Institut fr Theoretische Physik der Universitt Zrich

in conjunction with ETH Zrich

General Relativity
Autumn semester 2013

Prof. Philippe Jetzer

Original version by Arnaud Borde

Revision: Antoine Klein, Raymond Anglil, Cdric Huwyler

Last revision of this version: January 28, 2014

Sources of inspiration for this course include

S. Weinberg, Gravitation and Cosmology, Wiley, 1972
N. Straumann, General Relativity with applications to Astrophysics, Springer Verlag, 2004
C. Misner, K. Thorne and J. Wheeler, Gravitation, Freeman, 1973
R. Wald, General Relativity, Chicago University Press, 1984
T. Fliessbach, Allgemeine Relativittstheorie, Spektrum Verlag, 1995
B. Schutz, A first course in General Relativity, Cambridge, 1985
R. Sachs and H. Wu, General Relativity for mathematicians, Springer Verlag, 1977
J. Hartle, Gravity, An introduction to Einsteins General Relativity, Addison Wesley, 2002
H. Stephani, General Relativity, Cambridge University Press, 1990, and
M. Maggiore, Gravitational Waves: Volume 1: Theory and Experiments, Oxford University
Press, 2007.
A. Zee, Einstein Gravity in a Nutshell, Princeton University Press, 2013

As well as the lecture notes of

Sean Carroll (http://arxiv.org/abs/gr-qc/9712019),
Matthias Blau (http://www.blau.itp.unibe.ch/lecturesGR.pdf), and
Gian Michele Graf (http://www.itp.phys.ethz.ch/research/mathphys/graf/gr.pdf).

CONTENTS

Contents
I

Introduction

Newtons theory of gravitation

Goals of general relativity

II

Special Relativity

Lorentz transformations

3.1

Galilean invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2

Lorentz transformations

3.3

Proper time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

Relativistic mechanics

13

4.1

Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4.2
4.3

Energy and momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Equivalence between mass and energy . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Tensors in Minkowski space

14

Electrodynamics

17

Accelerated reference systems in special relativity

18

III
8

10

Towards General Relativity

20

The equivalence principle

20

8.1

About the masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

8.2

About the forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

8.3

Riemann space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

Physics in a gravitational field

25

9.1

Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

9.2

Christoffel symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

9.3

Newtonian limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

Time dilation

28

10.1

Proper time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

10.2

Redshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

10.3

Photon in a gravitational field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

CONTENTS

11

Geometrical considerations
11.1

IV
12

13

30

Curvature of space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

Differential Geometry

32

Differentiable manifolds

32

12.1

Tangent vectors and tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

12.2

The tangent map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

Vector and tensor fields

13.1

37

Flows and generating vector fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

14

Lie derivative

40

15

Differential forms

42

16

17

15.1

Exterior derivative of a differential form . . . . . . . . . . . . . . . . . . . . . . . . . 43

15.2

Stokes theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

15.3

The inner product of a p-form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

Affine connections: Covariant derivative of a vector field

48

16.1

Parallel transport along a curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

16.2

Round trips by parallel transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

16.3

Covariant derivatives of tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

16.4

Local coordinate expressions for covariant derivative . . . . . . . . . . . . . . . . . . 55

Curvature and torsion of an affine connection, Bianchi identities

17.1

57

Bianchi identities for the special case of zero torsion . . . . . . . . . . . . . . . . . . 59

18

Riemannian connections

60

General Relativity

64

19

Physical laws with gravitation

64

20

19.1

Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

19.2

Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

19.3

Energy-momentum tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

Einsteins field equations

20.1

21

66

The cosmological constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

The Einstein-Hilbert action

68

CONTENTS

22

23

Static isotropic metric

22.1

Form of the metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

22.2

Robertson expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

22.3

Christoffel symbols and Ricci tensor for the standard form . . . . . . . . . . . . . . 72

22.4

Schwarzschild metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

General equations of motion

23.1

VI

71

74

Trajectory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

Applications of General Relativity

80

24

Light deflection

80

25

Perihelion precession

83

25.1

Quadrupole moment of the Sun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

26

Lie derivative of the metric and Killing vectors

87

27

Maximally symmetric spaces

88

28

Friedmann equations

91

NEWTONS THEORY OF GRAVITATION

Part I

Introduction
1

Newtons theory of gravitation

In his book Principia in 1687, Isaac Newton laid the foundations of classical mechanics and made a
first step in unifying the laws of physics.
The trajectories of N point masses, attracted to each other via gravity, are the solutions to the equation
of motion
mi

N
X
d2~ri
mi mj (~ri ~rj )
=
G
dt2
|~ri ~rj |3
j=1

i = 1 . . . N,

(1.1)

j6=i

with ~ri (t) being the position of point mass mi at time t. Newtons constant of gravitation is determined
experimentally to be
G = 6.6743 0.0007 1011 m3 kg1 s2

(1.2)

The scalar gravitational potential (~r) is given by

(~r) = G

N
X
j=1

mj
= G
|~r ~rj |

d3 r0

(~r 0 )
,
|~r ~r 0 |

(1.3)

where it has been assumed that the mass is smeared out in a small volume d3 r. The mass is given by
dm = (~r 0 )d3 r0 , (~r 0 ) being the mass density. for point-like particles we have (~r 0 ) mj (3) (~r 0 ~rj ).
The gradient of the gravitational potential can then be used to produce the equation of motion:
m

d2~r
= m(~r).
dt2

(1.4)

According to (1.3), the field (~r) is determined through the mass of the other particles. The corresponding field equation derived from (1.3) is given by1
(~r) = 4G(~r)

(1.5)

The so called Poisson equation (1.5) is a linear partial differential equation of 2nd order. The source of
the field is the mass density. Equations (1.4) and (1.5) show the same structure as the field equation
of electrostatics:
e (~r) = 4e (~r),

(1.6)

and the non-relativistic equation of motion for charged particles

m

d2~r
= qe (~r).
dt2

(1.7)

Here, e is the charge density, e is the electrostatic potential and q represents the charge which acts
as coupling constant in (1.7). m and q are independent characteristics of the considered body. In
1

1
|~
r ~
r 0|

= 4 (3) (~
r~
r 0)

GOALS OF GENERAL RELATIVITY

analogy one could consider the gravitational mass (right side) as a charge, not to be confused with
the inertial mass (left side). Experimentally, one finds to very high accuracy ( 1013 ) that they
are the same. As a consequence, all bodies fall at a rate independent of their mass (Galileo Galilei).
This appears to be just a chance in Newtons theory, whereas in GR it will be an important starting
point.
For many applications, (1.4) and (1.5) are good enough. It must however be clear that these
equations cannot be always valid. In particular (1.5) implies an instantaneous action at a distance,
what is in contradiction with the predictions of special relativity. We therefore have to suspect that
Newtons theory of gravitation is only a special case of a more general theory.

Goals of general relativity

In order to get rid of instantaneous interactions, we can try to perform a relativistic generalization of
Newtons theory (eqs. (1.4) and (1.5)), similar to the transition from electrostatics (eqs. (1.6) and
(1.7)) to electrodynamics.
The Laplace operator is completed such as to get the DAlembert operator (wave equation)
=

1 2

c2 t2

(2.1)

Changes in e travel with the speed of light to another point in space. If we consider inertial coordinate
frames in relative motion to each other it is clear that the charge density has to be related to a
current density. In other words, charge density and current density transform into each other. In
electrodynamics we use the current density j ( = 0, 1, 2, 3):
e (e c, e v i ) = j ,

(2.2)

where the v i are the cartesian components of the velocity ~v (i = 1, 2, 3). An analogous generalization
can be performed for the potential:
e (e , Ai ) = A .

(2.3)

The relativistic field equation is then

e = 4e A =

4
j .
c

(2.4)

In the static case, the 0-component reduces to the equation on the left.
Equation (2.4) is equivalent to Maxwells equations (in addition one has to choose a suitable gauge
condition). Since electrostatics and Newtons theory have the same mathematical structure, one may
want to generalize it the same way. So in (1.5) one could introduce the change . Similarly one
generalizes the mass density. But there are differences with electrodynamics. The first difference is
that the charge q of a particle is independent on how the particle moves; this is not the case for the
mass: m = pm0 2 .
1 vc2

As an example, consider a hydrogen atom with a proton (rest mass mp , charge +e) and an electron
(rest mass me , charge e). Both have a finite velocity within the atom. The total charge of the atom
7

GOALS OF GENERAL RELATIVITY

is q = qe + qp = 0, but for the total mass we get mH 6= mp + me (binding energy). Formally this
means that charge is a Lorentz scalar (does not depend on the frame in which the measurement is
performed). Therefore we can assign a charge to an elementary particle, and not only a charge at
rest, whereas for the mass we must specify the rest mass.
q
Since charge is a Lorentz scalar, the charge density (e = V
) transforms like the 0-component
1
1

due to length contraction). The mass density

of a Lorentz vector ( V gets a factor =
2
2
1v /c

( =

m
V )

transforms instead like the 00-component of a Lorentz tensor, which we denote as the

energy-momentum tensor T . This follows from the fact that the energy (mass is energy E = mc2 )
is the 0-component of a 4-vector (energy-momentum vector p ) and transforms as such. Thus, instead
of (2.2), we shall have

c2

cv i

cv i

v i v j

!
T

i, j = 1, 2, 3

(2.5)

This implies that we have to generalize the gravitational potential to a quantity depending on 2
indices which we shall call the metric tensor g . Hence we get
= 4G g GT .

(2.6)

In GR one finds (2.6) for a weak gravitational field (linearized case), e.g. used for the description of
gravitational waves.
Due to the equivalence between mass and energy, the energy carried by the gravitational field is
also mass and thus also a source of the gravitational field itself. This leads to non-linearities. One can
note that photons do not have a charge and thus Maxwells equations can be linear.
To summarize:
1. GR is the relativistic generalization of Newtons theory. Several similarities between GR and
electrodynamics exist.
2. GR requires tensorial equations (rather than vectorial as in electrodynamics).
3. There are non-linearities which will lead to non-linear field equations.

LORENTZ TRANSFORMATIONS

Part II

Special Relativity
3

Lorentz transformations

A reference system with a well defined choice of coordinates is called a coordinate system. Inertial
reference systems (IS) are (from a practical point of view) systems which move with constant speed
with respect to distant (thus fixed) stars in the sky. Newtons equations of motion are valid in IS. NonIS are reference systems which are accelerated with respect to an IS. In this chapter we will establish
how to transform coordinates between different inertial systems.

3.1

Galilean invariance

Galilei stated that all IS are equivalent, i.e. all physical laws are valid in any IS: the physical laws
are covariant under transformations from an IS to another IS. Covariant means here form invariant.
The equations should have the same form in all IS.
With the coordinates xi (i = 1, 2, 3) and t, an event in an IS can be defined. In another IS, the
same event has different coordinates x0i and t0 . A general Galilean transformation can then be written
as:

x0i = i k xk + v i t + ai ,

(3.1)

t0 = t + ,

(3.2)

where
xi , v i and ai are cartesian components of vectors
~v = v i~ei where ~ei is a unit vector
we use the summation rule over repeated indices: i k xk =

i k xk

latin indices run on 1,2,3

greek indices run on 0,1,2,3
~v is the relative velocity between IS and IS
~a is a constant vector (translation)
i k is the relative rotation of coordinates systems, = (i k ) is defined by
i n (T )nk = i k

T = I i.e.

or

1 = T

(3.3)

LORENTZ TRANSFORMATIONS

The condition T = I ensures that the line element

ds2 = dx2 + dy 2 + dz 2

(3.4)

remains invariant. can be defined by giving 3 Euler angles. Eqs. (3.1) and (3.2) define a 10
(a = 3, v = 3, = 1 and = 3) parametric group of transformations, the so-called Galilean
group.
The laws of mechanics are left invariant under transformations (3.1) and (3.2). But Maxwells equations
are not invariant under Galilean transformations, since they contain the speed of light c. This led
Einstein to formulate a new relativity principle (special relativity, SR): All physical laws, including
Maxwells equations, are valid in any inertial system. This leads us to Lorentz transformations (instead
of Galilean), thus the law of mechanics have to be modified.

3.2

Lorentz transformations

We start by introducing 4-dimensional vectors, glueing time and space together to a spacetime. The
Minkowski coordinates are defined by
x0 = ct,

x1 = x,

x2 = y,

x3 = z.

(3.5)

x is a vector in a 4-dimension space (or 4-vector). An event is given by x in an IS and by x0 in an

IS. Homogeneity of space and time imply that the transformation from x to x0 has to be linear:
x0 = x + a ,

(3.6)

where a is a space and time translation. The relative rotations and boosts are described by the 4 4
matrix . Linear means in this context that the coefficients and a do not depend on x . In
order to preserve the speed of light appearing in Maxwells equations as a constant, the
have to be
such that the square of the line element
ds2 = dx dx = c2 dt2 d~r2
remains unchanged, with the Minkowski metric

0
=
0

(3.7)

0
.
0

(3.8)

Because of ds2 = ds02 c2 d 2 = c2 d 02 , the proper time is an invariant under Lorentz transformations.
 x 2
 
2
+dy 2 +dz 2
 and c =  d~x . Applying a Lorentz
Indeed for light d 2 = dt2 dx
= 0. Thus, c2 =  d~
dt
dt
 0 c2
 x
transformation results in c =  d~
This
has
the
important
consequence
that
the speed of light c is
.
0
dt
the same in all coordinate systems (what we intended by the definition of (3.7)).

10

LORENTZ TRANSFORMATIONS

A 4-dimensionial space together with this metric is called a Minkowski space. Inserting (3.6) into the
invariant condition ds2 = ds02 gives
ds02 = dx0 dx0
= dx dx
= dx dx = ds2 .

(3.9)

Then we get
=

or

T = .

(3.10)

Rotations are special subcases incorporated in : x0 = x with i k = i k , and 0 0 = 1,

i 0 = 0 i = 0. The entire group of Lorentz transformations (LT) is the so called Poincar group (and
has 10 parameters). The case a 6= 0 corresponds to the Poincar group or inhomogeneous Lorentz
group, while the subcase a = 0 can be described by the homogeneous Lorentz group. Translations and
rotations are subgroups of Galilean and Lorentz groups.
Consider now a Lorentz boost in the direction of the x-axis: x02 = x2 , x03 = x3 . v denotes the
relative velocity difference between IS and the boosted IS. Then

0 0 0 1 0 0

1
0 1 1 0 0
.
=
0
0
1 0

0
0
0 1

(3.11)

Evaluating eq. (3.10):

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

(0 0 )2 (1 0 )2 = 1

(3.12a)

(0 1 )2 (1 1 )2 = 1

(3.12b)

0 0 0 1 1 0 1 1 = 0

= (0, 1) or (1, 0)

(3.12c)

The solution to this system is

0 0

0 1

1 0

1 1

!
=

cosh

sinh

sinh

cosh

(3.13)

For the origin of IS we have x01 = 0 = 1 0 ct + 1 1 vt. This way we find

tanh =

1 0
v
= ,
0 0
c

(3.14)

and as a function of velocity:

1
0 0 = 1 1 = = q
1
v/c
0 1 = 1 0 = q
.
2
1 vc2
11

(3.15a)

v2
c2

(3.15b)

LORENTZ TRANSFORMATIONS

A Lorentz transformation (called a boost) along the x-axis can then be written explicitly as
x vt
,
x0 = q
2
1 vc2

(3.16a)

y 0 = y,

(3.16b)

z 0 = z,

(3.16c)

ct x vc
ct0 = q
,
2
1 vc2

(3.16d)

which is valid only for |v| < c. For |v|  c, (3.16) recovers the special (no rotation) Galilean transformation x0 = x vt, y 0 = y, z 0 = z and t0 = t. The parameter
v
c

= arctanh

(3.17)

is called the rapidity. From this we find for the addition of parallel velocities:
= 1 + 2
v1 + v2
1 + v1c2v2

v=

3.3

(3.18)

Proper time

The time coordinate t in IS is the time shown by clocks at rest in IS. We next determine the proper
time shown by a clock which moves with velocity ~v (t). Consider a given moment t0 an IS, which
moves with respect to IS with a constant velocity ~v0 (t0 ). During an infinitesimal time interval dt0 the
clock can be considered at rest in IS, thus:
r
0

d = dt =
0

t(1v02 /c2 )

Indeed (3.16) with x = v0 t gives t = q

v2
1 c20

q
=t 1

v02
dt.
c2
v02
c2

(3.19)

and thus (3.19).

At the next time t0 + dt, we consider an IS with velocity ~v0 = ~v (t0 + dt) and so on. Summing up
all infinitesimal proper times d gives the proper time interval:
Zt2

r
dt 1

v 2 (t)
c2

(3.20)

t1

This is the time interval measured by an observer moving at a speed v (t) between t1 and t2 (as given
by a clock at rest in IS). This effect is called time dilation.

12

RELATIVISTIC MECHANICS

Relativistic mechanics

Let us now perform the relativistic generalization of Newtons equation of motion for a point particle.

4.1

Equations of motion

The velocity ~v can be generalized to a 4-velocity vector u :

vi =
Since d =

ds
c ,

dx
dxi
u =
dt
d

(4.1)

d is invariant. With dx0 =

dx it follows that u transforms like dx :

u0 = u

(4.2)

All quantities which transforms this way are Lorentz vectors or form-vectors. The generalized equation
of motion is then
m
Both

du
d

du
= f .
d

(4.3)

and f are Lorentz vectors, therefore, (4.3) is a Lorentz vector equation: if we perform a
0

0
Lorentz transformation, we get m du
transformations
d = f . Eq. (4.3) is covariant under Lorentz


d~
v
and for
 v  c itreduces
 to Newtons equations. (left hand side becomes m 0, dt and the right hand
0 ~
~
side f , f = 0, K ). The Minkowski force f 0 is determined in any IS through a corresponding

1/2
1
2
01
LT: f 0 = f . For example ~v = v~e1 with = 1 vc2
, leads to f 00 = vK
= K 1 ,
c , f

f 02 = K 2 and f 03 = K 3 . For a general direction of velocity (~v ) we get:

f 00 =

4.2

~
~v K
,
c

~
~ + ( 1) ~v ~v K .
f~0 = K
2
v

(4.4)

Energy and momentum

The 4-momentum p = mu = m dx
d is a Lorentz vector. With (3.19) we get



i
mc
mv
= E , p~ .
p = q
,q
2
2
c
1 vc2
1 vc2

(4.5)

This yields the relativistic

energy :

mc2
E=q
= mc2
v2
1 c2

(4.6a)

momentum :

m~v
vc
p~ = q
= m~v p~ = m~v .
2
1 vc2

(4.6b)

With (4.4), the 0-component of (4.3) becomes (in the case v  c)

dE
=
dt

~
~v .K
|{z}

power given
to the particle

13

(4.7)

TENSORS IN MINKOWSKI SPACE

This justifies to call the quantity E = mc2 an energy. From ds2 = c2 d 2 = dx dx it follows
p p = m2 c2 and thus
E 2 = m2 c4 + c2 p~ 2 ,

(4.8)

the relativistic energy-momentum relation. The limit cases are

mc2 + p2 v  c or p  mc2
p
2m
2
4
2
2
E = m c + c p~
cp
v c or p  mc2

(4.9)

with p = |~
p|. For particles with no rest mass (photons): E = cp (exact relation).

4.3

Equivalence between mass and energy

One can divide the energy into the rest energy

E0 = mc2

(4.10)

and the kinetic energy Ekin = E E0 = E mc2 . The quantities defined in (4.6) are conserved when
more particles are involved. Due to the equivalence between energy and mass, the mass or the mass
density becomes a source of the gravitational field.

Tensors in Minkowski space

Let us discuss the transformation properties of physical quantities under a Lorentz transformation.
We have already seen how a 4-vector is transformed:
V V 0 = V .

(5.1)

This is a so-called contravariant 4-vector (indices are up). The coordinate system transforms according
to X X 0 = X . A covariant 4-vector is defined through
V = V .

(5.2)

Let us now define the matrix as the inverse matrix to :

= .
In our case

0 1
= =
0 0

0 0

(5.3)

0
0
1
0

0
.
0

(5.4)

With (5.3) we can express (5.2) equally as

V = V .

14

(5.5)

TENSORS IN MINKOWSKI SPACE

The transformation of a covariant vector is then given by

V ,
V0 = V 0 = V = V =

(5.6)

(5.7)

with
but one should be very careful in writings since 6= ). Thanks
(one can use instead of

to (3.10), we find
=   =   =

(5.8)

= . In matrix notation, we have

=
= I and thus
= 1 .
And similarly, we get

To summarize the transformations of 4-vectors:

A contravariant vector transforms with

A covariant vector transforms with 1 =

The scalar product of two vectors V and U is defined by
V U = V U = V U = V U

(5.9)

V U = V U .
and is invariant under Lorentz transformations: V 0 U0 =
| {z }

The operator x transforms like a covariant vector: x0 = x

0 x . Since x0 = x0 =
. We will now use the notations (covariant vector) and (contravariant

x
x
x

vector). The DAlembert operator can be written as  = = =

1 2
c2 t2

and is a

Lorentz scalar.
A quantity is a rank r contravariant tensor if its components transform like the coordinates x :
T 01 ...r = 1 1 . . . r r T 1 ...r

(5.10)

Tensors of rank 0 are scalars, tensors of rank 1 are vectors. For mixed tensors we have for example:

T 
T 0 =

The following operations can be used to form new tensors:

1. Linear combinations of tensors with the same upper and lower indices: T = aR + bS
2. Direct products of tensors: T V (works with mixed indices)
3. Contractions of tensors: T or T V (lowers a tensor by 2 in rank)
4. Differentiation of a tensor field: T (the derivative of any tensor is a tensor with one
additional lower index : T R )
5. Going from a covariant to a contravariant component of a tensor is defined like in (5.2) and (5.5)
(lowering and raising indices with , ).
15

TENSORS IN MINKOWSKI SPACE

One must be aware that

the order of the upper and lower indices is important,
is not a tensor.
can be considered as a tensor: = = is the Minkowski tensor.
0

(3.10)

(5.8)
=

=
=

appears in the line element (ds2 = dx dx ) and is thus the metric tensor in Minkowski space.
We also have = = = , and thus the Kronecker symbol is also a tensor.
We define the totally antisymmetric tensor or (Levi-Civita tensor) as

+1 (, , , ) is an even permutation of (0, 1, 2, 3)


= 1 (, , , ) is an odd permutation of (0, 1, 2, 3)

0
otherwise

(5.11)

Without proof we have: (det () = 1)

0 =  ,
0

 = 0 0 0 0 

0 0

=  .

The functions S(x), V (x), T . . . with x = (x0 , x1 , x2 , x3 ) are a scalar field, a vector field, or a tensor
field . . . respectively if:
S 0 (x0 ) = S(x)

V 0 (x0 ) =
V (x)

T 0 (x0 ) =
T (x)

..
.
Also the argument has to be transformed, thus x0 has to be understood as x0 = x .

16

ELECTRODYNAMICS

Electrodynamics

~ r, t), B(~
~ r, t), the charge density e (~r, t) and the current density
Maxwells equations relate the fields E(~
~(~r, t):

inhomogeneous

~ = 4c

div E

homogeneous

rot B
~ = 4 ~ + 1 E
c
c t

~ =0

div B

rot E
~ = 1 B
c t

(6.1)

The continuity equation

div ~ + c = 0 j = 0

(6.2)

with
Z j = (ce , ~) follows from the conservation of charge, which for an isolated system implies
t j 0 d3 r = 0. j is a Lorentz scalar. We can define the field strength tensor which is given

by the antisymmetric matrix

Ex
F =
E
y
Ez

Ex

Ey

Ez

Bz

Bz

By

Bx

By
.
Bx

(6.3)

Using this tensor we can rewrite the inhomogeneous Maxwell equations

4
F =
| {z }
c

4vector

j ,
|{z}

(6.4)

4vector

and also the homogeneous ones:

 F = 0.

(6.5)

Both equations are covariant under a Lorentz transformation. Eq. (6.5) allows to represent F as a
curl of a 4-vector A :
F = A A .

(6.6)

We can then reformulate Maxwells equations for A = (, Ai ). From (6.6) it follows that the gauge
transformation
A A +

(6.7)

of the 4-vector A leaves F unchanged, where (x) is an arbitrary scalar field. The Lorenz gauge
A = 0 leads to the decoupling of the inhomogeneous Maxwells equation (6.4) to
A =

4
j .
c

(6.8)

The generalized equation of motion for a particle with charge q is

m

du
q
= F u
d
c

17

(6.9)

ACCELERATED REFERENCE SYSTEMS IN SPECIAL RELATIVITY

The spatial components give the expression of the Lorentz force



d~
p
~ + ~v B
~ with p~ = m~v .
=q E
dt
c

The energy-momentum tensor for the electromagnetic field is



1
1

Tem
=
F F + F F
4
4

(6.10)



1
00
~2 + B
~ 2 and the
E
The 00-component represents the energy density of the field Tem
= uem = 8

i
c
0i
~ i = cTem
~ B
~ . In terms of these tensors, Maxwells
0i-components the Poynting vector S
= 4
E
1

equations are Tem

= F j . Tem
is symmetric and conserved: Tem
= 0. Setting = 0
c
k
leads to energy conservation whereas Tem = 0 leads to conservation of the k th component of the

momentum. One should note that Tem

= 0 is valid only if there is no external force, otherwise we

can write Tem

= f , where f is the external force density. Such an external force can often be

included in the energy-momentum tensor.

Accelerated reference systems in special relativity

Non inertial systems can be considered in the context of special relativity. However, then the physical
laws no longer have their simple covariant form. In e.g. a rotating coordinate system, additional terms
will appear in the equations of motion (centrifugal terms, Coriolis force, etc.).
Let us look at a coordinate system KS (with coordinates x0 ) which rotates with constant angular
speed with respect to an inertial system IS (x ):

x = x0 cos(t0 ) y 0 sin(t0 ),

y = x0 sin(t0 ) + y 0 cos(t0 ),

(7.1)

z = z0,

t = t0 ,
and assume that 2 (x02 + y 02 )  c2 . Then we insert (7.1) into the line element ds (in the known IS
form):
ds2 = dx dx = c2 dt2 dx2 dy 2 dz 2


= c2 2 (x02 + y 02 ) dt02 + 2y 0 dx0 dt0 2x0 dy 0 dt0 dx02 dy 02 dz 02
= g dx0 dx0 .

(7.2)

The resulting line element is more complicated. For arbitrary coordinates x0 , ds2 is a quadratic form
of the coordinate differentials dx0 . Consider a general coordinate transformation from x (in IS) to
x0 (in KS):
x x (x0 ) = x (x00 , x01 , x02 , x03 ),

18

(7.3)

ACCELERATED REFERENCE SYSTEMS IN SPECIAL RELATIVITY

then we get for the line element

ds2 = dx dx =
with

x x 0 0
dx dx = g (x0 )dx0 dx0 ,
x0 x0

(7.4)

x x
.
(7.5)
x0 x0
is the metric tensor of the KS system. It is symmetric (g = g ) and depends on
g (x0 ) =

The quantity g

x0 . It is called metric because it defines distances between points in coordinate systems.

In an accelerated reference system we get inertial forces. In the rotating frame we expect to
~ which can be written in terms of a centrifugal potential :
experience the centrifugal force Z,
=

2 02
(x + y 02 )
2

~ = m.
~
and Z

(7.6)

This enables us to see that g00 from (7.2) is

g00 = 1 +

2
.
c2

(7.7)

The centrifugal potential appears in the metric tensor. We will see later that the first derivatives of
the metric tensor are related to the forces in the relativistic equations of motion. To get the meaning
of t0 in KS we evaluate (7.2) at a point with dx0 = dy 0 = dz 0 = 0:
r
r
2 0
v2
dsclock

0
= g00 dt = 1 + 2 dt =
1 2
d =
c
c
| {z c }

dt

(7.8)

correspond to clocks
time computed in
an inertial system

represents the time of a clock at rest in KS.

In an inertial system we have g = and the clock moves with speed v = ( =

x02 + y 02 ).

With (7.6) we see that both expressions in (7.8) are the same.
The coefficients of the metric tensor g (x0 ) are functions of the coordinates. Such a dependence
will also arise when one uses curved coordinates. Consider for example cylindrical coordinates:
x00 = ct = x0 ,

x01 = ,

x02 = ,

x03 = z.

This results in the line element

ds2 = c2 dt2 dx2 dy 2 dz 2 = c2 dt2 d2 2 d2 dz 2 = g (x0 )dx0 dx0 .

(7.9)

Here g is diagonal:

1
2

(7.10)

The fact that the metric tensor depends on the coordinates can be either due to the fact that the
considered coordinate system is accelerated or that we are using non-cartesian coordinates.
19

THE EQUIVALENCE PRINCIPLE

Part III

Towards General Relativity

8

The equivalence principle

The principle of equivalence of gravitation and inertia tells us how an arbitrary physical system responds to an external gravitational field (with the help of tensor analysis). The physical basis of
general relativity is the equivalence principle as formulated by Einstein:
1. Inertial and gravitational mass are equal
2. Gravitational forces are equivalent to inertial forces
3. In a local inertial frame, we experience the known laws of special relativity without gravitation

8.1

About the masses

The inertial mass mt is the quantity appearing in Newtons law F~ = mt ~a which acts against acceleration by external forces. In contrast, the gravitational mass ms is the proportionality constant relating
the gravitational force between mass points to each other. For a particle moving in a homogeneous
gravitational field, we have the equation mt z = ms g, whose solution is
z(t) =

1 ms 2
gt
2 mt

(+v0 t + z0 ).

Galilei stated that all bodies fall at the same rate in a gravitational field, i.e.

(8.1)
ms
mt

is the same for

all bodies. Another experiment is to consider the period T of a pendulum (in the small amplitude


T 2
s l
approximation): 2
=m
mt g , where l is the length of the pendulum. Newton verified that this period
is independent on the material of the pendulum to a precision of about 103 . Etvs (1890), using
torsion balance, got a precision of about 5 109 . Todays precision is about 1011 1012 , this is
way we can make the assumption ms = mt on safe grounds.
Due to the equivalence between energy and mass (E = mc2 ), all forms of energy contribute to
mass, and due to the first point of the equivalence principle, to the inertial and to the gravitational
masses.

8.2

About the forces

As long as gravitational and inertial masses are equal, then gravitational forces are equivalent to inertial
forces: going to a well-chosen accelerated reference frame, one can get rid of the gravitational field. As
an example take the equation of motion in the homogeneous gravitational field at Earths surface:
mt

d2~r
= ms~g
dt2

20

(8.2)

THE EQUIVALENCE PRINCIPLE

This expression is valid for a reference system which is at rest on Earths surface ( to a good
approximation an IS). Then we perform the following transformation to an accelerated KS system:
1
~r = ~r 0 + gt02 ,
2

t = t0 ,

(8.3)

and we assume gt  c. The origin of KS ~r 0 = 0 moves in IS with ~r(t) = 21 gt2 . Then, inserting (8.3)
into (8.2) results in
d2
mt 02
dt



1 02
0
~r + gt
= ms~g
2
mt

d2~r 0
= (ms mt ) ~g .
dt02

If ms = mt , the resulting equation in KS is the one of a free moving particle

(8.4)
d2 ~
r0
dt02

= 0; the gravitational

force vanishes. As another example in a free falling elevator the observer does not feel any gravity.
Einstein generalized this finding postulating that (this is the Einstein equivalence principle) in a
free falling accelerated reference system all physical processes run as if there is no gravitational field.
Notice that the mechanical finding is now expanded to all types of physical processes (at all times
and places). Moreover also non-homogeneous gravitational fields are allowed. The equality of inertial
and gravitational mass is also called the weak equivalence principle (or universality of the free fall).
As an example of a freely falling system, consider a satellite in orbit around Earth (assuming that
the laboratory on the satellite is not rotating). Thus the equivalence principle states that in such a
system all physical processes run as if there would be no gravitational field. The processes run as in an
inertial system: the local IS. However, the local IS is not an inertial system, indeed the laboratory on
the satellite is accelerated compared to the reference system of the fixed distant stars. The equivalence
principle implies that in a local IS the rules of special relativity apply.
For an observer on the satellite laboratory all physical processes follow special relativity and
there are neither gravitational nor inertial forces.
For an observer on Earth, the laboratory moves in a gravitational field and moreover inertial
forces are present, since it is accelerated.
The motion of the satellite laboratory, i.e. its free falling trajectory, is such that the gravitational
forces and inertial forces just compensate each other (cf (8.4)). The compensation of the forces is
exactly valid only for the center of mass of the satellite laboratory. Thus the equivalence principle
applies only to a very small or local satellite laboratory (how small depends on the situation).
The equivalence principle can also be formulated as follows:
At every space-time point in an arbitrary gravitational field, it is possible to choose a
locally inertial coordinate system such that, within a sufficiently small region around the
point in question, the laws of nature take the same form as in non-accelerated Cartesian
coordinate systems in the absence of gravitation.2
2 Notice the analogy with the axiom Gauss took as a basis of non-Euclidean geometry: he assumed that at any point
on a curved surface we may erect a locally Cartesian coordinate system in which distances obey the law of Pythagoras.

21

THE EQUIVALENCE PRINCIPLE

The equivalence principle allows us to set up the relativistic laws including gravitation; indeed one can
just perform a coordinate transformation to another KS:
o
nrelativistic laws
special relativity laws
coordinate
with

without
gravitation

transf ormation

gravitation

The coordinate transformation includes the relative acceleration between the laboratory system and
KS which corresponds to the gravitational field. Thus from the equivalence principle we can derive the
relativistic laws in a gravitational field. However, it does not fix the field equations for g (x) since
those equations have no analogue in special relativity.
From a geometrical point of view the coordinate dependence of the metric tensor g (x) means
that space is curved. In this sense the field equations describe the connection between curvature of
space and the sources of the gravitational field in a quantitative way.

8.3

Riemann space

We denote with the Minkowski coordinates in the local IS (e.g. the satellite laboratory). From the
equivalence principle, the special relativity laws apply. In particular, we have for the line element
ds2 = d d .

(8.5)

Going from the local IS to a KS with coordinates x is given by a coordinate transformation =

(x0 , x1 , x2 , x3 ). Inserting this into (8.5) results in
ds2 =
and thus g (x) =
space.

dx dx = g (x)dx dx ,
x x

(8.6)

. A space with such a path element of the form (8.6) is called a Riemann
x x

The coordinate transformation (expressed via g ) also describes the relative acceleration between
KS and the local IS. Since at two different points of the local IS the accelerations are (in general)
different, there is no global transformation in the form (8.6) that can be brought to the Minkowski
form (8.5). We shall see that g are the relativistic gravitational potentials, whereas their derivatives
determine the gravitational forces.

22

THE EQUIVALENCE PRINCIPLE

Figure 1: An experimenter and his two stones freely floating somewhere in outer space, i.e. in the
absence of forces.

Figure 2: Constant acceleration upwards mimics the effect of a gravitational field: experimenter and
stones drop to the bottom of the box.

23

THE EQUIVALENCE PRINCIPLE

Figure 5: The experimenter and his stones in a

Figure 3: The effect of a constant gravitational

non-uniform gravitational field: the stones will ap-

field: indistinguishable for our experimenter from

proach each other slightly as they fall to the bot-

that of a constant acceleration as in figure 2.

tom of the elevator.

Figure 6: The experimenter and stones freely

falling in a non-uniform gravitational field: the ex-

Figure 4: Free fall in a gravitational field has the

same effect as no gravitational field (figure 1): ex- perimenter floats, so do the stones, but they move
closer together, indicating the presence of some
perimenter and stones float.
external forces.

24

9
9.1

PHYSICS IN A GRAVITATIONAL FIELD

Physics in a gravitational field

Equations of motion

According to the equivalence principle, in a local IS the laws of special relativity hold. For a mass
point on which no forces act we have
d2
= 0,
(9.1)
d 2
where the proper time is defined through ds2 = d d = c2 d 2 . We can also define the 4-velocity
as u =

d
d .

Solutions of (9.1) are straight lines

= a + b .

(9.2)

Light (or a photon) moves in the local IS on straight lines. However, for photons cannot be identified
with the proper time since on the light cone ds = cd = 0. Thus we denote by a parameter of the
trajectory of photons:
d2
= 0.
(9.3)
d2
Let us now consider a global coordinate system KS with x and metric g (x). At all points x , one can
locally bring ds2 into the form ds2 = d d . Thus at all points P there exists a transformation
(x) = (x0 , x1 , x2 , x3 ) between and x . The transformation varies from point to point. Consider
a small region around point P . Inserting the coordinate transformation into the line element, we get
ds2 = d d =

| x
{z x }

dx dx .

(9.4)

g (x) metric tensor

We write (9.1) in the form

d
0=
d
multiply it by

dx
x d

and make use of

x
x


=

d2 x
2 dx dx
+
,
x d 2
x x d d

= . This way we can solve for

d2 x
d 2

and get the following

equation of motion in a gravitational field

d2 x
dx dx
=
,
2
d
d d
with

(9.5)

x 2
.
(9.6)
x x
are called the Christoffel symbols and represent a pseudo force or fictive gravitational field
=

The

(like centrifugal or Coriolis forces) that arises whenever one describes inertial motion in non-inertial
coordinates. Eq. (9.5) is a second order differential equation for the functions x ( ) which describe

du
the trajectory of a particle in KS with g (x). Eq. (9.5) can also be written as m
= f , u = dx
.
d
d
Comparing with (4.3) one infers that the right hand side of (9.5) describes the gravitational forces.
Due to (9.4), the velocity

dx
d

has to satisfy the condition

c2 = g

dx dx
d d
25

(for m 6= 0)

(9.7)

PHYSICS IN A GRAVITATIONAL FIELD

(assume d 6= 0 and m 6= 0). Due to (9.7) only 3 of the 4 components of

dx
d

are independent (this

is also the case for the 4-velocity in special relativity). For photons (m = 0) one finds instead, using
(9.3), a completely analogous equation for the trajectory:

d2 x
dx dx
=

d2
d d

(9.8)

and since d = ds = 0, one has instead of (9.7):

0 = g

9.2

dx dx
d d

(for m = 0).

Christoffel symbols

The Christoffel symbols can be expressed in terms of the first derivatives of g . Consider with (9.4):


2
2
g
g


g

=
+



x
x
x
x x x

|x x
{z x }
1

2


+

x x x + x
x x

|
{z
}
2



2

2 


.

+


x



| x
|x x
{z x } 
{z x }
2

Using = this becomes

= 2

2
.
x x x

(9.9)

On the other hand

}|
{z
}|
{
x 2
=
x x x x
| {z }

2
x x x



1 g
g
g
=
+

.
2 x
x
x

(9.10)

We introduce the inverse matrix g such that g g = . Therefore we can solve with respect to
the Christoffel symbols:
=



1 g
g
g
g
+

.
2
x
x
x
26

(9.11)

PHYSICS IN A GRAVITATIONAL FIELD

Note that the s are symmetric in the lower indices = . The gravitational forces on the right
hand side of (9.6) are given by derivatives of g . Comparing with the equation of motion of a particle
in a electromagnetic field shows that the correspond to the field F , whereas the g correspond
to the potentials A .

9.3

Newtonian limit

Let us assume that v i  c and the fields are weak and static (i.e. not time dependent). Thus
dxi
d

dx0
d .

Inserting this into (9.5) leads to

2

d x
dx dx
=
d 2
d d

small
velocity

z}|{
00

dx0
d

2
.

(9.12)

For static fields we get from (9.11):

staticity

00

z}|{
g i g00
=
2 xi

(i = 1, 2, 3)

(9.13)

(the other terms contain partial derivative with respect to x0 which are zero by staticity). We write
the metric tensor as g = + h . For weak fields we have |h | = |g |  1. In this case the
coordinates (ct, xi ) are almost Minkowski coordinates. Inserting the expansion for g into (9.13)
(taking only linear terms in h) gives
00


=


1 h00 k
.
0,
2 xi i

(9.14)

Then, let us compute (9.12) for = 0, = j:

choice

z}|{
d2 t
dt
=0
= constant = 1,
2
d
d
 2
d2 xj
c2 h00 dt
=

.
d 2
2 xj d
| {z }

(9.15a)

(9.15b)

12

Taking (xj ) = ~r, we can write

d2~r
c2
h00 (~r),
=

dt2
2
which is to be compared with the Newtonian case

(9.16)

d2~r
= (~r). Therefore:
dt2

g00 (~r) = 1 + h00 (~r) = 1 +

2(~r)
.
c2

(9.17)

Notice that the Newtonian limit (9.16) gives no clue on the other components of h . The quantity
2
c2

is a measure of the strength of the gravitational field. Consider a spherically symmetric mass

27

10

TIME DILATION

distribution. Then
2(R)
1.4 109
c2

10

at Earth surface,

4 106

on the Sun (and similar stars),

3 104

on a white dwarf,

3 101

on a neutron star GR required.

Time dilation

We study a clock in a static gravitational field and the phenomenon of gravitational redshift.

10.1

Proper time

The proper time of the clock is defined through the 4-dimensional line element as
q

dsclock
1
g (x)dx dx
,
=
d =
c
c
clock

(10.1)

x = (x ) are the coordinates of the clock. The time shown by the clock depends on both the gravitational field and of its motion (the gravitational field being described by g ).
Special cases:
1. Moving clock in an IS without gravity :
r
d =

v2
dt
c2

(g = , dxi = v i dt, dx0 = cdt).

2. Clock at rest in a gravitational field (dxi = 0)
d =

g00 dt.

For a weak static field, one has with (9.17):

r
2(r)
d = 1 +
dt
c2

(||  c2 ).

(10.2)

The fact that is negative implies that a clock in a gravitational field goes more slowly than a
clock outside the gravitational field.

10.2

Redshift

Let us now consider objects which emit or absorb light with a given frequency. Consider only a static
gravitational field (g does not depend on time). A source in ~rA (at rest) emits a monochromatic
28

10

TIME DILATION

electromagnetic wave at a frequency A . An observer at ~rB , also at rest, measures a frequency B .

p
At source:
dA = g00 (~rA )dtA
(10.3)
p
At observer: dB = g00 (~rB )dtB
As a time interval we consider the time between two following peaks departing from A or arriving
at B. In this case dA and dB correspond to the period of the electromagnetic waves at A and B,
respectively, and therefore
dA =

1
,
A

dB =

1
.
B

(10.4)

Going from A to B needs the same time t for the first and the second peak of the electromagnetic
wave. Consequently, they will arrive with a time delay which is equal to the one with which they were
emitted, thus dtA = dtB . With (10.3) and (10.4) we get:
s
A
g00 (~rB )
A
B
=
,
with z =
1=
1.
B
g00 (~rA )
B
A

(10.5)

The quantity z is the gravitational redshift:

s
z=
For weak fields with g00 = 1 +

2
c2

g00 (~rB )
1.
g00 (~rA )

(10.6)

we have
z=

(~rB ) (~rA )
c2

(||  c2 ),

(10.7)

such a redshift is observed by measuring spectral lines from stars. As an example take solar light with
(10.7)
z=

(rA )
GM
(rB ) (rA )
2 = 2
2 106 ,
2
c
c
c R

with M 2 1030 kg and R 7 108 m. For a white dwarf we find z 104 and for a neutron star
z 101 . In general there are 3 effects which can lead to a modification in the frequency of spectral
lines:
1. Doppler shift due to the motion of the source (or of the observer)
2. Gravitational redshift due to the gravitational field at the source (or at the observer)
3. Cosmological redshift due to the expansion of the Universe (metric tensor is time dependent)

10.3

Photon in a gravitational field

Consider a photon with energy E = ~ = 2~, travelling upwards in the homogeneous gravity field
of the Earth, covering a distance of h = hB hA (h small). The corresponding redshift is
z=

A
(rB ) (rA )
g(hB hA )
gh
1=
=
= 2,
B
c2
c2
c
29

(10.8)

11

GEOMETRICAL CONSIDERATIONS

resulting in a frequency change = B A (A > B , B = ) and thus

gh
= 2.

(10.9)

The photon changes its energy by E = c2 gh (like a particle with mass c2 = m). This effect has
exp
been measured in 1965 (through the Mssbauer effect) as
= 1.00 0.01 (1% accuracy)3 .
th

11

Geometrical considerations

In general, the coordinate dependence of g (x) means that spacetime, defined through the line element
ds2 , is curved. The trajectories in a gravitational field are the geodesic lines in the corresponding
spacetime.

11.1

Curvature of space

The line element in an N -dimensional Riemann space with coordinates x = (x1 , . . . , xN ) is given as
ds2 = g dx dx

(, = 1, . . . , N ).

Let us just consider a two dimensional space x = (x1 , x2 ) with

ds2 = g11 dx1 dx1 + 2g12 dx1 dx2 + g22 dx2 dx2 .

(11.1)

Examples:
Plane with Cartesian coordinates (x1 , x2 ) = (x, y):
ds2 = dx2 + dy 2 ,

(11.2)

or with polar coordinates (x1 , x2 ) = (, ):

ds2 = d2 + 2 d2

(11.3)

Surface of a sphere with angular coordinates (x1 , x2 ) = (, ):

ds2 = a2 d2 + sin2 d2

(11.4)

The line element (11.2) can, via a coordinate transformation, be brought into the form (11.3). However,
there is no coordinate transformation which brings (11.4) into (11.2). Thus:
The metric tensor determines the properties of the space, among which is also the curvature.
The form of the metric tensor is not uniquely determined by the space, in other words it depends
on the choice of coordinates.
3 Pound,

R. V. and Snider, J. L., Effect of Gravity on Gamma Radiation, Physical Review, 140

30

11

GEOMETRICAL CONSIDERATIONS

The curvature of the space is determined via the metric tensor (and it does not depend on the coordinate
choice)4 . If gik = const then the space is not curved. In an Euclidian space, one can introduce Cartesian
coordinates gik = ik . For a curved space gik 6= const (does not always imply that space is curved).
For instance by measuring the angles of a triangle and checking if their sum amounts to 180 degrees
or differs, one can infer if the space is curved or not (for instance by being on the surface of a sphere).

4 Beside

the curvature discussed here, there is also an exterior curvature. We only consider intrinsic curvatures here.

31

12

DIFFERENTIABLE MANIFOLDS

Part IV

Differential Geometry
12

Differentiable manifolds

A manifold is a topological space that locally looks like the Euclidean Rn space with its usual topology.
A simple example of a curved space is the S 2 sphere: one can setup local coordinates (, ) which map
S 2 onto a plane R2 (a chart). Collections of charts are called atlases. There is no one-to-one map of
S 2 onto R2 ; we need several charts to cover S 2 .
Definition: Given a (topological) space M, a chart on M is a one-to-one map from an open subset
U M to an open subset (U) Rn , i.e. a map : M Rn . A chart is often called a coordinate
S
system. A set of charts with domain U is called an atlas of M, if U = M, { | I}.

Definition: dim M = n
Definition: Two charts 1 , 2 are C -related if both the maps 2 1
1 : 1 (U1 U2 ) 2 (U1 U2 )
and its inverse are C . 2 1
1 is the so-called transition function between the two coordinate charts.
A collection of C related charts such that every point of M lies in the domain of at least one chart
forms an atlas (C : derivatives of all orders exist and are continuous).
The collection of all such C -related charts forms a maximal atlas. If M is a space and A its maximal
atlas, the set (M, A) is a (C )-differentiable manifold. (If for each in the atlas the map : U Rn
has the same n, then the manifold has dimension n.)
Important notions:
A differentiable function f : M R belongs to the algebra F = C (M), sum and product of
such functions are again in F = C (M).
Fp is the algebra of C -functions defined in any neighbourhood of p M (f = g means f (q) =
g(q) in some neighbourhood of p).
A differentiable curve is a differentiable map : R M.
Differentiable maps F : M M0 are differentiable if 2 F 1
1 is a differentiable map for all
suitable charts 1 of M and 2 of M0 .
The notions have to be understood by means of a chart, e.g. f : M R is differentiable if x 7
f (p(x)) f (x) is differentiable. This is independent of the chart representing a neighbourhood of p.
|{z}
M

32

12

DIFFERENTIABLE MANIFOLDS

M
U1

U2
p

1
Rn

Rn

2 1
1

X
(Chart 1 ) Rn

(Chart 2 ) Rn

Figure 7: Manifold, charts and transition function.

12.1

Tangent vectors and tangent spaces

At every point p of a differentiable manifold M one can introduce a linear space, called tangent space
Tp (M). A tensor field is a (smooth) map which assigns to each point p M a tensor of a given type
on Tp (M).
Definition: a C -curve in a manifold M is a map of the open interval I = (a, b) R M such
that for any chart , : I Rn is a C map.

33

12

DIFFERENTIABLE MANIFOLDS

Let f : M R be a smooth function on M. Consider the map f : I R, t 7 f ((t)). This has

a well-defined derivative: the rate of change of f along the curve. Consider f 1
| {z }
n

R R
xi f (xi )
=f (1 (xi ))

use the chain rule:

X
d
(f ) =
dt
i=1

f (xi )
xi

and

IR
txi ((t))

dxi ((t))
.
dt

(12.1)


Thus, given a curve (t) and a function f , we can obtain a qualitatively new object

| {z }



d
,
(f ) 
dt
t=t0

the rate of change of f along the curve (t) at t = t0 .

Definition: The tangent vector p to a curve (t) at a point p is a map from the set of real functions
f defined in a neighbourhood of p to R defined by

p : f 7


d
(f ) = (f ) p = p (f ).
dt
p

Given a chart with coordinates xi , the components of p with respect to the chart are


d i
i

(x ) p =
x ((t)) .
dt
p

(12.2)

(12.3)

The set of tangent vectors at p is the tangent space Tp (M) at p.

Theorem: If the dimension of M is n, then Tp (M) is a vector space of dimension n (without proof).
We set (0) = p (t = 0), Xp = p , and Xp f = p (f ). Eq. (12.3) determines Xp (xi ), the components
of Xp with respect to a given basis:
Xp f = [f ] (0)


= f 1 (0)
=

n
X

d
(f 1 ) (xi )(0)
i
x
dt
i=1

(12.4)


X


1
n
=
f (x , . . . , x ) Xp (xi ) .
i
x
i
This way we see that
Xp =

Xp (x )

xi


,

(12.5)

span Tp (M). From (12.5) we see that Xp (xi ) are the components of Xp with
xi p
respect to the given basis (Xp (xi ) = Xpi or X i ).

and so the

34

12

DIFFERENTIABLE MANIFOLDS

Suppose that f, g are real functions on M and f g : M R is defined as f g(p) = f (p)g(p). If

Xp Tp (M), then (Leibniz rule)
Xp (f g) = (Xp f )g(p) + f (p)(Xp g).

(12.6)

Notation: (Xf )(p) = Xp f , p M.

Basis of Tp (M): Tp = TP (M) has dimension n. In any basis (e1 , . . . , en ) we have X = X i ei . Changes
of basis are given by
i = i k X k .
X

ei = i k ek ,

(12.7)

The transformations i k and i k are inverse transposed of each other. In particular, ei =

xi

is called

coordinate basis (with respect to a chart). Upon change of chart x 7 x

,
i k =

xk
,
x
i

i k =

x
i
.
xk

(12.8)

Definition: The cotangent space Tp (or dual space Tp of Tp ) consists of covectors Tp , which are
linear one-forms : X 7 (X) < , X > R ( : Tp R).
In particular for functions f , df : X 7 Xf is an element of Tp . The elements df = f,i dxi =

f
xi

dxi

form a linear space of dimension n, therefore all of Tp .

We can define a dual basis (e1 , . . . , en ) of Tp : = i ei . In particular the dual basis of a basis
(e1 , . . . , en ) of Tp is given by < ei , X >= X i or < ei , X j ej >= X j < ei , ej > = X i . Thus i =< , ei >.
| {z }
i j

Upon changing the basis, the i transform like the ei and the ei like the X i (see (12.7)). In particular
we have for the coordinate basis ei =

xi ,

i
ei = dxi (< ei , ej >=< dxi , x
j >= j ). The change of

basis is:

xk

=
= i k k
i
i
k
x

x
x
x
d
xi =

x
i k
dx = i k dxk
xk

(Similar to co- and contravariant vectors.)

Tensors on Tp are multilinear forms on Tp and Tp , i.e. a tensor T of type
T (, X, Y ) is a trilinear form on

Tp

1
2

(for short T 12 Tp ):

Tp Tp . The tensor product is defined between tensors of any

type, i.e. T (, X, Y ) = R(, X)S(Y ) : T = R S. In components:

T (, X, Y ) = T (ei , ej , ek )
{z
}
|
T i jk

ei

i X j Y k
| {z }

()ej (X)ek (Y

(12.9)

hence T = T i jk ei ej ek . Any tensor of any type can therefore be obtained as a linear combination
of tensor products X 0 with X Tp , , 0 Tp . A change of basis can be performed similarly
35

12

DIFFERENTIABLE MANIFOLDS

to the ones for vectors and covectors:

i
Tjk
= T i j k

(12.10)

Trace: any bilinear form b Tp Tp determines a linear form l (Tp Tp ) such that l(X ) = b(X, ).


In particular tr T is a linear form on tensors T of type 11 , defined by tr(X ) =< , X >. In
components with respect to a dual pair of bases we have: tr T = T . Similarly T i jk 7 Sk = T i ik




defines for instance a map from tensors of type 12 to tensors of type 01 .

12.2

The tangent map

and let p M, p = (p). Then induces a linear

Definition: Let be a differentiable map: M M
map (push-forward):

: Tp (M) Tp(M),
which we can describe in two ways:
(F: space of all smooth functions on M (or M),
that is C map f : M R):
(a) For any f Fp (M)
( X)f = X(f )
(b) Let be a representative of X (X = p , see (12.2) and (12.3)). Then let = be a


d
d
representative of X. This agrees with (a) since dt
f (
(t))t=0 = dt
(f )((t))t=0 .
this reads X
= X: X
i = ( )i X k
With respect to bases (e1 , . . . , en ) of Tp and (
e1 , . . . , en ) of Tp(M),
k
i
x

i
i
i
with ( ) k =< e , ek > or in case of coordinate bases: ( ) k =
.
xk
7
(= in
Definition: The adjoint map (or pull-back) of is defined as : Tp Tp ,
Tp ) with <
, X >=<
, X >. The same result is obtained from the definition
: df 7 d(f ),

f F(M).

(12.10a)

In components, =
reads k =
i ( )i k .
1
Consider (local) diffeomorphisms,
 i  i.e. maps such that exists in a neighbourhood of p. Note that
and det xj 6= 0. Then and , as defined above, are invertible and may be
dim M = dim M
x

extended to tensors of arbitrary types.

Example: tensor of type

1
1

= T (
),
( T )(
, X)
, 1 X
|{z} | {z }

( T)(, X) = T(( )1 , X ).
| {z } | {z }

36

13

VECTOR AND TENSOR FIELDS

Here, and are the inverse of each other and we have

(T S) = ( T ) ( S),
(12.11)
tr( T ) = (tr T ),
and similarly for . In components T = T reads
x
i x
Tki = T k
x x

(in a coordinate basis).

(12.12)

This is formally the same as for transformation (12.10) when changing basis.

13

Vector and tensor fields

Definition: If to every point p of a differentiable manifold M a tangent vector Xp Tp (M) is assigned,

then we call the map X: p 7 Xp a vector field on M.

Given a coordinate system xi and associated basis x

for each Tp (M), Xp has components Xpi with
i
p

Xp = Xpi x
and Xpi = Xp (xi ) (see (12.5)). Eq. (12.8) shows how Xpi transform under coordinate
i
p
transformations. The quantity Xf is called the derivative of f with respect to the vector field X. The
following rules apply:
X(f + g) = Xf + Xg,
(13.1)
X(f g) = (Xf )g + f (Xg)

(Leibnitz rule).

The vector fields on M form a linear space on which the following operations are defined as well:
X 7 f X

(multiplication by f F),

X, Y 7 [X, Y ] = XY Y X

(commutator).

[X, Y ], unlike XY , satisfies the Leibniz rule (13.1). The components of the commutator of two vector
fields X, Y relative to a local coordinate basis can be obtained by its action on xi . Thus using

k
X = X i x
we get
i and Y = Y
xk
j

[X, Y ] = (XY Y X)xj

Y xj = Y k

xj
= Y k j k = Y j
xk

XY j = X k

(Y j ) = X k Y j ,k
xk
|{z}
Y j
xk

XY j Y X j = X k Y j ,k Y k X j ,k
In a local coordinate basis, the bracket [k , j ] clearly vanishes (X k = 1 and Y k = 1, and thus Y,jk = 0).
The Jacobi identity holds:
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
37

(13.2)

13

VECTOR AND TENSOR FIELDS

Definition: Let Tp (M)rs be the set of all tensors of rank (r, s) defined on Tp (M) (contravariant of rank
r, covariant of rank s). If we assign to every p M a tensor tp Tp (M)rs , then the map t : p 7 tp


defines a tensor field of type rs .
Algebraic operations on tensor fields are defined point-wise; for instance the sum of two tensor fields
is defined by (t + s)p = tp + sp where t, s Tp (M)rs . Tensor products and contractions of tensor fields
are defined analogously. Multiplication by a function f F(M) is given by (f t)p = f (p)tp . In a
neighbourhood U of p, having coordinates (x1 , . . . , xn ) a tensor field can be expanded in the form



i1 ...ir
t=
t
...
dxj1 . . . dxjs .
(13.3)
j1 ...js
i
i
1
r
x
x
| {z }
components of t relative
to the coordinate system
(x1 , . . . , xn )

If the coordinates are transformed to (

x1 , . . . , x
n ) the components of t transform according to
ti1 ...ir

j1 ...js

tk1 ...kr l1 ...ls

x
i1
x
ir xl1
xls
.
.
.
.
.
.
.
xk1
xkr x
j1
x
js

(13.4)

(We shall consider C tensor fields). Covariant tensors of order 1 are also called one-forms. The set


of tensor fields of type rs is denoted by Tsr (M).
Definition: A pseudo-Riemannian metric on a differentiable manifold M is a tensor field g T20 (M)
having the properties:
(i) g(X, Y ) = g(Y, X) for all X, Y
(ii) For every p M, gp is a non-degenerate (6= 0) bilinear form on Tp (M). This means that
gp (X, Y ) = 0 for all X Tp (M) if and only if Y = 0.
The tensor field g T20 (M) is a (proper) Riemannian metric if gp is positive definite at every point p.
Definition: A (pseudo-)Riemannian manifold is a differentiable manifold M, together with a (pseudo-)
Riemannian metric g.

13.1

Flows and generating vector fields

A flow is a 1-parametric group of diffeomorphisms: t : M M, s, t R with t s = t+s . In

particular 0 = id. Moreover, the orbits (or integral curves) of any point p M, t 7 t (p) (t)
shall be differentiable. A flow determines a vector field X by means of


d
Xf =
(f t )
(13.5)
dt
t=0

d
(t)t=0 = (0)

(see (12.2) and (12.3)). (0)

is the tangent vector to at the point

i.e. Xp = dt
p = (0). At the point (t) we have then
(t)



d
d
t (p) =
(s t ) (p)
= Xt (p)
dt
ds
s=0
38

13

VECTOR AND TENSOR FIELDS

i.e. (t) solves the ordinary differential equation:

(t)

= X(t) ,

(0) = p.

(13.6)

The generating vector field determines the flow uniquely. Not always does (13.6) admit global solutions
(i.e. for all t R), however for most purposes, local flows are good enough.

39

14

14

LIE DERIVATIVE

Lie derivative

The derivative of a vector field V rests on the comparison of Vp and Vp0 at nearby points p, p0 . Since
Vp Tp and Vp0 Tp0 belong to different spaces their difference can be taken only after Vp0 has been
transported to Vp . This can be achieved by means of the tangent map (Lie transport). The Lie
derivative LX R of a tensor field R in direction of a vector field X is defined by

d 
,
(14.1)
t R
LX R =
dt
t=0


d
or more explicitly (LX R)p =
t Rt (p)  . Here t is the (local) flow generated by X, where
dt
t=0
t Rt (p) is a tensor on Tp depending on t.
R t (p)

Rp

t (R t (p)) = t (R t (p))

t (p)
p

1
d 
LX R =
= lim ( R R)
R
dt t t=0 t0 t t
d
t t (p) = (t); Xp = dt
(t)|t=0 = (0)

( is the inverse of )
Figure 8: Illustration of the Lie derivative

In order to express LX in components we write t in a chart: t : x 7 x

(t), and linearize it for small
2

x
t: x
i = xi + tX i (x) + O(t2 ), xi = x
i tX i (
x) + O(t2 ), thus x
= X i ,k at t = 0.
k t = x
k t


i
x

As an example, let R be of type 11 . By (12.12) we have (t R)i j (x) = R (

x) x
x
xj . Taking

(according to (14.1)) a derivative with respect to t at t = 0 yields

(LX R)i j = Ri j,k X k R j X i , + Ri X ,j
(first term:



x
k xi x

R
(
x
)
= Ri j,k X k ).

k
xj 

t=0
|
{z
} |{z}
R ,k (
x)

Xk

40

(14.2)

14

LIE DERIVATIVE

Properties of LX :
(a) LX is a linear map from tensor fields to tensor fields of the same type.
(b) LX (tr T ) = tr(LX T )
(c) LX (T S) = (LX T ) S + T (LX S)
(d) LX f = Xf

(f F(M))

(e) LX Y = [X, Y ] (Y vector field)

(proof: (a) follows from (14.1), (b) and (c) from (12.11), (d) from (13.5), whereas (e) is more involved).
Further properties of LX : if X,Y are vector fields and R, then
(i) LX+Y = LX + LY ,

LX = LX

(ii) L[X,Y ] = [LX , LY ] = LX LY LY LX

Proof of (ii): apply it to f F(M),

[LX , LY ]f = XY f Y Xf = [X, Y ]f = L[X,Y ] f,
then use (c) with T = X or Y and S = f :
LX (Y f ) = (LX Y )f + Y LX f = [X, Y ]f + Y Xf = XY f = LX (Y f )5
(e)

Next apply it on a vector field Z:

[LX , LY ]Z = [X, [Y, Z]] [Y, [X, Z]]

(e)

Jacobi
identity

[[X, Y ], Z].

For higher rank tensors the derivation follows from the use of (c). If [X, Y ] = 0 then LX LY = LY LX
and for and which are the flows generated by X and Y one finds: s t = t s .

5 note

that (LX Y )f 6= LX (Y f ).

| {z }
[X,Y ]f

| {z }
XY f

41

15

15

DIFFERENTIAL FORMS

Differential forms

Definition: A p-form is a totally antisymmetric tensor field of type

0
p

(X(1) , . . . , X(p) ) = (sign )(X1 , . . . , Xp )

for any permutation of {1, . . . , p} ( Sp (group of permutations)) with sign being its parity. For


p > dim M, 0. Any tensor field of type p0 can be antisymmetrized by means of the operation A:
1 X
(sign )T (X(1) , . . . , X(p) )
p!

(AT )(X1 , . . . , Xp ) =

(15.1)

Sp

with A2 = A. The exterior product of a p1 -form 1 with a p2 -form 2 is the (p1 + p2 )-form:
1 2 =

(p1 + p2 )!
A(1 2 )
p1 ! p2 !

(15.2)

Properties:
1 2 = (1)p1 p2 2 1
1 (2 3 ) = (1 2 ) 3 =

(p1 + p2 + p3 )!
A(1 2 3 )
p1 ! p2 ! p3 !

The components in a local basis (e1 , . . . , en ) of 1-forms are

= i1 ...ip ei1 . . . eip = A
= i1 ...ip A(ei1 . . . eip )
=

n
X
i1 =1

n
X

i1 ...ip

ip =1

1 i1
e . . . e ip
p!

i1 ...ip ei1 . . . eip

(15.3)

1i1 <...<ip n

A covariant tensor of rank p, which is antisymmetric under exchange of any pair of indices (i.e. is a


n!
independent components.
p-form), in n dimensions has np = (np)!p!
Examples:
For 1-forms A, B (vector fields) we have
(A B)ik = Ai Bk Ak Bi = (1)(B A)ik .
For a 2-form A and a 1-form B
(A B)ikl = Aik Bl + Akl Bi + Ali Bk ,

42

(15.4)

15

DIFFERENTIAL FORMS

since
AB =

(1 + 2)!
A(A B)
1! 2!

3!
1
(Aik Bl ) ei ek el
1! 2!
3!

1
(Aik Bl )ei ek el
2

11
(Aik Bl + cyclic permutations)ei ek el
23

= (Aik Bl + cyclic permutations)

1 i
e ek el .
3!

Thus by comparing with (15.3) we get (15.4).

15.1

Exterior derivative of a differential form

The derivative df of a 0-form f F is the 1-form df (X) = Xf : the argument X (vector) acts as
f
a derivation. In a local coordinate basis: df =
dxi . The exterior derivative is performed by an
xi
operator d applied to forms, converting p-forms to (p + 1)-forms. The derivative d of a 1-form is
given by
d(X1 , X2 ) = X1 (X2 ) X2 (X1 ) ([X1 , X2 ]).

(15.5)

This expression is verified as follows:

k
X1 (X2 ) = X1 h, X2 i = X1i
k X2k = X1i k,i X2k + X1i k X2,i
,
i
| {z }
x
|{z}
1-form

,i

i
X2 (X1 ) = X2k i,k X1i + X2k i X1,k
,

i
i
([X1 , X2 ]) = h, X1 X2 X2 X1 i = i (X1 X2 X2 X1 )i = i X1k X2,k
X2k X1,k
,
then
d(X1 , X2 ) = (k,i i,k )X1i X2k .
This is manifestly a 2-form (the coefficient also fits the expectations:

1 (1+1)!
2! 1!1!

= 1). One can easily

verify that
d(f X1 , X2 ) = f d(X1 , X2 ).
For f = f (as f is a 0-form), the product rule
d( f ) = d f df
43

(15.6)

15

DIFFERENTIAL FORMS

applies, as one can verify

d( f )(X1 , X2 ) = X1 (f )(X2 ) X2 (f )(X1 ) (f )([X1 , X2 ]),
(15.5)

and
X1 (f )(X2 ) = X1k

(f k X2k ) = f X1i i k X2k + X1i i k X2k .

i
x
x
x
|
{z
} |
{z
}
f X1 (X2 )

df (X1 )(X2 )

So
d(f )(X1 , X2 ) = f d(X1 , X2 ) + (X2 )df (X1 ) (X1 )df (X2 ) .
|
{z
} |
{z
}

(15.7)

df

df

Moreover we have d f = 0, since

d2 f (X1 , X2 )

=
(15.5)

X1 df (X2 ) X2 df (X1 ) df ([X1 , X2 ])

X1 X2 f X2 X1 f [X1 , X2 ]f = 0.
(15.8)

The generalization of the definition to a p-form gives

d(X1 , .. , Xp+1 )

p+1
X

i , .. , Xp+1 )
(1)i1 Xi (X1 , .. , X

i=1

p+1
X
i , .. , X
j , .. , Xp+1 ),
(1)i+j ([Xi , Xj ], X1 , .. , X

(15.9)

i<j

2 , X3 ) = (X1 , X3 ).
wheremeans omission, e.g. (X1 , X
One can show that the following properties hold:
(a) d is a linear map from p-forms to p + 1-forms,
(b) d(1 2 ) = d1 2 + (1)p1 1 d2 ,
(c) d2 = 0, i.e. d(d) = 0,
(d) df (X) = Xf (f F),
By means of (a)-(d) we have an alternative definition of d. By eq. (15.3) we have with respect to a
coordinate basis
=

1
i ...i dxi1 . . . dxip , and hence
p! 1 p

1
di1 ...ip dxi1 . . . dxip
d |{z}
=
p!
ddxip =0

44

(15.10a)
(15.10b)

15

DIFFERENTIAL FORMS

Components:
p! d = i1 ...ip ,i0 dxi0 dxi1 . . . dxip
= i0 i2 ...ip ,i1 dxi0 . . . dxip
= (1)k i0 ...ik ...ip ,ik dxi0 . . . dxip

(k = 0, . . . , p)

d =

1 1 X
(1)k i0 ...ik ...ip ,ik dxi0 . . . dxip
p! p + 1
| {z } k=0
{z
}
|
1
(p+1)!

(15.11)

(d)i0 ...ip

Examples:
p = 1:
(d)ik = k,i i,k

(15.12)

(d)ikl = ik,l + kl,i + li,k

(15.13)

p = 2:

Tp (M); then
and : Tp (M)
Consider a map : M M
d = d .

(15.14)

A proof is found by using (15.10), (12.11) and property (b). It suffices to verify (15.14) on 0-forms
and 1-forms. For 0-forms f, (15.14) is identical to (12.10a). For 1-forms which are differentials df,
due to (c) we have
( d)(df) = 0

(d2 f = 0),

(d )(df) = d( df)

d(d(f )) = d2 (f ) = 0.

(12.10a)
( df)
=d(f)

Setting = t (the flow generated by X) and forming (14.1) (LX R =

d 
dt t R t=0 ),

one obtains the

infinitesimal version of (15.14):

LX d = d LX .

(15.15)

Definition: A p-form with

= d is exact
d = 0 is closed
An exact p-form is always closed (d2 = 0), but the converse is not generally true (Poincar lemma
gives conditions under which the converse is valid).6

is not unique since gauge transformations 7 + d, with any (p 2)-form, leave d unchanged.
is a generalization of the results of three-dimensional vector analysis: rot grad f = 0 and div rot ~k = 0.

7 This

45

15

DIFFERENTIAL FORMS

The integral of an n-form:


M is orientable within an atlas of positively oriented charts, if det

x
i
xj


> 0 for any change of

coordinates. For an n-form (n = dim M):

= i1 ...in

1 i1
dx . . . dxin = 1...n dx1 . . . dxn
| {z }
n!

(15.16)

(x)

is determined by the single component (x); under a change of coordinates (x) transforms like
 i
xi1
xin
x

(
x) =
1...n = i1 ...in

= (x) det
.
(15.17)
| {z } x
1
x
n
x
j
totally
antisymmetric

The integral of a n-form is defined as follows:

Z
Z
= dx1 . . . dxn (x1 , . . . , xn )
(if the support of is contained in a chart U).
M

ThisZ integral is independent

 i  since in different coordinates
Z of the choice of coordinates,

x 
and (15.17) applies. 8 9
dx1 . . . dxn (x) = d
x1 . . . d
xn (x) det
x
j 

15.2

Stokes theorem

Let D be a region in a n dimensional differentiable manifold M. The boundary D consists of those

p D whose image x in some chart satisfies e.g. x1 = 0. One can show that D is a closed (n 1)
dimension submanifold of M. If M is orientable then D is also orientable. D shall have a smooth
is compact. Then for every (n 1)-form we have
boundary and be such that D
Z
Z
d =

(15.18)
D

15.3

The inner product of a p-form

Definition: Let X be a vector field on M. For any p-form we define the inner product as
(iX )(X1 , . . . , Xp1 ) (X, X1 , . . . , Xp1 )

(15.19)

(and zero if p = 0).

Properties:
8 Actually,

it is often impossible to cover the whole manifold with a single set of coordinates. In the general case it is
necessary to introduce different sets of coordinates in different overlapping patches of the manifold, with the constraint
that in the overlap between the patch with coordinate xi and another patch with coordinate x
i , the xi can be expressed
in a smooth one-to-one way as functions of x
i and vice-versa (orientable manifold).
9 The integral over a p-form over the overlap between two patches (xi and x
i ) can be evaluated using either coordinate
 
system, provided det

xi
x
j

> 0.

46

15

DIFFERENTIAL FORMS

(a) iX is a linear map from p-forms to (p 1)-forms,

(b) iX (1 2 ) = (iX 1 ) 2 + (1)p1 1 (iX 2 ),
(c) iX 2 = 0,
(d) iX df = Xf = hdf, Xi with f F(M),
(e) LX = iX d + d iX .
Proof of (e): for 0-forms f we have
LX f = Xf,


iX df + 
d iX
f = iX df = Xf,
=0

and for 1-forms df

LX df

=
(15.15)
LX d=dLX

d(LX f ) = d(Xf ),

 + d iX df = d(Xf ).
iX 
ddf
=0

Application: Gauss theorem

Let X be a vector field. Then d(iX ) is an n-form with dim M = n. is an n-form, and if p 6= 0 p
M, then is a volume form. A function div X F is defined through
(div X) = d(iX ) = LX .10
We can apply Stokes theorem since d(iX ) is an n-form and thus iX an (n 1)-form:
Z
Z
Z
d(iX ) = (div X) =
iX .
D

The standard volume form is given by =

10 d

|g| dx1 . . . dxn .

= 0, thus LX = iX d + d iX applied on gives LX = iX 

d
+ d(iX ).


47

(15.20)

(15.21)

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

Expression for div X in local coordinates:

Let = a(x) dx1 . . . dxn , X = X i x

i . Then since (div X) = LX , we have (using property (c)

of the Lie derivative):

LX = (Xa) dx1 . . . dxn + a

n
X

dx1 . . . d(Xxi ) . . . dxn .

i=1

Since d(Xxi ) = d(X k

i
x ) = dX i (x) = X i ,j dxj , but dx1 . . . dxj . . . dxn 6= 0 only if j = i
k
x
| {z }
i
k

(otherwise we have two identical dxi ) we find

LX = |{z}
Xa dx1 . . . dxn + a
Xi

a
xi

n
X

X i ,i dx1 . . . dxn

i=1

1
= (X i a,i + aX i ,i )
a

1 p
1
|g|X i
div X = (aX i ),i = p
a
,i
|g|

16

for the standard .

(15.22)

Affine connections: Covariant derivative of a vector field

Definition: An affine (linear) connection or covariant differentiation on a manifold M is a mapping

which assigns to every pair X, Y of C vector fields on M another C vector field X Y with the
following properties:

(i) X Y is bilinear in X and Y ,

(ii) if f F(M), then
f X Y = f X Y,
X (f Y ) = f X Y + X(f )Y.
(16.1)

Lemma: Let X and Y be vector fields. If X vanishes at the point p on M, then X Y also vanishes at
p.

Proof: Let U be a coordinate neighbourhood of p. On U we have the representation X = i x

i,

i F(U) with i (p) = 0. Then (X Y )p = i

xi

Y = i (p)[ i Y ]p = 0.
| {z } x
=0

Since X Y produces again a vector field, the result of the covariant differentiation can only be a linear
combination of again the basis in the current chart. This leads us to the following statement:
48

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

Definition: One sets, relative to a chart (X 1 , . . . , X n ) for U M:



xi

xj

= kij

xk

(16.2)

The n3 functions kij F(U) are called Christoffel symbols (or connection coefficients) of the connection
in a given chart.11
The Christoffel symbols are not tensors:


k

c =
c x .
a
=

ab
ab
x

x
b
x
c
x
c xk

(16.3)

If we use (16.1):


x
a

x
b


=

xi
x
a xi

xj
x
b xj


 j

xi xj k

+
x
a x
b ij xk
xi x
b xj

xi xj k
2 xj

+
.
ij
x
a x
b
xk
x
a x
b xj

Comparison with 16.3:

xk c
2 xk
xi xj k
ij +
ab =
c
a
b
x

x
x

x
a x
b
c =

ab

xi xj x
c k
x
c 2 xk
ij +
a
b
k
x
x
x
xk x
a x
b

(16.4)

The second term is not compatible with being a tensor. If for every chart there exist n3 functions
kij which transform according to (16.4) under a change of coordinates, then one can show that there
exists a unique affine connection on M which satisfies (16.3).
Definition: for every vector field X we can introduce the tensor X T11 (M) defined by
X(Y, ) h, Y Xi ,

(16.5)

where is a one-form. X is called the covariant derivative of X.

In a chart (x1 , . . . , xn ), let X = i i and X = i ;j dxj i (< dxi , i >= ki ):


i ;j = X(j , dxi ) = dxi , j X = dxi , k ,j k + k jk = i ,j + ijk k

11 For

12

(16.6)

a pseudo-Riemannian manifold, the corresponding connection coefficients are given by (9.6) or (9.11).
shall denote the covariant derivative (normal derivative + additional terms, that vanish in (cartesian)
Euclidean or Minkowski space)
12 semicolon

49

16

16.1

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

Parallel transport along a curve

and let X be a vector field on some

Definition: let : I M be a curve in M with velocity field (t),
open neighbourhood of (I). X is said to be autoparallel along if

The vector X is sometimes denoted as

i

coordinates, we have X = i , =

dx
dt

X = 0.

(16.7)

X
dt

(covariant derivative along ). In terms of

DX
dt

or

i (see (12.3)). With (16.1) and (16.2) we get

X = dxi ( k k )
dt

dxi
i ( k k )
dt
i
dxi h k j
=
ik j + i k k
dt
=


dxi  j k
ij k + i k k
dt
 k

i
d
k dx j
=
+ ij
k ,
dt
dt
=

(16.8)

dxi k
d k
=
. This shows that X only depends on the values of X along . In
dt xi
dt
terms of coordinates we get for (16.7)
where we used

d k
dxi j
+ kij
= 0.
dt
dt

(16.9)

For a curve and any two point (s) and (t) consider the mapping
t,s : T(s) (M) T(t) (M),
which transforms a vector v(s) at (s) into the parallel transported vector v(t) at (t). The mapping
t,s is the parallel transport along from (s) to (t). We have s,s = 1 and r,s s,t = r,t .
We can now give a geometrical interpretation of the covariant derivative that will be generalized
to tensors. Let X be a vector field along , then

d 
X((t)) =
t,s X((s)),
ds s=t

(16.10)

Proof: Lets work in a given chart. By construction, v(t) = t,s v(s) with v(s) T(s) (M) and due to
(16.8) it satisifies: v i + ikj x k v j = 0. If we write (t,s v(s))i = (t,s )i j v j (s) = v i (t) (with t,s = (s,t )1

50

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

v(t)

t,s

v(s)

(t)
(s)

Figure 9: Illustration of parallel transport.

and s,s = 1), we get


d 
v (s) =
v i (t)
dt t=s

h
i
d 
i j
=
(
)
v
(s)
t,s
j
dt t=s

 
d 
i
(t,s ) j v j (s)
=
dt t=s
i

= ikj x k v j (s).

Since t,s

(t,s )i j = ikj x k
dt t=s


d 
d
= (s,t )1 , ds
( )i = dt
( )i = ikj x k . Then
s=t t,s j
t=s t,s j


i
 
d 
d 
d 
i
j
[t,s X((s))] =
t,s
X +
X i ((s))

ds s=t
ds s=t
ds
s=t
j
=

ikj x k X j

+X

which is again (16.8) (X = i i and the second term gives

51

,j


dxj ((s)) 
,

ds
s=t

d i
dt ).

(16.11)

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

Definition: If X Y = 0, then Y is said to be parallel transported with respect to X.

Geometrical interpretation of parallel transport: Consider the differential dAi = Ai ,j dxj = Ai (x +
dx) Ai (x). In order that the difference of two vectors be a vector, we have to consider them at the
same position. The transport has to be chosen such that for cartesian coordinates there is no change
in transporting it. The covariant derivative exactly achieves this.
Definition: Let X be a vector field such that X X = 0. Then the integral curves of X are called
geodesics.
In local coordinates xi the curve is given by (using (12.3) and (13.6)) the requirement
i

X (x(t)). Inserting this into (16.8) and using

d x
dt2

dX
dt

d i
dt x (t)

, we get

x
k + kij x i x j = 0.

(16.12)

For a vector parallel transported along a geodesic, its length and angle with the geodesic does not
change.

16.2

Round trips by parallel transport

Consider (16.8) and denote i = v i , thus

v i = ikj x k v j .

(16.13)

Let : [0, 1] M be a closed path, wih (0) = p = (1). Displace a vector v0 Tp (M) parallel along
and obtain the field v(t) = t,0 v0 T(t) (M). We assume that the closed path is sufficiently small
(such that we can work in the image of some chart), thus we can expand ikj (x) around the point
x(0) = x0 on the curve:
ikj (x)

'

ikj (x0 )

x0 )

ikj

+ (x

(x)
+
kj

x
x=x0

(16.14)

Thus (16.13) is to first order in (xk xk0 ):

Zt

v dt = v (t)
0

v0i

Zt
=
0

v x dt
|{z}

ikj (x0 )v0j

'v0j

Zt

x k dt,

| {z }
xk (t)xk
0

taking only the first term in the expansion of . And hence,

v i (t) = v0i ikj (x0 )(xk (t) xk0 )v0j +

52

(16.15)

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

2
W
C

Figure 10: Illustration of the path dependence of parallel transport on a curved space: vector 1 at N
can be parallel transported along the geodesic N-S to C, giving rise to vector 2. Alternatively, it can
be first transported along the geodesic N-S to E (vector 3) and then along E-W to C to give the vector
4. Clearly these two are different. The angle between them reflects the curvature of the two-sphere.

By plugging (16.14) and (16.15) into (16.13), we obtain an equation valid to second order:
Z1

Z1

v dt =
0

v (1)

ikj x k v j dt

(16.16)

v0i

Z1 
'

ikj (x0 )

+ (x

x0 )

i
(x0 ) +
x kj

v0j jkj (x0 )(xk (t) xk0 )v0j + x k dt.
53

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

Multiplying out and discarding terms of third order or higher in xk xk0 , we get:

v (1) '

v0i

 Z1
i

j
i
x dt
(x0 ) kj (x0 )j (x0 ) v0j (x x0 )x k dt.
x kj
0
0
| {z }
Z1

ikj (x0 )v0j

xk (1)xk (0)
=0

R1
Since we are considering a closed path ( 0 x dt = xk (1) xk (0) = 0),

v i = v i (1) v i (0) =

 Z1
i
i
l

(x
)

(x
)
(x
)
v0j x x k dt,
0
0
kl 0
j
x kj
0

with
I1
0

I1
I1
d k
k
(x x ) dt x x dt = x xk dt,
x x dt =
dt
0
0
0
|
{z
}
I1

=0

antisymmetric in (, k). Then



Z1
i
1 i
j
i
l
i
l
kl j
+ l kj (x0 ) v0 x x k dt,
v =
2 x kj
xk j
{z
} 0
|
i

i
Rjk

1 i
v = Rjk
(x0 )v0j
2
i

Z1

x x k dt.

(16.17)

We shall see that

i
Rjk

is the curvature tensor.

i
Rjk
=

i
i
j
+ lj ikl lkj il
k
x
x kj

(16.18)

Thus an arbitrary vector v i will not change when parallel transported around an arbitrary small closed
i
curve at x0 if and only if Rjk
vanishes at x0 .

16.3

Covariant derivatives of tensor fields

The parallel transport is extended to tensors by means of the requirements:

s,t (T S) = (s,t T ) (s,t S),
s,t tr(T ) = tr(s,t T ),
s,t c = c
54

(c R).

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

For e.g. a covariant vector , hs,t , s,t Xi(s) = h, Xi(t) and for a tensor of type

1
1

: s,t T (s,t , s,t X) =

T (, X). In components:
(s,t T )

i
k

= T (s,t ) (s,t ) k

(16.19)

(i k is inverse transpose of i k ). The covariant derivative X (X vector field, T tensor field) associated
to is
(X T )p =



d
,
0,t T(t) 
dt
t=0

(16.20)

with (t) any curve with (0) = p and (0)

= Xp (generalization of (16.10)).
Properties of the covariant derivative:
(a) X is a linear map from tensor fields to tensor fields of the same type

r
r

(b) X f = Xf ,
(c) X (tr T ) = tr(X T ),
(d) X (T S) = (X T ) S + T (X S).
This follows from the properties of s,t . For a 1-form we have:
(X )(Y ) = tr(X Y )
= tr(X ( Y )) tr( X Y )
= X tr( Y ) (X Y )
= X(Y ) (X Y ).
General differentiation rule for a tensor field of type

1
r

(16.21)

(X T )(, Y ) = XT (, Y ) T (X , Y ) T (, X Y )

(16.22)

Due to (a)-(d), the operation X is completely determined by its action on vector fields Y , which are
the affine connections (see (16.1) and (16.2)).

16.4

Local coordinate expressions for covariant derivative

Let T Tpq (U) be a tensor of rank (p, q) with local coordinates (x1 , . . . , xn ) valid in a region U. We
have T i1 ...ip j1 ...jq i1 . . . ip dxj1 . . . dxjq and X = X k k . Let us use
XT i1 ...ip j1 ...jq = X k T i1 ...ip j1 ...jq ,k

(16.23)

X (i ) = X k k i = X k lki l .

(16.24)

and write (16.2):

55

16

AFFINE CONNECTIONS: COVARIANT DERIVATIVE OF A VECTOR FIELD

Moreover,
(X dxj )(i )

(16.21)


X dxj , i dxj , X i
| {z }
j i

{z

or

X dxj

X k jki ,

X k jki dxi .

(16.25)

Using (16.23), (16.24) and (16.25) for j = dxj , Yi = i we obtain the following expression for X T :

T i1 ...ip j1 ...jq ;k = T i1 ...ip j1 ...jq ,k + ikl1 T li2 ...ip j1 ...jq + . . . + klp T i1 ...ip1 l j1 ...jq
lkj1 T i1 ...ip lj2 ...jq . . . lkjq T i1 ...ip j1 ...jq1 l .

(16.26)

Examples:
Contravariant and covariant vector fields:
i ;k = i ,k + ikl l ,
i;k = i,k lki l ,
Kronecker tensor:
i j;k = 0,
Tensor

1
1

:
T i k;r = T i k,r + irl T l k lrk T i l .

The covariant derivative of a tensor is again a tensor. Consider the covariant derivative of the metric
g :
g; =
Inserting into this the expressions of

g
g g .
x
given by (9.11) leads us to
g; = 0.

(16.27)

(16.28)

This is not surprising since g; vanishes in locally inertial coordinates and being a tensor it is then
zero in all systems.
Covariance principle: Write the appropriate special relativistic equations that hold in the absence of
gravitation, replace by g , and replace all derivatives with covariant derivatives (, ;). The resulting
equations will be generally covariant and true in the presence of gravitational fields.

56

17

17

CURVATURE AND TORSION OF AN AFFINE CONNECTION, BIANCHI IDENTITIES

Curvature and torsion of an affine connection, Bianchi identities

Let an affine connection be given on M, let X, Y , Z be vector fields.

Definition:
T (X, Y ) = X Y Y X [X, Y ]

(17.1)

R(X, Y ) = X Y Y X [X,Y ]

(17.2)

T (X, Y ) is antisymmetric and f -linear in X, Y and then defines a tensor of type

(, X, Y ) h, T (X, Y )i is thus a 12 tensor field called the torsion tensor.

1
2

through:

f -linearity:
T (f X, gY ) = f gT (X, Y )

f, g, F(M).

In local coordinates, the components of the torsion tensor are given by:
*
+

k

k
k
T ij = dx , T (i , j ) = dx , i j j i [i , j ]
| {z }
| {z }
=lij l

=0

= kij kji

(17.3)


(using that dxk , l = k l ). In particular, we have T k ij = 0 kij = kji .
R(X, Y ) = R(Y, X) is antisymmetric in X,Y . The vector field R(X, Y )Z is f -linear in X, Y , Z:


(R(f X, gY ) hZ = f gh R(X, Y )Z; f, g, h F(M)). R determines a tensor of type 13 : the Riemann
tensor or curvature tensor.
(, Z, X, Y ) h, R(X, Y )Zi Ri jkl i Z j X k Y l
In components with respect to local coordinates:


Ri jkl = dxi , R(k , l )j = dxi , (k l l k )j 13


= dxi , k (slj s ) l (skj s )
= ilj,k ikj,l + slj iks skj ils .
13 Notice

that [ , ] j = 0.
k l

| {z }
=0

57

(17.4)

17

CURVATURE AND TORSION OF AN AFFINE CONNECTION, BIANCHI IDENTITIES

Eq. (17.4) is exactly the the same as defined in (16.18). It is antysymmetric in the last two indices:
Ri jkl = Ri jlk .
Definition: The Ricci tensor is the following contraction of the curvature tensor:
Rjl Ri jil = ilj,i iij,l + slj iis sij ils

(17.5)

The scalar curvature is the trace of the Ricci tensor:

R g lj Rjl = Rl l

(17.6)

Example: For a pseudo-Riemannian manifold the connection coefficients are given by (9.11). Consider
a two-sphere (which is a pseudo Riemannian manifold) with the metric ds2 = a2 (d2 + sin2 d2 ),
then:
2

g = a

sin2

1
= 2
a

1
sin2

The non-zero are:

= sin cos ,
= = cot .
The Riemann tensor is given by

R
= +

= (sin2 cos2 ) 0 + 0 ( sin cos ) cot

= sin2 .
The Ricci tensor has the following components:

R = R
+ R
= sin2 ,
| {z }
=0

R = 1,
R = R = 0.
The Ricci scalar is
R = g R + g R +g R +g R
|{z} |{z}
|{z} |{z}
|{z}
|{z}
1
a2

1
a2 sin2

sin2

1
1
+ 2 2 sin2
2
a
a sin

2
.
a2
58

17

CURVATURE AND TORSION OF AN AFFINE CONNECTION, BIANCHI IDENTITIES

The Ricci scalar is constant over this two-sphere and positive, thus the the sphere is positively curved.
14 15 16

17.1

Bianchi identities for the special case of zero torsion

X, Y and Z are vector fields, then

R(X, Y )Z + cyclic = 0

(1st Bianchi identity),

(17.7)

(X R)(Y, Z) + cyclic = 0

(2nd Bianchi identity).

(17.8)

Proof of the 1st identity: Torsion = 0 X Y Y X = [X, Y ]. Then

(X Y Y X )Z + (Z X X Z )Y
+ (Y Z Z Y )X [X,Y ] Z [Z,X] Y [Y,Z] X
= X (Y Z Z Y ) [Y,Z] X + cyclic
= [X, [Y, Z]] + cyclic
=0

due to the Jacobi identity (13.2).

(See textbooks for proof of the 2nd Bianchi identity.)

14 For a position independent metric (e.g. Cartesian coordinates) the Riemann tensor (and thus the scalar curvature)
vanishes as the vanish.


15 For

a plane with polar coordinates we get a position dependent metric

do not vanish. However the curvature vanishes.

16 The curvature does not depend on the choice of coordinates.

59

1
0

0
: ds2 = dr2 + r2 d2 and thus the
r2

18

18

RIEMANNIAN CONNECTIONS

Riemannian connections

Metric: Let M be equipped with a pseudo-Riemannian metric: a symmetric, non-degenerate tensor

field: g(X, Y ) = hX, Y i of type 02 .
Non-degenerate means that, for any point p M, X, Y Tp , one has gp (X, Y ) = 0 Y Tp
X = 0.
In components, hX, Y i = gik X i Y k , gik = gki (symmetric) and det gik 6= 0.
With the metric we can lower and raise indices:
i = gik X k ,
X

i = g ik k ,

where g ik denotes the inverse of gik . It also works for tensor fields of different types: T i k =
Tlk g il = T il glk .
Given a basis (e1 , . . . , en ) of Tp , the covectors of the dual basis (e1 , . . . , en ) become themselves
vectors; indeed ei = gij ej .


Riemann connection: The metric tensor g at a point p in M is a symmetric 02 tensor. It assigns a
p
magnitude |g(X, X)| to each vector X on Tp (M), denoted by d(X) and defines the angle between
any two vectors X, Y (6= 0) on Tp (M) via


g(X, Y )
.
(18.1)
a(X, Y ) = arccos
d(X)d(Y )
If a(X, Y ) =

then X and Y are orthogonal. Further observations:

The length of a curve with tangent vector X between t1 and t2 is L(t1 , t2 ) =

Rt2

d(X) dt.

t1

If (ea ) is a basis of Tp (M), the components of g with respect to this basis are gab = g(ea , eb ).
Like in special relativity we classify vectors at a point as timelike (g(X, X) > 0), null (g(X, X) =
0) and space like (g(X, X) < 0).

Definition: let (M, g) be a pseudo-Riemannian manifold. An affine connection is a metric connection

if parallel transport along any smooth curve on M preserve the inner product: for autoparallel fields
X(t), Y (t) (see (16.7)), g(t) (X(t), Y (t)) is independent of t along .
Theorem: an affine connection is metric if and only if (no proof)
g = 0.
Eq. (18.2) is equivalent for (g(Y, Z)) to
X g = 0 = Xg(Y, Z) g(X Y, Z) g(Y, X Z)
60

(18.2)

18

RIEMANNIAN CONNECTIONS

or
Xg(Y, Z) = g(X Y, Z) + g(Y, X Z).

(18.3)

Theorem: For every pseudo-Riemannian manifold (M, g), there exists a unique affine connection such
that
(a) has vanishing torsion ( is symmetric),
(b) is metric.
Proof: T = 0 (vanishing torsion) means X Y = Y X + [X, Y ]. Inserting this into (18.3) (and the
linearity of g) gives
Xg(Y, Z) = g(Y X, Z) + g([X, Y ], Z) + g(Y, X Z).

(18.4)

By cyclic permutations one obtains as well

Y g(Z, X) = g(Z Y, X) + g([Y, Z], X) + g(Z, Y X),

(18.5)

Zg(X, Y ) = g(X Z, Y ) + g([Z, X], Y ) + g(X, Z Y ).

(18.6)

Taking the linear combination (18.5) + (18.6) - (18.4), we get (Koszul formula):

2g(Z Y, X)

Xg(Y, Z) + Y g(Z, X) + Zg(X, Y )

g([Z, X], Y ) g([Y, Z], X) + g([X, Y ], Z).
(18.7)

The right hand side is independent of . Since g is non-degenerate, the uniqueness of follows from
(18.7).
Definition: the unique connection on (M, g) from the above theorem is called the Riemannian or
Levi-Civita connection.
We determine the Christoffel symbols for the Riemannian connection in a given chart (U, x1 , . . . , xn ).
For this purpose we take X = k , Y = j , Z = i in (18.7) and we use [i , j ] = 0 as well as
hi , j i = gij . The result is
hi j , k i = lij hl , k i,
| {z }
glk

2lij glk = k hi , j i +j hi , k i +i hk , j i,
| {z }
| {z }
| {z }
gik

gij

61

gkj

18

RIEMANNIAN CONNECTIONS

or
glk lij =

1
(gkj,i + gik,j gji,k ).
2

(18.8)

g ij denoting the inverse matrix of gij , we obtain

lij =

1 lk
g (gkj,i + gik,j gji,k ),
2

(18.9)

which is exactly equation (9.11).

Properties of the Riemannian connection:
(i) The inner product of any two vectors remains constant upon parallel transporting them along
any curve (g(X, Y )(t) = g(X, Y )(0) ).
i
(ii) The covariant derivative commutes with raising or lowering indices, e.g. Tk;l
= (gkm T im );l

= gkm T im ;l , because gkm;l = 0.

Riemann tensor: the curvature tensor of a Riemannian connection has the following additional symmetry properties (without proof):
hR(X, Y )Z, U i = hR(X, Y )U, Zi ,

(18.10)

hR(X, Y )Z, U i = hR(Z, U )X, Y i .

(18.11)

In coordinate expression the Riemann tensor satisfies the following symmetries:

Ri jkl = Ri jlk

is always the case,

(18.12)

Ri jkl = 0

1st Bianchi identity,

(18.13)

Ri jkl;m = 0

2nd Bianchi identity.

(18.14)

X
(jkl)

X
(klm)

Eqs. (18.13) and (18.14) are valid for vanishing torsion. Here

denotes the cyclic sum. Additionally,

(jkl)

Rijkl = Rjikl ,

(18.15)

Rijkl = Rklij ,

(18.16)

for the Riemannian connection with Rijkl = gis Rs jkl .

Ricci and Einstein tensor

62

18

RIEMANNIAN CONNECTIONS

Rik = Rj ijk
R = Ri i

Ricci tensor

(18.17)

scalar curvature

(18.18)

Einstein tensor

(18.19)

1
Gik = Rik Rgik
2

By symmetry, Rik = Rki , Gik = Gki and

Rk i;k =

1
R;i ,
2

(18.20)

Gk i;k = 0,

(18.21)

which are the contracted 2nd Bianchi identity.

Proof: Rik = g jl Rlijk = g jl Rjkli , 2nd Bianchi identity gives:
Ri jkl;m + Ri jlm;k + Ri jmk;l = 0.
Then we take the (ik)-trace:
Rjl;m + Ri jlm;i Rjm;l = 0,
| {z }
g ik Rjklm;i

Rj l;m g ik Rj klm;i Rj m;l = 0,

(jm)-trace:
Rj l;j + g ik Rkl;i R;l = 0.
|
{z
}
2Rj l;j

(18.20)
For (18.21):
1
1
Gk i = Rk i g k i R = Rk i k i R
2
2
1
1
1
Gk i;k = Rk i;k (ik R);k = R;i R;i = 0
2
|2{z } 2
(18.20)

Without proof in n dimensions, the Riemann tensor has cn =

(c1 = 0, c2 = 1, c3 = 6, c4 = 20).

63

n2 (n2 1)
12

independent components

19

PHYSICAL LAWS WITH GRAVITATION

Part V

General Relativity
19

Physical laws with gravitation

19.1

Mechanics

The physical laws are relations among tensors (scalars and vectors being tensors of rank 0 and 1
respectively). Thus the physical laws read the same in all coordinate systems (provided the physical
quantities are transformed suitably) and satisfy general covariance (same form). Practically, this means
that from the special relativity laws that hold in absence of gravitation, we have to replace by g
and replace derivation by covariant derivation.
In an inertial system, we have the equation of motion (see (4.3))
m

du
= f .
d

(19.1)

According to the equivalence principle, (19.1) holds in a local IS. f does not contain gravitational
forces as they would vanish in a local IS. We transform it to general KS (coordinate system), then the
Lorentz vector f gets transformed to f =
holds in a local IS reads then

x
f

Du
| d
{z }

( is in local IS, x is in KS). Equation (19.1)

= f ,

(19.2)

covariant derivative
given in (16.8)

(X

D
d )

with u (dt d and

dxi
d

= ui ) and thus

du
Du
=
+ u u .
d
d
Then equation (19.2) reads

du
= f m u u .
(19.3)
d
We see that on the right hand side there are now gravitational forces appearing explicitly (via ).
m

Equation (19.3) (or (19.2)) is covariant (it has the same form in all coordinate systems) and reduces for
g (thus = 0) to equation (19.1) (in a local IS). The components of u are not independent
but satisfy the condition g u u = c2 .

19.2

Electrodynamics

According to the equivalence principle, Maxwells equations (see (6.4) and (6.5))
F =

4
j
c

and

 F = 0

are valid in a local IS. Applying the covariance principle, they become as follows in a general KS:
F ; =

4
j
c

and
64

 F; = 0,

(19.4)

19

PHYSICAL LAWS WITH GRAVITATION

provided that going from coordinates in a local IS to the KS coordinates x we have

j j =

x
j

F =

and

x x
F .

Gravity enters via the in the covariant derivative. The continuity equation j = 0 translates
to j ; = 0. It can be shown that in the homogeneous equation the terms with vanish. Thus the
covariant derivative reduces to the ordinary derivative (,)

19.3

17

Energy-momentum tensor

For an ideal fluid, given by (in a local IS)


p
T = + 2 u u p,
c

(19.8)

with
u : four-velocity,
: proper energy density,
p: pressure of the fluid.
In a KS this becomes


p
(19.9)
T = + 2 u u g p.
c
In the IS the conservation law implies T , = 0 and in the KS T ; = 0 (explicitely, T ; =
T , + T + T = 0). With (19.5), =

g
1
g x ,

we get instead

1 gT
T ; =
+ T = 0.
g x

(19.10)

This is no longer a conservation law, as we cannot form any constant of motion from (19.10). This
should also not be expected, since the system under consideration can exchange energy and momentum
with the gravitational field.
17 g

= det(gik ) = i1 ...in g1i1 . . . gnin . Consider

gkm mr
g grik .
xl

g
xl

n
P

i1 ...in g1i1 . . .

k=1

gki

xl

. . . gnin and use

gki

xl

gkm m
ik
xl

Due to the antisymmetry of , only the term r = k survives. Thus

g
gkm mk
=
g g.
xl
xl

Plugging this into the definition of kkl (one contraction):

kkl =

g km
2

gmk
+
xl

gml
gkl

xk
xm

{z

ln g
g km gmk
1 g
=
=
.
l
l
2 x
x
g xl

(19.5)

vanish by interchanging
(mk)

With (19.5) one can show that the inhomogeneous Maxwell equation in KS can be written as

1 ( gF )
4
= 2 j,

g
x
c
and the continuity equation: j ; = 0 becomes

( gj )
x

65

= 0.

(19.6)

(19.7)

20

20

EINSTEINS FIELD EQUATIONS

Einsteins field equations

The field equations cannot be derived by using the covariance principle, since there is no equivalent
equation in a local IS. We have to make some requirements/assumptions.
Requirements:
The Newtonian limit is well confirmed through all observations: = 4G.
From the Newtonian limit of the equation of motion for a particle we derived (equation (9.17))
g00 1 + 2 c2 .
The non-relativistic limit should then be
g00 =

8G
T00 ,
c4

(20.1)

with T00 c2 (other Tij are small).

Thus a generalization should lead to something of type G =

8G
c4 T

where G has to satisfy the

following requirements:
(1) G is a tensor (T is tensor).
(2) G has the dimension of a second derivative. If we assume that no new dimensional constant
enter in G then it has to be a linear combination of terms which are either second derivatives of
the metric g or quadratic in the first derivative of g .
(3) Since T is symmetric, G also has to be symmetric and due to the fact that T is covariantly
conserved, i.e. T ; = 0, it follows that G must satisfy G = G and G ; = 0.
(4) For a weak stationary field we shall get (20.1), thus G00 ' g00 .
Conditions (1)-(4) determine G uniquely. (1) and (2) imply that G has to be a linear combination
G = aR + bRg

(20.2)

of R , the Ricci tensor, and R, the Ricci scalar18 . The symmetry of G is automatically satisfied.
The contracted Bianchi identity (18.20), (18.21) suggests that G ; = 0 on the Einstein tensor, what
implies b = a2 . Thus we find
1
8G
G = a(R g R) = 4 T .
(20.3)
2
c
The constant a has to be determined by performing the Newtonian limit. Consider weak fields:
g = + h , |h |  1 (non relativistic velocities: v i  c), then |Tik |  |T00 | |Gik |  |G00 |.
Compute the trace of G :

= a(R 2R) = aR




g G G00 = a R00 R g00 = a(R00 R/2)

2 |{z}

00

from (20.3)
.

(20.4)

=1

18 It

can be shown that indeed the Ricci tensor is the only tensor made of the metric tensor and first and second
deivatives of it, and which is linear in the second derivative.

66

20

EINSTEINS FIELD EQUATIONS

Compairing the two results gives R 2R00 , thus



R
G00 ' a R00
' 2aR00 .
2

(20.5)

For weak fields all terms quadratic in h can be neglected in the Riemann tensor; we get to leading
order:

R = R '

(|h |  1).
x
x
For weak stationary fields we find:
R00 =
Thus G00 2a

i00
xi

i00
xi

with

i00 =

1 g00
.
2 xi

= ag00 = g00 , therefore a = 1. Einsteins field equations are

19

(found 1915 by

Albert Einstein):
R

R
8G
g = 4 T
2
c

(20.6)

Together with the geodesic equation ((16.12) or (19.3)), these are the fundamental equations of general
relativity. By contraction of (20.6), we find also
R

8G
R
= R = 4 T.
2 |{z}
c

(20.7)

=4

R can be expressed in (20.6) in terms of T , and we get:

R

8G
= 4
c


T

T
g
2


(20.8)

an equivalent version of the field equations. For the vacuum case where T = 0 we have
R = 0.

(20.9)

Significance of the Bianchi identity

Einsteins equation constitutes a set of non-linear coupled partial differential equations whose general
solution is not known. Usually one makes some assumptions, for instance spherical symmetry. Because
the Ricci tensor is symmetric, the Einstein equations constitute a set of 10 algebraically independent
second order differential equations for g .
The Einstein equations are generally covariant, so that they can at best determine the metric
up to coordinate transformation ( 4 functions). Therefore we expect only 6 independent generally
covariant equations for the metric. Indeed the (contracted) Bianchi identities tell us that (equation
(18.21)) G ; = 0 and hence there are 4 differential relations among the Einsteins equations. Bianchi
identities can also be understood as a consequence of the general covariance of the Einstein equations.
19 Depending

on the convention used for the Riemann tensor, one could also encounter a minus in front of the energymomentum tensor, as for example in Weinberg.

67

21

20.1

THE EINSTEIN-HILBERT ACTION

The cosmological constant

As a generalization, one can relax condition (2) and have a linear term in g

20

. The field equations

become

R
8G
(20.10)
g g = 4 T ,
2
c
where is a constant: the cosmological constant ([] = L2 ). For point (4) the Newtonian limit of
R

(20.10) leads to
= 4G c2 .

(20.11)

The right-hand side can also be written as 4G( vacuum ), with

c2 vac =

c4
.
4G

(20.12)

corresponds to the (constant) energy density of empty space (vacuum). 1/2 (distance) has to be
much larger than the dimension of the solar system.

21

The Einstein-Hilbert action

The field equations (20.6) can be obtained from a covariant variational principle. The action for the
metric g is
Z
R(g) dv,

SD [g] =

(21.1)

where D M is a compact region space-time, R is a scalar curvature and dv a volume element:

p
(21.2)
dv = |g|d4 x
(g = det gik , d4 x in 4 dimensions). The Euler-Lagrange equations are the field equations in vacuum:
SD [g] = 0.
We have
Z
Z
Z
Z

4

4

R(g) dv = (g R g) d x = (R )g
g d x + R (g g) d4 x.
D

(21.3)

Consider first R :

R =
+ .

(21.4)

Let us compute the variation of R at any point p in normal coordinates, whose center is in p itself
(x(p) = 0, then
(0) = 0). Thus R reduces (at any such p) to

21
R =
, , .

(21.5)

Without proof one finds that

is indeed a tensor although is not a tensor. (21.5) is thus a

tensor equation, it holds in every coordinate system and we can also take the covariant derivative:

R =
(21.6)
; ;
20 Note

that g; = 0.
(with respect to g) and normal derivative commute

21 variation

68

21

THE EINSTEIN-HILBERT ACTION

(Palatini identity). Since g; = 0 we can write (21.6) as

g R = g
; g ;

(21.7)

= ;

|{z}

= , +

g
x

1 ( g )
=
.
g
x

(21.8)

Inserting this into the integral (21.3) and applying (15.21) (Gauss theorem), we get
Z
Z
(divg ) =
i ,
D

where divg =

and thus
Z
(R )g

g d x =

g d0 .

d0 is the coordinate normal to D and

= g

(21.9)

is a vector field. If the variations of g vanish outside a region contained in D, then the boundary
term vanishes as well.

As for the second term in (21.3) ( R (g g) d4 x), we recall that for an n n matrix A()
D

we have (see linear algebra):

i)
ii)

d
d

det A = det A tr A1 dA
d ,

d
1
)A
d (A

= A1 dA
d .

Thus (g )g = g g comes from ii) and g = gg g comes from i) with A1 = g . Hence

we find the desired expressions

1
1
g g g =
g g g ,
g =
2
2

1
(g g) = g g
g g g g .
2

(21.10)

And thus

Z
0=

R (g g) d4 x =

Z
D

g d4 x R g R g g g
| {z }
2 | {z } | {z }
dv

dv R
|


1
Rg g .
2
{z
}

=G =0

69

21

THE EINSTEIN-HILBERT ACTION

R
R
Therefore SD [g ] = 0 G = R 12 Rg = 0. Since
g d4 x =
g d4 x 12 g g =
D
D
R
g d4 x g g , it follows that if we have a cosmological constant , the Einsteins vacuum
21
D

equations are obtained from the action principle applied on

Z

SD [g] = (R 2) g d4 x.

(21.11)

The variational principle extends to matter described by any field = (A ) (A = 1, . . . , N ), (we include
also the electromagnetic field among the A ) transforming as a tensor under change of coordinates.
Consider an action of the form
Z
SD [] =

L(, g ) g d4 x,

(21.12)

where g is the Riemannian connection of the metric g. If we know L in flat space, the equivalence
principle prescribes to replace by g and replace ordinary derivatives by covariant ones.
Example: electromagnetic field
L=

1
1
F F =
F F g g ,
16
16

and the Euler-Lagrange equations in this case (F = A; A; = A, A, ) for the basic

4-potential A field read:
L
L

= 0,
A
A
in this case

L
A

= 0, and

L
A

with A = A; ;

1
= 4
F . The Euler-Lagrange equations are then F ; = 0, which

are the Maxwell equations for vanishing current j (F ; =

4
c j

1
and L = 16
F F 1c j A

with j A = g j A ).
Variations in (21.12) with respect to the fields A lead to the Euler-Lagrange equations, whereas
variations with respect to the metric (which is also a function and is determined by solving Einsteins
equations) gives (without proof)
Z
Z

1
4
T g g d4 x.
g L(, g ) g d x =
2

(21.13)

This term has to be added to the one proportional to g in Einsteins action:



Z

c4
1
g d4 x G
T g
16G 2
|
{z
}
D
=0

and thus G =

8G
c4 T .

For electrodynamics: T =

Lg ). And similarly for other matter fields.

70

1
4

F F 14 F F g (or T = F F

22

22

STATIC ISOTROPIC METRIC

Static isotropic metric

22.1

Form of the metric

For the gravity field of Earth and Sun we assume a spherically symmetric distribution of the matter
(rotation velocities v i  c). Thus we need a spherically symmetric and static solution for the metric
g (x). We first give the general form of such a metric (static and isotropic) which we then use as an
ansatz to solve the field equations. For r , the Newtonian gravitational potential = GM
r goes
to zero. Thus, asymptotically, the metric should be Minkowskian: ds2

= c2 dt2 dr2 r2 (d2 +

sin2 d2 ), in spherical coordinates r, , and t. Thus,

ds2 = B(r)c2 dt2 A(r)dr2 C(r)r2 (d2 + sin2 d2 ).

(22.1)

Due to isotropy and time independence, A, B and C cannot depend on , and t (and no linear terms
in d and d). Freedom in the choice of coordinates allows to introduce a new radial coordinate in
(22.1): C(r)r2 r2 , thus C(r) can be absorbed into r. We get the standard form:
ds2 = B(r)c2 dt2 A(r)dr2 r2 (d2 + sin2 d2 )

(22.2)

( and have the same significance as in Minkowski coordinates). Due to our asymptotic requirements
(r ) we can assume that B(r) 1 and A(r) 1.

22.2

Robertson expansion

Even without knowing the solution to the field equations, we can give an expansion of the metric
for weak fields outside the mass distribution. The metric can only depend on the total mass of the
considered object (Earth or Sun for instance), on the distance from it and on the constants G, c. Since
A and B are dimensionless, they can only depend on a combination of the dimensionless quantity
For

GM
c2 r

GM
c2 r .

 1 we can then have the following expansion:

B(r) = 1 2

GM
+ 2( )
c2 r

GM
c2 r

2
+ ...
(22.3)

GM
A(r) = 1 + 2 2 + . . .
c r
which is the Robertson expansion. The linear term in B(r) has no free parameter since it is constrained
by the Newtonian limit: g00 ' 1 + 2 c2 , = GM
(Newtonian potential), therefore B g00 . The
r
coefficient 2( ) comes from historical reasons, and are independent coefficients. In the solar
system,

GM
c2 r

GM
c 2 R

' 2 106 , then only linear terms in and play a role. For general relativity:

= = 1 (Newtonian gravity: = = 0).

71

22

22.3

STATIC ISOTROPIC METRIC

Christoffel symbols and Ricci tensor for the standard form

The metric tensor g is diagonal.

g 00 =

g22 = r2

g11 = A(r)

g00 = B(r)
1
B(r)

g 11 =

1
A(r)

The non-vanishing components of =

g 22 =

g
2

g
g
g
+

x
x
x

001 = 010 =

B0
2B

100 =

212 = 221 =

1
r

122 =

313 = 331 =

1
r

where 0 stands for

r .

1
r2

B0
2A
r
A

323 = 332 = cot

g33 = r2 sin2

(22.4)

1
sin2

(22.5)

r sin2
A

(22.6)

g 33 =

r2


are
111 =

A0
2A

133 =

233 = sin cos

With
g = r4 AB sin2

we get


=


 
A0
B0
ln g
2
+
, cot , 0 .
= 0, +
x
r
2A 2B

(22.7)

(22.8)

The Ricci tensor can then be calculated as

R =

+ ,
x
x

(22.9)

and we get as a result



B 00
A0 B 0
B 02
B0 2
A0
B0

+
+
+
,
2A
2A2
2AB
2A r
2A 2B


B 00
B 0 A0
B0
B0
=

+
,
+
2A 4A A
B
rA


B 00
B 0 A0
B0
A0
+
+
,
=
+
2B
4B A
B
rA
 0

r
A
B0
1
=1+

,
2A A
B
A

R00 =

R11

R22

R33 = R22 sin2 ,

(22.10)

(22.11)

(22.12)
(22.13)

The non-diagonal components R with 6= vanish.

72

22

22.4

STATIC ISOTROPIC METRIC

Schwarzschild metric

We assume a static, spherically symmetric, mass distribution with finite extension:

6= 0 r r
0
(r)
= 0 r > r

(22.14)

Similarly, the pressure P (r) is thought to vanish for r > r0 . The four velocity vector within the mass
distribution in the static case is u = (u0 = constant, 0, 0, 0). This way, the energy-momentum tensor
(describing matter) does not depend on time. We then adopt the ansatz for the metric elaborated
in (22.2): g = diag(B(r), A(r), r2 , r2 sin2 ). Outside the mass distribution (r r0 ), the Ricci
tensor vanishes: R = 0. We have already calculated the coefficients R in equations (22.10)
(22.13). For 6= , R = 0 is trivially satisfied while the diagonal components should be set to zero:
R00 = R11 = R22 = R33 = 0 (r r0 ).
 0

R00 R11
1
B
A0
d
+
=
+
(ln AB) = 0 (since rA 6= 0) or AB = constant
= 0 and thus
B
A
rA B
A
dr
1
(or ln AB = constant). For r we require A = B = 1, therefore AB = 1 A(r) = B(r)
.
Consider

Introducing this into R22 (22.12) and R11 (22.11) leads to

R22 = 1 rB 0 B = 0,
R11 =

(22.15)

B 00
B0
rB 00 + 2B 0
1 dR22

=
=
= 0.
2B
rB
2rB
2rB dr

(22.16)

With (22.15), (22.16) is automatically satisified (since R22 = 0 also its derivative vanishes). We write
(22.15) as
d
(rB) = 1.
dr
We integrate it and get rB = r + constant
| {z } = r 2a. Then

(22.17)

2a

B(r) = 1
A(r) =

2a
,
r

1
,
1 2a
r

(22.18)

for r r0 . This solution for the vacuum Einstein equations was found in 1916 by Schwarzschild. The
Schwarzschild solution is


2a 2 2
dr2
ds = 1
c dt
r2 (d2 + sin2 d2 )
r
1 2a
r
2

The constant can be determined by considering the Newtonian limit:

g00 = B(r) 1 + 2
r

GM
2a
=12 2 =1 .
2
c
c r
r

Thus one introduces the so called Schwarzschild radius:

rS = 2a =

73

2GM
c2

(22.19)

23

2GM
c 2 R

2GM
c2

' 3 km (M ' 2 1030 kg, R = 7 105 km) so

dt
22
. A clock at rest in r has the proper time d = B dt, thus d
diverges

The Schwarzschild radius of the Sun is rs, =

rS,
R

GENERAL EQUATIONS OF MOTION

' 4 106

at r rS . This implies that a photon emitted at r = rS will be infinitely redshifted (t is not a good
coordinate either for events taking place at r rS ). A star, whose radius rstar is smaller than rS , is
a black hole since photons emitted at its surface cannot reach regions with r > rS .
Expanding the Schwarzschild metric in power of

rS
r

and comparing it with the Robertson expansion

(22.3), one finds = = 1 for general relativity.

23

General equations of motion

We now consider the motion of a freely falling material particle or photon in a static isotropic gravitational field (e.g. motion of planets around the Sun). For the relativistic orbit xk () of a particle in
a gravitational field we have:

and

d2 xk
k dx dx
=

d2
d d

 2
 2 2

c
ds
d
dx dx
=
= c2
=
g
0
d d
d
d

(23.1)
m 6= 0,

(23.2)

m=0

For a massive particle we can take the proper time as a parameter for the trajectory or orbit (d = d ).
For massless particles one has to choose another parameter. For the spherically symmetric gravitational
field, we use the metric (r > r , radius of the star)
ds2 = B(r)c2 dt2 dr2 A(r) r2 (d2 + sin2 d2 ),

(23.3)

with the coordinates (x0 , x1 , x2 , x3 ) = (ct, r, , ). Equations (23.1) (23.3) define the relativistic
Kepler problem. Using the Christoffel symbols given in (22.6), we get for (23.1):
d2 x0
B 0 dx0 dr
=
,
2
d
B d d

2
 2
 2
 2
d2 r
B 0 dx0
A0 dr
r d
r sin2 d
=

+
+
,
d2
2A d
2A d
A d
A
d
d2
2 d dr
=
+ sin cos
2
d
r d d

d
d

(23.4)

(23.5)

2
,

(23.6)

d2
2 d dr
d d
=
2 cot
.
2
d
r d d
d d

(23.7)

Equation (23.6) can be solved by

= constant.
2
Without loss of generality we can choose the coordinate system such that =
=

lies on the plane with =

d
2 . d2

(23.8)

2,

this way the trajectory

= 0 corresponds to angular momentum conservation. With (23.8)

22 Apparently it seems that the Schwarzschild metric is singular for r = r , but this is not the case. It is only an
s
artefact of the coordinate choice. To be discussed later.

74

23

GENERAL EQUATIONS OF MOTION

we get for (23.7) :

1 d
r2 d


r

2 d

= 0,

(23.9)

which leads to

d
= constant = l.
(23.10)
d
l can be interpreted as the (orbital) angular momentum (per unit mass). Equations (23.8) and (23.10)
r2

follow from angular momentum conservation, which is a consequence of spherical symmetry (rotation
invariance).
Equation (23.4) can be written as (B = B(r())
  0

d
dx
ln
+ ln B = 0,
d
d
h 0  i
which can be integrated as ln dx
B = constant or
d
B

(23.11)

dx0
= constant = F.
d

(23.12)

In (23.5) we use (23.8), (23.10) and (23.12) and get:

d2 r
F 2B0
A0
+
+
2
2
d
2AB
2A
We multiply it with 2A

dr
d

dr
d

2

l2
= 0.
Ar3

and get
"  
#
2
dr
l2
F2
d
A
+ 2
= 0.
d
d
r
B

(23.13)

23

(23.14)

Integration gives

A

dr
d

2
+

l2
F2

=  = constant.
r2
B

(23.15)

Integrating it once more we get r = r(). Inserting then this result into (23.10) and (23.12), we obtain
with one more integration = () and t = t(). Next we eliminate and get r = r(t) and = (t).
Together with =

2,

this is then a complete solution (generally it has to be done numerically).

Equation (23.2) becomes

dx dx
g
=B
d d

dx0
d

2


A

dr
d

2
r

d
d

2

using (23.8), (23.10), (23.12) and (23.15). On the other hand

c2 (m 6= 0)
.
=
0 (m = 0)
We are left with two integration constants, F and l.
23 Notice:

dA
dr
= A0
d
d

75

r sin

d
d

2
= ,

(23.16)

23

23.1

GENERAL EQUATIONS OF MOTION

Trajectory

From (23.15) we get

s
dr
=
d

F2
B

rl 2 
,
A

(23.17)

and with (23.10),

d d
l
d
=
= 2
dr
d dr
r

s
F2
B

A
.
2
rl 2 

(23.18)

Thus,
Z
(r) =

dr
r2

s
F2
B(r)l2

A(r)
r12


l2

(23.19)

With this we can find the trajectory = (r) in the orbital plane. (Massive particles: 2 integration
constants

F2
l2

and


l2 ,

massless particles: only

F2
l2 ).

Trajectory in Schwarzschild metric:

rS
r

Insert Schwarzschild metric: B(r) = A1 (r) = 1

dt
t =
,
d

r =

=1

dr
,
d

2a
r

and write:

d
=
.
d

Then with (23.8), (23.10), (23.12) and (23.15) we get



2a

= F,
= , ct 1
2
r

r2 = l.

(23.20)

Multiplying (23.15) with B and using AB = 1, we have

r 2
a
l2
al2
F2 

+ 2 3 =
= constant.
2
r
2r
r
2

(23.21)

The radial component can be written as

r 2
+ Veff (r) = constant,
2
2GM
c2 ,

 = {c2 , 0})

GM + l2
r
2r 2
Veff (r) =
l2 GM l2

with the effective potential (2a =

(23.22)

2r 2

c2 r 3

GM l2
c2 r 3

(m 6= 0)

A formal solution r = r() of (23.22) is given through the following integral :

Z
dr
= p
.
2(constant Vef f (r))
Due to the

1
r3

(23.23)

(m = 0)

(23.24)

term (relativistic), this is an elliptical integral which has to be solved numerically.

For small values of r, centrifugal potential term dominates (as long as l is not too small), then for
even smaller values of r the attractive relativistic term takes over:
24 l
25 v
c

l2
v2
r2
2
104 vc2 108

rv

76

24 25

23

Veff

GENERAL EQUATIONS OF MOTION

collision
circular unstable
scattering

ellipse with precession

circular stable

Figure 11: Effective potential for massive particles in Schwarzschild metric

GM l2
GM v 2
'
2
2
r c r
r c2

Eq. (23.22) differs from the non-relativistic case by an additional

by terms of order

(23.25)
1
r3

term and r =

dr
d

differs from

dr
dt

v
c2 .

Observations:
Where Veff has a minimum there are bounded solutions, however due to the relativistic effects
there will be small deviations from the elliptical orbits (precession of the perihelion). As a special
case, with r = 0, the circular orbit is a possible solution (in which case the constant in (23.22) is
equal to the value of Veff at its minimum).
The solution at the maximum of Veff is an unstable circular orbit.
If the constant is positive one gets non-bounded trajectories (corresponding to hyperbolic solutions in the non-relativistic case).

77

23

GENERAL EQUATIONS OF MOTION

If the constant is larger than the maximum value of the potential, the particle falls into the
center.
At minimum and maximum we have

dVef f
dr

= 0. For m 6= 0 we get

r
c2 2
r 2 + 3 = 0.
2
l
rS
In order to have two real solutions, we need

3c2
l2

<

l lcrit =

1
2 .
rS

(23.26)

That means

3 rS c.

(23.27)

For l lcrit , the angular momentum barrier gets smaller and smaller until the maximum and
minimum fall together for l = lcrit . For l < lcrit , the potential decreases monotonically for r 0.
Veff

collision
circular unstable

scattering

Figure 12: Effective potential for massless particles in Schwarzschild metric

78

23

GENERAL EQUATIONS OF MOTION

Here both terms are proportional to l2 , thus the shape of Veff does not depend on l. At rmax = 32 rS
the potential has a maximum. At rmax the photons can move on a circular orbit, which is unstable. If
the constant in (23.22) is smaller than Veff (rmax ) then the incoming photon will be scattered, whereas
if the constant is bigger the photon will be absorbed at the center.

26 for

r rS the Schwarzschild solution is not applicable

79

26

24

LIGHT DEFLECTION

Part VI

Applications of General Relativity

24

Light deflection

The trajectory r = r() of a photon in the gravitational field is given by (23.19) ( = 0):
Zr
(r) = (r0 ) +

d
r
2
r

r0

A(
r)

F2
B(
r )l2

1
r2

(24.1)

r0

Sun
light ray
Figure 13: Deflection of light by the Sun
We will now show that light is deflected by a massive body, carrying through calculations for the Sun.
In fig. 13, the following quantities are defined: light is deflected by and r0 is the minimal distance
(or impact parameter) from the Sun. For simplification we assume also r0  rS .
As starting point of the integration we choose the minimum distance r0 , where we set (r0 ) = 0.
Going from r0 till r the angle changes by (). Along the drawn trajectory the radial vector turns
by 2(). If the trajectory would be a straight line, then 2() = .
Thus = = 0 for a straight line and in general ((r0 ) = 0):
= 2() .

(24.2)

At r0 , r() is a minimum, thus




dr 
= 0.
d r0

80

(24.3)

24

LIGHT DEFLECTION

light ray
(r) () =

(r) () =
Sun

Figure 14: Non-deflected ray of light

From (24.3) we get with (23.17) and (23.18) the condition

F2
B(r0 )
=
.
l2
r02

(24.4)

This way we can eliminate the constants F and l in terms of r0 with (24.1):
Z v
u
A(r)
dr u
t B(r ) 2
.
() =
r
0
r
2 1

(24.5)

B(r) r0

r0

2a
Let us compute the integral by inserting the Robertson expansion A(r) = 1 + 2a
r , B(r) = 1 r (with

a=

rs
2

GM
c2 ).

We get using

We keep terms up to

a
r

with



B(r0 ) r2
r2
1
1
1 ' 2 1 + 2a

1
B(r) r02
r0
r
r0


 2
2ar
r
.
= 2 1 1
r0
r0 (r + r0 )

1 + x = 1 + x2 ,
Z
() '

r0
r2 r02 r
dr

p
r0

"
= arccos

r 
0



a
ar
1+ +
r
r0 (r + r0 )

a
+
r0

#
p
r
r2 r02
a r r0
+
r
r0 r + r0
r0

a
a
+ + .
2
r0
r0

(24.6)

With (24.2) we get

=

4a
r0

1+
2


=

81

2rS
r0

1+
2


.

(24.7)

24

For general relativity, = 1, rS =

2GM
c2

LIGHT DEFLECTION

and thus
=

2rS
r0

For a light ray which just grazes the surface of the Sun (r0 = R = 7105 km) we get ( = 180360000 ):


1+
00
= 1.75
. 27
(24.8)
2
On May 29, 1919, an eclipse allowed experimental confirm of this result.

Figure 15: Gravitational lensing in the Abel 2218 galaxy cluster

27 cheating

with Newtons theory gives half this result that is 0.8400

82

25

25

PERIHELION PRECESSION

Perihelion precession

Consider the elliptical orbit of a planet around the Sun:

y
Mercury
r
p

rmax

rmin
Sun

Figure 16: Non-relativistic elliptical orbit of Mercury around the Sun

We will use the following notations: minimum distance r = rmin , maximum distance r+ = rmax ,
= (r ), A = A(r ), B = B(r ). The relativistic orbit follows from equation (23.19) for
r = r() with  = c2 . The integral gives for the change in angle between r and r+ :
Zr+
+ =

dr
r2

A(r)
F2
1
B(r)l2 r 2

Zr+
c2
l2

dr
r2

A(r)
.
K(r)

(25.1)

For a full orbit the angle is 2, i.e. twice the integral (25.1). The shift of the perihelion (per complete
orbit) is given by
= 2(+ ) 2.
The integrand in (25.1) is equal to

d
dr .

For r = r due to

dr
d

(25.2)
= 0,

K(r) r2 has to vanish, thus

K(r ) = 0:
F2
1
c2
= 2 + 2.
2
B l
r
l

83

(25.3)

25

PERIHELION PRECESSION

This way we can express F and l through r :

F2
=
l2

1
2
r+
1
B+

1
2
r

1
B

B+

2
2
r
r+

,
1
1
2 r2
r+
B+ B
r2

(25.4)

r2

2 r2
r+
c2
B+ B


.
=

1
1
l2
B+ B
2 r2

r+
B+
B

This leads us to an expression for K(r):




1
1
1
2
2
r
B(r) B r+ B(r)


K(r) =
1
1
2 r2
r+
B+ B

1
B+

(25.5)

1
.
r2

(25.6)

For A and B insert the Robertson expansion

A(r) = 1 +

2a
+ ...,
r

(25.7)

 a 2
2a
+ 2( )
+ ...,
r
r
 a 2
2a
1
=1+
+ 2(2 + )
+ ....
B(r)
r
r
B(r) = 1

With eq. (25.9), K(r) becomes a quadratic form in 1r . Since

d
dr

(25.8)
28

(25.9)

= for r = r , K+ = K = 0. This

determines K(r) up to a constant c:



1
1
K(r) = c

r
r



1
1

r
r+


.

(25.10)

c can be obtained by comparing with (25.6) for r . With (25.9) one gets


a
a
c = 1 (2 + )
+
.
r+
r

(25.11)

We get thus the following integral:

1
+ =
c

Zr+

dr
r2

a
1+
| {z r }




1
1

r
r



1
1

r
r+

 21
.

(25.12)

from A 1+ a
r

We perform the following substitution:





1
1 1
1
1 1
1
=
+
+

sin ;
r
2 r+
r
2 r+
r
28 v 2
c2

(25.13)

ar ; terms g00 u0 u0 Bc2 and g11 u1 u1 Av 2 Ac2 ar show up both and have thus to be expanded to the
same order in ar . Therefore B has to be expanded one order in ar more than A.

84

25

PERIHELION PRECESSION

and = 2 , respectively. With

 


1 1
1
1
1
d
= 2 dr =

cos d,
r
r
2 r+
r


1
1
1 1
1
=

(1 + sin ),
r
r
2 r
r+


1
1
1 1
1

(1 sin ),
r
r+
2 r
r+

r+ and r correspond to =

(25.14)

(25.15)

(25.16)

we get for the integral

1
+ =
c

Z2







1
a 1
1
a 1
+
+

sin .
d 1 +
2 r
r+
2 r+
r

(25.17)

Now introduce the parameter p of the ellipse (see figure 16):

2
1
1
=
+
.
p
r+
r

(25.18)

Integration of eq. (25.17) leads to








a
a
a
a

= 1 + (2 + )
1+
= 1 + (2 + 2) .
+ = 1 +
p
p
p
p
c

(25.19)

Precession per orbit for the perihelion is:

6a
= 2(+ ) 2 =
p
In general relativity = = 1 and so

2+2
3

2 + 2
3


.

= 1. Thus,

6a
p

Consider Mercury: p = 55 106 km, 2a 3 km, = 180 360000 which give =

0.104

00

(25.20)

6a
p

(per full orbit). In 100 years Mercury fulfills 415 orbits around the Sun, this way we get
00

= 43 (per century). For more distant planets (Venus, Earth,. . . ) is at most 500 per century.
Already in 1882, Newcomb found a perihelion precession of 4300 per century for Mercury. Full perihelion
precession amounts to 57500 per century of which 53200 are due to the influence of other planets (this
within Newtonian theory). One finds
2 + 2
= 1.003 0.005,
3

(25.21)

in good agreement with general relativity. So far, the parameters of the Robertson expansion are
constrained to | 1| < 3 104 and | 1| < 3 103 . More recently (radar echoes delay from
Cassini spacecraft): | 1| = (2.1 2.3) 105 .29
29 B.

Bertotti et al. Nature 425, 374 (2003)

85

25

PERIHELION PRECESSION

Sun

Mercury

Figure 17: Illustration of the perihelion precession of Mercury (effect strongly exagerated)

25.1

Quadrupole moment of the Sun

A quadrupole moment of the Sun could also influence a perihelion precession of Mercury, that is why
one has to study it. The mass quadrupole moment of the Sun (due to its rotation) is
2
Q = J2 M R 

with

J2 =

2 Rk R
,
5 R

(25.22)

and
Rk : orthogonal (to R ) radius,
R : radius orthogonal to the plane containing the planet orbits and parallel to the rotation axis
of the Sun.
The induced gravitational potential in the planets orbital plane (which is also the equatorial plane of
the Sun) is
GM
GQ
3.
r
2r
The additional term has the same r dependence than the additional relativistic term:
(r) =

Veff =

l2
GM l2
GM
+ 2 2 3 .
r
2r
c r
86

(25.23)

(25.24)

26

With l pv and v 2

GM
p

LIE DERIVATIVE OF THE METRIC AND KILLING VECTORS

we can compare the two terms (their relative strength)

2
2
2
J2 R
J2 R
J2 R
GQ

.
2
2
pa
p GM
GM cl 2
p2 vc2
c2

We see that the full expression for the perihelion precession is given by

2 
J2 R
6a 2 + 2
=
+
.
p
3
2ap
From observations one finds J2 (1 1.7) 107 ; thus the additional term is

(25.25)
2
J2 R 
2ap

5 104 ,

accounting for at most 1/10 of the error given in (25.21) and is thus negligible.

26

Lie derivative of the metric and Killing vectors

Consider the Lie derivative of the metric tensor g in the direction of the vector K. According to
equation (14.2) we get:
LK g = g, K + g K , + g K , .

(26.1)

To rewrite this expression we observe the identities

K = g K ,
K , g =

K
(K g )
K
g
g
g
=
K
=
K
.

x
x
x
x
x

Hence eq. (26.1) can also be written as

LK g =
=



K
g
g
K
g
+
+
K

x
x
x
x
x
K
K
+
2K
x
x

(26.2)

= K; + K; .
An infinitesimal coordinate transformation is a symmetry of the metric if LK g = 0, thus if
K; + K; = 0
Any 4-vector K (x) satisfying this equation will be said to form a Killing vector.

(26.3)
30

Example: Consider a stationary gravitational field, for which there exists coordinates {x } such that the
components of g do not depend on ct = x0 (for instance Schwarzschild metric). Let K = 0 with
the corresponding vector field 0 ( 0 ). Inserting K into (26.1) one gets LK g = g,0 + 0 + 0 =

x0 g

= 0 (since g does not depend on x0 ). K is a Killing vector or Killing field or an infinitesimal

isometry.
30 Named

after 19th century mathematician Wilhelm Killing

87

27

MAXIMALLY SYMMETRIC SPACES

Notice that, due to the properties of the Lie derivative, if K1 and K2 are Killing vectors, LK1 g = 0,
LK2 g = 0 then [K1 , K2 ] is also a Killing vector since
[LK1 , LK2 ]g = L[K1 ,K2 ] g = 0.

(26.4)

We are used to the fact that symmetries lead to conserved quantities: in classical mechanics the angular
momentum of a particle moving in a rotationally symmetric field is conserved. In the present context,
the concept of symmetries of a gravitational field is replaced by symmetries of the metric and we
therefore expect conserved quantities to be associated with the presence of Killing vectors.
Let K be a Killing vector and x ( ) be a geodesic. Then the quantity K x is constant along the
geodesic. Indeed,
D
(K x ) = ( K x )x + K ( x ) x
| {z }
d
=0 geodesic

1
( K + K ) x x = 0.
2|
{z
}

(26.5)

=0 (26.3)

T is the covariantly conserved symmetric energy-momentum tensor with T = 0. Then J =

T K is a covariantly conserved current:
1
J = ( T ) K + T K = T ( K + K ) = 0,
2
| {z }
{z
}
|
=0

=0 (26.3)

to which we can associate a conserved charge.

27

Maximally symmetric spaces

Maximally symmetric spaces are spaces that admit the maximal number of Killing vectors (which below
will turn out to be n(n+1)
for an n-dimensional space). In the context of the cosmological principle
2
such spaces, which are simultaneously homogeneous (the same at every point) and isotropic (the
same in every direction), provide a description of space in a cosmological space-time.
From equation (17.2) we had (from definition of Riemann tensor and covariant derivative)
([X , Y ] [X,Y ] )V = R X Y V ,

(27.1)

along with X = X , Y = Y and Y V = Y V = Y V , X Y V = X (Y V ) =

X ( Y ) V + X Y V etc, we get
[ , ]V = R V .

(27.2)

Taking into account the first Bianchi identity, it is possible to find that for a Killing vector K , one
has (no proof here)

K (x) = R
K (x)

88

(27.3)

27

MAXIMALLY SYMMETRIC SPACES

for x = x0 . Thus a Killing vector K (x) is completely determined everywhere by the values of K (x0 )
(i)

and K (x0 ) at a single point x0 (think of Taylor expansion). A set of Killing vectors {K (x)} is
said to be independent if any linear relation of the form
X
ci K(i) (x) = 0,

(27.4)

with constant coefficients ci ; implying ci = 0. Since in an n-dimensional space-time there can be

(i)

at most n linearly independent vectors K (x0 ) at a point, and at most

n(n1)
2

independent anti-

symmetric matrices ( K (x0 )), we reach the conclusion that an n-dimensional space-time can have
at most
n+

n(n 1)
n(n + 1)
=
2
2

(27.5)

independent Killing vectors.

Homogeneous space is meaning that the n-dimensional space(-time) admits n-translational Killing
vectors.
Isotropic space: K (x0 ) is an arbitrary anti-symmetric matrix ( rotation). We can choose
a set of

n(n1)
2

Killing vectors.

We define a maximally symmetric space to be a space with a metric with a maximal number of
n(n+1)
2

Killing vectors.

The Riemann curvature tensor of a maximally symmetric space becomes simpler. One can show (no
proof) that it becomes
Rijkl = k(gik gjl gil gjk )

(27.6)

for some constant k. The Ricci tensor then becomes

Rij (x) = (n 1)kgij .

(27.7)

R(x) = n(n 1)k,

(27.8)

The Ricci scalar can be obtained to be

and the Einstein tensor


1
n
Gik = Rik Rgik = k (n 1) 1
gik .
2
2

The Bianchi identity implies that k is a constant in order for Gik ;k to vanish. We shall deal with
space-times in which the metric is spherically symmetric and homogenous on each plane of constant
time. In our case n = 4 and the maximally symmetric subspace has 3 dimensions. Consider first the
metric on the 3-dimensional subspace
d 2 = A(r)dr2 + r2

d2
|{z}

d 2 +sin2 d2

For the Christoffel symbols, we use the ones for the general form of a static isotropic metric (22.6) with
B(r) = 0. Since the Christoffel symbols are invariant under an overall sign change of the metric, also
89

27

MAXIMALLY SYMMETRIC SPACES

the Ricci tensor is and thus one can apply (22.10)-(22.13) with B(r) = 0 for this three-dimensional
space, without caring for the sign in front of A. Hence we get for equation (27.7)
Rrr = R11 =
R = R22

A0
,
rA

1
rA0
.
= +1+
A
2A2

(27.9)

From eq. (27.7), we have Rrr = 2kA, and R = 2kg = 2kr2 . Thus from equating the two first
equation leads us to

A0
A0 = 2krA2 ,
rA

2kA =

(27.10)

while we get for the second one

2kr2 =

1
1
rA0
2kr2 A2
1
1
= +1+
= + 1 + kr2 kr2 = + 1,
+1+
2
A
2A
A
2A2
A
A

which leads to
A=

1
,
1 kr2

(27.11)

and solves also (27.10). Then the metric on the 3-dimensional subspace (maximally symmetric) is
d 2 =

dr2
+ r2 d2 .
1 kr2

(27.12)

It can be shown that k can have the following values: 0, 1.

+1 sphere, positive curvature

k = 1 hyperbola, negative curvature

0
plane, zero curvature
The full metric (with time coordinate) has then the form:
ds2 = c2 dt2 a2 (t)


dr2
2
2
2
2
+
r
(d
+
sin

d
)
1 kr2

(27.13)

where a(t) is the cosmic scale factor, which has to be determined by solving the Einsteins equations via
the matter content of the universe. This metric (first discovered by Friedmann, Lematre, Roberston
and Walker) is a reasonable ansatz for describing the universe. There is good evidence that the universe
(on large scales) is surprisingly homogeneous and isotropic (from redshift surveys of galaxies and cosmic
microwave background radiation).

90

28

28

FRIEDMANN EQUATIONS

Friedmann equations

We write the metric (27.13) as follows:

ds2 = c2 dt2 a2 (t)
gij dxi dxj ,

(28.1)

where tildes denote 3-dimensional quantities calculated with the metric gij . The Christoffel symbols
are given by (notice 00 = 0):
i ,
ijk =
jk

ij0 =

a i
j,
a

0ij = a a gij ,

(28.2)

where dot denotes derivation with respect to t. The relevant components of the Riemann tensor are:
a

Ri 0j0 = i j ,
a

ij + 2a 2 gij
Rk ikj = R

R0 i0j = a a
gij ,

(28.3)

ij = 2k
We can make use of R
gij (maximal symmetry of the 3-dimensional subspace) to compute R .
The non-zero components are then
R00

= 3 ,
a

Rij = (a
a + 2a + 2k)
gij =

a 2
k
+2 2 +2 2
a
a
a


gij ,

(28.4)

a + a 2 + k) and the non-zero components of

where gij = a2 gij . The Ricci scalar becomes R = a62 (a
the Einstein tensor are

G00 = 3

a 2
k
+ 2
a2
a


,

G0i = 0,

Gij =

k
a
a 2
2 + 2+ 2
a a
a


gij .

(28.5)

Next we have to specify the matter content. We treat here the universe as non-interacting particles or
a perfect fluid. A perfect fluid has energy-momentum tensor (equation (19.8))

p
(28.6)
T = 2 + u u g p,
c
p is the pressure, the energy density and u the velocity field of the fluid (u = (c, 0, 0, 0) in a
comoving coordinate system). The equation of state is p = p() and in particular one assumes
p = w,

(28.7)

where w is the equation of state parameter.

Examples:
For non-interacting particles (p = 0, w = 0), such matter is refered as dust. The trace of the
energy-momentum tensor is then (c = 1)
T = + 3p.

(28.8)

For radiation the energy-momentum tensor (like in Maxwell theory) is traceless and hence radiation has the equation of state
p=
thus w = 31 .
91

1
,
3

(28.9)

28

FRIEDMANN EQUATIONS

As we will see, a cosmological constant corresponds to a matter contribution with w = 1.

The conservation law T ; = 0 implies T 0 ; = 0 or T 0 + T 0 + 0 T = 0. For a perfect
fluid: t (t) + 0 + 000 + 0ij T ij = 0 (with i, j = 1, 2, 3). Inserting the expressions for the Christoffel
symbols (28.2) we get:
a
= 3( + p) .
a

(28.10)

a
= 3 .

(28.11)

For dust (p = 0):

Integration gives a3 = constant or a3 . For a radiation dominated universe we get

p=

= 4 .
3

(28.12)

Integration gives a4 = constant or a4 . More generally for (28.7) one gets:

a(t)3(1+w) = constant.

(28.13)

The Einstein equations with (equation (20.10)) are

G =

8G
T + g .
c4

Using (19.8) and that u = (c, 0, 0, 0) in a comoving coordinate system, let us write down the 00component and the ij-component of this equation:
 2

a
k
3
+ 2 = 8G + ,
a2
a


k
a
a 2
2 + 2 + 2 gij = (8Gp + ) gij .
a a
a

(28.14)

(28.15)

One has in addition equation (28.10) from the conservation law. Using the first equation to eliminate
a 2
a2

k
a2

from the second one, one obtains the Friedmann equations:

k
8G

a 2
+ 2 =
+ ,
2
a
a
3
3
3

(28.16)

= 4G( + 3p) ,
a

(28.17)

a
= 3( + p) .
a

(28.18)

Notice that one could also use the form (20.8) of the Einstein equations to derive the above equations.
Introducing the Hubble parameter: H(t) =

a(t)

a(t)

a(t)
and the deceleration parameter: q(t) = a(t)
a 2 (t) , with

their present day values denoted by H0 = H(t0 ) and q0 = q(t0 ), where t0 is the age of the universe;

92

28

FRIEDMANN EQUATIONS

we get instead
H2 =
q=

8G
k
2 +
3
a
3
1
(4G( + 3p) )
3H 2

(28.19)

d
(a3 ) = 3Hpa3
dt
In the case of = 0, we define a critical density crit =

3H 2
8G

and a density parameter =

crit .

Then

< crit k = 1 open universe

= crit k = 0
flat universe

>
crit k = +1 close universe
This density has contributions from
ordinary matter (baryons) baryons 0.05
dark matter DM 0.27
cosmological constant =

crit

0.68 with =

8G

(also called dark energy).

H01 is related to the age of the universe. H0 67.80 0.77

years.

km
sec/Mpc

gives an age of 13.8 billion

31

31 Planck

2013 results. XVI. Cosmological parameters - http://arxiv.org/abs/1303.5076

93