General Relativity Px436
General Relativity Px436
General Relativity Px436
Gareth P. Alexander
University of Warwick
email: G.P.Alexander@warwick.ac.uk
office: D1.09, Zeeman building
http://go.warwick.ac.uk/px436
Preface iii
2 Differential Geometry 27
2.1 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 The Metric . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.1 Lorentzian Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3 Geodesics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.4 Angles, Areas, Volumes, etc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.5 Vectors, 1-Forms, Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.6 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.7 Covariant Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.7.1 Continuity, Conservation and Divergence . . . . . . . . . . . . . . . . . 47
2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
i
5 Gravitational Collapse and Black Holes 77
5.1 Gravitational Collapse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 The Kerr Black Hole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.1 Geodesics of the Kerr Metric . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.2 Event Horizon and Ergoregion . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 Observational Evidence for Black Holes . . . . . . . . . . . . . . . . . . . . . . 82
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6 Gravitational Waves 87
6.1 Linearised General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.2 The Graviton is a Spin 2 Particle . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.3 Gravitational Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.4 Gravitational Wave Detectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7 Cosmology 98
7.1 de Sitter Universe and Cosmological Constant . . . . . . . . . . . . . . . . . . . 98
7.2 The Friedmann-Lematre-Robertson-Walker Metric . . . . . . . . . . . . . . . . 99
7.3 Dark Energy and -CDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
ii
Preface
The aim of this course is to give an account of gravity and general relativity, as I understand
it. I hope that you will find the subject as fascinating as I did when I was an undergraduate.
The first four weeks of the course will be devoted to an exposition of the theoretical foun-
dations of general relativity leading to a derivation of the Einstein equations. This will be
divided in equal parts into a discussion of the stress-energy-momentum tensor the source
of gravitation and curvature its influence in the gravitational field. For the remaining six
weeks we will study the predictions and implications of the Einstein equations, covering the
classical tests the motion of celestial bodies and gravitational lensing , stellar collapse and
black holes, gravitational waves and cosmology. The topics chosen, and emphasis of presenta-
tion, reflect my own knowledge and expertise and lie very much in the tradition of theoretical
physics. At the same time, I have tried to promote and give preference to those parts of the
theory that either have been tested against experiment and observation, or are the focus of
current experiments.
General relativity is a theory of the structure of space and time and as such makes consid-
erable use of (pseudo-)Riemannian geometry. It is lamentable that though geometry is taught
at school before one learns calculus, it is essentially ignored in university level physics courses
until one meets general relativity. This is to the diminishment of both subjects and makes the
presentation of both in a course such as this difficult. It is my conviction that the essential
ideas can be gained from the example of surfaces, of widespread relevance to physics in their
own right, which have the benefit of being directly visualisable. In addition to emphasising
this, I have also tried to adopt a more modern approach than in many textbooks, using intrinsic
descriptions in preference to the old-fashioned tensor calculus.
These lecture notes have been written in an informal and colloquial style. At times the
presentation is quite detailed; at others I have taken liberties. As a general rule, there is more
material in these notes than I will be able to cover in the lectures, although not too much more.
What may appear in the exam should be inferred from the official syllabus and what I say in
lectures rather than the fullness of these lecture notes. In preparing these notes I discovered
that images taken by NASA are generally not copyrighted and can be used freely for personal
or educational purposes. This is a most wonderful thing and I am grateful to be able to use
their inspiring images to illustrate these lecture notes. Their image galleries can be found at
this url http://www.nasa.gov/multimedia/imagegallery/index.html.
I am grateful to the students who took the course last year for their feedback and sugges-
tions, and to my colleagues who read my notes and spotted mistakes or points for improvement,
especially Nick dAmbrumenil and George Rowlands.
I shamelessly copy Newton in his Principia (1687): I heartily beg that what I have done
here may be read with candour; and that the defects in a subject so difficult be not so much
reprehended as kindly supplied, and investigated by new endeavours of my readers.
iii
Books and other reading
There are lots of good books. You are encouraged to read as many as you can. Some with
particular relevance to the course are listed below.
2. L.D. Landau and E.M. Lifshitz, The Classical Theory of Fields: Volume 2 of the Course of
Theoretical Physics, 4th edition, Butterworth-Heinemann, Oxford, 1975.
A wonderful account with the highest standard of theoretical physics. It is moderately dated and gives a
traditional tensor calculus approach to the differential geometry, but thereafter the treatment of physical
consequences of the theory is excellent. If you consider yourself a theorist, then this is the book I would like
to recommend.
3. M.P. Hobson, G.P. Efstathiou, and A.N. Lasenby, General Relativity: An Introduction for
Physicists, Cambridge University Press, Cambridge, 2006.
This was the previous lecturer, Elizabeth Stanways, recommended textbook. Being written relatively re-
cently it is modern in style and up to date. It should go without saying that it can still be recommended
highly.
5. B. Schutz, A First Course in General Relativity, 2nd edition, Cambridge University Press,
Cambridge, 2009.
A standard reference for undergraduate courses in general relativity that is often enthusiastically recom-
mended.
6. C.W. Misner, K.S. Thorne, and J.A. Wheeler, Gravitation, W.H. Freeman, 1973.
The big black coffee table, as it is affectionately known, is perhaps the most widely recommended text at
advanced level. Written by three who did so much to advance general relativity, and geometrical thinking
in particular, it is a classic. Its major drawback, however, is its length; 1279 pages.
iv
8. W. Rindler, Relativity: Special, General, and Cosmological, 2nd edition, Oxford University
Press, Oxford, 2006.
Another text that I read when first learning general relativity. As I recall, it is clear, pedagogical and
insightful.
9. S.W. Hawking and G.F.R Ellis, The Large Scale Structure of Space-Time, Cambridge Uni-
versity Press, Cambridge, 1973.
Hawking & Ellis is a mixture; certainly it is not the right book to learn the subject from for the first time,
but it is not intended for that purpose. There are many things it does not include, for instance any account
of the linearised theory and gravitational waves. However, the analysis of the global structure of the main
exact solutions is the benchmark for others to aspire to, as is the discussion of the singularity theorems.
10. J. Stewart, Advanced General Relativity, Cambridge University Press, Cambridge, 1993.
For many years John Stewart gave the Part III lectures on general relativity at Cambridge and this book
is based on those. I was fortunate enough to attend Johns lectures during my brief stint as a relativist;
they were excellent. The book can be recommended with one significant caveat; for Cambridge students this
course is the second time most of them have seen general relativity, there also being a Part II course.
11. S. Weinberg, Gravitation and Cosmology: Principles and Applications of the General Theory
of Relativity, John Wiley & Sons, New York, 1972.
Weinbergs book gives a treatment that adopts the style and emphasis of a particle theorist writing in the
early 1970s. Symmetry, behaviour under transformations and particle concepts are given preference over
geometric constructions. Many like this approach and Weinbergs book is frequently highly recommended,
but I have never warmed to it.
12. P.A.M. Dirac, General Theory of Relativity, Princeton University Press, Princeton, 1996.
Diracs book is now somewhat dated, although there is still much that can be learned from seeing how the
great man thought. Perhaps the most remarkable achievement is to give an account of general relativity in
only 68 pages. Compare with the 1279 pages of Misner-Thorne-Wheeler!
13. R.P. Feynman, Feynman Lectures on Gravitation, Penguin Books, London, 1999.
This is not a book to learn general relativity from. Having said that, it is fascinating. Written in 1962-1963
at the same time as Feynman was giving his more celebrated set of lectures, he does precisely what you
would want the leading quantum field theorist of the day to do; he attempts to reinvent general relativity
as a quantum field theory. Of course, he fails, but that is not the point; the insights provided by one of the
great theoretical physicists are a delight.
14. H. Hopf, Differential Geometry in the Large, 2nd edition, Springer-Verlag, New York, 1989.
For those who wish to see differential geometry done properly by a first rate mathematician, who originated
many important concepts, these lectures are highly accessible to undergraduates. The style is not the modern
one, but the clarity of presentation outweighs this in my opinion.
15. D. Hilbert and S. Cohn-Vossen, Geometry and the Imagination, AMS Chelsea Publishing,
New York, 1999.
One of my graduate students, who shall remain nameless, asked of this reference; is that Hilbert Hilbert?
I replied yes, but I believe his first name was David. The remarks made about Hopfs book apply equally
to Hilbert & Cohn-Vossen.
16. S. Chandrasekhar, The Mathematical Theory of Black Holes, Oxford University Press, Ox-
ford, 1998.
Everything that you (did not) want to know about black holes. A tour de force of mathematical physics.
v
General relativity has brought foundationally new concepts to physics, primarily on the largest
length scales, from the size of the solar system up to the entire universe. These new insights
have been rightly heralded; although Einstein did not receive the Nobel Prize for his theory of
general relativity, many others have for results related to, or impacting, its development. Here
are links to their Nobel Prize lectures, some closely related to topics in this course, others only
more tenuously, all of which you might find enlightening:
Albert Einstein (1921), for his services to Theoretical Physics, and especially for his discovery
of the law of the photoelectric effect.
Hans A. Bethe (1967), for his contributions to the theory of nuclear reactions, especially his
discoveries concerning the energy production in stars.
Sir Martin Ryle and Anthony Hewish (1974), for their pioneering research in radio astro-
physics: Ryle for his observations and inventions, in particular of the aperture synthesis
technique, and Hewish for his decisive role in the discovery of pulsars.
Arno A. Penzias and Robert W. Wilson (1978), for their discovery of cosmic microwave
background radiation.
Subrahmanyan Chandrasekhar and William A. Fowler (1983), Chandrasekhar: for his the-
oretical studies of the physical processes of importance to the structure and evolution of the
stars, and Fowler: for his theoretical and experimental studies of the nuclear reactions of
importance in the formation of the chemical elements in the universe.
Russell A. Hulse and Joseph H. Taylor (1993), for the discovery of a new type of pulsar, a
discovery that has opened up new possibilities for the study of gravitation.
Riccardo Giacconi (2002), for pioneering contributions to astrophysics, which have led to
the discovery of cosmic X-ray sources.
John C. Mather and George F. Smoot (2006), for their discovery of the blackbody form and
anisotropy of the cosmic microwave background radiation.
Saul Perlmutter, Brian P. Schmidt, and Adam G. Riess (2011), for the discovery of the
accelerating expansion of the Universe through observations of distant supernovae.
Great excitement has greeted the direct observation, just over a year ago, of gravitational
waves. The achievement of the LIGO experiment is hard to overstate and describing it will
be a highlight of these lectures. Their own webpage, with all of their scientific results, can
be found here. The original publications of the two events that have been confirmed so far
are already recognised as having historic significance; they can be recommended most strongly:
Observation of Gravitational Waves from a Binary Black Hole Merger, Physical Review Letters
116, 061102 (2016).
GW151226: Observation of Gravitational Waves from a 22-Solar-Mass Binary Black Hole
Coalescence, Physical Review Letters 116, 241103 (2016).
vi
Earthrise as seen from the moon, taken during the Apollo 8 mission on Christmas Eve 1968.
Image from NASA.
1
Chapter 1
... occasiond by the fall of an apple, as he sat in a contemplative mood. Why should it
not go sideways, or upwards? But constantly to the earths centre? Assuredly, the reason
is that the earth draws it. There must be a drawing power in matter.
1.1 Gravity
Anything that has mass interacts with everything else that has mass. That interaction is
always attractive and is known as gravity. It can be felt in the most mundane of everyday
things. Jump in the air and you come back to the ground; the rate at which you do so is
9.8 ms2 . It can be seen in the most spectacular of sights; the Milky Way on a cold, dark
night.
Observations of the night sky have provided the bulk of our knowledge about gravity. The
curious motion of the planets, as viewed from the Earth, was understood first by Copernicuss
heliocentric model and later by Newtons theory of gravity. Two bodies of masses M1 and M2
and separated by a distance r interact through an attractive force of magnitude
GM1 M2
Fgravity = , (1.1.1)
r2
directed along the straight line between them. G is Newtons gravitational constant, equal to
6.671011 m3 s2 kg1 in SI units. On this basis one predicts that the planets orbit about the
sun along elliptical trajectories, with the sun near one focus, sweeping out equal areas in equal
times and with the square of the orbital period varying as the cube of the semi-major axis of
the orbit. The agreement with observations is remarkable. On the basis of this understanding
we have been able to send men to the moon and satellites to all parts of the solar system. My
favourite result is the discovery of an entirely new planet Neptune, in 1846 predicted from
the discrepancy between Newtons theory and existing observations.
The agreement, however, is not perfect. There is a regular shift, or precession, in the
location of the perhelion of each of the planets. Most of it can be reconciled by accounting for
the gravitational pull of Jupiter, and the other planets, but a small discrepancy persists. In the
case of Mercury, where the effect is largest, the unaccounted for precession is 43 seconds of arc
per century. The precedent of Neptune suggested an undiscovered planet could be the cause
and a name Vulcan was even chosen. It was never found. When the resolution came, it
was by an altogether different conception: In Einsteins theory of general relativity the planets
move along geodesics in the space-time geometry created by the mass of the sun.
The source of gravity is mass. This same quantity also controls the dynamic response to
all forces via Newtons second law F = ma. The striking statement is that this is not true of
any other interaction; it applies only to gravity and is what makes the gravitational interaction
2
special. For comparison, consider a charged particle moving in a perpendicular magnetic field.
The Lorentz force qv B gives rise to an acceleration of magnitude qvB/m and the particle
moves along a circular trajectory of radius mv/qB. More massive particles move on circles of
greater radius. This simple observation allows many particles to be identified from their tracks
in accelerator detectors. Gravity behaves differently. In Galileos fabled experiment objects of
different mass, and composition, were dropped from the leaning tower of Pisa and observed to
all take the same time to impact the ground. A modern variant1 was used by MIT professor
Walter Lewin in his introductory lectures on mechanics. In an experiment to determine the
period of a pendulum, adding his own mass to that of the pendulum bob did not alter the
result. Precision tests of the equivalence of gravitational and inertial masses typically employ
more sensitive torsional balances. At the time Einstein developed his theory of relativity the
most precise measurements of this kind were due to Eotvos. The idea of the experiment is to
attach two equal masses to a solid bar suspended by a thread from its midpoint. The apparatus
should be isolated from vibrational and thermal disturbances and then carefully monitored. It
experiences both Coriolis forces and gravitational forces from the rotation and proximity of the
Earth. These couple to the inertial and gravitational masses, respectively, and any difference
would give rise to a rotation of the bar. The null results obtained in 1889 demonstrated
equivalence to at least 1 part in 106 . Eotvos extended and improved his experiment over the
next 30 years, improving the accuracy with which the equivalence could be established by three
orders of magnitude. Today it stands at better than 1 part in 1012 .
Einsteins equivalence principle, upon which he founded much of his general theory of
relativity, is sometimes stated in simple terms as asserting that the inertial and gravitational
masses are one and the same thing. However, to my mind this does not convey enough,
for Newton was certainly aware of what was implied when he stated that the gravitational
interaction between two bodies was proportional to the product of their masses. There is only
one mass defined in Newtons Principia, on page 12 . Einstein said, and perceived, so much
more. He perceived what Newton did not (and knew he did not), that the equivalence of inertial
and gravitational masses should be viewed as an equivalence between inertia and gravitation
as phenomena; strictly a local equivalence. Einstein was fond of thought experiments; it is
because of him that, even in the English language, we refer to them as gedanken experiments.
He gave first a simple thought experiment in which a collection of free particles in a(n inertial)
frame K are viewed from another frame K 0 uniformly accelerating with respect to K, say along
their common up direction. Being unable to say it better, I quote Einstein in his own words
at length:
Figure 1.1: Einsteins description of his equivalence principle: the frame K is inertial and K 0
is accelerating uniformly relatively to it along the vertical direction.
1
In fact both Galileo and Newton did experiments with pendulums, so it was not so much a variant as a
reboot.
2
It is Definition I. The quantity of matter is the measure of the same, arising from its density and bulk
conjunctly. Following the definition Newton writes It is this quantity that I mean hereafter everywhere under
the name of body or mass. And the same is known by the weight of each body; for it is proportional to the
weight, as I have found by experiments on pendulums, very accurately made, which shall be shewn hereafter.
3
Relatively to K 0 all the masses have equal and parallel accelerations; with respect to K 0 they
behave just as if a gravitational field were present and K 0 were unaccelerated. Overlooking
for the present the question as to the cause of such a gravitational field, which will occupy
us later, there is nothing to prevent our conceiving this gravitational field as real, that is,
the conception that K 0 is at rest and a gravitational field is present we may consider as
equivalent to the conception that only K is an allowable system of coordinates and no
gravitational field is present. The assumption of the complete physical equivalence of the
systems of coordinates, K and K 0 , we call the principle of equivalence.
Einsteins equivalence principle applies not just to phenomena involving massive particles
but also to light. Suppose an observer K 0 , experiencing a uniform gravitational field, is sending
a light pulse back and forth between two points of their lab. Let us suppose that it is being
sent horizontally, as we perceive it. On the grounds of the equivalence principle our situation
is identical in nature to the observations of a friend K who sees no gravitational field, but
instead the frame K 0 in a state of constant relative acceleration. To such an inertial observer
the trajectory of the light pulse is, as always, along a straight line. But then it cannot be so
for the observer K 0 ; in the presence of a gravitational field light must travel along a curved
arc, being deflected by the gravitational field. Now suppose we send our light pulse vertically
upwards. We are interested in its frequency when emitted and when it reaches the ceiling. Of
course, K and K 0 are in relative motion so that they do not see the same frequency; they differ
by the longitudinal Doppler shift. The new insight is that since K 0 is accelerating relative to
K the Doppler shift when the light leaves the floor is not the same as when it arrives at the
ceiling. In the presence of a gravitational field light experiences a gravitational redshift. If the
height of the room is h then the travel time for the light pulse is approximately h/c seconds,
during which time the relative speed between K and K 0 changes by gh/c = /c, where is
the gravitational potential of the ceiling relative to the floor. From the expression of special
relativity s
1 v/c
0 = , (1.1.2)
1 + v/c
for the longitudinal Doppler shift we can estimate the observed change in frequency due to
motion up the gravitational potential to be (/) /c2 . A precision test of this gravita-
tional redshift was performed in 1959 by Pound and Rebka using x-ray emission and absorption
from an 57 Fe sample and the vibrational motion of a loudspeaker to control the relative velocity
of the emitter and absorber. Recoil-less Mossbauer emission, discovered only the year before,
was needed to give sufficient sensitivity for observing the tiny gravitational effect.
A further thought experiment conveys the profundity of Einsteins principle of equivalence
and how it fundamentally alters our conception of gravity. Suppose you have a large collection
of identical metre sticks and you lay them out, at rest in frame K 0 , along a diameter and
circumference of a circle, so that they just fit snuggly end-to-end. We will suppose that there
are a number D along the diameter and a number C along the circumference. We want
to consider how the metre sticks, so laid out, appear in two frames, K and K 0 , in relative
acceleration. To set our bearings, first consider that there is no acceleration and both frames
are at rest. Then, from what we have learnt and been taught about inertial frames and about
circles and so forth, we know that C/D = . Now suppose that K 0 is rotating relative to
K with a constant angular velocity. From the point of view of K each of the rods that we
have laid out is undergoing an acceleration and there is no gravitational field. But Einsteins
principle of equivalence asserts that, from the point of view of K 0 , we may say that the rods3
are unaccelerated and behave as if a gravitational field were present. At the same time, we
know from the special theory of relativity that an observer in the frame K sees each of the rods
3
Each one, locally, not all of them simultaneously.
4
laid out along the circumference of the circle travel with some (instantaneous) velocity along
their length and so experience a Lorentz contraction, while the rods laid out along the diameter
experience no such contraction (along their length). It follows that in this case C/D > .
We therefore arrive at the result: the gravitational field influences and even determines
the metrical laws of the space-time continuum. If the laws of configuration of ideal rigid
bodies are to be expressed geometrically, then in the presence of a gravitational field the
geometry is not Euclidean.
Einstein arrived at these conclusions as early as 1907. It was a further eight years before he
fully understood their ramifications and gave a complete formulation of his general theory of
relativity.
The motion of celestial bodies tells us about the gravitational influence of the sun. Each of
the planets, asteroids and comets moves along their own individual trajectory, conveying the
local nature of the suns gravitational attraction at their positions. But the suns influence is
not restricted to just these locations; it extends everywhere. This is captured by the notion
of the suns gravitational field4 . In Newtons theory it is a scalar quantity, whose gradient
measures the gravitational force,
GM
sun = . (1.1.3)
r
Actually, it is the integral of a spherically symmetric mass density Newton was rightly proud
of this result; no matter how close we may be to the surface of the Earth, the force we feel
is the same as if its entire mass was concentrated at its centre rather than spread out over a
spherical volume. A distribution of matter with mass density (x0 ) creates at the point x a
gravitational potential (x) given by
(x0 ) 3 0
Z
(x) = G d x, (1.1.4)
|x x0 |
the integral being taken over the entire distribution of matter generating the gravitational field.
In Newtons theory mass acts as the source of gravity; the gravitational pull celestial bodies
experience comes from the sun and is generated by it.
A small amount of analysis allows this to be written equivalently as a differential equation
for . There are many ways in which one might do this; one that appeals to me is to observe
that the integral on the right-hand-side of (1.1.4) is a convolution and hence the inverse Fourier
transform of a product. We can then write the gravitational potential in Fourier space as
4
(k) = G 2 (k),
k (1.1.5)
2
k (k) = 4G (k).
2 = 4G . (1.1.6)
It is this equation that Einstein sought an improvement for in his general theory of relativity5 .
The principles that guided him were that while the source of gravity in Newtons theory is
4
This notion is quite significant, for it turns gravity into a field theory, even if the proper nature of the
gravitational field is not fully understood in Newtons formulation, as Newton was fully aware.
5
Loosely speaking, one may say that Feynman sought an improvement for the Fourier space relation
1
= 4G 2 ,
k
in his lectures on gravitation.
5
the mass density, in any relativistic theory this must be the energy-momentum content of a
continuous matter distribution, and from the equivalence principle the gravitational potential
should be considered as describing the metric properties of space and time. The equation
relating them should be a differential equation of second order, that reduces to Newtons in
a weak-field non-relativistic limit, and is consistent with all other known laws of physics. On
this basis Einstein arrived at his field equations of general relativity
1 8G
R R g = 4 T . (1.1.7)
2 c
i. If K is an inertial system, then every other system K 0 that moves uniformly and
without rotation relatively to K is also an inertial system; the laws of nature are in
concordance for all inertial systems. This statement we shall call the principle of
special relativity.
ii. Maxwells theory of electromagnetism shows that light propagates with a constant
speed c, at least in one inertial frame. According to the principle of special relativity,
it takes this same value in all inertial frames.
We deduce from these the form of the Lorentz transformations, again following Einsteins own
derivation. Suppose K and K 0 are two observers in relative motion. We may declare the
direction of relative motion to be the x-direction of Cartesian coordinate systems for each
observer, taken positive in the direction K 0 is moving away from K. Then K records the
motion of K 0 to be x = vt and likewise K 0 records the motion of K to be x0 = vt0 . Now
suppose a light pulse moves in the same direction as K 0 is moving away from K. Then our
two observers will record the motion of the light pulse as
A linear transformation between their coordinate systems is consistent with these two obser-
vations if it is of the form
x0 ct0 = x ct ,
(1.2.2)
for some , depending only on the speed of their relative motion, v. If instead the light pulse
is travelling along the direction in which K is moving away from K 0 then their observations
will be
and the linear transformation between their coordinate systems is required, by the principle of
special relativity, to take the form
x ct = x0 ct0 ,
(1.2.4)
6
for the same . These two relations are equivalent to
1 1
x0 = + 1 x 1 ct, (1.2.5)
2 2
0 1 1
1
ct = + ct 1 x, (1.2.6)
2 2
1+v/c 1/2
Using Ks observations of the location of K 0 we determine to be ( 1v/c ) and then rewrite
the Lorentz transformations in their standard form
x0 = x ct ,
(1.2.7)
0
ct = ct x , (1.2.8)
ct0 = ct x = cT.
(1.2.9)
From their perspective the time that has elapsed is longer; moving clocks run slow. Likewise,
if we are holding a stick of length L with one end at the origin and the other stretched out
along the x-direction then a moving observer, K 0 , will measure its length as follows. One end
is at the origin, x0 = 0. At the same instant (t0 = 0), from their perspective, the other end is
at the point
x0 = x ct = x 2 x = 1 L.
(1.2.10)
Thus K 0 records the length to be contracted by a factor 1 .
Measurements of time or distance are observer-dependent. However, appropriate combina-
tions of these measurements are not. They are known as the proper time, or proper length; the
time elapsed, or distance travelled, in the rest frame of the object being observed. Looking
back at Einsteins derivation of the Lorentz transformation it is easy to see that
x0 ct0 x0 + ct0 = x ct x + ct ,
(1.2.11)
so that the quantity (ct)2 + x2 + y 2 + z 2 is the same for all observers. This is known as the
invariant, or interval. It plays a special role in elucidating and understanding the nature of
space and time. It is the metric of special relativity.
Another physical quantity that we can identify as being agreed upon by all observers is
the phase of a (plane) wave. The phase of a wave records the number of crests and troughs.
Suppose in the passage of a wave a buoy bobs up and down three times. This corresponds
to a change in the phase of 3 2. All observers, regardless of their relative motion, will
record this same number, for they cannot claim that 57 waves passed if they are observing
the same event. For a plane wave, the phase may be expressed in terms of measurements of
position and time as = kx t. The quantity k is called the wavevector and the frequency.
For another observer the analogous expression will be k 0 x0 0 t0 . It follows from the Lorentz
transformation, expressing x0 and ct0 in terms of x and ct, that
h 0 i
= k0 , (1.2.12)
c c
h 0 i
0
k = k . (1.2.13)
c
7
Pay attention to the natural way that this arises. In the Lorentz transformation, reading left
to right, measurements of K 0 , ct0 and x0 , are expressed in terms of those of K. Here, reading
left to right, measurements of K, /c and k, are expressed in terms of those of K 0 , but this
interchange of the places of K and K 0 aside the form of the relation is precisely the same.
In the old language one says that the quantities (/c, k) transform covariantly, while the
quantities (ct, x) transform contravariantly6 . In modern parlance we say that (ct, x, y, z) form
the components of a 4-vector, while (/c, kx , ky , kz ) form the components of a 1-form, or
co-vector. We will discuss these things more formally in Chapter 2, but before that here is a
related and important example.
We learn in quantum mechanics that particles (indeed most things) can be described using
wavefunctions. The energy is given by Plancks formula, E = ~, and the momentum by de
Broglies, p = ~k. It follows that (minus) the energy and momentum of a particle transform
under Lorentz transformations as a 1-form7
E h E 0 i
= p0x , (1.2.14)
c c
h E 0 i
px = p0x . (1.2.15)
c
It follows from the properties of the Lorentz transformation that the quantity E 2 c2 p2 , where
p = |p| is the magnitude of the momentum, is an invariant associated to any particle; a
quantity that all observers agree takes the same value. This is the mass of the particle, or
more correctly m2 c4 .
1.2.1 Notation
Notation is hard won; it takes time and effort to figure out the nicest way to present material,
or ideas, or the most natural way to write relations. As such notation should be respected and
not sniffed at. However, it is just notation; certainly, it is not physics. For the most part,
I do not find it very interesting, so the description I give here is rather terse. It is essential,
therefore, that you do not let me befuddle you with something as bland as notation; if you
are unsure, ask me to explain I can assure you that the other students will be most grateful.
Either that or they will talk about you behind your back; one of the two.
We write x1 , x2 , x3 in place of x, y, z, respectively. Note that the indices here are super-
scripts. In the jargon they are called contravariant. We may use xi , with a lower-case Latin
index (i, j, k, . . . ), as a generic symbol for any spatial component. We write x0 in place of
ct, but continue to refer to it as time. We will write x , with a lower-case Greek index
(, , , . . . ), as a generic symbol for either a point in Minkowski space-time or for a 4-vector.
The summation convention, in which any repeated index is to be summed over, will be used
throughout.
We will write the interval as
00 = 1, 11 = 22 = 33 = 1, (1.2.17)
and all other components zero. It is called the Minkowski metric. The indices on the Minkowski
metric are both subscripts: in the jargon we say that they are covariant. It is an example of
a type ( 02 ) tensor. In the same jargon a vector is a type ( 10 ) tensor.
6
Which is covariant and which contravariant is a matter of historical convention.
7
It is nice to see that the 4-momentum is naturally a 1-form, as opposed to a vector, in agreement with
what we learn in classical mechanics.
8
It is especially common, even fundamental, to consider the interval between points that
are infinitesimally close. If the points are x and x + dx then the interval between them is
written
ds2 = (dx0 )2 + (dx1 )2 + (dx2 )2 + (dx3 )2 = dx dx , (1.2.18)
and again (even more forcefully) called the Minkowski metric.
Remark: There is no consensus over the signs in the metric, known as its signature. As
many people adopt the convention (, +, +, +), as I have, as adopt the converse (+, , , ).
Some even switch between the two when they think it is convenient to do so, although this has
to be done with great care. Informally speaking the choice of signature depends on whether
you prefer to think of the metric (interval) as measuring spatial distances or intervals of time
(or masses of particles).
x0 = x . (1.2.19)
The symbol has one superscript index and one subscript index. It is not a tensor. The
interval is invariant under Lorentz transformations. It follows that
x x = x0 x0 = x x , (1.2.20)
= , (1.2.21)
characterising how the Minkowski metric transforms under a Lorentz transformation. In many
treatments this is promoted to the status of defining a Lorentz transformation.
If x and y are 4-vectors then
x0 y 0 + x1 y 1 + x2 y 2 + x3 y 3 = x y = x y ,
(1.2.22)
is a Lorentz invariant. The final form is suggestive. We can think of x as an object that
acts on 4-vectors, in this case y , and returns a scalar (Lorentz invariant). It is evidently linear.
Linear maps from vectors to scalars are called 1-forms; you will also see the terms dual vectors
or covectors. They are also known as type ( 01 ) tensors. We create a more compact notation by
writing the 1-form x as x and think of this action of the metric as lowering indices. A
formal inverse operation of raising indices can be written as x = x . It is a linear map
taking a 1-form to a vector. For these to be consistent requires that
x = x = x = x , (1.2.23)
where is the Kronecker delta with components equal to 1 if = and zero otherwise. It is
a type ( 11 ) tensor. The object that raises indices is also a tensor, of type ( 20 ), and is called
the inverse metric. The lowering and raising of indices with the metric and inverse metric
extends to all tensors, of any rank, in a more or less obvious manner. In whimsical fashion it
is called the musical isomorphism and the operations called flat and sharp.
The transformation for 1-forms is important. Writing the invariant as x x , and using that
it is the same for all observers, we have
x0 x0 = x0 x = x x , (1.2.24)
x = x0 . (1.2.25)
Note that this is the opposite sense to the coordinates, or the components of a vector, as we
saw for (minus) the frequency and wavevector, k , or (minus) the energy and momentum, p .
9
Remark: The fact that some quantities transform one way and others oppositely is the
origin of the terms covariant and contravariant. Which type is co- and the other contra- is,
of course, a matter of historical convention.
Until now, I have not given any definition of a type ( kl ) tensor. If what you have gleaned
k
so far is that it is an object that looks like this T11 l
, having k contravariant (superscript)
indices and l covariant (subscript) indices, then that will be partly right. (Not everything that
looks like this is a tensor, is not.) For the time being, we take the following as a definition:
a type ( kl ) tensor is a multilinear map taking k 1-forms and l vectors and returning a number;
k
its components will be denoted T11 l
. Of course, in this and everything else we have written
here so far, we have been talking about the components of a vector, 1-form, or tensor but
have not specified which basis the components refer to. This is bad. We will remedy it in
Chapter 2 when we talk properly about geometry and physical quantities. Until then, please
do not panic; things may be underspecified but they are not wrong and (in my opinion) still
comprehensible.
endows Minkowski space-time with a causal structure. 4-vectors can be separated into differ-
ent types according to whether their magnitude (norm) squared under this inner product is
positive, negative, or zero. We say that a 4-vector with positive magnitude is space-like, one
with negative magnitude is time-like, and one with zero magnitude is light-like, or null
~ U
U ~ >0 space-like,
~ U
U ~ <0 time-like, (1.3.2)
~ U
U ~ =0 null.
Since the Minkowski inner product of two 4-vectors is invariant under Lorentz transformations,
the concept of space-like, time-like and null is well-defined and agreed upon by all observers.
The following results are fundamental. If U ~ is space-like then there exists a frame in which it
has no 0-component. Likewise, if U ~ is time-like then there exists a frame in which it only has
a 0-component. In both cases any frame for which this is true is called a rest frame (for the
4-vector). With a thorough understanding of the principle of special relativity these statements
are utterly trivial, but they can also be demonstrated algebraically. For instance, suppose the
4-vector U~ = (U 0 , U 1 , 0, 0) is time-like. Then under a Lorentz transformation the 1-component
becomes (U 1 U 0 ) and can be made to vanish by choosing = U 1 /U 0 . (The magnitude
of this is guaranteed to be less than 1 because the 4-vector is time-like.) Thus any time-like
4-vector is equivalent to (U 0 , 0, 0, 0) under a Lorentz transformation, for some U 0 . If U 0 is
positive the 4-vector is said to be future directed, while if it is negative it is said to be past
directed. The trajectory of any massive particle is time-like; at each point of its motion its
velocity is a time-like vector.
Through each point in Minkowski space one can draw the set of all light rays passing
through that point. The tangent vector to any light ray is of the form (U 0 , U 0 n) for a unit
magnitude 3-vector n corresponding to the direction of propagation. By a suitable rotation we
can find a frame in which its components are (U 0 , U 0 , 0, 0) but there is no rest frame. When
U 0 is positive we say the light ray is future directed. Likewise, when U 0 is negative we say the
light ray is past directed. In any case, passing through each point of Minkowski there is one
light ray for every direction in ordinary 3-space, or for each point on the surface of a 2-sphere.
The set of all such sweeps out a three-dimensional surface; the light cone through that point.
10
Figure 1.2: The structure of Minkowski space-time in the vicinity of any generic point p.
It separates the rest of Minkowski into disconnected parts. The set of points inside the light
cone through p are time-like separated from it and can be reached from p along a time-like
trajectory. We say that these points are causally connected to p. Points outside the light cone
are space-like separated from p; it is not possible for a massive, or massless, particle to pass
through both the given point and p.
The set of points x0 = 0 is space. What this means is that any two points in the
set are space-like separated. It is perhaps better to give a local characterisation: at any
point p of the space (x0 = 0), the set of tangent vectors are all space-like. Any subset
of Minkowski with this property is called space-like. The basic examples are the spaces
x0 0 = 0 as seen by different observers, but there are others. Here is a nice example. The set
(x0 )2 + (x1 )2 + (x2 )2 + (x3 )2 = 1 is a space-like hypersurface. (For simplicity restrict to
x0 > 0.) There are several ways of showing this; here is one. The points of the surface can be
given the explicit parameterisation
~s1 = [sinh(u), cosh(u) cos(v), cosh(u) sin(v) cos(w), cosh(u) sin(v) sin(w)]T , (1.3.4)
~s2 = [0, sin(v), cos(v) cos(w), cos(v) sin(w)]T , (1.3.5)
T
~s3 = [0, 0, sin(w), cos(w)] , (1.3.6)
Remark: The example we have described is an isometric embedding of the hyperbolic plane
H3 . Because of their familiarity with special relativity, physicists often find this picture
of the hyperbolic plane easier to grasp than the Poincare disc model, usually preferred by
mathematicians.
Remark: A time-like hypersurface is any subset of Minkowski with the property that at
each point there is a vector tangent to it that is time-like. Likewise, a null hypersurface is
any subset of Minkowski with the property that at each point there is one, and only one, null
vector tangent to it. This being said, you might like to show that the light cone through any
point is a null surface.
11
1.4 Particle Motion
In many situations we learn about the world by observing the motion and behaviour of test
particles; particles that experience their environment without appreciably disturbing it. On
the basis of observations we infer how test particles behave and then use them to deduce the
nature of any other environment (gravitational field, electric field, etc.) we put them in. One
of the most basic statements in mechanics is that, when left alone, test particles move along
straight lines at constant speed; Newtons first law8 . Recognising that a straight line is the
shortest distance between two points, or the path that takes least time, we can state this
alternatively that the trajectory of a free (test) particle is such as to make
Z final
S = mc d, (1.4.1)
init
take a minimum value, where is c times the proper time for the particle motion. The quantity,
S, is known as the action. The prefactor is an arbitrary constant that depends only on the
test particle; it is its mass. This way of thinking about particle motion, and physics in general,
is known as Hamiltons principle of least action. Not all of you may be familiar with it, so we
will use this example to illustrate and introduce it. The action is a number9 that depends on
the particle motion through its trajectory x ( ). Different trajectories yield different values
for the action. Hamiltons principle asserts that the observed physical trajectory is the one
that corresponds to the least value of the action, or more correctly to a critical point of the
action. It is one of the most profound and unifying concepts in all of physics.
Figure 1.3: Schematic of the motion of free test particles: Left; Newtons first law; such
particles travel along straight lines at constant speed, which is entirely equivalent to being
critical points of the distance, or time, between two points (the blue curve is shorter than
the red, for any ). Right; The symmetry of Minkowski implies that physical quantities are
unchanged by uniform translations . This leads to conservation of energy and momentum.
Let us see what this means for a free particle. For the trajectory x ( ) to be a critical
point means that the value of the action does not change to first order for all variations of
the trajectory. A variation of the trajectory between two points can be written x ( ) + ( ),
where is small and vanishes at the endpoints, but is otherwise arbitrary, and continues to
denote (c times) the proper time for the trajectory x ( ). The action is the time taken along
this trajectory. The infinitesimal interval between points of the trajectory x ( ) + ( ) is
d dx 2
ds2 = dx + d dx + d = d 2 + 2 d + O(2 ),
(1.4.2)
d d
remembering that is (c times) the proper time along the trajectory x ( ). It follows that
8
In his Principia, this is Definition III on page 1: The vis insita, or innate force of matter, is a power of
resisting, by which every body, as much as in it lies, endeavours to persevere in its present state, whether it be
of rest, or of moving uniformly forward in a right line.
9
Its dimensions are those of Plancks constant ~.
12
the change in the value of the action associated to this variation of the trajectory is
Z final
d dx
S[x + ] S[x] = mc d + O(2 ), (1.4.3)
init d d
dx final
Z final
d 2 x
= mc mc 2
d + O(2 ). (1.4.4)
d init init d
The first term is zero if the variation vanishes at the two endpoints, i.e. the trajectory is
always between the same initial and final points. In that case, we see that the change in the
action vanishes for arbitrary variations ( ), and the trajectory x ( ) corresponds to one of
least action, if
d2 x
= 0, (1.4.5)
d 2
or the test particle moves freely along a straight line at constant speed; precisely Newtons
first law.
Minkowski space has a fundamental symmetry; it is homogeneous and isotropic; it looks the
same at every point. This corresponds to the basic observation that physical phenomena appear
the same wherever you happen to be or whenever you happen to perform your experiment. A
pure translational shift x 7 x + , for constant , has no effect on physical phenomena.
In this case, and when x is the physical trajectory so that it satisfies the equation of motion,
the previous calculation for the difference in the value of the action (1.4.4) gives
dx final
S[x + ] S[x] = mc . (1.4.6)
d init
But this must be zero since a pure translation has no effect on any physical quantities, i.e.
S[x + ] = S[x]. The conclusion is significant: the quantity
dx
mc p , (1.4.7)
d
is conserved in any physical motion; it is called the 4-momentum10 . p0 = mc is (minus)
the energy divided by c and pi = m t xi is the 3-momentum. Each component is conserved
separately. So too is the magnitude of the 4-momentum
2
p p = E/c + pi pi = m2 c2 , (1.4.8)
which is the familiar relation between energy, 3-momentum and mass. The general relation
between symmetries and conserved quantities, illustrated here, is known as Noethers theorem.
Massive particles move along time-like trajectories. There is a frame of reference the rest
frame in which its speed is zero. Light is different; for light there is no such frame and
all observers record the same value c for its speed, regardless of how they are moving. It is
the same to say that the photon is massless. Nonetheless, like massive free particles, free
photons are observed to move along straight lines. This straight line may be described by its
wavevector k , the analogue of the momentum of a massive particle. As the photon is massless,
its wavevector is null, k k = 0. This can also be viewed as the condition that the wavefunction
eik x satisfy the wave equation. We do not develop an action principle for the trajectories
of massless particles, although this can be done; it is the branch of electromagnetism known
as geometrical optics and the action principle is Fermats principle of least time.
We may say, in summary, that the trajectory of a massive particle is a time-like straight
line, while that of a massless particle is a null straight line. We have never observed anything
that travels along space-like straight lines; this appears to be a deep property of the world.
10
Please note that I define this as a 1-form rather than a vector, which is why there is a negative sign for the
0-component; this is absent when the 4-momentum is considered a vector.
13
Remark: These basic properties of the motion of test particles Newtons first law, or the
equivalent Hamiltons principle and the quantities they have that are conserved along their
trajectories will continue to apply, in precisely the same form, to describe the motion of test
particles in general relativity. The description may seem more striking and profound, but the
underlying principles are no different. It is for this reason that we have taken the trouble to
describe the basic principles early on, and in a familiar setting.
dx final
2
mc = 0, (1.5.1)
d init
for any constant translation 11 . For a continuous gas of particles, or fluid, the mass of a
particle, m, should be replaced with a density, , that integrates over space to give the total
mass in any spatial region. Likewise, the velocity of an individual particle, dx /d , should be
replaced with the fluid velocity, u , of a material element just as in ordinary fluid dynamics.
The statement that the particle trajectory is parameterised by (c times) proper time then
becomes
u u = 1. (1.5.2)
That given, the natural replacement of conservation of energy for a particle should be the
expression Z Z
2 2
c u dvol3
c u dvol3 = 0, (1.5.3)
final init
where is a region of space, at time final or init , and dvol3 is the three-dimensional volume
element for that spatial region. A picture can help see the one below. The surfaces |final
and |init are not arbitrary; as for a particle, they are the start and end of the trajectory of
the fluid motion; so it is the direction of the fluid velocity that defines the spatial surface that
the energy is being transported through, or the unit normal to each surface is n = u . This
allows conservation of energy and momentum to be expressed as an integral over the boundary
of a region of space-time Z
c2 u u n dvol3 = 0, (1.5.4)
where n is the unit normal to the boundary of the region . The integrand T = c2 u u
is the stress-energy-momentum tensor for a perfect, pressureless fluid. For a perfect fluid with
pressure p, the corresponding expression is
T = c2 + p u u + p .
(1.5.5)
11
We will drop this from the following steps to avoid it being a distraction, but in truth it should really be
retained. The issue is: what does it mean for to be constant in a curved space?
14
Figure 1.4: Schematic of the motion of dust particles and conservation of energy-momentum.
You have to imagine that there are very many more particles than shown.
We will not consider viscous fluids. Note that in its rest frame, u = (1, 0, 0, 0), we have
T00 = c2 and Tij = pij . So, T00 has an interpretation as the energy density and Tij as the
pressure or isotropic stresses. The components T0i = Ti0 that are zero in the rest frame
have an interpretation as the momentum density. Finally, we have developed the expression
of conservation of energy and momentum in integral form; an application of Gauss divergence
theorem gives the equivalent differential expression
T = 0, (1.5.6)
t + j v j = 0,
(1.5.8)
neglecting terms of order p/c2 . Similarly, when = i we find the Navier-Stokes equations
t v i + j v i v j + i p = 0,
(1.5.9)
after a similar neglect of terms of order p/c2 . The form you may be more familiar with
t v i + v j j v i + i p = 0,
(1.5.10)
1.6 Electromagnetism
It is worth saying a few words about the general structure of the theory of electromagnetism
because it played such a formative role in the theory of relativity, because the philosophy of how
15
it is approached has such widespread resonance across all of theoretical physics and because
it provides a platform for learning how to approach general relativity. Also, by talking about
both general relativity and electromagnetism, which ultimately we formulate in different ways
but do not understand why, it is hoped that it will help encourage the creativity of a young
person to discover a better way than we currently know of thinking about and understanding
both subjects. The following treatment is inspired strongly by Landau.
Maxwells equations are traditionally written
1
E= , B = 0,
0 (1.6.1)
B 0 0 t E = 0 J, E + t B = 0.
These equations are Lorentz covariant; that observation was one of the strongest motivations
for Einstein in his development of special relativity. (His foundational 1905 paper is titled Zur
Electrodynamik Bewegter K orper.) However, this is not most readily apparent the way that
they are written. We consider how to rewrite them to make the Lorentz covariance manifest.
There is disparity in the four Maxwell equations; two have sources and two do not. Indeed,
we have searched extensively for magnetic monopoles and found none. There are two ways to
think about this. The first is that the absence of sources magnetic charges and currents
for the right-hand-pair of equations is because a forteriori there are no sources. The second
is that there are sources; we just havent found them yet. The right-hand-pair of equations
should really be extended to produce symmetry with the left-hand-pair by the introduction of
magnetic charges and currents. This is bold, but its consequences are profound; it was Paul
Dirac who first showed that this would immediately imply the quantisation of electric charge.
Thus, there is strong impetus to search for magnetic monopoles; the present status is that
we have searched extensively and found none. That being said, we will adopt the first line of
thought; a forteriori there are no magnetic charges or currents.
It is established in introductory courses on vector analysis that if a vector is divergence free
then it can be expressed as the curl of another vector. In other words, the Maxwell equation
B = 0 is equivalent to B = A. This result is known as the Poincare lemma for R3 .
There is an analogous statement for the curl of a vector. Namely, if a vector is a gradient then
its curl is automatically zero. Again, the Poincare lemma provides for the converse, that if
the curl of a vector is zero then it is the gradient of a function. This means that the Maxwell
equation E + t B = 0 is equivalent to
E = t A. (1.6.2)
is called the electrostatic potential and A the magnetic vector potential. These two potentials
form the electromagnetic gauge field A = (, cA). It is a natural relativistic object; it is also
properly viewed, as here, as a 1-form. From it, the electric and magnetic fields are recovered
by forming the antisymmetric derivative
F = A A , (1.6.3)
known as the Maxwell field strength tensor. It is an antisymmetric type ( 02 ) tensor, or in other
words a 2-form. You are strongly encouraged to verify that Fi0 = Ei and Fij = ijk cBk .
Remark: Sometimes you will see it said that the electromagnetic gauge field A is just a
mathematical trick and the real physical quantities are the electric and magnetic fields. This
is nonsense. The gauge field was deduced in precisely the manner laid out by Newton that in
natural philosophy particular propositions should be inferred from phenomena the observed
absence of magnetic charges and currents and afterwards rendered general by induction. It
is pure physics, in the very finest tradition.
16
The two source-free Maxwell equations are now encoded in the statement that the anti-
symmetric derivative of the Maxwell field strength tensor is identically zero
F + F + F = 0. (1.6.4)
These are called the Bianchi identities. To verify that they are true one simply replaces each
F with an antisymmetric derivative of the gauge field to get a sum of second partial derivatives
of A. One need then only note that each such derivative occurs twice and with opposite signs.
Remark: In the language of differential forms, the Maxwell field strength tensor is the
derivative of the gauge field, F = dA. The Bianchi identities are then the statement that
because F is exact it is automatically closed.
A = A0 . (1.6.5)
x0 0
= = 0 . (1.6.6)
x
It follows at once that the transformation of the Maxwell field strength tensor is
0
F = F . (1.6.7)
This allows one to deduce how electric and magnetic fields are perceived by different observers.
We describe explicitly only one situation and leave the rest to your own imaginations. Suppose
K 0 is holding an electron. In their frame the electron is at rest and there is an electric field
but no magnetic field. What this means is that Fi0 0 = E 0 = F 0 and all other components are
i 0i
zero. But from Ks point of view the electron is moving and moving charges generate magnetic
fields. We compute this magnetic field
ijk cBk = Fij = i j F
0
= li 0j 0i lj El0 , (1.6.8)
= cBx = 0, cBy = Ez0 , cBz = Ey0 , (1.6.9)
and merely record that the electric field perceived by K is Ex = Ex0 , Ey = Ey0 , Ez = Ez0 .
17
electromagnetic gauge field A and that it is a 1-form. That given, the theory follows from
Hamiltons principle of least action.
The way that we learn about electromagnetism is by observing the motion of test charges;
particles that have charge, but an amount of charge so small that they experience their envi-
ronment without disturbing it. We need to determine how their motion relates to the electro-
magnetic field, which means to give the addition to the free particle action that comes from
interaction with the gauge field A. The electromagnetic gauge field A that is generated by
charges is a 1-form. A 1-form is precisely the thing that you can integrate along a curve, and
the trajectory of a test particle is a curve. Thus the contribution to the action for a test particle
moving in an electromagnetic field is the integral of the gauge field A along its trajectory
Z final
dx
Z
q A dx = q A (x) d. (1.6.10)
init d
The prefactor is an arbitrary constant that depends only on the test particle; it is its charge. A
central feature of electromagnetism is already in evidence here. Namely, if A is the derivative
of a function, A = , then it has no influence on the test charge for it may be integrated
to the boundary and its contribution to the action depends only on the endpoints of the
motion, not on the trajectory. 1-forms that are the derivatives of functions are called exact.
In electromagnetism one says that the electromagnetic field A is pure gauge. This given, we
see that more is true. Two electromagnetic fields A and A0 that differ by the derivative of a
function
A0 A = , (1.6.11)
exert precisely the same influence on the motion of test charges. Their influence on physical
phenomena is identical; they both describe the same electromagnetic field. This is known
as gauge invariance; a transformation between two electromagnetic gauge fields of the form
(1.6.11) is called a gauge transformation.
Now let us see how the gauge field A influences the motion of test charges, using Hamiltons
principle of least action. If the trajectory should be changed to x + the value of the action
will change to
Z Z
q A (x + ) d(x + ) = q A dx + A dx + A d + O(2 ) , (1.6.12)
Z final Z
+ q A A dx + O(2 ).
= q A dx + q A
init
(1.6.13)
Again, the boundary term vanishes because is zero at the initial and final points of the
trajectory. Combining this variation with the variation of the action for a free particle (1.4.1)
we find that the principle of least action gives the trajectory of the test particle as
d2 x dx dx
mc = q A A = q F , (1.6.14)
d 2 d d
where F is the Maxwell field strength tensor. This is the equation of motion for a charged
particle in an electromagnetic field, the covariant version of the Lorentz force equation.
Remark: What is important here is that the motion of test charges is influenced by the
electromagnetic field through the Maxwell field strength tensor F rather than the gauge
field A directly. Thus, as we learn about electromagnetic phenomena through watching the
motion and behaviour of test charges what we will learn about are the classical electric and
magnetic fields, E and B.
18
Remark: If the gauge field A is exact, or pure gauge, the field strength tensor vanishes
identically, as one should expect. One can enquire about the converse; if the field strength
tensor vanishes is A necessarily pure gauge? The answer a version of the Poincare lemma
is no. A physical manifestation of this intriguing situation is provided by the Aharonov-
Bohm effect. It has become increasingly important in recent years with the growing interest
in topological properties of materials and phases of matter.
Finally, we turn to the dynamical equations for the electromagnetic field itself, i.e. we wish
to determine the action for the gauge field A. The action is required to be Lorentz invariant,
i.e. the same as measured by all observers, and the expression for it is constrained by the set
of invariants that we can find. There are three quantities that are Lorentz covariant
A A , F F , F F . (1.6.15)
Here is the fully antisymmetric Levi-Civita symbol with 0123 = 0123 = 1.
Remark: We run through how to establish Lorentz covariance for the second of these,
F F . One starts with the statement that x x = x x is an invariant and uses the
known transformation properties of the 1-form x to deduce that the inverse metric obeys
the relation
= .
It is then just a matter of transforming the expression from the frame K to the frame K 0
0 0
F F = F F ,
0 0
= F F ,
The third invariant, F F , is a total divergence and so does not contribute to the
field equations. That is not to say that it is unimportant; it is an invariant and it does
characterise the field configuration, but it does not contribute to determining it dynamically.
The first invariant, A A , is perhaps the most subtle one. Naively, it should be discarded
on the grounds that it is not invariant under gauge transformations, however, such a term is
permissible when it arises from the spontaneous symmetry breaking of a scalar field. This is
known as the Higgs mechanism. The term A A would then give a mass to the gauge field. This
is precisely what happens in a superconductor where it is known as the Meissner-Ochsenfeld
effect and has as its most dramatic consequence magnetic levitation, but in free space the
photon remains massless13 . This given, the action for the electromagnetic field comprises only
the term 14 F F , to be integrated over the entire extent of the electromagnetic field, i.e.
all of space-time. Finally, there should also be a coupling between the gauge field A and the
4-current J = (c, J) describing the sources of the electromagnetic field. The action for the
electromagnetic field is then
1
Z
4
S= d x F F + 0 cA J . (1.6.16)
R1,3 4
Remark: One should check that the coupling between the gauge field and the 4-current is
compatible with gauge invariance. A gauge transformation alters the coupling term as follows
A J 7 A + J = A J + J J .
The second term is a total divergence and does not enter the equations of motion, so it can
be safely ignored. The third term vanishes by conservation of charge, J = 0. Thus we see
that conservation of charge and gauge invariance are closely linked.
13
It is the W and Z bosons that acquire mass.
19
The form of the action for the electromagnetic field established, the dynamical equations
of motion are given by the critical points of the action with respect to variations of the fields,
A . We compute directly
h
1
Z
4
A A A A
S[A + A] = d x
R1,3 4
i (1.6.17)
4 A A A + + 0 c A + A J ,
Z h i
= S[A] + d4 x A F + 0 cJ + O(2), (1.6.18)
R1,3
and therefore the field equations corresponding to critical points of the action are
F = 0 cJ . (1.6.19)
This is the covariant form of Maxwells first and fourth equations; they should be viewed as a
system of linear second order partial differential equations for the gauge field A. I leave it to
each of you to check that Maxwells equations in their traditional form really are reproduced;
it is a good exercise.
where n is the unit normal to the boundary of the region , dvol3 is the three-dimensional
volume element for that boundary surface, and we write L = 41 F F for the Lagrangian
as short-hand. On the other hand, writing that A (x + ) = A (x) + A + O(2 ), we also
have
Z
L h i L
S[A0 ] S[A] = A + A + O(2 ) d4 x, (1.6.21)
A ( A )
Z
L L L
= A + A d4 x + O(2 ),
A ( A ) ( A )
(1.6.22)
Z
L
= A d4 x + O(2 ), (1.6.23)
( A )
since the first term in the second line vanishes by the field equations. Equating these two
expressions for the difference in the action, it follows that
Z
L
A L d4 x = 0. (1.6.24)
( A )
20
It is usual to denote the expression in square brackets by T and call it the stress-energy-
momentum tensor. The sign is a convention associated to our choice of signature for the metric.
Then, as is arbitrary, it must be that it is conserved
T = 0. (1.6.25)
However, this is a little too rash and there is in fact a stronger statement. We will write instead
L
A L = T + B , (1.6.26)
( A )
with B antisymmetric in , , i.e. B = B 14 . The upshot is that we still have the
conservation law
T = 0, (1.6.27)
but in addition the tensor T = T can always be chosen to be symmetric15 , i.e. T = T .
It is this symmetric, conserved quantity that is the stress-energy-momentum tensor.
For the electromagnetic field, a direct calculation gives
L 1
L = A F + F F ,
A (1.6.28)
( A ) 4
1
= F F A F + F F ,
(1.6.29)
4
1
= F F A F + A F + F F . (1.6.30)
4
By the field equations (for the Lagrangian 14 F F ) the term F vanishes and we can
then identify the stress-energy-momentum tensor as
1
T = F F F F , (1.6.31)
4
1
= F F F F . (1.6.32)
4
It is evidently symmetric and also traceless, T = T = 0. To get a feeling for what
it represents, we write the components in terms of the electric and magnetic fields, using
Fi0 = Ei and Fij = ijk cBk ,
1 2
E + c2 B 2 ,
T00 = (1.6.33)
2
T0i = ijk Ej cBk , (1.6.34)
1
Tij = Ei Ej c2 Bi Bj + ij E 2 + c2 B 2 .
(1.6.35)
2
These components have the following interpretation: T00 is the energy density in the field; T0i
is (minus) the energy flux of the field (it is the Poynting vector); and Tij is the stress in the
field. To see more, integrate the conservation law T = 0 over a region of the space-like
surface x0 = const, i.e. any region of space. We find, for the case = 0,
Z Z
3 1
t T00 + i T0i d3 x,
0= T0 d x = (1.6.36)
c
Z Z
1 1 2
E + c2 B 2 d3 x + ijk cEj Bk dAi .
= t (1.6.37)
c 2
14
It is known as the Belinfante-Rosenfeld tensor.
15
The rationale for using the freedom to ensure that the stress-energy-momentum tensor is symmetric comes
from the expression for conservation of angular momentum. We will say no more; the interested reader should
consult the excellent discussion in Landaus text.
21
This expresses precisely the conservation of energy for the electromagnetic field in any spatial
region.
Of course, there is nothing special about the electromagnetic field; all physical quantities
possess energy and fluxes of energy that they contribute to the stress-energy-momentum tensor
T . This tensor, that expresses the energy, momentum and internal stress content of the
matter fields acts as the source for the gravitational field in Einsteins theory of general
relativity, much the same way as charge acts as the source of the electromagnetic field in
Maxwells theory. It appears as the right-hand-side of the Einstein equations.
22
Problems
1. Solve the Kepler problem in Newtonian gravity: two bodies of masses M1 and M2 interacting
through their mutual gravitational attraction.
2. Estimate the perturbation to one planetary orbit caused by gravitational interaction with
another planet. If you have relatively little experience with perturbation theory, you might
find that this is difficult. And yet it was calculated by Newton: in Book III, Proposition XIV
Theorem XIV of his Principia he writes [I]f the aphelion of Mars, in the space of a hundred
years, is carried 330 2000 in consequentia, in respect of the fixed stars, the aphelions of the
Earth, of Venus, and of Mercury, will in a hundred years be carried forwards 170 4000 , 100 5300 ,
and 40 1600 , respectively. But these motions are so inconsiderable, that we have neglected
them in this Proposition. You might like to compare your calculations with Newtons.
[I wish to be clear about this problem: it does not in any way represent the sort of thing that will
appear in an exam paper. At the same time, it is precisely the sort of thing that is of foremost
importance in giving an accurate description of the motion of the planets and other celestial bodies.
Thus, and in keeping with my philosophy, the question is about physics and not exam questions. It
should also serve to illustrate the magnitude of what Newton achieved; Einstein too.]
3. Prove Newtons result that the gravitational interaction with a spherically symmetric mass
distribution is the same as that with a point object of the same total mass located at the
centre of the sphere. [If you need any help, consult Newtons Principia, Book I, Proposition
LXXI Theorem XXXI.]
4. The diagram below shows the design of a perpetual motion machine sent by an inventor
called Hermann Bondi to a patent office in Switzerland where a certain Albert Einstein
works: Atoms attached to a movable belt absorb photons at the top of their travel and
emit them at the bottom, with the photons directed back to the top-most atoms where
they are absorbed. When an atom absorbs a photon of energy E = ~ its mass increases
by m = ~/c2 , hence the excited atoms (filled circles) on the right of the belt always
outweigh the de-excited atoms (empty circles) on the left and hence perpetual clockwise
motion results which can solve all the worlds energy problems.
What flaw does Einstein spot in this madness?
5. In a frame K a photon travels in the x1 x2 -plane at an angle with the x1 -axis. Show that
in a frame K 0 , moving relatively to K with speed v along the x1 -direction, that the same
photon makes an angle 0 relative to the x0 1 -axis, where
cos()
cos 0 =
.
1 cos()
23
6. From the following set of expressions involving tensor components, identify those that are
meaningless:
A + B , R A + B , R = S , T = F , g g R .
7. Write out the equation = 0 in Cartesian coordinates (ct, x, y, z). How does the
equation change under a Lorentz transformation x 7 x0 = x ?
Show that the equation 2 2
~ c m2 c4 = 0,
j t)
has plane-wave solutions ei(kj x with (~)2 = ~2 c2 kj k j + m2 c4 .
9. Show that the subset of Minkowski that is unit space-like distance from the origin
is a time-like hypersurface.
10. Let p and q be two points of the 2+1-dimensional Minkowski space R1,2 that are space-like
separated. Argue (or simply convince yourself) that there exists a frame in which they have
coordinates (0, a, 0) and (0, a, 0) for some number a.
Consider the two light cones through these points, p and q. Show that their intersection is
a hyperbola. Show that it is a hyperbola for any inertial observer, i.e. consider how things
change under Lorentz transformations. Sketch the situation.
Now suppose p and q are time-like separated. By running a parallel analysis, show that
the light cones through p and q intersect in an ellipse and determine how the semi-major
and semi-minor axes depend on the invariant interval between p and q and on the Lorentz
transformation relating different inertial frames in which it is viewed. Sketch the situation.
[This example illustrates a way of understanding the classical conic sections as intersections
of light cones. You might like to think of a way of understanding parabolae in this fash-
ion. The generalisation to higher dimensions leads to a description of the beautiful Dupin
cyclides, which arise in smectic liquid crystals as the spectacular focal conic textures, and
also to the subject of Lie sphere geometry.]
11. Let A = (, cA) be the electromagnetic gauge field and F = A A the Maxwell
field strength tensor. Show that Ei = Fi0 is the electric field and cBi = 12 ijk Fjk is the
magnetic field. What are the components of the contravariant form of the Maxwell tensor,
F = F ? Express the Lorentz scalar F F in terms of the electric and magnetic
fields.
Let be the fully antisymmetric Levi-Civita symbol with 0123 = +1. The dual field
strength tensor is defined to be the antisymmetric type ( 02 ) tensor (2-form) with components
?F = 21 F . Express its components in terms of the electric and magnetic fields.
Express the pseudoscalar ?F F in terms of the electric and magnetic fields.
24
12. Show that the equations
F + F + F = 0, F = 0 c J ,
are equivalent to the four Maxwell equations when written in terms of the 3-component
vectors E and B.
13. A charged particle accelerates from rest in a uniform electric field. There is no magnetic
field. Describe its motion. Show that its trajectory is a time-like curve.
14. Find the stress-energy-momentum tensor for a scalar Klein-Gordon field for which the
action is Z h
1 2 2 i
S[] = ~ c + m2 c4 2 d4 x.
2
Show that the components T00 and T0i have their usual interpretation as the energy density
and momentum density, respectively.
R = R , R = R , R = R .
R + R + R = 0,
linearly independent constraints. Hence show that the total number of linearly indepen-
1 2 2
dent components of the tensor is 12 n (n 1). The Riemann curvature tensor has these
properties; this count gives the number of linearly independent curvatures in n dimensions.
25
Some examples of smooth manifolds for you to think about. Try to supply a variety of others
for yourself.
26
Chapter 2
Differential Geometry
People have very powerful facilities for taking in information visually or kinesthetically, and
thinking with their spatial sense. On the other hand, they do not have a very good built-
in facility for inverse vision, that is, turning an internal spatial understanding back into
a two-dimensional image. Consequently, mathematicians usually have fewer and poorer
figures in their papers than in their heads.
General relativity is a theory about the structure and geometry of space-time. There are
many different types of space and their study is a fascinating subject. It separates itself into
two main aspects; geometry and topology. In some ways they are very deeply intertwined.
It might surprise you, then, to be reminded that geometry dates back at least as far as the
ancient Greeks, while topology is essentially a 20th century conception1 . The disparity is very
much less severe if one recalls that a serious definition of curvature is only due to Bernhard
Riemann in 1854. In this course, we will focus on geometry and have more or less nothing to
say about topology.
Geometry is the study of lengths, areas, volumes and so forth, angles, parallelism and
curvature. For us, the aim is to understand what it means to say that space-time is curved,
how one describes this curvature and how one detects it through physical phenomena. Let
us be direct: we will really need to know two things in this course; what a geodesic is, and
what curvature is. A geodesic is a curve of shortest length between two given points. It is the
trajectory that a test particle travels along Newtons first law. Curvature is the following
thing. A small circle of geodesic radius r has a circumference C(r) that differs from 2r. The
difference gives a measure of the curvature of the space. It depends on the choice of circle.
In fact, there are several notions of curvature, which all turn out to be equivalent; this one is
called sectional curvature. It is an aim of mine in this chapter to give you a pedagogic account
of Riemanns immense achievement in properly understanding curvature.
2.1 Manifolds
There are many different types of space; a variety of examples are shown in the preceeding
plate. We shall confine our attention to those that are known as smooth manifolds. To see what
this is, it suffices to consider a description of the space that we live in, for it is an example;
the surface of the Earth is a sphere. One way to describe it is to divide it into four charts; the
Antarctic chart, the North American chart, the Chinese chart, and the European chart. Each
chart is a piece of paper that contains a detailed depiction of that part of the Earth that the
1
Although there were several prominent forerunners, most people acknowledge Henri Poincares 1895 paper
Analysis Situs as being foundational for topology in its modern sense. (I learn from Bob MacPherson that all
the features of homology theory can be found in Riemanns unpublished notes.)
27
Figure 2.1: Definition of a manifold in terms of local charts.
chart is named after. England is in the European chart and Australia in the Antarctic chart.
If you wish to visit Australia you can use the European chart to take you until there is overlap
with the coverage of the Antarctic chart at which point you switch to that and complete your
journey. Thats the basic structure; lets repeat what we have just said in technical language.
A manifold M is a space that locally looks like Rn for some n, called the dimension of
the space. What this means is that every point p of the space is contained in a little region
U such that there is a 1-to-1 smooth function, with smooth inverse, between the points of U
and the points x of an open subset V of Rn . We write the function as : U V and call
it a local chart or local coordinate system. The inverse construction 1 : V U is called a
parameterisation for a neighbourhood of the point p. A collection of local charts a : Ua Va
that covers every point of M is called an atlas. If a point p lies in two charts, a and b , then
one has the obvious transition function
b 1
a : Va Vb , (2.1.1)
which is required to be smooth and have smooth inverse. The collection of triples {(a , Ua , Va )},
with the stated properties, is the definition of a manifold2 . If the manifold also comes with
something that allows you to measure distances, called a metric, then it is a Riemannian
manifold. The study of Riemannian manifolds and their properties is what is meant by (Rie-
mannian) geometry.
Often the sphere is presented differently, as the subset
of Euclidean R3 . This is known as an embedding, a technical term which states that the sphere
sits inside the higher-dimensional space with no self-intersections. This situation gives rise to
another definition of a manifold. A manifold M is a smooth subset of RN , for some (large)
N , such that every point has a neighbourhood diffeomorphic3 to an open subset of Rn . This
2
Strictly, there is also a notion of equivalence to introduce, which I will not do.
3
The word diffeomorphic is a short-hand for what we wrote previously; there exists a smooth 1-to-1 function,
with smooth inverse, between the two spaces.
28
is not a bad definition; in fact it is the definition given by Milnor4 in his celebrated book
Topology from the Differentiable Viewpoint. A central result in the theory of manifolds is that
the two definitions are equivalent. This is known as the Whitney embedding theorem. More
is true; an embedding always exists that is isometric a result of John Nash, known as the
isometric embedding theorems. Loosely, what this means is that any intrinsic geometry that
the manifold has is the same as that it acquires from sitting inside a Euclidean space. The
upshot of all this is that one is not losing anything by viewing any given manifold as being
embedded in a Euclidean space, as every manifold can be viewed in such a way. Enormous
pedagogy is gained by thinking about two-dimensional manifolds surfaces embedded in R3 .
You can make them and look at them with your own eyes. I encourage you to do so. Direct
visualisation conveys things in a different way from pure algebra. I, for one, find it endlessly
helpful.
hypoteneuse is the sum of the squares on the other two sides. It leads immediately to the
expression for the distance between two points, (x1 , x2 , x3 ) and (x1 + dx1 , x2 + dx2 , x3 + dx3 )
in Cartesian coordinates, as
This is called the Euclidean metric5 . This given, we turn to the measure of distances between
points of a surface embedded in R3 . The surface can be described explicitly through an
embedding function X(u1 , u2 ) listing the points of R3 where the surface is. For instance, for
the sphere we might write
X(u1 , u2 ) = R sin(u1 ) cos(u2 )e1 + R sin(u1 ) sin(u2 )e2 + R cos(u1 )e3 . (2.2.2)
The parameters u1 , u2 in this case are the standard polar angles. The displacement between
two nearby points on the surface, X(u1 , u2 ) and X(u1 + du1 , u2 + du2 ), is
29
and the length squared of an infinitesimal line segment connecting them is
ds2 = dX dX = |u1 X|2 (du1 )2 + 2u1 X u2 X du1 du2 + |u2 X|2 (du2 )2 ,
= g11 (du1 )2 + g12 du1 du2 + g21 du2 du1 + g22 (du2 )2 , (2.2.4)
i j
= gij du du .
The object ds2 is called the metric and the quantities gij are its components for the given
parameterisation (or choice of coordinates). Please note that g12 = g21 , i.e. the metric is
symmetric. The metric is the square of the distance between nearby points of a space: if the
points are labelled by coordinates ui and ui + dui , then the square of the distance between
them is gij dui duj . For our example of the sphere, a simple calculation gives
where the quantities gij are the components of the metric in these coordinates.
As a terse but instructive example, consider the 3-sphere; the set of points distance R from
the origin in R4
(x1 )2 + (x2 )2 + (x3 )2 + (x4 )2 = R2 . (2.2.7)
It can be parameterised in many ways. For instance, Cartesian coordinates about the point
(0, 0, 0, R) can be introduced by the parameterisation (u1 , u2 , u3 , R2 r2 ), where r2 = (u1 )2 +
(u2 )2 + (u3 )2 . Such a parameterisation shows that the space is locally Euclidean, to second
order accuracy in the ui . An equivalent, but perhaps more convenient, parameterisation is to
take p
x1 , x2 , x3 , x4 = r sin() cos(), r sin() sin(), r cos(), R2 r2 .
(2.2.8)
With this choice, a short calculation determines the metric to be
dr2
ds2 = + r 2
d 2
+ sin 2
() d 2
. (2.2.9)
1 r2 /R2
Let us take the opportunity to illustrate a few things with this example. First, the param-
eterisation we have given
covers only half the sphere, not the whole space. This is because
we have written x4 = R2 r2 , which is therefore constrained to be non-negative, with the
parameter r taking values in the range [0, R]. If not already known, this is something that
should be learnt well: parameterisations, or coordinate systems, or local charts, cover only
portions of a manifold, not the whole space. Second, the metric is singular when r = R;
grr = (1 r2 /R2 )1 is formally infinite. Yet, there is nothing weird about those points of the
sphere: the problem is with our coordinate system, not with the space. This is an example
of a coordinate singularity6 . It warns us not to jump to unfounded conclusions purely on the
basis of the behaviour of coordinates; we must look at the behaviour of physical observables.
Finally, if we write r = R sin() we can give the metric on S 3 in its usual form, known as the
round metric,
ds2 = R2 d2 + R2 sin2 () d2 + sin2 () d2 . (2.2.10)
One should note that these coordinates cover a larger portion of the sphere and that the metric
is no longer singular at = /2 (corresponding to r = R). It is worth repeating; if not known
already, this is a lesson that should be learnt well. In tautological terms, we might say that
the way that we see a space depends on the way that we view it.
6
The eagle-eyed among you should observe that there is also a coordinate singularity at r = 0; there is
nothing wrong with that point either.
30
2.2.1 Lorentzian Metrics
The metric for Minkowski space-time
is fundamentally different from the metric for a Euclidean space; the difference is the minus
sign in front of the time-direction (and all that comes with it). The difference between the
number of space-like directions and time-like directions is called the signature of the metric.
If there is only one time-like dimension the metric is called Lorentzian the signature is then
n2. Space-time is a Lorentzian manifold. In general, we will write the metric for a Lorentzian
manifold as
ds2 = g dx dx , (2.2.12)
where x are coordinates in a local chart. The components g are symmetric, g = g , so
that if we think of them as the entries in a matrix then the matrix can be diagonalised at any
point of the manifold. That it is Lorentzian means that one of the diagonal entries is negative
and the other three positive. We will think of the negative entry as the time direction and
conventionally take it to be the component g00 .
The Lorentzian version of a sphere serves as a nice example. Lorentzian spheres come in
two flavours, space-like and time-like7 ; we describe only the former. The set of points a fixed
space-like distance from the origin in R1,4 ,
to second order accuracy in the local coordinates x ( = 0, 1, 2, 3). This result is general in
two respects. First, the same is true at every point of the de Sitter space, a consequence of its
homogeneity8 . Second, it is true at every point of every Riemannian, or pseudo-Riemannian,
manifold, although we will not prove this result. Coordinates in which the metric is Minkowski
to second order accuracy at a given point are called Riemann normal coordinates. In general
relativity they are known as local inertial frames9 .
Remark: Local inertial frames, or Riemann normal coordinates, are often used extensively
in the development of general relativity because Einstein did so. Because I wish to describe
things more intrinsically, defining concepts without reference to any coordinates at all, they
do not appear as prominently in these lecture notes as in many other textbooks. Of course,
the point of particular coordinate systems is to make calculations simple and in this regard
local inertial frames, or Riemann normal coordinates, are often invaluable.
7
They can also be null, which correspond to light-cones in the embedding space.
8
The easiest way to establish this is to note that there is a transitive action of the group SO(1, 4) corre-
sponding to the symmetries of the space.
9
Note that they are not unique; the Lorentz group SO(1, 3) acts (transitively) as a change of observer that
preserves the local inertial character of the coordinate system.
31
Figure 2.3: The two-dimensional de Sitter space dS2 can be visualised directly as a hyperboloid
in the Minkowski space R1,2 .
x0 = R sinh( ),
x1 = R cosh( ) sin() sin() cos(),
x2 = R cosh( ) sin() sin() sin(), (2.2.16)
3
x = R cosh( ) sin() cos(),
x4 = R cosh( ) cos(),
for any real number and , , the three standard spherical angles on S 3 10 . In these
coordinates the metric assumes the form
h i
ds2 = R2 d 2 + R2 cosh2 ( ) d2 + sin2 () d2 + sin2 () d2 . (2.2.17)
is a time-like coordinate. For any fixed value of , space is a 3-sphere with radius R cosh( )
that depends on the value of the time coordinate. So the picture we have is the following:
when is negative, but increasing, the space is a contracting 3-sphere; it reaches a minimum
size R when = 0 and thereafter expands again for positive values of . The picture of a
hyperboloid in three-dimensional Minkowski (dS2 ) conveys this quite nicely. An entertaining
problem is to imagine how we would recognise it from observations if the universe that we live
in turned out to be de Sitter space.
2.3 Geodesics
A geodesic is a curve of shortest length between two points. Their fundamental significance
in general relativity is that test particles move along geodesics. This is Newtons first law
of motion a particle moves along a straight line at constant speed unless acted upon by
an external force the only change being to recognise that straight line means the same
as geodesic. We first describe some basic properties of curves before deriving the equation
obeyed by geodesics and hence that characterises the motion of test particles.
Suppose is a curve in our surface. We might describe it explicitly by giving an embedding
function c( ), listing the points of R3 corresponding to the curve. It is clearly equivalent to
specify the ui , that parameterise our surface, as functions of , which parameterises our curve,
10
From this we see that the global topology of de Sitter space-time is R S 3 , which is different from that of
Minkowski, whose topology is R R3 .
32
since X(u1 ( ), u2 ( )) = c( ). Distance along the curve is given by either of
2
dc
ds = dc dc = d 2 ,
2
d
(2.3.1)
du i duj
ds2 = gij dui duj = gij d 2 .
d d
It is routine to verify that these are equivalent, as they must be. The length of the curve is
Z final r
dui duj
Z
ds = gij d. (2.3.2)
init d d
The same description, and formulae, applies mutatis mutandis for a curve in a manifold of any
dimension.
In giving an explicit description of our curve we introduced a parameter that labelled
its points. There is complete freedom over the choice of how we do this; any parameterisation
can be used. There is one, however, that is natural, for it has physical significance. That is to
parameterise by arc length, i.e. choose = s, so that the points are labelled by their actual
physical distance along the curve11 . With such a choice the tangent vector to the curve, dc/d
or dui /d , is properly normalised and has unit magnitude
2
dc dui duj
= 1,
d g ij = 1. (2.3.3)
d d
This parameterisation proves convenient in many calculations.
Let p and q be two points on our manifold and a curve connecting them. We can measure
the length of , the distance between p and q along the given curve. If 0 is another curve that
also connects p and q then we can measure its length too and ask if that distance is greater
or less than the distance along . We see that the distance between p and q depends on the
choice of curve connecting the two points; it is a function whose argument is a curve. A curve
that is a critical point of this distance functional is called a geodesic. It is a curve whose length
does not change to first order for any small change to the curve. Informally, it is the curve
that represents the shortest distance between p and q.
Let us give a local characterisation of a geodesic curve . In a local chart it can be expressed
as ui ( ), where is parameterisation by arc length. Then if 0 is a curve close to we can
describe it locally by the parameterisation ui ( ) + i ( ), where the i are all small and vanish
at the endpoints, since 0 also passes through the points p and q. We say that 0 is a variation
of the curve . Let us compute its length. The metric for 0 is
dui di duj dj
ds2 = gij 0 + + d 2 , (2.3.4)
d d d d
i j
dui dj di duj
k du du
= gij + k gij + O(2) + + + O(2) d 2 , (2.3.5)
d d d d d d
dui duj k dui dj
= 1 + k gij + 2gij + O(2) d 2 , (2.3.6)
d d d d
11
We use this choice on all of our road systems, as the helpful roadside signs tell us.
33
using the fact that is parameterised by arc length, the symmetry of gij and a Taylor expansion
of gij | 0 about the corresponding points of . It follows that the length of 0 is
final 1/2
dui duj k dui dj
Z Z
ds = 1 + k gij + 2gij + O(2) d, (2.3.7)
0 init d d d d
Z final
dui duj k dui dk
1
= 1 + k gij + gik + O(2) d, (2.3.8)
init 2 d d d d
Z final
dui duj dui
Z
1 d
= ds + k gij gik k d + O(2), (2.3.9)
init 2 d d d d
using the fact that the k all vanish at the endpoints. Since the k are otherwise arbitrary, the
condition for to be a curve of extremal length (critical point of the distance functional) is
that the integrand vanish12 . That is
d2 ui dui duj 1 dui duj
gik + j g ik k g ij = 0, (2.3.10)
d 2 d d 2 d d
d2 ui 1 h i dui duj
gik 2 + i gjk + j gik k gij = 0, (2.3.11)
d 2 d d
d2 ui 1 il h i duj duk
+ g j g lk + k g jl l gjk = 0, (2.3.12)
d 2 2 d d
where in the last step we have used the inverse metric g il and relabelled dummy indices. The
quantities that appear in this expression
1 h i
ijk = g il j glk + k gjl l gjk , (2.3.13)
2
are important; they are called the Christoffel symbols. To summarise, we have found that if
is a geodesic then it has a local parameterisation ui ( ) that satisfies the geodesic equation
d2 ui j
i du du
k
+ jk = 0. (2.3.14)
d 2 d d
In deriving this equation we made use of the fact that was parameterised by arc length,
meaning that its tangent vector has unit magnitude
dui duj
gij = 1. (2.3.15)
d d
The geodesic equation (2.3.14) is a system of second order ordinary differential equations.
Standard existence and uniqueness results in the theory of odes then tell us that if we specify
an initial point and tangent vector, there is an unique geodesic passing through that point
with the given initial tangent vector. Similarly, for points that are close enough together there
is an unique geodesic connecting them.
Of course, we are primarily interested in geodesics in a Lorentzian manifold. They are
almost entirely the same. If x ( ) is a local parameterisation of a geodesic then it satisfies the
geodesic equation
d2 x
dx dx
+ = 0, (2.3.16)
d 2 d d
with the Christoffel symbols given in terms of the components of the metric by the same
formula as above. The only difference is that the geodesic may be space-like, time-like, or null.
Particles with non-zero mass travel along time-like geodesics. What this means is that their
tangent vectors are time-like
dx dx
g = 1. (2.3.17)
d d
12
This is known as the fundamental lemma of the calculus of variations.
34
That the normalisation is 1 reflects the fact that the geodesic is parameterised by proper time
(actually proper time times the speed of light), which has been the motivation for using the
symbol . Massless particles, such as light, still satisfy the geodesic equation (2.3.16) but their
tangent vector is null
dx dx
g = 0. (2.3.18)
d d
But the magnitude of the tangent vector aside, the description of geodesics in Lorentzian
manifolds is the same as in the Riemannian setting.
dui1 duj2
dc1 dc2
cos() = , or cos() = gij , (2.4.1)
d1 d2 p d1 d2 p
where p is the point at which they cross. If they are not parameterised by arc length then
both expressions need to be divided by the magnitude of the tangent vectors to each curve at
the point p. Equivalently, if Y, Z are vectors at p then Y Z = |Y|.|Z|. cos() and an intrinsic
expression for the same formula is
q p
gij Y i Z j = gij Y i Y j . gkl Z k Z l . cos(). (2.4.2)
Think of a surface embedded in R3 ; the sphere will do. We can measure the area of a
patch of its surface. Consider a small patch of the surface with sides of length u1 X du1 and
u2 X du2 . Then it is well-known that the area of this patch can be expressed
35
The ideas presented here for surfaces in R3 apply mutatis mutandis to arbitrary n-dimensional
manifolds embedded in some Euclidean space RN . The n-dimensional volume element can be
written p
dvol = det g du1 dun , (2.4.6)
in terms of local coordinates for the manifold. To get at least a correct feeling for this, note
that the formula is certainly correct if the metric is diagonal.
In the Lorentzian case, the presence of a time-like direction minus sign in the metric
means that the determinant det g is negative. In recognition of this, there is a minor change
to how we write the volume element
p
dvol = | det g| dx0 dx1 dx2 dx3 , (2.4.7)
where, as usual, the x are local coordinates. This aside, there is no difference. This formula,
(2.4.7), for the volume element in general coordinates should be remembered; it will be needed.
dui
dc
= Zp , or = Zpi , (2.5.1)
d =0 d =0
to define the vector Zp . I will try to write consistently Zp without boldface when I am trying
to think of the manifold intrinsically and Zp with boldface if I am thinking of the same vector
from the perspective of some embedding space. The set of all possible tangents to curves
passing through p forms the tangent plane Tp M to the manifold at p. It is a linear space with
dimension n the same as the manifold M . A vector at p is any element of Tp M , an object with
magnitude and direction. The collection of all tangent planes for every point of the manifold
is called the tangent bundle T M . A choice of vector at every point p, varying smoothly with
p, is called a vector field. We will write it Z. The (surface) fluid velocity in the oceans of the
Earth, or the breeze of the air over the entire Earths surface are highly instructive examples
of vector fields on the sphere that are surely well-known to you.
Figure 2.4: Tangent plane to the sphere at a general point p. Coordinate curves are shown, as
well as the tangent vectors to these curves at p. Any vector at p can be expressed as a linear
combination of these two.
36
Let ui be coordinates for a local chart in the neighbourhood of a point p on some surface, say
with embedding function X(u1 , u2 ). The curves X(, 0) and X(0, ) can be called coordinate
curves passing through p. Their tangent vectors at p, u1 X and u2 X, form a basis for the
tangent space Tp M , since if c( ) is any curve passing through p then its tangent is
u1 u2 ui
dc
= 1 X + 2 X = i X. (2.5.2)
d =0 u u u
Evidently this is true in general, for any manifold; the vectors ui X that are tangent to
coordinate curves form a basis for the tangent space at each point of the local chart covered
by the coordinates. It is universal to denote these basis vectors by i when they are viewed
intrinsically, without reference to any embedding space. Although the symbol is the same, and
it will be treated much the same in many calculations, it is not the partial derivative; it is the
tangent vector to the coordinate curve ui . This notation is universal and has to be acclimatised
to. Thus any vector Zp can be expressed in a coordinate basis as
Zp = Zpi i , (2.5.3)
where the numbers Zpi are the components of the vector in the coordinate basis.
A function f is a rule that assigns a real number to every point p of a manifold, and varies
smoothly with p. We wish to convey what is meant by the derivative of a function. A good
question to ask is: in which direction? Given a direction Zp there is a curve c( ) passing
through p whose tangent vector at that point is Zp . We may then compare the values of the
function f as we move along the curve c and compute the derivative
df
, (2.5.4)
d =0
known as the directional derivative of f at the point p, in the direction Zp . Usually this is
denoted Zp (f ). The derivative of f , therefore, is an object that operates on vectors to obtain
directional derivatives. We denote it by
df : Tp M R
(2.5.5)
Zp df (Zp ) = Zp (f )
It is a linear map from the tangent space at p to the real line; such things are called 1-forms.
They form a linear vector space dual to the tangent space, called the cotangent space and
denoted Tp M . The collection of cotangent spaces for every point p of the manifold is called
the cotangent bundle. In addition to the derivative of a function, the electromagnetic gauge
field A is a commonly encountered example of a 1-form; as are the reciprocal lattice vectors
in crystallography; and in classical mechanics one learns that momentum should be properly
viewed as a 1-form.
In any local chart the coordinates ui are functions. Their derivatives dui are therefore
1-forms. Although the symbol is the same, and in many calculations they will be treated much
the same, these are not integration measures; they are the derivatives of coordinate functions.
This notation is universal and has to be acclimatised to. The 1-forms dui form a basis for the
cotangent space, for
f
df = dui . (2.5.6)
ui
This basis is dual to the coordinate basis i for the tangent space, since by unfolding the
various definitions
ui
dui j = j (ui ) = = ji .
(2.5.7)
uj
It is a good exercise to go through this result for yourself. In a coordinate basis any 1-form
may be written i dui , where i are the components of the 1-form in the coordinate basis.
37
All this is exactly the same in a Lorentzian manifold. If x is a local coordinate system then
is a basis for the tangent space and dx a basis for the cotangent space. These bases are
dual dx ( ) = . The geometric meaning is exactly the same and the only noticeable change
is the use of Greek indices rather than Latin ones13 . This given, we can define a general tensor.
In physics, a tensor is a natural physical quantity14 that generalises the concept of a vector
by exhibiting multilinearity. Examples include the stress tensor in a fluid, the dielectric tensor
in electromagnetism, the Maxwell field strength tensor, the conductivity tensor (electrical or
thermal) and the elasticity tensor in an elastic solid. Mathematically, a type ( kl ) tensor is a
multilinear map that at each point p of a manifold takes k 1-forms and l vectors and returns
a number, that number varying smoothly with p15
T : Tp M Tp M Tp M Tp M R
| {z } | {z }
k times l times (2.5.8)
p p Yp Zp T (p , , p , Yp , , Zp ).
for any three vectors Xp , Yp , Zp and function f . Multilinearity of a tensor means the same
thing; it is linear with respect to each object it acts on.
In traditional presentations in the physics literature the discussion of tensors is long and
centred around their components in a coordinate basis. I find that it is tiresome and does
not convey any physical content or understanding. Take any of the examples that I gave, say
conductivity17 . A potential difference is applied to a material so that it experiences an electric
field. The response, for small applied fields, is that a current flows through the material.
Conductivity is the material property that converts the applied electric field into the current
response; symbolically J = E. It acts on a vector (E) and returns a vector (J). So long as
the applied electric field is small the response is linear; double the electric field and the current
response doubles; superpose two electric fields to create a third and the current response is the
linear superposition of the response to each separate field. Thus the conductivity is a linear
map from a vector to another vector, or from a vector space to itself. It is naturally defined as
a material property entirely independently of any choice of coordinate system or basis for the
vector space. This is what is meant, in physics at least, by a type ( 11 ) tensor. The same holds
for anything that is a tensor, of any type. To my mind it is the physics that is important and
13
We should remember, or course, that in the Lorentzian setting tangent vectors can be space-like, time-like,
or null.
14
Not every physical quantity is a tensor; spin 12 particles like the electron are not.
15
When I first started as a postdoc, my supervisor came into my office and exclaimed: Hey, Gareth! Do you
know that tensors are toasters? It took about 5 seconds to parse what he had said before I replied: Yes, of
course I know that.
16
That Zp is the tangent means dx /d = Zp .
17
The quantum Hall effect is a beautiful experiment; think of it.
38
should be emphasised first and foremost. If you understand what the quantity is and what
it means to measure it; if you know what it depends on and what it influences; if you can
describe how changes in the experimental set-up affect the measurement that is made; if you
can explain all this to a fellow student, then you understand what a tensor is. If you cannot,
then you dont. Knowledge of raising and lowering of indices, contractions, outer products
and transformation laws for components under a change of coordinates seem of pale secondary
importance compared to genuine physical understanding.
For the vast majority of physical applications of general relativity only three tensors ever
appear; the metric, the stress-energy-momentum tensor, and the Ricci tensor. The latter
gives a characterisation of the curvature of space-time. I will try, in the next section, to
give a natural geometric description of curvature that emphasises how it manifests itself in
physical observations and measurements; the traditional tensor-calculus properties will fall out
by themselves, and, in my opinion, in the proper fashion.
2.6 Curvature
In an attempt to introduce curvature in a pedagogic way and try to promote geometric ways
of thinking (prior to the deduction of algebraic formulae), I want to describe four concepts of
curvature, each of them illustrated with the humble/venerable sphere. These four concepts
are: sectional curvature associated to the circumference of a circle of small geodesic radius;
Gaussian curvature associated to surfaces embedded in R3 ; geodesic deviation, or the failure
of parallel straight lines to maintain a fixed separation; and the rotation of a vector under
parallel transport. We begin with sectional curvature.
Take any point p on a surface and think of the geodesics passing through that point. If a
collection of small test particles should move distance r along those geodesics they will form
a small circle, of geodesic radius r. Geodesics on the sphere are great circles the intersection
of the sphere with planes through the origin in R3 . In standard polar coordinates the lines
of longitude are all geodesics passing through the north pole (and south pole). They can be
parameterised as
R sin(r/R) cos()e1 + sin()e2 + R cos(r/R)e3 , (2.6.1)
where r is arc length along the geodesic making angle with the Greenwich meridian. Fixing
r, arbitrarily small, and letting vary through a full 2 sweeps out a circle of geodesic radius
r. Its circumference is
Z 2 h 1 1 2 4
i
C(r) = R sin(r/R) d = 2R sin(r/R) = 2r 1 r + O(r ) . (2.6.2)
0 6 R2
The deviation of the circumference from the value 2r detects the curvature of the sphere18 .
18
Indeed, on the Earth (radius 6371 km) for a circle of geodesic radius 11 km (distance between the University
of Warwick and Leamington Spa) the shortfall is 3.4 cm.
39
This is illustrative of a general formula
h 1 i
C(r) = 2r 1 K r2 + O(r3 ) , (2.6.3)
6
for any such circle of geodesic radius r in a manifold of any dimension. The number K is
called a sectional curvature. It depends on the choice of circle as follows. Each point of the
circumference is connected to the centre, p, by an unique geodesic, which we can label by
its tangent vector at p. The set of all such tangent vectors associated to the circle define a
two-plane in the tangent space at p. Vice-versa, any choice of two-plane in the tangent space
at p allows us to construct a circle. The sectional curvature K depends on this two-plane.
The collection of all sectional curvatures, for every point of the manifold, gives a complete
characterisation of the curvature of the space.
By way of a simple example, geodesics on the 3-sphere, passing through the point (0, 0, 0, R),
can be parameterised as
R sin(r/R) n + R cos(r/R) e4 , (2.6.4)
where n is a unit vector orthogonal to e4 , i.e. a point on the 2-sphere. Any plane in the
tangent space at (0, 0, 0, R) intersects that 2-sphere in a great circle. If we parameterise n
by spherical polar angles such that the intersection is the equator, then the calculation of
the sectional curvature of the 2-plane is identical to the one we have just given, showing that
K = 1/R2 for every choice of 2-plane. The result also holds for every point of the 3-sphere;
there is nothing special about (0, 0, 0, R). It is not hard to see that the same analysis holds
for S n . Thus the sphere is a space of constant positive curvature. In fact this is an unique
characteristic; no other manifold has the same property.
The second concept of curvature I wish to describe is the Gaussian curvature, defined for
surfaces embedded in R3 . Let p be a point on such a surface and consider the direction that is
normal to the surface at p. This is called the Gauss map; an association to each point of the
surface of a point on S 2 corresponding to the direction of the surface normal at that point.
Now consider a small patch of the surface centred at p. To each point of this little patch we
associate its corresponding point on the 2-sphere under the Gauss map, its spherical image.
The ratio of the area of the spherical image19 to the area of the patch of surface, in the limit
where the size of the patch shrinks to nothing, is taken to define the Gaussian curvature
area spherical image
KG = lim . (2.6.5)
area0 area patch of surface
Gauss was so enamoured by his proof that KG = K that he called it the theorema egregium20
in modern parlance this theorem is sick! The demonstration for the sphere is elementary.
For a sphere of unit radius the Gauss map is the identity and KG = 1. By simple scaling
KG = 1/R2 for the sphere of radius R. You are challenged to find for yourself a general proof
for any surface embedded in R3 .
The third concept of curvature I wish to describe is geodesic deviation. The equator is a
great circle on the Earth. Pick two points on it, close to each other. The lines of longitude
through those two points are also great circles both of which meet the equator at right angles.
They are parallel at the equator. But this is not so if you move along them away from
the equator, either north or south. In both directions they bend towards each other and
eventually meet at the two poles. The failure of initially parallel geodesics to remain parallel,
or to maintain a fixed separation, is a signature of curvature, known as geodesic deviation. The
analogous convergence, or divergence, of test particles moving along geodesics of the space-
time metric in general relativity are known as tidal forces. They provide one means, at least
19
The area of the spherical image may be figured positive if the Gauss map in the neighbourhood of p is
orientation preserving and negative if it is orientation reversing.
20
The theorem is not usually stated as I have implied, but its content is the same; the Gaussian curvature
can be computed just using the metric.
40
in principle, of measuring curvature experimentally, and hence the metric, by observing the
motion of test particles. Let us describe the geodesic deviation for the sphere. Let the two
geodesics initially tangent at the equator be parameterised as
R cos(r/R) cos()e1 sin()e2 + R sin(r/R)e3 , (2.6.6)
where r is arc length along the geodesic measured from the equator. The geodesic connecting
corresponding points of these two lines of longitude is
R cos(s/R) cos()e1 + sin()e3 + R sin(s/R)e2 , (2.6.7)
where s is arc length along the geodesic measured from the point of latitude where it crosses
the Greenwich meridian. The geodesic separation between corresponding points of our lines of
longitude is therefore
s = 2R arcsin cos(r/R) sin() . (2.6.8)
If the initial separation s0 is very small, 1, then this is approximately
41
The Riemann curvature is a sophisticated thing. Let us compute it for the 2-sphere. Choose
for p the north pole (0, 0, R) and for Zp the vector e1 of R3 tangent to the sphere at the north
pole. A circle of geodesic radius r centred at the north pole can be described by the explicit
embedding function
c() = R sin(r/R) cos()e1 + sin()e2 + R cos(r/R)e3 . (2.6.11)
To get things started we first have to parallel transport the vector Zp from the north pole
to the circle. If we do it along the Greenwich meridian we will obtain the vector Z(0) =
cos(r/R)e1 sin(r/R)e3 . To describe its transport around the circle we introduce a set of
orthonormal basis vectors (for R3 ). The first choice you might think of are the spherical
polars
e = cos(r/R) cos()e1 + sin()e2 sin(r/R)e3 , (2.6.12)
e = sin()e1 + cos()e2 , (2.6.13)
n = sin(r/R) cos()e1 + sin()e2 + cos(r/R)e3 . (2.6.14)
This choice can be used and you will compute the correct value for the change in Zp under
parallel transport, however, the vectors e and e are not well-defined at the north pole (r = 0)
itself. It is better to use a basis without such a defect. One choice is to use
in their place. This basis is still orthonormal. In terms of it we can write the parallel transport
of Z(0) as
Z() = cos () s1 + sin () s2 . (2.6.17)
It remains to determine the function (). A direct calculation gives how Z() changes
dZ d d
= + 1 cos(r/R) sin() s1 + 1 cos(r/R) cos() s2
d d d (2.6.18)
h i
+ cos() sin() sin() cos() sin(r/R) n,
Finally, we need to parallel transport this back to the north pole, along the Greenwich meridian,
which produces the vector
1
cos 4 sin2 (r/2R) e1 + sin 4 sin2 (r/2R) e2 = e1 + r2 2 e2 + O(r4 ).
(2.6.21)
R
As promised, there is a change, proportional to r2 , with a coefficient that measures the
curvature of the sphere.
Remark: The rotation of a vector under parallel transport around a line of latitude on a
sphere is something that you can observe directly. It is the rotation of the plane of oscillation
of a Foucault pendulum; transport around a line of latitude is provided by the rotation of
the Earth. Hopefully this both piques your interest in Foucault pendulums and helps you to
appreciate what curvature means in concrete physical terms for at least one situation.
42
Taking what we have learnt we can now describe the general situation for any manifold. The
change in Zp depends on the choice of circle to transport it around. As described previously,
any such circle can be identified with a 2-plane in the tangent space at p. Any two orthogonal
unit vectors Xp , Yp define a 2-plane at p, so it is just the same to select any two such vectors.
The change in the vector Zp under parallel transport around a circle of geodesic radius r,
defined by the vectors Xp , Yp , can be written as
with R(Xp , Yp )Zp being a vector in the tangent space at p. It is linear in each of Xp , Yp , Zp
and so defines a type ( 13 ) tensor at p. This is the Riemann curvature tensor. The minus sign
that appears in the definition is chosen so that subsequent formulae have their conventional
signs. The Riemann tensor is antisymmetric in Xp , Yp since interchange of these reverses
the orientation of the 2-plane they define and hence the direction in which Zp is parallel
transported.
Remark: For any choice of circle to transport around, the Riemann curvature is a linear
transformation on the tangent space at p, defined by R(Xp , Yp ) : Zp R(Xp , Yp )Zp . This
linear transformation depends on two vectors Xp , Yp and is antisymmetric with respect to
interchange of them, R(Xp , Yp ) = R(Yp , Xp ). Therefore, the Riemann curvature is a 2-form
(a skew-symmetric bilinear on the tangent space), whose values when acting on a pair of
vectors, instead of being numbers, happen to be a linear transformation on Tp M . One can
say that it is an End(T M )-valued 2-forma . This way of thinking about curvature as a
2-form is common in many modern areas, particularly those involving geometric phases or
non-Abelian gauge theories.
a
The linear transformations of a vector space V to itself are called endomorphisms and the space of all
such is denoted End(V ).
The definition given for the Riemann curvature tensor is natural and geometric. It turns
out, like all things that are intrinsically geometric, that it can be computed directly from the
components of the metric. Knowledge of this is crucial in general relativity where the space-
time metric is determined by equating curvature with the stress-energy-momentum tensor of
the matter content the Einstein equations. We show now how curvature is related to the
metric. The calculation is identical in structure to what we just did for S 2 , so you should bear
that in mind to guide you through the details. As before, let Zp be a vector at p that we wish
to parallel transport around a circle of geodesic radius r defined by the 2-plane given by the
orthonormal vectors Xp , Yp . Now let si be an orthonormal basis for the tangent space to our
manifold in a local chart about p. Without loss of generality we can take s1 , s2 to correspond
to Xp , Yp , respectively, at p. Then, as before, on the circle we can write the parallel transport
of Zp as
Z() = Z i () si (), (2.6.23)
and compute its derivative along the circle to be
dZ i
i
dZ j dsj dZ j i
= si + Z = + Z j si + ( ) n, (2.6.24)
d d d d
where we have defined
dsj
ji = si . (2.6.25)
d
These are (essentially) the components of the connection. The change in the normal directions
( )n is unimportant; hence, we have not been careful in how we write it. The condition
for parallel transport of the vector is that
dZ i
+ ji Z j = 0. (2.6.26)
d
43
These equations would be easy to solve if there was only a single equation or the ji were
R i 0
diagonal, for then we would have Z i () = e j d Z j (0). The actual solution is often written
in a similar form to this as R i 0
Z i () = P e j d Z j (0), (2.6.27)
and called a path ordered exponential. The ordering is necessary because the ji are matrices
and matrices do not commute. These things arise in field theory or anything using the Feynman
path integral, non-Abelian gauge theory, and even in time-dependent perturbation theory in
quantum mechanics. We will not really need them here, but there is no harm in mentioning it in
passing. To convey what is meant, we solve the equation for parallel transport perturbatively,
thinking of ji as small, writing Z i = Z(0)
i + Z i + Z i + and solving order-by-order. At
(1) (2)
zeroth order we have
i
dZ(0)
i
= 0, Z(0) () = Z i (0) ; (2.6.28)
d
at first order
i
dZ(1) Z
j
+ ji Z(0) = 0, i
Z(1) () = ji d0 Z j (0) ; (2.6.29)
d 0
We find, then, that the parallel transport of Z i (0) around a small circle of geodesic radius r
returns
Z Z Z
i i i j i 0
Z (2) = Z (0) j d Z (0) + m j d d Z j (0) + O(r3 ).
m
(2.6.31)
C(r) C(r) 0
We compute the change, treating each of the two pieces in turn. In doing so, it is convenient
to introduce a local Cartesian coordinate system, x1 , x2 , for points of the geodesic disc D(r)
whose boundary is the circle C(r). In these local coordinates we write
The symbols ikj that appear here are the components of the connection. We use the same
notation for them as we did for the Christoffel symbols in anticipation of the result that such
an identification can be made in the final expression for the Riemann curvature tensor. It then
follows immediately from Stokes theorem that
Z Z Z h i
ji d = ikj dxk = 1 i2j 2 i1j dx1 dx2 , (2.6.33)
C(r) C(r) D(r)
h i
= r 1 2j 2 1j + O(r3 ).
2 i i
(2.6.34)
p
The second contribution to the parallel transport is a little more subtle. We first evaluate the
inner integral as21
Z Z x
m 0
j d = m l m l 2
lj dy = lj x + O(r ). (2.6.35)
0
21
This is integration by parts together with an estimate for the magnitude of the neglected term.
44
This given, the integral can again be computed with the help of Stokes theorem and we find
Z Z Z
i m 0
m j d d = ikm m l k
lj x dx + O(r ),
3
(2.6.36)
C(r) 0 C(r)
Z h i
= 1 i2m m lj x
l
2 i1m mlj x
l
dx1 dx2 + O(r3 ), (2.6.37)
D(r)
Z h i
= i2m m
1j i m 1 2
1m 2j dx dx + O(r ),
3
(2.6.38)
D(r)
h i
= r 2m 1j 1m 2j + O(r3 ).
2 i m i m
(2.6.39)
p
are the components of the Riemann curvature tensor in the given basis. Precisely the same
formula gives the components of the Riemann curvature tensor in the Lorentzian setting
R = + , (2.6.43)
the only noticeable change being to use Greek indices rather than Latin ones. From the
Riemann tensor we define two other measures of curvature, that ultimately are the ones that
appear in the Einstein equations. First, the components of the Ricci tensor are defined by
contraction of the components of Riemann
R = R = + . (2.6.44)
The Ricci tensor is a symmetric type ( 02 ) tensor, R = R . Second, the contraction of the
Ricci tensor with the inverse metric, g , defines the Ricci scalar
R = g R . (2.6.45)
Remark: The result (2.6.41) should probably be given the fanfare it deserves, for it is
a serious calculation. It represents the result of parallel transport of an arbitrary vector
around an arbitrary circle in an arbitrary Riemannian manifold of arbitrary dimension. It is
almost incredulous that it is even possible to calculate this.
Remark: We have, in fact, proved more than we claimed. We have worked exclusively with
parallel transport around a circle of geodesic radius r, but nowhere was it essential that is
was a circle. What was essential was that it is a closed loop associated to a choice of 2-plane
at p. The points of the loop do not all have to be at the same geodesic distance from p. The
only change this makes in the definition of the Riemann curvature is to replace r2 with the
area of the geodesic disc that the loop bounds.
45
2.7 Covariant Derivative
The way that I have introduced the Riemann curvature tensor is not the way you will see in
any of the textbooks. There, the following definition is given
R(X, Y )Z = X Y Y X [X,Y ] Z,
(2.7.1)
Z = Z ,
or R
for any smooth function f . I.e. the covariant derivative of a function is the ordinary derivative
of that function. A vector is said to be covariantly constant if Z = 0. The covariant derivative
of Z along the vector field X is a type of directional derivative; it is written X Z. For any
pair of vectors X, Z it is another vector whose basic properties are the Leibniz formula
X f Z = X(f ) Z + f X Z, (2.7.3)
f X+Y Z = f X Z + Y Z. (2.7.4)
46
In books employing the tensor calculus notation, prevalent in the majority of physics texts,
you will see this formula written as
Z = Z + Z , (2.7.8)
and the covariant derivative defined by this expression. I have chosen to give you its modern
mathematical definition; it seems appropriate to try to keep with the times.
The final thing that we shall do is show that the coefficients of the connection are the
same as the Christoffel symbols, as our notation has been anticipating. This is a significant
result, known as the fundamental theorem of Riemannian geometry. It follows from the metric
compatibility of the connection. The connection is said to be metric compatible if23
X g(Y, Z) = g(X Y, Z) + g(Y, X Z), (2.7.9)
for any three vectors X, Y, Z. I leave you to convince yourselves that this expresses precisely
the notion that parallel transport should preserve the magnitude of a vector and the angle
between two vectors. Choosing for X, Y, Z the basis vectors , , this becomes
g = g + g . (2.7.10)
g = g + g , (2.7.11)
g = g + g . (2.7.12)
g + g g = g + + g + g .
(2.7.13)
Finally, if the connection coefficients are taken to be symmetric, = , then they are
given by
1
= g g + g g .
(2.7.14)
2
There is an unique, symmetric connection that is metric compatible; its coefficients are the
Christoffel symbols that appear in the geodesic equation. This is the fundamental theorem of
Riemannian geometry. With it we see that the components of the Riemann curvature tensor
(2.6.43) can be expressed purely in terms of the components of the metric and their derivatives.
T = 0. (2.7.15)
This expression was obtained in flat Minkowski space, and in Cartesian coordinates. Its
general expression, for any manifold and any choice of coordinates has the partial derivative
replaced with the covariant derivative. Let us see why by obtaining an expression for the
23
The analogous expression in tensor calculus notation is
g = 0.
47
divergence of a vector field in a general coordinate system. The divergence of a vector field is
defined in terms of the flux of that vector through the sides of a small volume
R
Z n darea
div Z = lim ,
Z vol 0
Z vol (2.7.16)
or div Z dvol = gij Z i nj darea,
where n is the unit outward normal to the boundary of the region and darea is the area
element of this bounding surface. It will be sufficient to work in a local adapted coordinate
system where the metric is diagonal
gij = diag g1 , g2 , . . . , gn , (2.7.17)
and the region is a box whose sides are surfaces given by a constant value of any one
coordinate, x1 = a, x2 = b and so forth. In the coordinate basis, the unit outward normal
to the surface x1 = a is the vector with components
1
nj = , 0, . . . , 0 . (2.7.18)
g1
The unit outward normals to the other sides of the box are similar. Now, the induced metric
on the surface x1 = a is
2 2
ds2 = g2 dx2 + + gn dxn , (2.7.19)
with similar expressions for each of the other sides. We can then compute directly
1
Z Z
i j
gij Z n darea = g1 Z 1 g2 gn dx2 dxn
x1 =a g1
(2.7.21)
1
Z
+ g1 Z 1 g2 gn dx2 dxn + other sides,
1 g1
Z x =a
1 g1 g2 gn Z 1 dx1 dx2 dxn + other sides,
= (2.7.22)
Z
1 p
det g Z 1 dvol + other sides,
= 1 (2.7.23)
det g
Z
1 p
det g Z i dvol,
= i (2.7.24)
det g
and it follows that the divergence of a vector is given, in local coordinates, by the expression
1 p 1 p
det g Z i = i Z i + i det g Z i .
div Z = i (2.7.25)
det g det g
It is an exercise to show that the bracketed term is precisely the contracted Christoffel symbol
jji so that the divergence of a vector is expressed in terms of the covariant derivative as
div Z = i Z i . (2.7.26)
Naturally, all this is the same in a Lorentzian manifold, where the divergence of a vector is
expressed in local coordinates as Z . Conservation of the stress-energy-momentum tensor
is the statement that it is divergence free; the local expression for a general manifold is
T = 0. (2.7.27)
48
Remark: In obtaining conservation of energy for a particle we appealed to the change in
the action under a constant shift . We did the same for continuous fields, including the
electromagnetic field. This really led to the conservation law ( T ) = 0 but since
was a constant we could safely drop it. There are no non-zero constant vectors in a general
curved manifold; just think of the sphere. The best that we can arrange for is that the vector
is covariantly constant, so the appropriate generalisation is that the shift is covariantly
constant, = 0. From the Leibniz formula for ( T ) we then arrive at the stated
conservation law for the stress-energy-momentum tensor.
2.8 Summary
It seems worthwhile to summarise concisely the equations one uses in describing geometric
properties of a space-time in a local coordinate system. So, suppose x are coordinates in a
local chart of our space-time. In such a local chart we write the metric as
ds2 = g dx dx . (2.8.1)
d2 x
dx dx
+ = 0, (2.8.2)
d 2 d d
where the are the Christoffel symbols. They are given in terms of the components of the
metric by
1
= g g + g g .
(2.8.3)
2
Geodesics defined in this way are parameterised by arc length if they are space-like or proper
time if they are time-like. What this means is that for a time-like geodesic its tangent vector
has unit magnitude
dx dx
g = 1. (2.8.4)
d d
If the geodesic is null, corresponding to the path of light, then its tangent vector is a null vector
and has zero magnitude.
The complete description of the curvature of the space-time is conveyed by the Riemann
tensor. From it one can deduce the Ricci tensor and the Ricci scalar, appearing in the Ein-
stein equations. The curvature can be determined from the components of the metric. In a
coordinate basis the components of the Ricci tensor are
R = R = + , (2.8.5)
49
Problems
1. A torus can be described explicitly as the subset of Euclidean R3 given by
p 2
(x1 )2 + (x2 )2 R + (x3 )2 2 = 0,
where R and are the two radii of the torus. Sketch the surface, or plot it with a program
such as Mathematica. Give an explicit parameterisation of the surface. Determine the
metric for your parameterisation and find the total surface area of the torus.
2. A torus can be explicitly embedded in the three-dimensional sphere, itself thought of as a
subset of R4 , through the parameterisation
1 1 1 1
X(u1 , u2 ) = cos u1 , sin u1 , cos u2 , sin u2 .
2 2 2 2
Determine the metric and find the total surface area. This embedding is known as a flat
torus.
3. The diagram below defines stereographic projection between the 2-sphere S 2 and R2 .
It comes in two forms; given a point (u1 , u2 ) of R2 it can be mapped to an unique point of
the 2-sphere. Show that the coordinates (x1 , x2 , x3 ) of this point are
2(u1 + iu2 ) (u1 )2 + (u2 )2 1
x1 + ix2 = , x3 = .
(u1 )2 + (u2 )2 + 1 (u1 )2 + (u2 )2 + 1
The second form is that any point (x1 , x2 , x3 ) of S 2 R3 can be mapped to an unique point
(u1 , u2 ) of R2 . Show that the coordinates of this point are
x1 + ix2
u1 + iu2 = .
1 x3
The use of complex numbers in these expressions is supposed to suggest to you that there
is something to be gained by identifying R2 with C in these relations. Doing so gives (part
of) the description of the 2-sphere as a complex manifold, the Riemann sphere.
Show that the metric on the 2-sphere in stereographic coordinates is
4 h i
ds2 = 2 (du1 2
) + (du2 2
) .
(u1 )2 + (u2 )2 + 1
Compute the area of the unit 2-sphere using these coordinates.
Describe the 3-sphere using stereographic projection to R3 .
4. A soap film spanning two identical coaxial circular rings adopts the shape of a catenoid, a
famous minimal surface. It can be described by the explicit embedding
X(u1 , u2 ) = cosh(u1 ) cos(u2 ), cosh(u1 ) sin(u2 ), u1 .
Make the surface as a soap film. Sketch the surface, or plot it with a program such as
Mathematica. Determine the metric.
50
5. Any surface embedded in R3 can be described locally as the graph of a function, say with
the explicit embedding
X(u1 , u2 ) = u1 , u2 , h(u1 , u2 ) ,
where h(u1 , u2 ) is the height function. This description was first introduced by Gaspard
Monge and is sometimes known as Monge gauge. Determine the metric in Monge gauge
and the area element for the surface. Give a local description of the unit 2-sphere in this
way. How much of the surface can you cover with a single height function?
6. Consider the unit time-like sphere in Minkoswki space-time, i.e. the subset defined by
7. Explain why geodesics on the 2-sphere are great circles; or at least convince yourself that
they are. Parameterising the sphere by standard spherical polar coordinates, determine the
geodesic equations in these coordinates.
10. Determine the geodesic equations for the de Sitter space dS4 . Try to describe the null
geodesics. [Hint: do not try to solve the geodesic equations; instead recall that de Sitter
is a space-like sphere in a 4+1-dimensional Minkowski space-time (R1,4 ) and use what you
know about geodesics on ordinary spheres in Euclidean spaces.]
11. Anti-de Sitter space, AdS4 , can be defined as the subset of R2,3 given by
for a positive constant a. Construct a parameterisation of AdS4 and determine the metric
in your parameterisation. Show that the curve x0 = a cos( /a), x1 = a sin( /a), x2 = x3 =
x4 = 0 is a time-like geodesic.
12. Make sure you can explain what a geodesic is and derive the geodesic equation for a general
metric. It is important.
13. Consider the embedding of the hyperbolic plane H2 in 2+1-dimensional Minkowski space-
time as the subset
(x0 )2 + (x1 )2 + (x2 )2 = a2 ,
with x0 > 0 and for any positive constant a. Using this description, determine the sec-
tional curvature. [Hint: dont forget to exploit the fact that the space is isotropic and
homogeneous, so you need only do the calculation at any one point.]
Consider the circle of geodesic radius r (on the hyperbolic plane embedded in R1,2 as above)
given by
c() = a cosh(r/a), a sinh(r/a) cos(), a sinh(r/a) sin() .
Find the result of parallel transport of the vector Z = (0, 1, 0) = e1 , tangent to the surface
at the point (a, 0, 0), around this circle.
51
14. Consider the three metrics
2 2 2 2
dr + a sin (r/a) d ,
ds2 = dr2 + r2 d2 ,
dr + a2 sinh2 (r/a) d2 ,
2
where a is a fixed, positive constant. Determine the components of the Ricci tensor Rij and
hence the value of the Ricci scalar R = g ij Rij for each metric.
15. The Ricci tensor is symmetric, R = R . We use this property but did not establish it in
the notes. Here is a way to do so, using our approach. We gave an expression for the Ricci
tensor as
R = + ,
where are the Christoffel symbols and are symmetric in , , i.e. = . This
means we need only show symmetry for the second and fourth terms in the expression for
the Ricci tensor; the first and third are explicitly symmetric.
Show that = .
So we need only show that is symmetric. This follows from the fact that
p
= ln | det g|.
From the definition of the Christoffel symbols in terms of the metric, show that
1
= g g .
2
Now let M be a square, invertible matrix. We will suppose that it is symmetric and hence
diagonalisable. By diagonalising M , or otherwise, establish the matrix identity
det M = etr ln M ,
Taking for M the metric g, complete the demonstration that the Ricci tensor is symmetric.
A different strategy is as follows: Recall that the Ricci tensor can be expressed in terms of
the Riemann tensor as
R = R = R .
Show that the symmetries of the Riemann tensor
R = R , R = R , R = R ,
imply the symmetry of the Ricci tensor, R = R . [Of course, the lengthy part of this
approach, which we have skipped entirely, is to establish the given properties of the Riemann
tensor.]
16. In the lectures an expression was given for the components (in a coordinate basis) of the
covariant derivative of a vector
Z = Z + Z .
Show that the corresponding expression for the components of the covariant derivative of a
1-form A, with components A in a coordinate basis, is
A = A A .
52
Hence show that the components of the Maxwell field strength tensor, F = A A ,
are still given by A A .
What is the corresponding expression for the components (in a coordinate basis) of the
covariant derivative of a type ( 02 ) tensor, F ?
Write out in full the local continuity equation for the stress-energy-momentum tensor
T = 0.
17. In the notes, an expression was obtained for the divergence of a vector div Z = (det g)1/2 i (det g)1/2 Z i .
If the vector is the gradient of a scalar, Z i = i = g ij j , then this leads to an expression
for the Laplacian in a general coordinate system
1 p
2 = det g g ij j .
i
det g
Use this formula to give the form of the Laplacian in cylindrical and spherical coordinates.
In the Lorentzian setting, precisely the same formula yields the wave operator (often denoted
by the symbol and frequently called the dAlembertian)
1 p
| det g| g .
= p
| det g|
Verify this first for the usual Minkowski metric. What is the wave operator for the Schwarzschild
metric
2m 2 2 1 h i
ds2 = 1 c dt + dr 2
+ r 2
d 2
+ sin 2
() d 2
?
r 1 2m
r
Can you find any (non-zero) solutions of the wave equation = 0 in the Schwarzschild
space-time? [Hint: you should consider this to be very hard!!]
53
The original image of the Eagle Nebula taken by the Hubble Space Telescope was released in
1995 and called Pillars of Creation. It is one of the most iconic images of space and caught
the imagination of the general public, including myself. This image was taken in 2015 for the
20th anniversary. Image from NASA.
54
Chapter 3
The possibility of explaining the numerical equality of inertia and gravitation by the unity
of their nature gives to the general theory of relativity, according to my conviction, such a
superiority over the conceptions of classical mechanics, that all the difficulties encountered
in development must be considered as small in comparison.
I present in this chapter two derivations of the Einstein equations of general relativity; the first
using an action principle due to Hilbert, and the second an imitation of the line of reasoning
that led Einstein to his own derivation. Einsteins concept in general relativity is that the
matter content of the universe determines the space-time metric. How it does so can be
described by an action principle. The action should be a scalar quantity constructed out of the
metric and its derivatives; the field equations are obtained from the condition that the actual
metric corresponds to a critical point of the action. Now, in Newtons theory the gravitational
potential satisfies a differential equation of second order, so we look for an action that depends
on derivatives of the metric of at most second order and is linear in the second order derivatives.
There is only scalar quantity with these properties; the Ricci scalar R. So the action must be
proportional to it. The Einstein-Hilbert action in general relativity is1
c4 c4
Z Z p
S= R dvol = R | det g| d4 x, (3.1)
16G M 16G M
where R is the Ricci scalar, g is the metric, and the integral is taken over the entire space-time
manifold M . The prefactor turns out to be what is needed to produce the Newtonian theory
in the weak field regime with G Newtons gravitational constant. The Einstein equations are
a system of equations for the metric of space-time. They are obtained from the action by
the usual principle that the physical metric corresponds to a critical point of the action. Let
g + h be a variation of the metric g , i.e. the h are infinitesimally small
p and vanish
on the boundary of the space-time. The variation of the measure factor | det g| can be
determined by writing the perturbed metric as
g + h = g + h .
(3.2)
The determinant of the perturbed metric is then a product recall det AB = det A det B with
the second factor being the determinant of a near identity matrix, + h . By diagonalising
the matrix h we see that this latter determinant is equal to
det + h = 1 + h + O(2) = 1 + g h + O(2).
(3.3)
1
If the manifold has a boundary an additional term, known as the Gibbons-Hawking-York boundary term,
should be added to make the variational problem well-posed.
55
It follows that the variation of the measure factor is given by
p p q
| det g| 7 | det g| 1 + g h + O(2), (3.4)
p 1p
= | det g| + | det g| g h + O(2). (3.5)
2
The variation of the Ricci scalar, R = g R , can be broken into two parts; the variation of
the inverse metric g , and the variation of the Ricci tensor R . The former is easy. Since,
by definition of the inverse metric, g g = 4, the variation of the inverse metric must be
g 7 g h + O(2). (3.6)
We write the variation of the Ricci tensor as R 7 R + R + O(2). The variation of the
Einstein-Hilbert action is then
c4
Z p
1
S[g + h ] = S[g ] + g R h R R g | det g| d4 x + O(2). (3.7)
16G M 2
It turns out that the term involving the variation of the Ricci tensor makes no contribution;
it is a total divergence and integrates to the boundary where the variation vanishes. The
argument that is always given for this is very slick, making use of the existence of local inertial
frames, or Riemann normal coordinates. These are coordinates, adapted to any particular
point p, where the metric is the Minkowski one to second order accuracy about p. It follows
that the Christoffel symbols all vanish at the point p, although their variation does not.
The variation of the Ricci tensor can then be written, at p, as
h i
g R = , (3.8)
p
h i
= , (3.9)
h i
= div g g . (3.10)
p
Since this last expression is a natural geometric object the relation must be valid in any
coordinate system. This establishes that the variation of the Ricci tensor is a divergence and
therefore makes no contribution to the field equations. We find, therefore, that the condition
for g to be a critical point of the Einstein-Hilbert action is that
1
R R g = 0. (3.11)
2
These are the Einstein equations in the absence of matter content. If the action for the matter
content is Z Z p
Smatter = L dvol = L | det g| d4 x, (3.12)
M M
where L is the Lagrangian, then its change under a variation of the metric can be written,
formally, as
Z
1 p
Smatter [g + h ] Smatter [g ] h T | det g|d4 x, (3.13)
M 2
which serves as a definition of the stress-energy-momentum tensor. To see that it is the same
object as we introduced previously, we compute it explicitly for the electromagnetic field. There
the action is
1
Z p
Sem = F F g g | det g| d4 x, (3.14)
M 4
56
where the field strength tensor F = A A depends on the metric through the
connection . Computing the variation explicitly we find
Z
1
1
Sem [g + h ] = F F g h g h F F + O(2)
M 4 2
(3.15)
1
p 4
1 + g h + O(2) | det g| d x,
2
Z
1 1
= Sem [g ] + h g F F F F g
M 2 4
(3.16)
1 p 4
F F + O(2) | det g| d x.
2
Here the variation F F is that associated with the metric dependence of the connection .
It can be shown that this vanishes on account of Maxwells equations for the electromagnetic
field, F = 0. Thus, the stress-energy-momentum tensor for the electromagnetic field is
1
T = g F F F F g , (3.17)
4
precisely the expression that we obtained previously. This all being said, combining the varia-
tion of the Einstein-Hilbert action with that of the matter content the final Einstein equations
that determine the metric of space-time are
1 8G
R R g = 4 T . (3.18)
2 c
Remark: The derivation I have given of the Einstein equations is due to Hilbert (1915). I
would say that it is a modern approach, but given that it first appeared at almost exactly
the same time as Einsteins own paper deriving his field equations, Einstein included it in
his 1916 review article, and Landau used the same derivation in his Course of Theoretical
Physics from the 1930s, you would be justified in not believing me. It is not modern.
Nonetheless, the approach to physics based on action principles is not as widely accessible to
undergraduates in physics as I would like; it is to the enormous discredit of theoretical physics
as a profession that we have not managed to make our most central and unifying concept
more widely accessible to undergraduates. Inspired by the example laid out by Landau, and
Feynman amongst others, I have made a small effort to remedy this.
We also now give a stripped down and simplified version of Einsteins line of reasoning in
arriving at an improvement of Newtons theory of gravity, showing explicitly that Einsteins
equations reduce to Newtons in an appropriate limit. In Newtons theory we have the equations
d2 xi
= i , (3.19)
dt2
i i = 4G , (3.20)
for the motion of a test particle, along trajectory xi (t), in the gravitational field produced by a
mass distribution . According to Einsteins equivalence principle, gravity and inertia should
be considered equivalent as phenomena. The motion of the test particle, viewed in Newtons
theory as non-inertial motion in a gravitational field, can equally be viewed as free, meaning
geodesic, motion in a curved space. It should therefore be replaced by the geodesic equation
d2 x
dx dx
+ = 0. (3.21)
d 2 d d
In a non-relativistic regime where gravitational effects are weak and velocities of motion small,
we can expect the components dxi /d of the particle velocity to be O(v/c) and therefore small
compared to dx0 /d , which is O(1). Therefore, to leading order the geodesic equation is
1 dxi
+ i00 = 0, (3.22)
c2 dt2
57
so that we may identify c2 i00 with the gradient of the Newtonian potential. We can say more,
if we assume that the metric deviates only by small quantities from that of Minkowski space.
Then the Christoffel symbol may be approximated as
1 1
i00 = 0 gi0 + 0 g0i i g00 = i g00 , (3.23)
2 2
with the assumption that spatial derivatives of the metric are more significant than time
derivatives, as is appropriate for static gravitational situations like the solar system. From
this we identify
2
g00 = 1 2 , (3.24)
c
where is the Newtonian potential, in the weak field regime. The Newtonian potential appears
as the 00-component of the space-time metric.
Now we look for a replacement of Newtons field equation. The source is the mass density
in Newtons theory. From considerations of special relativity, we have seen that this should be
replaced with the stress-energy-momentum tensor T . Its 00-component is the energy density,
or mass density times c2 , so that such a replacement matches exactly in terms of character with
the identification of the Newtonian potential with the component g00 of the space-time metric.
The left-hand-side of Newtons field equation involves second derivatives of the potential; these
are now naturally viewed as representing the curvature of space-time. What we need to do
is find a symmetric type ( 02 ) tensor, corresponding to curvature, that can be identified with
the stress-energy-momentum tensor. The properties that guide us in finding it are that it
should involve derivatives of no greater than second order of the metric, it should be linear and
homogeneous in these second derivatives and it should be compatible with the conservation
property of the stress-energy-momentum tensor, T = 0. There is only one quantity that
can be constructed from the Riemann curvature tensor with these properties:
1
R R g . (3.25)
2
We therefore postulate the field equation
1
R R g = K T , (3.26)
2
for a constant K, to be determined by matching to Newtons theory in the weak field limit. It
will be convenient to make as much use of the stress-energy-momentum tensor as possible, as
we have more knowledge of it than of the components of the Ricci tensor. Taking the trace of
the field equations gives R = K T and therefore
1
R = K T T g . (3.27)
2
In the weak field limit we can approximate the 00-component of the right-hand-side as 12 Kc2 ,
so we only need an approximation for R00 to determine K. With the assumption that the
metric is static so that time derivatives can be neglected we find
1
R00 = i i00 = i i , (3.28)
c2
to leading order, and it follows that K = 8G/c4 . Thus we arrive at the Einstein field equations
of general relativity
1 8G
R R g = 4 T . (3.29)
2 c
58
A beautiful example of gravitational lensing in this Hubble Space Telescope image of the
galaxy cluster Abell 1689, taken in 2003. Such images vividly bring to life the basic content
of Einsteins theory, that mass distorts the structure of space and hence the path along which
light, or anything else, travels. Image from NASA.
59
Chapter 4
As you see, the war treated me kindly enough, in spite of the heavy gunfire, to allow me
to get away from it all and take this walk in the land of your ideas.
It is always nice to have exact solutions to work with. The first of these, for general relativity,
was provided by Karl Schwarzschild in late 1915. It is a solution for the vacuum region of
space-time outside a single, central massive body. As such, it can be used to model the motion
of the planets and other celestial bodies in the gravitational field of the sun and so determine
the corrections to Newtons theory predicted by general relativity; these are the classical tests.
It also contains a singularity, a point in the space-time at which the curvature is infinite, and
so is the progenitor of all studies of black holes.
with only two unknown functions f (r) and g(r), that depend only on the radial coordinate.
That they are independent of the time-coordinate t corresponds to the space-time being static.
It turns out that this is not an assumption; spherical symmetry is strong enough to imply this
property, a result known as Birkhoffs theorem. Therefore, (4.1.1) represents the most general
form of the metric compatible with spherical symmetry.
The unknown functions f and g are found by inserting this form of the metric into the
Einstein equations. So we need to determine the curvature of space-time represented by (4.1.1).
The usual formula for the components of the Ricci tensor gives them in terms of the Christoffel
symbols, so we begin by determining those. Although there is a formula in terms of derivatives
of the metric, it usually turns out to be more economical to read them off from the equation
for the geodesics, so what we really do is determine the equation for the geodesics. As usual,
let = ct( ), r( ), ( ), ( ) be a curve parameterised by arc length (c times proper time)
1
I will not describe properly what it means to say that a space-time has a certain symmetry, or isometry.
Therefore, I ask you to be pragmatic and take that spherical symmetry implies that the metric can be written
in such a way that part of it looks like the standard round metric for a sphere.
60
and consider a variation of it. By direct calculation we have that the metric for the variation
of is
d(x + ) d(x + ) 2
ds2 = g (x + ) d , (4.1.2)
d d
2 2
= 1 2f (r)c t 0 + 2g(r) r r + f 0 (r) c t + g 0 (r) r
2 2 r
+ 2r + sin2 () + 2r2
(4.1.3)
2
2
+ 2r sin() cos() + 2r sin () + O(2) d 2 ,
2 2
recalling that is (c times) proper time for , i.e. g (x) dx dx
d d = 1, and where the prime
denotes differentiation with respect to argument. The geodesics are critical points of the
distance function, or in other words curves for which
Z Z
ds ds = O(2), (4.1.4)
0
for all variations 0 of the curve preserving the endpoints. Using the form of the metric for the
variation 0 we find that the geodesics are given by the equations
0 = f (r)c t , (4.1.5)
1 2 1 2 h 2 2 i
0 = g(r) r + f 0 (r) c t g 0 (r) r r + sin2 () ,
(4.1.6)
2 2
2 2
2
0 = r r sin() cos() , (4.1.7)
0 = r2 sin2 () .
(4.1.8)
From the geodesic equations we can read off the Christoffel symbols; we record the complete
set that are non-zero (recall we denote ct by x0 )
f0 f0 g0 r r sin2 ()
00r = 0r0 = , r00 = , rrr = , r = , r = ,
2f 2g 2g g g
(4.1.9)
1 1 cos()
r = r = , = sin() cos(), r = r = , = = .
r r sin()
Next we want to determine the components of the Ricci tensor in our coordinate system.
Recall that they are given in terms of the Christoffel symbols by the general formula
R = + . (4.1.10)
The calculation is tedious, but not difficult. It turns out that only the diagonal components
are non-zero and they are given by
1 h 00 1 0 2 1 2 i
R00 = f f f 0g0 + f 0 , (4.1.11)
2g 2f 2g r
1 h 00 1 0 2 1 2f 0 i
Rrr = f f f 0g0 g , (4.1.12)
2f 2f 2g rg
1 r 0 r
R = 1 f + 2 g0, (4.1.13)
g 2f g 2g
R = sin2 ()R . (4.1.14)
The vacuum Einstein equations, R 21 Rg = (8G/c4 )T , imply that the Ricci scalar
vanishes; just take the trace of the Einstein equations. It follows that the components of Ricci
must themselves be zero. The vanishing of R00 and Rrr then combine to imply that
r f g = 0, (4.1.15)
61
or f g = 1. You might think that the equation only implies f g = const, but the constant can
be taken to be unity without loss of generality by the freedom to rescale t by any multiplicative
constant. Using this to simplify the equation R00 = 0 we find
2
f 00 + f 0 = 0, (4.1.16)
r
2m
f =a , (4.1.17)
r
for arbitrary integration constants a and m. Finally, the condition R = 0 demands that
a = 1 but places no constraints on m. At last we can write the Schwarzschild metric in its
standard form
2 2m 2 2 1 2 2
h
2 2 2
i
ds = 1 c dt + dr + r d + sin () d . (4.1.18)
r 1 2m
r
The sole free parameter m is the mass of the Schwarzschild black hole. The distance r = 2m,
where our coordinate system is badly behaved, is known as the Schwarzschild radius. It is the
event horizon of the black hole. However, it is not a singularity in space-time; it is only a
break-down of our coordinate system, called a coordinate singularity. On the other hand the
origin, r = 0, is a singularity of space-time. It is called a black hole.
Remark: The parameter m is always referred to as the mass of the Schwarzschild black hole,
even though it has dimensions of length; and r = 2m is properly called the Schwarzschild
radius. m is the mass in natural unitsa , being equal to GM/c2 in SI units, where M is the
actual mass. For an object with the mass of the sun, m 1.48 km.
a
Natural units are those where the speed of light c, Newtons gravitational constant G and Plancks constant
~ are all taken to be unity; it is always possible to restore the appropriate factors by dimensional analysis.
for a constant . In addition, we know that the geodesic is parameterised by (c times) proper
time, or is null, so that
2m 2 1 2
1 c t + 2m r = , (4.2.2)
r 1 r
where = 1 for the time-like geodesic of our explorer, and = 0 for the null geodesics of the
light signals they send to us. Eliminating t in favour of the constant this becomes
2 2m
r = 2 + . (4.2.3)
r
2
In my head she is called Jolene, but you may prefer to supply your own narrative.
62
Taking a square root, the trajectory of our friend (who is time-like so set = 1) is given
implicitly by the expression
dr0
Z R Z 1/r
du
= p = p , (4.2.4)
2 0
| 1 + 2m/r | 2 2
| 1 + 2mu|
r 1/R u
where is (c times) the proper time since our explorer set out at radius R. An explicit
expression for the explorers position r as a function of the proper time is not nearly as
important as the simple observation that this expression is perfectly regular through the event
horizon at r = 2m (or u = 1/2m). In fact, the explorer reaches the singularity at r = 0 in
finite proper time. A person falling freely inwards, approaching the horizon radially, will pass
straight through it; it is not a singularity of the space-time.
What do we see, from a distant position well away from the horizon? Expressing the motion
of the infalling explorer in terms of the coordinates we measure as r(t), (4.2.2) becomes
1 2 1 2 2
2m + 3 c2 t r = 1,
1 r 2m
1 r
(4.2.5)
2m 1/2
2m 2
t r = 1 1+ ,
c r r
and we obtain the formula
dr0
Z R Z 1/r
du
cT = p = p , (4.2.6)
0 2 0 2 2
r (1 2m/r ) | 1 + 2m/r | 1/R u (1 2mu) | 1 + 2mu|
for the time that we record it takes for the explorer to get to the radial distance r. The main
conclusion here is that this time diverges (logarithmically) as the explorer approaches the event
horizon at r = 2m. In other words, we never see them reach the horizon itself, according to
our own measure of time. It implies that we can never receive any information that such an
explorer obtains from the event horizon, or from the region of space-time inside it. It is hidden
from us. Even light does not make it to us in a finite time as we measure it. For suppose our
explorer pauses at radial position r to send us a signal, say a radio communication, that travels
to us along a radial null geodesic. We record the trajectory of the light signal as satisfying
2m 2 2 1
1 c dt = dr2 , (4.2.7)
r 1 2m
r
and this diverges as our explorer friend approaches the event horizon at r = 2m. The event
horizon is a barrier, concealing a region of space-time that we can never have any knowledge
of so long as we stay in the comfort and safety of the asymptotic far-field of the Schwarzschild
metric.
Some additional insight is gained from a change of perspective, or coordinates. The
Eddington-Finkelstein coordinates replace the asymptotic Schwarzschild time coordinate with
an affine coordinate parameterising the radial null geodesics. The ingoing Eddington-Finkelstein
coordinate u is defined by
2m 1
du = cdt + 1 dr, (4.2.9)
r
so that each trajectory u = const. (and = = const.) corresponds to an ingoing radial null
geodesic. By eliminating cdt, one finds directly that the Schwarzschild metric becomes
2 2m h i
ds = 1 du2 + 2dudr + r2 d2 + sin2 () d2 . (4.2.10)
r
63
In these coordinates the components of the metric no longer diverge on the surface r = 2m, a
reflection of what we learnt above that the event horizon is not a singularity of the space-time.
From these coordinates, however, we do learn something new, and interesting, about the event
horizon. This surface, r = 2m, is covered by Eddington-Finkelstein coordinates (u, , ). The
metric on the event horizon is
h i
ds2 = (2m)2 d2 + sin2 () d2 . (4.2.11)
r=2m
What this means is that the interval between two points of the event horizon that only differ
in the value of the u coordinate is zero, or in other words the event horizon is a surface that
contains a null direction; it is a null surface.
Analogous to the ingoing coordinates we have described, there are also outgoing Eddington-
Finkelstein coordinates (v, r, , ) with v defined by
2m 1
dv = cdt 1 dr, (4.2.12)
r
and such that each trajectory v = const. (and = = const.) corresponds to an outgoing
radial null geodesic. In terms of these coordinates the Schwarzschild metric becomes
2 2m 2 2
h
2 2 2
i
ds = 1 dv 2dvdr + r d + sin () d . (4.2.13)
r
The description of the Schwarzschild space-time that follows from a study with these coor-
dinates is an exact time-reversed version of that for the ingoing coordinates. This is to be
expected given the static nature of the space-time, i.e. invariance under t t.
64
Figure 4.1: Two views of the Schwarzschild space-time. Left: the Kruskal coordinates described
in the text, and Right: a compact version of the same picture, known as a Carter-Penrose
diagram. Its key feature is that null directions are preserved in the compactification so that the
causal structure can be correctly inferred from the diagram. Two typical time-like trajectories
are indicated; one that remains outside the event horizon and one that passes through it.
functions u(U ) and v(V ) so as to maximally extend the coverage of the space-time and, in
particular, to remove any coordinate singularities. Kruskals choice6 was to take
32m3 r/2m h 2 2 i h i
ds2 = e c dT + dX 2 + r2 d2 + sin2 () d2 . (4.3.7)
r
Tracing through the various definitions we find that the coordinates cT, X are related to the
original Schwarzschild coordinates ct, r by the expressions
cT + X
ct = 2m ln , (4.3.8)
cT X
r
1 er/2m = c2 T 2 + X 2 . (4.3.9)
2m
It follows that the entirety of the Schwarzschild time-coordinate < ct < is contained
within the subspace bounded by cT > X and cT < X of the Kruskal-Szekeres coordinates.
This corresponds to only one of four quadrants, separated by the boundaries cT = X
see figure (particularly left). The singularity at r = 0 corresponds to the set of points
c2 T 2 + X 2 = 1 in the Kruskal-Szekeres coordinates. It is a space-like set that exists at the
very end of space-time, as it should, being the end of all things7 . Region I is the asymptotic
far-field of the Schwarzschild coordinates. A radially infalling explorer crosses the event horizon
cT = X (both positive) and enters region II. The boundary the event horizon is a null
6
A different choice of coordinates, that also gives a maximal extension, is described by Landau & Lifshitz.
7
It is not the end of all things an observer that remains outside the event horizon does not fall into the
singularity but I could not resist the prose, even if it is inaccurate.
65
surface; even light emitted from within region II cannot escape it and necessarily arrives (in the
infinite future) at the singularity at c2 T 2 +X 2 = 1, T > 0. For this reason it is called a black
hole. Any time-like observer in region II reaches the singularity in finite proper time. Region
III is an exact time-reversed copy of region II. It contains a singularity at c2 T 2 + X 2 = 1,
T < 0. Any light signal emitted from within region III will cross the event horizon and leave
it; for this reason it is referred to as a white hole. The two regions are mirror images of
each other, a reflection of the symmetry t t associated with the space-time being static;
we give them different names because of the difference in the way that we perceive them and
because we have no experience of anything other than a particular direction for time. Finally,
region IV is a duplicate of region I, a second asymptotically flat region of space-time. They are
connected by a space-like portal or wormhole; a 2-sphere of radius 2m called the Einstein-
Rosen bridge. It is impossible for any time-like observer to travel between them or even to
communicate with the other side.
The first of these integrates immediately to give a first integral of the motion
c t = , (4.4.5)
1 2m/r
for a constant , that looks a bit like the asymptotic -factor of special relativity. This constant
of the motion is associated to the time-translation invariance (t t + const.) of the metric,
and, as such, is properly referred to as the energy (in natural units and per unit mass). As
in the non-relativistic Kepler problem, the two angular geodesic equations have the solution
= /2, r2 = `, (4.4.6)
where ` is a constant, and this one particular solution is enough to understand all of the
geodesics. The reason is the same; it is a consequence of the spherical symmetry, and typically
thought of in terms of conservation of angular momentum. This all given, the equation for
r( ) reduces to
2 m/r2 m/r2 2m `2
2
0 = r + r 1 . (4.4.7)
1 2m/r 1 2m/r r r3
66
However, it proves easier not to deal with this equation directly but to recall that the geodesic
is parameterised by (c times) proper time, which contributes the first integral
2m 2 2 1 2 h 2 2 i
1 = 1 c t + r + r2 + sin2 () , (4.4.8)
r 1 2m/r
2 1 2 `2
= + r + 2 . (4.4.9)
1 2m/r 1 2m/r r
This equation rearranges to the form
2m `2
2 2m
r + 2 1 = 2 1, (4.4.10)
r r r
and can then be thought of as an energy balance kinetic energy plus potential energy equals
total energy for a one-dimensional dynamical system with effective potential
2m `2 2m`2
2Veff = + 2 3 . (4.4.11)
r r r
The first two terms are those that appear in the Kepler problem of Newtonian gravity; it is
only the last term 2m`2 /r3 that is new in general relativity.
where
12m2 1/2 m 6m3
1
uc = 1 1 2 = 2 + 4 + O(m5 `6 ), (4.4.15)
6m ` ` `
and has an interpretation as the inverse radius of a circular orbit. Neglecting the quadratic
term, 3m(u uc )2 , the solution is
h i
u = uc 1 + e cos 1 6muc , (4.4.16)
where e is the eccentricity of the orbit. This should be contrasted against the Keplerian orbit,
obtainable by neglecting the general relativistic effects entirely
mh i
uKepler = 2 1 + e cos() . (4.4.17)
`
8
Our presentation is mildly idiosyncratic, being specially adapted to the particular problem; the formal
method for what we are doing is known as multiple time scales perturbation theory. I am grateful to George
Rowlands for explaining this to me.
67
The precession of the perihelion arising from general relativistic effects is given by the difference
from 2 of the change in angle between successive perihelia and is readily seen to be
2
= 2 = 6muc + O(m2 u2c ), (4.4.18)
1 6muc
6m2
= + O(m4 `4 ). (4.4.19)
`2
The orbit of the planet Mercury is not close to circular, the eccentricity being e = 0.21.
Nonetheless, for the purpose of estimates it suffices to take the value of the semi-major axis,
5.79107 km, and average orbital speed, 47.4 km s1 , in estimating the value of the constant `.
These values give ` = 9.15 103 km. Recall that the Schwarzschild mass m is equal to GM/c2 ,
where M is the actual mass. For the sun m = 1.48 km. The predicted contribution of general
relativistic effects to the precession of the perihelion of Mercury is therefore = 4.93 107
radians, or 0.102 arcseconds. This is the shift per Mercurian year. Mercury orbits the sun
with a period of 0.24 years, so that the precession amounts to
100
0.102 = 42.5 seconds of arc per century.
0.24
This prediction, which accurately accounts for the discrepancy between the observed precession
and the predictions of Newtons theory, was in 1915 the only experimental test of Einsteins
theory of general relativity.
Figure 4.2: A gallery of Einstein ring images taken by the Hubble Space Telescope. They are
the result of the gravitational lensing of light from a distant object by one closer to us along
the same line-of-sight. Image from NASA.
We have all seen the beautiful images of gravitational lensing of distant galaxies by those
closer to us along the same line of sight. They are wonderous and truly inspiring. The original
observation of gravitational deflection of light (or lensing) was made for stars behind the sun
by Sir Arthur Eddington during the total solar eclipse of 1919. The premise for it is much the
same as for gravitational lensing of distant galaxies; we are observing light travelling along null
geodesics and passing near to a massive object. The simplest account of this phenomenon is
therefore provided by the null geodesics of the Schwarzschild metric.
The derivation we gave for time-like geodesics does not pass over directly to null geodesics.
Nonetheless, they are geodesics and satisfy the same geodesic equation their tangent vector
68
is parallel transported along itself with the only difference being that this tangent vector is
a null vector. Therefore they are still characterised by constants
2m
1 c t = , r2 = `, (4.4.20)
r
much as before, with the help of which the condition that the tangent vector is null gives
2 `2 2m`2
r + = 2. (4.4.21)
r2 r3
Switching to a parameterisation in terms of , as before, and rewriting in terms of the inverse
radius u = 1/r, as before, we arrive at the equation
2 2
u + u2 2mu3 = 2 ,
` (4.4.22)
or u + u 3mu2 = 0.
The equation can be integrated exactly as an elliptic integral; as before we shall adopt a
perturbative approach, writing u = u(0) + u(1) + and solving order by order. The zeroth
order solution neglecting general relativistic effects is u(0) = b1 cos() and corresponds to the
straight line x = b just use x = r cos() and u = 1/r. The first order perturbation to this is
given by the particular integral of
2 3m
u(1) + u(1) = 3m u(0)
= 1 + cos(2) , (4.4.23)
2b2
which one may verify is
3m h 1 i
u(1) = 1 cos(2) . (4.4.24)
2b2 3
Thus the photon trajectory, incorporating the first order correction coming from general rela-
tivity, is described by
1 1 3m h 1 i
u = = cos() + 2 1 cos(2) + O(m2 b3 ). (4.4.25)
r b 2b 3
The closest approach of the photon to the origin (r = 0) still occurs when = 0 and is given by
the distance b(1 + m/b2 )1 . The asymptotic directions can be found by setting u = 0, leading
to r
8m2
b 2m
cos( ) = 1 1+ 2 = + O(m3 b3 ). (4.4.26)
2m b b
The total deflection angle is therefore given approximately by
4m
= , (4.4.27)
b
to leading order in general relativistic effects.
Let us estimate this deflection for light passing close to the sun as in the original obser-
vations of Eddington. Recall that the Schwarzschild mass m, in SI units, is equal to GM/c2
where M is the (proper) stellar mass and G is Newtons gravitational constant. For the sun
GM/c2 1.48 km, while the solar radius is approximately 6.96 105 km. It follows that light
that just grazes the solar surface undergoes a total deflection according to general relativity of
approximately
8.5 106 radians, (4.4.28)
or about 1.75 seconds of arc.
* * *
69
Gravitational lensing has become an important observational tool for inferring mass distribu-
tions of intervening objects that act as lenses for more distant galaxies or quasars. The idea
is to relate the deflection angle to the mass distribution of the lens from theory and then by
fitting an experimental image infer the mass distribution. The simplest description models
the lens by a weak-field metric
2 2 2 2 2 h 2 2 2 i
ds = 1 + 2 c dt + 1 2 dx1 + dx2 + dx3 , (4.4.29)
c c
where, as we learned in Chapter 3, may be identified with the Newtonian potential. Since
the light that reaches us travels along null geodesics we can say
1 2/c2 h 1 2 2 2
i
3 2
c2 dt2 =
dx + dx + dx ,
1 + 2/c2 (4.4.30)
or cdt = 1 2/c2 d`,
where d`2 = (dx1 )2 + (dx2 )2 + (dx3 )2 and assuming /c2 is small. We see that the Newtonian
potential acts much like an effective refractive index for the light as it travels to us through
the intervening matter distribution. With this in mind, the path that we observe the light ray
to follow will be the same as if the light were passing through a medium with refractive index
n(x) = 1 2(x)/c2 , and is hence given by Fermats principle of least time the path followed
by the light is such as to make the time taken as short as possible.
Let us orient our coordinate system so that the x3 -axis corresponds to the line of sight
to the distant quasar. Then, parameterising the path of the light by (x1 (z), x2 (z), z), the time
that it takes to reach us is
Z Earth h i1/2
2 2
ct = 1 2/c2 z x1 + z x2 + 1 dz, (4.4.31)
quasar
and by Fermats principle the path should be chosen so as to correspond to a critical point of
this functional; a geodesic of the so-called optical metric. Carrying out the variation we find
Z Earth z xi z i
ct[x + ] ct[x] = 1 2/c2 1/2
quasar (z x1 )2 + (z x2 )2 + 1
(4.4.32)
2 i h i1/2
1 2 2 2 2
2 i z x + z x +1 dz + O( ),
c
1 2/c2 z xi
Z Earth
i
= z 1/2
quasar (z x1 )2 + (z x2 )2 + 1
(4.4.33)
2i h i1/2
1 2 2 2 2
+ 2 z x + z x +1 dz + O( ),
c
and it follows that the path taken by the light satisfies
1 2/c2 z xi
2i h 1 2
2 2
i1/2
z = z x + z x + 1 . (4.4.34)
(z x1 )2 + (z x2 )2 + 1
1/2 c2
Integrating over z along the line of sight, from the quasar to ourselves, and recalling that the
distance element is d` = [(dx1 )2 + (dx2 )2 + dz 2 ]1/2 , we can then say
i Earth 2 Earth
Z
2 x
1 2/c = i d`. (4.4.35)
` quasar c2 quasar
For the left-hand-side we will assume that the Newtonian potential vanishes at both source
and observer, which are situated far from the matter distribution of the lensing galaxy it is
70
accounting for. It follows that the left-hand-side is the difference in the angle the light ray
makes with the line of sight at Earth and at the quasar, or in other words, the total deflection
angle i . For the right-hand-side we use Newtons formula for the gravitational potential
and find
2G Earth (x0 ) 3 0
Z Z
i = 2 i 0
d x d`, (4.4.36)
c quasar lens |x x |
2G Earth (x x0 )i (x0 ) 3 0
Z Z
= 2 d x d`. (4.4.37)
c quasar lens |x x0 |3
The integral over the line of sight can be done with a thin lens approximation for the lensing
galaxy, which seems completely reasonable, and that the path distance ` is very nearly the
same as the 3-component z `. This then leads to the lensing formula
4G (x x0 )i 2 (x0 ) 2 0
Z
i = 2 d x, (4.4.38)
c lens |x x0 |2
where 2 (x0 ) = (x0 , z) dz is the integral of the mass density over the thickness of the lensing
R
71
Problems
1. Calculate the general relativity contribution to the precession of the perihelion of the Earth.
2. Consider a satellite, of mass Ms , in circular orbit about the Earth and neglect all other
gravitational influences. Determine the constants ` and characterising the orbit.
Show that the rate at which time passes as measured by the satellite differs from the proper
time ( /c) measured by an observer at infinite distance by a factor
1/2
1 3m
= = 1 .
c t r
Compute this factor for a satellite at distance r = 2.66 104 km from the centre of the
Earth. Compute it also for the same satellite when it was being calibrated on the Earths
surface.
The GPS system gives highly accurate positional information for locations on the Earths
surface. In part, this depends on timing data sent between the satellites and the Earths
surface. Compute the total time discrepancy that accumulates over one day and estimate
the positional error that would be made if this was not accounted for in the operation of
the GPS system.
3. In the notes we obtained the equations describing geodesics of the Schwarzschild space-time
in the form
2m d ct d
1 = , r2 = `,
r d d
2
dr 2m `2 2m`2
+ 2 3 + 1 2 = 0,
d r r r
where and ` are constants, and with = /2 without loss of generality. Sketch the
form of the effective potential for different values of the angular momentum `. Show that
the effective potential has critical points (maximum and minimum) at values of the radial
coordinate given by
`2 12m2 1/2
r= 1 1 2 ,
2m `
and hence that there are no (stable) circular orbits if `2 12m2 .
Show that the maximum in the effective potential corresponds to the value V = 0 when
`2 = 16m2 . Non-relativistic particles with angular momenta less than this value therefore
fall into the black hole and are captured by it. Show that the capture cross section is
= 16m2 .
m 2
r = 1+ ,
2
and show that the metric then takes the form
m 2
1 2 m 4h 2
i
2 2 2
ds = m c dt + 1 + dx + dy 2 + dz 2 ,
1 + 2 2
72
where x, y, z are standard Cartesian coordinates for the spatial sections (t = const), with
x2 + y 2 + z 2 = 2 .
Show that the coordinate range m 2 < covers the region r 2m of the original
Schwarzschild coordinates. What region of the original Schwarzschild coordinates is covered
by the range 0 < m
2?
5. In 1964 Irwin Shapiro proposed a new solar system test of general relativity based on the
time taken to send radar signals to the inner planets Venus and Mercury and back.
This time should depend on the gravitational field that the signal passes through, so the
idea is to compare the time taken when Venus or Mercury are at different points of their
orbit so that the signal has a varying distance, b, of closest approach to the sun. Denote
the radial distance of the Earth from the sun by RE and that of Venus/Mercury by RP . We
may neglect the motion of either planet during the entire journey time of the radar signal.
(Why?) Equally, it suffices to take the trajectory of the radar signal as a straight line.
Using the Cartesian coordinate system of the previous problem, show that along the tra-
jectory of the radar signal
m 1 m 3
cdt = 1 1+ d`,
2 2
where d` is spatial distance in the Cartesian coordinate, and hence that the time taken
when the inner planet (Venus or Mercury) is near superior conjunction is given by
Z P
2m
ct 2 1+ d`,
E
q q
p p
=2 RE2 b2 + RP2 b2 + 2m arcsinh (R /b)2 1 + arcsinh (R /b)2 1 ,
E P
4RE RP
2 RE + RP + 2m ln .
b2
[For the last step use the formula arcsinh x = ln(x + x2 + 1), either with or without proof.]
The second term may be thought of as a delay due to the gravitational potential of the
sun. Estimate its magnitude. What is the analogous expression for the delay at inferior
conjunction?
In Shapiros original paper the time delay was calculated in Schwarzschild coordinates rather
than the Cartesian coordinates suggested here. Calculate the delay in Schwarzschild coor-
dinates.
[The variation in the delay as both Venus and Mercury moved between inferior and superior
conjunction was measured experimentally in 1967 and again in 1970. The latter results are
presented here.]
6. In 1936 Einstein published a short calculation on the Lens-like Action of a Star by the
Deviation of Light in the Gravitational Field, which is linked to here. This problem works
through that calculation. The figure (not to scale) illustrates the general setup: a star at
distance d from the observer acts as a lens for a more distant star, displaced off the direct
line of sight by a distance x. The deflection angle of light passing close to the lens is and
the observed angle of the lensed star, relative to the direction of the lensing star, is . Show
that the deflection angle is given by
b bx
= + ,
d D
73
and hence that the observation angle is (m is the usual parameter appearing in the
Schwarzschild metric)
2 1/2
x 4mD x
= + .
2(D + d) d(D + d) 2(D + d)
Why are there two values? What happens to the two images as x gets steadily larger?
Describe (qualitatively) what is seen when x 0 and estimate the magnitude of the effect
in this case.
2m 1/2
= 1 .
r0
If u = dx /d is the velocity of the observer (divided by c), show that their acceleration
(divided by c2 ) is
du du
a= = + u u ,
d d
and hence that the magnitude of the acceleration (divided by c2 ) is
2m 1/2
GM
1 .
c2 r02 r0
8. How large would a sphere of material with the same density as air have to be for its
Schwarzschild radius to exceed its own radius?
Were you to fall into a black hole, would you see the singularity at r = 0 once you crossed
the event horizon?
1 f d h i 1 f0 g
r 1 g 1 , 2 1 g 1 ,
R00 Rg00 = 2 Rrr Rgrr =
2 r dr 2 rf r
74
show that f and g are given by
2m 1 2
f = g 1 = 1 r .
r 3
Show that the time-like geodesics of this space-time, for the particular case = /2, are
characterised by
2m 1 2 d ct d
1 r = , r2 = `,
r 3 d d
2
dr 2m `2 2m`2 1
+ 2 3 r2 + `2 + 1 2 = 0,
d r r r 3
where and ` are constants. Thinking of the last of these in terms of an effective potential,
describe qualitatively the influence of the cosmological constant on planetary motion. Given
current estimates for the value of the cosmological constant of 1052 m2 , comment on
whether or not these effects are detectable in our solar system.
10. Allow for possible time dependence of the functions f and g in the metric for a spherically
symmetric space-time
h i
ds2 = f (r, t) c2 dt2 + g(r, t) dr2 + r2 d2 + sin2 () d2 .
Show that the Einstein equations imply that neither f nor g depend on time and hence the
only solution is still the Schwarzshild metric, i.e. establish Birkhoffs theorem.
11. It is often said that general relativity is a theory about the curvature of space-time; it is one
of its most evocative tag-lines. A large fraction of all gravitational phenomena that we have
been able to study may be understood in terms of properties of the Schwarzschild solution
this includes planetary orbits, deflection of light and lensing, gravitational redshift, Shapiro
delay, and many basic properties of black holes. In that solution one solves the vacuum
Einstein equations
1
R Rg = 0.
2
Show that R = 0 and hence that R = 0, i.e. that both the Ricci scalar and all components
of the Ricci curvature tensor vanish identically. Given this, what does it mean to say that
the Schwarzschild space-time is curved?
75
The active galaxy Centaurus A taken by the Hubble Space Telescope in 2011. The centres of
active galaxies, indeed all galaxies, are believed to contain supermassive black holes. Image
from NASA.
76
Chapter 5
It seems reasonable to try to attack the problem of stellar structure by the methods of
theoretical physics, i.e. to investigate the physical nature of stellar equilibrium.
77
stellar equilibrium, depends on its physical nature, which in our simple picture of things means
the equation of state relating its pressure, temperature and density. Now, I must confess that
I am not very knowledgeable about stars and their equations of state; certainly they are
tremendously complicated things and I am but a simple person. Nonetheless, it is still possible
to make generic statements and we will content ourselves with that.
When a star is young it burns hydrogen generating both heat and radiation that provide
the pressure that supports it against gravity. But what happens when the star is old and
cold? Forms of matter that we have some understanding of, such as atoms and nuclei the
things stars are made of , have a size, an Angstrom or a Fermi. This size is set by quantum
mechanics. The fundamental concept that underpins the structure and properties of simple
forms of condensed matter is the exclusion principle; a large body of fermions occupy quantum
states at most singly so that, at low temperatures, they fill up the available states to some
Fermi surface, characterised by a Fermi wavevector kF . If there are N fermions in a region of
size V then Z kF
V 2 V
N= 2
k dk = 2 kF3 . (5.1.4)
0 3
Assuming the fermions are non-relativistic, their energy will be roughly
Z kF
V 2 ~2 k 2 ~2 V 5 ~2 (3 2 )5/3 5/3 2/3
E' k dk = k F = N V , (5.1.5)
0 2 2m 10 2 m 10 2 m
and it follows that the pressure of such a degenerate Fermi gas is given by
E 32/3 4/3 ~2
P = ' (N/V )5/3 . (5.1.6)
V 5m
For stars that are composed of ordinary atoms, the degeneracy we have just described is
associated to electron states, so that the mass m is the electron mass and the number N is the
number of electrons in the star. This is the same as the number of protons, but only a fraction
of the total number of nucleons so that the total mass of the star is M = mH N where mH
is the mass of hydrogen and an average number of nucleons per electron in a typical aged
star, say about 2. This given, balancing the estimate of the degeneracy pressure against the
gravitational one requires that the stellar density should be roughly
125m3e (mH )5 G 3 2
' M . (5.1.7)
9 4 ~2
Stars that are able to maintain equilibrium conditions of this kind into their old age are known
as white dwarfs.
The density estimated here increases with the mass of the star; more massive stars are
more dense, meaning that the fermions they are made of are more tightly confined. If they are
confined tightly enough their typical momenta estimated from the uncertainty principle will
put them into a relativistic regime. The conclusion is then different, for their typical energy is
no longer the non-relativistic ~2 k 2 /2m but rather the relativistic estimate ~ck. The energy
of such a relativistic degenerate Fermi gas will be roughly
Z kF
V 2 ~cV 4 ~c(3 2 )4/3 4/3 1/3
E' k ~ck dk = k = N V , (5.1.8)
0 2 4 2 F 4 2
leading to an expression
(3 2 )1/3 ~c
P ' (N/V )4/3 , (5.1.9)
4
for the typical interior pressure of such stars. We may again assume electron degeneracy,
however, this time a balance against the gravitational requirements apparently leads to a
definite value for the stellar mass
3/2
3 ~c
M' (mH )2 . (5.1.10)
8 G
78
Substituting relevant values one finds 0.3M . This is not bad, but a little smaller than the
proper value. If one treats the hydrostatic equilibrium properly1 , for the degenerate Fermi gas
equation of state, one obtains a value about 5 times larger
known as the Chandrasekhar limit. Although the actual value is obviously important, the basic
concept and behaviour is correctly captured by our roughshod methods. There is an upper
bound to the mass of a star that can exist as a degenerate Fermi gas supported by electron
degeneracy pressure. The striking prediction that comes from this is that stars more massive
than the Chandrasekhar limit cannot possibly end their existence as white dwarfs; it is their
fate to suffer gravitational collapse to a different state of existence. If the mass is not too great
they will end up as a neutron star, their electronic content forcibly crushed by gravity into
recombination with protons to leave an extraordinary body of nuclear matter. The equation
of state of such a star may be even less well-known than that of white dwarfs, but it can still
be constrained by quantum mechanics and the degeneracy pressure that comes from being a
many-body state of fermions. The point is, that in the relativistic limit the degeneracy pressure
will again vary with the density as 4/3 , by precisely the same argument, and we again arrive
at an upper limit for the stellar mass, estimated this time at between 1.5 and 3 times the
solar mass. This is known as the Tolman-Oppenheimer-Volkoff limit. More massive stars are
unable to resist the force of gravity and presumably undergo continued gravitational collapse
until their outer surface recedes within their Schwarzschild radius at which point a singularity
is inevitable and a black hole is formed.
where M = mc2 /G is the mass of the black hole and J = aM c its angular momentum. When
the angular momentum a is zero it reduces to the Schwarzschild metric in Schwarzschild coor-
dinates. Unlike the Schwarzschild metric, the Kerr metric does not currently have a derivation
that can be widely considered physically intuitive, or peripatetic, or approachable at under-
graduate level. Even verifying that the metric, once given, is indeed an exact solution of the
Einstein equations is a lengthy and unenlightening process2 . I make no attempt here either to
motivate it, or verify it. However, for those who are interested, Roy Kerr has given a fascinat-
ing historical account of how he arrived at the rotating black hole metric [arXiv:0706.1109v2].
There is reason to believe the uniqueness theorems that the charged, rotating black hole
(Kerr-Newman metric) is the most general form of a stationary black hole. In other words,
we do not need to search for further generalisation of (5.2.1) (aside from allowing a non-zero
charge), and stationary black holes are described by just three numbers, their mass m, charge
1
Doing so leads to a Lane-Emden equation for a polytrope with index 3, whose solution identifies the correct
numerical prefactor.
2
Or so I am told; I have to confess that I, myself, have never gone through this calculation.
79
q, and angular momentum a. This is a strong statement; even if I do not justify it, I encourage
you to give it some thought3 .
80
The rotation of the black hole does not add fundamentally new terms to the effective potential
as compared to the Schwarzschild black hole, i.e. terms with different powers of 1/r, all it
does is alter the coefficients of the various different terms. It might seem that the effects of
the rotation are relatively minor; indeed the predictions of planetary orbits on the basis of the
Schwarzschild metric are already in good agreement with observations so that any modification
deriving from the rotation of the sun must indeed be minor. We leave a quantitative estimate
to the problems.
r2 2mr + a2 = 0. (5.2.8)
Whether there is a (real) solution or not depends on the value of the angular momentum a.
When |a| < m there are two solutions, andso two distinct surfaces on which grr is formally
divergent. The outer of these, r+ = m + m2 a2 , is the event horizon of the black hole;
massive, or massless, particles can cross it falling inwards, but can then never cross the same
surface again to get back out. As in the Schwarzschild space-time it is a null surface. When
|a| > m there are no real solutions and no coordinate singularities for r > 0. The curvature
singularity at r = 0 is, in this case, not hidden from a far-field observer, lying concealed behind
an event horizon, but sits openly in plain sight. It is a naked singularity. An open conjecture of
Roger Penrose cosmic censorship claims that such naked singularities are never produced in
any physical process, such as the gravitational collapse of a star. The borderline case |a| = m is
called an extremal Kerr black hole. If Penroses conjecture is correct it represents the maximum
value of the angular momentum that a black hole can acquire.
In the Schwarzschild metric two things happened at the event horizon; the metric compo-
nent grr diverged and, at the same time, the metric component g00 vanished. This is not so in
the Kerr metric. g00 vanishes on a different surface, defined by
and called the ergosphere. The region between the ergosphere and the event horizon is called the
ergoregion. To give an indication of why, consider the particle trajectory ct( ), r( ), ( ), ( ) =
(ct0 + , r0 , 0 , 0 ). This is not a geodesic, so it is not the (free) motion of a test particle,
but it is a special trajectory. It is an integral curve of a flow, generated by the vector field
K = (1, 0, 0, 0), that simply changes the t-coordinate and nothing else. As the metric does
not depend on t, it is unchanged under this flow, which corresponds to a symmetry of the
space-time. We say that the flow generates an isometry; the vector field K that generates it
is called a Killing vector. This vector, the velocity of our particle trajectory, has (squared)
magnitude g00 and so is only time-like outside the ergoregion. What this means is that this
trajectory can only correspond to the motion of an actual massive particle if it is outside the
ergoregion. What does the motion of a particle inside the ergoregion look like? Consider a
more general trajectory ct( ), r( ), ( ), ( ) = (ct0 + , r0 , 0 , 0 + ), corresponding to an
integral curve of the vector field K = (1, 0, 0, ). What is its magnitude? A short calculation
gives the answer
although it is more expedient to simply write g00 + 2g0 + g 2 . The vector is only time-like
if < < + where
g0 2 g00 1/2
g0
= . (5.2.11)
g g g
81
When g00 is positive both of these roots are positive (assuming a is positive) so that the
interval between them only includes strictly positive values of . Thus motion inside the
ergoregion, even for light, necessarily involves co-rotation with the black hole. It is impossible
to counter-rotate, or even to simply stand still.
To summarize, it seems that certainly some, and probably most, bodies of mass > ML will
eventually collapse within their Schwarzschild radius, and so give rise to a closed trapped
surface. There are at least 109 stars more massive than ML in our galaxy. Thus there are
a large number of situations in which theorem 2 predicts the existence of singularities.
This utterly striking theoretical prediction is eminently worthy of serious experimental atten-
tion and observational verification.
Observational evidence for black holes is indirect by necessity, for as we have seen it is
impossible to receive any signals from the event horizon, or the region of space-time within it,
so long as we stay safely in the asymptotic far-field. Thus we must infer; typically what we try
to infer is that a certain large amount of mass is confined to a certain small region of space
and yet cannot be seen directly as ordinary matter that we can otherwise identify. One then
invokes that we know of nothing that it could be other than a black hole, and we do know it
could be a black hole. On this basis, there is now widespread consensus that black holes are
common, as Hawking & Ellis said they should be.
The strongest evidence comes from observations of galactic nuclei; active nuclei and quasars
are amongst the most energetic phenomena known and supermassive black holes offer one
possible source for the enormous release of energy. Not only do they produce enormous amounts
of energy their luminosity can exceed 100 times that of the entire Milky Way but the source
is compact, with a size of only about a light year across. It is believed that their luminosity
derives from an accretion disc falling onto a supermassive black hole, with mass between 106
and 1010 M . The efficiency of this energy generation mechanism is impressive; of order 10% of
the mass is converted to light, which should be contrasted with the efficiency of nuclear fission
0.8%. More generally, observations of galactic nuclei and the behaviour of visible material
closest to them have led to the widespread view that all galaxies contain supermassive black
holes at their centres. Unsurprisingly, it is the one closest to us, in our own galaxy, that offers
the best observational evidence. The centre of the Milky Way is identified with a compact
radio source Sagittarius A . Over a period of about two decades the orbits of over 100 stars in
close proximity to the galactic centre have been monitored and those of at least 28 accurately
determined5 . One in particular, labelled S2, has an orbital period of 15 years, a pericentre
5
Numbers quoted from S. Gillessen et al. Astrophys. J. 692, 1075 (2009).
82
distance of only 1.81010 km, and has completed one full orbit during the period of observation.
The motion of all stars whose orbits have been determined are consistent with a single central
gravitating object with mass 4.3 106 M , localised within a distance certainly smaller than
the pericentre distance of the star S2.
Observational evidence for stellar-mass black holes derives primarily from x-ray observa-
tions of compact sources in binary systems. Typically, a visible star is seen in orbit with an
unseen, highly compact companion, which can be inferred to be a neutron star or a black hole.
The distinction between them depends on its mass; the Oppenheimer-Volkoff limit places an
upper bound on the mass of neutron stars, so that if its mass (significantly) exceeds this then
we may conclude that it is unlikely to be a neutron star and the presumption becomes that it
is a black hole. It is perhaps worth saying a few words about the observations that are made.
Orbital properties of the visible star are deduced from measurements of the Doppler shift of
its spectrum. These reflect variations in its line of sight velocity, the period of which gives
the orbital period P , and the semi-amplitude K gives the average component of the orbital
velocity projected onto the line of sight
2a sin(i)
K= , (5.3.1)
P
where a is the orbital radius of the visible star about the centre of mass and i is the angle
of inclination of the orbital plane as we see it on Earth6 . Using a Newtonian description of
binary orbits (2/P )2 = G(M + Mc )/r3 Keplers third law and the definition of centre of
mass M a = Mc ac , where Mc is the mass of the compact companion, one can then derive the
binary mass function
3
P K3 Mc sin(i)
= . (5.3.2)
2G (M + Mc )2
If in addition to the Doppler shift data one has estimates for the mass of the visible star M
and the angle of inclination i, then the mass of the companion can be deduced.
The first black hole candidate of this kind was the x-ray source Cygnus X-1, discovered in
the 1970s. Conservative estimates of its mass give a value of about 3.3 M that is only a little
above the Oppenheimer-Volkoff limit, but estimates that are not overly cautious, or pessimistic,
suggest a mass closer to 9 M . Thus Cygnus X-1 is a good candidate for a stellar-mass black
hole. Better estimates have come from so-called soft x-ray transients, which are strong x-ray
sources only for short intervals of time; during their quiet periods the accretion disc becomes
faint and accurate optical measurements of the visible companion can be made. These binaries
can have visible stars very much lighter than that in the Cygnus X-1 system, which is one
reason why they lead to better black hole candidates. One of the best such candidates is V404
Cygni where the compact source has a mass estimated at 12 2 M .
6
We have neglected the eccentricity of the orbit.
83
Problems
1. Derive the effective potential (5.2.7) characterising equatorial geodesics of the Kerr metric.
Find an explicit expression for elliptical planetary orbits in the form u = u(), where
u = 1/r, and comment on the changes associated to the non-zero rotation.
Find also an explicit expression for the deflection of light and again comment on the changes
associated to the non-zero rotation.
2. Although we do not show it, analogously to the Schwarzschild metric the Kerr metric has
a singularity at r = 0. However, the nature of this singularity is a little different and is
often referred to as a ring singularity. Show that, unless one approaches the singularity
in the equatorial plane = /2, the limit r 0 may be taken in the expression for the
metric. Subsequently taking /2 show that in any constant t slice the singularity has
the topology of a circle with circumference 2a.
3. Consider a crude model of a star where the density takes a constant value 0 out to some
radius a and is zero thereafter. For such a star, show that the Tolman-Oppenheimer-Volkoff
equation (5.1.2) becomes
dp 4Gr/c4 1 2
0 c p + 0 c2 ,
= 2 2
p +
dr 1 8G0 r /3c 3
and hence that the pressure at the centre of the star is given by
1 (1 2m/a)1/2 4G0 a3
p(0) = 0 c2 , where m= .
3(1 2m/a)1/2 1 3c2
Finally, show that the pressure at the centre diverges when m = 4a/9. [Stars that are
more massive than this cannot support themselves against gravity. This analysis is due to
Schwarzschild.]
4. In the notes we tried to estimate the maximum mass of a star that can support itself
against gravity. The result we obtained, 0.3M , is smaller than the value found by
Chandrasekhar, 1.44M , by enough to be considered significant. Read his own account of his
calculation in his Nobel Prize lecture, linked to here, and identify where our underestimate
comes from. [To say more, this is more than qualitative; the qualitative part is easy, the
difficult part is to determine accurately the numerical factor, which is of some importance
for observations. So the problem is a quantitative one; account for a discrepancy by a factor
of 5.]
5. Solutions of the Einstein equations for the interior region of a spherically symmetric star
can be obtained by writing the metric in the form
h i
ds2 = f (r) c2 dt2 + g(r) dr2 + r2 d2 + sin2 () d2 ,
and taking the stress-energy-momentum tensor to be that of an ideal fluid. Show that the
00 and rr Einstein equations are then
f d 1 8G 2
r 1 = c f,
r2 dr g c4
1 g 1 0 8G
2
2+ f = p g,
r r rf c4
respectively. Show that the solution of these equations can be written in the form
4G r
Z
1 2m(r)
=1 , m(r) = 2 (r0 ) r0 2 dr0 ,
g r c 0
m(r)c2 4r3 p(r)
2G 1
r ln f = 2 2 + .
c r 1 2m(r) G c2
r
84
The final linearly independent equation can be taken to be the continuity equation Tr =
0 for the stress-energy-momentum tensor. Show that it takes the form
1 0 2
r p + f c + p = 0,
2f
and hence obtain the Tolman-Oppenheimer-Volkoff equation (5.1.2) given in the notes.
85
The Cats Eye Nebula taken by the Hubble Space Telescope. The nebula is the result of a
series of large ejections of material from a massive star. Image from NASA.
86
Chapter 6
Gravitational Waves
87
a new choice y obtained by the near-identity transformation
x = y + , (6.1.2)
for functions that are small and slowly varying but otherwise arbitrary. In terms of our new
choice of coordinates the metric takes the form
x x
ds2 = g dy dy , (6.1.3)
y y
= + h + + dy dy ,
(6.1.4)
= + h + + + O(2) dy dy ,
(6.1.5)
suppressing terms of second order, or higher, in small quantities. In other words, a near-
identity coordinate transformation, which has no effect whatsoever on the space-time (only on
the way we view it), transforms the perturbation of the metric according to
h h + + . (6.1.6)
Remark: Another way of expressing this, which some of you may find preferable, is that, if
it happens that the perturbation is equal to the symmetric derivative of a function, h =
+ , then it is no perturbation at all but only a view of Minkowski using coordinates
that are close to inertial but not precisely so. A simple coordinate transformation then puts
things straight. One should say this more strongly; any part of the perturbation that is equal
to the symmetric derivative of a function can be eliminated by a coordinate transformation
and so carries no physical significance.
This preamble being said, let us proceed to determine the linearised form of the Ricci
tensor. To linear order, the Christoffel symbols are given by
1
= g g + g g ,
(6.1.7)
2
1
= h + h h + O(2), (6.1.8)
2
and it then follows that the components of the Ricci tensor, in our coordinate basis, are
R = + , (6.1.9)
1
= h + h h
2 (6.1.10)
1
h + h h + O(2),
2
1
= h + h + h h + O(2). (6.1.11)
2
This is the linearised Ricci tensor. It depends on second derivatives of the perturbation h ,
reflecting the fact that curvature is associated to second derivatives. It also means that constant
and linear terms in h carry no physical significance; they can be removed by coordinate
transformations.
One may verify easily that the components of the Ricci tensor are unchanged, to linear
order, under a gauge transformation h h + + , as they should be. Of course,
this reflects the fact that curvature is a directly observable physical quantity independent of
any choice of coordinate system. In particular, if the perturbation is pure gauge then the
space-time is Minkowski and is flat; the components of the Ricci tensor must then all be zero
regardless of the coordinate system.
88
6.2 The Graviton is a Spin 2 Particle
In electromagnetism, the gauge field A is a 1-form that allows for a manifestly Lorentz covari-
ant description of electromagnetic phenomena. Naively it conveys four independent compo-
nents at any point in space-time one for each value of the index . However, it is well-known
that the photon has only two linearly independent polarisation states. The description, there-
fore, necessarily contains redundancy. Part of this is gauge freedom the transformation
A A + changes the gauge field but has no effect whatsoever on any physically observ-
able quantities and part of it is more significant; the photon is a spin 1 particle but does not
have three distinct spin states because it is massless3 . In calculations, the redundancy is often
removed by imposing an additional condition on the gauge field; one that can always be sat-
isfied by freedom under gauge transformations. Any particular choice of additional condition
to impose is called gauge fixing. One of the most common choices is Lorenz gauge4
A = 0. (6.2.1)
This entire situation is repeated in the linear theory of gravity with perhaps the only difference
being that is it not so obvious, a priori, what the analogue of Lorenz gauge should be. We
study this now, in a presentation strongly inspired by Feynmans treatment.
Consider the form of the Ricci tensor for a plane wave h = e eiq x for some constants
e and a wavevector q . Away from masses the Ricci tensor vanishes and this leads to the
condition
1 2 1h
i
q e q q e + q q e q q e = 0, (6.2.2)
2 2
where, of course, q 2 = q q is the magnitude squared of the wavevector. Let us suppose
that q 2 6= 0. Then this equation can be rearranged into the form
1h 1
1
i
e = 2 q q e q e + q q e q e . (6.2.3)
q 2 2
But this is pure gauge. If h is pure gauge then we will have
h = + ,
(6.2.4)
e = q f + q f ,
for some choice of f . The choice f = q12 (q e 12 q e ) shows that h is pure gauge.
Thus plane wave modes for which q 2 6= 0 are always pure gauge and hence do not correspond
to anything physical. More positively, the only physical modes are those with q 2 = 0. Lets
add the fanfare that this result deserves; we have just shown that the graviton is a massless
particle.
When q 2 = 0 the vanishing of the Ricci tensor provides the constraint
h 1 i h 1 i
q q e e + q q e e = 0, (6.2.5)
2 2
which implies h 1 i
q e e = 0. (6.2.6)
2
Again, this deserves proper fanfare; we have just shown that the graviton is a transverse mode,
the same as the photon. Written in real-space this condition has the form of a divergence
h 1 i
h h = 0, (6.2.7)
2
3
The best description I know of this crucial part of (particle) physics is given in chapter 2 of Steven Weinbergs
book Quantum Theory of Fields, volume 1 (Cambridge University Press, Cambridge, 1996).
4
Lorenz gauge is named after the Danish physicist Ludvig Lorenz, not the Dutch physicist Hendrik Lorentz,
who is another person entirely. It would appear that confusion persists.
89
which is the analogue, for linearised gravity, of the Lorenz gauge condition in electromagnetism.
The discussion we have given of it led to the significant findings that gravitational waves
(gravitons) are massless and transverse, but it was also intended to bring out where this gauge
choice comes from, rather than simply plucking it from thin air.
The choice of Lorenz gauge does not completely fix the gauge freedom in the choice of the
metric perturbation h . Perhaps the easiest way to see this is through the Ricci tensor, since
it is an observable physical quantity. In Lorenz gauge the Ricci tensor reduces to
1
R = h , (6.2.8)
2
and is preserved by gauge transformations for which
= 0, (6.2.9)
in other words coordinate transformations that satisfy the wave equation. One may use this
additional freedom to impose an extra constraint on h and one that is frequently chosen is
that it is traceless,
h = 0. (6.2.10)
The choice of Lorenz gauge in which the metric perturbation is also traceless is called transverse
traceless gauge.
Perhaps the easiest way to see the final structure is to spell everything out in a specific
case. So lets take the wavevector q to be (k, k, 0, 0). The Lorenz gauge condition then reads
q e = 0, e0 + e1 = 0. (6.2.11)
The tensor is pure gauge if it is of the form e = q f + q f for some f . One may check
that each of the three tensors
1 1 0 0 0 0 1 0 0 0 0 1
1 1 0 0 0 0 1 0 0 0 0 1
,
1 1 0 0 ,
, (6.2.12)
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 1 0 0
is pure gauge. Finally, accounting for h being traceless we can write the gauge fixed form of
e for a plane wave perturbation as
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
e = h+
0 + h , (6.2.13)
0 1 0 0 0 0 1
0 0 0 1 0 0 1 0
and see that there are only two distinct polarisation states, both of which are transverse
traceless.
A polarisation state of the electromagnetic field is a direction in space (orthogonal to the
direction of propagation) along which the electric field points, and oscillates. A polarisation
state of a gravitational wave is a structure for the deformation of the metric in directions
orthogonal to the direction of propagation of the wave. These two deformations are
The former represents an extension of the arc length along y and equal compression along z.
The latter is the same distortion, but rotated by /4. A rotation of the coordinate system5
will mix the two polarisation states, in the same way that such a rotation mixes the two
5
In doing this it is important that we do not simultaneously rotate the gravitational wave since otherwise
we will have only succeeded in doing nothing at all.
90
linear polarisation states of the electromagnetic field. We compute this as follows. Under a
rotation through angle about the x-axis the coordinates y and z transform according to
y y cos() z sin(), and z y sin() + z cos(). (6.2.15)
It then follows that
dy 2 dz 2 cos(2) dy 2 dz 2 sin(2) 2dydz,
(6.2.16)
and we see that the rotation between the two polarisation states is at twice the rate of the
rotation. This was surely evident from the picture of the two polarisation states; we show it
algebraically purely for completeness. This behaviour is what is meant when one says that
gravitons are spin 2 particles.
where T is the Fourier transform of the stress-energy-momentum tensor, k = |k|, Gret (x, t)
is the retarded Green function for the wave operator, and we have made use of the convolution
theorem. The appearance of the retarded Green function comes from physical considerations
of the relevant boundary conditions; the matter content is a source for the gravitational radi-
ation and a compact source should produce outgoing waves rather than ingoing waves. The
gravitational disturbance produced by a compact source can therefore be given as
16G 1
Z
dt0 d3 x0 (t t0 ) |x x0 |/c T (x0 , t0 ),
h = 4 0
(6.3.6)
c R1,3 4|x x |
0 0
3 0 T (x , t |x x |/c)
Z
16G
= d x . (6.3.7)
c4 R3 4|x x0 |
6
In writing this we should note that the stress-energy-momentum tensor appearing on the right-hand-side is
also a leading order approximation, consistent with that taken for the curvature terms. We do not describe this
in any detail.
91
As a first application, we look at sources of gravitational waves. We would like to know
what properties the source should have so as to generate them. Gravitational waves, as we
have seen, have a transverse character and are associated with the spatial components hij of
the perturbed metric. If we focus on these components and look at distances far from the
source, |x| |x0 |, then
Z
4G
hij = 4 Tij (x0 , t |x|/c) d3 x0 . (6.3.8)
c |x| source
The integral over the source can be manipulated with the help of the continuity equation for
the stress-energy-momentum tensor, T = 0 or
1
t T0 = k Tk , (6.3.9)
c
to show that the source of gravitational radiation, in conformance with the nature of gravita-
tional waves, is characteristic of a mass quadrupole. To see this, first note that
Z Z
Tij d3 x = k xj Tik d3 x,
(6.3.10)
Z
= xj k Tik d3 x, (6.3.11)
1
Z
= t xj Ti0 d3 x. (6.3.12)
c
Since the stress-energy-momentum-tensor is symmetric, T = T , the same expression holds
with indices i and j interchanged. This observation allows us to proceed as follows
1
Z Z h i
Tij d3 x = t xj Ti0 + xi Tj0 d3 x, (6.3.13)
2c
1
Z h
i
xj k xi T0k + xi k xj T0k d3 x,
= t (6.3.14)
2c
Z
1
= t xi xj k T0k d3 x, (6.3.15)
2c
Z
1 2
= 2 t xi xj T00 d3 x. (6.3.16)
2c
To effectively generate gravitational waves a source should have a large moment of inertia
that is rapidly changing. Individual spinning stars, or other objects that are axisymmetric, are
therefore not good sources. Binary systems are rather better. As a crude description consider
92
a binary system of two equal mass stars with a circular orbit about their common centre of
mass of radius a and period 2/. The 00-component of the stress-energy-momentum tensor
can then be taken to be
h i
T00 = M c2 (x) y a cos(t) z a sin(t) + y + a cos(t) z + a sin(t) , (6.3.18)
choosing the orbital plane (yz) to orient the spatial coordinate system x, y, z. We then find
8GM 2 a2
hyy (x, t) = hzz (x, t) = 4
cos 2(t |x|/c) , (6.3.19)
c |x|
8GM 2 a2
hyz (x, t) = 4
sin 2(t |x|/c) . (6.3.20)
c |x|
Such a binary system is therefore expected to act as a source of gravitational waves that go
out in all directions with frequency twice that of the source and amplitude 8GM 2 a2 /c4 |x|
for both polarisation states, i.e. the radiation is not (completely) polarised7 .
The first observation of a binary system that could be studied carefully and accurately
and compared to these predictions for producing gravitational waves came in 1974 with the
discovery by Hulse and Taylor of the first binary neutron star system, PSR 1913 + 16, or,
informally, the Hulse-Taylor binary. It is a binary pulsar with a main star of mass 1.44M
and companion of mass 1.39M in a close orbit with semi-major axis of 1.95 106 km (about
3 times the radius of the sun) and eccentricity e = 0.62. The orbital period is only 0.32 days.
As a result it exhibits strong gravitational effects and provides an excellent test of general
relativity in a strong field regime. In particular, it should be a source for gravitational waves.
The distance of the Hulse-Taylor binary from the sun is about 21,000 light years, or 2 1017
km. It follows that the typical amplitude of gravitational waves produced by this source, when
they arrive at the Earth, can be estimated at about
8GM 2 a2
h 1.8 1022 . (6.3.21)
c4 |x|
This is tiny! The challenge for experimental groups to design and manufacture any sort of
instrument that could detect signals of such tiny amplitude cannot be overstated. And yet,
this is precisely what they have succeeded in doing. What a remarkable achievement!
The gravitational waves produced by binary neutron stars, such as the Hulse-Taylor binary,
have not yet been observed directly, but it is still possible to infer that gravitational waves
are being produced from the exquisite timing data that has been gathered over the past 40
years. The point is that gravitational waves carry energy so that their emission leads to a
loss of energy for the orbital motion of the two neutron stars and consequent slow inspiral.
In order to determine the change in orbital parameters we need to know the energy content
of the gravitational waves that are created. The analogous problem in electromagnetism is
quite straightforward you know about the energy density in the electromagnetic field and
the Poynting vector that describes its flux but for gravity it is a little more involved and we
will not present any of the analysis8 . A form that is especially convenient for comparison with
astrophysical observations was developed in the early 1960s by Peters and can be summarised
by the expressions9
64 G3 M1 M2 (M1 + M2 )
da 73 2 37 4
= 1 + e + e ,
dt 5 c5 a3 (1 e2 )7/2 24 96
(6.3.22)
304 G3 M1 M2 (M1 + M2 )
de 121 3
= e+ e ,
dt 15 c5 a4 (1 e2 )5/2 304
7
For details of the directional dependence see, as usual, Landau & Lifshitz.
8
This was originally done by Einstein as part of his initial formulation of general relativity and an account
can be found in, say, his 1916 review; it is also described in some detail by Landau & Lifshitz.
9
These are copied from P.C. Peters, Phys. Rev. 136, B1224 (1964).
93
for the average rate of change of orbital parameters, where M1 , M2 are the two stellar masses,
a is the semi-major axis, and e the eccentricity of the orbit. The observations of the Hulse-
Taylor binary fall on the general relativity prediction to a fractional accuracy within about
0.2%; for this result Hulse and Taylor shared the 1993 Nobel Prize. Since the discovery of
the Hulse-Taylor binary, several other binary neutron star systems have also been found and
observations of changes to their orbital parameters compared with the predictions of general
relativity. The best current test is for the binary pulsar J07373039 where agreement with the
predictions of general relativity is as good as 0.05% for some measurements.
94
Problems
1. One factor contributing to the feebleness of gravitational waves is the distance to the source,
which can often be thousands of light years (21,000 for the Hulse-Taylor binary). The closest
source is clearly the sun. Why do we not detect gravitational waves produced by the sun?
2. For circular orbits of two equal mass stars (mass M ), Peters formula (6.3.22) for the rate
of change of separation reduces to
da 128G3 M 3
= .
dt 5c5 a3
If the initial separation is a0 , show that the time taken for complete inspiral of the two stars
is
5a40 c5
t= .
512G3 M 3
Calculate this time for two neutron stars, each with 1.4M , starting from an orbital period
of 2 hours.
The LIGO experiment is expected to detect the final inspiral of binary neutron stars after
the gravitational waves they emit exceed a frequency of about 50 Hz (seismic noise makes
detection difficult at lower frequencies). Estimate how long a neutron star merger event
detected by LIGO will last for. Would a pair of merging black holes with equal masses of
10M produce a longer or shorter event?
Qualitatively describe the nature of the gravitational wave signal that will characterise such
mergers, at least while the two stars can be treated as point masses.
3. The only currently known double pulsar (binary system in which both stars are neutron
stars from which we receive radio pulses) was discovered in 2003 and is called PSR J0737-
3039. The masses of the two stars are 1.35M and 1.24M and the period of their orbit is
2.4 hours. The system is estimated to be between 3200 and 4500 light years from Earth.
Estimate the amplitude of the gravitational waves produced by this system when they arrive
at the Earth.
Use Peters formula for the rate of change of the semi-major axis (setting the eccentricity
to zero)
da 64G3 M1 M2 (M1 + M2 )
= ,
dt 5c5 a3
to determine how long it will be before the orbit completely shrinks and the neutron stars
merge. By how much does the orbit shrink per day?
Estimate the general relativistic contribution to the precession of the perihelion in PSR
J0737-3039.
4. If the sun was separated into two equal parts in circular orbit around each other with radius
twice that of the sun (R = 6.96105 km), what would be the amplitude of the gravitational
waves at the location of the Earth? What would the frequency of the gravitational waves
be?
T (x0 ) 3 0
Z
4G
h = 4 d x,
c |x x0 |
for the perturbation to the metric created by some matter distribution, here considered
to be stationary, i.e. independent of time. Taking for the stress-energy-momentum tensor
95
the expression for a perfect, pressure-less fluid, T = c2 u u , and with a non-relativistic
velocity u = (1, v i /c), show that
(x0 ) 3 0
Z
4G 4
h00 = 2 0
d x 2,
c |x x | c
4G
Z 0 i
(x )v 3 0
h0i = 3 d x i ,
c |x x0 |
(x0 )v i v j 3 0
Z
4G 2
hij = 4 d x = O |v/c| ,
c |x x0 |
has an interpretation as the Newtonian potential (it is defined by the same formula as
Newton gave). What is the physical interpretation of the vector i ?
96
The Ultra Deep Field image was taken by the Hubble Space Telescope in 2014. Almost
everything you see in this image is a galaxy, some of them the most distant ever imaged. The
few stars are easy to identify from their characteristic diffraction spikes. Image from NASA.
97
Chapter 7
Cosmology
This singularity is the most striking feature of the Robertson-Walker solutions. It occurs in
all models in which + 3p is positive and negative, zero, or with not too large a positive
value. It would imply that the universe (or at least that part of which we can have any
physical knowledge) had a beginning a finite time ago.
The solutions of the Einstein equations we have looked at so far have all been interpreted in
terms of local gravitational problems; planetary orbits, lensing by a massive object, stellar
collapse, individual black holes, and so on. The metric that arises as a particular solution of
the Einstein equations is then taken to represent a decent approximation of some local region
of the universe, for instance our own solar system. One can also conceive of an interpretation
of suitable solutions of the Einstein equations as providing a metric that describes the entire
universe, at some level. The study of the universe on its largest scales is called cosmology,
and solutions of the Einstein equations used to interpret it are known as cosmological models.
Their study originated with Einstein, in 1917.
whereas the anti-de Sitter space AdS4 is the subset of R2,3 a time-like distance a0 from the
origin
AdS4 : (x0 )2 (x1 )2 + (x2 )2 + (x3 )2 + (x4 )2 = a20 . (7.1.2)
Both are isotropic, homogeneous, four-dimensional Lorentzian manifolds with constant scalar
curvature. They are also exact solutions of the Einstein equations, provided a certain inter-
pretation is adopted. To see this, introduce coordinates (ct, , , ) for dS4 such that
98
After a wearisome calculation one arrives at expressions for the components of the Ricci tensor
in this coordinate basis. They may be conveniently summarised as R = (3/a20 )g , from
which it follows that the Ricci scalar is R = 12/a20 ; de Sitter space is a space-time of constant
positive scalar curvature. Now, for the Einstein equations to be satisfied we must have
1 8G 3c4
R R g = 4 T , T = g . (7.1.5)
2 c 8Ga20
In other words, the stress-energy-momentum tensor is a constant negative multiple of the
metric. This corresponds to a perfect fluid, T = (c2 + p)u u + pg , with constant negative
pressure, equal in magnitude to the constant positive energy density (p = c2 ). Ordinary
matter does not have this property, so this is not the stress-energy-momentum tensor of any
ordinary form of matter.
Historically, a term of this form a constant multiple of the metric was introduced into
the field equations by Einstein as part of his static universe cosmology; the constant prefactor
was referred to as the cosmological constant and denoted . It was not originally introduced as
a contribution to the stress-energy-momentum tensor, as it is now viewed, but as a modification
of the Einstein equations to
1 8G
R R g + g = 4 T . (7.1.6)
2 c
Nowadays, one moves it to the right-hand-side and interprets a contribution of the form
T = (c4 /8G)g to the stress-energy-momentum tensor as arising from non-ordinary
matter and calls it dark energy. Modern observations are compatible with simple cosmologi-
cal models based on the Einstein equations only if one assumes that a substantial fraction of
the total stress-energy-momentum tensor comes from dark matter (26.8%) and dark energy
(68.3%)1 . Such cosmologies, dominated by the cosmological constant, suggest that our uni-
verse will evolve to become increasingly like de Sitter space in the future. Despite the fact that
modern observations suggest that only 4.9% of the stress-energy-momentum tensor comes from
ordinary forms of matter that we know and understand, it is evident that one must study, and
understand, cosmological models in which the stress-energy-momentum tensor corresponds to
ordinary forms of matter. This is what we turn to now.
where d`2 is the metric for an isotropic, homogeneous, three-dimensional Riemannian manifold
of constant scalar curvature. There are precisely three of these, corresponding to the three
possibilities of positive scalar curvature, zero scalar curvature and negative scalar curvature.
These are, respectively, the 3-sphere, three-dimensional Euclidean space and three-dimensional
hyperbolic space H3 . The metric is commonly expressed in either of the forms
2 2 2 2 2
d + sin () d + sin ()d ,
d`2 = d2 + 2 d2 + sin2 () d2 , (7.2.2)
2
2 2 2 2
d + sinh () d + sin () d ,
dr2
d`2 = + r2 d2 + sin2 () d2 ,
2
k = +1, 0, 1. (7.2.3)
1 kr
1
These figures are derived from observations of the cosmic microwave background by the ESA Planck satellite,
released in 2013.
99
We will use the former, together with the symbol k = +1, 0, 1 to distinguish the three different
cases for the scalar curvature. Any metric of the form (7.2.1) is called a Friedmann-Lematre-
Robertson-Walker space-time.
Metrics of this form, and the cosmology they give rise to, were first studied in 1922 by
the Russian mathematical physicist Alexander Friedmann. The same results were obtained,
independently, by the Belgian priest and scientist Georges Lematre in 1927, who was also the
first to point out that the expansion factor a(t) could account for the observed redshifts of
distant galaxies, obtain the expression that subsequently became known as Hubbles law and
determine the present value of the Hubble constant from experimental data. The American
mathematical physicist Howard Robertson and British mathematician Arthur Walker made
their (independent) contributions in the mid 1930s. Their work established that the two
conditions of spatial isotropy and homogeneity were sufficient to imply that the metric could
always be written in the form (7.2.1), i.e. there is no loss of generality.
The assumption of spatial isotropy and homogeneity appears to be in reasonable accord
with observations of the universe on the largest scales. Every direction we look in we see
galaxies and clusters of galaxies; at least this is the impression one immediately gets from
looking at the Hubble Deep Field images. Of course we do not see galaxies in every direction,
nor do we see precisely the same galaxy, or cluster of galaxies, in every direction. There is
variation; some galaxies are spiral, some elliptical; some parts of any image are perfectly dark,
others are bright with the light of a star or galaxy. But on average the night sky looks
the same in all directions. Moreover, this is true at all wavelengths. In addition to optical
measurements we have been surveying the night sky in infrared, microwave, radio and x-ray
parts of the electromagnetic spectrum. In each case one can say the same; on average the sky
looks the same in every direction. Thus the FLRW cosmology has come to be the standard
cosmological model, used for the interpretation of the majority of observations. At the same
time, it is important to bear in mind and acknowledge that the universe is neither isotropic
nor homogeneous in a strict sense; it is patchy; there are galaxies and clusters of galaxies, and
within galaxies there are stars and planets and so forth. The point is that this patchiness can
be studied and one can try to learn new things from it. A major topic, of high current interest,
is the inhomogeneity, or fluctuations, in the cosmic microwave background. Nonetheless, here
we will content ourselves with a study of the isotropic, homogeneous FLRW cosmologies.
The cosmology envisaged by Friedmann is a solution of the Einstein equations for the metric
(7.2.1), with the stress-energy-momentum tensor of a perfect fluid. One finds by a tedious, but
straightforward, calculation that
1 3h 2 i
R00 R g00 = 2 c1 t a + k , (7.2.4)
2 a
1 1 h 2 i
Rij R gij = 2 2ac2 t2 a + c1 t a + k gij . (7.2.5)
2 a
Equating this to the stress-energy-momentum tensor for a perfect fluid, T = (c2 + p)u u +
pg , in its rest frame, yields
2
a kc2 8G 2
+ 2 c = 0, (7.2.6)
a a 3c2
a
4G 2
+ c + 3p = 0, (7.2.7)
a 3c2
where a = t a. These are known as the Friedmann equations. Sometimes the second is called
separately the acceleration equation. Either from these, by taking the time derivative of the
first equation, or from the continuity of the stress-energy-momentum tensor, T = 0, one
can derive the fluid equation
3a
t c2 + c2 + p = 0.
(7.2.8)
a
100
The Friedmann equations do not yield a closed system; one needs to give also an equation of
state for the fluid2 . We will consider a couple of approximations, or idealised equations of state,
corresponding to a dust dominated universe, where the pressure is negligible compared to the
energy density (p negligible compared to c2 ), and a radiation dominated universe, where the
pressure is taken to be one-third the energy density3 (p = 13 c2 ). But before this, some general
comments can be made.
Regardless of the equation of state for the fluid, the main feature of the Friedmann equations
is that the scale factor a is not a constant4 , i.e. the space-time is not static and space is
either expanding or contracting. This is an awe-inspiring prediction. It can, of course, be
tested against observation. Observations of distant galaxies reveal that they exhibit a generic
redshift in their spectral lines. Moreover, the value of the redshift depends only on the distance
of the galaxy from us, matching with the assumption that the universe appears isotropic to us
on the largest length scales. If the spectral redshift is attributed to a Doppler shift, then we
would say that all5 other galaxies are moving away from us, and those that are more distant
are receding faster. It was Georges Lematre who first recognised that these observations are
reproduced by the time-dependence of the scale factor a in the FLRW metric (7.2.1); the
universe is expanding isotropically. He recognised more than this; if the universe is expanding,
a > 0, and c2 + 3p is non-negative, then (7.2.7) shows that a 0, and the universe must
have been expanding at its present rate, or faster, in the past. So in the past the universe was
smaller, and smaller still. Extrapolating backwards its size must have been vanishingly small
at a time of no more than (a/a)| now H 1 long ago. This is the concept of a primevial atom
from which the universe originated, first introduced by Lematre in 1931; the name Big Bang
was coined by Fred Hoyle in 1949. It is perhaps the most striking feature of simple isotropic
homogeneous cosmologies.
To say more, consider a galaxy that at some instant in time, t = 0 say, is at the spatial
point (0 , 0, 0) and is travelling along a time-like geodesic6 . It is not difficult to verify, from
the geodesic equation, that its trajectory is
ct( ), ( ), ( ), ( ) = , 0 , 0, 0 . (7.2.9)
What is important here is that the spatial point where the galaxy is located is not changing;
it is at rest7 . Now consider two such galaxies, one at the spatial point (0 , 0, 0) the galaxy
we are observing, M31 for the sake of giving it a name and one at the spatial point (0, 0, 0)
our own galaxy, where we are observing from. When we look at M31 we see light that left it
and travelled to us along a radial null geodesic. Its trajectory therefore satisfies
the minus sign coming because it is travelling inwards from = 0 to = 0. The total time
2
For instance specifying the pressure as a function of density. As you will surely know, the equation of state
for a simple fluid or ideal gas is typically a relation between three thermodynamic state variables, not just two.
The third is the temperature or entropy. In such circumstances one then needs in addition an entropy balance
equation.
3
The reason for this is that then the stress-energy-momentum tensor is traceless as it is for the electromagnetic
field, however, one should say that this is only a proxy and certainly in the radiation dominated model we do
not employ the correct stress-energy-momentum tensor for the electromagnetic field, nor self-consistently solve
Maxwells equations.
4
Indeed, there are no solutions for constant a without a cosmological constant; this was Einsteins motivation
for introducing it.
5
Not quite all, some of those closest to us in the local cluster are moving towards us. But if we average
over length scales large compared to intergalactic distances then the observations are the same for all galaxies.
Remarks of this kind are implicit in any statement that the universe appears isotropic and homogeneous on
large scales.
6
It is a test particle for the cosmological metric.
7
You should recall that this condition was also used in deriving the Friedmann equations; the stress-energy-
momentum tensor for the fluid was given in its rest frame.
101
taken is given implicitly by Z tobs
cdt
0 = , (7.2.11)
temit a(t)
and should be thought of as an expression for the observation time tobs as a function of the
emission time temit . Differentiating with respect to the emission time we obtain
c tobs c
0= , (7.2.12)
a(t) t=tobs temit a(t) t=temit
or, since the derivative tobs /temit is the ratio emit /obs of the frequencies of the emitted
and observed light,
emit a(tobs ) obs
= = = 1 + z, (7.2.13)
obs a(temit ) emit
where z is the redshift. Thus the redshift is given by the difference between the value of the
scale factor a at the time the light is observed and the time the light is emitted. The expansion
of the universe accounts for the observed redshift of galactic spectra.
If the galaxies are close to each other, so that temit = tobs t with t small, then we may
approximate
a(tobs ) a(tobs ) a a(tobs )0
1+ t 1 + H(tobs ) , (7.2.14)
a(temit ) a(tobs ) t a(tobs ).t a tobs c
It follows that as the universe expands the energy density decreases, the energy density asso-
ciated with radiation decreasing more rapidly than that associated with dust, or the matter
content. Of course, going backwards in time to when the universe was very much smaller,
the converse is true and radiation would have dominated the energy content. A picture that
emerges is that at early times the universe was small and the main contribution to the energy
density was from (electromagnetic) radiation, but as it expanded that contribution died out
more rapidly than the energy density of the dust so that at some point there was a cross-over
8
Note that this is not the same as geodesic distance, but the difference is small under the assumptions made.
102
to a matter dominated regime in which the galaxies could form. The discovery of the cosmic
microwave background radiation by Penzias and Wilson in 1964 is interpreted as a remnant
of this transition from the early universe and rightly considered one of the most significant
observations of the evolutionary history of our universe.
Substituting the expression for the energy density into the Friedmann equation gives for
the scale factor
a 2 = Ka(1+3w) kc2 , (7.2.17)
where K is a constant that we do not give explicitly. In the dust filled regime, w = 0, the
solution is
c3
K 2 1
c 2 sin ((t)), 2 sin(2) = K t k = +1,
3 1/2 2/3
a(t) = 2K t k = 0, (7.2.18)
3
sinh2 ((t)), 12 sinh(2) = cK t k = 1.
K
c2
The most striking feature of the solution is that the scale factor increases without bound if the
universe is spatially flat (k = 0) or hyperbolic (k = 1) but not if there is positive curvature
(k = +1). If the spatial sections are 3-spheres then the universe expands to a maximum size
amax = K/c2 at a time tmax = K 2c3
after which it begins to contract until it eventually shrinks
back to nothing after a time tend = K c3
. This fate for the universe is sometimes called the
Big Crunch. Naturally, it is interesting to know the spatial curvature of our universe; current
observations suggest that it is flat, k = 0, but it is understandably very difficult to tell.
103
modified to read
2
a kc2 8G 2 1 2
+ 2 c c = 0, (7.3.2)
a a 3c2 3
a
4G 1
c2 + 3p c2 = 0,
+ (7.3.3)
a 3c2 3
while the fluid equation remains unchanged9 . A large enough value of the cosmological constant
allows for positive acceleration, a
> 0. If one neglects the matter density compared to the
cosmological constant, then, for a spatially flat universe k = 0, the expansion is exponential
a(t) = a(0) e /3 ct . (7.3.4)
This solution should not be that surprising, given how we started this chapter with a discussion
of the de Sitter universe. Indeed, setting k = +1, one obtains the de Sitter metric (7.1.4) with
a0 = (3/)1/2 . Since in this case (3/)1/2 is the radius of curvature, a tiny value for the
cosmological constant makes it exceedingly difficult to tell if spatial sections of the universe
are flat (k = 0) or curved (k = 1).
The -CDM cosmology in which the universe started with the Big Bang, was initially radi-
ation dominated until the recombination transition that left the cosmic microwave background
radiation, was then dominated by the matter content (dust) for a period and then more re-
cently by the cosmological constant, or dark energy, summarises the current view of the history
and nature of our universe.
9
So long as the cosmological constant does not depend on time.
104
Problems
1. The fluid equation in the FLRW cosmology can be described in terms of classical ther-
modynamics. As the universe is homogeneous and isotropic there can be no temperature
gradients and therefore no flow of heat, so that the first law of thermodynamics is simply
dU = p dV.
2. The universe at the current time is dominated by matter and the cosmological constant, with
the contribution of radiation negligible. Show that Friedmanns equation for the current
epoch can be written as
h i1/2
H(z) = H0 M (1 + z)3 + M + 1 (1 + z)2
,
where H(z) is the Hubble constant at redshift z, H0 is its present day value, and M ,
are the ratios of the current matter density and cosmological constant density to the critical
density, defined by crit = 3H02 /8G.
Which term in the relation above represents the spatial curvature of the universe?
In the lectures we made use of the relation
Z
c dt
= ,
a(t)
to find that the galactic redshift z was related to the scale factor a(t) by
a0
1+z = ,
a(t)
where a0 is the current value of the scale factor. By differentiating the latter relation, show
that
c dz 0
Z z
a0 = 0
.
0 H(z )
Hence show that a0 is a monotonically increasing function of . How does this relate to
the use of supernovae in cosmology?
3. Use the fluid equation to show that a cosmological fluid containing only dark energy (no
ordinary matter) has constant energy density. What does this imply about the homogeneity
of the space-time? Does it conflict with the scale factor a(t) having time-dependence?
4. For a universe with only cosmological constant (dark energy, no ordinary matter) solve the
Friedmann equations to find the possible forms that the universe may take and describe
their evolution. You should distinguish the cases > 0 and < 0, and consider all possible
spatial geometries.
[Hint: do not try to find the most general solution, but rather a (representative) solution
for each possibility, if such exists.]
105
5. Einsteins static universe is the cylinder
of the Minkowski space R1,4 with its usual metric. Show that it is an exact solution of the
Einstein equations with cosmological constant and perfect fluid form for the stress-energy-
momentum tensor (dust, no pressure).
[Hint: parameterise the space and determine the metric in your local coordinates; then
compute from it the components of the Ricci tensor; finally solve the Einstein equations.
This is easier than it sounds as if you make a sensible choice for the coordinates you will
find that the metric is of FLRW form, for which you have been given the calculation of the
Ricci tensor and solution of the Einstein equations in the notes.]
6. Show that the anti-de Sitter space is an exact solution of the Einstein equations with
cosmological constant and no ordinary matter content.
Show that there are closed time-like curves in the anti-de Sitter space.
106
Solutions to selected problems
I hope over time to include sketched solutions to as many of the end of chapter problems as I
can. This is a laborious process but I will try to get enough solutions written up quickly, so as
to be useful. As always, comments, suggestions, corrections, or help with producing solutions
are all most welcome. It is worth reiterating that the end of chapter problems are deliberately
designed to be distinct, both in content and style, from past exam problems, so as to minimise
overlap and retain the latter as a useful repository for exam preparation.
Chapter 1 Gravity
problem 1
The Kepler problem considers two masses m and M moving under their mutual gravita-
tional interaction. There is an implicit assumption that m is the mass of a planet, and may
be considered small, while M is the mass of the sun (or star), and may be considered large,
although the analysis is certainly not restricted to this scenario. If x1 , x2 denote the positions
of the two masses in some inertial frame, then Newtons law of motion reads
d2 x1 GM m x1 x2 d2 x2 GM m x2 x1
m 2
= , M = .
dt |x1 x2 |2 |x1 x2 | dt 2 |x1 x2 |2 |x1 x2 |
Adding, we find
d dx1 dx2
m +M = 0,
dt dt dt
with the interpretation that the total linear momentum is conserved. Dividing by m or M , as
appropriate, and subtracting we find
d2 (x1 x2 ) G(M + m) x1 x2
= ,
dt2 |x1 x2 |2 |x1 x2 |
The term in square brackets is the instantaneous relative velocity, which lies in a plane spanned
by the orthogonal vectors n and dn/dt (dont forget n is a unit vector). Its change is purely in
the direction n, which lies in this plane, and so the motion continues to lie in this same plane.
Introducing a standard basis {e1 , e2 } for this plane, and polar coordinates (r, ), we may write
x1 x2 = rn = r e1 cos() + e2 sin() ,
107
where n = sin() e1 + cos() e2 . By linear independence, the coefficients of n and n must
vanish separately. Taking the latter first we find
d2
dr d 1d 2 d
0=2 +r 2 = r ,
dt dt dt r dt dt
Finally, describe the motion by u = 1/r and think of u as parameterised by angle rather
than time t. We have
dr d du1 du
= = ` ,
dt dt d d
2 2
d r d d du 2 2d u
= ` = ` u ,
dt2 dt d d d2
d2 r `2
2
G(M + m) 2 2 d u G(M + m)
0= 2 3 + = ` u +u .
dt r r2 d2 `2
The solution is
G(M + m) h i
u= 1 + e cos() ,
`2
where e is the eccentricity. This is the equation of an ellipse with the centre of our coordinate
system (r, ) at one of the foci; Keplers first law. With this solution one can compute the
orbital period from the equation r2 d/dt = `
2
`3 2`3
Z
d
T = = .
0 G2 (M + m)2 (1 + e cos())2 G2 (M + m)2 (1 e2 )3/2
The integral can presumably be done a number of ways; I used contour methods. The semi-
major axis of the orbit is `2 /G(M + m)(1 e2 ) so that the square of the orbital period is
proportional to the cube of the semi-major axis, Keplers third law of planetary motion.
A Lagrangian approach is also insightful, especially given the method used in general
relativity where the planetary orbits are described using time-like geodesics of the Schwarzschild
metric. The Lagrangian is
2
dx2
2
m
dx 1
M
+ GM m ,
L=
+
2 dt
2 dt
|x1 x2 |
2
d(x1 x2 )
2
M + m
d mx 1 + M x 2
M m
+ GM m ,
=
+
2
dt M + m
2(M + m)
dt
|x1 x2 |
2 2 2
M + m
d mx1 + M x2
+ Mm dr d GM m
+ r2
= + ,
2
dt M + m
2(M + m) dt dt r
d
writing the relative position vector as x1 x2 = rn as before. It follows that dt (mx1 + M x2 )
M m 2 d
is a constant, as is the magnitude of the orbital angular momentum M +m r dt . [We implicitly
108
used the constancy of its direction in writing the Lagrangian in the form given above.] Calling
the latter ` we can the total energy of the relative motion as
2
`2
Mm dr GM m
+ 2 = E,
2(M + m) dt r r
which is a constant as the Lagrangian has no explicit time dependence. This energy equation
can be used to characterise the nature of planetary orbits circular, bound elliptical, unbound
hyperbolic without having to find the explicit form of the orbits. It should be contrasted with
the analogous expression obtained for the time-like geodesics of the Schwarzschild space-time.
problem 6
problem 7
= 00 0 0 + 11 1 1 + 22 2 2 + 33 3 3 ,
1
= 2 t2 + x2 + y2 + z2 .
c
Let the symbol 0 denote the partial derivative /x0 . Then, by the chain rule we have
x 0 x 0
= ,
x x
= 0 0 ,
= 0 0 ,
using the properties of the inverse metric and of the Lorentz transformation. It follows that
under a Lorentz transformation
= 0 7 0 0 = 0,
the wave equation takes the same form (assuming that the field transforms as a scalar).
From the foregoing we have that the wave operator acts on the plane wave
e j xj t) to give [(/c)2 kj kj ]. Thus the action of the Klein-Gordon operator gives
i(k
2 2
~ c m2 c4 = ~2 2 ~2 c2 kj kj m2 c4
and the plane wave satisfies the Klein-Gordon equation provided the relativistic dispersion
~2 2 = ~2 c2 kj kj + m2 c4 .
is satisfied.
problem 8
109
is the sum of the diagonal elements of , all of which are equal to +1. Thus = 4,
or n in general dimension.
For the full contraction of the Levi-Civita symbol we should note that 0123 = +1 but
0123 = 1. The indices () all need to be distinct, but other than that we are required to
sum over each of the 24 permutations of (0123); for each permutation the value of the product
of the two symbols is 1. Thus = 24.
Let F = F be antisymmetric and T = T symmetric. Then
F T = F T = F T = F T .
Here the first equality comes from the antisymmetry of F , the second from the symmetry of T
and the third from a simple relabelling of indices () 7 (). It follows that F T is equal
to its negative and so must be zero.
In the case where no symmetry conditions are assumed for T we need only note the
identity
1 1
T = T + T + T T ,
2 2
which writes T as the sum of a symmetric part and an antisymmetric part. The previous result
then gives the current one.
The symbol has three indices each of which can take any of n different values. There
are therefore n3 different components. If it is symmetric = then only 12 n(n + 1) of
the combinations of possibilities for the lower two indices give independent values. So there
are 21 n2 (n + 1) independent components in this case. [For instance, if n = 4 there are 40
independent components; this is the general situation for the Christoffel symbols.]
problem 11
F = F = F = F = F .
Then, accounting for the values of the components of the inverse metric, we have at once
110
The others follow by antisymmetry. The dual field strength tensor can therefore be thought
of as obtained from the field strength tensor by the replacements (E, cB) 7 (cB, E).
The Lorentz pseudoscalar ?F F is expressed in terms of the electric and magnetic fields
as
?F F = ?F0i F 0i + ?Fi0 F i0 + ?Fij F ij = 4cE B.
problem 12
For reference, we have Fi0 = Ei , Fij = ijk cBk and the components of the 4-current are
J = (c, J).
Consider first the Bianchi identities
F + F + F = 0.
F = 0 c J .
F 0 = 0 F 00 + 1 F 01 + 2 F 02 + 3 F 03 ,
= 0 + x Ex + y Ey + z Ez ,
and it follows that E = 0 c2 . Recalling that c2 = 1/0 0 we recover the first Maxwell
equation. Now set = i. Then
1
F i = 0 F i0 + j F ij = t Ei + j ijk cBk ,
c
h 1 i
= c B i 2 t Ei ,
c
and we recover Maxwells fourth equation B 0 0 t E = 0 J.
problem 13
The motion of the particle is given by the solution of the Lorentz force equation
d2 x dx
mc = qF ,
d 2 d
where the field strength tensor is that corresponding to a uniform electric field, say along the
x-direction, and no magnetic field. What this means is that F10 = F01 = E, while all other
111
components are zero. The Lorentz force equation then gives the motion of the charged particle
as the solution of the simultaneous equations
d2 x0 qE dx1 d2 x1 qE dx0
= , = .
d 2 mc d d 2 mc d
Viewed as a pair of first order equations for the velocities dx0 /d and dx1 /d , the solution is
dx0 dx1
qE qE
= cosh , = sinh ,
d mc d mc
since the particle starts from rest, so dx1 /d | =0 = 0, and the trajectory is parameterised by
dx
c times proper time, so that dx
d d = 1. Finally, the trajectory of the particle is
0 mc qE 1 mc qE mc
x ( ) = sinh , x ( ) = cosh + x1 (0) .
qE mc qE mc qE
It is already clear from the structure of the solution that the trajectory is time-like; the tangent
vector to the particle trajectory is a time-like vector, in fact with magnitude squared 1, since
it is parameterised by c times proper time.
problem 14
Writing the term in square brackets as T we can identify the stress-energy-momentum tensor
for the Klein-Gordon field as
1 h i
T = ~2 c2 + ~2 c2 + m2 c4 2 .
2
112
Chapter 2 Differential Geometry
problem 1
where 0 u1 , u2 < 2.
To compute the metric in these local coordinates, write the displacement between nearby
points on the surface as
dX = R cos(u2 ) sin(u1 ) e1 + cos(u1 ) e2 du1
problem 2
problem 6
113
of Minkowski R1,3 . A parameterisation of it in terms of local coordinates (, , ) can be given
by
x0 = cosh(),
x1 = sinh() cos(),
x2 = sinh() sin() cos(),
x3 = sinh() sin() sin().
To determine the metric in these coordinates note that the displacement between nearby
points can be written as
h i
dX = sinh() e0 + cosh() cos() e1 + sin() cos() e2 + sin() e3 d
+ sinh() sin() e1 + cos() cos() e2 + sin() e3 d
+ sinh() sin() sin() e2 + cos() e3 d,
One then finds that the metric is given by (the dot product is with respect to the Minkowski
metric of the embedding space)
problem 8
The metric is flat; it is the same as the standard Cartesian metric on R2 with Cartesian
coordinates (au1 , u2 ). This is enough to determine the geodesics, but we may also note that
since the components of the metric are all constants the Christoffel symbols are all zero. The
geodesic equations are therefore
d2 ui
=0 i = 1, 2,
d 2
and so are linear functions, u1 = c1 + d1 , u2 = c2 + d2 , with c1 , d1 and c2 , d2 all constants.
problem 10
114
in terms of which the metric is given by
h i
ds2 = c2 dt2 + a2 cosh2 (ct/a) d2 + sin2 () d2 + sin2 () d2 .
Now, let be a geodesic parameterised by c times proper time ( ) and 0 any nearby curve,
also parameterised by . By direct calculation the metric on 0 is found to be
h i
ds2 = d 2 2c dt d0 + 2a0 sinh(ct/a) cosh(ct/a) d2 + sin2 () d2 + sin2 () d2
+ a2 cosh2 (ct/a) 2d d + 2 sin() cos() d2 + sin2 () d2
h i
2 2 2
+ sin () 2d d + 2 sin() cos() d + 2 sin () d d + O(2).
From this one can read off the geodesic equations, which we summarise by giving the complete
list of non-zero Christoffel symbols
The hint recommends that we do not try to solve the geodesic equation but instead use
what we know about geodesics on spheres in the Euclidean setting. What we know there is
that the geodesics on S n Rn+1 are given by the intersection of the sphere with a 2-plane
through the origin of Rn+1 . In fact more is true; it suffices to consider the particular 2-plane
given by (x1 , x2 , 0, . . . , 0), or xk = 0 for k = 3, . . . , n. This intersection is the curve
115
The reason this is sufficient is because all other geodesics can be obtained from this one by
the action of the symmetry group of the sphere: the rotation group SO(n + 1) acts on Rn+1
preserving the sphere and also acts transitively on the 2-planes passing through the origin
(there is some rotation taking the particular 2-plane we have considered so far to any other).
[We do not prove this statement, but you may like to do so.]
The same is true in the Lorentzian setting; the geodesics are given by the intersection of
the de Sitter space with 2-planes through the origin of R1,4 . The only difference is the 2-plane
may be space-like, time-like or null. A typical space-like 2-plane is x0 = x3 = x4 = 0 and a
typical time-like 2-plane is x2 = x3 = x4 = 0. Their intersections with dS4 are a space-like and
a time-like geodesic, respectively. All others can be obtained from these two by the action of
the Lorentz group SO(1, 4), which is the symmetry group of dS4 and acts transitively on the
2-planes through the origin of R1,4 , preserving their type (space-like, time-like or null). [We
do not prove this statement, but you may like to do so.] Now a typical null 2-plane can be
given as x0 = x1 , x3 = x4 = 0. Its intersection with the de Sitter space can be expressed as
the light ray (, , a, 0, 0). All other light rays can be obtained from this one by the action
of the subgroup of SO(1, 4) isomorphic to SO(4) that corresponds to rotations of the spatial
coordinates x1 , x2 , x3 , x4 of R1,4 . [The action is transitive but of course not free; the isotropy
subgroup is isomorphic to SO(2) and corresponds to rotations acting only on the coordinates
x3 , x4 .]
problem 11
The displacement between nearby points of anti-de Sitter can then be written as
dX = a sinh() cos( /a) e0 + sin( /a) e1 d + a cosh() sin() cos() e2 + sin() e3 + cos() e4 d
+ cosh() sin( /a) e0 + cos( /a) e1 d + a sinh() cos() cos() e2 + sin() e3 sin() e4 d
+ a sinh() sin() sin() e2 + cos() e3 d,
and it follows that the metric on anti-de Sitter in these coordinates is
and its derivative along itself is ( is c times proper time along the curve, since along the curve
= 0)
d 1
T T = T= cos( /a) e0 + sin( /a) e1 .
d a
Now the expression given above for the displacement dX between nearby points in anti-de
Sitter allows one to identify a basis for the tangent space at each point and one can see directly
116
that the vector giving the direction that T is changing into is orthogonal to all of the tangent
vectors. [Dont forget that = 0 along the curve.]
problem 14
corresponding to standard round expressions for the metric on the three isotropic homoge-
neous two-dimensional manifolds of constant scalar curvature; the 2-sphere, the plane, and
the hyperbolic plane. I give the calculation for the 2-sphere only; the others are identical in
structure.
First we determine the Christoffel symbols from the geodesics. Let (r( ), ( )) be a geodesic
parameterised by arc length and let (r + r , + ) be a variation. The metric on this latter
curve is
dr dr 2 d d 2 d d 2
= d 2 + 2 d + 2a sin(r/a) cos(r/a) r d + 2a2 sin2 (r/a) d + O(2),
d d d d d d
from which it follows that
Z f h
dr dr d d i
Z
d d
ds = 1+ + a sin(r/a) cos(r/a) r + a2 sin2 (r/a) d + O(2),
0 i d d d d d d
Z Z f n h 2
d r d d i d
d o
= ds r a sin(r/a) cos(r/a) + a 2
sin2
(r/a) d + O(2).
i d 2 d d d d
The vanishing of the first order term yields the geodesic equations
d2 r d d
a sin(r/a) cos(r/a) = 0,
d 2 d d
d2 2 cos(r/a) dr d
+ = 0,
d 2 a sin(r/a) d d
from which we can read off the non-zero Christoffel symbols
1 cos(r/a)
r = a sin(r/a) cos(r/a), r = r = .
a sin(r/a)
For the components of the Ricci tensor we have by direct calculation
R = i i ii + i jji ij ji ,
= r r + r r r r r r ,
= sin2 (r/a).
117
Finally, the Ricci scalar is
1 1 2
R = g ij Rij = + sin2 (r/a) = 2 .
a2 a2 sin2 (r/a) a
problem 16
The idea is to use the fact that the covariant derivative of a function is the ordinary
derivative of a function, f = df , or in local coordinates
f = f.
Now choose for f the action of a 1-form A on the vector Z, which in terms of components is
given by A(Z) = A Z . Then by the Leibniz formula we have
A Z = A Z + A Z ,
= A Z + A Z + Z .
F = A A = A A A + A ,
= A A ,
F = F F F .
T A Z = T A Z + T A Z + A Z ,
T A Z = T A Z + T A Z + A Z .
T = T + T T .
118
Chapter 4 The Schwarzschild Solution
problem 3
`2 12m2 1/2
r = 1 1 2 .
2m `
These solutions only exist provided 12m2 /`2 < 1. (Equality is still a critical point, but it has
degenerate Hessian.)
Write the effective potential as follows
`2 `2 2 `2
2m h 2 2
i 2m 2
Veff (r) = 3 r r+` = 3 r +` 1 .
r 2m r 4m 16m2
This shows that the potential is non-positive if `2 /16m2 1 but positive for some range of
values of r if `2 /16m2 > 1. When `2 = 16m2 it takes its maximum value of 0 at r = `2 /4m
(ignore the point r ). One readily verifies that when `2 = 16m2 the previous calculation
yields
`2 (2 1)`2
1
r = 1 = ,
2m 2 4m
so that there is complete consistency. [The easiest way to see the answer to this question is to
sketch the form of the effective potential. Certainly this is how I obtained the solution outlined
above.]
problem 4
m 2
r = 1+ ,
2
in the Schwarzschild metric. One finds by direct calculation that
1 r (1 + m/2)2 (1 + m/2)2
= = = ,
1 2m
r
r 2m (1 + m/2)2 2m/ (1 m/2)2
m 2
m m m m
dr = 1 + d 1+ d = 1 + 1 d,
2 2 2 2
and it then follows swiftly that
dr2 m 4 2 m 4 2h 2
h i i
2 2 2 2
+ r d + sin () d =
1+ d + 1 + d + sin2 () d2 ,
1 2mr
2 2
4
m 2
dx + dy 2 + dz 2 .
= 1+
2
119
The form of the Minkowski metric given then follows at once.
To see the coordinate range, note that
1/2 1/2 2
m 2 m
r= + .
2 m 2
The fixed point of this inversion, 2/m = 1 corresponds to the minimum value of r covered by
the new coordinate chart. One finds by direct substitution that this minimum value is r = 2m.
problem 5
where 2 2 2 2 3
pd` = dx + dy + dz is the usual Euclidean metric for Cartesian coordinates on R and
2 2 2 2
= x + y + z . Along a null geodesic ds = 0 and hence
m 1 m 3
4m
cdt = 1 1+ d` 1 + d`.
2 2
We describe the trajectory just using the Cartesian coordinates x, y, z and standard Eu-
clidean geometry. If the distance of closest approach to the Sun is b then p the distance travelled
from that point to a point radial distance from the origin is ` = 2 b2 . It follows that
the total time taken along the photon trajectory is
Z R2 b2 Z RE Z R2 b2 Z RP
E 2m P 2m
ct = 2 d` + p d + d` + p d ,
0 b 2 b2 0 b 2 b2
q
RE RP i
q h
=2 2 2 2 2
RE b + RP b + 2m arccosh + arccosh ,
b b
r r
q
R 2 2
RP
R R
q
2 b2 + 2 b2 + 2m ln E E P
=2 RE RP + 1 + ln + 1 ,
b b2 b b2
q
4RE RP
q
2 2 2 2 2
RE b + RP b + 2m ln .
b2
problem 6
From the figure one has immediately that the deflection angle is
b bx
= arctan
+ arctan ,
d D
b bx
+ .
d D
The calculation using the Schwarzschild metric gives the deflection angle as = 4m/b. It then
follows that
b2 b2 xb d2
2 xd
4m = + = d+ ,
d D D D D
120
or
x 4mD
2 = 0.
d+D d(d + D)
The solutions are evidently
1/2
x2
x 4mD
= + .
2(d + D) d(d + D) 4(d + D)2
There are two values because there are two images, one either side of the lens. This matches
with Fermats principle of least time; one of the paths is a minimum of the time taken, the other
a local maximum (given the constraint that the deflection angle is given by general relativity).
As x increases one image tends to the direct line-of-sight location while the other disappears
behind the lens. As x 0 all directions going aroundp the lens become equivalent and the
image becomes an Einstein ring of angular radius 4mD/d(d + D).
problem 7
dx dx
2 2m
g = 1 .
d d r0
du d dx
a= = 0 = = 2 00 .
d d d 0
Looking up the Christoffel symbols for the Schwarzschild metric we find that only r00 is non-
zero and is given by
r m 2m
00 = 2 1 .
r r
It then follows that the acceleration of the observer is
m
a= r ,
r02
and since the magnitude squared of the basis vector r is the metric component grr we can say
that the magnitude of the acceleration is
2m 1/2
m
kak = 2 1 .
r0 r0
Its direction is everywhere tangent to the coordinate r curves, i.e. to the curves (ct0 , r, 0 , 0 )
or, more plainly, radial indeed radially outwards.
121
Chapter 5 Gravitational Collapse and Black Holes
problem 2
We wish to take both r 0 and /2. We need to do the former first if the limit r 0 is
to be non-singular in these coordinates. So, setting r = 0 we have
ds2 = a2 d2 .
What this tells us is that the coordinate (ct0 , 0, /2, ), which we might be tempted to associate
with the origin of a polar system since r = 0, describes a circle of radius a. Although we do
not show it, one finds by computing the Riemann curvature that the coordinate location r = 0
is a curvature singularity. Thus in the Kerr metric the singularity is not a single point, as it is
in the Schwarzschild space-time, but rather a ring of circumference 2a.
problem 3
4G r
Z
4G
m(r) = 2 (r0 )r0 2 dr0 = r3 ,
2 0
c 0 3c
for r a and takes the value m(a) for r a. We will denote this latter simply by m. The
Tolman-Oppenheimer-Volkoff equation then becomes
dp 4G r
2
1 2
= 4 p + 0 c p + 0 c .
dr c 1 8G
3c 2 0 r 2 3
to arrive at
dp dp 1 8G0 2
= d ln 1 r ,
p + 13 0 c2 p + 0 c2 2 3c2
Integrating and using that p(a) = 0 we obtain
1 2
0 c2 8G0 a2
3 0 c 1
ln ln = ln 1 ,
p(0) + 13 0 c2 p(0) + 0 c2 2 3c2
or
p(0) + 0 c2 2m 1/2
ln = ln 1 .
3p(0) + 0 c2 a
122
Exponentiating and rearranging we find the desired expression for the pressure at the centre
of the star 1/2
2 1 1 2m/a
p(0) = 0 c 1/2 .
3 1 2m/a 1
Evidently the pressure diverges when 3(1 2m/a)1/2 = 1, which is the same as m = 4a/9.
It is instructive to estimate this limiting mass for a star composed of ordinary matter,
like the Sun; i.e. take for 0 the average density of the Sun. Using that the total mass is
M = 4 3 2
3 0 a and that m = GM/c we obtain for the mass limit
123
Chapter 6 Gravitational Waves
problem 2
For two equal mass stars, mass M , in circular orbits (e = 0) with semi-major radius a, the
rate of decrease of the semi-major radius due to emission of gravitational waves is given by
da 128G3 M 3
= .
dt 5c5 a3
It follows by direct integration that
1h 4 i 128G3 M 3
a (t) a4 (0) = t
4 5c5
Writing a(0) = a0 , the time for complete inspiral a(t) = 0 is given by
5c5 a40
t= .
512G3 M 3
Recall Keplers third law; 2 a3 = G(M1 + M2 ), where M1 , M2 are the two masses and
= 2/P with P the orbital period. It follows that a0 = (2GM/2 )1/3 and hence
8/3 5/3 5/3
5c5 (2GM/2 )4/3 5c5/3
P 2GM 5P c
t= 3
= = .
64(2GM ) 64 2 c2 128 .2GM/c2
Substituting in the numbers given we find that the time to merger is
5 5/3
tmerge = 8.2 107 hours = 4.4 107 years.
64
If the frequency of the gravitational waves is 50 Hz, then the frequency of the orbit is 25
Hz. Continuing to assume stars of mass 1.4M this gives a time to merger of
tmerge = 13.4 s.
The qualitative features of the gravitational waves follow from the time dependence of the
orbital parameters of the binary star system and of the amplitude of the gravitational waves
they produce. From Peters formula we have that the semi-major axis of the orbit evolves with
the time to merger as
1/4
a tmerge t .
It then follows from Keplers third law, 2 a3 = G(M1 + M2 ) that the frequency of the orbital
motion scales as 3/8
tmerge t .
The gravitational waves have twice the frequency of the source, but this is simply a multiplica-
tive factor and they display the same scaling with time to merger. Finally, the amplitude of
the gravitational waves produced by such a binary source is proportional to 2 a2 and hence
scales as 1/4
h tmerge t ,
with the time to merger. These scalings are enough to give a rough sketch of the anticipated
chirp signal, at least in the early part of the merger.
problem 5
124
and the three formulae asked for follow immediately.
To find the metric, recall that h = h 12 ( h ). Since h = 4/c2 we have
4 1 4 2
h00 = 2
1 2 = 2 ,
c 2 c c
1 4
h0i = i 0 2 = i ,
2 c
1 4 2
hij = 0 ij 2 = 2 ij .
2 c c
It then follows that the metric is given by
2 2 2 2
ds = + h dx dx = 1 + 2 c dt i cdt dx + dx cdt + 1 2 dxi dxi .
2
i i
c c
125
Chapter 7 Cosmology
problem 2
To be added.
problem 5
ds2 = dX dX = 0 X 0 X c2 dt2 + X X d2 + X X d2 + X X d2 ,
= c2 dt2 + a20 d2 + a20 sin2 () d2 + a20 sin2 () sin2 () d2 ,
h i
2 2 2 2 2 2 2 2
= c dt + a0 d + sin () d + sin () d ,
where by the dot we mean the usual inner product on Minkowski space.
The metric for the Einstein static universe, in these coordinates, is of FLRW form with
scale factor a(t) equal to the constant a0 and for spatial metric of positive scalar curvature,
k = 1. We can therefore look up the components of the Ricci tensor
1 3
R00 Rg00 = 2 ,
2 a0
1 1
Rij Rgij = 2 gij .
2 a0
Including the cosmological constant and a pressure-free fluid (in its rest frame) the Einstein
equations read
3 8G
2 = 4 c2 ,
a0 c
1
gij + gij = 0.
a20
It follows that there is a solution with
c2
a0 = 1/2 , = .
4G
126