Scott A.

Hughes General relativity

Massachusetts Institute of Technology

Department of Physics
8.962 Spring 2024

Lecture 14
Solving the Einstein field equations: Weak field solutions

14.1 Solving the Einstein field equation: General considerations

For the remainder of 8.962, we will focus on solving the Einstein field equation Gµν = 8πGTµν and
studying the properties of its solutions (including the cosmological constant in our analysis for one
particular class of solutions).
When this equation is regarded as a statement about the geometry of spacetime and how it
relates to the flow and density of energy and momentum in this spacetime, terms like “simple” and
“beautiful” tend to be used. Indeed, as we saw in Lecture 13, Einstein’s general relativity is in a very
real sense the simplest possible theory of gravity that respects general covariance and the principle
of equivalence, arising from the simplest Lagrangian that can be constructed from curvature. The
result is a straightforward equivalence between a divergence-free tensor that describes spacetime
curvature and the flow of energy and momentum through spacetime.
On the other hand, if we regard the field equations as differential equations that specify the
spacetime metric — which is often what we want for the purpose of physical analysis — then they
are a rather horribly complicated mess. Consider the situation for a perfect fluid. On the left-hand
side, we have the Einstein tensor, which we can regard as a differential operator acting on the
metric of spacetime:
2
Gµν = Dµν [gαβ ] . (14.1)
2 is a 2nd-order, nonlinear partial differential operator. On the right, we
In this schematic form, Dµν
have
Tµν = (ρ + P )uµ uν + P gµν . (14.2)
The spacetime metric thus appears explicitly on the right-hand side, as well as under the operator
on the left-hand side. It is also present implicitly on the right-hand side via the requirement that
the four velocity be properly normalized: uµ uν gµν = −1.
So as a differential equation for the metric, the Einstein field equation sets us up for what
appears likely to be a rather challenging analysis. To make progress, we will implement three
approaches to solving this equation:

1. “Weak” gravity. Here we imagine that spacetime is not very different (a statement that must
be made quantitative) from that of special relativity, at least in some coordinate systems,
and we will linearize the field equations in that deviation from ηµν . Such an analysis can be
regarded as the first order in a perturbative expansion. We will discuss in schematic terms
how one can iterate this expansion to higher order in the last week of this course.

2. Impose a symmetry. By imposing a symmetry, the complexity of the operator Dµν 2 and of any

source terms is greatly reduced. The coupled nonlinear nastiness that remains is one that can
be handled without too much trouble; this is how some of the most important solutions in
general relativity have been constructed. By adding perturbations around these symmetric
but exact solutions, we can do a tremendous amount of useful analysis. (Indeed, a rather
large chunk of your lecturer’s career is based on studying perturbations around a particular
exact solution.)

87
 3. Just deal with it: Code up the equations and attack. Numerical analysis can be a very pow-
erful tool, provided we understand its limitations and can assess whether our solutions are
converging to a description of the correct physics. Especially since about 2005, numerical
methods for solving the Einstein field equations have been particularly productive; but it
took decades of foundational work to understand how to cast the equations into a form that
was amenable to this method of building solutions. We will discuss work in this vein in the
last week of this course.

We begin by studying the linearized field equations.

14.2 The “weak-field” limit: Linearized gravity

The weak-field limit is defined by the idea that spacetime is nearly flat. We choose coordinates
such that
gµν = ηµν + hµν , (14.3)
and in which the components describing the deviation from flatness are small enough that we
can linearize with negligible error: ||hµν || ≪ 1, so terms O(h2 ) can be neglected. Whether this
linearization is accurate or not is ultimately a matter of measurement precision. It is conceivable
that measurements which agree with the predictions of linearized theory will cease to agree if
measurement techniques are significantly improved. In this part of 8.962, we take the ansatz that
the linear approximation is acceptable, and develop this limit of general relativity in detail.
Coordinates in which one can write the spacetime metric in the form (14.3) are known as “nearly
Lorentz coordinates.” One can think of them as coordinates that hold in a convenient freely falling
frame and are then extended to cover the entire manifold of (nearly flat) spacetime. Note that we
have borrowed some of the ideas and results from this analysis in earlier sections of this course,
particularly in making sure that our field equation included a Newtonian limit. Our goal now is to
put this framework from which we have been borrowing onto solid ground.
Because the spacetime metric is a tensor, it of course is subject to the general rule for changing
representation:
∂xα ∂xβ
gµ̄ν̄ = gαβ . (14.4)
∂xµ̄ ∂xν̄
When the spacetime is flat, we picked out a special category of representation changes, the Lorentz
transformations, as playing an important role. One thing that makes the Lorentz transformations
special is that the metric has the same representation in all Lorentz frames:

ηµ̄ν̄ = Λα µ̄ Λβ ν̄ ηαβ , (14.5)

and both ηµ̄ν̄ and ηαβ are represented by the matrix diag(−1, 1, 1, 1).
The Lorentz transformation really only has physical meaning in special relativity — in general
relativity, it is not generally applicable except within the confines of a local Lorentz frame. However,
in the case of nearly flat spacetimes, we can extend this concept in a useful way. Begin by applying
the Lorentz transformation to a nearly flat spacetime metric:

gµ̄ν̄ = Λα µ̄ Λβ ν̄ (ηαβ + hαβ )

= ηµ̄ν̄ + Λα µ̄ Λβ ν̄ hαβ
≡ ηµ̄ν̄ + hµ̄ν̄ . (14.6)

The background flat spacetime is unchanged by the Lorentz transformation, but the “perturbation”
hαβ transforms just like any tensor field in special relativity. In this circumstance, we call this
transformation a background Lorentz transformation, since we are taking advantage of the fact that
the flat background to our perturbation respects Lorentz invariance.

88
 The relationship hµ̄ν̄ = Λα µ̄ Λβ ν̄ hαβ tells us that a useful fiction is to regard hαβ as simply a
tensor field that lives in flat spacetime. This is a fiction because in fact the spacetime is curved,
but it is a useful one for us because it gives us a calculational “narrative” that is very similar to
how we describe many field theories in special relativity.
Another operation that we need to work out is the form of the inverse metric in linearized
theory. We begin by raising indices:

hαβ = g αµ g βν hµν
= η αµ η βν hµν + O(h2 ) . (14.7)

We always find that indices related to spacetime tensors are raised and lowered using the background
metric to linear order, which is consistent with the idea that we can treat hµν as a tensor field in
a flat background. With this in hand, let us calculate the inverse metric:

δαγ = g αβ gβγ
= (η αβ + mαβ ) (ηβγ + hβγ ) . (14.8)

The first line here is just the definition of the metric inverse g αβ . On the second line, we have
introduced a tensor mαβ whose elements are of the same order as the elements of hαβ . We leave it
undetermined in order to see what value it must have in order to provide the correct inverse metric.
Expanding, we have
δ α γ + hα γ + mα γ + O(h2 ) = δ α γ , (14.9)
which tells us that
hα γ = −mα γ (14.10)
to linear order. Therefore,
g αβ = η αβ − hαβ (14.11)
to linear order in the deviation from ηαβ . This is a form that we used earlier without careful
justification in our analysis of the Newtonian limit, although this result should intuitively remind
you of the functional expansion 1/(1 + x) ≃ 1 − x for small x.

14.3 Gauge transformations in linearized gravity

Consider next making a coordinate shift, defined as
′
xα = xα + ξ α (xβ ) . (14.12)

It must be acknowledged that this is not the best notation — this form abuses the index notation
′
somewhat. Calling the new coordinate (x′ )α rather than xα would be more consistent with the
way we use indices in relativity. However, it is best to live with this inconsistency. Our goal here
is not to introduce a covariant relationship, but simply to indicate how two coordinate systems are
related to one another. In particular, one of the reasons that this somewhat irritatingly inconsistent
notation is commonly used is that we wish to introduce the coordinate transformation matrix
′
Lα β = δ α β + ∂ β ξ α . (14.13)

We imagine that both the new and the old coordinates are nearly inertial, and require the matrix of
partials to be very small: ||∂α ξ β || ≪ 1. We call this an “infinitesimal” coordinate transformation.
Using these properties, it is simple to show that the inverse transformation is given by

Lα β ′ = δ α β − ∂β ξ α + O[(∂ξ)2 ] . (14.14)

89
 Let us now examine how the metric changes under this coordinate transformation:

gµ′ ν ′ = Lα µ′ Lβ ν ′ gαβ
= (δ α µ − ∂µ ξ α )(δ β ν − ∂ν ξ β )(ηαβ + hαβ )
= ηµν + hµν − ∂µ ξν − ∂ν ξµ + O(h ∂ξ) . (14.15)

The effect of this coordinate transformation is to change the perturbation around flat spacetime:

hµν → hµν − ∂µ ξν − ∂ν ξµ . (14.16)

This should remind you of a gauge transformation in electrodynamics, in which one adjusts the
potential Aµ → Aµ − ∂µ Λ. We thus call this operation a gauge transformation in linearized gravity.
Just as the gauge transformation in electrodynamics changes potentials but leaves the field
unchanged, the linearized gravity gauge transformation has no effect on curvature. It is simple to
show that the Riemann tensor is given by
1
Rµανβ = (∂α ∂ν hµβ + ∂µ ∂β hαν − ∂α ∂β hµν − ∂µ ∂ν hαβ ) (14.17)
2
in linearized theory; just expand the formula given in terms of Christoffel symbols, discarding the
nonlinear terms and using the fact that we raise and lower indices with the flat spacetime metric
(which commutes with partial derivatives since it is a constant). If one changes the metric according
to (14.16), then Riemann is apparently shifted by

1
δRµανβ = −∂α ∂ν ∂µ ξβ − ∂α ∂ν ∂β ξµ − ∂µ ∂β ∂α ξν − ∂µ ∂β ∂ν ξα
2

+∂µ ∂ν ∂α ξβ + ∂α ∂β ∂ν ξµ + ∂α ∂β ∂µ ξν + ∂µ ∂ν ∂η ξα . (14.18)

Because partial derivatives commute, one can see that δRµανβ = 0. We will take advantage of this
property to adjust our gauge in such a way that curvature tensors, written as derivatives of the
metric, take a particularly convenient form for our analysis.

14.4 The linear gravity field equation and its leading solution
Having already built Riemann in order to examine its behavior under gauge transformations, let’s
build the various contractions that we need for the field equation. We start with the Ricci tensor:

Rαβ = η µν Rµανβ
1
= (∂α ∂ µ hµβ + ∂β ∂ µ hµα − ∂α ∂β h − □hαβ ) . (14.19)
2
Here □ = ∂ µ ∂µ = η µν ∂µ ∂ν is the wave operator in flat spacetime, and we have introduced h =
hµ µ = η µν hµν , the trace of the metric perturbation.
The Ricci scalar follows simply from this:

R = η αβ Rαβ = ∂ α ∂ µ hµα − □h . (14.20)

Finally, the linearized Einstein tensor is given by

1
Gαβ = Rαβ − ηαβ R
2
1
= (∂α ∂ hµβ + ∂β ∂ µ hµα − ∂α ∂β h − □hαβ + ηαβ □h − ηαβ ∂ µ ∂ ν hµν ) .
µ
(14.21)
2

90
 The presence of the trace h makes this expression somewhat unwieldy. Recalling that Gαβ is a
trace-reversed version of the Ricci curvature, it is worth seeing what happens if we build it using a
trace-reversed version of the metric perturbation. We define
1
h̄αβ = hαβ − ηαβ h ; (14.22)
2
notice that η αβ h̄αβ = h − 2h = −h. We then write hαβ = h̄αβ + 21 ηαβ h, and substitute this into the
Einstein tensor (14.21). The result is

1 1 1
Gαβ = ∂α ∂ µ h̄µβ + ηµβ ∂α ∂ µ h + ∂β ∂ µ h̄µα + ηµα ∂β ∂ µ h − ∂α ∂β h − □h̄αβ
2 2 2

1 1
− ηαβ □h + ηαβ □h − ηαβ ∂ µ ∂ ν h̄µν − ηαβ ηµν ∂ µ ∂ ν h . (14.23)
2 2

Carefully examining this expression, we see that all the terms involving the trace h cancel out: the
second, fourth, and fifth terms on the top line of the right-hand side sum to zero, as do the first,
second, and fourth expressions on the bottom line of the right-hand side. What remains is
1
∂α ∂ µ h̄µβ + ∂β ∂ µ h̄µα − ηαβ ∂ µ ∂ ν h̄µν − □h̄αβ .


Gαβ = (14.24)
2
It should be emphasized that this trick is really nothing more than algebraic sleight of hand;
however, it is no less useful for being essentially a bit of clever rearranging.
Notice that in what remains all the terms that are left except the very last term inside the
parentheses involve divergences of the metric perturbation: the first three terms are each of the
form ∂ µ h̄µν (or something involving a different index than ν). Just as in electromagnetic theory,
we can use our gauge freedom to clean this up. We would like to set

∂ µ h̄µν = 0 , (14.25)

a total of 4 conditions. The quantities ξν which generate the infinitesimal coordinate transformation
provide us with 4 four free functions to do this. Writing

hnew old
µν = hµν − ∂µ ξν − ∂ν ξµ , (14.26)

we trace reverse and find that

h̄new old α
µν = h̄µν − ∂µ ξν − ∂ν ξµ + ηµν ∂ ξα . (14.27)

Take the divergence of this:

∂ µ h̄new
µν = ∂ µ h̄old µ µ α
µν − □ξν − ∂ν ∂ ξµ + ηµν ∂ ∂ ξα
= ∂ µ h̄old
µν − □ξν . (14.28)

(To simplify, we used the metric to lower the index on the last term and recognized that µ and α
are dummy indices on the final two terms on the top line.) This tells us that if we choose our gauge
generators to satisfy
□ξν = ∂ µ h̄old
µν , (14.29)
then the metric in our new gauge will be divergence free. Solutions of (14.29) always exist as long as
h̄old
µν is well behaved, and this metric perturbation will always be well behaved in the weak gravity
applications to which we apply this framework. We call this gauge “Lorenz gauge,” following the

91
essentially identical gauge choice used in electrodynamics to make the potential 4-vector divergence
free. When we are in Lorenz gauge, the Einstein tensor is given by
1
Gαβ = − □h̄αβ , (14.30)
2
and we find the linearized Einstein field equation takes the form

□h̄αβ = −16πGTαβ (14.31)

This is starkly reminiscent of the sourced Maxwell’s equation written in terms of the potential in
Lorenz gauge, and very similar techniques can be used to construct its solution.

14.5 Static solutions

Let us begin by considering a perfect fluid source that is static (ut = 1 + O(h), ui = 0) and
non-relativistic (ρ ≫ P , corresponding to csound ≪ c). We can write the stress energy tensor

Tαβ ≃ ρuα uβ . (14.32)

Thanks to the static condition, the only non-negligible component is T00 = ρ + O(h). The only
non-trivial field equation component is then

□h̄00 = −16πGρ . (14.33)

Including the O(h) correction would lead to corrections to the solution h̄00 we find which are
themselves second-order in h.
Since the source is static, the differential operator simplifies to

∇2 h̄00 = −16πGρ . (14.34)

Solutions to this equation have a form reminiscent of a Coulomb potential. In the case of gravity,
we recognize it as the equation for the Newtonian potential Φ (modulo a factor −4), and deduce
that the solution is
h̄00 = −4Φ ; all other h̄µν = 0 . (14.35)
To get the metric perturbation, we trace reverse again:
1
hµν = h̄µν − ηµν h̄ , (14.36)
2
where the trace h̄ = η µν h̄µν = η 00 h̄00 = (−1)(−4Φ) = 4Φ. This tells us that

1
h00 = −4Φ − (−1)(4Φ)
2
= −2Φ , (14.37)
1
h11 = h22 = h33 = 0 − (1)(4Φ)
2
= −2Φ ; (14.38)

all off-diagonal components of hµν = 0 in this case. We see that

ds2 = −(1 + 2Φ)dt2 + (1 − 2Φ)(dx2 + dy 2 + dz 2 ) . (14.39)

This is the line element we used (without justification) to explore the Newtonian limit.

92
 On problem set 7, you will examine what happens when the body spins rigidly about an axis.
Just as the static problem leads to a potential that is a lot like the Coulomb potential governing
the electric field arising from the density field ρ, the rotating problem looks a lot like what we
get for the magnetic potential arising from a density flow J = ρv. The solution should remind
you quite a bit of problems you have already seen in an electromagnetic content, albeit with some
signs flipped, and various prefactors not quite the same. It should also be noted that linearity of
these field equations means that we can add any homogeneous solution of the field equation; i.e.,
any solution of (14.31) for which Tαβ = 0. Modulo some subtleties having to do with gauge, such
solutions can represent radiation. Hold that thought! — we throw away those solutions right now
because they are not static (our current focus), but will be a major focus of material we discuss
very soon.

93