QFT in Curved Spacetime

Lecture Scripts

Daniel Baumann
1
Institute of Theoretical Physics, University of Amsterdam
2
Center for Theoretical Physics, National Taiwan University
3
Leung Center for Cosmology and Particle Astrophysics
OUTLINE
Part I: Quantum Fields and Horizons
1. QFT in Flat Spacetime
2. QFT in Curved Spacetime
3. Rindler Space and Unruh Effect

i+

ũ ṽ
II
I+ I+

~ a
i0 III I i0 kB T =
c 2π

IV
I− I−

i−

4. Black Holes and Hawking Radiation

i+ Future singularity r = 0 i+

I+
II I+
H+
on
riz
ho

i0 III I i0

I− I−
ho

IV
riz

~c3
on

kB T =
H
−

8πGM
i− Past singularity r = 0 i−
Ac3 A
S= =
4~G 4L2P

1
5. Quantum Fields in De Sitter

Future infinity (η = 0)
0

Comoving observer (r = 0)
n
zo
ri
ho
e
ur
~

t
Fu
kB T =
2πL

Pa
ts
ho
T

ri
zo
n
−π
π ψ 0

6. Euclidean Quantum Gravity

Part II: Cosmological Correlations

1. Primordial Statistics
2. Inflationary Fluctuations
3. In-In Formalism
4. Wavefunction Approach
5. Cosmological Bootstrap

2
Chapter 1.
QFT IN FLAT SPACETIME
In this chapter, we will give a quick refresher of the canonical quantization of free (non-
interacting) fields in Minkowski space. We will present this familiar material in a way that
will prepare us for the generalization to curved spacetime.

1.1 Quantum Oscillators

Quantum field theory (QFT) is the theory of coupled harmonic oscillators. As a warmup
for QFT, we therefore begin by reviewing the simple harmonic oscillator (SHO) in quantum
mechanics. For simplicity, we consider a particle in one dimension.

Consider a SHO in one dimension.

Classical dynamics:
• Lagrangian:
1 1
L = q̇ 2 − ω 2 q 2 .
2 2
• Equation of motion:
 
d ∂L ∂L
= ⇒ q̈ + ω 2 q = 0 .
dt ∂ q̇ ∂q
• Solution:
q(t) = A0 sin(ωt) + B0 cos(ωt) ⇔ q(t) = q0 eiω+α0 .
• Hamiltonian:
1 1 ∂L
H = pq̇ − L = p2 + ω 2 q 2 ⇐ p≡ = q̇ .
2 2 ∂ q̇
Canonical quantization:
• Promote classical variables q, p to quantum operators q̂, p̂
• Impose canonical commutation relation
[q̂, p̂] ≡ q̂ p̂ − p̂q̂ = i~ ≡ i .
• Introduce annihilation and creation operators:
r
~
q̂ = (â + ↠) ,
2ω
r where [â, ↠] = 1 .
~ω
p̂ = −i (â − ↠) ,
2
3
 • Hamiltonian:
1 1
Ĥ = p̂2 + ω 2 q̂ 2
2 2  
1 1
= ~ω â↠+ ↠â = ~ω N̂ + where N̂ ≡ ↠â .


,
2 2
• Ground state:
1
â|0i = 0 ⇒ E0 ≡ h0|Ĥ|0i = ~ω .
2
• Excited states:
 
1 1
|ni = √ (↠)n |0i ⇒ En ≡ hn|Ĥ|ni = ~ω n + .
n! 2

Particle creation:
As we will see later, in a curved background, the harmonic oscillators that make up a quantum
field can receive time-dependent frequencies due to their couplings to the time-dependent
spacetime curvature. To prepare for this possibility, we now consider the simple harmonic
oscillator with a time-dependent frequency ω(t).

Consider a SHO with a time-dependent frequency:

ω 2 (t)

+ω02
q̈ + ω 2 (t)q = 0
t
−Ω20
0 T

• Asymptotic solutions:

t<0: qin (t) = Ain sin(ω0 t) ,

t>T : qout (t) = Aout sin(ω0 t + α) .

• Solution in the transition region:

0<t<T : q(t) = A cosh(Ω0 t) + B sinh(Ω0 t) .

4
 • Matching at t = 0:

q(0) = qin (0) ⇒ A = 0,

ω0
q̇(0) = q̇in (0) ⇒ B= Ain .
Ω0
• Matching at t = T :

q(T ) = qout (T ) ⇒ Aout sin(ω0 T + α) = B sinh(Ω0 T ) ,

Ω0
q̇(T ) = q̇out (T ) ⇒ Aout cos(ω0 T + α) = B cosh(Ω0 T ) .
ω0
This gives
2
 
Ω
A2out = B 2 sinh2 (Ω0 T ) + 20 cosh2 (Ω0 T )
ω0
2 2
 
Ω 0 2 ω 0 2
= B 2 2 cosh (Ω0 T ) + 2 sinh (Ω0 T )
ω0 Ω0
ω02
   
2 2
= Ain 1 + 1 + 2 sinh (Ω0 T ) .
Ω0

For Ω0 T  1, we then have

s
Ain Ω0 T ω02
Aout ≈ e 1 + 2  Ain .
2 Ω0

In QFT in curved spacetime, a similar phenomenon corresponds to particle

creation.

5
1.2 From Oscillators to Fields
Consider N coupled oscillators:
" N N
# N  
1 X X 1 X
L= q̇i2 − Mij qi qj → q̃˙α2 − ωα2 q̃α2 ,
2 i=1 i,j=1
2 α=1

where we have diagonalized Mij .

The oscillators have decoupled and can be treated separately.
Consider a field φ(t, x):
Z 
1
Z
3 2 3 3
L= d x φ̇ (t, x) − d xd y M (x, y) φ(t, x)φ(t, y) .
2

Note: φx (t) ⇔ qi (t).

• Lorentz symmetry:

M (x, y) = [−∇2x + m2 ]δ(x − y) ,

and hence
 
1
Z
3 2 2 2 2
L= d x φ̇ − (∇φ) − m φ , φ̇2 − (∇φ)2 = −η µν ∂µ φ∂ν φ .
2
• Fourier expansion:

d3 k
Z
φ(t, x) = 3
φk (t)eik·x .
(2π)

Note: φ∗ (t, x) = φ(t, x) ⇒ (φk )∗ = φ−k .

• Lagrangian:  
1
Z
L= d3 k φ̇k φ̇−k − ω 2 φk φ−k ,
2
where ω 2 ≡ k2 + m2 .

6
1.3 Classical Solutions
Consider
1 1
L = − ∂µ φ∂ µ φ − m2 φ2 .
2 2

• Equation of motion:
 
d ∂L ∂L
= ⇒ (∂µ ∂ µ + m2 )φ(x) = 0 .
dxµ ∂(∂µ φ) ∂φ

• Solution:
µ
φ(xµ ) = φ0 eikµ x = φ0 e−iωt+ik·x ,
where k µ = (ω, k), with ω 2 = k2 + m2 .

• Inner product:
Z
d3 x φ∗1 ∂t φ2 − φ2 ∂t φ∗1 ,


hφ1 , φ2 i ≡ i
Σt

so that
µ µ
heik1,µ x , eik2,µ x i
Z  
3 iω1 t−ik1 ·x −iω2 t+ik2 ·x −iω2 t+ik2 ·x iω1 t−ik1 ·x
=i dx e ∂t e −e ∂t e
Σt
Z
i(ω1 −ω2 )t
= (ω2 + ω1 )e d3 x e−i(k1 −k2 )·x
Σt

= (ω2 + ω1 )ei(ω1 −ω2 )t (2π)3 δ(k1 − k2 ) .

• Orthonormal mode functions:

µ
µ eikµ x
fk (x ) = p ⇒ (fk1 , fk2 ) = δ(k1 − k2 ) .
(2π)3 2ω

• Positive-frequency solutions:
p
∂t fk = −iωfk , ω ≡ k2 + m2 > 0 .

7
 • Negative-frequency solutions:
p
∂t fk∗ = +iωfk∗ , ω≡ k2 + m2 > 0 .
Note:
hfk1 , fk∗2 i = 0 ,
hfk∗1 , fk∗2 i = −δ(k1 − k2 ) .
• Mode expansion:
Z h i
φ(t, x) = 3
d k ak fk (t, x) + a∗k fk∗ (t, x) .

• Theory in a box:
Xh i
φ(t, x) = ai fi (t, x) + a∗i fi∗ (t, x) ,
i

where i labels the quantized wave vectors k = (2π/L)n.

1.4 Canonical Quantization

• Promote φ, π ≡ ∂L/∂ φ̇ = φ̇ to operators φ̂, π̂.
• Impose canonical commutation relations:
[φ̂(t, x), φ̂(t, x0 )] = 0 ,
[π̂(t, x), π̂(t, x0 )] = 0 ,
[φ̂(t, x), π̂(t, x0 )] = iδ(x − x0 ) .
• Mode expansion:
Z h i
φ̂(t, x) = 3
d k âk fk (t, x) + â†k fk∗ (t, x) ,

where
[âk , âk0 ] = 0 ,
[â†k , â†k0 ] = 0 ,
[âk , â†k0 ] = δ(k − k0 ) .
• Hamiltonian:
Z
H= d3 x (φ̇π − L)
 
1 1 1
Z
= d3 x φ̇2 + (∇φ)2 + m2 φ2 .
2 2 2

8
 Exercise: Show that
 
1 1
Z h i Z
† †
Ĥ = d3 k âk âk + âk âk ω = d3 k N̂k + δ(0) ω ,
2 2

where N̂k ≡ â†k âk .

• Vacuum state:
âk |0i = 0 , for all k ,
• Excited states:
1
|nk i = √ (â†k )nk |0i ,
nk !
1
|n1 , n2 , · · · , nN i = √ (â†k1 )n1 (â†k2 )n2 · · · (â†kN )nN |0i .
n1 !n2 ! · · · nN !
• Lorentz boost:

t0 = γt − γv · x , x0 = γx − γvt ,

t = γt0 + γv · x0 , x = γx0 + γvt0 .

√
where γ ≡ 1 − v2.
Note that
∂xµ
∂t0 fk = ∂µ fk
∂t0
= γ(−iω)fk + γv · (ik)fk
≡ −iω 0 fk ,

where
ω 0 ≡ γω − γv · k .

⇒ Functions are eigenfunctions of ∂t and ∂t0 .

⇒ All observers agree on separation into positive- and negative-frequency.
⇒ All observers agree on definition of creation and annihilation operators.
⇒ All observers agree on definition of vacuum and excited states.
⇒ The number of particles in a given state is an invariant.
All of this is not true in curved spacetime.

9
Chapter 2.
QFT IN CURVED SPACETIME
We are now ready to attack the problem of quantum fields living on a curved spacetime.
Although much of it will be a straightforward generalization of the canonical quantization of
fields in flat spacetime, we will also encounter some conceptual subtleties. Most importantly, in
a non-stationary spacetime, there is no unique separation into positive- and negative-frequency
modes. This means that different observers might not agree on the choice of vacuum and the
associated particle excitations. What one observer might declare to be the vacuum state,
another might find to be a state full of particles. This effect becomes especially interesting in
spacetimes with horizons.

2.1 From Flat to Curved

Consider  
1 1
Z
S= d4 x − η µν ∂µ φ∂ν φ − m2 φ2 .
2 2

To extend this to curved spacetime:

• ηµν → gµν
• ∂µ → ∇µ
√
• d4 x → −g d4 x
so that
√
 
1 1
Z
S= d4 x −g − g µν ∇µ φ∇ν φ − m2 φ2 .
2 2

• Equation of motion:
g µν ∇µ ∇ν + m2 φ = 0 .

• Curvature coupling:
√
 
1
Z
4 2
δS = d x −g − ξRφ ,
2

where ξ = 0 (minimal coupling) and ξ = 1/6 (conformal coupling).

10
2.2 Classical Solutions
Given two solutions φ1 and φ2 , we define their inner product as
√
Z  
µ ∗ ∗
hφ1 , φ2 i ≡ i dΣ φ1 ∇µ φ2 − φ2 ∇µ φ1 , dΣµ = d3 x γnµ .
Σ

• Orthonormal mode functions:

hfi , fj i = δij ,
hfi∗ , fj∗ i = −δij ,
hfi , fj∗ i = 0 .

• Mode expansion:
Xh i
µ µ ∗ ∗ µ
φ(x ) = ai fi (x ) + ai fi (x ) .
i

In curved spacetime, this expansion is not unique:

Xh i
µ µ ∗ ∗ µ
φ(x ) = bi gi (x ) + bi gi (x ) .
i

In the quantum theory, this leads to an ambiguity of the vacuum state.

2.3 Canonical Quantization

√
• Promote φ, π = ∂L/∂(∂0 φ) = −gg 0µ (∂µ φ) to operators φ̂, π̂.
• Impose canonical commutation relations:

[φ̂(t, x), φ̂(t, x0 )] = 0 ,

[π̂(t, x), π̂(t, x0 )] = 0 ,
[φ̂(t, x), π̂(t, x0 )] = iδ(x − x0 ) .

• Mode expansion:
Xh i
φ̂(x ) =µ µ
âi fi (x ) + â†i fi∗ (xµ ) ,
i

where [âi , â†j ] = δij .

11
 • Vacuum:
âi |0a i = 0 , for all i .

• Alternative mode expansion:

Xh † ∗ µ
i
µ µ
φ̂(x ) = b̂i gi (x ) + b̂i gi (x ) ,
i

where [b̂i , b̂†j ] = δij .

• Alternative vacuum:
b̂i |0b i = 0 , for all i .
What we have lost in the transition from flat to curved spacetime is any reason to prefer one set
of modes over any other. In flat spacetime, we defined positive-frequency modes with respect
to the time coordinate. Although the time coordinate is not unique, since it changes under
Lorentz transformations, we have seen that the vacuum state and the number operator are
invariant under such transformations. Every inertial observer therefore agrees on the vacuum
state and the number of particles in an excited state. This is not the case in a curved spacetime.

2.4 Bogolubov Transformations

Since fi are a complete set of functions, we can write
X 
∗
gi ≡ αij fj + βij fj ,
j

where
αij = hfj , gi i ,
βij = −hfj∗ , gi i .
Conversely, we may also write
X 
fi ≡ λij gj + µij gj∗ ,
j

where
λij = hgj , fi i = hfi , gj i∗ = αji
∗
,
µij = −hgj∗ , fi i = hfi∗ , gj i = −βji ,
so that
X 
∗
fi = αji gj − βji gj∗ .
j

12
Substituting this into the mode expansion, we get
Xh † ∗ µ
i
µ µ
φ̂(x ) = âi fi (x ) + âi fi (x )
i
Xh X 
†
X i
∗ ∗ ∗ ∗
= âi αji gj − βji gj + âi αji gj − βji gj
i j j
XhX  X  i
∗ ∗ † †
= αji âi − βji âi gj + αji âi − βji âi gj∗ ,
j i i

so that X 
b̂i = ∗
αij âj − βij∗ â†j .
j

Ex: Show that [b̂i , b̂†j ] = δij implies

X
∗ ∗
δij = (αik αjk − βik βjk ) .
k

In matrix form, this is α† α − β † β = 1.

Consider the a-vacuum |0a i. How many b-particles does it contain?

h0a |N̂b,i |0a i = h0a |b̂†i b̂i |0a i

X †
 
∗ ∗ †
= h0a | αij âj − βij âj αik âk − βik âk |0a i
j,k

)h0a |âj â†k |0a i

X
∗
= (−βij )(−βik
j,k

â†k âj + δjk |0a i
X
∗
= βij βik h0a |
j,k
X
= |βij |2 ,
j

where βij = −hfj∗ , gi i.

13
2.5 Cosmological Particle Production
Consider
ds2 = −dt2 + a2 (t)δij dxi dxj
dt
Z
= a2 (η) − dη 2 + δij dxi dxj ,
 
η= .
a(t)
We will study a massive scalar field in this spacetime.
• Action:
√
 
1 1
Z
S = d4 x −g − g µν ∇µ φ∇ν φ − m2 φ2
2 2
1
Z h i
3 2 0 2 2 2 2 2
= dη d x a (φ ) − (∇φ) − m a φ .
2
Define f ≡ a(η)φ, so that

a00
   
1
Z
3 0 2 2 2 2
S= dη d x (f ) − (∇f ) − m a − f2 .
2 a

• Equation of motion:

a00
 
00 2 2 2
f −∇ f + m a − f = 0,
a
00
  
a
fk00 + k 2 + m2 a2 − fk = 0 .
a

Same as in flat space, but with a time-dependent effective mass,

a00
m2eff (η) ≡ m2 a2 − .
a

• Particle creation:

m2eff (η)

+m20

−m20
0 η0

14
Asymptotic solutions:

(in) 1
fk (η) = √ e−iωk η , η < 0,
2ωk
(out) αk −iωk (η−η0 ) βk iωk (η−η0 )
fk (η) = √ e +√ e , η > η0 ,
2ωk 2ωk
p
where ωk ≡ k 2 + m20 .

Transition region:

Ak −iΩk η Bk iΩk η
fk (η) = √ e +√ e , 0 < η < η0 ,
2Ωk 2Ωk
p
where Ωk ≡ k 2 − m20 .

Matching at η = 0:

(in)
p
fk (0) = fk (0) ⇒ Ak + Bk = Ωk /ωk ,
(in)
f˙k (0) = f˙ (0)
p
k ⇒ Ak − Bk = ωk /Ωk ,

and hence
r r !
1 Ωk ωk
Ak = + ,
2 ωk Ωk
r r !
1 Ωk ωk
Bk = − .
2 ωk Ωk

Matching at η = η0 :
r
(out) ωk  −iΩk η0 iΩk η0

fk (η0 ) = fk (η0 ) ⇒ αk + βk = Ak e + Bk e ,
Ωk
r 
(out) Ωk 
f˙k (η0 ) = f˙k (η0 ) ⇒ αk − βk = Ak e −iΩk η0
− Bk e iΩk η0
,
ωk

15
and hence
r r !2 r r !2
1 ωk Ωk 1 ωk Ωk
αk = + e−iΩk η0 − − eiΩk η0 ,
4 Ωk ωk 4 Ωk ωk
 
1 ωk Ωk
βk = − sin(Ωk η0 ) .
2 Ωk ωk

⇒ The density of produced particles is

2
m40
q 
2
nk ≡ |βk | = 4 sin k2 − m20 η0 .
|k − m40 |

Limits:
m40 2
k  m0 : nk ≈ sin (kη0 )  1 ,
k4
k  m0 : nk ≈ sinh2 (m0 η0 )  1 .

16
Chapter 3.
RINDLER SPACE AND UNRUH EFFECT
Having introduced the general formalism of QFT in curved spacetime, we are now ready to
study some applications. We will begin with the so-called “Unruh effect”, which refers to the
phenomenon that an accelerating observer in Minkokwski space sees particles with a thermal
spectrum. For simplicity, we will compute this effect for massless particles in two spacetime
dimensions. We will see, in the next chapter, that this also provides a useful toy model for
Hawking radiation.

3.1 Accelerating Observer

Consider 2D Minkowski:

ds2 = −dt2 + dx2 ≡ −dτ 2 .

• Trajectory: xµ (τ )
• Velocity: uµ ≡ dxµ /dτ ⇒ uµ uµ = −1
• Acceleration: aµ ≡ duµ /dτ ⇒ aµ aµ = α 2 ⇒ aµ uµ = 0

In the instantaneous restframe, we have uµ = (1, 0) and αµ = (0, α).

Derive xµ (τ ) for the case of constant acceleration:

• Lightcone coordinates:

u ≡ t − x, v ≡ t + x.

we get
 
2 µ ν 0 −1/2
ds = −dudv ≡ gµν dx dx , gµν ≡ ,
−1/2 0

• Trajectory:

gµν uµ uν = −1 ⇒ u̇ v̇ = 1 ,
gµν aµ aν = α2 ⇒ ü v̈ = −α2 .

17
Using u̇ = 1/v̇, we have ü = −v̈/v̇ 2 and hence
 2
v̈ d ln v̇ A ατ
= α2 =⇒ =α =⇒ v(τ ) = e +B.
v̇ dτ α

Since u̇ = 1/v̇, we also have

1 −ατ
u(τ ) = − e +C.
Aα
• Use a Lorentz transformation u → λu and v → v/λ to set A ≡ 1.
• Shift the origin, to set B = C = 0.
Hence, we get
1 1
u(τ ) = − e−ατ t(τ ) = sinh(ατ )
α α 1
⇒ ⇒ x2 (τ ) − t2 (τ ) = 2 .
1 ατ 1 α
v(τ ) = e x(τ ) = cosh(ατ )
α α

18
3.2 Rindler Space
Consider
1
t = eaξ sinh(aη) ,
a
1
x = eaξ cosh(aη) , (x > |t|) .
a

t
u v

0)
r = constant

=
F ξ = const

(u
on
riz
ho
ηt =
= constant
const

L ho R x
riz
on

P
(v
=
0)

• Trajectory of the accelerated observer:

α
η(τ ) = τ,
a
1
ξ(τ ) = ln(a/α) .
a
For an observer with α = a, we have η = τ and ξ = 0.
More generally, α = ae−aξ .

• Metric:
ds2 = e2aξ − dη 2 + dξ 2 .

which is conformally equivalent to Minkowski space.

19
3.3 Field Quantization
Consider a massless scalar field.
• Equation of motion (in the right Rindler wedge):

φ = e−2aξ (−∂η2 + ∂ξ2 )φ = 0

• Positive-frequency solutions:
1
gkR (xµ ) = √ e−iωη+ikξ , with ω = |k| .
4πω
• Global modes:

 √ 1 e−iωη+ikξ

(R) ,
gkR (xµ ) = 4πω
0 (L) .


• Complete set: Add modes in the left Rindler wedge



 0 (R) ,
L µ
gk (x ) = 1
√
 e+iωη+ikξ (L) .
4πω
Positive-frequency with respect to the Killing vector −∂η .

• Mode expansion:
Z  
R R L L
φ̂ = dk b̂k gk + b̂k gk + h.c. .

Compare to expansion in the Minkowski basis:

e−iωt+ikx
Z  
µ
φ̂ = dk âk fk + h.c. , where fk (x ) = √ .
4πω

• Minkowski vacuum: âk |0M i = 0

• Rindler vacuum: b̂R L
k |0R i = b̂k |0R i = 0

20
3.4 Unruh Effect
Consider the Minkowski vacuum.
• Accelerating observer in the right Rindler wedge will see:
Z
R† R
R
hN̂k i ≡ h0M | b̂k b̂k |0M i = dk 0 |βk,k
R 2
0| .

R ∗ R
• Task: Compute βk,k 0 = −hfk , gk 0 i.

Positive-frequency Minkowski mode functions:

(
√ µ −iωt+ikx e−iωu right moving (u = t − x) k > 0
4πω fω (x ) = e =
e−iωv left moving (v = t + x) k < 0

Positive-frequency Rindler mode functions:

(
√ R µ −iωη+ikξ e−iωU right moving (U = η − ξ)
4πω gω (x ) = e =
e−iωV left moving (V = η + ξ)

Right- and left-moving mode evolve independently.

Focus on right-moving modes.
Note that −aU = ln(−au), so that
1 1
gωR (U ) = √ e−iωU = √ ei(ω/a) ln(−au) = gωR (u) .
4πω 4πω

Bogolubov transformation:
Z  
R 0 R R ∗
gω (u) = dω αω,ω0 fω0 (u) + βω,ω0 fω0 (u)
Z ∞ 0 r r 
dω R π −iω0 u R π iω0 u
= αω,ω0 e + βω,ω0 e .
0 2π ω0 ω0

Looks like a Fourier transform.

21
Consider
∞
dω 0 −iω0 u
Z
gωR (u) = e g̃ω (ω 0 )
−∞ 2π
Z ∞ 0 Z ∞ 0
dω −iω0 u 0 dω iω0 u
= e g̃ω (ω ) + e g̃ω (−ω 0 ) ,
0 2π 0 2π
so that r
ω0
R
αω,ω 0 = g̃ω (ω 0 ) ,
rπ
ω0
R
βω,ω 0 = g̃ω (−ω 0 ) , with ω 0 > 0 .
π
Task: Compute the Fourier transform of gωR (u).
Result: Explicit computation gives
r  0 iω/a
R 1 ω ω
αω,ω0 = 0
Γ(−iω/a) eπω/2a ,
2πa ω a
r  0 iω/a
1 ω ω
R
βω,ω 0 = −
0
Γ(−iω/a) e−πω/2a .
2πa ω a

Trick: Show that

g̃ω (−ω 0 ) = −e−πω/a g̃ω (ω 0 ) =⇒ R

|αω,ω 2
0| = e
2πω/a R
|βω,ω0 |2

Then use
Z   Z
0
δ(0) = dω R
|αω,ω 0|
2
− R
|βω,ω 0|
2
= (e 2πω/a
− 1) dω 0 |βω,ω
R 2
0| ,

so that
dω 0 |βω,ω
R 2
R
hN̂ω i 0| 1
nω ≡ = = 2πω/a .
V δ(0) e −1

22
Proof: Fourier transform:
Z ∞
0
0
g̃ω (ω ) ≡ du eiω u gωR (U (u))
−∞
Z 0
1 0
=√ du eiω u ei(ω/a) ln(−au) ,
4πω −∞
where we used that −aU = ln(−au).
Defining ū = −u, we get
Z ∞
1 0
g̃ω (ω 0 ) = √ dū e−iω ū ei(ω/a) ln(aū) ,
4πω Z0
∞
0 1 0
g̃ω (−ω ) = √ dū eiω ū ei(ω/a) ln(aū) .
4πω 0
We analyze these integrals in the complex plane:

Im(ū)

g̃ω (−ω 0 )
Re(ū)
g̃ω (ω 0 )

• For g̃ω (ω 0 ), we write ū = −iy, so that

Z ∞
0
g̃ω (ω 0 ) = −i dy e−ω y ei(ω/a) ln(−iay)
0 Z ∞
0
= −i eπω/(2a) dy e−ω y ei(ω/a) ln(ay) ≡ −i eπω/(2a) Iω,ω0 ,
0

where we have used

π
ln(−iay) = ln(ay) − i .
2
23
 • For g̃ω (−ω 0 ), we write ū = iy, so that
Z ∞
0
0
g̃ω (−ω ) = i dy e−ω y ei(ω/a) ln(iay)
0 Z ∞
0
−πω/(2a)
= ie dy e−ω y ei(ω/a) ln(ay) ≡ i e−πω/(2a) Iω,ω0 ,
0

where we have used

π
ln(iay) = ln(ay) + i .
2

• Hence, we have
g̃ω (−ω 0 ) = −e−πω/a g̃ω (ω 0 )

We have found that

hN̂ω i 1 ~
 
1 a a
nω ≡ = ω/T , where T = = .
V e −1 2π kB c 2π

Exercise: At what acceleration does water boil?

Answer: a ∼ 1023 m/s2

24
3.5 Particle Detectors
We have seen that the notion of a particle is observer-dependent. To clarify this further, we
now introduce a simple model of a particle detector and use it to analyze the Unruh effect.

Consider the following model of a particle detector:

Z
g dτ m̂(τ )φ̂(x(τ )) .

• Energy eigenstates: Ĥ0 |Ei i = Ei |Ei i

• Transitions: |0i ⊗ |Ei i → |ψi ⊗ |Ej i
• Transition amplitude:
Z ∞
A(Ei → Ej ) = g dτ hEj | ⊗ hψ| m̂(τ )φ̂(x(τ )) |0i ⊗ |Ei i .
−∞

Using m̂(τ ) = eiĤ0 τ m̂(0)e−iĤ0 τ , this becomes

Z ∞
A(Ei → Ej ) = gmji dτ ei(Ej −Ei )τ hψ| φ̂(x(τ )) |0i ,
−∞

where mij ≡ hEi |m̂(0)|Ej i.

• Transition probability:
X X
P (Ei → Ej ) = |A(Ei → Ej )|2 (using |ψihψ| = 1)
ψ ψ
Z ∞
0
= g 2 |mij |2 dτ dτ 0 e−i(Ej −Ei )(τ −τ ) G(x(τ ), x(τ 0 )) ,
−∞

where we have introduced the Wightman function:

G(x(τ ), x(τ 0 )) ≡ h0| φ̂(x(τ ))φ̂(x(τ 0 )) |0i .

• Detector’s response function:

Z ∞
0
F(E) ≡ dτ dτ 0 e−iE(τ −τ ) G(x(τ ), x(τ 0 )) .
−∞

Using ∆τ ≡ τ 0 − τ and τ̄ ≡ 21 (τ 0 + τ ), and dropping the (divergent)

τ̄ -integral, we get
Z ∞
f (E) = d∆τ e−iE∆τ G(∆τ ) .
−∞

25
Exercise: Show that the Wightman function of a massless scalar field in four-
dimensional Minkowski space is
1 1
G(x, x0 ) = − .
4π (t − t − i) − (x − x0 )2
2 0 2

• Inertial observer:
∞
1 e−iEτ
Z
f (E) = 2 dτ = 0.
4π −∞ (τ − i)2

Im(τ )

iϵ
Re(τ )

⇒ No particle detection.

• Accelerating observer:

t = a−1 sinh(aτ ) ,
x = a−1 cosh(aτ ) , y = z = 0,

so that
1 a2
G(x(τ ), x(0)) = − 2
4π (sinh(aτ ) − ia)2 − (cosh(aτ ) − 1)2
1 a2
=− .
16π 2 sinh2 [ 12 a(τ − i)]

⇒ Detector’s response:
∞
a2 e−iEτ
Z
f (E) = − dτ .
16π 2 −∞ sinh2 [ 12 a(τ − i)]

26
Exercise: Show that
∞ Z ∞
1 X e−iEτ
f (E) = − 2 dτ
4π n=−∞ −∞ [τ − i + i(2π/a)n]2

Im(τ )

iϵ
Re(τ )

−i(2π/a)n

and hence
∞
" #
1 X
f (E) = − 2
× (−2πi) × (−iE) × e−iEτn
4π n=1
∞
E X −2πEn/a
= e
2π n=1

E 1
= .
2π e2πE/a − 1

⇒ Thermal spectrum of particles with T = a/2π.

27
Chapter 4.
BLACK HOLES AND HAWKING RADIATION
One of the most famous applications of QFT in curved spacetimes is to black holes. In 1974,
Stephen Hawking discovered (to everyone’s surprise) that black holes radiate and therefore
have a temperature and an entropy. This addressed the concern that black holes may violate
the second law of thermodynamics by reducing the overall entropy of the Universe when an
object falls into a black hole. By carrying an entropy, black holes can balance the apparent
loss of entropy, thereby saving the second law. In this chapter, we will compute the thermal
spectrum of Hawking radiation.

4.1 Schwarzschild Black Holes

The metric around a spherically symmetric object of mass M is
   −1
2M 2M
ds2 = − 1 − dt2 + 1 − dr2 + r2 (dθ2 + sin2 θ dϕ2 ) ,
r r

where we have used units with G ≡ 1.

What is going on at r = 0 and r = 2M ?

Singularities
To decide whether a singularity is real or not, look at scalar curvatures:

R = g µν Rµν = 0 ,
Rµν Rµν = 0 ,
µνρσ 48M 2
R Rµνρσ = .
r6
⇒ r = 0 is a real singularity.
⇒ r = 2M is just a coordinate singularity.
Nevertheless, r = 2M is an interesting place!

28
Event horizon
• Time and space interchange for r < 2M .
• Nothing escapes r < 2M .
Consider an object of mass M⊕ = 6 × 1024 kg (Earth).
This gives RS,⊕ = 2GM⊕ /c2 = 8.9 mm:

Similarly, for M = 2 × 1030 kg (Sun) ⇒ RS, ≈ 3 km.

• For ordinary planets or stars, RS  R is not part of the spacetime.
• An object with R  RS is a black hole.

Near horizon limit

Let us look at the near horizon geometry by defining

r = 2M (1 + e2κξ ) ,

with 0 < e2κξ  1.

Exericse: Show that the Schwarzschild metric becomes

ds2 = e2κξ (−dt2 + dξ 2 ) + (2M )2 dΩ2 ,

| {z } | {z }
Rindler space S2

if we define
1
κ≡ (surface gravity) .
4M

29
Kruskal coordinates
In Schwarzschild coordinates, light cones “close up” as r → 2M :
t

r
2M

• To avoid this, we define the tortoise coordinate:

 −1
2M  r 
dr∗ ≡ 1 − dr ⇒ r∗ = r + 2M ln −1 ,
r 2M

so that the metric becomes

 
2M
ds2 = 1 − − dt2 + dr∗2 + r2 dΩ2 .


r

⇒ The light cones are then like in Minkowski:

r∗
r = 2GM
r∗ = −∞

• r∗ is only defined outside the black hole.

• r = 2M has been pushed to r∗ = −∞.

30
• We also define the tortoise lightcone coordinates :

U ≡ t − r∗ , V ≡ t + r∗ ,

and then the Kruskal coordinates :

 r 
2 r∗ /2M
κu = −e−κ U κ uv = −e =− − 1 er/2M ,
=⇒ 2M
κv = eκ V u −t/2M
= −e ,
v
where κ ≡ (4M )−1 . Black hole exterior corresponds to u < 0 and v > 0.

The Schwarzschild metric then becomes

 
2M
ds2 = − 1 − dU dV + r2 dΩ2
r
2M (4M )2
 
=− 1− dudv + r2 dΩ2
r −uv
 
2M r −1
=− 1− −1 e−r/2M dudv + r2 dΩ2
r 2M
2M −r/2M
= − e dudv + r2 dΩ2 .
r
⇒ Nothing special at r = 2M !
⇒ Extend the Schwarzschild coordinates (u < 0 and v > 0) to u, v ∈ R.

We can also define

1 2 2 r  r/2M
2

T = (v + u) , T − X = (4M ) 1 − e ,
2  2M
1 T t
X = (v − u) , = tanh ,
2 X 4M

and write the metric as

2M −r/2M
ds2 = e (−dT 2 + dX 2 ) + r2 dΩ2 .
r

31
Kruskal diagram
This is the Schwarzschild spacetime in Kruskal coordinates:

u T v

sin
gu 0
lar
ity r= r = constant
r = const

II
t = const
constant
III I
X

IV

• Region I: Outside the horizon

• Region II: Black hole
• Region III: Mirror universe
• Region IV: White hole

32
PENROSE DIAGRAMS
A black hole is defined as the region of space from which light cannot escape to infinity. The
boundary of that region is the event horizon. In the Kruskal diagram, infinity is still a large
distance away. A more precise way to capture the black geometry maps the points at infinity
to a finite distance. This leads to the famous Penrose diagram.

Two-Dimensional Minkowski
Consider
v=t+x v=tan ṽ 1
ds2 = −dt2 + dx2 −−−−−−→ −dudv −−−−−−→ − dũdṽ .
u=t−x u=tan ũ cos2 ũ cos2 ṽ

The last transformation maps u, v ∈ (−∞, ∞) to ũ, ṽ ∈ (−π/2, +π/2).

The overall factor does not affect null geodesics, with ds2 = 0.
The causal structure of ds2 is therefore the same as that of ds̃2 = −dũdṽ.
The Penrose diagram of R1,1 is

i+

ũ ṽ
II
I+ I+

i0 III I i0

IV
I− I−

i−

The boundaries of the diagram are different types of infinity:

• i± : past and future timelike infinity.
• i0 : spacelike infinity.
• I ± : past and future null infinity.
An accelerating observer is cut off from part of the spacetime!

33
Four-Dimensional Minkowski
Consider
v=t+r 1
ds2 = −dt2 + dr2 + r2 dΩ2 −−−−−→ −dudv + (u − v)2 dΩ2
u=t−r 4
sin2 (ũ − ṽ) 2

v=tan ṽ 1
−−−−−→ −dũdṽ + dΩ .
u=tan ũ cos2 ũ cos2 ṽ 4

Because r ≥ 0, we have v ≥ u and hence

π π
− ≤ ũ ≤ ṽ ≤ .
2 2
To draw a two-dimensional diagram, we suppress the angular coordinates.
The Penrose diagram of R1,3 is

i+

I+
r=0

i0

I−

i−

The vertical line corresponds to r = 0 and is not a boundary of the spacetime.

A null geodesic that starts on I − will simply be reflected at the vertical line.

34
Schwarzschild Black Hole
We are ready to return to the Schwarzschild geometry:
2M −r/2M
ds2 = − e dudv + r2 dΩ2 .
r
Define
κu = tan ũ ,
κv = tan ṽ ,
so that ũ, ṽ ∈ (−π/2, +π/2).
The metric becomes
3
 
1 32M
ds2 = 2 2
− e−r/2M dũdṽ + r2 cos2 ũ cos2 ṽ dΩ2 .
cos ũ cos ṽ r

The singularity at r = 0 (or κ2 uv = 1) now is at

tan ũ tan ṽ = 1 ⇒ sin ũ sin ṽ − cos ũ cos ṽ = 0

cos(ũ + ṽ) = 0 ⇒ ũ + ṽ = ±π/2 .

The Penrose diagram of a Schwarzschild black hole is

i+ Future singularity r = 0 i+

I+
II I+
H+
on
riz
ho

i0 III I i0

I− I−
ho

IV
riz
on
H
−

i− Past singularity r = 0 i−

35
Real Black Holes
The Penrose diagram for real black hole is a hybrid of the diagrams for the
Schwarzschild geometry and that of four-dimensional Minkowski space:

r=0 i+

H I+

i0
r=0

I−

i−

36
4.2 Hawking Radiation
We are now ready to consider the quantization of fields in the Schwarzschild geometry. We
will first do this by ignoring the angular dependence of the metric and study a simple model
for a “black hole” in two dimensions, where the analysis becomes the same as that for the
Unruh effect. After that, we will describe Hawking’s original derivation which is much more
involved, but leads to the same conclusion.

Toy Model
Consider the Schwarzschild metric, but drop the angular part:
 
2M
ds2 = 1 − − dt2 + dr∗2 ,


r
 
2M
=− 1− dU dV ,
r
2M −r/2M
=− e dudv .
r
• Lightcone tortoise coordinates: U, V (play the role of Rindler coordinates)
• Lightcone Kruskal coordinates: u, v (play the role of Minkowski coordinates)
Outgoing modes in region I:
1
fω (u) = √ e−iωu ,
4πω
1 1
gω(I) (U ) = √ e−iωU = √ ei(ω/κ) ln(−κu) ,
4πω 4πω
where we used −κU = ln(−κu).
Field operator: Z  
φ̂ = dω âω fω (u) + h.c.
Z  
(I)
= dω b̂ω gω (U ) + h.c. ,

• Boulware vacuum: b̂ω |0B i = 0 (analog of the Rindler vacuum)

• Kruskal vacuum: âω |0K i = 0 (analog of the Minkowski vacuum)

37
Note: Mode functions are the same as in Rindler space if we identify κ ↔ a.
⇒ Remote observer will see particles in the Kruskal vacuum:

hN̂ω i 1
nω ≡ = 2πω/κ .
V e −1
⇒ Black hole temperature:

κ 1 ~c3 1 M
TBH = = = = 61.7 nK .
2π 8πM kB 8πGM M

Black Hole Evaporation

Stefan–Boltzmann law:
dM dE 4
=− ≈ −σTBH A,
dt dt
where σ = π 2 /60 (in Planck units).
Substituting A = 4πRS2 = 16πM 2 and TBH = (8πM )−1 , we find
 1/3
dM 1 t
=− ⇒ M (t) = M0 1 − ,
dt 15360π M 2 tBH

where the black hole lifetime is

3
G2 M03

3 M0
tBH ≡ 5120π M0 = 5120π = 6 × 1074 s .
~c 4 M

Age of the Universe: t = 13.8 billion years = 4 × 1017 s.

38
Graybody Factor
Our analysis was greatly simplified by ignoring the angular part of the metric and focusing on
an effective two-dimensional spacetime. Due to conformal invariance, the mode functions were
the same as in flat space. Solving the real problem is significantly more complicated because
the mode functions are more complicated. Here, we just sketch how the result is modified.

Let
∞ X
l
X 1
φ(t, r, θ, ϕ) = φlm (t, r) Ylm (θ, ϕ) ,
r
l=0 m=−l

where φ = 0 becomes

∂2 ∂2
 
− + Vl (r) φlm (t, r) = 0 ,
∂t2 ∂r∗2

with   
2M l(l + 1) 2M
Vl (r) ≡ 1− + 3 .
r r2 r

0.4

reflected transmitted
0.3
Vl (r∗ )

outgoing
0.2

0.1

0
−15 −10 −5 0 5 10 15
H +
r∗ I +

⇒ Modified Hawking radiation:


 1 ω  1/M ,
Γ(ω) 
nω = ω/T , Γ(ω) = A
e BH − 1  ω2
 ω  1/M ,
4π
where Γ(ω) is the graybody factor.

39
Collapsing Matter
So far, we have only considered the case of an eternal Schwarzschild black hole. However, real-
istic black holes form by the collapse of matter. This complication was included in Hawking’s
original derivation. The result is the same. (See lecture notes for a sketch.)

gω

H u0
I+

(r)
gω

I−
0
(t)
gω
v0

We define
• In-modes fω (v) on I −
• Out-modes gω (u) on I +
• Out-modes hω (u) on H
The field operator is
Z


φ̂ = dω âω fω (v) + h.c.
Z


= dω b̂ω gω (u) + ĉω hω (u) + h.c. .

To find βω,ω0 = −hfω∗ , gω0 i, we trace gω (u) backwards in time.

• Interaction with the potential leads to

gω = gω(r) + gω(t) .
(t)
• Only gω contributes to βω,ω0 .

40
• Using a “geometric optics” approximation, one finds (see Reall’s lectures)

A(v) exp[i(ω/κ) log(−κv)] v < 0 ,
gω(t) (v) ≈
0 v > 0.

• The Bogolubov coefficients are

Z 0  ω 1/2
αω,ω0 = C dv 0
eiωv exp[i(ω 0 /κ) log(−κv)] ,
−∞ ω
Z 0  ω 1/2
βω,ω0 = C dv 0
e−iωv exp[i(ω 0 /κ) log(−κv)] .
−∞ ω

• Same integrals as in the Unruh effect, so we get

|αω,ω0 |2 = e2πω/κ |βω,ω0 |2 .

• The normalization of the transmitted modes is

(t)
Γ(ω) δ(ω − ω 0 ) ≡ hgω(t) , gω0 i
Z  Z
∗ ∗ ∗
βω0 ,λ = e2πω/κ − 1


= dλ αω,λ αω0 ,λ − βω,λ dλ βω,λ βω0 ,λ ,

and the density of particles observed on I + becomes

hN̂ω i h0|b̂†ω b̂ω |0i Γ(ω)

nω ≡ = = ω/T ,
V V e BH − 1

as before.

41
4.3 Black Hole Thermodynamics
Even before the discovery of Hawking radiation, it was noticed that the classical Einstein equa-
tions lead to laws of black hole mechanics that are “analogous” to the laws of thermodynamics.
The discovery of Hawking radiation then showed that these results are not merely an analogy,
but a deep property of the quantum nature of black holes.

Black Hole Entropy

Consider a Schwarzschild black hole:
 
1 A
A = 16πM 2 =⇒ dA = 32πM dM =⇒ dM = d .
8πM 4

Comparing to dE = T dS, with T = (8πM )−1 , we identify the black hole

entropy as
A Ac3 A
SBH = = = 2 ,
4 4~G 4`P
where `P ≡ (~G/c3 )1/2 ∼ 10−35 m is the Planck length.

For the Sun, we find

S ∼ 1077 (in units where kB ≡ 1)

77
What are the microstates? (N ∼ eS = 1010 )

Laws of Black Hole Mechanics

Classical GR lead to laws of black hole mechanics that are “analogous” to the
laws of thermodynamics:
• Zeroth law
Thermodynamics: The temperature T of a body in equilibrium is constant.
Black holes: The surface gravity κ is constant on the event horizon.
⇒ κ is “analogous” to T .

42
• First law
Thermodynamics: T dS = dE − ΦdQ − ΩdJ.
κ
Black holes: dA = dM − ΦdQ − ΩdJ.
8π
⇒ A is “analogous” to S.

• Second law
Thermodynamics: dS ≥ 0.
Black holes: dA ≥ 0 (area theorem).
⇒ A is “analogous” to S.

• Third law
Thermodynamics: Temperature can never be reduced to absolute zero.
Black holes: The surface gravity cannot be reduced to zero.
⇒ Claimed counterexamples exist.

Exercise: Consider the merger of two non-rotating black holes. Show that at
most 29% of the initial rest energy can be emitted in gravitational waves.

Exercise: Show that although you can decrease the energy of a charged black
hole (by throwing a charged particle into it), its horizon area still increases, so
that the second law is satisfied.

Exercise: Show that although you can decrease the mass of a rotating black
hole (by throwing a negative-energy particle into it), its horizon area still in-
creases, so that the second law is satisfied.

43
4.4 Holography and Information
The consequences of Hawking’s discovery continue to inspire important work in quantum
gravity. In this section, we will describe some of the more recent developments (with pointers
to the relevant literature).

Holography
Note that SBH ∝ Area, not Volume.
• Holographic principle (’t Hooft and Susskind)
• AdS/CFT (Maldacena):

AdS

CFT

Microstates
What is the microscopic origin of the black hole entropy?
• Naively: N ∼ eS is in tension with the “no-hair theorem”
• Resolution: String theory (Strominger and Vafa)

Information Paradox
Quantum mechanically black holes shrink.
• How is this consistent with the second law?
• Resolution: Missing entropy is carry away by Hawking radiation.
• New problem: Information paradox.
• Information appears to be lost in black hole evaporation
• States evolve from pure to mixed (in violation of unitarity)

44
Chapter 5.
QUANTUM FIELDS IN DE SITTER SPACE
Two periods of accelerated expansion: inflation and dark energy. Both are described by a de
Sitter spacetime. Given the importance of de Sitter space in the history of our Universe, we
will now study the behavior of quantum fields in this geometry.

De Sitter space:
• Maximally symmetric spacetime of positive curvature
• Solution of the vacuum Einstein equation with Λ > 0
• Good approximation to inflation and dark-energy era

5.1 Geometry of De Sitter

D-dimensional de Sitter as a hyperboloid in (D + 1)-dimensional Minkowski:
D
X
− X02 + Xi2 = L2 ,
i=1

where L is the de Sitter radius. [For our Universe: L ∼ 1026 m]

Global Coordinates
Let
X0 = L sinh(t/L) ,
Xi = L cosh(t/L)ωi ,
where −∞ < t < ∞ and ω i ωi = 1 (unit (D − 1)-sphere).

• Induced metric:
4
X
2
ds = −dX02 ++ dXi2
i=1

= −dt2 + L2 cosh2 (t/L) dΩ2D−1 .

Sphere shrinks, reaches minimal size at t = 0 and then grows.

45
For D = 4, we write
ω1 = cos ψ ,
ω2 = sin ψ cos θ ,
ω3 = sin ψ sin θ cos φ ,
ω4 = sin ψ sin θ sin φ ,
so that
dΩ23 = dψ 2 + sin2 ψ dθ2 + sin2 θ dφ2 .


| {z }
2
≡ dΩ2
In general spacetime dimensions, we have

dΩ2D−1 = dψ 2 + sin2 ψ dΩ2D−2 ,

where 0 ≤ ψ < π.

Penrose Diagram
Define cosh(t/L) = (cos T )−1 , where −π/2 ≤ T < π/2, so that

2 L2 2 2 2 2


ds = −dT + dψ + sin ψ dΩD−2 .
cos2 T
The associated Penrose diagram is:

π Future infinity
2
n
zo
ri

Comoving observer
ho
e
ur
North pole

South pole

t
FuPa
ts
ho

T
ri
zo
n

− π2
0 ψ Past infinity π

46
Planar Coordinates
Let
r̃2 t̃/L
X0 = L sinh(t̃/L) + e ,
2L
r̃2 t̃/L
X1 = L cosh(t̃/L) − e , Xi = et̃/L x̃i .
2L
so that the induced metric becomes

ds2 = −dt̃ 2 + e2t̃/L δij dx̃i dx̃j .

• These are the natural coordinates to describe inflation.

• Note: Coordinates only cover X0 + X1 = Let̃/L > 0 (half of the spacetime).

Conformal Coordinates
It is useful to work in conformal time:
Z t̃
dt0
η≡ 0
= −Le−t̃/L ,
−∞ a(t )

with −∞ < η < 0.

The metric becomes
L2 
2 2 i j

ds = 2 − dη + δij dx dx ,
η

which is conformally flat.

Static Coordinates
Let p
X0 = L2 − r2 sinh(t/L) ,
p
X1 = L2 − r2 cosh(t/L) , Xi = rωi ,
so that
−1
r2 r2
  
ds2 = − 1 − 2 dt2 + 1 − 2 dr2 + r2 dΩD−2 ,
L L

with 0 < r < L.

47
• Analog of the “Schwarzschild coordinates” for a black hole.
• Note: Coordinates only cover the “causal patch”.

De Sitter as a Sphere
Setting X0 = −iX0,E , this embedding constraint becomes
D
X
2
X0,E + Xi2 = L2 ,
i=1

which describes a four-sphere of radius L.

Equivalently, we can write the de Sitter metric in imaginary time:
t=−iτ
ds2 = −dt2 + L2 cosh2 (t/L) dΩ2D−1 −−−−→ L2 dΨ2 + sin2 Ψ dΩ2D−1 ,


| {z }
2
τ π ≡ dΩD
where Ψ ≡ + .
L 2

Geodesics
Consider a D-dimensional sphere:

δAB X A X B = L2

• For two points X and X 0 , we define

0 δAB X A X 0 B
P (X, X ) ≡ = cos θ .
L2
• The geodesic distance between the points is

d(x, x0 ) = Lθ = L cos−1 (P ) .

We want to do something similar for de Sitter.

48
Consider D-dimensional de Sitter:

ηAB X A X B = L2

• For two points X and X 0 , we define

ηAB X A X 0 B (η − η 0 )2 − (x − x0 )2
P (X, X 0 ) ≡ = 1 + .
L2 2ηη 0
• The geodesic distance between the points is
 −1
 L cosh (P ) timelike ,
d(x, x0 ) = L cos−1 (P ) spacelike ,
0 lightlike .


We will use this in Section 5.3.

5.2 Canonical Quantization

Consider the action
√
 
1 1 1
Z
S = dD x −g − g µν ∇µ φ∇ν φ − m2 φ2 − ξRφ2 ,
2 2 2

D(D − 1)
where a2 R = (in dS).
η2
D−2 ¯
Recall: For a massless field, the conformal coupling is ξ = ≡ ξ(D).
4(D − 1)
• Equation of motion
Defining f ≡ a(D−2)/2 (η)φ, we get

¯
m2 L2 + D(D − 1)(ξ − ξ)
00 2
f −∇ f + f =0 ,
η2
 ¯
m2 L2 + D(D − 1)(ξ − ξ)
fk00 + k + 2
fk = 0 .
η2
¯ this reduces to f 00 + k 2 fk = 0.
Note: For m2 = 0 and ξ = ξ, k

49
• Mode function
√
To solve this, we define fk (η) ≡ sh(s), with s ≡ −kη.
We then get the Bessel equation:

s2 h00 + sh0 + (s2 − ν 2 )h = 0 ,

where
1 D=4
¯ −− 9
ν2 ≡ − m2 L2 − D(D − 1)(ξ − ξ) −−→ − m2 L2 − 12ξ .
4 4
The general solution then is
p h i
fk (η) = k|η| Ak Jν (k|η|) + Bk Yν (k|η|) .

where Jν and Yν are Bessel functions of the first and second kind.

Solution for a single mode is

eik·x
φk (η, x) = p a(D−2)/2 (η)fk (η) .
(2π)D−1

Exercise: Show that hφk , φk0 i = δ(k − k0 ) leads to

iπ
Ak Bk∗ − A∗k Bk = − .
2k
To derive this, you have to use
2
W [Jν (s), Yν (s)] = ,
πs
where W [f, g] ≡ f g 0 − f 0 g.

50
• Vacuum choice
Consider the limit k|η|  1:
r 
2 Ak − iBk −i(kη+Φ) Ak + iBk i(kη+Φ)

νπ π
fk → e + e , Φ≡ + .
π 2 2 2 4
⇒ Set Ak + iBk ≡ 0 to select positive-frequency modes.
⇒ Normalization becomes
π
|Ak |2 = .
4k
⇒ Solution then reads
1p h i 1p
fk (η) = π|η| Jν (k|η|) + iYν (k|η|) ≡ π|η| Hν(1) (k|η|) ,
2 2
(1)
where Hν is the Hankel function of the first kind.
⇒ This leads to the Bunch–Davies vacuum.

Check: For a massless, conformally coupled scalar, we have ν = 1/2 and

 1/2
(1) 2 i
H1/2 (k|η|) = −i e−ikη ⇒ fk (η) = − √ e−ikη .
πk|η| 2k

5.3 Wightman Function

The Wightman function is
Z
0 0
G(x, x ) ≡ h0|φ̂(x)φ̂(x )|0i = dD−1 k a−(D−2) fk (x)fk∗ (x0 ) .

Hard to compute the integral ⇒ Use a trick.

Key idea: In a dS-invariant state, we have

G(x, x0 ) = G(P (x, x0 ))

Compute the Wightman function by solving

(x + m2 + ξR) G(P (x, x0 )) = 0 .

51
Exercise: Show that

µ 1 − P2 P
∇ P ∇µ P = , ∇µ ∇ν P = −gµν .
L2 L2
Using these identities, we get

x G(P ) = −g µν ∇µ ∇ν G
= −g µν ∇µ (G 0 (P )∇ν P )
= −G 00 (P )∇µ P ∇µ P − G 0 (P )g µν ∇µ ∇ν P
P 2 − 1 00 DP 0
= G (P ) + G (P ) ,
L2 L2
and the Wightman function satisfies the hypergeometric equation

(P 2 − 1)∂P2 G + DP ∂P G + (m2 L2 + D(D − 1)ξ)G = 0 .

The solution is
 
D 1+P
G(P ) = cm,D 2 F1 h+ , h− , , ,
2 2

where r
D−1 (D − 1)2
h± ≡ ± − (m2 L2 + D(D − 1)ξ) .
2 4

Check: For a massless, conformally coupled scalar, we have

 
D D D
2 F1 , − 1, , z = (1 − z)1−D/2 ,
2 2 2
and the Wightman function becomes
(D−2)/2
Γ((D − 2)/2) ηη 0

0
G(x, x ) = (−1)(D−2)/2
4π D/2 LD−2 (∆η)2 − (∆x)2
= a(η)−(D−2)/2 GM (x, x0 ) a(η 0 )−(D−2)/2 ,

which is the correct conformal rescaling of the Wightman function in Minkowski.

52
5.4 De Sitter Temperature
Recall that the response function of a particle detector is
Z ∞
f (E) = dτ e−iEτ G(τ ) ,
−∞

where G(x(τ ), x(0)) is the Wightman function.

Consider a massless, conformally coupled scalar field in D = 4 dimensions.

• Wigthman function:
1 ηη 0
G(x, x0 ) = − .
4π 2 L2 (∆η − i)2 − (∆x)2
• Geodesic distance:

0 −1 (∆η)2 − (∆x)2
d(x, x ) = ∆τ = L cosh (P ) , with P = 1 + ,
2ηη 0
so that
−1  −1
2ηη 0
   
∆τ 2 ∆τ
= cosh −1 = 2 sinh .
(∆η)2 − (∆x)2 L 2L

• Response function:
∞
1 e−iEτ
Z
f (E) = − dτ .
16π 2 L2 −∞ sinh2 [ 2L
1
(τ − i)]

⇒ Same as for the Unruh effect if a ↔ 1/L.

• De Sitter temperature:
1
TdS = .
2πL
⇒ Our Universe: TdS ∼ 10−30 K.
⇒ Entropy: SdS = πL2 /G.

53
OUTLINE
Part II: Cosmological Correlations
7. Primordial Statistics
8. Inflationary Fluctuations
9. In-In Formalism
10. Wavefunction Approach
11. Cosmological Bootstrap

One of the biggest questions in all of science is the origin of structure in the Universe. What
created everything that we see around us? An important clue lies in the fact that the struc-
tures in the Universe aren’t distributed randomly, but displays large-scale correlations. These
correlations are a fossil record of the early universe, and by measuring them we hope to uncover
how the cosmological perturbations formed and evolved. In this part of the course, we will
give an introduction to these “cosmological correlations”.

time

Inflation Reheating CMB LSS

54
Chapter 7.
PRIMORDIAL STATISTICS
In cosmology, we only predict the statistical properties of the Universe. This means that we
can’t predict whether a galaxy will be found at a specific position x. Instead, our theories can
only tell us about the probability P (x) of finding a galaxy at x, or the conditional probability
P (x2 , x1 ) that a second galaxy is at x2 given a galaxy at x1 . Using these probabilities, we can
derive spatial correlation functions, which are the main observables in cosmology.

7.1 Gaussian Random Fields

We model the statistical properties of the Universe using “random fields”,
which are a generalization of “random variables”.

Random Variable: r
• PDF: P (r)
R
• Normalization: P (r)dr = 1
R
• Average: hri ≡ rP (r)dr
• Moments: hrn i ≡ rn P (r)dr
R

• Variance: σr2 ≡ hr2 i − hri2

1 r2
 
1
• Gaussian PDF: PG (r) = p exp − 2 , with hri ≡ 0
2πσr2 2 σr
⇒ hrodd i = 0
⇒ hreven i = f (σr )

Random Field: Φ(t, x) = random variable at every point

• PDF: P (Φ1 , · · · , ΦN ) ⇐ cannot be measured in cosmology
• Correlation functions:
Z
hΦ1 Φ2 · · · ΦN i = Φ1 Φ2 · · · ΦN P (Φ1 , · · · , ΦN ) DΦ1 · · · DΦN ,

where Φi ≡ Φ(t, xi ).

55
• Ergodic hypothesis:
Average over statistical ensemble = spatial average over one realization.
1
Z Z
hΦ(t, x)i = ΦP (Φ)dΦ = d3 x Φ(t, x) ≡ Φ̄(t) .
V
↑
ergodicity

• Symmetries:
statistically homogeneous (invariant under translations): xij ≡ xi − xj
statistically isotropic (invariant under rotations): xij ≡ |xij |

hΦ(x1 )Φ(x2 )i = ξΦ (x12 ) = ξΦ (x12 ) ,

hΦ(x1 )Φ(x2 )Φ(x3 )i = BΦ (x12 , x13 ) = BΦ (x12 , x13 , x23 ) .
↑ ↑
homogeneity isotropy

+ ergodicity:

dΩr d3 r0
Z Z
ξˆΦ (r) = Φ(r0 )Φ(r0 + r)
4π V
• Fourier space:
Reasons cosmologists like Fourier space:
1) Fourier modes “decouple” (in linear theory)
2) Correlators “diagonalize”
Fourier expansion of the field is

d3 k ik·x
Z
Φ(t, x) = e Φk (t)
(2π)3
↑
complex : Φ∗k = Φ−k

⇒ Equal-time correlators: hΦk1 Φk2 · · · ΦkN i.

56
 ⇒ Two-point function:
Z
hΦk1 Φk2 i = d3 x1 d3 x2 e−ik1 ·x1 e−ik2 ·x2 hΦ(x1 )Φ(x2 )i
Z Z
−i(k1 +k2 )·x2
3
= d x2 e d3 x12 e−ik1 ·x12 ξΦ (x12 )
| {z } | {z }
3 PΦ (k1 )
(2π) δ(k1 + k2 )
↑ ↑
homogeneity power spectrum

• Power spectrum:

hΦx1 Φx2 i ≡ ξφ (|x1 − x2 |) ⇒ hΦk1 Φk2 i = (2π)3 δ(k1 + k2 )PΦ (k1 ) ,

where
Z
PΦ (k) = d3 x e−ik·x ξΦ (x) ,
d3 k k3
Z Z Z
ξΦ (0) = 3
PΦ (k) = d ln k 2 PΦ (k) ≡ d ln k ∆2Φ (k)
(2π) 2π
↑ ↑
dimensionless
variance
power spectrum

The spectrum is scale-invariant if ∆2Φ (k) is a constant.

• Gaussian random fields:
 
1 1
PG [Φ1 , · · · , ΦN ] = N/2
√ exp − Φ1 ξij−1 Φj ,
(2π) det ξ 2

where ξij ≡ hΦi Φj i.

In Fourier space:

1 |Φ(k)|2
 
Y 1
PG [Φk ] = p exp − .
2πPΦ (k) 2 PΦ (k)
k

All higher-order correlators are then simply functions of PΦ (k).

57
7.2 Cosmological Perturbations
Write the metric and stress-energy tensor as

gµν (t, x) = ḡµν (t) + δgµν (t, x) ,

Tµν (t, x) = T̄µν (t) + δTµν (t, x) ,

and then expand ∇µ Tµν = 0 and Gµν = 8πGTµν to linear order.

Metric Perturbations
Write
n  i jo
2 2 2 i

ds = a (η) − (1 + 2A)dt + 2Bi dx dη + (1 + 2C)δij + 2Eij dx dx ,

where A, C, Bi and Eij are functions of space and (conformal) time.

[Convention: Latin indices are raised and lowered with δij : e.g. E i i = δ ij Eij .]

Define Bi ≡ ∂i B + B̂i
scalar vector : ∂ i B̂i = 0

Eij ≡ 2∂hi ∂ji E + 2∂(i Êj) + 2Êij

scalar vector tensor
∂ i Êi = 0 ∂ i Êij = Ê i i = 0

where ∂hi ∂ji E ≡ (∂i ∂j − 13 δij ∇2 )E

∂(i Êj) ≡ 12 (∂i Êj + ∂j Êi )

10 = 4 scalar modes : A, B, C, E

+ 4 vector modes : B̂i , Êi

+ 2 tensor modes : Êij

Theorem: At first order, scalars, vectors, and tensors don’t mix!

⇒ We can treat them separately.

58
We will focus mostly on scalar modes (↔ density perturbations):

2 = 4 − 2
physical coordinate transformations:
scalar modes η → η + T and xi → xi + ∂ i L

Using
∂ x̃α ∂ x̃β
gµν (x) = g̃αβ (x̃) ,
∂xµ ∂xν
one finds

A 7→ A − T 0 − HT ,
B 7→ B + T − L0 ,
1
C 7→ C − HT − ∇2 L ,
3
E 7→ E − L ,

where f 0 = ∂f /∂η and H ≡ a0 /a .

• Newtonian gauge: B = E ≡ 0
h i
2 2 2 i j
ds = a (η) − (1 + 2Ψ)dη + (1 − 2Φ)δij dx dx .

Preferred gauge for studying structure formation.

• Spatially-flat gauge: C = E = 0
h i
2 2 2 i i j
ds = a (η) (1 + 2N )dη + 2Ni dx dη + δij dx dx .

A convenient gauge for computing inflationary perturbations.

59
Matter Perturbations
We write the perturbed stress tensor as

T 0 0 = −(ρ̄ + δρ) ≡ −ρ̄(1 + δ) ← δ: density contrast

T i 0 = −(ρ̄ + P̄ )v i ≡ −q i ← q i : momentum density
T 0 i = (ρ̄ + P̄ )(vi + Bi )
T i j = (P̄ + δP )δji + Πi j ← Πi j : anisotropic stress ≈ 0

For scalar modes, we can write v i ≡ ∂ i v and Πij ≡ ∂hi ∂ji Π.

Using
∂xµ ∂ x̃β α
T µ ν (x) = T̃ β (x̃) .
∂ x̃α ∂xν
one finds

δρ 7→ δρ − ρ̄ 0 T ,
δP 7→ δP − P̄ 0 T ,
v 7→ v + L 0 .

• Curvature perturbations:
The following variables are gauge-invariant:
1 δρ
ζ = −C + ∇2 E + H 0 ,
3 ρ̄
1
R = −C + ∇2 E − H(v + B) ,
3
where
 
1 2
a2 R(3) = 4∇2 −C + ∇ E .
3

On large scales, ζ = R = const.

• Uniform density gauge: δρ = E ≡ 0 ⇒ δgij = a2 (1 − 2ζ)δij

• Comoving gauge: v + B = E ≡ 0 ⇒ δgij = a2 (1 − 2R)δij

60
7.3 Observed Correlations
Angular separation
◦ ◦ ◦
90 18 1 0.2◦ 0.1◦ 0.07◦

6000

5000
Power [µK2 ]

4000

3000

2000

1000

0
2 10 30 500 1000 1500 2000 2500
Multipole

The shape of the CMB power spectrum depends on

{ H0 , ΩK , ΩΛ , Ωm , Ωb , Ων , As e−2τ , ns } .

104 galaxy
clustering
P (k) [(Mpc/h)3 ]

103 CMB

102

10 Lyα

1
10−4 10−3 0.01 0.1 1 10
k [h/Mpc]

The shape of the matter power spectrum depends on { Ωm , σ8 }.

61
7.4 Initial Condition
In linear perturbation theory, each observable O is related to the primordial
perturbations Rk (ti ) by a transfer function:

d3 k
Z
O(x, t) = 3
TO (k; t, ti ) Rk (ti ) eik·x .
(2π)

⇒ Fluctuations can be traced back to Rk (ti ).

⇒ Four important facts:

1. Superhorizon
The fluctuations spanned superhorizon scales at photon decoupling.
⇒ They must have been created before the hot Big Bang.

2. Scale-invariant
The primordial fluctuations were scale invariant.
⇒ This is a natural prediction of inflation.

3. Adiabatic
The fluctuations were adiabatic (i.e. no fluctuations in composition).
⇒ All fluctuations are sourced by a single scalar mode:
 
1 δρ
ζ(x) = , i = dark matter, baryons, photons
3 ρ̄ + P̄ i

4. Gaussian
The primordial fluctuations were highly Gaussian:
hRRRi
< 0.1%
hRRi3/2
⇒ This a natural prediction for quantum fluctuations of free fields.
⇒ A lot of the physics of inflation is encoded in non-Gaussian correlations.

62
Chapter 8.
INFLATIONARY FLUCTUATIONS
One of the most remarkable features of inflation is that it provides a natural mechanism for
creating the primordial density fluctuations that seeded the structure in the Universe. In this
chapter, we will derive the spectrum of quantum fluctuations produced during inflation.

8.1 Basics of Inflation

Consider the action
 2
√

M 1
Z
S = d4 x −g Pl
R − g µν ∂µ φ∂ν φ − V (φ) .
2 2

• Leads to accelerated expansion (ä > 0) when the potential V (φ) is flat.
• Inflation ends when the potential steepens.
• Quantum fluctuations in the field create density perturbations.

V (ϕ)

δϕ(x)
ϕ̄(t)

ϕ
inflation ϕe reheating

Varying the inflaton action with respect to the metric gives

 
2 δSφ 1 αβ
Tµν ≡ − √ = ∂µ φ∂ν φ − gµν g ∂α φ∂β φ − V (φ) .
−g δg µν 2

63
• Assuming φ = φ(t) for the homogeneous background, we get
1
ρφ = −T 0 0 = φ̇2 + V (φ) .
2
• The Friedmann equation becomes
 2  
ȧ ρ φ 1 1
H2 ≡ = 2 = 2 φ̇2 + V (φ) ,
a 3MPl 3MPl 2

where MPl ≡ (8πG)−1/2 .

• Slow-roll inflation occurs when 21 φ̇2  V .
• During inflation, H ≈ const ⇒ a(t) ≈ eHt .

Varying the inflaton action with respect to the field gives

dV
φ̈ + 3H φ̇ = − .
dφ

• Sustained slow-roll inflation occurs when φ̈  3H φ̇.

Accelerated expansion implies a shrinking comoving Hubble radius:

d d ä
(aH)−1 = (ȧ)−1 = − 2 < 0 .
dt dt (ȧ)
This has important consequences for the evolution of perturbations:
• Small-scale perturbations are stretched to large scales.
• Quantum fluctuations become classical perturbations.
• Perturbations are frozen on super-Hubble scales.
See diagram on the next page.

64
 Scales
1
(aH)

Cl B
Superhorizon
A C
B A
C
Quantum Density l
Fluctuations Fluctuations CMB Anisotropies

In ation Hot Big Bang

Time

The shrinking Hubble radius can also be written as

d ȧH + aḢ 1
(aH)−1 = − = − (1 − ε) ,
dt (aH)2 a
where the “slow-roll parameter” is

Ḣ
fl
ε≡− .
H2
• Inflation requires ε < 1.
• Scale-invariant perturbations require ε  1.

65
8.2 Free Scalar in De Sitter
During inflation, the spacetime is described by the de Sitter metric:
1
ds2 = a2 (η) −dη 2 + dx2 ,


a(η) ≈ − .
Hη
We consider a free, massless scalar in this background.

Classical dynamics
The field f = aφ satisfies

a00
 
00 2
2
f −∇ f − f =0 =⇒ fk00 2
+ k − 2 fk = 0 .
a η

• At early times (|kη|  1):

1
fk00 + k 2 fk = 0 =⇒ fk = √ e±ikη .
2k
• At late times (|kη|  1):
2
fk00 − fk = 0 =⇒ fk = Ak η −1 +Bk η 2 .
η2
• Most general solution:
   
1 i −ikη 1 i
fk (η) = Ck √ 1− e + Dk √ 1+ eikη .
2k kη 2k kη

Canonical quantization
The field operator is

d3 k h
Z i
†
fˆ(η, x) = 3
∗
fk (η)âk + fk (η)a−k eik·x ,
(2π)
where

[âk , â†k0 ] = (2π)3 δ(k − k0 ) =⇒ fk (η) ∂η fk∗ (η) − fk∗ (η)∂η fk (η) = i .

66
Bunch–Davies vacuum
Choose the minimum energy state at early times:
 
1 1 i
lim fk (η) = √ e−ikη =⇒ fk (η) = √ 1− e−ikη .
kη→−∞ 2k 2k kη

Zero-point fluctuations
The variance of the field operator is

h|fˆ|2 i ≡ h0|fˆ(η, 0)fˆ(η, 0)|0i

d3 k d3 k 0
Z Z
∗ †
∗ †

= h0| f k (η)â −k + f k (η)â k f k 0 (η)â k 0 + f k 0 (η)â −k0 |0i
(2π)3 (2π)3
d3 k d3 k 0
Z Z
∗ †
= f (η)f 0 (η) h0|[â , â
k k k −k0 ]|0i
(2π)3 (2π)3
d3 k
Z
= 3
|fk (η)|2
(2π)
k3
Z
= d ln k 2
|fk (η)|2 .
2π
We define the (dimensionless) power spectrum as

k3
∆2f (k, η) ≡ 2 |fk (η)|2 .
2π
Substituting the Bunch–Davies mode function, we find
2 h 2
∆2f (k, η)
 
H i kη→0 H
∆2φ (k, η) = = 1 + (kη) 2
−−−−→ .
a2 (η) 2π 2π

Since H ≈ const, the spectrum is scale-invariant.

67
8.3 Curvature Perturbations
Now consider the coupled inflaton-metric fluctuations.
• ADM metric:

ds2 = −N 2 dt2 + hij (N i dt + dxi )(N j dt + dxj ) ,

so that
−N 2 + Ni N i Nj Nj
   
1 −1
gµν = , g µν = 2 ,
Ni hij N N i N 2 hij − N i N j
√ √
and −g = N h.
(3)
• Intrinsic curvature: Rij
• Extrinsic curvature:
1   1
Kij ≡ ḣij − ∇i Nj − ∇j Ni ≡ Eij .
2N N
• Four-dimensional Ricci scalar:

R = R(3) + N −2 E ij Eij − E 2 + total derivative,

where E ≡ hij Eij .

• Inflaton action:
1
Z √ h (3)
d x hN R + N −2 E ij Eij − E 2
4


S=
2 i
−2 i

2 ij
+ N φ̇ − N ∂i φ − h ∂i φ∂j φ − 2V ,

where MPl ≡ 1.
• Varying the action with respect to N and N i , we find

R(3) − 2V − hij ∂i φ∂j φ − N −2 [Eij E ij − E 2 − (φ̇ − N i ∂i φ)2 ] = 0 ,

∇i [N −1 (Eji − Eδji )] = 0 .

These are constraint equations for N and N i .

68
We will work in comoving gauge:

hij = a2 e2R(t,x) δij , N ≡ 1 + α(t, x) , Ni ≡ ∂i β(t, x) .

Ex: Show that

R(3) = −2a−2 e−2R 2∂ 2 R + (∂R)2 ,

 

Eij = a2 e2R (H + Ṙ)δij − ∂(i Nj) + 2N(i ∂j) R − Nk ∂k Rδij .

Ex: Show that

Ṙ 2 ∂ 2R 2 φ̇
2
α= , ∂ β=− +a Ṙ .
H H 2H 2

Plugging these results into the action, expanding to second order, performing
integrations by parts and using the background equations of motion, we get
 
1
Z
S2 = dt d3 x a3 ε Ṙ2 − 2 (∂i R)2 .
a

• Note that L2 ∝ ε =⇒ R is pure gauge in the dS limit (ε → 0)

• Note that R has no mass term =⇒ frozen superhorizon modes
In conformal time, we have
1
Z
3 2
h
0 2 2
i √
S2 = dη d x z (R ) − (∂i R) , z ≡ a 2ε .
2
Defining f ≡ zR, the equation of motion is

z 00 z 00
 
00 2
f −∇ f − f =0 =⇒ fk00 2
+ k − fk = 0
z z

Ex: Show that

z 00
  
1 1
= 2 2+3 ε+ κ ,
z η 2
where κ ≡ ε0 /(Hε).

69
The equation of motion becomes

ν 2 − 1/4
 
00 2 3 1
fk + k − fk = 0 , where ν ≡ + ε + κ,
η2 2 2

whose Bunch–Davies solution is

√
π√
fk (η) = −η Hν(1) (−kη) .
2

Ex: Show that

1
z(η) = z∗ (η/η∗ ) 2 −ν ,
where η∗ = −k∗−1 .

The dimensionless power spectrum of R is

k 3 |fk (η)|2
∆2R (k) = 2 2
2π z (η)
k3 1 2ν−1 π
= 2 2
(−k∗ η) (−η)|Hν(1) (−kη)|2 .
2π 2ε∗ a∗ 4
• In the late-time limit, we get

2 1 H∗2
lim |Hν(1) (−kη)|2 ≈ (−kη)−2ν =⇒ ∆2R (k) = 2 2 (k/k∗ )
3−2ν
.
kη→0 π 8π ε∗ MPl

• The scalar tilt is

d ln ∆2R
ns − 1 ≡ = −2ε∗ − κ∗ ,
d ln k
which was first measured by WMAP: ns = 0.965 ± 0.004.

70
8.4 Gravitational Waves
Consider tensor metric perturbations:

ds2 = a2 (η) − dη 2 + (δij + hij )dxi dxj ,

 

where ∂ i hij = hi i = 0.
• Expanding the Einstein–Hilbert action to second order gives
2 Z 2 Z
MPl 4 √ MPl
dη d3 x a2 (hij0 )2 − (∇hij )2 + · · · .
 
S= d x −gR =
2 8
• Use rotational symmetry to write k = (0, 0, k) and
 
f+ f× 0
MPl
√ a hij ≡  f× −f+ 0  .
2
0 0 0

• The action becomes

00
 
1 X a
Z
S= dη d3 x (fλ0 )2 − (∇fλ )2 + fλ2 .
2 a
λ=+,×

• The equation of motion for each polarization mode is

a00
 
00 2
fk + k − fk = 0 ,
a
where the effective mass can be written as
a00 ν 2 − 1/4 3
= , with ν ≈ + ε.
a η2 2
• Bunch–Davies mode function is the same as before.
• Superhorizon limit of the power spectrum:
2
k3 2 H∗2

2 2
∆h (k) = 2 × lim 2
|fk (η)| = 2 2 (k/k∗ )3−2ν .
2
aMPl kη→0 2π π MPl

• Observations are expressed in terms of the tensor-to-scalar ratio:

∆2h (k∗ )
r≡ 2 = 16ε∗ .
∆R (k∗ )

71
 Planck

0.15
P+BK
m 2 ϕ2

Tensor-to-scalar ratio (r)

0.1
0.05

)
o s(ϕ/f
1−c
R2
0

0.94 0.95 0.96 0.97 0.98 0.99 1

Primordial tilt (ns )

8.5 Primordial Non-Gaussianity

The main diagnostic of primordial non-Gaussianity is the bispectrum:

hRk1 Rk2 Rk3 i = BR (k1 , k2 , k3 ) × (2π)3 δ(k1 + k2 + k3 ) .

5 BR (k, k, k)
• amplitude: fNL ≡
18 PR2 (k)

• shape:

local equilateral flattened

• effect: new particles new interactions excited states

• Planck loc equil flat

|fNL |<5 |fNL | < 40 |fNL | < 20
constraints:

72
 equilateral folded

squeezed

e+ikη
3
ϕ̇

In the notes, we describe three methods for computing these higher-point cor-
relations:

• In-In Formalism

• Wavefunction Approach

• Cosmological Bootstrap

In the lectures, I will only have time for the last two.

73
Chapter 10.
WAVEFUNCTION APPROACH
Cosmological correlations can also be computed by means of the so-called “wavefunction of the
universe.” The wavefunction is a slightly more primitive object than correlators themselves—
the relation between them is roughly the same as that between the S-matrix and scattering
cross-sections. As a result, the wavefunction is somewhat simpler than correlators in certain
ways that we will see.

10.1 Wavefunction of the Universe

φ(x) ≡ Φ(η∗ , x) : boundary field

η∗ = 0 Ψ[φ]

Φ(η, x) : bulk field

time

The “wavefunction of the universe” is Ψ[φ] ≡ hφ(x)|Ωi.

It defines boundary correlators
Z
hφ(x1 ) · · · φ(xN )i = Dφ φ(x1 ) · · · φ(xN ) |Ψ[φ]|2 .

The perturbative expansion of the wavefunction (in momentum space) is

∞
!
1
X Z
Ψ[φ] = exp − d3 k1 · · · d3 kN ΨN (k) φk1 · · · φkN ,
N!
N =2

where the “wavefunction coefficients” are

ΨN (k) = (2π)3 δ(k1 + · · · + kN ) ψN (k)
= (2π)3 δ(k1 + · · · + kN ) hOk1 · · · OkN i0 .
↑
dual operators: φ → O, γij → Tij

74
The relation between correlators and wavefunction coefficients is
1
hφφi0 = ,
2RehOOi0
0 2RehOOOi0
hφφφi = Q3 ,
2RehO O i0
a=1 a a

hOOOOi0 hOOXi3
hφφφφi0 = + .
(hOOi4 hXXihOOi4
The wavefunction has the following path integral representation:
Z
Ψ[φ] = DΦ eiS[Φ] ≈ eiS[Φcl ] ,
Φ(η∗ )=φ ↑
Φ(−∞− )=0 tree level
−
where −∞ ≡ −∞(1 − i). [Note: Opposite i to canonical quantization!]
To find the wavefunction, we therefore need to find the classical solution for
the bulk field with the correct boundary conditions.

10.2 Warmup in Quantum Mechanics

Consider our old friend the simple harmonic oscillator:
 
1 2 1 2 2
Z
S[Φ] = dt Φ̇ − ω Φ .
2 2

• The classical solution (with the correct boundary conditions) is

Φcl (t) = φeiωt .

• The on-shell action becomes

Z t∗  
1 1 2
S[Φcl ] = dt ∂t (Φ̇cl Φcl ) − Φcl (Φ̈cl + ω Φcl )
ti 2 2 | {z }
=0
1
= Φ̇cl Φcl
2 t=t∗

iω
= φ2 .
2

75
• The wavefunction then is
 ω 
Ψ[φ] ≈ exp(iS[Φcl ]) = exp − φ2 .
2
• The quantum variance of the oscillator therefore is
1
hφ2 i = .
2ω

• In QFT, the same result applies for each Fourier mode:

1
hφk φ−k i0 = ,
2ωk
√
where ωk = k 2 + m2 .

To make this more interesting, consider a time-dependent oscillator:

 
1 1
Z
S[Φ] = dt A(t)Φ̇2 − B(t)Φ2 .
2 2
• The classical solution is
K(0) = 1
Φcl = φK(t) , with
K(−∞) ∼ eiωt
• The on-shell action becomes
Z t∗  
1 1
S[Φcl ] = dt ∂t (AΦ̇cl Φcl ) − Φcl (∂t (AΦ̇cl ) + BΦcl )
ti 2 2 | {z }
=0
1
= AΦ̇cl Φcl
2 t=t∗

1
= Aφ2 ∂t log K .
2 t=t∗

• The wavefunction then is

 
i
Ψ[φ] ≈ exp(iS[Φcl ]) = exp (A∂t log K) φ2 ,
2 ∗

which implies

1
|Ψ[φ]|2 = exp −Im(A∂t log K) ∗ φ2 hφ2 i =


=⇒ .
2 Im(A∂t log K) ∗

76
10.3 Free Fields in de Sitter
Consider a massless field in de Sitter:
Z
S = dη d3 x a2 (η) (Φ0 )2 − (∇Φ)2
 

d3 k 2
 
1 1 k
Z
= dη 3 2
Φ0k Φ0−k − Φk Φ−k ,
2 (2π) (Hη) (Hη)2

which is the same as the time-dependent oscillator.

• The classical solution is
K(0) = 1
Φcl = φK(η) , with
K(−∞) ∼ eikη

The function K(η) is the bulk-to-boundary propagator.

• For a massless field, we have

K(η) = (1 − ikη)eikη ,
log K(η) = log(1 − ikη) + ikη ,

and hence
−ik
 
1
Im(A∂η log K) η=η∗
= Im + ik
(Hη∗ )2 1 − ikη∗
 2
k η∗ + ik 3 η∗2 k3

1 η∗ →0
= Im −
−−→ ,
(Hη∗ )2 1 + k 2 η∗2 H2

• The two-point function then is

H2
hφk φ−k i0 = .
2k 3

The result for a massive field is derived in the lecture notes.

77
10.4 Anharmonic Oscillator
Consider the following anharmonic oscillator:
 
1 2 1 2 2 1 3
Z
S[Φ] = dt Φ̇ − ω Φ − λΦ .
2 2 3

• The classical equation of motion is

Φ̈cl + ω 2 Φcl = −λΦ2cl .

• A formal solution is
Z
dt0 G(t, t0 ) −λΦ2cl (t0 ) ,


Φcl (t) = φK(t) + i

where
K(t) = eiωt ,
0 1  −iω(t−t0 ) 0 iω(t−t0 ) 0 iω(t+t0 )

G(t, t ) = e θ(t − t ) + e θ(t − t) − e .
2ω

• Computing the on-shell action is now a bit more subtle.

As before, we first write

t∗  
1 1 λ 3
Z
2
S[Φ] = dt ∂t (ΦΦ̇) − Φ(Φ̈ + ω Φ) − Φ
ti 2 2 3
 
1 1 λ 3
Z
2
= ΦΦ̇ + dt − Φ(Φ̈ + ω Φ) − Φ .
2 t=t∗ 2 3

Since

lim G(t, t0 ) = 0 ,
t→0
0
lim ∂t G(t, t0 ) = −ieiωt 6= 0 ,
t→0

the boundary term is

 
1 1
Z
0 iωt0

2 0
Φcl Φ̇cl = φ iωφ − iλ dt − ie Φcl (t )
2 t=t∗ 2
iω 2 λ
Z
0
= φ − φ dt0 eiωt Φ2cl (t0 ) .
2 2

78
The action then becomes
iω 2 λ
Z
S[Φcl ] = φ − φ dt eiωt Φ2cl
2 2
   
1  λ
Z Z
iωt 0 0 2 0 2 3
+ dt − φe − iλ dt G(t, t ) Φcl (t ) − λΦcl (t) − Φcl .
2 3

The terms linear in φ cancel.

• The final on-shell action is

iω λ iλ2
Z Z
S[Φcl ] = φ2 − 3
dt Φcl (t) − dt dt0 G(t, t0 )Φ2cl (t0 )Φ2cl (t) .
2 3 2

λn Φ(n) (t), where

P
• We then write the classical solution as Φcl (t) =

Φ(0) (t) = φeiωt ,

Z  
(1) 0 0 (0) 0 2
Φ (t) = i dt G(t, t ) −(Φ (t ))
φ2
Z  
0 0 2 2iωt0
= 2 e2iωt − eiωt .


=i dt G(t, t ) −φ e
3ω
• With this, the wavefunction becomes

λ2 4
 
ω λ
Ψ[φ] ≈ eiS[Φcl ] = exp − φ2 − φ3 + φ + ··· .
2 9ω 72ω 3

From this, we can compute hφ3 i, hφ4 i, etc.

79
10.5 Wavefunction in Field Theory
Back to field theory:

4 √
 
1 µν 1 2 2 1 3
Z
S[Φ] = d x −g − g ∂µ Φ∂ν Φ − m Φ − λΦ .
2 2 3

The analysis is similar to that of the anharmonic oscillator (⇒ lecture notes).

In the interest of time, we jump directly to Feynman rules for WF coefficients:
• bulk-to-boundary propagator K for every external line
• bulk-to-bulk propagator G for every internal line
• integrate each vertex over time.

Kk (t)
Z
iλ dt a3 (t)
G(kI ; t, t0 )

Given a mode function fk (t), the bulk-to-boundary and bulk-to-bulk propaga-

tors are
fk (t)
Kk (t) = ,
fk (t∗ )
0 fk∗ (t∗ )
G(k; t, t ) = fk∗ (t)fk (t0 ) θ(t −t)+0
fk∗ (t0 )fk (t) θ(t0 − t) − fk (t)fk (t0 ) .
| {z } fk (t∗ )
= GF (k; t, t0 )

80
10.6 Example in Flat Space
The wavefunction is an interesting object even in flat space:
 
1 λ 3
Z
4 2
S = d x − (∂Φ) − Φ .
2 3!

• Evaluate correlators at t∗ ≡ 0.
• The flat-space mode function is
1
fk (t) = √ eikt .
2k
• Using this, the relevant propagators are

Kk (t) = eikt ,
0 1  −ik(t−t0 ) 0 ik(t−t0 ) 0 ik(t+t0 )

G(k; t, t ) = e θ(t − t ) + e θ(t − t) − e .
2k
We will compute the simplest tree-level correlators in this theory.

• The three-point wavefunction coefficient in Φ3 theory is

hO1 O2 O3 i0 ≡

Z 0
= iλ dt ei(k1 +k2 +k3 )t
−∞
λ
= .
k1 + k2 + k3

• This is easily generalized to N -point wavefunction coefficients:

Z 0
0
hO1 O2 . . . QN i = iλ dt ei(k1 +k2 +···+kN )t
−∞
λ
=
k1 + k2 + · · · + kN

81
• The four-point wavefunction coefficient in Φ3 theory is

hO1 O2 O3 O4 i0 ≡

Z 0
0
= −λ 2
dtdt0 eik12 t G(kI ; t, t0 ) eik34 t
−∞
2 Z 0
λ 0 ik12 t

−ikI (t−t0 ) 0 ikI (t−t0 ) 0 ikI (t+t0 )
 0
= − dtdt e e θ(t − t ) + e θ(t − t) − e eik34 t
2kI −∞

Z 0 Z t
λ2 0
= − dt dt0 ei(k12 −kI )t ei(k34 +kI )t
2kI −∞ −∞
2 Z 0 Z t0
λ 0
− dt0 dt ei(k12 +kI )t ei(k34 −kI )t
2kI −∞ −∞
2 Z 0 Z 0
λ 0
+ dt dt0 ei(k12 +kI )t ei(k34 +kI )t
2kI −∞ −∞

λ2
 
1 1 1
= + −
2kI (k12 + k34 )(k34 + kI ) (k12 + k34 )(k12 + kI ) (k12 + kI )(k34 + kI )

λ2
= .
(k12 + k34 )(k12 + kI )(k34 + kI )

The answer has an interesting singularity structure. More about this later.

Exercise: Show that

hO1 O2 O3 O4 O5 i0 ≡

λ3
 
1 1
= + .
EEL EM ER k123 + kI0 k345 + kI

82
10.7 Examples in De Sitter
Consider a massive scalar in de Sitter.
• The equation of motion of u = a(η)Φ is
2 2
−
 
m /H 2
u00 + k 2 + u=0
η2

• For generic m, the solution is

r
π − iπ (1+2ν)
uk (η) = e 4 (−η)1/2 Hν(2) (−kη) .
4

• For m2 = 2H 2 , the field is conformally coupled and

eikη uk (η) eikη

uk (η) = √ =⇒ fk (η) = = H(−η) √ ,
2k a(η) 2k
In the following, we consider this special case.

• The three-point wavefunction coefficient in Φ3 theory is

Z η∗
iλ dη iKη iλ
hO1 O2 O3 i0 = − 4 e = log(−iKη∗ ) ,
H (−η∗ )3 −∞ η H 4 η∗3

where K = k1 + k2 + k3 .
• The four-point wavefunction coefficient in Φ3 theory is

hO1 O2 O3 O4 i0 ≡

0 0
λ2 dη ik12 η dη 0 ik34 η0 1 h −ikI |η−η0 |
Z Z i
ikI (η+η 0 )
=− 6 4 e 0
e e − e .
H η∗ −∞ η −∞ η 2kI

Using
∞
1
Z
−i dx eixη = eiaη ,
a η

83
 this can be written as
2 Z ∞ Z ∞ Z 0
λ 0 0
hO1 O2 O3 O4 i0 = − 6 4 dx dx0 dη dη 0 eixη eix η G(flat) (kI ; η, η 0 )
H η∗ k12 k34 −∞
Z ∞ Z ∞
1
= 6 4 dx dx0 hO1 O2 O3 O4 i0(flat) (x, x0 , kI )
H η∗ k12 k34
∞ Z ∞
λ2 1
Z
=− 6 4 dx dx0
H η∗ k12 k34 (x + x0 )(x + kI )(x0 + kI )
Z ∞ Z ∞
λ2 1
= − 6 4 dx dx0 .
H η∗ 0 0 (E + x + x0 )(EL + x)(ER + x0 )

Evaluating the integrals gives

2
− −
    
λ 1 k12 k I k34 kI
hO1 O2 O3 O4 i0 = Li2 + Li2
2H 6 η∗4 kI E E
2
    
EL ER π
+ log log − ,
E E 6

where Li2 is the dilogarithm.

10.8 Power Law Cosmology

Consider a conformally coupled scalar in a power-law cosmology:
" #
√ 1 1 X λp
Z
S = d4 x −g − (∂Φ)2 − RΦ2 − Φp ,
2 12 p>2
p!

where the metric is ds2 = a2 (η) −dη 2 + dx2 , with

 −(1+ε)
η
a(η) = .
η0

• ε = 0 (de Sitter)
• ε ≈ 0 (inflation)
• ε = −1 (Minkowski)
• ε = −2 (radiation)
• ε = −3 (matter).

84
Performing a Weyl transformation gµν 7→ a2 gµν , Φ 7→ a−1 Φ, we get
" #
1 λp (η) p
Z X
S = d4 x − (∂Φ)2 − Φ ,
2 p>2
p!

where
λp (η) ≡ λp a(η)4−p

∝ η (p−4)(1+ε)
Z ∞
p=3
= dx xα eixη , with α ≡ (4 − p)(1 + ε) − 1 −−−→ ε .
0

Example: One-site graph

k1 k2 k3

ψ(1) =
E
Z 0
=i dη eiEη λ3 (η)
−∞
∞ ∞
xε
Z Z Z
ε i(E+x)η
∝ dx x i dη e = dx = −π csc(πε) E ε .
0 0 E+x
| {z }
(1)
ψflat (E + x)

Example: Two-site graph

k1 k2 k3 k4

ψ(2) =
E1 kI E2
Z 0
0
=− dη dη 0 eiE1 η λ3 (η) G(kI ; η, η 0 ) λ3 (η 0 ) eiE2 η ,
−∞

85
which can be written as
Z ∞ Z
0 0
0 0 ε
ψ(2) ∝ dxdx (xx ) dη dη 0 ei(E1 +x)η G(kI ; η, η 0 ) ei(E2 +x )η
0
| {z }
(2)
ψflat (E1 + x, E2 + x0 , kI )
Z ∞
1
= − dxdx0 (xx0 )ε .
0 (E + x + x0 )(EL + x)(ER + x0 )

Twisted integrals like this can be computed using the method of differential
equations [see 2312.05303].
This particular integral can be done in Mathematica:

ψ(2) = c1 (ε) (EL ER )ε

− −
    
1 , ε k34 kI 1 , ε k12 kI
+ c2 (ε) E 2ε 1 − 2 F1 − 2 F1 ,
1−ε EL 1−ε ER

where

c1 (ε) = π 2 csc2 (πε) ,

√
c2 (ε) = 4−ε π csc(πε) Γ(ε)Γ( 21 − ε) .

86
Chapter 11.
COSMOLOGICAL BOOTSTRAP
In the previous two chapters, we presented two methods for computing inflationary correlations
by evolving the fluctuations through the bulk spacetime and evaluating the resulting correlators
on the future boundary. This involved complex time integrals (one for each vertex in a diagram)
that in general cannot be computed analytically. Numerical integration is possible, but comes
at the cost of masking the physical structure of the answers. In this chapter, we will describe
a new approach to deriving the correlations directly on the boundary by imposing consistency
with unitarity, locality and symmetry assumptions.

Reheating surface

End of inflation

particle
production
particle
decay

11.1 Cosmological Collider Physics

Consider the four-point function of a conformally coupled scalar mediated
by the exchange of a massive scalar:
dη dη 0 ik12 η ik34 η0
Z Z
0
F ≡ hO1 O2 O3 O4 i = −λ 2
2 02
e e G(kI ; η 0 , η 00 ) .
η η
↑
products of Hankel functions

The time integrals cannot be performed analytically.

87
Note: The propagator satisfies
2
m2
 
2 ∂ ∂
η 2
− 2η + kI η + 2 G(kI ; η, η 0 ) = −iH 2 η 2 η 02 δ(η − η 0 ) .
2 2
(*)
∂η ∂η H

We can use this to derive a differential equation for the correlator.

Consider
dη dη 0 ik12 η ik34 η0 2 ∂ 2 m2
 
∂
Z
Fe ≡ −λ 2
2 02
e e η 2
− 2η + kI η + 2 G(kI ; η, η 0 ) .
2 2
η η ∂η ∂η H

We evaluate this, in two different ways:

1. We use (*) to get

0
λ2 H 2
Z
Fe = iλ2 H 2 iEη
dη e = .
−∞ E

2. We integrate by parts and replace η by i∂k12 :

Z 0 0
 2

dη dη m ik12 η ik34 η 0
Fe = −λ2 2 02
(kI
2
− k 2
12 )η 2
+ 2ik12 η + 2
− 2 e e G(kI ; η, η 0 )
−∞ η η H
Z 0 0
 2

dη dη m ik12 η ik34 η 0
= −λ2 2 02
(k 2
12 − k 2 2
I )∂ k + 2k 12 ∂k 12
+ 2
− 2 e e G(kI ; η, η 0 )
−∞ η η H
12

2
 
2 m
= (k12 − kI2 )∂k212 + 2k12 ∂k12 + 2 − 2 F .
H

Hence, we have

m2 λ2 H 2
 
2 2 2
(k12 − kI )∂k12 + 2k12 ∂k12 + 2 − 2 F = .
H E

Defining F̂ = kI F , we can also write this as

m2
 
2 2 uv
u (1 − u )∂u2 − 2u ∂u + 2 − 2 F̂ = λ2 H 2
3
,
H u+v

where u ≡ kI /k12 and v = kI /k34 .

Note: Time has completely disappeared from the problem.

88
11.2 Bootstrap Philosophy

Principles Lagrangian Observables

Causality, unitarity, =⇒ Fields, Equations of motion, =⇒ Cross sections, decay rates,
locality, symmetries, … Feynman diagrams, … cosmological correlators, …

Bootstrap

S-matrix Bootstrap
⇒ Scattering amplitudes are fixed by Lorentz invariance, locality and unitarity:
2
 
X λ 2t
A(s, t) = anm sn tm + PS 1 + 2
s − M2 M

M, S

• No Lagrangian and Feynman diagrams are needed to derive this.

• Basic principles allow only a small menu of possibilities.

Cosmological Bootstrap
• What are the rules for consistent correlators?
Kinematics
Symmetries
Locality
Unitarity
• How are these rules implemented?

89
11.3 Symmetries
The metric of de Sitter space is
1
ds2 = 2 2


−dη + dx .
H 2η2
• Besides translations and rotations, the metric is invariant under dilatations
and special conformal transformations:

η → λη
D:
x → λx
η
η→
1 + 2(b · x) + b2 (x2 − η 2 )
SCT:
x + (x2 − η 2 )b
x→
1 + 2(b · x) + b2 (x2 − η 2 )
• The associated Killing vectors are

Pi = ∂i ,
Jij = xi ∂j − xj ∂i ,
D = −η∂η − xi ∂i ,
 
j 2 2 j
Ki = 2xi η∂η + 2x xi + (η − x )δi ∂j ,

which satisfy the de Sitter algebra so(4, 1):

[Jij , Jkl ] = δjk Jil − δik Jjl + δil Jjk − δjl Jik , [D, Pi ] = Pi ,
[Jij , Kl ] = δjl Ki − δil Kj , [D, Ki ] = −Ki ,
[Jij , Pl ] = δjl Pi − δil Pj , [Ki , Pj ] = 2δij D − 2Jij .

• The infinitesimal transformations of a scalar field Φ(t, x) then are

Pi Φ = ∂ i Φ ,
Jij Φ = (xi ∂j − xj ∂i )Φ ,
DΦ = −(η∂η + xi ∂i )Φ ,
 
j 2 2 j
Ki Φ = 2xi η∂η Φ + 2x xi + (η − x )δi ∂j Φ .

90
How do the boundary fields φ(x) transform?
Consider a massive scalar field:
2 m2 1
Φ00 − Φ0 − ∇2 Φ + 2 2 Φ = 0 .
η H η
At late times, the solution is

Φ(x, η → 0) = φ(x) η ∆ + φ̄(x) η 3−∆ , (*)

where r
3 9 m2
∆≡ − − .
2 4 H2
For “light fields”, with m < 32 H, the first term in (*) dominates at late times.

Using η∂η 7→ ∆, the boundary field profiles φ(x) transform as

P i φ = ∂i φ ,
Jij φ = (xi ∂j − xj ∂i )φ ,
Dφ = −(∆ + xi ∂i )φ ,
Ki φ = 2xi ∆ + 2xi xj ∂j − x2 ∂j φ ,



like a primary operator of conformal weight ∆ in a CFT.

The dual operator O transform in the same way, with ∆ 7→ 3 − ∆ ≡ ∆.

How does this constrain the boundary correlators? =⇒ Ward identities:

N  
X ∂
D: 0= ∆a + xja j hO1 · · · ON i ,
a=1 ∂xa
N  
X ∂ ∂
SCT : 0 = 2∆a xia + 2xia xja j − x2a hO1 · · · ON i .
a=1 ∂x a ∂x a,i

91
In Fourier space, we have

DO = ∆ + k i ∂ki O ,
 

Ki O = 2∆∂ki − ki ∂kj ∂kj + 2k j ∂kj ∂ki O ,

 

and the Ward identity becomes

N
" #
X
−3 + Da hO1 · · · ON i0 (k) = 0 ,
a=1
XN
Kai hO1 · · · ON i0 (k) = 0 .
a=1

Dilatation symmetry fixes the overall momentum scaling:

N
¯ t −3(N −1) ¯t ≡ ¯a.
X
0 ∆
hO1 · · · ON i ∝ k , where ∆ ∆
a=1

All the juice is then in the SCT Ward identity.

Exercise: Show that the two-point function of two scalar operators is

¯
hO1 O2 i0 ∝ k12∆1 −3 δ∆¯ 1 ,∆¯ 2 .

Exercise: Consider the three-point function of two conformally coupled

scalars Φ and a general scalar χ. Show that dilatation invariance implies
¯
hO1 O2 X3 i0 = k3∆−2 Ĝ(u) ,

where u ≡ k3 /(k1 + k2 ), and that special conformal invariance requires

2
 
m
u2 (1 − u2 )∂u2 − 2u3 ∂u + − 2 Ĝ(u) = 0 .
H2

This differential equation has a solution in terms of 2 F1 (u).

Four-point functions are less constrained by conformal symmetry alone.

We need additional input to bootstrap them.

92
11.4 Singularities
Cosmological correlators have an interesting singularity structure.
We have seen this in our results for the flat-space wavefunction:

hO1 O2 O3 i ≡

0
λ
Z
= iλ dt eiKt =
, where K ≡ k1 + k2 + k3 .
−∞ K
↑
total energy singularity

hO1 O2 O3 O4 i ≡

E ≡ k12 + k34 ,
λ2
= , where EL ≡ k12 + kI ,
EEL ER
ER ≡ k34 + kI .
↑
partial energy singularity

The residues of these singularities are scattering amplitudes:

λ2 1 λ2 1 λ2 A4
lim = = = ,
E→0 EEL ER E (k12 + kI )(−k12 + kI ) E s E
λ2 λ λ A3 × ψ̃3
lim = = ,
EL →0 EEL ER EL (k34 + kI )(k34 − kI ) EL

where ψ̃3 is the on-shell wavefunction coefficient:

1 h i
ψ̃3 ≡ ψ3 (k34 , −kI ) + ψ3 (k34 , kI ) .
2kI

93
Generalization to de Sitter:
Wavefunctions factorize on partial energy singularities:

1 (m) (n−m)
lim ψn = p ÃΓ × ψ̃Γ0 ,
EΓ →0 EΓ

where
ν1 − 12 νm − 12
ÃΓ ≡ k1 · · · km AΓ ,
 
ψ̃Γ0 ≡ P (kI ) ψΓ0 (ka , −kI ) − ψΓ0 (ka , kI ) .

Derived using: 1) Early-time limit of propagators, or 2) Unitarity cuts.

11.5 Cosmological Collider

We are now ready to return to the challenge of massive particle exchange:
The four-point function of conformally coupled scalars can be written as

k2
k3
kI 1
hO1 O2 O3 O4 i0 = k1 = F̂ (u, v) ,
k4 kI

where we have introduced

kI kI
u= , v= .
k1 + k2 k3 + k4
This ansatz automatically solves the dilatation Ward identity.

After some work, the conformal Ward identity can be written as

(∇u − ∇v )F̂ = 0 ,

where ∆u ≡ u2 (1 − u2 )∂u2 − 2u3 ∂u is our old friend.

We will classify solutions by their singularities.

94
Contact solutions
The simplest solutions correspond to contact interactions:
X cn (u, v)
F̂c = = ,
n
E 2n+1
↑
Φ4 , (∂µ Φ)4 , · · ·
which have poles at vanishing total energy
X u+v
E≡ kn = kI .
n
uv

(n) (0) (0)

Note that Fc = ∆nu Fc , where Fc ≡ uv/(u + v).

Exchange solutions
Next, we look for solutions corresponding to the tree-level exchange:

F̂e =

we try

(∆u + M 2 )F̂e = F̂c ,

(∆v + M 2 )F̂e = F̂c ,

where F̂c is a contact solution and M 2 ≡ µ2 + 41 .

Using the simplest contact interaction as a source, we have

  
2 2 1 uv
u (1 − u )∂u2 3 2
− 2u ∂u + µ + F̂e = λ2 . (*)
4 u+v

95
Equation (*) has interesting singularities:

• Flat-space limit: lim F̂e = A4 (u + v) log(u + v)

u→−v

• Factorization limit: lim F̂e = A3 log(1 + u) × A3 log(1 + v)

u,v→−1

• Folded limit: lim F̂e ∝ log(1 − u)

u→+1
1
• Collapsed limit: lim F̂e ∝ u 2 +iµ
u→0

Solution is fixed by: 1) No folded singularities and 2) Correct factorization.

Let’s do it:
• The homogeneous solution of (*) is
  21 ±iµ "
1 iµ 3 iµ
#
iu 4 ± 2, 4 ±
F̂± (u) = 2 F1
2
u2 ,
2µ 1 ± iµ

and equivalent solutions F̂± (v).

• To find the particular solution:
1) Find two series solutions around u = 0 and u = ∞.
2) Match these solutions at u = v:

Im(u)

F>

F<
factorization folded
Re(u)
−1 +1

u=v
matching

96
 3) Impose absence of folded singularities at u = 1 and v = 1.

lim F̂± (u) = α± (µ) log(1 − u) .

u→1

4) Demand correct factorization at u = v = −1:

lim F̂ = A3 log(1 + u) × A3 log(1 + v) ,

u,v→−1

where A3 ≡ 1.

The result is
∞
X π
F̂ (u, v) = cmn (µ) u2m+1 (u/v)n + ĝ(u, v) u≤v
m,n=0
2 cosh πµ

where we have defined

2m
(−1)n Y (n + i)
cmn (µ) ≡
n i=0 (n + 21 + i)2 + µ2

and h i
ĝ(u, v) ≡ (β − 1) F̂+ (u)F̂− (v) − αF̂− (u)F̂− (v)
h i
∗
+ (β + 1) F̂− (u)F̂+ (v) − α F̂+ (u)F̂+ (v) ,
with
2
β≡ ,
i sinh πµ
 2iµ 1
α+ i Γ( 2 − iµ)Γ(1 + iµ)
α≡ = .
α− µ Γ( 21 + iµ)Γ(1 − iµ)

In the soft limit, we have

 uv  21 +iµ Γ( 12 + iµ)2 Γ(−iµ)2
lim F̂ (u, v) = (1 + i sinh πµ) + c.c.
u,v→0 4 2π

97
 Summary:
The full solution has the form:

F̂ (u, v) = F̂EFT (u, v) + F̂pp (u, v) .

We are forced to have particle production.

This leads to oscillations in the collapsed limit:
strength
Correlationfunction

Four-point function
Cross section (nb)
Four-point

gy (GeV) Momentum ratio Centre-of-mass energy (GeV)

These oscillations are the analog of a resonance in collider physics:

• The frequency of the oscillations depends on the mass of the particles.
• The angular dependence of the signal depends on the spin of the particles.

kI / sin[M log kI ] PS (cos ✓)

✓

98
11.6 Outlook
The bootstrap approach is only the beginning of a systematic exploration of
cosmological correlators:

dSd+1 Rd,1 AdSd+1

under-
standing
(tree level) tree level nonperturbative
(some loops)

Many open questions remain:

• Beyond Feynman So far, the cosmological bootstrap has only been applied to individual
Feynman diagrams (which by themselves are unphysical). However, in the S-matrix bootstrap,
the real magic is found for on-shell diagrams. What is the on-shell formulation of the cosmo-
logical bootstrap? What magic will it reveal?
• Beyond Perturbation Theory So far, the bootstrap constraints (singularities, unitarity
cuts, etc.) are only formulated in perturbation theory and implemented only at tree level. In
contrast, we understand the conformal bootstrap nonperturbatively. Recently, some prelim-
inary progress was made to understand unitarity constraints in de Sitter nonperturbatively,
but much work remains. A nonperturbative bootstrap is required if we want to understand
the UV completion of cosmological correlators.
• Beyond Spacetime In quantum gravity, spacetime is an emergent concept. In cosmology,
the notation of “time” breaks down at the Big Bang. What replaces it? The cosmological
bootstrap provides a modest form of emergent time, but there should be a much more radical
approach where space and time are outputs, not inputs. For scattering amplitudes, such a
reformulation was achieved in the form of the amplituhedron. What is the analogous object in
cosmology? How does the Big Bang arise from it?

99