and hence the flux of the
incoming wave, leaving:
2 2
dσ e 2 3σT 2
= 2
|ˆ · ˆ0 | ≡ |ˆ · ˆ0 | (12.181)
dΩ mc 8π
as the differential cross section, where of course σT is the Thomson cross section.
The above derivation is based on Larmor formula and thus it is purely classical and
valid only for a non-relativistic movement of the electron and for photon energies
much smaller than the electron mass. If these conditions are not met, one should
employ the Klein-Nishina formula, see e.g. [Weinberg, 2005].
Let us assume σT = 1, for simplicity. We shall recover the correct dimensions
only at the end of our derivation via the scattering rate −τ 0 = ne σT a. The electric
field scattered in the direction p̂ is of the form:
E = A [p̂ × (p̂ × E0 )] , (12.182)
where A2 = 3/(8π). Let us use now the geometry in the scattering plane of Fig. 12.1
and calculate the double cross product.

E = −A x̂0 Ex0 cos2 β + ŷ 0 Ey0 + ẑ 0 Ex0 sin β cos β .

(12.183)
Since ŷ = ŷ 0 , we can already conclude that:

Ey = −AEy0 , (12.184)
i.e. the electric field contribution perpendicular to the scattering plane gains no
angular dependence.
In order to calculate Ex , which is the contribution of electric field parallel to
the scattering plane, we need to know x̂. One has:
x̂ = ŷ × p̂ , (12.185)
and
p̂ = (sin β, 0, − cos β) . (12.186)
Hence:
x̂ = −x̂0 cos β − ẑ 0 sin β . (12.187)
Therefore we have:
Ex = E · x̂ = AEx0 cos β . (12.188)
We are now in the position of relating the Stokes parameter of the outgoing wave
to those of the incoming one:
0
I = hEx2 i + hEy2 i = A2 hEx2 i cos2 β + A2 hEy20 i ≡ A2 Ix0 cos2 β + A2 Iy0 , (12.189)
Q = hEx2 i − hEy2 i = A2 Ix0 cos2 β − A2 Iy0 , (12.190)
U = 2hEx Ey i cos β = A2 U 0 cos β , (12.191)
V = 2hEx Ey i sin β = A2 V 0 cos β . (12.192)
342
Using now the definitions:
0
I 0 = hEx2 i + hEy20 i , (12.193)
02
Q0 = hEx i − hEy20 i , (12.194)
we have the complete transformation:

 1+cos2 β 2
− sin2 β I0
   
I 2
0 0
2β 2β
  Q0 
= 3  − 2
 Q sin 1+cos

 
2
0 0  0  . (12.195)
 U  8π  0 0 cos β 0  U 
V 0 0 0 cos β V0
These Stokes parameters have dimension of intensity. We now introduce Θ, the

temperature fluctuation instead of I and suitably redefine all the other Stokes
parameters. Moreover, we introduce the combination Q ± iU , since we know how
it transforms under a rotation in the polarisation plane, cf. Eq. (12.162). We shall
need this rotation in a moment. We have then:
2 2
 
1 + cos2 β − sin2 β − sin2 β Θ0
  
Θ
 Q + iU  = 3   − sin2 β (1+cos β)2 (1−cos β)2  
 Q0 + iU 0  ,
16π 2 2
Q − iU − sin2 β (1−cos β)2 (1+cos β) 2
Q0 − iU 0
2 2
(12.196)
and V does not mix up with the other Stokes parameter, so we consider it sepa-
rately:
3 cos β 0
V = V . (12.197)
16π
Recall that the above transformations hold true in the scattering plane and the
primed quantities depend on the incoming direction p̂0 whereas the unprimed quan-
tities from p̂.
We now have to transform to a generic reference frame in which the incident
wave comes from a direction (θ0 , φ0 ) and the outgoing wave goes along a direction
(θ, φ). This calculation is performed in [Chandrasekhar, 1960] but we follow [Hu
and White, 1997] because it is faster and takes advantage of the properties of the
spherical harmonics which we have seen earlier. In particular, we refer to Fig. 12.2.
We need to perform a rotation R(−α) in order to pass from the laboratory frame
to the scattering frame. Recall that the rotation is performed in the polarisation
plane, i.e. about −p̂0 . Hence it is a clockwise rotation and for this reason we have
−α. Then we apply the above matrix say S(β) in order to obtain the outgoing
Stokes parameters. Finally, we apply another rotation R(−γ) in order to obtain
the Stokes parameters in the laboratory frame. This rotation seems to be anti-
clockwise because it is in the direction opposite to R(−α). However, notice that
p̂ is now outgoing and thus the polarisation plane is reflected, giving then R(−γ)
instead of R(γ).
Exercise 12.26. Using the transformation of Eq. (12.162) and the fact that Θ is
343
z
(θ, φ)
γ
α
(θ0 , φ0 )
β
y
Figure 12.2: Scattering geometry.
invariant under rotation in the polarisation plane, show that:

2 2
 
1 + cos2 β − sin2 β e−2iα − sin2 β e2iα
3  β)2 2iα β)2 2iα 
R(−γ)S(β)R(−α) =  −e−2iγ sin2 β e−2iγ (1+cos 2
e e−2iγ (1−cos2
e  ,
16π 2iγ 2 2
2iγ (1−cos β) −2iα
2
2iγ (1+cos β) 2iα
−e sin β e 2
e e 2
e
(12.198)
Since also V is invariant under rotation in the polarisation plane, Eq. (12.200)
is valid even in the laboratory frame.
Exercise 12.27. Using the definitions of spherical harmonics given in Tab. 5.1,
write the above matrix as:
r  Y 0 + 2√5Y 0 − 3/2Y −2 − 3/2Y 2 
p p
1 4π  √ −2iγ 0
2 0
−2iγ
2
−2 −2iγ
2
2  .
R(γ)S(β)R(−α) = −√6e Y
2 2 3e Y
2 2 3e Y
2 2
8π 5
− 6e2iγ −2 Y20 3e2iγ −2 Y2−2 3e2iγ −2 Y22
(12.199)
The spherical harmonics inside this matrix are function of (β, α). For V show that:
r
1 3 0
V = Y (β, α)V 0 . (12.200)
4 4π 1
Now, using the addition theorem:

`
X
m∗ 0 0 m 2` + 1 −s1 −is2 γ
s1 Y` (θ , φ )s2 Y` (θ, φ) = s2 Y` e , (12.201)
m=−`
4π
we can introduce the matrix:
 m∗ m
p m∗ m
p m∗ m

√Y 2 Y 2 − 3/2 2 Y 2 Y 2 − 3/2 −2 Y 2 Y 2
P (m) (p̂, p̂0 ) =  −√ 6Y2m∗ 2 Y2m 32 Y2m∗ 2 Y2m 3−2 Y2m∗ 2 Y2m  ,
m∗ m m∗ m m∗ m
− 6Y2 −2 Y2 32 Y2 −2 Y2 3−2 Y2 −2 Y2
(12.202)
344
where the complex conjugates spherical harmonics depend on p̂0 whereas the others
depend on p̂. Hence, the scattered Stokes parameters can then be written as:
 0 
Θ0
   
Θ Θ 2
 Q + iU  = 1  0  + 1
X
P (m) (p̂, p̂0 )  Q0 + iU 0  , (12.203)
4π 10 m=−2
Q − iU 0 Q0 − iU 0
whereas for V we can write:
1
1 X m
V (p̂) = Y (p̂)Y1m∗ (p̂0 )V (p̂0 ) . (12.204)
4 m=−1 1
Integrating over all the incoming directions and multiplying by the Thomson scat-
tering rate:
− τ 0 = ne σ T a , (12.205)
we obtain part of the collisional terms for the Boltzmann equation (5.104). This
is the scattering rate into photons with momentum in the direction p̂, which we
have taken as the reference direction. There is also a contribution which takes
into account the scattering of photons into other directions different from p̂. This
generates the term:  
Θ
τ 0  Q + iU  . (12.206)
Q − iU
Finally, all the calculations performed here assume the electron fluid at rest. Per-
forming a boost, we get the Doppler shift term −τ 0 p̂·vb . Indeed, the photon energy
in the new frame is given by:
p̃ = p(1 − vb · p̂) , (12.207)
upon a boost with velocity vb . The Lorentz factor on the right hand side is unity
since |vb |2 is negligible, being a first-order quantity. The perturbed distribution
Fγ , being a scalar, transforms as:
Fγ (p) = F̃γ [p(1 − vb · p̂)] , (12.208)
and thus, developing the distribution function up to first-order, we get:
∂ f¯γ
Fγ (p) = F̃γ (p) − pvb · p̂ , (12.209)
∂p
and using Eq. (5.80) we have that:
Θ = Θ̃ + vb · p̂ . (12.210)
We have thus fully recovered the collisional term of Eq. (5.104).
Finally, consider initially unpolarised photons. Then, upon scattering:
   R d2 p̂0 
Θ 0
4π
Θ(p̂ )
 Q + iU  (p̂) =  0 
Q − iU 0
Ym
 
2
1 X  √2 m 
Z
+ −√ 62 Y2 (p̂) d2 p̂0 Y2m∗ (p̂0 )Θ(p̂0 ) . (12.211)
10 m=−2 m
− 6−2 Y2
345
Expanding the relative temperature fluctuation in spherical harmonics:
X
Θ(p̂0 ) = a`m Y`m (p̂0 ) , (12.212)
`m
due to the orthogonality properties of the latter, polarisation will be produced only
if the incident Θ(p̂0 ) has a quadrupole moment.
346
Bibliography
[Abbott et al., 2017a] Abbott, B. et al. (2017a). GW170817: Observation of

Gravitational Waves from a Binary Neutron Star Inspiral. Phys. Rev. Lett.,
119(16):161101.
[Abbott et al., 2016a] Abbott, B. P. et al. (2016a). GW151226: Observation of
Gravitational Waves from a 22-Solar-Mass Binary Black Hole Coalescence. Phys.
Rev. Lett., 116(24):241103.
[Abbott et al., 2016b] Abbott, B. P. et al. (2016b). Observation of Gravitational
Waves from a Binary Black Hole Merger. Phys. Rev. Lett., 116(6):061102.
[Abbott et al., 2017b] Abbott, B. P. et al. (2017b). A gravitational-wave standard
siren measurement of the Hubble constant. Nature, 551(7678):85–88.
[Abbott et al., 2017c] Abbott, B. P. et al. (2017c). Multi-messenger Observations
of a Binary Neutron Star Merger. Astrophys. J., 848(2):L12.
[Abbott and Wise, 1984] Abbott, L. F. and Wise, M. B. (1984). Constraints on
Generalized Inflationary Cosmologies. Nucl. Phys., B244:541–548.
[Abbott et al., 2017d] Abbott, T. M. C. et al. (2017d). Dark Energy Survey Year
1 Results: Cosmological Constraints from Galaxy Clustering and Weak Lensing.
[Abramowitz and Stegun, 1972] Abramowitz, M. and Stegun, I. A. (1972). Hand-
book of Mathematical Functions With Formulas, Graphs, and Mathematical Ta-
bles. Dover.
[Adam et al., 2016] Adam, R. et al. (2016). Planck 2015 results. I. Overview of
products and scientific results. Astron. Astrophys., 594:A1.
[Ade et al., 2016a] Ade, P. A. R. et al. (2016a). Planck 2015 results. XIII. Cosmo-
logical parameters. Astron. Astrophys., 594:A13.
[Ade et al., 2016b] Ade, P. A. R. et al. (2016b). Planck 2015 results. XVII. Con-
straints on primordial non-Gaussianity. Astron. Astrophys., 594:A17.
[Ade et al., 2016c] Ade, P. A. R. et al. (2016c). Planck 2015 results. XX. Con-
straints on inflation. Astron. Astrophys., 594:A20.
[Albrecht and Steinhardt, 1982] Albrecht, A. and Steinhardt, P. J. (1982). Cos-
mology for Grand Unified Theories with Radiatively Induced Symmetry Break-
ing. Phys. Rev. Lett., 48:1220–1223.
347
[Alpher et al., 1948] Alpher, R. A., Bethe, H., and Gamow, G. (1948). The origin
of chemical elements. Phys. Rev., 73:803–804.
[Amendola and Tsujikawa, 2010] Amendola, L. and Tsujikawa, S. (2010). Dark

energy: theory and observations. Cambridge University Press.
[Avelino and Kirshner, 2016] Avelino, A. and Kirshner, R. P. (2016). The dimen-
sionless age of the Universe: a riddle for our time. Astrophys. J., 828(1):35.
[Baqui et al., 2016] Baqui, P. O., Fabris, J. C., and Piattella, O. F. (2016). Cos-
mology and stellar equilibrium using Newtonian hydrodynamics with general
relativistic pressure. JCAP, 1604(04):034.
[Bardeen, 1980] Bardeen, J. M. (1980). Gauge Invariant Cosmological Perturba-

tions. Phys. Rev., D22:1882–1905.
[Bardeen et al., 1986] Bardeen, J. M., Bond, J. R., Kaiser, N., and Szalay, A. S.
(1986). The Statistics of Peaks of Gaussian Random Fields. Astrophys. J.,
304:15–61.
[Bardeen et al., 1983] Bardeen, J. M., Steinhardt, P. J., and Turner, M. S. (1983).
Spontaneous Creation of Almost Scale - Free Density Perturbations in an Infla-
tionary Universe. Phys. Rev., D28:679.
[Battye et al., 2015] Battye, R. A., Charnock, T., and Moss, A. (2015). Tension
between the power spectrum of density perturbations measured on large and
small scales. Phys. Rev., D91(10):103508.
[Berestetskii et al., 1982] Berestetskii, V. B., Lifshitz, E. M., and Pitaevskii, L. P.

(1982). Quantum Electrodynamics, volume 4 of Course of Theoretical Physics.
Pergamon Press, Oxford.
[Bernstein, 1988] Bernstein, J. (1988). Kinetic theory in the expanding universe.
[Bertone and Hooper, 2016] Bertone, G. and Hooper, D. (2016). A History of Dark
Matter.
[Betoule et al., 2014] Betoule, M. et al. (2014). Improved cosmological constraints

from a joint analysis of the SDSS-II and SNLS supernova samples. Astron.
Astrophys., 568:A22.
[Bolliet et al., 2017] Bolliet, B., Barrau, A., Martineau, K., and Moulin, F. (2017).
Some Clarifications on the Duration of Inflation in Loop Quantum Cosmology.
Class. Quant. Grav., 34(14):145003.
[Bond and Efstathiou, 1984] Bond, J. R. and Efstathiou, G. (1984). Cosmic back-
ground radiation anisotropies in universes dominated by nonbaryonic dark mat-
ter. Astrophys. J., 285:L45–L48.
[Bonometto, 2008] Bonometto, S. (2008). Cosmologia & cosmologie. Zanichelli.
348
[Bonvin et al., 2017] Bonvin, V. et al. (2017). H0LiCOW вҦҌ V. New COS-
MOGRAIL time delays of HE 0435вҾ1223: H0 to 3.8 per cent precision
from strong lensing in a flat ОCDM model. Mon. Not. Roy. Astron. Soc.,
465(4):4914–4930.
[Bouchet et al., 2011] Bouchet, F. R. et al. (2011). COrE (Cosmic Origins Ex-
plorer) A White Paper.
[Boylan-Kolchin et al., 2011] Boylan-Kolchin, M., Bullock, J. S., and Kaplinghat,

M. (2011). Too big to fail? The puzzling darkness of massive Milky Way sub-
haloes. Mon. Not. Roy. Astron. Soc., 415:L40.
[Brandenberger and Peter, 2017] Brandenberger, R. and Peter, P. (2017). Bounc-

ing Cosmologies: Progress and Problems. Found. Phys., 47(6):797–850.
[Brandenberger and Martin, 2013] Brandenberger, R. H. and Martin, J. (2013).

Trans-Planckian Issues for Inflationary Cosmology. Class. Quant. Grav.,
30:113001.
[Brans and Dicke, 1961] Brans, C. and Dicke, R. H. (1961). Mach’s principle and
a relativistic theory of gravitation. Phys. Rev., 124:925–935.
[Bucher et al., 2000] Bucher, M., Moodley, K., and Turok, N. (2000). The General
primordial cosmic perturbation. Phys. Rev., D62:083508.
[Bullock and Boylan-Kolchin, 2017] Bullock, J. S. and Boylan-Kolchin, M. (2017).

Small-Scale Challenges to the ΛCDM Paradigm. Ann. Rev. Astron. Astrophys.,
55:343–387.
[Butkov, 1968] Butkov, E. (1968). Mathematical Physics. Addison-Wesley Pub-

lishing Company, Incorporated.
[Camarena and Marra, 2016] Camarena, D. and Marra, V. (2016). Cosmological

constraints on the radiation released during structure formation. Eur. Phys. J.,
C76(11):644.
[Chandrasekhar, 1960] Chandrasekhar, S. (1960). Radiative transfer. New York:

Dover.
[Chluba, 2014] Chluba, J. (2014). Science with CMB spectral distortions. In Pro-
ceedings, 49th Rencontres de Moriond on Cosmology: La Thuile, Italy, March
15-22, 2014, pages 327–334.
[Chluba and Sunyaev, 2012] Chluba, J. and Sunyaev, R. A. (2012). The evolution
of CMB spectral distortions in the early Universe. Mon. Not. Roy. Astron. Soc.,
419:1294–1314.
[Clowe et al., 2006] Clowe, D., Bradac, M., Gonzalez, A. H., Markevitch, M., Ran-
dall, S. W., Jones, C., and Zaritsky, D. (2006). A direct empirical proof of the
existence of dark matter. Astrophys. J., 648:L109–L113.
349
[Coc, 2016] Coc, A. (2016). Primordial Nucleosynthesis. J. Phys. Conf. Ser.,
665(1):012001.
[Coleman and Weinberg, 1973] Coleman, S. R. and Weinberg, E. J. (1973). Radia-

tive Corrections as the Origin of Spontaneous Symmetry Breaking. Phys. Rev.,
D7:1888–1910.
[Crittenden et al., 1993] Crittenden, R., Bond, J. R., Davis, R. L., Efstathiou, G.,
and Steinhardt, P. J. (1993). The Imprint of gravitational waves on the cosmic
microwave background. Phys. Rev. Lett., 71:324–327.
[de Bernardis et al., 2000] de Bernardis, P. et al. (2000). A Flat universe from
high resolution maps of the cosmic microwave background radiation. Nature,
404:955–959.
[De Felice and Tsujikawa, 2010] De Felice, A. and Tsujikawa, S. (2010). f(R) the-
ories. Living Rev. Rel., 13:3.
[de Sitter, 1917] de Sitter, W. (1917). On the relativity of inertia. Remarks con-
cerning Einstein’s latest hypothesis. Koninklijke Nederlandsche Akademie van
Wetenschappen Proceedings, 19:1217–1225.
[de Sitter, 1918a] de Sitter, W. (1918a). Einstein’s theory of gravitation and its
astronomical consequences. Third paper. Mon. Not. Roy. Astron. Soc., 78:3–28.
[de Sitter, 1918b] de Sitter, W. (1918b). Further remarks on the solutions of the
field-equations of the Einstein’s theory of gravitation. Koninklijke Nederlandsche
Akademie van Wetenschappen Proceedings, 20:1309–1312.
[de Sitter, 1918c] de Sitter, W. (1918c). On the curvature of space. Koninklijke

Nederlandsche Akademie van Wetenschappen Proceedings, 20:229–243.
[Di Marco, 2017] Di Marco, A. (2017). Lyth Bound, eternal inflation and future
cosmological missions. Phys. Rev., D96(2):023511.
[Dodelson, 2003] Dodelson, S. (2003). Modern cosmology. Amsterdam, Nether-

lands: Academic Pr.
[Dodelson, 2017] Dodelson, S. (2017). Gravitational Lensing. Cambridge, UK:

Cambridge University Press, 2017.
[Dodelson and Widrow, 1994] Dodelson, S. and Widrow, L. M. (1994). Sterile-

neutrinos as dark matter. Phys. Rev. Lett., 72:17–20.
[Efstathiou et al., 1990] Efstathiou, G., Sutherland, W. J., and Maddox, S. J.

(1990). The cosmological constant and cold dark matter. Nature, 348:705–707.
[Eisenstein et al., 2005] Eisenstein, D. J. et al. (2005). Detection of the Baryon

Acoustic Peak in the Large-Scale Correlation Function of SDSS Luminous Red
Galaxies. Astrophys. J., 633:560–574.
350
[Eisenstein and Hu, 1998] Eisenstein, D. J. and Hu, W. (1998). Baryonic features
in the matter transfer function. Astrophys. J., 496:605.
[Etherington, 1933] Etherington, I. M. H. (1933). On the Definition of Distance

in General Relativity. Philosophical Magazine, 15.
[Fabris and Velten, 2012] Fabris, J. C. and Velten, H. (2012). Neo-Newtonian cos-
mology: An intermediate step towards General Relativity. RBEF, 4302.
[Fixsen et al., 1996] Fixsen, D. J., Cheng, E. S., Gales, J. M., Mather, J. C.,
Shafer, R. A., and Wright, E. L. (1996). The Cosmic Microwave Background
spectrum from the full COBE FIRAS data set. Astrophys. J., 473:576.
[Friedmann, 1922] Friedmann, A. (1922). Ueber die Kruemmung des Raumes . Z.

Phys., 10:377–386.
[Friedmann, 1924] Friedmann, A. (1924). Ueber die Moeglichkeit einer Welt mit
konstanter negativer Kruemmung des Raumes. Z. Phys., 21:326–332.
[Gaskins, 2016] Gaskins, J. M. (2016). A review of indirect searches for particle

dark matter. Contemp. Phys., 57(4):496–525.
[Gell-Mann et al., 1979] Gell-Mann, M., Ramond, P., and Slansky, R. (1979).
Complex Spinors and Unified Theories. Conf. Proc., C790927:315–321.
[Giblin et al., 2016] Giblin, J. T., Mertens, J. B., and Starkman, G. D. (2016).
Observable Deviations from Homogeneity in an Inhomogeneous Universe. As-
trophys. J., 833(2):247.
[Gorini et al., 2008] Gorini, V., Kamenshchik, A. Y., Moschella, U., Piattella,
O. F., and Starobinsky, A. A. (2008). Gauge-invariant analysis of perturba-
tions in Chaplygin gas unified models of dark matter and dark energy. JCAP,
0802:016.
[Grad, 1958] Grad, H. (1958). Principles of the kinetic theory of gases. In Ther-
modynamik der Gase/Thermodynamics of Gases, pages 205–294. Springer.
[Grieb et al., 2017] Grieb, J. N. et al. (2017). The clustering of galaxies in the
completed SDSS-III Baryon Oscillation Spectroscopic Survey: Cosmological im-
plications of the Fourier space wedges of the final sample. Mon. Not. Roy.
Astron. Soc., 467(2):2085–2112.
[Griffiths, 2017] Griffiths, D. J. (2017). Introduction to Electrodynamics. Cam-

bridge, UK: Cambridge University Press.
[Guth, 1981] Guth, A. H. (1981). The Inflationary Universe: A Possible Solution

to the Horizon and Flatness Problems. Phys. Rev., D23:347–356.
[Harrison, 1965] Harrison, E. R. (1965). Cosmology without general relativity.

Annals of Physics, 35:437–446.
351
[Hawking, 1966] Hawking, S. W. (1966). Perturbations of an expanding universe.
Astrophys. J., 145:544–554.
[Hogg, 1999] Hogg, D. W. (1999). Distance measures in cosmology.
[Hu and Sugiyama, 1996] Hu, W. and Sugiyama, N. (1996). Small scale cosmolog-
ical perturbations: An Analytic approach. Astrophys. J., 471:542–570.
[Hu and White, 1997] Hu, W. and White, M. J. (1997). CMB anisotropies: Total
angular momentum method. Phys. Rev., D56:596–615.
[Huang, 1987] Huang, K. (1987). Statistical Mechanics, 2nd Edition. Wiley-VCH.
[Hubble, 1929] Hubble, E. (1929). A relation between distance and radial velocity
among extra-galactic nebulae. Proc. Nat. Acad. Sci., 15:168–173.
[Hwang and Noh, 2013] Hwang, J.-c. and Noh, H. (2013). Newtonian Hydrody-
namics with General Relativistic Pressure. JCAP, 1310:054.
[Jackson, 1998] Jackson, J. D. (1998). Classical Electrodynamics, 3rd Edition.

Wiley-VCH.
[Jordan et al., 2009] Jordan, P., Ehlers, J., and Kundt, W. (2009). Republication
of: Exact solutions of the field equations of the general theory of relativity.
General Relativity and Gravitation, 41(9):2191–2280.
[Kaiser, 1983] Kaiser, N. (1983). Small-angle anisotropy of the microwave back-

ground radiation in the adiabatic theory. MNRAS, 202:1169–1180.
[Kaiser, 1984] Kaiser, N. (1984). On the Spatial correlations of Abell clusters.

Astrophys. J., 284:L9–L12.
[Kiefer and Polarski, 2009] Kiefer, C. and Polarski, D. (2009). Why do cosmolog-
ical perturbations look classical to us? Adv. Sci. Lett., 2:164–173.
[Kirby et al., 2014] Kirby, E. N., Bullock, J. S., Boylan-Kolchin, M., Kaplinghat,
M., and Cohen, J. G. (2014). The dynamics of isolated Local Group galaxies.
Mon. Not. Roy. Astron. Soc., 439(1):1015–1027.
[Klypin et al., 1999] Klypin, A. A., Kravtsov, A. V., Valenzuela, O., and Prada,
F. (1999). Where are the missing Galactic satellites? Astrophys. J., 522:82–92.
[Kodama and Sasaki, 1984] Kodama, H. and Sasaki, M. (1984). Cosmological Per-
turbation Theory. Prog. Theor. Phys. Suppl., 78:1–166.
[Kolb and Turner, 1990] Kolb, E. W. and Turner, M. S. (1990). The Early Uni-
verse. Front. Phys., 69:1–547.
[Kollmeier et al., 2017] Kollmeier, J. A., Zasowski, G., Rix, H.-W., Johns, M.,
Anderson, S. F., Drory, N., Johnson, J. A., Pogge, R. W., Bird, J. C., Blanc,
G. A., Brownstein, J. R., Crane, J. D., De Lee, N. M., Klaene, M. A., Kreckel, K.,
MacDonald, N., Merloni, A., Ness, M. K., O’Brien, T., Sanchez-Gallego, J. R.,
352
Sayres, C. C., Shen, Y., Thakar, A. R., Tkachenko, A., Aerts, C., Blanton,
M. R., Eisenstein, D. J., Holtzman, J. A., Maoz, D., Nandra, K., Rockosi, C.,
Weinberg, D. H., Bovy, J., Casey, A. R., Chaname, J., Clerc, N., Conroy, C.,
Eracleous, M., Gänsicke, B. T., Hekker, S., Horne, K., Kauffmann, J., McQuinn,
K. B. W., Pellegrini, E. W., Schinnerer, E., Schlafly, E. F., Schwope, A. D.,
Seibert, M., Teske, J. K., and van Saders, J. L. (2017). SDSS-V: Pioneering
Panoptic Spectroscopy. ArXiv e-prints.
[Kosowsky, 1996] Kosowsky, A. (1996). Cosmic microwave background polariza-

tion. Annals Phys., 246:49–85.
[Landau and Lifschits, 1975] Landau, L. D. and Lifschits, E. M. (1975). The Clas-
sical Theory of Fields, volume Volume 2 of Course of Theoretical Physics. Perg-
amon Press, Oxford.
[Landau and Lifshits, 1991] Landau, L. D. and Lifshits, E. M. (1991). Quan-

tum Mechanics, volume v.3 of Course of Theoretical Physics. Butterworth-
Heinemann, Oxford.
[Lemaitre, 1927] Lemaitre, G. (1927). A Homogeneous Universe of Constant Mass

and Growing Radius Accounting for the Radial Velocity of Extragalactic Nebu-
lae. Annales Soc. Sci. Brux. Ser. I Sci. Math. Astron. Phys., A47:49–59.
[Lemaitre, 1931] Lemaitre, G. (1931). The Expanding Universe. Mon. Not. Roy.
Astron. Soc., 91:490–501.
[Lemaı̂tre, 1997] Lemaı̂tre, G. (1997). The expanding universe. Gen. Rel. Grav.,
29:641–680.
[Lesgourgues, 2011] Lesgourgues, J. (2011). The Cosmic Linear Anisotropy Solving

System (CLASS) I: Overview.
[Lesgourgues and Pastor, 2006] Lesgourgues, J. and Pastor, S. (2006). Massive

neutrinos and cosmology. Phys. Rept., 429:307–379.
[Liddle et al., 1994] Liddle, A. R., Parsons, P., and Barrow, J. D. (1994). Formal-
izing the slow roll approximation in inflation. Phys. Rev., D50:7222–7232.
[Lifshitz, 1946] Lifshitz, E. (1946). On the Gravitational stability of the expanding

universe. J. Phys. (USSR), 10:116.
[Lifshitz and Khalatnikov, 1963] Lifshitz, E. M. and Khalatnikov, I. M. (1963).

Investigations in relativistic cosmology. Adv. Phys., 12:185–249.
[Lima et al., 1997] Lima, J. A. S., Zanchin, V., and Brandenberger, R. H. (1997).
On the Newtonian cosmology equations with pressure. Mon. Not. Roy. Astron.
Soc., 291:L1–L4.
[Linde, 2017] Linde, A. (2017). On the problem of initial conditions for inflation.
In Black Holes, Gravitational Waves and Spacetime Singularities Rome, Italy,
May 9-12, 2017.
353
[Linde, 1982] Linde, A. D. (1982). A New Inflationary Universe Scenario: A Pos-
sible Solution of the Horizon, Flatness, Homogeneity, Isotropy and Primordial
Monopole Problems. Phys. Lett., 108B:389–393.
[Liu et al., 2017] Liu, J., Chen, X., and Ji, X. (2017). Current status of direct dark
matter detection experiments. Nature Phys., 13(3):212–216.
[Lovell et al., 2012] Lovell, M. R., Eke, V., Frenk, C. S., Gao, L., Jenkins, A.,
Theuns, T., Wang, J., White, D. M., Boyarsky, A., and Ruchayskiy, O. (2012).
The Haloes of Bright Satellite Galaxies in a Warm Dark Matter Universe. Mon.
Not. Roy. Astron. Soc., 420:2318–2324.
[Lukash, 1980] Lukash, V. N. (1980). Production of phonons in an isotropic uni-

verse. Sov. Phys. JETP, 52:807–814. [Zh. Eksp. Teor. Fiz.79,1601(1980)].
[Lyth, 1997] Lyth, D. H. (1997). What would we learn by detecting a gravitational

wave signal in the cosmic microwave background anisotropy? Phys. Rev. Lett.,
78:1861–1863.
[Lyth and Liddle, 2009] Lyth, D. H. and Liddle, A. R. (2009). The primordial
density perturbation: Cosmology, inflation and the origin of structure.
[Ma and Bertschinger, 1995] Ma, C.-P. and Bertschinger, E. (1995). Cosmologi-
cal perturbation theory in the synchronous and conformal Newtonian gauges.
Astrophys. J., 455:7–25.
[Maartens, 1996] Maartens, R. (1996). Causal thermodynamics in relativity.
[Macciò et al., 2015] Macciò, A. V., Mainini, R., Penzo, C., and Bonometto, S. A.
(2015). Strongly Coupled Dark Energy Cosmologies: preserving LCDM success
and easing low scale problems II - Cosmological simulations. Mon. Not. Roy.
Astron. Soc., 453:1371–1378.
[Macciò et al., 2012] Macciò, A. V., Paduroiu, S., Anderhalden, D., Schneider, A.,
and Moore, B. (2012). Cores in warm dark matter haloes: a Catch 22 problem.
Mon. Not. Roy. Astron. Soc., 424:1105–1112.
[Maddox et al., 1990] Maddox, S. J., Efstathiou, G., Sutherland, W. J., and Love-
day, J. (1990). Galaxy correlations on large scales. Mon. Not. Roy. Astron. Soc.,
242:43–49.
[Malik and Matravers, 2013] Malik, K. A. and Matravers, D. R. (2013). Comments

on gauge-invariance in cosmology. Gen. Rel. Grav., 45:1989–2001.
[Marra et al., 2013] Marra, V., Amendola, L., Sawicki, I., and Valkenburg, W.
(2013). Cosmic variance and the measurement of the local Hubble parameter.
Phys. Rev. Lett., 110(24):241305.
[Martin, 2012] Martin, J. (2012). Everything You Always Wanted To Know About
The Cosmological Constant Problem (But Were Afraid To Ask). Comptes Ren-
dus Physique, 13:566–665.
354
[Martin et al., 2014] Martin, J., Ringeval, C., Trotta, R., and Vennin, V. (2014).
The Best Inflationary Models After Planck. JCAP, 1403:039.
[McCrea, 1951] McCrea, W. H. (1951). Relativity Theory and the Creation of

Matter. Proceedings of the Royal Society of London Series A, 206:562–575.
[McCrea and Milne, 1934] McCrea, W. H. and Milne, E. A. (1934). Newtonian

Universes and the curvature of space. The Quarterly Journal of Mathematics, 5.
[McVittie, 1962] McVittie, G. C. (1962). Appendix to The Change of Redshift and

Apparent Luminosity of Galaxies due to the Deceleration of Selected Expanding
Universes. The Astrophysical Journal, 136:334.
[Meszaros, 1974] Meszaros, P. (1974). The behaviour of point masses in an ex-

panding cosmological substratum. Astron. Astrophys., 37:225–228.
[Milne, 1934] Milne, E. A. (1934). A Newtonian expanding Universe. The Quar-

terly Journal of Mathematics, 5.
[Milne, 1935] Milne, E. A. (1935). Relativity, gravitation and world-structure. Ox-

ford, The Clarendon press.
[Moffat and Toth, 2011] Moffat, J. W. and Toth, V. T. (2011). Comment on “The
Real Problem with MOND” by Scott Dodelson, arXiv:1112.1320. ArXiv e-prints.
[Moore, 1994] Moore, B. (1994). Evidence against dissipationless dark matter from
observations of galaxy haloes. Nature, 370:629.
[Muñoz et al., 2017] Muñoz, J. B., Kovetz, E. D., Raccanelli, A., Kamionkowski,
M., and Silk, J. (2017). Towards a measurement of the spectral runnings. JCAP,
1705:032.
[Mukhanov, 2005] Mukhanov, V. (2005). Physical foundations of cosmology. Cam-

bridge, UK: Univ. Pr.
[Mukhanov, 1985] Mukhanov, V. F. (1985). Gravitational Instability of the Uni-

verse Filled with a Scalar Field. JETP Lett., 41:493–496. [Pisma Zh. Eksp. Teor.
Fiz.41,402(1985)].
[Mukhanov, 2004] Mukhanov, V. F. (2004). CMB-slow, or how to estimate cos-

mological parameters by hand. Int. J. Theor. Phys., 43:623–668.
[Mukhanov et al., 1992] Mukhanov, V. F., Feldman, H. A., and Brandenberger,

R. H. (1992). Theory of cosmological perturbations. Part 1. Classical pertur-
bations. Part 2. Quantum theory of perturbations. Part 3. Extensions. Phys.
Rept., 215:203–333.
[Newman and Penrose, 1966] Newman, E. T. and Penrose, R. (1966). Note on the
Bondi-Metzner-Sachs group. J. Math. Phys., 7:863–870.
[Novello and Bergliaffa, 2008] Novello, M. and Bergliaffa, S. E. P. (2008). Bounc-

ing Cosmologies. Phys. Rept., 463:127–213.
355
[Olbers, 1826] Olbers, W. (1826). Edinburgh new phil. J, 1:141.
[Peccei and Quinn, 1977] Peccei, R. D. and Quinn, H. R. (1977). CP Conservation

in the Presence of Instantons. Phys. Rev. Lett., 38:1440–1443.
[Peebles, 1968] Peebles, P. J. E. (1968). Recombination of the Primeval Plasma.

Astrophys. J., 153:1.
[Peebles, 1980] Peebles, P. J. E. (1980). The large-scale structure of the universe.

Princeton university press.
[Peebles and Yu, 1970] Peebles, P. J. E. and Yu, J. T. (1970). Primeval adiabatic
perturbation in an expanding universe. Astrophys. J., 162:815–836.
[Penzias and Wilson, 1965] Penzias, A. A. and Wilson, R. W. (1965). A Measure-

ment of excess antenna temperature at 4080- Mc/s. Astrophys. J., 142:419–421.
[Percival et al., 2007] Percival, W. J. et al. (2007). The shape of the SDSS DR5
galaxy power spectrum. Astrophys. J., 657:645–663.
[Perlmutter et al., 1999] Perlmutter, S. et al. (1999). Measurements of Omega and

Lambda from 42 High-Redshift Supernovae. Astrophys. J., 517:565–586.
[Piattella et al., 2016] Piattella, O. F., Casarini, L., Fabris, J. C., and de Fre-
itas Pacheco, J. A. (2016). Dark matter velocity dispersion effects on CMB and
matter power spectra. JCAP, 1602(02):024.
[Piattella and Giani, 2017] Piattella, O. F. and Giani, L. (2017). Redshift drift of
gravitational lensing. Phys. Rev., D95(10):101301.
[Piattella et al., 2014] Piattella, O. F., Martins, D. L. A., and Casarini, L. (2014).
Sub-horizon evolution of cold dark matter perturbations through dark matter-
dark energy equivalence epoch. JCAP, 1410(10):031.
[Piattella et al., 2013] Piattella, O. F., Rodrigues, D. C., Fabris, J. C., and de Fre-
itas Pacheco, J. A. (2013). Evolution of the phase-space density and the Jeans
scale for dark matter derivedfrom the Vlasov-Einstein equation. JCAP, 1311:002.
[Polnarev, 1985] Polnarev, A. G. (1985). Polarization and Anisotropy Induced

in the Microwave Background by Cosmological Gravitational Waves. Soviet
Astronomy, 29:607–613.
[Profumo, 2017] Profumo, S. (2017). An Introduction to Particle Dark Matter.

Advanced textbooks in physics. World Scientific.
[Profumo et al., 2006] Profumo, S., Sigurdson, K., and Kamionkowski, M. (2006).
What mass are the smallest protohalos? Phys. Rev. Lett., 97:031301.
[Racah, 1942] Racah, G. (1942). Theory of complex spectra. ii. Physical Review,
62(9-10):438.
356
[Riess et al., 1998] Riess, A. G. et al. (1998). Observational Evidence from Super-
novae for an Accelerating Universe and a Cosmological Constant. Astron. J.,
116:1009–1038.
[Robertson, 1935] Robertson, H. P. (1935). Kinematics and World-Structure. As-

trophys. J., 82:284.
[Robertson, 1936] Robertson, H. P. (1936). Kinematics and World-Structure III.

[Ryden, 2003] Ryden, B. (2003). Introduction to cosmology. San Francisco, USA:

Addison-Wesley (2003) 244 p.
[Sachs and Wolfe, 1967] Sachs, R. K. and Wolfe, A. M. (1967). Perturbations

of a cosmological model and angular variations of the microwave background.
Astrophys. J., 147:73–90.
[Sandage, 1958] Sandage, A. (1958). Current Problems in the Extragalactic Dis-

tance Scale. ApJ, 127:513.
[Sandage, 1962] Sandage, A. (1962). The Change of Redshift and Apparent Lu-
minosity of Galaxies due to the Deceleration of Selected Expanding Universes.
The Astrophysical Journal, 136:319.
[Sarkar and Pandey, 2016] Sarkar, S. and Pandey, B. (2016). An information the-
ory based search for homogeneity on the largest accessible scale. Mon. Not. Roy.
Astron. Soc., 463(1):L12–L16.
[Sasaki, 1986] Sasaki, M. (1986). Large Scale Quantum Fluctuations in the Infla-
tionary Universe. Prog. Theor. Phys., 76:1036.
[Schneider et al., 2014] Schneider, A., Anderhalden, D., Macciò, A., and Diemand,
J. (2014). Warm dark matter does not do better than cold dark matter in solving
small-scale inconsistencies. Mon. Not. Roy. Astron. Soc., 441:6.
[Schutz, 1985] Schutz, B. F. (1985). A First Course In General Relativity. Cam-

bridge, Uk: Univ. Pr.
[Schwarz et al., 2016] Schwarz, D. J., Copi, C. J., Huterer, D., and Starkman,
G. D. (2016). CMB Anomalies after Planck. Class. Quant. Grav., 33(18):184001.
[Sciama, 2012] Sciama, D. W. (2012). The unity of the universe. Courier Corpo-
ration.
[Seljak and Zaldarriaga, 1996] Seljak, U. and Zaldarriaga, M. (1996). A Line of

sight integration approach to cosmic microwave background anisotropies. Astro-
phys. J., 469:437–444.
[Silk, 1967] Silk, J. (1967). Fluctuations in the primordial fireball. Nature,

215(5106):1155–1156.
357
[Silk et al., 2010] Silk, J. et al. (2010). Particle Dark Matter: Observations, Models
and Searches. Cambridge Univ. Press, Cambridge.
[Slipher, 1917] Slipher, V. M. (1917). Nebulae. Proc. Am. Phil. Soc., 56:403–409.
[Smoot et al., 1992] Smoot, G. F. et al. (1992). Structure in the COBE differential
microwave radiometer first year maps. Astrophys. J., 396:L1–L5.
[Sofue and Rubin, 2001] Sofue, Y. and Rubin, V. (2001). Rotation curves of spiral
galaxies. Ann. Rev. Astron. Astrophys., 39:137–174.
[Sofue et al., 1999] Sofue, Y., Tutui, Y., Honma, M., Tomita, A., Takamiya, T.,
Koda, J., and Takeda, Y. (1999). Central rotation curves of spiral galaxies.
[Sotiriou and Faraoni, 2010] Sotiriou, T. P. and Faraoni, V. (2010). f(R) Theories
Of Gravity. Rev. Mod. Phys., 82:451–497.
[Spergel and Steinhardt, 2000] Spergel, D. N. and Steinhardt, P. J. (2000). Obser-

vational evidence for selfinteracting cold dark matter. Phys. Rev. Lett., 84:3760–
3763.
[Starobinsky, 1979] Starobinsky, A. A. (1979). Spectrum of relict gravitational

radiation and the early state of the universe. JETP Lett., 30:682–685. [Pisma
Zh. Eksp. Teor. Fiz.30,719(1979)].
[Stewart, 1990] Stewart, J. M. (1990). Perturbations of Friedmann-Robertson-

Walker cosmological models. Class. Quant. Grav., 7:1169–1180.
[Stewart and Walker, 1974] Stewart, J. M. and Walker, M. (1974). Perturbations

of spacetimes in general relativity. Proc. Roy. Soc. Lond., A341:49–74.
[Tram and Lesgourgues, 2013] Tram, T. and Lesgourgues, J. (2013). Optimal po-
larisation equations in FLRW universes. JCAP, 1310:002.
[Trotta, 2017] Trotta, R. (2017). Bayesian Methods in Cosmology.
[Tsujikawa, 2014] Tsujikawa, S. (2014). Distinguishing between inflationary models

from cosmic microwave background. PTEP, 2014(6):06B104.
[Valkenburg et al., 2014] Valkenburg, W., Marra, V., and Clarkson, C. (2014).
Testing the Copernican principle by constraining spatial homogeneity. Mon.
Not. Roy. Astron. Soc., 438:L6–L10.
[van den Bergh, 2011] van den Bergh, S. (2011). The Curious Case of Lemaître’s
Equation No. 24. JRASC, 105:151.
[Velten et al., 2014] Velten, H. E. S., vom Marttens, R. F., and Zimdahl, W.
(2014). Aspects of the cosmological вҦЁcoincidence problemвҦ№. Eur. Phys.
J., C74(11):3160.
358
[Verde et al., 2013] Verde, L., Protopapas, P., and Jimenez, R. (2013). Planck and
the local Universe: Quantifying the tension. Phys. Dark Univ., 2:166–175.
[Viel et al., 2013] Viel, M., Becker, G. D., Bolton, J. S., and Haehnelt, M. G.
(2013). Warm dark matter as a solution to the small scale crisis: New constraints
from high redshift Lyman- forest data. Phys. Rev., D88:043502.
[Vogelsberger et al., 2014] Vogelsberger, M., Zavala, J., Simpson, C., and Jenkins,
A. (2014). Dwarf galaxies in CDM and SIDM with baryons: observational probes
of the nature of dark matter. Mon. Not. Roy. Astron. Soc., 444:3684.
[Wagoner, 1973] Wagoner, R. V. (1973). Big bang nucleosynthesis revisited. As-

trophys. J., 179:343–360.
[Walker, 1937] Walker, A. G. (1937). On milne’s theory of world-structure. Pro-

ceedings of the London Mathematical Society, 2(1):90–127.
[Wands et al., 2000] Wands, D., Malik, K. A., Lyth, D. H., and Liddle, A. R.
(2000). A New approach to the evolution of cosmological perturbations on large
scales. Phys. Rev., D62:043527.
[Wands et al., 2016] Wands, D., Piattella, O. F., and Casarini, L. (2016). Physics
of the Cosmic Microwave Background Radiation. Astrophys. Space Sci. Proc.,
45:3–39.
[Warren et al., 2006] Warren, M. S., Abazajian, K., Holz, D. E., and Teodoro,
L. (2006). Precision determination of the mass function of dark matter halos.
Astrophys. J., 646:881–885.
[Way and Nussbaumer, 2011] Way, M. J. and Nussbaumer, H. (2011). Lemaître’s

Hubble relationship. Phys. Today, 64N8:8.
[Weinberg, 1972] Weinberg, S. (1972). Gravitation and Cosmology: Principles and

Applications of the General Theory of Relativity. Wiley - New York.
[Weinberg, 1989] Weinberg, S. (1989). The cosmological constant problem. Re-

views of Modern Physics, 61:1–23.
[Weinberg, 1992] Weinberg, S. (1992). Dreams of a final theory: The Search for
the fundamental laws of nature.
[Weinberg, 2002] Weinberg, S. (2002). Cosmological fluctuations of short wave-

length. Astrophys. J., 581:810–816.
[Weinberg, 2005] Weinberg, S. (2005). The Quantum theory of fields. Vol. 1: Foun-
dations. Cambridge University Press.
[Weinberg, 2006] Weinberg, S. (2006). A No-Truncation Approach to Cosmic Mi-

crowave Background Anisotropies. Phys. Rev., D74:063517.
[Weinberg, 2008] Weinberg, S. (2008). Cosmology. Oxford, UK: Oxford Univ. Pr.
359
[Weinberg, 2013] Weinberg, S. (2013). The quantum theory of fields. Vol. 2: Mod-
ern applications. Cambridge University Press.
[Weinberg, 2015] Weinberg, S. (2015). Lectures on Quantum Mechanics. Cam-

bridge University Press.
[Wilczek, 2015] Wilczek, F. (2015). Particle physics: A weighty mass difference.

Nature, 520:303–304.
[Williams et al., 1996] Williams, R. E., Blacker, B., Dickinson, M., Dixon,
W. V. D., Ferguson, H. C., Fruchter, A. S., Giavalisco, M., Gilliland, R. L.,
Heyer, I., Katsanis, R., Levay, Z., Lucas, R. A., McElroy, D. B., Petro, L.,
Postman, M., Adorf, H.-M., and Hook, R. (1996). The Hubble Deep Field:
Observations, Data Reduction, and Galaxy Photometry. AJ, 112:1335.
[Wilson and Silk, 1981] Wilson, M. L. and Silk, J. (1981). On the Anisotropy of the
cosomological background matter and radiation distribution. 1. The Radiation
anisotropy in a spatially flat universe. Astrophys. J., 243:14–25.
[Wu et al., 1999] Wu, K. K. S., Lahav, O., and Rees, M. J. (1999). The large-scale
smoothness of the Universe. Nature, 397:225–230. [,19(1998)].
[Zeldovich, 1984] Zeldovich, Y. B. (1984). Structure of the Universe. Astrophysics

and Space Physics Reviews, 3:1.
[Zimdahl, 1996] Zimdahl, W. (1996). Bulk viscous cosmology. Phys. Rev.,

D53:5483–5493.
[Zlatev et al., 1999] Zlatev, I., Wang, L.-M., and Steinhardt, P. J. (1999).
Quintessence, Cosmic Coincidence, and the Cosmological Constant. Phys. Rev.
Lett., 82:896–899.
[Zwicky, 1933] Zwicky, F. (1933). Die Rotverschiebung von extragalaktischen

Nebeln. Helvetica Physica Acta, 6:110–127.
360
Subject Index
ΛCDM model, 6, 30 Collisional term, 70

Age of the universe, 30 Collisionless, 68
Scale factor solution, 37 Coupled to FLRW metric, 69
σ8 , 316 Force term, 132
f (R) gravity, 222 Scalar perturbations, 134
Tensor perturbations, 134
Absolute magnitude, 307 Vector perturbations, 135
Acceleration equation, 24 Moments, 69
Acoustic oscillations, 267 Non-relativistic case, 65
Adiabatic speed of sound, 115, 230 Perturbation, 131
Age of the universe, 26 Relativistic case, 68
Alternatives to Λ, 10 Bose-Einstein distribution, 54, 320,
Angular diameter distance, 43 321
Anisotropic stress, 106, 124 Bouncing cosmology, 194
Anthropic principle, 10 Bulk viscosity, 106
Apparent magnitude, 306 Bullet cluster, 5
Associated Legendre polynomials,
328 Chemical equilibrium, 46, 75
Axion, 6 Christoffel symbols
Perturbation, 100
Bardeen’s potentials, 115 CMBFAST, 279
Baryogenesis, 47 Cold Dark Matter, 29
Baryon Acoustic Oscillations, 239, Density parameter, 30
268 Perturbed Boltzmann equation,
Baryon feedback, 11 136
Baryon loading, 272 Small-scale anomalies, 11
Baryon-to-photon ratio, 48, 75 Comoving coordinates, 17
Baryons, 29 Comoving curvature perturbation,
Boltzmann equation, 155 116
Density parameter, 30 Conservation, 325
Bias, 225, 317 Comoving distance, 38
Big Bang Nucleosynthesis Comoving momentum, 53
Temperature, 75 Comoving-gauge density
Big-Bang, 12 perturbation, 116
Bispectrum, 183 Confidence level, 308
Reduced, 183 Conformal Hubble factor, 25
Bolometric magnitude, 306 Conformal time, 17
Boltzmann equation Conformal transformation, 222
361
Continuity equation, 28, 69 Dark Energy Survey, 8
Perturbation, 137 Dark Matter
Using thermodynamics laws, 46 Cold Dark Matter, 6
Copernican principle, 12, 188 Hot Dark Matter, 6
Core/Cusp problem, 11 Warm Dark Matter, 6
Correlation function, 177, 315 Dark matter, 4
Cosmic coincidence, 10 Dark matter decoupling, 48
Cosmic Microwave Background Dark matter searches, 9
CTST,` spectrum, 187 de Sitter universe, 32
T
CBB,` spectrum, 304 Deceleration parameter, 34
S
CEE,` spectrum, 300 Deceleration parameter, 27
T
CEE,` spectrum, 304 Density contrast
CTSE,` spectrum, 300 Definition, 107
CTTE,` spectrum, 304 Density parameter
CTTT,` spectrum, 296 Closure relation, 28
Acoustic oscillations, 261 Density parameter Ω, 27
Anomalies, 12, 188 Deuterium bottleneck, 76
Doppler effect, 261 Diffusion damping, 266, 275
Integrated Sachs-Wolfe effect, Distance modulus, 307
261 Distribution function, 50
Large scale anisotropies, 262 Energy density, 51
Observation, 6 Energy-momentum tensor, 52
Planck TT spectrum, 257 Particle number density, 51
Polarisation spectra, 188 Perturbation, 108, 130
Power spectrum, 186 Pressure, 52
Primary anisotropies, 261 Dust, 29
Spectral distortions, 56 Dust-dominated universe
Temperature, 57 Scale factor solution, 36
Temperature fluctuations
expansion, 186 Effective number of relativistic
Cosmic time, 17 degrees of freedom, 80, 93
Cosmic variance, 178 Effective speed of sound, 115
Angular power spectrum, 188 Einstein equations
CMB power spectrum, 190 Scalar perturbations, 122
Power spectrum, 182, 183 Tensor perturbations, 125
Cosmography, 39 Vector perturbations, 128
Cosmological constant Einstein frame, 222
Density parameter, 30 Einstein static universe, 32
Problem, 10 Einstein tensor
Cosmological observations, 6 Perturbation, 104
Cosmological principle, 14 Einstein-de Sitter universe, 35
Coulomb scattering, 155 Electron-positron annihilation, 49
Critical density, 27 Electroweak phase transition, 48
Cross section Energy momentum tensor
Thermally averaged, 74 Imperfect fluid, 106
Kinetic theory, 109
Dark Energy, 4 Perturbations, 107
362
Ensemble, 176 Problem, 110
Ensemble average, 176 Gauge transformation, 111
Ensemble variance, 177 Energy-momentum tensor, 112
Entropy density, 54 Metric, 111
Entropy modes, 162 Gaussian filter, 312
Entropy perturbation, 115 Grand unified theory (GUT), 47
Equation of state, 29 Gravitational waves
Equivalence wavenumber, 234 Equation, 126
Ergodic theorem, 178, 192 Helicity, 127
Etherington’s distance duality, 44 Observatories, 8
Euclid (experiment), 8 Sum over helicities, 191
Euler equation Green’s function, 333
Perturbation, 138 Growth factor, 237
Event horizon, 41
Heat transfer, 106
Evolution of perturbation
Helium mass fraction, 83
Super-horizon, 229
Helmholtz theorem, 322
Evolution of perturbations
Horizon crossing, 160
Λ-dominated epoch, 232
Horizon problem, 196
Mészáros equation, 245
Hot Big Bang
Matter-dominated epoch, 238
Thermal history, 47
Radiation-dominated epoch, 243
Hot dark matter, 29
Fermi-Dirac distribution, 54, 319, 320 Hubble constant, 1, 25, 26
Fine structure constant, 49 Hubble parameter, 18
Fine tuning, 195 Hubble radius, 2, 20
Finite thickness effect, 283 Hubble’s law, 1
Flatness problem, 27, 194 Derivation, 39
FLRW metric, 17 IceCube, 9
Christoffel symbols, 22, 101 Inflation, 195
Light-cone structure, 18 e-folds number, 196
Perturbation, 99 Energy scale, 196
Perturbed Christoffel symbols, Horizon problem, 196
102 Inflaton, 198
Perturbed Einstein tensor, 104 Lyth bound, 203
Perturbed Ricci scalar, 104 Maximum number of e-folds, 200
Perturbed Ricci tensor, 103 Planck constraints, 220
Ricci tensor, 24 Power-law, 221
Four-velocity Slow-roll, 198
Perturbations, 108 Starobinsky model, 222
Fourier transform, 120 Trans-Planckian problem, 201
Free electron fraction, 84 Interacting dark matter, 12
Free-streaming Interaction rate, 46
Solution, 261 Isocurvature modes, 162
Freeze out, 47 Isometry, 111
Friedmann equation, 24
Jerk parameter, 40
Gauge, 99 Jordan frame, 222
363
Killing equations, 111 Mass constraints, 61
Kinetic equilibrium, 46 Massive neutrinos density
Kodama-Sasaki equation, 229 parameter, 63
Perturbed Boltzmann equation,
Landau damping, 283 139
Landau damping scale, 286 Relative neutrino heat flux, 164
Larmor formula, 341 Right-handed, 58
Last scattering surface, 84 Sterile, 58
Legendre polynomials Temperature, 60
Expansion, 140 Temperature equations hierarchy,
Orthogonality relation, 140 140
Recurrence relation, 140 Neutron abundance, 80
Legendre polynomials expansion, 149 Newtonian cosmology, 13
Leptogenesis, 47 Newtonian gauge, 118
Likelihood, 308
Non-Gaussianity, 183
Line-of-sight, 259
Local type, 183
Unit vector, 185
Planck constraint on fN L , 183
Line-of-sight integration, 279
Normal mode decomposition, 119
Linearised Einstein equations, 109
Number density
Liouville operator, 65
Equilibrium, 72
Liouville theorem, 65
Proof, 66 Olbers’s paradox, 2
Lithium problem, 12 Open problems in cosmology, 12
Lookback time, 39 Optical depth, 146
Lukash variable, 116
Luminosity distance, 42 Partial wave expansion, 140, 149
Particle horizon, 40
Marginalisation, 308
Particle number conservation, 69
Mass variance, 311
Particles
Matter-radiation equality, 64
Thermal production, 6
Maximally symmetric space, 15
Particles in a thermal bath, 62
Maxwell equations, 334
Perfect fluid, 24
Maxwell-Boltzmann distribution, 319
Periodic boundary conditions, 51
Method of characteristics, 66
Phase-space
Milne universe, 38
Volume of the fundamental cell,
Missing satellites problem, 11
51
Mukhanov-Sasaki equation, 216
Multiverse, 11 Photon density matrix, 339
Photons
Neutralino, 6 Decoupling from electrons, 84
Neutrino decoupling, 48 Density parameter, 30, 57
Neutrinos Energy density, 56
Density parameter, 30 Free-streaming, 258
Effective number of families Neff , Number density, 57
61 Perturbed Boltzmann equation
Energy density, 58 Collisional term, 144
Fraction, 165 Scalar perturbations, 149
Isocurvature velocity mode, 164 Tensor perturbations, 151
364
Vector perturbations, 153 Ricci tensor
Relative temperature Definition, 103
fluctuations, 143 Perturbation, 103
Poisson distribution, 322 Riemann ζ function, 56
Polarisation ellipse, 337 Rodrigues’ formula, 329
Polarisation vectors, 128, 340
Power spectrum, 175, 180 Sachs-Wolfe effect, 259
Amplitude, 218 Sachs-Wolfe plateau, 264
Dimensionless, 181 Saha equation, 75
Scale invariant, 210 Scalar field
Present time t0 , 26 Klein-Gordon equation, 199
Primordial modes, 160 Scalar-Vector-Tensor decomposition,
Adiabatic, 166 113
Baryon density isocurvature, 169 Scale factor, 17, 23
Cold Dark Matter density See-saw mechanism, 58
isocurvature, 169 Shot noise, 226, 315
Neutrino density isocurvature, Silk damping, 266
169 Silk length, 277
Neutrino velocity isocurvature, Size of the visible universe, 26
171 Sloan Digital Sky Survey, 7
Planck constraints, 172 Slow-roll
Primordial plasma, 47 Conditions, 199
Primordial tilt, 264 Parameters, 200
Proper distance, 38 Snap parameter, 40
Proper momentum, 22, 53, 109 Spatial curvature density, 28
Proper radius, 18 Spectral index
Running, 218
Quantum-to-classical transition, 205 Scalar, 217
Tensor, 211
Radiation, 29 Spherical Harmonics, 186
Radiation plus dust universe Spherical harmonics
Scale factor solution, 37 Addition theorem, 330
Radiation-dominated universe Completeness relation, 329, 332
Scale factor solution, 34 Complex conjugation, 329
Random field, 176 Normalisation, 329
Gaussian, 180 Orthonormality, 329, 332
Reality condition, 179, 186 Parity, 330
Recombination, 84 Plane wave expansion, 141, 330
Redshift, 9, 23 Spin-weighted, 331
photometric, 9 Spurious gauge modes, 115, 118
spectroscopic, 9 Standard candle, 41
Redshift drift, 26 Standard ruler, 43, 273
Reheating, 204 Statistical homogeneity, 176
Reionisation, 89, 281 Statistical isotropy, 177
Relativistic Poisson equation, 123 Stefan-Boltzmann law, 57
Ricci scalar Sterile neutrino, 6
Perturbation, 104 Stewart-Walker lemma, 112
365
Stochastic initial conditions, 184 Tight-coupling, 163, 265
Stokes parameters, 337 Too big to fail problem, 11
Boltzmann equation, 147 Top Hat filter, 312
Structure formation Top quark, 48
Bottom-up scenario, 237 Transfer function, 174, 184
Top-down scenario, 237 BBKS, 248
Sudden recombination, 265, 281 Gravitational waves, 256
Synchronicity problem, 30 Matter, 248
Synchronous gauge, 118
Vector perturbation, 113
Visibility function, 281
Tensions in cosmology, 12
Tensor perturbation, 114 Warm dark matter, 11
Tensor-to-scalar ratio, 219 Wigner 3j-symbols, 294
Thermal distributions, 318 Wigner D-matrix, 294, 331
Thermal relic, 90 WIMP, 6
Thomson scattering Miracle, 96
Cross section, 49, 342 WIMPless miracle, 96
Rate, 87 Wronskian, 334
366

Lecture Notes in Cosmology: Oliver F. Piattella

Uploaded by

Copyright:

Available Formats

Lecture Notes in Cosmology: Oliver F. Piattella

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture Notes in Cosmology: Oliver F. Piattella

Uploaded by

Copyright:

Available Formats

Lecture Notes in Cosmology

Núcleo Cosmo-UFES and Physics Department,

Together we are gold

2 The universe in expansion 13

5 Perturbed Boltzmann equations 130

6 Initial conditions 159

7 Stochastic Properties of Cosmological Perturbations 174

9 Evolution of perturbations 225

10 Anisotropies in the Cosmic Microwave Background 257

Subject Index 361

• S. Dodelson, Modern Cosmology, [Dodelson, 2003],

• V. Mukhanov, Principles of physical Cosmology, [Mukhanov, 2005],

• S. Weinberg, Cosmology, [Weinberg, 2008],

The signature employed for the metric is (−, +, +, +).

Spatial 3-vectors are indicate in boldface, e.g. v.

Except on vector and tensor, a 0 subscript means that a time-dependent quantity

With k is denoted the comoving wavenumber.

And that inverted Bowl we call the Sky,

In this Chapter we present an overview of cosmology, addressing its most impor-

1.1 The expanding universe and its content

where v is the recessional velocity, r is the distance and H0 (usually pronounced

1.1.1 Olbers’s paradox

1.1.2 The accelerated expansion of the universe and Dark

1.1.3 Dark Matter

1.2 Cosmological observations

1.2.1 The cosmic microwave background

1.2.2 Redshift surveys

• The Apache Point Observatory Galaxy Evolution Experiment (APOGEE-2)

• The Mapping Nearby Galaxies at Apache Point Observatory (MaNGA) study

1.2.3 Gravitational waves observatories

1.2.4 Neutrino observation

1.2.5 Dark matter searches

1.4 Open problems in cosmology

1.4.1 Cosmological constant and dark energy

1.4.2 Dark matter and small-scale anomalies

2. The Missing satellites problem [Klypin et al., 1999]. Numerical simulations

Possible solutions to these small-scale anomalies are the following:

• To solve the Core/Cusp problem is necessary a mass of ∼ 0.1 keV [Macciò

Interacting Dark Matter. CDM anomalies could perhaps be understood by

1.4.3 Other problems

The universe in expansion

oras ubicumque locaris extremas, quaeram: quid telo denique fiet?

2.1 Newtonian cosmology

Exercise 2.1. Integrate Eq. (2.2) and show that:

2.2 Relativistic cosmology

2.2.1 Friedmann-Lemaı̂tre-Robertson-Walker metric

2. ds23 = |dx|2 + dz 2 , with the constraint z 2 + |x|2 = a2 . This is a 3-sphere of

3. ds23 = |dx|2 − dz 2 , with the constraint z 2 − |x|2 = a2 . This is a 3-hypersphere,

Let us write in a compact form the above metrics as follows:

ds23 = |dx|2 ± dz 2 , z 2 ± |x|2 = a2 . (2.8)

Differentiating z 2 ± |x|2 = a2 , one gets:

Now put this back into ds23 :

In a more compact form:

dΩ2 ≡ dθ2 + sin2 θdφ2 , (2.13)

and letting a to be a function of time, we finally get the FLRW metric:

2.2.2 The conformal time

2.2.4 Light-cone structure of the FLRW space

Cosmic time-comoving distance

Conformal time-comoving distance

��