Notes for PX436, General Relativity

Tom Marsh and Elizabeth Stanway

Updated: March 16, 2012

1
Foreword:

These notes mostly show the essentials of the lectures, i.e. what I write on
the board. The exception to the rule is when I write pieces of text like this
(outside of the examples). These represent information that I may have said
but not written during lectures. I use them when I think it would help you
follow the notes.
The notes are very terse, and brief to the point of grammatical inaccuracy.
This is because they are notes and are not intended to replace books. I make
them available in case you had to miss a lecture or find it difficult to make
notes during lectures, but if you rely on these notes only and do not read
books, you will struggle.

Lecture 1
Introduction to GR
Objectives:
Presentation of some of the background to GR
Reading: Rindler chapter 1, Weinberg chapter 1, Foster & Nightingale
introduction.

1.1

Introduction

Newtonian gravity is clearly inconsistent with Special Relativity (SR). Consider Poissons equation for the gravitational potential
2 = 4G,
= density. No time derivative = gravity instantaneous, and not a
Lorentz-invariant.

1.2

What makes gravity special?

Same problems apply to 2 = /0 from electrostatics, but full Maxwells

equations are Lorentz-invariant.
Something odd about gravity. Consider:
F = mI a,
for the force acting on a mass accelerating at rate a and
F = mG g,
2

LECTURE 1. INTRODUCTION TO GR

for the force acting on the same mass in a gravitational field g.

Why is mI = mG ? In Newtons theory this is a remarkable coincidence.

1.2.1

How remarkable?

Galileo, Newton: mI /mG same to 1 part in 103 (pendulum experiments)

Eotvos (1889): mI /mG same to 1 part in 109

View from North

celestial pole

ah

ah

Figure: Eotvoss experiment. Two masses A and B are in

balance on a beam suspended by a torsion fibre. If they have
a different ratio of inertial and gravitational mass, the horizontal component of centripetal acceleration due to Earths
rotation will cause a torque. None could be measured.
If two masses gravitationally balance, but mI /mG differs, there will be a
torque on the fibre due to the centripetal acceleration from Earths rotation.
Dicke et al (1960s): mI /mG varies by < 1 part in 1012

1.3

Inertial frames

Definition: in the absence of forces, particles move with constant velocity in

inertial frames (straight, at constant speed).
In EM neutral particles can be used to spot an inertial frame, but there are
no neutral particles in gravity. Are there inertial frames in a gravitational
field, even in thought experiments?
What defines inertial frames (as important in Special Relativity as in Newtonian gravity)?

This is why we
can talk about
the acceleration
due to gravity

LECTURE 1. INTRODUCTION TO GR

Newton: water in a bucket at the North Pole has a curved surface because
it rotates relative to the fixed stars Earth not an inertial frame.
Ernst Mach (1893): what if there were no fixed stars? Thought that Earth
would define its own inertial frame Machs Principle water surface
would be flat. Real physical consequences. e.g. expect acceleration in direction of rotation near massive rotating object, dragging of inertial frames.
No quantitative content however.
Does the weather
on Earth require
the rest of the
Universe?
1.4 Principle of Equivalence
Einstein explained mI = mG with his principle of equivalence:
The physics in a freely-falling small laboratory is that of special relativity (SR).
Equivalently, one cannot tell whether a laboratory on Earth is not actually
in a rocket accelerating at 1 g.
Has real physical content:
e.g. Predicts that light moves in a straight line at v = c in a freely-falling
laboratory. It is a locally inertial frame and gravity disappears.

Freefall lab view

Earth observers view

l

Light

g
Figure: Light sent across a freely-falling laboratory on the
right appears straight, but must appear to bend according
to an Earth-based observer since the laboratory accelerates
downwards as the light travels across it.
The light takes time
t=

l
c

to cross the lab. Therefore

1
gl2
h = gt2 = 2 .
2
2c

LECTURE 1. INTRODUCTION TO GR

e.g. l = 1 km then h = 0.055 nm on Earth, 10 m on a neutron star.

Laboratory must be small because gravity is not constant. e.g. No single
inertial frame can apply to the whole Earth.
Gravitational time dilation:

h
g

Figure: Light sent downwards in a freely-falling laboratory

will be unchanged in frequency, but an Earth-based observer
will see a higher freequency at the bottom since the lab is
moving downwards by the time the light reaches the floor.
Assume lab is dropped at same time as light leaves ceiling. Light takes time
t

h
c

to reach floor, by which time lab is moving down at speed

v=

gh
.
c

From the EP, the frequency unchanged in lab, so according to Earth observer,
the frequency at the floor is





v
gh

1 0 1 +
= 0 1 + 2 = 0 1 + 2 .
c
c
c
Clocks at ceiling run fast by factor 1 + /c2 cf floor! [read up on Pound &
Rebka experiment].
This
gravitational
time dilation is
significant for
atomic clocks on
Earth.

Lecture 2
Special Relativity I.
Objectives:
To recap some basic aspects of SR
To introduce important notation.
Reading: Schutz chapter 1; Hobson chapter 1; Rindler chapter 1.

2.1

Introduction

The equivalence principle makes Special Relativity (SR) the starting point
for GR. Familiar SR equations define much of the notation used in GR.
A defining feature of SR are the Lorentz transformations (LTs), from frame
S to S 0 which moves at v in the +ve x-direction relative to S:

vx 
(2.1)
t0 = t 2 ,
c
x0 = (x vt),
(2.2)
0
y = y,
(2.3)
0
z = z,
(2.4)
where the Lorentz factor

=

v2
1 2
c

1/2
.

(2.5)

Defining x0 = ct, x1 = x, x2 = y and x3 = z, these can be re-written more

LECTURE 2. SPECIAL RELATIVITY I.

symmetrically as
0

x0
x

10

= x0 x1 ,


= x1 x0 ,

20

(2.6)
(2.7)

x = x,
0
x3 = x3 ,

(2.8)
(2.9)

where = v/c, so = (1 2 )1/2 .

NB. The indices here are written as superscripts; do not confuse with exponents! The dashes for the new frame are applied to the indices following Schutz.
More succinctly we have
0

x =

=3
X

x ,

=0
0

for = 0, 1, 2 or 3, where the coefficients

from frame S to S 0 . Can write as a matrix:

0
0

=
0
0
1
0
0
0

represent the LT taking us

0
0
0
1

(2.10)

with 0 the row index and the column index. Better still, using Einsteins
summation convention write simply:
0

x = x .

(2.11)

NB. The summation convention here is special: summation implied only when the repeated index appears
0
once up, once down. The LT coefficients have been carefully written with a subscript to allow this. This helps keep
0
0
track of indices by making some expressions, e.g. x ,
invalid.
LT from S 0 to S is easily seen to be
0

x = 0 x ,

(2.12)

where

0 =

0 0
0 0
0
0 1 0
0
0 0 1

(2.13)

LECTURE 2. SPECIAL RELATIVITY I.

It is easily shown

0
0
0
0

that
0
0
1
0

0
0
0
1

Prove this.

0 0

0 0
0
0
1 0
0
0
0 1

1
0
0
0

0
1
0
0

0
0
1
0

0
0
0
1

Defining the Kronecker delta = 1 if = , = 0 otherwise, this equation

can be written:
0
0 = .
(2.14)
Guarantees that after LTs from S to S 0 then back to S we get x again
since
0
0 x = x = x .
Prove each step of
this equation.
Note the use of dummy index 0 to avoid a clash with or .

2.2

Nature of LTs

In SR the coefficients of the LT are constant and thus

0

x = x ,
is a linear transform, mathematically very similar to spatial rotations such
as
!
!
!
0
x
c s
x
=
,
0
y
s c
y
where c = cos , s = sin , c2 + s2 = 1. A defining feature of rotations is that
lengths are preserved, i.e.
2

l2 = (x0 ) + (y 0 ) = x2 + y 2 .
Q: What general linear transform
x0 = x + y,
y 0 = x + y,
where , , and are constants, preserves lengths?
Since




2
2
(x0 ) + (y 0 ) = 2 + 2 x2 + 2 ( + ) xy + 2 + 2 y 2 ,

LECTURE 2. SPECIAL RELATIVITY I.

then
2 + 2 = 1,
+ = 0,
2 + 2 = 1.
These are satisfied by = and = , so
x0 =
x + y,
y 0 = x + y,
with 2 + 2 = 1.
Thus the requirement to preserve length defines the linear
transform representing rotations.
The interval
s2 = (ct)2 x2 y 2 z 2 ,
plays the same role in SR.

Lecture 3
Special Relativity II.
Objectives:
Four vectors
Reading: Schutz chapter 2, Rindler chapter 5, Hobson chapter 5

3.1

The interval of SR

To cope with shifts of origin, restrict to the interval between two events
s2 = (ct2 ct1 )2 (x2 x1 )2 (y2 y1 )2 (z2 z1 )2 ,
or
s2 = c2 t2 x2 y 2 z 2 ,
or finally with infinitesimals:
ds2 = c2 dt2 dx2 dy 2 dz 2 .

(3.1)

ds2 is the same in all inertial frames. It is a Lorentz scalar. Writing

ds2 = c2 d 2 ,
defines the proper time , which is the same as the coordinate time t when
dx = dy = dz = 0. i.e. proper time is the time measured on a clock travelling
with an object.
Introducing x0 = ct, etc again, we can write
ds2 = c2 d 2 = dx dx ,
10

(3.2)

LECTURE 3. SPECIAL RELATIVITY II.

11

where

1
0
0
0
0 1
0
0
0
0 1
0
0
0
0 1

(3.3)

The interval is the SR equivalent of length corresponding to the relation for

lengths in Euclidean 3D
dl2 = dx2 + dy 2 + dz 2 .

NB There is no standard sign convention for the interval and

. Make sure you know the convention used in textbooks.

3.2

The grain of SR

The mixture of plus and minus signs in the definition of ds2 means there are
three distinct types of interval:
ds2 > 0 timelike intervals. Intervals between events on the wordlines of massive
particles are timelike.
ds2 = 0 Null intervals. Intervals between events on the wordlines of massless
particles (photons) are null.
ds2 < 0 Spacelike intervals which connect events out of causal contact.
These impose a distinct structure on spacetime.

LECTURE 3. SPECIAL RELATIVITY II.

12

ct

l
ul

ul
l

Worldline

Future
Elsewhere

Elsewhere

E
Past
x
Figure: The invariant interval of SR slices up spacetime relative to an event E into past, future and elsewhere, the
latter being the events not causally connected to E.
These so-called light-cones are preserved in GR but are distorted according
to the coordinates used.

3.3

Four-vectors

~ = (x0 , x1 , x2 , x3 ) is called
Any quantity that transforms in the same way as X
a four vector (or often just a vector). Thus V~ is defined to be a vector
if and only if
0
0
V = V .
Useful because:
Four vectors can often be identified easily
The way they transform follows from the LTs.
Lead to Lorentz scalars equivalent to ds2 .

LECTURE 3. SPECIAL RELATIVITY II.

3.3.1

13

Four-velocity

The four-velocity is one of the most important four-vectors. Consider

~
~
~
~ = lim X( + ) X( ) = dX .
U
0

d
~ is a four-vector and is a scalar, U
~ is clearly a four-vector.
Since X
From time dilation, d = dt/, so
~
~ = dX = (c, v),
U
dt
where v is the normal three-velocity and is shorthand for the spatial components of the four-velocity.

3.3.2

Scalars from four-vectors

If V~ is a four-vector, then the equivalent of the interval ds2 = dx dx is

V~ V~ = |V~ |2 = V V

(3.4)

This defines the invariant length or modulus of a four-vector. It is a

scalar under LTs.
2

This relation is fundamental. Note that |V~ |2 6= (V 0 ) +

2
2
2
(V 1 ) + (V 2 ) + (V 3 ) . SR and GR are not Euclidean.
~.
Example 3.1 Calculate the scalar equivalent to the four-velocity U
Answer 3.1 Long way

2

2

2

2
U U = U 0 U 1 U 2 U 3 ,


= 2 c2 vx2 vy2 vz2 ,


= 2 c2 v 2 ,
c2
= 2 2 = c2 .

Short way: since it is invariant, calculate its value in a frame for which
~ U
~ = c2 .
v = 0 and = 1, from which immediately U
~ U
~ = c2 is an important relation. It means that U
~ is a
U
timelike four-vector.

Lecture 4
Vectors
Objectives:
Contravariant and covariant vectors, one-forms.
Reading: Schutz chapter 3; Hobson chapter 3

4.1

Scalar or dot product

We have had
V~ V~ = V V .
~ and B
~ are four-vectors then V~ with components
If A
V = A + B ,
is also a four-vector. Therefore
V~ V~

See problems

= (A + B ) A + B

= A A + A B + B A + B B ,
~A
~+A
~B
~ +B
~ A
~+B
~ B.
~
= A
~B
~ =B
~ A,
~ so
Since is symmetric then A
~A
~ + 2A
~B
~ +B
~ B.
~
V~ V~ = A
~A
~ and B
~ B
~ are all scalars, then
Since V~ V~ , A
~B
~ = A B
A

(4.1)

is also a scalar, i.e. invariant between all inertial frames. This defines the
scalar product of two vectors.
~B
~ = 0 = A
~ and B
~ orthogonal. Null vectors are self-orthogonal.
A
14

LECTURE 4. VECTORS

4.2

15

Basis vectors

With the following basis vectors (4D versions of ~i, ~j, ~k):
~e0
~e1
~e2
~e3

=
=
=
=

(1, 0, 0, 0),
(0, 1, 0, 0),
(0, 0, 1, 0),
(0, 0, 0, 1),

we can write for frames S and S 0 :

~ = A~e = A0 ~e0 .
A
These express the frame-independent nature of any fourvector, just as we write a to represent a three-vector.
Substituting

Note that indices

are lowered on
basis vectors to fit
raised indices on
components.

A = 0 A ,
then

0 A ~e = A ~e0 ,
and re-labelling dummy indices, 0 0 , ,

0
~e0 0 ~e A = 0.
~ is arbitrary, the term in brackets must vanish, i.e.
Since A
~e0 = 0 ~e .
Comparing with

(4.2)

A = A ,
we see that the components transform oppositely to the basis vectors,
hence these are often called contravariant vectors and superscripted indices
are called contravariant indices.

4.3

Covariant vectors or one-forms

Consider the gradient = (/x0 , /x1 , /x2 , /x3 ), where is

a scalar function of the coordinates. Is it a vector?
The chain rule gives
d =

dx ,
x

LECTURE 4. VECTORS
and on differentiating wrt x

16
0

x
=
.
x0
x x0
0

But x = 0 x so
x
0

0 = 0 0 = 0 .

x
Therefore

(4.3)
0 = 0

x
x
Thus the components of the gradient do not transform like the components
of four-vectors, instead they transform like basis vectors.
Quantities like are called covariant vectors or covectors or one-forms, the latter emphasizing their difference
from vectors.
I will write one-forms with tildes such as p. Like vectors, one-forms can be
defined by their transformation, i.e. if quantities p transform as
p0 = 0 p .

(4.4)

then they are components of a one-form p.

One-forms are written with subscripted indices, also known
as covariant indices. Do not confuse with Lorentz covariance.
~ consider the quantity:
Given a one-form p and a vector A,
p A .

Because of the contra and co transformations, this is a scalar. In a

~ Thus a one-form
more frame-independent way we can write this as p(A).
is a machine that produces a scalar from a vector. Equally, a vector is a
~ p).
machine that produces a scalar from a one-form, A(
One-forms are best thought of as a series of parallel surfaces. The number of
such surfaces crossed by a vector is the scalar. One-forms cannot be thought
as arrows because they do not transform in the same way as vectors. Oneforms do not crop up in orthonormal bases (e.g. Cartesian coordinates or
because in that one case they transform
unit vectors in polar coordinates r, )
identically to vectors. They cannot be avoided in GR.

p A is one
number. Why?

LECTURE 4. VECTORS

4.4

17

Basis one-forms

A set of basis vectors ~e define a natural set of basis one-forms

:

(~e ) = ,

(4.5)

because then


~ = [p
p(A)
] A ~e ,
= p A
(~e ) ,
= p A ,
= p A ,
as required.
One can then show that basis one-forms transform like vector components,
i.e.
0
0
.

=
(4.6)

4.5

Summary of transformations

~ = A~e and one-form p = p

Given a vector A
the four transformations
are:
0

A
0

p0
~e0

=
=
=
=

A ,
0

,
0 p ,
~e .

As long as you remember that vector components have superscripted indices

and one-form components have subscripted indices, and balance free and
dummy indices properly, it should be straightforward to remember these
relations.

Lecture 5
Tensors
Objectives:
Introduction to tensors, the metric tensor, index raising and lowering
and tensor derivatives.
Reading: Schutz, chapter 3; Hobson, chapter 4; Rindler, chapter 7

5.1

Tensors

Not all physical quantities can be represented by scalars, vectors or oneforms. We will need something more flexible, and tensors fit the bill.
Tensors are machines that produce scalars when
! operating on multiple
N
vectors and one-forms. More specifically an
tensor produces a scalar
M
given N one-form and M vector arguments.
!
3
e.g. if T (
p, V~ , q, r) is a scalar then T is a
tensor.
1
!
1
Since vectors acting on one-forms produce scalars, vectors are
tensors;
0
!
!
0
0
similarly one-forms are
tensors and scalars are
tensors.
1
0

18

LECTURE 5. TENSORS

5.2

19

Tensor components

Components of a tensor in a given frame are found by feeding it basis vectors

and one-forms. e.g.
T (
, ~e ,
,
) = T .
(NB 3 up indices, 1 down matching the rank.) However, like vectors and
one-forms, T exists independently of coordinates.
It is straightforward to show that for arbitrary arguments
~ q, r) = T p A q r .
T (
p, A,
All indices are dummy, so this is a single number.
For it to be a scalar the tensor components must transform appropriately.
Using transformation properties of p , A etc, one can show that
0

T 0

0 0

= 0 T .

Extends in an obvious manner for different indices. This is often used as the
definition of tensors, similar to our definition of vectors.

5.3

Why tensors?

Consider a

1
1

!
tensor such that T (V~ , p) is a scalar. Now consider
T (V~ , ),

i.e. one unfilled slot is available for a one-form, with which it will give a
scalar = this is a vector, i.e.
~ = T (V~ , ),
W
or in component form
W = T V .
This is one reason why tensors appear in physics, e.g. to relate D to E in
EM, or stress to strain in solids. More importantly:
Tensors allow us to express mathematically the frameinvariance of physical laws. If S and T are tensors and
S = T is true in one frame, then it is true in all frames.

LECTURE 5. TENSORS

5.4

20

The metric tensor

Recall the scalar product

~B
~ = A B .
A
~B
~ is a scalar while A
~ and B
~ are vectors. is therefore a
A

0
2

!
tensor

producing a scalar given two vector arguments:



~
~
~
~
A B = A, B .
are thus components of a tensor, the metric tensor.

5.4.1

Index raising and lowering

The metric tensor arises directly from the physics of spacetime. This gives
it a special place in associating vectors and one-forms. Consider as before
an unfilled slot, this time with :
~ ).
(A,
Fed a vector, this returns a scalar, so it is a one-form. We define this as the
~
one-form equivalent to the vector A:
~ ),
A = (A,
or in component form
A = A .
Thus can be used to lower indices, as in
T = T ,
or
T = T .
If we define by
= ,
then applying it to an arbitrary one-form
A =

In SR = .

A ,

= ( ) A ,
= A ,
= A ,
so it raises indices.
The metric tensor in its covariant and contravariant forms,
and , can be used to switch between one-forms and
vectors and to lower or raise any given index of a tensor.

e.g. /x is
a gradient vector.

LECTURE 5. TENSORS

5.5

21

Derivatives of tensors

Derivatives of scalars, such as /x = give one-forms but what about

derivatives of vectors, V /x ?
Work out how they transform:
0

V = V
thus
0

V
x0

because the

h 0 i
V ,
x0

0 V
= 0 ,
x
=

are constant in SR (but not in GR!).

Using the chain-rule

x

=
,
x0
x0 x
and as in the last lecture

Therefore

x
= 0 .
x0
0

V
V
0

0
=

0
x
x
!
1
This is the transformation rule of a
tensor. Key point:
1
The derivatives of tensors are also tensors we dont need to
introduce a new type of quantity phew!

Lecture 6
Stressenergy tensor
Objectives:
To introduce the stressenergy tensor
Conservation laws in relativity
Reading: Schutz chapter 4; Hobson, chapter 8; Rindler, chapter 7.

6.1

Numberflux vector

Consider a cloud of particles (dust) at rest in frame S0 , the instantaneous

rest frame or IRF with number density n0 .
Lorentz contraction means that a cube dx0 , dy0 , dz0 in S0 transforms to
dx = dx0 /, dy = dy0 , dz = dz0 in a frame S in which the particles move,
while particle numbers are conserved, so in S the particle density n is given
by
n = n0 .
n is not a scalar or a four-vector and so cannot be part of form-invariant
relations. Consider instead
~ = n0 U
~.
N
This is a four-vector because
~ = (c, v) is a four-vector
The four velocity U
n0 is a scalar (defined in the IRF so all observers agree on it).

22

LECTURE 6. STRESSENERGY TENSOR

23

The time component N 0 = n0 c = nc gives the number density. The spatial

components N i = n0 v i = nv i , i = 1, 2, 3 are the fluxes (particles/unit
area/unit time) across surfaces of constant x, y and z.
Even N 0 is a flux across a surface, a surface of constant time:
Sketch this:

ct

C
(ct)
A
B

x
x

Figure: World lines of dust particles travelling at speed v

in the x-direction crossing surfaces of constant t (AB) and
constant x (BC).
Worldlines crossing CB represent the flux across constant x, N 1 = nv
Same worldlines crossing AB represent flux across constant t. Scaling by
ratio of sides of triangle we get a flux:
N1

CB
(ct)
c
= N1
= N 1 = N 0,
AB
x
v

so N 0 is the particle flux across a surface of constant time.

6.2

Conservation of particle numbers

N
~ ) (one-form
acting on N
~ ). Written out in full:
Consider the scalar (

N
~ ) = N ,
(
x
N 0 N 1 N 2 N 3
=
+
+
+
,
x0
x1
x2
x3
nc nvx nvy nvz
=
+
+
+
.
ct
x
y
z

LECTURE 6. STRESSENERGY TENSOR

24

This can be written as

n
+ (nv).
t
Compare with the continuity equation of fluid mechanics:

+ (v) = 0,
t
based on (Newtonian) conservation of mass . = if particles are conserved:
n
+ (nv) = 0.
t
Thus conservation of particle numbers can be expressed as:

N
~ ) = N = N = N , = 0,
(
x

(6.1)

introducing the short-hand = /x , and the even shorter-hand comma notation

for derivatives.

6.3

Stressenergy tensor

If the mass density in the IRF is 0 , then due to Lorentz contraction and
relativistic mass increase, in any other frame it becomes:
= 2 0 ,
Now consider
T = 0 U U ,
then since U 0 = c,
T 00 = 2 0 c2 = c2 .
From E = mc2 , T 00 must therefore be the energy density.
T is a tensor because
~ is a four-vector
The four velocity U
0 is a scalar (defined in the IRF)
T is called the stressenergy tensor.

LECTURE 6. STRESSENERGY TENSOR

6.3.1

25

Physical meaning

T is the flux of the -th component of four-momentum across a surface of

constant x , so:
T 00 = flux of 0-th component of four-momentum (energy) across the
time surface (cf N 0 ) = energy density
T 0i = T i0 = energy flux across surface of constant xi (heat conduction
in IRF)
T ij = flux of i-momentum across j surface = stress.

6.4

Perfect fluids

Definition: a perfect fluid has (i) no heat conduction and (ii) no viscosity.
In the IRF (i) implies T 0i = T i0 = 0, while (ii) implies T ij = 0 if i 6= j.
For T ij to be diagonal for any orientation of axes = T ij = p0 ij where p0
is the pressure in the IRF. Therefore in the IRF:

T =

0 c2 0 0 0
0 p0 0 0
0
0 p0 0
0
0 0 p0

Convince yourself
of this.

But this can be written:


p0 
T = 0 + 2 U U p0 ,
c
and since all terms are tensors, this is true in any frame remembering that The sign of the p0
0 and p0 are defined in the IRF.
term can vary
according to
Just as conservation of particles implies N , = 0, so energymomentum convention
conservation gives
adopted for
T

T
=
= 0.
x

This equation plays a key role in GR where the stressenergy

tensor replaces the simple density, , of Newtonian gravity.

Lecture 7
Generalised Coordinates
Objectives:
Generalised coordinates
Transformations between coordinates
Reading: Schutz, 5 and 6; Hobson, 2; Rindler, 8.
Consider the following situation:

Figure: A freely falling laboratory with two small masses

floating within it.

The masses can

Lab falls freely with two small masses within it. The masses accelerate be made as small
towards centre of mass M . Therefore they will end up moving towards each as one likes, so
their movement is
other.
not because of
Equivalence principle says SR in a small freely-falling lab, but clearly not their mutual
true over large region.
gravitational
attraction.
Einsteins remarkable insight was that this was similar to the following:
26

LECTURE 7. GENERALISED COORDINATES

27

Equator

Figure: Two people set off due North from the equator on
Earth.
Two people at Earths equator travel due North, i.e. parallel to each other.
Although they stick to straight paths, they find that they move towards
each other, and ultimately meet at the North pole.
Einstein replaced Newtonian gravity by the curvature of
spacetime. Although particles travel in straight lines in spacetime, the warping of spacetime by large masses can cause initially parallel paths to converge. There is no gravitational
force in GR!

7.1

Coordinates

We have to be able to cope with general coordinates covering potentially

curved spaces = differential geometry developed by Gauss, Riemann and
many others.
Start by defining a set of coordinates covering an N -dimensional space
(manifold) by x1 , x2 , x3 , . . . xN . [Temporary suspension of 0 index to
avoid N 1 everywhere.]

7.2

Curves

A curve can be defined by the N parametric equations

x = x (),

LECTURE 7. GENERALISED COORDINATES

28

for each , where is a parameter marking position along the curve. e.g.
x = , y = 2 is a parabola in 2D. independent of coordinates = scalar.

Figure: A curve parameterised by parameter .

7.3

Coordinate transforms

Coordinates can always be re-labelled:

0

x = x (x1 , x2 , . . . , x , . . . xN ),
0

or x = x (x ) for short. This is a coordinate transformation.

Example 7.1 In Euclidean 2D

1/2
r = x2 + y 2
,
1
= cos (x/r),
transforms from Cartesian to polar coordinates.
Recall the SR equation:

x = x .
Compare with:

dx =

x
dx ,
x
0

then the N N partial derivatives x /x define the transformation matrix:

0
0
0
x1 /x1 x1 /x2 . . . x1 /xN

0
0
0
x2 /x1 x2 /x2 . . . x2 /xN
,
L=
..
..
..
..

.
.
.
.

N0
1
N0
2
N0
N
x /x x /x . . . x /x

LECTURE 7. GENERALISED COORDINATES

29
0

a generalisation of the LT matrix . The L are not constant unlike

in SR; the transformation also only applies to infinitesimal displacements.
0

Good news: With x /x instead of , the transformation formulae for vectors, one-forms and tensors are otherwise
unchanged.

7.4

The general metric tensor

In a freely-falling frame (SR), let coordinates be w , so the interval is

ds2 = dw dw .
Replacing w with x using
dw =

w
w

dx
and
dw
=
dx ,
x
x

avoiding clashing indices, gives

ds2 =

w w
dx dx .
x x

Setting
w w
g = ,
x x
we therefore have the very important relation
ds2 = g dx dx .

(7.1)

g is the generalised version of the SR metric tensor and replaces it.

~ satisfies
e.g. In general coordinates, the four-velocity U
g U U = c2 .

(7.2)

The first part of the transition from SR to GR is to replace

every occurrence of by g .
e.g. index raising lowering A = A becomes A = g A . g is symmetric but not necessarily diagonal; is a special case. Similarly =
becomes g g = , so the up coefficients come from the matrix-inverse
of the down ones.

Lecture 8
Metrics
Objectives:
More on the metric and how it transforms.
Reading: Hobson, 2.

8.1

Riemannian Geometry

The interval
ds2 = g dx dx ,
is a quadratic function of the coordinate differentials.
This is the definition of Riemannian geometry, or more correctly, pseudo-Riemannian
geometry to allow for ds2 < 0.
Example 8.1 What are the coefficients of the metric tensor in 3D Euclidean
space for Cartesian, cylindrical polar and spherical polar coordinates?
Answer 8.1 The interval in Euclidean geometry can be written in Cartesian coordinates as
Introducing an
2
2
2
2
ds = dx + dy + dz .
obvious notation
with x standing
The metric tensors coefficients are therefore given by
for the x
coordinate index,
gxx = gyy = gzz = 1,
etc.
with all others = 0.
In cylindrical polars:
ds2 = dr2 + r2 d2 + dz 2 ,
30

LECTURE 8. METRICS

31

so grr = 1, g = r2 , gzz = 1 and all others = 0.

Finally spherical polars:
ds2 = dr2 + r2 d2 + r2 sin2 d2 ,
gives grr = 1, g = r2 and g = r2 sin2 .
Example 8.2 Calculate the metric tensor in 3D Euclidean space for the
coordinates u = x + 2y, v = x y, w = z.
Answer 8.2 The inverse transform is easily shown to be x = (u + 2v)/3,
y = (u v)/3, z = w, so
1
du +
3
1
dy =
du
3
dz = dw,

dx =

2
dv,
3
1
dv,
3

so
ds

2 
2
2
1
1
1
du + dv +
du dv + dw2 ,
=
3
3
3
3
2 2 5 2 2
=
du + dv + dudv + dw2 .
9
9
9


We can immediately write guu = 2/9, gvv = 5/9, gww = 1, and guv = gvu =
1/9 since the metric is symmetric. This metric still describes 3D Euclidean
flat geometry, although not obviously.

8.2

Metric transforms

The method of the example is often the easiest way to transform metrics,
however using tensor transformations, we can write more compactly:
g0 0 =

x x
g .
x0 x 0

This shows how the components of the metric tensor transform under coordinate transformations but the underlying geometry does not change.
Example 8.3 Use the transformation of g to derive the metric components
in cylindrical polars, starting from Cartesian coordinates.

LECTURE 8. METRICS

32

Answer 8.3 We must compute terms like x/r, so we need x, y and z in

terms of r, , z:
x = r cos ,
y = r sin ,
z = z.
Find x/r = cos , y/r = sin , z/r = 0. Consider the grr component:
grr =

xi xj
gij ,
r r

where i and j represent x, y or z. Since gij = 1 for i = j and 0 otherwise,

and since z/r = 0, we are left with:
2  2
x
y
=
+
r
r
2
2
= cos + sin = 1.


grr

Similarly

g =

2


+

2

= (r sin )2 + (r cos )2 = r2 ,

and gzz = 1, as expected.

This may seem a very difficult way to deduce a familiar result, but the point is
that it transforms a problem for which one otherwise needs to apply intuition
and 3D visualisation into a mechanical procedure that is not difficult at
least in principle and can even be programmed into a computer.

8.3

First curved-space metric

We can now start to look at curved spaces. A very helpful one is the surface
of a sphere.

LECTURE 8. METRICS

33

Figure: Surface of a sphere parameterised by distance r from

a point and azimuthal angle
The sketch shows
Two coordinates are needed to label the surface. e.g. the distance from a the surface
point along the surface, r, and the azimuthal angle , similar to Euclidean embedded in
3D. This is a
polar coords.
priviledged view
The distance AP is given by R sin , so a change d corresponds to distance that is not always
R sin d. Thus the metric is
possible. You
need to try to
ds2 = dr2 + R2 sin2 d2 .
imagine that you
or since r = R,
are actually stuck
r
2
2
2
2
2
in the surface
ds = dr + R sin
d .
R
with no height
This is the metric of a 2D space of constant curvature.
dimension.
Circumference of circle in this geometry: set dr = 0, integrate over
r
C = 2R sin < 2r.
R
e.g. On Earth (R = 6370 km), circle with r = 10 km shorter by 2.6 cm than
if Earth was flat.
Exactly the same is possible in 3D. i.e we could find that a circle radius r
has a circumference < 2r owing to gravitationally induced curvature.

8.4

2D spaces of constant curvature

Can construct metric of the surface of a sphere as follows. First write the
equation of a sphere in Euclidean 3D
x2 + y 2 + z 2 = R 2 .

LECTURE 8. METRICS

34

If we switch to polars (r, ) in the xy plane, this becomes

r 2 + z 2 = R2 .
In the same terms the Euclidean metric is
dl2 = dr2 + r2 d2 + dz 2 .
But we can use the restriction to a sphere to eliminate dz which implies
2r dr + 2z dz = 0,
and so
dl2 = dr2 + r2 d2 +
which reduces to
dl2 =

r2 dr2
,
z2

dr2
+ r2 d2 .
1 r2 /R2

Defining curvature k = 1/R2 , we have

dl2 =

dr2
+ r2 d2 ,
1 kr2

the metric of a 2D space of constant curvature. k > 0 can be embedded

in 3D as the surface of a sphere; k < 0 cannot, but it is still a perfectly valid
geometry. [A saddle shape has negative curvature over a limited region.]
A very similar procedure can be used to construct the spatial part of the
metric describing the Universe.

Lecture 9
The connection
Objectives:
The connection
Reading: Schutz 5; Hobson 3; Rindler 10.
Apart from the change from to its more general counterpart, g , we have
not had to change much in moving from SR to more general coordinates, but
this comes to an end when we look again at derivatives.

9.1

Covariant derivatives of vectors

We showed that V /x are components of a tensor in SR; this is not true

in GR. Consider the derivative of V~ = V ~e :
V~
V
~e
=
~e + V .

x
x
x
~e /x , the change in a vector is still a vector, and hence can be expanded
over the basis:
~e
= ~e
(9.1)

x
where the are a set of coefficients dependent upon position. They are
called variously the connection coefficients or Christoffel symbols. This
equation defines the coefficients .
Sometimes
Christoffel
Swapping indices and , we can write
symbols of the


second kind
V~
V

~e .
=
+ V
(9.2)
x
x
35

LECTURE 9. THE CONNECTION

36

The derivative of a vector must be a tensor, so

V
+ V ,
x
are the components of a tensor, called the covariant derivative, written in
frame-independent notation as V~ with components
V = V + V .

(9.3)

V ; = V , + V ,

(9.4)

or equivalently
introducing the semi-colon notation to represent the covariant derivative.
The final notation has the advantage that the index is last in every term.
Otherwise, try to remember that whichever component you take the derivative
with respect to goes last on the connection coefficients.
The two terms V and V do not transform as tensors, only their
sum does; in SR V are tensor components while = 0.
V comes from the change of components with position, V comes
from the change of basis vectors with position.
Example 9.1 Calculate the connection coefficients in Euclidean polar coordinates r, .
Answer 9.1 Start from Cartesian basis vectors ~ex and ~ey . Using the transformation rule for basis vectors:
~e0 =
we have
~er =

x
~e ,
x0

x
y
~ex +
~ey ,
r
r

and since x = r cos , y = r sin ,

~er = cos ~ex + sin ~ey .
Similarly
~e = r sin ~ex + r cos ~ey .
Prove this.

LECTURE 9. THE CONNECTION

37

Taking derivatives, remembering that the Cartesian vectors are constant, we

have
~er
r
~er

~e
r
~e

= 0,
= sin ~ex + cos ~ey ,
= sin ~ex + cos ~ey ,
= r cos ~ex r sin ~ey ,

which we can re-write as

~er
r
~er

~e
r
~e

= 0,
1
~e ,
r
1
= ~e ,
r
=

= r~er .

Hence the Christoffel symbols are r = r = 1/r, r = r, and r rr =

rr = r r = r r = = 0.
Note that the final set of relations does not involve Cartesian vectors. The
Christoffel symbols allow one to work in complex coordinate systems without
reference to Cartesian coordinates, and to derive such well-known formulae
such as the Laplacian in spherical coordinates see Schutz or Hobson for
this.
The way we calculated the connection above is tedious and indirect, but there
is a better way.

9.2

The Levi-Civita Connection

One can show that

See handout ??
1
= g (g, + g, g, ) ,
2

which is known as the Levi-Civita connection and shows that the connection can be calculated from the metric alone without recourse to Cartesian
coordinates.

LECTURE 9. THE CONNECTION

38

Example 9.2 Calculate the connection coefficients in polar coordinates (r, ).

Answer 9.2 The metric is ds2 = dr2 + r2 d2 , so grr = g rr = 1, g = r2 ,
g = 1/r2 , while all gr = 0.
Thus
1
g (gr, + g,r gr, ) ,
2
1
=
g g,r ,
2
11
=
2r,
2 r2
1
.
=
r
This agrees with the value found earlier, and although algebraically tricky, is
more straightforward.
r =

9.3

Covariant derivatives of one-forms

What is the equivalent for one-forms of

V ; = V , + V ?
Consider the scalar = p V , then , is a tensor and
, = p V , + p, V .
Writing
, = p (V , + V ) + (p, p ) V ,
or
, = p V ; + (p, p ) V .
All terms outside brackets are tensors and therefore
p; = p, p ,
is a tensor, the covariant derivative of the one-form.
These results generalise to general tensors, e.g.
T ; = T , + T + T T T
i.e one +ve term for each contravariant index, one ve term for each covariant one, derivative index always last on connection.
This chapter/lecture has introduced the important concept of the covariant
derivative which allows us to write frame-invariant tensor derivatives in
GR.

Lecture 10
Parallel transport
Objectives:
Parallel transport
Geodesics
Equations of motion
Reading: Schutz 6; Hobson 3; Rindler 10.
In this lecture we are finally going to see how the metric determines the
motion of particles. First we discuss the concept of parallel transport.

10.1

Parallel transport

In SR, the equation for force-free motion of a particle is

~
~ = dU = 0,
A
d
~
i.e a straight line through spacetime as well as 3D space with the vector U
remaining constant along the line parameterised by .
This is extended to the curved spacetime of GR by the notion of parallel
transport in which a vector is moved along a curve staying parallel to
itself and of constant magnitude.

39

LECTURE 10. PARALLEL TRANSPORT

40

Figure: Parallel transport of a vector from A to B, keeping

it parallel to itself and of constant length at all points.
Consider the change of a vector V~ = V e~ along a line parameterised by
d~e
dV~
dV
=
~e + V
.
d
d
d
We can write

d~e
~e dx
=
.
d
x d
Using this and the definition of the connection
~e
= ~e ,

x
gives
dV~
dV
dx

=
~e + V
~e .
d
d
d
Swapping dummy indices and in the second term finally leads to
dV~
=
d


dV
dx

+
V
~e .
d
d

This is a vector with components

DV
dx
dV
=
+
V ,
D
d
d
and is known variously as the intrinsic, absolute or total derivative.
One also sometimes sees the vector written as
dV~
= U~ V~ ,
d
where U = dx /d is the tangent vector pointing along the line (= fourvelocity if = ).

LECTURE 10. PARALLEL TRANSPORT

41

The components are very similar to the covariant derivative

V ; = V , + V .
In fact if we write

(a cheat: V

dV
V dx
V
=
=
U ,

d
x d
x
might only be defined on the line) then we can write

DV /D is to
dV /d as V ; is
to V , .

DV
= V ; U .
D
Parallel transport: if a vector V~ is parallel transported along a line then
dV~
= 0,
U~ V~ =
d
or in component form:
DV
dV
dx
=
+
V = 0.
D
d
d

10.2

Straight lines or geodesics

With parallel transport we can extend the idea of straight lines to curved
spaces:
Definition: a line is straight if it parallel transports its own
tangent vector.
~ = 0 or,
In other words straight lines in curved spaces are defined by U~ U

setting V = U = dx /d
d2 x
dx dx

= 0.

d2
d d
More compactly
x + x x = 0,
using the dot notation for derivatives wrt .
These are force-free equations of motion
~ = dU
~ /d = 0 to GR.
Extends SR A

Shows how V
must change for
V~ to remain
constant.

LECTURE 10. PARALLEL TRANSPORT

42

In GR, gravity is not a force but a distortion of spacetime

Metric g particle motion.
Straight lines are often called geodesics. Great circles are geodesics
on spheres.

10.2.1

Affine parameters

~ = kU
~ , i.e. the tangent vector
We could have defined straight by U~ U
changes by a vector parallel to itself. However in such cases one can always
transform to a new parameter, say = (), such that U~ 0 U~ 0 = 0, where
U~ 0 is the new tangent vector. is then called an affine parameter. Proper
time is affine for massive particles.
I will always
assume affine
parameters.

10.3

Example: motion under a central force

Consider motion under Newtonian gravity

GM
dV~
= 2 ~r.
dt
r
In general coordinates the left-hand side is
dV
+ V V .
dt

In polar coordinates V~ = (r,

).
From last time r = r, r = r = 1/r with all others = 0. Therefore:
GM
dV r
+ r V V = 2 ,
dt
r
and

dV
+ r V r V + r V V r = 0.
dt

These give
GM
r r2 = 2 ,
r
and

2
+ r = 0.
r
The second can be integrated to give the well known conservation of angular
momentum r2 = h.

LECTURE 10. PARALLEL TRANSPORT

43

These two equations are the equations of planetary motion which lead to
ellipses and Keplers laws. The point here is how the connection allows one
to cope with familiar equations in awkward coordinates. In much of physics
such coordinates can be avoided, but not in GR where there is no sidestepping
the connection. Note here how the centrifugal term, r2 , appears via the
connection.

Lecture 11
Geodesics
longish lecture
Objectives:
Variational approach to geodesics
Reading: Schutz, 5, 6 & 7; Hobson 5, 7; Rindler 9, 10

11.1

Extremal Paths

Straight lines are also the shortest. In GR path length is

Z
Z q
g dx dx .
S = ds =
Parameterising by :

In GR S is
actually
maximum for
straight paths, as
a consequence of
the minus signs in
the metric.

Z r
dx dx
S=
g
d.
d d
Minimisation of S is a variational problem solvable with the Euler-Lagrange
equations:
See handout ??


d L
L
= 0,
dt x
x
where x = dx /d and the Lagrangian is
r
ds
dx dx
L=
= g
.
d
d d
The square root is inconvenient; consider instead using L0 = (L)2 as the
Lagrangian. Then the Euler-Lagrange equations would be


d
L
L
2L 2L = 0.
d
x
x
44

LECTURE 11. GEODESICS

Now if satisfies
then

45

ds
= L = constant,
d
 


d
L
L
2L
= 0,
d x
x

so

dx dx
,
d d
leads to the same equations as L provided is chosen so that ds/d is
constant (L works for any ).
L0 = (L)2 = g

The constraint on is another way to define affine parameters.

ds/d = c, the speed of light, a constant, proper time is affine.

Since

But remember,
proper time
Can show that Euler-Lagrange equations are equivalent to equations of mo- cannot be used
tion derived before, i.e.
for photons.
x + x x = 0.

11.2

Why use the Lagrangian approach?

Application of the Euler-Lagrange equations is often easier than calculating

the 40 coefficients of the Levi-Civita connection.
Example 11.1 Calculate the equations of motion for the Schwarzschild metric




2GM
dr2
2
2
ds = c 1 2
dt2
r2 d2 + sin2 d2 ,
2
cr
1 2GM/c r
using the Euler-Lagrange approach.
dr r,
Answer 11.1 Setting dt t,
d and d in ds2 , the
Lagrangian is given by




r 2
2GM 2
2 2
2
2
2 .
t

+
sin

r
L=c 1 2
cr
1 2GM/c2 r
Consider, say, the component of the E-L equations:
 
d L
L

= 0.
d

This gives

d 
2r2 + 2r2 sin cos 2 = 0,
d
much more directly than the connection approach.

LECTURE 11. GEODESICS

11.3

46

Conserved quantities

If L does not depend explicitly on a coordinate x say, then L/x = 0,

and so the E-L equations show that
L
= 2g x = 2x = constant.
x
In other words the covariant component of the corresponding velocity is
conserved.
e.g. The metric of the example does not depend upon so
r2 sin2 () = constant.
When motion confined to equatorial plane = /2, r2 = h, a constant:
GR equivalent of angular momentum conservation.

11.4

Slow motion in a weak field

Consider equations of motion at slow speeds in weak, slowly varying (spatially) fields (Newtonian case). Mathematically x i 0 for i = 1, 2 or 3, and
g = + h where |h |  1 and |h, |  1. The equations of motion
x + x x = 0,
reduce to
x + 00 x 0 x 0 = 0.
The time velocities x 0 are never negligible, and in fact for = , are
d(ct)/d c.
The derivative terms in the Levi-Civita equations are first order in h, so
retaining only terms first order in h we can write
1
00 = (g0,0 + g0,0 g00, ).
2
If the metric is stationary, all time derivatives (, 0 terms) are zero, and so,
remembering that 00 = +1,
1
0 00 = (g00,0 + g00,0 g00,0 ) = 0.
2
Therefore x0 = 0 or
x 0 = c

dt
= constant.
d

LECTURE 11. GEODESICS

47

Since ii = 1 for each i, the spatial components become

1
1
i 00 = (gi0,0 + g0i,0 g00,i ) = g00,i ,
2
2
after dropping time derivatives, hence
xi =

d 2 xi
1
= g00,i x 0 x 0 .
2
d
2

Since x 0 = cdt/d is constant, we finally obtain

d 2 xi
1 2
c g00,i ,
=

dt2
2
or equivalently
1
r = c2 g00 .
2
(dots now derivatives wrt t not ). What is g00 ? Consider a clock at rest:
ds2 = c2 d 2 = g00 c2 dt2 .
But from the equivalence principle



d = 1 + 2 dt,
c
where is the Newtonian potential ( < 0 so d < dt: grav. time dilation).
Thus
2

2

=1+ 2,
g00 = 1 + 2
c
c
at the level of approximation we are using here. Therefore:
r = ,
the equation of motion in Newtonian gravity! is the Newtonian equivalent
to the g00 component of the metric. At slow speeds in weak fields, none of
the other 9 components of the metric matter.
This finally completes the loop of establishing that motion in a curved spacetime can give rise to what until now we have called the force of gravity. On
Earth h00 109 . It is amazing that so tiny a wrinkle of spacetime leads
to the phenomenon of gravity. We must next see how mass determines the
metric.

Lecture 12
Curvature
Objectives:
Curvature and geodesic deviation
Reading: Schutz, 6; Hobson 7; Rindler 10.

12.1

Local inertial coordinates

The metric determines particle motion, and Newtons Law of Gravity, 2 =

4G, suggests that mass must fix the metric. Thus we seek a tensor built
from the metric and/or its derivatives that can substitute for 2 in Newtons theory.
g alone is no good because coordinates can always be found such that
g = , the Minkowski mertic. This clearly cannot simultaneously describe situations with and without mass.
Proof: there are 10 independent coefficients of g but 16 degrees of freedom
0
in the transformation matrix, x /x .
The first derivatives g /x = g, are not enough either, because it can
be shown that coordinates can always be found in which
g, = 0.
In these coordinates, the Levi-Civita equation implies
= 0,
so that A = dU /d = 0. These are locally inertial or geodesic coordinates,
the freely-falling frames of the equivalence principle.
48

LECTURE 12. CURVATURE

49

Corollary: in an inertial frame, covariant derivative ordinary partial derivative =

g; = g, = 0,
g; = 0 is tensorial, so the metric is covariantly constant, g = 0.
Conclusion: we need a tensor involving at least second derivatives of the
metric, as suggested by 2 and g00 1 + 2/c2 .

12.2

Curvature tensor

Consider the expression

V = V; ,
where V is an arbitrary one-form. This is a tensor (derivatives are covariant)
which contains second derivatives of the metric. Expanding the covariant
derivative with respect to :
V; = [V; ]; ,
= V;, V; V; .
Each of the three covariant derivatives, V; etc, can be expanded similarly
and one ends up with an expression of the form
see handout 5
V; = [. . .]V, + [. . .]V, + [. . .]V .
The terms in brackets involve second derivatives of g. Unfortunately although the sum is a tensor, we cannot assert that the individual terms are
tensors: we need just one term involving V alone.
If instead we consider the tensor V; V; , the derivatives in V cancel
and we find
Where the square
brackets indicate
V; V; = [ , ] V = R V .
a commutator
where R is the Riemann curvature tensor and is given by
R = , , + .
In flat spacetime, one can find a coordinate system in which the connection
and its derivatives = 0, and so
R = 0.
i.e. the Riemann tensor vanishes in flat spacetime. (i.e. covariant differentiation is commutative in flat space.)

Do not try to
memorise this!!

LECTURE 12. CURVATURE

12.3

50

Understanding the curvature tensor

Pictorially the relation

V; V; = R V ,
corresponds to the following:

Figure: Vector parallel transported two ways around the

same loop does not match up at the end if there is curvature
Vector V~ is first parallel transported A C D, associated with V ; .
Then the same vector is taken A B D, associated with V ; . Curvature causes the vectors at D to differ.
Related to this, a vector parallel-transported around a loop in a curved space
changes, e.g.

Figure: Vector parallel transported on a sphere A to B to

C to A has changed by the time it gets back to A.

LECTURE 12. CURVATURE

12.4

51

Geodesic Deviation

Figure: Two nearby geodesics deviate from each other because of curvature
Consider the relative distance w
~ between two nearby particles at P and Q
undergoing geodesic motion (free-fall). Can show that
D2 w
+ R x x w = 0,
D2
where x = dx /d etc. This is a tensor equation, the equation of geodesic deviation.
Here the capital Ds indicate absolute or total derivatives, i.e. derivatives
that allow for variations in components caused purely by curved coordinates,
so that we expect
D2 w
= 0,
D2
in the absence of gravity.
The second term therefore represents the effect of gravity that is not removed
by free-fall, i.e. it is the tidal acceleration. In Newtonian physics tides are
caused by a variation in the gravitational field, g, and since g = , tides
are related to 2 . This is another indication of the connection between
curvature and the left-hand side of 2 = 4G.
This is the quantitative version of the notion from chapter 7 of two particles
falling towards a gravitating mass moving on initially parallel-paths in spacetime which remain straight and yet ultimately meet.

Lecture 13
Einsteins field equations
Objectives:
The GR field equations
Reading: Schutz, 6; Hobson 7; Rindler 10.

13.1

Symmetries of the curvature tensor

With 4 indices, the curvature tensor has a forbidding 256 components. Luckily several symmetries reduce these substantially. These are best seen in fully
covariant form:
R = g R ,
for which symmetries such as
R = R ,
and
R = R .

swaps order of
covarient
derivatives

can be proved. These relations reduce the number of independent components to 20.
Handout 6
These symmetries also mean that there is only one independent contraction
R = R ,
because others are either zero, e.g.
R = g R = 0,
52

LECTURE 13. EINSTEINS FIELD EQUATIONS

53

or the same to a factor of 1. R is called the Ricci tensor, while its

contraction
R = g R = R ,
is called the Ricci scalar.

13.2

The field equations

NB Signs vary
between books. I
follow Hobson et
al and Rindler.

We seek a relativistic version of the Newtonian equation

2 = 4G.
The relativistic analogue of the density is the stressenergy tensor T .
is closely related to the metric, and 2 suggests that we look for some
tensor involving
the second derivatives of the metric, g, , which should be
!
2
a
tensor like T .
0
The contravariant form of the Ricci tensor satisfies these conditions, suggesting the following:
R = kT ,
where k is some constant. (NB both R and T are symmetric.)
However, in SR T satisfies the conservation equations T , = 0 which in
GR become
T ; = 0,
whereas it turns out that
1
R ; = R, g 6= 0,
2
where R is the Ricci scalar. Therefore R = kT cannot be right.
Fix by defining a new tensor, the Einstein tensor
1
G = R Rg ,
2
because then



1
1
1

= R Rg
= R ; R; g Rg ; = 0,
2
2
2
;

since g = 0 and R; = R, . Therefore we modify the equations to

1
R Rg = kT .
2
These are Einsteins field equations.

Handout ??

LECTURE 13. EINSTEINS FIELD EQUATIONS

13.3

54

The Newtonian limit

The equations must reduce to 2 = 4G in the case of slow motion in weak

fields. To show this, it is easier to work with an alternate form: contracting
the field equations with g then
1
g R Rg g = kg T ,
2
and remembering the definition of R and defining T = g T ,
1
R R = R = kT,
2
since = 4. Therefore
R



1

= k T Tg
.
2

Easier still is the covariant form:




1
= k T T g .
2

The stressenergy tensor is

T

p
= + 2 U U pg .
c


In the Newtonian case, p/c2  , and so

T U U .
Therefore
T = g T = g U U = c2 .
Weak fields imply g , so g00 1. For slow motion, U i  U 0 c, Only T00
and so U0 = g0 U g00 U 0 c too. Thus
significant
T00 c2 ,
is the only significant component.
The 00 cpt of R is:
R00 = 0,0 00, + 0 0 00 .
All are small, so the last two terms are negligible. Then assuming timeindependence,
R00 i 00,i .

LECTURE 13. EINSTEINS FIELD EQUATIONS

55

But, from the lecture on geodesics,

i 00 =

,i
.
c2

Thus

1 2
1
1
R00 2 ,ii = 2 i i = 2 2 .
c
c x x
c
Finally, substituting in the field equations


1 2
1 2
2
2 = k c c ,
c
2
or

kc4
.
2
Therefore if k = 8G/c4 , we get the Newtonian equation as required, and
the field equations become
2 =

8G
1
R Rg = 4 T .
2
c
Key points:
The field equations are second order, non-linear differential equations
for the metric
10 independent equations replace 2 = 4G
By design they satisfy the energy-momentum conservation relations
T ; = 0
The constant 8G/c4 gives the correct Newtonian limit
Although derived from strong theoretical arguments, like any physical
theory, they can only be tested by experiment.

No longer
balancing
up/down indices
since we are
referring to
spatial
components only
in nearly-flat
space-time.

Lecture 14
Schwarzschild geometry
Objectives:
Schwarzschilds solution
Reading: Schutz, 10; Hobson 9; Rindler 11; Foster & Nightingale 3.

14.1

Isotropic metrics

It is hard to solve the field equations. Symmetry arguments are essential.

The first such solution to the field equations was derived by Schwarzschild
in 1916 for spherical symmetry.
Consider first the Minkowski interval


ds2 = c2 dt2 dr2 r2 d2 + sin2 d2 .
The term in brackets expresses spherical symmetry or isotropy (no preference
for any direction). Any spherically symmetric metric must have a term of
this form. Thus a general isotropic metric can be written


ds2 = A dt2 B dt dr C dr2 D d2 + sin2 d2 .
c.f. Kerr metric
which has ddt
Expect symmetry under , so no cross terms with
dr d or d dt.
A, B, C and D cannot depend on or otherwise isotropy is broken
= functions of r and t only.

56

LECTURE 14. SCHWARZSCHILD GEOMETRY

57

We can define a new radial coordinate r0 such that (r0 )2 = D, and so the
metric becomes


ds2 = A0 dt2 B 0 dt dr0 C 0 (dr0 )2 (r0 )2 d2 + sin2 d2 .
This metric is still general.
Dropping the primes, with this radial coordinate, the area of a sphere is still
4r2 , but r is not necessarily the ruler distance from the origin.
Finally we can transform the time coordinate using
dt = f dt0 + g dr,
choosing f and g such that dt is an exact differential and so that the cross
terms in dr dt0 cancel. We are left with
Dropping primes


ds2 = A(r, t) dt2 B(r, t) dr2 r2 d2 + sin2 d2 .
as the general form of an isotropic metric.

14.2

Schwarzschild metric

We specialise further by looking for time-independent metrics, i.e.

ds2 = A(r) dt2 B(r) dr2 r2 d2 + sin2 d2 .
This is also static as it is invariant under the transform t t.
We want to find the metric around a star such as the Sun, i.e. in empty
space where T = 0 and T = T = 0 = R = 0, so the field equations


1
8G
R Rg = 4 T ,
2
c
reduce to
R = 0.
R comes from
R = , , + ,
while

1
= g (g, + g, g, ) .
2
Unfortunately there are no more short-cuts from this point. Work out
then R. Much algebra leads to coupled, ordinary differential equations for See Q4.8, Q5.5,
Q6.1

LECTURE 14. SCHWARZSCHILD GEOMETRY

58

A and B (e.g. Hobson et al p200) and one finds



k
A(r) = 1 +
,
r

1
k
B(r) =
1+
,
r
and k constants.
In weak fields we know that


2
A(r) c 1 + 2 ,
c
2

so = c2 and k = 2GM/c2 . We arrive at the Schwarzschild metric:




1


2GM
2GM
2
ds = c 1 2
dt 1 2
dr2 r2 d2 + sin2 d2 .
cr
cr
2

This applies outside a spherically-symmetric object, e.g. for motions of the

planets but not inside the Sun.
Schwarzschilds solution is important as the first exact solution of the field
equations.

14.3

Birkhoff s theorem

If one does not impose time-independence, i.e. A = A(r, t), B = B(r, t), and
solves R = 0, one still finds Schwarzschilds solution (Birkhoff 1923), i.e.
The geometry outside a spherically symmetric distribution of
matter is the Schwarzschild geometry.
This means spherically symmetric explosions cannot emitt gravitational waves.
It also means that spacetime inside a hollow spherical shell is flat since it
must be Schwarzschild-like but have M = 0. Flat implies no gravity, the GR
equivalent of Newtons iron sphere theorem.
Used in semi-Newtonian justifications of the Friedmann equations.

14.4

Schwarzschild radius

The Schwarzschild metric has a singularity at

LECTURE 14. SCHWARZSCHILD GEOMETRY

r = RS =

59

2GM
M
km.
= 2.9
2
c
M

Usually this is irrelevant, because the Schwarzschild radius lies well inside
typical objects where the metric does not apply, e.g. for the Sun RS  R =
7 105 km, for Earth RS 1 cm.
However, it is easy to conceive circumstances where objects have R < RS ,
e.g consider the Galaxy as 1011 Sun-like stars. Then
RS = 2.9 1011 km,
50 size of Solar system. Mean distance between N stars in a sphere
radius RS
1/3

4RS3
= 1.00 108 km.
d=
3N
Comparing with R = 7 105 km, the stars have plenty of space: do not
require extreme density.
Finally, as a hint of things to come, consider the interval for r < RS . Then
gtt = c2 (1 RS /r) < 0 and grr = (1 Rs /r)1 > 0. Massive particles
must have ds2 > 0, but, ignoring and ,
ds2 = c2 d 2 = gtt dt2 + grr dr2 > 0.
Given that gtt < 0 and grr > 0, we must have dr 6= 0 for r < RS to
give ds2 > 0. The passing of proper time therefore requires a change in
radial coordinate; the future points inwards. This leads to a collapse to
a singularity at r = 0. There is no such thing as a stationary observer for
r < RS .

Lecture 15
Schwarzschild equations of
motion
Objectives:
Planetary motion, start.
Reading: Schutz, 11; Hobson 9; Rindler 11.

15.1

Equations of motion

Writing = GM/c2 , the Schwarzschild metric becomes




1


2
2
2
ds = c 1
dt 1
dr2 r2 d2 + sin2 d2 ,
r
r
2

and the corresponding Lagrangian is



1



2
2 2
2
2
2 2
2
2

t 1
L=c 1
r r + sin .
r
r
There is no explicit dependence on either t or , and thus L/ t and L/
are constants of motion, i.e


2
1
t = k,
r
r2 sin2 = h,
where k and h are constants. h is the GR equivalent of angular momentum
per unit mass.
60

LECTURE 15. SCHWARZSCHILD EQUATIONS OF MOTION

61

For k, recall that for ignorable coordinates such at t and , the corresponding covariant velocity is conserved , i.e.
x 0 = g0 x = g00 x 0 = constant,
where the third term follows from diagonal metric. Now x0 = ct, while
g00 = 1 2/r, so


2
ct = kc.
x 0 = 1
r
Now p0 = mx 0 , where p0 is the time component of the four-momentum, and
in flat spacetime p0 = E/c where E is the energy, so
E = p0 c = x 0 mc = kmc2 ,
is the total energy for motion in a Schwarzschild metric.

p0 = mc = E/c

NB k can be < 1, because in Newtonian terms it contains

potential energy as well as kinetic and rest mass energy.
For the r component we have
d
d

L
r

L
= 0,
r

which gives
!

1 !
2



2c2 2
2
2 2
d
2
2r
1
r 2r 2 + sin2 2
.
t + 1
d
r
r2
r
r2
while the component leads to:
 

d 
2r 2r2 sin cos 2 = 0.
d
The last equation is satisfied for = /2, i.e. motion in the equatorial plane.
By symmetry, we need not consider any other case, leaving


2
1
t = k,
r
1

2

2
2
c2 2
2
r r 2 = 0,
1
r + 2 t 1
r
r
r
r2
r2 = h.
For circular motion, r = r = 0, the second equation reduces to
c2 2
t = r 2 ,
r2
and defining = d/dt and remembering = GM/c2 , we get
GM
,
r3
Keplers third law! . . . somewhat luckily because of the choice of r and t.
2 =

LECTURE 15. SCHWARZSCHILD EQUATIONS OF MOTION

15.2

62

An easier approach

Rather than use the radial equation above, it is easier to use another constant
of geodesic motion:
~ U
~ = g x x = constant.
U
This is effectively a first integral which comes from the affine constraint, or,
~ = 0. It side-steps the r term.
equivalently, from U~ U
More specifically we have
g x x = c2 ,
for massive particles with = , and
g x x = 0,
for photons.

15.3

Motion of massive particles

The equations to be solved in this case are thus



2
t = k,
1
r


1

2
2 2
2
t 1
r 2 r2 2 = c2 ,
c 1
r
r
r2 = h.
Substituting for t and in the second equation and multiplying by (1
2/r) gives
h2
r + 2
r
2

2
1
r

2c2
= c2 k 2 1 .
r

This has the form of an energy equation with a kinetic energy term, r 2
plus a function of r, potential energy equalling a constant.
Thus the motion in the radial coordinate is exactly equivalent to a particle
moving in an effective potential V (r) where


h2
2
c2
V (r) = 2 1

,
2r
r
r
or, setting = GM/c2 ,
h2
V (r) = 2
2r



2GM
GM
1 2

.
cr
r

Dots are wrt

proper time

LECTURE 15. SCHWARZSCHILD EQUATIONS OF MOTION

63

One can learn much about Schwarzschild orbits from this potential.
The equivalent in Newtonian mechanics is easy to derive:
r 2 + r2 2

2E
2GM
=
,
r
m

and r2 = h. Thus
r 2 +

h2 2GM
2E
=
,

2
r
r
m

so
VN (r) =

h2
GM
.

2
2r
r

GR introduces an extra term in 1/r3 in addition to the Newtionian 1/r

gravitational potential and 1/r2 centrifugal barrier terms.

15.4

Schwarzschild orbits (Not in lectures)

Three movies of orbits in Schwarzschild geometry were shown in the lecture.

Movies illustrate the following key differences between GR and Newtonian
predictions:
Apsidal precession of elliptical orbits
Instability of close-in circular orbits
Capture orbits

Lecture 16
Schwarzschild orbits
Note - lectures
16/17/18 may be
compressed

Objectives:
Planetary motion
Reading: Schutz, 11; Hobson 9; Rindler 11

16.1

Newtonian orbits
VN (r) =

h2
2r2

Figure: Newtonian effective potential: centrifugal barrier

always wins
show variation
with h
Centrifugal barrier always dominates as r 0
2 types of orbits: unbound, hyperbolic E > 0; bound, elliptical E < 0.
64

GM
r

LECTURE 16. SCHWARZSCHILD ORBITS

65

Circular: r = 0, r = rC such that r = 0 = dV /dr = V 0 (r) = 0.

Newtonian elliptical orbits do not precess.
To see last point, expand potential around r = rC :
1
V (r) V (rc ) + V 00 (rC )(r rC )2 .
2
2
cf potential/unit mass of a spring kx /2m, then r must oscillate with angular
frequency (epicyclic frequency)
r2 = V 00 (rC ).
Given the Newtonian effective potential
V (r) =

h2
GM
,

2
2r
r

so

h2 GM
+ 2 .
r3
r
V 0 (rC ) = 0 = h2 = GM rC , therefore
V 0 (r) =

V 00 (rC ) =

GM
3h2 2GM
3 = 3 .
4
rC
rC
rC

However, 2 = GM/rC3 , thus r = .

same , so no precession.

16.2

= always reach minimum r at

Schwarzschild orbits

Case 1. Large angular momentum h

Reminder:
V (r) =


2
h2

2
2r
r

Units of h on
plots are c.

Figure: Schwarzschild effective potential for a large values

of h

c2
.
r

LECTURE 16. SCHWARZSCHILD ORBITS

Essentially Newtonian behaviour as small r is inaccessible.
This case applies to the planets. e.g. for Earth h 104 c.
Case 2. Intermediate angular momentum h

Figure: Schwarzschild effective potential for an intermediate

value of h
Bound near-elliptical and circular orbits still exist
Qualitatively different capture orbits possible.
Case 3. Low angular momentum h

Figure: Schwarzschild effective potential for a low value of

h

66

LECTURE 16. SCHWARZSCHILD ORBITS

67

No bound orbits.

16.2.1

Instability of circular orbits

The Schwarzschild effective potential is



2
c2
h2

.
V (r) = 2 1
2r
r
r
At the radius of circular orbits, dV (r)/dr = V 0 (r) = 0 =
V 0 (r) =

h2 3h2 c2
+ 4 + 2 = 0,
r3
r
r

or
c2 r2 h2 r + 3h2 = 0,
so
rC =

h2

p
h4 12h2 2 c2
.
2c2

The smaller root is a maximum of V and unstable. The larger root is stable
while h2 > 122 c2 , but once h2 122 c2 there are no more stable circular orbits.
At this point
6GM
h2
= 6 =
= 3RS .
rC =
2
2c
c2
In accretion discs around non-rotating black-holes no more energy is available
from within this radius. Calculate energy lost using E = kmc2 .
Since r = 0, r = 6 and h2 = 122 c2 :


h2
2
2c2
2 2
c (kC 1) = 2 1
,

r
r
r


122 c2
2
2c2
,
=
1

362
6
6
1
= c2 .
9
Thus kC2 = 8/9. A mass dropped from rest at r = starts with k = 1, and
thus 1 kC = 5.7 % of the rest mass must be lost to radiation. Compare cf Newtonian
with 0.7 % H He fusion.
value of
GM/6RS =
Accretion power from black-holes is thus a conservative hypothesis in many 1/12 = 8.3%.
cases as it requires much less fuel than fusion, e.g. 1 star per week rather than
7 or 8. Rotating black-holes can be more efficient still, with a maximum of
42% (Kerr metrics). In realistic cases it is thought that about 30% efficiency
is possible.

Lecture 17
Precession and Photon orbits
Note - lectures
16/17/18 may be
compressed

Objectives:
Precession of perihelion
Start on orbits of photons
Reading: Schutz, 10 & 11; Hobson 9 & 10; Rindler 11.

17.0.2

Precession in the Schwarzschild geometry

As for Newton, oscillations in r occur at r2 = V 00 (rc ) but now, setting

= GM/c2 ,




h2
2
c2
1

c2
2
V (r) = 2 1

=h

.
2r
r
r
2r2 r3
r
First obtain a condition on h for circular orbits of radius r from V 0 (r) = 0:


1
3
c2
0
2
V (r) = h 3 + 4 + 2 = 0,
r
r
r
thus
h2 =

c2 r2
.
r 3

68

LECTURE 17. PRECESSION AND PHOTON ORBITS

69

The second derivative is then



3
2c2
12
00
2
V (r) = h
5 3 ,
r4
r
r


2 2
c r
3
2c2
12
=

,
r 3 r4
r5
r3
c2
(3r 12 2(r 3)) ,
= 3
r (r 3)
c2 (r 6)
=
.
r3 (r 3)
Thus
r2 =

r 6
r 3

c
.
r3

cf Newton c2 /r3

NB r2 0 as r 6 = 6GM/c2 as expected for the last circular orbit.

Therefore successive close approaches to the star (periastron) occur on a
period of
2
Pr =
,
r
measured in terms of the proper time of the orbiting particle. During this
time the azimuthal angle increases by
NB = d/d 6=
d/dt
2 2 h

Pr =
=
radians.
r
r r2
Therefore, subtracting 2, the periastron precesses by an amount
" 
#
1/2 
1/2  3 1/2
1
c2 r2
r 3
r
= 2 2
1 ,
r
r 3
r 6
c2
"
#
1/2
r
= 2
1 rads/orbit
r 6
If r  this can be approximated as 6/r rads/orbit, or

6GM
rads/orbit.
c2 r

The precession is in the direction of the orbit (prograde).

LECTURE 17. PRECESSION AND PHOTON ORBITS

70

Figure: Prograde precession of an orbit started at r =

100GM/c2 at its most distant point.

17.1

Precession of the perihelion of Mercury

The orbit of Mercury is observed to precess at about 5600 arcseconds/century.

All but 42.98 0.04 /century can be explained by Newtonian effects pre- 1 arcsec = 1/3600
cession of the Earths axis causing the reference frame to change (5025) of a degree
and perturbations from other planets (532). Discrepancy known in 19th
century and ascribed to a new planet Vulcan.
This bears certain
similarities to
What does GR predict? rM = 5.55 107 km, and since GM/c2 = 1.47 km
dark matter.
6 1.47
=
= 0.103 arcsec/orbit.
5.55 107
Mercurys orbital period PM = 0.24 yr, so GR predicts a precession of 100
0.103/0.24 = 43 /century!
This is one of the classic experimental tests of GR. The same effect is seen
with dramatic effect in the orbits of binary pulsars where precession rates as
high as 17 /year have been measured. Then used to measure the masses.
When Einstein developed GR, the anomalous precession of Mercurys orbit was the only experimental evidence against Newtons theory. Einstein
included the GR prediction in his 1916 paper presenting GR. Solving this
problem so beautifully must have been supremely satisfying. Consider the
beauty of GR here compared to alternatives such as altering Newtons Law of
Gravity to 1/r2.00000016 as was also proposed . . . there is no contest!

LECTURE 17. PRECESSION AND PHOTON ORBITS

17.2

71

Equations of motion for photons

The equations of motion for photons read:



2
t = k,
1
r
r2 = h,



1
2 2
2
c2 1
t 1
r 2 r2 2 = 0.
r
r
The only difference is the last equation which ends in c2 for massive particles.
(Remember it comes from g x x = 0 for null paths.)
Substituting for t and in the second equation gives an energy equation
for photons:
h2
r + 2
r
2

2
1
r

= c2 k 2 .

The effective potential for light is thus



h2
2
V (r) = 2 1
.
2r
r
The Newtonian potential term GM/r does not appear at all!

Figure: Effective potential for photons

Key points:
Photons have equivalents of hyperbolic, circular and capture orbits.

LECTURE 17. PRECESSION AND PHOTON ORBITS

There are no elliptical orbits for photons.
The circular orbits are always unstable (maximum of V (r)).

72

Lecture 18
Deflection of light
Note - lectures
16/17/18 may be
compressed

Objectives:
Deflection of light
Reading: Schutz, 10 & 11; Hobson 9 & 10; Rindler 11.

18.1

Circular photon orbits

0

Circular orbits: r = rC such that V (rC ) = 0, or

3
1
+ 4 = 0,
3
rC
rC

i.e.

3GM
.
c2
3 the Newtonian result rC = GM/c2 , problem sheet 1.
rC =

18.2

Deflection of light by the Sun

Orbits with r  GM/c2 suffer a small deflection which is experimentally

measurable.

73

Show
gravitational
lensing pictures.
V =
h2 /2r2 (1 2/r).

LECTURE 18. DEFLECTION OF LIGHT

74

Figure: Deflection of light by a mass. Black circle shows the

event horizon, so the deflection in this case is large.
To calculate light deflection, we need an equation relating r and without
the affine parameter, . Can obtain this by noting:
r =

dr
dr d
dr
h dr
=
=
= 2 .
d
d d
d
r d

Then the energy equation becomes

 2


h2 dr
h2
2
+ 2 1
= c2 k 2 .
r4 d
r
r
Making the substitution r = 1/u (also used for Newtonian orbits):
2

c2 k 2
1 du
4
u 2
+ u2 (1 2u) = 2 ,
u d
h
and so

2
du
c2 k 2
+ u2 2u3 = 2 .
d
h
Finally, differentiating with respect to and dividing by 2du/d:


d2 u
+ u = 3u2 .
d2
For large radii, r  , u  1 , the RHS can be neglected and we have the
SHM equation, thus:
u = a sin + b cos ,
where a and b are constants, or, without loss of generality, simply
u = a sin ,
or r sin = 1/a = r0 , a constant. This is the equation of a straight line with
impact parameter r0 . As r , u 0 gives = 0 or .

LECTURE 18. DEFLECTION OF LIGHT

75

Figure: Straight from = 0 to = with polar equation

r sin = a1 .
Now look for a better approximation u = u0 + u0 where |u0 |  u0 = a sin .
Then
d2 u 0
3a2
0
2
2
2
2
(1 cos 2),
+
u
=
3u

3u
=
3a
sin

=
0
d2
2
neglecting small terms on the right. Particular integral is


1
3a2
0
1 + cos 2 ,
u =
2
3
so a better solution is
3a2
u = a sin +
2


1
1 + cos 2 .
3

Now r = u = 0 =
3a
sin =
2

1
1 + cos 2
3


2a,

since cos 2 1 for = 0, . Therefore

2a, or + 2a.
Thus light is deflected by
= 4a =

4GM
.
c2 r0

This is 2 the Newtonian result (pure SR predicts zero).

For light grazing the Sun
=

4GM
4 6.67 1011 2 1030
=
= 8.47106 rads = 1.75 arcsec.
c2 R
(3 108 )2 7 108

Confirmed from observations of radio sources to 2 parts in 104 Deflection of See Shapiro et al
light now an important tool in astronomy, gravitational lensing.
in reading.
Famously tested by British astrophysicist Eddington in 1919 using observations of stars near the Sun during a total eclipse. Made Einstein famous.
Eddington the source of the well-known quote Interviewer: Professor Eddington, is it true that only three people understand Einsteins theory? Eddington: Who is the third?

Lecture 19
Schwarzschild Black holes
Lecture can get a
bit rushed

Objectives:
Beyond the Schwarzschild horizon
Reading: Schutz 11; Hobson 11; Rindler 12

19.1

The Schwarzschild horizon

The Schwarzschild metric



1

2
2
2
2
2
dr2 r2 d2 ,
dt 1
ds = c 1
r
r
(d2 short-hand for angular terms) is singular at
r = RS = 2 =

2GM
.
c2

This is a coordinate singularity, similar to the singularity of the 2-sphere

metric
dr2
ds2 =
+ r2 d2 ,
2
2
1 r /R
when r = R at the equator.
Consider radially moving particles for which d = d = 0. Then we have
ds2 = gtt dt2 + grr dr2 .
For r < 2, gtt < 0, grr > 0. For massive particles a time-like interval
ds2 > 0 therefore requires dr 6= 0 and so r can never change sign.
76

LECTURE 19. SCHWARZSCHILD BLACK HOLES

77

A particle which enters the event horizon will keep r < 0,

until r = 0 no escape once inside event horizon.
Now consider photons (ds = 0):


2
c dt = 1
r

1
dr,

+ for outgoing, for incoming. Integrating


Z
Z 
Z
r dr
r 2
2
dr
=
=
+
dr,
ct =
1 2/r
r 2
r 2 r 2
thus
ct = (r + 2 ln |r 2|) + constant.
Spacetime diagram:

Figure: Spacetime diagram in r and t coordinates representing a series of in- and out-going photon worldlines. On the
left, ingoing worldlines move down the ct axis. Wavy line
represents the singularity at r = 0. The dashed line is the
event horizon at r = RS . The green line shows the path of
the same ingoing photon on each side of r = RS .
At any event E, the future lies between the worldlines of ingoing and
outgoing photons, on the same side as their direction of travel.

See reading on
web pages on the
Shapiro delay
for an
experimental
measurement of
this.

LECTURE 19. SCHWARZSCHILD BLACK HOLES

78

As r RS , lightcones are squeezed; worldlines take infinite coordinate time t

to reach RS .
For r < RS , lightcones are rotated and point towards r = 0. Particles
crossing r = RS can never again be seen from r > RS , thus the event
horizon.

19.2

Maximum free-fall time

For massive particles and r < RS = 2 can write


1


2
2
2
2
2
2
1
dr c
1 dt2 r2 d2 .
c d =
r
r
d maximum if dt = d = 0. Thus the maximum time one has before
reaching the singularity from r = RS is
1
m =
c

Z
0

1/2



GM
2
M
6
1
dr =
=
= 15 10
sec .
r
c
c3
M

e.g. 4.2 hours for M = 109 M . Any use of a rocket shortens this!

19.3

Kruskal-Szekeres coordinates

Schwarzschild coordinates are singular at r = RS and ill-behaved for r < RS .

In 1961 Kruskal found better-behaved coordinates. Consider the incoming/outgoing photon worldlines:
ct = r 2 ln |r 2| + p,
ct = +r + 2 ln |r 2| + q,
where p and q are integration constants. We can label events with p and q.
Photon paths then form a rectangular grid in (p, q) and the interval becomes Sketch grid


2
2
ds = 1
dp dq r2 d2 .
r
The following transform removes the awkward 1 2/r:
p = + exp(p/4),
q = exp(q/4).
A rotation gives time- and space-like rather than null coords:

see Q8.6

LECTURE 19. SCHWARZSCHILD BLACK HOLES

79

v = (
p + q)/2,
u = (
p q)/2.
These are Kruskal-Szekeres coordinates. The interval becomes


323 r/2
dv 2 du2 r2 d2 ,
ds =
e
r
2

where
2

u v =


r
1 er/2 .
2

Null radial paths, ds2 = d2 = 0 =

v = u + constant,
i.e. 45 like Minkowski!
r = 0 = v 2 u2 = 1, i.e. hyperbolae.

Figure: Spacetime diagram in u, v Kruskal coordinates.

Light-cones now have same structure as Minkowski, so the
future of any even is the region over it within 45 of the vertical.
Kruskal spacetime diagram:
Region I is the region r > RS in which we live; region II represents
r < RS .
Future of any event is contained in 45 lightcone directed upwards.
Once inside region II, the future ends on the upper r = 0 singularity.
Can pass from I to II but not back again.

LECTURE 19. SCHWARZSCHILD BLACK HOLES

80

Region III similar to I but disconnected from it: a different Universe

Lower shaded line is a past singularity, out of which particles emerge.
Once you have entered region II you can never leave; once you have
left IV you can never return: a white hole
Whether regions
III and IV have
any reality is
unclear, but
seems unlikely.

Lecture 20
The FRW metric
Lecture tends to
overrun

Objectives:
Friedmann-Robertson-Walker metric
Reading: Schutz 12; Hobson 14; Rindler 16

20.1

Isotropy and homogeneity

On large scales, the Universe looks similar in all directions, and, in addition,
assuming that ours is not a special location (Copernican principle), we
assert that on large scales the Universe is
isotropic: no preferred direction
homogeneous: the same everywhere.

20.2

Cosmic time

Homogeneity implies a synchronous time t can be defined so that at a given

t, physical parameters such as density and temperature are the same everywhere. Thus we can write the interval
ds2 = c2 dt2 dl2 ,
where
dl2 = gij dxi dxj ,

81

LECTURE 20. THE FRW METRIC

82

i.e. spatial terms only. g0i = 0 because isotropy = no preferred direction

(cf Schwarzschild). For dl2 we look for a 3D-space of constant curvature,
analagous to the surface of a sphere (2D-space of constant curvature).
Consider the surface of a sphere in Euclidean 4D. Using Cartesian coordinates (x, y, z, w), but replacing (x, y, z) by spherical polars (, , ), we Using for the
have
radial coord, to
2
2
2
2
2
dl = d + d + dw ,
save r for later;
see below.
where d2 is short-hand for the angular terms. Also
x2 + y 2 + z 2 + w 2 = 2 + w 2 = R 2 ,
and so
d + w dw = 0.
Therefore
dw2 =

2 d2
2 d2
=
,
w2
R 2 2

and so
dl2 = d2 +

2 d2
+ 2 d2 ,
R2 2

giving
dl2 =

d2
+ 2 d2 .
2
1 (/R)

This is a homogeneous, isotropic 3D space of (positive) curvature 1/R2 .

Negative and zero curvature are also possible, and setting = Rr, all three
cases can be expressed as


dr2
2
2
2
2
+ r d ,
dl = R
1 kr2
where k = 1, 0 or +1.
In general we must allow for R to be an arbitrary function of time R(t) (not
position since that would destroy homogeneity), thus we arrive at
2

ds = c dt R (t)


dr2
2
2
2
2
2
+ r d + r sin d .
1 kr2

This is the Friedmann-Robertson-Walker metric. It was first derived by

Friedmann in 1922, and then more generally by Robertson and Walker in
1935. It applies to any metric theory of gravity, not just GR.

20.3

Geometry of the Universe

Three cases:

LECTURE 20. THE FRW METRIC

83

k = 1 Positive curvature, closed universe.

k = 0 Zero curvature, flat universe (flat space, not flat spacetime)
k = 1 Negative curvature, open Universe.
An alternative form of the metric is often useful. For k = 1, setting r = sin ,
the interval becomes
ds2 = c2 dt2 R2 (t)(d2 + sin2 d2 ).
R
The circumference of a circle of proper radius rp = R d = R is then
clearly
r 
p
C = 2R sin = 2R sin
,
R
while the area of a sphere of the same radius is
r 
p
A = 4 (R sin )2 = 4R2 sin2
,
R
and its volume is



Z


1
2rp
2
2
3 rp
sin
V =
4R sin R d = 2R
.
R 2
R
0
As rp R, C and A 0, and V 2 2 R3 is finite, hence a closed
universe, directly analogous to the surface of a sphere.
In a closed
Universe you
In general we can write the alternative FRW metric as
could keep


travelling in one
ds2 = c2 dt2 R2 (t) d2 + Sk2 () d2 ,
direction and yet
return to where
where

you started.

sin , for k = 1,
Sk () =

20.4

,
for k = 0,
sinh , for k = 1.

Redshift

The wavelength of light from astronomical sources is a crucial, easily measured observable. Consider two pulses of light emitted at times t = te and
t = te + te by an object at towards an observer at the origin who picks
them up at t = to t = to + to .
For photons travelling towards the origin, since ds = 0
c dt = R(t) d,

as a comoving
coordinate

LECTURE 20. THE FRW METRIC

Therefore
Z

to

=
te

c dt
=
R(t)

84

to +to

te +te

c dt
.
R(t)

Subtracting the first integral from the second:

Z to +to
Z te +te
c dt
c dt

= 0.
R(t)
R(t)
to
te
For small intervals R(t) is almost constant, so
to
te
=
.
R(to )
R(te )
Therefore we define redshift z by
1+z =

o
e
to
R(to )
=
=
=
.
e
o
te
R(te )

1 + z is thus the factor by which the Universe has expanded in between

emission and reception of the light.

20.5

Hubbles Law

The universal fluid (= galaxies) is at rest in comoving coordinates r or ,

and . Expansion of the Universe is contained in the size factor R(t).
Consider the proper distance to a galaxy at radius
Z
dP =
R(t) d = R(t),
0

Since is fixed, the rate of recession of the galaxy is

v=

d
= R dP .
(dP ) = R
dt
R

Identifying

H(t) = R/R,
we have
v = H(t)dP
which is Hubbles Law, while H(t) is Hubbles constant = H(t0 ) = H0
today.

Hubbles Law is
thus a direct
outcome of
homogeneity and
isotropy.

Lecture 21
Dynamics of the Universe
Objectives:
The Friedmann equations
Reading: Schutz 12; Hobson 14; Rindler 16

21.1

Friedmanns equation

The evolution of the Universe in GR is determined as follows:

1. The FRW interval = the metric, e.g. grr = R2 /(1 kr2 )
2. The metric = , the connection.
3. The metric and connection = R , the Ricci tensor.
4. The Ricci tensor and field equations = differential equations for the
size factor R and the fluid density .
Jumping straight in at step 4, consider

R
Rtt = 3 .
R
Use field equations in the form

1
= k T T g .
2


R
Assume perfect fluid:


p
T = + 2 U U pg .
c
85

See handout 7

LECTURE 21. DYNAMICS OF THE UNIVERSE

86

Fluid is static in co-moving coordinates of FRW metric so U i = 0 and

g U U = gtt U t U t = c2 ,
so since gtt = c2 , U t = 1. Hence
Ut = gtt U t = c2 ,
and


p
Ttt = + 2 c4 pc2 = c4 ,
c

while


p
p
T = g T = + 2 g U U pg g = + 2 c2 4p = c2 3p.
c
c
Therefore




1 2
R
4
2
3 = k c (c 3p)c .
R
2

Putting k = 8G/c4 we obtain

= 4G
R
3

3p
+ 2
c


R.

(21.1)

This is the acceleration equation.

Similarly the rr, and components all lead to:
8G 2
R 2 + kc2 =
R .
3

(21.2)

This is the Friedmann equation.

Finally, taking the time derivative of the Friedmann equation and substitut from the acceleration equation it is simple to show:
ing for R
Prove this
+

3R 
p
+ 2 = 0.
R
c

(21.3)

which is the fluid equation. Alternatively this comes from T ; = 0.

21.1.1

Newtonian interpretation [Not in Lectures]

Each of Eqs 21.1, 21.2 and 21.3 has an approximate Newtonian interpretation. If one considers an expanding uniform density sphere then
= 4G R.
R
3

LECTURE 21. DYNAMICS OF THE UNIVERSE

87

There is no Newtonian explanation for the pressure term in the acceleration

equation. Conserving energy for a particle on the edge of such a sphere gives:
1 2 4G 2 E
R = .
R
2
3
m
Newtonian equivalent for curvature term kc2 is total energy per unit mass.
Finally the fluid equation follows directly from
T dS = dU + p dV,
setting dS = 0 (reversible adiabatic, no temperature gradients) and using massenergy equivalence. Such Newtonian interpretations are a fudge:
Eqs 21.1, 21.2 and 21.3 are relativistic.

21.2

The cosmological constant

In 1917 Einstein modified the field equations to read

1
R Rg + g = kT ,
2
where is the cosmological constant. Still satisifes T ; = 0 since g ; = 0.
Nowadays, it is usual to place the new term on the right as the stressenergy
tensor of the vacuum.


1

R Rg = k T g
.
2
k
Second term in brackets on the right has the form of a perfect fluid

p 
+ 2 U U p g ,
c
if
+
and
p =

p
= 0,
c2

c4
=
,
k
8G

and thus

c2
.
8G
i.e. a fluid of constant density and negative pressure.
=

This is dark
energy, perhaps
the most puzzling
problem in
modern physics.

LECTURE 21. DYNAMICS OF THE UNIVERSE

21.2.1

88

Einsteins static universe

Negative pressure allows a static Universe. From



3p
4G

+ 2 R,
R=
3
c
can be zero if
R

3p
= 0.
c2
Here and p are the sums of contributions from all components. Considering
matter and only, for matter pM  M c2 so
+

3p
3p
M + + 2 = M 2 .
2
c
c

Thus

equation of state
4G
=
p = wc2
(M 2 ) R,
R
3
which is zero if M = 2 . This is Einsteins static universe. Unfortunately
it would not be static for long since it is unstable. Consider a perturbation
M = 2 + 0 , R = R0 + R0 . To first order
4G 0
R0 .
R0 =
3
If R0 > 0 we expect 0 < 0 since matter is diluted as the universe expands,
hence R0 > 0 and the perturbation will grow = instability. The universe
either contracts or expands away from R = R0 .
therefore can give a static but not a stable universe. Had Einstein realised
this, he could have predicted an expanding or contracting universe. Perhaps
this was why he once referred to the cosmological constant as my greatest
blunder (as quoted by Gamow, 1970).

Lecture 22
Cosmological distances
Objectives:
Friedmann-Robertson-Walker metric
Reading: Schutz 12; Hobson 14 and 15; Rindler 17

22.1

Distances

There is no one distance in cosmology. Using the metric

ds2 = c2 dt2 R2 (t) d2 + Sk2 () d2 ,
the easiest to define is the ruler or proper distance dP
dP = R0 ,
where R0 = R(t0 ) is the present size factor of the Universe.
A more practical measure is the luminosity distance defined as the distance
at which the observed flux f from an object equals the standard Euclidean
formula:
L
,
f=
4d2L
where L is the luminosity.
Consider a source S at the origin (can always shift origin) and an observer
O at . When light reaches O at time to , it is spread equally (isotropy) over
an area
Sketch O and S
2 2
A = 4R0 Sk ().

89

LECTURE 22. COSMOLOGICAL DISTANCES

90

The flux (power/unit area) observed is therefore

f=

L
4R02 Sk2 ()(1

+ z)2

The (1 + z)2 factor comes from the redshift which reduces both the energy
and arrival rate of the photons. The R2 (t)Sk2 () comes from the angular
terms of the FRW metric. Therefore
dL = R0 Sk ()(1 + z).

The angular diameter distance dA is defined such that

=

l
,
dA

where is the angle subtended by an object of size l.

Sketch:

Figure: The angular size defined at emission is preserved

during expansion because the photons travel along radial
paths towards the origin.
Photons travel from source to observer along radial paths. Angular size
defined at time of emission. From the FRW metric,
l = R(te )Sk (),
and therefore
dA = R(te )Sk () =
since
1+z =

R0 Sk ()
,
1+z

R0
.
R(te )

In each case we need which is connected to the time of emission te and

observation t0 through
Z t0
c dt
.
=
te R(t)

LECTURE 22. COSMOLOGICAL DISTANCES

91

We can replace t by z(t) where

1 + z(t) =
so
dz =
and hence
Z

R0
,
R(t)

R0
R dt,
R2

cR2 1
dz,
R0 R R

=
z

so, remembering H = R/R,

Z
R0 =
0

c dz
.
H(z)

Thus , and hence the distances, are sensitive to the expansion history of
the Universe encoded in H(z). e.g. flux vs redshift Hubble diagrams of
supernovae = a cosmological constant.
Show SN Hubble
diagram

22.2

The future of our Universe

Observations suggest that our Universe is 74% cosmological constant, 26%

matter (5% baryonic). In the future will dominate since M R3 while
is constant, so the Friedman equation tends to

1/2
8G

R =
R,
3
the curvature term being constant becomes negligible compared to the above
terms. This equation describes a de Sitter universe in which there is only a
cosmological constant. Clearly
R = R0 exp(t/ ),
where t is measured from the present and
1/2

3
= 1.6 1010 yr,
=
8G
for our Universe.
Outrunning a photon: consider a photon emitted at time t = te (counting
from the present). By time t it will have reached comoving radius given
by
Z t
Z t


c dt
c
c te /
(t) =
=
et/ dt =
e
et/ .
R0 te
R0
te R(t)

LECTURE 22. COSMOLOGICAL DISTANCES

92

As t , the photon has reached a proper distance as measured in todays

Universe (R = R0 ) of
dP = R0 = c ete / .
Implication: photons in a de Sitter unverse never catch up distant parts of
the Universe. The later a photon is emitted, the shorter the distance it
travels in todays terms. Put differently, we see no photons that a galaxy at
proper distance dP emitts after a time
te = ln

c
.
dP

Were we to observe a clock in such a galaxy, we would see it get slower

and slower, never quite making it to te . The galaxy meanwhile becomes
increasing redshifted and ever fainter. This is an external event horizon in
fact.
As a consequence, in the future, all galaxies now in the Hubble flow away
from us will disappear from our view, unless the dark energy driving the
expansion runs out of steam.

Lecture 23
Linear GR
Objectives:
Linearised GR
Reading: Schutz 8; Hobson 17; Rindler 15

23.1

Approximating GR

The non-linearity of GR makes it difficult to solve in most situations. It is

useful to develop an approximate form of the field equations for the common
case of weak fields.
In weak fields we can assume that there are coordinates x in which the
metric can be written
g = + h ,
where |h |  1. Using this the field equations
1
R Rg = kT ,
2
can be approximated to first order in h.
e.g. the connection
1
g (g, + g, g, ) ,
2
1
(h, + h, h, ) ,
=
2

is first-order in h, so the Riemann tensor boils down to

R = , , .
93

LECTURE 23. LINEAR GR

94

Eventually one finds:

h, + 2h (h, + h, ) (2h h , ) = 2kT ,
where h = h and
2 = = =

1 2
2 ,
c2 t2

is the DAlembertian or wave operator.

23.2

Lorenz Gauge

The choice of h is not unique; it depends on the underlying coordinates.

This can be used to simplify the linearised equations. For instance consider
the coordinate transform
x0 = x +  ,
with  and its derivatives  1 (easier here not to put primes on indices;
h is not a tensor). Then
x0 x0 0
g ,
=
x x

0
= ( +  , ) +  , g
,

so
+ h = ( +  , ) +  ,

+ h0 .

Thus (to first order in h)

+ h = +  , +  , + h0 ,
and so
h0 = h , , .
Very similar to gauge transformation of EM where the physics is invariant
to transforms of the 4-potential of the form
A0 = A + , ,
where is some scalar field.
Choose  to simplify field equations. In particular choosing coordinates
such that
1
h , = h, ,
2
(Lorenz gauge), then the field equations reduce to
1
2h 2h = 2kT .
2

Do not try to
remember this!
Also see Q5.8

LECTURE 23. LINEAR GR

95

Further simplification comes from defining

= h 1 h ,
h
2
= h). The Lorenz gauge condition becomes
(trace reversal since h
= 0,
h
,
while the field equations reduce to
= 2kT ,
2h
or in full:


1 2
16G
2
h = 4 T .
2
2
c t
c

There is still some remaining freedom: the same relations survive coordinate
transforms x0 = x +  provided
2 = 0.

23.3

Newtonian limit [not in lectures]

Consider a time-independent, weak-field. Setting k = 8G/c4 , and 2 =

2 , the field equations become
=
2 h

16G
T ,
c4

which has the form of Poissons equation. If all mass is stationary, then only
T 00 = c2 is significant so we have
00 =
2 h

16G
,
c2

and by analogy with

2 = 4G,
we can immediately write
00 = 4 ,
h
c2
where is the Newtonian potential. All other components = 0.
= 4/c2 , and since
From this we deduce h = h
+ 1 h ,
h = h
2

see Q10.5

LECTURE 23. LINEAR GR

96

we find

2
,
c2
Finally, since g = + h , and lowering indices we find






2
2
2
2
2
ds = c 1 + 2 dt 1 2
dx2 + dy 2 + dz 2 .
c
c
h00 = h11 = h22 = h33 =

This approximate metric is useful for studying gravitational lensing around

anything more complex than a point mass, e.g. a star plus planets, or clusters
of galaxies.

Lecture 24
Gravitational waves
Objectives:
Linearised GR and Gravitational Waves
Reading: Schutz 9; Hobson 17; Rindler 15

24.1

Gravitational waves

In the vacuum, T = 0, and so



2
1

= 0.
=
h
2h
2
2
c t
This is the wave equation for waves that travel at the speed of light c. It has
solution
= A exp(ik x ).
h
Remembering that
2 = ,
and substituting the solution into the wave equation gives
= 0.
k k h
For non-zero solutions we must have
k k = k k = ~k ~k = 0,
i.e. ~k is a null vector. This is the wave vector and usually written ~k =
(/c, k). k k = 0 is then just the familiar = ck.

97

LECTURE 24. GRAVITATIONAL WAVES

24.2

98

Gauge conditions

Our solution must satisfy the Lorenz gauge

= 0,
h
,
which leads to the four conditions:
A k = 0.

(24.1)

Four more conditions come from our freedom to make coordinate transformations with any vector field  satisfying
This allows us to
remove waves in
2 = 0.
the coordinates.
The standard choice is called the transversetraceless (TT) gauge in which
A = 0,

(24.2)

Ati = 0.

(24.3)

which makes A traceless, and

These introduce a further four constraints.

Eq. 24.1 can be written as
At kt + Ai ki = 0,
and setting = t, Eq. 24.3 = Att = 0, thus At = At = 0.
Specialising to a wave in the z-direction, k = (kt , 0, 0, kz ), then Eq. 24.1
shows that
At kt + Az kz = Az kz = 0,
so
Az = 0,
hence transverse. Finally, since Att = Azz = 0, Eq. 24.2 shows that
Axx + Ayy = 0,
and so

A =

0
0
0
0

0 0 0
a b 0
b a 0
0 0 0

where a and b are arbitrary constants.

The 2 degrees of freedom represented by a and b correspond
to 2 polarisations of gravitational waves.

LECTURE 24. GRAVITATIONAL WAVES

99

Figure: The two polarisations can be separated into tidal

distortions at 45 to each other. The figure shows the extremes of the distortion that occur to a ring of freely floating
particles as a gravitational wave passes (directly in or out of
the page). The extent of the distortion is very exaggerated
compared to reality!
The two polarisations give varying tidal distortions perpendicular to the
direction of travel.

24.3

Generation of gravitational waves

The equation
= 2kT
2h
is analagous to the equation in the Lorenz gauge in EM
2 =
which has solution

Z
(t, r) =

,
0
[]
dV,
40 R

where [] = (t R/c, x), R = |r x|. Thus by analogy:

Z  
T
h = 2k
dV
4R

Retarded
potential (i.e.
potential at r due
to charge at x)

If the origin is inside the source, and |r| = r  |x| (compact source), we are
left with the far-field solution
Z
h (t, r) 2k
T (t r/c, x) dV.
4r
Using the energy-momentum conservation relation T , = 0 one can then
show that
See Q 2.16, uses
2 ij
2G
d
I
Gausss theorem
ij
h
,
c4 r dt2

LECTURE 24. GRAVITATIONAL WAVES

where
ij

I =

100

xi xj dV,

is the moment-of-inertia or quadrupole tensor.

No gravitational dipole
because conservation of moR radiation
i
mentum means that x dV is constant.

24.3.1

Estimate of wave amplitude

Consider two equal masses M separated by a in circular orbits in the xy

plane of angular frequency around their centre of mass. Then
Z
2 1
a
xx
cos t = M a2 (1 + cos 2t) .
I = x2 dV = 2M
2
4
Differentiating twice gives
2 2
xx = 2GM a cos 2t.
h
c4 r

Other terms similar. Consequences:

Gravitational wave has twice frequency of the source (quadrupole radiation)
Amplitude GM a2 2 /c4 r.
Example: M = 10 M , a = 1 R , at r = 8 kpc (Galactic centre). Then
Kepler3
G(M1 + M2 )
2 =
= 7.8 104 rad2 s2 .
a3
(Orbital period 38 mins, GW period 19 mins).
Find h 2 1021 . This is a tiny distortion of space, < 0.1 mm in the
distance from us to the nearest star.

Note:M a2 2 =
mv 2 i.e. K.E.

Lecture 25
Detection of gravitational
waves
Objectives:
GRW detection
Reading: Schutz 9; Hobson 17; Rindler 15

25.1

Detecting Gravitational waves

The decreasing orbital period of binary pulsar provides strong but indirect
evidence of gravitational waves. Direct detection of gravitational waves is
one of the greatest challenges of modern experimental physics. The main
possible sources are:
Very close pairs of stars: white dwarfs, neutron stars and black-holes
in orbits of a few minutes and down to milliseconds.
Mergers of super-massive black-holes at the centres of galaxies. Most
powerful events of all 4% of total mass in gravitational waves. e.g.
could release 107 M of energy within about an hour, L 1024 L 
rest of observable Universe!
Asymmetric rapidly rotating neutron stars, e.g. in X-ray binaries.
Supernovae
Fluctuations of the very early Universe
GWs may be the only way to observe the 96% of the Universe that doesnt
couple to electromagnetic radiation.
101

LECTURE 25. DETECTION OF GRAVITATIONAL WAVES

25.2

102

Detectors

Two types:
1. Resonant bars (Joseph Weber in the 1960s, experiments like AURIGA
in the 2000s).
2. Michelson interferometers (suspended mirrors act as test masses). Mirrors > 99.999% reflection. Existing (main ones):
(a)
(b)
(c)
(d)

LIGO: 2 interferometers in the USA with 4 km long arms

VIRGO: France/Italy, 3 km arms
GEO600: Germany/UK, 600 m arms
TAMA300: Japan, 300 m arms

Planned: LISA, 2 million km space-based interferometer

Planned: Einstein Telescope, underground interferometer.
Multiple detectors vital for believable result.

25.3

Ground-based detection

LIGO: 4 km-long arms = detect l 1018 m for h 1021 .

Advantages:
Short arms good for high-frequency inspirals. e.g. neutron star pairs
reach 1 kHz.
High laser power possible.
Can be upgraded.
Disadvantages:
Seismic noise limits low frequencies, so most common sources undetectable
Short arms require very high precision
Events are very short lived (< 1 second), making them hard to detetect
Current LIGO can detect merging neutron stars out to 10 Mpc. However,
no detection to date: such events are probably rare.
Advanced LIGO will raise max distance to 100 Mpc, 1000 increase in volume. Expect several events per year.

LECTURE 25. DETECTION OF GRAVITATIONAL WAVES

25.4

103

Space-based detection

Space offers:
Potentially long interferometer arms
No seismic noise so sensitive to much lower frequencies, e.g. early
Universe, merger of supermassive black-holes, early detection of lower
mass mergers and commoner types of binary star.
but
low laser power limits high frequency sensitivity.
costs of space-based science are very high.
LISA is a proposed interferometer with spacecraft 2 million km apart.

25.5

Numerical relativity

At low signal-to-noise, one needs to know the shape of the waveform to

detect it. Thus computer simulations are part of the detection effort. Good
progress has been made in understanding the merger of two black-holes.
Prospects for the first direct detection are good; its now down to the Universe
to give us some observable events.
Watch this space!