Astro 408

Set 1: Relativistic Perturbation Theory

Wayne Hu
 Covariant Perturbation Theory
• Covariant = takes same form in all coordinate systems
• Invariant = takes the same value in all coordinate systems
• Fundamental equations are covariant: Einstein equations, covariant
conservation of stress-energy tensor:

Gµν = 8πGTµν
∇µ T µν = 0

• Components such as ρ, velocity, curvature etc are not invariant

under a coordinate change
• Between any two fully specified coordinates, Jacobian ∂xµ /∂ x̃ν is
invertible - so perturbations in given gauge can be written in a
covariant manner in terms of perturbations in an arbitrary gauge:
called “gauge invariant” variables
 Covariant Perturbation Theory
• In evolving perturbations we inevitably break explicit covariance
by evolving conditions forward in a given time coordinate
• Retain implicit covariance by allowing the freedom to choose an
arbitrary time slicing and spatial coordinates threading constant
time slices
• Exploit covariance by choosing the specific slicing and threading
(or “gauge”) according to what best matches problem
• Preserve general covariance by keeping all free variables: 10 for
each symmetric 4×4 tensor but blocked into 3 + 1 “ADM” form

1 2 3 4
5 6 7
8 9
10
 ADM 3+1 Split
• Since Einstein equations dynamically evolve the spacetime, to
solve the initial value problem choose a slicing for the foliation
and evolve the spatial metric forward: 3+1 ADM split
• Define most general line element: lapse N , shift N i , 3-metric hij
ds2 = −N 2 dφ2 + hij (dxi + N i dφ)(dxj + N j dφ)
or equivalently the metric

g00 = −N 2 + N i Ni , g0i = hij N j ≡ Ni , gij = hij

and its inverse g µα gαν = δ µν

g 00 = −1/N 2 , g 0i = N i /N 2 , g ij = hij − N i N j /N 2
• Time coordinate x0 = φ need not be cosmological time t - could
be any parameterization, e.g. conformal time, scalar field, . . .
.

ADM 3+1 Split

• Useful to define the unit xi=-Nidφ xi=0

normal timelike vector nµ nµ = −1, Ndφ φ+dφ

orthogonal to constant time surfaces nµ

xi=0
nµ ∝ ∂µ φ φ

nµ = (−N, 0, 0, 0), nµ = (1/N, −N i /N )

where we have used nµ = g µν nν

• Interpretation: lapse of proper time along normal, shift of spatial
coordinates with respect to normal
• In GR (and most scalar-tensor EFT extensions), the lapse and shift
are non-dynamical and just define the coordinates or gauge
• Dynamics in evolving the spatial metric forwards
 ADM 3+1 Split
• Projecting 4D tensors onto the normal direction utilizes nµ nν , e.g.

−nµ nν V ν

• Projecting 4D tensors onto the 3D tensors involves the

complement through the induced metric

hµν = gµν + nµ nν ,
hµν V ν = (δ µν + nµ nν )V ν = V µ + nµ nν V ν

e.g. in the preferred slicing

Ṽ µ = hµν V ν = (δ µν + nµ nν )V ν = (0, V i + N i V 0 )

whose spatial indices are raised an lowered by hij :

Ṽi = giν Ṽ ν = hij Ṽ j
.

ADM 3+1 Split

• 3-surface embedded in 4D, so there is Ki jδxi

both an intrinsic curvature associated φ+δφ

nµ nµ+δnµ
with hij and an extrinsic curvature
δxi
which is the spatial projection of the φ

gradient of nµ

Kµν = hµα hν β nα;β

• Kµν symmetric since the antisymmetric projection (or vorticity)

vanishes by construction since nµ = −N φ;µ
 ADM 3+1 Split
• Likewise split the spacetime curvature (4) R into intrinsic and
extrinsic pieces via Gauss-Codazzi relation
(4)
R = Kµν K µν − (Kµ µ )2 + (3) R + 2(Kν ν nµ − nα nµ;α );µ

Last piece is total derivative so Einstein Hilbert action is equivalent

to keeping first three pieces
• No explicit dependence on slicing and threading N , N i - any
preferred slicing is picked out by the matter distribution not by
general relativity
• Beyond GR we can embed a preferred slicing by making the
Lagrangian an explicit function of N - will return to this in the
effective field theory of inflation, dark energy
 ADM 3+1 Split
• Trace Kµ µ = nµ;µ ≡ θ is expansion
• Avoid confusion with FRW notation for intrinsic curvature:
(3)
R = 6K/a2
• The anisotropic part is known as the shear

θ
σµν = Kµν − hµν
3
• For the FRW background the shear vanishes and the expansion
θ = 3H
 ADM 3+1 Split
• Fully decompose the 4-tensor nµ;ν by adding normal components

nµ;ν = Kµν − nµ nα hν β nα;β − hµα nν nβ nα;β + nµ nα nν nβ nα;β

= Kµν − hµα nν nβ nα;β = Kµν − nν nβ nµ;β − nµ nα nν nβ nα;β
= Kµν − nν nβ nµ;β = Kµν − aµ nν

where we have used [(nµ nµ );ν = 0 → nµ nµ;ν = 0]

• Here the directional derivative of the normal along the normal or
“acceleration” is

aµ = (nµ;β )nβ
 ADM 3+1 Split
• Since

ni;j = Kij = −Γµij nµ = Γ0ij N

in terms of the ADM variables

1
Kij = (∂t hij − Nj|i − Ni|j )
2N
where | denotes the covariant derivative with respect to hij
• Extrinsic curvature acts like a “velocity” term for hij moving the
metric from one slice to another with the coordinate freedom of the
lapse and shift
• Initial value problem in GR: define hij and ḣij on the spacelike
surface and integrate forwards, with lapse and shift defining the
temporal and spatial coordinates
 ADM 3+1 Split
• Beyond GR we can extend this logic by constructing a general
theory with some scalar whose constant surfaces define the normal
and the time coordinate - build the most general action that retains
spatial diffeomorphism invariance out of the ADM geometric
objects
→ EFT of inflation and dark energy: return to this in inflation
discussion
• For linear perturbation theory in GR, ADM looks simpler since we
can linearize metric fluctuations and take out the global scale
factor in the spatial tensors for convenience hij = a2 γij
• ADM language useful in defining the geometric meaning of gauge
choices in defining the time slicing and spatial threading
 Metric Perturbations
• ADM on the conformal metric g̃µν with gµν = a2 g̃µν , recall FRW
background

ds̃2 = g̃µν dxµ dxν = −dη 2 + γij dxi dxj

= −dη 2 + dD2 + DA
2
(dθ2 + sin2 θdφ2 )

• Background lapse N̄ = 1, and shift N̄ i = 0 so define perturbations

N = (1 + A), shift N i = −B i

g̃00 = −(1 + 2A) ,

g̃0i = −γij B j ≡ −Bi

where to linear order indices on 3-tensors raised and lowered by γij

• This absorbs 1+3=4 free variables in the metric
 Metric Perturbations
• Remaining 6 is in the spatial surfaces which we parameterize as

g̃ij = γij + 2HL γij + 2HT ij

here (1) HL a perturbation to the scale factor; (5) HTij a trace-free
distortion to spatial metric
• Curvature perturbation on the 3D slice, hereafter ∇2 is the
3-Laplacian using covariant derivatives with respect to γij

6K 4 2
δ R = 2 − 2 ∇ + 3K HL + 2 ∇i ∇j HTij
(3) 2


a a a
where recall that K characterizes the background intrinsic
curvature
• Curvature perturbation is a 3-scalar in the ADM split and a Scalar
in the SVT decomposition
 Matter Tensor
• Likewise expand the matter stress energy tensor around a
homogeneous density ρ and pressure p:

T 00 = −ρ − δρ ,
T 0i = (ρ + p)(vi − Bi ) ,
T0i = −(ρ + p)v i ,
T ij = (p + δp)δ ij + pΠij ,

• (1) δρ a density perturbation; (3) vi a vector velocity, (1) δp a

pressure perturbation; (5) Πij an anisotropic stress perturbation
• So far this is fully general and applies to any type of matter or
coordinate choice including non-linearities in the matter, e.g.
scalar fields, cosmological defects, exotic dark energy.
 Counting Variables

20 Variables (10 metric; 10 matter)

−10 Einstein equations
−4 Conservation equations
+4 Bianchi identities
−4 Gauge (coordinate choice 1 time, 3 space)
——
6 Free Variables
• Without loss of generality these can be taken to be the 6
components of the matter stress tensor
• For the background, specify p(a) or equivalently
w(a) ≡ p(a)/ρ(a) the equation of state parameter.
 Homogeneous Einstein Equations
• Einstein (Friedmann) equations:
 2  2
1 da K 8πG 1 ȧ
= − 2+ ρ [= = H 2]
a dt a 3 aa
1 d2 a 4πG 1 d ȧ 1 d
2
= − (ρ + 3p) [= 2
= 2 (aH)]
a dt 3 a dη a a dη
so that w ≡ p/ρ < −1/3 for acceleration
• Conservation equation ∇µ Tµν = 0 implies
ρ̇ ȧ
= −3(1 + w)
ρ a
overdots are conformal time but equally true with coordinate time
 Homogeneous Einstein Equations
• Counting exercise:

20 Variables (10 metric; 10 matter)

−17 Homogeneity and Isotropy
−2 Einstein equations
−1 Conservation equations
+1 Bianchi identities
——
1 Free Variables

without loss of generality choose ratio of homogeneous & isotropic

component of the stress tensor to the density w(a) = p(a)/ρ(a).
Acceleration Implies Negative Pressure
• Role of stresses in the background cosmology
• Homogeneous Einstein equations Gµν = 8πGTµν imply the two
Friedmann equations (flat universe, or associating curvature
ρK = −3K/8πGa2 )
 2
1 da 8πG
= ρ
a dt 3
1 d2 a 4πG
2
= − (ρ + 3p)
a dt 3
so that the total equation of state w ≡ p/ρ < −1/3 for acceleration
• Conservation equation ∇µ Tµν = 0 implies
ρ̇ ȧ
= −3(1 + w)
ρ a
so that ρ must scale more slowly than a−2
 Scalar, Vector, Tensor
• In linear perturbation theory, perturbations may be separated by
their transformation properties under 3D rotation and translation.
• The eigenfunctions of the Laplacian operator form a complete set
∇2 Q(0) = −k 2 Q(0) S,
2(±1) (±1)
2
∇ Qi = −k Qi V,
(±2) (±2)
∇2 Qij = −k 2 Qij T,
• Vector and tensor modes satisfy divergence-free and
transverse-traceless conditions
(±1)
∇i Qi =0
(±2)
∇i Qij =0
(±2)
γ ij Qij =0
 Vector and Tensor Quantities
• A scalar mode carries with it associated vector (curl-free) and
tensor (longitudinal) quantities
• A vector mode carries and associated tensor (trace and divergence
free) quantities
• A tensor mode has only a tensor (trace and divergence free)
• These are built from the mode basis out of covariant derivatives
and the metric
(0)
Qi = −k −1 ∇i Q(0) ,
(0) −2 1
Qij = (k ∇i ∇j + γij )Q(0) ,
3
(±1) 1 (±1) (±1)
Qij = − [∇i Qj + ∇j Qi ] ,
2k
 Perturbation k-Modes
• For the kth eigenmode, the scalar components become
A(x) = A(k) Q(0) , HL (x) = HL (k) Q(0) ,
δρ(x) = δρ(k) Q(0) , δp(x) = δp(k) Q(0) ,
the vectors components become
X 1 1
X
(m) (m)
Bi (x) = B (m) (k) Qi , vi (x) = v (m) (k) Qi ,
m=−1 m=−1
and the tensors components
2
X 2
X
(m) (m) (m)
HT ij (x) = HT (k) Qij , Πij (x) = Π(m) (k) Qij ,
m=−2 m=−2

• Note that the curvature perturbation only involves scalars

(3) 4 2 (0) 1 (0) (0)
δ[ R] = 2 (k − 3K)(HL + HT )Q
a 3
 Spatially Flat Case
• For a spatially flat background metric, harmonics are related to
plane waves:

Q(0) = exp(ik · x)
(±1) −i
Qi = √ (ê1 ± iê2 )i exp(ik · x)
2
r
(±2) 3
Qij = − (ê1 ± iê2 )i (ê1 ± iê2 )j exp(ik · x)
8
where ê3 k k. Chosen as spin states, cf. polarization.
• For vectors, the harmonic points in a direction orthogonal to k
suitable for the vortical component of a vector
 Spatially Flat Case
• Tensor harmonics are the transverse traceless gauge representation
• Tensor amplitude related to the more traditional

h+ [(e1 )i (e1 )j − (e2 )i (e2 )j ] , h× [(e1 )i (e2 )j + (e2 )i (e1 )j ]

as
√ (∓2)
h+ ± ih× = − 6HT
(±2)
• HT proportional to the right and left circularly polarized
amplitudes of gravitational waves with a normalization that is
convenient to match the scalar and vector definitions
 Covariant Scalar Equations
• DOF counting exercise

8 Variables (4 metric; 4 matter)

−4 Einstein equations
−2 Conservation equations
+2 Bianchi identities
−2 Gauge (coordinate choice 1 time, 1 space)
——
2 Free Variables

without loss of generality choose scalar components of the stress

tensor δp, Π .
 Covariant Scalar Equations
• Einstein equations (suppressing 0) superscripts
1 ȧ ȧ ȧ
(k 2 − 3K)[HL + HT ] − 3( )2 A + 3 ḢL + kB =
3 a a a
= 4πGa2 δρ , 00 Poisson Equation
ȧ 1 K
A − ḢL − ḢT − 2 (kB − ḢT )
a 3 k
= 4πGa2 (ρ + p)(v − B)/k , 0i Momentum Equation
"  2 #
2
 
ä ȧ ȧ d k d ȧ 1
2 −2 + − A− + (ḢL + kB)
a a a dη 3 dη a 3
2 1
= 4πGa (δp + δρ) , ii Acceleration Equation
3
 
1 d ȧ
k 2 (A + HL + HT ) + +2 (kB − ḢT )
3 dη a
= −8πGa2 pΠ , ij Anisotropy Equation
 Covariant Scalar Equations
• Poisson and acceleration equations are the perturbed
generalization of the Friedmann equations
• Momentum and anisotropy equations are new to the perturbed
metric
• Poisson and momentum equations in the ADM language take the
form of constraints on the shift and lapse respectively - leaving the
spatial metric components as dynamical
• Like the Friedmann equations, the 4 equation are redundant given
the 2 energy-momentum conservation equations
• Choose a gauge and set of equations to simplify the given problem
 Covariant Scalar Equations
• Conservation equations: continuity and Navier Stokes
 
d ȧ ȧ
+ 3 δρ + 3 δp = −(ρ + p)(kv + 3ḢL ) ,
dη a a
  
d ȧ (v − B) 2 K
+4 (ρ + p) = δp − (1 − 3 2 )pΠ + (ρ + p)A ,
dη a k 3 k

• Equations are not independent since ∇µ Gµν = 0 via the Bianchi

identities.
• Related to the ability to choose a coordinate system or “gauge” to
represent the perturbations.
 Covariant Vector Equations
• Einstein equations
(±1)
(1 − 2K/k 2 )(kB (±1) − ḢT )
= 16πGa2 (ρ + p)(v (±1) − B (±1) )/k ,
 
d ȧ (±1)
+ 2 (kB (±1) − ḢT )
dη a
= −8πGa2 pΠ(±1) .

• Conservation Equations
 
d ȧ
+ 4 [(ρ + p)(v (±1) − B (±1) )/k]
dη a
1
= − (1 − 2K/k 2 )pΠ(±1) ,
2
• Gravity provides no source to vorticity → decay
 Covariant Vector Equations
• DOF counting exercise

8 Variables (4 metric; 4 matter)

−4 Einstein equations
−2 Conservation equations
+2 Bianchi identities
−2 Gauge (coordinate choice 1 time, 1 space)
——
2 Free Variables

without loss of generality choose vector components of the stress

tensor Π(±1) .
 Covariant Tensor Equation
• Einstein equation
2
 
d ȧ d 2 (±2) 2 (±2)
+ 2 + (k + 2K) H T = 8πGa pΠ .
dη 2 a dη
• DOF counting exercise

4 Variables (2 metric; 2 matter)

−2 Einstein equations
−0 Conservation equations
+0 Bianchi identities
−0 Gauge (coordinate choice 1 time, 1 space)
——
2 Free Variables

wlog choose tensor components of the stress tensor Π(±2) .

 Arbitrary Dark Components
• Total stress energy tensor can be broken up into individual pieces
• Dark components interact only through gravity and so satisfy
separate conservation equations
• Einstein equation source remains the sum of components.
• To specify an arbitrary dark component, give the behavior of the
stress tensor: 6 components: δp, Π(i) , where i = −2, ..., 2.
• Many types of dark components (dark matter, scalar fields,
massive neutrinos,..) have simple forms for their stress tensor in
terms of the energy density, i.e. described by equations of state.
• An equation of state for the background w = p/ρ is not sufficient
to determine the behavior of the perturbations.
 Geometry of Gauge Choice
• Geometry of the gauge or time slicing and spatial threading
• For perturbations larger than the horizon, a local observer should
just see a different (separate) FRW universe
• Scalar equations should be equivalent to an appropriately
remapped Friedmann equation
• ADM recap: unit normal vector nµ to constant time hypersurfaces
nµ dxµ = n0 dη, nµ nµ = −1, to linear order in metric

n0 = −a(1 + AQ), ni = 0
n0 = a−1 (1 − AQ), ni = −BQi

• Intrinsic 3-geometry of δgij , changes in the normal vector nµ;ν that

define the extrinsic curvature
 Geometric Quantities
• Expansion of spatial volume per proper time is given by
4-divergence
k 3
nµµ ≡ θ = 3H(1 − AQ) + BQ + ḢL Q
a a
• Other pieces of nµ;ν give the vorticity, shear and acceleration
θ
nµ;ν ≡ ωµν + σµν + hµν − aµ nν
3

hµν = gµν + nµ nν
ωµν = hµα hν β (nα;β − nβ;α ) = 0
1 α β 1
σµν = h h (nα;β + nβ;α ) − θhµν
2 µ ν 3
aµ = nµ;α nα

• Recall nµ is a special timelike vector normal to the constant time

surfaces, the vorticity vanishes by construction
 Geometric Quantities
• Remaining perturbed quantities are the spatial shear and
acceleration (0 components vanish)

σij = a(ḢT − kB)Qij

ai = −kAQi

• Recall that the extrinsic curvature Kij = σij + θhµν /3

• Intrinsic curvature of the 3-surface determined by 3-metric hij
4 2 HT
δ[(3) R] = 2
(k − 3K)(H L + )
a 3
• E-foldings of the local expansion ln aL are given
Z Z  
1 ȧ 1
ln aL = dτ θ = dη + ḢL Q + kBQ
3 a 3

where we have used dτ = (1 + AQ)adη

 Separate Universe
• Notice that
d ḢT 1
δ ln aL = ḢL + − (ḢT − kB)
dη 3 3
so that if the shear is negligible the change in efolds tracks the
change in curvature
• Shear vanishes in the FRW background; uniform efolding gives
constant curvature
• Underlying principle: local observer should find long wavelength
perturbations are indistingishable from a change in the background
FRW quantities
• Perturbation equations take the form of Friedmann equations once
rescaled
 Time Slicing
• Constant time surfaces can be defined according to what geometry
is helpful for the problem at hand
• Common choices:
Uniform efolding: ḢL + kB/3 = 0
Shear free: ḢT − kB = 0
Zero lapse pert or acceleration, A = 0
Uniform expansion: −3HA + (3ḢL + kB) = 0
Comoving: v = B
• For the background all of these conditions hold.
• For perturbations each define a choice of slicing
• Can define the validity of the separate universe principle as the
coexistence of comoving and zero lapse slicing
 Time Slicing
• Comoving slicing is more formally called velocity orthogonal
slicing since constant time surfaces are orthogonal to the matter
4-velocity V µ :

hµν V ν = (δ µν + nµ nν )V ν = (0, V i + N i V 0 ) = 0

→ V i = vQi = B i = BQi

• Should not be confused with comoving (threading) where the

3-velocity v = 0 unless the shift B also vanishes
 Gauge
• Metric and matter fluctuations take on different values in different
coordinate system
• No such thing as a “gauge invariant” density perturbation!
• General coordinate transformation:

η̃ = η + T
x̃i = xi + Li

free to choose (T, Li ) to simplify equations or physics –

corresponds to a choice of slicing and threading in ADM.
• Decompose these into scalar T , L(0) and vector harmonics L(±1) .
 Gauge
• gµν and Tµν transform as tensors, so components in different
frames can be related
α β
∂x ∂x
g̃µν (η̃, x̃i ) = µ ν
gαβ (η, xi
)
∂ x̃ ∂ x̃
∂xα ∂xβ i i
= µ ν
gαβ (η̃ − T Q, x̃ − LQ )
∂ x̃ ∂ x̃
• Fluctuations are compared at the same coordinate positions (not
same space time positions) between the two gauges
• For example with a T Q perturbation, an event labeled with
η̃ =const. and x̃ =const. represents a different time with respect to
the underlying homogeneous and isotropic background
 Gauge Transformation
• Scalar Metric:
ȧ
à = A − Ṫ − T ,
a
B̃ = B + L̇ + kT ,
k ȧ
H̃L = HL − L − T ,
3 a
1 1 ȧ
H̃T = HT + kL , H̃L + H̃T = HL + HT − T
3 3 a
curvature perturbation depends on slicing not threading
• Scalar Matter (Jth component):

δ ρ̃J = δρJ − ρ̇J T ,

δ p̃J = δpJ − ṗJ T ,
ṽJ = vJ + L̇,

density and pressure likewise depend on slicing only

 Gauge Transformation
• Vector:

B̃ (±1) = B (±1) + L̇(±1) ,

(±1) (±1)
H̃T = HT + kL(±1) ,
(±1) (±1)
ṽJ = vJ + L̇(±1) ,

• Spatial vector has no background component hence no dependence

on slicing at first order
Tensor: no dependence on slicing or threading at first order
• Gauge transformations and covariant representation can be
extended to higher orders
• A coordinate system is fully specified if there is an explicit
prescription for (T, Li ) or for scalars (T, L)
 Slicing
Common choices for slicing T : set something geometric to zero
• Proper time slicing A = 0: proper time between slices
corresponds to coordinate time – T allows c/a freedom
• Comoving (velocity orthogonal) slicing: v − B = 0, slicing is
orthogonal to matter 4 velocity - T fixed
• Newtonian (shear free) slicing: ḢT − kB = 0, expansion rate is
isotropic, shear free, T fixed
• Uniform expansion slicing: −(ȧ/a)A + ḢL + kB/3 = 0,
perturbation to the volume expansion rate θ vanishes, T fixed
• Flat (constant curvature) slicing, δ (3) R = 0, (HL + HT /3 = 0),
T fixed
• Constant density slicing, δρI = 0, T fixed
 Threading
• Threading specifies the relationship between constant spatial
coordinates between slices and is determined by L
Typically involves a condition on v, B, HT
• Orthogonal threading B = 0, constant spatial coordinates
orthogonal to slicing (zero shift), allows δL = c translational
freedom
• Comoving threading v = 0, allows δL = c translational
freedom.
• Isotropic threading HT = 0, fully fixes L
 Newtonian (Longitudinal) Gauge
• Newtonian (shear free slicing, isotropic threading):

B̃ = H̃T = 0
Ψ ≡ Ã (Newtonian potential)
Φ ≡ H̃L (Newtonian curvature)
L = −HT /k
T = −B/k + ḢT /k 2

Good: intuitive Newtonian like gravity; matter and metric

algebraically related; commonly chosen for analytic CMB and
lensing work
Bad: numerically unstable
 Newtonian (Longitudinal) Gauge
• Newtonian (shear free) slicing, isotropic threading B = HT = 0 :
 
ȧ
(k 2 − 3K)Φ = 4πGa2 δρ + 3 (ρ + p)v/k Poisson + Momentum
a
k 2 (Ψ + Φ) = −8πGa2 pΠ Anisotropy

so Ψ = −Φ if anisotropic stress Π = 0 and

 
d ȧ ȧ
+ 3 δρ + 3 δp = −(ρ + p)(kv + 3Φ̇) ,
dη a a
 
d ȧ 2 K
+ 4 (ρ + p)v = kδp − (1 − 3 2 )p kΠ + (ρ + p) kΨ ,
dη a 3 k

• Newtonian competition between stress (pressure and viscosity)

and potential gradients
• Note: Poisson source is the density perturbation on comoving
slicing
 Comoving Gauge
• Comoving gauge (comoving slicing, isotropic threading)

B̃ = ṽ (Ti0 = 0)
HT = 0
ξ = Ã
R = H̃L (comoving curvature)
∆ = δ̃ (total density pert)
T = (v − B)/k
L = −HT /k

Good: Algebraic relations between matter and metric;

comoving curvature perturbation obeys conservation law
Bad: Non-intuitive threading involving v
 Comoving Gauge
• Euler equation becomes an algebraic relation between stress and
potential
 
2 3K
(ρ + p)ξ = −δp + 1 − 2 pΠ
3 k
• Einstein equation lacks momentum density source

ȧ K
ξ − Ṙ − 2 kv = 0
a k
Combine: R is conserved if stress fluctuations negligible, e.g.
above the horizon if |K|  H 2
   
ȧ δp 2 3K p
Ṙ + Kv/k = − + 1− 2 Π →0
a ρ+p 3 k ρ+p
 “Gauge Invariant” Approach
• Gauge transformation rules allow variables which take on a
geometric meaning in one choice of slicing and threading to be
accessed from variables on another choice
• Functional form of the relationship between the variables is gauge
invariant (not the variable values themselves! – i.e. equation is
covariant)
• E.g. comoving curvature and density perturbations

1 ȧ
R = HL + HT − (v − B)/k
3 a
ȧ
∆ρ = δρ + 3(ρ + p) (v − B)/k
a
 Newtonian-Comoving Hybrid
• With the gauge in(or co)variant approach, express variables of one
gauge in terms of those in another – allows a mixture in the
equations of motion
• Example: Newtonian curvature and comoving density

(k 2 − 3K)Φ = 4πGa2 ρ∆
ordinary Poisson equation then implies Φ approximately constant
if stresses negligible.
• Example: Exact Newtonian curvature above the horizon derived
through comoving curvature conservation
Gauge transformation
ȧ v
Φ=R+
ak
Hybrid “Gauge Invariant” Approach
Einstein equation to eliminate velocity

ȧ
Ψ − Φ̇ = 4πGa2 (ρ + p)v/k
a
Friedmann equation with no spatial curvature
 2
ȧ 8πG 2
= aρ
a 3
With Φ̇ = 0 and Ψ ≈ −Φ
ȧ v 2
=− Φ
ak 3(1 + w)
 Newtonian-Comoving Hybrid
Combining gauge transformation with velocity relation

3 + 3w
Φ= R
5 + 3w
Usage: calculate R from inflation determines Φ for any choice of
matter content or causal evolution.
• Example: Scalar field (“quintessence” dark energy) equations in
comoving gauge imply a sound speed δp/δρ = 1 independent of
potential V (φ). Solve in synchronous gauge.
 Sachs-Wolfe Effect
• On superhorizon scales kη  1, ∆ρ/ρ  R and ξ  R in
comoving gauge for adiabatic perturbations p(ρ)
• Both R and Ψ, Φ are constant when w =const.
• Derive the observed CMB temperature fluctuation for
superhorizon fluctuations at recombination: Sachs-Wolfe effect
• Time shift from Newtonian lapse Ψ to comoving lapse ξ  Ψ
Z
T ≡ δη = a−1 δt ≈ a−1 aΨdη ≈ a−1 Ψt
R
• Fractional time shift δt/t = Ψ, and t = d ln a/H ∝ a3(1+w)/2 so

δt 3 δa
= (1 + w) = Ψ
t 2 a
 Sachs-Wolfe Effect
• TCMB ∝ a−1 so
δTCMB δa 2
=− =− Ψ
TCMB local a 3(1 + w)
• Correct for gravitational redshift from climbing out of Ψ
δTCMB δTCMB 1 + 3w
= +Ψ=
TCMB obs TCMB local 3(1 + w)Ψ
• So in matter domination Ψ/3 = R/5 and in radiation domination
Ψ/2 = R/3
• So measurement of δTCMB /TCMB ≈ 10−5 implies the initial
comoving curvature R ≈ 5 × 10−5 or |R|2 = As ≈ 2.5 × 10−9
• In the deceleration epoch the comoving horizon grows, so this can
only have come about if the early universe had a period of
acceleration
 Synchronous Gauge
• Synchronous: (proper time slicing, orthogonal threading )

à = B̃ = 0
1
ηT ≡ −H̃L − H̃T
3
hL ≡ 6HL
Z
T = a−1 dηaA + c1 a−1
Z
L = − dη(B + kT ) + c2

Good: stable, the choice of numerical codes and separate

universe constructs
Bad: residual gauge freedom in constants c1 , c2 must be
specified as an initial condition, intrinsically relativistic,
threading conditions breaks down beyond linear regime if c1 is
fixed to CDM comoving.
 Synchronous Gauge
• The Einstein equations give
K 2 v
η̇T − 2 (ḣL + 6η̇T ) = 4πGa (ρ + p) ,
2k k
ȧ
ḧL + ḣL = −8πGa2 (δρ + 3δp) ,
a
2 1 ȧ
−(k − 3K)ηT + ḣL = 4πGa2 δρ
2a
[choose (1 & 2) or (1 & 3)] while the conservation equations give
 
d ȧ ȧ 1
+ 3 δρJ + 3 δpJ = −(ρJ + pJ )(kvJ + ḣL ) ,
dη a a 2
 
d ȧ vJ 2 K
+ 4 (ρJ + pJ ) = δpJ − (1 − 3 2 )pJ ΠJ .
dη a k 3 k
 Synchronous Gauge
• Lack of a lapse A implies no gravitational forces in Navier-Stokes
equation. Hence for stress free matter like cold dark matter, zero
velocity initially implies zero velocity always.
• Choosing the momentum and acceleration Einstein equations is
good since for CDM domination, curvature ηT is conserved and ḣL
is simple to solve for.
• Choosing the momentum and Poisson equations is good when the
equation of state of the matter is complicated since δp is not
involved. This is the choice of CAMB.
Caution: since the curvature ηT appears and it has zero CDM
source, subtle effects like dark energy perturbations are important
everywhere
 Spatially Flat Gauge
• Spatially Flat (flat slicing, isotropic threading):

H̃L = H̃T = 0
L = −HT /k
à , B̃ = metric perturbations
 −1  
ȧ 1
T = HL + HT
a 3

Good: eliminates spatial metric evolution in ADM and

perturbation equations ; useful in inflationary calculations
(Mukhanov et al)
Bad: non-intuitive slicing (no curvature!) and threading
• Caution: perturbation evolution is governed by the behavior of
stress fluctuations and an isotropic stress fluctuation δp is gauge
dependent.
 Uniform Density Gauge
• Uniform density: (constant density slicing, isotropic threading)

HT = 0 ,
ζI ≡ HL
BI ≡ B
AI ≡ A
δρI
T =
ρ̇I
L = −HT /k

Good: Curvature conserved involves only stress energy

conservation; simplifies isocurvature treatment
Bad: non intuitive slicing (no density pert! problems beyond
linear regime) and threading
 Uniform Density Gauge
• Einstein equations with I as the total or dominant species
 2
ȧ ȧ ȧ
(k 2 − 3K)ζI − 3 AI + 3 ζ̇I + kBI = 0 ,
a a a
ȧ K 2 v − BI
AI − ζ̇I − BI = 4πGa (ρ + p) ,
a k k
• The conservation equations (if J = I then δρJ = 0)
 
d ȧ ȧ
+ 3 δρJ + 3 δpJ = −(ρJ + pJ )(kvJ + 3ζ̇I ) ,
dη a a
 
d ȧ vJ − B I 2 K
+ 4 (ρJ + pJ ) = δpJ − (1 − 3 2 )pJ ΠJ + (ρJ + pJ )AI .
dη a k 3 k
 Uniform Density Gauge
• Conservation of curvature - single component I
ȧ δpI 1
ζ̇I = − − kvI .
a ρI + pI 3
• Since δρI = 0, δpI is the non-adiabatic stress and curvature is
constant as k → 0 for adiabatic fluctuations pI (ρI ).
• Note that this conservation law does not involve the Einstein
equations at all: just local energy momentum conservation so it is
valid for alternate theories of gravity
• Curvature on comoving slices R and ζI related by
1 ρ∆I
ζI = R + .
3 (ρI + pI ) comoving

and coincide above the horizon for adiabatic fluctuations

 Uniform Density Gauge
• Simple relationship to density fluctuations in the spatially flat
gauge
1 δ ρ̃I
ζI = .
3 (ρI + pI ) flat

• For each particle species δρ/(ρ + p) = δn/n, the number density

fluctuation
• Multiple ζJ carry information about number density fluctuations
between species
• ζJ constant component by component outside horizon if each
component is adiabatic pJ (ρJ ).
 Unitary Gauge
• Given a scalar field φ(xi , η), choose a slicing so that the field is
spatially uniform φ(xi , η) = φ(η) via the transformation

δφ
δ φ̃ = δφ − φ̇T → T =
φ̇0
• Specify threading, e.g. isotropic threading L = −HT /k
Good: Scalar field carried completely by the metric; EFT of
inflation and scalar-tensor theories of gravity. Extensible to
nonlinear perturbations as long as ∂µ φ remains timelike
Bad: Preferred slicing retains only the spatial diffeomorphism
invariance; can make full covariance and DOF counting obscure
• For a canonical scalar field, unitary and comoving gauge coincide
 EFT of Dark Energy and Inflation
• Beyond linear theory, unitary gauge and ADM is useful to define
most general Lagrangian and interaction terms for a scalar-tensor
theory of gravity: so-called Effective Field Theory (EFT)
• Rule: broken temporal diffeomorphisms (preferred slicing) but
spatial diffeomorphism invariance means explicit functions of
unitary time and ADM spatial scalars allowed
• Typically also want second order in time derivatives to avoid
Ostrogradsky ghost, lapse and shift non-dynamical

L(N, K ij , Ri j , ∇i ; t)
where the function is constructed out of spatial scalars (3D
Riemann tensor can be expressed through 3 Ricci tensor and metric
• Recall that the extrinsic curvature carries a first time derivative of
the spatial metric and spatial gradients of the shift
 EFT of Dark Energy and Inflation
• This class includes quintesence, k-essence, f (R), Horndeski,
“beyond Horndeski”, Horava-Liftshiz gravity, ghost condensate
• Does not include theories where derivatives of the lapse N appear
but the shift is still nondynamical due to hidden constraints - can
be generalized
• GR: time diffeomorphism not broken so the Einstein-Hilbert
Lagrangian in EFT language is given by the Gauss-Codazzi
relation
(4)
R → Kµν K µν − K 2 + (3) R
 EFT of Dark Energy and Inflation
• Now consider the scalar field to pick out a particular foliation
• Simplest example k-essence where L(X, φ) where
X = g µν ∂µ φ∂ν φ (sometimes −1/2× to resemble kinetic energy)
• In unitary gauge φ is a function of the temporal coordinate only so
X = −φ̇2 /N 2 so that

L(X, φ) → L(N, t)

• We will return to this case when considering inflationary

non-Gaussianity: note that g 00 = −1/N 2 so the EFT literature
sometimes writes L(g 00 , . . . , t) (Cheung et al 2008)
• Unifying description for “building blocks” of dark energy
(Gleyzes, Langois, Vernizzi 2015)
 Vector Gauges
• Vector gauge depends only on threading L
• Poisson gauge: orthogonal threading B (±1) = 0, leaves constant L
translational freedom
(±1)
• Isotropic gauge: isotropic threading HT = 0, fixes L
• To first order scalar and vector gauge conditions can be chosen
separately
• More care required for second and higher order where scalars and
vectors mix