Lectures on Cosmological Correlations

Daniel Baumann and Austin Joyce

time

Inflation Reheating CMB LSS

OUTLINE

I. Motivation

II. In-In Formalism

III. Wavefunction Approach

IV. Cosmological Bootstrap

V. Outlook

Lecture notes and lecture scripts can be found at:

[Link]
I. MOTIVATION
1.1. Cosmological Correlations
In cosmology, we measure spatial correlations:

hδρ(x1 ) · · · δρ(xN )i

hδT (θ1 ) · · · δT (θN )i

These correlations can be traced back to the origin of the hot Big Bang:

hζ(x1 ) · · · ζ(xN )i

?
Where did the primordial correlations come from?
• Clue 1: The correlations span superhorizon scales.
• Clue 2: They are scale-invariant.

This suggests that the fluctuations were created before the hot Big Bang,
during a phase of approximate time-translation invariance.

2
1.2. Inflation and De Sitter Space
The observed correlations can be explained by an extended period of acceler-
ated expansion (= inflation):

1 da
H(t) ≡ ≈ const.
a dt
Ḣ
ε(t) ≡ − 2  1 .
H

The spacetime during inflation is approximately de Sitter space.

In conformal time, dη = dt/a(t), the de Sitter metric is

−dη 2 + dx2
ds2 = ,
(Hη)2
where −∞ < η < 0. The primordial correlations live on the future boundary
of the de Sitter space at η∗ ≈ 0.

3
1.3. Quantum Fluctuations
Consider a massless scalar field during inflation:
Z
1

S= dη d3 x a2 ϕ̇2 − (∂i ϕ)2
2  
Z
1 ä 2
= dη d x u̇ − (∂i u) + u , where u(η, x) ≡ a(η)ϕ(η, x) .
3 2 2
2 a

Classical dynamics
The classical equation of motion (for each Fourier mode) is
 
2
ük + k 2 − 2 uk = 0 .
η

• At early times (−kη  1), we have

1
ük + k 2 uk = 0 =⇒ uk ∼ √ e±ikη .
2k
• At late times (−kη → 0), we have
2
ük − uk = 0 =⇒ uk ∼ c1 η −1 + c2 η 2 → c1 η −1 .
η2
The exact solution is
  ±ikη
i e
uk (η) = a±
k 1± √ .
kη 2k

Canonical quantization
• Promote classical fields u, π = u̇ to quantum operators û, π̂, with

[û(η, x), π̂(η, x0 )] = iδD (x − x0 ) =⇒ [ûk (η), π̂k′ (η)] = i(2π)3 δD (k + k0 ) .

• Define the mode expansion

ûk (η) = u∗k (η)âk + uk (η)â†−k ,

where [âk , â†k′ ] = (2π)3 δD (k − k0 ) and u̇k u∗k − uk u̇∗k = i (Wronksian).

4
• Define the vacuum by (see Exercise 2.1)
  ikη
i e
âk |0i = 0 and uk (η) = 1 + √ (BD) .
kη 2k

• Compute the two-point function of zero-point fluctuations:

  
∗ † ∗ †
h0|ûk (η)ûk′ (η)|0i = h0| uk (η) âk + uk (η) â−k uk′ (η) âk′ + uk′ (η) â−k′ |0i

= (2π)3 δD (k + k0 ) |uk (η)|2 .

• The power spectrum of ϕ is

|uk (η)|2 H2
kη→0 H 2

Pϕ (k) = = 1 + k 2 2
η −−−−→ .
a2 (η) 2k 3 2k 3
This is the famous scale-invariant spectrum of a massless field in de Sitter.

Curvature perturbations
The initial conditions of the hot Big Bang are typically defined in terms of the
comoving curvature perturbation:

δgij = a2 (η)e2ζ(η,x) δij ,

which is related to the inflaton fluctuations (in spatially flat gauge) by

H
ζ=− δϕ .
ϕ̇
The power spectrum of ζ then is
 2
H H2
Pζ (k) = ≡ As k ns −1 .
ϕ̇ 2k 3 k=aH

Massive fields
Massive fields are also produced during inflation, but don’t survive until late
times (see Exercise 2.2). Their imprints can, however, be found in the corre-
lations of the light fields (= “cosmological collider physics”)

5
1.4. Primordial Non-Gaussianity
The main diagnostic of primordial non-Gaussianity is the bispectrum:

hζk1 ζk2 ζk3 i = Bζ (k1 , k2 , k3 ) × (2π)3 δD (k1 + k2 + k3 ) .

5 Bζ (k, k, k)
• amplitude: fNL ≡
18 Pζ2 (k)

• shape:

local equilateral flattened

• effect: new particles new interactions excited states

• Planck equil
|fNL
loc
|<5 |fNL | < 40 |fNL
flat
| < 20
constraints:

equilateral folded

squeezed

3
e−ikη
φ̇

Massive particles can be created by the inflationary expansion.

The decay of particles produces distinct correlations.
These correlations are tracers of the inflationary dynamics.

6
In the lectures notes, we describe three methods for computing these higher-
point correlations:

• In-In Formalism

• Wavefunction Approach

• Cosmological Bootstrap

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

7
II. WAVEFUNCTION APPROACH
2.1. 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

Z
Ψ[ϕ] ≡ hϕ(x)|0i = DΦ eiS[Φ] ≈ eiS[Φcl ] .
Φ(η∗ )=ϕ
Φ(−∞)=0

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

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

where the “wavefunction coefficients” are

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

The relation between correlators and wavefunction coefficients is

1
hϕϕi = ,
2RehOOi
2RehOOOi
hϕϕϕi = Q3 ,
n=1 2RehOn On i
hOOOOi hOOXi3
hϕϕϕϕi = + .
hOOi4 hXXihOOi4

8
2.2. A Warmup
As a warmup, we apply the wavefunction approach to a harmonic oscillator:
Z  
1 2 1 2 2
S[Φ] = dt Φ̇ − ω Φ .
2 2
The classical solution is
Φcl = ϕeiωt ,
and the on-shell action becomes
Z t∗  
1 1
S[Φcl ] = dt ∂t (Φ̇cl Φcl ) − Φcl (Φ̈cl + ω Φcl )
2
ti 2 2 | {z }
=0
1
= Φ̇cl Φcl
2 t=t∗

iω
= ϕ2 .
2
The wavefunction then is
 ω 
Ψ[ϕ] ≈ exp(iS[Φcl ]) = exp − ϕ2 ,
2
which implies
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 .

9
To make this more interesting, consider a time-dependent oscillator:
Z  
1 1
S[Φ] = dt A(t)Φ̇ − B(t)Φ .
2 2
2 2
The classical solution is
K(0) = 1
Φcl = ϕK(t) , with
K(−∞) ∼ eiωt

and 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) ∗

10
2.3. Free Fields in de Sitter
Consider a massless field in de Sitter:
Z
 
S = dη d3 x a2 (η) (Φ0 )2 − (∇Φ)2
Z  
1 d3 k 1 k 2
= 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
 
1 −ik
Im(A∂η log K) = Im + ik
η=η∗ (Hη∗ )2 1 − ikη∗
 2 
1 k η + ik 3 η∗2 η∗ →0 k3
= Im −−−→ ,
(Hη∗ )2 1 + k 2 η∗2 H2

The two-point function then is

0 H2
hϕk ϕ−k i = 3 .
2k

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

11
2.4. Anharmonic Oscillator
Consider the following anharmonic oscillator:
Z  
1 2 1 2 2 1 3
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


Φcl (t) = ϕK(t) + i dt0 G(t, t0 ) −λΦ2cl (t0 ) ,

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

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

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

Since

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

the boundary term is

 Z 
1 1 ′ iωt′

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

12
The action then becomes
Z
iω 2 λ
S[Φcl ] = ϕ − ϕ dt eiωt Φ2cl
2 2
Z   Z   λ 
1 ′ ′ 2 ′
+ dt − ϕe − iλ dt G(t, t ) Φcl (t )
iωt
− λΦcl (t) − Φcl .
2 3
2 3

The terms linear in ϕ cancel.

The final on-shell action is

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

To evaluate this, we write the classical solution as

Φcl (t) = Φ(0) (t) + λΦ(1) (t) + λ2 Φ(2) (t) + · · · .

where

Φ(0) (t) = ϕeiωt ,

Z  
0 0 0
(1)
Φ (t) = i dt G(t, t ) −(Φ (t ))
(0) 2

Z  
0 0 2 2iωt′ ϕ2

=i dt G(t, t ) −ϕ e = e 2iωt
− e iωt
.
3ω 2
With this, the wavefunction becomes
 
ω 2 λ 3 λ2 4
Ψ[ϕ] ≈ eiS[Φcl ] = exp − ϕ − ϕ + ϕ + ··· .
2 9ω 72ω 3

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

13
2.5. Interactions in Field Theory
Back to field theory.
Z  
4 √ 1 µν 1 2 2 1 3
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 wavefunction
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∗ )
f ∗ (t∗ )
G(k; t, t0 ) = fk∗ (t)fk (t0 ) θ(t − t0 ) + fk∗ (t0 )fk (t) θ(t0 − t) − k fk (t)fk (t0 ) .
| {z } fk (t∗ )
0
= GF (k; t, t )

14
2.6. Correlators in Flat Space
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−t′ ) 0 ik(t−t′ ) 0 ik(t+t′ )

G(k; t, t ) = e θ(t − t ) + e θ(t − t) − e .
2k
The three- and four-point wavefunction coefficients in Φ3 theory are

hO1 O2 O3 i ≡

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

hO1 O2 O3 O4 i ≡

Z 0
′ ′′
= −λ 2
dt0 dt00 eik12 t G(kI ; t0 , t00 ) eik34 t , knm ≡ kn + km
−∞

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

In the lecture notes, we show in detail how this reproduces the correct in-in
correlators.

15
2.7. A Challenge
In general, correlators in de Sitter cannot be computed analytically (some ex-
ceptions are presented in the lecture notes).
Consider the four-point function of a conformally coupled scalar (with
m2 = 2H 2 ) mediated by the exchange of a massive scalar:

F ≡ hO1 O2 O3 O4 i ≡

Z Z
dη 0 dη 00 ik12 η′ ik34 η′′
= −λ 2
e e G(kI ; η 0 , η 00 ) .
η 02 η 002
↑
products of Hankel functions

The time integrals cannot be performed analytically.

Is there a way to obtain an analytic understanding of this (and other) correla-
tors in de Sitter space?

16
III. COSMOLOGICAL BOOTSTRAP
Reheating surface

End of inflation

particle
production
particle
decay

3.1. 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:
X 2
 
g 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?
• How are these rules implemented?

17
3.2. Symmetries
The metric of de Sitter is
1

ds2 = −dη 2
+ dx2
.
H 2η2
The isometries of the metric are:
• Pi : translations
• Jij : rotations
• D: dilatations D = −η∂η − xi ∂i ,

• Ki : SCTs Ki = 2xi η∂η + 2xi xj ∂j + (η 2 − x2 )∂i .

We would like to see how these symmetries act on the boundary fields.
Consider a massive scalar field in de Sitter space:

2 m2 Φ
Φ00 − Φ0 − ∇2 Φ + 2 2 = 0 .
η H η
At late times, the solution is
r
3 9 m2
Φ(x, η → 0) = ϕ(x) η ∆ + ϕ̄(x) η 3−∆ , ∆≡ − − . (⋆)
2 4 H2
↑
dominates at late times (η → 0), for m < 23 H

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

Dϕ = −(∆ + xi ∂i )ϕ ,


Ki ϕ = 2xi ∆ + 2xi xj ∂j − x2 ∂i ϕ ,

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

The dual operator O transform in the same way, with ∆ 7→ 3 − ∆ ≡ ∆.
In cosmology, we work in Fourier space:
 
DO = ∆ + k i ∂ki O ,
 
Ki O = 2∆∂ki − ki ∂kj ∂kj + 2k j ∂kj ∂ki O .

18
This leads to the following Ward identity for the wavefunction coefficients:
" #
X
N
−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:

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

All the juice is then in the SCT Ward identity.

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

hO1 O2 i0 ∝ k12∆1 −3 δ∆1 ,∆2 .

Ex: 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 that

 
m2
u (1 − u
2 2
)∂u2 − 2u ∂u +
3
− 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.

19
3.3. Singularities
Cosmological correlators have an interesting singularity structure.
We can see this in our results for the flat-space wavefunction:

hO1 O2 O3 i ≡

Z 0
λ
= 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 shifted wavefunction coefficient:

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

The same holds for de Sitter correlators:

• Correlators have singularities at points of would-be energy conservation.
• The residues of these singularities are scattering amplitudes.
We will use these singularities as an input for the bootstrap.

20
3.4. Massive Exchange
We are now ready to return to the challenge of massive particle exchange:
The four-point function of conformally coupled scalars (in the s-channel) can
be written as

k2 k3
0 1
F ≡ hO1 O2 O3 O4 i = k1 = F̂ (u, v) ,
k4
kI
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 .

We will classify solution by their singularities.

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

Note that F̂c,n = ∆nu F̂c,0 , where F̂c,0 ≡ uv/(u + v).

21
Exchange solutions
For tree exchange, we try
(∆u + M 2 )F̂e = Ĉ
2
⇔ (s + m2 )Ae = Ac (s, t)
(∆v + M )F̂e = Ĉ

where Ĉ = M 2 F̂c is a contact solution and M 2 ≡ m2 /H 2 − 2.

Using the simplest contact interaction as a source, we have
  
1 uv
u2 (1 − u2 )∂u2 − 2u3 ∂u + µ2 + F̂ = , (⋆)
4 u+v
p
where µ ≡ m2 /H 2 − 9/4.

Singularities
The equation has a number of interesting singularities:
• Factorization limit: lim F̂ ∝ A3 log(1 + u) × A3 log(1 + v)
u,v→−1

• Folded limit: lim F̂ ∝ log(1 − u)

u→+1

k3
k2 This singularity should be absent
in the standard vacuum.
k1 k4

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

k3
k2 This non-analyticity corresponds to
k4 spontaneous particle production.
k1

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

u→−v

Factorization and folded limits uniquely fix the solution.

The remaining singularities become consistency checks.

22
 Cosmological collider
The full solution can be written as (see lecture notes for explicit expressions)

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 ✓)

✓

This is the analog of

g2
A= PS (cos θ)
s − M2

23
IV. OUTLOOK
There are many interesting things I didn’t have time to describe in the lectures.
Some more can be found in the lecture notes.
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?

24