General Relativity Fall 2019

Lecture 15: Linearized Einstein field equations

Yacine Ali-Haı̈moud
October 17th 2019

SUMMARY FROM PREVIOUS LECTURE

We are considering nearly flat spacetimes with nearly globally Minkowski coordinates: gµν = ηµν + hµν ,
with |hµν |  1. Such coordinates are not unique. First, we can make Lorentz transformations and keep a
globally-Minkowski coordinate system, with hµ0 ν 0 = Λµµ0 Λνν 0 hµν , so that hµν can be seen as a Lorentz tensor
field on flat spacetime. Second, if we make small changes of coordinates, xµ → xµ − ξ µ , with |∂µ ξ ν |  1, the
metric perturbation remains small and changes as hµν → hµν + 2ξ(µ,ν) . By analogy with electromagnetism, we can
see these small coordinate changes as gauge transformations, leaving the Riemann tensor unchanged at linear order.

Since we will linearize the relevant equations, we may work in Fourier space: each Fourier mode satisfies an
independent equation. We denote by ~k the wavenumber and by k̂ its direction and k its norm.

We have decomposed the 10 independent components of the metric perturbation according to their transformation
properties under spatial rotations: there are 4 independent “scalar” components, which can be taken, for instance,
to be h00 , k̂ i h0i , hii , and k̂ i k̂ j hij – or any 4 linearly independent combinations thereof. There are 2 independent
transverse “vector” components, each with 2 independent components: ilm k̂l h0m and ilm k̂l hmj k̂ j – these are
proportional to the curl of h0i and to the curl of the divergence of hij , and are divergenceless (transverse to the
Fourier wavenumber ~k). Finally, there is a “tensor” mode hTT ij , which is the transverse-trace-free part of hij ,
~
and is obtained by double-projecting hij transverse to k, and subtracting the trace. This piece has 2 independent
components. Explicitly, we have

hTT TT
ij = Pijmn hmn , (1)

where the TT-projection operator is given by

TT 1
Pijmn ≡ Pim Pjn − Pij Pmn , Pij ≡ δij − k̂i k̂j . (2)
2
Let us write explicitly the gauge transformation equations in Fourier space: replace ∂j by ikj :
h00 → h00 + 2∂0 ξ0 , h0j → h0j + ∂0 ξj + ikj ξ0 , hjl → hjl + 2ik(j ξl) . (3)

GAUGE-INVARIANT METRIC PERTURBATIONS

While a gauge transformation in electromagnetism amounts to providing one scalar function, a gauge transforma-
tion in linearized GR amounts to providing 4 functions ξ 0 , ξ. ~ These 4 functions can be decomposed in 2 scalars
ξ 0 , k̂i ξ i and 1 transverse vector ijk k̂ j ξ k . Therefore, we expect that out of the 4 scalars components, only 2
linear combinations are gauge-invariant. Similarly, out of the 2 vector modes, only 1 linear combination
is gauge-invariant. Finally, since there is no way to make a TT mode out of scalars and vectors, we expect, and
will show explicitly, that the “tensor” mode is gauge-invariant.

The two scalar gauge-invariant variables are not unique (any linear combination is also gauge-invariant). We’ll see
that these two make expressions particularly simple (for reference these are related to the gauge-invariant Bardeen
potentials in cosmological perturbation theory):
"  #
1  1 2ik̂j 3 2 1
Ψ≡ k̂j k̂l hjl − hjj , Φ ≡ − h00 + ∂0 h0j − 2 ∂0 k̂j k̂l hjl − hjj . (4)
4 2 k 2k 3

I encourage you to explicitly check gauge-invariance – and please report likely typos!
 2

There is only one gauge-invariant transverse vector mode, defined up to a normalization constant:
 
i
v i ≡ ilm k̂l h0m + ∂0 hmj k̂j (5)
k
Indeed, under a gauge transformation, the change in the parenthesis is a pure gradient, so its curl is zero.

Finally, since a gauge transformation cannot add a TT part to the metric: the TT part of the metric pertur-
bation is gauge-invariant, much like the transverse-vector part of the vector potential is in electromagnetism. I
encourage you to show this explicitly using the TT projection operator applied to a gauge transformation.

So, to summarize, gauge freedom implies that there are only 6 physical degrees of freedom in the metric
perturbation (that we could tell right away just from counting the number of coordinates). For linearized GR, we
can moreover explicilty identify these degrees of freedom and classify them as 2 scalar modes, 1 transverse vector
mode, and 1 transverse-traceless “tensor” mode. Note: it is always possible to set the metric to be Minkowski and
with vanishing first derivatives at any given point by using a LICS. So it is no surprise that the gauge-invariant
variables are defined with at least two derivatives of the metric.

LINEARIZED EINSTEIN TENSOR

The Riemann tensor takes the form Riemann ∼ ∂Γ + ΓΓ, where Γ are the Christoffel symbols. To linear order in
hµν , we only need to keep the first term. Furthermore, we only need to compute the Christoffel symbol at linear order
in hµν . This implies
1
Rαµβν = (∂µ ∂β hαν + ∂ν ∂α hβµ − ∂µ ∂ν hαβ − ∂α ∂β hµν ) , (6)
2
which is the same expression as that in a LICS, derived in lecture 11. From this we obtain the Ricci tensor, by
contracting the first and third indices:
1
Rµν = Rαµαν = (∂µ ∂ α hαν + ∂ν ∂ α hαµ − ∂µ ∂ν h − hµν ) , (7)
2
where h ≡ hµµ ≡ η µν hµν is the trace of hµν (obtained using the Minkowski metric) and  ≡ ∂ µ ∂µ = η µν ∂µ ∂ν is the
D’Alembertian operator.
The Ricci scalar is obtained by taking the trace:

R = Rµµ = ∂ α ∂ β hαβ − h. (8)

From this we obtain the Einstein tensor at linear order in metric perturbations:
1
∂µ ∂ α hαν + ∂ν ∂ α hαµ − ∂µ ∂ν h − hµν − ηµν (∂ α ∂ β hαβ − h) .


Gµν = (9)
2

EINSTEIN FIELD EQUATIONS (EFES)

We first note that we can also decompose Gµν in scalars, vectors and tensors, and that each kind can only include
the components of hµν of the same kind: at linear order one cannot make a transverse vector mode out of a scalar,
and vice-versa.

Let us now explicitly write the EFEs. The 00 equation can be rewritten in terms of our gauge-invariant variable Ψ,

∇2 Ψ = 4πT00 . (10)

This is just Poisson’s equation. As in the case of electromagnetism, the (linearized) Bianchi identity G0i,i = G00,0 ,
consistent with the conservation of the stress-energy tensor Tµν,ν = 0 (at linear order), implies that the divergence
of the 0i equation is nothing but the time-derivative of the 00 equation, and therefore does not carry any additional
information. The curl of the 0i equation can be written in terms of the gauge-invariant vector field as

∇2 v i = 16πilm k̂l T0m (11)

 3

Out of the 6 purely spatial equations, 3 are redundant with the G0i equations, again, from the contracted Bianchi
identity. We are left with three independent equations. The first one can be taken to be Gij − 31 δij Gkk , which, upon
taking the double gradient, gives us [please report typos!]
 
4 1
∇ (Φ − Ψ) = −12π∂i ∂j Tij − δij Tkk . (12)
3

The right-hand-side is proportional to the anisotropic stress, i.e. the stress tensor Tij minus its isotropic (i.e. diag-
onal) part, proportional to the pressure. This equation shows that Φ = Ψ if there are no anisotropic stresses.

Finally, the transverse-trace-free part of Gij gives us the following equation for the TT part of the metric pertur-
bation:

hTT TT
ij = −16πTij , TijTT ≡ Pijmn
TT
Tmn . (13)

As expected, the EFEs only provide information about the 6 gauge-invariant, physical degrees of freedom. To fix all
10 components of hµν one needs to impose 4 additional and freely specifiable gauge (or coordinate) conditions.

Constraints and dynamics

Now, let us consider the character of these equations. The first three are constraint equations: they do not
involve any time derivatives. It is easier to see this explicitly in the transverse gauge, which is the generalization
of the Coulomb gauge in electromagnetism: this gauge is defined by the 4 conditions
 
1
∂i h0i = 0 = ∂i hij − δij hkk . (14)
3

First, one needs to show explicitly that such a gauge choice is indeed possible, and it is: starting from a coordinate
system in which the gauge condition is not satisfied, make a gauge transformations such that ∇2 ξ i + 31 ∂j ξl,l =
−2∂j (hij − 13 δij hkk ) and ∇2 ξ 0 = ∂i h0i + ∂0 ξ,ii . The metric perturbation in these new coordinates will satisfy the
transverse gauge conditions.

In the transverse gauge, we have Ψ = − 16 hkk and Φ = − 12 h00 , and v i = ilm k̂l h0m , i.e. no time derivatives appear,
and the equations for Φ, Ψ and v i are clearly purely spatial, constraint equations.

Finally, the TT part is a truly dynamical equation for the gauge-invariant tensor mode hTT ij . Therefore,
just like electromagnetism, GR has two dynamical degrees of freedom, the TT part of the metric. These are
the famous gravitational waves, which propagate at the speed of light (they satisfy the wave equation in vaccum),
and can exist even in vacuum, while all other components can be set to zero in vacuum by appropriate gauge
choices.

While we have shown all this in linearized gravity, let’s now mention how this carries over to non-linear GR. First,
it remains true at the non-linear level that there are 6 physical degrees of freedom, due to the 4 coordinate degrees of
freedom. However, one can no longer classify them into “scalars”, “vectors” and “tensors” without a flat background.
At the non-linear level, all these mix-up; for instance, the non-linear quantity hTT TT
ij hij is a scalar.
Second, it remains true even in non-linear GR that the 0µ Einstein field equations (with both indices up)
are constraint equations: G0µ do not contain any second-time derivative (in fact, G00 contains no time derivative
at all). This can be seen using the Bianchi identity: ∂0 G0µ = −∂i Giµ . Giµ contains at most two time derivatives (it
contains at most second derivatives of the metric), thus ∂i Giµ also contains at most two time derivatives, hence G0µ
contains at most one time derivative. It also remains true that the metric only has two dynamical degrees of freedom,
though, again, one cannot explicitly identify them in non-linear gravity.
 4

THE HARMONIC GAUGE

We define the trace-reversed metric perturbation

1 1
hµν ≡ hµν − ηµν h ⇒ hµν = hµν + ηµν h. (15)
2 2
The Einsetein tensor takes the following expression in terms of hµν :

1 1
Gµν = − hµν + ∂ α hα(µ,ν) − ηµν ∂ α ∂ β hαβ . (16)
2 2
The harmonic or Lorenz gauge is the generalization of the Lorenz gauge in electromagnetism, and consists of the 4
conditions

∂ µ hµν = 0 . (17)

Again, those are 4 conditions, which can imposed by appropriately chosing 4 functions ξ µ given some initial coordinates
in which they are not, in general, satisfies.
We see that in this gauge the Einstein tensor simplifies to Gµν = − 21 hµν , so the Einstein field equations take on
the simple form

hµν = −16πTµν . (18)