Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Gravitational Waves and Black Holes: J.W. Van Holten

Download as pdf or txt
Download as pdf or txt
You are on page 1of 97

a

r
X
i
v
:
g
r
-
q
c
/
9
7
0
4
0
4
3
v
1


1
5

A
p
r

1
9
9
7
NIKHEF/97-017
HD-THEP-97-6
Gravitational Waves and Black Holes
An Introduction to General Relativity
J.W. van Holten
NIKHEF, P.O. Box 41882
1009 DB Amsterdam NL
Abstract
In these lectures general relativity is outlined as the classical eld theory of gravity, emphasizing
physical phenomena rather than formalism. Dynamical solutions representing traveling waves as
well as stationary elds like those of black holes are discussed. Their properties are investigated
by studying the geodesic structure of the corresponding space-times, as representing the motion
of point-like test particles. The interaction between gravitational, electro-magnetic and scalar
elds is also considered.
c
1997
Lectures presented at the University of Heidelberg, february 1997
Contents
1 Gravity and Geometry 1
1.1 The gravitational force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Geometrical interpretation of gravity . . . . . . . . . . . . . . . . . . . . . 3
1.4 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 The Einstein equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6 The action principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Geodesics 11
2.1 Curves and geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Canonical formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Action principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Symmetries and Killing vectors . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Phase-space symmetries and conservation laws . . . . . . . . . . . . . . . 22
2.6 Example: the rigid rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Dynamics of space-time 27
3.1 Classical solutions of the gravitational eld equations . . . . . . . . . . . . 27
3.2 Plane fronted waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Nature of the space-time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Scattering of test particles . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 General planar waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.6 The gravitational eld of a light wave . . . . . . . . . . . . . . . . . . . . . 44
3.7 Symmetry breaking as a source of gravitational oscillations . . . . . . . . 50
4 Black holes 57
4.1 Horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 The Schwarzschild solution . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4 The interior of the Schwarzschild sphere . . . . . . . . . . . . . . . . . . . 63
4.5 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.6 Extended Schwarzschild geometry . . . . . . . . . . . . . . . . . . . . . . 70
4.7 Charged black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
i
ii
4.8 Spinning black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.9 The Kerr singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.10 Black-holes and thermodynamics . . . . . . . . . . . . . . . . . . . . . . . 85
Chapter 1
Gravity and Geometry
1.1 The gravitational force
Gravity is the most universal force in nature. As far as we can tell from observations and
experiments every object, every particle in the universe attracts any other one by a force
proportional to its mass. For slow moving bodies at large distances this is a central force,
inversely proportional to the square of the distance. As the action is reciprocal, and since
according to Newton action and reaction forces are equal in magnitude, the expression for
the gravitational force between two objects of mass M
1
and M
2
at a distance R is then
determined to have the unique form
F = G
M
1
M
2
R
2
. (1.1)
The constant of proportionality, Newtons constant of gravity, has dimensions of accelera-
tion per unit of mass times an area. Therefore its numerical value obviously depends on
the choice of units. In the MKS system this is
G = 6.672 59(85) 10
11
m
3
kg
1
s
2
. (1.2)
It is also possible, and sometimes convenient, to x the unit of mass in such a way that
Newtons constant has the numerical value G = 1. In the natural system of units, in which
also the velocity of light and Plancks constant are unity: c = h = 1, this unit of mass is
the Planck mass m
P
:
m
P
=
_
hc/G = 2.176 71 10
8
kg
= 1.221 047 10
19
GeV/c
2
.
(1.3)
Newtons law of gravity (1.1) is valid for any two massive bodies, as long as they are far
apart and do not move too fast with respect to one another. In particular, it describes
the motions of celestial bodies like the moon circling the earth, or the planets orbiting the
sun, as well as those of terrestrial objects like apples falling from a tree, or canon balls
1
2
in free ight. Ever since Newton this unication of celestial and terrestial mechanics has
continued to impress people and has had a tremendous impact on our view of the universe.
It is the origin and basis for the belief in the general validity of physical laws independent
of time and place.
1.2 Fields
Although as a force gravity is universal, Newtons law (1.1) itself has only limited validity.
Like Coulombs law for the electrostatic force between two xed charges, Newtons law
holds strictly speaking only for static forces between bodies at rest. Moreover, and unlike
the electric forces, there are modications at smaller, nite distances which can be observed
experimentally.
For example, if the gravitational force would have a pure 1/R
2
dependence, the orbits of
particles around a very heavy central body would be conic sections: ellipses, parabolas or
hyperbolas, depending on the energy and angular momentum, in accordance with Keplers
laws. The observation of an excess in the precession of the perihelion of the orbit of Mercury
around the sun by LeVerrier in 1845, and improved by Newcomb in 1882 [1], was one of
the rst clear indications that this is actually not the case, and that the gravitational force
is more complicated.
The exact form of the gravitational forces exerted by moving bodies is a problem with
many similarities to the analogous problem in electrodynamics. The understanding of
electrodynamical phenomena greatly improved with the introduction of the concept of
local eld of force. This concept refers to the following characteristics of electrodynamical
forces:
the inuence of electric charges and currents, and of magnetic poles, extends through-
out empty space;
the force on a standard test charge or test magnet at any given time depends on its
location with respect to the source of the eld and the relative state of motion;
changes in the sources of the elds give rise to changes in the force on test objects at
a later time, depending on the distance; the speed of propagation of disturbances in
empty space is nite.
In the case of electrodynamics this speed turned out to be a universal constant, the speed
of light c. One of the most striking consequences of these properties, which follow directly
from the mathematical description of the elds as expressed by Maxwells equations, is the
existence of electromagnetic waves, local variations in the elds which transport energy,
momentum and angular momentum over large distances through empty space. Maxwells
predictions were magnicently veried in the experiments of Hertz, conrming the reality
and characteristics of electromagnetic waves [2]. From these results it became clear, for
3
example, that light is just a special variety of such waves. Therefore optical and electro-
magnetic phenomena have a common origin and can be described in a single theoretical
framework.
The concept of eld has found other applications in physics, for example in the phe-
nomenological description of uids and gases. Treating them as continua, one can describe
the local pressure and temperature as elds, with local variations in these quantities prop-
agating through the medium as sound waves or heat waves.
As in electrodynamics, also in gravity the concept of eld has taken a central place.
What Maxwell achieved for electromagnetism, was accomplished by Einstein in the case
of gravity [3]: to obtain a complete description of the forces in terms of space- and time-
dependent elds. This eld theory is known as the general theory of relativity. One of the
most remarkable aspects of this theory is, that it provides an interpretation of gravitational
phenomena in terms of the geometry of space-time. Many detailed presentations of the
geometrical description of space-time can be found in the literature, including a number of
books included in the references at the end of these lecture notes [6]-[14]. A general outline
of the structure of the theory is presented in the following sections.
1.3 Geometrical interpretation of gravity
The rst step in describing a local eld, both in electrodynamics and gravity, is to specify
the potentials. The force on a test particle at a certain location is then computed from
the gradients of these potentials. In the case of gravity these quantities have not only a
dynamical interpretation, but also a geometrical one. Whereas the electromagnetic eld is
described in terms of one scalar and one vector potential, together forming the components
of a four vector
1
A

(x), the gravitational eld is described by 10 potentials, which can be


assembled in a symmetric four tensor g

(x). This tensor has a geometrical interpretation


as the metric of space-time, determining invariant space-time intervals ds in terms of local
coordinates x

:
ds
2
= g

(x) dx

dx

. (1.4)
In the absence of gravity, space-time obeys the rules of Minkowski geometry well-known
from special relativity. In cartesian coordinates (ct, x, y, z) the metric of Minkowski space
is

= diag(1, +1, +1, +1). (1.5)


Then the invariant ds
2
takes the form
ds
2
=

dx

dx

= c
2
dt
2
+ dx
2
+ dy
2
+ dz
2
. (1.6)
1
We use greek letters , , ..., taking values (0,1,2,3), to denote the space-time components of four
vectors and tensors.
4
For a test particle, there is a simple physical interpretation of invariant space-time
intervals measured along its world line, in terms of proper time measured by a clock in the
local rest frame of the particle. As world lines are time-like, any interval along a world line
satisies ds
2
< 0, and the proper time interval d measured by a clock at rest with respect
to the particle is given by
c
2
d
2
= ds
2
. (1.7)
The local matrix inverse of the metric is denoted by g

(x):
g

(x) g

(x) =

, (1.8)
where

is the Kronecker delta, and the Einstein summation convention has been used.
This convention implies that one should automatically perform a full summation over any
repeated upper and lower indices, like on the left-hand side of this equation.
From the gradients of the potentials, i.e. of the metric tensor, one constructs a quantity
known as the connection, with components

given by the expression:

=
1
2
g

) . (1.9)
At this point this expression may simply be considered a denition. It is a useful denition
because this combination of gradients of the metric occurs very often in equations in
general relativity. The deeper mathematical reason for this stems from the transformation
properties of expressions for physical quantities under local coordinate transformations.
However, it is not necessary to discuss these aspects at this point.
A physical argument for the importance of the connection is, that it gives directly the
gravitational acceleration of a test particle in a given coordinate system by the following
prescription, due to Einstein: the world lines of point-like test particles in general relativity
are time-like geodesics of space-time, the latter being considered as a continuous pseudo-
euclidean space
2
with metric g

. Time-like geodesics are lines along which the total proper


time between two xed events
_
d is an extremum; such world lines are solutions of the
equation of motion
d
2
x

d
2
+

dx

d
dx

d
= 0. (1.10)
A geometrical interpretation of the connection is therefore that it denes geodesics. As
concerns its dynamical interpretation, there is some similarity of eq.(1.10) with the Lorentz
force in electrodynamics in that the proper acceleration depends on the four velocity x

(we
use an overdot to denote a derivative w.r.t. proper time). However, whereas the Lorentz
force is linear in the four velocity, the gravitational acceleration depends quadratically
on the four velocity. The coecients of this quadratic term are given by the connection,
2
The term pseudo-euclidean here refers to the fact that the metric is not positive-denite, as it would
be in a true euclidean space; space-time has a lorentzian signature (, +, +, +), which implies that a
general metric g

has one negative (time-like) and three positive (space-like) eigenvalues.


5
which therefore directly determines the force. Notice, that the gravitational acceleration
does not involve the mass of the test particle. This is the mathematical content of the
famous equivalence principle, which states that in a xed gravitational eld all bodies fall
with the same acceleration, independent of their mass.
Eqs.(1.10) and (1.9) express the force on a test particle in terms of a eld (the connec-
tion), and the eld in terms of potentials (the metric). This provides the basis for the eld
theory of gravity formulated by Einstein in his theory of general relativity.
What is still lacking here of course is a set of equations determining the elds or
potentials in terms of given sources, like Maxwells equations allow one to compute the
electromagnetic elds in terms of charges and currents. General relativity provides such
equations; they are known as the Einstein equations, and they determine the gravitational
elds arising from given distributions of mass-energy and momentum. These distributions,
the sources of the gravitational elds, are described by a symmetric tensor T

(x) with 10
independent components, equal to the number of independent components of the metric;
T

is called the energy-momentum tensor. In order to give a concise formulation of


Einsteins equations, it is useful to rst introduce one more geometrical concept, that of
curvature.
1.4 Curvature
Curvature is a important concept, because it characterizes the space-time geometry in a
way which is basically coordinate independent. In order to illustrate the idea, we rst
present a heuristic discussion. Then we turn to mathematical denitions, which allow
us to write Einsteins equations in a compact way with a direct and elegant geometrical
interpretation.
Curvature is dened in terms of the properties of surfaces; it is a measure of their
deviation from a at plane. Curvature can be expressed conveniently in terms of the change
of a vector when transported parallel to itself around a loop enclosing a two-dimensional
surface element. In a plane, displacement of a vector parallel to itself will always result
in the same vector when you return to the starting point: there is absolute parallelism,
but no curvature. In a curved surface this is no longer true. Consider the example of a
sphere, a surface of constant curvature. If a vector is moved southward from the north pole
parallel to itself, along a great circle, it ends up on the equator cutting it at a right angle.
Now transport it along the equator over an arc of angular extension , always keeping at a
right angle. Finally, move it back to the north pole along a great circle, keeping it parallel
to the great circle and to itself. It then ends up at the north pole rotated by an angle
with respect to its original orientation. Thus parallel transport of a vector along a closed
loop on a curved surface does not leave the vector invariant in general.
In the example of the sphere, the rotation of the vector is proportional to the distance
traveled along the equator, as measured by the angle . This is also proportional to the
area enclosed by the loop consisting of the two sections of great circles through the north
pole and the arc of the equator. For a sphere with radius R, the change dV in a vector V
6
when the loop encloses an area dA = R
2
d is
dV = V d =
V
R
2
dA KV dA. (1.11)
The constant of proportionality K = 1/R
2
between the relative change in the vector dV/V
and the change in the area dA is a measure of the intrinsic curvature of the surface, known
as the Gaussian curvature. Note, that the Gaussian curvature is large when the radius
of the sphere is small, and that it vanishes in the limit R , as one would expect
intuitively.
This idea can now be generalized to situations in higher-dimensional spaces. Of course,
in a space of three or more dimensions, there are many surfaces one can draw through a
given point. Each of these surfaces may have its own curvature. Our rst conclusion is
therefore, that in general the curvature has many components.
Fortunately, it is only necessary to consider curvature components in surfaces which are
linearly independent. To this end, let us choose a locally non-degenerate coordinate system
x

and consider in particular the surfaces dened by keeping all coordinates constant except
two, say x

and x

.
Let d

= d

be the area of an oriented surface element in the x

-x

-plane which
contains the point with coordinates (x
0
). Any vector with components V

transported
parallel to itself around the loop dened by the circumference of this surface element now
changes to rst order by an amount
dV

=
1
2
R

(x
0
) V

. (1.12)
The constants of proportionality R

(x
0
) dene a tensor known as the Riemann curvature
tensor at the location (x
0
). A comparison with eq.(1.11) shows, that the Riemann tensor
is a higher-dimensional generalization of the Gaussian curvature. From eq.(1.12) it follows
immediately, that a space-time is completely at if and only if all components of the
Riemann curvature tensor vanish.
Dening parallel transport of a vector along a line element dx

in terms of the Riemann-


Christoel connection (1.9):
dV

= dx

V

, (1.13)
a direct computation for an innitesimal closed loop gives the result that the components
of the Riemann tensor can be expressed as
R

, (1.14)
where the square brackets denote an ordinary matrix commutator of two connection com-
ponents, considered as 4 4 matrices (

, the rst one with = , the second one with


= .
7
Using eq.(1.9) for the connection, we can ultimately express the curvature in terms of
the metric and its rst and second derivatives. It may then be checked from this explicit
expression that it identically satises the relation
R

+ R

+ R

= 0. (1.15)
This relation is known as the Bianchi identity. It is analogous to the homogeneous Maxwell-
equation, which implies that the electric and magnetic elds can be obtained from a four-
vector potential A

. In a similar way the Bianchi identity for the curvature tensor implies
that the curvature tensor can be obtained from a symmetric tensor g

.
We wish to emphasize here, that the curvature of space-time is not a measure for the
strength of the gravitational elds or forces; rather, it is a measure for the variation of the
gravitational elds in space and time, the gradients.
1.5 The Einstein equations
The Einstein equations are second order non-linear partial dierential equations for the
metric of space-time in terms of a given energy-momentum distribution. As a result, these
equations describe variations of the elds in space and time, rather than the elds them-
selves. In view of the role of curvature as a measure for such variations and the geometrical
interpetation of gravitational forces, it is quite natural that the Einstein equations should
admit an interpretation as equations for the curvature of space-time. However, rather than
the full Riemann tensor, it turns out that these equations involve only its contracted forms,
the Ricci tensor R

and the curvature scalar R:


R

= R

, R = R

= g

. (1.16)
Note that we adhere to the general convention that contraction with an inverse metric
can be used to raise an index of a vector or tensor, turning a covariant component into a
contravariant one; similarly contraction of an upper index on a vector or tensor with the
metric itself is used to lower it to make a covariant component out of a contravariant one.
The reason that only the contracted forms of the Riemann curvature appear in the
Einstein equations is, that in four-dimensional space-time the Riemann curvature tensor
has too many components, twice as many as the metric. Therefore xing the metric
components in terms of the full Riemann tensor would generally lead to an overdetermined
system of equations. In contrast, the Ricci tensor is symmetric: R

= R

, and has 10
independent components, precisely equal to the number of metric components to be solved
for. It is therefore to be expected that the Einstein equations take the form of an equation
for the Ricci tensor. Indeed, Einsteins equations for the gravitational elds can be written
in the simple form
R


1
2
g

R =
8G
c
4
T

, (1.17)
or equivalently:
8
R

=
8G
c
4
_
T


1
2
g

T
_
, (1.18)
where T = T

= g

. Notice the appearance of Newtons constant on the right-hand


side of the equations. Of course, in empty space-time T

= 0; the second version of


Einsteins equations (1.18) then implies that the Ricci tensor vanishes:
R

= 0. (1.19)
Space-time geometries which satisfy these conditions are some-times called Ricci-at. The
most interesting aspect of these equations is, that an empty space-time can still support
non-trivial gravitational elds: the Riemann curvature tensor can have non-zero compo-
nents even if its contracted form, the Ricci tensor, is zero everywhere.
This observation is important, because it lies at the heart of the theoretical arguments
for the existence of gravitational waves. Indeed, as we discuss in more detail later on, one
can nd non-trivial solutions of the Einstein equations representing lumps in the gravita-
tional eld which travel at the speed of light and transport a nite amount of energy and
momentum per unit of volume from one at part of empty space to another at part of
empty space.
There is a modied version of Einsteins equations in which the gravitational eld
itself creates a non-vanishing energy-momentum tensor, even in the absence of matter.
This gravitational eld behaves like a homogeneous continuum with constant density and
pressure (one of them negative), satisfying the equations
R

=
8G
c
4
g

, (1.20)
with a constant. Such an equation arises if the energy-momentum tensor is
T

= g

. (1.21)
The constant is known as the cosmological constant. Contraction of eq.(1.20) with the
inverse metric gives
R =
32G
c
4
. (1.22)
This shows that in the presence of a cosmological constant empty space-time has a constant
non-zero scalar curvature. Depending on the sign of the cosmological constant, eq.(1.20)
then admits solutions of Einsteins equations corresponding to empty space-times in contin-
uous expansion or contraction. This modied version of Einsteins equations is especially
interesting with regard to cosmological applications.
Of course in the general case there is a contribution to the energy-momentum tensor
from all matter and energy which happens to be present in the part of the universe we
wish to describe. The precise form of the energy-momentum tensor depends on the kind
of matter or radiation which constitutes the source of the gravitational eld. However,
9
it is very often justied and convenient to treat matter and radiation as an ideal uid,
the energy-momentum tensor of which depends only on the mass- and pressure-density
and its ow, as decribed by the local four velocity of an element of uid at location with
coordinates (x). Denoting the mass density by (x), the pressure density by p(x), and the
local four velocity by u

(x), the energy-momentum tensor then takes the form


T

= pg

+ (p + ) u

. (1.23)
Comparison with eq.(1.21) shows, that the cosmological constant can be reobtained from
an ideal uid with p = = .
1.6 The action principle
The Einstein equations (1.17) can be obtained from a variational principle; that is, they
follow by requiring an action S to be stationary under variations g

of the metric: S = 0.
The action for the gravitational eld equations, including a coupling to material energy-
momentum, is
S =
1
2c
_
d
4
x

g
_
c
4
8G
R + g

_
. (1.24)
Here

(x) represents the specic source term, but generally it is not the energy-momentum
tensor itself; rather, the energy-momentum tensor is given in terms of

and its covariant


trace =

by
T

1
2
g

. (1.25)
However, we consider

to be the independent variable, hence by at there is no contri-


bution from this tensor under variations with respect to the metric. Moreover, in eq.(1.24)
g(x) = det g

(x), and the minus sign is necessary because of the lorentzian signature of
the metric, which implies the product of its eigenvalues to be negative everywhere. The
proof of the action principle can be stated in a few lines, but it requires some intermediate
results which we discuss rst.
To begin with, in stead of considering variations of the metric g

, one can equivalently


use the variations of the inverse metric g

. Indeed, because g

, it follows that
g

+ g

= 0, (1.26)
or
g

= g

. (1.27)
Thus the variation of the metric can be expressed in terms of the variation of the inverse
metric and vice versa.
10
Next, we need a rule for computing the variation of the determinant g. This is obtained
from the formula
log det M = Tr log M, (1.28)
which holds for symmetric matrices M, as can easily be checked by diagonalizing M and
using the rule that the logarithm of the product equals the sum of the logarithms of the
eigenvalues. From this equation we then derive
1
g
(g) = g

= g

. (1.29)
Finally, we observe that the Ricci tensor R

is an expression involving only the connection

. Hence
R = g

+ g

, (1.30)
with the variation R

expressible purely terms of the variation of the connection


.
The total variation of the action now becomes
S =
1
2c
_
d
4
x

g g

__
c
4
8G
R

1
2
g

_
c
4
8G
R +
__

1
2
_
d
4
x

g
R

.
(1.31)
The last term is to be interpreted in the sense that we vary R with respect to the metric
only via its dependence on the connection. However, it turns out that if we insert the
expression (1.9) for the connection, this whole variation adds up to a total derivative. As
the action principle applies to variations which vanish on the boundary, this term does not
contribute to S. If we then require the variation of the action to vanish, we indeed obtain
the Einstein equations:
c
4
8G
_
R


1
2
g

R
_
+


1
2
g

=
c
4
8G
_
R


1
2
g

R
_
+ T

= 0.
(1.32)
The equation with the cosmological constant is obtained by making the replacement
g

= 2 in the action.
Chapter 2
Geodesics
2.1 Curves and geodesics
In this chapter we discuss how basic information about the geometry of spaces or space-
times generically manifolds can be obtained from studying their geodesics. The
geodesics of euclidean dierentiable manifolds are smooth curves of stationary proper
length, meaning that to rst order the proper length of an arc does not change under
small variations of the curve. More generally and more precisely, given a metric on a man-
ifold (euclidean or lorentzian) the geodesics are those smooth curves for which the proper
interval between any two xed points P and P

through which the curve passes is both


well-dened and stationary. The interval among the xed points can be either a maximum,
a minimum or a saddle point.
A direct reason for geodesics to be important is that time-like geodesics represent
the allowed trajectories of test particles in a xed background space-time. But in general
geodesics tell us interesting things about the symmetries and topology of a manifold, which
often have wider implications for physics.
First we derive the fundamental equation characterizing geodesics. The proper interval
s [C
PP
] of the arc C
PP
of an arbitrary smooth curve C between the two xed points
(P, P

) is given by the integral


s [C
PP
] =
_
C
PP

ds =
_
C
PP

dx

d
dx

. (2.1)
This holds for all geodesics of euclidean manifolds, and time- or space-like geodesics
of lorentzian manifolds; is a continuous parameter on some interval of the real line,
parametrizing the curve between the points P and P

in any convenient way, provided


the coordinates x

() of points on the curve are monotonic dierentiable functions of this


parameter: the label of any point must be unique and the tangent vector x

= dx

/d
must be well-dened. The parameter is then called an ane parameter. Note that the
integral does not depend on the particular choice of ane parameter: it is invariant under
dierentiable changes of parametrization

(). This freedom to choose for any


11
12
smooth parametrization of the curve is often useful, but requires some care in interpreting
results explicitly containing . As any invariant geometrical results with physical inter-
pretation can not depend on the choice of ane parameter, the best way to proceed is to
remove any -dependence from equations before drawing physical conclusions.
The proper interval s[C
PP
] is stationary if the integral does not change to rst order
under arbitrary smooth variations of the path while keeping the end points of the integral
xed:

_
C
PP

d w[x()] = 0, (2.2)
where the integrand is w[x()] =
_
|g

|. Now after a partial integration with van-


ishing boundary terms (the end points remain xed) the variation of the integrand can be
written
w = x

d
d
_
g

w
dx

d
_
+
1
2w
g

dx

d
dx

d
_
, (2.3)
provided w = 0. The sign on the r.h.s. is positive in euclidean spaces or for space-like
curves in lorentzian space-times, and negative for time-like curves in lorentzian space-
times. Light-like curves in lorentzian space-time need special consideration, as ds
2
= 0 is
equivalent to wd = 0.
The variational principle (2.2) now implies that for any smooth local variation x

the induced variation w must vanish if the curve x

() is to describe a geodesic. For


w = 0 we can divide the right-hand side of eq.(2.3) by w once more, and use the fact that
wd = ds, the innitesimal proper interval which is measured in the laboratory frame and
independent of the choice of ane parameter . We can also replace the partial derivative
of the metric in the second term by the connection; recall, that the connection was dened
in eq.(1.9) as

=
1
2
g

_
g

+
g

_
. (2.4)
This expression is actually the solution of the equation
Dg

Dx

=
g

= 0, (2.5)
stating that the metric is covariantly constant, a condition known as the metric postulate.
From the equation above it follows that
g

. (2.6)
The vanishing of w is now seen to imply a dierential equation of the form (1.10) for the
coordinates of a point moving on the curve as a function of the proper interval parameter
s:
13
d
2
x

ds
2
+

dx

ds
dx

ds
= 0. (2.7)
Note, that the above procedure eliminating the ane parameter is equivalent to a choice
of rate of ow of the ane parameter d equal to the proper interval ds measured in the
coordinate frame {x

}. If the signature of the manifold is lorentzian and the curve time-


like this ane parameter is the proper time , and eq.(2.7) is the equation of motion of a
test particle of arbitrary non-vanishing mass, in a xed space-time with ane connection

(x). Note also that the choice of d = ds implies an identity


g

dx

ds
dx

ds
= 1, (2.8)
where again the plus sign holds for euclidean or space-like lorentzian curves, and the minus
sign for time-like curves in lorentzian space-times. This is a constraint on the solutions of
the geodesic equation (2.7).
That this constraint is consistent with the geodesic equation follows by observing that
H(s) =
1
2
g

dx

ds
dx

ds
(2.9)
is a constant of motion, or more generally a constant of geodesic ow:
dH
ds
= g

dx

ds
_
d
2
x

ds
2
+

dx

ds
dx

ds
_
= 0. (2.10)
Finally light-like curves (also called null-curves) by denition obey
g

dx

d
dx

d
= 0. (2.11)
In this case the ane parameter can not be identied with proper time or distance. As
eq.(2.11) is to hold on all points of the light-like curve, the quantity
H() =
1
2
g

dx

d
dx

d
= 0 (2.12)
must be a constant of light-like motion. Therefore eq.(2.10) still holds with s replaced by
. From this it follows that it is correct to interpret the solutions of the geodesic equation
d
2
x

d
2
+

dx

d
dx

d
= 0, (2.13)
satisfying the constraint H() = 0 as light-like geodesics.
Remark. Although we have shown that the geodesic equation implies the constancy of
H() for all types of geodesics, the inverse statement is not true. In particular, for non-
null geodesics (euclidean, and space-like or time-like lorentzian) the more general condition
14
d
2
x

d
2
+

dx

d
dx

d
= B

(x)
dx

d
, (2.14)
with B

= B

an anti-symmetric tensor, is sucient for H to be conserved. This


happens for example for charged particles in an electro-magnetic eld with eld strength
F

:
B

(x) =
q
m
F

(x), (2.15)
where q/m is the charge-to-mass ratio of the particle. Similar extensions exists for particles
with spin or non-abelian charges.
For light-like geodesics the equation can be further generalized to read
d
2
x

d
2
+

dx

d
dx

d
= B

(x)
dx

d
+
dx

d
, (2.16)
with an arbitrary constant. It should be noted that if = 0, then H is only conserved
by virtue of the fact that it vanishes.
Summary
The results of this discussion an be summarized as follows. All geodesic curves, in euclidean
or lorentzian manifolds, satisfy the equation
d
2
x

d
2
+

dx

d
dx

d
= 0, (2.17)
where is an ane parameter. The quantity
H() =
1
2
g

dx

d
dx

d
, (2.18)
is conserved by the geodesic ow. In physical applications three cases are to be distin-
guished:
1. For euclidean manifolds and space-like geodesics of lorentzian manifolds one can take
d = ds, the proper distance; these geodesics are then characterized by the condition
2H(s) = 1.
2. Time-like geodesics of lorentzian manifolds can be parametrized by the proper time:
d = d; such geodesics satisfy 2H() = 1.
3. Light-like (null) geodesics admit no such identication of with invariant measures
of the arc length in an inertial frame (laboratory frame); null-geodesics are charac-
terized by H() = 0.
15
2.2 Canonical formulation
In D-dimensional space-time eq.(2.17) represents a set of D second order dierential equa-
tions in the ane parameter providing a covariant description of geodesics. It is possible
to replace this set by a set of 2D rst order dierential equations by going over to the
canonical formulation of the dynamics of test particles, in which the number of indepen-
dent variables is doubled. Introducing the momentum variables
p

() = g

(x)
dx

d
, (2.19)
the geodesic equation (2.17) becomes
dp

d
=

. (2.20)
The equations (2.19) and (2.20) constitute a pair of rst-order dierential equations equiv-
alent to the second order geodesic equation (2.17). A powerful result is now obtained
by observing that these two equations can be derived from the constant of motion H(),
eq.(2.18) as a pair of canonical hamiltonian equations with H() in the role of hamiltonian:
H() =
1
2
g

. (2.21)
With this denition
dx

d
=
H
p

,
dp

d
=
H
x

. (2.22)
Then any function F(x

, p

) on the 2D-dimensional phase space spanned by the coordinates


and momenta changes along a geodesic according to
dF
d
=
dx

d
F
x

+
dp

d
F
p

= {F, H} , (2.23)
where the Poisson bracket of two functions (F, G) on the phase space is dened generally
as
{F, G} = {G, F} =
F
x

G
p

F
p

G
x

. (2.24)
The anti-symmetry of the Poisson bracket automatically guarantees the conservation of
the hamiltonian H by
{H, H} = 0. (2.25)
A similar property holds for all quantities which are constant along geodesics:
dF
d
= 0 {F, H} = 0. (2.26)
16
Formal properties of the Poisson bracket include its anti-symmetry and its linearity in each
of the arguments:
{F,
1
G
1
+
2
G
2
} =
1
{F, G
1
} +
2
{F, G
2
} . (2.27)
Another important property is that it satises the Jacobi identity:
{F, {G, K}} +{G, {K, F}} +{K, {F, G}} = 0. (2.28)
This property implies, that the Poisson bracket of two constants of geodesic ow is again
a constant of geodesic ow: if {F, H} = 0 and {G, H} = 0, then
{{F, G} , H} = 0. (2.29)
As H is itself constant on geodesics, the above results are sucient to establish that the
constants of geodesic ow form a Lie algebra, with the Poisson bracket as the Lie bracket.
The denition of the Poisson bracket (2.24) is not manifestly covariant, and therefore
at rst sight it seems to destroy the build-in covariance of the tensor calculus of dierential
geometry, and of general relativity in particular. However, note that in our denition
the connection (2.4) is alway symmetric in its lower indices; therefore we can rewrite the
Poisson bracket formula as
{F, G} = D

F
G
p

F
p

G, (2.30)
where for scalar functions F and G on the phase space we dene the covariant derivative
D

F =

F +

F
p

. (2.31)
This equation preserves manifest covariance for scalar functions F and G; for example, if
J(x, p) = J

(x)p

, then
D

J =
_

_
p

= (D

J)

. (2.32)
The manifestly covariant form of the Poisson bracket can also be extended to all completely
symmetric tensors T

1
...n
of rank n by contracting all n indices with the momentum p

to
obtain the scalar
T(x, p) =
1
n!
T

1
...n
(x)p

1
...p
n
. (2.33)
Inserting this scalar into the covariant Poisson bracket, the bracket for a particular tensor
component T

1
...n
is obtained by taking, in the expression resulting on the right-hand side,
the coecient of the n-nomial in the momenta that one is interested in. For anti-symmetric
index structures a similar result can be achieved using dierential forms, or more generally
Grassmann algebras [18, 19, 20].
17
2.3 Action principles
The canonical formulation of geodesic ow presented here can be cast into the form of a
standard hamiltonian action principle. Namely, the canonical eqs.(2.22) dene the critical
points of the phase-space action
S =
_
2
1
d ( x p H(x, p)) . (2.34)
By using the hamilton equation
x

=
H
p

,
the momentum variable can be eliminated in favour of the velocity of geodesic ow, to give
the lagrangian form of the action
S =
_
2
1
d
1
2
g

(x) x

. (2.35)
We observe, that this is not the action for the geodesics that we started from, eq.(2.1). In
particular, it is not reparametrization invariant. A comparison shows, that the lagrangian
in (2.35) is actually the square of w() used earlier, eq.(2.2). Clearly, the critical points are
therefore the same, with the possible exception of light-like curves. Actually, the variational
principle for the original action was not even well-dened in the case of light-like curves.
In this section we examine somewhat more closely the relation between the various
action principles, and in particular the role of reparametrization invariance. It makes clear
how one can derive the lagrangian and hamiltonian actions presented above within a unied
framework.
The key to the general formulation is to incorporate reparametrization invariance in
an action of the type (2.35) [21]. This is achieved by introduction of a new variable
e(), a function of the ane parameter, which acts as a gauge eld in one dimension for
local reparametrizations of the worldline. More precisely, the variable e() by denition
transforms under reparametrizations

as
e() e

) = e()
d
d

. (2.36)
This is the transformation rule for the square root of a metric in one dimension (on a curve),
for which reason the variable e() is called the einbein. Written in a slightly dierent
way, eq.(2.36) states that e()d is dened to be invariant under reparametrizations. This
makes it straightforward to write a reparametrization-invariant form of the quadratic action
(2.35):
S[e; x] =
_
2
1
d
_
1
2e
g

(x)
dx

d
dx

d

c
2
2
e
_
. (2.37)
18
The last term, proportional to the constant c
2
, does not depend on the coordinates x

.
However, it is reparametrization invariant and its inclusion is important for reasons given
below.
Variation of the full action S[e; x] w.r.t. the coordinates x

reproduces the geodesic


equation (2.17) provided everywhere the dierentials d are replaced by ed. This replace-
ment makes the geodesic equation reparametrization invariant, as ed = e

. On the
other hand, the substitution e() = 1, which breaks reparametrization invariance, makes
the old and new equations fully identical. Such a substitution is allowed because it only
serves to dene the ane parameter in terms of quantities measurable by an observer in
the locally euclidean or lorentzian laboratory frame, like proper length or proper time. Of
course, such a choice necessarily breaks the reparametrization invariance, but it preserves
the physical content of the equations. It amounts to a choice of gauge for reparametriza-
tions in the true sense of the expression, with the result that modulo a constant the action
S[e; x] reduces to the simple quadratic action (2.35).
However, the action S[e; x] contains more information, as one can also consider its
variation w.r.t. the einbein e. Requiring stationarity under this variation gives a constraint
on the solutions of the geodesic equation, which after multiplication by 2e
2
takes the form
g

(x) x

+ e
2
c
2
= 0. (2.38)
With e = 1, comparison with the denition of H in eq.(2.9) shows that this constraint
expresses the constancy of the hamiltonian along a geodesic:
2H() = c
2
. (2.39)
For euclidean manifolds one can always choose units such that c
2
= 1, but for lorentzian
space-times c
2
= (1, 0, +1) corresponds to space-like, light-like or time-like geodesics
respectively; in the latter case the constant c then represents the speed of light, which is
unity when expressed in natural units.
Finally we derive the original reparametrization-invariant geodesic action (2.1) from
S[e; x]. This is achieved by using eq.(2.38) to solve for e, instead of xing a gauge:
e =
_
|g

/c
2
|. (2.40)
Substitution of this expression for e into the action S[e; x] then leads back (modulo a
possible sign) to the original action (2.1). Therefore we conclude that both this action and
the quadratic action (2.35) follow from the same starting point provided by eq.(2.37).
2.4 Symmetries and Killing vectors
Often space-times of interest possess special symmetries. Such symmetries are useful in
nding solutions of the geodesic equations, especially if they are continuous symmetries as
these generate constants of geodesic ow. This is the content of Noethers theorem applied
to the motion of test particles and its generalization to space-like curves; see for example
19
refs.[11]-[15], [18, 19]. In this section the role of continuous symmetries is explored in some
detail.
A manifold possesses a continuous symmetry, or isometry, if there are innitesimal
coordinate transformations leaving the metric invariant. Finite transformations can be
obtained by reiteration of the innitesimal transformations, and donot require separate
discussion at this point. Consider an innitesimal coordinate transformation characterised
by local parameters

(x):
x

x

= x

(x). (2.41)
For the line element ds
2
to be invariant the metric must transform as a tensor; therefore
by denition
g

(x) = g

(x

)
x

x

x

x

. (2.42)
On the other hand, invariance of the metric under a coordinate transformation implies
g

(x) = g

(x). (2.43)
Combining the two equations (2.42) and (2.43) to eliminate the transformed metric g

,
and inserting the innitesimal transformation (2.41), expansion to rst order in

leads
to the covariant condition
D

+ D

= 0. (2.44)
This equation is known as the Killing equation, and the vector elds dened by the solutions

(x) are called Killing vectors. The existence of a Killing vector is equivalent with the
existence of an innitesimal coordinate transformation (2.41) leaving the metric invariant.
Let us pause for a moment to reect on the interpretation of this result. If the coordinate
transformation is considered to be a passive transformation, the new coordinates represent
a relabeling of the points of the manifold: x

and x

in eq.(2.41) refer to the same point
P. Now consider the point P

, which in the new coordinate system has the same value of


the coordinates as the point P in the old coordinate system: x

(P

) = x

(P). Eq.(2.43)
states that the new metric at the point P

has the same functional dependence on the


new coordinates as the old metric on the old coordinates at the point P. Thus in the
new coordinate system the neighborhood of P

looks identical to the neighborhood of P


in the old coordinates. In particular it is possible to map the points in the neighborhood
of P one-to-one to points in the neighborhood of P

in such a way that all distances


(intervals) are preserved. As this is in fact a coordinate-invariant statement, it implies
a local indistinguishability of the two neighborhoods of the manifold (as long as it is
not endowed with additional structures). This then is what is implied by a continuous
symmetry of the manifold generated by the solutions of the Killing equation.
Now if two neighborhoods of the manifold are similar in the sense described above,
then this similarity should also extend to the geodesics passing through the points mapped
20
on each other by the symmetry operation. Indeed, as the new metric at the point P

has the same functional dependence on the new coordinates there as the old metric at P
has on the old coordinates, and not just the same value, the connections and curvature
components there also become identical after the coordinate transformation. The same
holds for the solutions of the geodesic equations. Thus the symmetry transformation maps
geodesics through the point P to geodesics through the point P

, and the geodesic ow in


the neighborhood of these points is completely similar.
The similarity of geodesic structure generated by symmetry operations implies the con-
servation of certain dynamical quantities under geodesic ow. This is most conveniently
discussed in the canonical formalism of the geodesic ow presented above in sect.2.2. The
quantity conserved is the generator of the symmetry transformations, obtained by contrac-
tion of the Killing vector with the canonical momentum:
J[] =

(x)p

. (2.45)
This is a scalar quantity, hence its value is coordinate invariant. It generates the innites-
imal coordinate transformations through the Poisson bracket (2.24):
x

(x) = {J[], x

} . (2.46)
One can similarly determine the variation of the momentum under the transformation
generated by J[]:
p

= {J[], p

}
=

J[] = T

(x)p

(2.47)
with T

. This represents a coordinate-dependent linear transformation of the


momentum.
That the quantity J[] is conserved now follows from the observation that the hamil-
tonian is invariant under the phase-space transformations dened in eqs.(2.46), (2.47).
Indeed, taking into account the canonical equations of motion (2.23) and (2.26), and using
the Killing equation (2.44) we obtain
dJ[]
d
= {J[], H} = 0. (2.48)
This result was to be expected, as the hamiltonian depends on the coordinates only through
the metric, which is invariant by denition of an isometry.
As the Killing equation (2.44) is linear it follows that the Killing vectors dene a linear
vector space: any linear combination of two Killing vectors is again a Killing vector. Let
the dimension of this vector space be r; then any Killing vector can be expanded in terms
of r linearly independent basis vectors e
i
(x), i = 1, ..., r:
(
i
) =
1
e
1
(x) + ... +
r
e
r
(x). (2.49)
21
Equivalently, any innitesimal symmetry transformation (2.41) depends linearly on r pa-
rameters
i
, to which correspond an equal number of conserved generators
J[] =
1
J
1
+ ... +
r
J
r
, where J
i
(x, p) = e

i
(x)p

. (2.50)
We now show that these generators dene a Lie algebra through their Poisson brackets.
The key observation is, that the Poisson bracket of any two constants of geodesic ow is
itself a constant of geodesic ow. This is implied by the Jacobi identity for the Poisson
bracket, which can be written as:
{{F, G} , H} = {{F, H} , G} {{G, H} , F} . (2.51)
Thus, if the Poisson brackets of F and G with the hamiltonian H vanish, then the left-hand
side vanishes as well, and {F, G} is itself a constant of geodesic ow.
A second important observation is, that for two constants of motion which are linear
in the momentum, like the J
i
, the Poisson bracket is also linear in the momentum:
{J
i
, J
j
} =
_
e

i
e

j
_
p

. (2.52)
As the {J
i
}
r
i=1
are supposed to form a complete set, the expression on the right-hand side
can be expanded in terms of the basis elements, with the result that with a certain set of
coecients f
k
ij
= f
k
ji
we have
{J
i
, J
j
} = f
k
ij
J
k
. (2.53)
An equivalent statement is that the basis vectors of the Killing space satisfy
e

i
e

j
= f
k
ij
e

k
. (2.54)
Eq.(2.53), combined with the observations that the J
i
form a linear vector space and that
the Poisson bracket is bilinear and anti-symmetric in its arguments complete the proof
that the generators of the symmetry transformations dene a Lie algebra with structure
constants f
k
ij
.
Of course the linear p.d.e.(2.54) encodes the same information. It can be interpreted
in terms of the Lie-derivative, which is the operator comparing the value of a function at
the point P with that at the point P

which has the same value of its coordinates after an


arbitrary transformation of the form (2.41) as the point P had before the transformation:
x

(P

) = x

(P). For a scalar (x), for which

(x

) = (x), this gives


L

[] (x) lin

(x) (x)] =

(x), (2.55)
where the lin

[Q] denotes the part of the expression Q linear in . For a vector v

(x), for
which
v

(x

) = v

(x)
x

x

, (2.56)
22
one nds similarly
(L

[v])

(x) lin

[v

(x) v

(x)] =

(x)

(x). (2.57)
For higher-rank tensors the construction of the Lie-derivative gives the result
(L

[T])

1
...
k
(x) lin

[T

1
...
k
(x) T

1
...
k
(x)]
=

1
...
k
(x)

1
T
...
k
(x) ...

k
T

1
...
(x).
(2.58)
Now comparing eqs.(2.54) and (2.57) we see that the Lie-algebra property of the Killing
vectors may be expressed as
L
e
j
[e
i
] = L
e
i
[e
j
] = f
k
ij
e
k
. (2.59)
Finally, in the phase space where by denition

= 0, we can also write the transfor-


mation (2.47) for the momentum as

i
p

{J
i
, p

} = (L
e
i
[p])

. (2.60)
Thus we have established a close relationship between Poisson brackets and Lie derivatives,
and between symmetries, Killing vectors and Lie algebras.
2.5 Phase-space symmetries and conservation laws
In the previous section it was established that continuous symmetries of manifolds are
generated by Killing vectors and imply conservation laws for certain dynamical quantities
(the generators J) along geodesics. However, not every constant of geodesic ow is neces-
sarily connected with a Killing vector. This is most easily seen from the example of the
hamiltonian itself, which is conserved but not of the form (2.45). Rather the hamiltonian
is quadratic in the momenta. This leads us to consider the possibility of conservation laws
for quantities which are higher order expressions in the momenta.
Let J(x, p) be any function on the phase space of coordinates x

() and momenta
p

(), associated with a geodesic, which is non-singular in the momenta. In others words,
we suppose it has an expansion of the type
J(x, p) =

k=0
1
k!
J
(k)
1
...
k
(x) p

1
...p

k
. (2.61)
Here the coecients J
(k)
can be taken to be completely symmetric in all k indices. We ask
under what conditions such a function can be a constant of geodesic ow:
dJ
d
= {J, H} = 0. (2.62)
23
To obtain the answer, insert the expansion (2.61) into this equation and compare terms at
equal order of powers of momentum. At zeroth order one nds

J
(0)
= 0, (2.63)
i.e. J
(0)
is a constant. Such a constant can always be absorbed in the denition of J and
is of no consequence. From now on it will be dropped.
At rst order one nds back the Killing equation (2.44):
D

J
(1)

+ D

J
(1)

= 0.
These terms we have already discussed in detail.
Finally, for k 2 we nd a generalization of the Killing equation for the co-ecients
J
(k)
1
...
k
(x):
D
(
1
J
(k)

2
...
k+1
)
= 0. (2.64)
The parentheses denote full symmetrization over all component indices enclosed. The so-
lutions of these generalised Killing equations are called Killing tensors. It is now obvious
that the hamiltonian is included, because the metric g

is covariantly constant, and there-


fore automatically a solution of eq.(2.64) for k = 2. Moreover, this solution always exists,
independent of the particular manifold considered. In addition, for specic manifolds (spe-
cic metrics g

) other solutions may exist as well. For example, in spherically symmetric


spaces the square of the angular momentum:
L
2
=
1
2
_
x
2
p
2
(x p)
2
_
, (2.65)
is conserved and quadratic in the momenta by construction. Other examples will be en-
countered later on.
Just as the Killing vectors, the Killing tensors generate symmetry transformations, not
of the manifold but rather of the corresponding phase space spanned by the coordinates
and the momenta (tangent vectors). In particular:
x

= {J(x, p), x

} = J
(1)
(x)

k=2
1
(k 1)!
J
(k)
2
...
k
(x)p

2
...p

k
. (2.66)
We observe that the transformations generated by the higher-order Killing tensors are
velocity dependent [18].
Finally we address the algebraic properties of the constants of geodesic ow. We have
already remarked, that the Poisson brackets of two such constants yields another constant
by the Jacobi identity (2.51). In general the rank of quantities on the left-hand side are
additive minus one:
_
J
(k)
, J
(l)
_
J
(k+l1)
. (2.67)
24
Now for l = 1 the rank of J
(k+l1)
is the same as that of J
(k)
. It follows, that
(i) for the constants of geodesic ow linear in the momentum, corresponding to Killing
vectors (k = l = 1), the algebra is closed in itself: it is a Lie-algebra G;
(ii) for any xed value of k, the brackets of a J
(1)
and a J
(k)
close on the J
(k)
; however, the
brackets of two elements J
(k)
generates an element of order (2k 1). Hence in general the
algebra (2.67) is not closed, unless one includes elements of arbitrary rank, i.e. all possible
k. This implies that once higher-order invariants appear, one may expect an innite series
of them. Sometimes this happens indeed, as with the Virasoro algebra for the conformal
transformation in two dimensions [22]. However, in many cases the series stops because
the higher order invariants are Casimir-type of operators, or they become trivial in that
they are just products of commuting lower-order elements (commuting here is meant in
the sense of vanishing Poisson brackets).
The Lie algebra G of Killing vectors is characterized by its structure constants f
k
ij
.
As the Poisson brackets of J
(k)
with all elements of the Lie algebra spanned by the J
(1)
are again of order k, one may choose a basis
_
e
(k)
1
...
k
A
(x)
_
for the Killing tensors of order
k, labeled by the index A, and decompose the right-hand side of the Poisson bracket again
in terms of the corresponding conserved charges J
(k)
A
:
_
J
(1)
i
, J
(k)
A
_
= g
B
iA
J
(k)
B
. (2.68)
This denes a set of higher-order structure constants. The Jacobi identities then imply,
that the matrices (T
i
)
B
A
= g
B
iA
dene a representation of the Lie algebra (2.53):
g
C
jA
g
B
iC
g
C
iA
g
B
jC
= f
k
ij
g
B
kA
[T
i
, T
j
]
B
A
= f
k
ij
T
B
kA
. (2.69)
Therefore the constants of geodesic ow
J
(k)
A
(x, p) =
1
k!
e
(k)
1
...
k
A
(x)p

1
...p

k
, (2.70)
span a representation space of the Lie algebra G for the representation {T
i
}. The discussion
of symmetries shows that group theory and Lie algebras can be important tools in the
analysis of geodesic motion and the structure of manifolds.
2.6 Example: the rigid rotor
The above concepts and procedures can be illustrated by the simple example of a rigid rotor,
which has a physical interpretation as a model for the low-energy behaviour of diatomic
molecules. As such the example also serves to emphasize the usefulness of geometric
methods in physics outside the context of general relativity.
Consider two particles of mass m
1
and m
2
, interacting through a central potential V (r)
depending only on the relative distance r = |r
2
r
1
|. The lagrangian for this system is
25
L =
m
1
2
r
2
1
+
m
2
2
r
2
2
V (r). (2.71)
Dening as usual the total mass M = m
1
+ m
2
and the reduced mass = m
1
m
2
/M, the
lagrangian is separable in the center of mass coordinates dened by
MR = m
1
r
1
+ m
2
r
2
, r = r
2
r
1
. (2.72)
In terms of (R, r) the expression (2.71) becomes
L =
M
2

R
2
+

2
r
2
V (r). (2.73)
As the center of mass moves like a free particle of mass M, it is convenient to work in
the rest frame of the center of mass: R = 0. Then we are only left with the problem of
describing the relative motion of the masses, described by the three coordinates represented
by the vector r.
Now suppose that the potential V (r) has a minimum for some separation r = r
0
, and
rises steeply for all values of r near this minimum. Then the rst excited vibration state of
the molecule has an energy well above the ground state, and at low temperatures the only
degrees of freedom that play a role in the dynamics are rotations of the molecule at xed
interatomic distance r
0
. Thus the distance is frozen out as one of the dynamical degrees of
freedom, and we are eectively left with only two coordinates: the angles (, ) describing
the relative orientation of the two masses; the potential V (r) = V (r
0
) is constant and may
be ignored. This leaves as the eective action of the system
L
eff
=

2
r
2
0
_

2
+ sin
2

2
_
. (2.74)
This is the lagrangian of a particle of mass moving on a spherical surface of radius
r
0
. In mathematical physicists parlance the eective low energy degrees of freedom (, )
are sometimes called the modular parameters or moduli of the system, and the sphere
representing the eective low-energy conguration space is called the moduli space.
Up to an overall scale factor r
2
0
, the lagrangian L
eff
represents an action of the form
(2.35) in a 2-dimensional space with euclidean signature, described by the coordinates
x
1
= , x
2
= and metric
g
ij
=
_
1 0
0 sin
2

_
. (2.75)
As the geodesics on a sphere are great circles, the great circles provide the general solution
to the equations of motion. The spherical symmetry implies that the system is invariant
under 3-dimensional rotations (rotations about each of three independent body axes of the
sphere) and hence there are three Killing vectors generating these rotations. The corre-
sponding constants of motion are the 3 components of angular momentum L = (L
1
, L
2
, L
3
),
given explicitly by
26
L
1
= sin p

cot cos p

,
L
2
= cos p

cot sin p

,
L
3
= p

.
(2.76)
They span the Lie-algebra so(3) of the 3-dimensional rotation group:
{L
i
, L
j
} =
ijk
L
k
. (2.77)
As the Poisson brackets of L with the hamiltonian vanish, the hamiltonian can depend only
on the Casimir constant of the rotation group, i.e. the total angular momentum squared L
2
.
A direct computation shows, that up to the constant potential the eective hamiltonian is
in fact proportional to the Casimir invariant:
H
eff
=
L
2
2r
2
0
. (2.78)
Chapter 3
Dynamics of space-time
3.1 Classical solutions of the gravitational eld equa-
tions
The Einstein equations (1.17) describe the gravitational elds generated by a specic dis-
tribution of material sources, as described by the energy-momentum tensor T

. In two or
three dimensions, these elds carry at best a nite number of physical degrees of freedom,
related to the sources and/or the topology [23]. In four or higher-dimensional space-time
the gravitational eld is a fully dynamical system with innitely many degrees of freedom.
In particular, dynamical gravitational elds can exist in the absence of material sources,
in empty space-time. In the geometrical interpretation of general relativity such elds
represent the intrinsic dynamics of space-time itself.
In eq.(1.12) the curvature tensor was dened in terms of the parallel transport of a
vector around a closed curve. A manifold is called at if after parallel transport around an
arbitrary closed curve in the manifold the image of any vector coincides with the original.
Therefore a necessary and sucient condition for a manifold to be at near a given point
is that the components of the Riemann tensor R

vanish there. This condition is more


general than to require that the metric takes the form
ds
2
= dx
2
0
+
n

i=1
dx
2
i
, (3.1)
with the sign depending on the euclidean or lorentzian signature of the manifold. Indeed, in
any dierentiable manifold one can always nd a local coordinate system in the neighbor-
hood of a given point, in which the line element can be diagonalized to this form, but this
is not generally the case globally. Even in cases in which it can be done globally, eq.(3.1) is
not the most general solution of the Einstein equations for at empty space-time, because
it is still allowed to perform an arbitrary local change of co-ordinates in this metric. The
atness-criterion introduced above, amounting to the vanishing of the Riemann-curvature,
is more useful as it holds in any co-ordinate system and in any topology.
27
28
The existence of non-trivial solutions of the Einstein equations with vanishing cosmo-
logical constant in matter-free regions of space-time is possible because vanishing of the
Ricci-tensor does not imply a vanishing of all components of the Riemann-tensor. A for-
mal mathematical criterion for distinguishing solutions of the Einstein equations in empty
space from those with material sources is through the Weyl curvature tensor, dened as
the completely traceless part of the Riemann tensor:
W

= R


1
(d 2)
(g

+ g

)
+
1
(d 1)(d 2)
(g

) R.
(3.2)
Here d is the dimensionality of the manifold. If we similarly dene W

as the traceless
part of the Ricci tensor:
W

= R


1
d
g

R, W

= W

, W

= 0, (3.3)
then we obtain a complete decomposition of the Riemann tensor in terms of traceless
components and the Riemann scalar:
R

= W

+
1
(d 2)
(g

+ g

)
+
1
d(d 1)
(g

) R.
(3.4)
This decomposition allows a classication of the solutions of the Einstein equations ac-
cording to which of the independent components (W

, W

, R) vanish and which dont.


In particular, metrics for which only the Weyl tensor is non-vanishing are solutions of the
source-free Einstein equations R

= 0. On the other hand, there also exist manifolds for


which the only non-zero component is the Riemann scalar R, whilst the Weyl tensor and
traceless Ricci tensor vanish: the n-spheres x
2
0
+x
2
1
+. . . +x
2
n
=
2
, and all two-dimensional
surfaces. For the n-spheres this follows essentially from the spherical symmetry, which
implies the absence of any prefered direction, in combination with the permutation sym-
metries between the components of the Riemann curvature tensor.
For two-dimensional manifolds, it is always possible to write the Riemann tensor as
R

=
1
2
g

R =
1
2
(g

) R. (3.5)
For the (torsion-free) riemannian and pseudo-riemannian manifolds we consider in higher
dimensions (d 3), the number of independent components of the Riemann tensor R

,
taking account of all its symmetries, is given by:
29
N[R

] =
d
2
(d
2
1)
12
. (3.6)
This is counted as follows: the Riemann tensor is anti-symmetric in each of the index
pairs [] and []. Hence R
[][]
may be considered as a square matrix of dimension
n = d(d 1)/1. Also as an (n n)-matrix it is symmetric under interchange of these
index pairs. As such it has n(n +1)/2 independent elements. Finally the cyclic symmetry
in the rst three (lower) indices of the fully covariant components of the Riemann tensor
expressed by the Bianchi identity (1.15) adds d(d1)(d2)(d3)/4! algebraic constraints,
eliminating an equal number of independent components. Combining these results one nds
precisely the number N[R

] in eq.(3.6) above.
For d 3 the number of components of the traceless Ricci tensor and Riemann scalar
is
N[W

] =
d(d + 1)
2
1. N[R] = 1. (3.7)
(for d = 2 the Riemann curvature tensor has only one independent component, correspond-
ing to the Riemann scalar). Therefore for d 3 the number of independent components
of the Weyl tensor is
N[W

] =
1
12
d(d + 1)(d + 2)(d 3). (3.8)
Clearly this vanishes for d = 3, and is positive non-zero only for d 4. Thus we nd
indeed that non-trivial dynamical solutions to the source-free Einstein equations can exist
only in space-times of dimensions d 4. As the minimal case d = 4 happens to be the
number of dimensions of our (macroscopic) world, we observe that for this case the Weyl
tensor and the Ricci tensor each have 10 algebraically independent components. Therefore
half of the components of the curvature are determined by the material sources, whilst the
other half describes the intrinsic dynamics of space-time.
In the geometrical interpretation of the gravitational eld and its dynamics, Einsteins
famous principle of the equivalence of accelerations and local gravitational elds can be
formulated as the hypothesis that space-time is a manifold, in each point of which one
can establish a locally at (cartesian) coordinate system of the form (3.1). Of course a
fully geometric interpretation of all of physics, and also the equivalence principle, breaks
down in those points or regions where material sources are located
1
. But also solutions
of the source-free Einstein equations can violate the equivalence principle locally if there
are singularities for example points where curvature invariants become innite and/or
the number of dimensions changes. Physics at these points is no longer described, at least
not completely, by the Einstein equations, and space-time is not necessarily a manifold in
the strict sense. In some cases it can still be a manifold with boundaries, where specic
1
Attempts to explain matter as a manifestation of physics in higher-dimensional space-time, as in
Kaluza-Klein models or the more sophisticated string theories, aim to redress this situation.
30
boundary conditions hold; but these boundary conditions then represent external data and
the gravitational eld is not determined by general relativity alone.
Nevertheless singular solutions are important because they are relevant for physics as
we observe it in the universe: cosmological solutions of the Friedmann-Robinson-Walker
type with a Big Bang-singularity, and black-hole solutions describing stellar collapse. The
singularities in these cases are not necessarily physical, as the short-distance physics may be
regulated by Planck-scale modications of the theory, which are presently unobservable.
However that may be, here we simply accept the possibility of singularities and discuss
examples of both non-singular and singular geometries arising as solutions of the Einstein
equations. They illustrate many interesting aspects of gravitational physics.
3.2 Plane fronted waves
To show that in four dimensions non-trivial gravitational elds can indeed propagate
through empty space-time, we now consider a class of solutions known in the literature as
plane-fronted gravitational waves [4, 5]. This name derives from the property that there
is a co-ordinate system in which these eld congurations (i) move with the speed of light
along a straight line; (ii) have a at planar wave front, and (iii) are of nite duration.
When discussing the physical properties of these solutions it will become clear that these
wave-like solutions actually behave more like dynamical domain walls.
We construct a planar wave solution with the direction of motion along the x-axis, and
the transverse directions dened by the (y, z)-plane. Thus we look for solutions of the
Einstein equations of the form
g

(x, t) =
_
_
_
_
_
1
1
0
0
f
2
(x, t)
g
2
(x, t)
_
_
_
_
_
. (3.9)
The corresponding space-time line-element is
ds
2
= c
2
dt
2
+ dx
2
+ f
2
(x, t)dy
2
+ g
2
(x, t)dz
2
. (3.10)
We observe the following properties of these metrics:
in the (x, t)-plane space-time is at (Minkowskian);
in the y- and z-directions space-time is generally not at, but the value of the metric
depends only on the distance along the x-axis and on time, not on the values of y or z
themselves. Hence in the co-ordinate system chosen the components of these gravitational
potentials are the same everywhere in the (y, z)-plane. In particular they do not fall o to
unity at spatial innity, showing that the space-time is not asymptotically Minkowskian at
large space-like distances. Of course, they share this property with plane electromagnetic
waves, which at a xed time also have the same amplitude everywhere in the transverse
directions;
31
the metric has no inverse if f = 0 or g = 0; then the connections are singular. We must
take special care when this happens.
With the Ansatz (3.9) for the metric one can compute the components of the connection
using eq.(1.9). Using the notation x

= (ct, x, y, z) for = (0, 1, 2, 3), the non-vanishing


components have been assembled in the following list:

2
02
=
2
20
=
f
t
cf
,
0
22
=
1
c
ff
t
,

3
03
=
3
30
=
g
t
cg
,
0
33
=
1
c
gg
t
,

2
12
=
2
21
=
f
x
f
,
1
22
= ff
x
,

3
13
=
3
31
=
g
x
g
,
1
33
= gg
x
.
(3.11)
The next step in solving the Einstein equations is to compute the components of the
Riemann curvature tensor using eq.(1.14). The only independent covariant components
R

= R

not identically zero are the following ones:


R
0202
=
1
c
2
ff
tt
, R
0303
=
1
c
2
gg
tt
,
R
1212
= ff
xx
, R
1313
= gg
xx
,
R
0212
=
1
c
ff
tx
, R
0313
=
1
c
gg
tx
,
R
2323
= fg(f
x
g
x

1
c
2
f
t
g
t
).
(3.12)
Before discussing the general solution of the Einstein equations for this class of metrics, we
rst use the above results to answer the question which of the metrics singled out by the
Ansatz (3.9) describe at Minkowski space. If we require all components of the curvature
tensor (3.12) to vanish independently, the most general solution for f(x, t) and g(x, t) is
f(x, t) = a + px + rct,
g(x, t) = b + qx + lct,
(3.13)
with the additional constraint
pq = rl. (3.14)
32
Thus metrics of the form (3.9), and corresponding line-elements (3.10), describe a at
space-time if f and g are linear functions of x and t, with the additional constraint (3.14)
between their co-ecients.
The components of the Ricci tensor R

now follow by taking a covariant trace as in


eqs.(1.16). The non-vanishing ones are
R
00
=
1
c
2
_
f
tt
f
+
g
tt
g
_
,
R
01
= R
10
=
1
c
_
f
tx
f
+
g
tx
g
_
,
R
11
=
f
xx
f
+
g
xx
g
,
R
22
= f
_
f
xx

1
c
2
f
tt
_
+
f
g
_
f
x
g
x

1
c
2
f
t
g
t
_
,
R
33
= g
_
g
xx

1
c
2
g
tt
_
+
g
f
_
f
x
g
x

1
c
2
f
t
g
t
_
,
(3.15)
In the absence of material sources Einsteins equations require the Ricci tensor to van-
ish. Assuming f and g themselves not to vanish identically, some manipulation of these
equations then leads to the following ve necessary and sucient conditions:
f
xx

1
c
2
f
tt
= 0,
g
xx

1
c
2
g
tt
= 0,
f
x
g
x

1
c
2
f
t
g
t
= 0,
gf
xx
+ fg
xx
= 0,
gf
tx
+ fg
tx
= 0.
(3.16)
The rst two equations are free wave equations in two-dimensional (x, t) space. They imply
that f and g can be decomposed into left- and right-moving waves:
33
f(x, t) = f
+
(x + ct) + f

(x ct),
g(x, t) = g
+
(x + ct) + g

(x ct),
(3.17)
where (f

(), g

()) and (f
+
(), g
+
() are functions only of the single lightcone variable
= x ct, and = x +ct, respectively. Each of these components then satises the wave
equation by itself, and at this stage they can be taken to be completely independent. The
only ambiguity here is that a possible constant term can be divided in an arbitrary way
between the left- and right-moving components of the solution.
If we now substitute this decomposition of f and g in the third equation (3.16), and
assume that the rst derivatives of the components with respect to u or v (i.e. f

and
g

) are all non-zero, then the only solutions of the Einstein equations correspond to at
Minkowski space. This can be seen by dividing the equation by g

+
g

to obtain
f

+
g

+
=
f

= , (3.18)
where on the right-hand side of the last equality is a nite, non-zero constant. This last
equality holds because the two previous expressions are functions of dierent variables,
hence they can be identically equal only if they are constant. It then follows, that
f

+
= g

+
, f

= g

. (3.19)
Inserting these results into the last two equations (3.16), addition and subtraction gives
the equivalent conditions
(g + f) g

+
= 0, (g f) g

= 0. (3.20)
The solution g

+
= g

= 0 implies f

+
= f

= 0. This leads to linear dependence of f and


g on x and t: they are of the form (3.13), (3.14) and correspond to at Minkowski space.
On the other hand, in case one of the second derivatives g

is non-zero, one of the co-


ecients (g f) must vanish; note that they cannot vanish simultaneously for non-zero
f and g. Hence eq.(3.20) gives either
f = g and g

+
= f

+
= 0, (3.21)
or
f = g and g

= f

= 0. (3.22)
But combining this with the results (3.19), we nd again that all second derivatives must
vanish and that the geometry is that of at space-time.
We conclude that in order to get non-trivial solutions, at least one of the rst derivatives
f

or g

must vanish. The last three equations (3.16) then imply that, after absorbing any
constants terms in the remaining components, the only solutions are
34
f(x, t) = f
+
(x + ct), g(x, t) = g
+
(x + ct), (3.23)
or
f(x, t) = f

(x ct), g(x, t) = g

(x ct), (3.24)
with the additional condition
f

f
+
g

g
= 0. (3.25)
Although f and g represent waves traveling either to the right or to the left at the velocity
of light, the non-linear character of the Einstein equations exemplied by the additional
condition (3.25) do not allow us to make arbitrary superpositions of these solutions. In-
terestingly, viewed as a two-dimensional theory in (x, t)-space f and g represent chiral
bosons. Wherever necessary in the following we chose to work with right-moving elds
f(x, t) = f

() and g(x, t) = g

() for deniteness.
An example of a simple solution of eq.(3.25) is [9]:
f(x, t) = cos k(x ct) = cos k,
g(x, t) = cosh k(x ct) = cosh k.
(3.26)
Note that at = (2n + 1)/2k the component f has a zero, whilst the solution for g(x, t)
grows indenitely with time or distance. However, this does neither imply that the geom-
etry is singular, nor that the gravitational elds become arbitrarily strong: the apparent
problems turn out to be coordinate artifacts. This point is discussed in detail in the next
section.
Let us consider a wave which reaches x = 0 at time t = 0, and extends to x = L, i.e.
the solution (3.26) holds for L 0. If |kL| < /2 the metric components f
2
and g
2
are strictly positive. For 0 space-time is at; we can choose a co-ordinate system in
which f = g = 1. Note that the wave solution (3.26) assumes these same values at = 0,
whilst the rst derivatives match as well. As a result the solutions are smoothly connected
and satisfy the Einstein equations even at the boundary = 0.
For L space-time is again at, but now we have to use the general solution (3.13):
f(x, t) = a + p(x ct),
g(x, t) = b + q(x ct),
(x ct) L. (3.27)
The co-ecients (a, b; p, q) in this domain of (x, t) values are to be determined again by
requiring the metric to be continuous and dierentiable at the boundary = L between
the dierent domains of the solution. This condition is sucient to guarantee that the
Einstein equations are well-dened and satised everywhere. Imposing these requirements
we nd
35
cos kL = a pL,
cosh kL = b qL,
k sin kL = p
k sinh kL = q.
(3.28)
For f and g in the domain (x ct) L we obtain the result
f(x, t) = cos kL + k (L + x ct) sin kL,
g(x, t) = cosh kL k (L + x ct) sinh kL.
(3.29)
Conclusion: we have constructed a solution of the Einstein equations in empty space which
interpolates between two Minkowski half-spaces, connected by a nite region of non-zero
curvature, where the gravitational elds have non-zero gradients and test particles are
accelerated with respect to the initial inertial frame. After passage of the wave the test
particles have acquired a nite, but constant velocity. This shows that they have been
accelerated, but the acceleration has ceased after t > (x + L)/c.
The plane-fronted wave constructed here is a well-dened solution of the equation R

= 0
everywhere, including the wave fronts x = ct and x = ct L, as the contributions of
the metric components g
yy
= f
2
and g
zz
= g
2
always compensate each other in eq.(3.25).
However the Riemann curvature itself, which coincides with the Weyl curvature in this
case, does not vanish everywhere. Indeed, we nd from eq.(3.12)
R
0202
= R
1212
= R
0212
= ff

,
R
0303
= R
1313
= R
0313
= gg

,
R
2323
= 0.
(3.30)
Inserting the explicit solution for f and g for each of the three regions x > ct, L <
x ct < 0 and x < ct L one nds that the curvature is zero in the two Minkowski
half-spaces before and after the wave, as expected, but nite and non-zero in the interior
of the wave. Indeed, the second derivatives f

and g

have a nite discontinuity on the


wave front of magnitude k
2
at x = ct, and k
2
cos kL, resp. k
2
cosh kL, at x = ct L.
This nite discontinuity then also appears in the curvature components, where it must be
attributed to a nite discontinuity in the Weyl curvature, and thus can not be attributed
to a (singular) distribution of sources. But the Riemann-Christoel connections, which
depend only on the rst derivatives of the metric, are continuous. Then the geodesics and
the motions of test particles in the space-time described by the wave are continuous, and
it follows that the jump in the Weyl curvature presents no physical inconsistencies.
36
3.3 Nature of the space-time
The planar wave constructed above can be characterised as a region of non-vanishing
curvature sandwiched between two at space-times and propagating with the speed of
light either in the positive or the negative x-direction such that one of the two at
space-time regions grows at the expense of the other. In this section we establish the
relation between the two at space-times, before and after the wave. The analysis shows
that, although they are of course related by a general co-ordinate transformation, they
are not related by a simple Lorentz transformation: a set of equidistant test particles
carrying synchronised ideal clocks and thus dening an inertial co-ordinate system before
the wave are neither equidistant nor synchronised in any inertial co-ordinate system after
the wave. In this sense the two Minkowski space-times on either side of the region of non-
zero curvature are inequivalent: in each the stability group of the vacuum, the Lorentz
group of rigid transformations leaving the in-product of four-vectors invariant in each of
the at regions is embedded dierently in the group of local co-ordinate transformations
which constitutes the symmetry of the full theory.
This situation is also encountered in scalar eld theories with broken symmetry, when
there is a boundary separating two domains with dierent vacuum expectation values of the
eld. One can even show that in some models these two forms of spontaneous symmetry
breaking of a scalar eld and of the gravitational eld are connected [27]. Therefore we
nd a new interpretation for the planar gravitational waves: they act as domain boundaries
separating dierent, inequivalent at-space solutions (vacua) of the Einstein equations.
These domain boundaries are dynamical: they are not static but move at the speed of light;
it is shown below that these particular solutions are block waves of constant curvature.
Regions of space-time enclosed by such dynamical boundaries clearly grow or shrink
at the speed of light. If we consider our planar wave solution as an approximation to
a more general domain-wall geometry with a radius of curvature large compared to the
characteristic scale of the region of space-time we are interested in, then we can use it to
describe the dynamics of gravitational domain structure, which has interesting implications
for the physics of the early universe.
Having sketched this general picture, I now turn to describe the sandwich structure
of the planar wave geometry more precisely. A convenient procedure is to introduce new
co-ordinates X

= (cT, X, Y, Z) related to the original x

= (ct, x, y, z) by
c T = ct +

2
,
X = x +

2
,
Y = fy,
Z = gz,
(3.31)
37
where in terms of the original coordinates x

(x

) = y
2
ff

+ z
2
gg

. (3.32)
This coordinate transformation takes an even simpler form when expressed in the (nor-
malized) light-cone variables (U, V ):
U = (cT X) /

2 = (ct x) /

2 = u,
V = (cT + X) /

2 = (ct + x + ) /

2 = v + /

2.
(3.33)
Indeed the line-element (3.10) now becomes
ds
2
= c
2
dt
2
+ dx
2
+ f
2
dy
2
+ g
2
dz
2
= 2dUdV 2K(U, Y, Z)dU
2
+ dY
2
+ dZ
2
,
(3.34)
with
K(U, Y, Z) =
f

f
Y
2

g
Z
2
=
g

g
_
Y
2
Z
2
_
. (3.35)
Both for t < x/c and t > (x + L)/c we have f

= g

= 0, and therefore the line element


(3.34) is manifestly at Minkowskian on both sides of the planar wave. However, inside
our particular wave solution one has
f

f
=
g

g
= k
2
. (3.36)
Then K = k
2
(Y
2
Z
2
), independent of U. The coordinate system X

with the line-element


(3.34) is the closest one to an inertial frame at all times for an observer at rest in the origin
one can construct. Indeed, the point x = y = z = 0 corresponds to X = Y = Z = 0
at all times, and both before and after the wave the X

dene a global inertial frame.


Furthermore, inside the wave the new coordinate system deviates from an inertial frame
only for large values of (Y
2
Z
2
). We also observe that in this coordinate system the line
element (3.34) never becomes singular, showing the previous zeros of det g, corresponding
to zeros of f(u), to be coordinate artifacts indeed.
Computing the various geometrical quantities: metric, connections and curvature com-
ponents, in the new coordinates produces the following summary of results:
The only non-vanishing components of the inverse metric are:
g
UV
= g
V U
= 1, g
V V
= 2K = 2k
2
(Y
2
Z
2
),
g
Y Y
= g
ZZ
= 1.
(3.37)
The only non-vanishing components of the connection are:
38

Y
UU
=
V
Y U
=
V
UY
= 2k
2
Y,

Z
UU
=
V
ZU
=
V
UZ
= 2k
2
Z.
(3.38)
The only non-vanishing covariant components of the curvature are
R
UY UY
= R
UZUZ
= 2k
2
. (3.39)
Thus in these coordinates the curvature components are constant inside the wave, and
we conclude that our particular planar wave solution is actually like a blockwave: at the
boundaries U = 0, U = L the curvature components jump from zero to the values
(3.39) and back to zero, whilst in between they are constant. This makes it easy to
visualize how our particular solution (3.26) can be used to construct general planar wave
solutions with polarization along the Y - and Z-axes: essentially it amounts to glueing
together suciently narrow block waves of xed curvature to obtain solutions in which the
curvature components vary according to some arbitrary prole.
The result (3.39) also allows one to verify once more by a one-line computation that
the Ricci tensor vanishes identically for these planar-wave geometries.
With these results in hand, our next aim is to establish the relation between the at half-
spaces before and after the wave. We have already observed that the origin x = y = z = 0
becomes the origin X = Y = Z = 0 for all times t or T, respectively. Next observe,
that the coordinate transformation (3.31) is not a Lorentz transformation: the coordinates
(t, x, y, z), dening an inertial frame before the passage of the wave, represent a non-inertial
frame after passage. As a check, note that the jacobian of the transformation is

= fg = 1, for = ct x = 0. (3.40)
Indeed, a system of synchronized, equidistant clocks at rest in the rst coordinate system,
which operationally denes an inertial frame before the wave arrives, is neither synchro-
nized nor equidistant in the nal inertial frame. Moreover, in general it cannot be made
so by a simple Lorentz transformation: the clocks have been accelerated with respect to
each other by the wave, and ultimately dene a non-inertial frame at late times. This is
veried by the explicit calculation of geodesics in the next section.
From this analysis it will be clear, that in principle it is possible to determine the
passage, and measure the characteristics, of a planar gravitational wave by measuring the
acceleration it imparts to test masses in the laboratory. Indeed, such measurements can
determine the functions (, f, g) from the relation between (t, x, y, z) and (T, X, Y, Z).
Although in this form the statement holds only for waves of the type discussed, the princi-
ples for other types of waves are the same, even if the functions characterizing the relation
between the initial and nal Minkowski spaces are generally dierent.
39
3.4 Scattering of test particles
In order to study the response of test masses to passing gravitational waves, it is necessary
to solve for the geodesics of space-time in the gravitational eld generated by the wave.
We are particularly interested in the worldlines of particles initially at rest in a specic
inertial frame (the laboratory). Observing their motion in the nal inertial frame gives the
scattering data, from which the properties of the gravitational wave can then be inferred
relatively easily.
In principle it is very easy to nd geodesics: they correspond to the worldlines of
particles in free fall. Consider therefore a particle at rest in the initial Minkowski frame;
its four co-ordinates are
x

() = (c,
1
,
2
,
3
), (3.41)
where the three vector = (
1
,
2
,
3
) xing the position of the particle is constant; note
also that by denition for a particle at rest in an inertial frame the co-ordinate time equals
the proper time .
In the co-ordinate system we use, a particle with co-ordinates (3.41) is actually always in
free fall, its co-ordinates in the x

-frame do not change if it enters a non-trivial gravitational


eld. Of course, in an arbitrary gravitational eld the x

-frame is not an inertial frame,


and even if at late times space-time is at again the particle will have acquired a nite,
generally non-vanishing four velocity. However, its position in the nal Minkowski frame
can be found straightforwardly by applying the co-ordinate transformation (3.31).
Let us rst conrm the above statements by a simple mathematical argument. Then we
turn to apply the result to nd the actual motion of the test particle in the nal Minkowski
frame. We begin by noting that in the x

-frame the metric for the gravitational wave is


always of the form (3.9), and therefore the only non-vanishing components of the connection
are those given in eq.(3.11). In particular, in the x

-frame the components



00
always
vanish.
On the other hand, for a particle with the worldline (3.41), the four velocity is purely
time-like and constant:
u

=
dx

d
= (c, 0, 0, 0). (3.42)
Then the four acceleration in this frame vanishes identically, as does the gravitational force:

= 0. (3.43)
As a result the geodesic equation (1.10) is satised, proving that (3.41) represents the
worldline of a particle at all times.
What we are particularly interested in, is the position and velocity of the particle at the
moment it comes out of the gravitational wave. It has already been remarked that the
origin = 0 remains at rest and coincides with the origin of the nal Minkowski frame.
Therefore at late times the origin has laboratory coordinates
40
X

0
() = (c, 0, 0, 0) . (3.44)
Let us compare this with the four co-ordinates of a particle away from the origin in the
(y, z)-plane: = (0,
2
,
3
) = 0. For such a particle the values of the metric components
at late times are
f(0, ) = a pc,
g(0, ) = b qc,
c L. (3.45)
Here (a, b; p, q) have the values quoted before, eq.(3.28). For one nds in the same
domain c L:
= ap
2
2
+ bq
2
3

_
p
2

2
2
+ q
2

2
3
_
c. (3.46)
As a result, the Minkowski co-odinates of this particle at late proper times are
X

1
() = (cT
1
, X
1
, Y
1
, Z
1
) , (3.47)
with
c T
1
=
1
2
_
ap
2
2
+ bq
2
3
_
+
c
2
_
2 + p
2

2
2
+ q
2

2
3
_
,
X
1
=
1
2
_
ap
2
2
+ bq
2
3
_
+
c
2
_
p
2

2
2
+ q
2

2
3
_
,
Y
1
= a
2
p
2
c,
Z
1
= b
3
q
3
c.
(3.48)
Note, that clocks which were originally synchronized same value of t = at early proper
time are no longer so after passage of the gravitational wave: T
1
is not equal to T
0
= at
late proper time. Before we can calculate the position of the test particle at synchronized
co-ordinate time T = , in the nal Minkowski frame, we rst have to correct for this
gravitationally induced position-dependent time shift.
To determine the instantaneous proper distance between the particle with worldine
X

1
() and the origin, we have to ask at what proper time the time co-ordinate T
1
equals
the time T
0
measured by a clock at rest in the origin:
T
1
( ) = , (3.49)
with the result
c =
2c + ap
2
2
+ bq
2
3
2 + p
2

2
2
+ q
2

2
3
. (3.50)
The position at this proper time,

X

1
= X

1
( ), can now be evaluated:
41

X
1
=
(ap
2
2
+ bq
2
3
) + c (p
2

2
2
+ q
2

2
3
)
2 + p
2

2
2
+ q
2

2
3
,

Y
1
=
2
_
(aq bp) q
2
3
+ 2 (a pc)
2 + p
2

2
2
+ q
2

2
3
_
,

Z
1
=
3
_
(aq bp) p
2
2
+ 2 (b qc)
2 + p
2

2
2
+ q
2

2
3
_
,
(3.51)
Finally we can also determine the velocity of the test particle:
v
c
=
1
2 + p
2

2
2
+ q
2

2
3
_
p
2

2
2
+ q
2

2
3
, 2p
2
, 2q
3
_
, (3.52)
and as a result the kinetic energy is
1
2
Mv
2
= Mc
2
(p
2

2
2
+ q
2

2
3
) (4 +p
2

2
2
+ q
2

2
3
)
(2 + p
2

2
2
+ q
2

2
3
)
2
. (3.53)
From this analysis we draw the conclusion, that not only in general test particles are
accelerated by the gravitational wave, but also that the acceleration and hence the nal
velocity depend on the position of the particle. Therefore no Lorentz transformation with
any xed boost velocity v can bring all test particles initially at rest w.r.t. one another
back to rest simultaneously after passage of the wave. This result explicitly demonstrates
the inequivalence of the initial and nal Minkowski space-times.
3.5 General planar waves
Having studied in great detail a very specic planar wave solution of the Einstein equations,
we now briey consider the general planar wave solution dened by metrics of the form
(3.34):
ds
2
= 2dUdV 2K(U, Y, Z)dU
2
+ dY
2
+ dZ
2
. (3.54)
It was mentioned below eq.(3.39) that solutions of this type can be found by glueing
together block waves of constant curvature. The general solution is obtained analytically
by allowing for arbitrary dependence of K(U, Y, Z) on its arguments. The expressions for
the corresponding connection components are

V
UU
= K
,U
,
V
Y U
=
Y
UU
= K
,Y
,
V
ZU
=
Z
UU
= K
,Z
, (3.55)
all others being zero. Up to permutation of indices, the only non-vanishing components of
the Riemann curvature tensor then turn out to be
42
R
UY UY
= K
,Y Y
, R
UZUZ
= K
,ZZ
,
R
UY UZ
= R
UZUY
= K
,Y Z
.
(3.56)
As a result the Ricci tensor takes the simple form
R
MN
= R
UU

U
M

U
N
, (3.57)
with
R
UU
= (K
,Y Y
+ K
,ZZ
) =
trans
K. (3.58)
Here
trans
denotes the laplacian in the transverse Y -Z-plane. It is convenient to introduce
complex co-ordinates in this plane:
= (Y + iZ) /

2,

= (Y iZ) /

2. (3.59)
The Einstein equations in empty space-time then reduce to the linear second order wave
equation for K:

trans
K = K
,

= 0, (3.60)
The general solution of this equation is a sum of holomorphic and anti-holomorphic func-
tions of (,

). Expanding in Fourier modes it can be written as
K
0
(U, ,

) =

n=0
_

dk
2

n
(k)e
ikU

n
+
n
(k)e
ikU

n
_
. (3.61)
From eqs.(3.56) it follows that the constant and linear terms in and

correspond to
a vanishing Riemann curvature tensor and hence describe at space-time. Therefore it
should always be possible to nd a coordinate system in which these terms are absent. In
this system the expansion (3.61) in Fourier modes then starts at n = 2.
The proper-time equations of motion for a point mass take the form
d
2
U
d
2
= 0,
d
2
V
d
2
=
2
K
,U
2
dY
d
K
,Y
2
dZ
d
K
,Z
,
d
2
Y
d
2
=
2
K
,Y
,
d
2
Z
d
2
=
2
K
,Z
.
(3.62)
Here we have already inserted the conservation law for the proper U-velocity
dU
d
= constant. (3.63)
The origin of proper time can be chosen on the light-cone U(0) = 0, with the result that
U = . It is then possible to exchange the proper-time for the light-cone time U, which
43
sometimes leads to simplications. Next, the conservation law for the particle hamiltonian,
generating time-like geodesic ow, gives

dV
d
+
2
K =
1
2
_
_
_
dY
d
_
2
+
_
dZ
d
_
2
+ c
2
_
_
. (3.64)
We can combine the last two equations to nd equivalent expressions for the proper-time
derivatives of (T, X). Adding the square of the X-component of the proper velocity to the
left- and right-hand side of eq.(3.64) gives
c
2
2
_
dT
d
_
2
=
1
2
_

2 +
dX
d
_
2
=
1
2
_
_
c
2
+
_
dR
d
_
2
_
_

2
K.
(3.65)
Dening the velocity in the laboratory frame as v = dR/dT, eq.(3.65) leads to the result
dT
d
=

_
1 2
2
K/c
2
1 v
2
/c
2
. (3.66)
This establishes the physical interpretation of K as the gravitational potential determining
the gravitational contribution to the time-dilation (redshift). We can invert this relation
to establish an expression for the constant in terms of the gravitational potential K and
the velocity components:

2 =
dT
d
_
1
v
x
c
_
=
1

1 v
2
/c
2
(1 v
x
/c)
2
+ K
. (3.67)
An immediate corrolary of this result is that the quantity
h = K +
1 v
2
/c
2
(1 v
x
/c)
2
(3.68)
is a constant of motion. This result takes the place of conservation of energy for test
particles in the eld of the gravitational wave. It results from the light-cone co-ordinate V
being cyclic.
Thus the equations for (U, V ) can be integrated in general for arbitrary K(U, Y, Z).
Then it only remains to solve the equations for (Y, Z) on the second line of eq.(3.62). Of
course, the explicit solutions of these equations depend on the specic form of K(U, Y, Z).
44
3.6 The gravitational eld of a light wave
So far we have considered gravitational waves in empty space-time. But in nature grav-
itational waves will most likely be associated with other physical phenomena, with the
dynamical behaviour of matter and energy. In fact we expect gravitational waves to ac-
company almost any time-varying distributions of matter and energy, provided the motions
are not too trivial (like uniform acceleration).
In this respect is instructive to study a relative simple and well-known process, which
is nevertheless quite non-trivial: the propagation of electro-magnetic waves. Waves in the
electro-magnetic eld provide an example of a time-varying energy and momentum distri-
bution in space-time, accompanied by a co-traveling gravitational wave eld. The physical
presence of this gravitational eld traveling at the speed of light can be demonstrated
in principle by considering the motion of an electrically neutral point-like test particle.
Although it is totally insensitive to the electro-magnetic wave, it does feel the associated
gravitational wave and will therefore start to oscillate. These orbits of neutral point-masses
are quite non-trivial; in general they are non-periodic, except for the case of zero-frequency:
a pair of constant crossed electric and magnetic elds generates pure harmonic motion of
a neutral point mass by its accompanying gravitational eld. This is a quite unexpected
and strange phenomenon, which may best be compared with the Josephson eect in su-
perconductivity, in which a constant potential generates an oscillating current. In fact we
will demonstate an even closer analogy in the case of scalar elds in the next section.
With the conventional energy-momentum tensor for the electro-magnetic eld, the
Einstein-Maxwell equations describing the coupling between gravity and electro-magnetism
read [24]
R

=
8
0
G
c
2
_
F


1
4
g

F
2
_
, (3.69)
and
D

= 0. (3.70)
Because the energy-momentum tensor of the electro-magnetic eld is traceless, the rst
equation implies that the scalar curvature vanishes, R = 0.
A free monochromatic plane electromagnetic wave of angular frequency = kc and
traveling in the x-direction is described in the Coulomb gauge by a vector potential

A = a sin k(x ct)), (3.71)


where a is a constant vector in the y-z-plane: a
x
= 0. The corresponding electric and
magnetic elds are

E =


A
t
= a cos k(x ct),

B =


A =

k a cos k(x ct). (3.72)
45
The wave-vector

k = (k, 0, 0) is orthogonal to a, hence

k a

= |ka| = ka, and

= E
0
cos k(x ct), E
0
= a, (3.73)
whilst

E

B = 0. The wave (3.71) remains a solution in general relativity provided the
covariant derivative in eq.(3.70) acts as an ordinary derivative. Now from the symmetry
of the Riemann-Christoel connection

and the anti-symmetry of the electro-


magnetic eld-strength tensor F

= F

it follows that the covariant divergence of


eq.(3.70) can be written explicitly as
D

= 0. (3.74)
Therefore the covariant derivative turns into an ordinary derivative if the contracted con-
nection vanishes:

= 0. This happens if the metric satises g = det g

= constant.
The symmetry of the plane-wave conguration implies that the energy momentum
of the electro-magnetic eld varies only in the x- and t-directions; thus we expect the
curvature to change only in those directions as well. Moreover, to keep the electric and
magnetic components of the wave perpendicular we require the metric in the y-z-plane to
be trivial. Finally, as the plane-wave conguration depends only on the variable (ct x)
it is convenient to switch to light-cone coordinates
u =
ct x

2
, v =
ct + x

2
. (3.75)
The above requirements can be satised by taking a metric of the form
ds
2
= 2dudv 2K(u, y, z)du
2
+ dy
2
+ dz
2
. (3.76)
This Ansatz for the metric is the same as that for free gravitational waves, eqs.(3.34),
(3.54). It has constant determinant g = 1 and is manifestly at in the transverse plane.
The non-vanishing components of the Riemann-Christoel connection are the same as
in eq.(3.55):

v
uu
= K
,u
,
v
yu
=
y
uu
= K
,y
,
v
zu
=
z
uu
= K
,z
. (3.77)
From these expressions one can explicitly verify that the contracted form of the connection
vanishes:

= 0; this implies that Maxwells equations are satised indeed.


As before, the non-zero components of the Riemann curvature are
R
yuyu
= K
,yy
, R
zuzu
= K
,zz
,
R
yuzu
= R
zuyu
= K
,yz
.
(3.78)
Then the Ricci tensor reads
R

= R
uu

, (3.79)
46
where again
R
uu
= (K
,yy
+ K
,zz
) . (3.80)
From eq.(3.72), the energy-momentum tensor of the electro-magnetic wave takes the form
F


1
4
g

F
2

=
_
F
2
ux
+ F
2
uy
_

u

= 2k
2
a
2
cos
2
_

2ku
_

u

, (3.81)
with
F
ui
= A
i,u
=
1
c

2
(E
i
+
ij
cB
j
) =

2 ka
i
cos

2ku, (3.82)
with
ij
the 2-dimensional anti-symmetric epsilon symbol in the transverse (y, z)-plane.
Comparing eqs.(3.81) and (3.79), and introducing as in eq.(3.59) the complex coordi-
nates = (y + iyz) /

2 and

= (y iz) /

2, we see that they match indeed provided


the single condition:
K
,

=
4
0
G
c
4
_

E
2
+ c
2

B
2
_
=
8
0
G
c
4
E
2
0
cos
2

2ku, (3.83)
is satised. Remarkably, this equation is linear in the metric component K. Therefore its
general solution can be given as
K(u, y, z) =
8
0
G
c
4

E
2
0
cos
2
_

2ku
_
+ f(u; ) +

f(u;

), (3.84)
where the rst term on the right is a special solution of the inhomogeneous Einstein equa-
tion, whilst f(u; ) is an arbitrary complex analytic function of u and , representing the
general solution of the homogeneous equation [25]. Thus the general solution of Einsteins
equations is seen to be a superposition of a special gravitational wave, generated by the
energy-momentum of the electro-magnetic eld, and an arbitrary free planar gravitational
wave traveling in the same direction.
It is straightforward to extend the solution to the case of a pulse of radiation described
by a vector potential

A(x, t) =
_
dka
k
sin k(x ct), (3.85)
where the direction of polarization determined by a
k
may depend on k provided it is
transverse to the axis of propagation:

k a
k
= 0. The energy momentum tensor of the
pulse is similar to that of the monochromatic wave (3.81):
F


1
4
g

F
2

=
_
F
2
ux
+ F
2
uy
_

u

=
u

1
c
2
_
dk
_
dq

E
k


E
q
_
cos

2(k + q)u + cos

2(k q)u
_
.
(3.86)
47
The corresponding metric is again of the form (3.76), with
K(u, ,

) = K
0
(u, ,

) +
+
4
0
G
c
4

_
dk
_
dq

E
k


E
q
_
cos

2(k + q)u + cos

2(k q)u
_
.
(3.87)
The physical nature of the special gravitational wave generated by the electro-magnetic eld
can be investigated by solving the equations for time-like geodesics, representing the orbits
of electrically neutral massive test particles. Parametrizing the geodesics by the proper
time and using an overdot to denote proper-time derivatives, the geodesic equation is
x

= 0. (3.88)
Consider the simplest case of a monochromatic electro-gravitational wave, with f(u; ) = 0;
then the metric component K can be written as
K
em
(u, x, y) =

2
0
2
_
1 + cos 2

2ku
_ _
y
2
+ z
2
_
. (3.89)
In this expression we have dened the constant

2
0
=
4
0
G
c
4
E
2
0
. (3.90)
Using this in the connections (3.77), the rst geodesic equation for u() becomes
u = 0. (3.91)
It follows that the momentum u is a constant of motion:
u = . (3.92)
In addition we can write down two other constants of motion. The rst one is the hamil-
tonian
H =
m
2
g

=
mc
2
2
, (3.93)
where we have used that for time-like geodesics the geodesic interval is measured by proper
time: ds
2
= c
2
d
2
. From the expression for the metric in eq.(3.76) it then follows that
v =
1
2
_
c
2
+ y
2
+ z
2
_
K. (3.94)
The other constant of motion is the angular momentum in the z-direction
J = y z z y. (3.95)
48
It remains to solve for y() and z(). The geodesic equations for these coordinates take
the form
y =

2
0
2
(1 + cos 2
1
) y, z =

2
0
2
(1 + cos 2
1
) z, (3.96)
with the frequencies
0
and
1
dened by

0
=
0

2,
1
= k

2. (3.97)
In the limit of static electric and magnetic elds, k =
1
= 0 with E
0
and B
0
xed, they
reduce to ordinary oscillator equations, with purely harmonic solutions of frequency
0
.
Hence we nd that a neutral test particle in static crossed electric and magnetic elds
executes a harmonic motion in the y-z-plane, as a result of an induced gravitational wave
[28]. As pointed out in the introductory remarks, this phenomenon as well as a similar
one for the gravitational eect of a scalar eld with constant gradient [27, 28], recall the
Josephson eect in superconductivity.
For
1
= 0 eqs.(3.96) are particular instances of the Mathieu equation, the solutions of
which are well-known [26]. In general they are not periodic. For the present application
it is convenient to proceed as follows. Introduce the new ane parameter =
1
; then
writing =
0
/
1
for the ratio of the two frequencies we obtain the equation of motion
d
2
y
d
2
+
1
2

2
(1 + cos 2) y = 0, (3.98)
and a similar equation for z(). In practice we always have 1 therefore we can expand
y() in powers of
2
:
y() =

n=0
c
n
()
2n
. (3.99)
This can be solved recursively for the c
n
():
d
2
c
0
d
2
= 0,
d
2
c
n+1
d
2
=
1
2
(1 + cos 2) c
n
. (3.100)
The solution of these equations is straightforward. We x the constants of integration such
that y(0) y
0
= c
0
(0) and y

(0)
y
= c

0
(0), and therefore
c
k
(0) = c

k
(0) = 0, k = 1, 2, ... (3.101)
Here the prime is used to denote a derivative with respect to . The solutions then take
the form
y() = y
0
a() +
y
b(),
z() = z
0
a() +
z
b(),
(3.102)
49
where the universal functions a() and b() contain all the dependence on the ane pa-
rameter . They possess the expansions
a() = 1 +

2
4
a
1
() +

4
16
a
2
() + ...
b() =
_
1 +

2
4
b
1
() +

4
16
b
2
() + ....
_
,
(3.103)
with
a
1
() =
1
2

2
+
1
2
cos 2,
a
2
() =
25
32
+

2
4
+

4
6
+
_
3
4


2
2
_
cos 2 + sin 2 +
1
32
cos 4,
b
1
() =
1
2


2
3

sin 2
2
+
1
2
cos 2,
b
2
() =
45
32


2
4
+

4
30

_
5
2
+
2
_
sin 2
2
+
_
5
4


2
6
_
cos 2

3
16
sin 4
4
+
1
32
cos 4.
(3.104)
A non-trivial check on this result follows from the conservation of the x-component of
angular momentum. Its value is xed by the initial positions and velocities as
J = y
0
v
z
z
0
v
y
, (3.105)
where v
i
=
1

i
. It is constant because the functions a() and b() satisfy the relation
a b

b a

= 1. (3.106)
This may be checked order by order in powers of
2
. In particular, one nds
a
1
a

1
+ b
1
+ b

1
= 0,
a
2
a

2
+ b
2
+ b

2
+ a
1
b
1
+ (a
1
b

1
b
1
a

1
) = 0.
(3.107)
Thus the conservation of angular momentum relates the functions a() and b().
The expansion (3.99) assumes < 1. From the explicit results for a() and b() we
can infer that the coecients remain small for values of such that =
0
< 1, i.e. for
proper times < 1/
0
. Numerically
50

0
(rad/sec) = 0.29 10
18

2
c
E
0
(V/m). (3.108)
For non-relativistic motion

2 c, whilst the highest electric elds in the laboratory


are generated by lasers, with E
0
10
11
V/m [29]. This limits
0
in practice to values
less than 10
7
rad/sec, and therefore almost always
0

1
. Note that on top of the
slow oscillation we observed for the case of static (zero frequency) electro-magnetic elds,
there is now a modulation with frequencies 2n
1
, where n is an integer number, but unless

1

0
the amplitude of these oscillations is very small.
A theoretical upper limit on the frequency
0
is provided by quantum electro-dynamics,
as the vacuum breaks down due to electron-positron pair creation for values of the order
[30, 31]
E
crit
=
m
e
c
2
e
e
= 4.1 10
18
V/m. (3.109)
Here m
e
is the electron mass, and
e
its reduced Compton wave length; for details we refer
to ref.[32]. The critical eld strength corresponds to a frequency upper limit of
0
1
rad/sec.
From the solution (3.102) of the equations of motion it follows that a point mass at
rest in the origin remains at rest at all times. Now clearly, the choice of origin in the
y-z-plane is arbitrary due to translation invariance. Therefore the non-trivial nature of
the solution becomes manifest only if we compare the relative motion of test particles.
Fortunately, using one test particle initially at non-zero (y
0
, z
0
) and one at rest in the
origin, our solutions are seen to determine the relative acceleration between these point
masses. As a result they provide a true measure for the curvature of space-time generated
by a light-wave.
3.7 Symmetry breaking as a source of gravitational
oscillations
The planar gravitational wave solutions described earlier in this chapter as block-waves,
arise as transition regions between at domains of space-time, which are related by a
coordinate transformation that cannot be reduced to a (linear) Lorentz transformation.
We have therefore interpreted them as domain boundaries between regions of space-time
trapped in inequivalent classical vacua of the gravitational eld (in the absence of a cos-
mological constant). This is possible because at space (zero curvature) is associated with
a value of the gravitational eld, the metric g

, which in the appropriate Minkowskian co-


ordinate system (Minkowski gauge) is constant, non-vanishing and Lorentz-invariant, but
not invariant under general coordinate transformations. In this sense the general coordi-
nate invariance of the Einstein equations is broken spontaneously by the vacuum solutions
representing at Minkowski space-time.
51
A domain structure arising from the presence of degenerate but inequivalent classical
vacua is a well-known consequence of spontaneous symmetry breaking in scalar eld theo-
ries. In this section we take up this aspect of the the eld theory and show, that dynamical
domain boundaries in scalar eld theories (scalar-type shock waves, if one prefers) are also
sources of planar gravitational waves of the kind analysed in detail above [27]. This is
another instance of the general observation that almost any type of dynamical situation
with time-dependent mass-energy distributions is accompanied by gravitational elds.
The simplest scalar eld theory with a continuously degenerate set of classical vacua is
that of a complex scalar eld with a mexican hat type of potential, as described by the
lagrangian
1

g
L = g



2
_

_
2
, (3.110)
In this lagrangian hc is a dimensionless coupling constant and is an inverse length
( h/c is a mass). In the natural system of units in which h = c = 1 we can take to be
dimensionless and to represent a mass. In the following equations one then can substitute
c = 1 everywhere.
This lagrangian (3.110) is invariant under rigid U(1) phase shifts e
i
, with
constant. As a result for
2
> 0 there is an innitely degenerate set of solutions of the
equations of motion minimizing the potential (refered to as classical vacua), of the form
=

e
i
, (3.111)
with an arbitrary real constant. Clearly the solutions themselves are not invariant un-
der the U(1) transformations: the classical vacuum solutions break the U(1) symmetry
spontaneously. However, because of the U(1) symmetry of the dynamics these solutions
are physically indistinguishable, and the absolute value of is unobservable. On the other
hand, variations in the vacuum angle between dierent regions of space-time can have
observable eects [33, 34].
Parametrizing the complex scalar eld as (x) = (x)e
i(x)
/

2, the lagrangian becomes


1

g
L =
1
2
g



8
_

2
2

_
2

1
2

2
g

. (3.112)
This shows that in the spontaneously broken mode of the U(1) symmetry the theory
desribes actually two real scalar elds: the eld =
_
2/, obtained by shifting by its
classical vacuum value
_
2/, with a mass (or inverse coherence length) equal to , which
can become arbitrarily large; and the strictly massless eld =
_
2/, representing the
Goldstone excitations arising from spontaneous symmetry breaking [33, 34].
Clearly at energies small compared to the scale set by the mass of the eld the
excitations of this massive eld play no role and one only has to consider the Goldstone
eld , described by the eective lagrangian
52
1

g
L =
c
4
16G
R +
1
2
g

, (3.113)
where we have added the Einstein-Hilbert action for the gravitational eld as this also de-
scribes a low-energy degree of freedom. Note that with the potential dened in eqs.(3.110),
(3.112) there is no cosmological constant in the presence of the vacuum value of the scalar
eld, =
_
2/ or equivalently = 0.
The eld equations derived from this lagrangian are the massless Klein-Gordon equation
for the Goldstone eld :
2
cov
=
1

gg

= 0. (3.114)
and the Einstein-equations with a source term from the energy-momentum of the Goldstone
eld:
R

=
8G
c
4

. (3.115)
This system of equations admits exact planar wave solutions similar to those of the electro-
magnetic waves discussed above. Using the same Ansatz (3.54), (3.76) for the metric we nd
running planar gravitational waves described by a metric component g
uu
= 2K(u, y, z)
in combination with a scalar wave (u) running in the same direction:
2
cov
=
_
2
u

v
2
v
K
v
+
2
y
+
2
z
_
= 0, (3.116)
because both K and are independent of v. The inhomogeneous Einstein equation now
takes the form
R
uu
=
trans
K =
8G
c
4
(
u
)
2
, (3.117)
with the solution in complex transverse coordinates (,

):
K(u, y, z) =
4G
c
4
(
u
)
2

+ f(u; ) +

f(u;

), (3.118)
where the (anti-)holomorphic terms K
0
= f(u; ) +c.c. denote the free gravitational waves
that are allowed in the background. As the functional dependence of (u) on u is quite
general, we do not have to restrict ourselves to monochromatic waves, but can consider
arbitrary wave packets.
The net eect of a scalar wave on the motion of test particles is the same as that of an
electro-magnetic wave. For the simple case of a monochromatic wave of wave number k:
(u) =
0
sin

2ku,
u
(u) = k
0

2 cos

2ku, (3.119)
the gravitational potential is of the form (3.89) with
53

0
2
=
4G
c
4
k
2

2
0
. (3.120)
It follows that the solutions of the equations of motion for a point mass in a scalar wave is
also given by expressions (3.102)-(3.104).
In addition we also observe that scalar elds with zero frequency and constant gradient

u
(u) = 2/

represent a simple -wave


(u) =
0
e
iu

2
=

e
i(xct)
. (3.121)
Such a wave can interpolate between two xed values of the Goldstone eld on a nite
interval L x ct 0:
(u) =
0
, for u 0;
(u) =
0
+
2u

, for 0 u L/

2;
(u) =
1
=
0
+ L

, for u L/

2.
(3.122)
The wave number of the scalar eld (u) is given by the slope of in eq.(3.122) as
=

1

0
L

2

2

=

1

0
L/

2
. (3.123)
From the geodesic equation (3.96) with
1
= 0 we nd, that in the scalar eld with constant
non-zero gradient in the region 0 u L/

2 a point mass will execute a harmonic


motion in response to the gravitational eld generated by the uniform scalar potential
dierence. This response is the same as to the gravitational eld of a uniform static
electro-magnetic eld. But here it is generated by a Goldstone eld of constant growth,
which really represents a constant change in the phase of a complex scalar eld in the
broken symmetry mode; as such the oscillations driving test masses through the associated
gravitational eld represent not only a mathematical, but also a physical analogue of the
Josephson eect in superconductivity.
From eqs.(3.118) and (3.78) it follows that these simple -waves on the nite interval
0 u L/

2 represent block-waves of constant curvature


R
uyuy
= R
uzuz
=
1
2
R
uu
=
8G
c
4
_

0
L
_
2
=
16G
c
4

. (3.124)
We have seen before that such block waves can be interpreted as dynamical gravitational
domain boundaries; we observe that in this case they coincide with analogous domain
boundaries of the scalar eld. Stated dierently, the transition between dierent scalar
vacua also induces a transition between dierent gravitational vacua.
54
The simple treatment in terms of block waves can be regularized by considering more
sophisticated eld congurations interpolating between dierent vacua; for example the
regular scalar eld
(u) =
1
2
(
1
+
0
) +
(
1

0
)

arctan
_
2

0
_
, (3.125)
interpolates between
0
and
1
; it also generates a smooth gravitational eld:
R
uyuy
= R
uzuz
=
1
2
R
uu
=
16G
c
4

1
_
1 +

2

4
2
u
2
(
1

0
)
2
_
2
. (3.126)
The geodesic equations describing the motion of test masses can be solved analytically
in closed form for this space-time geometry. Observing that as before u = = const.,
the equations of motion in the transverse plane can be written conveniently in terms of u
rather than proper time as ane parameter. Then
d
2
y
du
2
=

2
0
y
(1 +
2
u
2
)
2
,
d
2
z
du
2
=

2
0
z
(1 +
2
u
2
)
2
, (3.127)
where the parameters in the equation are given by the explicit expressions

2
0
=
16G
c
4

,
2
=
4
2

2
(
1

0
)
2
. (3.128)
We will also dene the ratio of these quantities as a new parameter

2
=

2
0

2
. (3.129)
We give the solution of the equation for y(u), that of z(u) being similar:
y(u) =

1 +
2
u
2
_
y
0
cos
_
_
1 +
2
arctanu
_
+
+
v
y

_
1 +
2
sin
_
_
1 +
2
arctan u
_
_
_
.
(3.130)
Here y
0
and v
y
represent the value of the y-coordinate and the velocity in the y-direction
at u = 0. Note that the velocity at u = is given by
lim
u
dy
du
= lim
u
1

dy
d
=
_
_
y
0
cos
_

2
_
1 +
2
_
+
v
y

_
1 +
2
sin
_

2
_
1 +
2
_
_
_
.
(3.131)
55
Clearly the change in proper velocity depends on the position y
0
:
u
y
=
dy
d

dy
d

= 2y
0
cos
_

2
_
1 +
2
_
. (3.132)
From this we see once more that a system of particles at xed distances and at rest
w.r.t. each other at some early time
1
will be scattered by the wave and have
all dierent velocities at a very late time
2
: no Minkowski frame at early times
remains a Minkowski frame at late times. This conrms our previous conclusion that
the gravitational vacuum states, the Minkowski spaces before and after the wave, are
inequivalent and that space-time can have a gravitational domain structure originating in
the breaking of general co-ordinate invariance down to Poincare invariance, in addition to
a domain structure originating in broken internal symmetry.
56
Chapter 4
Black holes
4.1 Horizons
The example of gravitational shock waves/domain walls has shown that space-time in
four dimensions is a truely dynamical arena for physics even in the absence of matter,
although adding matter and other elds of force for example in the form of scalar or
electromagnetic elds makes the possible structures even richer. In this chapter we
turn to another class of very interesting structures in four-dimensional space-time: static
or quasi-static solutions of the Einstein equations which represent gravitating extended
bodies, whose static elds become strong enough to capture permanently anything that
gets suciently close to the central core. As these quasi-static objects even capture light
they are called black holes, but it has turned out that their properties are much more
interesting and peculiar than this ominous but dull sounding name suggests.
Black holes possess two characteristic features. The rst, which has given them their
name, is the existence of a horizon: a closed surface through which objects can fall without
ever being able to return. The second is the presence of a singularity, a locus of points
inside the horizon where the curvature becomes innite. It has even been conjectured by
some authors that these two properties are so closely linked that any space-time singularity
should always be hidden behind a horizon. This conjecture bears the name of the Cosmic
Censorship hypothesis. Although as an unqualied mathematical theorem the hypothesis
is certainly not correct, it could be true for singularities that arise in realistic processes
from the gravitational collapse of macroscopic bodies, like imploding massive stars.
In this chapter we restrict our attention to the special black-hole solutions that might
be called classical eternal black holes, in the sense that at least outside the horizon
they describe stationary gravitational elds, and do not require for the description of their
behaviour the inclusion of their formation from collapsing matter. In fact they are solutions
of the source-free Einstein or coupled Einstein-Maxwell equations and are characterised for
an outside observer completely in terms of their mass, their angular momentum and their
electro-magnetic charge. One might alternatively think of them as black holes that have
been formed in the innite past and settled in a stationary state in which all trace of their
57
58
previous history has been lost, e.g. by emitting gravitational radiation.
Because of this restriction these lectures do not include a detailed account of the process
of gravitational collapse as studied by astrophysicists, but they do allow one to isolate and
study in detail the special aspects of the physics of black-holes. It may even be that these
very special simple solutions are relevant to elementary particle physics near the Planck
scale, but that leads one to consider the problems of quantum gravity, a subject we will
not deal with at length in these lectures. The interested reader can nd more information
on many aspects of black-hole structure and formation in the literature [35, 36].
4.2 The Schwarzschild solution
The simplest of all stationary black-hole solutions of the source-free Einstein equations is
that for the static spherically symmetric space-time, asymptotically at at spatial innity,
rst described in the literature by Schwarzschild
1
[37]. We present a derivation of this
solution making full use of the symmetries and their connection to Killing vectors explained
in chapter 2.
In view of the spherical symmetry we introduce polar coordinates (r, , ) in three-
dimensional space and add a time coordinate t, measuring asymptotic Minkowski time at
r . The Schwarzschild solution is obtained by requiring that there exists at least one
coordinate system parametrized like this in which the following two conditions hold:
(i) the metric components are t-independent;
(ii) the line-element is invariant under the three-dimensional rotation group SO(3) acting
on three-vectors r in the standard (linear) way.
The rst requirement is of course equivalent to invariance of the metric under time-shifts
t t + t. Thus formulated both conditions take the form of a symmetry requirement.
In chapter (2) we discussed how symmetries of the metric are related to Killing vectors.
Rotations of three-vectors are generated by the dierential operators L of orbital angular
momentum:
L
1
= sin

cot cos

, L
2
= cos

cot sin

,
L
3
=

.
(4.1)
Similarly, time-shifts are generated by the operator
L
0
=

t
. (4.2)
According to the discussion in sect.(2.4) the coecients of the symmetry operators L
A
=
L

A

are to be components of a set of Killing vectors L


A
, A = 0, 1, 2, 3:
1
The solution was found independently at the same time in [38].
59
D

L
A
+ D

L
A
= 0. (4.3)
These can equivalently be characterised as a set of constants of geodesic ow as in eq.(2.45):
J
A
= L

A
p

. (4.4)
The relation with the metric is, that the Poisson brackets of these constants of geodesic
ow with the hamiltonian are to vanish:
{J
A
, H} = 0, H =
1
2
g

. (4.5)
The general solution of these conditions for the hamiltonian H is t-independent, whilst it
can depend on the angular coordinates only through the Casimir invariant
L
2
= p
2

+
p
2

sin
2

. (4.6)
It follows that the metric must be of the form
2
ds
2
= h
2
(r)dt
2
+ g
2
(r)dr
2
+ k
2
(r)d
2
, (4.7)
where d
2
= d
2
+sin
2
d
2
is the angular distance on the unit sphere, and the coecients
h(r), g(r) and k(r)) are functions of the radial coordinate r only. Indeed, the hamiltonian
derived from this metric is
H =
1
2h
2
p
2
t
+
1
2g
2
p
2
r
+
1
2k
2
_
p
2

+
1
sin
2

p
2

_
=
1
2h
2
p
2
t
+
1
2g
2
p
2
r
+
L
2
2k
2
.
(4.8)
Note that the set of functions (h, g, k) can not be unique, as one can always perform a
coordinate transformation r r(r) to another metric of the same general form (4.7) with
coecients (

h, g,

k). There are several standard options to remove this freedom; to begin
with, we choose one in which the spherical symmetry is manifest and the spatial geometry
is as close to that of at space as possible. That is, we take r as the solution of
k
2
(r) = r
2
g
2
(r). (4.9)
Then the line element becomes
ds
2
= h
2
(r)dt
2
+ g
2
(r)
_
dr
2
+ r
2
d
2
_
= h
2
(r)dt
2
+ g
2
(r)dr
2
. (4.10)
Clearly it is rotation invariant, and the three-dimensional space-like part of the line element
is conformal to that of at space in polar coordinates, with the conformal factor g(r)
2
In this chapter we employ natural units with c = 1.
60
depending only on the radius, and not on the orientation or the time. In the literature
the coordinates chosen here to parametrize the Schwarzschild metric are called isotropic
coordinates.
Up to the freedom of radial reparametrizations, the form of the line element (4.10) is
completely determined by the symmetry requirements. To x the radial dependence of
the remaining coecients h(r) and g(r) we nally substitute the metric into the Einstein
equations. For the Einstein tensor G

= R

1/2 g

R we nd that it is diagonal with


components
G
tt
=
h
g
3
_
2h(g

+
2
r
g

)
hg
2
g
_
,
G
rr
=
1
gh
_
2h

+
hg
2
g
+
2
r
(hg)

_
,
G

=
G

sin
2

=
r
2
gh
_
h

g + hg

+
1
r
(hg)

hg
2
g
_
,
(4.11)
In the absence of sources or a cosmological constant these expressions have to vanish. As
boundary conditions at innity, where space-time must become asymptotically at, we
require that r implies h 1 and g 1. Then we have three dierential equations
for two unknown functions, and therefore the equations must be degenerate. One can
check, that this is the case if the following two relations are satised:
hg = g + rg

,
_
r
3
(hg)

= 0. (4.12)
Indeed, in this case the various components of G

become proportional:
G
rr
=
g
2
h
2
G
tt
=
1
r
2
G

=
1
r
2
sin
2

. (4.13)
The conditions (4.12) are met for any g(r) of the form
g(r) = a +
b
r
+
c
r
2
, h(r) =
g + rg

g
. (4.14)
It turns out that eqs.(4.14) indeed provide solutions of the Einstein equations satisfying
the correct boundary conditions. They are most easily obtained from the rst Einstein
equation (4.11): for hg = 0 it is possible to factor out all h-dependence and obtain an
equation for g only:
gg

+
2
r
gg

1
2
g
2
= 0. (4.15)
There is a unique solution with the required normalization for r :
g(r) =
_
1 +
m
2r
_
2
, (4.16)
61
where m is an undetermined constant of integration. Eq.(4.12) then immediately provides
us with the solution for h:
h(r) =
2r m
2r + m
. (4.17)
The complete solution for the static, spherically symmetric space-time in isotropic coordi-
nates therefore is
ds
2
=
_
2r m
2r + m
_
2
dt
2
+
_
1 +
m
2r
_
4 _
dr
2
+ r
2
d
2
_
. (4.18)
Clearly this solution matches Minkowski space at r by construction, and h and g
change monotonically (decreasing and increasing, respectively) as a function of r untill
r = m/2. At that value h = 0, and the metric has a zero mode. Then the derivation of the
solution we have presented breaks down, and its continuation to values r < m/2 requires a
careful interpretation, discussed below. In any case, it turns out that this singularity of the
metric is not physical, as invariants of the curvature remain nite and well-dened there.
Indeed, in the following we construct coordinate systems in which the metric is perfectly
regular at these points and can be continued to regions inside the surface r = m/2. In
the isotropic coordinate system the apparent singularity of the metric for r = 0 is another
coordinate artifact. It disappears if one brings spatial innity to a nite distance by an
appropriate coordinate transformation. Details are given in the next section.
4.3 Discussion
Modulo radial reparametrizations r r(r) the line-element (4.18) is the most general
static and spherically symmetric solution of the source-free Einstein equations matching at
Minkowski space-time at spatial innity (r ). Therefore in the non-relativistic weak-
eld range (v 1, r m) it should reproduce Newtons law for the gravitational eld of
a spherically symmetric point mass of magnitude M, according to which the acceleration
of a test particle at distance r is
d
2
r
dt
2
=
GM
r
3
r (4.19)
As the spherically symmetric solutions of the Einstein equations (4.18) are distinguished
only by a parameter m, in the Newtonian limit this parameter must be related to the mass
M, thereby representing the equivalent gravitational mass of the object described by the
spherically symmetric metric (4.18).
We rst show that our solutions reproduce Newtons law in the limit r m, v 1
(= c). The equation of motion for a test particle is the geodesic equation (2.7). For the
spatial coordinates x
i
this becomes
62
d
2
x
i
d
2
=
_
dx
0
d
_
2

i
00
+ (terms O(v/c))

h
g
2

i
h =
m
r
3
x
i
h
g
3
.
(4.20)
Note that in the limit v 1 one has dx
0
= dt d, whilst for r the factor h/g
3
1.
Upon the identication m = GM we thus indeed reproduce Newtons equation (4.19).
Taking into account the dimensions of Newtons constant, as follows from (4.19) and
is expressed in eq.(1.2), and our natural units in which c = 1, we nd that m has the
dimension of a length, consistent with the observation that m/r should be dimensionless.
We have already noted that in isotropic coordinates the metric component h vanishes at
half this distance: r = r
0
= m/2. Presently this only signals that the isotropic system
cannot be extended beyond this range, but the surface r = r
0
does have a special physical
signicance: it denes the location of the horizon of a spherically symmetric black hole.
This characteristic value r
0
= m/2 of the radial coordinate is known as the Schwarzschild
radius.
For the range m/2 < r < the Schwarzschild solution in the form (4.18) is well-dened,
and it matches the asymptotic Minkowski metric for large values of r. We noted that on
the Schwarzschild sphere r = m/2 the metric is singular, but as a solution of the Einstein
equations it can be continued to smaller values of r and then it is again well-dened in the
domain 0 < r < m/2. However, it turns out that this domain of r-values does not in any
sense represent the physical interior of the horizon, as one might expect naively. Rather,
it easy to establish that the solution (4.18) is invariant under the transformation
r
m
2
4r
, (4.21)
which maps the domain (0, m/2) to (m/2, ). Therefore the isotropic coordinate system
actually represents a double cover of the region between the Schwarzschild radius and the
asymptotic region r . Note that under this transformation h(r) h(r), which can
be compensated by an accompanying time-reversal t t. From these observations it
follows that the apparent singularity for r = 0 disappears as it simply describes another
asymptotic region like that for r in the original coordinate system.
The existence of such a coordinate system presenting a double cover of the space outside
the black hole might lead one to suspect, that a test particle thrown in radially towards the
spherical Schwarzschild surface after a long enough time would return to an asymptotic
Minkowski region. However, this does not happen; in the rst place, as a simple calculation
to be presented later shows, in terms of the time measured by an observer at asymptotic
innity it takes an innite period of time for a test particle starting at any nite radial
distance r > m/2 with any nite velocity even to reach the Schwarzschild surface; certainly
such an observer will never see it start up and set o towards innity again. Apart from
this, leaving the Schwarzschild surface would also take an innite amount of asymptotic
63
Minkowski-time. Clearly, these eects are due to the vanishing of h(r) on the horizon,
which implies that an innite amount of asymptotic Minkowski time t has to pass during
any nite period of proper time . But this is only the minor part of the argument; more
important is the second reason, to wit that time-like geodesics do not ow from values
r > m/2 to values r < m/2. Instead, as will become clear they ow into a new region of
space-time described by complex values of r.
Some calculational details are given later, but the above arguments at least indicate
that an observer at spatial innity is essentially disconnected from the domain of space-
time beyond the Schwarzschild sphere. Therefore the Schwarzschild surface is called a
horizon: from spatial innity one cannot look beyond it. In contrast, things are radically
dierent for an observer moving with the test particle towards the horizon. For such an
observer in free fall, co-moving with the test particle on a radial geodesic and starting
from an arbitrary point at a nite distance from the Schwarzschild surface, only a nite
amount of proper time passes until the radius r = m/2 is reached, although for large radial
distances the required time grows linearly with the distance.
Moreover, having reached the horizon the particle (and the observer) can continue
falling inwards without noticing anything particular, at least initially. The only remarkable
eect they could discover once they have passed beyond the Schwarzschild surface is the
impossibility to turn back to spatial innity, no matter how powerful the engines they have
at their disposal to accelerate and propel them. Therefore their passage to the inside of
the Schwarzschild sphere represents an irreversible event: they are permanently trapped
inside the Schwarzschild sphere. For this reason the Schwarzschild space-time is called a
black hole: nothing, not even radiation, can get out once it has fallen through the horizon,
at least not in the domain of classical relativity.
To describe this motion of test particles (i.e. the ow of geodesics) through the Schwarzschild
sphere, one must pass to a dierent coordinate system, one better suited to particles in free
fall rather than to the description of the space-time from the point of view of an asymptotic
Minkowskian observer at rest at spatial innity. We make this passage in two steps. First
we show that in other coordinate systems there is a dierent continuation of space-time
beyond the Schwarzschild surface, one in which the interior geometry inside the horizon
is distinct from the outside geometry and can not be mapped back isometrically to the
exterior. After that we show by explicit calculation of the geodesics, that this alternative
continuation can be taken to decribe the true physical situation for a test particle falling
through the Schwarzschild sphere.
4.4 The interior of the Schwarzschild sphere
To go beyond the region of space-time covered by the isotropic coordinates, we perform
a radial coordinate transformation which instead of making the spatial part of the metric
conformal to at space, as in (4.9), (4.10), makes the angular part directly isomorphic to
the two-sphere of radius r:
64
r = rg(r),

k
2
( r) = r
2
. (4.22)
Then the metric takes the form
ds
2
=

h
2
( r) dt
2
+ g
2
( r) d r
2
+ r
2
d
2
. (4.23)
The explicit form of the radial transformation (4.22) gives for the coecients

h
2
( r) =
1
g
2
( r)
= 1
2m
r
. (4.24)
This leads to the standard form of the Schwarzschild solution as presented in the original
papers [37, 38]:
ds
2
=
_
1
2m
r
_
dt
2
+
1
_
1
2m
r
_ d r
2
+ r
2
d
2
. (4.25)
As to lowest order the transformation (4.22) does not include a rescaling, we have for large
r and r:
1
r
=
1
r
+ O(
1
r
2
). (4.26)
Therefore in the asymptotic region the Newtonian equation of motion for a testparticle in
the eld of a point mass M still holds in the same form after replacing r by r, and in the
non-relativistic long-distance limit the two coordinate systems are indistinguishable.
Note that the Schwarzschild radius r = m/2 is now located at r = 2m. Again the
metric components are singular there, but now there is a zero mode in the (tt)-component
and an innity in the (rr)-component, in such a way that the determinant of the metric
remains nite on the horizon. On the other hand, the determinant g does have a real zero
for r = 0. This metric singularity was not present in the isotropic coordinate system, as
it did not cover this region of space-time, but it turns out to represent a real physical
singularity: the curvature becomes innite there in a coordinate-independent way.
A peculiarity of the metric (4.23) is that the components

h
2
and g
2
are not positive
denite, but change sign at the Schwarzschild radius. This implies that the role of time
and radial distance are interchanged for r < 2m: r is the time coordinate and t the
radial coordinate. At the same time the metric components have become explicitly time
dependent (they are functions of r), and therefore the metric is no longer static inside the
horizon. In return, the metric is now invariant under radial shifts t t + t. Hence it is
the radial momentum p
t
which now commutes with the Hamiltonian (4.8), the generator
of proper-time translations, and is conserved.
This strange result is the price one has to pay for continuing the Schwarzschild geometry
to the interior of the Schwarzschild sphere. It is quite clear that this could never heve been
achieved in the isotropic coordinate system (4.18), at least not for real values of r. Together
with the appearance of the curvature singularity at r = 0 this shows beyond doubt that
65
the region inside the Schwarzschild sphere r < 2m has no counter part in the standard
isotropic coordinate system.
To gain a better understanding of the dierence between the continuation to r < m/2 in
isotropic coordinates and r < 2m in Schwarzschild coordinates it is useful to perform yet
another radial coordinate transformation to a dimensionless coordinate dened by [39]
tanh

2
= h(r), (4.27)
with the result that
r = rg(r) = 2m cosh
2

2
. (4.28)
Then the Schwarzschild metric becomes
ds
2
= tanh
2

2
dt
2
+ 4m
2
cosh
4

2
_
d
2
+ d
2
_
. (4.29)
As cosh /2 1 this expression for the metric, like the isotropic coordinate system, is valid
only outside the horizon r 2m. Moreover, continuing to negative values of we reobtain
the double cover of the exterior of the Schwarzschild black hole:
tanh
_

2
_
= h(r) = h(
m
2
4r
). (4.30)
Now the interior of the Schwarzschild sphere can also be reparametrized in a similar way,
taking account of the fact that h
2
has become negative:
r = m(1 + cos ) = 2mcos
2

2
, tan
2

2
=

h
2
( r). (4.31)
This parametrization clearly exists only for 0 r 2m. In these coordinates the metric
takes the form
ds
2
= tan
2

2
dt
2
+ 4m
2
cos
4

2
_
d
2
+ d
2
_
. (4.32)
The time-like nature of now is obvious. It is also clear, that this coordinate system
presents an innitely repeated periodic covering of the interior of the Schwarzschild sphere:
the fundamental domain may be chosen to be 0 , corresponding to 2m r 0;
but continuation of beyond this region leads to repeated covering of the same set of
r-values.
Comparing these two parametrizations of the exterior and the interior of the Schwarzschild
sphere, it is now obvious that they are related by analytic continuation [39]: the interior
of the sphere is not described by negative values of , which correspond to r < m/2 in
isotropic coordinates, but by imaginary values of :
= i, (4.33)
66
which is like a Wick rotation changing the space-like radial coordinate to the time-like
coordinate . In terms of the original isotropic coordinates it can be described as the
analytic continuation of r to complex values
r =
m
2
e
i
. (4.34)
Thus in the interior of the Schwarzschild surface the modulus of r is constant, but its
phase changes, whilst outside its phase is constant and the modulus changes. Note that
whereas the period of the fundamental domain of in describing the interior region of the
Schwarzschild sphere is , the values of the isotropic radial coordinate r are periodic in
with period 2. Again it seems that the isotropic coordinates cover the Schwarzschild
solution twice. We have more to say about that in the following. Finally we observe, that
as indicated in eq.(4.33) there are two ways to perform the analytic continuation from
to , i.e. from the exterior to the interior of the Schwarzschild space-time. Also this fact
has non-trivial signicance for the geometry of the Schwarzschild space-time.
4.5 Geodesics
The analysis presented so far makes it clear that the geometry of the Schwarzschild space-
time is quite intricate. Indeed, at the Schwarzschild sphere r = m/2 (or r = 2m and
= = 0, respectively) several branches of the space-time meet: two branches of the
exterior connected to an asymptotic Minkowski region, and two branches of the interior
connecting the horizon with the space-time singularity. It is therefore of importance to
understand the ow of geodesics in the neighborhood of the horizon, and see which branches
test particles moving in the Schwarzschild space-time follow. We begin with time-like
geodesics, representing the motion of test particles of unit mass. For such a particle one
has in asymptotic Minkowski space:
2H = g

= 1. (4.35)
This relation was already argued from the denition of proper time in sect.(2.1), eq.(2.8)
and below. As H is conserved, the relation holds everywhere on the particles worldline.
The solution of the geodesic equations for Schwarzschild space-time is much simplied
by the large number of its symmetries. They provide us with suciently many additional
constants of motion to allow a complete solution of the equations of motion of test particles
in terms of their energy and angular momentum. The rst of these constants of motion is
the momentum component p
t
, conserved because the metric is t-independent:
p
t
= h
2
(r)
dt
d
=
dt
d
=

h
2
(r)
. (4.36)
At space-like innity h
2
1; also, with dened as proper time, dt = d for a particle at
rest there. Therefore a particle at rest at r has = 1, equal to the (unit) rest mass of
the test particle; moreover, as is intuitively obvious and argued more precisely below, if it
67
is not at rest it must have > 1. Indeed, particles with < 1 cannot reach spatial innity
and in this sense they are in bound states. The factor gives the usual special relativistic
(kinematical) time-dilation for a moving particle in asymptotic Minkowski space.
In contrast, for nite radial distance r
0
< r < the factor h
2
(r) describes the grav-
itational redshift: the time dilation due only to the presence of a gravitational eld. It
represents a universal eect, not inuenced by the state of motion of the particle.
The conservation of angular momentum gives us three conservation laws, which are
however not independent:
J
1
= sin p

cot cos p

,
J
2
= cos p

cot sin p

,
J
3
= p

,
(4.37)
with J
2
1
+ J
2
2
+ J
2
3
= L
2

2
, a constant. Here the separate momentum components are
p

= k
2
(r)
d
d
, p

= k
2
(r) sin
2

d
d
. (4.38)
If one orients the coordinate system such that the angular momentum is in the x
3
-direction,
then = /2 = constant, and therefore p

= 0 and p

= J
3
= . Then the angular
coordinates as a function of proper time are given by
= /2,
d
d
=

k
2
(r)
. (4.39)
It remains to solve for the radial coordinate
3
r. For this we use the conservation of the
hamiltonian, setting 2H = 1:
p
2
r
=
_
g
2
dr
d
_
2
= g
2
_

2
h
2


2
k
2
1
_
. (4.40)
Note, that consistency of this equation in the region outside the horizon requires

2
h
2


2
k
2
+ 1. (4.41)
It shows, why in the region r the time dilation factor has to be greater or equal
to unity: 1. At the same time, for non-zero values of angular momentum the
radial distance r can in general not become arbitrary low: like in the ordinary Kepler
problem there is a centrifugal barrier to be overcome. For details we refer to the literature
[11, 35, 36].
Combining eq.(4.40) with that for the angle , by eliminating we directly obtain the
orbital equation
3
Note that each of these equations is still true for arbitrary radial parametrization.
68
dr
d
=
k
2
g

2
h
2


2
k
2
1. (4.42)
It is clear that an orbit can have a point of closest approach, or furthest distance, only
if dr/d = dr/d = 0. The only physical solutions are points where relation (4.41)
becomes an equality. When the equality holds identically, for all points of the orbit, the
orbit is circular. One can show that the smallest radius allowed for a stable circular orbit
corresponds in Schwarzschild coordinates to r = 2m
2
0
= 6m.
To the set of time-like trajectories we add the light-like geodesics, representing the orbits
of massless particles. In that case the proper-time interval vanishes identically on any
geodesic, that is to say H = 0. We parametrize the trajectories with an ane parameter
and obtain
p
t
= h
2
dt
d
, (4.43)
allowing one to replace the ane parameter by t through
d =
h
2

dt. (4.44)
One can again choose the plane of motion to be = /2, and the conservation of angular
momentum holds in the form
r
2
h
2
d
dt
= r
2
d
d
= , (4.45)
where is a constant. The radial equation now becomes
h
2
= g
2
_
dr
dt
_
2
+ k
2
_
d
dt
_
2
+ k
2
sin
2

_
d
dt
_
2
, (4.46)
or using the choice = /2 above:
g
2
_
dr
dt
_
2
= h
2
_
1

2
h
2
k
2

2
r
4
_
. (4.47)
Like in the massive case, if = 0 one can eliminate the ane parameter also in favour
of instead of t.
An interesting case, as concerns the exploration of the horizon and the connection between
the interior and the exterior of the Schwarzschild sphere, is that of radial motion: = 0 or
= 0, respectively. Let us rst consider the time-like geodesic for a particle starting from
rest near r = , which implies = 1 and dr/dt 0. Then
dt
d
=
1
h
2
,
dr
d
=
1
gh

1 h
2
. (4.48)
69
It is easy to solve these equations in the Schwarzschild coordinates; indeed, with (1

h
2
) =
2m/ r one nds a solution valid in the whole range 0 < r < , both outside and inside the
horizon:
r()
2m
=
_
3
4m
(
0
)
_
2
3
, (4.49)
where
0
is the proper time at which the particle reaches the singularity at r = 0. Taking
this time as xed, denotes the proper time, prior to
0
, at which the particle is at radial
distance r. Observe, that this time is nite for any nite radial distance r; only when
starting from innity it takes an innite proper time to reach the singularity. Nothing
special happens at the horizon r = 2m: it is crossed at proper time

H
=
0

4m
3
, (4.50)
a proper period 4m/3 before the singularity is reached (of course, in terms of the time-
coordinate r inside the horizon, it takes a nite period 2m).
Solving for t as a function of proper time, we have to distinguish two cases: r > 2m
and r < 2m. Namely, introducing the new variable
y =
_
3
4m
(
0
)
_
1
3
, (4.51)
we obtain the dierential equation

1
4m
dt
d
=
y
4
y
2
1
, (4.52)
the solution of which depends on whether the r.h.s. is positive or negative. We nd for the
two cases the solutions
t
0
t
4m
=

0

4m
+
_
3
4m
(
0
)
_
1
3
+
_

_
arccoth
_
3
4m
(
0
)
_
1
3
, r > 2m;
arctanh
_
3
4m
(
0
)
_
1
3
, r < 2m.
(4.53)
In both cases the left-hand side becomes innite at the horizon, i.e. at proper time
H
.
This conrms that an external observer at innity will never actually see the particle reach
the horizon. Similarly, the distance measured from the horizon to the singularity at r = 0
is innite in terms of the coordinate t.
Now consider what happens in isotropic coordinates. In terms of a new variable
x =

2r
m
, (4.54)
70
where we choose the positive branch of the square root, integration of the equation for r()
is elementary and gives
2
m
(
0
) =
1
3
_
x +
1
x
_
3
. (4.55)
Clearly this is symmetric under interchange x 1/x, corresponding to inversion w.r.t. the
horizon r = m/2. Indeed, both at r and r = 0 we see that . This conrms
our earlier remark that a particle released from space-like innity does not return there.
Instead, it will cross the horizon x = 1 at =
H
, as in the previous calculation. Then
continuing r to complex values (4.34), we nd
x = e
i
2

,

0

4m
=
1
3
cos
3

2
. (4.56)
The singularity = is then reached for =
0
, as expected. In terms of the radial
coordinate , related by the pseudo-Wick rotation (4.33) the equation for the radial distance
in terms of proper time outside the horizon becomes

4m
=
1
3
cosh
3

2
. (4.57)
The analytic continuation of coordinates in the complex plane on crossing the horizon is
manifest.
Next we consider incoming light-like radial geodesics. The equation is
dr
dt
=
h
g
. (4.58)
Let t
0
refer to the time at which the massless particle (e.g., a photon) passes a xed point
on the radius, e.g. r
0
in Schwarzschild coordinates. Then the solution of the geodesic
equation for t is
t t
0
2m
=
r
0
r
2m
ln

r 2m
r
0
2m

. (4.59)
With the absolute value as argument of the logarithm this equation is valid both outside
and inside the horizon. The time for the photon to reach the horizon from any point outside
diverges, as does the distance it has to move from horizon to the curvature singularity, as
measured by the coordinate t.
4.6 Extended Schwarzschild geometry
As we have already noticed that for a massive particle falling across the horizon nothing
very special happens, it is unsatisfactory that we do not have a non-singular description of
the same process for massless particles. To improve the description of light-like geodesics at
the end of the last section, it is not sucient to perform a radial coordinate transformation:
71
one has to eliminate the t-coordinate and replace it by one which does not become innite
on the horizon. Moreover, to describe light-like geodesics it would be convenient to have
a metric where the light-cone structure is manifest by having the radial part of the metric
conformal to radial Minkowski space: a coordinate system in which (r, t) are replaced by
coordinates (u, v) such that the radial part of the metric becomes
h
2
(r)dt
2
+ g(r)
2
dr
2
f(u, v)
_
du
2
+ dv
2
_
. (4.60)
Such a coordinate system was rst proposed by Kruskal and Szekeres [40]; dening the
new coordinates such as to be dimensionless, the line element is written in the form
ds
2
4m
2
= f(u, v)
_
dv
2
+ du
2
_
+ g(u, v) d
2
, (4.61)
with the standard choice for (u, v) dened in terms of the Schwarzschild coordinates ( r, t)
by:
u
2
v
2
=
_
r
2m
1
_
e
r
2m
,
u
v
=
_

_
tanh
t
4m
, if |v| < |u|;
coth
t
4m
, if |v| > |u|.
(4.62)
In these coordinates the horizon r = 2m, t = is located at u = v, whilst the curvature
singularity r = 0 corresponds to the two branches of the hyperbola u
2
v
2
= 1, one in
the past and one in the future of an observer at space-like innity. More generally, from
the rst result one concludes that the surfaces r = constant are mapped to hyperbolas
u
2
v
2
= constant, where the last constant is positive outside and negative inside the
horizon. These hyperbolas are worldlines of particles in circular orbit, i.e. being subject
to a constant acceleration.
Similarly hypersurfaces t = constant correspond to u = constant v, represented by
straight lines in a (u, v)-diagram. Note also that the singularity r = 0 consists of a set of
two space-like surfaces in (u, v)-coordinates.
More explicitly, we can write (u, v) in terms of (r, t) as
u =

r
2m
1 e
r
4m
cosh
t
4m
,
v =

r
2m
1 e
r
4m
sinh
t
4m
,
(4.63)
outside the horizon ( r 2m); and
72
u =

1
r
2m
e
r
4m
sinh
t
4m
,
v =

1
r
2m
e
r
4m
cosh
t
4m
,
(4.64)
within the horizon ( r 2m). We note once again, that these coordinates produce two
solutions for each region of the space-time geometry. In the above conventions the dimen-
sionless coecient functions are given implicitly by
f(u, v) =
8m
r
e

r
2m
, g(u, v) =
r
2
4m
2
. (4.65)
We can also compare with the coordinate systems (4.29) and (4.32), describing explicitly
the double cover of the Schwarzschild space-time:
u = e
1
2
cosh
2
2
sinh

2
cosh
t
4m
, v = e
1
2
cosh
2
2
sinh

2
sinh
t
4m
, (4.66)
for the exterior region |v| < |u|; and
u = e
1
2
cos
2
2
sin

2
sinh
t
4m
, v = e
1
2
cos
2
2
sin

2
cosh
t
4m
, (4.67)
for the interior region |v| > |u|. From these expressions we infer, that the exterior regions
> 0 and < 0, corresponding to r > m/2 and r < m/2 in isotropic coordinates, and
the two interior regions > 0 and < 0 as well, coincide with the regions (u, v) > 0
and (u, v) < 0 in Kruskal-Szekeres coordinates, respectively. Hence the double cover of
Schwarzschild space-time given by the two solutions for the Kruskal-Szekeres coordinates
above correspond exactly to the double cover we found in isotropic and (, )-coordinates.
The Kruskal-Szekeres coordinates were introduced, and are especially useful, because
they represent the light-like geodesics in a very simple way. In particular, the radial light-
like geodesic are given by
du
2
= dv
2

du
dv
= 1. (4.68)
These correspond to straight lines in the (u, v)-diagram parallel to the diagonals. The
special light rays u = v are the asymptotes of the singularity and describe the horizon
itself: they can be interpreted as photons permanently hovering on the horizon. Any other
light-like radial geodesic eventually hits the singularity, either in the past or in the future,
at nite (u, v), although this is in the region before t = or after t = in terms of the
time of an observer at spatial innity.
As the light cones have a particularly simple representation in the (u, v)-diagram, for all
regions inside and outside the horizon, it follows that time-like and space-like geodesics are
distinguished by having their derivatives |du/dv| > 1 for the time-like case, and |du/dv| < 1
for the space-like ones. One can check by explicit computation, that a particle starting from
73
rest at any distance r and falling into the black hole has a time-like world line everywhere,
also inside the horizon: du/dv can change on the world line of a particle, but never from
time-like to space-like or vice versa. Therefore all time-like geodesics must cross the horizon
(light-like) and the singularity (space-like) sooner or later. If they dont reach or start from
spatial innity, they must come from the past singularity across the past horizon, and then
cross the future horizon to travel towards the nal singularity, all within a nite amount
of proper time.
The Kruskal-Szekeres coordinate system also teaches us more about the topology of the
Schwarzschild geometry: how the various regions of the space-time are connected. First of
all it shows that there is a past and a future singularity inside a past and a future horizon
(|u| < |v|), depending on v being positive or negative. In this respect it is helpful to observe,
that according to eq.(4.64) for r < 2m the sign of v is xed for all t. In the literature these
regions are sometimes refered to as the black hole, the interior of the horizon containting
the future singularity, and the white hole containing the past singularity. In a standard
convention, these regions of space-time are classied as regions II (future Schwarzschild
sphere) and IV (past Schwarzschild sphere), respectively.
There are also two exterior regions connected to asymptotic Minkowski space. They
are distinguished similarly by u > 0 and u < 0, as for r > 2m the sign of u is xed for
all t: one cannot reach negative u starting from positive u or vice versa by any time-like
or light-like geodesic. In the same standard convention these regions are refered to as
region I and region III, respectively. Being connected by space-like geodesics only, the two
sheets of the exterior of the Schwarzschild space-time are not in causal contact and cannot
physically communicate. However, the surfaces t = constant do extend from u = to
u = ; indeed, the regions are space-like connected at the locus u = v = 0, where
the two branches of the horizon intersect. To understand the topology of Schwarzschild
space-time, we must analyze this connection in some more detail; this can be done using
the technique of embedding surfaces, as explained e.g. in ref.[11].
Consider a surface t = constant and x a plane = /2. We wish to understand how
the regions u > 0 and u < 0 are connected for this plane. The Schwarzschild line element
restricted to this plane is
ds
2
=
d r
2
1
2m
r
+ r
2
d
2
. (4.69)
Now we embed this two-dimensional surface in a three-dimensional at, euclidean space:
ds
2
= dx
2
+ dy
2
+ dz
2
, (4.70)
by taking x = r cos , y = r sin , and dening the surface z = F( r) such that
ds
2
=
_
1 + F
2
( r)
_
d r
2
+ r
2
d
2
. (4.71)
Comparison of eqs. (4.71) and (4.69) then implies
74
z
2
16m
2
=
_
F( r)
4m
_
2
=
r
2m
1. (4.72)
This denes a surface of revolution, obtained by taking the parabola z
2
= 8mx 16m
2
in
the (x, z)-plane and rotating it around the z-axis. It is a single surface, with looks like a
throat, of which the upper half z > 0 corresponds to

r
2m
1 0, or u 0, (4.73)
whilst the lower half z < 0 describes the branch

r
2m
1 0, or u 0, (4.74)
The two regions are connected along the circle r = 2m at z = 0. Of course, the spherical
symmetry implies that the same topology would be observed for any choice of angle ;
any plane with u > 0 is connected to one with u < 0 in this way. Although no time-like
travel is possible classically between the regions on each side of the throat, the possibility
of particles tunneling via quantum processes between the two regions poses a curious and
interesting question to theories of quantum gravity.
4.7 Charged black holes
The spherically symmetric solution of the source-free Einstein equations can be extended
to a solution of the coupled Einstein-Maxwell equations including a static spherically sym-
metric electric eld (in the rest frame of the black hole)[41]. As the only solution of this
type in at Minkowski space is the Coulomb eld, we expect a solution of the coupled
equations to approach the Coulomb potential at space-like innity:
A = A

dx

=
q
4rp(r)
dt, (4.75)
with p(r) 1 for r . This is a solution of the Maxwell equation
D

= 0, (4.76)
in a spherically symmetric static space-time geometry. The Maxwell eld strength appear-
ing in this equations is
F =
1
2
F

dx

dx

=
q
4
p + rp

r
2
p
2
dt dr. (4.77)
In isotropic coordinates (4.10) the energy-momentum tensor of this electro-magnetic eld
reads
75
T

[F] =
_
F


1
4
g

F
2
_
=
q
2
32
2
(p + rp

)
2
r
4
h
2
p
4
diag
_
h
2
g
2
, 1, r
2
, r
2
sin
2

_
. (4.78)
This has the same general structure as the Einstein tensor (4.13):
G

=
1
gh
_
2h

+
hg
2
g
+
2
r
(hg)

_
diag
_
h
2
g
2
, 1, r
2
, r
2
sin
2

_
, (4.79)
which was derived under the conditions (4.12), with the general solution (4.14). Now
requiring the coupled Einstein-Maxwell equations
G

= 8GT

[F], (4.80)
this reduces to a single condition on the radial functions (h, g, p), with the solution
p(r) = g(r) = 1 +
m
r
+

2
4r
2
,
h(r) =
g + rg

g
=
4r
2

2
4r
2
+ 4mr +
2
.
(4.81)
Here the constant
2
is dened by

2
= m
2
e
2
, e
2
=
q
2
G
4
. (4.82)
It may be checked that this solution satises the Maxwell equations in this gravitational
elds as well. As g(r) can also be written in the form
g(r) =
_
1 +
m
2r
_
2

e
2
4r
2
, (4.83)
we nd that the Schwarzschild solution is reobtained in the limit e
2
0. For
2
> 0 the
zero of the metric coecient h
2
is now shifted to
r =

2
. (4.84)
As in the Schwarzschild case this is a point where two asymptotically at regions of space-
time meet: the metric is again invariant under the transformation
r

2
4r
, t t, (4.85)
under which h(r)dt h(r)dt. Moreover, also like in the Schwarzschild geometry, the
sphere r = /2 actually denes a horizon and time-like geodesics are continued into the
interior of this horizon, which is not covered by the (real) isotropic coordinate system.
76
A Schwarzschild-like coordinate system which allows continuation into the interior of
the horizon exists. It was found by Reissner and Nordstrom [41] and is related to the
isotropic coordinates by a radial coordinate transformation
r = rg(r) = r + m +

2
4r
. (4.86)
Then the metric of the charged spherically symmetric black hole takes the form
ds
2
=

h
2
dt
2
+ g
2
d r
2
+ r
2
d
2
, (4.87)
where the coecients (

h, g) are again inversely related:

h
2
( r) =
1
g
2
( r)
= 1
2m
r
+
e
2
r
2
=
_
1
m
r
_
2


2
r
2
. (4.88)
The horizon is now seen to be located at r = m + ; in fact, there is a second horizon at
r = m , provided m > e. As in general

h
2
and g
2
are not positive denite, at each of
these horizons the role of radial and time coordinate is interchanged. Only for = 0, i.e.
for m = e, the situation is dierent, as the two horizons coincide and no interchange of
time- and space-like coordinates occurs. This special case of Reissner-Nordstrom solution
is refered to in the literature as an extremal charged black hole.
The result of the second horizon for non-extremal charged black holes is, that the real
curvature singularity that is found to exist at r = 0 is a time-like singularity, for m > e.
This is expected from the Coulomb solution in at Minkowski space, which contains a
time-like singularity of the electric eld at r = 0. Therefore such a black hole looks more
like a point particle than the Schwarzschild type of solution of the Einstein equations.
The interior region between the horizons can be obtained from the isotropic coordinates
by complex analytic continuation of r from the outer horizon by
r =

2
e
i
. (4.89)
The domain of the argument is chosen as [0, ]. Indeed, for these values of r the radial
Reissner-Nordstrom coordinate is
r = m + cos , (4.90)
and takes all values between the two horizons r = m precisely once. In this parametriza-
tion the isotropic Reissner-Nordstrom metric becomes
ds
2
=

2
sin
2

(m + cos )
2
dt
2
+ (m+ cos )
2
_
d
2
+ d
2
_
. (4.91)
Again we observe the role of as the time-like variable, with t space-like between the
horizons. For = the isotropic coordinate r becomes negative:
= r =

2
. (4.92)
77
Thus we discover that the Reissner-Nordstrom space-time can be extended consistently to
negative values of the radial coordinate. In particular the curvature singularity at r = 0 is
seen to be mapped to
r =
me
2
. (4.93)
This shows rst of all, that there are two (time-like) branches of the singularity, and
secondly that r can become negative for non-extremal black holes with m > e. The two
branches are found also upon analytic continuation of in (4.91) to imaginary values:
i. (4.94)
This leads to a parametrization of the Reissner-Nordstrom geometry for the exterior regions
of the form
ds
2
=

2
sinh
2

(m cosh )
2
dt
2
+ (m cosh )
2
_
d
2
+ d
2
_
, (4.95)
where the plus sign holds for the exterior region connected to asymptotic at Minkowski
space, and the minus sign to the region connected to the curvature singularity. Because
of the double-valuedness of the analytic continuation (4.94) there are two copies of each of
these exterior regions involved.
4.8 Spinning black holes
The Schwarzschild and Reissner-Nordstrom black holes are static and spherically symmetric
w.r.t. an asymptotic Minkowski frame at spatial innity. There also exist extensions of
these solutions of Einstein and coupled Einstein-Maxwell equations which are neither static
nor spherically symmetric, but represent rotating black holes with or without charge [42,
43]. These solutions have an axial symmetry around the axis of rotation, and are stationary:
in the exterior region the metric is time-independent, and the rate of rotation and all other
observable properties of the black hole are constant in time.
The standard choice of coordinates for these metrics, with the axis of rotation being
taken as the z-axis and reducing to those of Schwarzschild, or Reissner and Nordstrom for
the spherically symmetric case of no rotation, is that of Boyer and Lindquist [44]. Denoting
these coordinates by (t, r, , ) to emphasize their relation with the spherical coordinate
systems for the non-rotating metrics, the line element for reads
ds
2
=

2
r
2
+ a
2
cos
2

_
dt a sin
2
d
_
2
+
r
2
+ a
2
cos
2

2
_
d r
2
+

2
d
2
_
+
( r
2
+ a
2
)
2
sin
2

r
2
+ a
2
cos
2

_
d
a
r
2
+ a
2
dt
_
2
.
(4.96)
78
Here a is a constant parametrizing the deviation of this line element from the usual diagonal
form of the Schwarzschild-Reissner-Nordstrom metric; its physical interpretation is that of
the total angular momentum per unit of mass:
J = ma. (4.97)
As one can always choose coordinates such that the angular momentum points along the
positive z-axis, the angular momentum per unit of mass may be taken positive: a > 0. For
convenience of discussion this will be assumed in the following.
The quantity

2
in the line element (4.96) is short hand for

2
= ( r m)
2
+ a
2
+ e
2
m
2
. (4.98)
Note that, like the metric coecients

h
2
and g
2
we have encountered earlier, this quantity
is not positive denite everywhere. Also, the zeros of this function determine the location
of the horizons, of which there are two (an outer and an inner one) for m
2
> a
2
+ e
2
:
r
H
= m

m
2
a
2
e
2
. (4.99)
The gravitational eld represented by the metric (4.96) is accompanied by an electro-
magnetic eld, the vector potential of which can be chosen as the one-form
A =
q r
4
2
_
dt a sin
2
d
_
. (4.100)
For convenience of notation we have here introduced the function dened as

2
= r
2
+ a
2
cos
2
. (4.101)
These elds are solutions of the coupled Einstein-Maxwell equations (4.76) and (4.80), with
the Maxwell eld strength given by the two form
F =
q
4
4
_
r
2
a
2
cos
2

_
dr
_
dt a sin
2
d
_
+
qa
2
r cos sin

4
d
_
adt ( r
2
+ a
2
)d
_
,
(4.102)
and with the identication of the quantity e
2
as in eq.(4.82):
e
2
=
q
2
G
4
.
To determine the properties of the space-time described by the Kerr-Newman line element
we need to analyse the geodesic ow. Because the spherical symmetry of the Schwarzschild
and Reissner-Nordstrom solutions is absent, and only axial symmetry remains, there is one
less independent constant of motion from angular momentum: only the component along
the axis of rotation (the z-axis, say) is preserved. Fortunately this is compensated by the
79
appearance of a new constant of motion, quadratic in momenta, which replaces the Casimir
invariant of total angular momentum [46, 47]. As a result one can still completely solve
the geodesic equations in terms of rst integrals of motion. We now give some details.
First we write down the geodesic hamiltonian, eectively representing the inverse metric
g

:
H =
1
2
2
_

2
p
2
r
+ p
2

+
_
a sin p
t
+
1
sin
p

_
2

2
_
( r
2
+ a
2
) p
t
+ a p

_
2
_
. (4.103)
Clearly, to be well-dened we must assume = (0, ), as the z-axis represents a coordinate
singularity in the Boyer-Lindquist system where the angle is not well-dened. The above
hamiltonian is a constant of geodesic ow; recall, that for time-like geodesics it takes the
value 2H = 1, whilst for light-like ones H = 0.
Next there are two Killing vectors, respresenting invariance of the geometry under
time-shifts and rotations around the z-axis:
= p
t
= g
tt
dt
d
g
t
d
d
,
= p

= g
t
dt
d
+ g

d
d
.
(4.104)
Of course, for time-like geodesics the ane parameter maybe taken to be proper time
. These equations can be viewed as establishing a linear relation between between the
constants of motion (, ) and the velocities (dt/d, d/d):
_

_
=

G
_

t

_
, (4.105)
with the dot denoting a derivative w.r.t. the ane parameter . The determinant of this
linear transformation is
det

G = g
tt
g

+ g
2
t
=

2
sin
2
, (4.106)
and changes sign whenever

2
does. Therefore the zeros of

2
dene the locus of points
where one of the eigenvalues of

G changes sign, from a time-like to a space-like variable
or vice-versa; at the same time, the coecient of p
2
r
in the hamiltonian (4.103) changes
sign, conrming that these values represent the horizons of the Kerr-Newman space-time.
It follows from this analysis, that outside the outer horizon, in the region connected to
space-like innity, det

G > 0.
Now if from eqs.(4.104) the angular proper velocity d/d is eliminated, one nds the
relation
=
det

G
g

dt
d
, (4.107)
80
where the quantity with the dimensions of an angular velocity is dened by
=
g
t
g

= a
r
2
+ a
2

2
( r
2
+ a
2
)
2

2
a
2
sin
2

=
a
r
2
+ a
2

1
1 +

2

2
( r
2
+ a
2
)
1
( r
2
+ a
2

2
)
1
.
(4.108)
Observe, that it is proportional to a with a constant of proportionality which for large but
nite r is positive. Hence represents an angular velocity with the same orientation as
the angular momentum J. With a canonical choice of coordinate system this quantity is
positive, and therefore > 0 as well.
It also follows from eq.(4.107) that for time-like geodesics outside the horizon, with
= and therefore dt/d > 0, an inequality holds of the form
> 0. (4.109)
For large r near innity one nds in fact the stronger inequality > 1.
Returning to the solution of the geodesic equations, we take into account the existence of
a constant of motion K such that 2K = K

, with K

the contravariant components


of a Killing tensor:
D
(
K
)
= 0. (4.110)
The explicit expression for K is
K =
1
2
2
_

2
a
2
cos
2
p
2
r
+ r
2
p
2

_
+
r
2
sin
2

2
2
_
ap
t
+
p

sin
2

_
2
+
a
2
cos
2

2
2

2
_
( r
2
+ a
2
)p
t
+ ap

_
2
.
(4.111)
That K denes a constant of geodesic ow is most easily established by checking that its
Poisson bracket with the hamiltonian H (4.103) vanishes:
{K, H} = 0. (4.112)
Using these constants of motion, it is straightforward to establish the solutions for time-like
geodesics in the form
81

2
dt
d
= a
_
1
r
2
+ a
2

2
_
+
( r
2
+ a
2
)
2

2
a
2
sin
2

2
,

2
d
d
=

2
a
2
sin
2

2
sin
2

a
_
1
r
2
+ a
2

2
_
,

4
_
dr
d
_
2
=

2
_
2K + r
2
_
+
_
a ( r
2
+ a
2
)
_
2
,

4
_
d
d
_
2
= 2K a
2
cos
2
a
2
sin
2

_

a sin
2


_
2
.
(4.113)
Note that for a particle starting from r = one has 1. These solutions are well-dened
everywhere except on the horizons and at points where
2
= 0. At the latter locus the
curvature invariant R

becomes innite and there is a real physical singularity.


At the horizons both t and change innitely fast, signalling innite growth of these
coordinates w.r.t. an observer in asymptotic at Minkowski space.
As with the innite amount of coordinate time t to reach the horizon in the Schwarzschild
and Reissner-Nordstrom solutions, this merely signies a coordinate singularity. However,
here it is not only t but also which becomes innite. To gain a better understanding of
the reason for this rotation of geodesics, rst notice that according to the second equation
(4.113) even for = 0 the proper angular velocity d/d does not vanish in general, be-
cause = 0. Hence there are no purely radial geodesics: any incoming test particle picks
up a rotation. Now computing the angular velocity as measured by an observer at rest at
spatial innity from the rst two equations (4.113) one nds:
d
dt
=
g
t
+ g
tt
g

+ g
t
. (4.114)
It follows immediately, that for a geodesic without orbital angular momentum: = 0, the
observed angular velocity equals the special reference value (4.108):
d
dt

=0
= . (4.115)
At the horizon r = r
+H
, where

2
= 0, this becomes
d
dt

_
= 0
r = r
+H
_
=
H
=
a
r
2
+H
+ a
2
. (4.116)
Comparison with the expression for , eq.(4.108), shows that outside the horizon the
inequality

H
, (4.117)
82
holds, with equality only on the horizon. As these quantities are nite, the time t can
increase near the horizon at an innite rate only if increases at a proportional rate. For
this reason both t and have to grow without bound as r approaches r
+H
.
At space-like innity the Killing vectors (4.104) dene time translations and plane
rotations. However, their nature changes if one approaches the horizon. Consider the time
translation generated by p
t
= . The corresponding Killing vector has contravariant and
covariant components, given by

t
=

t

t
= g
t
. (4.118)
Here

is the usual Kronecker delta symbol. Similarly the components of the Killing
vector for plane rotations, generated by p

, are

= g

. (4.119)
It follows, that these Killing vectors have norms given by
(
t
)
2
= g
tt
=

2
+ a
2
sin
2


2
, (4.120)
and
(

)
2
= g

= sin
2

( r
2
+ a
2
)
2

2
a
2
sin
2


2
, (4.121)
whilst their inner product is

= g
t
= a sin
2

r
2
+ a
2

2

2
. (4.122)
From these equations several observations follow. First, as the norms and inner products of
Killing vectors are coordinate independent quantities, we can write the expressions for the
reference angular velocity , eq.(4.108), and for the observed angular velocity, eq.(4.114),
in a coordinate independent form:
=

t

,
d
dt
=

t


2
t

. (4.123)
The last equality is equivalent to
( )
d
dt
=

2
t

=
_


2
t

_
. (4.124)
Second, with a > 0 and excluding the z-axis = (0, ) where the Boyer-Lindquist
coordinates are not well-dened and vanishes and the invariant inner product
t

are positive outside the horizon. Also the positivity of det



G in the exterior region,
eq.(4.106), can be expressed in the invariant form
83


2
t


2
t

1, (4.125)
with the equality holding only on the horizon. Then for 0 we nd from eq.(4.124)
that d/dt , whilst for < 0 the angular velocity is less than this reference value:
d/dt < . As long as
t
is time-like this is all that can be established; in fact, with < 0
and
2
t
< 0, the absolute value of the angular velocity d/dt can become arbitrarily large.
However, it is easily established that the norm of
t
changes sign on the surface
r() = r
0
() = m +

m
2
e
2
a
2
cos
2
, (4.126)
which, with the exception of the points where the surface cuts the z-axis: = (0, ),
lies outside the horizon. The region between this surface and the horizon is called the
ergosphere, and
t
is space-like there:
2
t
> 0. It then follows directly from eqs.(4.124) and
(4.125) that in this region the angular velocity is always positive:
d
dt
> 0, for
2
t
> 0. (4.127)
Thus we have established, that inside the ergosphere any test particle is dragged along
with the rotation of the black hole and no free-falling observer can be non-rotating there.
Related to this, both
t
and

are space-like Killing vectors there. Nevertheless, outside the


horizon there still exists time-like combinations of these vectors even inside the ergosphere:
=
t

, (4.128)
which has non-positive norm:
2
0 as a result of the inequality (4.125), with equality
holding only on the horizon r = r
H
.
4.9 The Kerr singularity
We continue the analysis of black-hole geometry with a brief discussion of the nature of
the singularity of spinning black holes. To begin, we have already noticed that, like the
spherical charged black hole, the spinning Kerr-Newman black holes have two horizons, an
innner and an outer one. As the singularity
2
= 0 is located inside the inner horizon, we
expect the singularity to have a time-like character.
In order to elucidate the topological structure of the space-time near the singularity,
we introduce new coordinates in two steps. First we dene the Kerr coordinates (

V , ) as
the solution of the dierential equations
d

V = dt +
( r
2
+ a
2
)

2
d r, d = d +
a

2
d r. (4.129)
With these denitions the Kerr-Newman line element becomes
84
ds
2
=

2

2
_
d

V a sin
2
d
_
2
+
sin
2


2
_
( r
2
+ a
2
)d ad

V
_
2
+2
_
d

V a sin
2
d
_
d r +
2
d
2
.
(4.130)
Now we turn from quasi-spherical coordinates to quasi-Cartesian ones, known as Kerr-
Schild coordinates, by dening
x = ( r cos a sin ) sin ,
y = ( r sin + a cos ) sin
z = r cos ,
(4.131)
and we reparametrize the time coordinate t

t using
d

t = d

V d r. (4.132)
We then nd that at xed (, ) the transformation
x x

= x 2 r cos ,
y y

= y 2 r sin ,
z z

= z,
(4.133)
is equivalent to r r

= r. This shows, that like the Reissner-Nordstrom case, the


negative values of r are to be taken as part of the physical domain of values.
Next one introduces the coecients k
i
(i = 1, 2, 3) by the relation
k dr
r (xdx + ydy) + a (xdy ydx)
r
2
+ a
2

zdz
r
d

t, (4.134)
and scalar function u(r):
u(r) =

2
+ r
2
+ a
2

2
=
2m r e
2
r
2
+
a
2
z
2
r
2
. (4.135)
With these denitions the metric can be written in the hybrid form
ds
2
= dx
2

+ u(r) (k dr)
2
, (4.136)
where the rst term on the right-hand side is to be interpreted as in at space-time, with
x

= (x, y, z,

t). Therefore u and k parametrize the deviation from at space-time.


Having introduced this new parametrization, the physical singularity
2
= 0, equivalent
to r = 0 and = /2, is now seen to be mapped to the ring
85
x
2
+ y
2
= a
2
, z = 0, (4.137)
with the solution
x = a sin , y = a cos . (4.138)
A consequence of this structure is, that in contrast to the Schwarzschild geometry, geodesic
crossing the horizon do not necessarily encounter the singularity, but can pass it by.
4.10 Black-holes and thermodynamics
When a test particle falls into the ergosphere of a black hole, the constant of motion

t
p = becomes a momentum component, rather than an energy. This provides a means
of extracting energy from a spinning black hole, as rst proposed by Penrose [48]: if a
particle enters the ergosphere with energy , this becomes a momentum; during its stay in
the ergosphere the momentum can be changed, for example by a decay of the particle into
two or more fragments, in such a way that the nal momentum of one of the fragments
is larger than the original momentum. If this fragment would again escape, which is
allowed as the ergosphere is located outside the horizon, it leaves with a dierent energy

corresponding to its changed momentum. This energy is now larger than the energy
with which the original particle entered. By this process it is possible to extract energy
from a black hole, even in classical physics.
It would seem that the Penrose process violates the conservation of energy, if there would
be no limit to the total amount of energy that can be extracted this way. However, it can
be established that such a violation is not possible and that the process is self-limiting:
emission of energy is always accompanied by a decrease in the angular momentum (and/or
electric charge) of the black hole in such a way, that after extraction of a nite amount of
energy the black hole becomes spherical. After that there is no longer an ergosphere and
no energy extraction by the Penrose process is possible.
We will now show how this comes about for the simplest case of an electrically neutral
black hole of the Kerr type. In this case the location of the horizon is at
r
H
= m +

m
2
a
2
= m +
1
m

m
4
J
2
. (4.139)
Then the angular momentum of test particles moving tangentially to the horizon is

H
=
a
r
2
H
+ a
2
=
J
2m

1
m
2
+

m
4
J
2
. (4.140)
This quantity can be interpreted in terms of the area of the spherical surface formed by
horizon of the black hole:
A
H
=
_ _
r= r
H
dd

= 4
_
r
2
H
+ a
2
_
=
4a

H
. (4.141)
86
Eq.(4.140) thus provides an expression for the area of the horizon in terms of the mass and
the angular momentum of the black hole:
A
H
8
= m
2
+

m
4
J
2
. (4.142)
Now we observe that
(i) the area A
H
increases with increasing mass;
(ii) at xed area A
H
the mass reaches a minimum for J = 0;
(iii) by absorbing infalling test particles (including Penrose-type processes) the black-hole
area A
H
never decreases.
The rst two statements are obvious from eq.(4.142). The last one follows if we consider
the eect of changes in the mass and angular momentum of the black hole. We assume
that the absorption of a test particle with energy and angular momentum changes a
Kerr black hole of total mass m and angular momentum J to another Kerr black hole of
mass m

= m+ and angular momentum J

= J + . This assumption is justied by the


uniqueness of the Kerr solution for stationary rotating black holes.
Now under arbitrary changes (m, J) the corresponding change in the area is
A
H
8
=
2mm

m
4
J
2
_
m
2
+

m
4
J
2
_

JJ

m
4
J
2
=
J

m
4
J
2
(m
H
J) .
(4.143)
For an infalling test particle only subject to the gravitational force of the black hole we
have derived the inequality (4.109). Moreover, on the horizon such a particle has been
shown to have the angular velocity
H
w.r.t. a stationary observer at innity, independent
of the value of . Hence with m = and J = we obtain
A
H
8
=
J

m
4
J
2
(
H
) > 0. (4.144)
Thus we can indeed conclude that by such adiabatic processes, adding innitesimal amounts
of energy and angular momentum at a time, the area of the black hole never decreases.
Casting the relation (4.143) in the form
m =

8
A
H
+
H
J, (4.145)
where is given by:
=
4
mA
H

m
4
J
2
, (4.146)
it bears a remarkable resemblance to the laws of thermodynamics which states that changes
of state described by equilibrium thermodynamics are such that for isolated systems vari-
ations in energy U, entropy S and angular momentum J are related by
87
U = TS + J, (4.147)
where T is the temperature and the total angular momentum, and moreover that in
such processes one always has S > 0. The important step needed in a full identication
of these relations is to associate the area A
H
with the entropy and the quantity , called
the surface gravity, with the temperature:
S = cA
H
, T =

8c
, (4.148)
where c is an unknown constant of proportionality. Now if black holes have a nite tem-
perature, they should emit thermal radiation. Using a semi-classical approximation, it was
shown by Hawking [45], that quantum uctuations near the horizon of a black hole indeed
lead to the emission of particles with a thermal spectrum. This is true irrespective of the
existence of an ergosphere, therefore it covers the case of vanishing angular momentum as
well. The temperature of the radiation Hawking found was
T
H
=

2
. (4.149)
This xes the unknown constant above to the value
c =
1
4
. (4.150)
It seems therefore, that the thermodynamical description of black holes is more than a
mere analogy. The issue of interpreting the entropy
S =
A
H
4
+ S
0
, (4.151)
where S
0
is an unknown additive constant, in statistical terms by counting microstates is
hotly debated and has recently received a lot of impetus from developments in superstring
theory.
The extension of these results to the case of charged black holes is straightforward. The
relation between horizon area A
H
and angular velocity
H
remains the as in (4.141), but
in terms of the macroscopic observables (m, J, e) associated with the black hole the area is
now
A
H
8
= m
2

e
2
2
+

m
4
J
2
m
2
e
2
. (4.152)
The changes in these observable quantities by absorbing (possibly charged) test particles
are then related by
m =

8
A
H
+
H
J +
H
e, (4.153)
where the electric potential at the horizon is
88

H
=
4e r
H
A
H
, (4.154)
and the constant is now dened as
=
4
mA
H

m
4
J
2
m
2
e
2
. (4.155)
Again, by sending test particles into the black hole the area can never decrease: A
H
> 0.
In this way the thermodynamical treatment is generalized to include processes in which
the electric charge changes.
We conclude this discussion with establishing the gravitational interpretation of the con-
stant , associated with the temperature in the thermodynamical description. This quan-
tity is known as the surface gravity, because in the spherically symmetric case it represents
the (outward) radial acceleration necessary to keep a test particle hovering on the horizon
without actually falling into the black hole.
For the case of the Schwarzschild black hole this is easily established. Dene the
covariant acceleration as

=
Du

D
. (4.156)
As the four velocity of a test particle has negative unit norm: u
2
= 1, it follows that in
the four-dimensional sense the acceleration is orthogonal to the velocity:
u = 0. (4.157)
Also, a particle at rest in the frame of an asymptotic inertial observer at innity, has a
velocity four-vector of the form
u

=
_
1

g
tt
, 0, 0, 0
_
. (4.158)
Combining these two equations and taking into account the spherical symmetry, a particle
at rest must have a purely radial acceleration given by

r
=
1
2

r
g
tt
=
m
r
2
. (4.159)
Remarkably, we recover Newtons law! It follows in particular that on the horizon

r
H
=
1
4m
= . (4.160)
This quantity is actually a proper scalar invariant, as can be seen from writing it as
=
_

2
t
_
r
H
, (4.161)
89
with the invariant magnitude of the four acceleration:
=
_
g

. (4.162)
For the case of the spinning black holes the proper generalization of this result gives the
surface gravity as in eq.(4.155).
90
Bibliography
[1] S. Newcomb, Astronomical Papers Am. Ephemeris 1 (1882), 472
[2] H. Hertz, Wied. Ann. 31 (1887), 421
[3] A. Einstein, Ann. d. Physik 49 (1916) 769; reproduced in: The Principle of Relativity,
ed. A. Sommerfeld (Dover publ., 1952)
[4] H. Bondi, Nature 179 (1957), 1072
[5] J. Ehlers and W. Kundt, in: Gravitation: An Introduction to Current Research
(Wiley, N.Y.; 1962), ed. L. Witten
[6] A. Einstein, The Meaning of Relativity (Chapman and Hall; 6th ed., 1956)
[7] W. Pauli, Theory of Relativity (Engl.transln.: Pergamon Press, 1958; Dover Publ.
Inc., 1981)
[8] H. Weyl, Space-Time-Matter (Engl.transln.: Dover Publ. Inc., 1952)
[9] W. Rindler, Essential Relativity (Springer Verlag; 1977)
[10] S. Weinberg, Gravitation and Cosmology (J. Wiley, N.Y.; 1972)
[11] C.W. Misner, K.S. Thorn and J.A. Wheeler, Gravitation (W.H. Freeman, San Fran-
sisco; 1973)
[12] A. Papapetrou, Lectures on General Relativity (Reidel, Dordrecht; 1974)
[13] S.W. Hawking and G.F. Ellis, The Large-Scale Structure of of Spacetime (Cambridge
Univ. Press; 1973)
[14] R. Wald, General Relativity (Univ. of Chicago Press; 1984)
[15] D. Kramer, H. Stephani, M. MacCallum and E. Herlt, Exact Solution of Einsteins
Field Equations (Cambridge Univ. Press; 1980)
[16] R. Arnowitt, S. Deser and C.W. Misner, in: Gravitation, an Introduction to Current
Research (Wiley, N.Y.; 1962), ed. L. Witten
91
92
[17] M. Veltman, in: Methods in Field Theory, Les Houches 1975 (North-Holland; 1976);
eds. R. Balian and J. Zinn-Justin
[18] R.H. Rietdijk and J.W. van Holten, Class. Quantum Grav. 7 (1990), 247
[19] R.H. Rietdijk and J.W. van Holten, J. of Geom. and Physics 11 (1993), 559
[20] G.W. Gibbons, R.H. Rietdijk and J.W. van Holten, Nucl. Phys. B404 (1993), 42
[21] L. Brink, P. Di Vecchia and P. Howe, Phys. Lett. B65 (1976), 471
[22] M.B. Green, J.H. Schwarz and E. Witten, Superstring Theory, 2 vols. (Cambridge
Univ. Press; 1986)
[23] E. Witten, Nucl. Phys. B311 (1988), 46
[24] A. Einstein, Sitzber. Preus. Akad. Wiss. (1915), 844; Ann. d. Physik 49 (1916), 769
[25] D. Kramer et al., ibid., ch. 21
[26] P.M. Morse and H. Feshbach, Methods in Theoretical Physics, (McGraw-Hill, 1953)
[27] J.W. van Holten, Phys. Lett. B352 (1995), 220
[28] J.W. van Holten, preprint NIKHEF/95-007 (1995)
[29] C.J. Joshi and P.B. Corkum, Physics Today 48 (1995), 36
[30] W. Heisenberg and H. Euler , Z. Phys. 98 (1936), 718
[31] V. Weisskopf, K. Dan. Vidensk. Selsk. Mat. Fys. Medd. 14, no. 6 (1936)
[32] W. Greiner, B. M uller and J. Rafelski, Quantum Electrodynamics of Strong Fields
(Springer Verlag, 1985)
[33] J. Goldstone, Nuov. Cim. 19 (1961), 154
[34] Y. Nambu and G. Jona-Lasinio, Phys. Rev. 122 (1961), 345
[35] S. Chandrasekhar, The Mathematical Theory of Black Holes, (Oxford Univ. Press;
1983)
[36] I.D. Novikov and V.P. Frolov, Physics of Black Holes, (Kluwer Ac. Publ..; 1989)
[37] K. Schwarzschild, Sitzber. Deut. Akad. Wiss. (Berlin, 1916), 189
[38] J. Droste, Het zwaartekrachtsveld van een of meer lichamen volgens de theorie van
Einstein, PhD thesis (Leiden, 1916)
93
[39] K. Peeters, C. Schweigert and J.W. van Holten, Class. Quantum Grav. 12 (1995),
173
[40] M.D. Kruskal, Phys. Rev. 119 (1960), 1743
G. Szekeres, Publ. Mat. Debrecen 7 (1960), 285
[41] H. Reissner, Ann. Physik 50 (1916), 106
G. Nordstrom, Proc. Kon. Ned. Ak. Wet. 20 (1918), 1238
[42] R.P. Kerr, Phys. Rev. Lett. 11 (1963), 237
[43] E.T. Newman, E. Couch, K. Chinnapared, A. Exton, A. Prakash and R. Torrence,
J. Math. Phys. 6 (1965), 918
[44] R.H. Boyer and R.W. Lindquist, J. Math. Phys. 8 (1967), 265
[45] S.W. Hawking, Nature 248 (1974), 30
Comm. Math. Phys. 43 (1975), 199
[46] B. Carter, Phys. Rev. 174 (1968), 1559
[47] M. Walker and R. Penrose, Comm. Math. Phys. 18 (1970), 265
[48] R. Penrose, Nuov. Cim. 1 (1969), 252

You might also like