CHAPTER 6

EINSTEIN EQUATIONS

You will be convinced of the

general theory of relativity once
you have studied it. Therefore I
am not going to defend it with a
single word.

A. Einstein

6.1 The energy-momentum tensor

Having decided that our description of the motion of test particles and light in a gravitational field
should be based on the idea of curved space times with a metric, we must now complete the theory by
postulating a law to say how the sources of the gravitational field determine the metric. To construct
this gravitational field equation we must first find a covariant way of expressing the source term ρ in
the Poisson equation
∇2 Φ = 4πGρ . (6.1)
It is clear that the relativistic generalization of Eq. (6.1) cannot simply involve ρ as the source of
the relativistic gravitational field, since ρ is the energy density measured by only one observer, that
at rest with respect to the fluid element. It is not the first time that we find this kind of situation.
The relativistic formulation of Maxwell equations needed the combination of the charge density ρe
and the charge current J i into a 4-vector J µ = (ρ, J i ) with the right transformation properties. Can
we do something similar here? The most naive trial would be a
combination of the energy density
ρ with someR energy flux si into a 4-vector, let’s say sµ = ρ, si . However, the total energy in this
case, E = ρ d3 x, is not a Lorentz invariant quantity

1
, due to its combination with the linear 3-
i µ i
momentum p into the 4-momentum p = E, p . We are forced therefore to look for a higher rank
object encoding the relation among the energy density, the energy flux, the momentum density and
the momentum flux or stress. This quantity is the so-called energy-momentum-stress tensor. Let’s
construct it.
1 Note ρe d3 x is Lorentz invariant.
R
that in the electromagnetic case, the total electric charge Q =
6.1 The energy-momentum tensor 84

6.1.1 Newtonian fluids

While point particles are characterized by their energy and momentum, the motion of continuous
matter is usually characterized by two quantities: the mass density ρ(t, xi ) and the velocity of the
fluid v(t, xi ), which generally depend on space and time. The evolution of a continuous system is
determined by two equations:

i) A continuity equation

∂ρ ∂ ρv j
+ = 0, (6.2)
∂t ∂xj
reflecting the fact that mass is neither created or destroyed in Classical Mechanics (the flowing
of mass out from a volume is equal to the loss of mass in it).
ii) A Newton’s 2nd law for fluids

∂v i ∂v i
 
i i
f = ρa = ρ + vj j , (6.3)
∂t ∂x

with
v i (t + ∆t, x + ∆x) − v i (t, x)
ai = lim , (6.4)
∆t→0 ∆t
and f i = f i (t, x) the total force per unit volume around a point x at time t. The so-called total
derivative of the velocity field
Dv i ∂v i ∂v i
≡ + vj j (6.5)
dt ∂t ∂x
contains two pieces, the local derivative ∂v/∂t, which gives the change of the velocity v as a
function of time at a given point in space, and the so-called convective derivative, (v · ∇) v, which
represents the change of v for a moving fluid particle due to the inhomogeneity of the fluid vector
field.
If we assume that there are not other forces apart from those exerted by the fluid on itself, we
are left with internal forces like pressure or friction acting only between neighboring regions of
matter. Consider a infinitesimal volume dV with surface area dA centered at a point x at time
t. Let us denate by nj the normal vector to the surface. In a perfect fluid2 , the force F i exerted
by the matter on the area is proportional to the area itself F i = p(t, x)δ ij nj dA, with p(t, x) the
pressure at that point at time t. In the most general case, we will also have shear forces

F i (t, x) = T ij (t, x)nj dA , (6.6)

due to the tendency of fluid elements moving with different velocities to drag adjacent matter.
The coefficients T ij are the components of the so-called stress tensor, which must be symmetric,
T ij = T ji .

Exercise
Consider the 3-component of the torque acting on an infinitesimal cube of a material of
density ρ and side length L. Compare it with the moment of inertia of the cube I = 61 ρL5 .
What happens if T ij 6= T ji in the limit L → 0?
2 A perfect fluid is defined as one for which there are no forces between the particles, no heat conduction and no

viscosity.
6.1 The energy-momentum tensor 85

The total force exerted per unit area in a given direction3 can be transformed into a total force
by unit volume via the Gauss’ theorem
∂T ij
Z Z
ij
∂j T ij dV fi = − j .


− T nj dA = − −→ (6.7)
A V ∂x
Plugging in this result into the Newton 2nd law (cf. Eq. (6.3))
 i
∂v i ∂T ij

∂v
ρ + vj j + = 0, (6.8)
∂t ∂x ∂xj
and using the continuity equation to write




∂v i ∂ ρv i i ∂ρ ∂ ρv i i ∂ ρv
j
ρ = −v = +v (6.9)
∂t ∂t ∂t ∂t ∂xj
The previous result and the continuity equation (6.2), the Newton’s 2nd law (6.3) for this partic-
ular case (f i = −∂j T ij ) can be written as


∂ ρv i ∂
ρv i v j + T ij = 0 ,


+ (6.10)
∂x0 ∂xj
which is the so-called Euler equation.

6.1.2 Relativistic fluids

Eqs. (6.2) and (6.10) can be unified into a single equation in the framework of Special Relativity.

To

see this, note that the 3-velocity v i is contained in the relativistic 4-velocity

uµ
= u0
, ui
= γ, γv i
.
µ i
Taking into account the non-relativistic limit of this relation, u = 1, v , we can rewrite (6.2) and
(6.10) as



∂ ρu0 u0 ∂ ρu0 uj ∂ ρui u0 ∂
ρui uj + T ij = 0 ,


0
+ j
= 0, 0
+ j
(6.11)
∂x ∂x ∂x ∂x
which can be considered as parts of the single equation

∂ν T µν = 0 , T µν = ρuµ uν + tµν , (6.12)

with tµν = diag 0, T . The quantity T µν is the so-called energy-momentum-stress tensor or in a
ij

shorter version the energy-momentum tensor 4 or the stress-energy tensor. It is a rank-2 symmetric
tensor encoding all the information about energy density, momentum density, stress, pressure . . . .
The ten components of this tensor have the following interpretation:

• T 00 is the local energy density, including any potential contribution from forces between particles
and their kinetic energy.
• T 0i is the energy flux in the i direction. This includes not only the bulk motion but also any
other processes giving rise to transfers of energy, as for instance heat conduction.
• T i0 is the density of the momentum component in the i direction, i.e. the 3−momentum density.
As the previous case, it also takes into account the changes in momentum associated to heat
conduction.
3 The minus sign appears because we are considering the force exerted on matter inside the volume by the matter

outside
4 This name can be sometimes misleading as it can be confused with the energy-momentum 4-vector pµ in sentences

including things like “the energy-momentum conservation equation. . . ”. The difference should be always clear from the
context.
6.2 The microscopic description 86

• T ij is the 3-momentum flux or stress tensor, i.e the rate of flow of the i momentum component
per unit area in the plane orthogonal to the j-direction. The component T ii encodes the isotropic
pressure in the i direction while the components T ij with i 6= j refer to the viscous stresses of
the fluid.

6.1.3 Relativistic perfect fluids

A relativistic perfect fluid is defined to be one in which the tµν part of the stress-energy tensor T µν ,
as seen in a local reference frame moving along with the fluid, has same form as the non-relativistic
perfect fluid  
0 0 0 0
 0 p 0 0 
tµν = 
 0 0 p 0 .
 (6.13)
0 0 0 p
Heat conduction, viscosity or any other transport or dissipative processes in this case are negligible.
The form of Eq. (6.13) in an arbitrary inertial frame can be obtained by performing a general Lorentz
transformation
γv i
  0
ui
 
γ u
Λµ ν = = , (6.14)
γv i δ ij + v i v j (γ − 1) /v 2 ui δ ij + ui uj /(1 + γ)

moving from the rest frame uµ = (1, 0) to one in which the fluid moves with 3-velocity v i . We get

t̄µν = Λν ρ Λν σ tρσ = p (η µν + uµ uν ) , (6.15)

with uµ the 4-velocity vector field tangent to the worldines of the fluid particles. Taking into account
this result, the full stress-energy tensor (6.12) takes the form

T µν = (ρ + p)uµ uν + pη µν . (6.16)

The resulting equation is manifestly covariant and can be easily generalized to arbitrary coordinate
systems or curved spacetimes by simply replacing the local metric η µν by a general metric g µν

T µν = (ρ + p)uµ uν + pg µν . (6.17)

The conservation law ∂ν T µν = 0 in Eq. (6.12) becomes a local conservation law

∇ν T µν = 0 (6.18)

in which the standard derivative ∂µ is replaced by the covariant derivative ∇µ . The word local is, as
always in this course, important. Eq. (6.18) is not a conservation law, nor should it be. As we will
see, energy is not conserved in the presence of dynamical spacetime curvature but rather changes in
response to it.

Exercise
Prove Eq. (6.15).

6.2 The microscopic description

The relation between ρ and p is usually characterized by an equation of state p = p(ρ) which depends on
the microscopic particles involved in the fluid. In order to get some insight about the possible equations
6.2 The microscopic description 87

of state, let me consider a macroscopic collection of N structureless point particles interacting through
spatially localized collisions. The energy density associated to any of them is given by

Tn00 = En δ (3) (x − xn (t)) = mn γn δ (3) (x − xn (t)) , (6.19)

p
with γn = 1/ 1 − vn2 and n = 1, . . . , N a label selecting the particular particle we are referring to.
Taking into account the identity
Z +∞ Z +∞
dτ δ (4) (x − x(τ )) = dτ δ (t − t(τ )) δ (3) (x − x(τ ))
−∞ −∞
dτ (3) 1
= δ (x − x(t)) = δ (3) (x − x(t)) , (6.20)
dt γ
the non-Lorentz invariant 3-dimensional Dirac delta appearing in Eq. (6.23) can be transformed into
a Lorentz invariant 4-dimensional Dirac delta5
Z +∞
00
Tn = mn dτn u0n u0n δ (4) (x − xn (τn )) . (6.21)
−∞

The same procedure can be applied to the spatial momentum density (or energy current) of the particle

Tn0i = pin δ (3) (x − xn (t)) = mn γn vni δ (3) (x − xn (t)) = En vni δ (3) (x − xn (t)) , (6.22)

and to the flux of the i momentum component in the j direction (or viceversa)

Tnij = pin vnj δ (3) (x − xn (t)) = pjn vni δ (3) (x − xn (t)) . (6.23)

We obtain
Z +∞ Z +∞
Tn0i = mn dτn u0n uin δ (4) (x − xn (τn )) , Tnij = mn dτn uin ujn δ (4) (x − xn (τn )) . (6.24)
−∞ −∞

Eqs. (6.21) and (6.24) can be rewritten in a very compact way in terms of the stress-energy-momentum
tensor T µν
Z +∞ Z +∞
pµ pν
Tnµν = mn dτn uµn uνn δ (4) (x − xn (τn )) = dτn n n δ (4) (x − xn (τn )) , (6.25)
−∞ −∞ mn
which is manifestly symmetric and Lorentz invariant since uµn uνn is a tensor under Lorentz transfor-
mations and both mn and dτn δ (4) (x − xn (τn )) are Lorentz scalars. The total energy density of the
whole system of particles can be written as the sum of the individual contributions, namely
N
X
T µν = Tnµν . (6.26)
n=1

6.2.1 Energy-momentum tensor conservation and geodesics

Let us see under which conditions the total energy momentum tensor (6.26) is conserved. Taking the
derivative with respect to the coordinates we get
N
X Z +∞
µν
∂µ T = mn dτn uµn uνn ∂µ δ (4) (x − xn (τn )) , (6.27)
n=1 −∞

5 The fact that the 4-Dimensional Dirac delta δ (4) (x) is Lorentz invariant follows directly from the definition

d4 xδ (4) (x) = 1 and the fact that the volume element d4 x is Lorentz invariant.
R
6.2 The microscopic description 88

which using
dxµn ∂ (4)
uµn ∂µ δ (4) (x − xn (τn )) = δ (x − xn (τn )) = −d/dτn δ (4) (x − xn (τn )) , (6.28)
dτn ∂xµ
can be written as
N Z +∞ N Z +∞
X d  ν (4)  X
∂µ T µν = − mn dτn un δ (x − xn (τn )) + mn dτn u̇νn δ (4) (x − xn (τn )) .
n=1 −∞ dτn n=1 −∞

The first term in the right hand side of the previous expression disappears in the particles are stable,
i.e. if the orbits are closed or come from negative infinite time and disappear into positive infinite
time. We are left then with the second term, which can be written as
N Z +∞ N
µν
X dpνn (4) X dpνn (3)
∂µ T = dτn δ (x − xn (τn )) = δ (x − xn ) , (6.29)
n=1 −∞ dτn n=1
dt

with pνn = mn uνn the 4-momentum of the individual particles. The local energy momentum conser-
vation ∂µ T µν = 0 requires the particles to be free. Or in other words, the condition ∂µ T µν = 0 is
equivalent to the geodesic equation in Minkowski space-time, dpµn /dτ = 0. This will be also the case
in curved spacetime.

6.2.2 The fluid limit

On distances d much larger than the typical mean free path a, the number of particles is large and
the statistical fluctuations about the mean properties of the fluid are expected to be small6 . Imagine
a comoving observer exploring distances d  a. If the fluid is isotropic7 , the average value of the
T 0i ∝ u0 ui component measured by this observer will be zero since the vector ui points in all possible
directions. In this case, the fluid can be characterized in terms of two quantities: its mean density
and pressure over the volume ∆V = d3
DX E 1 X D X i i (3) E
ρ= En δ (3) (x − xn ) , p= pn vn δ (x − xn ) . (6.30)
n
∆V 3 i n
∆V

A simple inspection of Eqs. (6.30) reveals that, for standard matter, 0 ≤ p ≤ ρ/3. In any other
reference frame, the energy-momentum tensor for the perfect fluid reads

T µν (x) = (ρ(x) + p(x))uµ (x)uν (x) + p(x)η µν , (6.31)

with uµ (x) denoting now the average value of the 4-velocities uµi of the individual particles NR inside
the volume8 . The perfect fluid form (6.31) can be used to model very different physical situations
that often fall into one of the following categories:
p
1. Non-relativistic matter: For small velocities the dispersion relation En = m2n + p2n can be
approximated by En ' mn +p2n /2mn , which plugged back into (6.30) gives rise to ρ ' mn n+ 32 p.
Taking into account that the statistical definition of temperature T is twice the energy possessed
by each degree of freedom and assuming a monoatomic gas with 3 kinetic degrees of freedom,
we can write T = (2/3) × p2n /2mn and therefore ρ ' mn n + 23 T .
6 Remember that, when we later apply the Equivalence Principle, we will have another scale into play: the scale

L at which the gravitational effects start to be important. If this scale happens to be much larger than the scale d
(L  d  a), the mean properties of the fluid can be safely considered as constant over the region.
7 i.e if the fluid is perfect.
8 Note that, when writing uµ (x), ρ(x) and p(x) we are explicitly taking into account that the averages can vary from

one region to another.

6.2 The microscopic description 89

2. Dust: A perfect fluid with zero pressure. p = 0, tµν = 0, T µν = ρ diag (1, 0, 0, 0).

3. Radiation: A perfect highly relativistic fluid. In this case En ' |pn |  mn and therefore9
ρ ' 3p. The energy momentum for radiation is traceless, T = T µ µ = ηµν T µν = −ρ + 3p = 0.

A worked-out example: The electromagnetic field

The paradigmatic case of a fluid with a radiation equation of state is the electromagnetic
field. To see this explicitly, consider the energy density of the electromagnetic field
1
E2 + B2 ,


T00 = (6.32)
2
and write it in the way seen by an observer moving with 4-velocity uµ . The electron field
seen by that observer is given by
Eµ = Fµν uν . (6.33)
Using this expression we get the following covariant expression for the square of the electric
field
E2 = Fµν uν F µ ρ uρ . (6.34)
A similar expression for the magnetic field square can be obtained from the explicit ex-
pression for the square of the electromagnetic field strength tensor

Fµν F µν = −2 E 2 − B 2 .


(6.35)

We obtain
1
B2 = Fµν uν F µ ρ uρ + Fµν F µν . (6.36)
2
Putting Eqs. (6.34) and (6.36) together, the covariant generalization of the energy density
(6.32) becomes  
1
ρ ρσ
ρ = Fρµ F ν − Fρσ F ηµν uµ uν , (6.37)
4
where we have inserted a factor uµ uµ = −1. The work is basically done. The quantity in
parenthesis is the sought-for energy-momentum tensor for the electromagnetic field!
1
Tµν = Fρµ F ρ ν − Fρσ F ρσ ηµν . (6.38)
4

Exercise
• Compute the T 0i in terms of the electric and magnetic fields. Do you recognize the
result?
• Prove that the electromagnetic energy-momentum is symmetric Tµν = Tνµ and
traceless, T µ µ = 0. The electromagnetic field behaves as a fluid with equation of
state p = 1/3ρ.

9 Note |pn |2
pin vn
i in Eq. (6.30) can be written as pin vn
i = i vi =
P P P
that the quantity i i i mγn vn n En
, which goes to
|pn | when En ' |pn |.
6.3 Einstein equations: Heuristic derivation 90

6.3 Einstein equations: Heuristic derivation

We have finally all the tools needed to derive the Einstein field equations for the gravitational field. In
the Poisson equation, the gravitational field is determined by the matter distribution. The relativistic
version of the matter distribution10 , the energy-momentum tensor Tµν M
, must be somehow equated11
to some tensor Kµν depending of the metric gµν and its first and second derivatives12
Kµν = κ2 Tµν
M
, (6.39)
with κ2 a proportionality constant to be determined. But, what tensor? Einstein got the answer to
this question trough a complicated process of intuition, trial and error; superhuman exertions in his
own words. As claimed above, the left-hand side of Eq.(6.39) should contain a second order differential
operator acting on the metric. We already found some quantities with this property in the previous
chapter: the Riemann tensor Rµ νρσ and its contractions. The most natural tentative for Kµν would
be the Ricci tensor Rµν , since this is the contraction appearing in the Newtonian limit of the geodesic
deviation equation (Ri 0i0 = E i i ). This was also one of the first trial and error choices of Einstein
Rµν ≈ κ2 Tµν
M
. (6.40)
Note however that this choice is inconsistent, since the divergence ∇ν Rµν of the Ricci tensor is, in gen-
eral, different from zero and, according to our minimal coupling prescription, the energy-momentum
should be locally conserved, ∇ν TµνM
= 0. Indeed, making use of the Bianchi identity (5.71) we can
write ∇ Rµν = 1/2∇ν R, which together with the trace of Eq. (6.40), R = κ2 g µν Tµν
µ M
= κ2 T M , im-
plies the condition ∇µ T = 0. Since the covariant derivative of the scalar quantity T M is just the
M

partial derivative, we should necessarily have a constant T M throughout the whole spacetime, which
is highly implausible, since, as we know, T M = 0 for the electromagnetic field and T M > 0 for stan-
dard matter. On top of that, Eq. (6.40) hides 10 differential equations for 6 physical unknowns: the
components of the metric that cannot be freely changed by performing coordinates transformations
in the 4 coordinates. We have to try harder.
The most general combination of symmetric tensors involving up to two derivatives of the metric is
Kµν = Rµν + agµν R + Λgµν (6.41)
with a and Λ some unknown constants to be determined13 . Imposing the local conservation of the
energy-momentum tensor ∇µ Tµν
M
=0 in Eq. (6.39) we get
∇µ Kµν = ∇µ (Rµν + agµν R) = 0 , (6.42)
where we have taken into account that the covariant derivative ∇µ is metric compatible and therefore
∇µ (Λg µν ) = 0. Our situation now is much better than that of Einstein, we are aware of the contracted
form of the Bianchi identities14 (5.71) and know the precise value of a that satisfies Eq. (6.42), namely
a = 1/2. Taking this into account, we can rewrite Eq. (6.39) as
Gµν + Λgµν = κ2 Tµν
M
, (6.43)
with
1
Gµν ≡ Rµν − gµν R , (6.44)
2
10 Matter should be understood in a broad sense, meaning really matter, radiation etc. . .
11 A relativistic generalization should take the form of an equation between tensors.
12 The requirement of having derivatives only up to second order is certainly reasonable. If this were not the case, one

would have to specify for the Cauchy problem not only the value of the metric and its first derivative, but also higher
derivatives on a spacelike surface.
13 A possible proportionality constant in front of R
µν has been factored out and incorporated in the still unknown
factors κ and Lambda in the right hand side of Eq. (6.39).
14 He wasn’t.
6.3 Einstein equations: Heuristic derivation 91

the Einstein tensor defined in previous chapter (cf. Eq. (5.72)) and Λ the famous cosmological constant
term. Writing this cosmological constant term in the right hand side of the equation, we can interpret
it as the energy-momentum tensor of a fluid with a weird equation of state p = −ρ
Λ
Gµν = κ2 Tµν
M Λ Λ


+ Tµν , Tµν =− gµν . (6.45)
κ2
M Λ
Defining Tµν ≡ Tµν + Tµν , we can write

Gµν = κ2 Tµν . (6.46)

Even though our derivation was quite heuristic, the solution that we have obtained is unique (Lovelock
theorem). The resulting tensorial equation is a set of ten differential equations15 for the metric gµν (x)
given the energy-momentum tensor Tµν (x). However, due to the existence of the Bianchi identities,
not all the components are longer independent. There are only 6 independent equations to determine
6 independent components of the metric tensor.
As differential equations they are very complicated, even in vacuum. Both the Ricci scalar and
the scalar curvature involve derivatives and products of Christoffel symbols, which in turn involve
derivatives of the metric tensor. There is also some dependence on the metric hidden in the energy-
momentum tensor. On top of that, the equations are not linear, as it should be expected, since,
according to the Equivalence Principle, every form of energy, including the gravitational self-energy,
must be a source of the gravitational field16 . The non-linearity of the equation forbids us to apply the
superposition principle, given two known solutions they cannot be combined to get a new one.

The Einstein equation in words

The physical meaning of Einstein equations can be clarified by considering an observer with
velocity uµ . The energy density as measured in the energy frame of such an observer is given by
ρ = Tµν uµ uν . Taking this into account, together the interpretation of the Einstein tensor that
we developed in the previous Chapter, the physical content of (6.43) can be summarized as

Gµν − κ2 Tµν uµ uν = 0 ,


(6.47)

which in words reads

 
Scalar curvature of the spatial  
 = 2κ2 Energy density measured by
sections measured by an .
an observer with 4-velocity uµ

observer with velocity uµ

15 Both sides of the equation are symmetric rank-2 tensors.

16 Note however that they have a well-posed initial-value structure, i.e. they determine the future values of gµν from
given initial data. This consideration is of key importance for the study of systems evolving in time from some initial
state, as for instance, gravitational waves.
6.4 The linearized theory of gravity 92

Newton Einstein

Newton 2nd law Geodesic equation

d 2 xi ∂Φ d 2 xµ ρ
dxν
dt2 = −δ ij ∂xj d2 σ = −Γµ νρ ∂x
dσ dσ

Tidal deviation Geodesic deviation

d2 ξ i i j D2 ξµ
dt2 = −E j ξ dσ 2 = −Rµ νρσ uν uσ ξ ρ

1st Bianchi identity 1st Bianchi identitiy

Eij = Eji Rµνρσ + Rµρσν + Rµσνρ = 0

2nd Bianchi identity 2nd Bianchi identitiy

E i [j,l] = 0 ∇κ R νρσ + ∇σ Rµ νκρ + ∇ρ Rµ νσκ = 0
µ

mass density Energy-momentum tensor

ρ Tµν

Poisson equation Einstein equation

E i i = 4πGρ Gµν = 8πGTµν

single elliptic equation 10 coupled equations

4 elliptic and 6 hyperbolic

boundary data required initial and boundary data required

Table 6.1: Newtonian vs Einstenian description of gravity.

6.4 The linearized theory of gravity

Equation (6.46) looks very promising but we have still to prove that it is able to reproduce the
Newtonian theory of gravity and determine the value of the unknown constants κ and Λ. The fastest
way to obtain the Newtonian limit is to use the assumptions discussed in Section 3.6. Let me however
relax these assumptions and obtain the general expression for the Einstein equation in the so-called
weak field limit. This limit is defined by the condition

gµν = ηµν + hµν , with |hµν |  1 . (6.48)

The quantity hµν is then understood as a small perturbation on top of the Minkowski background.
Consistently with this point of view, we will raise and lower its indices with the flat Minkowski metric
ηµν , namely hµ σ = η µρ hρσ , hµν = η νσ hµ σ .
In order to compute the expression for the Einstein tensor Gµν at the lowest order in perturbation
6.4 The linearized theory of gravity 93

theory we must first determine the linearized version of the Ricci tensor and the scalar curvature,
which are functions of the metric connection Γµ νρ . Inserting the expansion (6.48) into the definition
of the metric connection, we get
1 µσ
Γµ νρ = η (∂ν hσρ + ∂ρ hσν − ∂σ hρν ) + O(h2µν ) . (6.49)
2
The next step is to compute the 4 pieces of Riemman tensor, which, written in a very schematic way,
···
have the structure R··· ∼ ∂Γ − ∂Γ + ΓΓ + ΓΓ. Taking into account (6.49), we realize that only the
first two terms (∼ ∂Γ) give a contribution to the leading order
1 1
Rµ νρσ = ∂ρ (∂σ hµ ν + ∂ν hµ σ − ∂ µ hνσ ) − (ρ ↔ σ) = (∂ν ∂ρ hµ σ + ∂σ ∂ µ hνρ − (ρ ↔ σ)) . (6.50)
2 2
The linearized version of the Ricci tensor and the scalar of curvature can be computed by simply
performing contractions in the previous expression. Denoting respectively by h ≡ hµ µ and 2 = ∂ µ ∂µ
the trace of the perturbation tensor and the d’Alambertian operator and contracting the indices µ
and σ in Eq. (7.17), we get17
1
Rνρ = − (2hνρ + ∂ν ∂ρ h − ∂ν ∂σ hσ ρ − ∂ρ ∂σ hσ ν ) , (6.51)
2
which can be further contracted in the indices ν and ρ to obtain

R = Rν ν = η νρ Rνρ = −2h + ∂ν ∂σ hνσ . (6.52)

Collecting all the terms and inserting them into the definition of the Einstein tensor (6.44), we get
1
Gνρ = − (∂ν ∂ρ h + 2hνρ − ∂ν ∂σ hσ ρ − ∂ρ ∂σ hσ ν − ηνρ 2h + ηνρ ∂µ ∂σ hµσ )
2
1 
= − 2h̃νρ + ηνρ ∂µ ∂σ h̃µσ − ∂ν ∂σ h̃σ ρ − ∂ρ ∂σ h̃σ ν , (6.53)
2
where in the last step we have defined the so-called trace reverse
1 1
h̃µν ≡ hµν − ηµν h , hµν = h̃µν − ηµν h̃ , (6.54)
2 2
which keeps track of the extra terms obtained when passing from Rνρ to Gνρ . The name trace reverse
comes from the property h̃ ≡ h̃µ µ = −h. Note also the useful properties

˜
h̃µν = hµν , Gµν = R̃µν . (6.55)

The linearized Einstein equations becomes finally

 
2h̃νρ + ηνρ ∂µ ∂σ h̃µσ − ∂ν ∂σ h̃σ ρ − ∂ρ ∂σ h̃σ ν = −2κ2 Tνρ . (6.56)

The resulting expression is rather involved, but fortunately we still have some freedom to play with:
the gauge freedom.
17 The global minus sign comes from the permutation of the last two indices two construct the Ricci scalar.
6.4 The linearized theory of gravity 94

Gauge fixing
Eqs. (7.17) and (6.53), and therefore (6.56), are invariant under the transformation

hνρ −→ hνρ − ∂ν ξρ − ∂ρ ξν , (6.57)

as can be easily verified be performing the explicit computation. This kind of change is
called a gauge transformation, due to the strong analogy with the gauge transformations in
the electromagnetic theory. The simplest way to understand this gauge freedom is to trace
it back to the transformation of the full metric gµν . Consider an infinitesimal transformation
xµ → x̄µ = xµ + ξ µ . Under such a transformation the metric changes to
∂ x̄µ ∂ x̄ν
ḡ µν (xρ + ξ ρ ) = g ρσ (xρ ) (6.58)
∂xρ ∂xσ
= g (δ ρ + ∂ρ ξ µ ) (δ ν σ + ∂σ ξ ν )
ρσ µ

= g µν (xρ ) + g µσ ∂σ ξ ν + g νρ ∂ρ ξ µ .

Expanding the left-hand side of this equation in a Taylor series in ξ ρ and retaining only the
terms up to linear order, we get

ḡ µν (xρ ) = g µν (xρ ) + δg µν , (6.59)

with
δg µν ≡ −ξ ρ ∂ρ g µν + g µρ ∂ρ ξ ν + g νρ ∂ρ ξ µ = ∇ν ξ µ + ∇µ ξ ν . (6.60)
In the particular case in which the perturbation is performed around the Minkowski background,
gµν = hµν + ηµν , the covariant derivatives in (6.59) become standard derivatives and we recover
the transformation law (6.57). The linearized theory is invariant under (6.57) because the
full nonlinear theory is invariant under general coordinate transformations! This is extremely
interesting, since it allows us to further simplify the linearized version of the Einstein tensor by
simply performing infinitesimal coordinates transformations, or in other words, changes from
a splitting gµν = ηµν + hold to a different splitting gµν = ηµν + hnew . A simple inspection of
Eq. (6.56) reveals that an interesting condition to be satisfied by the trace reverse tensor in
the new coordinate system would be the tensor analog of the Lorenz gauge ∂µ Aµ = 0 in the
electromagnetic theorya , namely
∂ρ h̃νρ
new = 0 . (6.61)
Let us see if we are allowed to choose such a gauge. The change in the trace reverse tensor h̃µν
follows directly from Eqs. (6.54) and (6.57)
νρ
h̃νρ ν ρ ρ ν νρ µ
new = h̃old − ∂ ξ − ∂ ξ + η ∂µ ξ . (6.62)

Taking the derivative of this equation we get

νρ
∂ρ h̃νρ ν
new = ∂ρ h̃old − 2ξ . (6.63)

In order to satisfy the gauge fixing (6.61), ξ ν must be a solution of the inhomogeneous wave
equation
2ξ ν = ∂ρ h̃νρ
old . (6.64)
The existence of a solution transforming from an arbitrary hµν to the so-called Lorenz gauge
νρ
∂ρ h̃νρ b
new = 0 is guaranteed for sufficiently well behaved ∂ρ h̃old . In fact, the choice is not unique
ν
since we can always add to it any solution of the homogeneous wave equation 2ξH = 0 and the
ν ν νρ νρ
result will still obey 2 (ξ + ξH ) = ∂ρ h̃old . The Lorenz gauge ∂ρ h̃new = 0 is actually a set of
gauges.
a It“kills” three of the four terms in (6.53).
b Asyou learnt in your electrodynamic course, the solution of this equation can be obtained by means of the
retarded Green functions of the d’Alambertian operator.
6.4 The linearized theory of gravity 95

O
x
P

x'
x- x'

In view of the previous discussion, we realize that most of the terms in the left-hand side of Eq. (6.56)
merely serve to maintain gauge invariance. When the Hilbert gauge condition 18 ∂ρ h̃νρ = 0 is imposed,
the linearized version of the Einstein equation simplifies dramatically
2h̃µν = −2κ2 Tµν . (6.65)
This equation is formally identical to the Maxwell equations in the Lorenz gauge and can be solved
by using the Green’s function method.

Green’s functions
Consider a differential wave equation of the form

2f (t, x) = s(t, x) , (6.66)

with f (t, x) a radiation field and s(t, x) a source term. A Green’s function G(t, x; t0 , x0 ) is
defined as the field generated at the point (t, x) by a delta function source at (t0 , x0 ). i.e.

2G(t, x; t0 , x0 ) = δ(t − t0 )δ(x − x0 ) . (6.67)

The field due the actual source s(t, x) can be obtained by integrating the Green’s function
against s(t, x): Z
f (t, x) = dt0 d3 x0 G(t, x; t0 , x0 ) s(t0 , x0 ) . (6.68)

Physically the Green’s function approach merely reflects the fact that (6.66) is a linear equation.
The full solution of the equation can be obtained by solving for a point source and adding the
resulting waves from each point inside the source.

The Green’s function associated with the wave operator 2 is very well known (see for instance the
Jackson’s book on electrodynamics.):
δ(t0 − [t − |x − x0 |])
G(t, x; t0 , x0 ) = − . (6.69)
4π|x − x0 |

Exercise
Derive this equation in case you haven’t done it before.

Using (6.69) into (6.65), we get19

κ2 Tµν (t − |x − x0 |, x0 ) 3 0
Z
h̃µν = d x , (6.70)
2π |x − x0 |
18 This gauge is also called Einstein gauge, harmonic gauge, de Donder gauge, Fock gauge or, in analogy with electro-

magnetism, Lorenz gauge.

19 Note that we can always add to this particular solution an arbitrary solution of the homogeneous wave equation

(vacuum). As in electromagnetism, the metric perturbation consists of the field generated by the source plus wave-like
vacuum solutions propagating at the speed of light.
6.4 The linearized theory of gravity 96

which is analogous to the relation between the vector potential Aµ and the current Jµ in electromag-
netism. Note the argument t − |x − x0 | = t − |x − x0 |/c. Eq. (6.70) is a retarded solution 20 , taking
into account the lag associated with the propagation of information from events at x to position x0 .
Gravitational influences propagate at the finite speed of light. Action at a distance is gone forever!
We will be back to this point at the next chapter , but before let me finish our main task: determining
the value of the constants κ2 and Λ. For doing that let me consider the case we know better: the grav-
itational field created by a static spherical mass distribution of total mass M . The energy-momentum
tensor for such a system has only one non-vanishing component (cf. Eq. (6.45))
 
Λ
T 00 = ρ + 2 diag (1, 0, 0, 0) . (6.72)
κ
Plugging this into the time independent version of Eq. (6.70), we get
κ2 ρ (x0 ) 3 0
Z Z
1 Λ
h̃00 = 0
d x + d3 x0 , h̃0i = 0 , h̃ij = 0 . (6.73)
2π |x − x | 2π |x − x0 |
If the mass distribution is concentrated around the origin (x0 = 0), the component h00 evaluated at a
distance r = |x − x0 | becomes21
κ2 ρ (x0 ) 3 0 κ2 M
Z Z
1 Λ 3 0 2
h̃00 = d x + d x = + Λr2 (6.74)
2π r 2π r 2π r 3
with Z
M= ρ (x0 ) d3 x0 (6.75)

the total mass of our spherical distribution. Taking now into account that h̃ = η µν h̃µν = −h̃00 and
using the definition (6.54) we get
κ2 M 1
h00 = h11 = h22 = h33 = + Λr2 . (6.76)
4πr 3
Comparing this result with that obtained by performing the weak field limit of the geodesic equation in
the Λ = 0 case, hΛ=0
00 = −2Φ = 2GM/r, allows us to identify the sought-for proportionality constant
κ2 = 8πG . (6.77)
When Λ 6= 0, the Newtonian potential becomes modified at long distances
GM Λ
Φ=− − r2 (6.78)
r 6
and line element takes the form
   
2 2GM 1 2 2 2GM 1 2
ds = − 1 − − Λr dt + 1 + + Λr dX 2 , (6.79)
r 3 r 3
with dX 2 ≡ dx2 + dy 2 + dz 2 . In Newtonian terms, a positive cosmological constant (Λ > 0) gives rise
to a repulsive force per unit mass whose strength increases linearly with the distance
GM Λ
f =− ur + r ur , (6.80)
r2 3
20 The retarded solution is obtained by imposing the Kirchoff-Sommerfeld “no-incoming radiation” boundary condition

at past null infinity

lim (∂r + ∂t ) (rh̄µν ) = 0 , (6.71)
t→∞
with the limit taken along any surface with ct + r =constant, together with the condition that rh̃µν and r∂ρ h̃µν are
bounded in this limit.
21 Note that the integral in over the prime variables!
6.4 The linearized theory of gravity 97

Cosmological constant
If Λ 6= 0, it must be at least very small, ρΛ  ρmatter , to avoid any observational effect in
those situations in which the Newton’s theory of gravity successfully explains the observations.
Taking into account, for instance, that we do not see any modification of the Newtonian theory
of gravity within the solar system, we can set the limit

|Λ| 3M
|ρΛ | = ≤ ρSolar −→ |ρΛ | ≤ 3 ' 10−29 GeV4 (6.81)
8πG 4πRPluto

which, as assumed, makes the contribution of Λ completely negligible on the scale of the systems
we will be interested in in this coursea .
a It will play however a fundamental role at larger scales, as those you will considered in your Cosmology

course.

Linearized Gravity Electromagnetism

Field equation Einstein equation with Maxwell equations

gµν = hµν + hµν

Basic potentials Linearized metric 4-vector potential

hµν (x) Aµ = (Φ, A)

Sources Energy-momentum tensor 4-vector current

T µν J µ = (ρ, J)

Lorenz gauge ∂µ h̃µν = 0 ∂µ Aµ = 0

h̃µν = hµν − 21 ηµν h

Sourced wave equation 2h̃µν = −16πGTµν 2Aµ = Jµ

R [Tµν ]ret 3 0 1
R [Jµ ]ret 3 0
Solution h̃µν = 4G |x−x0 | d x õ = 4π |x−x0 | d x

Table 6.2: Linearized Einstein equations vs Maxwell equations.