Physik-Institut der Universität Zürich

in conjunction with ETH Zürich

Applications of General Relativity

in Astrophysics and Cosmology
Spring semester 2017

Prof. Philippe Jetzer

Original version by Felix Hähl

Revision: Tobias Baldauf and Raymond Angélil

Last revision of this version: June 25, 2019

 CONTENTS

Contents
1 Time Delay of Radar Echoes 1

2 Geodetic Precession 4
2.1 Geodetic Precession . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Linearized Field Equations 10

3.1 The Energy Momentum Tensor of the Grav. Field . . . . . . . . . . . . . . . . . . . 10
3.2 Linearized Field Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4 Lense-Thirring Effect 14
4.1 Metric of the rotating Earth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.2 Gravitomagnetic Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Gravitational Waves 19
5.1 Electromagnetic Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.2 The Case of Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.3 Particles in the Field of a Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.4 Energy and Momentum of a Gravitational Wave . . . . . . . . . . . . . . . . . . . . 24
5.5 Quadrupole Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

6 Sources of Gravitational Waves 33

6.1 Rigid Rotator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.2 The PSR 1913+16 System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.3 Waveform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.4 Quadrupole radiation from a mass in circular orbit . . . . . . . . . . . . . . . . . . 41

7 The post-Newtonian Approximation 44

7.1 The 1PN Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.2 Motion of Test Particles in the 1PN Metric . . . . . . . . . . . . . . . . . . . . . . . 48

8 The Kerr Solution 55

8.1 Interpretation of the Parameters a and m . . . . . . . . . . . . . . . . . . . . . . . . 55
8.2 Kerr-Newman Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
8.3 Equations of Motion for Test Particles . . . . . . . . . . . . . . . . . . . . . . . . . 57

ii
 CONTENTS

Preface
This course is a continuation of the General Relativity course taught by Prof. Philippe Jetzer in fall
2017. The script covering this course can be found on the webpage www.physik.uzh.ch/en/teaching/
PHY519/FS2017.html. When referring to equations in the General Relativity script, we use the prefix
GR I , as for example in Eq. (GR I 1.1). These lecture notes are far from being a complete treatment
of the subject but should enable the student or interested reader to access specialized literature on the
subject. For further reading we included a list of useful textbooks at the end. If you find any mistakes,
please report them to boetzel@physik.uzh.ch.

Zürich, February 2017

iv
 CONTENTS

1 Time Delay of Radar Echoes

In 1964 Shapiro proposed a new test of GR consisting of a measurement of the time delay of radar
signals transmitted from Earth through a region near the Sun to another planet or spacecraft and
reflected back to Earth. Since the radar signal is affected by the gravitational field of the Sun, it will
return to Earth with a time delay.
The radar signal is emitted from Earth and then sent back by a reflector as sketched in Fig. 1.
π
In calculations we set θ = 2. The trajectory of the signal is curved. This is not shown in the figure
because the curvature is, of course, only a tiny effect and the light ray seems almost straight. Let’s
compute the time that the signal needs to get from Earth to the reflector. To this end we use Eq. (GR I
23.15) for the orbit in a static isotropic gravitational field with ε = 0 (since m = 0):
2
l2 F2

dr
A + − = 0. (1.1)
dλ r2 B

If we divide this by l2 and use x0 = ct, we find

dr 1 dr dx0 (GR I 23.12) 1 dr F

= = . (1.2)
dλ c dt dλ c dt B
This yields
2
AF 2 F2

dr 1
+ − 2 = 0. (1.3)
c B 2 l2
2 dt r 2 Bl

For the minimal distance r0 (from the Sun) it holds that

dr F2 B(r0 )
=0 ⇒ = . (1.4)
dt l2 r02

We insert this into Eq. (1.3) to obtain

2
r02 B(r)

A dr
2
+ − 1 = 0. (1.5)
c B dt r2 B(r0 )

Sun
ϕE ϕR
rE r0 rR

Earth Reflector
Figure 1: Sketch of the system as described in the text. The angles ϕE and ϕR and the distances rE , rR
describe the position of the Earth and the reflector relative to the Sun. The whole setup is chosen
π
to lie in the θ = 2
-plane.

1
 CONTENTS

This differential equation is solved by the following integral:

r  −1/2
1 r 0 A r02 B(r0 )
Z
t(r, r0 ) = dr 1 − 02 , (1.6)
c r0 B r B(r0 )

where t(r, r0 ) is the time that the radar signal needs to travel from r0 to r. Note that this is the
time which would be shown by a clock at rest at infinity (as space is asymptotically Minkowskian at
infinity). This actually forces us to introduce a correction since our clock rests at Earth, not at infinity.
However, the correction which is needed to compensate this effect is much smaller than the time delay
and can thus be neglected.
Using the Robertson expansion from Eq. (GR I 22.3),
2a 2a
A(r) = 1 + γ + ..., B(r) = 1 − + ..., (1.7)
r r
we get

r02 B(r) r02

  
1 1
1− = 1 − 1 + 2a −
r2 B(r0 ) r2 r0 r

r2
  
2ar0
= 1 − 02 1 − . (1.8)
r r(r + r0 )

Inserting Eqs. (1.7), (1.8) into (1.6) and expanding, we get

r −1/2 
r02
Z  
1 0 ar0 a
t(r, r0 ) ' dr 1 − 02 1+ 0 0 + (1 + γ) 0
c r0 r r (r + r0 ) r
p p !
r2 − r02 r2 − r02
r
a r − r0 a r+
= + + (1 + γ) log . (1.9)
c c r + r0 c r0
√
r 2 −r02
The first term c corresponds to the travelling time assuming a straight trajectory in Euclidean
space as can easily be seen from the figure and Pythagoras’ theorem. The other terms account for the
general relativistic time delay due to the gravitational field of the Sun. For the system drawn in the
figure, the total delay is
" p p #
2 − r2
rE 2 − r2
rR
0 0
δt = 2 t(rE , r0 ) + t(rR , r0 ) − − , (1.10)
c c

where the factor of 2 accounts for the fact that the signal travels from Earth to the reflector and back
again.
Significant delays occur if the radar signal passes nearby the Sun, i.e. if r0 is of the order of some
Sun radii. In this case we have rE , rR  R and also rE , rR  r0 . We thus can perform the following
approximations in Eq. (1.10):1
   
4a 1+γ 4rE rR
δt ' 1+ log . (1.11)
c 2 r02
1 I. I. Shapiro, Phys. Rev. Lett. 13, 789 (1964)

2
 CONTENTS

We see that δt is maximal if the signal just grazes the surface of the Sun, i.e. r0 = R .

In order to see the orders of magnitude, we use rE ∼ rR ∼ 108 km, R ' 7 · 105 km. This yields
2GM
2a
c = c3 ' 10−5 s and thus
   
4a 1+γ 4rE rR
δtmax = 1+ log 2
c 2 R

' 2 · 10−4 s. (1.12)

Performing the measurements has been a very difficult task since the distances rE and rR were not
known with sufficient precision. Nevertheless, in the seventies these measurements were performed
using Venus and Mercury and later again using spacecrafts as reflectors (e.g. the Vikings which landed
on Mars or, more recently, the Cassini spacecraft2 ). The results are

Vikings: γ = 1.000 ± 0.001,

Cassini: γ = 1 + (2.1 ± 2.3) · 10−5 . (1.13)

Another experimental verification of the Shapiro delay is the measurement of the PSR J1614-2230
system. It consists of a pulsar which emitts signals in very regular time intervals, and a white dwarf
that orbits the pulsar. When the white dwarf is in front of the pulsar and the light of the pulsar
arrives at Earth by passing close to the white dwarf, then the signal arrives with a delay. Measuring
the Shapiro delay3 , one can infer the mass of the white dwarf to be 0.500 ± 0.006M . With an orbital
period of 8.7 days, this yields a neutron star mass of 1.97 ± 0.04M . This result is important for the
modelling of neutron stars since the largest neutron stars (so far) had masses of about 1.4M .

2 B. Bertotti et al., Nature 425, 374 (2003)

3 P. Demorest et al., Nature 467, 1081 (2010)

3
 CONTENTS

2 Geodetic Precession
Consider a particle with a “classical” angular momentum (for instance the intrinsic angular momentum
of a rigid body like a gyroscope). In the local inertial system in which the body is at rest, the spin
(i.e. angular momentum) is given by S = S i ei . To the three-vector S i we assign a Lorentz vector S α .
Consider now a locally inertial coordinate frame IS’ which is momentarily at rest with respect to the
rigid body (or particle):

S 0α = (0, S 0i ). (2.1)

We can transform this to some arbitrary inertial system IS by means of a Lorentz transformation. In
the rest frame IS’, the velocity of the body is

u0α = (c, 0).

Therefore we have in IS’

u0α S 0α = 0.

Since this quantity is a Lorentz scalar, uα S α = 0 in any arbitrary IS.

Consider first the case without any forces acting on the particle and no torque acting on its spin.
In an arbitrary IS we have
dS α
= 0. (2.2)
dτ
We define the Riemann vector
∂xµ α
Sµ ≡ S , (2.3)
∂ξ α
which describes the transition from the coordinate system IS with coordinates (ξ α ) to a general system
(xµ ). According to the covariance principle, the generalization of Eq. (2.2) reads
DS µ dS µ
=0 ⇔ = −Γµ νλ uν S λ . (2.4)
dτ dτ
This equation describes the spin precession in a gravitational field. The condition uα S α = 0 reads
uµ S µ = 0 in the general coordinate system. Note that Sµ S µ =const., which implies that (2.4) describes
the rotation or precession of the spin vector. Since we assumed no external forces, Eq. (2.4) contains
only gravitational effects. We conclude that Eq. (2.4) describes the precession of the spin of a particle
which is freely falling in a gravitational field as, for example, the precession of a rotating satellite (or
gyroscope).

If there are other external forces f µ besides gravity, then one finds instead of Eq. (2.4)
DS ν 1 Duµ
=− 2 Sµ uν (2.5)
dτ c dτ
which is also called Fermi transport. It describes the spin precession of an accelerated particle on
which a gravitational field acts (c.f. Eq. (GR I 19.2), Duµ /dτ = f µ /m). (The special case f µ = 0 is

4
 CONTENTS

just parallel transport.)

Based on the above considerations, we shall study the following effects (in the gravitational field of
the Earth):

1. Geodetic precession: the precession of a freely falling gyroscope. In order to simplify the
analysis we will assume the gravitational field to be isotropic and static.

2. Lense-Thirring effect: the precession of a gyroscope in the gravitational field of the Earth
which is due to the rotation of the Earth. This is a smaller effect.

2.1 Geodetic Precession

Gyroscopes are rigid bodies which can also perform rotations described by S µ besides the movement
of its center of mass. To compute the geodetic precession of a gyroscope we use Eq. (2.4). In the local
rest frame of the satellite (orbiting the Earth) we have for the spin vector:

S 0α = (0, l) , (2.6)

where l describes the angular momentum of the gyroscope. We use the standard form of the static
and isotropic metric in spherical coordinates (cf. chapter 22.1 of GR I ):

xµ = (ct, r, θ, ϕ)

g = diag B(r), −A(r), −r2 , −r2 sin2 θ .

Assume that the satellite is on a circular orbit, i.e.

π
r = const., θ= , ϕ = ω0 τ. (2.7)
2
Therefore the velocity of the satellite reads
dxµ
uµ = = u0 = const., 0, 0, u3 = ω0 = const. .


(2.8)
dτ
π
If we insert θ = 2 into the Christoffel symbols from Eq. (GR I 22.6), we obtain for the non-zero
components

B0 A0 r r
Γ1 00 = , Γ1 11 = , Γ1 22 = − , Γ1 33 = −
2A 2A A A
B0 1 1
Γ0 01 = Γ0 10 = , Γ2 12 = Γ2 21 = , Γ3 13 = Γ3 31 = . (2.9)
2B r r

5
 CONTENTS

Eq. (2.4) reads then

dS 0
= −Γ0 01 u0 S 1 (2.10)
dτ
dS 1
= −Γ1 00 u0 S 0 − Γ1 33 u3 S 3 (2.11)
dτ
dS 2
=0 (2.12)
dτ
dS 3
= −Γ3 31 u3 S 1 . (2.13)
dτ
We can immediately solve the third of these equations:

S 2 (τ ) = const. (2.14)

this is the component of the spin (or angular momentum) of the gyroscope which is perpendicular to
the satellite’s orbit (θ-direction) is constant.
Because of r =const., all coefficients of the system of linear differential equations (2.10)-(2.13) are
constants. We differentiate Eq. (2.11) with respect to τ and insert (2.10) and (2.13) on the right-hand
side. This yields
d2 S 1
= Γ1 00 Γ0 01 (u0 )2 + Γ1 33 Γ3 31 (u3 )2 S 1
 
dτ 2

≡ −ω 2 S 1 . (2.15)

With (u3 )2 = ω02 and inserting the Christoffel symbols from (2.9), we get
"  2 #
2 2 B 02 u0 1
ω = ω0 − + . (2.16)
4AB u3 A

u0
In order to understand the ratio u3 better, we look at the equation for the trajectory of the satellite
(geodesic equation):
duµ
= −Γµ νλ uν uλ . (2.17)
dτ
The µ = 1 component of this equation is (u1 = 0)

du1
0= = −Γ1 00 (u0 )2 − Γ1 33 (u3 )2 (2.18)
dτ
from which we infer
2
u0

2r
= . (2.19)
u3 B0
Inserting this into (2.16), we find
s
rB 0
 
1
ω = ω0 1− . (2.20)
A 2B

6
 CONTENTS

~ey

~eϕ
~er

ωτ
~ex

Figure 2: Projection of the motion orbit of precession of the angular momentum vector onto the orbital plane
of the satellite. The precession takes place in the (r, ϕ)-plane since the θ-component of the spin
vector (i.e. the component perpendicular to the orbital plane) is constant.

Using the Schwarzschild solution (B = A−1 = 1 − 2a

r ), this yields
r
3a
ω = ω0 1− , (2.21)
r

or in terms of the Robertson expansion

r
a
ω = ω0 1 − (1 + 2γ) . (2.22)
r
Since r =const., Eq. (2.15) is of type

S̈ 1 + ω 2 S 1 = 0 (2.23)

which describes a harmonic oscillator. With initial conditions S 1 (0) = S and Ṡ 1 (0) = 0 the solution
reads

S 1 (τ ) = S cos(ωτ ), S 2 = const.. (2.24)

Inserting this into Eq. (2.13), we obtain by integration

Sω0
S 3 (τ ) = − sin(ωτ ). (2.25)
rω
We proceed by studying the time dependence of the projection (S 1 , S 3 ) (or (r, ϕ)-components) of
π
the spin vector onto the orbital plane (θ = 2 ), c.f. Fig. 2. Consider the constant vector ex along the
orbit of the satellite:

ex = cos(ω0 τ )er − sin(ω0 τ )eϕ . (2.26)

|{z} |{z}
=ϕ =ϕ

7
 CONTENTS

2π
The orbital period of the satellite is τ0 = ω0 . After each orbit, the argument ϕ in (2.26) increases by
τ0 ω0 = 2π whereas the argument in (2.24) or (2.25) increases by τ0 ω which differs slightly from 2π.
The phase difference after one orbit is given by (note that in GR γ = 1)

∆α = τ0 (ω0 − ω)
r
(1 + 2γ)a
= 2π − 2π 1−
r

∼ (1 + 2γ)a
=π . (2.27)
r
Consider the vector S which is the projection of S µ onto the orbital plane:

S = Sr er + Sϕ eϕ . (2.28)

The components of this vector are given by

(Sr )2 = −g11 S 1 S 1 and (Sϕ )2 = −g33 S 3 S 3 (2.29)

with

− g33 = r2 sin2 θθ= π = r2 .

−g11 = A(r) and (2.30)
2

Therefore Sr ∝ cos(ωτ ) and Sϕ ∝ sin(ωτ ). For τ = 0, S is thus parallel to ex . However, after an orbit,
ωτ differs slightly from 2π as we have calculated in Eq. (2.27). The geodetic precession after one orbit
is given by

3πa (1 + 2γ)
∆α = . (2.31)
r 3

Consider the concrete example of a satellite in a circular orbit around the Earth. We have
 1/2
GME RE
ω02 RE = 2 =g and τ0 = 2π . (2.32)
RE g
t
After one year (i.e. after τ0 orbits with t = 1 year), we find (assuming r ' RE )

t 3πgRE t
∆α(t) = ∆α = ' 800 .4 yr−1 . (2.33)
c2 2π RE
q
τ0
g

5/2
For a general radius r, one finds ∆α(t) ' 800 .4 (RE /r) yr−1 .
On April 20, 2004 the satellite Gravity Probe B has been launched to measure the geodetic preces-
sion. At an altitude of 642 km, general relativity predicts a geodetic precession of −6606.1 mas yr−1 .
The measured result was −6601.8 ± 18.3 mas yr−1 (1 mas = 1 milliarcsecond). This result therefore
matches the predictions reasonably well.4
4 C. Everitt et al., Classical and Quantum Gravity 25, (2008) 114002, C. Everitt et al., Phys. Rev. Lett. 106, (2011)
221101.

8
 CONTENTS

2.1.1 De Sitter Precession of the Moon

The Earth-Moon system can be considered as a “gyroscope” with an axis “perpendicular” to the orbital
plane in an orbit around the Sun. We denote by l the angular momentum of the Earth-Moon system
with respect to the common center of mass (l is essentially the orbital angular momentum of the Moon
because the common center of mass almost coincides with the center of the Earth).
The angular momentum l can be decomposed in a component l⊥ perpendicular to the Earth’s
orbital plane around the Sun and a parallel component l|| . The parallel component does not vanish
because the orbital plane of the Moon around the Earth is tilted by 5◦ with respect to the orbital
plane of the Earth around the Sun. In terms of the previously defined quantities, l⊥ corresponds to
S 2 and stays constant. The component l|| corresponds to S and lies in the orbital plane of the Earth
around the Sun. It is this parallel component which is affected by geodetic precession. Therefore the
orbital plane of the Moon rotates slightly (this was first noticed by De Sitter in 1916).
We can calculate this precession per century
3πa
∆αDe Sitter = 100 ' 200 per century (2.34)
rEarth-Sun

where a = 1.5 km and rEarth-Sun = 1 AU ≈ 150 × 106 km. Additionally we have a Newtonian
precession (as in any three body system) with a period of 18.6 yr. This Newtonian effect is 107 times
larger than the De Sitter precession. Nevertheless the De Sitter precession has been measured. A
laser beam has been sent to the the Moon where it was reflected back to Earth by mirrors previously
brought to the Moon by the Apollo mission (1969 and following years). The measurements that have
been performed from 1970 till 1986 confirmed the De Sitter precession with a precision of about 1%.5

5 Shapiro et al., Phys. Rev. Lett. 61, 2643 (1988) and Müller et al., Astrophys. J. 382, L101 (1991)

9
 CONTENTS

3 Linearized Field Equations

In order to find solutions to the Einstein field equations in the weak field regime, one can linearize the
equations. The results of this section will, for example, be used to calculate the Lense-Thirring effect
and to describe gravitational waves.
Since the field equations are non-linear, there is no standard procedure to solve the equations
given a source for the fields Tµν (for example, in electrodynamics we didn’t have to struggle with such
difficulties). Besides numerical methods, there are essentially three possibilities as for which type of
solutions one can find:

• Exact solutions assuming simplifying conditions (as, for example, staticity, isotropy, ...). An
example for this case is the Schwarzschild solution.

• Solutions of the linearized field equations for weak gravitational fields.

• Systematic expansion of the field equations and the equations of motion for weak fields and
small velocities. This method is also called post-Newtonian approximation. For example,
v2 φ
in planetary systems we have c2 ∼ c2 which is very small. Such results should reproduce the
Newtonian limit in lowest order (linearized in φ).

We shall now elaborate the second possibility.

3.1 The Energy Momentum Tensor of the Gravitational Field

The field itself is a form of energy and thus also a source of the field. This effect is purely due to
the non-linearities, of course. Considering a weak field, we can work with small deviations from the
Minkowski metric:

gµν = ηµν + hµν with |hµν |  1. (3.1)

One proceeds as follows. First, Gµν has to be expanded in powers of hµν . The first order terms
will lead to a linear wave equation. Neglecting terms of third order, the second order terms give the
energy-momentum tensor of the gravitational field.
The expansion of the Ricci tensor can be written as

(1) (2)
Rµν = Rµν + Rµν + ... (3.2)
(0)
with Rµν = 0. In order to compute the first order term of (3.2), we write down the expansion of the
6
curvature tensor:
1
Rρµσν = (gρσ,µ,ν + gµν,ρ,σ − gµσ,ν,ρ − gρν,σ,µ ) + O(h2 ) (3.3)
2
6 Notice
that we adopt here a different notation with respect to GR I: we adopt the “minus” convention of the EFE
(i.e. Rµν − R g = − 8πG
2 µν c4
Tµν ), which simply follows from a change of sign of the Riemann tensor. The Riemann tensor
from which (3.3) is derived differs thus from (GR I, 17.4) by a factor −1.

10
 CONTENTS

where the derivatives are non-covariant (the additional terms due to covariant derivatives are of higher
order). We can thus write the first order Ricci tensor in terms of hµν :

(1) 1
Rµν = (hµν + hρ ρ,µ,ν − hρ µ,ρ,ν − hρ ν,ρ,µ ) . (3.4)
2
The d’Alembert operator can be used instead of ∂µ ∂ µ because in the approximation (3.1) the coordi-
nates are “almost” Minkowskian, so ∂µ ∂ µ =  + O(h). The first order Ricci scalar is given by
(1)
R(1) = η λρ Rλρ . (3.5)

We proceed by considering the second order equations. The left-hand side of the field equations
can be written in terms of the quantity tµν which is defined by
 (2)
(2) Rgµν 8πG
Rµν − =: tµν . (3.6)
2 c4

We take these terms to the right-hand side of Einstein’s equations and find at second order in hµν :

(1) R(1) 8πG

Rµν − ηµν = − 4 (Tµν + tµν ). (3.7)
2 c
This can be interpreted as a wave equation linear in hµν with source terms

τµν = Tµν + tµν . (3.8)

We have to think of τµν as being the energy-momentum tensor which also includes the contribution of
the gravitational field itself.
We interpret (3.7) as follows: since Gµν ;ν = 0 (Bianchi identity), we find for the left-hand side of
(3.7):

R(1)
 
∂ (1)
Rµν − ηµν = 0. (3.9)
∂xν 2

Therefore the right-hand side satisfies

∂τµν
= 0. (3.10)
∂xν
This gives the momentum
Z
Pµ = d3 r τµ0 = const. (3.11)

which is conserved (in time). We can thus interpret τµ0 as the momentum density and τµν as an energy-
momentum tensor (indeed we know that T µν ;ν = 0 but so far we did not necessarily conclude τ µν ;ν =
0). Since Tµν includes all non-gravitational sources and τµν is interpreted as the “complete” energy-
momentum tensor, tµν clearly describes energy-momentum which is purely due to the gravitational
field:
(2) !
c4

Rgµν
tgrav.
µν = (2)
Rµν − (|hµν |  1). (3.12)
8πG 2

11
 CONTENTS

3.2 Linearized Field Equations

With Eq. (3.3) we find for the field equations at first order in h
 
16πG T
hµν + hρ ρ,µ,ν − hρ µ,ρ,ν − hρ ν,ρ,µ = − Tµν − ηµν . (3.13)
c4 2

We use ηµν instead of gµν in this equation because both sides are already of order h. Since the field
equations are covariant, we are free to perform a coordinate transformation. But note that since
|hµν |  1 we can only perform coordinate transformations which deviate only slightly from Minkowski
coordinates:

xµ −→ x0µ = xµ + εµ (x) with ε  1. (3.14)

∂x0µ ∂x0ν λρ ∂x0µ µ

From g 0µν = ∂xλ ∂xρ
g we infer how hµν transforms. With ∂xλ
= δλµ + ∂x
∂ε
λ inserted into g
0µν
we get

g 0µν = η µν − h0µν

∂εµ ∂εν
  
δλµ + ν
η λρ − hλρ


= λ
δ ρ + ρ
(3.15)
∂x ∂x

where we used that from gµν = ηµν + hµν it follows g µν = η µν − hµν . From Eq. (3.15) we infer

∂εµ ∂εν
h0µν = hµν − − . (3.16)
∂xν ∂xµ

Since this is already a first order equation (in h), we can raise and lower indices with gµν ' ηµν and
g µν ' η µν . Thus
∂εµ ∂εν
h0µν = hµν − − . (3.17)
∂xν ∂xµ
In analogy to electrodynamics this transformation of the “potentials” gµν is called a gauge transfor-
mation. We can choose four functions εµ (x) which give four constraints on the “potentials” hµν . For
instance,

2hµ ν,µ = hµ µ,ν . (3.18)

We insert the gauge condition (3.18) into (3.13) and obtain the decoupled linearized field equa-
tions:
 
16πG T
hµν = − Tµν − ηµν . (3.19)
c4 2

This can easily be seen if we differentiate (3.18) (i.e. hρ ρ,µ = 2hρ µ,ρ ) with respect to xν :

hρ ρ,µ,ν = 2hρ µ,ρ,ν = hρ µ,ρ,ν + hρ ν,ρ,µ (3.20)

(we used hµν = hνµ ). This implies

−hρ µ,ρ,ν − hρ ν,ρ,µ + hρ ρ,µ,ν = 0 , (3.21)

12
 CONTENTS

which is just another form of our gauge condition from which it can be seen that (3.13) indeed reduces
to (3.19).
Furthermore, it can be shown that from (3.17) it follows that if hµν does not satisfy (3.18), then
we can find a transformed h0µν that does so. This can be done using the coordinate transformation
(3.14) with εν = hµ ν,µ − 12 hµ µ,ν .

The linearized field equation Eq. (3.19) has the same structure as the field equations in electrody-
namics. We can therefore immediately write down the well-known solution for the retarded potentials:
0
 
4G
Z Sµν r 0 , t − |r−r
c
|

hµν (r, t) = − 4 d3 r0 (3.22)

c |r − r 0 |

T
with Sµν = Tµν − ηµν .
2
The interpretation is the same as in electrodynamics: a change in Sµν at podition r 0 does not affect
|r−r 0 |
the position r before some time c has passed.

13
 CONTENTS

4 Lense-Thirring Effect
The Lense-Thirring effect is the precession of a gyroscope in the gravitational field of the Earth due
to the Earth’s rotation. To set up an analogy with electrodynamics, we note that the gravitational
field of the Schwarzschild metric corresponds to the Coulomb field outside of a static, spherical charge
distribution. If a charge distribution rotates with constant angular velocity, this results in the presence
of a static, non-isotropic magnetic field. Similarly, the rotation of the Earth will cause a gravitomagnetic
field.
We will treat this problem by using the linearized field equations. Another approach would be to
start from the exact Kerr solution (i.e. the metric outside of a rotating black hole) and apply the weak
field limit.

4.1 Metric of the rotating Earth

We assume a weak field caused by a slowly rotating planet, so |hµν |  1. The linearized field equations
read
 
16πG T
hµν = − Tµν − ηµν . (4.1)
c4 2
The coordinates xµ = (x0 , ..., x3 ) are Minkowski coordinates up to corrections of O(h). In the energy-
momentum tensor (19.9) we can neglect the pressure since p  ρc2 . Since the velocities (rotation of

2
the Earth) are small compared to c, we neglect terms of order vc . The energy-momentum tensor
thus reads
!
vi
1
Tµν ' ρc2 vi
c
. (4.2)
c 0
The terms proportional to vi generate the gravitomagnetic field. This has to be compared to electro-
dynamics where magnetic fields are generated by currents.
The mass distribution of the Earth can be approximated as follows:

ρ (r ≤ R )
0 E
ρ(r) = (4.3)
0 (r > R )
E

The angular velocity of the Earth is

2π
ω = ωe3 with ω = . (4.4)
1 day
We consider the Earth as a rigid body, so we can write the velocity field as

v(r) = ω ∧ r or vi = εijk ω j rk . (4.5)

Because this velocity is constant, Tµν does not depend on time and therefore the field equations (4.1)
have stationary solutions. We can thus replace  by −∆ and Eq. (4.1) becomes
8πG
∆hµµ = ρ(r), (4.6)
c2
16πG
∆h0i = ρ(r)εijk ω j rk (4.7)
c3

14
 CONTENTS

T ρc2
where we used ηµν = diag(1, −1, −1, −1), 2 = 2 and T00 = ρc2 , Tii = 0.
Using
1
∆ = −4πδ (3) (r − r 0 ), (4.8)
|r − r 0 |

we can immediately solve the equations (4.6) and (4.7):

ρ(r 0 )
Z
2G
hµµ (r) = − 2 d3 r0 , (4.9)
c |r − r 0 |

ρ(r 0 )x0n
Z
4G
h0i (r) = − 3 εikn ω k d3 r0 (4.10)
c |r − r 0 |

where x0n denotes the n-th component of r 0 .

For the region outside of the mass distribution (r > RE ), we can use the expansion

1 X 4π r0l ∗ 0
= Y (r̂ )Ylm (r̂) (r > r0 )
|r − r 0 | (2l + 1) rl+1 lm
l,m

1 xj x0j
= − 3 + ... (4.11)
r r
where Ylm are the spherical harmonic functions and r̂, r̂ 0 denote unit vectors in directions of r and
r 0 , respectively. Note that for the Cartesian components we have xi = gik xk = −xi + O(h). The
corrections of O(h) can thus be neglected because the right-hand sides of (4.9) and (4.10) are already
of first order in h. Since ρ(r) is spherically symmetric, only the first term of (4.11) contributes in
Eq. (4.9):
Z
2G 2GME
hµµ (r) = − d3 r0 ρ(r 0 ) = − (r ≥ RE ) . (4.12)
c2 r c2 r

Since ρx0n is proportional to Y1m , only terms with l = 1 contribute to (4.10):

ω j xk
Z
4G
h0i (r) = εijn d3 r0 ρ(r 0 )x0n x0k
c3 r3
2
4GME RE ω j xn
=− ε ijn (r ≥ RE ). (4.13)
5c3 r3
02
Since ρ(r) is spherically symmetric, we integrated using x0n x0k = −δkn r3 and furthermore used ρ0 =
3
3ME /(4πRE ). Considering h0i as a vector, i.e. h0i → h = h0i ei , we can write (4.7) as

16πG
∆h(r) = ρω ∧ r (4.14)
c3
and (4.13) becomes

2
4GME RE ω∧r 2GI ω ∧ r
h(r) = − 3
=− 3 (4.15)
5c r3 c r3

where I = 25 ME RE
2
is the moment of inertia of a homogeneous sphere.

15
 CONTENTS

We want to consider the analogy with electrodynamics again. In magnetostatics, the vector poten-
tial A of a homogeneously charged rotating sphere with radius R and total charge q satisfies
4π
∆A = − ρe ω ∧ r (4.16)
c
qR2 ω ∧ r
⇒ A= (4.17)
5c r3
which has the same form as Eq. (4.15).
Eqs. (4.12) and (4.13) determine the metric of the rotating Earth (valid for r ≥ RE ):
   
2 2GME 2 2 2GME
ds = 1 − c dt − 1 + dr 2 + 2ch0i dxi dt (4.18)
c2 r c2 r
GME


where dr 2 = −dxi dxi . Note that this metric at O c2 r and for ω = 0 does not reduce to the
Schwarzschild metric since Eq. (4.1) implies that we chose other coordinates as compared to the
standard form. However, the metric (4.18) asymptotically (r → ∞) becomes the Minkowski metric.
One can perform a coordinate transformation such that dr 2 has the usual angular dependence, i.e.
dr 2 → dr2 + r2 (dθ2 + sin2 θdϕ2 ). Distant “fixed” stars (which live in the asymptotic Minkowski
spacetime) have constant values for the angles (θ, ϕ). Therefore changes in the angles due to spin
precession refer to rotations with respect to the distant fixed stars.

4.1.1 Rotation of the Local IS

The metric (4.18) implies a rotation of the local inertial system IS. In order to see this, consider the
axis of a gyroscope which is described by
dS µ
= −Γµ κν S κ uν . (4.19)
dτ
A freely falling gyroscope would not only show a precession linked to the Lense-Thirring effect (that
can be loosely defined as “precession due to the rotation of the central body”), but would also be
subjected, because of its motion in a curved metric, to the geodesic precession derived in Section 2.
However, we want here to artificially separate these two different contributions, and only look at the
pure Lense-Thirring component. This can be done with a simple trick: we consider the gyroscope as
being at rest in the external frame, and just let it free to change its rotational axis. Thus we have, in
the coordinate system defined by Eq. (4.18),

S µ = (0, S i ), uµ = (c, 0). (4.20)

dS i dS i
Because we are taking into account only terms up to O(h), we can replace dτ by dt . From
Eqs. (4.19) and (4.20) it follows that
dS i
= −cΓi 0j S j . (4.21)
dt
For time independent hµν we find at first order

η ik ∂h0k
 
i ∂h0j 1
∂j h0 i − ∂ i h0j .


Γ 0j = j
− k
= (4.22)
2 ∂x ∂x 2

16
 CONTENTS

Inserting this into (4.21) and lowering the index i, we find

dSi c
= − (∂j h0i − ∂i h0j ) S j
dt 2
c
= − εikl εkmn ∂m h0n S l


2

= εikl Ωk S l (4.23)

with εikl εkmn = δlm δin − δln δim and the gravitomagnetic field
c c
Ωk = − εkmn ∂m h0n or Ω(r) = − ∇ ∧ h(r) . (4.24)
2 2
The indices i, j, m run over 1, 2, 3. We call h the gravitomagnetic potential. Using h from Eq. (4.15)
we obtain the following expression for the angular velocity in the local IS:
2
2GME RE 3(ω · r)r − ωr2
Ω(r) = 2
. (4.25)
5c r5

Note that Ω has the same form as B = ∇ ∧ A in electrodynamics. In vector form, Eq. (4.23) reads

dS
=Ω∧S or dS = (Ωdt) ∧ S . (4.26)
dt
This implies a precession of the spin of the gyroscope’s axis with angular velocity Ω. This precession
is due to the rotation of the Earth with angular velocity ω. The precession of the gyroscope’s axis is
equivalent to the rotation of the local IS because in the local IS we have S =const.. Therefore the
local IS rotates with angular velocity Ω as compared to the global coordinate system described by
(4.18). For r → ∞ Eq. (4.18) becomes Minkowskian, i.e. it describes a system which doesn’t rotate
with respect to the fixed star system.
To summarize, the physical meaning of the angular velocity Ω is that the local IS rotates with Ω
with respect to the fixed star system. The rotation of the Earth drags the local IS (“frame dragging”).

Now that we derived geodetic precession and the Lense-Thirring effect, we note that in fact both
effects take place at the same time and thus sum up. Inserting r = RE in Eq. (4.25) we get at the
north pole and at the equator, respectively:

2GME 2 (North pole, θ = 0)
Ω= 2 ω· (4.27)
5c RE (−1) (equator, θ = π
2)

where we used |ω · r| = ωr cos θ. This evaluates to

2GME
ω ∼ 10−9 ω. (4.28)
5c2 RE
We conclude that the Lense-Thirring precession affects a Foucault pendulum located at the north
pole. The rotation with respect to the distant stars amounts to
4GME
∆φ = ΩLT · 1yr = 2π · 365 = 0.200 per year. (4.29)
5c2 RE

17
 CONTENTS

This effect was first computed by Lense and Thirring in 1918.7

If one chooses the spin perpendicular to the orbital plane, the Lense-Thirring effect vanishes for
dS
equatorial orbits since in this case we have S||w||Ω and Eq. (4.26) gives dt = 0. The expected total
00
frame dragging is at most of the order of 0.05 per year. The NASA satellite Gravity Probe B, with
the aim to measure both geodetic precession and Lense-Thirring precession, was launched in 2004 in a
polar orbit around Earth. In May 2011 the final results of Gravity Probe B have been released and they
indeed confirm the GR predictions for the Lens-Thirring effect. The measured value being: −37.2 ± 7.2
mas/yr (1 mas = 1 milliarcsecond) to be compared with the GR prediction of −39.2 mas/yr8 . However
there are claims that the effect has been measured on the orbit of the LAGEOS satellites.9

4.2 Gravitomagnetic Forces

In the discussion of the Lense-Thirring precession, we saw a formal equivalence to electromagnetism:

h ↔ A

Ω ↔ B.

This analogy persists even for the equation of motion of a particle in the metric (4.18),
duµ
= −Γµ γν uγ uν . (4.30)
dτ


Neglecting terms O v 2 /c2 we have dτ ≈ dt and uµ = (c, v i ). Therefore Eq. (4.30) reads

dv i
 2
v
= −Γi 00 c2 − 2cΓi 0j v j + O . (4.31)
dt c2
The first term on the right-hand side corresponds to the gradient of the Newtonian potential. In
analogy to to (4.21)-(4.23), the second term can be shown to give

−Γi 0j v j = (Ω ∧ v)i . (4.32)

This implies that Eq. (4.31) can be written as

dv
= −gradφ + 2Ω ∧ v. (4.33)
dt
This is the equation of motion in the presence of gravitomagnetic forces and it has the same structure
as the equation for the Lorentz force
 
1
K =q E+ v∧B . (4.34)
c
This analogy is the origin of the notion of gravitomagnetism.
Note, however, that this analogy is only true if we consider the linearized field equations of general
relativity. In the above identification the electromagnetic fields are full solutions to Maxwell’s equations
whereas the gravitomagnetic potential and field are approximations. Furthermore, the analogy is quite
formal and certainly not complete due to the absence of negative “gravitational charges”.
7 Lense and Thirring, Phys. Zeitschr. 19, 156 (1918)
8 C. W. F. Everitt et al., Phys. Rev. Lett. 106, 221101 (2011)
9 I. Ciufolini and Pavlis, Nature 431, (2004) 958

18
 CONTENTS

5 Gravitational Waves
On 14 September 2015 the two LIGO detectors simultaneously observed a transient gravitational wave
(GW) signal, which has been interpreted as due to the merger of two black holes with masses of about
36 M and 29 M , respectively. This being the first direct detection of GW, which was announced last
11 February 2016 10 . In the data of the first Advanced LIGO run at least one more GW event has been
11
found as has been announced on 15 June 2016 . Moreover, the satellite LISA Pathfinder, with the
aim to test the technology needed to build LISA, a GW detector in space, was successfully launched
on 3 December 2015. On 7 June 2016 the first results which showed that the performance of LISA
Pathfinder was much better than expected and almost already at the level of the LISA requirements
12
were released . Thus the year 2016, 100 years after Einstein’s first paper on GW as a consequence
of his theory of General Relativity, has thus seen dramatic advancements in the field of GW.

For weak gravitational fields (i.e. |hµν | = |gµν − ηµν |  1) the Einstein field equations read
 
16πG T
hµν = − 4 Tµν − ηµν . (5.1)
c 2

In the vacuum (Tµν = 0) the equation reduces to

hµν = 0 (5.2)

which has plane waves as its simplest solution. The above equation is quite similar to the wave
equation in electromagnetism, Aµ = 0 with the electromagnetic vector potential Aµ . As we will see,
the solutions are similar, as well. Note that the wave equation in electromagnetism is exact whereas
the general relativistic wave equation arises from the approximate linearized field equations.

5.1 Electromagnetic Waves

Physical fields are invariant under gauge transformations

Aµ → A0µ = Aµ + ∂ µ χ (5.3)

so that we can choose ∂µ Aµ = 0 (Lorenz gauge) and get

4π µ
Aµ = j . (5.4)
c
Due to the gauge conditions, only three out of four components of Aµ are independent. While
leaving the Lorenz gauge unaltered, we still have the freedom to perform an additional gauge trans-
formation satisfying χ = 0. Since in vacuum j µ = 0, this allows us to set A0 = 0. Finally we are left
10 LIGO Scientific and Virgo Collaborations (B.P. Abbott (Caltech) et al.), Observation of Gravitational Waves from
a Binary Black Hole Merger, Phys.Rev.Lett. 116, (2016) 061102
11 LIGO Scientific and Virgo Collaborations (B.P. Abbott (LIGO Lab., Caltech) et al.), GW151226: Observation of

Gravitational Waves from a 22-Solar-Mass Binary Black Hole Coalescence, Phys.Rev.Lett. 116, (2016) 241103
12 The LISA Pathfinder collaboration, M. Armano et al. Sub-Femto- g Free Fall for Space-Based Gravitational Wave

Observatories: LISA Pathfinder Results, Phys.Rev.Lett. 116, (2016) 231101

19
 CONTENTS

with two degrees of freedom (polarizations). The conditions read then

Aµ = 0, A0 = 0, ∂i Ai = 0. (5.5)

This is solved by the ansatz

Aµ = eµ exp[−ikν xν ] + c.c. , (5.6)

where kµ k µ = 0 and ei k i = 0 (polarizations are transverse to propagation direction).

5.2 The Case of Gravity

Due to the symmetry hµν = hνµ , 10 out of 16 components of hµν are independent. With a gauge
transformation of the form (3.18) we can impose four additional conditions. This leaves us with 6
degrees of freedom that are truly independent. If we consider the vacuum case

hµν = 0, (5.7)

in addition to (3.17) we can perform a further transformation of the form

hµν → h0µν = hµν − ∂µ εν − ∂ν εµ (5.8)

provided that εµ satisfies

εµ = 0. (5.9)

Such a transformation leaves Eq. (5.7) and the gauge condition (3.18) invariant (this is in complete
analogy to electromagnetism, of course). With these four additional conditions we are left with two
independent components of hµν . The solution to (5.7) can be written in terms of plane waves

hµν = eµν exp[−ikκ xκ ] + c.c. (5.10)

where
ω2
η λν kλ kν = k ν kν = 0 ⇔ k02 = = k2 = k 2 . (5.11)
c2
The amplitude of the wave eµν is called polarization tensor. Inserting (5.10) into the gauge condition
(3.18) (2hµ ν,µ = hµ µ,ν ) leads to

2kµ η µρ eρν = kν η µρ eρµ . (5.12)

Clearly eµν inherits the symmetry of hµν , thus eµν = eνµ . Let us choose a wave travelling along the
x3 -axis. This yields the wave solution

hµν = eµν exp ik(x3 − ct)

 
(5.13)

where we used Eq. (5.11). The components of the wave vector are then
ω
k1 = k2 = 0, k0 = −k3 = k = . (5.14)
c

20
 CONTENTS

In this case the gauge condition (5.12) reads

1
e00 + e30 = (e00 − e11 − e22 − e33 ), (5.15)
2

e01 + e31 = 0, (5.16)

e02 + e32 = 0, (5.17)

1
e03 + e33 = − (e00 − e11 − e22 − e33 ). (5.18)
2
With eµν = eνµ and these four conditions, the polarization tensor is fully determined by six compo-
nents. All the other components can be expressed in terms of the six independent components

e00 , e11 , e33 , e12 , e13 and e23 . (5.19)

The other components are given by

1
e01 = −e31 = −e13 , e02 = −e32 , e22 = −e11 , e03 = − (e00 + e33 ). (5.20)
2
We can perform yet another transformation (3.14) (x0µ = xµ +εµ ) with functions εµ satisfying εµ = 0.
The functions are solutions of the wave equation, therefore we can write them as

εµ (x) = δ µ exp [−ikµ xµ ] + c.c.. (5.21)

As noted before, such a transformation with arbitrary δ µ does not violate the gauge condition (3.18).
We choose k µ in (5.21) equal to the wave vector of a given gravitational wave. Using (5.21) in (3.16) we
obtain a new solution h0µν in which all the terms have the same exponential dependence of exp[−ikµ xµ ].
Thus only the amplitudes transform as

e011 = e11 , (5.22)

e012 = e12 , (5.23)

e013 = e13 − iδ1 k, (5.24)

e023 = e23 − iδ2 k, (5.25)

e033 = e33 − 2iδ3 k, (5.26)

e000 = e00 + 2ikδ0 . (5.27)

We can choose δµ such that e000 = e013 = e033 = e023 = 0. This new solution is equivalent to the old one.
From the physical point of view, only polarizations corresponding to e011 and e012 are relevant.
Neglecting primes in our notation from now on, we get for the gravitational wave propagating in
3
x -direction, after gauging away all redundancies
 
0 0 0 0
 
0 e11 e12 0
 · exp ik(x3 − ct) + c.c.
 
hµν = 
 (5.28)
0 e12 −e11 0

0 0 0 0

21
 CONTENTS

5.2.1 Helicity

The direction of k is the x3 -axis. We ask now the question how (5.28) transforms under a rotation
around this axis. Since we are in an almost Minkowskian metric we can realize this transformation as
a Lorentz transformation described by the matrix
 
1 0 0 0
 
µ
0 cos ϕ sin ϕ 0
Λ̄ ν = 
 . (5.29)
0 − sin ϕ cos ϕ 0

0 0 0 1

Therefore the polarization tensor transforms as

e0µν = Λ̄ρ µ Λ̄σ ν eρσ . (5.30)

This yields

e011 = e11 cos(2ϕ) + e12 sin(2ϕ), (5.31)

e012 = −e11 sin(2ϕ) + e12 cos(2ϕ). (5.32)

If we consider e± ≡ e11 ± ie12 instead, we thus have

e0± = e±2iϕ e± . (5.33)

The vectors e± have helicity ±2, whereas the wave solutions in electrodynamics have helicity ±1.
Generalizing from the electromagnetic field, which is quantized using a spin 1 particle, the photon,
one can thus expect the quanta of the gravitational field to be spin 2 particles. While there is neither
evidence for their existence nor a closed theory of quantum gravity, the hypothetical quanta of the
gravitational field are commonly dubbed gravitons .

5.3 Particles in the Field of a Wave

Similarly to electromagnetic waves also gravitational waves exert forces on massive particles. We want
to explore how the positions of particles are affected in the field of a gravitational wave. For this
analyis consider a plane wave along the x3 -direction
 
0 0 0 0
 
3
0 e11 e12 0
 exp ik(x3 − ct) + c.c. .


hµν (x , t) = 
 (5.34)
0 e12 −e11 0

0 0 0 0

The corresponding metric has the form

ds2 = ηµν + hµν (x3 , t) dxµ dxν .

(5.35)

The trajectory xσ (τ ) of a particle in the gravitational field satisfies the equation of motion
d2 xσ σ dxµ dxν
= −Γ µν (5.36)
dτ 2 dτ dτ

22
 CONTENTS

where Γσ µν can be taken from Eq. (5.28). We assume that there are no other forces but gravity which
act on the particles. Inserting (5.34) into
 
σ 1 σλ ∂hνλ ∂hµλ ∂hµν
+ O h2


Γ µν = η + − (5.37)
2 ∂xµ ∂xν ∂xλ

it follows
 
i 1 ∂h0i ∂h0i ∂h00
Γ 00 =− + − = 0. (5.38)
2 ∂x0 ∂x0 ∂xi

As initial conditions we choose

dxi 

i
ẋ (0) = = 0. (5.39)
dτ τ =0

This implies

d2 xi 

(5.65)
= −Γiµν ẋµ (0)ẋν (0) = 0. (5.40)
dτ τ =0

Therefore the acceleration vanishes. This means that the velocities of the particles don’t change. The
solution of the equations of motion (5.36) thus reads

dxi
=0 ⇒ xi (τ ) = const. (5.41)
dτ
In the chosen coordinates the particles in the field of the gravitational wave can thus be described by
constant spatial coordinates. However, this does not mean that the particles are at rest. In fact their
distances vary due to the time dependence of the metric tensor gµν as in Eq. (5.35).
Consider now particles which are arranged on a circle in the x1 -x2 -plane. The particles are initially
at rest on a circle ((x1 )2 + (x2 )2 = R2 ). We want to examine the effect of an incident gravitational
wave along the x3 -axis. To do so, we write (5.35) in the form

ds2 = c2 dt2 − dl2 − (dx3 )2

with dl2 = (δmn − hmn (t)) dxm dxn (m, n = 1, 2). (5.42)

In the x1 -x2 -plane we have

!
3 e11 e12
hmn (t) = hmn (x = 0, t) = exp(−iωt) + c.c. (5.43)
e12 −e11

where ω 2 = c2 k 2 . With (5.42) we can compute the physical distance ρ of a particle p from the center
of the circle. According to Eq. (5.41) the coordinates x1p and x2p of the particle are constant. We insert
in (5.42) the finite values of the coordinates xm
p of p instead of dx
m
(this is allowed because the metric
coefficients do not depend on x1 and x2 ):

ρ2 = (δmn − hmn (t)) xm n

p xp (m, n = 1, 2) (5.44)

with x1p = R cos ϕ, x2p = R sin ϕ. (5.45)

23
 CONTENTS

Using Eqs. (5.43)-(5.45) we find the solution


1 − 2h cos(2ϕ) cos(ωt) if e = h, e = 0,
11 12
ρ2 = R 2 (5.46)
1 − 2h sin(2ϕ) cos(ωt) if e = 0, e = h.
11 12

The cos(ωt) term comes from e−iωt + c.c., whereas the cos(2ϕ) term, for example, comes from cos2 ϕ −
sin2 ϕ = cos(2ϕ). The distinction that we made in (5.46) concerns the two possible linear polarization
states.
Unlike to the coordinates x1 , x2 which are constant, the physical variables x = ρ cos ϕ and
y = ρ sin ϕ describe the distance relative to the center. The physical oscillations lead to a particle
configuration which is an ellipse with very tiny eccentricities (h  1). From the type of oscillation
one can infer the polarization of the incoming wave. The two independent polarization states form an
π
angle of 4. Therefore the oscillations correspond to a quadrupole moment of the mass distribution:
the gravitational waves induce a quadrupole oscillation of the mass distribution. Conversely we ex-
pect that mass distributions with oscillating quadrupole moment should emit gravitational waves. In
order to study this phenomenon further, we have to learn more about the energy and momentum of a
gravitational wave.

5.4 Energy and Momentum of a Gravitational Wave

We now want to determine the energy-momentum tensor of a gravitational wave, which is a solution
of the free field equation up tofirst order in h (|hµν |  1)

(1)
Rµν = 0. (5.47)

The solution to this equation is the wave solution that we derived before:

hµν = eµν exp −ikλ xλ + c.c.

(5.48)

The energy-momentum tensor of a gravitational field is known from Eq. (3.12):

!
(2)
grav. c4 (2) (gµν R)
tµν = Rµν − (5.49)
8πG 2

(0) (1)
with gµν = ηµν + hµν . Using Rµν = Rµν = 0 and R = g ρσ Rρσ we get

c4 h (2) i
tgrav.
µν = 2Rµν − ηµν η ρσ Rσρ
(2)
+ ηµν hρσ Rσρ
(1)
− hµν η ρσ Rσρ
(1)
16πG
c4 h (2) i
= 2Rµν − ηµν η ρσ Rσρ
(2)
(5.50)
16πG
where we used g µν = η µν − hµν which follows from g µν gνλ = δλµ . In order to consider this expression
(2)
further, we need the Ricci tensor Rµν . In order to be able to calculate the Ricci tensor

(2)

(2) (2) (1)
Rµκ = g λν Rλµνκ = η λν Rλµνκ − hλν Rλµνκ (5.51)

24
 CONTENTS

we need the Riemann tensor which is given by:

 2
∂ 2 gµκ ∂ 2 gµν ∂ 2 gλκ

1 ∂ gλν
Rλµνκ = + − − +
2 ∂xµ ∂xκ ∂xλ ∂xν ∂xλ ∂xκ ∂xµ ∂xν

+ gησ (Γη νλ Γσ µκ − Γη κλ Γσ µν ) . (5.52)

The first line of this expression (which contains only first order terms in h) gives rise to the second
term in (5.51) with gij = hij . The second line (which is of second order in h) gives rise to the first
term in (5.51) if we insert the Christoffel symbols

1 ∂hσ µ ∂hσ ν
 
∂hµν
Γσ (1)
µν = + − . (5.53)
2 ∂xν ∂xµ ∂xσ

This yields the first term in (5.51) if we multiply with ηµν . In total, we get the following expression
for the Ricci tensor that we searched for in order to evaluate Eq. (5.50):

hλν ∂ 2 hλν ∂ 2 hµκ ∂ 2 hµν ∂ 2 hλκ

 
(2)
Rµκ =− + − −
2 ∂xµ ∂xκ ∂xλ ∂xν ∂xλ ∂xκ ∂xµ ∂xν

1 ∂hν σ ∂hν σ ∂hν ν

 σ
∂hσ κ
 
∂h µ ∂hµκ
+ + − + −
4 ∂xν ∂xν ∂xσ ∂xκ ∂xµ ∂xσ
 σ
∂hσλ ∂hλ µ
 
1 ∂hσκ ∂hσλ ∂hλκ ∂h µ
− + − + − . (5.54)
4 ∂xλ ∂xκ ∂xσ ∂xλ ∂xµ ∂xσ

The first term in the second line of this expression vanishes because of the gauge condition (3.18). The
remaining terms are quadratic in h and of the form

eµν exp(−ikλ xλ ) + c.c. eσκ exp(−ikλ xλ ) + c.c.

  

= eµν eσκ exp(2ixλ k λ ) + 2 + exp(−2ixλ k λ ) .

 
(5.55)

We can see that there appear on the one hand oscillating terms of the form exp(±2ikλ xλ ) and on the
other hand there are also terms which do not depend on the coordinates xµ at all. If we average over
time, the oscillating terms drop out (their average over time is zero) such that we are left with terms
of the form
 
eµν exp(−ikλ xλ ) + c.c. eσκ exp(−ikλ xλ ) + c.c. = 2< e∗µν eσκ
  

, (5.56)

where h·i denotes time-average and < is the real part. For plane wave solutions, derivatives correspond
to multiplication with k, so
∂hµν
= −ikλ hµν . (5.57)
∂xλ

25
 CONTENTS

We can now plug all these terms which are quadratic in h into Eq. (5.54), replacing all partial derivatives
by factors of k:

(2)
hRµκ i = < (eλν )∗ (kµ kκ eλν + kλ kν eµκ − kλ kκ eµν − kµ kν eλκ )

 ∗
1
λ
+ e ρ kλ − eλ λ kρ (kµ eρ κ + kκ eρ µ − k ρ eµκ ) (5.58)
2

1 ∗
(kλ eσκ + kκ eσλ − kσ eκλ ) · k λ eσ µ + kµ eσλ − k σ eλ µ


− .
2

Using the relation: 2∂µ hµ ν = ∂ν hµ µ (which follows from the gauge condition (3.18)) one can simplify
the above expressions. For instance:
1 λ ∗ µ 1 2
(eλµ )∗ kκ kλ eµν = (e λ ) k kκ eµν = kκ kν eλ λ  .

(5.59)
2 4
Imposing the null condition kµ k µ = 0 we obtain
 
(2) 1 λν ∗ 1 λ 2
hRµκ i = kµ kκ (e ) eλν − |e λ | . (5.60)
2 2

Thus the energy-momentum tensor (5.49) or (5.50) of the gravitational wave reads

c4
 
λκ ∗ 1 λ 2
tgrav.
µν = kµ kν (e ) eλκ − |e λ | (5.61)
16πG 2

where we used that

ηµν η ρσ hRρσ
(2)
i ∝ ηµν η ρσ kρ kσ = ηµν k σ kσ = 0. (5.62)

We can further simplify the energy-momentum tensor by specializing to the case of linearly polarized
waves with either e11 = −e22 = h, e12 = e21 = 0 or e11 = −e22 = 0, e12 = e21 = h:

c4
tgrav.
µν = kµ kν h2 . (5.63)
8πG

Energy in this formula, being proportional to frequency squared is exactly the type of relation that
we would intuitively expect. Furthermore it is clear that tµν ∝ kµ kν because t0i is the current of
momentum which should be proportional to ki . We see immediately that measuring such energies
will be extremely difficult because h2 is very small. A wave propagating in the x3 -direction has the
wavevector kµ = ωc , 0, 0, ωc . The energy current density

c5 0 3 2
Φgrav. = ct03
grav. = k k h (5.64)
8πG
for such a wave is then given by

energy c3 2 2
Φgrav. = = ω h . (5.65)
time · surface 8πG

26
 CONTENTS

Figure 3: Sketch of the waves that are emitted by a dipole in electromagnetism.

5.5 Quadrupole Radiation

Oscillating charge distributions emit electromagnetic waves. In analogy we expect oscillating mass dis-
tributions to emit gravitational waves. We quickly repeat the case of electromagnetic dipole radiation
before turning to the case of oscillating mass distributions.

5.5.1 Dipole Radiation in Electromagnetism

In electromagnetism one finds that an oscillating dipole moment

p(t) = p0 exp(−iωt) + c.c. (5.66)

emits electromagnetic waves whose power P per solid angle is given by

dP ω4
= |p|2 sin2 θ (5.67)
dΩ 8πc3
where θ is the angle between p and k where k is the direction of propagation. This is sketched in
Fig. 3. The total emitted power can be obtained by integrating in θ:

ω4 2
P = |p| . (5.68)
3c3

5.5.2 Gravitational Quadrupole Radiation

The computation of the emitted gravitational radiation is similar to electromagnetism but also more
involved since the source terms are rank 2 tensors. We will proceed with the following steps:

1. Calculate the asymptotic fields emitted by a source Tµν .

2. Reduce the result to spatial components.

27
 CONTENTS

3. Apply the long wavelength approximation.

The setup is sketched in Fig. 4. In contrast to the electromagnetic case there is no gravitational dipole
radiation. The density is given by
Z
ρ(r, t) = ρ(r) exp(−iωt) + c.c. ⇒ p= d3 r rρ(r) = M Rc.m. (5.69)

where M is the total mass and Rc.m. is the center of mass. If we choose the center of mass system
as the inertial system then p = 0. Consequently p = 0 in all inertial systems. This is not possible in
electromagnetism. We shall now assume an oscillatory mass distribution of the form

6= 0 if r ≤ r
0
Tµν (r, t) = Tµν (r) exp(−iωt) + c.c. (5.70)
= 0 otherwise.

This is only a single Fourier component. Thus a generalization is possible by integrating over ω.
According to (3.22) the retarded potentials are given by
exp(ik|r − r 0 |)
Z
4G
hµν (r, t) = − 4 exp(−iωt) d3 r0 Sµν (r 0 ) + c.c. (5.71)
c |r − r 0 |
where we used
|r − r 0 |
 
−iωtr = −iω t − = −iωt + ik|r − r 0 | (5.72)
c
to obtain the phase factors. Furthermore we have introduced
1
Sµν (r) = Tµν (r) − ηµν T (r). (5.73)
2
ω 2π
We now assume r0  r with k = c = For large distances we have |r|  r0 and thus
λ .

r · r0
  0 
r
|r − r 0 | = r − + ... = r 1 + O (5.74)
r r

~r

~r ′

r0

Figure 4: Sketch of the setup for gravitational wave emission. The source has spatial extent r0 . The observer
is at position r.

28
 CONTENTS

and
  0 
0 0 r
exp [ik|r − r |] = exp[ikr] exp[−ikr ] 1 + O (5.75)
r
where we defined k = k rr = ker . This way we obtain for (5.71)
Z
4G 1
hµν (r, t) = − 4 exp −ikλ xλ d3 r0 Sµν (r 0 ) exp [−ikr 0 ] +c.c.
 
(5.76)
c r
| {z }
=:Sµν (k)

where Sµν (k) is the spatial Fourier transform of Sµν (r). This yields

hµν (r, t) = eµν (r, ω) exp −ikλ xλ + c.c..

 
(5.77)

The amplitudes are defined as

4G 1 4G 1 h ηµν i
eµν (r, ω) = − 4
Sµν (k) = − 4 Tµν (k) − T (k) (5.78)
c r c r 2
1 r
and are proportional to r. They depend on r = er and ω via k = ker . The energy current passing
2
through a surface element r dΩ is given by
xi 2
dP = ctgrav.
0i df i = ctgrav.
0i r dΩ. (5.79)
r
Plugging in Eq. (5.61) for tgrav.
µν we obtain

c4 k0 ki xi 2
 
dP λν ∗ 1 λ 2
=c r (e ) eλν − |e λ | . (5.80)
dΩ 16πG r 2
1
In the derivation of (5.61) we assumed eµν = const., whereas here we have eµν ∝ r. The energy-
momengum tensor tgrav.
µν contains partial derivatives of the hµν which would lead to additional terms
1 1
∝ r. With eµν = const., the derivatives just lead to factors of kµ ∝ λ. In the far field and distant
1
observer approximation we have r  λ and can thus neglect the additional terms since r  λ1 . Using

ki xi k·r ω
= =k= (5.81)
r r c
in Eq. (5.80) we get
Gω 2
 
dP µν ∗ 1 2
= T (k) Tµν (k) − |T (k)| (5.82)
dΩ πc5 2
where Tµν (k) is the Fourier transform of the source distribution.
We proceed with the second step as outlined in the beginning: we want to reduce our results to
spatial components. The source distribution can be written as
Z
1
d3 k T µν (k) exp −ikλ xλ + c.c..
 
Tµν (r, t) = 3
(5.83)
(2π)
For weak fields the covariant derivative in the energy-momentum conservation simplifies to an ordinary
derivative and the continuity equation reads

kµ T µν (k) = 0. (5.84)

29
 CONTENTS

In particular we find for ν = 0 and ν = i

k0 T 00 = −kj T 0j and k0 T i0 = −kj T ij . (5.85)

ki
We can define a three dimensional unit vector k̂i = k and obtain for (5.85)

T 0i = T i0 = −k̂j T ij , (5.86)

T 00 = k̂i k̂j T ij . (5.87)

All non-spatial components in (5.82) can thus be eliminated and we calculate

T µν∗ Tµν = ηµρ ηνσ T µν∗ T ρσ

X X
= T 00∗ T 00 − 2 T 0i∗ T 0i + T ij∗ T ij
i i,j

= k̂i k̂j k̂l k̂m T ij∗ T lm − 2k̂j k̂m δil T ij∗ T lm + δil δjm T ij∗ T lm , (5.88)

T λ λ = ηλρ T ρλ
X
= T 00 − T ii
i

= k̂i k̂j T ij − δij T ij , (5.89)

|T λ λ |2 = k̂i k̂j k̂l k̂m T ij∗ T lm − δij k̂l k̂m T ij∗ T lm − δlm k̂i k̂j T ij∗ T lm . (5.90)

Inserting these expressions into (5.82) we get

dP Gω 2
= Λij,lm T ij (k)∗ T lm (k) (5.91)
dΩ πc5
where we introduced
1 1 1 1
Λij,lm (θ, ϕ) = δil δjm − δij δlm − 2δil k̂j k̂m + δij k̂l k̂m + δlm k̂i k̂j + k̂i k̂j k̂l k̂m . (5.92)
2 2 2 2
Having reduced the formula for the radiated power to spatial components, we now turn to the last
step that we outlined in the beginning: we apply the long wavelength approximation, i.e. we assume
λ  r0 which simplifies the energy-momentum tensor as follows:
Z
T (k) = d3 r T ij (r) exp(−ikr)
ij

Z
= d3 r T ij (r)(1 − ikr + ...)

ω 2 ij
Z
≈ d3 r T ij (r) =: − Q . (5.93)
2

30
 CONTENTS

The object Qij will turn out to be a quadrupole tensor. From covariant conservation of energy-
momentum, T µν ,ν = 0, we get

∂j T ij (r, t) = −∂0 T i0 (r, t) and ∂i T 0i (r, t) = −∂0 T 00 (r, t). (5.94)

Using Eq. (5.70) we obtain

ω 2 00
∂i ∂j T ij (r, t) = ∂02 T 00 (r, t) = − T (r, t) (5.95)
c2
ω 2 00
⇒ ∂i ∂j T ij (r) = − T (r). (5.96)
c2
Since we are in the non-relativistic regime (λ  r0 , v  c) we have T 00 ' ρc2 . Therefore Eq. (5.93)
yields

ω2
Z Z Z
d3 r T ij (r) = d3 r xi xj ∂k ∂l T lk (r) = − 2 d3 r xi xj T 00 (r)


2 (5.97)
c
where we integrated by parts twice in the first step and used Eq. (5.96) in the second step. According
to the definition in Eq. (5.93), we find
Z Z
ij 3 i j 1
Q = d r x x ρ(r) = 2 d3 r xi xj T 00 (r) (5.98)
c

which we can obviously interpret as the quadrupole tensor of the mass distribution13 . Because we
are in almost Minkowskian spacetime, we can compute Qij in three-dimensional Cartesian coordinate.
Inserting (5.93) into (5.91) we get

dP Gω 6
= Λij,lm Qij∗ Qlm . (5.99)
dΩ 4πc5

We observe that the corresponding formula in electrodynamics looks very similar but it depends on ω 4
rather than ω 6 . This is just a reflection of the fact that the electromagnetic radiation is dipole radiation
whereas gravitational dipole radiation does not exist (the dipole moment of any mass distribution
vanishes in the center of mass system).
Furthermore we note that Qij are constants (they do not depend on θ or ϕ). The complete angular
dependence is encoded in Λij,lm in which the vector k̂ appears. This is the unit vector which indicates
the direction from the mass distribution to the observer and therefore clearly depends on θ and ϕ:

(k̂ i ) = (k̂x , k̂y , k̂z ) = (sin θ cos ϕ, sin θ sin ϕ, cos θ). (5.100)

This simplifies calculations, of course, because the quadrupole moment can be calculated once and
forever and the specific angular dependence is only to be considered in the form of Λij,lm .

As an example we consider a quadrupole mass distribution in the principal axis system: the only
non-vanishing elements are the diagonal, Qij = diag(Q11 , Q22 , Q33 ). In this case the only non-vanishing
13 In the literature one also finds different definitions of this tensor. For example, one can define a traceless version

where xi xj is replaced by xi xj − 13 r2 δ ij .

31
 CONTENTS

terms in (5.99) are those with i = j and l = m. This means that there appear only even powers of k̂x ,
k̂y , k̂z in Eq. (5.92). The emitted power has an angular dependence of the form

dP
∼ a1 cos4 θ + a2 cos2 θ sin2 θ + a3 sin4 θ. (5.101)
dΩ
To get the total emitted power, we have to integrate Eq. (5.99) over dΩ:
Z
2π
dΩ Λij,lm = (11δil δjm − 4δij δlm + δim δjl ) (5.102)
15
where we used
Z
4π
dΩ k̂i k̂j = δij , (5.103)
3
Z
4π
dΩ k̂i k̂j k̂l k̂m = (δij δlm + δil δjm + δim δjl ) . (5.104)
15

Inserting this result into (5.99) we obtain for the total emitted power
 
3  3 2
2Gω 6  X ij 2 1  X ii  
Z
dP
P = dΩ = |Q | −  Q  . (5.105)
dΩ 5c5 i,j=1
3 i=1

Note that Qij can be defined traceless such that the second sum in the brackets vanishes. Furthermore
one can assume a more general time dependence than just e−iωt . If we had defined the quadrupole
moments in a traceless form,
Z  
ij 3 i j1 2 ij
Q (t) = d x xx − r δ ρ(t, x), (5.106)
3

then instead of Eq. (5.105) we would have found

G ...kl ...
P = hQ Qkl i (5.107)
5c5
...
where h·i denotes a time average, for instance over one orbital period.14 The third time derivatives Qij
...
in the above equation can be easily evaluated for a plane wave Qij ∝ exp [−iωt] and yield Qij ∝ ω 3 .

14 In G
...kl ...
the literature, one finds also the formula P = 45c 5 h Q Q kl i. The different prefactor arises if one defines the

quadrupole tensor with an additional factor of 3 in the second term of Eq. (5.106).

32
 CONTENTS

6 Sources of Gravitational Waves

We want to consider different physical systems whose dynamics leads to the emission of gravitational
waves.

6.1 Rigid Rotator

As a first example we consider the emission of gravitational waves by a rigid rotating body. Consider a
coordinate system KS’ with coordinates x0m in which the body is fixed. In KS’ the mass density ρ0 (r 0 )
is time independent. We choose KS’ such that the quadrupole tensor Θ0ij is diagonal:
 
Z I1 0
Θ0ij = d3 r0 x0i x0j ρ0 (r 0 ) =  0 I2 0  . (6.1)
 

0 0 I3
We assume that the body rotates with angular velocity Ω around the x03 -axis. The orthogonal trans-
formation to an inertial system IS with coordinates xn can be written as
 
cos Ωt − sin Ωt 0
xn = αm n (t)x0m with αm n (t) =  sin Ωt cos Ωt 0 . (6.2)
 

0 0 1
The tensor Θij in IS reads
Z
Θij (t) = d3 r xi xj ρ(r, t)
Z
= d3 r0 (αn i x0n ) (αm j x0m ) ρ0 (r 0 )

= (α(t)Θ0 αT (t))ij (6.3)

where we used d3 r = d3 r0 and ρ0 (r 0 ) = ρ(r) since the density transforms as a scalar quantity. With
Eqs. (6.1) and (6.2) we can compute (6.3):
I1 + I2 I1 − I2
Θ11 (t) = + cos(2Ωt)
2 2
I1 − I2
Θ12 (t) = sin(2Ωt)
2
(6.4)
I1 + I2 I1 − I2
Θ22 (t) = − cos(2Ωt)
2 2

Θ33 (t) = I3 , Θ13 (t) = Θ31 (t) = 0.

This is of the form

Θij = const. + [Qij exp(−2iΩt) + c.c.] (6.5)

 
1 i 0
I1 − I2 
with Qij = i −1 0 . (6.6)

4
0 0 0

33
 CONTENTS

Comparing Eqs. (6.3) and (6.5) we conclude that a rotating rigid body can be interpreted as a mass
distribution whose rotational frequency Ω leads to gravitational waves of frequency ω = 2Ω. We
introduce the moment of inertia I with respect to the rotation axis and the ellipticity of the body ε,
I1 − I2
I = I1 + I2 , ε= . (6.7)
I1 + I2
We can then write
32GΩ6 2 2
P = ε I (6.8)
5c5
which is clearly the type of formula that one expects for quadrupole radiation (Q ∼ I).

6.1.1 Example: The Double Star System

As an example we consider a binary star system (masses M1 and M2 ) rotating on a Keplerian ellipse.
Assuming a circle with constant radius r, we can consider the system as a rigid rotator:
M 1 M2 2
I ' I1 = r , I2 ' 0, ε ' 1. (6.9)
M1 + M2
The circular orbit is characterized by
M1 M2 GM1 M2 GM
Ω2 r = ⇒ Ω2 = (6.10)
M1 + M2 r2 r3
| {z }
=µ

where µ is the reduced mass and M = M1 + M2 . Inserting this into Eq. (6.8) we find

32G4 M12 M22 (M1 + M2 ) 32 G4

P = = M 3 µ2 . (6.11)
5c5 r5 5 c5 r5
It is convenient to express the emitted power in terms of the Planck luminosity

c5
LP = = 3.63 × 1059 erg s−1
G

= 2.03 × 105 M c2 s−1 (6.12)

which yields
5
µ2

32 GM
P = LP . (6.13)
5 c2 r M2
GM
For example, coalescing neutron stars in the final stage have r ≈ RS and thus c2 r ∼ O(1) and
P ≈ LP . In general, for order of magnitude estimates one can use
 5
GM
P = LP (6.14)
c2 R

for systems of typical scale R.

34
 CONTENTS

Due to the emission of gravitational waves the system loses energy and thus its distance R shrinks,
until the two bodies coalesce after a time tspir. (inspiral time). In the Kepler problem the total energy
is
GM1 M2
E=− . (6.15)
2r
During the inspiral process, the system loses potential energy which is emitted in the form of gravita-
tional waves. Thus dE = −P dt and
dE GM1 M2 dr 32 G4 M12 M22 (M1 + M2 )
P =− =− = . (6.16)
dt 2r2 dt 5 c5 r5
Upon substituting x(t) = [r(t)/r(0)]4 we can rewrite the last equality in (6.16) as

dx 256G3 M1 M2 (M1 + M2 ) 1
=− ≡− . (6.17)
dt 5c5 r4 (0) tspir.

This is solved by x = 1 − t/tspir. such that we find

 1/4
t
r(t) = r(0) 1 − . (6.18)
tspir.

Next, we want to calculate the strain of such a system. We evaluate the expression (5.78) for eµν
in analogy to what we did with Eq. (5.79). We can express Tµν in terms of its spatial components
(k0 T 0i = kj T ij and k0 T 00 = ki T 0i ) which in turn can be expressed in terms of Qij :

ω2
Tij (k) = − Qij . (6.19)
2
With the definitions in (6.6) and (6.8) we obtain
1 GIε 1
e11 = 4
(2Ω)2 (6.20)
2 c D
where D is the distance between source and observer.
If there are two polarizations and e11 = e12 then we have for the dimensionless strain
q √ GIε 1
h = e211 + e212 = 2 2 4 Ω2 . (6.21)
c D
For a binary system characterized by masses M̃ = M1 = M2 on an orbit of radius r, this yields

GM̃ M̃ r2 RS2
r 2 Ω2 = , I= ⇒ h∼ (6.22)
r 2 Dr
where RS = 2GM̃ /c2 is the Schwarzschild radius of the system.
For two neutron stars with M̃ = 1.4M , r = 100 km (i.e. T ≈ 10−2 s, Ω ≈ 6 × 102 Hz) at a
distance D = 30000 ly ≈ 10 kpc we get

h ≈ 10−18 and P ≈ 1052 erg s−1 . (6.23)

This strain is the relative amplitude of the oscillation of a ruler’s length when the gravitational wave
passes through Earth.

35
 CONTENTS

6.2 The PSR 1913+16 System

The Hulse-Taylor binary system is a famous example of the confirmation of the gravitational wave
theory. Russell Alan Hulse and Joseph Taylor were awarded the 1993 Nobel Prize for the discovery
and measurement of this system. The system consists of two neutron stars, one of which is a pulsar
whose radio signals have been observed for many years. From the phase shift of the pulsar one can
infer the orbital parameters

τ = 0.06 s,

T = 27906.980894 ± 0.000002 s,

M1 = (1.442 ± 0.003)M ,

M2 = (1.386 ± 0.003)M ,

where T is the orbital time of the stars. Because of these very accurate pulse times, very precise
astrophysical measurements can be done with this system. Let us ignore ellipticity of the orbit for an
order of magnitude estimate. We obtain for the spiral time

tspir. ∼ 109 yr. (6.24)

From Kepler’s law we have T 2 ∝ r3 such that

dT dr
2 =3 . (6.25)
T r
Furthermore Eq. (6.18) implies that
 
dr 1 r(0)
=− . (6.26)
dt 0 4 tspir.
Therefore
dT 3 dr T 3 T
= =− ∼ 10−12 . (6.27)
dt 2 dt r 8 tspir.
Through the phase shift of the pulses one can measure the variation in the orbital time and finds
experimentally
dT
= −(2.43 ± 0.03) × 10−12 . (6.28)
dt
This is the same order of magnitude as the above simplified GR estimate.

So far we approximated the actual orbit by a circular one. We want to refine the analysis by taking
into account that the real orbit is actually elliptical in order to make more precise predictions. As
we will see, this more detailed analysis yields very good agreement with the experimental data (6.28).
Accounting for elliptical orbits (see the textbooks of Maggiore or Straumann for derivations), we have
for the semi-major axis (G = c = 1)
M 1 M2
a=− (6.29)
2E

36
 CONTENTS

where E is the total energy. Denoting by L the angular momentum, one finds for the squared eccen-
tricity of the orbit
2EL2 (M1 + M2 )
e2 = 1 + . (6.30)
M13 M23
The orbit can be described as
a(1 − e2 )
r= . (6.31)
1 + e cos ϑ
The emitted power reads
8M12 M22

12(1 + e cos ϑ)2 + e2 sin2 ϑ ϑ̇
 
P = (6.32)
15a2 (1 − e2 )2
where h·i denotes the time average over a period T ,

1 T 1 2π dϑ
Z Z
h·i = (·)dt = (·) . (6.33)
T 0 T 0 ϑ̇
Using Kepler’s third law
2πa3/2
T = , (6.34)
(M1 + M2 )1/2
after averaging one finds
32G4 M12 M22 (M1 + M2 )
 
73 2 37 4
P = 1+ e + e . (6.35)
5c5 a5 (1 − e2 )7/2 24 96
Then averaging
da M1 M2 dE
= (6.36)
dt 2E 2 dt
over time, one finds
2a2 64G4 M1 M2 (M1 + M2 )
     
da dE 73 2 37 4
= =− 5 1+ e + e . (6.37)
dt M1 M2 dt 5c a3 (1 − e2 )7/2 24 96
Using Kepler’s third law again, this yields
96 G4
   
Ṫ 3 ȧ
= = − 5 4 M1 M2 (M1 + M2 )f (e) (6.38)
T 2 a 5c a
 
73 37 1
where f (e) := 1 + e2 + e4 . (6.39)
24 96 (1 − e2 )7/2
If we replace a by T in the right-hand side of Eq. (6.38), we conclude
 
Ṫ 96 M1 M2
=− 5 f (e). (6.40)
T 5c (T /2π)8/5 (M1 + M2 )1/3
Plugging in the measured ellipticity of e ≈ 0.617, one gets

Ṫtheoretical = (−2.40243 ± 0.00005) × 10−12 (6.41)

which is in perfect agreement with the measured data. This is indirect evidence for the existence of
gravitational waves.

37
 CONTENTS

6.3 Waveform
ki
Let k̂i = k = ni , i = 1, 2, 3 and introduce the projector tensor

Pij (n) = δij − ni nj , (6.42)

which is symmetric, it transforms as ni Pij (n) = 0 and its trace is Pii = 2. With the help of Pij we
can construct Λij,kl as given by eq. (5.92) as follows
1
Λij,kl = Pik Pjl − Pij Pkl . (6.43)
2
Furthermore, it is transverse on all indices

ni Λij,kl = nj Λij,kl = 0 , (6.44)

and it is traceless with repeat to the (ij) and (kl) indices

Λii,kl = Λij,kl = 0 . (6.45)

Choosing n in the z direction, the transverse, traceless gauge (TT gauge) takes the following form
 
h+ h× 0   z 
hTijT (t, z) = h× −h+ 0 cos ω t − . (6.46)
 
c
0 0 0
ij

Using Λij,kl as defined in eq. (5.92) where k̂i = ni , it can be shown that the gravitational waves are
given in terms of the spacial components hij of hµν as

hTijT = Λij,kl hkl , (6.47)

where the summation over the k and l indices is implied. In the TT gauge, we have T = 0, indeed
from eq. (3.22) we see that it we take the trace, we get
T µ
0 = hµ µ ∼ S µ µ = T µ µ − δ µ = −T . (6.48)
2
Therefore we get with eq. (6.47), (3.22) and T = 0

Tkl (t − |r−r|
c ,r )
0
Z
4G
hTT
ij (t, r) = − 4 Λij,kl d3 r0 . (6.49)
c |r − r 0 |

As we are interested in the value of hTijT at large distances from the source, we take r → ∞. To leading
order we get
Tkl 1 0 r r 0 n̂
∼ Tkl (r , t − + ), (6.50)
|r − r 0 | r c c
since
|r − r 0 | r r 0 n̂ d2
 
= − +O , (6.51)
c c c r
where d is the size of the source → r0 . Furthermore
r r 0 n̂ r r0i ni 1
Tkl (r 0 , t − + ) ' Tkl (r 0 , t − ) + ∂0 Tkl + 2 r0i r0j ni nj ∂02 Tkl + ... , (6.52)
c c c c 2c

38
 CONTENTS

where all derivatives are evaluated at the point (r 0 , t − rc ). We define the momenta of the stress tensor
T ij as following
Z
S ij (tret ) = d3 xT ij (tret , x) , (6.53)
Z
S ij,k (tret ) = d3 xT ij (tret , x)xk , (6.54)
Z
S ij,kl (tret ) = d3 xT ij (tret , x)xk xl . (6.55)

In this notation the comma sparates the spatial indices of T ij from the indices coming from xi , ... .
Furthermore, note that the separated indices from T ij with the ones from xk , xl can be interchanged,
i.e. i ↔ j and k ↔ l. Inserting the expression (6.52) into eq. (6.50), we get
 
TT 1 4G kl 1 kl,m 1 kl,mp
hij (r, t) = Λij,kl (n̂) S + nm Ṡ + 2 nm np S̈ + ... , (6.56)
r c4 c 2c

where S kl and its derivatives are evaluated at the retarded time tret = t − rc . The physical meaning of
the various terms in the expansion becomes more clear if we eliminate the momenta of T ij in favor of
T 0i
the momenta of the energy density T 00 , and of the linear momentum c . We define the momenta of
T 00
c2 by
Z
1
M (t) = d3 xT 00 (t, x) , (6.57)
c2
Z
i 1
M (t) = 2 d3 xT 00 (t, x)xi , (6.58)
c

etc. (6.59)
T 0i
and the momenta of the momentum density c by
Z
1
P i (t) = d3 xT 0i (t, x) , (6.60)
c
Z
i,j 1
P (t) = d3 xT 0i (t, x)xj , (6.61)
c

etc. (6.62)

In the flat spacetime approximation T µν satisfies ∂µ T µν = 0. Thus, ∂0 T 00 = −∂i T 0i and recalling

that Ṁ = c∂0 M , it follows
Z Z Z
3 00 3 0i
cṀ = d x∂0 T =− d x∂i T =− dS i T 0i = 0 , (6.63)
V V ∂V

where in the last equality we used that T 0i vanishes on the boundary ∂V , as we have taken the volume
V larger than the volume of the source. Similarly, neglecting mass loss of the source, we obtain
Z Z Z Z Z
cṀ i = d3 xxi ∂0 T 00 = − d3 xi ∂j T 0j = d3 x(∂j xi )T 0j = d3 xδji T 0j = d3 xT 0i = cP i .
V V
(6.64)

39
 CONTENTS

One can show in the same ways that the following relations hold

Ṁ = 0 , Ṁ i = P i , Ṁ ij = P i,j + P j,i Ṁ ijk = P i,jk + P j,ki + P k,ij , (6.65)

where Ṁ = 0 is the mass conservation and

Ṗ i = 0 , Ṗ i,j = S ij , Ṗ i,jk = S ij,k + S ik,j , (6.66)

where Ṗ i = 0 is the momentum conservation of the source. Using the above relations, one gets
1 ij ... ijk
S ij = M̈ , M = 2(Ṡ ij,k + Ṡ ik,j + Ṡ jk,i ) , P̈ i,jk = Ṡ ij,k + Ṡ ik,j , (6.67)
2
1 ijk 1 i,jk
⇒ Ṡ ij,k =
M̈ + (P̈ + P̈ j,ik − 2P̈ k,ij ) . (6.68)
6 3
These relations can be inserted into eq. (6.56). One can proceed similarly with higher order terms
(∼ S̈ kl,mp ). To leading order, quadrupole radiation, eq. (6.56) becomes
1 2G r
hTijT = Λij,kl (n̂)M̈ kl (t − ) , (6.69)
r c4 c
with
1 1
M kl = (M kl − δ kl Mii ) + δ kl Mii , (6.70)
3 3
where Mii is the trace of Mij . The first term is traceless since Λij,kl gives zero when contracted with
δkl , only the first term contributes. We introduce the quadrupole moment
Z
1 1
Qij = M ij − δ ij Mkl = d3 x(xi xj − r2 δ ij )ρ(t, x) , (6.71)
3 3
then eq. (6.69) becomes
1 2G r
hTijT (t, r) =
Λij,kl (n̂)Q̈kl (t − ) . (6.72)
r c4 c
Assume a wave traveling along the z direction, thus n̂ = ẑ, then it can be shown that
 
(M̈11 − M̈22 )/2 M̈12 0
Λij,kl M̈kl =  M̈21 −(M̈11 − M̈22 )/2 0 , (6.73)
 

0 0 0

and thus
1G
h+ = (M̈11 − M̈22 ) , (6.74)
r c4
2G
h× = M̈12 , (6.75)
r c4
where it is understood that the right-hand side is computed at the retarded time tret = t − rc .
For a generic direction ni = (sin θ sin φ, sin θ cos φ, cos θ). Then the components ni and n0i (for
instance n0i in z-axis), are related by the rotation matrix Rij : ni = Rij n0j with
  
cos φ sin φ
1 0 0 0
R = − sin φ cos φ 0 0 cos θ sin θ  (6.76)
  

0 0 1 0 − sin θ cos θ

40
 CONTENTS

0
and Mij = Rik Rjl Mkl . We get then
1 G
h+ (t : θ, φ) = M̈11 (cos2 φ − sin2 φ cos2 θ) + M̈22 (sin2 φ − cos2 φ cos2 θ)+
r c4

− M̈33 sin2 θ − M̈12 sin 2φ(1 + cos2 θ) + M̈13 sin φ sin 2θ + M̈23 cos φ sin 2θ , (6.77)

1 G
h× (t : θ, φ) = (M̈11 − M̈22 ) sin 2φ cos θ + 2M̈12 cos 2φ cos θ − 2M̈13 cos φ sin θ+
r c4

+ 2M̈23 sin φ sin θ , (6.78)

and remember that M̈ij is evaluated at t − rc . This equation allows us to compute the angular
distribution of the quadrupole radiation once Mij is given.

6.4 Quadrupole radiation from a mass in circular orbit

We consider a binary system with masses m1 and m2 on a circular orbit of radius R. We neglect any
back-reaction on the motion due to gravitational waves emission. We chose our coordinate system such
that the orbit lies in the (x, y) plane and it is given by
 π
x0 (t) = R cos ωs t + , (6.79)
2
 π
y0 (t) = R sin ωs t + , (6.80)
2

z0 (t) = 0 , (6.81)
π m1 m2
where the phase 2 is a useful choice of the origin of time. We use the reduced mass µ = m1 +m2 of the
system to express the second mass moment

M ij = µxi0 (t)xj0 (t) , (6.82)

in the center of mass. Using the trigonometric identity cos 2α = 2 cos2 α − 1 and the coordinates given
in eq. (6.81), we get
1 − cos(2ωs t)
M11 = µR2 , (6.83)
2
1 + cos(2ωs t)
M22 = µR2 , (6.84)
2
1
M12 = − µR2 sin 2ωs t , (6.85)
2
while the other components vanish and we also compute

M̈11 = 2µR2 ωs2 cos(2ωs t) , (6.86)

M̈12 = 2µR2 ωs2 sin(2ωs t) , (6.87)

M̈11 = −M̈22 . (6.88)

41
 CONTENTS

Inserting these relations into eq. (6.77) and eq. (6.78), we get

1 4Gµωs2 R2 1 + cos2 θ
 
h+ (t, θ, φ) = cos(2ωs tret + 2φ) , (6.89)
r c4 2

1 4Gµωs2 R2
h× (t, θ, φ) = cos θ sin(2ωs tret + 2φ) , (6.90)
r c4
v2 Gm
where the orbital frequency ωs is related to the orbital radius R by v = ωs R and R = R2 where
Gm
m = m1 + m2 thus ωs = R3 . Note that the quadrupole radiation given in eqs. (6.89) and (6.90) is at
twice the frequency ωs of the source.
We now introduce the chirp mass
3
3 2 (m1 m2 ) 5
Mc = µ 5 m 5 = 1 , (6.91)
(m1 + m2 ) 5
ωGW
moreover we define fGW := 2π where ωGW = 2ωs . Then eqs. (6.89) and (6.90) become
 53   23
1 + cos2 θ

4 GMc πfGW
h+ (t, θ, φ) = cos(2πfGW tret + 2φ) (6.92)
r c2 c 2
  53   23
4 GMc πfGW
h× (t, θ, φ) = cos θ sin(2πfGW tret + 2φ) , (6.93)
r c2 c

From eq. (6.17) and eq. (6.18) we had that the typical inspiral time, i.e. time to coalescence, is

5 c5 R04
tspir = . (6.94)
256 G3 m2 µ
2π
Expressing the initial radius R0 in terms of the initial orbital period T0 = ωs (tspir ) through Kepler’s
third law R03 2
= Gm(T0 /2π) , we find
  83   32  
T0 M M
tspir ' 9.8 × 106 yr . (6.95)
1 hour m µ
Thus, for masses ∼ M in circular orbits: only binaries which had at formation an initial period of
less than about one day can have coalesced by emission of gravitational waves within the present age
of galaxies.
A particle that moves on a quasi-circular orbit in the (x, y) plane with a radius R = R(t) and angular
velocity ωs = ωs (t) has cartesian coordinates x(t) = R(t) cos(Φ(t)/2) and y(t) = R(t) sin(Φ(t)/2) where
we have defined Z t Z t
0 0
Φ(t) = 2 dt ωs (t ) = dt0 ωGW (t0 ) . (6.96)
t0 t0

When computing the gravitational waves production in the quadrupole approximation in the time
derivatives we have additional terms. This implies the following:

• in the argument of the trigonometric functions, ωGW t has to be replaced by Φ(t);

• in the factors in front of the trigonometric functions, ωGW or fGW is replaced by ωGW (t) or
fGW (t);

42
 CONTENTS

• one should also include the contributions coming from the derivatives of R(t) and ωGW (t) however
Ṙ is negligible as long as ω̇s  ωs2 . In the following we shall assume that this condition is fulfilled
and thus neglect those terms.

Eqs. (6.92) and (6.93) become then

 53   23 
1 + cos2 θ
 
4 GMc πfGW (tret )
h+ (t) = cos Φ(tret ) , (6.97)
r c2 c 2
  53   23
4 GMc πfGW (tret )
h× (t) = cos θ sin Φ(tret ) , (6.98)
r c2 c

where fGW → fGW (tret ) and 2πfGW tret + 2φ → 2ωs tret + 2φ. In the coalescence phase fGW increases
1
to fGW ∼ T → ∞, however when the black holes are very closed all the approximations we used break
down. The functional dependence of h+ or h× on t is shown in Figure 5.

Figure 5: Time evolution of the gravitational wave amplitude in the inspiral phase of a binary system.

43
 CONTENTS

7 The post-Newtonian Approximation

Astrophysical sources of gravitational radiation are held together by gravitational forces. For a self-
gravitating system of mass m we have
 v 2 RS
∼ (7.1)
c d
where RS is the Schwarzschild radius and d is a typical size of the system. This relation follows
immediately from v 2 ≈ Gm/r ≈ Gm/d. We note that RS /d is a measure of the strength of the
gravitational field close to the source. As soon as we consider v/c corrections to the orbital motion,
for consistency we have to consider also (RS /d)1/2 corrections to the metric (corrections to the flat
background).
Moderately relativistic systems require a post-Newtonian treatment. The assumptions that we
will use are

• The systems under consideration are moving slowly, weakly self-gravitating systems such that an
expansion in v/c or (RS /d)1/2 is possible.

• The energy-momentum tensor Tµν has a spatially compact support (Tµν (r) = 0 for r ≥ d).

If ωS is a typical frequency of the system, then typical velocities are v ≈ ωS d. As we saw before,
the frequency of the radiation of the emitted gravitational radiation is ω = 2ωS ≈ 2v/d. In non-
relativistic systems we have v  c, thus c/v  1 and the wavelength of the emitted radiation satisfies
λ = c/ω ∼ cd/v  d.
In analogy to the electromagnetic case, for non-relativisic sources it is convenient to distinguish
between:

• near field regime (r, d  λ) where retardation is negligible and potentials are static,

• far field regime (r  λ) where retardation is crucial and we have waves.

The small parameter in powers of which we will perform an expansion, is15

 1/2
RS v
ε∼ ∼ . (7.2)
d c
We demand that
|T ij |
∼ O(ε2 ). (7.3)
T 00
For a fluid with pressure p and energy density ρ, we thus have
p
∼ ε2 . (7.4)
ρ
We expand the metric and the energy-momentum tensor in the near field regime in powers of ε.
As long as the emission of radiation is neglected, a classical system subject to conservative forces is
invariant under time reversal. Under time reversal, g00 and gij are even (i.e. there appear even powers
15 Note that some references also use the convention ε ∼ (v/c)2 .

44
 CONTENTS

of v and thus of ε, as well) while g0i is odd (i.e. only odd powers of v and thus of ε appear). By
inspection of Einstein’s field equations, one finds that in order to work consistently to a given order ε,
if we expand g00 up to O(εn ), then we have to expand g0i up to O(εn−1 ) and gij up to O(εn−2 ).
The metric is expanded as follows:

g00 = −1 + (2) g00 + (4) g00 + (6) g00 + ...

(3)
g0i = g0i + (5) g0i + ... (7.5)

gij = δij + (2) gij + (4) gij + ...

(n)
where gµν denotes terms of O(εn ). Similarly, for the energy-momentum tensor:

T 00 = (0) T 00 + (2) T 00 + ...

T 0i = (1)
T 0i + (3) T 0i + ... (7.6)

T ij = (2)
T ij + (4) T ij + ...

We now want to insert these expansions into Einstein’s field equations and equate terms of the same
order in ε. Considering v  c, the time derivatives of the metric are smaller than the spatial derivatives
by O(ε):

∂ ∂
= O(v) or ∂0 ∼ O(ε)∂i , (7.7)
∂t ∂x
where we used that ∂0 = 1c ∂t . The d’Alembert operator applied to the metric, to lowest order becomes
the Laplacian:

1 ∂2
 
= O(ε2 ) + 1 ∆.


− +∆ (7.8)
c2 ∂t2

This means that retardation effects are small corrections.

Consequently, we also have to expand the geodesic equation

d2 xi µ
i dx dx
ν
= −Γ µν . (7.9)
dτ 2 dτ dτ
In chapter 9.3 of GR I , we considered the Newtonian limit where we just had to use g00 = −1 + (2) g00 ,
g0i = 0 and gij = δij . It thus follows that the terms (4) g00 , (3) g0i and (2) gij give the first post-Newtonian
(6) (5) (4)
order for which we use the notation 1PN. The terms g00 , g0i , gij give the 2PN approximation
(which is already highly complicated), and so on.

7.1 The 1PN Approximation

It is useful to choose a simplifying gauge condition right from the beginning. A convenient choice is
the de Donder gauge condition
√
∂µ ( −gg µν ) = 0. (7.10)

45
 CONTENTS

(this is a harmonic gauge condition, i.e. the coordinate functions satisfy the d’Alembert equation).
The next step is to insert Eqs. (7.5), (7.6) into Einstein’s equations (together with (7.7) and (7.10)).
We skip the explicit computations but state only the results16 . One finds for (2)
g00 the Newtonian
equation
h
(2)
i 8πG
∆ g00 = − 4 (0) T 00 (7.11)
c
while the 1PN correction to the metric yields
h i
(2) 8πG
∆ gij = − 4 δij (0) T 00 , (7.12)
c
h i 16πG
∆ (3) g0i = (1) 0i
T , (7.13)
c4
h i h i h i h i h i
∆ (4) g00 = ∂02 (2) g00 + (2) gij ∂i ∂j (2) g00 − ∂i (2) g00 ∂i (2) g00

8πG h(2) 00 (2) ii i

− 4
T + T − 2(2) g00 (0) T 00 , (7.14)
c
where ∆ = δ ij ∂i ∂j and the sum over repeated spatial indices is performed with δ ij .
The solution of (7.11) with the boundary condition that the metric is asymptotically flat, is
(0)
T 00 (x0 , t)
Z
G
(2)
g00 = −2φ with φ(t, x) = − 4 d3 x0 (7.15)
c |x − x0 |

where the Newtonian potential is U = −c2 φ. Similarly Eqs. (7.12) and (7.13) are solved by

(2)
gij = −2φδij (7.16)

(3)
g0i = ξi (7.17)

where we defined
(1)
T 0i (x0 , t)
Z
4G
ξi (x, t) = − d3 x0 . (7.18)
c4 |x − x0 |
(2)
In order to solve (7.14), we replace g00 on the right-hand side by −2φ and 2 gij by −2φδij .
Furthermore we use the identity
1
(∇φ)2 = ∂i φ∂i φ = ∆(φ2 ) − φ∆φ. (7.19)
2
and we introduce a new potential ψ such that

(4)
g00 = −2(φ2 + ψ). (7.20)

Eq. (7.14) then reads

4πG (2) 00 (2) ii 

∆ψ = ∂02 φ + T + T . (7.21)
c4
16 For details, see section 9.1. in Weinberg’s book or section 5.2 in Straumann’s.

46
 CONTENTS

Using the boundary condition that ψ vanishes at spatial infinity, ψ can be written as
d3 x0
Z  i
1 2 G h(2) 00 0 (2) ii 0
ψ(x, t) = − ∂ φ+ 4 T (x , t) + T (x , t) . (7.22)
|x − x0 | 4π 0 c

Notice that φ and ξi are not independent due to the gauge condition (7.10) which imposes the
contraint

4∂0 φ + ∇ · ξ = 0. (7.23)

From Eqs. (7.15) and (7.18) one can see that this condition is indeed satisfied due to energy-momentum
conservation at the 1PN order (since T µν is covariantly conserved in the exact solution, it has to be
conserved at all post-Newtonian orders independently).
We observe that φ, ψ, ξi are instantaneous potentials. Our order of approximation is thus insensitive
to retardation effects.
Note also that g00 can be expressed very simply as
2
g00 = −e−2V /c + O(ε6 ) (7.24)

where V = −c2 (φ + ψ). This follows immediately if we expand the exponential and write (7.24) as

2V 2
 
2V 1
g00 = −1 + 2 − 4 + O 6
c c c

= −1 − 2(φ + ψ) − 2(φ + ψ)2 + O(ε6 )

= −1 − 2φ − 2(φ2 + ψ) − 2(ψ 2 + 2φψ) + O(ε6 ). (7.25)

Using that φ = O(ε2 ) and ψ = O(ε4 ), we see that this is just

g00 = −1 + (2) g00 + (4) g00 + O(ε6 ). (7.26)

(2)
Putting together (7.11) (with g00 = −2φ) and (7.21), we have
4πG h i
∆(φ + ψ) = ∂02 φ + 4 (0) T 00 + (2) T 00 + (2) T ii . (7.27)
c
To this order we can set ∂02 φ = ∂02 (φ + ψ) and replace ∆ by . We then obtain
4πG  00 4πG
T + T ii ≡ − 2 σ

V = − 2
(7.28)
c c
(0)
where we replaced T 00 + (2) T 00 → T 00 and (2)
T ii → T ii . The solution of Eq. (7.28) is given by a
retarded integral
|x − x0 | 0
Z  
3 0 1 00 ii


V (x, t) = G d x T +T t− ,x (7.29)
|x − x0 | | {z } c
=σ

and similarly for ξi defined as Vi , given by

|x − x0 | 0
Z  
1
Vi (x, t) = G d3 x0 σi t − ,x (7.30)
|x − x0 | c

47
 CONTENTS

where σi ≡ T 0i .
To summarize, in harmonic coordinates the 1PN solution can be written in terms of two functions
V and Vi in the following way:
 
2 2 2 1
g00 = −1 + V − V + O ,
c2 c4 c6
 
4 1
g0i = − 3 Vi + O 5 , (7.31)
c c
   
2 1
gij = δij 1 + 2 V + O 4 .
c c

7.2 Motion of Test Particles in the 1PN Metric

To get the equations of motion of a particle of mass m in the near zone, we have to solve the geodesic
equation which is encoded in the following action which reads in curved background
1/2
dxµ dxν
Z 
S = −mc dt −gµν
dt dt
1/2
vi vi vj
Z 
2
= −mc dt −g00 − 2g0i − gij 2 + . . . . (7.32)
c c

We are interested in the equations of motion for a binary system. If we restrict ourselves to the lowest
PN corrections, it is possible to treat the two masses as point-like.
In curved space the energy-momentum tensor of a set of point-like particles is given by
N
1 X dxµ dxν
T µν = √ γa ma a a δ (3) (x − xa (t)) . (7.33)
−g a=1 dt dt

where the masses are denoted by ma and xµa are the coordinates (a = 1, ..., N ). Furthermore we used
the definitions
dτa dxµa
q
γa−1 = = −g00 − gij vai vaj /c2 and pµa = γa ma . (7.34)
dt dt
In an N -body system (a > 2) the metric felt by a particle b is obtained by taking the energy-momentum
P P
tensor of all other particles as a source. This amounts to replacing a by a(6=b) in Eq. (7.33). We
(2)
expand the determinant of the metric to second order and using g00 = −2φ, we get
X
−g = 1 − (2) g00 + (2)
gii = 1 − 4φ. (7.35)
i

48
 CONTENTS

Therefore the expansion of (7.33) gives

X
(0)
T 00 (x, t) = ma c2 δ (3) (x − xa (t))
a(6=b)

 
(2) 00
X 1 2
T (x, t) = ma v + φc δ (3) (x − xa (t))
2
2 a
a(6=b)
(7.36)
X
(1) 0i
T (x, t) = c ma vai δ (3) (x − xa (t))
a(6=b)

X
(2)
T ij (x, t) = ma vai vaj δ (3) (x − xa (t)).
a(6=b)

Inserting these expressions into Eqs. (7.15), (7.18) and (7.22), one can obtain the metric in which the
particle b propagates. Inserting this metric into (7.32), one can calculate its action Sb . The total
P
action can then be obtained by summing over all particles, S = b Sb . Expanding the square root
that appears in the integral of the action and demanding consistency of the expansion, only terms up


to O (v/c)4 give the 1PN correction.
In terms of the Lagrangian, one can verify the following results for the two-body system:
1
L = L0 + L1 (7.37)
c2
1 1 Gm1 m2
with L0 = m1 v12 + m2 v22 + , (7.38)
2 2 r
1 1
L1 = m1 v14 + m2 v24 +
8 8
 
Gm1 m2 G(m1 + m2 )
+ 3(v12 + v22 ) − 7v1 v2 − (r̂v1 )(r̂v2 ) − . (7.39)
2r r
where r denotes the separation vector between the two particles, r = |r| and r̂ = rr . The Lagrangians
L0 and L1 describe the Newtonian part and the first post-Newtonian correction, respectively.
The same can be obtained for N particles (Einstein-Infeld-Hoffmann Lagrangian) :
X1 X Gma mb
L0 = ma va2 + , (7.40)
a
2 2rab
a(6=b)

X1 X Gma mb
L1 = ma va4 − [7va vb + (r̂ab va )(r̂ab vb )] +
a
8 4rab
a(6=b)

3G X X ma mb va2 G2 X X X ma mb mc
+ − , (7.41)
2 a rab 2 a rab rac
b6=a b6=a c6=a

where a = 1, ..., N labels the particles, rab is the distance between particles a and b, and r̂ab is
the corresponding unit vector. From this Lagrangian one can derive the Euler-Lagrange equations
of the N particle system including 1PN corrections. These equations are also called the Einstein-
Infeld-Hoffmann equations. Denoting rab = |xa − xb | and xab = xa − xb , one finds after lengthy

49
 CONTENTS

calculations
  
X xab X mc X 1 xab xbc
v˙a = −G mb 3 1 − 4G +G mc − + 3 −
rab rac rbc 2rbc
b6=a c6=a c6=a,b

 2 
ma 3 vb xab
− 5G + va2 − 4va vb + 2vb2 − −
rab 2 rab
  X  
7 X mb mc xbc X xab
− G G 3 + G m b 3 (4va − 3vb ) · (va − vb ). (7.42)
2 rab rbc rab
b6=a c6=b,a b6=a

7.2.1 Two Body Problem in the 1PN Approximation

The Einstein-Infeld-Hoffmann equations for the two body problem imply that the center of mass
m∗1 x1 + m∗2 x2
X= (7.43)
m∗1 + m∗2
with
1  v 2 1 m m G
a a b
m∗a := ma + ma − (a, b = 1, 2, a 6= b) (7.44)
2 c 2 rab c2
is not accelerated, i.e.
d2 X
= 0. (7.45)
dt2
We can choose X = 0 such that
 
m2 µδm  2 m 
x1 = + v − x (7.46)
m 2m2 r
 
m1 µδm  2 m 
x2 = − + v − x (7.47)
m 2m2 r
m1 m2
where x = x1 − x2 , v = v1 − v2 , m = m1 + m2 , δm = m1 − m2 , µ = m (reduced mass).
For the relative motion we obtain from Eqs. (7.38), (7.39) with (7.46), (7.47) after dividing by µ:

Lrel = L0 + L1 (7.48)

with the Newtonian part L0 and the post-Newtonian perturbation L1 :

1 2 Gm
L0 = v + (7.49)
2 r
3µ v 4 G2 m2
   
3 Gm 2 µ 2 µ  vx 2
L1 = 1− 2
+ 2
3v + v + − 2 2. (7.50)
8 m c 2rc m m r 2r c
The corresponding Euler-Lagrange equation is (c = 1)
       
Gm Gm 2µ 3µ 3µ  vx 2
v̇ = − 3 x 1 − 4+ + 1+ v2 − +
r r m m 2m r
 
Gm 2µ
+ 3 v(vx) 4 − . (7.51)
r m

50
 CONTENTS

~r ϕ
a
x
S

Figure 6: Elliptical Kepler orbit.

Consider the Kepler problem with motion in the plane z = 0 and the periastron lying on the x-axis.
Without 1PN corrections one has an elliptic orbit (c.f. fig. 6) where e is the eccentricity and

a(1 − e2 )
r= . (7.52)
1 + e cos ϕ
One finds

r = a(1 − e cos u) where u − e sin u = ω0 t (7.53)

2π
(Kepler’s equation). Here ω0 = T with orbital period T (t = 0 for passage at perihelion) and u is
the so called eccentric anomaly (ω0 t is the mean anomaly). We define f = ϕ, the true anomaly,
such that (c.f. fig. 7)
cos u − e
cos f = cos ϕ = . (7.54)
1 − e cos u
The invariance under time translations and space rotations of the Lagrangian Lrel implies four constants
of motion – the reduced energy ε and the three components of the reduced angular momentum j:
E ∂Lrel
ε= =v· − Lrel
µ ∂v

v4
   
1 2 Gm 3 3µ
Gm  µ 2 µ  vx 2 Gm
= v − + 1−
+ 3 + v + + , (7.55)
2 r 8 c2 2rc2
m m m r r

3µ v 2 
   
J ∂Lrel 1 µ  Gm
j= =x∧ = (x ∧ v) 1 + 1− + 3 + . (7.56)
µ ∂v 2 m c2 m rc2

After some lengthy calculations (including an integration over time), one can get an equation which is

51
 CONTENTS

~r

u ϕ
x
S

Figure 7: The variable u takes the role of the angle as measured from the center of the ellipse.

analogous to Kepler’s equation but with different coefficients:

2π
t = u − et sin u, (7.57)
Tb

r = ar (1 − er cos u), (7.58)

where
Gm h µ  ε i
ar = − 1− −7 (7.59)
2ε M 2c2
2ε h  µ  ε i µ  G2 m2 
2 2
er = 1 + 2 2 1 + 5 − 15 j + −6 (7.60)
G m m 2c2 m c2

2µ G2 m2
   
2ε h  µ ε i 2
e2t = 1 + 2 2 1 + 17 − 7 j + 2 − (7.61)
G m m 2c2 m c2

2π (−2ε)3/2 h µ  ε i
= 1− − 15 , (7.62)
Tb Gm m 4c2
with j = |j|. Notice that, in a bound orbit, ε is negative. The eccentricity e of the Keplerian orbit is
now split into a “radial eccentricity” er and a “time eccentricity” et . The Newtonian limit is found by
considering c → ∞:
 Gmµ
ar c→∞ = (7.63)
−2E
2EL2
e2r c→∞ = e2t c→∞ = 1 + 2 2 3 ,
 
(7.64)
G m µ
where L simply denotes the Newtonian limit of |J |, which is nothing but the classical angular momen-
tum L ≡ µ |r ∧ v|.

52
 CONTENTS

One finds for the true anomaly

" 1/2 #
cos u − eθ 1 + eθ u
ϕ(u) = or ϕ(u) = Aeθ (u) = 2 arctan tan (7.65)
1 − eθ cos u 1 − eθ 2

where eθ = e for Kepler orbits. This yields

3Gm
ϕ(u) = ω0 + (1 + k)Aeθ (u) with k = (7.66)
c2 a(1 − e2 )
with ω02 = (2π/T )2 = Gm/a3 . The quantity eθ is called “angular eccentricity” and it satisfies
 ε i 6G2 m2

2ε h µ
e2θ = 1 + 2 2 1 + − 15 j 2
− . (7.67)
G m m 2c2 c2
For c → ∞, we have to make the replacement eθ → e.
One can also find the perihelion precession (or periastron precession) per orbit (c.f. section 25 of
GR I ). After some calculations one finds:

6πG(m1 + m2 )
δϕ = . (7.68)
c2 a(1 − e2 )

Note that a in this formula is the semi-major axis (a(1 − e2 ) = p); in section 25 of GR I it was half of
Gm
the Schwarzschild radius, a = c2 . The relevant difference as compared to section 25 is the fact that
we had only one mass (the Sun) in section 25. In (7.68) m = m1 + m2 is the sum of the two masses.
In view of the non-linearities that are involved in the description of the system, this simple result is
far from obvious.

For the binary pulsar PSR 1913+16 (c.f. section 6.2), the measured periastron shift is

ω̇obs. = 4.226607 ± 0.00007 deg yr−1 (7.69)

The GR prediction which follows from Eq. (7.68) and the known orbital element (given a period of
about 7.75 hours, i.e. ω̇ ∼ δϕ × 1130 deg yr−1 ) is
 2/3
m1 + m2
ω̇GR = 2.11 deg yr−1 (7.70)
M
where we used
 1/3
G
a= (m1 + m2 ) (7.71)
ω02
due to Kepler’s third law. If we set ω̇obs. = ω̇GR , it follows m1 + m2 = 2.83M .
In order to characterize the pulsar system, several parameters are relevant:

• The parameters which characterize the pulsar itself: right ascension α, declination δ, proper
motion, the initial pulse time φ0 , the pulse frequency ν, and the spindown parameter.

• The five Keplerian parameters (c.f. fig. 8): Tb (period), t0 (time of passage at periastron),
x = ar sin i (i denotes the inclination angle of the orbital plane with respect to the observer), e
(eccentricity), ω (angular position of periastron as measured from the ascending node).

53
 CONTENTS

~r

Figure 8: Sketch of spatial arrangement of post-Keplerian orbit relative to observer.

• There are eight independent measurable post-Keplerian parameters (we state only the five main
parameters):

– ω̇ (periastron shift),
– γ (not to be confused with γ from Robertson expansion),
– ∆E (Einstein time delay) which is related to the transformation from the pulsar proper time
to the coordinate time of the pulsar-companion barycenter system. One finds

∆E = γ sin u (7.72)
1/3
G2/3 m2 (m1 + 2m2 )

Tb
where γ = e
2π c2 a(m1 + m2 )
   −4/3
m2 m1 + 2m2 m1 + m2
= 2.93696 ms · (7.73)
M M M

where m2 is the mass of the companion and m1 is the mass of the emitting pulsar.
Gm2 a
– r= c3 (Shapiro time delay) which corresponds to c = ν in Eq. (1.12),
– Ṫb see Eq. (6.38) due to emission of gravitational waves and depends also on the masses m1
and m2 .
Tb −2/3
– s = sin i = cG−1/3 m2/3 m−1


2π 2
eθ −et G 7 2


– δθ = et = c2 am 2 m1 + 6m1 m2 + 2m22 where we used Eqs. (7.67) and (7.61).

Seven parameters are needed to fully specify the dynamics of the two-body system (up to un-
interesting rotation about the line of sight). Therefore, the measurement of any two post-Keplerian
parameters (besides five Keplerian parameters) allows to predict the remaining ones. These parameters
thus constitute a consistency check for GR.

54
 CONTENTS

8 The Kerr Solution

The Kerr solution is one of the most important solutions to Einstein’s (vacuum) equations. It describes
stationary rotating black holes, and was found in 1963 by R. Kerr17 . Later, it was generalized to the
Kerr-Newman solution which describes rotating, electrically charged black holes.
In this chapter we will give an overview over the Kerr solution and the Kerr-Newman solution
without detailed calculations.
The Kerr solution is axisymmetric and stationary. We use the so-called Boyer-Lindquist coor-
dinates (t, r, θ, ϕ) and we use the following abbreviations:

∆ = r2 − 2mr + a2 , ρ2 = r2 + a2 cos2 θ (8.1)

where we set G = c = 1. The Kerr metric reads


1
ds = 2 − (∆ − a2 sin2 θ)c2 dt2 + 2a sin2 θ(∆ − r2 − a2 )c dtdϕ+
2
ρ
  2  (8.2)
2 2 2 2 2 2 2 dr

2 2
+ sin θ (r + a ) − ∆a sin θ dϕ + ρ + dθ .
∆

8.1 Interpretation of the Parameters a and m

In order to interpret a and m we look at the asymptotic form of the metric (8.2) for large “radial
coordinate” r:
     
2 2m 1 2 4am 2 1
ds = − 1 − +O dt − sin θ + O dtdϕ+
r r2 r r2
  
1
dr + r2 (dθ2 + sin2 θ dϕ2 ) .
 2 
+ 1+O (8.3)
r
The examination is easier by transforming to asymptotically Lorentzian coordinates (Cartesian coor-
dinates):

x = r sin θ cos ϕ, y = r sin θ sin ϕ, z = r cos θ. (8.4)

This yields the following form of the asymptotic metric:

     
2m 1 4am 1
ds2 = − 1 − +O dt2
− + O (x dy − y dx)dt+
r r2 r3 r4
  
1  2
dx + dy 2 + dz 2 ,

+ 1+O (8.5)
r

where we used r2 sin2 θ dϕ = x dy − y dx.

In section 4 we computed the metric of the rotating Earth assuming slow rotation. In Eq. (4.18)
we found
   
2GME 2 2 2GME
ds2 = 1 − c dt − 1 + dr 2 + 2ch0i dxi dt. (8.6)
c2 r c2 r
17 Phys. Rev. Lett. 11, 237 (1963)

55
 CONTENTS

GME
Identifying c2 → m, dr 2 = dx2 + dy 2 + dz 2 and including an overall sign (since we used another
sign convention in section 4), this metric should coincide with (8.5).
Using Eq. (4.1),

ω k xj
Z
4G
h0i = 3
εikn 3 d3 re x
en ρ(e
r )e
xj , (8.7)
c r
we infer that T 0i is proportional to ρ vci where vi = εikn ω k xn with ω k being the angular velocity of the
rotating body. We define
Z
Sk = εklm d3 x xl T m0 (8.8)

which is the intrinsic angular momentum of the rotating body. Therefore,

4G xm S k
h0i = ε ikm . (8.9)
c3 r3
With the sign convention of (8.5), Eqs. (8.5) and (8.6) indeed take the common form
     
2m 1 2m 1
ds2 = − 1 − +O 2
dt2
+ 1 + + O δij dxi dxj −
r r r r2

S k xl
  
1
− 4εikl 3 + O dtdxi . (8.10)
r r4

Clearly S k is proportional to the body’s mass M and its angular momentum (cf. Eq. (4.13)):
aGM ∂ GM  unit vector along polar axis 
Sk = = a × of Boyer-Lindquist coordinates . (8.11)
c2 ∂z c2
Therefore, m is just the mass, and a can be interpreted as the angular momentum (0 ≤ a ≤ 1).

8.2 Kerr-Newman Solution

The Kerr-Newman solution is the extension of the Kerr solution which also describes electrically
charged black holes (c = G = 1). We use the three parameters

∆ = r2 − 2M r + a2 + Q2 (8.12)

ρ2 = r2 + a2 cos2 θ (8.13)

Σ2 = (r2 + a2 )2 − a2 ∆ sin2 θ (8.14)

where M is the total mass, a the intrinsic angular momentum and Q the total charge. The metric
coefficients of the Kerr-Newman metric are

ρ2 Σ2
grr = , gθθ = ρ2 , gϕϕ = sin θ,
∆ ρ2

2M r − Q2 2M r − Q2
gtt = −1 + , gtϕ = −a sin2 θ
ρ2 ρ2

where we assumed a > 0 without loss of generality. This metric contains the following special cases:

56
 CONTENTS

• Q = a = 0: Schwarzschild solution,

• a = 0: Reissner-Nordstrøm solution,

• Q = 0: Kerr solution.

The electromagnetic field of the Kerr-Newman solution is

F = Qρ−4 (r2 − a2 cos2 θ) dr ∧ (dt − a sin2 θ dϕ)+

+ 2Qρ−4 ar cos θ sin θ dθ ∧ (r2 + a2 )dϕ − adt

(8.15)

where ∧ denotes the exterior product defined in section 15 of GR I .

From (8.16) one can deduce the asymptotic expressions for electric and magnetic fields (in r, θ, ϕ
directions):
 
Q 1
Er = Frt = + O
r2 r3
 
Eθ Fθt 1
= =O
r r r4

Eϕ Fθt
= =0
r sin θ r sin θ
  (8.16)
Fθϕ 2Qa 1
Br = 2 = 3 cos θ + O
r sin θ r r4
 
Fϕr Qa 1
Bθ = = 3 sin θ + O
r sin θ r r4

Frθ
Bϕ = = 0.
r
We see immediately that asymptotically the electric field is a Coulomb field.

8.3 Equations of Motion for Test Particles

Let a test particle with electric charge e and rest mass µ move in the external fields of a Kerr-Newman
black hole. The equations of motion are

d2 xα α dxβ dxγ α dx
β
+ Γ βγ = eF β (8.17)
dλ2 dλ dλ dλ
The best way to solve this equation turns out to be the Hamiltonian formalism.18
We can simplify the analysis by assuming that the metric is that of a Kerr black hole (Q = 0)
and that the motion is confined to the equatorial plane (θ = π/2).19 In this case, the metric has the
18 Detailscan be found in Straumann’s book or the book by Misner, Thorne, Wheeler.
19 Note that the equatorial plane of a rotating black hole is distinguished. When we solved the geodesic equations in
the Schwarzschild background, any arbitrary plane which includes the origin was equivalent.

57
 CONTENTS

following non-vanishing components:

 
2GM
gtt = − 1 − 2 , (8.18)
c r

a 2GM
gtϕ = − , (8.19)
r c2
a2 2GM
 
2 2
gϕϕ = r +a + . (8.20)
r c2
E
Denoting by K = µc the total energy and by l the angular momentum of the particle, one finds

1 2
µṙ + µVeff. = const. (8.21)
2
with the effective potential
2GM 2 2GM 2
c2 c l2 − a2 (K 2 − c2 ) c2 (l − aK)
Veff. = − + −
2r 2µ2 r2 2µ2 c2 r3

GM l2 − a2 (K 2 − c2 ) GM (l − aK)2
=− + − . (8.22)
r 2µ2 r2 µ2 c2 r3
For a = 0 this reduces to the Schwarzschild case.

For a black hole which is spinning extremely fast, it can be shown that for a particle which spirals
in towards the
 black hole
 in an accretion disc from very far away to the innermost circular stable orbit,
the fraction 1 − √1 of its rest energy is set free. The innermost stable circular orbit can easily be
3
determined from Eq. (8.22). Thus a rotating black hole allows a gravitational energy conversion with
an efficiency up to ≈ 42.3%!
These considerations are of astrophysical importance since quasars in the center of galaxies are
supermassive rotating black holes.

58
Literature
• T. Fliessbach, Allgemeine Relativitätstheorie, Spektrum Verlag, 1995

• S. Weinberg, Gravitation and Cosmology, Wiley, 1972

• M. Maggiore, Gravitational Waves: Volume 1: Theory and Experiments, Oxford University

Press, 2007

• N. Straumann, General Relativity with Applications to Astrophysics, Springer Verlag, 2004

• C. Misner, K. Thorne and J. Wheeler, Gravitation, Freeman, 1973

• R. Wald, General Relativity, Chicago University Press, 1984

• B. Schutz, A first course in General Relativity, Cambridge, 1985

• R. Sachs and H. Wu, General Relativity for mathematicians, Springer Verlag, 1977

• J. Hartle, Gravity, An introduction to Einstein’s General Relativity, Addison Wesley, 2002

• H. Stephani, General Relativity, Cambridge University Press, 1990