You are on page 1of 201

The Cosmic Microwave

Background: basic theory


Lecture 1
Yacine Ali-Haïmoud
Johns Hopkins University

Shanghai 2017 Spring School on Cosmology


Review of cosmology basics
Friedmann-Lemaître-Robertson-Walker (FLRW) metric for a
homogeneous and isotropic (and spatially flat) Universe:

2
ds = 2 2
dt + a (t)d~x 2 a(t): scale factor (a=1 today)

2
⇥ 2 2
⇤ x : comoving coordinate
= a (⌘) d⌘ + d~x
η: conformal time

For comoving separation Δx, physical length Δxphys = a Δx

dN 3
=> Number densities of particles n= 3 / 1/a
d xphys
Particle momenta redshift: p / 1/a ⌘ 1 + z
Review of cosmology basics
Einstein’s equations Gµν = 8π G Tµν => Friedmann’s equation:
✓ ◆2
2 ȧ 8⇡G H: Hubble expansion rate
H ⌘ = ⇢ H0 = H(t0)
a 3c2
⇢ = ⇢m + ⇢r + ⇢⇤ ρ: energy density

2 3
non-relativistic matter: ⇢m = nm mc / a
4
radiation: ⇢r = nr hEi = nr hpci / a

cosmological constant: ⇢⇤ = constant


Review of cosmology basics
Non-relativistic matter
• “baryons”: atoms, ions, and electrons
All are non-relativistic at T ≲ MeV

Big Bang Nucleosynthesis produces Helium and trace


amounts of heavier elements.

Mass fractions: 76% Hydrogen, 24% Helium


Number fractions: 92% Hydrogen, 8% Helium
✓ ◆3
3 1+z
Hydrogen abundance: nH ⇠ 200 cm
1000
• cold dark matter: unknown non-interacting matter assumed to
be non-relativistic at all times of interest.
Review of cosmology basics
Neutrinos
3 flavors (νe ,νµ ,ντ) = linear combinations of 3 mass eigenstates (ν1 ,ν2 ,ν3)

From cosmological measurements we know that mi . 0.1 eV

Decouple from the rest of the plasma at T ~ 1 MeV, i.e. while relativistic.
From that moment keep a Fermi-Dirac distribution.
✓ ◆1/3
4
Decouple before e- - e+ annihilation => T⌫ = T ⇡ 0.7 T
11

behave like radiation for T ≳ 0.1 eV


and non-relativistic matter for T ≲ 0.1 eV
Review of cosmology basics
Photons
Expanding Universe was hotter and denser early on.
Early enough photons were in thermal equilibrium:
2
dn 8⇡E 1 blackbody
= 3 E/T spectrum
dE (hc) e 1
Z
dn 4
⇢ = dEE = aR T aR : radiation constant
dE
If photons stop interacting when they fall out of equilibrium, they keep
a blackbody spectrum with temperature 1
T ⇠ hE i / a
Prediction of the CMB
Using measured Helium abundance + Big Bang Nucleosynthesis
theory, Alpher & Herman (1948) first suggested that the Universe was
initially radiation-dominated.

They estimated the required CMB temperature to be ~5 K today.

In the 1960’s H0 measured to ≲ 100 km/s/Mpc


=> Friedmann’s equation imply maximum energy density at t0:
2
3c 3
⇢0 = H0 . 2 ⇥ 10
2 8
erg/cm
8⇡G
=> Maximum photon temperature today:
T ,0 . (⇢0 /aR ) 1/4
. 40 K
Discovery of the CMB
Roll & Wilkinson were attempting to measure this radiation…

… but it was discovered by chance by Penzias & Wilson (1965)

Measured excess noise in their


antenna, at temperature

T = 3.5 ± 1 K

Dicke, Peebles, Roll &


Wilkinson (1965) suggest this is
Cosmic Blackbody radiation
Additional measurements are required to
for the Coulomb energy and the center-of-mass January 1966 (unpublished).

Discovery
confirm the of the CMB
interpretation
COSMIC BACKGROUND RADIATION AT 3.2 cm —SUPPORT FOR COSMIC BLACK-BODY RADIATION*

REVIEWby Roll
AL Confirmed LETTERS
&
P. G. Rollt and David T. Wilkinson 7 MARcH 1966
Wilkinson (1966)
Palmer Physical Laboratory, Princeton University, Princeton, New Jersey
(Received 27 January 1966}
of
Dicke et al. ' have suggested that the universe which would be emitted by a black body at 3.0
oldmay be filled with black-body
'4- radiation which +0.5 K. A more complete description of the
(fl lo
I)originated a time when the matter and radi-
at V) experiment will appear elsewhere.
QJ
ation were in a hot,e highly contracted, state Figure 1 shows a schematic diagram of the
of —the primordialI- fireball.
CL As the universe ex- instrument. It is a Dicke-type radiometer'
O
s panded, the cosmological red shift would have P R in which the receiver input is periodically switched
l6
K ~lO
NC E TON
between a horn antenna and a reference source
I

cooled the cosmic black-body radiation to the


er- extent that one should now look for it in the (cold load). The output of the receiver at the
(3.5
entsmicrowave band.O Concurrent with this sugges- switching
(3.
i
frequency is synchronously detect-
tion, Penzias and tAWilson' reported the discov- ed and recorded. The record is a measure
r- ery of an excess background
lO-is radiation at a, wave- of the difference between the temperature of
xper- 0Xx E
length of 7.35 cm. The measurement of the the reference source and the apparent temper-
spectrum of this tOnew microwave background ature of the radiation collected by the antenna.
re provides a severeO p test of the cosmic black-body-
0Z~g) The horn antenna is shielded to exclude radi-
ation from the ground and has a main lobe half-
0
radiation hypothesis. This Letter reports a
measurement of thel O-20microwave background at angle (10 dB down) of 10'. The cold-load ter-
a wavelength of 3.2 cm; the flux found is that mination is immersed in liquid helium to es-
l
lo' IO IO I !0 '
405
to WAVELENGTH (c m )
sis-
First frequency spectrum measurement
1990ApJ...354L..37M COBE FIRAS (Mather et al 1990)

Intensity

Frequency
First frequency spectrum measurement
584 FIXSEN ET AL. Vol. 473

Intensity

Frequency
FIG. 4.ÈUniform spectrum and Ðt to Planck blackbody (T ). Uncertainties are a small fraction of the line thickness.

of the spectra derived from the DIRBE templates in the can be performed on the unknown parameters p, G , and
Final FIRAS results (1996):
principal Ðt, g (l), and Ðt for their intensities at each pixel.
k
The CMBR temperature assumes a Planck spectrum. These
T = 2.728 ± 0.004 K 0
*T . The Ðrst two terms are the Planck blackbody spectrum,
with the temperature T ] *T . It is important to have the
spectra were chosen to match the shape of the spectrum in second term in order to0 properly estimate the uncertainty
Perfect blackbody up to precision of calibrator
the data. This yields maps of the CMBR temperature and
dust intensities. A monopole plus three dipole components
are then Ðtted to the resulting temperature map.
. 5 ⇥ 10 5
since the *T is strongly correlated with the resulting p (95%
in the case of the Bose-Einstein distortion). The third term
allows for Galactic contamination to remain in the mono-
The vector sum of the dipole coefficients points in the pole spectrum. The Ðnal term is the modeled deviation. We
direction (l, b) \ (264¡.14 ^ 0.15, 48¡.26 ^ 0.15), consistent Ðt either the Kompaneets parameter or the chemical poten-
Best measurements to date; planned missions will improve by ~103
with the direction from the DMR results. Data for
o b o \ 10¡ were excluded from the dipole Ðt because of the
tial, but the two are too similar to Ðt simultaneously. The
uncertainties are propagated from the template Ðts, and the
potential inaccuracy of the model of the Galaxy. The direc- correlation between the g(l) and LS /Lp increases the uncer-
c
tion is particularly sensitive to the Galaxy because it is tainty of G and p.
First anisotropy measurement

COBE DMR
(Smoot et al 1990)

Map of temperature fluctuations (after subtracting foregrounds)

T 5
⇠ 10
T
CMB temperature anisotropies
CMB polarization

Planck collaboration
Overview

1990ApJ...354L..37M
I- CMB frequency
spectrum

Frequency spectrum probes thermal history of the Universe since z


~ 2x106: spectral distortions, i.e. deviations from a perfect
blackbody probe energy or photon injection in the plasma.
Proposed instruments (PIXIE, PRISM) will measure CMB spectral
distortions of ~10-8, so it is important to understand implications

We will discuss the underlying physics in Lecture 2.


Overview

II- CMB anisotropies

CMB anisotropies are a linear transformation of the initial


perturbations. They carry information about the initial conditions
and the contents of the Universe.

Currently the most precise observational test of modern cosmology.


Anisotropies: problem setup
Early-Universe theories predict the statistical properties of initial metric and
density perturbations δ0 (δ0 represents all relevant initial perturbations).

The lowest-order statistic is the power spectrum of primordial perturbations:

~ ⇤ ~0
h 0 (k) 0 (k )i = (2⇡) 3 ~
Dirac (k
0 ~k)P0 (k)

Z 3 Z
2 d k 2
h 0 (~
x)i = P 0 (k) = d ln k 0 (k),
(2⇡)3
3
2 k
0 (k) ⌘ P 0 (k) Variance per logarithmic
2⇡ 2 interval in k
Anisotropies: problem setup
A map of CMB temperature perturbations (CMB anisotropies) Θ = ΔT/T
can be decomposed into spherical harmonics:
X
⇥(n̂) = ⇥`m Y`m (n̂)
`m
Z
2 ⇤
⇥`m ⌘ d n̂ Y`m (n̂)⇥(n̂)

(equivalent of Fourier transforming on a sphere)

The angular power spectrum is defined as



h⇥`m ⇥`0 m0 i ⌘ ``0 mm0 C`

X 2` + 1 Z
2 `(` + 1)
h⇥ (n̂)i = C` ⇡ d ln ` C`
4⇡ 2⇡
`
Anisotropies: problem setup
Anisotropy power spectrum: variance of temperature
fluctuations as a function of angular scale

`(` + 1)
C`
2⇡
Anisotropies: problem setup
small perturbations => anisotropies are linearly related to initial conditions.
=> angular power spectrum is linearly related to the primordial power
spectrum:
Z
`(` + 1) 2
C` = d ln k T (k, `) 0 (k)
2⇡
depends on cosmological parameters

Estimation of the angular power spectrum allows to estimate (i) the


primordial power spectrum and (ii) other cosmological parameters
Anisotropies: problem setup
CMB Theory from Nucleosynthesis to Recombination 29

100 (a) Curvature (b) Dark Energy Density (c) Equation of


State
80

60

40

20
wDE

0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 -0.8 -0.6 -0.4 -0.2 0
10 100 1000 10 100 1000 10 100 1000
30 l l l

dark(a)energy.
Fig. 14. Curvature and 100 BaryonsGiven a fixed physical (b) Matter
scale for the acoustic peaks (fixed ⌦b h2 and
⌦m h2 ) the observed angular position of the peaks provides a measure of the angular diameter distance and the
parameters it depends on:80 curvature, dark energy density and dark energy equation of state. Changes at low
` multipoles are due to the decay of the gravitational potential after matter domination from the integrated
Sachs-Wolfe e↵ect.
60

the baryon-photon momentum density ratio


40
!✓ ◆
3 ⇢b ⌦b h2 a⇤
R⇤ ⌘ = 0.729 , (129)
4⇢ 0.024 10 3
a⇤
Hu 2008
20

and the expansion rate prior


0.02to
0.04 recombination
0.06 which is determined by the matter radiation ratio
0.1 0.2 0.3 0.4 0.5
10 100 1000 10 100 1000
! 1✓ ◆ 1 l
⇢r l ⌦m h 2 a⇤
Anisotropies: problem setup
small perturbations => anisotropies are linearly related to initial conditions.
=> angular power spectrum is linearly related to the primordial power
spectrum:
Z
`(` + 1) 2
C` = d ln k T (k, `) 0 (k)
2⇡
depends on cosmological parameters

Estimation of the angular power spectrum allows to estimate (i) the


primordial power spectrum and (ii) other cosmological parameters

=> Goal: compute T (k, `) as a function of cosmological parameters

The adequate formalism is relativistic perturbation theory. But we will


start by simple Newtonian approximations.
Preliminary: Thomson scattering
Main non-gravitational interaction of photons: Thomson
scattering with free electrons.

Thomson scattering cross section : σΤ = 6.65 x 10-25 cm2


Thomson scattering rate: ne c σΤ

ne: abundance of free electrons, ne = xe nH, where nH is the


total abundance of hydrogen and xe is the free-electron fraction

• As long as ne c σΤ >> H, photons and baryons form a


tightly coupled fluid.

• Once ne c σΤ ≲ H, photons propagate freely


Preliminary: Thomson scattering
Tomorrow we will discuss recombination theory and how to compute xe(z).
The calculation shows that last scattering occurs at z ~ 1100 (± 100)

=> observed photon temperature


anisotropies ≈ photon temperature on a
the last-scattering ``surface” at z ~ 1100

=> 2 steps:

(i) solve the photon-baryon fluid


equations at z ≳ 1100
z ⇡ 1100 ± 100
(ii) convert spatial fluctuations at
z~1100 into angular fluctuations.
Review: ideal fluid equations
• Fluid equations for non-relativistic gas, in a non-expanding Universe:

@⇢ ~ continuity (mass conservation)


+ r · (⇢~v ) = 0,
@t
@~v ~
rP
~ v=
+ (~v · r)~ ~
r momentum equation
@t ⇢
⇢ = ⇢ ⇥ (1 + ), ⇢ = constant, ⌧1
Now suppose
v / small quantity
= 0 + , ⌧ 0

linearize in δ, φ and v ˙+r~ · ~v = 0,


~ 2~
rP = cs r⇢ ˙~v + c2 r
~ = ~
r
s
Review: ideal fluid equations
Z
i~
F̃ (~k) = 3
d xe k·~
x
F (~x),
small k = large
Fourier transform Z
d3 k i~k·~x ~ wavelength
F (~x) = 3
e F̃ (k),
(2⇡)
˙+r~ · ~v = 0, ˙ + i~k · ~v = 0,
˙~v + c2 r
~ = ~
r ˙~v + c2 i~k = i~k
s
s
~ ! i~k
r ¨ + c2 k 2 = 2
s k
Poisson equation:
r2 = 4⇡G⇢, ¨ + c2 (k 2 2
kJ ) =0
s
2
r = 4⇡G⇢ p
4⇡G⇢
k 2 = 4⇡G⇢ kJ ⌘ Jeans scale
cs
Review: ideal fluid equations
p
¨ + c2 (k 2 2
kJ ) =0 kJ ⌘
4⇡G⇢
s
cs
Two different behaviors depending on whether k < kJ or k > kJ :
q
• If k < kJ , define ⌧ ⌘ cs kJ2 k2 ¨ = ⌧2
t/⌧ t/⌧
) = +e + e
=>for large scales k < kJ ``gravity wins” and perturbations grow
q
• If k > kJ , define ! ⌘ cs k2 kJ2 ¨= ! 2

) = A cos(!t) + B sin(!t)
=>for small scales k > kJ ``pressure wins” and perturbations oscillate
Sound waves
Consider the limit k >> kJ (i.e. gravity is negligible) ) ! ⇡ kcs
) (~k, t) = A cos(kcs t) + B sin(kcs t)
Suppose ˙ (~k, t = 0) = 0 ) B = 0
(~k, t = 0) = A
˙ (~k, t = 0) = kcs B ) (~k, t) = (~k) cos(kc t) 0 s

In 1-dimension, can easily show that


1 h i
(1d) (1d) (1d)
(x, t) = 0 (x cs t) + 0 (x + cs t)
2

Slightly more complicated but similar in 3-d: an initial


density “bump” propagates at the sound speed
Sound waves
���� cst =
0

���� 4
0.1 0.5 1 2

����
r (r, t)
����

-����

� � � � � �
r/cs
Photon-baryon acoustic oscillations
Expanding Universe: same thing, use comoving wavenumbers k and
conformal time η instead of physical scales and physical time.
dt
d⌘ =
a
Before last-scattering, photons and baryons are tightly coupled by
Thomson scattering and together form a fluid
1 2 1
c2s (photon)= c (P = ⇢ )
3 3
2 nb T T
cs (baryon) = ⇡ ⇠ 10 10 c2 negligible
n b mb mH
p p
Comoving 4⇡G⇢ 12⇡G⇢ 3 aH
kJ = a =a =p
Jeans scale: cs c 2 c
This is comparable to the Hubble scale aH, so our
Newtonian approximation breaks down anyway for k ≲ kJ
Photon-baryon acoustic oscillations
The equations we derived before hold for photons n

for the photon number density perturbation n

p
Assuming ˙ (⌘ = 0) = 0 (~k, ⌘) = ~
,0 (k) cos(kc⌘/ 3)

We define: z⇤ ⌘ 1100 ⌘⇤ ⌘ ⌘(z⇤ )

The sound horizon is the comoving distance travelled by a sound wave


from η = 0 to decoupling: p
c⌘⇤ / 3 ⇡ 280 Mpc
p
3 1
We define: k⇤ ⌘ ⇡ ⇡ 0.02 Mpc
c⌘⇤
Photon-baryon acoustic oscillations
p
p 3 1
(~k, ⌘) = ~
,0 k) cos(kc⌘/ 3)
( k⇤ ⌘ ⇡
c⌘⇤
⇡ 0.02 Mpc

(⌘, ~k)/ 0 (~k)


���
k = 2k⇤
���
δγ (�� �)

���
3
k = k⇤
2
-���
k = k⇤
-���
��� ��� ��� ��� ��� ��� ⌘/⌘⇤
�/����
Photon-baryon acoustic oscillations
p
p 3 1
(~k, ⌘) = ~
,0 (k) cos(kc⌘/ 3) k⇤ ⌘ ⇡
c⌘⇤
⇡ 0.02 Mpc

~ ~
[ (⌘, k)/ 0 (k)] 2

���
k = k⇤
���
k = 2k⇤
[δγ (�� �)]�

���

���

��� 3
k = k⇤
2
���
��� ��� ��� ��� ��� ��� ⌘/⌘⇤
�/����
Photon-baryon acoustic oscillations
Squared perturbation at the time of last scattering η*
~ ~
[ (⌘⇤ , k)/ 0 (k)] 2

���

���
[δγ (�� ���� )]�

���

���

���

���
� � � � � � k/k⇤
�/����
Diffusion damping

Photons are only an ideal fluid on scales >> mean-free path


1
mfp = (comoving)
ane T
On smaller scales they are a viscous fluid.
Diffusion damping
Drag force if velocities are
non-uniform:
d~v
⇠ ne c T (~v h~v imfp )
dt
0 @~v
~v (~x ) = ~v (~x) + xi i
@x
1 @ 2~v
+ xi xj i j
2 @x @x
0 1 2 2
) h~v (~x )imfp = ~v (~x) + mfp r ~v
5
d~v 2 2 2
) ⇠ ne c T mfp r ~v ⇠c mfp r ~v
dt
Diffusion damping
=> On small enough scales,
~
rP
photons are viscous:
~v˙ = ~
r +c mfp r
2
~v

2
The wave equation gets ¨ + c 2˙ c
mfp k + =0
a damping term: 3
 Z ⌘
WKB approximation: / exp i !d⌘ 0
!˙ ⌧ ! 2
2
c 2
2 2
) ! = k + c mfp k i!
3
kc 1 2
In the limit k λmfp<< 1 ! ⇡ ±p + i mfp ck
3 2
 Z
p 1 2 ⌘
) (~k, ⌘) = ~
,0 k) cos(k⌘c/ 3) exp
( k c mfp (⌘
0
)d⌘ 0
2
Diffusion damping
p k2 /kD
2
(~k, ⌘) = ~
,0 k) cos(k⌘c/ 3)e
(
Z ⌘ Z ⌘ 0
2 0 0 2 d⌘
kD ⇠ mfp (⌘ )d⌘ = mfp
⌘scat

mfp
⌘scat ⌘ : mean time between scattering events
c

1 dNscat 2 2
)
⌘scat
=
d⌘ 0
) kD ⇠ Nscat mfp

=> Perturbations are damped on scales smaller than the


characteristic scale photons can travel by random walk
Diffusion damping

=> Perturbations are damped on scales smaller than the


characteristic scale photons can travel by random walk
Diffusion damping
An accurate calculation accounting for baryon inertia and polarization gives
Z ⌘ 0 2 16
2 1 d⌘ R + + R)
15 (1 3⇢b
kD = R⌘
6 ane (1 + R) 2 4⇢
T

1
���� kD (z) (Mpc )
����
Numerically,
����
1
kD (⌘⇤ ) ⇡ 0.1 Mpc
����
⇡ 6 k⇤ ����

����
��� ��� ���� ���� ����

Diffusion damping
Squared perturbation at the time of last scattering η*
~ ~
[ (⌘⇤ , k)/ 0 (k)] 2

��� no damping
���

���

���

���
with damping
���
� � � � �
k/k⇤
Observed temperature anisotropies
T T line of sight
(n̂) = (⌘⇤ , r⇤ n̂) last
T obs T r⇤ scattering

surface
T 3 1
n /T ) =
T 3

Comoving wavelength λ δθ λ
corresponds to angle ✓=
r⇤

Equivalent of Fourier transforming for a 2-D map on the sky: spherical


harmonic transform. Perturbations with angular period δθ have multipole
2⇡ 2⇡
`= = kr⇤ k=

Temperature angular power spectrum
The angular power spectrum is the equivalent of the 3-d power spectrum:
for large l, the variance of temperature perturbations per ln(l) interval is

`(` + 1) 2
C` ⇡ T (⌘⇤ , k = `/r⇤ )
2⇡
 ✓ ◆ 2
`⌘⇤ 2`2 /`2D 2
⇡ cos p e 0 (`/r⇤ )
3r⇤

`D ⌘ r⇤ kD (r⇤ ) ~ 1500
Temperature angular power spectrum
 ✓ ◆
`(` + 1) `⌘⇤
2
2`2 /`2D 2
C` ⇡ cos p
3r⇤
e 0 (`/r⇤ )
2⇡
=> Angular power spectrum of ΔT/T has acoustic peaks at multipoles
p
3r⇤
`peak ⇡ n⇡ ⇡ 270 n n: integer
c⌘⇤

=> Its amplitude is damped by photon diffusion at l ≳ 1000


(Silk damping tail)

=> It is proportional to the primordial power spectrum


Temperature angular power spectrum

acoustic peaks

diff
usio
nd
am
pin
g ta
il
Summary

The CMB angular power spectrum quantifies the variance of temperature


fluctuations as a function of angular scale.

It is linearly related to the primordial power spectrum.

To derive the exact linear relationship requires a full general relativistic


treatment (Lecture 3).

Some qualitative features can be understood with Newtonian theory:


before last scattering, photons undergo acoustic oscillations, damped by
photon diffusion. After last-scattering, they propagate freely.
Program for next lectures

Tomorrow:

- Recombination theory: how exactly did the Universe become neutral


and when did CMB photons last scatter?

- CMB frequency spectrum: why is the CMB a blackbody, and what


would deviations from a blackbody tell us?

Friday:

- Relativistic perturbation theory for CMB anisotropies.


- Introduction to CMB polarization
- Introduction to parameter estimation
The Cosmic Microwave
Background: basic theory
Lecture 2
Yacine Ali-Haïmoud
Johns Hopkins University

Shanghai 2017 Spring School on Cosmology


Plan of this lecture
I- Recombination theory: evolution of the free-electron fraction
xe(z), determines the redshift at which photons last scatter.

II- The Boltzmann equation: describes the evolution of the photon


distribution. We will compute the Compton scattering collision operator,
needed for both CMB frequency spectrum calculation and CMB
anisotropy calculation.

III- The photon thermalization problem: why is the CMB


frequency spectrum a blackbody and why would deviations
from a blackbody (spectral distortions) tell us?

IV- Compton (Thomson) scattering collision operator in the


Boltzmann equation for temperature anisotropies.
I- Cosmological recombination

� ne
xe (z) ⌘
nH
�����

�����

�����

��� ��� z
��� ���� ����
Recombination history
2 3/2
Simplest assumption is xe (2⇡me T ) 13.6eV/T
that recombination = e
proceeds in equilibrium:
1 xe Saha h3 n H


�����
�����
�� [����]

�����
��-�
��-�
��-�
��� ��� ���� ���� ���� ���� ���� ����

Recombination history
Equilibrium only holds as long as recombination rate >> H
r
10 4 K
Recombination 13
↵rec ⇠ 5 ⇥ 10 cm s 3 1
coefficient: T
2 2
Rate of recombinations per unit volume: e p rec
n n ↵ = n H xe ↵rec

Rate of recombinations per nucleus: ẋe = nH ↵rec x2e

=> Equilibrium assumption only holds as long as

ẋe
= nH ↵rec xe H
xe
Recombination history
Saha
nH ↵rec xe
����
H

��

�����

�����
��� ���� ���� ���� ����

=> Equilibrium assumption fails at the very least at z ≲ 1000.
In fact it fails even earlier on.
Hydrogen atom

p+ + e-
13.6
n` eV
n2

2s 2p
13.6 eV
(n = 2, ` = 0) (n = 2, ` = 1)

10.2 eV

1s
Recombination theory
• Direct recombinations to p+ + e-
the ground state produce
photons with E > 13.6 eV n`

• Photoionization cross
section at threshold:
18 2
⇡ 6 ⇥ 10 cm
pi
2s 2p
• Time before photon is re-
absorbed by other atom:
treabs ⇠ 1/(nH x1s c pi ) 1s
4
✓ ◆3
10 1000
⇠ sec
x1s 1 + z => Direct recombinations to the ground
state lead to immediate re-ionization.
• As soon as x1s > 10-9, treabs < tH They are very inefficient
Recombination theory
• Recombinations proceed to the p+ + e-
excited states. Population of n=2 n`
with respect to n=1 suppressed by
10.2 eV/T 17
e ⇠ 10
=> No re-absorption problem 2s 2p
• From excited states, hydrogen
rapidly decays to n = 2. 1s

• ``Case-B” recombination
coefficient: => Recombination rate:
X
↵B = ↵n` 2
ẋe rec
= nH ↵ B x e
n 2
Recombination theory
p+ + e-
• Recombinations are
counterbalanced by n`
photoionizations from n = 2:
ẋe photo
= B x2

x2 ⌘
n2 2s 2p
nH

• Net recombination rate: 1s


2
ẋe = (nH ↵B xe B x2 )

=> Need to calculate the population


of the first excited state x2
Recombination theory
From 2p, allowed transition to ground state 2s 2p
(Lyman-α line). Lyman-α photons are almost
certainly re-absorbed by other atoms

Photons redshift: ⌫ / 1/a 1s

=> there is a small probability that a Lyman-α photon redshifts out


of resonance and escapes before being re-absorbed
3 2
Lyman-α absorption cross section: (⌫) = A21 (⌫)
8⇡
8 1 Z
A21 ⇡ 6 ⇥ 10 s (⌫)
(⌫)d⌫ = 1
Spontaneous decay rate line profile
Recombination theory
Optical depth for a photon emitted at ν: 2s 2p
Z
⌧ (⌫) = n1s (⌫(t))cdt
Z ⌫ 0

= 0
n1s (⌫ )c
d⌫ d⌫
= H⌫
1s
H⌫ 0 dt
0
Z ⌫
3 3 A21 0 0
⇡ n1s (⌫ )d⌫ (assuming line is narrow)
8⇡ H 0
⌧ (⌫)
Escape probability for photons emitted at ν: Pesc (⌫) = e
=> Average escape probability:
Z
8⇡ H 1 independent of
Pesc = d⌫ (⌫)Pesc (⌫) = 3 line profile
3 A21 n1s
Recombination theory
Net transition rate in Lyman-α, 2s 2p
accounting for re-absorptions:
⇣ ⌘
E21 /T
ẋ2 Ly↵
= A21 Pesc x2p 3x1s e
1s
E21 = 10.2 eV
1
A21 Pesc ⇠ 1 10 s

=> ``Forbidden” two-photon transitions 2s 2p


from 2s have a comparable rate:
1
⇤2 ⇡ 8.22 s

E21 /T
⌘ 1s
ẋ2 2
= ⇤2 x2s x1s e
Recombination theory
p+ + e-
The total rate of change of
n`
population of n = 2 is
2
ẋ2 = nH ↵ B x e B x2
3 ⇣ ⌘ 2s 2p
E21 /T
A21 Pesc x2 4x1s e
4
1 ⇣ ⌘
E21 /T
⇤2 x2 4x1s e
4 1s
Rate multiplying x2 is at least ~1-10 sec-1
This is >> H => x2 is in quasi-steady state: ẋ2 ⇡ 0
2 E21 /T
nH ↵ B x e + (3A21 Pesc + ⇤2 )x1s e
x2 = 3 1
B + 4 A 21 P esc + ⇤
4 2
Recombination theory
2
Final equations: ẋe = (nH ↵B xe B x2 )

2 E21 /T
nH ↵ B x e + (3A21 Pesc + ⇤2 )x1s e
x2 = 3 1
B + 4 A21 Pesc + 4 ⇤2

⇣ ⌘
2 E21 /T
ẋe = C nH ↵ B x e 4 B x1s e

=> 3 1
4 A21 Pesc + 4 ⇤2
C⌘ 3 1 x1s ⇡ 1 xe
B + 4 A21 Pesc + 4 ⇤2

This simple effective 3-level model was developed by


Peebles (1968) and Zeldovich, Kurt & Sunyaev (1968)
Recombination theory

�����
��

�����
Correct
calculation Saha
����� equilibrium

�� -�
��� ��� ��� ��� ���� ���� ���� ����

Recombination theory
The high-precision of Planck measurements required a
0.1% accuracy of the recombination history.

=> Modern developments:


• Account for stimulated recombinations and all radiative
transitions between excited states.

• Precise radiative transfer calculation in Lyman-α line

• Helium recombination

Modern recombination codes, with ~0.1% accuracy:


CosmoRec (Chluba & Thomas 2011)
HyRec (Ali-Haïmoud & Hirata 2011)
The visibility function
Thomson Scattering dPscat d⌧
= ne c T ⌘
probability per unit time : dt dt

Z t0
Thomson optical depth 0
⌧⌘ dt ne c T
between t and t0: t

Probability to not scatter ⌧


between t and t0:
e

=> probability to last scatter g(t) t ⌘ ne c e ⌧


t
T
between t and t + dt:

g(t) is called the visibility function


The visibility function
Visibility function in redshift (last- ne c T ⌧
g(z) ⌘ e
scattering probability per z-interval): (1 + z)H(z)
g(z)
���
max[g] peaks at z ⇡ 1080
��� the last-scattering ``surface”

���
has a finite width
��� z ⇠ 200

���

���
��� ��� ��� ��� ���� ���� ����

II- The Boltzmann equation and
the Compton scattering operator
Phase-space density
dN Number of particles per
F (t, ~x, p~) = 3 3 unit phase-space volume
d xd p
Example 1: consider electrons with Maxwell-Boltzmann velocity
distribution fe with temperature Te(t, x) and bulk velocity vb(t, x)

1 1 me (~v ~vb )2
fe (~v ) = 3/2
exp
(2⇡Te /me ) 2 Te

If the electron density is ne(t, x) their phase-space density is


1
Fe (t, ~x, p
~) = ne (t, ~x) 3 fe (~
p/me )
me
Phase-space density
dN Number of particles per
F (t, ~x, p~) = 3 3 unit phase-space volume
d xd p
Example 2: photons with a blackbody distribution at
temperature Tγ(t, x) have phase-space density
2 1
F (t, ~x, p~) = 3 p/T
h e 1
The photon phase-space density Fγ is proportional to the photon
occupation number fγ (number of photons per quantum state)
2
F = 3 ⇥f
h
2 polarizations, each quantum state occupies phase-space volume h3
Phase-space density and fluid moments
Photon phase-space distribution can also be described by the
(infinite number of) moments of the phase-space density Fγ:
Z
3
energy density ⇢ (t, ~x) = d p p F (t, ~x, p~)
Z
momentum density ~ (t, ~x) =
P 3
d p p~ F (t, ~x, p~)
(or energy flux)
Z i j
p p
stress tensor (j-th component T ij (t, ~x) = 3
d p F (t, ~x, p~)
of the flux of i-th component of p
momentum) etc…
1 ii 1
photon pressure: P = T = ⇢
3 3
ij 1 ij
anisotropic stress: ⌃ij = T P
3
Liouville’s theorem
In a non-expanding Universe, for non-interacting particles:
p~i
p
~=p
~i = const ~x = ~xi + (t ti )
m
px ΔxΔpx = constant

x => d3xd3p = constant

=> phase-space volumes are conserved along trajectories


dN
F (t, ~x, p~) = 3 3
d xd p
dN = constant => F is constant along particles trajectories
Liouville’s theorem
Remains true if particles evolve in a potential:
d~x p~ 0 p~
= , ~x = ~x + t ,
dt m => after Δt: m
d~p ~ p~0 = p~ t mr ~
= mr
dt
Jacobian of the transformation: x0 p0x = J x px ,
@x0 @x0
! ✓ ◆
t
@x @px 1
J = det @p0x @p0x
= det 2
@
m
= 1 + O( t)2
m t @x2 1
@x @px

d x px
) =0
dt
Remains true in expanding Universe and for massless particles.
Boltzmann equation
=> in the absence of non-gravitational interactions (collisions),
the phase-space density F is constant along particles trajectories

This is expressed by the collisionless Boltzmann equation:


dF @F d~x @F d~
p @F
⌘ + · + · =0
dt traj @t dt traj @~x dt traj @~p
Describes the evolution of non-interacting particles, like
neutrinos or dark matter.
Photons are collisional: they interact with electrons.
Their occupation number evolves according to

df @f d~x @f d~
p @f
⌘ + · + · = C[f ]
dt traj @t dt traj @~x dt traj @~p
Boltzmann equation
df @f d~x @f d~
p @f
⌘ + · + · = C[f ]
dt traj @t dt traj @~x dt traj @~p

We decompose fγ into a homogenous and isotropic part and a


small non-homogenous, anisotropic perturbation:
0
f (t, ~x, p~) = f (t, p) + f (t, ~x, p~)

• CMB spectrum: study the momentum (or energy or frequency)


dependence of fγ0

• CMB anisotropies: study the evolution of perturbations δfγ

The Compton-collision operator C[fγ] is relevant for both


Compton collision operator
p~ 0
p~ p~ p~0
0
0 n̂ ⌘ , n̂ ⌘ 0
~v p p
e ~v
in out
C[f ](~
p) = C[f ](~
p) C[f ](~
p)
0 0
p~ ! p~ p~ ! p~
0
Given p~, ~v , n̂ conservation of energy and momentum (4 equations)
) uniquely determine ~v 0 and p0
Z Z
out 3 dP
2 0 0
C[f ](~
p) = ne T d vfe (~v ) d n̂ 2 0 f (~
p)[1 + f (~
p )]
d n̂
scattering
rate electron velocity angular distribution stimulated
distribution of scattering scattering
Compton collision operator
C[f ](~ p)in
p) = C[f ](~ p)out
C[f ](~
Z Z
out 3 2 0 dP 0
C[f ](~
p) = ne T d vfe (~v ) d n̂ 2 0 f (~
p)[1 + f (~
p )]
d n̂
C[fγ]in is in principle an integral over p’ and v’.
p, ~v , n̂0 ), p0 and ~v 0 are entirely determined
given (~

=> we can write C[fγ]in as an integral over ~v and n̂0


Z Z
in 3 2 0 0 0 0
C[f ](~
p) = ne T d v d n̂ (~
p, ~v , n̂ )fe (~v )f (~
p )[1 + f (~
p)]

In principle χ can be determined from kinematics/ change of variable


considerations, but we can find it by detailed balance argument:
Compton collision operator
If photons and electrons are in thermal C[f T , f T ]in = C[f T , f T ]out
e e
equilibrium at the same temperature T,
dP
This must hold for any temperature T. ) =
dn̂0
Z Z
3 2 0 dP
C[f ](~
p) = n e T d v d n̂ 2 0
d n̂
n o
0 0 0
⇥ fe (~v )f (~p )[1 + f (~
p)] fe (~v )f (~
p)[1 + f (~
p )]


me v 2
Check: if fe (~v ) / exp
2T C[f ] = 0
⇣ ⌘ 1 then
and p) = ep/T
f (~ 1
III- The Photon
thermalization problem
1990ApJ...354L..37M
Intensity

Frequency
Baryon thermal equilibrium
Cross section for neutral-neutral collisions (the least frequent),
is σ ~ a02 ~ 10-16 cm2 (a0 = 1Å = Bohr radius).
Characteristic time between collisions:
3
✓ ◆1/2
1 100 cm 3000 K
tcoll = ⇠ 1 yr 16 cm2
nH v nH 10 T
Always much shorter than the characteristic expansion timescale
1
tcoll ⌧ tH ⇠ H
=> Baryons quickly achieve thermal equilibrium, i.e. have a
Maxwell-Boltzmann velocity distribution
 2
1 m(~v ~vb )
f (~v ) / exp
2 T
Photon equilibrium distributions

Bose-Einstein distribution is achieved if photons can efficiently


change their energies at constant total energy and photon number

2
dn 8⇡E 1
= µ :chemical potential
dE (hc)3 eE/T +µ 1

A blackbody is a Bose-Einstein distribution with µ = 0.

Achieved if photons are in addition efficiently absorbed and


emitted at all energies (i.e. photon number can be changed)
Photon thermalization processes
• Compton scattering e + $e +
Cannot change photon number, but can change photon energy

We will see that it efficiently establishes a


Bose-Einstein distribution at z ≳ 5x104

• Free-free (Bremsstrahlung) e +p$e +p+


0 00
• Double-Compton scattering e + $e + +

Can change photon number


We will see that they efficiently establishes a
blackbody distribution at z ≳ 2x106
Compton scattering
e + !e +
0 0
~v p~ ~v p~
0 0
Conservation of momentum: p~ p~ = me (~v ~v )
0 1 me 2 02
Conservation of energy: p p= (v v )
2 c
Combine and obtain, in the limit p << me c and v⌧c
0
p p p 1 0
= ⇡ ~v · (n̂ n̂) n̂ : direction of photon
p p c
p~ In this configuration
Makes sense ~v
intuitively: 0 p v
p~ e =2
p c
Compton scattering
e + !e +
0 0
~v p~ ~v p~
0
p p p 1 0
= ⇡ ~v · (n̂ n̂)
p p c
*✓ ◆2 +
p 1 2 2 Te
= hv ih(n̂0 n̂) i = 2
p 3 me

=> Compton scattering makes photons diffuse in energy space.


Compton scattering
*✓ ◆2 +
p Te
) =2 per scattering
p me c2
Photon momentum (or energy Eγ = pc) random-walks

2 Te2
h( p) i = Nscat ⇥ 2p
me c2
2
me c
Number of scatterings required to get Δp ~ p: Nscat ⇠
Te
1
Time between 2 scatterings: tscat ⇠
ne c T
me c 2 1
Time to change ΔEγ ~ Eγ : t ⇠ Nscat tscat ⇠
Te ne c T
Compton scattering
2
me c 1
Time to change ΔEγ ~ Eγ tTh ⇠
by Thomson scattering: T e ne c T

Time to change ΔEγ ~ Eγ 1


by Hubble expansion: tH ⇠ H
✓ ◆
4 2
tTh 5 ⇥ 10
Numerically: ⇡
tH 1+z

=> For z ≳ 5 x104 (t ≲ 300 yr after the Big Bang), Thomson


scattering efficiently redistributes photon energies
Compton collision operator
Z Z
3 2 0 dP
C[f ](~
p) = n e T d v d n̂ 2 0
d n̂
n o
0 0 0
⇥ fe (~v )f (~p )[1 + f (~
p)] fe (~v )f (~
p)[1 + f (~
p )]
We now consider the case where fe and fγ are isotropic: we
assume that electrons have no bulk velocity, and that

f (~
p) = f (p) isotropic = no dependence on n̂

0 Ee0 /Te (Ee Ee0 )/Te (p0 p)/Te


fe (v ) / e =e fe (v) = e fe (v)

(by conservation of energy)


Compton collision operator
Z Z
3 dP
2 0
C[f ](p) = ne T d v d n̂ 2 0 fe (v)
d n̂
n 0 o
(p p)/Te 0 0
⇥ e f (p )[1 + f (p)] f (p)[1 + f (p )]

(i) Taylor-expand {…} to second order in (p’- p):


n o 0
✓ 0
◆2
0 p p 0 00 p p
...... = function(f , f ) + function(f , f , f )
p p

0
(ii) Express (p’- p) in p p 0
= ~v · (n̂ n̂) + ...
terms of p, ~
v , n̂ 0
: p
dP
(iii) Find 2 0 in the frame where the electron has velocity ~v
d n̂
Compton collision operator
At the end of the calculation:
⇢ ✓ ◆
ne T d 4 df
C[f ](p) ⇡ p Te + f (p)[1 + f (p)]
me p2 dp dp

(i) detailed balance: C[fγ] = 0 if fγ is a blackbody at temperature Te

(ii) conserves photon number:


Z
dn 2 2
Compt
= 3 4⇡p dpC[f ](p) = 0
dt h
This is called the Kompaneets collision operator
It is a particular instance of a Fokker-Planck operator
The Fokker-Planck equation
Consider a random process X, changed by stochastic collisions
occurring with rate Γ(X), such that at every collision,

h Xi ⌘ d(X) ⌧ X

h( X)2 i h Xi2 ⌘ D(X) ⌧ X 2

The probability distribution of X, N(X) evolves according to


the following diffusion equation:

@N 2
dN @ 1 @
= [ d N (X)] + [ D N (X)]
@tdt @X 2 @X 2
The Fokker-Planck equation
Example: Brownian motion

@N @ 1 2 @N
= XN (X) + 0
(X) = constant @t @X 2 @X
d(X) = X equilibrium distribution:
2
D(X) = 0 1 X 2 / 02
Neq (X) = p e
⇡ 0
If N(X) is initially a Dirac function at Xi, the solution at time t is

1 (X X(t))2
N (t, X) = p exp
⇡ (t) (t)2
t/⌧
X(t) = Xi e
⇣ ⌘ => the distribution reaches the
2 2 2t/⌧
(t) = 0 1 e equilibrium of the collision
operator in a timescale t ~ τ
⌧ = 1/( )
The Fokker-Planck equation

t/⌧ = 0
� N (X)
t/⌧ = 0.01
� => Equilibrium distribution
reached on timescale τ

t/⌧ = 0.1
� t/⌧ = 1
t/⌧ = 1
X

-� -� � � � 0
Compton collision operator
The Compton-collision (Kompaneets) operator is a
particular instance of a Fokker-Planck operator.
X!p ! ne T
2 2 2 2 2 2Te
N (X) ! 3 4⇡p f (p) h X i ! h( p) i = p
h me
2
dN @ 1 @
= [ d N (X)] + 2
[ D N (X)]
dt @X 2 @X
⇢ ✓ ◆
ne T d 4 df
C[f ](p) ⇡ p Te + f (p)[1 + f (p)]
me p2 dp dp
✓ ◆ 1
T
characteristic timescale: ⌧Compt ⇠ ne T
me
Compton collision operator
✓ ◆ 1
=> The photon distribution reaches ⌧ T
Compt ⇠ ne T
equilibrium on a timescale me
✓ ◆
4 2
⌧Compt 5 ⇥ 10

tH 1+z
⇢ ✓ ◆
Equilibria of the d 4 df
p Te + f (p)[1 + f (p)] =0
Kompaneets operator: dp dp
1
) f (p) = , µ = constant
ep/Te +µ 1

=> For z ≳ 5 x104 , Compton scattering establishes a Bose-Einstein


distribution at temperature Te, regardless of initial conditions.
Chemical potential distortion
2
dn 8⇡E 1
= 3
Suppose µ << 1
dE (hc) eE/T +µ 1
Z
dn 3
Number density of photons: n = dE / T (1 1.37µ)
dE
Z
dn 4
Energy density of photons: ⇢ = dEE / T (1 1.11µ)
dE

Take log and differentiate: n T


=3 1.37 µ
n T
⇢ T
=4 1.11 µ
⇢ T
Chemical potential distortion
n T ⇢ T
=3 1.37 µ =4 1.11 µ
n T ⇢ T
✓ ◆
⇢ 4 n
Solve for Δµ: µ = 1.40
⇢ 3 n
=> if energy and photons are injected at z ≳ 5 x104 with rate
⇢˙ , ṅ
get a Bose-Einstein distribution with chemical potential
Z ✓ ◆
⇢˙ 4 ṅ
µtot = 1.40 dt
z&5⇥104 ⇢ 3 n

Now we study what happens at z ≲ 5 x104


Compton-y distortion
If some process injects photons, one has to solve the equation
df df df
= +
dt dt Compt dt other

In general, has to be solved numerically, depends on injection process


Consider the special case where some process heats the baryons
(so Te ≠ Tγ), but does not inject photons.
Suppose the photon occupation number is close to a blackbody
⇣ ⌘ 1
T
f =f + f T
f ⌘ e p/T
1 T / 1/a

) C[f ] ⇡ C[f T ] at z ≲ 5 x104

(at z ≲ 5 x104 , one can neglect the correction ΔC[fγ] << H Δfγ )
Compton-y distortion
Inserting the blackbody distribution in the Compton
collision operator, we find:

T Te T
C[f ] ⇡ C[f ] = ne T Y(p/T )
me
✓ ◆
1 d 4 ex
Y(x) ⌘ x Compton-y distortion
x2 dx (ex 1)2

Integrate the Boltzmann equation along photon


trajectories (p/Tγ = constant) to obtain the distortion:
Z t 
0 Te T
f = y Y(p/T ) y⌘ dt ne T
me t0
Compton-y distortion
p3 f (p) (arbitrary units)
���
���
���
y = 0.05
Z t 
��� y = 0.025 0 Te T
y⌘ dt ne T
��� y=0 me t0
���
���
���
� � � � � ��
p/T
The Compton-y distortion is a universal shape resulting
from Te ≠ Tγ and inefficient Comptonization at z ≲ 5 x104
Now we estimate the y-parameter
Compton-y distortion
Rate of change of photon energy density by Compton scattering:
Z
d⇢ 2 3 Te T
= 3 d p p C[f ] ⇡ 4ne T⇢
dt Compt h me

Total energy density of photons + electrons + baryons is conserved by


Compton + Coulomb scattering. Baryon number is also conserved.

3 dTe d⇢
) nb =
2 dt Comp dt Comp

dTe 8 ne ⇢
) = T (T Te )
dt Comp 3 nb me
Compton-y distortion
Now suppose some process heats the baryons with rate ⇢˙
3 T Te 3
nb Ṫe = 4ne T⇢ + ⇢˙ nb 2HTe
2 me 2
adiabatic cooling:
nb H for z & 200 2 2
Te / ve /a
At z ≳ 200, baryon temperature can be obtained in quasi-steady-state:
Te T 1 ⇢˙ (z ≳ 200)
) ne T ⇡
me 4 ⇢
=> If baryons are heated at 200 ≲ z ≲ 5 x104, Compton-y parameter is
Z Z
Te T 1 ⇢˙
y= dtne T ⇡ dt
me 4 ⇢
Bremsstrahlung and Double-Compton
Free-free (Bremsstrahlung)
e +p$e +p+
Classical picture: electrons are e-
accelerated by protons
=> Electric dipole radiation p

Double-Compton scattering
0 00
e + $e + +
Classical picture (Melrose 1972): electrons are accelerated by
electromagnetic wave (=photons). Linear response: Compton.
Non-linear response: double-Compton
Bremsstrahlung and Double-Compton
Br and DC collision terms in the photon Boltzmann equation:

1
C[f ]Br,DC (p) = K(p, Te , T ) p/T f
e e 1
They both tend to establish a blackbody distribution at temperature Te
✓ ◆
We saw before dµ ⇢˙ 4 ṅ
⇡ 1.4
that at z ≲ 5 x104 dt ⇢ 3 n
Bremsstrahlung and DC can change photon number and have
to be included in the right-hand side. One can show that
✓ ◆
⇢˙ 4 ṅ
1.4 ⇡ µ
⇢ 3 n Br,DC
Bremsstrahlung and Double-Compton
���
Blackbody at Te
���
���
��� Bose-Einstein at
Te with µ = 0.2
���
���
���
���
� � � � � ��
For µ > 0 there is a deficit of photons with respect to blackbody
=> creation of photons by Bremsstrahlung and DC emission
Bremsstrahlung and Double-Compton
If some other process injects energy/photons, net evolution of µ is
✓ ◆
dµ ⇢˙ 4 ṅ
= 1.4 µ
dt ⇢ 3 n other

=> When Γ >> H, the chemical potential is efficiently erased by


Bremsstrahlung and DC emission.
✓ ◆
6 5/2
2 ⇥ 10
Numerically: ⇡
H 1+z
=> Bremsstrahlung and DC (in combination with Compton
scattering) erase any spectral distortion for energy/photon
injection at z ≳ 2 x106.
Recap
Given some process heating the baryons and/or injecting photons,
photon distribution is obtained by solving the Boltzmann equation

⇢ ✓ ◆
df ne T 1 d 4 @f Energy diffusion by
= p Te + f [1 + f ] Compton scattering
dt me p2 dp @p

1 Bremsstrahlung and
+ KBr,DC (p, Te , T ) p/T f
e e 1 Double-Compton
+ f˙ [other processes] emission/absorption

Must be solved simultaneously with the heat equation for baryons:

3 Te T 3
nb Ṫe = 4ne T⇢ nb 2HTe + ⇢˙ b
2 me 2
Recap
In general this must be solved numerically… Green’s function 3
5

temperature-shift
4 5
µ-distortion at zh ~ 3 x 10
W m Hz sr ]

y-distortion
-1

t
hif
-1

s
re
atu
2
-2

per
te m
1
-18
Gth(ν, zh, 0) [ 10

on
y-distorti
0

µ-d
-1 isto
rtio
n
-2

Chluba 2013
-3
1 10 100 1000
ν [GHz]
Figure 1. Numerical results for the Green’s function of the cosmological thermalization problem for various heating redshifts, zh ∈ [103 , 5 × 106 ]. Energy
… but simple solutions exist in some limiting cases
release at very high redshifts causes an increase in the effective temperature of the photon field, while at low redshifts photons only partially up-scatter, creating
a y-distortion. Around zh ≃ 3 × 105 a pure µ-distortion is created. All intermediate stages are roughly (precision ≃ 10% − 30%) represented by a superposition
of these extreme cases, however, the small residuals provide a principle possibility to distinguish different thermal histories at redshifts 104 ! z ! 3 × 105 .
Recap
• At z ≳ 2 x106, fγ is a blackbody spectrum, no matter what
energy/photon injection is at work.

• At 5 x104 ≲ z ≲ 2 x106, any energy or photon injection results


in a chemical potential (µ-type) distortion
Z ✓ ◆
⇢˙ 4 ṅ (5x104 ≲ z ≲ 2 x106)
µ ⇡ 1.40 dt
⇢ 3 n
• At z ≲ 5 x104, heating of the baryons results in a Compton-y
distortion. Compton-y parameter:
Z Z
Te T 1 ⇢˙
y= dtne T ⇡ dt (200 ≲ z ≲ 5 x104)
me 4 ⇢
Spectral distortions as a probe of new physics

COBE FIRAS limits to spectral distortions:

|y| < 1.5 x 10-5 , |µ| < 9 x 10-5

Proposed experiments [PIXIE, PRISM] will probe spectral


distortions at the level of µ, y ~ 10-8

=> What are the characteristic µ, y expected in the standard


cosmological scenario?

=> What could we learn from detecting µ, y ~ 10-8 ?


Adiabatic cooling of baryons
If left to cool adiabatically in the expanding Universe,
baryon temperature
2/3 2
Te ad
/⇢ /a

However, until z ~ 200, Compton scattering maintains Te ≈ Tγ ∝ a-1

=> Heat transferred from photons to baryons:

3 3 3
⇢˙ b = nb (Ṫe Ṫe ad ) = nb (2HTe HTe ) = nb HTe
2 2 2
3
) ⇢˙ ⇡ nb HT
2
Adiabatic cooling of baryons
3
⇢˙ ⇡ nb HT
2
Z 2⇥106 Z 2⇥106
d ln(1 + z) ⇢˙ nb T
µ⇡ 1.4 = 2.1 d ln(1 + z)
5⇥104 H ⇢ 5⇥104 ⇢
Z 5⇥104
3 nb T
y⇡ d ln(1 + z)
8 200 ⇢
nb 10
⇢ ⇡ 2.7 n T ⇠ 10
n

µ ≈ - 3x10-9 y ≈ - 5x10-10
Dissipation of acoustic waves
Energy density of photon acoustic waves:
Z
d3 k
⇢wave = ⇢ hv 2 i = ⇢ c2s h 2 i = ⇢ c2s P (k)
(2⇡)3
Recall from previous lecture: k2 /kD
2
(k, t) = 0 (k) cos(kcs t)e

1 2k2 /kD
2
) P (k) ⇡ P 0 (k)e
2
=> Rate of energy deposition due to the dissipation of acoustic
waves by photon diffusion (cs2 = 1/3):
Z
⇢˙ ⇢˙ wave 1 d d3 k 2k2 /kD
2
= = P (k)e
⇢ ⇢ 6 dt (2⇡)3 0

"Z #
kD (t)
1 d d3 k
⇡ 3
P 0 (k)
6 dt (2⇡)
Dissipation of acoustic waves
Z tf Z kD (ti ) 3
⇢˙ 1 d k
) dt ⇡ 3
P 0 (k)
ti ⇢ 6 kD (tf ) (2⇡)

If the power spectrum is approximately scale-invariant:


Z tf D E ✓ ◆
⇢˙ 2 kD (ti ) 9
) dt ⇠ ( T /T ) log ⇠ 10
ti ⇢ kD (tf )

However, we do not have a direct measurement of the primordial


power spectrum beyond the diffusion scale at z ~ 1000
Dissipation of acoustic waves

1
kD (z⇤ ) ⇠ 0.1 Mpc

diff
usio
nd
am
pin
g ta
il
Dissipation of acoustic waves
Z tf Z kD (ti ) 3
⇢˙ 1 d k
) dt ⇡ 3
P 0 (k)
ti ⇢ 6 kD (tf ) (2⇡)

If the power spectrum is approximately scale-invariant:


Z tf D E ✓ ◆
⇢˙ 2 kD (ti ) 9
) dt ⇠ ( T /T ) log ⇠ 10
ti ⇢ kD (tf )

However, we do not have a direct measurement of the primordial


power spectrum beyond the diffusion scale at z ~ 1000

=> Spectral distortions can be used to set limits on the power


spectrum at very small scales (and also on non-gaussianity)
Dissipation of acoustic waves
Z tf Z kD (ti ) 3
⇢˙ 1 d k
) dt ⇡ 3
P 0 (k) 2
ti ⇢ 6 kD (tf ) (2⇡)
not
en Pajer & Zaldarriaga 2012
his
he 2.0
pli-
9
DRHkL ¥10

ne
ti- 1.5
2

on
cal
CMBêLSS m kD (z = 2 ⇥ 106 )
mic 1.0 kD (z = 5 ⇥ 104 )
ese
ain
ic- 10-4 0.01 1 100 104
k Mpc
Recombination radiation
p+ + e-
n`

2s 2p

1s

As hydrogen recombines, a few photons per atom are emitted at


z~1000 << 5x104
=> Compton scattering no longer redistributes photon energies
efficiently.

No simple analytic solution: calculate the distortion numerically.


Recombination radiation
10-29
0.1 1.0 10.0 100.0 1000.0
The resulting spectral distortion(GHz) appears as broad lines due the
finite
FIG. 4. time(notextent
Bound-bound of!recombination.
including 2s 1s and Ly-↵ photons) and Overall amplitude:
free-bound spectra for various values of the cuto↵
principal quantum number nmax .
nH 9
few ⇥ ⇠ 10
n
10-26
bound-bound
free-bound
2s-1s
Ly-
total

10-27
I (J m-2 s-1 Hz-1 sr-1)

10-28

-29
Ali-Haimoud 2013
10
0.1 1.0 10.0 100.0 1000.0
(GHz)
Sunyaev-Zeldovich effect
So far we have assumed Te -Tγ << Tγ and focused on z ≳ 200.

The Kompaneets equation also describes the Sunyaev-


Zeldovich effect at low-redshift.

In galaxy clusters the gas has temperature Te ~107-108 K >> Tγ

cluster
CMB
Te

y-parameter obtained by integrating along cluster’s line of sight:


Z
Te 6 4 Comparable to
ycluster = d` ne T ⇠ 10 10 CMB anisotropy
me
Sunyaev-Zeldovich effect
Converting photon energy into frequency as observed today:

I⌫ (arbitrary units)
��� ���
217 GHz
��� ���

��� ���

��� ���

�� ��� y=0

� ��� ��� ��� ��� ��� ⌫(GHz)
Sunyaev-Zeldovich effect 8 Planck Collaboration: T

I⌫
� ��� 217 GHz
� ���

� ���

� ���

� ��� y=0

� ��� ��� ��� ��� ���
⌫(GHz)

Cluster appear as cold spots


for ν < 217 GHz and hot
spots for ν > 217 GHz
Sunyaev-Zeldovich effect
8 Planck Collaboration:

Unresolved clusters and hot


intergalactic gas
=> global y ~ 10-6 (Hill et al 2016)

=> y-distortion is not a ``clean”


probe of early-Universe physics.

=> but µ ≳ 10-9 would be an indication


of new physics in the early Universe!

Figure 6. From left to right and top to bottom: observed stack


intensity maps at 30, 44, 70, 100, 143, 217, 353, 545, 857, 30
5000, and 12 000 GHz at the positions of confirmed SZ cl
Example: decaying new particles
Consider a particle with mass mX and number density nX decaying with rate

d(a3 nX ) a 3 nX
= tX : lifetime of the particle
dt tX
3
(a nX )i t/tX
The solution for the density is nX = 3
e
a
Each decay releases mX c2 of energy. Suppose all of the annihilation
products can be deposited in the plasma (e.g. photons, electrons/
positrons, not neutrinos):
1
⇢˙ = mX c2 nX
tX
Z tfµ t/tX
2 3 dt e
µ = 1.4 mX c ai nXi
tiµ tX a 3 ⇢
Example: decaying new particles
74 CHAPTER 3. THERMALIZATION AND
=> Given FIRAS limits, one can set limits on the abundance of
new particles as a function of their mass and lifetime:

Hu & Silk 1993


posure any mo
action cross-section of dark matter particles with nucleons. tuations are exp
Example: interacting light dark matter
While this one-dimensional method does not rely on any as- for 10,000 Mo
sumption on the background, it exploits differences between background mo
the measured (see figure 6) and the expected energy spec- contour is shad
There are
trumstringent
(see sectionlimits
8). on dark matter interactions with In CRESST
nuclei, but only for masses ≳ GeV sorber crystals
more, we expe
Firstly more lig
light detector c
hanced energy
the background
ture an upgrad
with absorber c
icantly lower l
Combining the
tion power, a d
pected.
In this lette
the key require
particles of O(
progress explor
Fig. 8 Parameter space for elastic spin-independent dark matter-
Example: interacting light dark matter
If a dark matter particle can scatter elastically with either photons or
baryons, it exchanges heat with them.

Heat exchange is efficient until redshift zdec which depends on (σ, mDM)

Same process as for adiabatic cooling of baryons (for baryons zdec = 200)
Z 2⇥106
nDM
µ⇡ 2.1 d ln(1 + z)
n zdec
⇢DM
nDM = => Lighter DM particles result in larger distortions
mDM
=> Given limits on µ, one can set upper bounds on DM-photon or
DM-baryon cross section, as a function of DM mass.
Example: interacting light dark matter
FIRAS PIXIE [proposed, Kogut et al. 2011]

- DM-proton
] [cm2]
( ) [section

DM-electron
Maximum cross

tellites
Way sa
-
DM-photon Milky etectio n
Direct d
-
Ali-Haïmoud, Chluba & Kamionkowski 2015

Dark matter[ mass][MeV]


The Cosmic Microwave
Background: basic theory
Lecture 3
Yacine Ali-Haïmoud
Johns Hopkins University

Shanghai 2017 Spring School on Cosmology


Plan of this lecture

I- Compton scattering collision term for anisotropies

II- Relativistic perturbation theory for CMB anisotropies

III- Introduction to CMB polarization

IV- Introduction to parameter estimation


Compton scattering operator: anisotropic part
Full Compton scattering collision operator:
Z Z
3 2 0 dP
C[f ](~
p) = ne T d v d n̂ 2 0
d n̂
n o
⇥ fe (~v 0 )f (~
p 0 )[1 + f (~
p)] fe (~v )f (~ p 0 )]
p)[1 + f (~


Baryons have a me (~v ~vb )2
fe (~v ) / exp
bulk velocity: 2Te

We first compute C[fγ] in the frame where baryons are at rest.

in the baryon rest-frame, (p0 p)/Te (b)


fe(b) (v 0 ) = e fe (v)
fe is isotropic as before:
Compton scattering operator: anisotropic part
Z Z
(b) 3 2 0dP (b)
C[f ] (~
p) = n e T d v d n̂ 2 0 fe (v)
d n̂
n 0 o
p 0 )[1 + f (b) (~
⇥ e(p p)/Te f (b) (~ p)] f (b) (~ p 0 )]
p)[1 + f (b) (~

Earlier we Taylor-expanded in (p’-p). p~ ! (p, n̂)


If fγ is isotropic, the zero-th order term 0 0 0 0
vanishes. But not if fγ is anisotropic. p
~ ! (p , n̂ ) ⇡ (p, n̂ )
Z Z
2 0 dP
C[f ] (p, n̂) = ne T d v d n̂ 2 0 fe(b) (v)
(b) 3
d n̂
n o
⇥ f (b) (p, n̂0 )[1 + f (b) (p, n̂)] f (b) (p, n̂)[1 + f (b) (p, n̂0 )] + O(p0 p)

We see that (i) the stimulated scattering terms cancel


Z out and (ii)
nothing depends on electron velocities so
d3 vfe (v) ! 1
Compton scattering operator: anisotropic part
Z n o
dP
C[f ](b) (p, n̂) = ne T d2 n̂0 2 0 f (b) (p, n̂0 ) f (b) (p, n̂)
d n̂
This is a perturbation: is vanishes if fγ = fγ0(p) is isotropic.
Re-express things in terms of fγ in the original frame:
(b) @f
f (~ p + p~vb ) ⇡ f (~
p) = f (~ p) + p~vb ·
@~p
~vb and the anisotropic part of f are perturbations

=> To first order in the perturbations (fγ0(p) = unperturbed part of fγ ):


0 0
@f @f
f (b) (~ p) + p~vb ·
p) = f (~ p) + p(~vb · n̂)
= f (~
@~p @p
Compton scattering operator: anisotropic part

(Z )
2 0 dP
@f 0
C[f ](p, n̂) = ne T d n̂ 2 0 f (p, n̂0 ) f (p, n̂) (~vb · n̂)p
d n̂ @p

n̂ n̂0
Compton scattering operator: anisotropic part

(Z )
2 0 dP
@f 0
C[f ](p, n̂) = ne T d n̂ 2 0 f (p, n̂0 ) f (p, n̂) (~vb · n̂)p
d n̂ @p


Compton scattering operator: anisotropic part

(Z )
2 0 dP
@f 0
C[f ](p, n̂) = ne T d n̂ 2 0 f (p, n̂0 ) f (p, n̂) (~vb · n̂)p
d n̂ @p

~vb

Even if photon distribution is isotropic, scattered radiation


is beamed in the direction of electron’s motion.
Compton scattering operator: anisotropic part
(Z )
0
dP @f
C[f ](p, n̂) = ne T d n̂ 2 0 f (p, n̂0 )
2 0
f (p, n̂) (~vb · n̂)p
d n̂ @p

dP 3 ⇥ 0 2

2 0
= 1 + (n̂ · n̂ )
d n̂ 16⇡

Photon momentum density:


Z Z Z
~ 2 3 2
P ⌘ 3 d p p~ f (~
p) = 3 p2 dp d2 n̂ p n̂ f (p, n̂)
h h
~ Z Z
dP 2
= 3 p2 dp d2 n̂ p n̂ C[f ](p, n̂)
dt Comp h
( )
~ 4
= ne T 0 P + ⇢ ~vb
3
Compton scattering operator: anisotropic part
~ ✓ ◆
dP 4 ~
= ne T ⇢ ~vb P
dt Comp 3
~ = ~0
~v : velocity with respect to the frame where P
~ 4
can show that P = ⇢ ~v
3
~
dP 4
= ⇢ ne T (~vb ~v )
dt Comp 3
Compton + Coulomb scattering conserve the total photon +
electron + baryon momentum, as well as baryon density.

d~vb 4⇢
= ne T (~v ~vb )
dt Comp 3 ⇢b
Anisotropic Compton scattering: recap
(Z )
2 0 dP
@f 0
C[f ](p, n̂) = ne T d n̂ 2 0 f (p, n̂0 ) f (p, n̂) (~vb · n̂)p
d n̂ @p

dP 3 ⇥ 0 2

2 0
= 1 + (n̂ · n̂ )
d n̂ 16⇡

d~vb 4⇢
= ne T (~v ~vb )
dt Comp 3 ⇢b
d~v
= ne T (~vb ~v )
dt Comp

Baryons and photons exert a drag force on


one another through Compton scattering
Metric and gauge

The Universe is approximately homogenous and isotropic:


there exists coordinates (x0 = ⌘, xi ) in which the metric is

2 2
⇥ 2 2 µ ⌫

ds = a (⌘) d⌘ + d~x + hµ⌫ dx dx , |hµ⌫ | = O(✏) ⌧ 1

and the stress-energy tensor of any species takes the form

✓ ◆
µ ⇢(⌘) 0
T ⌫ = + T µ⌫, | T µ ⌫ /⇢| = O(✏) ⌧ 1
0 P (⌘) ij
Metric and gauge
Such coordinates are not unique: if we make a small
coordinate (or gauge) transformation

@⇠ µ
x̃µ = xµ + ⇠ µ (x), with ⌫
= O(✏) ⌧ 1
@x

the metric still has the same form with a di↵erent


h̃µ⌫ = O(✏), and similarly for the stress-energy tensor.

Non-unicity of coordinates (or gauge) ) quantities like the


density or velocity are gauge-dependent.

The observables we want to predict have to be


gauge-independent! Multipoles ⇥`m with ` 2 are
gauge-independent.
Metric and gauge
) we are free to choose the most convenient gauge to
compute the observable ⇥`m , ` 2

4 coordinates
) We can set up to 4 (of the 10) components of hµ⌫ to 0

Conformal Newtonian gauge:

ds2 = a2 (1 + 2 )d⌘ 2 + Bi d⌘dxi + [(1 2 ) ij + ij ]dx i


dx j
,

@i Bi = 0, ii = 0, @i ij =0
6 independent components: 2 scalar modes and , 1
vector mode Bi (2 components), 1 tensor mode (primordial
gravitational wave) ij (2 components).
Metric and gauge
To linear order in small perturbations:
• Scalar, vector (i.e. divergence-free) and tensor
(i.e. trace-free, divergence-free) modes are not coupled to
each other
• Vector mode Bi can be shown to decay. We will assume
Bi = 0
• Tensor mode is gauge-independent. If tensor modes
vanish initially, remain 0 at all times. We will assume
ij = 0 for now.

) We work with the metric

ds2 = gµ⌫ dxµ dx⌫ = a2 (⌘) (1 + 2 )d⌘ 2 + (1 2 )d~x2


Ideal Fluid equations: Newtonian limit
Newtonian fluid equations in a non-expanding Universe:
@⇢ ~
+ r · (⇢~v ) = 0 [mass conservation]
@t
@~v ~
rP F~
~ v+
+ (~v · r)~ +r ~ = [momentum equation],
@t ⇢ ⇢
~ external force per unit volume.
F:
Write ⇢ = ⇢(1 + ), ⌧ 1, ~v = O( ) and linearize:

@ @~v r~ P
~ · ~v = 0,
+r + ~ = 0.
+r
@t @t ⇢
In Fourier space:
@ @~v P
+ i~k · r
~ · ~v = 0, ~
+ ik + i~k = 0.
@t @t ⇢
Relativistic ideal fluid equations
Relativistic generalization: conservation of the
stress-energy tensor:
µ⌫
X
µ⌫ µ
r⌫ Ttotal = 0, r⌫ Ts = Fs , Fsµ = 0.
s

Fsµ accounts for energy and momentum exchange between


di↵erent fluids.
Stress tensor of an ideal fluid: Tµ⌫ = (⇢ + P )uµ u⌫ + P gµ⌫ .
Perturbed stress tensor for a non-relativistic ideal fluid
(v ⌧ c, P ⌧ ⇢):
✓ ◆
µ ⇢(⌘)(1 + ) ⇢vi
T⌫= + O(v 2 , P/⇢)
⇢vi (P (⌘) + P ) ij
Relativistic ideal fluid equations
Relativistic fluid equations (no energy exchange F 0 = 0):

@ 1 @ P ~
F
( 3 ) + i~k · ~v = 0, (a~v ) + i~k + i~k =
@⌘ a @⌘ ⇢ ⇢

where ~k is a comoving wavenumber.


• Simple explanation for the continuity equation: if ~v = ~0,
particles remain at the same comoving coordinate )
number of particles per comoving volume is constant.
✓ 3 ◆ ✓ 3 ◆
dN dN d xphys d xphys
constant = m 3 = m 3 3
=⇢
dx d xphys dx d3 x
dxphys = a(1 )dx ) ⇢a3 (1 3 ) = const
) ⇢a3 (1 + 3 ) = const if ~v = ~0.
Relativistic ideal fluid equations

• Momentum equation: first term accounts for ~v / 1/a for


freely-moving massive particles in a homogenous expanding
Universe.
) Fluid equations for baryons (Ẋ ⌘ @X/@⌘)

˙b 3 ˙ + i~k · ~vb = 0,
˙~vb + aH~vb + i~k Pb + i~k 4⇢
= ane T (~
v ~vb )
⇢b 3⇢b
Null geodesics

To write the Boltzmann equation for photons we need to


compute photon geodesics:

dpµ µ p↵ p µ1 µ⌫
= ↵ 0
, = g (@↵ g⌫ + @ g↵⌫ @⌫ g↵ ) .

d⌘ p 2
p p
define q ⌘ a pi pi = a p0 p0 = a2 (1 + )p0 ,
Observers with fixed xi have 4-velocity
uµ = ( a(1 + ), 0, 0, 0). They measure energy

µ 0 q
Eobs = uµ p = u0 p = ) q ⌘ aEobs
a
) q = energy measured by comoving observers at a = 1.
Null geodesics
Also define
i
p
n̂i ⌘ p , i j
ij n̂ = 1

j k
jk p p

Null geodesics:

dxi
= n̂i + O(✏),
d⌘
dq
= ( ˙ ni @i )q + O(✏2 ),
d⌘
dn̂i
= O(✏)
d⌘

We will see that we don’t need to explicitly compute


dxi /d⌘ and dn̂i /d⌘.
Photon Boltzmann equation
The photon Boltzmann equation takes the form
df df
=a = aC[f ]
d⌘ traj dt traj

We are free to use any coordinates for pi . We choose (q, n̂i ):

df @f i @f dq @f dn̂i @f
= +n̂ + +
d⌘ traj @⌘ i
x ,q,n̂ i @xi ⌘,q,n̂ i d⌘ @q i
⌘,x ,n̂ i d⌘ @ n̂i ⌘,xi ,q

In unperturbed homogeneous and isotropic Universe,


f = f (⌘, q) does not depend on n̂i ) last term is O(✏2 )

df @f i @f ˙ @f
) = + n̂ + ( n̂i @i )q = aC[f ]
d⌘ traj @⌘ @xi @q
Photon Boltzmann equation
Recall the Compton collision term:
@ f i@ f ˙ @f
+ n̂ i
+( n̂i @i )q = ane T
@⌘ @x @q
(Z )
2 0 dP 0 @f
⇥ d n̂ 2 0 f (q, n̂ ) f (q, n̂) (~vb · n̂)q
d n̂ @q

We can define ⇥(⌘, xi , q, n̂i ) = O(✏) ⌧ 1 such that


✓ ◆
i i q
f (⌘, x , q, n̂ ) = f
1+⇥

If f is a blackbody, ⇥ is the temperature perturbation.


With this definition
@f
f= ⇥q + O(✏2 )
@q
Photon Boltzmann equation
In terms of ⇥, the Boltzmann equation becomes

˙ i @⇥ ˙ = ane T
⇥ + n̂ i
+ n̂ i @ i
(Z @x )
2 0 dP
⇥ d n̂ 2 0 ⇥(n̂0 ) ⇥(n̂) + ~vb · n̂
d n̂

There is no explicit dependence on q ) if ⇥ does not


depend on q initially (i.e. assuming a local blackbody
spectrum, possibly direction-dependent), it will always
remain independent of q.

Intuitive explanation: Thomson scattering does not change


photon frequency (up to O(T /me )).
Photon Boltzmann equation in Fourier space
Fourier convention:
Z
d3 k i~k·~x ~
F (~x) = 3
e F (k),
(2⇡)
Z
i~k·~
F (~k) = 3
d xe x
F (~k)
@F
! i~kF (~k)
@~x
The Boltzmann equation becomes

˙ + i(~k · n̂)⇥ + i(~k · n̂)


⇥ ˙ = ane T
(Z )
2 0 dP
⇥ d n̂ 2 0 ⇥(n̂0 ) ⇥(n̂) + ~vb · n̂
d n̂
Photon Boltzmann hierarchy
We can define photon temperature moments:
Z 2 ⇣ ⌘
d n̂
⇥` (~k) ⌘ i` P` k̂ · n̂ ⇥(n̂),
4⇡
where P` are the Legendre polynomials:
1
P0 (x) ⌘ 1, P1 (x) ⌘ x, P2 (x) ⌘ (3x2 1), etc.
2
The first few moments have simple physical meanings:
1 n
⇥0 ⌘ ⌘ : number density perturbation,
3 n
1
⇥1 ⌘ k̂ · ~v ⌘ v : bulk velocity
3i
⇥2 / anisotropic stress
Photon Boltzmann hierarchy
The Boltzmann equation is exactly equivalent to the
infinite series of equations for the moments ⇥` (called
“Boltzmann hierarchy”):

[continuity] ˙ 3 ˙ + ikv = 0,
1
[momentum] v̇ + ik 2ik⇥2 + ik = |⌧˙ |(vb v ),
3
˙ k 2 9
⇥2 + (3⇥3 iv ) = |⌧˙ |⇥2 ,
5 3 10
˙ k
⇥` + ((` + 1)⇥`+1 `⇥` 1 ) = |⌧˙ |⇥` , ` 3.
2` + 1
Recall the baryon fluid equations:
˙b 3 ˙ + ikvb = 0
4⇢
v̇b + aHvb = |⌧˙ | (v vb ).
3⇢b
Dark matter and neutrinos
Cold dark matter and neutrinos are collisionless. We treat
the cold dark matter as a pressureless fluid:
˙c 3 ˙ + ikvc = 0, v̇c + aHvc + ik = 0.

The phase-space density of neutrinos satisfies the


collisionless Boltzmann equation. In the limit that T⌫ m⌫
(e↵ectively massless neutrinos), the neutrino hierarchy is
identical to that of the photons without Compton collisions:
˙⌫ 3 ˙ + ikv⌫ = 0,
1
v̇⌫ + ik ⌫ 2ik⇥⌫2 + ik = 0,
3
˙ ⌫` + k
⇥ ((` + 1)⇥⌫,`+1 `⇥⌫,` 1 ) = 0, ` 2.
2` + 1
Einstein’s field equations
These equations must be closed with Einstein’s equations,
governing the evolution of the metric given the matter
content:
Gµ⌫ = 8⇡G Tµ⌫ .
Unperturbed piece of this equation: H 2 = 8⇡G
3
⇢. 1st order
in perturbations: 2 independent equations:
!
2 2 aH X
k = 4⇡Ga ⇢tot + 3i (⇢s + P s )vs ,
k s
2 2
⇥ ⇤
k ( ) = 8⇡Ga ⇢ ⇥ 2 + ⇢⌫ ⇥⌫2 (if T⌫ m⌫ )

First equation: generalization of Poisson equation


r2 = 4⇡G⇢. Second equation: = unless there is
significant anisotropic stress.
Initial conditions

For every mode ~k, we have a linear system of equations for


, v, ⇥` , for baryons, photons, neutrinos and CDM. To
completely describe it we need initial conditions at a ! 0.
Well-behaved initial conditions (in this gauge) have
v = ⇥` 2 = 0, implying = ⌘ 0 . Solving the
Einstein-field equations with photon and neutrino density
perturbations only, one can show that

2⇢ + ⇢⌫ ⌫
0 = ,
3 ⇢ + ⇢⌫ 0

) There are 4 linearly-independent initial conditions for


, b, c, , ⌫ .
Initial conditions

• “Adiabatic” initial conditions: all number density


fluctuations are equal, i.e. the baryon-to-photon,
cdm-to-photon, and neutrino-to-photon ratios are the same
everywhere:
3
b0 = c0 = 0 = ⌫0 = 0 6= 0
2
n
[ ⌘ is the number density perturbation]
n
Most “natural” initial conditions, arise in single-field
inflation.
Initial conditions

• “Isocurvature” initial conditions: have vanishing initial


potential 0 = 0, but non-vanishing initial :

baryon isocurvature : ⌫0 = 0 = c0 = 0, b0 6= 0
cdm isocurvature : ⌫0 = 0 = b0 = 0, c0 6= 0
⇢⌫
neutrino isocurvature : b0 = c0 = 0, 0 = ⌫0 6= 0

Arise e.g. in multi-field inflation models. Are observed to


be subdominant.
Line-of-sight integral

⇥˙ + i(~k · n̂)⇥ + i(~k · n̂) ˙ = ane T


(Z )
2 0 dP
⇥ d n̂ 2 0 ⇥(n̂0 ) ⇥(n̂) + ~vb · n̂
d n̂

Define ⇥e↵ (n̂) ⌘ ⇥(n̂) + ,


Z
2 0 dP
S(n̂) ⌘ d n̂ 2 0 ⇥(n̂0 ) + + ~vb · n̂,
d n̂
Z ⌘0
⌧ (⌘) ⌘ d⌘ 0 (ane T )(⌘ 0 ), ⌘0 = ⌘(today)

The Boltzmann equation becomes (using ⌧˙ = ane T < 0)


˙ e↵ + i(~k · n̂)⇥e↵
⇥ ⌧˙ ⇥e↵ = ˙ + ˙ + |⌧˙ |S
Line-of-sight integral

Integral solution (drop (⌘0 ): doesn’t contribute to ` 2)


Z ⌘0 h i
~ i~k·n̂(⌘0 ⌘) ˙
⇥(⌘0 , k, n̂) = d⌘e ⌧ (⌘)
e + ˙ + |⌧˙ |S
0

Fourier transform to obtain the temperature perturbation


at present time and at the origin ⇥(⌘0 , ~x = ~0, n̂):
Z ⌘0 n ⇣ ⌘ o
⇥(⌘0 , ~0, n̂) = d⌘ e ⌧ (⌘) ˙ + ˙ + g(⌘)S
0 ~
x= (⌘0 ⌘)n̂

This integral formalizes and generalizes our statement that


“the observed anisotropy is the temperature at the last
scattering surface”.
Line-of-sight integral

• Let’s first understand the term / S.


Z
⌧ (⌘)
g(⌘) ⌘ e |⌧˙ | is the visibility function. d⌘ g(⌘) = 1.

g(⌘) ⌘ = probability of last scattering at [⌘, ⌘ + ⌘] (|⌧˙ | ⌘:


proba of scattering, e ⌧ : probability of not scattering
between ⌘ and ⌘0 ).
Line-of-sight integral

If Thomson scattering was isotropic dP/d2 n̂0 = 1/(4⇡),

S(n̂) = ⇥0 + + ~vb · n̂,


Z 2 0
d n̂
⇥0 ⌘ ⇥(n̂0 ) : temperature monopole
4⇡
: gravitational redshifting: (Sachs-Wolfe term)
overdense regions with < 0 appear colder
~vb · n̂ : Doppler e↵ect:
regions moving towards the observer appear hotter
Line-of-sight integral

• The other term is the integrated Sachs-Wolfe (ISW)


e↵ect:
Z ⌘0 ⇣ ⌘
⇥ISW (n̂) = d⌘ e ⌧ (⌘) ˙ + ˙
0 ~
x= (⌘0 ⌘)n̂

Also a gravitational redshifting e↵ect. During matter


domination ˙ = ˙ = 0. Contributions: “early ISW”, from
radiation contribution to total energy budget, “late ISW”,
from dark energy contribution.

Only contributes on large angular scales.


acoustic peaks

ISW

dampi
ng tail
SW ``plateau”
Angular power spectrum
Consider only the term ⇥0 + in S (to simplify):

Z ⌘0
⇥(n̂) = d⌘g(⌘)S(⌘, ~x = ⌘n̂) ( ⌘ ⌘ ⌘0 ⌘)
Z 0 ⌘0 Z
d3 k i ⌘~k·n̂
= d⌘g(⌘) 3
e S(⌘, ~k)
0 (2⇡)
X
= ⇥`m Y`m (n̂)
`m

Plane wave expansion:


~
X

e i ⌘k·n̂ = 4⇡ ( i)` j` (k ⌘)Y`m (k̂)Y`m (n̂)
`m
Angular power spectrum

Z ⌘0 Z
d3 k ⇤ ~k)
) ⇥`m = 4⇡( i)` d⌘g(⌘) j ` (k ⌘)Y `m ( k̂)S(⌘,
0 (2⇡)3

Write S(⌘, ~k) = T (⌘, k) 0 (~k)


Z 3 Z ⌘0
2
2 dk
C` = (4⇡) d⌘j` (k ⌘)T (⌘, k) P0 (k)
(2⇡)3 0
CMB Boltzmann codes
Examples: CMBFast, CAMB, CLASS.
• Inputs: basic cosmological parameters (⌦b , ⌦c , ⌦⇤ , T 0 , H0 )
+ amplitude and shape parameters for the primordial
power spectrum.
(i) Compute the recombination history xe (a). This is done
by independent recombination codes (e.g. HyRec and
CosmoRec).
(ii) Solve coupled linear fluid + Einstein equations for
every mode k = |~k|. This is done by solving a truncated
Boltzmann hierarchy, up to `max = a few tens. Extract the
source function S for the line-of-sight integral.
(iii) Use the line-of-sight integral to compute the angular
power spectrum.
CMB polarization

Planck map showing the polarization pattern of the CMB


Polarization
The total intensity in an electromagnetic wave is I / hE 2 i.
This can be generalized to an intensity tensor I
Iab / hEa Eb i, n̂a Iab (n̂) = 0.
Total intensity I = Iaa . The remainder is the symmetric,
trace-free polarization tensor P:
1
Iab = I ( ab n̂a n̂b ) + Pab , I = Iaa , Paa = 0.
2
Pab has 2 independent components: in a fixed basis ? n̂,
✓ ◆
Q U
Pab =
U Q
Can diagonalize Pab in a basis ?n̂:
✓ ◆
P 0
Pab =
0 P
Thomson scattering with polarization
The simplest expression for scattering from n̂0 to n̂ is
I(n̂) / I(n̂0 )?n̂ projection of I(n̂0 ) perpendicular to n̂

(T?n̂ )ab ⌘ Tab n̂a n̂c Tcb Tad n̂d n̂b + (n̂c Tcd n̂d )n̂a n̂b
exact expression:
 Z
dI(n̂) 3
= ne T d2 n̂0 I(n̂0 )?n̂ I(n̂)
dt Thom 8⇡
From this we can find dI/dt by taking the trace:
( Z
dI(n̂) 3
= ne T d2 n̂0 [1 + (n̂ · n̂0 )2 ]I(n̂0 ) I(n̂)
dt Thom 16⇡
Z )
3
n̂a n̂b d2 n̂0 Pab (n̂0 )
8⇡
Thomson scattering with polarization
We can also find the evolution of Pab by subtracting the
trace. The expression is not very enlightening, but we get
that the quadrupole moment of the total intensity sources
polarization:
dP
/ ne T Q?n̂,tf + terms / P [tf = trace-removed],
dt Thom
Z 2 0
d n̂
Qab ⌘ ( ab 3n̂0a n̂0b )I(n̂0 ) / ⇥2
4⇡

One can show that ⇥2 ⇠ ⌧k˙ v when photons and baryons


are tightly coupled.
) Generation of polarization can only happen in regions
where the optical depth is high enough to have Thomson
scattering but not so high as to wash out the quadrupole.
Isotropic radiation field
Quadrupolar radiation field

net polarization
Temperature and polarization power spectra
 1/2
( + 1)
C (µK)
2

���
Temperature

��
Polarization
(E-mode)
� reionization

���

� �� �� ��� ��� ����


Polarization: E and B modes

A vector field V~ (~x) can be decomposed into a gradient


mode G and a curl mode C ~

V~ = rG
~ ~ ⇥ C,
r ~ ~ ·C
r ~ =0

~ are uniquely defined by


The components G and C
~ · V~ ,
r2 G ⌘ r ~ ⌘r
r2 C ~ ⇥ V~ .
Polarization: E and B modes
-5

The same can be done for a polarization map Pab (n̂) on the
sky. We define the gradient (E) and curl (B) modes as
follows:

r2? E ⌘ ra? (rb? Pab ), ~ ? )a (rb Pab ),


r2? B ⌘ (n̂ ⇥ r ?

~ ? is perpendicular to the line of sight:


where r

ra? = ( ab
na nb )rb .
Polarization: E and B modes

Properties of the vector rb? Pab :

na (rb? Pab ) = rb? (na Pab ) Pab rb? na


= rb? ( 0 ) Pab ( ab na nb ) = 0.

) rb? Pab is orthogonal to n̂.


Let us consider scalar initial conditions with a single
wavenumber ~k. The only way to build a vector orthogonal
to n̂ is
rb? Pab = G(µ)(ka µna ).
One can show that this vector is curl-free, i.e. B = 0
) For any scalar mode ~k, the B-mode vanishes.
~k
rb? Pab


Polarization: E and B modes
) By linear superposition, B-modes vanish for purely
scalar initial conditions. ) B-modes are a signature of
primordial gravitational waves.
BICEP2: E signal Simulation: E from

1.7µK
−50

−55

−60

−65

BICEP2: B signal Simulation: B from

0.3µK
−50
Declination [deg.]

−55

−60

−65

50 0 −50 50
Right ascension [deg.]
Cosmological parameter estimation
Given observed anisotropies ⇥`m , we want to construct an
~ est for the underlying true cosmological
estimator ⌦
~ and give an error bar on that estimator.
parameters ⌦,
If initial conditions are Gaussian, the power spectrum
contains all the information, so we start by constructing an
estimate of the power spectrum:

1 X̀
C`est ⌘ |⇥`m |2
2` + 1 m= `

By definition of C` , the expectation value of C`est is C` :

hC`est i = C` .
Cosmological parameter estimation
This estimator for the power spectrum has a covariance:
P 2 2
est est m,m 0 h|⇥ `m | |⇥ `m i
0 0 |
hC` C`0 C` C`0 i = 0
C` C`0
(2` + 1)(2` + 1)

Gaussian initial conditions ) ⇥`m is a Gaussian field:


⌦ ↵
2
|⇥`m | |⇥`0 m0 | 2
= h⇥`m ⇥⇤`m ⇥`0 m0 ⇥⇤`0 m0 i
= h⇥`m ⇥⇤`m i h⇥`0 m0 ⇥⇤`0 m0 i
+ h⇥`m ⇥`0 m0 i h⇥⇤`m ⇥⇤`0 m0 i
+ h⇥`m ⇥⇤`0 m0 i h⇥⇤`m ⇥`0 m0 i
= C` C`0 + ``0 C`2 [ m, m0 + mm0 ]

2 ``0 2
) hC`est C`est
0 C` C`0 i = C`
2` + 1
Cosmological parameter estimation

h(C`est C` )(C`est
0 C`0 )i 2
) = .
C`2 2` + 1

This is called cosmic variance: even with a noiseless


instrument, one cannot estimate the angular power
spectrum to arbitrary precision.

This is because we have a single sky and finite number of


independent modes (2` + 1 modes per multipole `).
Cosmological parameter estimation
The probability distribution of the estimator C`est is not
Gaussian (C`est is the sum of squares of Gaussians).
For large `, it is approximately Gaussian (central-limit
theorem). Let’s approximate it as a Gaussian for all `.
) Given cosmological parameters ⌦, ~ the probability
distribution of {C2est , ...C`est } is approximately

h i
~ / exp
P({C2est , ..., C`est } | ⌦) 2 ~ ,
(⌦)
1 X 2 [C est C` (⌦)]
~ 2
2 ~ `
(⌦) ⌘
2 ` 2` + 1 C`2
Cosmological parameter estimation
What is the likelihood of a given set of cosmological
~ are the true underlying parameters ⌦
parameters ⌦ ~ true ?
We use Bayes’ theorem:

P(A)
P(A|B) = P(B|A) :
P(B)

P(parameters)
P(parameters|data) = P(data|parameters) .
P(data)
P(parameters) is called the prior, and includes additional
knowledge about parameters (for instance, from other data
~
sets). If we take a constant prior, then the probability of ⌦
given the observed data is
h i
P(⌦~ | data) / exp 2 ~
(⌦) (“likelihood”)
Cosmological parameter estimation

~ est for the true underlying parameters ⌦


Our estimator ⌦ ~ true
maximizes the likelihood, or minimizes the 2 :
2
@ ~ est ) = 0.
) (⌦
~
@⌦
We can find this minimum numerically for instance. This
should not be far from (but most likely not equal to!) the
~ true . Taylor-expand:
true cosmological parameters ⌦

@ 2 ~ est @ 2 ~ true @ 2 2
~ true )
0= (⌦ ) ⇡ (⌦ ) + (⌦est
j ⌦true
j ) ( ⌦
@⌦i @⌦i @⌦i @⌦j
Cosmological parameter estimation

0 ⇡ Vi + Fij (⌦est
j ⌦true
j )
@ 2 ~ true X 2 1 h est ~ true
i @C
`
Vi ⌘ (⌦ ) ⇡ 2
C` C` ( ⌦ ) ,
@⌦i `
2` + 1 C` @⌦i
@ 2 2 ~ true X 2 1 @C` @C`
Fij ⌘ (⌦ ) ⇡ 2
.
@⌦i @⌦j `
2` + 1 C` @⌦i @⌦j

From this we find

⌦est
i ⌦true
i = (F 1
)ij Vj .

~ true ), the estimator is unbiased:


Since hC`est i = C` (⌦

h⌦est
i i = ⌦true
i .
Cosmological parameter estimation

So on average, if we were given many Universes, we would


get the right estimate of cosmological parameters. But we
are only given one, so we want to know what is the
characteristic error we make. We compute the covariance:
~ est
⌃ij ⌘ h(⌦ i ⌦true
i )( ~
⌦ est
j ⌦true
j )i = (F 1
)ik (F 1
)jl hVk Vl i.

Using the covariance of C`est that we computed before, we


arrive at hVk Vl i = Fkl . So our final covariance matrix is

~ est
⌃ij = h(⌦ ⌦true )( ~ est
⌦ ⌦true )i = (F 1
)ij .
i i j j

The matrix Fij is called the Fisher matrix and the simple
error estimate we made is called a Fisher analysis.
Cosmological parameter estimation
In our simple noise-free approximation,

@2 2 X (2` + 1) @C` @C`


Fij = = .
@⌦i @⌦j `
2C` @⌦i @⌦j

This has to be cuto↵ at `max corresponding to the


instrument angular resolution, `max ⇠ 2⇡/ ✓.
A Fisher analysis is good to get approximate error bars.
To include non-Gaussian noise and foreground
contamination, and other “real-life” issues, in general one
needs to find the best-fit parameter and their error bar by
sampling the Likelihood function with Monte Carlo Markov
chains.
~ `m ) = probability of ⌦
L(⌦|⇥ ~ given data.
Useful references

General cosmology:
- Lecture notes by Chris Hirata
- Lecture notes by Daniel Baumann

CMB spectrum and spectral distortions:


- PhD thesis by Wayne Hu
- CUSO lecture notes by Jens Chluba

CMB anisotropies:
- Wayne Hu’s tutorials
- Lecture notes by Wayne Hu [arXiv:0802.3688]
- Recombination: Y. Ali-Haïmoud’ PhD thesis

You might also like