You are on page 1of 42

Chapter 4

Linear Perturbations in the Universe

4.1 Perturbations

Clearly, the Universe is not locally homogeneous and isotropic, and only exhibits such symmetries
on sufficiently large scales of ∼ 100 Mpc. We now consider departures from isotropy and homo-
geneity on the largest cosmological scales. These are caused by gravitational collapse and manifest
themselves as space-time perturbations both in the FRW metric as well as in the energy-momentum
tensor components, e.g.

δgµν (x, t) : Ψ(x, t), Φ(x, t) (4.1)


δTµν (x, t) : δρ(x, t), vi (x, t), δP (x, t), Πij (x, t) (4.2)

These perturbations can be treated using linear perturbation theory, which is the subject of
this chapter.

4.2 Fourier Transform

Let us define the conventions for Fourier Transforms that we will use throughout. The Fourier
transform of a field δ(x, t) is denoted F(δ(x, t)) = δ(k, t) and given by

Z
δ(k, t) = d3 x e−ik·x δ(x, t) (4.3)

and the inverse relation is F −1 (δ(k, t)) = δ(x, t)

Z
d3 k ik·x
δ(x, t) = e δ(k, t) (4.4)
(2π)3

121
122 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

Similarly the Fourier Transform of spatial derivatives can be obtained integrating by parts, i.e.
  Z
∂ ∂
F δ(x, t) = d3 x e−ik·x δ(x, t)
∂xi ∂xi
Z Z

= d2 xj e−ikj ·xj dxi e−iki xi δ(x, t)
j6=i ∂xi
 
Z Z
d2 xj e−ikj ·xj e−iki xi δ(x, t)|∞ dxi e−iki xi δ(x, t)
 
= −∞ + iki
j6=i | {z }
0
Z
= iki d3 x e−ik·x δ(x, t) = iki δ(k, t) (4.2)

In linear perturbation theory, it becomes very convenient to work with the various fields (density,
potential, etc) in Fourier space. As we will see, in linear theory, the Fourier components of the
field evolve independently, greatly simplifying the analysis. In addition, spatial derivatives become
simple multiplications by factors of ki or k, turning partial differential equations (in space and
time) into simple ordinary differential equations (in time) for each Fourier mode k.
In summary we have the following conversions to Fourier space:
δ(x) → δ(k) (4.3)

δ(x) → iki δ(k) (4.4)
∂xi
∇2 δ(x) → −k 2 δ(k) (4.5)
Z
d3 x′ δ(x′ )W (x − x′ ) → δ(k)W (k) (4.6)

As above, we often drop the t dependence, and also the tilde ∼ sometimes used in the Fourier
components.

4.3 Geometric Perturbations


FRW metric with scalar perturbations in the Newtonian conformal gauge:
ds2 = gµν dxµ dxν = −(1 + 2Ψ)dt2 + a2 (1 + 2Φ)dx2 (4.7)
so that

g00 = −(1 + 2Ψ)


g0i = gi0 = 0
gij = δij a2 (1 + 2Φ)
(4.5)
and
g 00 = −(1 − 2Ψ)
g 0i = g i0 = 0
δij
g ij = (1 − 2Φ)
a2
(4.3)
4.3. GEOMETRIC PERTURBATIONS 123

where Ψ(x, t) is the Newtonian gravitational potential (time-time metric perturbation) and Φ(x, t)
is the curvature potential (space-space metric perturbation).

The most general perturbations include scalar, vector or tensor perturbations. It turns out that
these decouple and do not affect each other. Scalar perturbations are the only ones that couple
to matter and the most important to couple with photons. Vector perturbations can be shown to
decay away and become unimportant, at least in standard considerations. Tensor perturbations are
related to gravity waves, and are relevant to characterize the polarization of photon perturbations,
but their strength is much smaller than scalar perturbations.

The gauge choice is related to the freedom of choosing coordinates in General Relativity and still
have the same Physics. Here it may be useful to recall gauges in Electromagnetism (e.g. Lorenz
Gauge ∂µ Aµ = 0, or Coulomb gauge ∇ · A = 0) which produced the same observed E and B
fields, and were a mere question of choice and convenience. In GR, this freedom also exists because
we can make coordinate transformations that do not change the Physics described by the metric
perturbations, e.g. still leave the line element unchanged i.e. ds2 = gµν dxµ dxν = gµν
′ (dx′ )µ (dx′ )ν .

Here we start considering only scalar perturbations in this specific gauge.

The Christoffel symbols are:


1 σρ
Γσµν = g (∂µ gνρ − ∂ρ gµν + ∂ν gρµ ) (4.4)
2
Inserting the metric and keeping terms to first order in perturbations, i.e. linear in Ψ and Φ,
we obtain its components as

Γ000 = Ψ̇ (4.5)

and

∂Ψ
Γ00i = Γ0i0 = (4.6)
∂xi

and

h i
Γ0ij = δij a2 H + 2H(Φ − Ψ) + Φ̇ (4.7)

and

1 ∂Ψ
Γi00 = (4.8)
a2 ∂xi

and
124 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

 
Γi0j = Γij0 = δij H + Φ̇ (4.9)

and

Γijk = δki ∂j Φ − δjk ∂i Φ + δij ∂k Φ (4.10)

The Ricci tensor is given by

Rµν = Γαµν,α + Γαµν Γβαβ − Γαµα,ν − Γαµβ Γβαν (4.11)

Therefore, its components are

ä 1
R00 = −3 + 2 ∇2 Ψ + 3H(Ψ̇ − 2Φ̇) − 3Φ̈ (4.12)
a a

and

R0i = −2∂i (Φ̇ − HΨ) (4.13)

and
h   i
Rij = δij (aä + 2a2 H 2 )[1 + 2(Φ − Ψ)] + a2 H 6Φ̇ − Ψ̇ + a2 Φ̈ − ∇2 Φ − ∂i ∂j (Φ + Ψ) (4.14)

The Ricci scalar is

R = g µν Rµν = g 00 R00 + g ij Rij (4.15)

and it becomes
   
ä 2 ä
R=6 + H2 − 2 ∇2 (Ψ + 2Φ) − 6H(Ψ̇ − 4Φ̇) + 6Φ̈ − 12Ψ + H2 (4.16)
a a a

Summary
We can summarize these results in Fourier space as

Γ000 = Ψ̇ (4.17)

Γ00i = Γ0i0 = iki Ψ (4.18)

h i
Γ0ij = δij a2 H + 2H(Φ − Ψ) + Φ̇ (4.19)
4.4. PERTURBED BOLTZMANN EQUATIONS 125

iki
Γi00 = Ψ (4.20)
a2

 
Γi0j = Γij0 = δij H + Φ̇ (4.21)

Γijk = iΦ(δki kj − δjk ki + δij kk ) (4.22)

ä k 2
R00 = −3 − 2 Ψ + 3H(Ψ̇ − 2Φ̇) − 3Φ̈ (4.23)
a a

R0i = −2iki (Φ̇ − HΨ) (4.24)

h   i
2 2 2 2 2
Rij = δij (aä + 2a H )[1 + 2(Φ − Ψ)] + a H 6Φ̇ − Ψ̇ + a Φ̈ + k Φ + ki kj (Φ + Ψ) (4.25)

   
ä 2k 2 ä
R=6 + H2 + 2 (Ψ + 2Φ) − 6H(Ψ̇ − 4Φ̇) + 6Φ̈ − 12Ψ + H2 (4.26)
a a a

4.4 Perturbed Boltzmann Equations


We have seen that the Boltzmann equation describing the evolution of the distribution function
f (x, p, t) is given by
 
df ∂f dxi ∂f dp ∂f dp̂i ∂f ∂f
= + + + = . (4.27)
dt ∂t dt ∂xi dt ∂p dt ∂ p̂i ∂t C

where p2 = gij P i P j and p̂i p̂i = δij p̂i p̂j = 1. Before we considered the simpler case where
f (x, p, t) = f (p, t) was the equilibrium distribution. Our goal now is to consider this equation
for different species (photons, dark matter, baryons) for the case of a metric with linear pertur-
bations in Conformal Newtonian gauge. These perturbations will distort the usual equilibrium
distribution function adding spatial dependence

f (x, p, t) = fEQ (p, t) + δf (x, p, t) . (4.28)

We need to rewrite the Boltzmann equation in a way that allows it to be evolved in linear
theory. First, we see that the last term is zero in linear perturbations. That is because:
∂f
is first order, since fEQ has no p̂i dependence
∂ p̂i
dp̂i
is also first order, since γ’s move in straight-lines in the absence of potentials (perturbations)
dt
Therefore the product is second-order and neglected.
126 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

4.4.1 Calculating dxi /dt


Recall the definition 4-momentum P µ :
dxµ
Pµ = = (P 0 , P i ) (4.27)

and for photons we have

P 2 = gµν P µ P ν = g00 P 0 P 0 + gij P i P j = −(1 + 2Ψ)(P 0 )2 + p2 = −m2 (4.28)


| {z }
≡p2

p
Therefore defining the energy E in the usual way: E = p 2 + m2

E
P0 = √ ≈ E(1 − Ψ) (4.29)
1 + 2Ψ

On the other hand

dxi dxi dλ Pi
= = 0 (4.30)
dt dλ dt P

We have P 0 above, so we need to express P i in terms of p and p̂i . Since they must be proportional,

P i = C p̂i (4.31)

we just need to find C using the definition of p2 :

p2 = gij P i P j = C 2 a2 (1 + 2Φ) δij p̂i p̂j (4.32)


| {z }
=1

so
p p(1 − Φ)
C= √ ≈ (4.33)
a 1 + 2Φ a

and
p(1 − Φ) i
Pi = p̂ (4.34)
a
so finally, the coefficient

dxi Pi p(1 − Φ) i p̂i p


= 0 = p̂ = (1 + Ψ − Φ) (4.35)
dt P aE(1 − Ψ) aE

The combination Ψ − Φ is the lensing potential, which is related to the deflection of photons in
gravitational potentials. Here we can neglect these terms, since dxi /dt is multiplying ∂f /∂xi which
is a first-order term (absence in the zero-order case). So the Boltzmann equation becomes so far
 
∂f p̂i p ∂f dp ∂f ∂f
+ i
+ = . (4.36)
∂t a E ∂x dt ∂p ∂t C
4.4. PERTURBED BOLTZMANN EQUATIONS 127

4.4.2 Calculating dp/dt


In order to estimate dp/dt we need the Geodesics equation, whose zero component now becomes

dP 0
= −Γ0αβ P α P β (4.37)

Using dt/dλ = P 0 we may replace derivatives in the first term:

dP 0 dP 0 d
= P0 = E(1 − Ψ) [E(1 − Ψ)] (4.38)
dλ dt dt
so the Geodesics becomes
d P αP β
[E(1 − Ψ)] = −Γ0αβ (1 + Ψ) (4.39)
dt E
Then differentiating the left-hand side (recall Ψ = Ψ(x, t)), we have
d dE dΨ
[E(1 − Ψ)] = (1 − Ψ) −E
dt dt dt 
dE dxi ∂Ψ
= (1 − Ψ) − E Ψ̇ +
dt dt ∂xi
 
dE p̂i p ∂Ψ
= (1 − Ψ) − E Ψ̇ + (4.38)
dt a E ∂xi
so the Geodesics become
 
dE p̂i p ∂Ψ α β
0 P P
(1 − Ψ) − E Ψ̇ + = −Γαβ (1 + Ψ) (4.39)
dt a E ∂xi E
or
 
dE p̂i p ∂Ψ α β
0 P P
= E Ψ̇ + − Γαβ (1 + 2Ψ) (4.40)
dt a E ∂xi E
So we just need to evaluate the last term on the right:

P αP β P 0P 0 P 0P i P iP j
Γ0αβ = Γ000 + 2 Γ00i + Γ0ij
E |{z} E |{z} E |{z} E
Ψ̇ ∂Ψ/∂xi δij a2 [H+2H(Φ−Ψ)+Φ̇]

∂Ψ p(1 − Ψ − Φ)p̂i 2 p2 (1 − 2Φ)p̂i p̂j


≈ Ψ̇E(1 − 2Ψ) + 2 + δ ij a [H + 2H(Φ − Ψ) + Φ̇]
∂xi a Ea2
∂Ψ pp̂ i p 2 p 2 p 2
≈ E Ψ̇ + 2 i + H(1 − 2Φ) + 2 H(Φ − Ψ) + Φ̇
∂x a E E E
p2 ∂Ψ pp̂i p2
≈ E Ψ̇ + Φ̇ + 2 i + H(1 − 2Ψ)
E ∂x a E
(4.37)

and
P αP β p2 ∂Ψ pp̂i p2
Γ0αβ (1 + 2Ψ) = E Ψ̇ + Φ̇ + 2 i + H (4.38)
E E ∂x a E
128 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

so the Geodesics gives


   
dE p̂i p ∂Ψ p2 ∂Ψ pp̂i p2
= E Ψ̇ + − E Ψ̇ + Φ̇ + 2 i + H
dt a E ∂xi E ∂x a E
i
 2 i 2

p̂ p ∂Ψ p ∂Ψ pp̂ p
= i
− Φ̇ + 2 i + H (4.38)
a ∂x E ∂x a E

and finally,
 
dE p2 ∂Ψ pp̂i p2
= − Φ̇ + i + H (4.39)
dt E ∂x a E

or, equivalently, using EdE = pdp:


 
dp ∂Ψ E p̂i
= −p Φ̇ + i +H (4.40)
dt ∂x p a

Therefore the Boltzmann equation becomes


   
∂f p̂i p ∂f ∂f ∂Ψ E p̂i ∂f
+ i
−p Φ̇ + i +H = .
∂t a E ∂x ∂p ∂x p a ∂t C
(4.40)

Equivalently, using E instead of p, we have:


 
∂f p̂i p ∂f dE ∂f ∂f
+ + = ,
∂t a E ∂xi dt ∂E ∂t C
(4.40)

or finally
   
∂f p̂i p ∂f ∂f p2 ∂Ψ pp̂i p2 ∂f
+ i
− Φ̇ + i + H = .
∂t a E ∂x ∂E E ∂x a E ∂t C
(4.40)

4.4.3 Photons
Collisionless Equation

For photons, E = p, so
 
dp ∂Ψ p̂i
= −p Φ̇ + i + H (4.41)
dt ∂x a

and the Boltzmann equation becomes


 
df ∂f p̂i ∂f ∂f ∂Ψ p̂i
= + −p H + Φ̇ + i (4.42)
dt ∂t a ∂xi ∂p ∂x a
4.4. PERTURBED BOLTZMANN EQUATIONS 129

Temperature Perturbation
Considering the perturbation in the distribution function such that the zero-th order term is the
equilibrium Planck distribution f 0 (p, t) = fEQ (p, t), we have:

f (x, p, t) = f 0 (p, t) + δf (x, p, t) (4.43)

We can parametrize this perturbation in terms of a perturbation in the temperature field

T (x, p̂, t) = T (t) + δT (x, p̂, t) = T (t) [1 + Θ(x, p̂, t)] (4.44)

where Θ(x, p̂, t) = δT (x, p̂, t)/T (t), such that the Planck format is preserved:
   −1
p
f (x, p, t) = exp −1 (4.45)
T (x, p̂, t)
Expanding this relation perturbatively in δT we have
   −1
p ∂f (x, p, t)
f (x, p, t) = exp −1 + δT
T (t) ∂T |{z}
T (x,p̂,t)=T (t) T Θ
| {z } | {z }
f 0 (p,t) ∂f 0 0
∂T
=− Tp ∂f
∂p

∂f 0
= f 0 (p, t) − p Θ (4.45)
∂p
so
∂f 0
δf (x, p, t) = −p Θ (4.46)
∂p

Zero-order
As seen in the previous Chapter, keeping only zero-th order terms in the collision less equation, we
have

df ∂f 0 ∂f 0
= − p H=0 (4.47)
dt 0 order ∂t ∂p
which implies
1
T (t) ∝ (4.48)
a

First-order
Keeping now only first-order terms (after using the zero-th order solution to get rid of zero-order
terms), we have
 
df ∂δf p̂i ∂δf ∂δf ∂f 0 ∂Ψ p̂i
= + −p H −p Φ̇ + i (4.49)
dt 1st order ∂t a ∂xi ∂p ∂p ∂x a
and inserting δf from above, we have
     
df ∂ ∂f 0 p̂i ∂f 0 ∂Θ ∂ ∂f 0 ∂f 0 ∂Ψ p̂i
= −p Θ −p +pHΘ p −p Φ̇ + i (4.50)
dt 1st order ∂t ∂p a ∂p ∂xi ∂p ∂p ∂p ∂x a
130 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

The first term can be expanded as


   
∂ ∂f 0 ∂f 0 ∂ ∂f0
−p Θ = −p Θ̇ − pΘ (4.51)
∂t ∂p ∂p ∂t ∂p

Then using Ṫ /T = ṗ/p = −H and ∂f0 /∂T = −(p/T )∂f0 /∂p, we have
        p  ∂f   
∂ ∂f0 ∂ ∂f0 ∂ ∂f0 ∂ 0 ∂ ∂f0
= = Ṫ = −(T H) − =H p (4.52)
∂t ∂p ∂p ∂t ∂p ∂T ∂p T ∂p ∂p ∂p

So
   
∂ ∂f 0 ∂f 0 ∂ ∂f 0
−p Θ = −p Θ̇−pHΘ p (4.53)
∂t ∂p ∂p ∂p ∂p

Inserting this back into the first-order equation, the terms in blue cancel out, and the left-over
terms all have a common factor of −p∂f 0 /∂p so that we have
 
df ∂f 0 p̂i ∂Θ p̂i ∂Ψ
= −p Θ̇ + + Φ̇ + (4.54)
dt 1st order ∂p a ∂xi a ∂xi

Collision Term
Compton scattering of photon off electrons is the main interaction for photons:

e− (q) + γ(p) ↔ e− (q′ ) + γ(p′ ) (4.55)

and contributes with a collision term that is at least first-order in perturbations (in zero-th order
the distribution is Planck and the collision term is zero). The collision term associated to this
process for the incoming photon integrates over the momenta of the other species, i.e.
  Z Z Z
∂f (p) 1 d3 q d3 q ′ d 3 p′
= |M|2
∂t C p (2π)3 2Ee (q) (2π)3 2Ee (q ′ ) (2π)3 2Ee (p′ )
 
× (2π)4 δ 3 (p + q − p′ − q′ )δ E(p) + Ee (q) − E(p′ ) − Ee (q ′ ) fe (q′ )f (p′ ) − fe (q)f (p)
(4.54)

After using energy conservation, making a few approximations related to the electron being non-
relativistic, and performing an integral we find
  Z Z  
∂f (p) π d3 q d 3 p′ 2 ′ (p − p′ ) · q ∂δ(p − p′ )  
= fe (q) |M| δ(p − p ) + f (p′ ) − f (p)
∂t C 4m2e p (2π)3 (2π) 3 me ∂p ′

(4.54)

The Compton scattering amplitude can be computed from Feynman diagrams producing
 

|M|2 = 6πσT m2e 1 + (p̂ · p̂′ )2  (4.55)


| {z }
cos2 θ
4.4. PERTURBED BOLTZMANN EQUATIONS 131

Expanding in Legendre polynomials the piece:


 
 2
 2 2 (3 cos2 θ − 1)
6 1 + cos θ = 6 + 6 cos (θ) = (8 − 2) + 6 cos θ = 8 × |{z}
1 +4
2
P0 (cos θ) | {z }
P2 (cos θ)
= 8P0 (cos θ) + 4P2 (cos θ)

To simplify the calculation, we take only the first term in the expansion, assuming a constant
amplitude:

|M|2 = 8πσT m2e (4.54)

but the second term can be considered in a similar fashion (exercise). Inserting this amplitude,
using the distribution expansion to first order (Eq. 6.149), making some further approximations to
perform the extra integrals, and defining
Z
d3 q
ne = fe (q) (4.55)
(2π)3
Z
d3 q q
ne v b = 3
fe (q) (4.56)
(2π) me
Z
1
Θ0 = dΩ Θ(x, p̂, t) (4.57)

where vb is the baryonic velocity, we obtain


 
∂f (p) ∂f 0
= −p ne σT [Θ0 − Θ + p̂ · vb ] (4.58)
∂t C ∂p

More generally we define the lth multipole moment Θl of the temperature fluctuation field Θ as
Z 1
1 dµ
Θl = Pl (µ) Θ (4.59)
(−i)l −1 2

where Pl (µ) is the Legendre polynomial of order l. For l = 0, P0 (µ) = 1 and using dΩ = 2π sin θdθ =
2πd(cos θ) = 2πdµ, we have
Z 1 Z
1 dµ 1
Θ0 = P0 (µ) Θ = dΩ Θ (4.60)
(−i)0 −1 2 4π

Full Equation
Setting the left-hand side of the Boltzmann equation to the collision term we find

p̂i ∂Θ p̂i ∂Ψ
Θ̇ + i
+ Φ̇ + = ne σT [Θ0 − Θ + p̂ · vb ] (4.61)
a ∂x a ∂xi
Changing from physical time t to conformal time η via dt = adη, we have

∂Θ ∂Ψ
Θ′ + p̂i i
+ Φ′ + p̂i i = ne σT a [Θ0 − Θ + p̂ · vb ] (4.62)
∂x ∂x
132 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

Changing to Fourier space, we have ∂/∂xi → iki , so

Θ′ + ip̂i ki Θ + Φ′ + ip̂i ki Ψ = ne σT a [Θ0 − Θ + p̂ · vb ] (4.63)

Let θ be the angle between p̂ (photon propagation) and k (wave vector). We define its cosine:

k · p̂ ki p̂i
µ = cos(θ) = = → p̂i ki = µk (4.64)
k k
We also assume the velocity is irrotational, or divergent1 , i.e. points in the same direction as k or
(vb = vb k/k) so that

Θ′ + ikµΘ + Φ′ + ikµΨ = ne σT a [Θ0 − Θ + vb µ] (4.65)

It is then useful to define the optical depth τ (η):

Number of interactions a photon suffer (with e− s) from η to today (η0 )

or similarly the differential optical depth dτ (η):

Number Ne of e− s a photon encounters in a volume dV (η) defined by σT and depth dL(η):

dτ (η) = Ne = ne (η)dV (η) = ne (η)σT dL(η) = ne (η)σT adη (4.66)

so that the finite optical depth becomes the integral


Z η0 Z η0

τ= dτ (η ) = ne (η ′ )σT adη ′ (4.67)
η η

so that

τ′ ≡ = −ne σT a (4.68)

and the Boltzmann equation in Fourier space becomes

Θ′ + ikµΘ + Φ′ + ikµΨ′ = −τ ′ [Θ0 − Θ + µvb ] (4.69)

where Θ(k, p̂, η) and Φ(k, η).

4.4.4 Dark Matter


We assume that dark matter decouples and become non-relativistic much before matter domina-
tion and therefore can be considered collisionless and non-relativistic for cosmological purposes of
structure formation. Dark matter behaves effectively as a fluid, i.e. its evolution is described by
only the first two moments of δf , i.e. δdm = δρ/ρ̄ and vdm . As a result, we can obtain its equations
of motion simply by energy-momentum covariant conservation:

∇µ Tµν = 0 (4.70)
1 ~ × vb = 0) is a pure gradient vb = ∇φ.
An irrotational vector (∇ ~ Therefore ṽb = ikφ̃ and ṽb = ikφ̃, so ṽb = kṽb /k.
4.4. PERTURBED BOLTZMANN EQUATIONS 133

Here however, we will derive the same results from the formalism of the Boltzmann equa-
tion. Going back to the general boltzmann equation, we apply it to the dark matter distribution
fdm (x, E, p̂, t) with no collision term:

 
∂fdm p̂i p ∂fdm ∂fdm p2 p2 pp̂i ∂Ψ
+ − H + Φ̇ + = 0
∂t a E ∂xi ∂E E E a ∂xi
(4.70)

For photons, we knew that the zero-th order solution was Bose-Einstein (or Planck).
For dark matter, we do not know its origin, but we don’t really need to know it anyway. The only
useful information is that it is non-relativistic, so that E ∼ m and the factors p/E ∼ mv/m ∼ v
which are first-order velocity perturbations (as opposed to photons where p/E = 1).
We proceed by taking moments of the distribution. We already did this in the previous chapter
in zeroth order, but let us repeat it here to separate zeroth and first order terms. Multiplying the
equation by d3 p/(2π)3 and integrating (the zeroth moment) we have
Z Z i Z d3 p ∂f p2 Z
∂ d3 p 1 ∂ d3 p pp̂i h dm 1 ∂Ψ d3 p ∂fdm i
f dm + f dm − H + Φ̇ − pp̂ = 0
∂t (2π)3 a ∂xi (2π)3 E (2π)3 ∂E E ∂x}i
a {z
| (2π)3 ∂E
| {z } | {z } | {z } | {z }
ndm i
ndm vdm R d3 p ∂fdm first-order first-order
p =−3n dm
(2π)3 ∂p | {z }
2nd-order=0
(4.70)

In the third integral we used dE/dp = p/E and integrated by parts.


Therefore the zeroth-moment of the distribution gives
i )
∂ndm 1 ∂(ndm vdm h i
+ + 3 H + Φ̇ ndm = 0 (4.71)
∂t a ∂xi
Now we consider the number density itself (or more generally the energy density) as a back-
ground (zeroth order) term plus a perturbation:

ndm (x, t) = n0dm + δndm (x, t) = n0dm [1 + δ(x, t)] (4.72)

A similar definition follows for the energy density since ρdm = mndm so the mass drops out:

ρdm (x, t) = ρ̄dm [1 + δ(x, t)] (4.73)

so we want to describe the zeroth order number density n0dm and the perturbation δ:

δρdm
δ(x, t) = (4.74)
ρ̄dm
The zeroth order terms had already given us in Chapter 2:

∂n0dm
+ 3Hn0dm = 0 (4.75)
∂t
or
d(n0dm a3 )
= 0 → n0dm ∝ a−3 and ρ̄dm ∝ a−3 (4.76)
dt
134 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

Collecting now the first-order terms we have


i (x, t)
∂δ(x, t) n0dm ∂vdm
n0dm + + 3Φ̇(x, t)n0dm = 0 (4.77)
∂t a ∂xi
or canceling out the common n0dm :
i
∂δ 1 ∂vdm
+ + 3Φ̇ = 0 (4.78)
∂t a ∂xi
This is the linear continuity equation for dark matter in the presence of expansion a and a gravi-
tational field Φ.
Doing a similar procedure for the first moment of the equation, we multiply it by d3 p/(2π)3 (pp̂i /E),
we end up obtaining to first-order (at zeroth order we obviously obtain nothing, since all terms are
at least first-order)
i
∂vdm i 1 ∂Ψ
+ Hvdm + =0 (4.79)
∂t a ∂xi
This is the linear Euler equation expressing conservation of momentum of the perturbations.
Higher moments of the equation will lead to terms of order (p/E)2 ∼ v 2 , which we are neglecting
since dark matter is cold. Obviously for photons, this cannot be done. In fact, for photons we need
all moments of the distribution to solve the equation we derived before.
Notice that the equation for the zero-th moment (δ) we need the first-moment quantity (v i ).
Fortunately, the equation for the first-moment does not depend on the second-moment for cold dark
matter, but in general it does, i.e. the equation for the lth moment of the distribution depends on
the (l + 1)th moment, so we always need to go one moment further (and truncate at some high
moment in which the scales we’re studying do not require further moments). This is one technique
to solve the Boltzmann equations for photons.
Writing the equations in conformal time η and in Fourier space:

δc′ + iki vci + 3Φ′ = 0 (4.80)


 ′
i ′ a
(vc ) + vci + iki Ψ = 0 (4.81)
a

Assuming the velocity field is irrotational i.e. ∇~ × vc = 0 (or similarly that it is divergent and
points in the direction of ki ), we must have vc = (k i /k)vc so that ki vci = (k 2 /k)vc = kvc and
i

δc′ + ikvc + 3Φ′ = 0 (4.82)


 ′
a
vc′ + vc + ikΨ = 0 (4.83)
a
~ · vc (x, t) or in Fourier space θc = iki v i = ikvc and multiplying
Defining the variable θc (x, t) = ∇ c
the second equation by ik, we have

δc′ + θc + 3Φ′ = 0 (4.84)


 
a′
θc′ + θc − k 2 Ψ = 0 (4.85)
a
So once we have equations for the evolution of Φ and Ψ (which will come from the perturbed Einstein
equations), we can solve this system of equations above for δc (k, t) and θc (k, t) = −ikvc (k, t).
4.4. PERTURBED BOLTZMANN EQUATIONS 135

Taking the time derivative of the first equation and multiplying the second by −1, we have:

δ ′′ + θ′ + 3Φ′′ = 0 (4.86)
 c′  c
a
−θc′ − θc + k 2 Ψ = 0 (4.87)
a

Adding the two equations


 
a′
δc′′ − θc +3Φ′′ + k 2 Ψ = 0 (4.88)
a |{z}
(−δc′ −3Φ′ )

so that we obtain a second-order equation for δ(k, t) in terms of the gravitational potentials.
 ′   ′ 
a a
′′
δc + ′ ′′
δc + 3 Φ + Φ′ = −k 2 Ψ (4.89)
a a

4.4.5 Baryons
For baryons we must consider the interactions between electrons and protons (coupled by Coulomb
scattering) and between electrons and photons (coupled by Compton scattering):

e(q) + p(Q) → e(q ′ ) + p(Q′ ) (4.90)


′ ′
e(q) + γ(p) → e(q ) + γ(p ) (4.91)

The interactions bring electrons and protons to a common overdensity δb , which we denote the
baryon overdensity, and to a common velocity vb

δρe δρp
δb = = (4.92)
ρ̄e ρ̄p
vb = ve = vp (4.93)

and our job is to derive evolution equations for δb and vb , as we did for dark matter.
Here the Boltzmann equation for electrons and protons become

dfe (x, q, t)
= hcep iQQ′ q′ + hceγ ipp′ q′ (4.94)
dt
dfp (x, Q, t)
= hcep iqq′ Q′ + hcpγ ipp′ Q′ (4.95)
dt | {z }
≈0

where the angular brackets mean integration over the momenta in the subscripts:
Z Z Z
d3 p d 3 p′ d3 q ′
h...ipp′ q′ = ... (4.96)
(2π)3 (2π)3 (2π)3

and the unintegrated collision terms are, e.g. for the Compton scattering

|M|2  
ceγ ≡ (2π)4 δ 4 (p + q − p′ − q ′ ) ′ ′
fe (q ′ )fγ (p′ ) − fe (q)fγ (p) (4.97)
8E(p)E(p )Ee (q)Ee (q )
136 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

and similar for the Coulomb scattering. The main difference in theses processes is the amplitude,
which is proportional to the mass inverse for non-relativistic particles. Therefore the term for the
Compton scattering of protons is much smaller than electrons.
Proceeding similarly to the case of dark matter, but accouting for the collision terms, we obtain

δb′ + ikvb + 3Φ′ = 0 (4.98)


a′ 4ργ
vb′ + vc + ikΨ = τ ′ [3iΘ1 + vb ] (4.99)
a 3ρb

where, from Eq. (4.59), we have


Z 1

Θ1 ≡ i µΘ(µ) (4.100)
−1 2

So the density evolution is similar to that of cold dark matter, and the velocity is modified by the
interaction term which accounts for the momentum exchange between photons and baryons. To
first order, the interaction affects the momentum of baryons. But since they are non-relativistic and
Eb ∼ mb , the energy over density evolution is not affected and proceeds just like for non-interaction
dark matter.

4.4.6 Neutrinos
For massless neutrinos, the evolution equations are similar to photons, except that they have a
different temperature T ν and there is no collision term. Defining the temperature perturbation
N = δTν /Tν , we immediately have

N ′ + ikµN + Φ′ + ikµΨ′ = 0 (4.101)


p
For massive neutrinos, E = p2 + m2ν . Their evolution starts as that of photons while neutrinos
are relativistic and make a transition to that of dark matter once they become non-relativistic. Look
at Ma & Bertschinger 1995 for a careful description of linear perturbations, not only for massless and
massive neutrinos, but also for the other species and for the Einstein Equations in both Conformal
Newtonian Gauge and Synchronous Gauge. They also provide a technique to solve the equations
for photons and neutrinos in terms of a multipole expansion in Legendre polynomials.

4.5 Perturbed Energy-Momentum Tensor


We had seen that for a perfect fluid, Tµν was simply given by

T µν = (ρ + P )U µ U ν + P g µν (4.102)

In a local frame K′ in which the fluid is isotropic the components becomes simply

(T 00 )′ = ρ (4.103)
0i ′
(T ) = 0 (4.104)
ij ′ ij
(T ) = Pδ (4.105)
4.5. PERTURBED ENERGY-MOMENTUM TENSOR 137

Making a transformation to a frame moving with small velocity v i = dxi /dτ relative to the isotropic
frame changes the energy-momentum tensor to

T 00 = γ(ρ + P v 2 ) (4.106)
0i 2 i
T = γ (ρ + P )v (4.107)
ij 2 ij
T = γ (ρ + P )vi vj + P δ (4.108)

If the velocity if small v i ≪ 1, γ ≈ 1 so that T 0i = (ρ + P )v i can be seen as a small perturbation.


Obviously, perturbations can arise from other causes (not just because of a change of frame
we), so we define more generally the perturbations in the fluid case as

T 0 0 = −(ρ̄ + δρ) (4.109)


0
T i = (ρ̄ + P̄ )vi = −T i0 (4.110)
i
T j = (P̄ + δP )δ i j + Σi j , i
Σi=0 (4.111)

where we allow for pressure perturbations and for anisotropic stress perturbation Σi j .
More generally, from the definition of the energy momentum tensor:
X Z dP1 dP2 dP3 1 P µ Pν
µ
T ν= gi √ fi (x, p, t) (4.112)
(2π)3 −g P0
i

we may define first-order perturbations in T µ ν by account for all perturbations that enter this
expression, i.e. perturbations in the metric δgµν and in the distribution function δf .
We have

−g = (1 + 2Ψ)(1 + 2Φ)a6 (4.113)


−1/2 −3
→ (−g) = (1 − Ψ − 3Φ)a (4.114)

Also P 0 = E(1 − Ψ), P0 = g00 P 0 = −E(1 + Ψ), P i = ap (1 − Φ)p̂i and Pi = gij P j = ap(1 + Φ)p̂i .
Therefore

dP1 dP2 dP3 = a3 (1 + 3Φ)d3 p (4.115)

and

1 3 3 1 d3 p
dP1 dP2 dP3 (−g)−1/2 = a (1 + 3Φ)d p(1 − Ψ − 3Φ)a −3
= (4.116)
P0 E(1 − Ψ) E

Therefore
Z
µ
X d 3 p P µ Pν
T ν = gi fi (x, p, t) (4.117)
(2π)3 E
i

For µ = ν = 0, since P 0 P0 = −E 2 to first order, we have


Z
X d3 p
T00 = − gi E(p) [f0 (p) + δfi (x, p, t)] = −(ρ̄ + δρ) = −ρ̄(1 + δ) = T̄ 0 0 + δT 0 0(4.118)
(2π)3
i
138 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

For relativistic species (e.g. photons), E = p and δf = −p ∂f 0


∂p Θ, so

Z Z 1 Z  
0 d3 p dp(2π)p2 ∂f0
δT 0 = −g Eδf = −g dµ p −p Θ (4.119)
(2π)3 −1 (2π)3 ∂p
Z ∞ Z
(2π)p4 ∂f0 1 dµ
= 2g dp Θ (4.120)
(2π)3 ∂p −1 2
| 0 {z } | {z }
R (2π)4p3 Θ0
−2g dp f0
(2π)3
 Z 
(4π)p2
= −4 g dp pf0 Θ0 = −4ρ̄Θ0 (4.121)
(2π)3
| {z }
R d3 p
g pf0 =ρ̄
(2π)3

This is expected since for photons ρ ∝ a−4 ∝ T 4 , so

δT
δρ ∝ 4T 3 δT = 4ρ = 4ρΘ (4.122)
T

Similarly, P 0 Pi = Eapi (1 − Ψ + Φ), where pi = pp̂i , so we have

Z
0 d3 p
T i = g api (1 − Ψ + Φ) [f0 (p) + δfi (x, p, t)] (4.123)
(2π)3

For non-relativistic particles, their small velocity v i = pi /m is a first-order perturbation. Therefore,


in this case we neglect the gravitational potentials and pick only f0 :
Z
d3 p
δT 0 i = g api f0 (p) (4.124)
(2π)3

Therefore, using pi = mvi , we have


Z
d3 p
δT 0 i = am g vi f0 (p) = a |{z}
mn̄ vi = aρ̄vi (4.125)
(2π)3
| {z } ρ̄
n̄vi

This agrees with Eq. (4.110) for a = 1 and P̄ = 0 (non-relativistic matter). Finally, using the fact
that vi is irrotational (k i vi = kv) we have

k i δT 0 i = aρ̄k i vi = aρ̄kv (4.126)

For relativistic particles on the other hand, p = E is not small, so the perturbation comes from δf :
Z Z 1 Z  
0 d3 p dp(2π)p2 ∂f0
δT i = g app̂i (1 − Ψ + Φ) δf = ag dµ pp̂i −p Θ (4.127)
(2π)3 −1 (2π)3 ∂p
4.6. PERTURBED EINSTEIN EQUATIONS 139

Therefore
Z 1 Z  
i 0 dp(2π)p2 i ∂f0
k δT i = −ag dµ p k p̂i −p Θ (4.128)
−1 (2π)3 |{z} ∂p

Z ∞ Z 1
(2π)p4
∂f0 dµ
= 2akg dp 3
µΘ (4.129)
0 (2π) ∂p −1 2
| {z } | {z }
R (2π)4p3 iΘ1
−2akg dp f0
(2π)3
 Z 
(4π)p2
= −4ak g dp pf0 (iΘ1 ) = −4iak ρ̄Θ1 (4.130)
(2π)3
| {z }
R d3 p
g pf0 =ρ̄
(2π)3

So in general,

k i δT 0 i = ak ρ̄(vi − 4iΘ1 ) (4.131)

So the dipole term −4iΘ1 plays the role of a non-relativistic fluid velocity for radiation.

4.6 Perturbed Einstein Equations


Using the derived expressions for Rµν and R we may write down the perturbed Einstein Tensor in
Conformal Newtonian Gauge:
1
Gµν = Rµν − gµν R (4.132)
2
so that for the (00) component
1
G00 = g 00 G00 = g 00 R00 − g 00 g00 R (4.133)
2
and for the perturbed part in first order:

2 2  
δG00 = ∇ Φ − 6H Φ̇ − HΨ (4.134)
a2

In terms of conformal time η, such that dt = adη, whose derivative we denote with a prime, i.e.
′ = d/dη:
 
2 6 a′ a′
δG00 = 2 ∇2 Φ − 2 ′
Φ − Ψ (4.135)
a a a a

Since δT 00 = −δρ, we have the perturbed Einstein Equation

δG00 = 8πGδT 00
(4.135)
140 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

and we obtain the generalized Poisson equation:


 
2 a′ a′
∇ Φ−3 ′
Φ − Ψ = −4πGa2 δρ (4.136)
a a

Obviously, this reduces to the usual Poisson equation ∇2 Φ = −4πGδρ for a non-expanding
universe (a′ = 0 and a = 1).
In Fourier space, this equation becomes

 
2 a′ a′
−k Φ − 3 ′
Φ − Ψ = −4πGa2 δρ (4.137)
a a

For the 0i component

1
G0i = R0i − g0i R = R0i
2
(4.137)

so

G0i = g 00 G0i = g 00 R0i = −R0i (4.138)

and

δG0i = −R0i = 2∂i (Φ̇ − HΨ) (4.139)

and the Einstein Equation becomes, with δT 0i = (ρ + P )vi :

δG0i = 8πGδT 0i
→ ∂i (Φ̇ − HΨ) = 4πGa(ρ + P )vi (4.139)

Multiplying both sides by ∂ i , we get

∇2 (Φ̇ − HΨ) = 4πGa(ρ + P )∂ i vi (4.140)

In terms of conformal time, we have

a′
∇2 (Φ′ − Ψ) = 4πGa2 (ρ + P )∂ i vi (4.141)
a

In Fourier space, this equation becomes

a′
−k 2 (Φ′ − Ψ) = 4πGa2 (ρ + P )(ik i vi ) (4.142)
a

For the (ij) component


4.6. PERTURBED EINSTEIN EQUATIONS 141

1
Gij = Rij − gij R
2
(4.142)

so
1
Gi j = g ik Gkj = g ik Rkj − g ik gkj R (4.143)
2
and we find for the perturbed part
   
i ä 1 2 1
δG j = δij 2 2 + H Ψ + 2H(Ψ̇ − 3Φ̇) − 2Φ̈ + 2 ∇ (Ψ + Φ) − 2 ∂ i ∂j (Ψ + Φ)
2
(4.144)
a a a

Let us first consider the case i = j. Using δii = 3, we have


   
i ä 2 1 2
δG i = −6 Φ̈ − H(Ψ̇ − 3Φ̇) − 2 + H Ψ − 2 ∇ (Ψ + Φ) (4.145)
a 3a

Using δT ii = 3δP , we can get another Einstein Equation:

δGi i = 8πGδT ii
(4.145)

which becomes
 
ä 1
Φ̈ − H(Ψ̇ − 3Φ̇) − 2 + H Ψ − 2 ∇2 (Ψ + Φ) = −4πGδP
2
(4.146)
a 3a

We can express this equation in terms of conformal time derivatives. First derivatives simply
change by the addition of a factor of a, since dt = a dη. However one has to be careful about
second derivatives. For instance, for the third term we need:

da 1 da a′
ȧ = = =
dt a dη a
2
d a 1 da da 1 d da
ä = =− 2 +
dt2 a dt dη a dt dη
   
1 1 da 2 1 d2 a 1 ′′ (a′ )2
= − 2 + 2 2 = 2 a − (4.145)
a a dη a dη a a
so
"  2 #
ä 1 a′′ a′
= −
a a2 a a
 2  
ȧ 1 a′ 2
=
a a2 a
142 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

and
  "  ′ 2 #
ä 1 a′′ a
2 + H2 = 2 2 − (4.144)
a a a a

For the derivatives on Φ we have


dΦ 1 dΦ Φ′
Φ̇ = = =
dt a dη a
2
 
d Φ 1 da dΦ 1 d dΦ 1 ′′ a′ ′
Φ̈ = = − + = Φ − Φ (4.144)
dt2 a2 dt dη a dt dη a2 a
Therefore the Einstein Equation becomes
"  ′ 2 #
a ′ a ′′ a 1
Φ′′ + (2Φ′ − Ψ′ ) − 2 − Ψ − ∇2 (Ψ + Φ) = −4πGa2 δP (4.145)
a a a 3

In Fourier space, this is given by


"  ′ 2 #
a ′ a ′′ a k2
Φ′′ + (2Φ′ − Ψ′ ) − 2 − Ψ + (Ψ + Φ) = −4πGa2 δP (4.146)
a a a 3

Finally, for i 6= j, we have

1 i
δGi j = − ∂ ∂j (Ψ + Φ) (i 6= j) (4.147)
a2
In Fourier space, this is given by
1 i
δGi j = k kj (Ψ + Φ) (4.148)
a2
Applying the operator

1
Πi j = k̂i k̂j − δij (4.149)
3
on δGi j we get

2 2
Πi j δGi j = k (Ψ + Φ) (4.150)
3a2
Applying Πi j on δT ij we get
8ρ̄Θ2
Πi j δT ij = − (4.151)
3

= Πi j δT ij
Πi j δGi j (4.152)
 
2 2 8ρ̄Θ2
k (Ψ + Φ) = 8πG − (4.153)
3a2 3
4.6. PERTURBED EINSTEIN EQUATIONS 143

or
k 2 (Ψ + Φ) = −32πGa2 ρ̄Θ2 (4.154)
All these expressions agree with Ma and Bertschinger 98 after changing Φ → −Φ.
Summary:
k2  
Φ + 3H Φ̇ − HΨ = 4πGδρ (4.155)
a2
k 2 (Φ̇ − HΨ) = −4πG(ρ + P )ik i vi (4.156)
 
ä k2
Φ̈ − H(Ψ̇ − 3Φ̇) − 2 + H 2 Ψ + 2 (Ψ + Φ) = −4πGδP (4.157)
a 3a
k 2 (Ψ + Φ) = −32πGa2 ρ̄Θ2 (4.158)
In the limit of a non-expanding universe and static fields, we may add the first (00) Einstein
Equation to 3 times the third (ii) Einstein equation to obtain, in physical coordinates (∇2 /a2 →
∇2 ):

∇2 (Ψ − Φ)
∇2 Φ − 3 = −4πGδρ − 4πG3δP (4.159)
3
or

∇2 Ψ = 4πG(δρ + 3δP ) (4.160)


So, in General Relativity, pressure is also a source to the gravitational potential Ψ.
The Einstein equations can also be expressed as
Rµν = 4πG (2Tµν − T gµν ) (4.161)
whose perturbed version is

δRµν = 4πG 2δTµν − δT gµν − T̄ δgµν (4.162)
and since T = g µν Tµν = T µµ = −ρ + 3P , the 00 component of the perturbed equation yields
R00 = 4πG (2δT00 − δT g00 − T δg00 )
1 2
∇ Ψ + 3H(Ψ̇ − 2Φ̇) − 3Φ̈ = 4πG [2δρ − (−δρ + 3δP )(−1) − (−ρ + 3P )(2Ψ)]
a2
= 4πG(δρ + 3δP ) + 4πG(ρ − 3P )(2Ψ) (4.161)
or, finally, since 3(ä/a + ȧ2 /a2 ) = 4πG(ρ − 3P ) we have
 
1 2 ä 2
∇ Ψ + 3H( Ψ̇ − 2 Φ̇) − 3 Φ̈ − 6 + H Ψ = 4πG(δρ + 3δP ) (4.162)
a2 a
Because non-relativistic particles move much slower than light (δx ≪ cδt), in sub-horizon scales
(x ≪ H −1 ) spatial gradients are much larger than time derivatives
∇Ψ ≫ ȧΨ ≫ Ψ̇ (4.163)
2 2
∇ Ψ ≫ äΨ ∼ ȧ Ψ ≫ ȧΨ̇ ≫ Ψ̈ (4.164)
and, in this case,
∇2 Ψ = 4πGa2 (δρ + 3δP ) (4.165)
144 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

4.7 Newtonian Linear Perturbation Theory


Let us work within the Newtonian approach and show that the results we obtained before properly
recover the Newtonian results in the appropriate limits.
We start by writing down the hydrodynamical equations for a fluid, namely the Continuity
Equation, representing energy conservation, and the Euler Equation, representing momentum con-
servation, as well as the Poisson equation, accounting for the gravitational interaction:
∂ρ
+ ∇ · (ρv) = 0 (4.166)
∂t
∂v ∇P
+ (v · ∇)v + = −∇φ (4.167)
∂t ρ
∇2 φ = −4πGρ (4.168)

These equations have the interpretations above, and can be easily derived in Newtonian physics
by simply imposing mass conservation of the a volume element of the fluid as well as the total
change in the element velocity due to the gravitational and pressure forces (see Appendix F).
Let us first consider the case where the fluid simply follows the Hubble flow. We change from
physical r to comoving x = r/a coordinates, so that

r = ax (4.169)
v = ṙ = ȧx (4.170)
1
∇r = ∇x (4.171)
a
1 2
∇2r = ∇ (4.172)
a2 x
(4.173)

Then with ρ(x, t) = ρ̄(t), the continuity equation gives


∂ ρ̄ ρ̄
+ ∇x · v = 0
∂t a | {z }
ȧ∇x ·x=3ȧ
∂ ρ̄
→ + 3H ρ̄ = 0
∂t
→ ρ̄(t) ∝ a−3 (4.172)

Moreover, the velocity equation gives for P = P̄ (t) and φ = φ̄

∂ ȧ ȧ2 1
x + (x · ∇x ) x = − ∇x φ̄ (4.173)
∂t a a
The x component of this equation gives
 
ȧ2 ∂ ∂ ∂ ∂φ
xä + x +y +x x=−
a ∂x ∂x ∂x ∂x
1 ∂φ
→ x(ä + ȧ2 ) = −
" a ∂x
 2 #
ä ȧ 1 ∂φ 1 ∂φ
→ + =− 2 =− (4.172)
a a a x ∂x rx ∂rx
4.7. NEWTONIAN LINEAR PERTURBATION THEORY 145

Then using the Friedmann’s equations for dust


 2
ä ȧ 4πG 8πG 4πG ä
+ =− ρ̄ + ρ̄ = ρ̄ = − (4.173)
a a 3 3 3 a
and therefore
∂ φ̄ ä
= rx
∂rx a
1 ä 2
φ̄ = r + f (ry , rz ) (4.173)
2a x
Since similar results are obtained for the y and z equations we find
1 ä 2
φ̄(r, t) = r (4.174)
2a
Finally, we may check that the Poisson equation is satisfied, since
 
2 ∂ φ̄ ä ä
∇ φ̄ = ∇ · (∇φ̄) = ∇ · r̂ = ∇ · r} = 3 = −4πGρ̄ (4.175)
∂r a | {z a
3

Now we consider perturbations around this solution that follows the Hubble flow. For instance,
the velocity perturbation may arise due to peculiar velocities induced by time-changes in the co-
moving coordinates:

v = ṙ = ȧx + |{z}
aẋ (4.176)
u(x,t)

So we introduce

ρ(x, t) = ρ̄(t) + δρ(x, t) (4.177)


P (x, t) = P̄ (t) + δP (x, t) (4.178)
v(x, t) = ȧx + u(x, t) (4.179)
φ(x, t) = φ̄(r, t) + Φ(x, t) (4.180)

where the second terms are perturbations. Using δρ = ρ̄δ(x, t), introducing these expansions into
the original equations, changing to comoving coordinates and keeping only first order terms, we
find
∂(ρ̄δ) ρ̄ ȧ
+ ∇ · u + ρ̄ ∇ · (δ x) = 0
∂t a a
∂ ρ̄ ∂δ ρ̄ ȧ ȧ
δ + ρ̄ + ∇ · u + ρ̄δ∇ · x + ρ̄ x · ∇δ = 0
∂t ∂t  a  a a 
∂ ρ̄ ȧ ∂δ 1 ȧ
δ + 3 ρ̄ +ρ̄ + ∇ · u + x · ∇δ = 0
∂t a ∂t a a
| {z }
0
(4.178)

so
∂δ ȧ 1
+ (x · ∇)δ + ∇ · u = 0 (4.179)
∂t a a
146 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

For the velocity, at first order, we have


∂u ȧ ȧ 1 ∇δP 1
+ (u · ∇)x + (x · ∇)u + = − ∇Φ (4.180)
∂t a a a ρ̄ a
so defining the adiabatic sound speed c2s = δP/δρ, and using (u · ∇)x = u, we have
∂u ȧ ȧ c2 1
+ (x · ∇)u + u + s ∇δ = − ∇Φ (4.181)
∂t a a a a
where above we used
 
∇δP 1 δP δρ
= ∇( δρ) = c2s ∇ = c2s ∇δ (4.182)
ρ̄ ρ̄ δρ ρ̄
|{z}
c2s

Finally, the Poisson equation becomes simply

∇2x Φ = −4πGa2 ρ̄δ (4.183)

Now the time derivatives for constant r is different from the partial time derivatives for constant
x, and must be changed. We have
     
∂f (r = ax) ∂f (r) ∂ri ∂f
= + (4.184)
∂t x ∂t r ∂t ∂ri t
|{z}
ȧxi
 
∂f (r) ȧ
= + x · ∇x f (4.185)
∂t r a
or
   
∂ ∂ ȧ
= + (x · ∇x ) (4.186)
∂t x ∂t r a
Therefore the continuity and Euler equations get simplified and we have
∂δ 1
+ ∇·u = 0 (4.187)
∂t a
∂u ȧ c2 1
+ u + s ∇δ = − ∇Φ (4.188)
∂t a a a
∇2x Φ = −4πGa2 ρ̄δ (4.189)

Differentiating the first equation with respect to time and applying the divergence ∇r = ∇x /a to
the second, we have
ȧ 1
δ̈ − ∇·u+ ∇ · u̇ = 0
a2 a
1 ȧ c2s 2 1
∇ · u̇ + 2 ∇ · u + 2
∇ δ = − 2 ∇2 Φ (4.189)
a a a a
so that subtracting the second from the first:
ȧ 1 c2 1
δ̈ − 2 ∇ · u − s2 ∇2 δ + 2 ∇ 2
Φ =0 (4.190)
a |a {z } a a |{z}
−4πGa2 ρ̄δ
−δ̇
4.7. NEWTONIAN LINEAR PERTURBATION THEORY 147

so
c2s 2
δ̈ + 2H δ̇ − ∇ δ − 4πGρ̄δ = 0 (4.191)
a2
In Fourier space:
 
c2s k 2
δ̈ + 2H δ̇ + − 4πGρ̄ δ = 0 (4.192)
a2

The Jeans length λJ = 2π/kJ is defined such that the parenthesis is zero:

c2s kJ2
= 4πGρ̄
a2
1 c2s
→ 2 = (4.192)
kJ 4πGa2 ρ̄
or
r
2π cs π
λJ = = (4.193)
kJ a Gρ̄

The physical Jeans wavelength


r
π
λph
J = aλJ = cs (4.194)
Gρ̄
1
Since for a matter-dominated Universe, a ∝ t2/3 , and ρ̄ ∝ a−3 ∼ t−2 , we have ρ̄(t) = 6πGt2
, and

λph
J ∼ cs t (4.195)

Now since H 2 = 8πGρ̄/3, we have


 
c2s k 2 3H 2
δ̈ + 2H δ̇ + − δ=0 (4.196)
a2 2

For a pressureless fluid cs = 0 and

3H 2
δ̈ + 2H δ̇ − δ=0 (4.197)
2

For matter domination, a ∼ t2/3 and H = (2/3)t−1 . One can check that δ = a ∼ t2/3 and δ ∼ H(t)
are solutions. The first corresponds to the growing mode and the latter to a decaying mode. For
the growing mode:

δ = At2/3 (4.198)
2
δ̇ = At−1/3 (4.199)
3
2
δ̈ = − At−4/3 (4.200)
9
(4.201)
148 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

so the equation is satisfied:


 
2 2 2 3 4 −2 2/3 2 8 6
− At−4/3 + 2 t−1 At−1/3 − t At = − + − At−4/3 = 0 (4.202)
9 3 3 29 9 9 9

and for the decaying mode:

δ = At−1 (4.203)
−2
δ̇ = −At (4.204)
−3
δ̈ = 2At (4.205)
(4.206)

so the equation is satisfied:


 
−3 2 3 4 −2 −1 18 12 6
2At − 2 t−1 At−2 − t At = − − At−3 = 0 (4.207)
3 29 9 9 9

so the solution is

δ(t) = At2/3 + Bt−1 (4.208)

Finally, notice that in terms of conformal time adη = dt, we have


 
ȧ a′ δ′ 1 ′′ a′ ′
H = = 2 and δ̇ = and δ̈ = 2 δ − δ (4.209)
a a a a a
so
 
1 a′
δ̈ + 2H δ̇ = 2 ′′
δ + δ′ (4.210)
a a

and with Ψ = −Φ, Eq. 4.197 becomes

a′ ′
δ ′′ + δ = −k 2 Ψ (4.211)
a
which can be compared to Eq. 4.89 for static fields (Φ′ = Φ′′ = 0).

4.8 Tensor Perturbations


We have been considering only scalar perturbations thus far. Now we consider tensor perturbations.

ds2 = −dt2 + a2 (t)[δij + Hij ] dl2 (4.212)


| {z }
gij

where
 
h+ h× 0
Hij =  h× −h+ 0  (4.213)
0 0 0
4.8. TENSOR PERTURBATIONS 149

and
 
1 + h+ h× 0
gij = a2 (t)  h× 1 − h+ 0  (4.214)
0 0 1

and we must solve for two functions h+ and h× , which are the two components of a tensor Hij that
is divergenceless (k i Hij = 0), traceless (Hi i = 0) and symmetric (Hi j = Hji ). This choice assumes
perturbations in the xy plane, i.e. we choose the k vector in the z direction (notice this guarantees
k i Hij = 0).
Computing the Christoffel symbols, we find

Γ000 = 0 (4.215)
Γ00i = Γ0i0 = 0 (4.216)
a2 Ḣ ij
Γ0ij = Hgij + (4.217)
2
1
Γi 0j = Hδij + Ḣij (4.218)
2
i
Γi jk = [kk Hij + kj Hik − ki Hjk ] (4.219)
2
The Ricci tensor components are

R00 = −3 (4.220)
a
R0i = 0 (4.221)
 
ä 3 a2 k2
Rij = gij + 2H 2 + a2 H Ḣij + Ḧij + Hij (4.222)
a 2 2 2
and Ricci scalar is
 

R=6 + H2 (4.223)
a
for this metric. Notice that only Rij has perturbations to first order for tensor perturbations.
Therefore the only Einstein Equation to give a contribution to first order is

δGi j = δRi j = 8πGT ij (4.224)

Keeping first order terms, using g ik gkj = δ i j , the left hand side becomes
 
i ik 3 1 k2
δG j =δ H Ḣkj + Ḧkj + 2 Hkj (4.225)
2 2 2a
Now since
 
3 1 k2
δG11 = H Ḣ11 + Ḧ11 + H11 (4.226)
2 2 2a2
 
3 1 k2
δG22 = H Ḣ22 + Ḧ22 + H22 (4.227)
2 2 2a2
(4.228)
150 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

and since H11 = −H22 = h+ , we can take the difference

k2
δG11 − δG22 = 3H ḣ+ + ḧ+ + h+ (4.229)
a2

Change to conformal time ḣ+ = h′+ /a and ḧ+ = h′′+ /a2 − (a′ /a)h′+ /a2 , and

  a′
a2 δG11 − δG22 = h′′+ + 2 + k 2 h+ (4.230)
a
One can show (exercise) that T 11 = T 22 and therefore

a′ ′
h′′+ + 2 h + k 2 h+ = 0, (4.231)
a +
Considering the 12 component, we can show (exercise) h× satisfies the exact same equation and
therefore:
a′ ′
h′′α + 2 h + k 2 hα = 0, α = +, × (4.232)
a α
The
√ solutions to these equations are called gravity waves. During radiation domination, we have
a = Ωr η and a′ /a = 1/η so assuming a solution hα = ebη , we find
2
b2 + b + k 2 = 0 (4.233)
η
whose solution is
q
− η2 ± 4
η2
− 4k 2 1 p 
b= =− 1 ± 1 − k2 η2 (4.234)
2 η
Let us consider the solutions at sufficiently early times, i.e. kη ≪ 1, so that
1 2
b = − (1 ± 1) = 0, − (4.235)
η η
both of which implies hα ∝ const. On the other for kη ≫ 1, we have
1
b = − (1 ± ikη) (4.236)
η

so we have oscillations at sufficiently late times (large η) or for sufficiently small scales (large k).

4.9 Decomposition
Perturbations can be decomposed into scalar, vector and tensor perturbations. We have just seen
scalar perturbations in conformal Newtonian gauge and also tensor perturbations. Vector pertur-
bations can be shown to decay fast and are not important for our studies. Scalar perturbations are
the most important since they are sourced and source density perturbations. Tensor perturbations
are smaller, but still important, specially at large scales, because these modes enter the horizon
later, and decay much later. They affect the CMB spectrum on large scales for instance.
4.10. GAUGE TRANSFORMATIONS 151

Figure 4.1: Evolution of gravity waves h as a function of conformal time η, scaled to it present valeu η0 .
At sufficiently early times all modes are constant. Small scales kη0 ≫ 1 decay and oscillate earlier. Larger
scales remain constant for a longer time. (Dodelson).

It turns out that these scalar, vector and tensor perturbations evolve independently and do not
couple. Let us see that specifically for 2 Einstein equations.
For G00 , since it depends on R00 and R and none carry tensor perturbations. Therefore tensor
perturbations do not affect the previous scalar perturbation results.
The only equation to worry about is Gi j . Let us consider the projection operator on δGi j as
given by tensor perturbations. Since k̂i = δi3 , we have
    
j 1 i 1 3 1 k2
k̂i k̂ − δij G j = δi3 δj3 − δij H Ḣij + Ḧij + 2 Hij = 0 (4.237)
3 3 2 2 2a

since H33 = 0 and δij Hij = h+ − h+ = 0. Therefore, these tensor perturbations have no effect on
the corresponding equation for scalar perturbations derived earlier.

4.10 Gauge Transformations


Let us go back to scalar perturbations only. Recall that scalar, vector and tensor perturbations
decouple. The most general scalar perturbations on the metric are

ds2 = −(1 + 2A) dt2 −2aB,i dtdxi + a2 [δij (1 + 2ψ) − 2E,ij ] dxi dxj (4.238)
| {z } | {z } | {z }
g00 2g0i gij

where A, B, ψ, E are scalar functions of space-time. Notice for instance that B,i = ∂i B is the
gradient of the scalar B and not an independent vector, which would have to appear on vector
perturbations. Similarly E,ij is not a tensor, which would have to appear on tensor perturbations.
Conformal Newtonian gauge corresponds to a choice: A = Ψ, ψ = Φ, B = E = 0.
152 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

Let us consider the relation of perturbations in different gauges, i.e. different coordinate systems.
The most general coordinate transformation is

xµ → x̃µ = xµ + dµ (xν ) (4.239)

with dµ = (ξ 0 , δ ij ξ,j ) and we take the two functions ξ 0 and ξ to be small first-order perturbations.
In components

t → t̃ = t + ξ 0 (t, x) (4.240)
i i i ij
x → x̃ = x + δ ξ,j (t, x) (4.241)

To related one gauge to another, we use the invariance of the line element:

g̃αβ (x̃)dx̃α dx̃β = gµν (x)dxµ dxν (4.242)


∂ x̃α
Since the differential on the left-hand side can be written dx̃α = µ
∂xµ dx , we have

∂ x̃α ∂ x̃β
g̃αβ (x̃) = gµν (x) (4.243)
∂xµ ∂xν
Consider the µν = 00 component on the left-hand side:
 2
∂ x̃α ∂ x̃β ∂ t̃
g̃αβ (x̃) = g̃00 (x̃) + second order terms (4.244)
∂t ∂t ∂t
 2
∂ξ 0
= −(1 + 2Ã) 1 + = −1 − 2Ã − 2ξ˙0 (4.245)
∂t

Therefore

−1 − 2Ã − 2ξ˙0 = −1 − 2A (4.246)

or
1
à = A − (ξ 0 )′ (4.247)
a
Proceeding similarly to all other components we find that under the coordinate transformations
above, the scalar perturbations change as

1
A → Ã = A − (ξ 0 )′ (4.248)
a
ψ → ψ̃ = ψ − Hξ 0 (4.249)
ξ0
B → B̃ = B − + ξ′ (4.250)
a
E → Ẽ = E + ξ (4.251)

There are 4 functions (A, ψ, B, E) which characterize scalar perturbations, but we may specify
two other functions (ξ 0 , ξ) that characterizes the gauge and can be chosen to manipulate the 4
functions. There are only 4-2=2 independent functions to choose, with no remaining gauge freedom.
4.11. CORRELATION FUNCTION AND POWER SPECTRUM 153

Bardeen identified two gauge-invariant variables:


1 ∂  
ΦA = A + a(E ′ − B) (4.252)
a ∂η
ΦH = −ψ + aH(B − E ′ ) (4.253)

For instance

Φ̃H = −ψ̃ + aH(B̃ − Ẽ ′ ) (4.254)


ξ0
= −ψ + Hξ 0 + aH[(B − + ξ ′ ) − (E ′ + ξ ′ )] (4.255)
a
= −ψ + aH(B − E ′ ) = ΦH (4.256)

It turns out that we can use these to transform the Einstein equations we derived before to
any other gauge. Those equations are valid in the conformal Newtonian gauge for ΦA = Ψ and
ΦH = −Φ. But that means they are valid in all gauges, so we just need to insert the definitions
above to obtain the Einstein equations in any other gauge.
Similarly, under a general transformation of coordinates, the energy-momentum tensor trans-
forms as
∂ x̃α ∂ x̃β
T̃αβ (x̃) = Tµν (x) (4.257)
∂xµ ∂xν
Again, Bardeen found combinations of Tµν that do not change and are gauge-invariant:

k̂ i T 0 i
v = ikB + (4.258)
(ρ + P )a
T00 3H
ǫm = −1 − + 2 ki T 0 i (4.259)
ρ k ρ

4.11 Correlation Function and Power Spectrum


4.11.1 Configuration Space
Let us consider a statistically homogeneous and isotropic background with matter density ρ̄m ,
and a local density field ρ(x) that changes at each point x in space. We define the overdensity
δρ(x) = ρ(x) − ρ̄m and the density contrast δ(x)

δρ(x)
δ(x) = (4.260)
ρ̄m
For discrete galaxy counts, the number density contrast can be similarly defined as

δng (x)
δg (x) = (4.261)
n̄g

Notice that when we go from the smooth dark matter density field δ to the discrete galaxy field
δg , we must introduce a bias b factor that connects both, δg = bδ. The bias appears because is
the galaxy field is a biased tracer of the smooth density field. Here we neglect the bias, implicitly
taking b = 1 and identifying both contrasts, i.e. δg = δ, or ng ∝ ρ.
154 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

The two-point correlation function ξ(r) is defined as the mean value in the ensemble of δ(x) at
two distinct positions,

ξ(r) ≡ hδ(x)δ(x + r)i , (4.262)

which depends only upon r ≡ |r| due to the statistical homogeneity and isotropy.
In order to understand the meaning of the correlation function, let us consider two volume
elements dV1 and dV2 separated by distance r, containing dN1 and dN2 galaxies each. If galaxies
were simply randomly distributed (according to a Poisson process) in the sky, the average number
Poisson , one in dV and one in dV would be
of pairs of galaxies dNpairs 1 2

Poisson
dNpair = hdN1 dN2 i = hn2g dV1 dV2 i = n̄2g dV1 dV2 (Poisson Distribution) (4.263)

However, since galaxies have intrinsic clustering due to the gravitational collapse, we must consider
the local numbers ng (x) = n̄g (1 + δ(x)) above to get:

dNpair = hdN1 dN2 i = hng (x)dV1 ng (x + r)dV2 i


= hn̄2g [1 + δ(x) + δ(x + r) + δ(x)δ(x + r)]dV1 dV2 i (4.263)

Since by definition the average hδ(x)i = 0, we end up with

dNpair (r) = n̄2g [1 + ξ(r)]dV1 dV2 = dNpair


Poisson
[1 + ξ(r)] (4.264)

where in the last line we used Eq. 4.263. This interpretation allows us to write an estimator for
ξ(r) as

ˆ = dNpair DD(r)
ξ(r) Poisson
−1= −1 (4.265)
dNpair RR(r)

where we used a typical notation where DD(r) represents the number of pairs separated by r in
a real catalog and RR(r) the number of pairs that would result if the distribution was random.
Other more interesting estimators exist, see e.g. Landy & Szalay 1993.
Similarly, if we fix dV1 such that n̄g dV1 = 1 contains one galaxy (call it i), dNpair above
represents the number dN of galaxies in volume dV = dV2 a distance r around galaxy i:

dN (r) = n̄g [1 + ξ(r)]dV (4.266)

For a sufficiently small dV such that n̄g dV is much smaller than unit, dP (r) = dN (r)/(n̄g V ) (i.e.
the fraction of galaxies at r relative to the total number in volume V ) represents the probability of
finding a galaxy at a distance r from a fixed (but random) galaxy:
dV
dP (r) = [1 + ξ(r)] (4.267)
V
Since this probability must integrate to 1 if one integrates over all space d3 x = dV , we have
Z Z Z
dV 1
1 = dP = [1 + ξ(r)] =1+ d3 x ξ(r) (4.268)
V V
i.e. the integral of ξ(r) over all space must be zero:
Z
d3 x ξ(r) = 0 (4.269)
4.11. CORRELATION FUNCTION AND POWER SPECTRUM 155

or similarly
Z
dr r2 ξ(r) = 0 (4.270)

Since we expect ξ(r) to be positive and decreasing with r at short distances (galaxies positively
correlated due to gravitational clustering), the above equation means that ξ(r) must eventually
become negative at some (large) r (to allow for a zero integral) and then must approach zero when
r → 0 (no causal connection between fluctuations separated by very large distances).

4.11.2 Fourier Space


In Fourier space, the overdensity field is
Z
δ̃(k) = d3 xδ(x)eik·x , (4.271)

and the reverse relation is given by


Z
d3 k
δ(x) = δ̃(k)e−ik·x . (4.272)
(2π)3

The overdensity field is evidently real, which means δ(x) = δ ∗ (x). Thus, inserting Eq. 4.272 in
Eq. 4.262 and defining the power spectrum P (k) as

hδ̃(k)δ̃ ∗ (k′ )i ≡ (2π)3 δD (k − k′ )P (k) , (4.273)

we find that the power spectrum is the Fourier transform of the correlation function, and therefore
Z Z
d3 k dφd(cos θ)dk 2
ξ(r) = 3
P (k)e −ik·x
= k P (k)e−ikr cos θ
(2π) (2π)3
Z
2π sin(kr)
= 3
dk k 2 P (k) 2 .
(2π) kr
Z
1 sin(kr)
= 2
dk k 2 P (k) . (4.272)
2π kr
where we took k = kẑ to use spherical coordinates and perform the angular part of the integral at
the second to last step. Due to isotropy, P (k) depends only on k = |k|.
For r = 0, the correlation function becomes the variance of the density field, σ 2 = ξ(r = 0),
which can be expressed as
Z Z  
2 1 2 dk k 3 P (k)
σ = 2 dk k P (k) = . (4.273)
2π k 2π 2

An interesting statistical quantity is ∆2 , defined as

k 3 P (k)
∆2 ≡ , (4.274)
2π 2
which measures the power per logarithmic scale. Scales in which non-linear effects become impor-
tant are defined as those in which ∆ & 1.
156 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

For a gaussian density field, the power spectrum contains all the statistical information. How-
ever, gravitational collapse causes an initially gaussian distribution to develop a skewness. That
happens because hδ(x)i = 0 and, although δ(x) may assume arbitrarily large values (e.g. inside
galaxy clusters, black holes, etc.), it has a minimum value of −1 inside voids. Deviations from
gaussianity can be characterized by higher order correlations (e.g. 3-point correlations) and their
counterparts in Fourier space (bi-spectrum, etc.).
The matter power spectrum can be calculated in linear theory by evolving the coupled Einstein-
Boltzmann equations (Ma and Bertschinger 1995) for the various components in the Universe. Ad-
vanced codes such as CMBFAST (Seljak and Zaldarriaga 1996, Zaldarriaga et al. 1998, Zaldarriaga
and Seljak 2000) and CAMB (Lewis 2000) make this calculation fast and precisely.

4.11.3 Filtering and Window Function


In the galaxy cluster framework, it will be useful to consider not the field δ(x) per se, but the
field δR (x) filtered by a window-function W (x, R). Mathematicaly, the field δR (x) is given by the
convolution of δ(x) and W (x, R), i.e.
Z
δR (x) = d3 x′ δ(x′ )W (x − x′ , R) . (4.275)

The simplest window function is called Top-Hat, being given as



1 if r < R ,
W (x, R) = (4.276)
0 if r > R ,

where r = |x|. Using δR instead of δ in the previous definitions, the filtered overdensity field
variance σ(R) = ξ(r = 0, R) becomes a function of scale R as
Z
1
2
σ (R) = 2 dk k 2 P (k) |W̃ (kR)|2 , (4.276)

where W̃ (y) is the Fourier transform of the window function. Considering the Top-Hat window
function, we have
 
3 sin y
W̃ (y) = 2 − cos y . (4.277)
y y

The variance can be expressed as a function of the mass M = 4πR3 ρ̄m /3, contained in a sphere
of radius R at the background matter density ρ̄m , i.e. σ(M ) = σ(R = [3M/4π ρ̄m ]1/3 ).

4.12 Initial Conditions


In order to evolve the differential equations for δ and v for dark matter, baryons, radiation and
neutrinos as well as the metric perturbations Ψ and Φ, we need initial conditions.
We will see that these variables are all related and we will need to specify conditions for only
one of them, e.g. the gravitational potential Ψ.
Let us consider early times, such that a comoving scale of interest λ/a ∼ (ak)−1 is ”outside”
the comoving horizon DH ∼ η, i.e. k −1 ≫ η or kη ≪ 1. In practice we set terms with k to be much
smaller than terms not having k in the perturbation equations.
4.12. INITIAL CONDITIONS 157

Recall that we decomposed Θ(k, µ, η) as

X
Θ(k, µ, η) = (−i)l (2l + 1)Θl (k, η)Pl (µ) = Θ0 − 3iµΘ1 + ... (4.278)
l

such that
Z 1
1 dµ
Θl (k, η) = Pl (µ)Θ(k, µ, η) (4.279)
(−i)l −1 2
The equation for photons

 
1
Θ′ + ikµΘ +Φ′ + ikµΨ′ = −τ ′ Θ0 − Θ + µvb − P2 (µ)Θ2 (4.280)
| {z } | {z } 2
∼0 ∼0

At those very early times, the perturbations were very small, so an observer would see a nearly
uniform sky, so higher moments are negligible compared to the monopole: (Θ1 , Θ2 , ...) ≪ Θ0 . Then

Θ′0 + Φ′ = −τ ′ [Θ0 − Θ + µvb ] (4.281)

Also, Compton scattering is very strong early on, so that τ ′ is large, driving Θ0 − Θ + µvb ∼ 0
in order to the equation valid. Since Θ ≈ Θ0 − 3iµΘ1 or similarly Θ0 − Θ ∼ 3iµΘ1 we have

3iΘ1 = −vb (4.282)

and then the equation for photons reduce to

Θ′0 + Φ′ = 0 (4.283)

A similar analysis for neutrinos (simplified since there’s no collision term) also leads to

N0′ + Φ′ = 0 (4.284)

Now looking at the overdensity equations for cold dark matter and baryons:

ikv = −3Φ′
δ ′ + |{z} (4.285)
∼0
δb′ + ikvb = −3Φ′ (4.286)
|{z}
∼0

gives

δ ′ = −3Φ′ (4.287)
δb′ = −3Φ ′
(4.288)

Looking at the equations for the velocities:


a′
v′ + v = −ikΨ (4.289)
a
a′ τ′
vb′ + vb = −ikΨ + (vb + 3iΘ1 ) (4.290)
a R
158 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

we already concluded that vb = −3iΘ1 , which turns both equations equal and therefore

v = vb = −3iΘ1 = −3iN1 (4.291)

If we now consider the perturbed Einstein equation at early times (dominated by radiation:
photons+neutrinos):
 
2 a′ a′

| {z Φ} −3 a Φ − a Ψ
−k = −4πGa2 δρr (4.292)
≈0

Using δργ = ρ̄γ δγ = 4ρ̄γ Θ0 and similarly for neutrinos, this equation at early times becomes
 
a′ a′
3 ′
Φ − Ψ = 16πGa2 (ρ̄γ Θ0 + ρ̄ν N0 ) (4.293)
a a

At radiation domination, (a′ /a2 )2 = H02 Ωr a−4 , so a′ = H0 Ωr . Therefore, a ∝ η and a′ /a =
1/η. The Einstein equation then becomes
 
1 Ψ 16πG 2
Φ′ − = a (ρ̄γ Θ0 + ρ̄ν N0 ) (4.294)
η η 3

Multiplying and dividing by ρ̄r = ρ̄γ + ρ̄ν , we have


 
Φ′ Ψ 16πGρ̄r 2 ρ̄γ ρ̄ν
− 2 = a Θ 0 + N0 (4.295)
η η 3 ρ̄r ρ̄r

From the Friedmann equation, (a′ /a2 )2 = (8πG/3)ρ̄r , so


 2
16πGρ̄r a2 a′ 2
=2 = (4.296)
3 a η2
so
 
Φ′ Ψ 2 ρ̄γ ρ̄ν
− 2 = Θ 0 + N0 (4.297)
η η η2 ρ̄r ρ̄r

Define
ρ̄ν ρ̄ν
fν = = (4.298)
ρ̄r ρ̄γ + ρ̄ν

and notice that fν is constant (not a function of η), so we have

ηΦ′ − Ψ = 2 [(1 − fν )Θ0 + fν N0 ] (4.299)

Differentiate both sides with respect to η and use Eqs. (4.283) and (4.284):
 

ηΦ′′ + Φ′ − Ψ′ = 2 (1 − fν ) Θ′0 +fν N0′  = −2Φ′


 
(4.300)
|{z} |{z}
−Φ′ −Φ′
4.12. INITIAL CONDITIONS 159

Now from the other Einstein Equation,

k 2 (Ψ + Φ) = −32πGa2 ρ̄ Θ2 → Ψ ≈ −Φ (4.301)
|{z}
≈0

So we have

ηΦ′′ + 4Φ′ = 0 (4.302)

Proposing a solution Φ = η p , i.e. Φ′ = pη p−1 and Φ′′ = p(p − 1)η p−2 , we have

η p(p − 1)η p−2 + 4pη p−1 = 0 (4.303)


p(p − 1) + 4p = 0 (4.304)

Whose solutions are p = 0 and p = −3. But p = −3 is a decaying mode. If it is excited, it will
die out quickly and will not generate structure. Let us focus on p = 0 then. Since Φ =const, then
Φ′ = 0 and from Eq. 4.299 and Ψ = −Φ we have:

Φ = 2 [(1 − fν )Θ0 + fν N0 ] (4.305)

From Eq. 4.283, we have Θ′0 = N0′ = 0. Therefore Θ0 and N0 are both constant in time as well.
Let us assume that whatever caused the perturbations did not distinguish between photons and
neutrinos, so that they are initially equal (true for adiabatic perturbations, see § 4.12.1 below):

Θ 0 = N0 (4.306)

Therefore

Φ = 2 [(1 − fν )Θ0 + fν Θ0 ] = 2Θ0 (4.307)

Now combining Eq. 4.283 and Eq. 4.287 we have

δ ′ = −3Φ′ = 3Θ′0 (4.308)

Therefore

δ = 3Θ0 + C (4.309)

and an identical equation for δb .


Perturbations are divided into those in which C = 0 (adiabatic, see § 4.12.1) and C 6= 0
(isocurvature). CMB data seems to indicate that perturbations in the universe are mainly adiabatic,
with negligible (if any) amounts of isocurvature perturbations. For adiabatic perturbations then:

δ = 3Θ0 (4.310)

Since δγ = 4Θ0 , we have


3
δ = 3Θ0 = δγ (4.311)
4
Finally, multiplying the velocity equation for cdm by a = Aη, we get

av ′ + a′ v = −ikΨa (4.312)

(av) = −ikΨAη (4.313)
160 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

Integrating (Ψ ∼const.)

η2
a v = −ikΨA (4.314)
|{z} 2

or
η
v = −ikΨ (4.315)
2
This can be written for θ = ikv as in Ma & Bertschinger (1995):
1
θ = (k 2 η)Ψ (4.316)
2
or similary, since a′ = aȧ, we have η = a/a′ = 1/ȧ = (aH)−1 and Ψ = −Φ and then
kΨη kΨ kΦ
iv = = =− (4.317)
2 2aH 2aH
or
ivb iv kΦ
Θ 1 = N1 = = =− . (4.318)
3 3 6aH

4.12.1 Adiabatic Perturbations


As seen above, perturbations are divided into those in which C = 0 (adiabatic perturbations are
those in which C = 0. Let us see what this means in a couple of different ways.

Constant number density ratio


One way to understand adiabatic perturbations is that they have a constant matter to radiation
number ratio in every point and time
ndm
= const. (4.319)

Since ρdm = ρ̄dm (1 + δ) and ρdm = mndm , we have

ndm = n̄dm (1 + δ) (4.320)

What about nγ ? Let us write

nγ = n̄γ + δnγ = n̄γ (1 + δnγ /n̄γ ) (4.321)

where
Z Z Z   Z Z
d3 p dp (2π)p2 ∂f0 dp (2π)p3 ∂f0 dµ
δnγ = δf = dµ −p Θ = −2 Θ = 3n̄γ Θ0 (4.322)
(2π)3 (2π)3 ∂p (2π) 3 ∂p 2
| {z } | {z }
R dp (4π)p2 Θ0
3 f0 =3n̄γ
(2π)3

Therefore

nγ = n̄γ (1 + 3Θ0 ) (4.323)


4.12. INITIAL CONDITIONS 161

And the ratio becomes


ndm n̄dm (1 + δ) n̄dm
= ≈ (1 + δ − 3Θ0 ) = const. (4.324)
nγ n̄γ (1 + 3Θ0 ) n̄γ

Since n̄dm ∝ a−3 and n̄γ ∝ a−3 , the pre-factor is constant. For ratio to be constant, the parenthesis
must be equal to 1 and we must have

δ = 3Θ0 (4.325)

Since δγ = 4Θ0 , we have


3
δ = 3Θ0 = δγ (4.326)
4

Time Shift
Another way to think about adiabatic perturbations is that they can be obtained from a time shift
δt such that

δρi = ρ̇i δt (4.327)

For different species i and j, using ρ̇i = 3Hρi (1 + wi ), we have


δρi δi δj
δt = = = (4.328)
ρ̇i 3H(1 + wi ) 3H(1 + wj )

Therefore
δi δj
= (4.329)
(1 + wi ) (1 + wj )

For matter and radiation wi = 0 and wj = 1/3 and

δγ 3
δ= = δγ (4.330)
1 + 1/3 4

No Entropy Change
Recall that at the background level, we have seen that the expansion is adiabatic, since dS = 0
implies the usual energy conservation equation ρ̇ = 3H(1 + w)ρ.

Now at the perturbative level , adiabatic perturbations are those that generate no entropy fluctu-
ation. In fact, since U = ρV , we may compute the entropy change δS due to density perturbations
δρ and number perturbations δn from each component. We have

T δS = δU + P δV (4.331)
= δ(ρV ) + P δV = V δρ + (ρ + P )δV (4.332)

Therefore
T δS δV
= δρ + (ρ + P ) (4.333)
V V
162 CHAPTER 4. LINEAR PERTURBATIONS IN THE UNIVERSE

The density and pressure indicated above are obviously the total ones, i.e. sums over all species.
For simplicity, let us assume the Universe has only matter and photons. Since Pγ = (1/3)ργ , we
have
 
T δS 4 δV
= δρ + δργ + ρ + ργ (4.334)
V 3 V
 
4 δV
= ρδ + ργ δγ + ρ + ργ (4.335)
3 V
Now, for constant number Ni of particle species i, since ni = Ni /V , we have
Ni δni δV
δni = − δV → =− . (4.336)
V2 ni V
So we may replace the volume fluctuation by minus the number fluctuation of any particular species
(or even sums of them). Let us replace it by the matter number density, i.e.
δV δndm
=− = −δ (4.337)
V n̄dm
where the last equality follows from Eq. 4.320. Therefore the entropy change becomes
 
T δS 4
= ρδ + ργ δγ − ρ + ργ δ (4.338)
V 3
 
4
= ργ δγ − δ (4.339)
3
Therefore, the entropy does not change δS = 0 for
4
δγ = δ (4.340)
3
Likewise, we could have replaced the volume by the photon number density:
δV δnγ 3
=− = −3Θ0 = − δγ (4.341)
V n̄γ 4
where now we used Eq. 4.323 and 4Θ0 = δγ . In this case we would have
 
T δS 4 3
= ρδ + ργ δγ − ρ + ργ δγ (4.342)
V 3 4
 
3
= ρ δ − δγ (4.343)
4
again leading to the same result. If we include neutrinos and baryons above, a similar argument
implies Θ0 = N0 and δ = δb . In this case, we set δV /V = −δ and find
 
T δS 4 4
= ρδ + ρb δb + ργ δγ + ρν δν − ρ + ρb + ργ + ρν δ (4.344)
V 3 3
   
4 4
= ρb (δb − δ) + ργ δγ − δ + ρν δν − δ (4.345)
3 3
so for δS = 0 independently of the values of ρb , ργ and ρν , we must have
3 3
δ = δb = δγ = δν (= 3Θ0 = 3N0 ) (4.346)
4 4

You might also like