General Relativity - Benjamin Crowell

2
General Relativity
Benjamin Crowell
www.lightandmatter.com
Fullerton, California
www.lightandmatter.com
Copyright c _2009 Benjamin Crowell
rev. February 15, 2010
Permission is granted to copy, distribute and/or modify this docu-
ment under the terms of the Creative Commons Attribution Share-
Alike License, which can be found at creativecommons.org. The
license applies to the entire text of this book, plus all the illustra-
tions that are by Benjamin Crowell. All the illustrations are by
Benjamin Crowell except as noted in the photo credits or in paren-
theses in the caption of the gure. This book can be downloaded
free of charge from www.lightandmatter.com in a variety of formats,
including editable formats.
Brief Contents
1 A Geometrical Theory of Spacetime 11
2 Geometry of Flat Spacetime 35
3 Differential Geometry 71
4 Tensors 103
5 Curvature 129
6 Vacuum Solutions 165
7 Sources 193
8 Gravitational Waves 229
5
6
Contents
1 A Geometrical Theory of Spacetime 11
1.1 Time and causality . . . . . . . . . . . . . . . . 12
1.2 Experimental tests of the nature of time . . . . . . . . 14
The Hafele-Keating experiment, 14.Muons, 16.Gravitational
redshifts, 16.
1.3 Non-simultaneity and the maximum speed of cause and
effect . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4 Ordered geometry. . . . . . . . . . . . . . . . . 18
1.5 The equivalence principle. . . . . . . . . . . . . . 20
Proportionality of inertial and gravitational mass, 20.Geometrical
treatment of gravity, 21.Eotvos experiments, 22.Equivalence
of gravitational elds and accelerations, 23.The equivalence principle,
23.Locality of Lorentz frames, 25.Gravitational redshifts, 27.
The Pound-Rebka experiment, 29.
Problems . . . . . . . . . . . . . . . . . . . . . . 33
2 Geometry of Flat Spacetime 35
2.1 Afne properties of Lorentz geometry. . . . . . . . . 35
2.2 Relativistic properties of Lorentz geometry . . . . . . 40
2.3 The light cone . . . . . . . . . . . . . . . . . . 49
Velocity addition, 51.Logic, 53.
2.4 Experimental tests of Lorentz geometry. . . . . . . . 53
Dispersion of the vacuum, 54.Observer-independence of c, 54.
Lorentz violation by gravitational forces, 56.
2.5 Three spatial dimensions . . . . . . . . . . . . . . 57
Lorentz boosts in three dimensions, 58.Gyroscopes and the equiv-
alence principle, 58.Boosts causing rotations, 60.An experi-
mental test: Thomas precession in hydrogen, 67.
Problems . . . . . . . . . . . . . . . . . . . . . . 68
3 Differential Geometry 71
3.1 The afne parameter revisited, and parallel transport . . 72
The ane parameter in curved spacetime, 72.Parallel transport,
73.
3.2 Models . . . . . . . . . . . . . . . . . . . . . 74
3.3 Intrinsic quantities . . . . . . . . . . . . . . . . . 79
Coordinate independence, 80.
3.4 The metric . . . . . . . . . . . . . . . . . . . . 82
The Euclidean metric, 83.The Lorentz metric, 87.Isometry, in-
ner products, and the Erlangen Program, 87.Einsteins carousel,
89.
3.5 The metric in general relativity. . . . . . . . . . . . 95
The hole argument, 95.A Machian paradox, 96.
3.6 Interpretation of coordinate independence. . . . . . . 97
Is coordinate independence obvious?, 97.Is coordinate indepen-
dence trivial?, 97.Coordinate independence as a choice of gauge,
7
99.
Problems . . . . . . . . . . . . . . . . . . . . . . 100
4 Tensors 103
4.1 Lorentz scalars . . . . . . . . . . . . . . . . . . 103
4.2 Four-vectors . . . . . . . . . . . . . . . . . . . 104
The velocity and acceleration four-vectors, 104.The momentum
four-vector, 106.The frequency four-vector and the relativistic
Doppler shift, 111.A non-example: electric and magnetic elds,
112.The electromagnetic potential four-vector, 113.
4.3 The tensor transformation laws . . . . . . . . . . . 114
4.4 Experimental tests . . . . . . . . . . . . . . . . 117
Universality of tensor behavior, 117.Speed of light diering from
c, 118.Degenerate matter, 118.
4.5 Conservation laws. . . . . . . . . . . . . . . . . 122
No general conservation laws, 122.Conservation of angular mo-
mentum and frame dragging, 123.
Problems . . . . . . . . . . . . . . . . . . . . . . 126
5 Curvature 129
5.1 Tidal curvature versus curvature caused by local sources 130
5.2 The energy-momentum tensor . . . . . . . . . . . 131
5.3 Curvature in two spacelike dimensions . . . . . . . . 132
5.4 Curvature tensors . . . . . . . . . . . . . . . . . 136
5.5 Some order-of-magnitude estimates . . . . . . . . . 138
The geodetic eect, 139.Deection of light rays, 140.
5.6 The covariant derivative . . . . . . . . . . . . . . 141
The covariant derivative in electromagnetism, 142.The covariant
derivative in general relativity, 143.
5.7 The geodesic equation . . . . . . . . . . . . . . . 146
Characterization of the geodesic, 147.Covariant derivative with
respect to a parameter, 147.The geodesic equation, 148.Uniqueness,
148.
5.8 Torsion . . . . . . . . . . . . . . . . . . . . . 149
Are scalars path-dependent?, 149.The torsion tensor, 152.Experimental
searches for torsion, 153.
5.9 From metric to curvature . . . . . . . . . . . . . . 156
Finding given g, 156.Numerical solution of the geodesic equation,
157.The Riemann tensor in terms of the Christoel symbols,
159.Some general ideas about gauge, 159.
Problems . . . . . . . . . . . . . . . . . . . . . . 162
6 Vacuum Solutions 165
6.1 Event horizons . . . . . . . . . . . . . . . . . . 165
The event horizon of an accelerated observer, 165.Information
paradox, 167.Radiation from event horizons, 168.
6.2 The Schwarzschild metric . . . . . . . . . . . . . 169
The zero-mass case, 170.Geometrized units, 172.A large-r limit,
172.The complete solution, 173.Geodetic eect, 175.Orbits,
178.Deection of light, 182.
8
6.3 Black holes. . . . . . . . . . . . . . . . . . . . 185
Singularities, 185.Event horizon, 186.Expected formation, 186.
Observational evidence, 186.Singularities and cosmic censorship,
187.Black hole radiation, 188.
Problems . . . . . . . . . . . . . . . . . . . . . . 191
7 Sources 193
7.1 Sources in general relativity. . . . . . . . . . . . . 193
Point sources in a background-independent theory, 193.The Ein-
stein eld equation, 194.The uniform gravitational eld revisited,
195.Energy conditions, 199.The cosmological constant, 201.
7.2 Cosmological solutions. . . . . . . . . . . . . . . 203
Evidence for expansion of the universe, 206.A singularity at the
Big Bang, 207.Observability of expansion, 207.The vacuum-
dominated solution, 212.The matter-dominated solution, 217.
Observation, 219.
7.3 Machs principle revisited . . . . . . . . . . . . . . 221
The Brans-Dicke theory, 221.Predictions of the Brans-Dicke theory,
224.Hints of empirical support, 225.Machs principle is false.,
225.
Problems . . . . . . . . . . . . . . . . . . . . . . 227
8 Gravitational Waves 229
8.1 The speed of gravity . . . . . . . . . . . . . . . . 229
8.2 Gravitational radiation . . . . . . . . . . . . . . . 230
Empirical evidence, 230.Expected properties, 231.Some exact
solutions, 233.Energy content, 235.Rate of radiation, 236.
Problems . . . . . . . . . . . . . . . . . . . . . . 239
Appendix 1: Excerpts from three papers by Einstein . . . . . . . . . 240
On the electrodynamics of moving bodies . . . . . . . . . . . . . . . . . . . 240
Does the inertia of a body depend upon its energy content? 253
The foundation of the general theory of relativity . . . . . . . . . . . 255
Appendix 2: Hints and solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
9
10
Chapter 1
A Geometrical Theory of
Spacetime
I always get a slight brain-shiver, now [that] space and time appear
conglomerated together in a gray, miserable chaos. Sommerfeld
This is a book about general relativity, at a level that is meant
to be accessible to advanced undergraduates.
This is mainly a book about general relativity, not special rel-
ativity. Ive heard the sentiment expressed that books on special
relativity generally do a lousy job on special relativity, compared to
books on general relativity. This is undoubtedly true, for someone
who already has already learned special relativity but wants to
unlearn the parts that are completely wrong in the broader context
of general relativity. For someone who has not already learned spe-
cial relativity, I strongly recommend mastering it rst, from a book
such as Taylor and Wheelers Spacetime Physics. Even an advanced
student may be able to learn a great deal from a masterfully writ-
ten, nonmathematical treatment at an even lower level, such as the
ones in Hewitts, Conceptual Physics or the inexpensive paperback
by Gardner, Relativity Simply Explained.
In the back of this book Ive included excerpts from three papers
by Einstein two on special relativity and one on general relativity.
They can be read before, after, or along with this book. There are
footnotes in the papers and in the main text linking their content
with each other.
I should reveal at the outset that I am not a professional rela-
tivist. My eld of research was nonrelativistic nuclear physics until
I became a community college physics instructor. I can only hope
that my pedagogical experience will compensate to some extent for
my shallow background, and that readers who nd mistakes will be
kind enough to let me know about them using the contact informa-
tion provided at http://www.lightandmatter.com/area4author.
html.
11
1.1 Time and causality
Updating Platos allegory of the cave, imagine two super-intelligent
twins, Alice and Betty. Theyre raised entirely by a robotic tutor
on a sealed space station, with no access to the outside world. The
robot, in accord with the latest fad in education, is programmed to
encourage them to build up a picture of all the laws of physics based
on their own experiments, without a textbook to tell them the right
answers. Putting yourself in the twins shoes, imagine giving up
all your preconceived ideas about space and time, which may turn
out according to relativity to be completely wrong, or perhaps only
approximations that are valid under certain circumstances.
Causality is one thing the twins will notice. Certain events re-
sult in other events, forming a network of cause and eect. One
general rule they infer from their observations is that there is an
unambiguously dened notion of betweenness: if Alice observes that
event 1 causes event 2, and then 2 causes 3, Betty always agrees that
2 lies between 1 and 3 in the chain of causality. They nd that this
agreement holds regardless of whether one twin is standing on her
head (i.e., its invariant under rotation), and regardless of whether
one twin is sitting on the couch while the other is zooming around
the living room in circles on her nuclear fusion scooter (i.e., its also
invariant with respect to dierent states of motion).
You may have heard that relativity is a theory that can be inter-
preted using non-Euclidean geometry. The invariance of between-
ness is a basic geometrical property that is shared by both Euclidean
and non-Euclidean geometry. We say that they are both ordered
geometries. With this geometrical interpretation in mind, it will
be useful to think of events not as actual notable occurrences but
merely as an ambient sprinkling of points at which things could hap-
pen. For example, if Alice and Betty are eating dinner, Alice could
choose to throw her mashed potatoes at Betty. Even if she refrains,
there was the potential for a causal linkage between her dinner and
Bettys forehead.
Betweenness is very weak. Alice and Betty may also make a
number of conjectures that would say much more about causality.
For example: (i) that the universes entire network of causality is
connected, rather than being broken up into separate parts; (ii) that
the events are globally ordered, so that for any two events 1 and 2,
either 1 could cause 2 or 2 could cause 1, but not both; (iii) not only
are the events ordered, but the ordering can be modeled by sorting
the events out along a line, the time axis, and assigning a number t,
time, to each event. To see what these conjectures would entail, lets
discuss a few examples that may draw on knowledge from outside
Alice and Bettys experiences.
Example: According to the Big Bang theory, it seems likely that
the network is connected, since all events would presumably connect
12 Chapter 1 A Geometrical Theory of Spacetime
back to the Big Bang. On the other hand, if (i) were false we might
have no way of nding out, because the lack of causal connections
would make it impossible for us to detect the existence of the other
universes represented by the other parts disconnected from our own
universe.
Example: If we had a time machine, we could violate (ii), but
this brings up paradoxes, like the possibility of killing ones own
grandmother when she was a baby, and in any case nobody knows
how to build a time machine.
Example: There are nevertheless strong reasons for believing
that (ii) is false. For example, if we drop Alice into one black hole,
and Betty into another, they will never be able to communicate
again, and therefore there is no way to have any cause and eect
relationship between Alices events and Bettys.
1
Since (iii) implies (ii), we suspect that (iii) is false as well. But
Alice and Betty build clocks, and these clocks are remarkably suc-
cessful at describing cause-and-eect relationships within the con-
nes of the quarters in which theyve lived their lives: events with
higher clock readings never cause events with lower clock readings.
They announce to their robot tutor that theyve discovered a uni-
versal thing called time, which explains all causal relationships, and
which their experiments show ows at the same rate everywhere
within their quarters.
Ah, the tutor sighs, his metallic voice trailing o.
I know that ah, Tutorbot, Betty says. Come on, cant you
just tell us what we did wrong?
You know that my pedagogical programming doesnt allow that.
Oh, sometimes I just want to strangle whoever came up with
those stupid educational theories, Alice says.
The twins go on strike, protesting that the time theory works
perfectly in every experiment theyve been able to imagine. Tutor-
bot gets on the commlink with his masters and has a long, inaudible
argument, which, judging from the hand gestures, the twins imagine
to be quite heated. He announces that hes gotten approval for a
eld trip for one of the twins, on the condition that she remain in a
sealed environment the whole time so as to maintain the conditions
of the educational experiment.
Who gets to go? Alice asks.
Betty, Tutorbot replies, because of the mashed potatoes.
But I refrained! Alice says, stamping her foot.
Only one time out of the last six that I served them.
1
This point is revisited in section 6.1.
Section 1.1 Time and causality 13
The next day, Betty, smiling smugly, climbs aboard the sealed
spaceship carrying a duel bag lled with a large collection of clocks
for the trip. Each clock has a duplicate left behind with Alice. The
clock design that theyre proudest of consists of a tube with two
mirrors at the ends. A ash of light bounces back and forth between
the ends, with each round trip counting as one tick, one unit of
time. The twins are convinced that this one will run at a constant
rate no matter what, since it has no moving parts that could be
aected by the vibrations and accelerations of the journey.
Bettys eld trip is dull. She doesnt get to see any of the outside
world. In fact, the only way she can tell shes not still at home is that
she sometimes feels strong sensations of acceleration. (Shes grown
up in zero gravity, so the pressing sensation is novel to her.) Shes
out of communication with Alice, and all she has to do during the
long voyage is to tend to her clocks. As a crude check, she veries
that the light clock seems to be running at its normal rate, judged
against her own pulse. The pendulum clock gets out of synch with
the light clock during the accelerations, but that doesnt surprise
her, because its a mechanical clock with moving parts. All of the
nonmechanical clocks seem to agree quite well. She gets hungry for
breakfast, lunch, and dinner at the usual times.
When Betty gets home, Alice asks, Well?
Great trip, too bad you couldnt come. I met some cute boys,
went out dancing, . . .
You did not. What about the clocks?
They all checked out ne. See, Tutorbot? The time theory still
holds up.
That was an anticlimax, Alice says. Im going back to bed
now.
Bed? Betty exclaims. Its three in the afternoon.
The twins now discover that although all of Alices clocks agree
among themselves, and similarly for all of Bettys (except for the
ones that were obviously disrupted by mechanical stresses), Alices
and Bettys clocks disagree with one another. A week has passed
for Alice, but only a couple of days for Betty.
1.2 Experimental tests of the nature of time
1.2.1 The Hafele-Keating experiment
In 1971, J.C. Hafele and R.E. Keating
2
of the U.S. Naval Obser-
vatory brought atomic clocks aboard commercial airliners and went
around the world, once from east to west and once from west to east.
(The clocks had their own tickets, and occupied their own seats.) As
2
Hafele and Keating, Science, 177 (1972), 168
a / The atomic clock has its
own ticket and its own seat.
in the parable of Alice and Betty, Hafele and Keating observed that
there was a discrepancy between the times measured by the trav-
eling clocks and the times measured by similar clocks that stayed
at the lab in Washington. The result was that the east-going clock
lost an amount of time t
E
= 59 10 ns, while the west-going
one gained t
W
= +273 7 ns. This establishes that time is not
universal and absolute.
Nevertheless, causality was preserved. The nanosecond-scale ef-
fects observed were small compared to the three-day lengths of the
plane trips. There was no opportunity for paradoxical situations
such as, for example, a scenario in which the east-going experimenter
arrived back in Washington before he left and then proceeded to
convince himself not to take the trip.
Hafele and Keating were testing specic quantitative predictions
of relativity, and they veried them to within their experiments
error bars. At this point in the book, we arent in possession of
enough relativity to be able to make such calculations, but, like
Alice and Betty, we can inspect the empirical results for clues as to
how time works.
The opposite signs of the two results suggests that the rate at
which time ows depends on the motion of the observer. The east-
going clock was moving in the same direction as the earths rotation,
so its velocity relative to the earths center was greater than that of
the ones that remained in Washington, while the west-going clocks
velocity was correspondingly reduced.
3
The signs of the ts show
that moving clocks were slower.
On the other hand, the asymmetry of the results, with [t
E
[ , =
[t
W
[, implies that there was a second eect involved, simply due
to the planes being up in the air. Relativity predicts that times
rate of ow also changes with height in a gravitational eld. The
deeper reasons for such an eect are given in section 1.5.8 on page
29.
Although Hafele and Keatings measurements were on the ragged
edge of the state of the art in 1971, technology has now progressed
to the point where such eects have everyday consequences. The
satellites of the Global Positioning System (GPS) orbit at a speed
of 1.9 10
3
m/s, an order of magnitude faster than a commercial
jet. Their altitude of 20,000 km is also much greater than that of
an aircraft. For both these reasons, the relativistic eect on time is
stronger than in the Hafele-Keating experiment. The atomic clocks
aboard the satellites are tuned to a frequency of 10.22999999543
MHz, which is perceived on the ground as 10.23 MHz. (This fre-
3
These dierences in velocity are not simply something that can be eliminated
by choosing a dierent frame of reference, because the clocks motion isnt in
a straight line. The clocks back in Washington, for example, have a certain
acceleration toward the earths axis, which is dierent from the accelerations
experienced by the traveling clocks.
Section 1.2 Experimental tests of the nature of time 15
quency shift will be calculated in example 7 on page 46.)
1.2.2 Muons
Although the Hafele-Keating experiment is impressively direct,
it was not the rst verication of relativistic eects on time, it did
not completely separate the kinematic and gravitational eects, and
the eect was small. An early experiment demonstrating a large and
purely kinematic eect was performed in 1941 by Rossi and Hall,
who detected cosmic-ray muons at the summit and base of Mount
Washington in New Hampshire. The muon has a mean lifetime of
2.2 s, and the time of ight between the top and bottom of the
mountain (about 2 km for muons arriving along a vertical path)
at nearly the speed of light was about 7 s, so in the absence of
relativistic eects, the ux at the bottom of the mountain should
have been smaller than the ux at the top by about an order of
magnitude. The observed ratio was much smaller, indicating that
the clock constituted by nuclear decay processes was dramatically
slowed down by the motion of the muons.
1.2.3 Gravitational redshifts
The rst experiment that isolated the gravitational eect on time
was a 1925 measurement by W.S. Adams of the spectrum of light
emitted from the surface of the white dwarf star Sirius B. The grav-
itational eld at the surface of Sirius B is 4 10
5
g, and the gravi-
tational potential is about 3,000 times greater than at the Earths
surface. The emission lines of hydrogen were redshifted, i.e., re-
duced in frequency, and this eect was interpreted as a slowing of
time at the surface of Sirius relative to the surface of the Earth. His-
torically, the mass and radius of Sirius were not known with better
than order of magnitude precision in 1925, so this observation did
not constitute a good quantitative test.
The rst such experiment to be carried out under controlled
conditions, by Pound and Rebka in 1959, is analyzed quantitatively
in example 4 on page 106.
The rst high-precision experiment of this kind was Gravity
Probe A, a 1976 experiment
4
in which a space probe was launched
vertically from Wallops Island, Virginia, at less than escape veloc-
ity, to an altitude of 10,000 km, after which it fell back to earth and
crashed down in the Atlantic Ocean. The probe carried a hydro-
gen maser clock which was used to control the frequency of a radio
signal. The radio signal was received on the ground, the nonrela-
tivistic Doppler shift was subtracted out, and the residual blueshift
was interpreted as the gravitational eect eect on time, matching
the relativistic prediction to an accuracy of 0.01%.
4
Vessot at al., Physical Review Letters 45 (1980) 2081
b / Gravity Probe A.
1.3 Non-simultaneity and the maximum speed
of cause and effect
Weve seen that time ows at dierent rates for dierent observers.
Suppose that Alice and Betty repeat their Hafele-Keating-style ex-
periment, but this time they are allowed to communicate during
the trip. Once Bettys ship completes its initial acceleration away
from Betty, she cruises at constant speed, and each girl has her own
equally valid inertial frame of reference. Each twin considers herself
to be at rest, and says that the other is the one who is moving.
Each one says that the others clock is the one that is slow. If they
could pull out their phones and communicate instantaneously, with
no time lag for the propagation of the signals, they could resolve
the controversy. Alice could ask Betty, What time does your clock
read right now? and get an immediate answer back.
By the symmetry of their frames of reference, however, it seems
that that Alice and Betty should not be able to resolve the contro-
versy during Bettys trip. If they could, then they could release two
radar beacons that would permanently establish two inertial frames
of reference, A and B, such that time owed, say, more slowly in B
than in A. This would violate the principle that motion is relative,
and that all inertial frames of reference are equally valid. The best
that they can do is to compare clocks once Betty returns, and verify
that the net result of the trip was to make Bettys clock run more
slowly on the average.
Alice and Betty can never satisfy their curiosity about exactly
when during Bettys voyage the discrepancies accumulated or at
Section 1.3 Non-simultaneity and the maximum speed of cause and effect 17
what rate. This is information that they can never obtain, but
they could obtain it if they had a system for communicating in-
stantaneously. We conclude that instantaneous communication is
impossible. There must be some maximum speed at which signals
can propagate or, more generally, a maximum speed at which
cause and eect can propagate and this speed must for example
be greater than or equal to the speed at which radio waves propa-
gate. It is also evident from these considerations that simultaneity
itself cannot be a meaningful concept in relativity.
1.4 Ordered geometry
Lets try to put what weve learned into a general geometrical con-
text.
Euclids familiar geometry of two-dimensional space has the fol-
lowing axioms,
5
which are expressed in terms of operations that can
be carried out with a compass and unmarked straightedge:
E1 Two points determine a line.
E2 Line segments can be extended.
E3 A unique circle can be constructed given any point as its center
and any line segment as its radius.
E4 All right angles are equal to one another.
E5 Parallel postulate: Given a line and a point not on the line,
no more than one line can be drawn through the point and
parallel to the given line.
6
The modern style in mathematics is to consider this type of
axiomatic system as a self-contained sandbox, with the axioms, and
any theorems proved from them, being true or false only in relation
to one another. Euclid and his contemporaries, however, believed
them to be empirical facts about physical reality. For example, they
considered the fth postulate to be less obvious than the rst four,
because in order to verify physically that two lines were parallel,
one would theoretically have to extend them to an innite distance
and make sure that they never crossed. In the rst 28 theorems of
the Elements, Euclid restricts himself entirely to propositions that
can be proved based on the more secure rst four postulates. The
more general geometry dened by omitting the parallel postulate is
known as absolute geometry.
5
These axioms are summarized for quick reference in the back of the book on
page 272.
6
This is a form known as Playfairs axiom, rather than the version of the
postulate originally given by Euclid.
What kind of geometry is likely to be applicable to general rel-
ativity? We can see immediately that Euclidean geometry, or even
absolute geometry, would be far too specialized. We have in mind
the description of events that are points in both space and time.
Conning ourselves for ease of visualization to one dimension worth
of space, we can certainly construct a plane described by coordi-
nates (t, x), but imposing Euclids postulates on this plane results
in physical nonsense. Space and time are physically distinguishable
from one another. But postulates 3 and 4 describe a geometry in
which distances measured along non-parallel axes are comparable,
and gures may be freely rotated without aecting the truth or
falsehood of statements about them; this is only appropriate for a
physical description of dierent spacelike directions, as in an (x, y)
plane whose two axes are indistinguishable.
We need to throw most of the specialized apparatus of Euclidean
geometry overboard. Once weve stripped our geometry to a bare
minimum, then we can go back and build up a dierent set of equip-
ment that will be better suited to relativity.
The stripped-down geometry we want is called ordered geometry,
and was developed by Moritz Pasch around 1882. As suggested by
the parable of Alice and Betty, ordered geometry does not have
any global, all-encompassing system of measurement. When Betty
goes on her trip, she traces out a particular path through the space
of events, and Alice, staying at home, traces another. Although
events play out in cause-and-eect order along each of these paths,
we do not expect to be able to measure times along paths A and B
and have them come out the same. This is how ordered geometry
works: points can be put in a denite order along any particular
line, but not along dierent lines. Of the four primitive concepts
used in Euclids E1-E5 point, line, circle, and angle only the
non-metrical notions of point (i.e., event) and line are relevant in
ordered geometry. In a geometry without measurement, there is no
concept of measuring distance (hence no compasses or circles), or of
measuring angles. The notation [ABC] indicates that event B lies
on a line segment joining A and C, and is strictly between them.
The axioms of ordered geometry are as follows:
7
7
The axioms are summarized for convenient reference in the back of the book
on page 272. This is meant to be an informal, readable summary of the system,
pitched to the same level of looseness as Euclids E1-E5. Modern mathematicians
have found that systems like these actually need quite a bit more technical
machinery to be perfectly rigorous, so if you look up an axiomatization of ordered
geometry, or a modern axiomatization of Euclidean geometry, youll typically
nd a much more lengthy list of axioms than the ones presented here. The
axioms Im omitting take care of details like making sure that there are more
than two points in the universe, and that curves cant cut through one another
intersecting. The classic, beautifully written book on these topics is H.S.M.
Coxeters Introduction to Geometry, which is introductory in the sense that
its the kind of book a college math major might use in a rst upper-division
Section 1.4 Ordered geometry 19
b / Stephen Hawking (1942-).
a / Axioms O2 (left) and O3
(right).
O1 Two events determine a line.
O2 Line segments can be extended: given A and B, there is at
least one event such that [ABC] is true.
O3 Lines dont wrap around: if [ABC] is true, then [BCA] is false.
O4 Betweenness: For any three distinct events A, B, and C lying
on the same line, we can determine whether or not B is between
A and C (and by statement 3, this ordering is unique except
for a possible over-all reversal to form [CBA]).
O1-O2 express the same ideas as Euclids E1-E2. Not all lines
in the system will correspond physically to chains of causality; we
could have a line segment that describes a snapshot of a steel chain,
and O3-O4 then say that the order of the links is well dened. But
O3 and O4 also have clear physical signicance for lines describing
causality. O3 forbids time travel paradoxes, like going back in time
and killing our own grandmother as a child. O4 says that events
are guaranteed to have a well-dened cause-and-eect order only
if they lie on the same line. This is completely dierent from the
attitude expressed in Newtons famous statement: Absolute, true
and mathematical time, of itself, and from its own nature ows
equably without regard to anything external . . .
If youre dismayed by the austerity of a system of geometry with-
out any notion of measurement, you may be more appalled to learn
that even a system as weak as ordered geometry makes some state-
ments that are too strong to be completely correct as a foundation
for relativity. For example, if an observer falls into a black hole, at
some point he will reach a central point of innite density, called a
singularity. At this point, his chain of cause and eect terminates,
violating statement 2. It is also an open question whether O3s pro-
hibition on time-loops actually holds in general relativity; this is
Stephen Hawkings playfully named chronology protection conjec-
ture. Well also see that in general relativity O1 is almost always
true, but there are exceptions.
1.5 The equivalence principle
1.5.1 Proportionality of inertial and gravitational mass
What physical interpretation should we give to the lines de-
scribed in ordered geometry? Galileo described an experiment (which
he may or may not have actually performed) in which he simultane-
ously dropped a cannonball and a musket ball from a tall tower. The
two objects hit the ground simultaneously, disproving Aristotles as-
sertion that objects fell at a speed proportional to their weights. On
a graph of spacetime with x and t axes, the curves traced by the two
course in geometry.
a / The cannonball and the mus-
ketball have identical parabolic
world-lines. On this type of
space-time plot, space is conven-
tionally shown on the horizontal
axis, so the tower has to be
depicted on its side.
b / A piece of string held taut
on a globe forms a geodesic
from Mexico City to London.
Although it appears curved, it
is the analog of a straight line
in the non-Euclidean geometry
conned to the surface of the
Earth. Similarly, the world-lines of
gure a appear curved, but they
are the analogs of straight lines
in the non-Euclidean geometry
used to describe gravitational
elds in general relativity.
objects, called their world-lines, are identical parabolas. (The paths
of the balls through x y z space are straight, not curved.) One
way of explaining this observation is that what we call mass is re-
ally two separate things, which happen to be equal. Inertial mass,
which appears in Newtons a = F/m, describes how dicult it is
to accelerate an object. Gravitational mass describes the strength
with which gravity acts. The cannonball has a hundred times more
gravitational mass than the musket ball, so the force of gravity act-
ing on it is a hundred times greater. But its inertial mass is also
precisely a hundred times greater, so the two eects cancel out, and
it falls with the same acceleration. This is a special property of the
gravitational force. Electrical forces, for example, do not behave
this way. The force that an object experiences in an electric eld
is proportional to its charge, which is unrelated to its inertial mass,
so dierent charges placed in the same electric eld will in general
have dierent motions.
1.5.2 Geometrical treatment of gravity
Einstein realized that this special property of the gravitational
force made it possible to describe gravity in purely geometrical
terms. We dene the world-lines of objects acted on by gravity
to be the lines described by the axioms of the geometry. Since we
normally think of the lines described by Euclidean geometry and
its kin as straight lines, this amounts to a redenition of what it
means for a line to be straight. By analogy, imagine stretching a
piece of string taut across a globe, as we might do in order to plan
an airplane ight or aim a directional radio antenna. The string
may not appear straight as viewed from the three-dimensional Eu-
clidean space in which the globe is embedded, but it is as straight
as possible in the sense that it is the path followed by a radio wave,
8
or by an airplane pilot who keeps her wings level and her rudder
straight. The world-line of an object acted on by nongravita-
tional forces is not considered to be a straight line in the sense of
O1-O4. When necessary, one eliminates this ambiguity in the over-
loaded term line by referring to the lines of O1-O4 as geodesics.
The world-line of a low-mass object acted on only by gravity is one
type of geodesic.
9
We can now see the deep physical importance of statement O1,
that two events determine a line. To predict the trajectory of a
golf ball, we need to have some initial data. For example, we could
measure event A when the ball breaks contact with the club, and
event B an innitesimal time after A.
10
This pair of observations can
be thought of as xing the balls initial position and velocity, which
8
Radio waves in the HF band tend to be trapped between the ground and
the ionosphere, causing them to curve over the horizon, allowing long-distance
communication.
9
For more justication of this statement, see ch. 8, problem 1, on page 239.
10
Regarding innitesimals, see p. 77.
Section 1.5 The equivalence principle 21
c / Lor and E otv os (1848-1919).
d / If the geodesics dened
by an airplane and a radio wave
differ from one another, then
it is not possible to treat both
problems exactly using the same
geometrical theory. In general
relativity, this would be analogous
to a violation of the equivalence
principle. General relativitys
validity as a purely geometrical
theory of gravity requires that the
equivalence principle be exactly
satised in all cases.
should be enough to predict a unique world-line for the ball, since
relativity is a deterministic theory. With this interpretation, we can
also see why it is not necessarily a disaster for the theory if O1
fails sometimes. For example, event A could mark the launching of
two satellites into circular orbits from the same place on the Earth,
heading in opposite directions, and B could be their subsequent
collision on the opposite side of the planet. Although this violates
O1, it doesnt violate determinism. Determinism only requires the
validity of O1 for events innitesimally close together. Even for
randomly chosen events far apart, the probability that they will
violate O1 is zero.
1.5.3 E otv os experiments
Einsteins entire system breaks down if there is any violation, no
matter how small, of the proportionality between inertial and grav-
itational mass, and it therefore becomes very interesting to search
experimentally for such a violation. For example, we might won-
der whether neutrons and protons had slightly dierent ratios of
gravitational and inertial mass, which in a Galileo-style experiment
would cause a small dierence between the acceleration of a lead
weight, with a large neutron-to-proton ratio, and a wooden one,
which consists of light elements with nearly equal numbers of neu-
trons and protons. The rst high-precision experiments of this type
were performed by Eotvos around the turn of the twentieth century,
and they veried the equivalence of inertial and gravitational mass
to within about one part in 10
8
. These are generically referred to
as Eotvos experiments.
Figure e shows a strategy for doing Eotvos experiments that al-
lowed a test to about one part in 10
12
. The top panel is a simplied
version. The platform is balanced, so the gravitational masses of
the two objects are observed to be equal. The objects are made
of dierent substances. If the equivalence of inertial and gravita-
tional mass fails to hold for these two substances, then the force
of gravity on each mass will not be exact proportion to its inertia,
and the platform will experience a slight torque as the earth spins.
The bottom panel shows a more realistic drawing of an experiment
by Braginskii and Panov.
11
The whole thing was encased in a tall
vacuum tube, which was placed in a sealed basement whose tem-
perature was controlled to within 0.02
C. The total mass of the

platinum and aluminum test masses, plus the tungsten wire and
the balance arms, was only 4.4 g. To detect tiny motions, a laser
beam was bounced o of a mirror attached to the wire. There was
so little friction that the balance would have taken on the order of
several years to calm down completely after being put in place; to
stop these vibrations, static electrical forces were applied through
the two circular plates to provide very gentle twists on the ellipsoidal
11
V.B. Braginskii and V.I. Panov, Soviet Physics JETP 34, 463 (1972).
e / An E otv os experiment.
Top: simplied version. Bottom:
realistic version by Braginskii and
Panov. (Drawing after Braginskii
and Panov.)
mass between them.
1.5.4 Equivalence of gravitational elds and accelerations
One consequence of the Eotvos experiments null results is that
it is not possible to tell the dierence between an acceleration and
a gravitational eld. At certain times during Bettys eld trip, she
feels herself pressed against her seat, and she interprets this as ev-
idence that shes in a space vessel that is undergoing violent accel-
erations and decelerations. But its equally possible that Tutorbot
has simply arranged for her capsule to be hung from a rope and
dangled into the gravitational eld of a planet. Suppose that the
rst explanation is correct. The capsule is initially at rest in outer
space, where there is no gravity. Betty can release a pencil and a
lead ball in the air inside the cabin, and they will stay in place. The
capsule then accelerates, and to Betty, who has adopted a frame
of reference tied to its deck, ceiling and walls, it appears that the
pencil and the ball fall to the deck. They are guaranteed to stay
side by side until they hit the deckplates, because in fact they arent
accelerating; they simply appear to accelerate, when in reality its
the deckplates that are coming up and hitting them. But now con-
sider the second explanation, that the capsule has been dipped into
a gravitational eld. The ball and the pencil will still fall side by
side to the oor, because they have the same ratio of gravitational
to inertial mass.
1.5.5 The equivalence principle
This leads to one way of stating a central principle of relativity
known as the equivalence principle: Accelerations and gravitational
elds are equivalent. There is no experiment that can distinguish
one from the other.
12
To see what a radical departure this is, we need to compare with
the completely dierent picture presented by Newtonian physics and
special relativity. Newtons law of inertia states that Every object
perseveres in its state of rest, or of uniform motion in a straight
line, unless it is compelled to change that state by forces impressed
thereon.
13
Newtons intention here was to clearly state a contra-
diction of Aristotelian physics, in which objects were supposed to
naturally stop moving and come to rest in the absence of a force. For
Aristotle, at rest meant at rest relative to the Earth, which repre-
sented a special frame of reference. But if motion doesnt naturally
stop of its own accord, then there is no longer any way to single out
one frame of reference, such as the one tied to the Earth, as being
special. An equally good frame of reference is a car driving in a
straight line down the interstate at constant speed. The earth and
the car both represent valid inertial frames of reference, in which
12
This statement of the equivalence principle is summarized, along with some
other forms of it to be encountered later, in the back of the book on page 273.
13
paraphrased from a translation by Motte, 1729
Newtons law of inertia is valid. On the other hand, there are other,
noninertial frames of reference, in which the law of inertia is vio-
lated. For example, if the car decelerates suddenly, then it appears
to the people in the car as if their bodies are being jerked forward,
even though there is no physical object that could be exerting any
type of forward force on them. This distinction between inertial and
noninertial frames of reference was carried over by Einstein into his
theory of special relativity, published in 1905.
But by the time he published the general theory in 1915, Einstein
had realized that this distinction between inertial and noninertial
frames of reference was fundamentally suspect. How do we know
that a particular frame of reference is inertial? One way is to verify
that its motion relative to some other inertial frame, such as the
Earths, is in a straight line and at constant speed. But how does
the whole thing get started? We need to bootstrap the process
with at least one frame of reference to act as our standard. We
can look for a frame in which the law of inertia is valid, but now
we run into another diculty. To verify that the law of inertia
holds, we have to check that an observer tied to that frame doesnt
see objects accelerating for no reason. The trouble here is that by
the equivalence principle, there is no way to determine whether the
object is accelerating for no reason or because of a gravitational
force. Betty, for example, cannot tell by any local measurement
(i.e., any measurement carried out within the capsule) whether she
is in an inertial or a noninertial frame.
f / Wouldnt it be nice if we could dene the meaning of an inertial frame of reference? Newton makes
it sound easy: to dene an inertial frame, just nd some object that is not accelerating because it is not being
acted on by any external forces. But what object would we use? The earth? The xed stars? Our galaxy?
Our supercluster of galaxies? All of these are accelerating relative to something.
We could hope to resolve the ambiguity by making non-local
measurements instead. For example, if Betty had been allowed to
look out a porthole, she could have tried to tell whether her capsule
was accelerating relative to the stars. Even this possibility ends up
not being satisfactory. The stars in our galaxy are moving in circular
orbits around the galaxy. On an even larger scale, the universe is
expanding in the aftermath of the Big Bang. It spent about the
g / Two local Lorentz frames.
rst half of its history decelerating due to gravitational attraction,
but the expansion is now observed to be accelerating, apparently
due to a poorly understood phenomenon referred to by the catch-all
term dark energy. In general, there is no distant background of
physical objects in the universe that is not accelerating.
The conclusion is that we need to abandon the entire distinction
between inertial and noninertial frames of reference. The best that
we can do is to single out certain frames of reference dened by the
motion of objects that are not subject to any nongravitational forces.
A falling rock denes such a frame of reference. In this frame, the
rock is at rest, and the ground is accelerating. The rocks world-line
is a straight line of constant x = 0 and varying t. Such a free-falling
frame of reference is called a Lorentz frame. The frame of reference
dened by a rock sitting on a table is an inertial frame of reference
according to the Newtonian view, but it is not a Lorentz frame.
In Newtonian physics, inertial frames are preferable because they
make motion simple: objects with no forces acting on them move
along straight world-lines. Similarly, Lorentz frames occupy a privi-
leged position in general relativity because they make motion simple:
objects move along straight world-lines if they have no nongravita-
tional forces acting on them.
1.5.6 Locality of Lorentz frames
It would be convenient if we could dene a single Lorentz frame
that would cover the entire universe, but we cant. In gure g, two
girls simultaneously drop down from tree branches one in Los An-
geles and one in Mumbai. The girl free-falling in Los Angeles denes
a Lorentz frame, and in that frame, other objects falling nearby will
also have straight world-lines. But in the LA girls frame of refer-
ence, the girl falling in Mumbai does not have a straight world-line:
she is accelerating up toward the LA girl with an acceleration of
about 2g.
A second way of stating the equivalence principle is that it is
always possible to dene a local Lorentz frame in a particular neigh-
borhood of spacetime.
14
It is not possible to do so on a universal
basis.
The locality of Lorentz frames can be understood in the anal-
ogy of the string stretched across the globe. We dont notice the
curvature of the Earths surface in everyday life because the radius
of curvature is thousands of kilometers. On a map of LA, we dont
notice any curvature, nor do we detect it on a map of Mumbai, but
it is not possible to make a at map that includes both LA and
Mumbai without seeing severe distortions.
14
This statement of the equivalence principle is summarized, along with some
other forms of it, in the back of the book on page 273.
h / Chiaos paradox: a charged
particle and a neutral particle are
in orbit around the earth. Will the
charged particle radiate, violating
the equivalence principle?
Chiaos paradox
The remainder of this subsection deals with the subtle ques-
tion of whether and how the equivalence principle can be applied to
charged particles. You may wish to skip it on a rst reading. The
short answer is that using the equivalence principle to make con-
clusions about charged particles is like the attempts by slaveholders
and abolitionists in the 19th century U.S. to support their positions
based on the Bible: you can probably prove whichever conclusion
was the one you set out to prove.
The equivalence principle is not a single, simple, mathemati-
cally well dened statement. As an example of an ambiguity that is
still somewhat controversial, 90 years after Einstein rst proposed
the principle, consider the question of whether or not it applies to
charged particles. Raymond Chiao
15
proposes the following thought
experiment, which Ill refer to as Chiaos paradox. Let a neutral
particle and a charged particle be set, side by side, in orbit around
the earth. Assume that the earth has no electric or magnetic eld
(or, more realistically, that the charged particle is surrounded by a
Faraday cage and mu-metal shielding). If the equivalence principle
applies regardless of charge, then these two particles must go on
orbiting amicably, side by side. But then we have a violation of
conservation of energy, since the charged particle, which is acceler-
ating, will radiate electromagnetic waves (with very low frequency
and amplitude). It seems as though the particles orbit must decay.
The resolution of the paradox, as demonstrated by hairy cal-
culations
16
is interesting because it exemplies the local nature of
the equivalence principle. When a charged particle moves through a
gravitational eld, in general it is possible for the particle to experi-
ence a reaction from its own electromagnetic elds. This might seem
impossible, since an observer in a frame momentarily at rest with
respect to the particle sees the radiation y o in all directions at
the speed of light. But there are in fact several dierent mechanisms
by which a charged particle can be reunited with its long-lost elec-
tromagnetic ospring. An example (not directly related to Chiaos
scenario) is the following.
Bring a laser very close to a black hole, but not so close that it
has strayed inside the event horizon, which is the spherical point of
no return from within which nothing can escape. Example 10 on
page 50 gives a plausibility argument based on Newtonian physics
15
arxiv.org/abs/quant-ph/0601193v7
16
The rst detailed calculation appears to have been by Cecile and Bryce
DeWitt, Falling Charges, Physics, 1 (1964) 3. This paper is unfortunately
very dicult to obtain now. A more recent treatment by Grn and Nss is
accessible at arxiv.org/abs/0806.0464v1. A full exposition of the techniques
is given by Poisson, The Motion of Point Particles in Curved Spacetime, www.
livingreviews.org/lrr-2004-6.
i / 1. A photon is emitted upward
from the oor of the elevator. The
elevator accelerates upward. 2.
By the time the photon is de-
tected at the ceiling, the elevator
has changed its velocity, so the
photon is detected with a Doppler
shift.
that the radius
17
of the event horizon should be something like r
H
=
GM/c
2
, and section 6.3.2 on page 186 derives the relativistically
correct factor of 2 in front, so that r
H
= 2GM/c
2
. It turns out
that at r = (3/2)R
H
, a ray of light can have a circular orbit around
the black hole. Since this is greater than R
H
, we can, at least in
theory, hold the laser stationary at this value of r using a powerful
rocket engine. If we point the laser in the azimuthal direction, its
own beam will come back and hit it.
Since matter can experience a back-reaction from its own elec-
tromagnetic radiation, it becomes plausible how the paradox can be
resolved. The equivalence principle holds locally, i.e., within a small
patch of space and time. If Chiaos charged and neutral particle are
released side by side, then they will obey the equivalence principle
for at least a certain amount of time and for at least a certain
amount of time is all we should expect, since the principle is local.
But after a while, the charged particle will start to experience a
back-reaction from its own electromagnetic elds, and this causes
its orbit to decay, satisfying conservation of energy. Since Chiaos
particles are orbiting the earth, and the earth is not a black hole, the
mechanism clearly cant be as simple as the one described above,
but Grn and Nss show that there are similar mechanisms that
can apply here, e.g., scattering of light waves by the nonuniform
gravitational eld.
It is worth keeping in mind the DeWitts caution that The ques-
tions answered by this investigation are of conceptual interest only,
since the forces involved are far too small to be detected experimen-
tally (see problem 7, p. 34).
1.5.7 Gravitational redshifts
Starting on page 15, we saw experimental evidence that the rate
of ow of time changes with height in a gravitational eld. We can
now see that this is required by the equivalence principle.
By the equivalence principle, there is no way to tell the dierence
between experimental results obtained in an accelerating laboratory
and those found in a laboratory immersed in a gravitational eld.
18
In a laboratory accelerating upward, a photon emitted from the
oor and would be Doppler-shifted toward lower frequencies when
observed at the ceiling, because of the change in the receivers ve-
locity during the photons time of ight. The eect is given by
E/E = f/f = ay/c
2
, where a is the labs acceleration, y is the
height from oor to ceiling, and c is the speed of light.
17
Because relativity describes gravitational elds in terms of curvature of
spacetime, the Euclidean relationship between the radius and circumference of
a circle fails here. The r coordinate should be understood here not as the radius
measured from the center but as the circumference divided by 2.
18
Problem 4 on p. 33 veries, in one specic example, that this way of stating
the equivalence principle is implied by the one on p. 20.
j / An electromagnetic wave
strikes an ohmic surface. The
waves electric eld excites an
oscillating current density J. The
waves magnetic eld then acts
on these currents, producing
a force in the direction of the
waves propagation. This is a
pre-relativistic argument that light
must possess inertia.
Self-check: Verify this statement.
By the equivalence principle, we nd that when such an experi-
ment is done in a gravitational eld g, there should be a gravitational
eect on the energy of a photon equal to E/E = gy/c
2
. Since the
quantity gy is the gravitational energy per unit mass, the photons
fractional loss of energy is the same as the (Newtonian) loss of en-
ergy experienced by a material object of mass m and initial kinetic
energy mc
2
.
The interpretation is as follows. Classical electromagnetism re-
quires that electromagnetic waves have inertia. For example, if a
plane wave strikes an ohmic surface, as in gure j, the waves elec-
tric eld excites oscillating currents in the surface. These currents
then experience a magnetic force from the waves magnetic eld,
and application of the right-hand rule shows that the resulting force
is in the direction of propagation of the wave. Thus the light wave
acts as if it has momentum. The equivalence principle says that
whatever has inertia must also participate in gravitational interac-
tions. Therefore light waves must have weight, and must lose energy
when they rise through a gravitational eld.
Self-check: Verify the application of the right-hand rule de-
scribed above.
Further interpretation:
The quantity mc
2
is famous, even among people who dont
know what m and c stand for. This is the rst hint of where
it comes from. The full story is given in section 4.2.2.
The relation p = E/c between the energy and momentum
of a light wave follows directly from Maxwells equations, by
the argument above; however, we will see in section 4.2.2 that
according to relativity this relation must hold for any massless
particle
What we have found agrees with Niels Bohrs correspondence
principle, which states that when a new physical theory, such
as relativity, replaces an older one, such as Newtonian physics,
the new theory must agree with the old one under the experi-
mental conditions in which the old theory had been veried by
experiments. The gravitational mass of a beam of light with
energy E is E/c
2
, and since c is a big number, it is not sur-
prising that the weight of light rays had never been detected
before Einstein trying to detect it.
This book describes one particular theory of gravity, Einsteins
theory of general relativity. There are other theories of grav-
ity, and some of these, such as the Brans-Dicke theory, do
just as well as general relativity in agreeing with the presently
available experimental data. Our prediction of gravitational
Doppler shifts of light only depended on the equivalence princi-
ple, which is one ingredient of general relativity. Experimental
tests of this prediction only test the equivalence principle; they
do not allow us to distinguish between one theory of gravity
and another if both theories incorporate the equivalence prin-
ciple.
Chiaos paradox revisited Example: 1
The equivalence principle says that electromagnetic waves have
gravitational mass as well as inertial mass, so it seems clear that
the same must hold for static elds. In Chiaos paradox (p. 34), the
orbiting charged particle has an electric eld that extends out to
innity. When we measure the mass of a charged particle such as
an electron, there is no way to separate the mass of this eld from
a more localized contribution. The electric eld falls through the
gravitational eld, and the equivalence principle, which is local,
cannot guarantee that all parts of the eld rotate uniformly about
the earth, even in distant parts of the universe. The electric eld
pattern becomes distorted, and this distortion causes a radiation
reaction which back-reacts on the particle, causing its orbit to de-
cay.
1.5.8 The Pound-Rebka experiment
The 1959 Pound-Rebka experiment at Harvard
19
was the rst
quantitative test of the equivalence principle to be carried out under
controlled conditions, and in this section we will discuss it in detail.
When y is on the order of magnitude of the height of a building,
the value of E/E = gy/c
2
is 10
14
, so an extremely high-
precision experiment is necessary in order to detect a gravitational
redshift. A number of other eects are big enough to obscure it en-
tirely, and must somehow be eliminated or compensated for. These
are listed below, along with their orders of magnitude in the exper-
imental design nally settled on by Pound and Rebka.
19
Phys. Rev. Lett. 4 (1960) 337
k / The Pound-Rebka experi-
ment.
l / Emission of 14 keV gamma-
rays by
57
Fe. The parent nucleus
57
Co absorbs an electron and
undergoes a weak-force decay
process that converts it into
57
Fe,
in an excited state. With 85%
probability, this state decays to a
state just above the ground state,
with an excitation energy of 14
keV and a half-life of 10
7
s. This
state nally decays, either by
gamma emission or emission of
an internal conversion electron,
to the ground state.
(1) Classical Doppler broadening due to temper-
ature. Thermal motion causes Doppler shifts of
emitted photons, corresponding to the random
component of the emitting atoms velocity vec-
tor along the direction of emission.
10
6
(2) The recoil Doppler shift. When an atom
emits a photon with energy E and momentum
p = E/c, conservation of momentum requires
that the atom recoil with momentum p = E/c
and energy p
2
/2m. This causes a downward
Doppler shift of the energy of the emitted pho-
ton. A similar eect occurs on absorption, dou-
bling the problem.
10
12
(3) Natural line width. The Heisenberg uncer-
tainty principle says that a state with a half-life
must have an uncertainty in its energy of at
least h/, where h is Plancks constant.
10
12
(4) Special-relativistic Doppler shift due to tem-
perature. Section 1.2 presented experimental ev-
idence that time ows at a dierent rate depend-
ing on the motion of the observer. Therefore
the thermal motion of an atom emitting a pho-
ton has an eect on the frequency of the photon,
even if the atoms motion is not along the line of
emission. The equations needed in order to cal-
culate this eect will not be derived until section
2.2; a quantitative estimate is given in example
9 on page 48. For now, we only need to know
that this leads to a temperature-dependence in
the average frequency of emission, in addition
to the broadening of the bell curve described by
eect (1) above.
10
14
per
degree C
The most straightforward way to mitigate eect (1) is to use
photons emitted from a solid. At rst glance this would seem like
a bad idea, since electrons in a solid emit a continuous spectrum of
light, not a discrete spectrum like the ones emitted by gases; this
is because we have N electrons, where N is on the order of Avo-
gadros number, all interacting strongly with one another, so by the
correspondence principle the discrete quantum-mechanical behavior
must be averaged out. But the protons and neutrons within one
nucleus do not interact much at all with those in other nuclei, so
the photons emitted by a nucleus do have a discrete spectrum. The
energy scale of nuclear excitations is in the keV or MeV range, so
these photons are x-rays or gamma-rays. Furthermore, the time-
scale of the random vibrations of a nucleus in a solid are extremely
short. For a velocity on the order of 100 m/s, and vibrations with an
amplitude of 10
10
m, the time is about 10
12
s. In many cases,
this is much shorter than the half-life of the excited nuclear state
m / Top: A graph of velocity
versus time for the source. The
velocity has both a constant
component and an oscillating one
with a frequency of 10-50 Hz.
The constant component v
o
was
used as a way of determining the
calibration of frequency shift as a
function of count rates. Data were
acquired during the quarter-cycle
periods of maximum oscillatory
velocity, 1 and 2. Bottom: Count
rates as a function of velocity,
for v
o
= 0 and v
1
,= 0. The
dashed curve and black circles
represent the count rates that
would have been observed if
there were no gravitational effect.
The gravitational effect shifts
the resonance curve to one side
(solid curve), resulting in an
asymmetry of the count rates
(open circles). The shift, and the
resulting asymmetry, are greatly
exaggerated for readability; in
reality, the gravitational effect was
500 times smaller than the width
of the resonance curve.
emitting the gamma-ray, and therefore the Doppler shift averages
out to nearly zero.
Eect (2) is still much bigger than the 10
14
size of the eect to
be measured. It can be avoided by exploiting the Mossbauer eect,
in which a nucleus in a solid substance at low temperature emits or
absorbs a gamma-ray photon, but with signicant probability the
recoil is taken up not by the individual nucleus but by a vibration
of the atomic lattice as a whole. Since the recoil energy varies as
p
2
/2m, the large mass of the lattice leads to a very small dissipation
of energy into the recoiling lattice. Thus if a photon is emitted and
absorbed by identical nuclei in a solid, and for both emission and
absorption the recoil momentum is taken up by the lattice as a
whole, then there is a negligible energy shift. One must pick an
isotope that emits photons with energies of about 10-100 keV. X-
rays with energies lower than about 10 keV tend to be absorbed
strongly by matter and are dicult to detect, whereas for gamma-
ray energies 100 keV the Mossbauer eect is not sucient to
eliminate the recoil eect completely enough.
If the Mossbauer eect is carried out in a horizontal plane, reso-
nant absorption occurs. When the source and absorber are aligned
vertically, k, gravitational frequency shifts should cause a mismatch,
destroying the resonance. One can move the source at a small ve-
locity (typically a few mm/s) in order to add a Doppler shift onto
the frequency; by determining the velocity that compensates for the
gravitational eect, one can determine how big the gravitational
eect is.
The typical half-life for deexcitation of a nucleus by emission
of a gamma-ray with energy E is in the nanosecond range. To
measure an gravitational eect at the 10
14
level, one would like to
have a natural line width, (3), with E/E 10
14
, which would
require a half-life of 10 s. In practice, Pound and Rebka found
that other eects, such as (4) and electron-nucleus interactions that
depended on the preparation of the sample, tended to put nuclei
in one sample out of tune with those in another sample at the
10
13
-10
12
level, so that resonance could not be achieved unless
the natural line width gave E/E 10
12
. As a result, they settled
on an experiment in which 14 keV gammas were emitted by
57
Fe
nuclei (gure l) at the top of a 22-meter tower, and absorbed by
57
Fe
nuclei at the bottom. The 100-ns half-life of the excited state leads
to E/E 10
12
. This is 500 times greater than the gravitational
eect to be measured, so, as described in more detail below, the
experiment depended on high-precision measurements of small up-
and-down shifts of the bell-shaped resonance curve.
The absorbers were seven iron lms isotopically enhanced in
57
Fe, applied directly to the faces of seven sodium-iodide scintil-
lation detectors (bottom of gure k). When a gamma-ray impinges
on the absorbers, a number of dierent things can happen, of which
n / Pound and Rebka at the
top and bottom of the tower.
we can get away with considering only the following: (a) the gamma-
ray is resonantly absorbed in one of the
57
Fe absorbers, after which
the excited nucleus decays by re-emission of another such photon (or
a conversion electron), in a random direction; (b) the gamma-ray
passes through the absorber and then produces ionization directly
in the sodium iodide crystal. In case b, the gamma-ray is detected.
In case a, there is a 50% probability that the re-emitted photon will
come out in the upward direction, so that it cannot be detected.
Thus when the conditions are right for resonance, a reduction in
count rate is expected. The Mossbauer eect never occurs with
100% probability; in this experiment, about a third of the gammas
incident on the absorbers were resonantly absorbed.
The choice of y = 22 m was dictated mainly by systematic er-
rors. The experiment was limited by the strength of the gamma-ray
source. For a source of a xed strength, the count rate in the de-
tector at a distance y would be proportional to y
2
, leading to sta-
tistical errors proportional to 1/
count rate y. Since the eect

to be measured is also proportional to y, the signal-to-noise ratio
was independent of y. However, systematic eects such as (4) were
easier to monitor and account for when y was fairly large. A lab
building at Harvard happened to have a 22-meter tower, which was
used for the experiment. To reduce the absorption of the gammas
in the 22 meters of air, a long, cylindrical mylar bag full of helium
gas was placed in the shaft, k.
The resonance was a bell-shaped curve with a minimum at the
natural frequency of emission. Since the curve was at a minimum,
where its derivative was zero, the sensitivity of the count rate to
the gravitational shift would have been nearly zero if the source had
been stationary. Therefore it was necessary to vibrate the source
up and down, so that the emitted photons would be Doppler shifted
onto the shoulders of the resonance curve, where the slope of the
curve was large. The resulting asymmetry in count rates is shown
in gure m. A further eort to cancel out possible systematic eects
was made by frequently swapping the source and absorber between
the top and bottom of the tower.
For y = 22.6 m, the equivalence principle predicts a fractional
frequency shift due to gravity of 2.46 10
15
. Pound and Rebka
measured the shift to be (2.56 0.25) 10
15
. The results were in
statistical agreement with theory, and veried the predicted size of
the eect to a precision of 10%.
Problems
1 In classical mechanics, one hears the term the acceleration of
gravity, which doesnt literally make sense, since it is objects that
accelerate. Explain why this terms usefulness is dependent on the
equivalence principle.
2 The New Horizons space probe communicates with the earth
using microwaves with a frequency of about 10 GHz. Estimate the
sizes of the following frequency shifts in this signal, when the probe
ies by Pluto in 2015, at a velocity of 10 A.U./year: (a) the
Doppler shift due to the probes velocity; (b) the Doppler shift due
to the Earths orbital velocity; (c) the gravitational Doppler shift.
3 Euclids axioms E1-E5 (p. 18) do not suce to prove that
there are an innite number of points in the plane, and therefore
they need to be supplemented by an extra axiom that states this
(unless one nds the nonstandard realizations with nitely many
points to be interesting enough to study for their own sake). Prove
that the axioms of ordered geometry O1-O4 on p. 19 do not have
this problem. Solution, p. 260
4 In the science ction novel Have Spacesuit Will Travel, by
Robert Heinlein, Kip, a high school student, answers a radio distress
call, encounters a ying saucer, and is knocked out and kidnapped
by aliens. When he wakes up, he nds himself in a locked cell
with a young girl named Peewee. Peewee claims theyre aboard an
accelerating spaceship. If this was a spaceship, Kip thinks. The
oor felt as solid as concrete and motionless.
The equivalence principle can be stated in a variety of ways. On
p. 20, I stated it as (1) gravitational and inertial mass are always
proportional to one another. An alternative formulation (p. 27)
is (2) that Kip has no way, by experiments or observarions inside
his sealed prison cell, to determine whether hes in an accelerating
spaceship or on the surface of a planet, experiencing its gravitational
eld.
(a) Show that any violation of statement 1 also leads to a violation of
statement 2. (b) If wed intended to construct a geometrical theory
of gravity roughly along the lines of axioms O1-O4 on p. 19, which
axiom is violated in this scenario? Solution, p. 260
5 Clock A sits on a desk. Clock B is tossed up in the air from
the same height as the desk and then comes back down. Compare
the elapsed times. Hint, p. 260 Solution, p. 260
6 Consider the following physical situations: (1) a charged
object lies on a desk on the planet earth; (2) a charged object orbits
the earth; (3) a charged object is released above the earths surface
and dropped straight down; (4) a charged object is subjected to a
Problems 33
constant acceleration by a rocket engine in outer space. In each case,
we want to know whether the charge radiates. Analyze the physics
in each case (a) based on conservation of energy; (b) by determining
whether the objects motion is inertial in the sense intended by Isaac
Newton; (c) using the most straightforward interpretation of the
equivalence principle (i.e., not worrying about the issues discussed
on p. that surround the ambiguous denition of locality).
Solution, p. 260
7 Consider the physical situation depicted in gure h, p. 26.
Let a
g
be the gravitational acceleration and a
r
the acceleration of
the charged particle due to radiation. Then a
r
/a
g
measures the vi-
olation of the equivalence principle. The goal of this problem is to
make an order-of-magnitude estimate of this ratio in the case of a
neutron and a proton in low earth orbit.
(a) Let m the mass of each particle, and q the charge of the charged
particle. Without doing a full calculation like the ones by the De-
Witts and Grn and Nss, use general ideas about the frequency-
scaling of radiation (see section 8.2.5, p. 236) to nd the proportion-
ality that gives the dependence of a
r
/a
g
on q, m, and any convenient
parameters of the orbit.
(b) Based on considerations of units, insert the necessary universal
constants into your answer from part a.
(c) The result from part b will still be o by some unitless factor, but
we expect this to be of order unity. Under this assumption, make
an order-of-magnitude estimate of the violation of the equivalence
principle in the case of a neutron and a proton in low earth orbit.
Solution, p. 261
a / Hendrik Antoon Lorentz
(1853-1928)
Chapter 2
Geometry of Flat
Spacetime
2.1 Afne properties of Lorentz geometry
The geometrical treatment of space, time, and gravity only requires
as its basis the equivalence of inertial and gravitational mass. That
equivalence holds for Newtonian gravity, so it is indeed possible
to redo Newtonian gravity as a theory of curved spacetime. This
project was carried out by the French mathematician Cartan, as
summarized very readably in section 17.5 of The Road to Reality by
Roger Penrose. The geometry of the local reference frames is very
simple. The three space dimensions have an approximately Eu-
clidean geometry, and the time dimension is entirely separate from
them. This is referred to as a Euclidean spacetime with 3+1 dimen-
sions. Although the outlook is radically dierent from Newtons, all
of the predictions of experimental results are the same.
The experiments in section 1.2 show, however, that there are
real, experimentally veriable violations of Newtons laws. In New-
tonian physics, time is supposed to ow at the same rate everywhere,
which we have found to be false. The ow of time is actually depen-
dent on the observers state of motion through space, which shows
that the space and time dimensions are intertwined somehow. The
geometry of the local frames in relativity therefore must not be as
simple as Euclidean 3+1. Their actual geometry was implicit in
Einsteins 1905 paper on special relativity, and had already been
developed mathematically, without the full physical interpretation,
by Hendrik Lorentz. Lorentzs and Einsteins work were explicitly
connected by Minkowski in 1907, so a Lorentz frame is often referred
to as a Minkowski frame.
To describe this Lorentz geometry, we need to add more struc-
ture on top of the axioms O1-O4 of ordered geometry, but it will not
be the additional Euclidean structure of E3-E4, it will be something
dierent.
To see how to proceed, lets consider the bare minimum of geo-
metrical apparatus that would be necessary in order to set up frames
of reference. The following argument shows that the main miss-
ing ingredient is merely a concept of parallelism. We only expect
Lorentz frames to be local, but we do need them to be big enough
35
b / Objects are released at
rest at spacetime events P and
Q. They remain at rest, and their
world-lines dene a notion of
parallelism.
c / There is no well-dened
angular measure in this ge-
ometry. In a different frame of
reference, the angles are not
right angles.
d / Simultaneity is not well
dened. The constant-time lines
PQ and RS from gure b are
not constant-time lines when
observed in a different frame of
reference.
to cover at least some amount of spacetime. If Betty does an Eotvos
experiment by releasing a pencil and a lead ball side by side, she
is essentially trying to release them at the same event A, so that
she can observe them later and determine whether their world-lines
stay right on top of one another at point B. That was all that was
required for the Eotvos experiment, but in order to set up a Lorentz
frame we need to start dealing with objects that are not right on top
of one another. Suppose we release two lead balls in two dierent
locations, at rest relative to one another. This could be the rst step
toward adding measurement to our geometry, since the balls mark
two points in space that are separated by a certain distance, like
two marks on a ruler, or the goals at the ends of a soccer eld. Al-
though the balls are separated by some nite distance, they are still
close enough together so that if there is a gravitational eld in the
area, it is very nearly the same in both locations, and we expect the
distance dened by the gap between them to stay the same. Since
they are both subject only to gravitational forces, their world-lines
are by denition straight lines (geodesics). The goal here is to end
up with some kind of coordinate grid dening a (t, x) plane, and on
such a grid, the two balls world-lines are vertical lines. If we release
them at events P and Q, then observe them again later at R and
S, PQRS should form a rectangle on such a plot. In the gure, the
irregularly spaced tick marks along the edges of the rectangle are
meant to suggest that although ordered geometry provides us with a
well-dened ordering along these lines, we have not yet constructed
a complete system of measurement.
The depiction of PQSR as a rectangle, with right angles at its
vertices, might lead us to believe that our geometry would have
something like the concept of angular measure referred to in Euclids
E4, equality of right angles. But this is too naive even for the
Euclidean 3+1 spacetime of Newton and Galileo. Suppose we switch
to a frame that is moving relative to the rst one, so that the balls
are not at rest. In the Euclidean spacetime, time is absolute, so
events P and Q would remain simultaneous, and so would R and
S; the top and bottom edges PQ and RS would remain horizontal
on the plot, but the balls world-lines PR and QS would become
slanted. The result would be a parallelogram. Since observers in
dierent states of motion do not agree on what constitutes a right
angle, the concept of angular measure is clearly not going to be
useful here. Similarly, if Euclid had observed that a right angle
drawn on a piece of paper no longer appeared to be a right angle
when the paper was turned around, he would never have decided
that angular measure was important enough to be enshrined in E4.
In the context of relativity, where time is not absolute, there is
not even any reason to believe that dierent observers must agree on
the simultaneity of PQ and RS. Our observation that time ows dif-
ferently depending on the observers state of motion tells us speci-
36 Chapter 2 Geometry of Flat Spacetime
cally to expect this not to happen when we switch to a frame moving
to the relative one. Thus in general we expect that PQRS will be
distorted into a form like the one shown in gure d. We do expect,
however, that it will remain a parallelogram; a Lorentz frame is one
in which the gravitational eld, if any, is constant, so the properties
of spacetime are uniform, and by symmetry the new frame should
still have PR=QS and PQ=RS.
With this motivation, we form the system of ane geometry by
adding the following axioms to set O1-O4.
1
The notation [PQRS]
means that events P, Q, S, and R form a parallelogram, and is
dened as the statement that the lines determined by PQ and RS
never meet at a point, and similarly for PR and QS.
A1 Constructibility of parallelograms: Given any P, Q, and R,
there exists S such that [PQRS], and if P, Q, and R are distinct
then S is unique.
A2 Symmetric treatment of the sides of a parallelogram: If [PQRS],
then [QRSP], [QPSR], and [PRQS].
A3 Lines parallel to the same line are parallel to one another: If
[ABCD] and [ABEF], then [CDEF].
The following theorem is a stronger version of Playfairs axiom
E5, the interpretation being that ane geometry describes a space-
time that is locally at.
Theorem: Given any line and any point P not on the line, there
exists a unique line through P that is parallel to .
This is stronger than E5, which only guarantees uniqueness, not
existence. Informally, the idea here is that A1 guarantees the exis-
tence of the parallel, and A3 makes it unique.
2
Although these new axioms do nothing more than to introduce
the concept of parallelism lacking in ordered geometry, it turns out
that they also allow us to build up a concept of measurement. Let
be a line, and suppose we want to dene a number system on this
1
The axioms are summarized for convenient reference in the back of the book
on page 272. This formulation is essentially the one given by Penrose, The Road
to Reality, in section 14.1.
2
Proof: Pick any two distinct points A and B on , and construct the uniquely
determined parallelogram [ABPQ] (axiom A1). Points P and Q determine a line
(axiom O1), and this line is parallel to (denition of the parallelogram). To
prove that this line is unique, we argue by contradiction. Suppose some other
parallel m to exist. If m crosses the innite line BQ at some point Z, then both
[ABPQ] and [ABPZ], so by A1, Q=Z, so the and m are the same. The only
other possibility is that m is parallel to BQ, but then the following chain of
parallelisms holds: PQ AB m BQ. By A3, lines parallel to another line are
parallel to each other, so PQ BQ, but this is a contradiction, since they have
Q in common.
Section 2.1 Afne properties of Lorentz geometry 37
f / Afne geometry gives a
well-dened centroid for the
triangle.
e / Construction of an afne
parameter.
line that measures how far apart events are. Depending on the type
of line, this could be a measurement of time, of spatial distance, or
a mixture of the two. First we arbitrarily single out two distinct
points on and label them 0 and 1. Next, pick some auxiliary point
q
0
not lying on . By A1, construct the parallelogram 01q
0
q
1
. Next
construct q
0
1q
1
2. Continuing in this way, we have a scaolding of
parallelograms adjacent to the line, determining an innite lattice of
points 1, 2, 3, . . . on the line, which represent the positive integers.
Fractions can be dened in a similar way. For example,
1
2
is dened
as the the point such that when the initial lattice segment 0
1
2
is
extended by the same construction, the next point on the lattice is
1.
The continuously varying variable constructed in this way is
called an ane parameter. The time measured by a free-falling
clock is an example of an ane parameter, as is the distance mea-
sured by the tick marks on a free-falling ruler. Since light rays travel
along geodesics, the wave crests on a light wave can even be used
analogously to the rulers tick marks.
Centroids Example: 1
The afne parameter can be used to dene the centroid of a set
of points. In the simplest example, nding the centroid of two
points, we simply bisect the line segment as described above in
the construction of the number
1
2
. Similarly, the centroid of a tri-
angle can be dened as the intersection of its three medians, the
lines joining each vertex to the midpoint of the opposite side.
Conservation of momentum Example: 2
In nonrelativistic mechanics, the concept of the center of mass
is closely related to the law of conservation of momentum. For
example, a logically complete statement of the law is that if a sys-
tem of particles is not subjected to any external force, and we
pick a frame in which its center of mass is initially at rest, then its
center of mass remains at rest in that frame. Since centroids are
well dened in afne geometry, and Lorentz frames have afne
properties, we have grounds to hope that it might be possible to
generalize the denition of momentum relativistically so that the
generalized version is conserved in a Lorentz frame. On the other
hand, we dont expect to be able to dene anything like a global
Lorentz frame for the entire universe, so there is no such natural
expectation of being able to dene a global principle of conser-
vation of momentum. This is an example of a general fact about
relativity, which is that conservation laws are difcult or impossible
to formulate globally.
Although the ane parameter gives us a system of measurement
for free in a geometry whose axioms do not even explicitly mention
measurement, there are some restrictions:
The ane parameter is dened only along straight lines, i.e.,
g / Example 3. The area of
the viola can be determined
by counting the parallelograms
formed by the lattice. The area
can be determined to any desired
precision, by dividing the parallel-
ograms into fractional parts that
are as small as necessary.
h / Example 4.
geodesics. Alices clock denes an ane parameter, but Bettys
does not, since it is subject to nongravitational forces.
We cannot compare distances along two arbitrarily chosen
lines, only along a single line or two parallel lines.
The ane parameter is arbitrary not only in the choice of its
origin 0 (which is to be expected in any case, since any frame
of reference requires such an arbitrary choice) but also in the
choice of scale. For example, there is no fundamental way of
deciding how fast to make a clock tick.
We will eventually want to lift some of these restrictions by
adding to our kit a tool called a metric, which allows us to de-
ne distances along arbitrary curves in space time, and to compare
distances in dierent directions. The ane parameter, however, will
not be entirely superseded. In particular, well nd that the metric
has a couple of properties that are not as nice as those of the ane
parameter. The square of a metric distance can be negative, and the
metric distance measured along a light ray is precisely zero, which
is not very useful.
Self-check: By the construction of the ane parameter above,
ane distances on the same line are comparable. By another con-
struction, verify the claim made above that this can be extended to
distances measured along two dierent parallel lines.
Area and volume Example: 3
It is possible to dene area and volume in afne geometry. This
is a little surprising, since distances along different lines are not
even comparable. However, we are already accustomed to mul-
tiplying and dividing numbers that have different units (a concept
that would have given Euclid conniptions), and the situation in
afne geometry is really no different. To dene area, we extend
the one-dimensional lattice to two dimensions. Any planar gure
can be superimposed on such a lattice, and dissected into paral-
lelograms, each of which has a standard area.
Area on a graph of v versus t Example: 4
If an object moves at a constant velocity v for time t, the distance
it travels can be represented by the area of a parallelogram in an
afne plane with sides having lengths v and t. These two lengths
are measured by afne parameters along two different directions,
so they are not comparable. For example, it is meaningless to
ask whether 1 m/s is greater than, less than, or equal to 1 s. If
we were graphing velocity as a function of time on a conventional
Cartesian graph, the v and t axes would be perpendicular, but
afne geometry has no notion of angular measure, so this is irrel-
evant here.
Self-check: If multiplication is dened in terms of ane area,
Section 2.1 Afne properties of Lorentz geometry 39
prove the commutative property ab = ba and the distributive rule
a(b +c) = ab +bc from axioms A1-A3.
2.2 Relativistic properties of Lorentz geometry
We now want to pin down the properties of the Lorentz geome-
try that are left unspecied by the ane treatment. This can ap-
proached either by looking for an appropriate metric, or by nding
the appropriate rules for distorting parallelograms when switching
from one frame of reference to another frame is in motion relative
to the rst. In either case, we need some further input from exper-
iments in order to show us how to proceed. We take the following
as empirical facts about at spacetime:
3
L1 Spacetime is homogeneous and isotropic. No time or place
has special properties that make it distinguishable from other
points, nor is one direction in space distinguishable from an-
other.
4
L2 Inertial frames of reference exist. These are frames in which
particles move at constant velocity if not subject to any forces.
We can construct such a frame by using a particular particle,
which is not subject to any forces, as a reference point.
L3 Equivalence of inertial frames: If a frame is in constant-velocity
translational motion relative to an inertial frame, then it is also
an inertial frame. No experiment can distinguish one preferred
inertial frame from all the others.
L4 Causality: If an observer in certain inertial frame observes
that event A causes event B, then observers in other inertial
frames agree that A causes B, not the other way around.
L5 No simultaneity: The experimental evidence in section 1.2
shows that observers in dierent inertial frames do not agree
on the simultaneity of events.
Dene ane parameters t and x for time and position, and con-
struct a (t, x) plane. Although ane geometry treats all directions
symmetrically, were going beyond the ane aspects of the space,
and t does play a dierent role than x here, as shown, for example,
by L4 and L5.
In the (t, x) plane, consider a rectangle with one corner at the
origin O. We can imagine its right and left edges as representing the
3
These facts are summarized for convenience on page 272 in the back of the
book.
4
For the experimental evidence on isotropy, see http://www.
edu-observatory.org/physics-faq/Relativity/SR/experiments.html#
Tests_of_isotropy_of_space.
a / Two objects at rest have
world-lines that dene a rect-
angle. In a second frame of
reference in motion relative to the
rst one, the rectangle becomes
a parallelogram.
world-lines of two objects that are both initially at rest in this frame;
they remain at rest (L2), so the right and left edges are parallel.
We now dene a second frame of reference such that the origins
of the two frames coincide, but they are in motion relative to one
another with velocity v. The transformation L from the rst frame
to the second is referred to as a Lorentz boost with velocity v. L
depends on v.
By homogeneity of spacetime (L1), L must be linear, so the
original rectangle will be transformed into a parallelogram in the
new frame; this is also consistent with L3, which requires that the
world-lines on the right and left edges remain parallel. The left edge
has inverse slope v. By L5 (no simultaneity), the top and bottom
edges are no longer horizontal.
For simplicity, let the original rectangle have unit area. Then
the area of the new parallelogram is still 1, by the following argu-
ment. Let the new area be A, which is a function of v. By isotropy
of spacetime (L1), A(v) = A(v). Furthermore, the function A(v)
must have some universal form for all geometrical gures, not just
for a gure that is initially a particular rectangle; this follows be-
cause of our denition of ane area in terms of a dissection by
a two-dimensional lattice, which we can choose to be a lattice of
squares. Applying boosts +v and v one after another results in a
transformation back into our original frame of reference, and since
A is universal for all shapes, it doesnt matter that the second trans-
formation starts from a parallelogram rather than a square. Scaling
the area once by A(v) and again by A(v) must therefore give back
the original square with its original unit area, A(v)A(v) = 1, and
since A(v) = A(v), A(v) = 1 for any value of v. Since A(0) = 1,
we must have A(v) = 1 for all v. The argument is independent
of the shape of the region, so we conclude that all areas are pre-
served by Lorentz boosts. The argument is also purely one about
ane geometry (it would apply equally well to a Euclidean space),
so there is no reason to expect the area A in the (t, x) plane to have
any special physical signicance in relativity; it is simply a useful
mathematical tool in the present discussion.
If we consider a boost by an innitesimal velocity dv, then the
vanishing change in area comes from the sum of the areas of the four
innitesimally thin slivers where the rectangle lies either outside the
parallelogram (call this negative area) or inside it (positive). (We
dont worry about what happens near the corners, because such
eects are of order dv
2
.) In other words, area ows around in the
x t plane, and the ows in and out of the rectangle must cancel.
Let v be positive; the ow at the sides of the rectangle is then to the
right. The ows through the top and bottom cannot be in opposite
directions (one up, one down) while maintaining the parallelism of
the opposite sides, so we have the following three possible cases:
Section 2.2 Relativistic properties of Lorentz geometry 41
b / Flows of area: (I) a shear
that preserves simultaneity, (II) a
rotation, (III) upward ow at all
edges.
I There is no ow through the top and bottom. This case cor-
responds to Galilean relativity, in which the rectangle shears
horizontally under a boost, and simultaneity is preserved, vi-
olating L5.
II Area ows downward at both the top and the bottom. The
ow is clockwise at both the positive t axis and the positive
x axis. This makes it plausible that the ow is clockwise ev-
erywhere in the (t, x) plane, and the proof is straightforward.
5
As v increases, a particular element of area ows continually
clockwise. This violates L4, because two events with a cause
and eect relationship could be time-reversed by a Lorentz
boost.
III Area ows upward at both the top and the bottom.
Only case III is possible, and given case III, there must be at least
one point P in the rst quadrant where area ows neither clockwise
nor counterclockwise.
6
The boost simply increases Ps distance from
the origin by some factor. By the linearity of the transformation,
the entire line running through O and P is simply rescaled. This
special lines inverse slope, which has units of velocity, apparently
has some special signicance, so we give it a name, c. Well see later
that c is the maximum speed of cause and eect whose existence
we inferred in section 1.3. Any world-line with a velocity equal to
c retains the same velocity as judged by moving observers, and by
isotropy the same must be true for c.
For convenience, lets adopt time and space units in which c = 1,
and let the original rectangle be a unit square. The upper right
tip of the parallelogram must slide along the line through the origin
with slope +1, and similarly the parallelograms other diagonal must
have a slope of 1. Since these diagonals bisected one another on
5
Proof: By linearity of L, the ow is clockwise at the negative axes as well.
Also by linearity, the handedness of the ow is the same at all points on a ray
extending out from the origin in the direction . If the ow were counterclockwise
somewhere, then it would have to switch handedness twice in that quadrant, at
1 and 2. But by writing out the vector cross product r dr, where dr is the
displacement caused by L(dv), we nd that it depends on sin(2+), which does
not oscillate rapidly enough to have two zeroes in the same quadrant.
6
This follows from the fact that, as shown in the preceding footnote, the
handedness of the ow depends only on .
c / Unit square PQRS is Lorentz-
boosted to the parallelogram
P
.
the original square, and since bisection is an ane property that
is preserved when we change frames of reference, the parallelogram
must be equilateral.
We can now determine the complete form of the Lorentz transfor-
mation. Let unit square PQRS, as described above, be transformed
to parallelogram P
in the new coordinate system (x
, t
). Let
the t
coordinate of R
be , interpreted as the ratio between the

time elapsed on a clock moving from P
to R
and the corresponding

time as measured by a clock that is at rest in the (x
, t
) frame. By
the denition of v, R
has coordinates (v, ), and the other geo-

metrical facts established above place Q
symmetrically on the other

side of the diagonal, at (, v). Computing the cross product of vec-
tors P
and P
, we nd the area of P
to be
2
(1 v
2
),
and setting this equal to 1 gives
=
1
1 v
2
.
Self-check: Interpret the dependence of on the sign of v.
d / The behavior of the factor.
The result for the transformation L, a Lorentz boost along the
x axis with velocity v, is:
t
= t +vx
x
= vt +x
The symmetry of P
with respect to reection across the

diagonal indicates that the time and space dimensions are treated
symmetrically, although they are not entirely interchangeable as
they would have been in case II. Although we dened in terms
of the time coordinate of R
, we could just as easily have used the

spatial coordinate of Q
, so represents a factor of both time dila-

tion and length contraction. (Clearly it wouldnt have made sense to
distort one quantity without distorting the other, since the invariant
velocity c represents a ratio of a distance to a time.) In summary, a
clock runs fastest according to an observer who is at rest relative to
the clock, and a measuring rod likewise appears longest in its own
rest frame.
The lack of a universal notion of simultaneity has a similarly
symmetric interpretation. In prerelativistic physics, points in space
have no xed identity. A brass plaque commemorating a Civil War
battle is not at the same location as the battle, according to an
observer who perceives the Earth has having been hurtling through
space for the intervening centuries. By symmetry, points in time
have no xed identity either.
In everyday life, we dont notice relativistic eects like time di-
lation, so apparently 1, and v 1, i.e., the speed c must be
very large when expressed in meters per second. By setting c equal
to 1, we have chosen a the distance unit that is extremely long in
proportion to the time unit. This is an example of the correspon-
dence principle, which states that when a new physical theory, such
as relativity, replaces an old one, such as Galilean relativity, it must
remain backward-compatible with all the experiments that ver-
ied the old theory; that is, it must agree with the old theory in
the appropriate limit. Despite my coyness, you probably know that
the speed of light is also equal to c. It is important to emphasize,
however, that light plays no special role in relativity, nor was it
necessary to assume the constancy of the speed of light in order to
derive the Lorentz transformation; we will in fact prove on page 53
that photons must travel at c, and on page 107 that this must be
true for any massless particle.
On the other hand, Einstein did originally develop relativity
based on a dierent set of assumptions than our L1-L5. His treat-
ment, given in his 1905 paper On the electrodynamics of moving
bodies, is reproduced on p. 240. It starts from the following two
postulates:
P1 The principle of relativity: . . . the phenomena of electrody-
namics as well as of mechanics possess no properties corre-
sponding to the idea of absolute rest.
P2 . . . light is always propagated in empty space with a denite
velocity c which is independent of the state of motion of the
e / Example 5. Flashes of
light travel along P
and Q
.
The observer in this frame of
reference judges them to have
been emitted at different times,
and to have traveled different
distances.
emitting body.
Einsteins P1 is essentially the same as our L3 (equivalence of iner-
tial frames). He implicitly assumes something equivalent to our L1
(homogeneity and isotropy of spacetime). In his system, our L5 (no
simultaneity) is a theorem proved from the axioms P1-P2, whereas
in our system, his P2 is a theorem proved from the axioms L1-L5.
Example: 5
Let the intersection of the parallelograms two diagonals be T in
the original (rest) frame, and T
in the Lorentz-boosted frame. An

observer at T in the original frame simultaneously detects the
passing by of the two ashes of light emitted at P and Q, and
since she is positioned at the midpoint of the diagram in space,
she infers that P and Q were simultaneous. Since the arrival of
both ashes of light at the same point in spacetime is a concrete
event, an observer in the Lorentz-boosted frame must agree on
their simultaneous arrival. (Simultaneity is well dened as long
as no spatial separation is involved.) But the distances traveled
by the two ashes in the boosted frame are unequal, and since
the speed of light is the same in all cases, the boosted observer
infers that they were not emitted simultaneously.
Example: 6
A different kind of symmetry is the symmetry between observers.
If observer A says observer Bs time is slow, shouldnt B say that
As time is fast? This is what would happen if B took a pill that
slowed down all his thought processes: to him, the rest of the
world would seem faster than normal. But this cant be correct
for Lorentz boosts, because it would introduce an asymmetry be-
tween observers. There is no preferred, correct frame corre-
sponding to the observer who didnt take a pill; either observer
can correctly consider himself to be the one who is at rest. It may
seem paradoxical that each observer could think that the other
was the slow one, but the paradox evaporates when we consider
the methods available to A and B for resolving the controversy.
They can either (1) send signals back and forth, or (2) get to-
gether and compare clocks in person. Signaling doesnt estab-
lish one observer as correct and one as incorrect, because as
well see in the following section, there is a limit to the speed of
propagation of signals; either observer ends up being able to ex-
plain the other observers observations by taking into account the
nite and changing time required for signals to propagate. Meet-
ing in person requires one or both observers to accelerate, as in
the original story of Alice and Betty, and then we are no longer
dealing with pure Lorentz frames, which are described by non-
accelerating observers.
GPS Example: 7
In the Hafele-Keating experiment using atomic clocks aboard air-
planes (p. 14), both gravity and motion had effects on the rate
of ow of time. Similarly, both effects must be considered in the
case of the GPS system. The gravitational effect was found on
page 28 to be E/E = gy (with c = 1), based on the equiva-
lence principle. The special-relativistic effect can be found from
the Lorentz transformation. Lets determine the directions and
relative strengths of the two effects in the case of a GPS satellite.
A radio photon emitted by a GPS satellite gains energy as it falls
to the earths surface, so its energy and frequency are increased
by this effect. The observer on the ground, after accounting for
all non-relativistic effects such as Doppler shifts and the Sagnac
effect (p. 58), would interpret the frequency shift by saying that
time aboard the satellite was owing more quickly than on the
ground.
However, the satellite is also moving at orbital speeds, so there is
a Lorentz time dilation effect. According to the observer on earth,
this causes time aboard the satellite to ow more slowly than on
the ground.
We can therefore see that the two effects are of opposite sign.
Which is stronger?
For a satellite in low earth orbit, we would have v
2
/r = g, where
r is only slightly greater than the radius of the earth. Expand-
ing the Lorentz gamma factor in a Taylor series, we nd that the
relative effect on the ow of time is 1 v
2
/2 = gr /2. The
gravitational effect, approximating g as a constant, is gy, where
y is the satellites altitude above the earth. For such a satellite,
the gravitational effect is down by a factor of 2y/r , so the Lorentz
time dilation dominates.
GPS satellites, however, are not in low earth orbit. They orbit at
an altitude of about 20,200 km, which is quite a bit greater than
the radius of the earth. We therefore expect the gravitational ef-
fect to dominate. To conrm this, we need to generalize the equa-
tion E/E = gy to the case where g is not a constant. Integrating
the equation dE/E = gdy, we nd that the time dilation factor is
equal to e
, where =
_
gdy is the gravitational potential per
unit mass. When is small, this causes a relative effect equal
to . The total effect for a GPS satellite is thus (inserting fac-
tors of c for calculation with SI units, and using positive signs for
blueshifts)
1
c
2
_
+
v
2
2
_
= 5.2 10
10
0.9 10
10
,
where the rst term is gravitational and the second kinematic. A
more detailed analysis includes various time-varying effects, but
this is the constant part. For this reason, the atomic clocks aboard
the satellites are set to a frequency of 10.22999999543 MHz be-
fore launching them into orbit; on the average, this is perceived
on the ground as 10.23 MHz. A more complete analysis of the
general relativity involved in the GPS system can be found in the
review article by Ashby.
7
Self-check: Suppose that positioning a clock at a certain dis-
tance from a certain planet produces a fractional change in the
rate at which time ows. In other words, the time dilation factor
is 1 + . Now suppose that a second, identical planet is brought
into the picture, at an equal distance from the clock. The clock is
positioned on the line joining the two planets centers, so that the
gravitational eld it experiences is zero. Is the fractional time dila-
tion now approximately 0, or approximately 2? Why is this only
an approximation?
f / Apparatus used for the test
of relativistic time dilation de-
scribed in example 8. The promi-
nent black and white blocks are
large magnets surrounding a cir-
cular pipe with a vacuum inside.
(c) 1974 by CERN.
Large time dilation Example: 8
The time dilation effect in the Hafele-Keating experiment was very
small. If we want to see a large time dilation effect, we cant do
it with something the size of the atomic clocks they used; the
kinetic energy would be greater than the total megatonnage of
all the worlds nuclear arsenals. We can, however, accelerate
subatomic particles to speeds at which is large. An early, low-
precision experiment of this kind was performed by Rossi and Hall
in 1941, using naturally occurring cosmic rays. Figure f shows a
1974 experiment
8
of a similar type which veried the time dilation
predicted by relativity to a precision of about one part per thou-
sand.
7
N. Ashby, Relativity in the Global Positioning System, http://www.
livingreviews.org/lrr-2003-1
8
Bailey at al., Nucl. Phys. B150(1979) 1
h / The change in the frequency
of x-ray photons emitted by
57
Fe
as a function of temperature,
drawn after Pound And Rebka
(1960). Dots are experimental
measurements. The solid curve
is Pound and Rebkas theoretical
calculation using the Debye the-
ory of the lattice vibrations with
a Debye temperature of 420 de-
grees C. The dashed line is one
with the slope calculated in the
text using a simplied treatment
of the thermodynamics. There is
an arbitrary vertical offset in the
experimental data, as well as the
theoretical curves.
g / Muons accelerated to nearly c
undergo radioactive decay much
more slowly than they would
according to an observer at rest
with respect to the muons. The
rst two data-points (unlled
circles) were subject to large
systematic errors.
Muons were produced by an accelerator at CERN, near Geneva.
A muon is essentially a heavier version of the electron. Muons un-
dergo radioactive decay, lasting an average of only 2.197 s be-
fore they evaporate into an electron and two neutrinos. The 1974
experiment was actually built in order to measure the magnetic
properties of muons, but it produced a high-precision test of time
dilation as a byproduct. Because muons have the same electric
charge as electrons, they can be trapped using magnetic elds.
Muons were injected into the ring shown in gure f, circling around
it until they underwent radioactive decay. At the speed at which
these muons were traveling, they had = 29.33, so on the av-
erage they lasted 29.33 times longer than the normal lifetime. In
other words, they were like tiny alarm clocks that self-destructed
at a randomly selected time. Figure g shows the number of ra-
dioactive decays counted, as a function of the time elapsed af-
ter a given stream of muons was injected into the storage ring.
The two dashed lines show the rates of decay predicted with and
without relativity. The relativistic line is the one that agrees with
experiment.
Time dilation in the Pound-Rebka experiment Example: 9
In the description of the Pound-Rebka experiment on page 29, I
postponed the quantitative estimation of the frequency shift due
to temperature. Classically, one expects only a broadening of
the line, since the Doppler shift is proportional to v
/c, where
v
, the component of the emitting atoms velocity along the line

of sight, averages to zero. But relativity tells us to expect that if
the emitting atom is moving, its time will ow more slowly, so the
frequency of the light it emits will also be systematically shifted
downward. This frequency shift should increase with tempera-
ture. In other words, the Pound-Rebka experiment was designed
as a test of general relativity (the equivalence principle), but this
special-relativistic effect is just as strong as the relativistic one,
and needed to be accounted for carefully.
In Pound and Rebkas paper describing their experiment,
9
they
refer to a preliminary measurement
10
in which they carefully mea-
sured this effect, showed that it was consistent with theory, and
pointed out that a previous claim by Cranshaw et al. of having
measured the gravitational frequency shift was vitiated by their
failure to control for the temperature dependence.
It turns out that the full Debye treatment of the lattice vibrations is
not really necessary near room temperature, so well simplify the
thermodynamics. At absolute temperature T, the mean transla-
tional kinetic energy of each iron nucleus is (3/2)kT. The velocity
is much less than c(= 1), so we can use the nonrelativistic expres-
sion for kinetic energy, K = (1/2)mv
2
, which gives a mean value
9
Phys. Rev. Lett. 4 (1960) 337
10
Phys. Rev. Lett. 4 (1960) 274
i / The light cone in 2+1 di-
mensions.
j / The circle plays a privi-
leged role in Euclidean geometry.
When rotated, it stays the same.
The pie slice is not invariant as
the circle is. A similar privileged
place is occupied by the light
cone in Lorentz geometry. Under
a Lorentz boost, the spacetime
parallelograms change, but the
light cone doesnt.
for v
2
of 3kT/m. In the limit of v 1, time dilation produces a
change in frequency by a factor of 1/, which differs from unity
by approximately v
2
/2. The relative time dilation is therefore
3kT/2m, or, in metric units, 3kT/2mc
2
. The vertical scale
in gure h contains an arbitrary offset, since Pound and Rebkas
measurements were the best absolute measurements to date of
the frequency. The predicted slope of 3k/2mc
2
, however, is
not arbitrary. Plugging in 57 atomic mass units for m, we nd
the slope to be 2.4 10
15
, which, as shown in the gure is an
excellent approximation (off by only 10%) near room temperature.
2.3 The light cone
Given an event P, we can now classify all the causal relationships in
which P can participate. In Newtonian physics, these relationships
fell into two classes: P could potentially cause any event that lay in
its future, and could have been caused by any event in its past. In a
Lorentz spacetime, we have a trichotomy rather than a dichotomy.
There is a third class of events that are too far away from P in space,
and too close in time, to allow any cause and eect relationship, since
causalitys maximum velocity is c. Since were working in units in
which c = 1, the boundary of this set is formed by the lines with
slope 1 on a (t, x) plot. This is referred to as the light cone, and in
the generalization from 1+1 to 3+1 dimensions, it literally becomes
a (four-dimensional) cone. The terminology comes from the fact
that light happens to travel at c, the maximum speed of cause and
eect. If we make a cut through the cone dened by a surface of con-
stant time in Ps future, the resulting section is a sphere (analogous
to the circle formed by cutting a three-dimensional cone), and this
sphere is interpreted as the set of events on which P could have had
a causal eect by radiating a light pulse outward in all directions.
Events lying inside one anothers light cones are said to have
a timelike relationship. Events outside each others light cones are
spacelike in relation to one another, and in the case where they lie
on the surfaces of each others light cones the term is lightlike.
The light cone plays the same role in the Lorentz geometry that
the circle plays in Euclidean geometry. The truth or falsehood of
propositions in Euclidean geometry remains the same regardless of
how we rotate the gures, and this is expressed by Euclids E3 as-
serting the existence of circles, which remain invariant under rota-
tion. Similarly, Lorentz boosts preserve light cones and truth of
propositions in a Lorentz frame.
Self-check: Under what circumstances is the time-ordering of
events P and Q preserved under a Lorentz boost?
In a uniform Lorentz spacetime, all the light cones line up like
soldiers with their axes parallel with one another. When gravity is
Section 2.3 The light cone 49
a / Light cones tip over for
two reasons in general relativity:
because of the presence of
masses, which have gravita-
tional elds, and because of
the cosmological constant. The
time and distance scales in the
bottom gure are many orders of
magnitude greater than those in
the top.
b / Example 10. Matter is
lifted out of a Newtonian black
hole with a bucket. The dashed
line represents the point at which
the escape velocity equals the
speed of light.
present, however, this uniformity is disturbed in the vicinity of the
masses that constitute the sources. The light cones lying near the
sources tip toward the sources. Superimposed on top of this gravi-
tational tipping together, recent observations have demonstrated a
systematic tipping-apart eect which becomes signicant on cosmo-
logical distance scales. The parameter that sets the strength of
this eect is known as the cosmological constant. The cosmologi-
cal constant is not related to the presence of any sources (such as
negative masses), and can be interpreted instead as a tendency for
space to expand over time on its own initiative. In the present era,
the cosmological constant has overpowered the gravitation of the
universes mass, causing the expansion of the universe to accelerate.
Self-check: In the bottom panel of gure a, can an observer look
at the properties of the spacetime in her immediate vicinity and
tell how much her light cones are tipping, and in which direction?
Compare with gure g on page 25.
A Newtonian black hole Example: 10
In the case of a black hole, the light cone tips over so far that
the entire future timelike region lies within the black hole. If an
observer is present at such an event, then that observers en-
tire potential future lies within the black hole, not outside it. By
expanding on the logical consequences of this statement, we ar-
rive at an example of relativitys proper interpretation as a theory
of causality, not a theory of objects exerting forces on one an-
other as in Newtons vision of action at a distance, or Lorentzs
original ether-drag interpretation of the factor , in which length
contraction arose from a physical strain imposed on the atoms
composing a physical body.
Imagine a black hole from a Newtonian point of view, as proposed
in 1783 by geologist John Michell. Setting the escape velocity
equal to the speed of light, we nd that this will occur for any grav-
itating spherical body compact enough to have M/r > c
2
/2G.
(A fully relativistic argument, as given in section 6.2, agrees on
M/r c
2
/G, which is xed by units. The correct unitless factor
depends on the denition of r , which is exible in general relativ-
ity.) A ash of light emitted from the surface of such a Newtonian
black hole would fall back down like water from a fountain, but
it would nevertheless be possible for physical objects to escape,
e.g., if they were lifted out in a bucket dangling from a cable. If the
cable is to support its own weight, it must have a tensile strength
per unit density of at least c
2
/2, which is about ten orders of mag-
nitude greater than that of carbon nanotube bers. (The factor of
1/2 is not to be taken seriously, since it comes from a nonrela-
tivistic calculation.)
The cause-and-effect interpretation of relativity tells us that this
Newtonian picture is incorrect. A physical object that approaches
c / The rapidity, = tanh
1
v, as
a function of v.
to within a distance r of a concentration of mass M, with M/r
sufciently large, has no causal future lying at larger values of r .
The conclusion is that there is a limit on the tensile strength of
any substance, imposed purely by general relativity, and we can
state this limit without having to know anything about the physical
nature of the interatomic forces. Cf. homework problem 3 and
section 3.4.4.
2.3.1 Velocity addition
In classical physics, velocities add in relative motion. For exam-
ple, if a boat moves relative to a river, and the river moves relative
to the land, then the boats velocity relative to the land is found
by vector addition. This linear behavior cannot hold relativistically.
For example, if a spaceship is moving at 0.60c relative to the earth,
and it launches a probe at 0.60c relative to itself, we cant have the
probe moving at 1.20c relative to the earth, because this would be
greater than the maximum speed of cause and eect, c. To see how
to add velocities relativistically, we start be rewriting the Lorentz
transformation as the matrix
_
cosh sinh
sinh cosh
_
,
where = tanh
1
v is called the rapidity. We are guaranteed that
the matrix can be written in this form, because its area-preserving
property says that the determinant equals 1, and cosh
2
sinh
2
=
1 is an identity of the hyperbolic trig functions. It is now straight-
forward to verify that multiplication of two matrices of this form
gives a third matrix that is also of this form, with =
1
+
2
. In
other words, rapidities add linearly; velocities dont. In the example
of the spaceship and the probe, the rapidities add as tanh
1
.60 +
tanh
1
.60 = .693 + .693 = 1.386, giving the probe a velocity of
tanh 1.386 = 0.88 relative to the earth. Any number of velocities
can be added in this way,
1
+
2
+. . . +
n
.
Self-check: Interpret the asymptotes of the graph in gure c.
Bells spaceship paradox Example: 11
A difcult philosophical question is whether the time dilation and
length contractions predicted by relativity are real. This de-
pends, of course, on what one means by real. They are frame-
dependent, i.e., observers in different frames of reference dis-
agree about them. But this doesnt tell us much about their reality,
since velocities are frame-dependent in Newtonian mechanics,
but nobody worries about whether velocities are real. John Bell
(1928-1990) proposed the following thought experiment to physi-
cists in the CERN cafeteria, and found that nearly all of them got
it wrong. He took this as evidence that their intuitions had been
misguided by the standard way of approaching this question of
the reality of Lorentz contractions.
Section 2.3 The light cone 51
d / Example 11.
Let spaceships A and B accelerate as shown in gure d along a
straight line. Observer C does not accelerate. The accelerations,
as judged by C, are constant for both ships. Each ship is equipped
with a yard-arm, and a thread is tied between the two arms. Does
the thread break, due to Lorentz contraction? (We assume that
the acceleration is gentle enough that the thread does not break
simply because of its own inertia.)
The popular answer in the CERN cafeteria was that the thread
would not break, the reasoning being that Lorentz contraction is
a frame-dependent effect, and no such contraction would be ob-
served in A and Bs frame. The ships maintain a constant dis-
tance from one another, so C merely disagrees with A and B
about the length of the thread, as well as other lengths like the
lengths of the spaceships.
The error in this reasoning is that the accelerations of A and B
were specied to be equal and constant in Cs frame, not in A and
Bs. Bells interpretation is that the frame-dependence is a distrac-
tion, that Lorentz contraction is in some sense a real effect, and
that it is therefore immediately clear that the thread must break,
without even having to bother going into any other frame. To con-
vince his peers in the cafeteria, however, Bell presumably needed
to satisfy them as to the specic errors in their reasoning, and this
requires that we consider the frame-dependence explicitly.
We can rst see that it is impossible, in general, for different ob-
servers to agree about what is meant by constant acceleration.
Suppose that A and B agree with C about the constancy of their
acceleration. Then A and B experience a voyage in which the
rapidities of the stars around them (and of observer C) increase
linearly with time. As the rapidity approaches innity, both C and
the stars approach the speed of light. But since A and C agree on
the magnitude of their velocity relative to one another, this means
that As velocity as measured by C must approach c, and this
contradicts the premise that C observes constant acceleration for
both ships. Therefore A and B do not consider their own acceler-
ations to be constant.
A and B do not agree with C about simultaneity, and since they
also do not agree that their accelerations are constant, they do
not consider their own accelerations to be equal at a given mo-
ment of time. Therefore the string changes its length, and this
is consistent with Bells original, simple answer, which did not re-
quire comparing different frames of reference. To establish that
the string comes under tension, rather than going slack, we can
apply the equivalence principle. By the equivalence principle, any
experiments done by A and B give the same results as if they
were immersed in a gravitational eld. The leading ship B sees A
as experiencing a gravitational time dilation. According to B, the
slowpoke A isnt accelerating as rapidly as it should, causing the
string to break.
These ideas are closely related to the fact that general relativity
does not admit any spacetime that can be interpreted as a uni-
form gravitational eld (see problem 5, p. 162).
2.3.2 Logic
The trichotomous classication of causal relationships has in-
teresting logical implications. In classical Aristotelian logic, every
proposition is either true or false, but not both, and given proposi-
tions p and q, we can form propositions such as p q (both p and
q) or p q (either p or q). Propositions about physical phenomena
can only be veried by observation. Let p be the statement that
a certain observation carried out at event P gives a certain result,
and similarly for q at Q. If PQ is spacelike, then the truth or false-
hood of p q cannot be checked by physically traveling to P and
Q, because no observer would be able to attend both events. The
truth-value of pq is unknown to any observer in the universe until
a certain time, at which the relevant information has been able to
propagate back and forth. What if P and Q lie inside two dierent
black holes? Then the truth-value of p q can never be determined
by any observer. Another example is the case in which P and Q
are separated by such a great distance that, due to the accelerating
expansion of the universe, their future light cones do not overlap.
We conclude that Aristotelian logic cannot be appropriately ap-
plied to relativistic observation in this way. Some workers attempt-
ing to construct a quantum-mechanical theory of gravity have sug-
gested an even more radically observer-dependent logic, in which
dierent observers may contradict one another on the truth-value of
a single proposition p
1
, unless they agree in advance on the list p
2
,
p
3
, . . . of all the other propositions that they intend to test as well.
Well return to these questions on page 189.
2.4 Experimental tests of Lorentz geometry
Weve already seen, in section 1.2, a variety of evidence for the non-
classical behavior of spacetime. Were now in a position to discuss
tests of relativity more quantitatively. An up-to-date review of such
tests is given by Mattingly.
11
One such test is that relativity requires the speed of light to
be the same in all frames of reference, for the following reasons.
Compare with the speed of sound in air. The speed of sound is not
the same in all frames of reference, because the wave propagates
at a xed speed relative to the air. An observer at who is moving
relative to the air will measure a dierent speed of sound. Light, on
11
livingreviews.org/lrr-2005-5
Section 2.4 Experimental tests of Lorentz geometry 53
a / An artists conception of a
gamma-ray burst, resulting from
a supernova explosion.
the other hand, isnt a vibration of any physical medium. Maxwells
equations predict a denite value for the speed of light, regardless
of the motion of the source. This speed also cant be relative to
any medium. If the speed of light isnt xed relative to the source,
and isnt xed relative to a medium, then it must be xed relative
to anything at all. The only speed in relativity that is equal in all
frames of reference is c, so light must propagate at c. We will see
on page 107 that there is a deeper reason for this; relativity requires
that any massless particle propagate at c. The requirement of v = c
for massless particles is so intimately hard-wired into the structure
of relativity that any violation of it, no matter how tiny, would be of
great interest. Essentially, such a violation would disprove Lorentz
invariance, i.e., the invariance of the laws of physics under Lorentz
transformations. There are two types of tests we could do: (1)
test whether photons of all energies travel at the same speed, i.e.,
whether the vacuum is dispersive; (2) test whether observers in all
frames of reference measure the same speed of light.
2.4.1 Dispersion of the vacuum
Some candidate quantum-mechanical theories of gravity, such as
loop quantum gravity, predict a granular structure for spacetime at
the Planck scale,
_
G/c
3
= 10
35
m, which would naturally lead to
deviations from v = 1 that would become more and more signicant
for photons with wavelengths getting closer and closer to that scale.
Lorentz-invariance would then be an approximation valid only at
large scales.
Presently the best experimental tests of the invariance of the
speed of light with respect to wavelength come from astronomical
observations of gamma-ray bursts, which are sudden outpourings of
high-energy photons, believed to originate from a supernova explo-
sion in another galaxy. One such observation, in 2009,
12
collected
photons from such a burst, with a duration of 2 seconds, indicating
that the propagation time of all the photons diered by no more
than 2 seconds out of a total time in ight on the order of ten bil-
lion years, or about one part in 10
17
! A single superlative photon in
the burst had an energy of 31 GeV, and its arrival within the same
2-second time window demonstrates Lorentz invariance over a vast
range of photon energies, ruling out some versions of loop quantum
gravity.
2.4.2 Observer-independence of c
The constancy of the speed of light for observers in all frames of
reference was originally detected in 1887 when Michelson and Morley
set up a clever apparatus to measure any dierence in the speed of
light beams traveling east-west and north-south. The motion of
the earth around the sun at 110,000 km/hour (about 0.01% of the
12
http://arxiv.org/abs/0908.1832
speed of light) is to our west during the day. Michelson and Morley
believed that light was a vibration of a physical medium, the ether,
so they expected that the speed of light would be a xed value
relative to the ether. As the earth moved through the ether, they
thought they would observe an eect on the velocity of light along
an east-west line. For instance, if they released a beam of light in
a westward direction during the day, they expected that it would
move away from them at less than the normal speed because the
earth was chasing it through the ether. They were surprised when
they found that the expected 0.01% change in the speed of light did
not occur.
b / The Michelson-Morley experi-
ment, shown in photographs, and
drawings from the original 1887
paper. 1. A simplied draw-
ing of the apparatus. A beam of
light from the source, s, is par-
tially reected and partially trans-
mitted by the half-silvered mirror
h
1
. The two half-intensity parts of
the beam are reected by the mir-
rors at a and b, reunited, and ob-
served in the telescope, t. If the
earths surface was supposed to
be moving through the ether, then
the times taken by the two light
waves to pass through the mov-
ing ether would be unequal, and
the resulting time lag would be
detectable by observing the inter-
ference between the waves when
they were reunited. 2. In the real
apparatus, the light beams were
reected multiple times. The ef-
fective length of each arm was
increased to 11 meters, which
greatly improved its sensitivity to
the small expected difference in
the speed of light. 3. In an
earlier version of the experiment,
they had run into problems with
its extreme sensitiveness to vi-
bration, which was so great that
it was impossible to see the in-
terference fringes except at brief
intervals . . . even at two oclock
in the morning. They therefore
mounted the whole thing on a
massive stone oating in a pool of
mercury, which also made it pos-
sible to rotate it easily. 4. A photo
of the apparatus. Note that it is
underground, in a room with solid
brick walls.
Although the Michelson-Morley experiment was nearly two dec-
ades in the past by the time Einstein published his rst paper on
relativity in 1905, and Einstein did know about it,
13
its unclear how
much it inuenced him. Michelson and Morley themselves were un-
certain about whether the result was to be trusted, or whether sys-
tematic and random errors were masking a real eect from the ether.
There were a variety of competing theories, each of which could
claim some support from the shaky data. Some physicists believed
that the ether could be dragged along by matter moving through it,
which inspired variations on the experiment that were conducted on
mountaintops in thin-walled buildings, (gure), or with one arm of
the apparatus out in the open, and the other surrounded by massive
13
J. van Dongen, http://arxiv.org/abs/0908.1545
Section 2.4 Experimental tests of Lorentz geometry 55
lead walls. In the standard sanitized textbook version of the history
of science, every scientist does his experiments without any pre-
conceived notions about the truth, and any disagreement is quickly
settled by a denitive experiment. In reality, this period of confu-
sion about the Michelson-Morley experiment lasted for four decades,
and a few reputable skeptics, including Miller, continued to believe
that Einstein was wrong, and kept trying dierent variations of the
experiment as late as the 1920s. Most of the remaining doubters
were convinced by an extremely precise version of the experiment
performed by Joos in 1930, although you can still nd kooks on the
internet who insist that Miller was right, and that there was a vast
conspiracy to cover up his results.
c / Dayton Miller thought that the result of the Michelson-Morley ex-
periment could be explained because the ether had been pulled along by
the dirt, and the walls of the laboratory. This motivated him to carry out a
series of experiments at the top of Mount Wilson, in a building with thin
walls.
Before Einstein, some physicists who did believe the negative
result of the Michelson-Morley experiment came up with explana-
tions that preserved the ether. In the period from 1889 to 1895, both
Lorentz and George FitzGerald suggested that the negative result
of the Michelson-Morley experiment could be explained if the earth,
and every physical object on its surface, was contracted slightly by
the strain of the earths motion through the ether. Thus although
Lorentz developed all the mathematics of Lorentz frames, and got
them named after himself, he got the interpretation wrong.
2.4.3 Lorentz violation by gravitational forces
The tests described in sections 2.4.1 and 2.4.2 both involve the
behavior of light, i.e., they test whether or not electromagnetism
really has the exact Lorentz-invariant behavior contained implicitly
in Maxwells equations. In the jargon of the eld, they test Lorentz
invariance in the photon sector. Since relativity is a theory of
gravity, it is natural to ask whether the Lorentz invariance holds
for gravitational forces as well as electromagnetic ones. If Lorentz
invariance is violated by gravity, then the strength of gravitational
forces might depend on the observers motion through space, rela-
tive to some xed reference frame analogous to that of the ether.
e / The matter interferometer
used by Chung et al. Each atoms
wavefunction is split into two
parts, which travel along two
different paths (solid and dashed
lines).
d / The results of the measure-
ment of g by Chung et al., sec-
tion 2.4.3. The experiment was
done on the Stanford University
campus, surrounded by the Pa-
cic ocean and San Francisco
Bay, so it was subject to vary-
ing gravitational from both astro-
nomical bodies and the rising and
falling ocean tides. Once both of
these effects are subtracted out
of the data, there is no Lorentz-
violating variation in g due to
the earths motion through space.
Note that the data are broken up
into three periods, with gaps of
three months and four years sep-
arating them. (c) APS, used un-
der the U.S. fair use exception to
copyright.
Historically, gravitational Lorentz violations have been much more
dicult to test, since gravitational forces are so weak, and the rst
high-precision data were obtained by Nordtvedt and Will in 1957,
70 years after Michelson and Morley. Nordtvedt and Will measured
the strength of the earths gravitational eld as a function of time,
and found that it did not vary on a 24-hour cycle with the earths
rotation, once tidal eects had been accounted for. Further con-
straints come from data on the moons orbit obtained by reecting
laser beams from a mirror left behind by the Apollo astronauts.
A recent high-precision laboratory experiment was done in 2009
by Chung et al.
14
They constructed an interferometer in a verti-
cal plane that is conceptually similar to a Michelson interferometer,
except that it uses cesium atoms rather than photons. That is,
the light waves of the Michelson-Morley experiment are replaced by
quantum-mechanical matter waves. The roles of the half-silvered
and fully silvered mirrors are lled by lasers, which kick the atoms
electromagnetically. Each atoms wavefunction is split into two
parts, which travel by two dierent paths through spacetime, even-
tually reuniting and interfering. The result is a measurement of g
to about one part per billion. The results, shown in gure d, put a
strict limit on violations of Lorentz geometry by gravity.
2.5 Three spatial dimensions
New and nontrivial phenomena arise when we generalize from 1+1
dimensions to 3+1.
14
arxiv.org/abs/0905.1929
Section 2.5 Three spatial dimensions 57
b / Inertial devices for main-
taining a direction in space: 1.
A ring laser. 2. The photon in
a perfectly reective spherical
cavity. 3. A gyroscope.
a / A boost along x followed
by a boost along y results in
tangling up of the x and y coordi-
nates, so the result is not just a
boost but a boost plus a rotation.
2.5.1 Lorentz boosts in three dimensions
How does a Lorentz boost along one axis, say x, aect the other
two spatial coordinates y and z? We have already proved that area
in the (t, x) plane is preserved. The same proof applies to volume
in the spaces (t, x, y) and (t, x, z), hence lengths in the y and z
directions are preserved. (The proof does not apply to volume in,
e.g., (x, y, z) space, because the x transformation depends on t, and
therefore if we are given a region in (x, y, z), we do not have enough
information to say how it will change under a Lorentz boost.) The
complete form of the transformation L(v x), a Lorentz boost along
the x axis with velocity v, is therefore:
t
= t +vx
x
= vt +x
y
= y
z
= z
Based on the trivial nature of this generalization, it might seem
as though no qualitatively new considerations would arise in 3+1
dimensions as compared with 1+1. To see that this is not the case,
consider gure a. A boost along the x axis tangles up the x and
t coordinates. A y-boost mingles y and t. Therefore consecutive
boosts along x and y can cause x and y to mix. The result, as
well see in more detail below, is that two consecutive boosts along
non-collinear axes are not equivalent to a single boost; they are
equivalent to a boost plus a spatial rotation. The remainder of this
section discusses this eect, known as Thomas precession, in more
detail; it can be omitted on a rst reading.
Self-check: Apply similar reasoning to a Galilean boost.
2.5.2 Gyroscopes and the equivalence principle
To see how this mathematical fact would play out as a physical
eect, we need to consider how to make a physical manifestation of
the concept of a direction in space.
In two space dimensions, we can construct a ring laser, b/1,
which in its simplest incarnation is a closed loop of optical ber
with a bidirectional laser inserted in one place. Coherent light tra-
verses the loop simultaneously in both directions, interfering in a
beat pattern, which can be observed by sampling the light at some
point along the loops circumference. If the loop is rotated in its
own plane, the interference pattern is altered, because the beam-
sampling device is in a dierent place, and the path lengths traveled
by the two beams has been altered. This phase shift is called the
Sagnac eect, after M. Georges Sagnac, who observed the eect in
1913 and interpreted it, incorrectly, as evidence for the existence of
the aether.
15
The loop senses its own angular acceleration relative
15
Comptes rendus de lAcademie des science 157 (1913) 708
to an inertial reference frame. If we transport the loop while always
carefully adjusting its orientation so as to prevent phase shifts, then
its orientation has been preserved. The atomic clocks used in the
Hafele-Keating atomic-clock experiment described on page 14 were
sensitive to Sagnac eects, and it was not practical to maintain their
orientations while they were strapped into seats on a passenger jet,
so this orientational eect had to be subtracted out of the data at
the end of the experiment.
In three spatial dimensions, we could build a spherical cavity
with a reective inner surface, and release a photon inside, b/2.
In reality, the photon-in-a-cavity is not very practical. The pho-
ton would eventually be absorbed or scattered, and it would also be
dicult to accurately initialize the device and read it out later. A
more practical tool is a gyroscope. For example, one of the classic
tests of general relativity is the 2007 Gravity Probe B experiment
(discussed in detail on pages 139 and 175), in which four gyro-
scopes aboard a satellite were observed to precess due to special-
and general-relativistic eects.
The gyroscope, however, is not so obviously a literal implemen-
tation of our basic concept of a direction. How, then, can we be sure
that its behavior is equivalent to that of the photon-in-a-cavity? We
could, for example, carry out a complete mathematical development
of the angular momentum vector in relativity.
16
The equivalence
principle, however, allows us to bypass such technical details. Sup-
pose that we seal the two devices inside black boxes, with identical
external control panels for initializing them and reading them out.
We initialize them identically, and then transport them along side-
by-side world-lines. Classically, both the mechanical gyroscope and
the photon-gyroscope would maintain absolute, xed directions in
space. Relativistically, they will not necessarily maintain their ori-
entations. For example, weve already seen in section 2.5.1 that
there are reasons to expect that their orientations will change if
they are subjected to accelerations that are not all along the same
line. Because relativity is a geometrical theory of spacetime, this
dierence between the classical and relativistic behavior must be
determinable from purely geometrical considerations, such as the
shape of the world-line. If it depended on something else, then we
could conceivably see a disagreement in the outputs of the two in-
struments, but this would violate the equivalence principle.
Suppose there were such a discrepancy. That discrepancy would
be a physically measurable property of the spacetime region through
which the two gyroscopes had been transported. The eect would
have a certain magnitude and direction, so by collecting enough data
we could map it out as vector eld covering that region of spacetime.
16
This is done, for example, in Misner, Thorne, and Wheeler, Gravitation, pp.
157-159.
c / Classically, the gyroscope
should not rotate as long as the
forces from the hammer are all
transmitted to it at its center of
mass.
This eld evidently causes material particles to accelerate, since it
has an eect on the mechanical gyroscope. Roughly speaking (the
reasoning will be lled in more rigorously on page 117), the fact
that this eld acts dierently on the two gyroscopes is like getting a
non-null result from an Eotvos experiment, and it therefore violates
the equivalence principle. We conclude that gyroscopes b/2 and
b/3 are equivalent. In other words, there can only be one uniquely
dened notion of direction, and the details of how it is implemented
are irrelevant.
2.5.3 Boosts causing rotations
As a quantitative example, consider the following thought ex-
periment. Put a gyroscope in a box, and send the box around the
square path shown in gure c at constant speed. The gyroscope de-
nes a local coordinate system, which according to classical physics
would maintain its orientation. At each corner of the square, the
box has its velocity vector changed abruptly, as represented by the
hammer. We assume that the hits with the hammer are transmitted
to the gyroscope at its center of mass, so that they do not result in
any torque. Classically, if the set of gyroscopes travels once around
the square, it should end up at the same place and in the same ori-
entation, so that the coordinate system it denes is identical with
the original one.
For notation, let L(v x) indicate the boost along the x axis de-
scribed by the transformation on page 58. This is a transformation
that changes to a frame of reference moving in the negative x direc-
tion compared to the original frame. A particle considered to be at
rest in the original frame is described in the new frame as moving
in the positive x direction. Applying such an L to a vector p, we
calculate Lp, which gives the coordinates of the event as measured
in the new frame. An expression like MLp is equivalent by asso-
ciativity to M(Lp), i.e., ML represents applying L rst, and then
M.
In this notation, the hammer strikes can be represented by a
series of four Lorentz boosts,
T = L(v x) L(v y) L(v x) L(v y) ,
where we assume that the square has negligible size, so that all four
Lorentz boosts act in a way that preserves the origin of the coordi-
nate systems. (We have no convenient way in our notation L(. . .) to
describe a transformation that does not preserve the origin.) The
rst transformation, L(v y), changes coordinates measured by the
original gyroscope-dened frame to new coordinates measured by
the new gyroscope-dened frame, after the box has been acceler-
ated in the positive y direction.
The calculation of T is messy, and to be honest, I made a series
of mistakes when I tried to crank it out by hand. Calculations in
d / A page from one of Einsteins
notebooks.
relativity have a reputation for being like this. Figure d shows a page
from one of Einsteins notebooks, written in fountain pen around
1913. At the bottom of the page, he wrote zu umstaendlich,
meaning too involved. Luckily we live in an era in which this sort
of thing can be handled by computers. Starting at this point in the
book, I will take appropriate opportunities to demonstrate how to
use the free and open-source computer algebra system Maxima to
keep complicated calculations manageable. The following Maxima
program calculates a particular element of the matrix T.
1 /* For convenience, define gamma in terms of v: */
2 gamma:1/sqrt(1-v*v);
3 /* Define Lx as L(x-hat), Lmx as L(-x-hat), etc.: */
4 Lx:matrix([gamma, gamma*v, 0],
5 [gamma*v, gamma, 0],
6 [0, 0, 1]);
7 Ly:matrix([gamma, 0, gamma*v],
8 [0, 1, 0],
9 [gamma*v, 0, gamma]);
10 Lmx:matrix([gamma, -gamma*v, 0],
11 [-gamma*v, gamma, 0],
12 [0, 0, 1]);
13 Lmy:matrix([gamma, 0, -gamma*v],
14 [0, 1, 0],
15 [-gamma*v, 0, gamma]);
16 /* Calculate the product of the four matrices: */
17 T:Lx.Ly.Lmx.Lmy;
18 /* Define a column vector along the x axis: */
19 P:matrix([0],[1],[0]);
20 /* Find the result of T acting on this vector,
21 expressed as a Taylor series to second order in v: */
22 taylor(T.P,v,0,2);
Statements are terminated by semicolons, and comments are writ-
ten like /* ... */ On line 2, we see a symbolic denition of the
symbol gamma in terms of the symbol v. The colon means is de-
ned as. Line 2 does not mean, as it would in most programming
languages, to take a stored numerical value of v and use it to cal-
culate a numerical value of . In fact, v does not have a numerical
value dened at this point, nor will it ever have a numerical value
dened for it throughout this program. Line 2 simply means that
whenever Maxima encounters the symbol gamma, it should take it as
an abbreviation for the symbol 1/sqrt(1-v*v). Lines 5-16 dene
some 33 matrices that represent the L transformations. The basis
is

t, x, y. Line 18 calculates the product of the four matrices; the
dots represent matrix multiplication. Line 23 denes a vector along
the x axis, expressed as a column matrix (three rows of one column
each) so that Maxima will know how to operate on it using matrix
multiplication by T.
Finally line 26 outputs
17
the result of T acting on P:
19 [ 0 + . . . ]
20 [ ]
21 (%o9)/T/ [ 1 + . . . ]
22 [ ]
23 [ 2 ]
24 [ - v + . . . ]
In other words,
T
_
_
0
1
0
_
_
=
_
_
0
1
v
2
_
_
+. . . ,
where . . . represents higher-order terms in v. Suppose that we use
the initial frame of reference, before T is applied, to determine that
a particular reference point, such as a distant star, is along the x
axis. Applying T, we get a new vector TP, which we nd has a non-
vanishing y component approximately equal to v
2
. This result is
17
Ive omitted some output generated automatically from the earlier steps in
the computation. The (%o9) indicates that this is Maximas output from the
ninth and nal step.
entirely unexpected classically. It tells us that the gyroscope, rather
than maintaining its original orientation as it would have done clas-
sically, has rotated slightly. It has precessed in the counterclockwise
direction in the xy plane, so that the direction to the star, as mea-
sured in the coordinate system dened by the gyroscope, appears
to have rotated clockwise. As the box moved clockwise around the
square, the gyroscope has apparently rotated by a counterclockwise
angle v
2
about the z axis. We can see that this is a purely
relativistic eect, since for v 1 the eect is small. For historical
reasons discussed in section 2.5.4, this phenomenon is referred to as
the Thomas precession.
The particular features of this square geometry are not necessary.
I chose them so that (1) the boosts would be along the Cartesian
axes, so that we would be able to write them down easily; (2) it is
clear that the eect doesnt arise from any asymmetric treatment
of the spatial axes; and (3) the change in the orientation of the
gyroscope can be measured at the same point in space, e.g., by
comparing it with a twin gyroscope that stays at home. In general:
A gyroscope transported around a closed loop in at space-
time changes its orientation compared with one that is not
accelerated.
This is a purely relativistic eect, since a classical gyroscope
does not change its axis of rotation unless subjected to a
torque; if the boosts are accomplished by forces that act at
the gyroscopes center of mass, then there is no classical ex-
planation for the eect.
The eect can occur in the absence of any gravitational elds.
That is, this is a phenomenon of special relativity.
The composition of two or more Lorentz boosts along dierent
axes is not equivalent to a single boost; it is equivalent to a
boost plus a spatial rotation.
Lorentz boosts do not commute, i.e., it makes a dierence what
order we perform them in. Even if there is almost no time lag
between the rst boost and the second, the order of the boosts
matters. If we had applied the boosts in the opposite order,
the handedness of the eect would have been reversed.
Self-check: If Lorentz boosts did commute, what would be the
consequences for the expression L(v x) L(v y) L(v x) L(v y)?
The velocity disk
Figure e shows a useful way of visualizing the combined eects
of boosts and rotations in 2+1 dimensions. The disk depicts all
possible states of motion relative to some arbitrarily chosen frame
g / A round-trip involving ul-
trarelativistic velocities. All three
legs are at constant acceleration.
e / The velocity disk.
f / Two excursions in a rocket-
ship: one along the y axis and
one along x.
of reference. Lack of motion is represented by the point at the
center. A point at distance v from the center represents motion at
velocity v in a particular direction in the x y plane. By drawing
little axes at a particular point, we can represent a particular frame
of reference: the frame is in motion at some velocity, with its own
x and y axes are oriented in a particular way.
It turns out to be easier to understand the qualitative behavior
of our mysterious rotations if we switch from the low-velocity limit
to the contrary limit of ultrarelativistic velocities. Suppose we have
a rocket-ship with an inertial navigation system consisting of two
gyroscopes at right angles to one another. We rst accelerate the
ship in the y direction, and the acceleration is steady in the sense
that it feels constant to observers aboard the ship. Since it is rapidi-
ties, not velocities, that add linearly, this means that as an observer
aboard the ship reads clock times
1
,
2
, . . . , all separated by equal
intervals , the ships rapidity changes at a constant rate,
1
,
2
,
. . . . This results in a series of frames of reference that appear closer
and closer together on the diagram as the ship approaches the speed
of light, at the edge of the disk. We can start over from the center
again and repeat the whole process along the x axis, resulting in
a similar succession of frames. In both cases, the boosts are being
applied along a single line, so that there is no rotation of the x and
y axes.
Now suppose that the ship were to accelerate along a route like
the one shown in gure g. It rst accelerates along the y axis at a
constant rate (again, as judged by its own sensors), until its velocity
is very close to the speed of light, A. It then accelerates, again at
a self-perceived constant rate and with thrust in a xed direction
as judged by its own gyroscopes, until it is moving at the same
ultrarelativistic speed in the x direction, B. Finally, it decelerates
in the x direction until it is again at rest, O. This motion traces out
a clockwise loop on the velocity disk. The motion in space is also
clockwise.
We might naively think that the middle leg of the trip, from A
to B, would be a straight line on the velocity disk, but this cant be
the case. First, we know that non-collinear boosts cause rotations.
Traveling around a clockwise path causes counterclockwise rotation,
and vice-versa. Therefore an observer in the rest frame O sees the
ship (and its gyroscopes) as rotating as it moves from A to B. The
ships trajectory through space is clockwise, so according to O the
ship rotates counterclockwise as it goes A to B. The ship is always
ring its engines in a xed direction as judged by its gyroscopes, but
according to O the ship is rotating counterclockwise, its thrust is
progressively rotating counterclockwise, and therefore its trajectory
turns counterclockwise. We conclude that leg AB on the velocity
disk is concave, rather than being a straight-line hypotenuse of a
triangle OAB.
h / In the limit where A and
B are ultrarelativistic velocities,
leg AB is perpendicular to the
edge of the velocity disk. The
result is that the x y frame
determined by the ships gyro-
scopes has rotated by 90 degrees
by the time it gets home.
We can also determine, by the following argument, that leg AB
is perpendicular to the edge of the disk where it touches the edge of
the disk. In the transformation from frame A to frame O, y coordi-
nates are dilated by a factor of , which approaches innity in the
limit were presently considering. Observers aboard the rocket-ship,
occupying frame A, believe that their task is to re the rockets
engines at an angle of 45 degrees with respect to the y axis, so as
to eliminate their velocity with respect to the origin, and simulta-
neously add an equal amount of velocity in the x direction. This
45-degree angle in frame A, however, is not a 45-degree angle in
frame O. From the stern of the ship to its bow we have displace-
ments x and y, and in the transformation from A to O, y
is magnied almost innitely. As perceived in frame O, the ships
orientation is almost exactly antiparallel to the y axis.
18
As the ship travels from A to B, its orientation (as judged in
frame O) changes from y to x. This establishes, in a much more
direct fashion, the direction of the Thomas precession: its handed-
ness is contrary to the handedness of the direction of motion. We
can now also see something new about the fundamental reason for
the eect. It has to do with the fact that observers in dierent
states of motion disagree on spatial angles. Similarly, imagine that
you are a two-dimensional being who was told about the existence
of a new, third, spatial dimension. You have always believed that
the cosine of the angle between two unit vectors u and v is given by
the vector dot product u
x
v
x
+u
y
v
y
. If you were allowed to explore
a two-dimensional projection of a three-dimensional scene, e.g., on
the at screen of a television, it would seem to you as if all the
angles had been distorted. You would have no way to interpret the
visual conventions of perspective. But once you had learned about
the existence of a z axis, you would realize that these angular dis-
tortions were happening because of rotations out of the xy plane.
Such rotations really conserve the quantity u
x
v
x
+u
y
v
y
+u
z
v
z
; only
because you were ignoring the u
z
v
z
term did it seem that angles
were not being preserved. Similarly, the generalization from three
Euclidean spatial dimensions to 3+1-dimensional spacetime means
that three-dimensional dot products are no longer conserved.
The general low-v limit
Lets nd the low-v limit of the Thomas precession in general,
not just in the highly articial special case of v
2
for the example
involving the four hammer hits. To generalize to the case of smooth
18
Although we will not need any more than this for the purposes of our present
analysis, a longer and more detailed discussion by Rhodes and Semon, www.
bates.edu/
~
msemon/RhodesSemonFinal.pdf, Am. J. Phys. 72(7)2004, shows
that this type of inertially guided, constant-thrust motion is always represented
on the velocity disk by an arc of a circle that is perpendicular to the disk at its
edge. (We consider a diameter of the disk to be the limiting case of a circle with
innite radius.)
i / If the crack between the
two areas is squashed at, the
two pieces of the path on the
interior coincide, and their contri-
butions to the precession cancel
out (v v, but a +a, so
a v a v). Therefore the
precession obtained by going
around the outside is equal to the
sum
1
+
2
of the precessions
that would have been obtained by
going around the two parts.
acceleration, we rst note that the rate of precession d/dt must
have the following properties.
It is odd under a reversal of the direction of motion, v v.
(This corresponds to sending the gyroscope around the square
in the opposite direction.)
It is odd under a reversal of the acceleration due to the second
boost, a a.
It is a rotation about the spatial axis perpendicular to the
plane of the v and a vectors, in the opposite direction com-
pared to the handedness of the curving trajectory.
It is approximately linear in v and a, for small v and a.
The only rotationally invariant mathematical operation that has
these symmetry properties is the vector cross product, so the rate
of precession must be ka v, where k > 0 is nearly independent of
v and a for small v and a.
To pin down the value of k, we need to nd a connection between
our two results: v
2
for the four hammer hits, and d/dt kav
for smooth acceleration. We can do this by considering the physical
signicance of areas on the velocity disk. As shown in gure i, the
rotation due to carrying the velocity around the boundary of a
region is additive when adjacent regions are joined together. We
can therefore nd for any region by breaking the region down into
elements of area dA and integrating their contributions d. What is
the relationship between dA and d? The velocity disks structure
is nonuniform, in the sense that near the edge of the disk, it takes a
larger boost to move a small distance. But were investigating the
low-velocity limit, and in the low-velocity region near the center of
the disk, the disks structure is approximately uniform. We there-
fore expect that there is an approximately constant proportionality
factor relating dA and d at low velocities. The example of the
hammer corresponds geometrically to a square with area v
2
, so we
nd that this proportionality factor is unity, dA d.
To relate this to smooth acceleration, consider a particle per-
forming circular motion with period T, which has [av[ = 2v
2
/T.
Over one full period of the motion, we have =
_
k[a v[dt =
2kv
2
, and the particles velocity vector traces a circle of area A =
v
2
on the velocity disk. Equating A and , we nd k = 1/2. The
result is that in the limit of low velocities, the rate of rotation is

1
2
a v ,
where is the angular velocity vector of the rotation. In the special
case of circular motion, this can be written as = (1/2)v
2
, where
= 2/T is the angular frequency of the motion.
j / States in hydrogen are la-
beled with their and s quantum
numbers, representing their
orbital and spin angular momenta
in units of . The state with
s = +1/2 has its spin angular
momentum aligned with its orbital
angular momentum, while the
s = 1/2 state has the two
angular momenta in opposite
directions. The direction and
order of magnitude of the splitting
between the two = 1 states
is successfully explained by
magnetic interactions with the
proton, but the calculated effect
is too big by a factor of 2. The
relativistic Thomas precession
cancels out half of the effect.
2.5.4 An experimental test: Thomas precession in hydrogen
If we want to see this precession eect in real life, we should look
for a system in which both v and a are large. An atom is such a
system.
The Bohr model, introduced in 1913, marked the rst quantita-
tively successful, if conceptually muddled, description of the atomic
energy levels of hydrogen. Continuing to take c = 1, the over-all
scale of the energies was calculated to be proportional to m
2
, where
m is the mass of the electron, and = ke
2
/ 1/137, known as the
ne structure constant, is essentially just a unitless way of express-
ing the coupling constant for electrical forces. At higher resolution,
each excited energy level is found to be split into several sub-levels.
The transitions among these close-lying states are in the millime-
ter region of the microwave spectrum. The energy scale of this ne
structure is m
4
. This is down by a factor of
2
compared to the
visible-light transitions, hence the name of the constant. Uhlenbeck
and Goudsmit showed in 1926 that a splitting on this order of mag-
nitude was to be expected due to the magnetic interaction between
the proton and the electrons magnetic moment, oriented along its
spin. The eect they calculated, however, was too big by a factor
of two.
The explanation of the mysterious factor of two had in fact been
implicit in a 1916 calculation by Willem de Sitter, one of the rst
applications of general relativity. De Sitter treated the earth-moon
system as a gyroscope, and found the precession of its axis of rota-
tion, which was partly due to the curvature of spacetime and partly
due to the type of rotation described earlier in this section. The
eect on the motion of the moon was noncumulative, and was only
about one meter, which was much too small to be measured at the
time. In 1927, however, Llewellyn Thomas applied similar reason-
ing to the hydrogen atom, with the electrons spin vector playing
the role of gyroscope. Since gravity is negligible here, the eect has
nothing to do with curvature of spacetime, and Thomass eect cor-
responds purely to the special-relativistic part of de Sitters result.
It is simply the rotation described above, with = (1/2)v
2
. Al-
though Thomas was not the rst to calculate it, the eect is known
as Thomas precession. Since the electrons spin is /2, the energy
splitting is (/2), depending on whether the electrons spin is in
the same direction as its orbital motion, or in the opposite direc-
tion. This is less than the atoms gross energy scale by a factor
of v
2
/2, which is
2
. The Thomas precession cancels out half of
the magnetic eect, bringing theory in agreement with experiment.
Uhlenbeck later recalled: ...when I rst heard about [the Thomas
precession], it seemed unbelievable that a relativistic eect could
give a factor of 2 instead of something of order v/c... Even the
cognoscenti of relativity theory (Einstein included!) were quite sur-
prised.
Problems
1 Suppose that we dont yet know the exact form of the Lorentz
transformation, but we know based on the Michelson-Morley exper-
iment that the speed of light is the same in all inertial frames, and
weve already determined, e.g., by arguments like those on p. 58,
that there can be no length contraction in the direction perpendic-
ular to the motion. We construct a light clock, consisting simply
of two mirrors facing each other, with a light pulse bouncing back
and forth between them.
(a) Suppose this light clock is moving at a constant velocity v in the
direction perpendicular to its own optical arm, which is of length L.
Use the Pythagorean theorem to prove that the clock experiences a
time dilation given by = 1/
1 v
2
, thereby xing the time-time
portion of the Lorentz transformation.
(b) Why is it signicant for the interpretation of special relativity
that the result from part a is independent of L?
(c) Carry out a similar calculation in the case where the clock moves
with constant acceleration a as measured in some inertial frame. Al-
though the result depends on L, prove that in the limit of small L,
we recover the earlier constant-velocity result, with no explicit de-
pendence on a.
Remark: Some authors state a clock postulate for special relativity, which
says that for a clock that is suciently small, the rate at which it runs de-
pends only on v, not a (except in the trivial sense that v and a are related
by calculus). The result of part c shows that the clock postulate is really a
theorem, not a statement that is logically independent of the other postulates
of special relativity. Although this argument only applies to a particular fam-
ily of light clocks of various sizes, one can also make any small clock into an
acceleration-insensitive clock, by attaching an accelerometer to it and apply-
ing an appropriate correction to compensate for the clocks observed sensitiv-
ity to acceleration. (Its still necessary for the clock to be small, since other-
wise the lack of simultaneity in relativity makes it impossible to describe the
whole clock as having a certain acceleration at a certain instant.) Some peo-
ple get confused by this acceleration-independent property of small clocks and
think that it contradicts the equivalence principle. For a good explanation, see
http://math.ucr.edu/home/baez/physics/Relativity/SR/clock.html.
Solution, p. 261
2 (a) On p. 66 (see gure i), we showed that the Thomas pre-
cession is proportional to area on the velocity disk. Use a similar
argument to show that the Sagnac eect (p. 58) is proportional to
the area enclosed by the loop.
(b) Verify this more directly in the special case of a circular loop.
(c) Show that a light clock of the type described in problem 1 is
insensitive to rotation with constant angular velocity.
(d) Connect these results to the commutativity and transitivity as-
sumptions in the Einstein clock synchronization procedure described
on p. 242. Solution, p. 262
3 Example 10 on page 50 discusses relativistic bounds on the
properties of matter, using the example of pulling a bucket out of a
black hole. Derive a similar bound by considering the possibility of
sending signals out of the black hole using longitudinal vibrations of
a cable, as in the childs telephone made of two tin cans connected
by a piece of string. (Surprisingly subtle issues can arise in such
calculations. See A.Y. Shiekh, Can. J. Phys. 70, 458 (1992).)
4 The Maxima program on page 61 demonstrates how to mul-
tiply matrices and nd Taylor series. Apply this technique to the
following problem. For successive Lorentz boosts along the same
axis with rapidities
1
and
2
, nd the matrix representing the com-
bined Lorentz transformation, in a Taylor series up to the rst non-
classical terms in each matrix element. A mixed Taylor series in
two variables can be obtained simply by nesting taylor functions.
The taylor function will happily work on matrices, not just scalars.
Solution, p. 262
Problems 69
Chapter 3
Differential Geometry
General relativity is described mathematically in the language of
dierential geometry. Lets take those two terms in reverse order.
The geometry of spacetime is non-Euclidean, not just in the
sense that the 3+1-dimensional geometry of Lorentz frames is dif-
ferent than that of 4 interchangeable Euclidean dimensions, but also
in the sense that parallels do not behave in the way described by
E5 or A1-A3. In a Lorentz frame, which describes space without
any gravitational elds, particles whose world-lines are initially par-
allel will continue along their parallel world-lines forever. But in
the presence of gravitational elds, initially parallel world-lines of
free-falling particles will in general diverge, approach, or even cross.
Thus, neither the existence nor the uniqueness of parallels can be
assumed. We cant describe this lack of parallelism as arising from
the curvature of the world-lines, because were using the world-lines
of free-falling particles as our denition of a straight line. Instead,
we describe the eect as coming from the curvature of spacetime it-
self. The Lorentzian geometry is a description of the case in which
this curvature is negligible.
What about the word dierential ? The equivalence principle
states that even in the presence of gravitational elds, local Lorentz
frames exist. How local is local? If we use a microscope to zoom in
on smaller and smaller regions of spacetime, the Lorentzian approx-
imation becomes better and better. Suppose we want to do experi-
ments in a laboratory, and we want to ensure that when we compare
some physically observable quantity against predictions made based
on the Lorentz geometry, the resulting discrepancy will not be too
large. If the acceptable error is , then we should be able to get the
error down that low if were willing to make the size of our labora-
tory no bigger than . This is clearly very similar to the Weierstrass
style of dening limits and derivatives in calculus. In calculus, the
idea expressed by dierentiation is that every smooth curve can be
approximated locally by a line; in general relativity, the equivalence
principle tells us that curved spacetime can be approximated locally
by at spacetime. But consider that no practitioner of calculus ha-
bitually solves problems by lling sheets of scratch paper with ep-
silons and deltas. Instead, she uses the Leibniz notation, in which dy
and dx are interpreted as innitesimally small numbers. You may
be inclined, based on your previous training, to dismiss innitesi-
71
mals as neither rigorous nor necessary. In 1966, Abraham Robinson
demonstrated that concerns about rigor had been unfounded; well
come back to this point in section 3.2. Although it is true that any
calculation written using innitesimals can also be carried out using
limits, the following example shows how much more well suited the
innitesimal language is to dierential geometry.
Areas on a sphere Example: 1
The area of a region S in the Cartesian plane can be calculated
as
_
S
dA, where dA = dxdy is the area of an innitesimal rectan-
gle of width dx and height dy. A curved surface such as a sphere
does not admit a global Cartesian coordinate system in which the
constant coordinate curves are both uniformly spaced and per-
pendicular to one another. For example, lines of longitude on the
earths surface grow closer together as one moves away from the
equator. Letting be the angle with respect to the pole, and the
azimuthal angle, the approximately rectangular patch bounded by
, + d, , and + d has width r sind and height r d, giv-
ing dA = r
2
sindd. If you look at the corresponding derivation
in an elementary calculus textbook that strictly eschews innites-
imals, the technique is to start from scratch with Riemann sums.
This is extremely laborious, and moreover must be carried out
again for every new case. In differential geometry, the curvature
of the space varies from one point to the next, and clearly we
dont want to reinvent the wheel with Riemann sums an innite
number of times, once at each point in space.
3.1 The afne parameter revisited, and parallel
transport
3.1.1 The afne parameter in curved spacetime
An important example of the dierential, i.e., local, nature of our
geometry is the generalization of the ane parameter to a context
broader than ane geometry.
Our construction of the ane parameter with a scaolding of
parallelograms depended on the existence and uniqueness of paral-
lels expressed by A1, so we might imagine that there was no point in
trying to generalize the construction to curved spacetime. But the
equivalence principle tells us that spacetime is locally ane to some
approximation. Concretely, clock-time is one example of an ane
parameter, and the curvature of spacetime clearly cant prevent us
from building a clock and releasing it on a free-fall trajectory. To
generalize the recipe for the construction (gure a), the rst obsta-
cle is the ambiguity of the instruction to construct parallelogram
01q
0
q
1
, which requires us to draw 1q
1
parallel to 0q
0
. Suppose we
construe this as an instruction to make the two segments initially
parallel, i.e., parallel as they depart the line at 0 and 1. By the time
72 Chapter 3 Differential Geometry
a / Construction of an afne
parameter in curved spacetime.
they get to q
0
and q
1
, they may be converging or diverging.
Because parallelism is only approximate here, there will be a
certain amount of error in the construction of the ane parameter.
One way of detecting such an error is that lattices constructed with
dierent initial distances will get out of step with one another. For
example, we can dene
1
2
as before by requiring that the lattice
constructed with initial segment 0
1
2
line up with the original lattice
at 1. We will nd, however, that they do not quite line up at
other points, such as 2. Lets use this discrepancy = 2 2
as
a numerical measure of the error. It will depend on both
1
, the
distance 01, and on
2
, the distance between 0 and q
0
. Since
vanishes for either
1
= 0 or
2
= 0, and since the equivalence
principle guarantees smooth behavior on small scales, the leading
term in the error will in general be proportional to the product
2
. In the language of innitesimals, we can replace
1
and
2
with innitesimally short distances, which for simplicity we assume
to be equal, and which we call d. Then the ane parameter
is dened as =
_
d, where the error of order d
2
is, as usual,
interpreted as the negligible discrepancy between the integral and
its approximation as a Riemann sum.
3.1.2 Parallel transport
If you were alert, you may have realized that I cheated you at
a crucial point in this construction. We were to make 1q
1
and 0q
0
initially parallel as they left 01. How should we even dene this
idea of initially parallel? We could try to do it by making angles
q
0
01 and q
1
12 equal, but this doesnt quite work, because it doesnt
specify whether the angle is to the left or the right on the two-
dimensional plane of the page. In three or more dimensions, the
issue becomes even more serious. The construction workers building
the lattice need to keep it all in one plane, but how do they do that
in curved spacetime?
A mathematicians answer would be that our geometry lacks
some additional structure called a connection, which is a rule that
species how one locally at neighborhood is to be joined seamlessly
onto another locally at neighborhood nearby. If youve ever bought
two maps and tried to tape them together to make a big map, youve
formed a connection. If the maps were on a large enough scale,
you also probably noticed that this was impossible to do perfectly,
because of the curvature of the earth.
Physically, the idea is that in at spacetime, it is possible to
construct inertial guidance systems like the ones discussed on page
59. Since they are possible in at spacetime, they are also possible
in locally at neighborhoods of spacetime, and they can then be
carried from one neighborhood to another.
In three space dimensions, a gyroscopes angular momentum vec-
Section 3.1 The afne parameter revisited, and parallel transport 73
c / Bad things happen if we
try to construct an afne param-
eter along a curve that isnt a
geodesic. This curve is similar
to path ABC in gure b. Par-
allel transport doesnt preserve
the vectors angle relative to
the curve, as it would with a
geodesic. The errors in the
construction blow up in a way
that wouldnt happen if the curve
had been a geodesic. The fourth
dashed parallel ies off wildly
around the back of the sphere,
wrapping around and meeting
the curve at a point, 4, that is
essentially random.
b / Parallel transport is path-
dependent. On the surface of
this sphere, parallel-transporting
a vector along ABC gives a
different answer than transporting
it along AC.
tor maintains its direction, and we can orient other vectors, such as
1q
1
, relative to it. Suppose for concreteness that the construction
of the ane parameter above is being carried out in three space di-
mensions. We place a gyroscope at 0, orient its axis along 0q
0
, slide
it along the line to 1, and then construct 1q
1
along that axis.
In 3+1 dimensions, a gyroscope only does part of the job. We
now have to maintain the direction of a four-dimensional vector.
Four-vectors will not be discussed in detail until section 4.2, but
similar devices can be used to maintain their orientations in space-
time. These physical devices are ways of dening a mathematical
notion known as parallel transport, which allows us to take a vector
from one point to another in space. In general, specifying a notion
of parallel transport is equivalent to specifying a connection.
Parallel transport is path-dependent, as shown in gure b.
Ane parameters dened only along geodesics
In the context of at spacetime, the ane parameter was dened
only along lines, not arbitrary curves, and could not be compared
between lines running in dierent directions. In curved spacetime,
the same limitation is present, but with along lines replaced by
along geodesics. Figure c shows what goes wrong if we try to
apply the construction to a world-line that isnt a geodesic. One
denition of a geodesic is that its the course well end up following
if we navigate by keeping a xed bearing relative to an inertial guid-
ance device such as gyroscope; that is, the tangent to a geodesic,
when parallel-transported farther along the geodesic, is still tangent.
A non-geodesic curve lacks this property, and the eect on the con-
struction of the ane parameter is that the segments nq
n
drift more
and more out of alignment with the curve.
3.2 Models
A typical rst reaction to the phrase curved spacetime or even
curved space, for that matter is that it sounds like nonsense.
How can featureless, empty space itself be curved or distorted? The
concept of a distortion would seem to imply taking all the points
and shoving them around in various directions as in a Picasso paint-
ing, so that distances between points are altered. But if space has
no identiable dents or scratches, it would seem impossible to deter-
mine which old points had been sent to which new points, and the
distortion would have no observable eect at all. Why should we
expect to be able to build dierential geometry on such a logically
dubious foundation? Indeed, historically, various mathematicians
have had strong doubts about the logical self-consistency of both
non-Euclidean geometry and innitesimals. And even if an authori-
tative source assures you that the resulting system is self-consistent,
its mysterious and abstract nature would seem to make it dicult
a / Tullio Levi-Civita (1873-
1941) worked on models of
number systems possessing
innitesimals and on differential
geometry. He invented the
tensor notation, which Einstein
learned from his textbook. He
was appointed to prestigious
endowed chairs at Padua and the
University of Rome, but was red
in 1938 because he was a Jew
and an anti-fascist.
for you to develop any working picture of the theory that could
play the role that mental sketches of graphs play in organizing your
knowledge of calculus.
Models provide a way of dealing with both the logical issues and
the conceptual ones. Figure a on page 73 pops o of the page,
presenting a strong psychological impression of a curved surface ren-
dered in perspective. This suggests nding an actual mathematical
object, such as a curved surface, that satises all the axioms of a
certain logical system, such as non-Euclidean geometry. Note that
the model may contain extrinsic elements, such as the existence of
a third dimension, that are not connected to the system being mod-
eled.
Lets focus rst on consistency. In general, what can we say
about the self-consistency of a mathematical system? To start with,
we can never prove anything about the consistency or lack of consis-
tency of something that is not a well-dened formal system, e.g., the
Bible. Even Euclids Elements, which was a model of formal rigor for
thousands of years, is loose enough to allow considerable ambiguity.
If youre inclined to sco at the silly Renaissance mathematicians
who kept trying to prove the parallel postulate E5 from postulates
E1-E4, consider the following argument. Suppose that we replace
E5 with E5
, which states that parallels dont exist: given a line and

a point not on the line, no line can ever be drawn through the point
and parallel to the given line. In the new system of plane geometry
E
consisting of E1-E4 plus E5
, we can prove a variety of theorems,

and one of them is that there is an upper limit on the area of any g-
ure. This imposes a limit on the size of circles, and that appears to
contradict E3, which says we can construct a circle with any radius.
We therefore conclude that E
lacks self-consistency. Oops! As your

high school geometry text undoubtedly mentioned in passing, E
is
a perfectly respectable system called elliptic geometry. So whats
wrong with this supposed proof of its lack of self-consistency? The
issue is the exact statement of E3. E3 does not say that we can
construct a circle given any real number as its radius. Euclid could
not have intended any such interpretation, since he had no notion of
real numbers. To Euclid, geometry was primary, and numbers were
geometrically constructed objects, being represented as lengths, an-
gles, areas, and volumes. A literal translation of Euclids statement
of the axiom is To describe a circle with any center and distance.
1
Distance means a line segment. There is therefore no contradic-
tion in E
, because E
has a limit on the lengths of line segments.

Now suppose that such ambiguities have been eliminated from
the systems basic denitions and axioms. In general, we expect
it to be easier to prove an inconsistent systems inconsistency than
to demonstrate the consistency of a consistent one. In the former
1
Heath, pp. 195-202
Section 3.2 Models 75
case, we can start cranking out theorems, and if we can nd a way
to prove both proposition P and its negation P, then obviously
something is wrong with the system. One might wonder whether
such a contradiction could remain contained within one corner of
the system, like nuclear waste. It cant. Aristotelian logic allows
proof by contradiction: if we prove both P and P based on certain
assumptions, then our assumptions must have been wrong. If we
can prove both P and P without making any assumptions, then
proof by contradiction allows us to establish the truth of any ran-
domly chosen proposition. Thus a single contradiction is sucient,
in Aristotelian logic, to invalidate the entire system. This goes by
the Latin rubric ex falso quodlibet, meaning from a falsehood, what-
ever you please. Thus any contradiction proves the inconsistency
of the entire system.
Proving consistency is harder. If youre mathematically sophisti-
cated, you may be tempted to leap directly to Godels theorem, and
state that nobody can ever prove the self-consistency of a mathemat-
ical system. This would be a misapplication of Godel. Godels theo-
rem only applies to mathematical systems that meet certain techni-
cal criteria, and some of the interesting systems were dealing with
dont meet those criteria; in particular, Godels theorem doesnt
apply to Euclidean geometry, and Euclidean geometry was proved
self-consistent by Tarski and his students around 1950. Further-
more, we usually dont require an absolute proof of self-consistency.
Usually were satised if we can prove that a certain system, such
as elliptic geometry, is at least as self-consistent as another system,
such as Euclidean geometry. This is called equiconsistency. The
general technique for proving equiconsistency of two theories is to
show that a model of one can be constructed within the other.
Suppose, for example, that we construct a geometry in which the
space of points is the surface of a sphere, and lines are understood
to be the geodesics, i.e., the great circles whose centers coincide at
the spheres center. This geometry, called spherical geometry, is
useful in cartography and navigation. It is non-Euclidean, as we
can demonstrate by exhibiting at least one proposition that is false
in Euclidean geometry. For example, construct a triangle on the
earths surface with one corner at the north pole, and the other
two at the equator, separated by 90 degrees of longitude. The sum
of its interior angles is 270 degrees, contradicting Euclid, book I,
proposition 32. Spherical geometry must therefore violate at least
one of the axioms E1-E5, and indeed it violates both E1 (because
no unique line is determined by two antipodal points such as the
north and south poles) and E5 (because parallels dont exist at all).
A closely related construction gives a model of elliptic geometry,
in which E1 holds, and only E5 is thrown overboard. To accomplish
this, we model a point using a diameter of the sphere,
2
and a line as
the set of all diameters lying in a certain plane. This has the eect
of identifying antipodal points, so that there is now no violation of
E1. Roughly speaking, this is like lopping o half of the sphere, but
making the edges wrap around. Since this model of elliptic geometry
is embedded within a Euclidean space, all the axioms of elliptic
geometry can now be proved as theorems in Euclidean geometry. If a
contradiction arose from them, it would imply a contradiction in the
axioms of Euclidean geometry. We conclude that elliptic geometry
is equiconsistent with Euclidean geometry. This was known long
before Tarskis 1950 proof of Euclidean geometrys self-consistency,
but since nobody was losing any sleep over hidden contradictions
in Euclidean geometry, mathematicians stopped wasting their time
looking for contradictions in elliptic geometry.
Innitesimals Example: 2
Consider the following axiomatically dened system of numbers:
1. It is a eld, i.e., it has addition, subtraction, multiplication, and
division with the usual properties.
2. It is an ordered geometry in the sense of O1-O4, and the
ordering relates to addition and multiplication in the usual way.
3. Existence of innitesimals: There exists a positive number d
such that d < 1, d < 1/2, d < 1/3, . . .
A model of this system can be constructed within the real number
system by dening d as the identity function d(x) = x and forming
the set of functions of the form f (d) = P(d)/Q(d), where P and Q
are polynomials with real coefcients. The ordering of functions f
and g is dened according to the sign of lim
x0
+ f (x) g(x). Ax-
ioms 1-3 can all be proved from the real-number axioms. There-
fore this system, which includes innitesimals, is equiconsistent
with the reals. More elaborate constructions can extend this to
systems that more of the properties of the reals, and a browser-
based calculator that implements such a system is available at
lightandmatter.com/calc/inf. Abraham Robinson extended
this in 1966 to all of analysis, and thus there is nothing intrinsically
nonrigorous about doing analysis in the style of Gauss and Eu-
ler, with symbols like dx representing innitesimally small quanti-
ties.
3
Besides proving consistency, these models give us insight into
whats going on. The model of elliptic geometry suggests an in-
sight into the reason that there is an upper limit on lengths and
2
The term elliptic may be somewhat misleading here. The model is still
constructed from a sphere, not an ellipsoid.
3
More on this topic is available in, for example, Keislers Elementary Calcu-
lus: An Innitesimal Approach, Stroyans A Brief Introduction to Innitesimal
Calculus, or my own Calculus, all of which are available for free online.
Section 3.2 Models 77
b / An Einsteins ring is formed
when there is a chance alignment
of a distant source with a closer
gravitating body. Here, a quasar,
MG1131+0456, is seen as a
ring due to focusing of light by
an unknown object, possibly a
supermassive black hole. Be-
cause the entire arrangement
lacks perfect axial symmetry, the
ring is nonuniform; most of its
brightness is concentrated in two
lumps on opposite sides. This
type of gravitational lensing is
direct evidence for the curvature
of space predicted by gravita-
tional lensing. The two geodesics
form a lune, which is a gure
that cannot exist in Euclidean
geometry.
areas: it is because the space wraps around on itself. The model of
innitesimals suggests a fact that is not immediately obvious from
the axioms: the innitesimal quantities compose a hierarchy, so that
for example 7d is in nite proportion to d, while d
2
is like a lesser
ea in Swifts doggerel: Big eas have little eas/ On their backs
to ride em,/ and little eas have lesser eas,/And so, ad innitum.
Spherical and elliptic geometry are not valid models of a general-
relativistic spacetime, since they are locally Euclidean rather than
Lorentzian, but they still provide us with enough conceptual guid-
ance to come up with some ideas that might never have occurred to
us otherwise:
In spherical geometry, we can have a two-sided polygon called
a lune that encloses a nonzero area. In general relativity, a
lune formed by the world-lines of two particles represents mo-
tion in which the particles separate but are later reunited,
presumably because of some mass between them that created
a gravitational eld. An example is gravitational lensing.
Both spherical models wraps around on themselves, so that
they are not topologically equivalent to innite planes. We
therefore form a conjecture there may be a link between cur-
vature, which is a local property, and topology, which is global.
Such a connection is indeed observed in relativity. For exam-
ple, cosmological solutions of the equations of general relativ-
ity come in two avors. One type has enough matter in it to
produce more than a certain critical amount of curvature, and
this type is topologically closed. It describes a universe that
has nite spatial volume, and that will only exist for a nite
time before it recontracts in a Big Crunch. The other type,
corresponding to the universe we actually inhabit, has innite
spatial volume, will exist for innite time, and is topologically
open.
There is a distance scale set by the size of the sphere, with its
inverse being a measure of curvature. In general relativity,
we expect there to be a similar way to measure curvature
numerically, although the curvature may vary from point to
point.
Self-check: Prove from the axioms E
that elliptic geometry, un-

like spherical geometry, cannot have a lune with two distinct ver-
tices. Convince yourself nevertheless, using the spherical model of
E
, that it is possible in elliptic geometry for two lines to enclose a

region of space, in the sense that from any point P in the region,
a ray emitted in any direction must intersect one of the two lines.
Summarize these observations with a characterization of lunes in
elliptic geometry versus lunes in spherical geometry.
3.3 Intrinsic quantities
Models can be dangerous, because they can tempt us to impute
physical reality to features that are purely extrinsic, i.e., that are
only present in that particular model. This is as opposed to intrinsic
features, which are present in all models, and which are therefore
logically implied by the axioms of the system itself. The existence
of lunes is clearly an intrinsic feature of non-Euclidean geometries,
because intersection of lines was dened before any model has even
been proposed.
Curvature in elliptic geometry Example: 3
What about curvature? In the spherical model of elliptic geom-
etry, the size of the sphere is an inverse measure of curvature.
Is this a valid intrinsic quantity, or is it extrinsic? It seems sus-
pect, because it is a feature of the model. If we try to dene
size as the radius R of the sphere, there is clearly reason for
concern, because this seems to refer to the center of the sphere,
but existence of a three-dimensional Euclidean space inside and
outside the surface is clearly an extrinsic feature of the model.
There is, however, a way in which a creature conned to the sur-
face can determine R, by constructing geodesic and an afne
parameter along that geodesic, and measuring the distance ac-
cumulated until the geodesic returns to the initial point. Since
antipodal points are identied, equals half the circumference of
the sphere, not its whole circumference, so R = /, by wholly
intrinsic methods.
Extrinsic curvature Example: 4
Euclids axioms E1-E5 refer to explicit constructions. If a two-
dimensional being can physically verify them all as descriptions of
the two-dimensional space she inhabits, then she knows that her
space is Euclidean, and that propositions such as the Pythagorean
theorem are physically valid in her universe. But the diagram in
a/1 illustrating illustrating the proof of the Pythagorean theorem in
Euclids Elements (proposition I.47) is equally valid if the page is
rolled onto a cylinder, 2, or formed into a wavy corrugated shape,
3. These types of curvature, which can be achieved without tear-
ing or crumpling the surface, are extrinsic rather than intrinsic. Of
the curved surfaces in gure a, only the sphere, 4, has intrinsic
curvature; the diagram cant be plastered onto the sphere without
folding or cutting and pasting.
Self-check: How would the ideas of example 4 apply to a cone?
Example 4 shows that it can be dicult to sni out bogus ex-
trinsic features that seem intrinsic, and example 3 suggests the de-
sirability of developing methods of calculation that never refer to
any extrinsic quantities, so that we never have to worry whether a
Section 3.3 Intrinsic quantities 79
a / Example 79.
symbol like R staring up at us from a piece of paper is intrinsic.
This is why it is unlikely to be helpful to a student of general rel-
ativity to pick up a book on dierential geometry that was written
without general relativity specically in mind. Such books have a
tendency to casually mix together intrinsic and extrinsic notation.
For example, a vector cross product a b refers to a vector poking
out of the plane occupied by a and b, and the space outside the
plane may be extrinsic; it is not obvious how to generalize this op-
eration to the 3+1 dimensions of relativity (since the cross product
is a three-dimensional beast), and even if it were, we could not be
assured that it would have any intrinsically well dened meaning.
3.3.1 Coordinate independence
To see how to proceed in creating a manifestly intrinsic notation,
consider the two types of intrinsic observations that are available in
general relativity:
1. We can tell whether events and world-lines are incident:
whether or not two lines intersect, two events coincide, or an
event lies on a certain line.
Incidence measurements, for example detection of gravitational lens-
ing, are global, but they are the only global observations we can
do.
4
If we were limited entirely to incidence, spacetime would be
described by the austere system of projective geometry, a geometry
without parallels or measurement. In projective geometry, all propo-
sitions are essentially statements about combinatorics, e.g., that it
is impossible to plant seven trees so that they form seven lines of
three trees each.
But:
2. We can also do measurements in local Lorentz frames.
4
Einstein referred to incidence measurements as determinations of space-
time coincidences. For his presentation of this idea, see p. 259.
b / A series of Lorentz boosts
acts on a square.
This gives us more power, but not as much as we might expect.
Suppose we dene a coordinate such as t or x. In Newtonian me-
chanics, these coordinates would form a predened background, a
preexisting stage for the actors. In relativity, on the other hand,
consider a completely arbitrary change of coordinates of the form
x x
= f(x), where f is a smooth one-to-one function. For ex-

ample, we could have x x +px
3
+q sin(rx) (with p and q chosen
small enough so that the mapping is always one-to-one). Since the
mapping is one-to-one, the new coordinate system preserves all the
incidence relations. Since the mapping is smooth, the new coordi-
nate system is still compatible with the existence of local Lorentz
frames. The dierence between the two coordinate systems is there-
fore entirely extrinsic, and we conclude that a manifestly intrinsic
notation should avoid any explicit reference to a coordinate system.
That is, if we write a calculation in which a symbol such as x ap-
pears, we need to make sure that nowhere in the notation is there
any hidden assumption that x comes from any particular coordinate
system. For example, the equation should still be valid if the generic
symbol x is later taken to represent the distance r from some center
of symmetry. This coordinate-independence property is also known
as general covariance.
Frames moving at c?
A good application of these ideas is to the question of what the
world would look like in a frame of reference moving at the speed
of light. This question has a long and honorable history. As a
young student, Einstein tried to imagine what an electromagnetic
wave would look like from the point of view of a motorcyclist riding
alongside it. We now know, thanks to Einstein himself, that it really
doesnt make sense to talk about such observers.
The most straightforward argument is based on the positivist
idea that concepts only mean something if you can dene how to
measure them operationally. If we accept this philosophical stance
(which is by no means compatible with every concept we ever discuss
in physics), then we need to be able to physically realize this frame
in terms of an observer and measuring devices. But we cant. It
would take an innite amount of energy to accelerate Einstein and
his motorcycle to the speed of light.
Since arguments from positivism can often kill o perfectly in-
teresting and reasonable concepts, we might ask whether there are
other reasons not to allow such frames. There are. Recall that
we placed two technical conditions on coordinate transformations:
they are supposed to be smooth and one-to-one. The smoothness
condition is related to the inability to boost Einsteins motorcycle
into the speed-of-light frame by any continuous, classical process.
(Relativity is a classical theory.) But independent of that, we have
a problem with the one-to-one requirement. Figure b shows what
Section 3.3 Intrinsic quantities 81
b / The vectors dx
and dx
are duals of each other.

a / The tick marks on the line
dene a coordinate measured
along the line. It is not possible to
set up such a coordinate system
globally so that the coordinate is
uniform everywhere. The arrows
represent changes in the value
coordinate; since the changes in
the coordinate are all equal, the
arrows are all the same length.
happens if we do a series of Lorentz boosts to higher and higher
velocities. It should be clear that if we could do a boost up to a ve-
locity of c, we would have eected a coordinate transformation that
was not one-to-one. Every point in the plane would be mapped onto
a single lightlike line.
3.4 The metric
Applying these considerations to the creation of a manifestly intrin-
sic notation, consider a coordinate x dened along a certain curve,
which is not necessarily a geodesic. For concreteness, imagine this
curve to exist in two spacelike dimensions, which we can visualize
as the surface of a sphere embedded in Euclidean 3-space. These
concrete features are not strictly necessary, but they drive home
the point that we should not expect to be able to dene x so that
it varies at a steady rate with elapsed distance; it is not possible
to dene this type of uniform, Cartesian coordinate system on the
surface of a sphere. In the gure, the tick marks are therefore not
evenly spaced. This is perfectly all right, given the coordinate in-
variance of general relativity. Since the incremental changes in x
are equal, Ive represented them below the curve as little vectors
of equal length. They are the wrong length to represent distances
along the curve, but this wrongness is an inevitable fact of life in
relativity.
Now suppose we want to integrate the arc length of a segment
of this curve. The little vectors are innitesimal. In the integrated
length, each little vector should contribute some amount, which is
a scalar. This scalar is not simply the magnitude of the vector,
ds ,=
dx dx, since the vectors are the wrong length. We therefore

need some mathematical rule, some function, that accepts a vector
as its input and gives a scalar as its output. This function is a locally
adjustable fudge factor that compensates for the wrong lengths of
the little vectors. Since the space is locally at and uniform, the
function must be linear, and from linear algebra, we know that the
most general function of this kind is an inner product. If the little
arrow is a row vector, then the function would be represented by
taking the row vectors inner product with some column vector to
give ds
2
. Of course the distinction between row and column vectors
is pointless in a one-dimensional space, but it should be clear that
this will provide an appropriate foundation for the generalization to
more than one coordinate. The row and column vectors are referred
to as one anothers duals. Figure b shows the resulting picture.
Anticipating the generalization to four-dimensional spacetime with
coordinates (x
0
, x
1
, x
2
, x
3
), well start referring to x as x
, although
in our present one-dimensional example = 0 is xed. The reason
for the use of the odd-looking superscripts, rather than subscripts,
will become clear shortly.
The vectors drawn below the curve are called the contravariant
vectors, notated dx
, and the ones above it are the covariant vectors,

dx
. Its not particularly important to keep track of which is which,

since the relationship between them is symmetric, like the relation-
ship between row and column vectors. Each is the dual of the other.
The arc length is given by
_
ds =
_ _
dx
dx
, or, equivalently we
say ds
2
= dx
dx
. (Remember, = 0.)
Given a dx
, how do we nd its dual dx
, and vice versa? In

one dimension, we simply need to introduce a real number g as a
correction factor. If one of the vectors is shorter than it should be
in a certain region, the correction factor serves to compensate by
making its dual proportionately longer. The two possible mappings
(covariant to contravariant and contravariant to covariant) are ac-
complished with factors of g and 1/g. The number g is called the
metric, and it encodes all the information about distances. For ex-
ample, if represents longitude measured at the arctic circle, then
the metric is the only source for the datum that a displacement d
corresponds to 2540 km per radian.
Now lets generalize to more than one dimension. Because glob-
ally Cartesian coordinate systems cant be imposed on a curved
space, the constant-coordinate lines will in general be neither evenly
spaced nor perpendicular to one another. If we construct a local
set of basis vectors lying along the intersections of the constant-
coordinate surfaces, they will not form an orthonormal set. We
would like to have an expression of the form ds
2
= dx
dx
for the
squared arc length, and in dierential geometry we practice the con-
venient notational convention, introduced by Einstein, of assuming
a summation when an index is repeated, so this becomes
ds
2
= dx
dx
.
3.4.1 The Euclidean metric
In a Euclidean plane, where the distinction between covariant
and contravariant vectors is irrelevant, this expression for ds
2
is
simply the Pythagorean theorem, summed over two values of i for
the two coordinates:
ds
2
= dx
i
dx
i
= dx
2
+ dy
2
The symbols dx, dx
0
, and dx
0
are all synonyms, and likewise for
dy, dx
1
, and dx
1
.
In the non-Euclidean case, the Pythagorean theorem is false; dx
and dx
are no longer synonyms, so their product is no longer simply

the square of a distance. To see this more explicitly, lets write the
expression so that only the covariant quantities occur. By local
atness, the relationship between the covariant and contravariant
vectors is linear, and the most general relationship of this kind is
given by making the metric a symmetric matrix g
. Substituting
Section 3.4 The metric 83
dx
= g
, we have
ds
2
= g
dx
dx
,
where there are now implied sums over both and . Notice how
implied sums occur only when the repeated index occurs once as
a superscript and once as a subscript; other combinations are un-
grammatical.
Self-check: Why does it make sense to demand that the metric
be symmetric?
In an introductory course in Newtonian mechanics, one makes
a distinction between vectors, which have a direction in space, and
scalars, which do not. These are specic examples of tensors, which
can be expressed as objects with m superscripts and n subscripts.
A scalar has m = n = 0. A covariant vector has (m, n) = (0, 1),
a contravariant vector (1, 0), and the metric (0, 2). We refer to the
number of indices as the rank of the tensor. Tensors are discussed
in more detail, and dened more rigorously, in chapter 4. For our
present purposes, it is important to note that just because we write
a symbol with subscripts or superscripts, that doesnt mean it de-
serves to be called a tensor. This point can be understood in the
more elementary context of Newtonian scalars and vectors. For ex-
ample, we can dene a Newtonian vector u = (m, T, e), where m
is the mass of the moon, T is the temperature in Chicago, and e is
the charge of the electron. This creature u doesnt deserve to be
called a vector, because it doesnt behave as a vector under rota-
tion. Similarly, a tensor is required to behave in a certain way under
rotations and Lorentz boosts.
When discussing the symmetry of rank-2 tensors, it is convenient
to introduce the following notation:
T
(ab)
=
1
2
(T
ab
+T
ba
)
T
[ab]
=
1
2
(T
ab
T
ba
)
Any T
ab
can be split into symmetric and antisymmetric parts. This
is similar to writing an arbitrary function as a sum of and odd
function and an even function. The metric has only a symmetric
part: g
(ab)
= g
ab
, and g
[ab]
= 0. This notation is generalized to
ranks greater than 2 on page 152.
Self-check: Characterize an antisymmetric rank-2 tensor in two
dimensions.
A change of scale Example: 5
How is the effect of a uniform rescaling of coordinates repre-
sented in g?
If we change our units of measurement so that x
, while
demanding that ds
2
come out the same, then we need g
2
g
.
Polar coordinates Example: 6
Consider polar coordinates (r , ) in a Euclidean plane. The const-
ant-coordinate curves happen to be orthogonal everywhere, so
the off-diagonal elements of the metric g
r
and g
r
vanish. In-
nitesimal coordinate changes dr and d correspond to innitesi-
mal displacements dr and r d in orthogonal directions, so by the
Pythagorean theorem, ds
2
= dr
2
+ r
2
d
2
, and we read off the
elements of the metric g
r r
= 1 and g
= r
2
.
Notice how in example 6 we started from the generally valid
relation ds
2
= g
dx
dx
, but soon began writing down facts like

g
= r
2
that were only valid in this particular coordinate system.
To make it clear when this is happening, we adopt a convention in-
troduced by Roger Penrose known as the abstract index notation. In
this convention, Latin superscripts and subscripts indicate that an
equation is of general validity, without regard to any choice of coor-
dinate system, while Greek ones are used for coordinate-dependent
equations. For example, we can write the general expression for
squared dierential arc length with Latin indices,
ds
2
= g
ij
dx
i
dx
j
,
because it holds regardless of the coordinate system, whereas the
vanishing of the o-diagonal elements of the metric in Euclidean
polar coordinates has to be written as g
= 0 for ,= , since it
would in general be false if we used a dierent coordinate system to
describe the same Euclidean plane. The advantages of this notation
became widely apparent to relativists starting around 1980, so for
example it is used in the text by Wald (1984), but not in Misner,
Thorne, and Wheeler (1970). Some of the older literature uses a
notation in which the Greek and Latin indices are instead used to
distinguish between timelike and spacelike components of a vector,
but this usage is dying out, since it inappropriately singles out a
distinction between time and space that is not actually preserved
under a Lorentz boost.
Oblique Cartesian coordinates Example: 7
Oblique Cartesian coordinates are like normal Cartesian coor-
dinates in the plane, but their axes are at at an angle ,= /2 to
one another. Find the metric in these coordinates. The space is
globally Euclidean.
Since the coordinates differ from Cartesian coordinates only in
the angle between the axes, not in their scales, a displacement
dx
i
along either axis, i = 1 or 2, must give ds = dx, so for the diag-
onal elements we have g
11
= g
22
= 1. The metric is always sym-
metric, so g
12
= g
21
. To x these off-diagonal elements, consider
a displacement by ds in the direction perpendicular to axis 1. This
changes the coordinates by dx
1
= ds cot and dx
2
= ds csc .
c / Example 7.
We then have
ds
2
= g
i j
dx
i
dx
j
= ds
2
(cot
2
+ csc
2
2g
12
cos csc )
g
12
= cos .
Area Example: 8
In one dimension, g is a single number, and lengths are given
by ds =

gdx. The square root can also be understood through
example 5 on page 84, in which we saw that a uniform rescaling
x x is reected in g
2
g
.
In two-dimensional Cartesian coordinates, multiplication of the
width and height of a rectangle gives the element of area dA =
g
11
g
22
dx
1
dx
2
. Because the coordinates are orthogonal, g is di-
agonal, and the factor of
g
11
g
22
is identied as the square root
of its determinant, so dA =
_
[g[dx
1
dx
2
. Note that the scales on
the two axes are not necessarily the same, g
11
,= g
22
.
The same expression for the element of area holds even if the co-
ordinates are not orthogonal. In example 7, for instance, we have
_
[g[ =
_
1 cos
2
= sin, which is the right correction factor
corresponding to the fact that dx
1
and dx
2
form a parallelepiped
rather than a rectangle.
Area of a sphere Example: 9
For coordinates (, ) on the surface of a sphere of radius r , we
have, by an argument similar to that of example 6 on page 85,
g
= r
2
, g
= r
2
sin
2
, g
= 0. The area of the sphere is

A =
_
dA
=
_ _
_
[g[dd
= r
2
_ _
sindd
= 4r
2
Inverse of the metric Example: 10
Relate g
i j
to g
i j
.
The notation is intended to treat covariant and contravariant
vectors completely symmetrically. The metric with lower indices
g
i j
can be interpreted as a change-of-basis transformation from a
contravariant basis to a covariant one, and if the symmetry of the
notation is to be maintained, g
i j
must be the corresponding in-
verse matrix, which changes from the covariant basis to the con-
travariant one. The metric must always be invertible.
3.4.2 The Lorentz metric
In a locally Euclidean space, the Pythagorean theorem allows us
to express the metric in local Cartesian coordinates in the simple
form g
= +1, g
= 0, i.e., g = diag(+1, +1, . . . , +1). This is

not the appropriate metric for a locally Lorentz space. The axioms
of Euclidean geometry E3 (existence of circles) and E4 (equality of
right angles) describe the theorys invariance under rotations, and
the Pythagorean theorem is consistent with this, because it gives the
same answer for the length of a vector even if its components are
reexpressed in a new basis that is rotated with respect to the original
one. In a Lorentzian geometry, however, we care about invariance
under Lorentz boosts, which do not preserve the quantity t
2
+ x
2
.
It is not circles in the (t, x) plane that are invariant, but light cones,
and this is described by giving g
tt
and g
xx
opposite signs and equal
absolute values. A lightlike vector (t, x), with t = x, therefore has
a magnitude of exactly zero,
s
2
= g
tt
t
2
+g
xx
x
2
= 0 ,
and this remains true after the Lorentz boost (t, x) (t, x). It
is a matter of convention which element of the metric to make pos-
itive and which to make negative. In this book, Ill use g
tt
= +1
and g
xx
= 1, so that g = diag(+1, 1). This has the advan-
tage that any line segment representing the timelike world-line of a
physical object has a positive squared magnitude; the forward ow
of time is represented as a positive number, in keeping with the
philosophy that relativity is basically a theory of how causal rela-
tionships work. With this sign convention, spacelike vectors have
positive squared magnitudes, timelike ones negative. The same con-
vention is followed, for example, by Penrose. The opposite version,
with g = diag(1, +1) is used by authors such as Wald and Misner,
Thorne, and Wheeler.
Our universe does not have just one spatial dimension, it has
three, so the full metric in a Lorentz frame is given by
g = diag(+1, 1, 1, 1).
3.4.3 Isometry, inner products, and the Erlangen Program
In Euclidean geometry, the dot product of vectors a and b is
given by g
xx
a
x
b
x
+ g
yy
a
y
b
y
+ g
zz
a
z
b
z
= a
x
b
x
+ a
y
b
y
+ a
z
b
z
, and in
the special case where a = b we have the squared magnitude. In
the tensor notation, a
= a
1
b
1
+ a
2
b
2
+ a
3
b
3
. Like magnitudes,
dot products are invariant under rotations. This is because know-
ing the dot product of vectors a and b entails knowing the value
of a b = [a[[a[ cos
ab
, and Euclids E4 (equality of right angles)
implies that the angle
ab
is invariant. the same axioms also entail
invariance of dot products under translation; Euclid waits only until
the second proposition of the Elements to prove that line segments
can be copied from one location to another. This seeming triviality is
actually false as a description of physical space, because it amounts
to a statement that space has the same properties everywhere.
The set of all transformations that can be built out of succes-
sive translations, rotations, and reections is called the group of
isometries. It can also be dened as the group
5
that preserves dot
products, or the group that preserves congruence of triangles.
In Lorentzian geometry, we usually avoid the Euclidean term
dot product and refer to the corresponding operation by the more
general term inner product. In a specic coordinate system we have
a
= a
0
b
0
a
1
b
1
a
2
b
2
a
3
b
3
. The inner product is invariant under
Lorentz boosts, and also under the Euclidean isometries. The group
found by making all possible combinations of continuous transfor-
mations
6
from these two sets is called the Poincare group. The
Poincare group is not the symmetry group of all of spacetime, since
curved spacetime has dierent properties in dierent locations. The
equivalence principle tells us, however, that space can be approxi-
mated locally as being at, so the Poincare group is locally valid,
just as the Euclidean isometries are locally valid as a description of
geometry on the Earths curved surface.
The triangle inequality Example: 11
In Euclidean geometry, the triangle inequality [b + c[ < [b[ + [c[
follows from
([b[ + [c[)
2
(b + c) (b + c) = 2([b[[c[ b c) 0 .
The reason this quantity always comes out positive is that for two
vectors of xed magnitude, the greatest dot product is always
achieved in the case where they lie along the same direction.
In Lorentzian geometry, the situation is different. Let b and c be
timelike vectors, so that they represent possible world-lines. Then
the relation a = b+c suggests the existence of two observers who
take two different paths from one event to another. A goes by a
direct route while B takes a detour. The magnitude of each time-
like vector represents the time elapsed on a clock carried by the
observer moving along that vector. The triangle equality is now
reversed, becoming [b + c[ > [b[ + [c[. The difference from the
5
In mathematics, a group is dened as a binary operation that has an identity,
inverses, and associativity. For example, addition of integers is a group. In the
present context, the members of the group are not numbers but the transforma-
tions applied to the Euclidean plane. The group operation on transformations
T1 and T2 consists of nding the transformation that results from doing one and
then the other, i.e., composition of functions.
6
The discontinuous transformations of spatial reection and time reversal are
not included in the denition of the Poincare group, although they do preserve
inner products. General relativity has symmetry under spatial reection (called
P for parity), time reversal (T), and charge inversion (C), but the standard
model of particle physics is only invariant under the composition of all three,
CPT, not under any of these symmetries individually.
Euclidean case arises because inner products are no longer nec-
essarily maximized if vectors are in the same direction. E.g., for
two lightlike vectors, b
i
c
j
vanishes entirely if b and c are paral-
lel. For timelike vectors, parallelism actually minimizes the inner
product rather than maximizing it.
7
In his 1872 inaugural address at the University of Erlangen, Felix
Klein used the idea of groups of transformations to lay out a gen-
eral classication scheme, known as the Erlangen program, for all
the dierent types of geometry. Each geometry is described by the
group of transformations, called the principal group, that preserves
the truth of geometrical statements. Euclidean geometrys principal
group consists of the isometries combined with arbitrary changes of
scale, since there is nothing in Euclids axioms that singles out a
particular distance as a unit of measurement. In other words, the
principal group consists of the transformations that preserve simi-
larity, not just those that preserve congruence. Ane geometrys
principal group is the transformations that preserve parallelism; it
includes shear transformations, and there is therefore no invariant
notion of angular measure or congruence. Unlike Euclidean and
ane geometry, elliptic geometry does not have scale invariance.
This is because there is a particular unit of distance that has special
status; as we saw in example 3 on page 79, a being living in an el-
liptic plane can determine, by entirely intrinsic methods, a distance
scale R, which we can interpret in the hemispherical model as the
radius of the sphere. General relativity breaks this symmetry even
more severely. Not only is there a scale associated with curvature,
but the scale is dierent from one point in space to another.
3.4.4 Einsteins carousel
Non-Euclidean geometry observed in the rotating frame
The following example was historically important, because Ein-
stein used it to convince himself that general relativity should be
described by non-Euclidean geometry.
8
Its interpretation is also
fairly subtle, and the early relativists had some trouble with it.
Suppose that observer A is on a spinning carousel while observer
B stands on the ground. B says that A is accelerating, but by the
equivalence principle A can say that she is at rest in a gravitational
7
Proof: Let b and c be parallel and timelike, and directed forward in time.
Adopt a frame of reference in which every spatial component of each vector
vanishes. This entails no loss of generality, since inner products are invariant
under such a transformation. Since the time-ordering is also preserved under
transformations in the Poincare group, but each is still directed forward in time,
not backward. Now let b and c be pulled away from parallelism, like opening
a pair of scissors in the x t plane. This reduces bt ct , while causing bx cx to
become negative. Both eects increase the inner product.
8
The example is described in Einsteins paper The Foundation of the General
Theory of Relativity. An excerpt, which includes the example, is given on
p. 255.
d / Observer A, rotating with
the carousel, measures an
azimuthal distance with a ruler.
eld, while B is free-falling out from under her. B measures the
radius and circumference of the carousel, and nds that their ratio
is 2. A carries out similar measurements, but when she puts her
meter-stick in the azimuthal direction it becomes Lorentz-contracted
by the factor = (1
2
r
2
)
1/2
, so she nds that the ratio is greater
than 2. In As coordinates, the spatial geometry is non-Euclidean,
and the metric diers from the Euclidean one found in example 6
on page 85.
Observer A feels a force that B considers to be ctitious, but
that, by the equivalence principle, A can say is a perfectly real
gravitational force. According to A, an observer like B is free-falling
away from the center of the disk under the inuence of this gravita-
tional eld. A also observes that the spatial geometry of the carousel
is non-Euclidean. Therefore it seems reasonable to conjecture that
gravity can be described by non-Euclidean geometry, rather than as
a physical force in the Newtonian sense.
At this point, you know as much about this example as Einstein
did in 1912, when he began using it as the seed from which general
relativity sprouted, collaborating with his old schoolmate, mathe-
matician Marcel Grossmann, who knew about dierential geometry.
The remainder of subsection 3.4.4, which you may want to skip on a
rst reading, goes into more detail on the interpretation and math-
ematical description of the rotating frame of reference. Even more
detailed treatments are given by Grn
9
and Dieks.
10
.
Ehrenfests paradox
Ehrenfest
11
described the following paradox. Suppose that ob-
server B, in the lab frame, measures the radius of the disk to be r
when the disk is at rest, and r
when the disk is spinning. B can

also measure the corresponding circumferences C and C
. Because
B is in an inertial frame, the spatial geometry does not appear non-
Euclidean according to measurements carried out with his meter
sticks, and therefore the Euclidean relations C = 2r and C
= 2r
both hold. The radial lines are perpendicular to their own motion,
and they therefore have no length contraction, r = r
, implying
C = C
. The outer edge of the disk, however, is everywhere tan-

gent to its own direction of motion, so it is Lorentz contracted, and
therefore C
< C. The resolution of the paradox is that it rests on

the incorrect assumption that a rigid disk can be made to rotate.
If a perfectly rigid disk was initially not rotating, one would have
to distort it in order to set it into rotation, because once it was
rotating its outer edge would no longer have a length equal to 2
times its radius. Therefore if the disk is perfectly rigid, it can never
9
Relativistic description of a rotating disk, Am. J. Phys. 43 (1975) 869
10
Space, Time, and Coordinates in a Rotating World, http://www.phys.uu.
nl/igg/dieks
11
P. Ehrenfest, Gleichformige Rotation starrer Korper und Relativit atstheorie,
Z. Phys. 10 (1909) 918
e / Einstein and Ehrenfest.
be rotated. As discussed on page 50, relativity does not allow the
existence of innitely rigid or innitely strong materials. If it did,
then one could violate causality. If a perfectly rigid disk existed, vi-
brations in the disk would propagate at innite velocity, so tapping
the disk with a hammer in one place would result in the transmis-
sion of information at v > c to other parts of the disk, and then
there would exist frames of reference in which the information was
received before it was transmitted. The same applies if the hammer
tap is used to impart rotational motion to the disk.
Self-check: What if we build the disk by assembling the building
materials so that they are already rotating properly before they are
joined together?
The metric in the rotating frame
What if we try to get around these problems by applying torque
uniformly all over the disk, so that the rotation starts smoothly and
simultaneously everywhere? We then run into issues identical to the
ones raised by Bells spaceship paradox (p. 51). In fact, Ehrenfests
paradox is nothing more than Bells paradox wrapped around into
a circle. The same question of time synchronization comes up.
To spell this out mathematically, lets nd the metric according
to observer A by applying the change of coordinates
= t.
First we take the Euclidean metric of example 6 on page 85 and
rewrite it as a (globally) Lorentzian metric in spacetime for observer
B,
[1] ds
2
= dt
2
dr
2
r
2
d
2
.
Applying the transformation into As coordinates, we nd
[2] ds
2
= (1
2
r
2
)dt
2
dr
2
r
2
d
2
2r
2
d
dt .
Recognizing r as the velocity of one frame relative to another, and
(1
2
r
2
)
1/2
as , we see that we do have a relativistic time dilation
eect in the dt
2
term. But the dr
2
and d
2
terms look Euclidean.
Why dont we see any Lorentz contraction of the length scale in the
azimuthal direction?
The answer is that coordinates in general relativity are arbi-
trary, and just because we can write down a certain set of coordi-
nates, that doesnt mean they have any special physical interpreta-
tion. The coordinates (t, r,
) do not correspond physically to the

quantities that A would measure with clocks and meter-sticks. The
tip-o is the d
dt cross-term. Suppose that A sends two cars driv-

ing around the circumference of the carousel, one clockwise and one
counterclockwise, from the same point. If (t, r,
) coordinates cor-
responded to clock and meter-stick measurements, then we would
expect that when the cars met up again on the far side of the disk,
their dashboards would show equal values of the arc length r
on
their odometers and equal proper times ds on their clocks. But this
is not the case, because the sign of the d
dt term is opposite for the

two world-lines. The same eect occurs if we send beams of light
in both directions around the disk, and this is the Sagnac eect (p.
58).
This is a symptom of the fact that the coordinate t is not prop-
erly synchronized between dierent places on the disk. We already
know that we should not expect to be able to nd a universal time
coordinate that will match up with every clock, regardless of the
clocks state of motion. Suppose we set ourselves a more modest
goal. Can we nd a universal time coordinate that will match up
with every clock, provided that the clock is at rest relative to the
rotating disk?
The spatial metric and synchronization of clocks
A trick for improving the situation is to eliminate the d
dt cross-
term by completing the square in the metric [2]. The result is
ds
2
= (1
2
r
2
)
_
dt +
r
2
1
2
r
2
d
_
2
dr
2
r
2
1
2
r
2
d
2
.
The interpretation of the quantity in square brackets is as follows.
Suppose that two observers situate themselves on the edge of the
disk, separated by an innitesimal angle d
. They then synchronize

their clocks by exchanging light pulses. The time of ight, measured
in the lab frame, for each light pulse is the solution of the equation
ds
2
= 0, and the only dierence between the clockwise result dt
1
and the counterclockwise one dt
2
arises from the sign of d
. The
quantity in square brackets is the same in both cases, so the amount
by which the clocks must be adjusted is dt = (dt
2
dt
1
)/2, or
dt =
r
2
1
2
r
2
d
.
Substituting this into the metric, we are left with the purely spatial
metric
[3] ds
2
= dr
2
r
2
1
2
r
2
d
2
.
The factor of (1
2
r
2
)
1
=
2
in the d
2
term is simply the
expected Lorentz-contraction factor. In other words, the circumfer-
ence is, as expected, greater than 2r by a factor of .
Does the metric [3] represent the same non-Euclidean spatial
geometry that A, rotating with the disk, would determine by meter-
stick measurements? Yes and no. It can be interpreted as the
one that A would determine by radar measurements. That is, if
A measures a round-trip travel time dt for a light signal between
points separated by coordinate distances dr and d
, then A can say

that the spatial separation is dt/2, and such measurements will be
described correctly by [3]. Physical meter-sticks, however, present
some problems. Meter-sticks rotating with the disk are subject to
Coriolis and centrifugal forces, and this problem cant be avoided
simply by making the meter-sticks innitely rigid, because innitely
rigid objects are forbidden by relativity. In fact, these forces will in-
evitably be strong enough to destroy any meter stick that is brought
out to r = 1/, where the speed of the disk becomes equal to the
speed of light.
It might appear that we could now dene a global coordinate
T = t +
r
2
1
2
r
2
,
interpreted as a time coordinate that was synchronized in a con-
sistent way for all points on the disk. The trouble with this inter-
pretation becomes evident when we imagine driving a car around
the circumference of the disk, at a speed slow enough so that there
is negligible time dilation of the cars dashboard clock relative to
the clocks tied to the disk. Once the car gets back to its original
position,
has increased by 2, so it is no longer possible for the

cars clock to be synchronized with the clocks tied to the disk. We
conclude that it is not possible to synchronize clocks in a rotating
frame of reference; if we try to do it, we will inevitably have to have
a discontinuity somewhere. This problem is present even locally, as
demonstrated by the possibility of measuring the Sagnac eect with
apparatus that is small compared to the disk. The only reason we
were able to get away with time synchronization in order to establish
the metric [3] is that all the physical manifestations of the impossi-
bility of synchronization, e.g., the Sagnac aect, are proportional to
the area of the region in which synchronization is attempted. Since
we were only synchronizing two nearby points, the area enclosed by
the light rays was zero.
GPS Example: 12
As a practical example, the GPS system is designed mainly to
allow people to nd their positions relative to the rotating surface
of the earth (although it can also be used by space vehicles). That
is, they are interested in their (r ,
, ) coordinates. The frame of

reference dened by these coordinates is referred to as ECEF, for
Earth-Centered, Earth-Fixed.
The system requires synchronization of the atomic clocks carried
aboard the satellites, and this synchronization also needs to be
extended to the (less accurate) clocks built into the receiver units.
It is impossible to carry out such a synchronization globally in the
rotating frame in order to create coordinates (T, r ,
, ). If we
tried, it would result in discontinuities (see problem 2, p. 100).
Instead, the GPS system handles clock synchronization in coor-
dinates (t , r ,
, ), as in equation [2]. These are known as the

Earth-Centered Inertial (ECI) coordinates. The t coordinate in
this system is not the one that users at neighboring points on
the earths surface would establish if they carried out clock syn-
chronization using electromagnetic signals. It is simply the time
coordinate of the nonrotating frame of reference tied to the earths
center. Conceptually, we can imagine this time coordinate as one
that is established by sending out an electromagnetic tick-tock
signal from the earths center, with each satellite correcting the
phase of the signal based on the propagation time inferred from
its own r . In reality, this is accomplished by communication with a
master control station in Colorado Springs, which communicates
with the satellites via relays at Kwajalein, Ascension Island, Diego
Garcia, and Cape Canaveral.
Impossibility of rigid rotation, even with external forces
The determination of the spatial metric with rulers at rest rel-
ative to the disk is appealing because of its conceptual simplicity
compared to complicated procedures involving radar, and this was
presumably why Einstein presented the concept using ruler measure-
ments in his 1916 paper laying out the general theory of relativity.
12
In an eort to recover this simplicity, we could propose using exter-
nal forces to compensate for the centrifugal and Coriolis forces to
which the rulers would be subjected, causing them to stay straight
and maintain their correct lengths. Something of this kind is car-
ried out with the large mirrors of some telescopes, which have active
systems that compensate for gravitational deections and other ef-
fects. The rst issue to worry about is that one would need some
way to monitor a rulers length and straightness. The monitoring
system would presumably be based on measurements with beams
of light, in which case the physical rulers themselves would become
superuous.
In addition, we would need to be able to manipulate the rulers in
order to place them where we wanted them, and these manipulations
would include angular accelerations. If such a thing was possible,
then it would also amount to a loophole in the resolution of the
Ehrenfest paradox. Could Ehrenfests rotating disk be accelerated
and decelerated with help from external forces, which would keep it
from contorting into a potato chip? The problem we run into with
such a strategy is one of clock synchronization. When it was time to
impart an angular acceleration to the disk, all of the control systems
would have to be activated simultaneously. But we have already
seen that global clock synchronization cannot be realized for an
object with nite area, and therefore there is a logical contradiction
in this proposal. This makes it impossible to apply rigid angular
acceleration to the disk, but not necessarily the rulers, which could
12
The paper is reproduced in the back of the book, and the relevant part is
on p. 257.
a / Einsteins hole argument.
in theory be one-dimensional.
3.5 The metric in general relativity
So far weve considered a variety of examples in which the metric
is predetermined. This is not the case in general relativity. For
example, Einstein published general relativity in 1915, but it was
not until 1916 that Schwarzschild found the metric for a spherical,
gravitating body such as the sun or the earth.
When masses are present, nding the metric is analogous to
nding the electric eld made by charges, but the interpretation is
more dicult. In the electromagnetic case, the eld is found on
a preexisting background of space and time. In general relativity,
there is no preexisting geometry of spacetime. The metric tells us
how to nd distances in terms of our coordinates, but the coordi-
nates themselves are completely arbitrary. So what does the metric
even mean? This was an issue that caused Einstein great distress
and confusion, and at one point, in 1914, it even led him to pub-
lish an incorrect, dead-end theory of gravity in which he abandoned
coordinate-independence.
With the benet of hindsight, we can consider these issues in
terms of the general description of measurements in relativity given
on page 80:
1. We can tell whether events and world-lines are incident.
2. We can do measurements in local Lorentz frames.
3.5.1 The hole argument
The main factor that led Einstein to his false start is known as
the hole argument. Suppose that we know about the distribution of
matter throughout all of spacetime, including a particular region of
nite size the hole which contains no matter. By analogy
with other classical eld theories, such as electromagnetism, we ex-
pect that the metric will be a solution to some kind of dierential
equation, in which matter acts as the source term. We nd a metric
g(x) that solves the eld equations for this set of sources, where x is
some set of coordinates. Now if the eld equations are coordinate-
independent, we can introduce a new set of coordinates x
, which is
identical to x outside the hole, but diers from it on the inside. If
we reexpress the metric in terms of these new coordinates as g
(x
),
then we are guaranteed that g
(x
) is also a solution. But further-

more, we can substitute x for x
, and g
(x) will still be a solution.

For outside the hole there is no dierence between the primed and
unprimed quantities, and inside the hole there is no mass distribu-
tion that has to match the metrics behavior on a point-by-point
basis.
Section 3.5 The metric in general relativity 95
b / A paradox? Planet A has
no equatorial bulge, but B does.
What cause produces this effect?
Einstein reasoned that the cause
couldnt be Bs rotation, because
each planet rotates relative to the
other.
We conclude that in any coordinate-invariant theory, it is impos-
sible to uniquely determine the metric inside such a hole. Einstein
initially decided that this was unacceptable, because it showed a
lack of determinism; in a classical theory such as general relativity,
we ought to be able to predict the evolution of the elds, and it
would seem that there is no way to predict the metric inside the
hole. He eventually realized that this was an incorrect interpreta-
tion. The only type of global observation that general relativity lets
us do is measurements of the incidence of world-lines. Relabeling all
the points inside the hole doesnt change any of the incidence rela-
tions. For example, if two test particles sent into the region collide
at a point x inside the hole, then changing the points name to x
doesnt change the observable fact that they collided.

3.5.2 A Machian paradox
Another type of argument that made Einstein suer is also re-
solved by a correct understanding of measurements, this time the
use of measurements in local Lorentz frames. The earth is in hy-
drostatic equilibrium, and its equator bulges due to its rotation.
Suppose that the universe was empty except for two planets, each
rotating about the line connecting their centers.
13
Since there are
no stars or other external points of reference, the inhabitants of each
planet have no external reference points against which to judge their
rotation or lack of rotation. They can only determine their rotation,
Einstein said, relative to the other planet. Now suppose that one
planet has an equatorial bulge and the other doesnt. This seems to
violate determinism, since there is no cause that could produce the
diering eect. The people on either planet can consider themselves
as rotating and the other planet as stationary, or they can describe
the situation the other way around. Einstein believed that this ar-
gument proved that there could be no dierence between the sizes
of the two planets equatorial bulges.
The aw in Einsteins argument was that measurements in local
Lorentz frames do allow one to make a distinction between rotation
and a lack of rotation. For example, suppose that scientists on
planet A notice that their world has no equatorial bulge, while planet
B has one. They send a space probe with a clock to B, let it stay
on Bs surface for a few years, and then order it to return. When
the clock is back in the lab, they compare it with another clock that
stayed in the lab on planet A, and they nd that less time has elapsed
according to the one that spent time on Bs surface. They conclude
that planet B is rotating more quickly than planet A, and that the
motion of Bs surface was the cause of the observed time dilation.
This resolution of the apparent paradox depends specically on the
Lorentzian form of the local geometry of spacetime; it is not available
13
The example is described in Einsteins paper The Foundation of the General
Theory of Relativity. An excerpt, which includes the example, is given on
p. 255.
in, e.g., Cartans curved-spacetime description of Newtonian gravity
(see page 35).
Einsteins original, incorrect use of this example sprang from his
interest in the ideas of the physicist and philosopher Ernst Mach.
Mach had a somewhat ill-dened idea that since motion is only a
well-dened notion when we speak of one object moving relative
to another object, the inertia of an object must be caused by the
inuence of all the other matter in the universe. Einstein referred
to this as Machs principle. Einsteins false starts in constructing
general relativity were frequently related to his attempts to make his
theory too Machian. Section 7.3 on p. 221 discusses an alternative,
more Machian theory of gravity proposed by Brans and Dicke in
1951.
3.6 Interpretation of coordinate independence
This section discusses some of the issues that arise in the inter-
pretation of coordinate independence. It can be skipped on a rst
reading.
3.6.1 Is coordinate independence obvious?
One often hears statements like the following from relativists:
Coordinate independence isnt really a physical principle. Its
merely an obvious statement about the relationship between math-
ematics and the physical universe. Obviously the universe doesnt
come equipped with coordinates. We impose those coordinates on
it, and the way in which we do so can never be dictated by nature.
The impressionable reader who is tempted to say, Ah, yes, that
is obvious, should consider that it was far from obvious to New-
ton (Absolute, true and mathematical time, of itself, and from its
own nature ows equably without regard to anything external . . . ),
nor was it obvious to Einstein. Levi-Civita nudged Einstein the di-
rection of coordinate independence in 1912. Einstein tried hard to
make a coordinate-independent theory, but for reasons described in
section 3.5.1 (p. 95), he convinced himself that that was a dead
end. In 1914-15 he published theories that were not coordinate-
independent, which you will hear relativists describe as obvious
dead ends because they lack any geometrical interpretation. It seems
to me that it takes a highly rened intuition to regard as intuitively
obvious an issue that Einstein struggled with like Jacob wrestling
with Elohim.
3.6.2 Is coordinate independence trivial?
It has also been alleged that coordinate independence is trivial.
To gauge the justice of this complaint, lets distinguish between two
reasons for caring about coordinate independence:
1. Coordinate independence tells us that when we solve problems,
Section 3.6 Interpretation of coordinate independence 97
we should avoid writing down any equations in notation that
isnt manifestly intrinsic, and avoid interpreting those equa-
tions as if the coordinates had intrinsic meaning. Violating
this advice doesnt guarantee that youve made a mistake, but
it makes it much harder to tell whether or not you have.
2. Coordinate independence can be used as a criterion for judging
whether a particular theory is likely to be successful.
Nobody questions the rst justication. The second is a little trick-
ier. Laying out the general theory systematically in a 1916 paper,
14
Einstein wrote The general laws of nature are to be expressed by
equations which hold good for all the systems of coordinates, that is,
are covariant with respect to any substitutions whatever (generally
covariant). In other words, he was explaining why, with hindsight,
his 1914-1915 coordinate-dependent theory had to be a dead end.
The only trouble with this is that Einsteins way of posing the
criterion didnt quite hit the nail on the head mathematically. As
Hilbert famously remarked, Every boy in the streets of Gottingen
understands more about four-dimensional geometry than Einstein.
Yet, in spite of that, Einstein did the work and not the mathemati-
cians. What Einstein had in mind was that a theory like Newtonian
mechanics not only lacks coordinate independence, but would also
be impossible to put into a coordinate-independent form without
making it look hopelessly complicated and ugly, like putting lipstick
on a pig. But Kretschmann showed in 1917 that any theory could
be put in coordinate independent form, and Cartan demonstrated in
1923 that this could be done for Newtonian mechanics in a way that
didnt come out particularly ugly. Physicists today are more apt to
pose the distinction in terms of background independence (mean-
ing that a theory should not be phrased in terms of an assumed ge-
ometrical background) or lack of a prior geometry (meaning that
the curvature of spacetime should come from the solution of eld
equations rather than being imposed by at). But these concepts as
well have resisted precise mathematical formulation.
15
My feeling
is that this general idea of coordinate independence or background
independence is like the equivalence principle: a crucial conceptual
principle that doesnt lose its importance just because we cant put
it in a mathematical box with a ribbon and a bow. For example,
string theorists take it as a serious criticism of their theory that it is
not manifestly background independent, and one of their goals is to
show that it has a background independence that just isnt obvious
on the surface.
14
see p. 259
15
Giulini, Some remarks on the notions of general covariance and background
independence, arxiv.org/abs/gr-qc/0603087v1
a / Since magnetic eld lines
can never intersect, a magnetic
eld pattern contains coordinate-
independent information in the
form of the knotting of the lines.
This gure shows the mag-
netic eld pattern of the star
SU Aurigae, as measured by
Zeeman-Doppler imaging (Petit
at al.). White lines represent
magnetic eld lines that close
upon themselves in the immedi-
ate vicinity of the star; blue lines
are those that extend out into the
interstellar medium.
3.6.3 Coordinate independence as a choice of gauge
It is instructive to consider coordinate independence from the
point of view of a eld theory. Newtonian gravity can be described
in three equivalent ways: as a gravitational eld g, as a gravitational
potential , or as a set of gravitational eld lines. The eld lines are
never incident on one another, and locally the eld satises Poissons
equation.
The electromagnetic eld has polarization properties dierent
from those of the gravitational eld, so we describe it using either
the two elds (E, B), a pair of potentials,
16
or two sets of eld
lines. There are similar incidence conditions and local eld equations
(Maxwells equations).
Gravitational elds in relativity have polarization properties un-
known to Newton, but the situation is qualitatively similar to the
two foregoing cases. Now consider the analogy between electromag-
netism and relativity. In electromagnetism, it is the elds that are
directly observable, so we expect the potentials to have some extrin-
sic properties. We can, for example, redene our electrical ground,
+ C, without any observable consequences. As discussed in
more detail in section 5.6.1 on page 142, it is even possible to modify
the electromagnetic potentials in an entirely arbitrary and nonlinear
way that changes from point to point in spacetime. This is called a
gauge transformation. In relativity, the gauge transformations are
the smooth coordinate transformations. These gauge transforma-
tions distort the eld lines without making them cut through one
another.
16
There is the familiar electrical potential , measured in volts, but also a
vector potential A, which you may or may not have encountered. Briey, the
electric eld is given not by but by A/t, while the magnetic eld
is the curl of A. This is introduced at greater length in section 4.2.5 on page
113.
Section 3.6 Interpretation of coordinate independence 99
Problems
1 Consider a spacetime that is locally exactly like the stan-
dard Lorentzian spacetime described in ch. 2, but that has a global
structure diering in the following way from the one we have im-
plicitly assumed. This spacetime has global property G: Let two
material particles have world-lines that coincide at event A, with
some nonzero relative velocity; then there may be some event B in
the future light-cone of A at which the particles world-lines coincide
again. This sounds like a description of something that we would
expect to happen in curved spacetime, but lets see whether that
is necessary. We want to know whether this violates the at-space
properties L1-L5 on page 272, if those properties are taken as local.
(a) Demonstrate that it does not violate them, by using a model in
which space wraps around like a cylinder.
(b) Now consider the possibility of interpreting L1-L5 as global state-
ments. Do spacetimes with property G always violate L3 if L3 is
taken globally? Solution, p. 262
2 Example 12 on page 93 discusses the discontinuity that would
result if one attempted to dene a time coordinate for the GPS sys-
tem that was synchronized globally according to observers in the
rotating frame, in the sense that neighboring observers could verify
the synchronization by exchanging electromagnetic signals. Calcu-
late this discontinuity at the equator, and estimate the resulting
error in position that would be experienced by GPS users.
3 Resolve the following paradox.
Equation [3] on page 92 claims to give the metric obtained by an ob-
server on the surface of a rotating disk. This metric is shown to lead
to a non-Euclidean value for the ratio of the circumference of a circle
to its radius, so the metric is clearly non-Euclidean. Therefore a lo-
cal observer should be able to detect violations of the Pythagorean
theorem.
And yet this metric was originally derived by a series of changes
of coordinates, starting from the Euclidean metric in polar coor-
dinates, as derived in example 6 on page 85. Section 3.3 (p. 79)
argued that the intrinsic measurements available in relativity are
not capable of detecting an arbitrary smooth, one-to-one change of
coordinates. This contradicts our earlier conclusion that there are
locally detectable violations of the Pythagorean theorem.
Solution, p. 263
4 This problem deals with properties of the metric [3] on page
92. (a) A pulse of collimated light is emitted from the center of
the disk in a certain direction. Does the spatial track of the pulse
form a geodesic of this metric? (b) Characterize the behavior of the
geodesics near r = 1/. (c) An observer at rest with respect to the
surface of the disk proposes to verify the non-Euclidean nature of
the metric by doing local tests in which right triangles are formed
out of laser beams, and violations of the Pythagorean theorem are
detected. Will this work? Solution, p. 263
Problems 101
Chapter 4
Tensors
We now have enough machinery to be able to calculate quite a bit of
interesting physics, and to be sure that the results are actually mean-
ingful in a relativistic context. The strategy is to identify relativistic
quantities that behave as Lorentz scalars and Lorentz vectors, and
then combine them in various ways. The notion of a tensor has been
introduced on page 84. A Lorentz scalar is a tensor of rank 0, and
a Lorentz vector is a rank-1 tensor.
4.1 Lorentz scalars
A Lorentz scalar is a quantity that remains invariant under both spa-
tial rotations and Lorentz boosts. Mass is a Lorentz scalar.
1
Elec-
tric charge is also a Lorentz scalar, as demonstrated to extremely
high precision by experiments measuring the electrical neutrality of
atoms and molecules to a relative precision of better than 10
20
; the
electron in a hydrogen atom has typically velocities of about 1/100,
and those in heavier elements such as uranium are highly relativis-
tic, so any violation of Lorentz invariance would give the atoms a
nonvanishing net electric charge.
The time measured by a clock traveling along a particular world-
line from one event to another is something that all observers will
agree upon; they will simply note the mismatch with their own
clocks. It is therefore a Lorentz scalar. This clock-time as measured
by a clock attached to the moving body in question is often referred
to as proper time, proper being used here in the somewhat archaic
sense of own or self, as in The Vatican does not lie within Italy
proper. Proper time, which we notate , can only be dened for
timelike world-lines, since a lightlike or spacelike world-line isnt
possible for a material clock.
More generally, when we express a metric as ds
2
= . . ., the
quantity ds is a Lorentz scalar. In the special case of a timelike
world-line, ds and d are the same thing. (In books that use a
+ ++ metric, one has ds = d.)
Even more generally, ane parameters, which exist independent
of any metric at all, are scalars. As a trivial example, if is a
particular objects proper time, then is a valid ane parameter,
1
Some older books dene mass as transforming according to m m, which
can be made to give a self-consistent theory, but is ugly.
103
but so is 2 +7. Less trivially, a photons proper time is always zero,
but one can still dene an ane parameter along its trajectory. We
will need such an ane parameter, for example, in section 6.2.7,
page 182, when we calculate the deection of light rays by the sun,
one of the early classic experimental tests of general relativity.
Another example of a Lorentz scalar is the pressure of a perfect
uid, which is often assumed as a description of matter in cosmo-
logical models.
Innitesimals and the clock postulate Example: 1
At the beginning of chapter 3, I motivated the use of innitesimals
as useful tools for doing differential geometry in curved space-
time. Even in the context of special relativity, however, innitesi-
mals can be useful. One way of expressing the proper time accu-
mulated on a moving clock is
s =
_
ds
=
_
_
g
i j
dx
i
dx
j
=
_

1
_
dx
dt
_
2
_
dy
dt
_
2
_
dz
dt
_
2
dt ,
which only contains an explicit dependence on the clocks veloc-
ity, not its acceleration. This is an example of the clock postulate
referred to in the remark at the end of homework problem 1 on
page 68. Note that the clock postulate only applies in the limit of
a small clock. This is represented in the above equation by the
use of innitesimal quantities like dx.
4.2 Four-vectors
4.2.1 The velocity and acceleration four-vectors
Our basic Lorentz vector is the spacetime displacement dx
i
. Any
other quantity that has the same behavior as dx
i
under rotations
and boosts is also a valid Lorentz vector. Consider a particle moving
through space, as described in a Lorentz frame. Since the particle
may be subject to nongravitational forces, the Lorentz frame can-
not be made to coincide (except perhaps momentarily) with the
particles rest frame. Dividing the innitesimal displacement by an
innitesimal proper time interval, we have the four-velocity vector
v
i
= dx
i
/d, whose components in a Lorentz coordinate system
are (, v
1
, v
2
, v
3
), where v
, = 1, 2, 3, is the ordinary three-

component velocity vector as dened in classical mechanics. The
four-velocitys squared magnitude v
i
v
i
is always exactly 1, even if
the particle is not moving at the speed of light.
When we hear something referred to as a vector, we usually
take this is a statement that it not only transforms as a vector, but
104 Chapter 4 Tensors
also that it adds as a vector. But we have already seen in section
2.3.1 on page 51 that even collinear velocities in relativity do not
add linearly; therefore they clearly cannot add linearly when dressed
in the clothing of four-vectors. Weve also seen in section 2.5.3 that
the combination of non-collinear boosts is noncommutative, and is
generally equivalent to a boost plus a spatial rotation; this is also
not consistent with linear addition of four vectors. At the risk of
beating a dead horse, a four-velocitys squared magnitude is always
1, and this is not consistent with being able to add four-velocity
vectors.
A zero velocity vector? Example: 2
Suppose an object has a certain four-velocity v
i
in a certain
frame of reference. Can we transform into a different frame in
which the object is at rest, and its four-velocity is zero?
No. In general, the Lorentz transformation preserves the mag-
nitude of vectors, so it can never transform a vector with a zero
magnitude into one with zero magnitude. We can transform into a
frame in which the object is at rest, but an object at rest does not
have a vanishing four-velocity. It has a four-velocity of (1, 0, 0, 0).
The four-acceleration is found by taking a second derivative with
respect to proper time. Its squared magnitude is only approximately
equal to minus the squared magnitude of the classical acceleration
three-vector, in the limit of small velocities.
Constant acceleration Example: 3
Suppose a spaceship moves so that the acceleration is judged
to be the constant value a by an observer on board. Find the
motion x(t ) as measured by an observer in an inertial frame.
Let stand for the ships proper time, and let dots indicate
derivatives with respect to . The ships velocity has magnitude
1, so
t
2
x
2
= 1 .
An observer who is instantaneously at rest with respect to the
ship judges is to have a four-acceleration (0, a, 0, 0) (because the
low-velocity limit applies). The observer in the (t , x) frame agrees
on the magnitude of this vector, so
t
2
x
2
= a
2
.
The solution of these differential equations is t =
1
a
sinha,
x =
1
a
cosha, and eliminating gives
x =
1
a
__
1 + a
2
t
2
1
_
.
As t approaches innity, dx/dt approaches the speed of light.
Section 4.2 Four-vectors 105
4.2.2 The momentum four-vector
In subsection 4.2.1, we obtained the velocity four-vector. Mul-
tiplying by the particles mass, we have the four-momentum p
i
=
mv
i
, which in Lorentz coordinates is (m, mv
1
, mv
2
, mv
3
). The
spacelike components look like the classical momentum vector mul-
tiplied by a factor of , the interpretation being that to an observer
in this frame, the moving particles inertia is increased relative to
its classical value. This is why particle accelerators are so big and
expensive. As the particle approaches the speed of light, diverges,
so greater and greater forces are needed in order to produce the
same acceleration.
The momentum four-vector has locked within it the reason for
Einsteins famous E = mc
2
, which in our relativistic units becomes
simply E = m. To see why, consider the experimentally measured
inertia of a physical object made out of atoms. The subatomic par-
ticles are all moving, and many of the velocities, e.g., the velocities
of the electrons, are quite relativistic. This has the eect of in-
creasing the experimentally determined inertial mass, by a factor
of 1 averaged over all the particles. The same must be true for
the gravitational mass, based on the equivalence principle as veried
by Eotvos experiments. If the object is heated, the velocities will
increase on the average, resulting in a further increase in its mass.
Thus, a certain amount of heat energy is equivalent to a certain
amount of mass. But if heat energy contributes to mass, then the
same must be true for other forms of energy. For example, sup-
pose that heating leads to a chemical reaction, which converts some
heat into electromagnetic binding energy. If one joule of binding
energy did not convert to the same amount of mass as one joule of
heat, then this would allow the object to spontaneously change its
own mass, and then by conservation of momentum it would have
to spontaneously change its own velocity, which would clearly vio-
late the principle of relativity. We conclude that mass and energy
are equivalent, both inertially and gravitationally. In relativity, nei-
ther is separately conserved; the conserved quantity is their sum,
referred to as the mass-energy, E. The timelike component of the
four-momentum, m, is interpreted as the mass-energy of the par-
ticle, consisting of its mass m plus its kinetic energy m( 1). An
alternative derivation, by Einstein, is given in example 11 on page
111.
Gravitational redshifts Example: 4
Since a photons energy E is equivalent to a certain gravitational
mass m, photons that rise or fall in a gravitational eld must
lose or gain energy, and this should be observed as a redshift
or blueshift in the frequency. We expect the change in gravita-
tional potential energy to be E, giving a corresponding op-
posite change in the photons energy, so that E/E = . In
metric units, this becomes E/E = /c
2
, and in the eld near
the Earths surface we have E/E = gh/c
2
. This is the same
result that was found in section 1.5.7 based only on the equiva-
lence principle, and veried experimentally by Pound and Rebka
as described in section 1.5.8.
Since the momentum four-vector was obtained from the magni-
tude-1 velocity four-vector through multiplication by m, its squared
magnitude p
i
p
i
is equal to the square of the particles mass. Writing
p for the magnitude of the momentum three-vector, and E for the
mass-energy, we nd the useful relation E
2
p
2
= m
2
.
A common source of confusion for beginners in relativity is the
distinction between quantities that are conserved and quantities that
are the same in all frames. There is nothing relativistic about this
distinction. Before Einstein, physicists already knew that observers
in dierent frames of reference would agree on the mass of a particle.
That is, m was known to be frame-invariant. They also knew that
energy was conserved. But just because energy was conserved, that
didnt mean that it had to be the same for observers in all frames
of reference. The kinetic energy of the chair youre sitting in is
millions of joules in a frame of reference tied to the axis of the
earth. In relativity, m is frame-invariant (i.e., a Lorentz scalar), but
the conserved quantity is the momentum four-vector, which is not
frame-invariant.
Applying E
2
p
2
= m
2
to the special case of a massless particle,
we have [p[ = E, which demonstrates, for example, that a beam of
light exerts pressure when it is absorbed or reected by a surface.
2
A massless particle must also travel at exactly the speed of light,
since [p[ E requires mv m; conversely, a massive particle
always has [v[ < 1.
Constraints on polarization Example: 5
We observe that electromagnetic waves are always polarized
transversely, never longitudinally. Such a constraint can only ap-
ply to a wave that propagates at c. If it applied to a wave that
propagated at less than c, we could move into a frame of refer-
ence in which the wave was at rest. In this frame, all directions in
space would be equivalent, and there would be no way to decide
which directions of polarization should be permitted. For a wave
that propagates at c, there is no frame in which the wave is at rest
(see p. 81).
The Dirac sea Example: 6
A great deal of physics can be derived from the T.H. Whites
principle that whatever is not forbidden in compulsory orig-
inally intended for ants but applied to particles by Gell-Mann.
In quantum mechanics, any process that is not forbidden by a
conservation law is supposed to occur. The relativistic relation
2
cf. p. 28
E =
_
p
2
+ m
2
has two roots, a positive one and a negative one.
The positive-energy and negative-energy states are separated by
a no-mans land of width 2m, so no continuous classical process
can lead from one side to the other. But quantum-mechanically, if
an electron exists with energy E = +
_
p
2
+ m
2
, it should be able to
make a quantum leap into a state with E =
_
p
2
+ m
2
, emitting
the energy difference of 2E in the form of photons. Why doesnt
this happen? One explanation is that the states with E < 0 are all
already occupied. This is the Dirac sea, which we now interpret
as being full of electrons. A vacancy in the sea manifests itself as
an antielectron.
Massive neutrinos Example: 7
Neutrinos were long thought to be massless, but are now believed
to have masses in the eV range. If they had been massless, they
would always have had to propagate at the speed of light. Al-
though they are now thought to have mass, that mass is six or-
ders of magnitude less than the MeV energy scale of the nuclear
reactions in which they are produced, so all neutrinos observed
in experiments are moving at velocities very close to the speed of
light.
No radioactive decay of massless particles Example: 8
A photon cannot decay into an electron and a positron, e
+
+
e
, in the absence of a charged particle to interact with. To see

this, consider the process in the frame of reference in which the
electron-positron pair has zero total momentum. In this frame,
the photon must have had zero (three-)momentum, but a photon
with zero momentum must have zero energy as well. This means
that relativistic four -momentum has been violated: the timelike
component of the four-momentum is the mass-energy, and it has
increased from 0 in the initial state to at least 2mc
2
in the nal
state.
To demonstrate the consistency of the theory, we can arrive at the
same conclusion by a different method. Whenever a particle has
a small mass (small compared to its energy, say), it must travel
at close to c. It must therefore have a very large time dilation,
and will take a very long time to undergo radioactive decay. In
the limit as the mass approaches zero, the time required for the
decay approaches innity. Another way of saying this is that the
rate of radioactive decay must be xed in terms of proper time,
but there is no such thing as proper time for a massless particle.
Thus is is not only this specic process that is forbidden, but any
radioactive decay process involving a massless particle.
Massive photons Example: 9
Continuing in the same vein as example 7, we can consider the
possibility that the photon has some nonvanishing mass. A 2003
experiment by Luo et al.
3
has placed a limit of about 10
54
kg
on this mass. This is incredibly small, but suppose that future ex-
perimental work using improved techniques shows that the mass
is less than this, but actually nonzero. A naive reaction to this
scenario is that it would shake relativity to its core, since relativity
is based upon the assumption that the speed of light is a con-
stant, whereas for a massive particle it need not be constant. But
this is a misinterpretation of the role of c in relativity. As should
be clear from the approach taken in section 2.2, c is primarily a
geometrical property of spacetime, not a property of light.
In reality, such a discovery would be more of a problem for parti-
cle physicists than for relativists, as we can see by the following
sketch of an argument. Imagine two charged particles, at rest,
interacting via an electrical attraction. Quantum mechanics de-
scribes this as an exchange of photons. Since the particles are
at rest, there is no source of energy, so where do we get the
energy to make the photons? The Heisenberg uncertainty prin-
ciple, Et h, allows us to steal this energy, provided that we
give it back within a time t . This time limit imposes a limit on
the distance the photons can travel, but by using photons of low
enough energy, we can make this distance limit as large as we
like, and there is therefore no limit on the range of the force. But
suppose that the photon has a mass. Then there is a minimum
mass-energy mc
2
required in order to create a photon, the max-
imum time is h/mc
2
, and the maximum range is h/mc. Rening
these crude arguments a little, one nds that exchange of zero-
mass particles gives a force that goes like 1/r
2
, while a nonzero
mass results in e
r
/r
2
, where
1
= /mc. For the photon,
the best current mass limit corresponds to
1
10
11
m, so the
deviation from 1/r
2
would be difcult to measure in earthbound
experiments.
Now Gausss law is a specic characteristic of 1/r
2
elds. It would
be violated slightly if photons had mass. We would have to modify
Maxwells equations, and it turns out
4
that the necessary change
to Gausss law would be of the form E = (. . .) (. . .)
2
,
where is the electrical potential, and (. . . ) indicates factors
that depend on the choice of units. This tells us that , which
in classical electromagnetism can only be measured in terms of
differences between different points in space, can now be mea-
sured in absolute terms. Gauge symmetry has been broken. But
gauge symmetry is indispensible in creating well-behaved rela-
3
Luo et al., New Experimental Limit on the Photon Rest Mass with a Ro-
tating Torsion Balance, Phys. Rev. Lett. 90 (2003) 081801. The interpretation
of such experiments is dicult, and this paper attracted a series of comments. A
weaker but more universally accepted bound is 8 10
52
kg, Davis, Goldhaber,
and Nieto, Phys. Rev. Lett. 35 (1975) 1402.
4
Goldhaber and Nieto, Terrestrial and Extraterrestrial Limits on The Pho-
ton Mass, Rev. Mod. Phys. 43 (1971) 277
a / Example 10.
tivistic eld theories, and this is the reason that, in general, parti-
cle physicists have a hard time with forces arising from the ex-
change of massive particles. The hypothetical Higgs particle,
which may be observed at the Large Hadron Collider in the near
future, is essentially a mechanismfor wriggling out of this difculty
in the case of the massive W and Z particles that are responsible
for the weak nuclear force; the mechanism cannot, however, be
extended to allow a massive photon.
Dust and radiation in cosmological models Example: 10
In cosmological models, one needs an equation of state that re-
lates the pressure P to the mass-energy density . The pressure
is a Lorentz scalar. The mass-energy density is not (since mass-
energy is just the timelike component of a particular vector), but
in a coordinate system without any net ow of mass, we can ap-
proximate it as one.
The early universe was dominated by radiation. A photon in a
box contributes a pressure on each wall that is proportional to
[p
[, where is a spacelike index. In thermal equilibrium, each of

these three degrees of freedom carries an equal amount of en-
ergy, and since momentum and energy are equal for a massless
particle, the average momentum along each axis is equal to
1
3
E.
The resulting equation of state is P =
1
3
. As the universe ex-
panded, the wavelengths of the photons expanded in proportion
to the stretching of the space they occupied, resulting in a
1
,
where a is a distance scale describing the universes intrinsic cur-
vature at a xed time. Since the number density of photons is
diluted in proportion to a
3
, and the mass per photon varies as
a
1
, both and P vary as a
4
.
Cosmologists refer to noninteracting, nonrelativistic materials as
dust, which could mean many things, including hydrogen gas,
actual dust, stars, galaxies, and some forms of dark matter. For
dust, the momentum is negligible compared to the mass-energy,
so the equation of state is P = 0, regardless of . The mass-
energy density is dominated simply by the mass of the dust, so
there is no red-shift scaling of the a
1
type. The mass-energy
density scales as a
3
. Since this is a less steep dependence on
a than the a
4
, there was a point, about a thousand years after
the Big Bang, when matter began to dominate over radiation. At
this point, the rate of expansion of the universe made a transition
to a qualitatively different behavior resulting from the change in
the equation of state.
In the present era, the universes equation of state is dominated
by neither dust nor radiation but by the cosmological constant
(see page 201). Figure a shows the evolution of the size of the
universe for the three different regimes. Some of the simpler
cases are derived in sections 7.2.4 and 7.2.5, starting on page
212.
4.2.3 The frequency four-vector and the relativistic Doppler
shift
Frequency is to time as the wavenumber k = 1/ is to space,
so when treating waves relativistically it is natural to conjecture
that there is a four-frequency f
a
made by assembling (f, k), which
behaves as a Lorentz vector. This is correct, since we already know
that
a
transforms as a covariant vector, and for a scalar wave of
the form A = A
o
exp [2if
a
x
a
] the partial derivative operator is
identical to multiplication by 2f
a
.
As an application, consider the relativistic Doppler shift of a light
wave. For simpicity, lets restrict ourselves to one spatial dimension.
For a light wave, f = k, so the frequency vector in 1+1 dimensions
is simply (f, f). Putting this through a Lorentz transformation, we
nd
f
= (1 +v)f =
_
1 +v
1 v
f ,
where the second form displays more clearly the symmetic form
of the relativistic relationship, such that interchanging the roles of
source and observer is equivalent to ipping the sign of v. That is,
the relativistic version only depends on the relative motion of the
source and the observer, whereas the Newtonian one also depends
on the sources motion relative to the medium (i.e., relative to the
preferred frame in which the waves have the right velocity). In
Newtonian mechanics, we have f
= f/(1 v) for a moving source.

Relativistically, there is also a time dilation of the oscillation of the
source, providing an additional factor of 1/.
This analysis is extended to 3+1 dimensions in problem 10.
Einsteins derivation of E = mc
2
Example: 11
On page 106, we showed that the celebrated E = mc
2
follows di-
rectly from the form of the Lorentz transformation. An alternative
derivation was given by Einstein in one of his classic 1905 pa-
pers laying out the theory of special relativity; the paper is short,
and is reproduced in English translation on page 253 of this book.
Having laid the groundwork of four-vectors and relativistic Doppler
shifts, we can give an even shorter version of Einsteins argument.
The discussion is also streamlined by restricting the discussion to
1+1 dimensions and by invoking photons.
Suppose that a lantern, at rest in the lab frame, is oating weight-
lessly in outer space, and simultaneously emits two pulses of
light in opposite directions, each with energy E/2 and frequency
f . By symmetry, the momentum of the pulses cancels, and the
lantern remains at rest. An observer in motion at velocity v rel-
ative to the lab sees the frequencies of the beams shifted to
f
= (1 v)f . The effect on the energies of the beams can

be found purely classically, by transforming the electric and mag-
b / Magnetism is a purely rel-
ativistic effect.
netic elds to the moving frame, but as a shortcut we can ap-
ply the quantum-mechanical relation E
ph
= hf for the energies of
the photons making up the beams. The result is that the mov-
ing observer nds the total energy of the beams to be not E but
(E/2)(1 + v) + (E/2)(1 v) = E.
Both observers agree that the lantern had to use up some of the
energy stored in its fuel in order to make the two pulses. But
the moving observer says that in addition to this energy E, there
was a further energy E( 1). Where could this energy have
come from? It must have come from the kinetic energy of the
lantern. The lanterns velocity remained constant throughout the
experiment, so this decrease in kinetic energy seen by the moving
observer must have come from a decrease in the lanterns inertial
mass hence the title of Einsteins paper, Does the inertia of a
body depend upon its energy content?
To gure out how much mass the lantern has lost, we have to
decide how we can even dene mass in this new context. In
Newtonian mechanics, we had K = (1/2)mv
2
, and by the corre-
spondence principle this must still hold in the low-velocity limit.
Expanding E( 1) in a Taylor series, we nd that it equals
E(v
2
/2) + . . ., and in the low-velocity limit this must be the same
as K = (1/2)mv
2
, so m = E. Reinserting factors of c to get
back to nonrelativistic units, we have E = mc
2
.
4.2.4 A non-example: electric and magnetic elds
It is fairly easy to see that the electric and magnetic elds cannot
be the spacelike parts of two four-vectors. Consider the arrangement
shown in gure b/1. We have two innite trains of moving charges
superimposed on the same line, and a single charge alongside the
line. Even though the line charges formed by the two trains are
moving in opposite directions, their currents dont cancel. A nega-
tive charge moving to the left makes a current that goes to the right,
so in frame 1, the total current is twice that contributed by either
line charge.
In frame 1 the charge densities of the two line charges cancel out,
and the electric eld experienced by the lone charge is therefore zero.
Frame 2 shows what wed see if we were observing all this from a
frame of reference moving along with the lone charge. Both line
charges are in motion in both frames of reference, but in frame
1, the line charges were moving at equal speeds, so their Lorentz
contractions were equal, and their charge densities canceled out. In
frame 2, however, their speeds are unequal. The positive charges
are moving more slowly than in frame 1, so in frame 2 they are
less contracted. The negative charges are moving more quickly, so
their contraction is greater now. Since the charge densities dont
cancel, there is an electric eld in frame 2, which points into the
wire, attracting the lone charge.
c / The charged particle fol-
lows a trajectory that extremizes
_
f
b
dx
b
compared to other
nearby trajectories. Relativis-
tically, the trajectory should be
understood as a world-line in
3+1-dimensional spacetime.
We appear to have a logical contradiction here, because an ob-
server in frame 2 predicts that the charge will collide with the wire,
whereas in frame 1 it looks as though it should move with constant
velocity parallel to the wire. Experiments show that the charge does
collide with the wire, so to maintain the Lorentz-invariance of elec-
tromagnetism, we are forced to invent a new kind of interaction, one
between moving charges and other moving charges, which causes the
acceleration in frame 2. This is the magnetic interaction, and if we
hadnt known about it already, we would have been forced to invent
it. That is, magnetism is a purely relativistic eect. The reason a
relativistic eect can be strong enough to stick a magnet to a re-
frigerator is that it breaks the delicate cancellation of the extremely
large electrical interactions between electrically neutral objects.
Although the example shows that the electric and magnetic elds
do transform when we change from one frame to another, it is easy
to show that they do not transform as the spacelike parts of a rela-
tivistic four-vector. This is because transformation between frames
1 and 2 is along the axis parallel to the wire, but it aects the com-
ponents of the elds perpendicular to the wire. The electromagnetic
eld actually transforms as a rank-2 tensor.
4.2.5 The electromagnetic potential four-vector
An electromagnetic quantity that does transform as a four-vector
is the potential. On page 99, I mentioned the fact, which may or
may not already be familiar to you, that whereas the Newtonian
gravitational elds polarization properties allow it to be described
using a single scalar potential or a single vector eld g = ,
the pair of electromagnetic elds (E, B) needs a pair of potentials,
and A. Its easy to see that cant be a Lorentz scalar. Elec-
tric charge q is a scalar, so if were a scalar as well, then the
product q would be a scalar. But this is equal to the energy of
the charged particle, which is only the timelike component of the
energy-momentum four-vector, and therefore not a Lorentz scaler
itself. This is a contradiction, so is not a scalar.
To see how to t into relativity, consider the nonrelativistic
quantum mechanical relation q = hf for a charged particle in a
potential . Since f is the timelike component of a four-vector in
relativity, we need to be the timelike component of some four
vector, A
b
. For the spacelike part of this four-vector, lets write A,
so that A
b
= (, A). We can see by the following argument that
this mysterious A must have something to do with the magnetic
eld.
Consider the example of gure c from a quantum-mechanical
point of view. The charged particle q has wave properties, but lets
say that it can be well approximated in this example as following a
specic trajectory. This is like the ray approximation to wave optics.
A light ray in classical optics follows Fermats principle, also known
d / The magnetic eld (top)
and vector potential (bottom) of
a solenoid. The lower diagram is
in the plane cutting through the
waist of the solenoid, as indicated
by the dashed line in the upper
diagram. For an innite solenoid,
the magnetic eld is uniform
on the inside and zero on the
outside, while the vector potential
is proportional to r on the inside
and to 1/r on the outside.
as the principle of least time, which states that the rays path from
point A to point B is one that extremizes the optical path length
(essentially the number of oscillations). The reason for this is that
the ray approximation is only an approximation. The ray actually
has some width, which we can visualize as a bundle of neighboring
trajectories. Only if the trajectory follows Fermats principle will
the interference among the neighboring paths be constructive. The
classical optical path length is found by integrating k ds, where k
is the wavenumber. To make this relativistic, we need to use the
frequency four-vector to form f
b
dx
b
, which can also be expressed
as f
b
v
b
d = (f k v)d. If the charge is at rest and there
are no magnetic elds, then the quantity in parentheses is f =
E/h = (q/h). The correct relativistic generalization is clearly
f
b
= (q/h)A
b
.
Since A
b
s spacelike part, A, results in the velocity-dependent
eects, we conclude that A is a kind of potential that relates to the
magnetic eld, in the same way that the potential relates to the
electric eld. A is known as the vector potential, and the relation
between the potentials and the elds is
E =
A
t
B = A .
An excellent discussion of the vector potential from a purely classical
point of view is given in the classic Feynman Lectures.
5
Figure d
shows an example.
4.3 The tensor transformation laws
We may wish to represent a vector in more than one coordinate
system, and to convert back and forth between the two represen-
tations. In general relativity, the transformation of the coordinates
need not be linear, as in the Lorentz transformations; it can be any
smooth, one-to-one function. For simplicity, however, we start by
considering the one-dimensional case, and by assuming the coordi-
nates are related in an ane manner, x
= ax
+ b. The addition
of the constant b is merely a change in the choice of origin, so it
has no eect on the components of the vector, but the dilation by
the factor a gives a change in scale, which results in v
= av
for a
contravariant vector. In the special case where v is an innitesimal
displacement, this is consistent with the result found by implicit dif-
ferentiation of the coordinate transformation. For a contravariant
vector, v
=
1
a
v
. Generalizing to more than one dimension, and to

5
The Feynman Lectures on Physics, Feynman, Leighton, and Sands, Addison
Wesley Longman, 1970
a possibly nonlinear transformation, we have
v
= v
[1]
v
= v
. [2]
Note the inversion of the partial derivative in one equation compared
to the other.
Self-check: Recall that the gauge transformations allowed in gen-
eral relativity are not just any coordinate transformations; they
must be (1) smooth and (2) one-to-one. Relate both of these re-
quirements to the features of the vector transformation laws above.
In equation [2], appears as a subscript on the left side of the
equation, but as a superscript on the right. This would appear
to violate our rules of notation, but the interpretation here is that
in expressions of the form /x
i
and /x
i
, the superscripts and
subscripts should be understood as being turned upside-down. Sim-
ilarly, [1] appears to have the implied sum over written ungram-
matically, with both s appearing as superscripts. Normally we
only have implied sums in which the index appears once as a super-
script and once as a subscript. With our new rule for interpreting
indices on the bottom of derivatives, the implied sum is seen to be
written correctly. This rule is similar to the one for analyzing the
units of derivatives written in Leibniz notation, with, e.g., d
2
x/dt
2
having units of meters per second squared.
A quantity v that transforms according to [1] or [2] is referred
to as a rank-1 tensor, which is the same thing as a vector.
The identity transformation Example: 12
In the case of the identity transformation x
= x
, equation [1]
clearly gives v
= v, since all the mixed partial derivatives x
/x
with ,= are zero, and all the derivatives for = equal 1.

In equation [2], it is tempting to write
x
=
1
x
(wrong!) ,
but this would give innite results for the mixed terms! Only in the
case of functions of a single variable is it possible to ip deriva-
tives in this way; it doesnt work for partial derivatives. To evalu-
ate these partial derivatives, we have to invert the transformation
(which in this example is trivial to accomplish) and then take the
partial derivatives.
The metric is a rank-2 tensor, and transforms analogously:
g
= g
Section 4.3 The tensor transformation laws 115

Self-check: Write the similar expressions for g
, g
, and g
,
which are entirely determined by the grammatical rules for writing
superscripts and subscripts. Interpret the case of a rank-0 tensor.
An accelerated coordinate system? Example: 13
Lets see the effect on Lorentzian metric g of the transformation
t
= t x
= x +
1
2
at
2
.
The inverse transformation is
t = t
x = x
1
2
at
2
.
The tensor transformation law gives
g
t
= 1 (at
)
2
g
x
= 1
g
t
= at
.
Clearly something bad happens at at
= 1, when the relative

velocity surpasses the speed of light: the t
component of the
metric vanishes and then reverses its sign. This would be physi-
cally unreasonable if we viewed this as a transformation from ob-
server As Lorentzian frame into the accelerating reference frame
of observer B aboard a spaceship who feels a constant acceler-
ation. Several things prevent such an interpretation: (1) B cannot
exceed the speed of light. (2) Even before B gets to the speed
of light, the coordinate t
cannot correspond to Bs proper time,

which is dilated. (3) Due to time dilation, A and B do not agree
on the rate at which B is accelerating. If B measures her own
acceleration to be a
, A will judge it to be a < a
, with a 0 as B
approaches the speed of light. There is nothing invalid about the
coordinate system (t
, x
), but neither does it have any physically

interesting interpretation.
Physically meaningful constant acceleration Example: 14
To make a more physically meaningful version of example 13, we
need to use the result of example 3 on page 105. The some-
what messy derivation of the coordinate transformation is given
by Semay.
6
The result is
t
=
_
x +
1
a
_
sinhat
x
=
_
x +
1
a
_
coshat
Applying the tensor transformation law gives (problem 6, page
126):
g
t
= (1 + ax
)
2
g
x
= 1
6
arxiv.org/abs/physics/0601179
Unlike the result of example 13, this one never misbehaves.
The closely related topic of a uniform gravitational eld in general
relativity is considered in problem 5 on page 162.
Accurate timing signals Example: 15
The relation between the potential A and the elds E and B given
on page 114 can be written in manifestly covariant form as F
i j
=
[i
A
j ]
, where F, called the electromagnetic tensor, is an antisym-
metric rank-two tensor whose six independent components corre-
spond in a certain way with the components of the E and B three-
vectors. If F vanishes completely at a certain point in spacetime,
then the linear form of the tensor transformation laws guarantees
that it will vanish in all coordinate systems, not just one. The GPS
system takes advantage of this fact in the transmission of timing
signals from the satellites to the users. The electromagnetic wave
is modulated so that the bits it transmits are represented by phase
reversals of the wave. At these phase reversals, F vanishes, and
this vanishing holds true regardless of the motion of the users
unit or its position in the earths gravitational eld.
4.4 Experimental tests
4.4.1 Universality of tensor behavior
The techniques developed in this chapter allow us to make a vari-
ety of new predictions that can be tested by experiment. In general,
the mathematical treatment of all observables in relativity as ten-
sors means that all observables must obey the same transformation
laws. This is an extremely strict statement, because it requires that
a wide variety of physical systems show identical behavior. For ex-
ample, we already mentioned on page 59 the 2007 Gravity Probe
B experiment (discussed in detail on pages 139 and 175), in which
four gyroscopes aboard a satellite were observed to precess due to
special- and general-relativistic eects. The gyroscopes were compli-
cated electromechanical systems, but the predicted precession was
entirely independent of these complications. We argued that if two
dierent types of gyroscopes displayed dierent behaviors, then the
resulting discrepancy would allow us to map out some mysterious
vector eld. This eld would be a built-in characteristic of space-
time (not produced by any physical objects nearby), and since all
observables in general relativity are supposed to be tensors, the eld
would have to transform as a tensor. Lets say that this tensor was
of rank 1. Since the tensor transformation law is linear, a nonzero
tensor can never be transformed into a vanishing tensor in another
coordinate system. But by the equivalence principle, any special,
local property of spacetime can be made to vanish by transforming
into a free-falling frame of reference, in which the spacetime is has a
generic Lorentzian geometry. The mysterious new eld should there-
Section 4.4 Experimental tests 117
fore vanish in such a frame. This is a contradiction, so we conclude
that dierent types of gyroscopes cannot dier in their behavior.
This is an example of a new way of stating the equivalence prin-
ciple: there is no way to associate a preferred tensor eld with space-
time.
7
4.4.2 Speed of light differing from c
In a Lorentz invariant theory, we interpret c as a property of
the underlying spacetime, not of the particles that inhabit it. One
way in which Lorentz invariance could be violated would be if dif-
ferent types of particles had dierent maximum velocities. In 1997,
Coleman and Glashow suggested a sensitive test for such an eect.
8
Assuming Lorentz invariance, a photon cannot decay into an
electron and a positron, e
+
+ e
(example 8, page 108).

Suppose, however, that material particles have a maximum speed
c
m
= 1, while photons have a maximum speed c
p
> 1. Then the pho-
tons momentum four-vector, (E, E/c
p
) is timelike, so a frame does
exist in which its three-momentum is zero. The detection of cosmic-
ray gammas from distant sources with energies on the order of 10
TeV puts an upper limit on the decay rate, implying c
p
1 10
15
.
An even more stringent limit can be put on the possibility of
c
p
< 1. When a charged particle moves through a medium at a speed
higher than the speed of light in the medium, Cerenkov radiation
results. If c
p
is less than 1, then Cerenkov radiation could be emitted
by high-energy charged particles in a vacuum, and the particles
would rapidly lose energy. The observation of cosmic-ray protons
with energies 10
8
TeV requires c
p
1 10
23
.
4.4.3 Degenerate matter
The straightforward properties of the momentum four-vector
have surprisingly far-reaching implications for matter subject to ex-
treme pressure, as in a star that uses up all its fuel for nuclear fusion
and collapses. These implications were initially considered too ex-
otic to be taken seriously by astronomers. For historical perspective,
consider that in 1916, when Einstein published the theory of gen-
eral relativity, the Milky Way was believed to constitute the entire
universe; the spiral nebulae were believed to be inside it, rather
than being similar objects exterior to it. The only types of stars
whose structure was understood even vaguely were those that were
roughly analogous to our own sun. (It was not known that nuclear
fusion was their source of energy.) The term white dwarf had not
been invented, and neutron stars were unknown.
An ordinary, smallish star such as our own sun has enough hy-
7
This statement of the equivalence principle, along with the others we have
encountered, is summarized in the back of the book on page 273.
8
arxiv.org/abs/hep-ph/9703240
drogen to sustain fusion reactions for billions of years, maintaining
an equilibrium between its gravity and the pressure of its gases.
When the hydrogen is used up, it has to begin fusing heavier el-
ements. This leads to a period of relatively rapid uctuations in
structure. Nuclear fusion proceeds up until the formation of ele-
ments as heavy as oxygen (Z = 8), but the temperatures are not
high enough to overcome the strong electrical repulsion of these nu-
clei to create even heavier ones. Some matter is blown o, but nally
nuclear reactions cease and the star collapses under the pull of its
own gravity.
To understand what happens in such a collapse, we have to un-
derstand the behavior of gases under very high pressures. In gen-
eral, a surface area A within a gas is subject to collisions in a time t
from the n particles occupying the volume V = Avt, where v is the
typical velocity of the particles. The resulting pressure is given by
P npv/V , where p is the typical momentum.
Nondegenerate gas: In an ordinary gas such as air, the parti-
cles are nonrelativistic, so v = p/m, and the thermal energy
per particle is p
2
/2m kT, so the pressure is P nkT/V .
Nonrelativistic, degenerate gas: When a fermionic gas is sub-
ject to extreme pressure, the dominant eects creating pres-
sure are quantum-mechanical. Because of the Pauli exclu-
sion principle, the volume available to each particle is V/n,
so its wavelength is no more than (V/n)
1/3
, leading to
p = h/ h(n/V )
1/3
. If the speeds of the particles are still
nonrelativistic, then v = p/m still holds, so the pressure be-
comes P (h
2
/m)(n/V )
5/3
.
Relativistic, degenerate gas: If the compression is strong enough
to cause highly relativistic motion for the particles, then v c,
and the result is P hc(n/V )
4/3
.
As a star with the mass of our sun collapses, it reaches a point
at which the electrons begin to behave as a degenerate gas, and
the collapse stops. The resulting object is called a white dwarf. A
white dwarf should be an extremely compact body, about the size
of the Earth. Because of its small surface area, it should emit very
little light. In 1910, before the theoretical predictions had been
made, Russell, Pickering, and Fleming discovered that 40 Eridani B
had these characteristics. Russell recalled: I knew enough about
it, even in these paleozoic days, to realize at once that there was
an extreme inconsistency between what we would then have called
possible values of the surface brightness and density. I must have
shown that I was not only puzzled but crestfallen, at this exception
to what looked like a very pretty rule of stellar characteristics; but
Pickering smiled upon me, and said: It is just these exceptions
a / Subrahmanyan Chan-
drasekhar (1910-1995)
that lead to an advance in our knowledge, and so the white dwarfs
entered the realm of study!
S. Chandrasekhar showed in that 1930s that there was an upper
limit to the mass of a white dwarf. We will recapitulate his calcu-
lation briey in condensed order-of-magnitude form. The pressure
at the core of the star is P gr GM
2
/r
4
, where M is the total
mass of the star. The star contains roughly equal numbers of neu-
trons, protons, and electrons, so M = Knm, where m is the mass of
the electron, n is the number of electrons, and K 4000. For stars
near the limit, the electrons are relativistic. Setting the pressure at
the core equal to the degeneracy pressure of a relativistic gas, we
nd that the Chandrasekhar limit is (hc/G)
3/2
(Km)
2
= 6M
.
A less sloppy calculation gives something more like 1.4M
. The self-
consistency of this solution is investigated in homework problem 13
on page 127.
What happens to a star whose mass is above the Chandrasekhar
limit? As nuclear fusion reactions icker out, the core of the star be-
comes a white dwarf, but once fusion ceases completely this cannot
be an equilibrium state. Now consider the nuclear reactions
n p +e
+
p +e
n + ,
which happen due to the weak nuclear force. The rst of these re-
leases 0.8 MeV, and has a half-life of 14 minutes. This explains
why free neutrons are not observed in signicant numbers in our
universe, e.g., in cosmic rays. The second reaction requires an input
of 0.8 MeV of energy, so a free hydrogen atom is stable. The white
dwarf contains fairly heavy nuclei, not individual protons, but sim-
ilar considerations would seem to apply. A nucleus can absorb an
electron and convert a proton into a neutron, and in this context the
process is called electron capture. Ordinarily this process will only
occur if the nucleus is neutron-decient; once it reaches a neutron-
to-proton ratio that optimizes its binding energy, neutron capture
cannot proceed without a source of energy to make the reaction go.
In the environment of a white dwarf, however, there is such a source.
The annihilation of an electron opens up a hole in the Fermi sea.
There is now an state into which another electron is allowed to drop
without violating the exclusion principle, and the eect cascades
upward. In a star with a mass above the Chandrasekhar limit, this
process runs to completion, with every proton being converted into a
neutron. The result is a neutron star, which is essentially an atomic
nucleus (with Z = 0) with the mass of a star!
Observational evidence for the existence of neutron stars came
in 1967 with the detection by Bell and Hewish at Cambridge of a
mysterious radio signal with a period of 1.3373011 seconds. The sig-
nals observability was synchronized with the rotation of the earth
relative to the stars, rather than with legal clock time or the earths
rotation relative to the sun. This led to the conclusion that its origin
was in space rather than on earth, and Bell and Hewish originally
dubbed it LGM-1 for little green men. The discovery of a second
signal, from a dierent direction in the sky, convinced them that it
was not actually an articial signal being generated by aliens. Bell
published the observation as an appendix to her PhD thesis, and
it was soon interpreted as a signal from a neutron star. Neutron
stars can be highly magnetized, and because of this magnetization
they may emit a directional beam of electromagnetic radiation that
sweeps across the sky once per rotational period the lighthouse
eect. If the earth lies in the plane of the beam, a periodic signal
can be detected, and the star is referred to as a pulsar. It is fairly
easy to see that the short period of rotation makes it dicult to
explain a pulsar as any kind of less exotic rotating object. In the
approximation of Newtonian mechanics, a spherical body of density
, rotating with a period T =
_
3/G, has zero apparent gravity
at its equator, since gravity is just strong enough to accelerate an
object so that it follows a circular trajectory above a xed point on
the surface (problem 12). In reality, astronomical bodies of plane-
tary size and greater are held together by their own gravity, so we
have T 1/
G for any body that does not y apart spontaneously

due to its own rotation. In the case of the Bell-Hewish pulsar, this
implies 10
10
kg/m
3
, which is far larger than the density of nor-
mal matter, and also 10-100 times greater than the typical density
of a white dwarf near the Chandrasekhar limit.
An upper limit on the mass of a neutron star can be found in a
manner entirely analogous to the calculation of the Chandrasekhar
limit. The only dierence is that the mass of a neutron is much
greater than the mass of an electron, and the neutrons are the only
particles present, so there is no factor of K. Assuming the more
precise result of 1.4M
for the Chandrasekhar limit rather than

our sloppy one, and ignoring the interaction of the neutrons via the
strong nuclear force, we can infer an upper limit on the mass of a
neutron star:
1.4M
_
Km
e
m
n
_
2
5M
The theoretical uncertainties in such an estimate are fairly large.

Tolman, Oppenheimer, and Volko originally estimated it in 1939
as 0.7M
, whereas modern estimates are more in the range of 1.5

to 3M
. These are signicantly lower than our crude estimate of

5M
, mainly because the attractive nature of the strong nuclear

force tends to push the star toward collapse. Unambiguous results
are presently impossible because of uncertainties in extrapolating
the behavior of the strong force from the regime of ordinary nuclei,
where it has been relatively well parametrized, into the exotic envi-
ronment of a neutron star, where the density is signicantly dierent
and no protons are present.
a / Two spacelike surfaces.
b / We dene a boundary
around a region whose charge
we want to measure.
c / This boundary cuts the
sphere into equal parts.
d / The circumference of this
boundary is small, but the region
it contains is large.
For stars with masses above the Tolman-Oppenheimer-Volko
limit, theoretical predictions become even more speculative. A va-
riety of bizarre objects has been proposed, including gravastars,
fuzzballs, black stars, quark stars, Q-balls. It seems likely, how-
ever, both on theoretical and observational grounds, that objects
with masses of about 3 to 20 solar masses end up as black holes; see
section 6.3.3.
4.5 Conservation laws
4.5.1 No general conservation laws
Some of the rst tensors we discussed were mass and charge, both
rank-0 tensors, and the rank-1 momentum tensor, which contains
both the classical energy and the classical momentum. Physicists
originally decided that mass, charge, energy, and momentum were
interesting because these things were found to be conserved. This
makes it natural to ask how conservation laws can be formulated
in relativity. Were used to stating conservation laws casually in
terms of the amount of something in the whole universe, e.g., that
classically the total amount of mass in the universe stays constant.
Relativity does allow us to make physical models of the universe as
a whole, so it seems as though we ought to be able to talk about
conservation laws in relativity.
We cant.
First, how do we dene stays constant? Simultaneity isnt
well-dened, so we cant just take two snapshots, call them initial
and nal, and compare the total amount of, say, electric charge in
each snapshot. This diculty isnt insurmountable. As in gure
a, we can arbitrarily pick out three-dimensional spacelike surfaces
one initial and one nal and integrate the charge over each
one. A law of conservation of charge would say that no matter what
spacelike surface we picked, the total charge on each would be the
same.
Next theres the issue that the integral might diverge, especially
if the universe was spatially innite. For now, lets assume a spa-
tially nite universe. For simplicity, lets assume that it has the
topology of a three-sphere (see section 7.2 for reassurance that this
isnt physically unreasonable), and we can visualize it as a two-
sphere.
In the case of the momentum four-vector, what coordinate sys-
tem would we express it in? In general, we do not even expect to
be able to dene a smooth, well-behaved coordinate system that
covers the entire universe, and even if we did, it would not make
sense to add a vector expressed in that coordinate system at point
A to another vector from point B; the best we could do would be
to parallel-transport the vectors to one point and then add them,
e / A relativistic jet.
but parallel transport is path-dependent. (Similar issues occur with
angular momentum.) For this reason, lets restrict ourselves to the
easier case of a scalar, such as electric charge.
But now were in real trouble. How would we go about actually
measuring the total electric charge of the universe? The only way to
do it is to measure electric elds, and then apply Gausss law. This
requires us to single out some surface that we can integrate the ux
over, as in b. This would really be a two-dimensional surface on the
three-sphere, but we can visualize it as a one-dimensional surface
a closed curve on the two-sphere. But now suppose this curve is
a great circle, c. If we measure a nonvanishing total ux across it,
how do we know where the charge is? It could be on either side.
You might protest that this is an articial example. In reality
wouldnt we conduct a charge survey by chopping up space into
small regions, as in b? Figure d shows that this wont work either.
The boundary curve has a small circumference, but it contains a
very large interior region.
The conclusion is that conservation laws only make sense in rel-
ativity under very special circumstances. We do not have anything
like over-arching principles of conservation.
As an example, section 6.2.6 shows how to dene conserved quan-
tities, which behave like energy and momentum, for the motion of
a test particle in a particular metric that has a certain symmetry.
4.5.2 Conservation of angular momentum and frame
dragging
Another special case where conservation laws work is that if
the spacetime were studying gets very at at large distances from a
small system were studying, then we can dene a far-away boundary
that surrounds the system, measure the ux through that boundary,
and nd the systems charge. For such asymptotically at space-
times, we can also get around the problems that crop up with con-
served vectors, such as momentum. If the spacetime far away is
nearly at, then parallel transport loses its path-dependence, so we
can unambiguously dene a notion of parallel-transporting all the
contributions to the ux to one arbitrarily chosen point P and then
adding them. Asymptotic atness also allows us to dene an ap-
proximate notion of a global Lorentz frame, so that the choice of P
doesnt matter.
As an example, gure e shows a jet of matter being ejected from
the galaxy M87 at ultrarelativistic elds. The blue color of the jet in
the visible-light image comes from synchrotron radiation, which is
the electromagnetic radiation emitted by relativistic charged parti-
cles accelerated by a magnetic eld. The jet is believed to be coming
from a supermassive black hole at the center of M87. The emission
of the jet in a particular direction suggests that the black hole is not
Section 4.5 Conservation laws 123
spherically symmetric. It seems to have a particular axis associated
with it. How can this be? Our suns spherical symmetry is broken
by the existence of externally observable features such as sunspots
and the equatorial bulge, but the only information we can get about
a black hole comes from its external gravitational (and possibly elec-
tromagnetic) elds. It appears that something about the spacetime
metric surrounding this black hole breaks spherical symmetry, but
preserves symmetry about some preferred axis. What aspect of the
initial conditions in the formation of the hole could have determined
such an axis? The most likely candidate is the angular momentum.
We are thus led to suspect that black holes can possess angular mo-
mentum, that angular momentum preserves information about their
formation, and that angular momentum is externally detectable via
its eect on the spacetime metric.
What would the form of such a metric be? Spherical coordinates
in at spacetime give a metric like this:
ds
2
= dt
2
dr
2
r
2
d
2
r
2
sin
2
d
2
.
Well see in chapter 6 that for a non-rotating black hole, the metric
is of the form
ds
2
= (. . .)dt
2
(. . .)dr
2
r
2
d
2
r
2
sin
2
d
2
,
where (. . .) represents functions of r. In fact, there is nothing spe-
cial about the metric of a black hole, at least far away; the same
external metric applies to any spherically symmetric, non-rotating
body, such as the moon. Now what about the metric of a rotating
body? We expect it to have the following properties:
1. It has terms that are odd under time-reversal, corresponding
to reversal of the bodys angular momentum.
2. Similarly, it has terms that are odd under reversal of the dif-
ferential d of the azimuthal coordinate.
3. The metric should have axial symmetry, i.e., it should be in-
dependent of .
Restricting our attention to the equatorial plane = /2, the sim-
plest modication that has these three properties is to add a term
of the form
f(. . .)L d dt ,
where (. . .) again gives the r-dependence and L is a constant, inter-
preted as the angular momentum. A detailed treatment is beyond
the scope of this book, but solutions of this form to the relativistic
eld equations were found by New Zealand-born physicist Roy Kerr
in 1963 at the University of Texas at Austin.
f / Two light rays travel in the
earths equatorial plane from A to
B. Due to frame-dragging, the ray
moving with the earths rotation
is deected by a greater amount
than the one moving contrary to
it. As a result, the gure has an
asymmetric banana shape. Both
the deection and its asymmetry
are greatly exaggerated.
g / Gravity Probe B veried
the existence of frame-dragging.
The rotational axis of the gy-
roscope precesses in two
perpendicular planes due to the
two separate effects: geodetic
and frame-dragging.
The astrophysical modeling of observations like gure e is com-
plicated, but we can see in a simplied thought experiment that if
we want to determine the angular momentum of a rotating body
via its gravitational eld, it will be dicult unless we use a measur-
ing process that takes advantage of the asymptotic atness of the
space. For example, suppose we send two beams of light past the
earth, in its equatorial plane, one on each side, and measure their
deections, f. The deections will be dierent, because the sign of
ddt will be opposite for the two beams. But the entire notion of a
deection only makes sense if we have an asymptotically at back-
ground, as indicated by the dashed tangent lines. Also, if spacetime
were not asymptotically at in this example, then there might be
no unambiguous way to determine whether the asymmetry was due
to the earths rotation, to some external factor, or to some kind of
interaction between the earth and other bodies nearby.
It also turns out that a gyroscope in such a gravitational eld
precesses. This eect, called frame dragging, was predicted by Lense
and Thirring in 1918, and was nally veried experimentally in 2008
by analysis of data from the Gravity Probe B experiment, to a pre-
cision of about 15%. The experiment was arranged so that the rela-
tively strong geodetic eect (6.6 arc-seconds per year) and the much
weaker Lense-Thirring eect (.041 arc-sec/yr) produced precessions
in perpendicular directions. Again, the presence of an asymptoti-
cally at background was involved, because the probe measured the
orientations of its gyroscopes relative to the guide star IM Pegasi.
Section 4.5 Conservation laws 125
Problems
1 The Large Hadron Collider is designed to accelerate protons
to energies of 7 TeV. Find 1 v for such a proton.
Solution, p. 264
2 Prove that a photon in a vacuum cannot absorb a photon.
(This is the reason that the ability of materials to absorb gamma-
rays is strongly dependent on atomic number Z. The case of Z = 0
corresponds to the vacuum.)
3 (a) For an object moving in a circle at constant speed, the
dot product of the classical three-vectors v and a is zero. Give
an interpretation in terms of the work-kinetic energy theorem. (b)
In the case of relativistic four-vectors, v
i
a
i
= 0 for any world-line.
Give a similar interpretation. Hint: nd the rate of change of the
four-velocitys squared magnitude.
4 Starting from coordinates (t, x) having a Lorentzian metric
g, transform the metric tensor into reected coordinates (t
, x
) =
(t, x), and verify that g
is the same as g.
5 Starting from coordinates (t, x) having a Lorentzian metric g,
transform the metric tensor into Lorentz-boosted coordinates (t
, x
),
and verify that g
is the same as g.
6 Verify the transformation of the metric given in example 14
on page 116.
7 A skeptic claims that the Hafele-Keating experiment can only
be explained correctly by relativity in a frame in which the earths
axis is at rest. Prove mathematically that this is incorrect. Does it
matter whether the frame is inertial? Solution, p. 264
8 Assume the metric g = diag(+1, +1, +1). Which of the fol-
lowing correctly expresses the noncommutative property of ordinary
matrix multiplication?
A
j
i
B
jk
,= B
jk
A
j
i
A
j
i
B
jk
,= B
j
i
A
jk
9 Example 6 on page 107 introduced the Dirac sea, whose
existence is implied by the two roots of the relativistic relation
E =
_
p
2
+m
2
. Prove that a Lorentz boost will never trans-
form a positive-energy state into a negative-energy state.
Solution, p. 264
10 On page 111, we found the relativistic Doppler shift in 1+1
dimensions. Extend this to 3+1 dimensions, and check your result
against the one given by Einstein on page 253.
Solution, p. 264
11 For gamma-rays in the MeV range, the most frequent mode of
interaction with matter is Compton scattering, in which the photon
is scattered by an electron without being absorbed. Only part of
the gammas energy is deposited, and the amount is related to the
angle of scattering. Use conservation of four-momentum to show
that in the case of scattering at 180 degrees, the scattered photon
has energy E
= E/(1+2E/m), where m is the mass of the electron.

12 Derive the equation T =
_
3/G given on page 121 for the
period of a rotating, spherical object that results in zero apparent
gravity at its surface.
13 Section 4.4.3 presented an estimate of the upper limit on the
mass of a white dwarf. Check the self-consistency of the solution
in the following respects: (1) Why is it valid to ignore the contri-
bution of the nuclei to the degeneracy pressure? (2) Although the
electrons are ultrarelativistic, spacetime is approximated as being
at. As suggested in example 10 on page 50, a reasonable order-of-
magnitude check on this result is that we should have M/r c
2
/G.
14 The laws of physics in our universe imply that for bodies with
a certain range of masses, a neutron star is the unique equilibrium
state. Suppose we knew of the existence of neutron stars, but didnt
know the mass of the neutron. Infer upper and lower bounds on the
mass of the neutron.
Problems 127
b / Two rocks are dropped
side by side. The curvatures of
their world-lines are not intrinsic.
In a free-falling frame, both would
appear straight. If initially parallel
world-lines became non-parallel,
that would be evidence of intrinsic
curvature.
a / The expected structure of
the eld equations in general
relativity.
Chapter 5
Curvature
General relativity describes gravitation as a curvature of spacetime,
with matter acting as the source of the curvature in the same way
that electric charge acts as the source of electric elds. Our goal is
to arrive at Einsteins eld equations, which relate the local intrin-
sic curvature to the locally ambient matter in the same way that
Gausss law relates the local divergence of the electric eld to the
charge density. The locality of the equations is necessary because
relativity has no action at a distance; cause and eect propagate at
a maximum velocity of c(= 1).
The hard part is arriving at the right way of dening curvature.
Weve already seen that it can be tricky to distinguish intrinsic
curvature, which is real, from extrinsic curvature, which can never
produce observable eects. E.g., example 4 on page 79 showed that
spheres have intrinsic curvature, while cylinders do not. The mani-
festly intrinsic tensor notation protects us from being misled in this
respect. If we can formulate a denition of curvature expressed using
only tensors that are expressed without reference to any preordained
coordinate system, then we know it is physically observable, and not
just a supercial feature of a particular model.
As an example, drop two rocks side by side, b. Their trajectories
are vertical, but on a (t, x) coordinate plot rendered in the Earths
frame of reference, they appear as parallel parabolas. The curva-
ture of these parabolas is extrinsic. The Earth-xed frame of refer-
ence is dened by an observer who is subject to non-gravitational
forces, and is therefore not a valid Lorentz frame. In a free-falling
Lorentz frame (t
, x
), the two rocks are either motionless or moving

at constant velocity in straight lines. We can therefore see that the
curvature of world-lines in a particular coordinate system is not an
intrinsic measure of curvature; it can arise simply from the choice
of the coordinate system. What would indicate intrinsic curvature
would be, for example, if geodesics that were initially parallel were
to converge or diverge.
Nor is the metric a measure of intrinsic curvature. In example
14 on page 116, we found the metric for an accelerated observer to
be
g
t
= (1 +ax
)
2
g
x
x
= 1 ,
where the primes indicate the accelerated observers frame. The fact
that the timelike element is not equal to 1 is not an indication of
129
a / Tidal forces disrupt comet
Shoemaker-Levy.
intrinsic curvature. It arises only from the choice of the coordinates
(t
, x
) dened by a frame tied to the accelerating rocket ship.

The fact that the above metric has nonvanishing derivatives, un-
like a constant Lorentz metric, does indicate the presence of a grav-
itational eld. However, a gravitational eld is not the same thing
as intrinsic curvature. The gravitational eld seen by an observer
aboard the ship is, by the equivalence principle, indistinguishable
from an acceleration, and indeed the Lorentzian observer in the
earths frame does describe it as arising from the ships accelera-
tion, not from a gravitational eld permeating all of space. Both
observers must agree that I got plenty of nothin that the
region of the universe to which they have access lacks any stars,
neutrinos, or clouds of dust. The observer aboard the ship must de-
scribe the gravitational eld he detects as arising from some source
very far away, perhaps a hypothetical vast sheet of lead lying billions
of light-years aft of the ships deckplates. Such a hypothesis is ne,
but it is unrelated to the structure of our hoped-for eld equation,
which is to be local in nature.
Not only does the metric tensor not represent the gravitational
eld, but no tensor can represent it. By the equivalence princi-
ple, any gravitational eld seen by observer A can be eliminated by
switching to the frame of a free-falling observer B who is instanta-
neously at rest with respect to A at a certain time. The structure of
the tensor transformation law guarantees that A and B will agree on
whether a given tensor is zero at the point in spacetime where they
pass by one another. Since they agree on all tensors, and disagree
on the gravitational eld, the gravitational eld cannot be a tensor.
We therefore conclude that a nonzero intrinsic curvature of the
type that is to be included in the Einstein eld equations is not
encoded in any simple way in the metric or its rst derivatives.
Since neither the metric nor its rst derivatives indicate curvature,
we can reasonably conjecture that the curvature might be encoded
in its second derivatives.
5.1 Tidal curvature versus curvature caused
by local sources
A further complication is the need to distinguish tidal curva-
ture from curvature caused by local sources. Figure a shows Comet
Shoemaker-Levy, broken up into a string of fragments by Jupiters
tidal forces shortly before its spectacular impact with the planet in
1994. Immediately after each fracture, the newly separated chunks
had almost zero velocity relative to one another, so once the comet
nished breaking up, the fragments world-lines were a sheaf of
nearly parallel lines separated by spatial distances of only 1 km.
These initially parallel geodesics then diverged, eventually fanning
130 Chapter 5 Curvature
b / Tidal forces cause the ini-
tially parallel world-lines of the
fragments to diverge. The space-
time occupied by the comet has
intrinsic curvature, but it is not
caused by any local mass; it is
caused by the distant mass of
Jupiter.
c / The moons gravitational
eld causes the Earths oceans to
be distorted into an ellipsoid. The
sign of the sectional curvature is
negative in the x t plane, but
positive in the y t plane.
d / A cloud of test masses is
released at rest in a spherical
shell around the earth, shown
here as a circle because the z
axis is omitted. The volume of
the shell contracts over time,
which demonstrates that the
local curvature of spacetime is
generated by a local source
the earth rather than some
distant one.
out to span millions of kilometers.
If initially parallel lines lose their parallelism, that is clearly an
indication of intrinsic curvature. We call it a measure of sectional
curvature, because the loss of parallelism occurs within a particular
plane, in this case the (t, x) plane represented by gure b.
But this curvature was not caused by a local source lurking in
among the fragments. It was caused by a distant source: Jupiter.
We therefore see that the mere presence of sectional curvature is not
enough to demonstrate the existence of local sources. Even the sign
of the sectional curvature is not a reliable indication. Although this
example showed a divergence of initially parallel geodesics, referred
to as a negative curvature, it is also possible for tidal forces exerted
by distant masses to create positive curvature. For example, the
ocean tides on earth oscillate both above and below mean sea level,
c.
As an example that really would indicate the presence of a local
source, we could release a cloud of test masses at rest in a spheri-
cal shell around the earth, and allow them to drop, d. We would
then have positive and equal sectional curvature in the t x, t y,
and t z planes. Such an observation cannot be due to a distant
mass. It demonstrates an over-all contraction of the volume of an
initially parallel sheaf of geodesics, which can never be induced by
tidal forces. The earths oceans, for example, do not change their
total volume due to the tides, and this would be true even if the
oceans were a gas rather than an incompressible uid. It is a unique
property of 1/r
2
forces such as gravity that they conserve volume
in this way; this is essentially a restatement of Gausss law in a
vacuum.
5.2 The energy-momentum tensor
In general, the curvature of spacetime will contain contributions
from both tidal forces and local sources, superimposed on one an-
other. To develop the right formulation for the Einstein eld equa-
tions, we need to eliminate the tidal part. Roughly speaking, we
will do this by averaging the sectional curvature over all three of the
planes tx, ty, and tz, giving a measure of curvature called the
Ricci curvature. The roughly speaking is because such a prescrip-
tion would treat the time and space coordinates in an extremely
asymmetric manner, which would violate local Lorentz invariance.
To get an idea of how this would work, lets compare with the
Newtonian case, where there really is an asymmetry between the
treatment of time and space. In the Cartan curved-spacetime the-
ory of Newtonian gravity (page 35), the eld equation has a kind of
scalar Ricci curvature on one side, and on the other side is the den-
sity of mass, which is also a scalar. In relativity, however, the source
Section 5.2 The energy-momentum tensor 131
a / This curve has no intrin-
sic curvature.
b / A surveyor on a mountaintop
uses a heliotrope.
c / A map of a triangulation
survey such as the one Gauss
carried out. By measuring the
interior angles of the triangles,
one can determine not just the
two-dimensional projection of
the grid but its complete three-
dimensional form, including both
the curvature of the earth (note
the curvature of the lines of lat-
itude) and the height of features
above and below sea level.
term in the equation clearly cannot be the scalar mass density. We
know that mass and energy are equivalent in relativity, so for exam-
ple the curvature of spacetime around the earth depends not just
on the mass of its atoms but also on all the other forms of energy it
contains, such as thermal energy and electromagnetic and nuclear
binding energy. Can the source term in the Einstein eld equations
therefore be the mass-energy E? No, because E is merely the time-
like component of a particles momentum four-vector. To single it
out would violate Lorentz invariance just as much as an asymmetric
treatment of time and space in constructing a Ricci measure of cur-
vature. To get a properly Lorentz invariant theory, we need to nd a
way to formulate everything in terms of tensor equations that make
no explicit reference to coordinates. The proper generalization of
the Newtonian mass density in relativity is the energy-momentum
tensor T
ij
(also known as the stress-energy tensor), whose 16 el-
ements measure the local density of mass-energy and momentum,
and also the rate of transport of these quantities in various direc-
tions. If we happen to be able to nd a frame of reference in which
the local matter is all at rest, then T
00
represents the mass density.
For the purposes of the present discussion, its not necessary to
introduce the explicit denition of T; the point is merely that we
should expect the Einstein eld equations to be tensor equations,
which tells us that the denition of curvature were seeking clearly
has to be a rank-2 tensor, not a scalar. The implications in four-
dimensional spacetime are fairly complex. Well end up with a rank-
4 tensor that measures the sectional curvature, and a rank-2 Ricci
tensor derived from it that averages away the tidal eects. The
Einstein eld equations then relate the Ricci tensor to the energy-
momentum tensor in a certain way. The energy-momentum tensor
is discussed further in section 7.1.2 on page 195.
5.3 Curvature in two spacelike dimensions
Since the curvature tensors in 3+1 dimensions are complicated, lets
start by considering lower dimensions. In one dimension, a, there
is no such thing as intrinsic curvature. This is because curvature
describes the failure of parallelism to behave as in E5, but there is
no notion of parallelism in one dimension.
The lowest interesting dimension is therefore two, and this case
was studied by Carl Friedrich Gauss in the early nineteenth century.
Gauss ran a geodesic survey of the state of Hanover, inventing an
optical surveying instrument called a heliotrope that in eect was
used to cover the Earths surface with a triangular mesh of light
rays. If one of the mesh points lies, for example, at the peak of a
mountain, then the sum of the angles of the vertices meeting at
that point will be less than 2, in contradiction to Euclid. Although
the light rays do travel through the air above the dirt, we can think
e / Example 2.
d / Example 1.
of them as approximations to geodesics painted directly on the dirt,
which would be intrinsic rather than extrinsic. The angular defect
around a vertex now vanishes, because the space is locally Euclidean,
but we now pick up a dierent kind of angular defect, which is that
the interior angles of a triangle no longer add up to the Euclidean
value of .
A polygonal survey of a soccer ball Example: 1
Figure d applies similar ideas to a soccer ball, the only difference
being the use of pentagons and hexagons rather than triangles.
In d/1, the survey is extrinsic, because the lines pass below the
surface of the sphere. The curvature is detectable because the
angles at each vertex add up to 120 + 120 + 110 = 350 degrees,
giving an angular defect of 10 degrees.
In d/2, the lines have been projected to form arcs of great circles
on the surface of the sphere. Because the space is locally Eu-
clidean, the sum of the angles at a vertex has its Euclidean value
of 360 degrees. The curvature can be detected, however, be-
cause the sum of the internal angles of a polygon is greater than
the Euclidean value. For example, each spherical hexagon gives
a sum of 6 124.31 degrees, rather than the Euclidean 6 120.
The angular defect of 6 4.31 degrees is an intrinsic measure of
curvature.
Angular defect on the earths surface Example: 2
Divide the Earths northern hemisphere into four octants, with
their boundaries running through the north pole. These octants
have sides that are geodesics, so they are equilateral triangles.
Assuming Euclidean geometry, the interior angles of an equilat-
eral triangle are each equal to 60 degrees, and, as with any tri-
angle, they add up to 180 degrees. The octant-triangle in gure
e has angles that are each 90 degrees, and the sum is 270. This
shows that the Earths surface has intrinsic curvature.
This example suggests another way of measuring intrinsic curva-
ture, in terms of the ratio C/r of the circumference of a circle to
its radius. In Euclidean geometry, this ratio equals 2. Let be
the radius of the Earth, and consider the equator to be a circle
centered on the north pole, so that its radius is the length of one
of the sides of the triangle in gure e, r = (/2). (Dont confuse
r , which is intrinsic, with , the radius of the sphere, which is ex-
trinsic and not equal to r .) Then the ratio C/r is equal to 4, which
is smaller than the Euclidean value of 2.
Let = be the angular defect of a triangle, and for
concreteness let the triangle be in a space with an elliptic geometry,
so that it has constant curvature and can be modeled as a sphere of
radius , with antipodal points identied.
Self-check: In elliptic geometry, what is the minimum possible
Section 5.3 Curvature in two spacelike dimensions 133
g / Gaussian normal coordi-
nates on a sphere.
f / Proof that the angular defect
of a triangle in elliptic geometry
is proportional to its area. Each
white circle represents the en-
tire elliptic plane. The dashed
line at the edge is not really a
boundary; lines that go off the
edge simply wrap back around.
In the spherical model, the white
circle corresponds to one hemi-
sphere, which is identied with
the opposite hemisphere.
value of the quantity C/r discussed in example 2? How does this
dier from the case of spherical geometry?
We want a measure of curvature that is local, but if our space
is locally at, we must have 0 as the size of the triangles ap-
proaches zero. This is why Euclidean geometry is a good approx-
imation for small-scale maps of the earth. The discrete nature of
the triangular mesh is just an artifact of the denition, so we want
a measure of curvature that, unlike , approaches some nite limit
as the scale of the triangles approaches zero. Should we expect this
scaling to go as ?
2
? Lets determine the scaling. First
we prove a classic lemma by Gauss, concerning a slightly dierent
version of the angular defect, for a single triangle.
Theorem: In elliptic geometry, the angular defect = ++
of a triangle is proportional to its area A.
Proof: By axiom E2, extend each side of the triangle to form a line,
gure f/1. Each pair of lines crosses at only one point (E1) and
divides the plane into two lunes with its two vertices touching at
this point, gure f/2. The two interior angles at the vertex are the
same (Euclid I.15). The area of a lune is proportional to its interior
angle, as follows from dissection into narrower lunes; since a lune
with an interior angle of covers the entire area P of the plane, the
constant of proportionality is P/. The sum of the areas of our three
lunes is (P/)(+ +), but these three areas also cover the entire
plane, overlapping three times on the given triangle, and therefore
their sum also equals P + 2A. Equating the two expressions leads
to the desired result.
This calculation was purely intrinsic, because it made no use of
any model or coordinates. We can therefore construct a measure
of curvature that we can be assured is intrinsic, K = /A. This is
called the Gaussian curvature, and in elliptic geometry it is constant
rather than varying from point to point. In the model on a sphere
of radius , we have K = 1/
2
.
Self-check: Verify the equation K = 1/
2
by considering a tri-
angle covering one octant of the sphere, as in example 2.
It is useful to introduce normal coordinates, dened as follows.
Through point O, construct perpendicular geodesics, and dene
ane coordinates x and y along these. For any point P o the axis,
dene coordinates by constructing the lines through P that cross
the axes perpendicularly. For P in a suciently small neighborhood
of O, these lines exist and are uniquely determined. Gaussian polar
coordinates can be dened in a similar way.
Here are two useful interpretations of K.
1. The Gaussian curvature measures the failure of parallelism in
the following sense. Let line be constructed so that it crosses the
normal y axis at (0, dy) at an angle that diers from perpendicular
by the innitesimal amount d (gure h). Construct the line x
=
h / 1. Gaussian curvature
can be interpreted as the failure
of parallelism represented by
d
2
/dxdy.
i / 2. Gaussian curvature as
L ,= r .
dx, and let d
be the angle its perpendicular forms with . Then

1
the Gaussian curvature at O is
K =
d
2
dxdy
,
where d
2
= d
d.
2. From a point P, emit a fan of rays at angles lling a certain
range of angles in Gaussian polar coordinates (gure i). Let the
arc length of this fan at r be L, which may not be equal to its
Euclidean value L
E
= r. Then
2
K = 3
d
2
dr
2
_
L
L
e
_
.
Lets now generalize beyond elliptic geometry. Consider a space
modeled by a surface embedded in three dimensions, with geodesics
dened as curves of extremal length, i.e., the curves made by a piece
of string stretched taut across the surface. At a particular point
P, we can always pick a coordinate system (x, y, z) such that the
surface z =
1
2
k
1
x
2
+
1
2
k
2
y
2
locally approximates the surface to the
level of precision needed in order to discuss curvature. The surface
is either paraboloidal or hyperboloidal (a saddle), depending on the
signs of k
1
and k
2
. We might naively think that k
1
and k
2
could be
independently determined by intrinsic measurements, but as weve
seen in example 4 on page 79, a cylinder is locally indistinguishable
from a Euclidean plane, so if one k is zero, the other k clearly cannot
be determined. In fact all that can be measured is the Gaussian
curvature, which equals the product k
1
k
2
. To see why this should
be true, rst consider that any measure of curvature has units of
inverse distance squared, and the ks have units of inverse distance.
The only possible intrinsic measures of curvature based on the ks
are therefore k
2
1
+k
2
2
and k
1
k
2
. (We cant have, for example, just k
2
1
,
because that would change under an extrinsic rotation about the z
axis.) Only k
1
k
2
vanishes on a cylinder, so it is the only possible
intrinsic curvature.
Eating pizza Example: 3
When people eat pizza by folding the slice lengthwise, they are
taking advantage of the intrinsic nature of the Gaussian curva-
ture. Once k
1
is xed to a nonzero value, k
2
cant change without
varying K, so the slice cant droop.
1
Proof: Since any two lines cross in elliptic geometry, crosses the x axis. The
corollary then follows by application of the denition of the Gaussian curvature
to the right triangles formed by , the x axis, and the lines at x = 0 and x = dx,
so that K = d/dA = d
2
/dxdy, where third powers of innitesimals have been
discarded.
2
In the spherical model, L = sin u, where u is the angle subtended at the
center of the sphere by an arc of length r. We then have L/LE = sin u/u, whose
second derivative with respect to u is 1/3. Since r = u, the second derivative
of the same quantity with respect to r equals 1/3
2
= K/3.
Section 5.3 Curvature in two spacelike dimensions 135
k / A ea on the football can-
not orient himself by intrinsic,
local measurements.
j / A triangle in a space with
negative curvature has angles
that add to less than .
Elliptic and hyperbolic geometry Example: 4
Weve seen that gures behaving according to the axioms of el-
liptic geometry can be modeled on part of a sphere, which is a
surface of constant K > 0. The model can be made into global
one satisfying all the axioms if the appropriate topological prop-
erties are ensured by identifying antipodal points. A paraboloidal
surface z = k
1
x
2
+ k
2
y
2
can be a good local approximation to
a sphere, but for points far from its apex, K varies signicantly.
Elliptic geometry has no parallels; all lines meet if extended far
enough.
A space of constant negative curvature has a geometry called
hyperbolic, and is of some interest because it appears to be the
one that describes the the spatial dimensions of our universe on
a cosmological scale. A hyperboloidal surface works locally as a
model, but its curvature is only approximately constant; the sur-
face of constant curvature is a horn-shaped one created by re-
volving a mountain-shaped curve called a tractrix about its axis.
The tractrix of revolution is not as satisfactory a model as the
sphere is for elliptic geometry, because lines are cut off at the
cusp of the horn. Hyperbolic geometry is richer in parallels than
Euclidean geometry; given a line and a point P not on , there
are innitely many lines through P that do not pass through .
A ea on a football Example: 5
We might imagine that a ea on the surface of an American foot-
ball could determine by intrinsic, local measurements which di-
rection to go in order to get to the nearest tip. This is impossible,
because the ea would have to determine a vector, and curvature
cannot be a vector, since z =
1
2
k
1
x
2
+
1
2
k
2
y
2
is invariant under the
parity inversion x x, y y. For similar reasons, a measure
of curvature can never have odd rank.
Without violating reection symmetry, it is still conceivable that the
ea could determine the orientation of the tip-to-tip line running
through his position. Surprisingly, even this is impossible. The
ea can only measure the single number K, which carries no
information about directions in space.
5.4 Curvature tensors
The example of the ea suggests that if we want to express curvature
as a tensor, it should have even rank. Also, in a coordinate system
in which the coordinates have units of distance (they are not angles,
for instance, as in spherical coordinates), we expect that the units of
curvature will always be inverse distance squared. More elegantly,
we expect that under a uniform rescaling of coordinates by a factor
of , a curvature tensor should scale down by
2
.
Combining these two facts, we nd that a curvature tensor should
have one of the forms R
ab
, R
a
bcd
, . . . , i.e., the number of lower in-
a / The denition of the Riemann
tensor. The vector v
b
changes
by dv
b
when parallel-transported
around the approximate parallel-
ogram. (v
b
is drawn on a scale
that makes its length comparable
to the innitesimals dp
c
, dq
d
, and
dv
b
; in reality, its size would be
greater than theirs by an innite
factor.)
dices should be two greater than the number of upper indices. The
following denition has this property, and is equivalent to the earlier
denitions of the Gaussian curvature that were not written in tensor
notation.
Denition of the Riemann curvature tensor: Let dp
c
and dq
d
be two innitesimal vectors, and use them to form a quadrilateral
that is a good approximation to a parallelogram.
3
Parallel-transport
vector v
b
all the way around the parallelogram. When it comes back
to its starting place, it has a new value v
b
v
b
+ dv
b
. Then the
Riemann curvature tensor is dened as the tensor that computes dv
a
according to dv
a
= R
a
bcd
v
b
dp
c
dq
d
. (There is no standardization in
the literature of the order of the indices.)
A symmetry of the Riemann tensor Example: 6
If vectors dp
c
and dq
d
lie along the same line, then dv
a
must van-
ish, and interchanging dp
c
and dq
d
simply reverses the direction
of the circuit around the quadrilateral, giving dv
a
dv
a
. This
shows that R
a
bcd
must be antisymmetric under interchange of the
indices c and d, R
a
bcd
= R
a
bdc
.
In local normal coordinates, the interpretation of the Riemann
tensor becomes particularly transparent. The constant-coordinate
lines are geodesics, so when the vector v
b
is transported along them,
it maintains a constant angle with respect to them. Any rotation
of the vector after it is brought around the perimeter of the quadri-
lateral can therefore be attributed to something that happens at
the vertices. In other words, it is simply a measure of the angular
defect. We can therefore see that the Riemann tensor is really just
a tensorial way of writing the Gaussian curvature K = d/dA.
In normal coordinates, the local geometry is nearly Cartesian,
and when we take the product of two vectors in an antisymmetric
manner, we are essentially measuring the area of the parallelogram
they span, as in the three-dimensional vector cross product. We can
therefore see that the Riemann tensor tells us something about the
amount of curvature contained within the innitesimal area spanned
by dp
c
and dq
d
. A nite two-dimensional region can be broken
down into innitesimal elements of area, and the Riemann tensor
integrated over them. The result is equal to the nite change v
b
in a vector transported around the whole boundary of the region.
Curvature tensors on a sphere Example: 7
Lets nd the curvature tensors on a sphere of radius .
Construct normal coordinates (x, y) with origin O, and let vec-
tors dp
c
and dq
d
represent innitesimal displacements along x
and y, forming a quadrilateral as described above. Then R
x
yxy
represents the change in the x direction that occurs in a vector
that is initially in the y direction. If the vector has unit magni-
3
Section 5.8 discusses the sense in which this approximation is good enough.
Section 5.4 Curvature tensors 137
b / The change in the vector
due to parallel transport around
the octant equals the integral
of the Riemann tensor over the
interior.
tude, then R
x
yxy
equals the angular decit of the quadrilateral.
Comparing with the denition of the Gaussian curvature, we nd
R
x
yxy
= K = 1/
2
. Interchanging x and y, we nd the same result
for R
y
xyx
. Thus although the Riemann tensor in two dimensions
has sixteen components, only these two are nonzero, and they
are equal to each other.
This result represents the defect in parallel transport around a
closed loop per unit area. Suppose we parallel-transport a vector
around an octant, as shown in gure b. The area of the octant
is (/2)
2
, and multiplying it by the Riemann tensor, we nd that
the defect in parallel transport is /2, i.e., a right angle, as is also
evident from the gure.
The above treatment may be somewhat misleading in that it may
lead you to believe that there is a single coordinate system in
which the Riemann tensor is always constant. This is not the
case, since the calculation of the Riemann tensor was only valid
near the origin O of the normal coordinates. The character of
these coordinates becomes quite complicated far from O; we end
up with all our constant-x lines converging at north and south
poles of the sphere, and all the constant-y lines at east and west
poles.
Angular coordinates (, ) are more suitable as a large-scale de-
scription of the sphere. We can use the tensor transformation law
to nd the Riemann tensor in these coordinates. If O, the origin
of the (x, y) coordinates, is at coordinates (, ), then dx/d =
sin and dy/d = . The result is R
= R
x
yxy
(dy/d)
2
= 1
and R
= R
y
xyx
(dx/d)
2
= sin
2
. The variation in R
is not due to any variation in the spheres intrinsic curvature; it

represents the behavior of the coordinate system.
The Riemann tensor only measures curvature within a particular
plane, the one dened by dp
c
and dq
d
, so it is a kind of sectional cur-
vature. Since were currently working in two dimensions, however,
there is only one plane, and no real distinction between sectional
curvature and Ricci curvature, which is the average of the sectional
curvature over all planes that include dq
d
: R
cd
= R
a
cad
. The Ricci
curvature in two spacelike dimensions, expressed in normal coordi-
nates, is simply the diagonal matrix diag(K, K).
5.5 Some order-of-magnitude estimates
As a general proposition, calculating an order-of-magnitude estimate
of a physical eect requires an understanding of 50% of the physics,
while an exact calculation requires about 75%.
4
Weve reached
4
This statement is itself only a rough estimate. Anyone who has taught
physics knows that students will often calculate an eect exactly while not un-
derstanding the underlying physics at all.
a / The geodetic effect as
measured by Gravity Probe B.
the point where its reasonable to attempt a variety of order-of-
magnitude estimates.
5.5.1 The geodetic effect
How could we conrm experimentally that parallel transport
around a closed path can cause a vector to rotate? The rotation
is related to the amount of spacetime curvature contained within
the path, so it would make sense to choose a loop going around
a gravitating body. The rotation is a purely relativistic eect, so
we expect it to be small. To make it easier to detect, we should
go around the loop many times, causing the eect to accumulate.
This is essentially a description of a body orbiting another body. A
gyroscope aboard the orbiting body is expected to precess. This is
known as the geodetic eect. In 1916, shortly after Einstein pub-
lished the general theory of relativity, Willem de Sitter calculated
the eect on the earth-moon system. The eect was not directly
veried until the 1980s, and the rst high-precision measurement
was in 2007, from analysis of the results collected by the Gravity
Probe B satellite experiment. The probe carried four gyroscopes
made of quartz, which were the most perfect spheres ever manu-
factured, varying from sphericity by no more than about 40 atoms.
Lets estimate the size of the eect. The rst derivative of the
metric is, roughly, the gravitational eld, whereas the second deriva-
tive has to do with curvature. The curvature of spacetime around
the earth should therefore vary as GMr
3
, where M is the earths
mass and G is the gravitational constant. The area enclosed by a
circular orbit is proportional to r
2
, so we expect the geodetic eect
to vary as nGM/r, where n is the number of orbits. The angle of
precession is unitless, and the only way to make this result unitless
is to put in a factor of 1/c
2
. In units with c = 1, this factor is un-
necessary. In ordinary metric units, the 1/c
2
makes sense, because
it causes the purely relativistic eect to come out to be small. The
result, up to unitless factors that we didnt pretend to nd, is

nGM
c
2
r
.
We might also expect a Thomas precession. Like the spacetime
curvature eect, it would be proportional to nGM/c
2
r. Since were
not worrying about unitless factors, we can just lump the Thomas
precession together with the eect already calculated.
The data for Gravity Probe B are r = r
e
+(650 km) and n 5000
(orbiting once every 90 minutes for the 353-day duration of the
experiment), giving 3 10
6
radians. Figure b shows the
actual results
5
the four gyroscopes aboard the probe. The precession
was about 6 arc-seconds, or 3 10
5
radians. Our crude estimate
5
http://einstein.stanford.edu/content/final_report/GPB_Final_
NASA_Report-020509-web.pdf
Section 5.5 Some order-of-magnitude estimates 139
was on the right order of magnitude. The missing unitless factor on
the right-hand side of the equation above is 3, which brings the two
results into fairly close quantitative agreement. The full derivation,
including the factor of 3, is given on page 175.
b / Precession angle as a function of time as measured by the four gyroscopes aboard Gravity Probe B.
5.5.2 Deection of light rays
In the discussion of the momentum four vector in section 4.2.2,
we saw that due to the equivalence principle, light must be aected
by gravity. There are two ways in which such an eect could occur.
Light can gain and lose momentum as it travels up and down in
a gravitational eld, or its momentum vector can be deected by
a transverse gravitational eld. As an example of the latter, a ray
of starlight can be deected by the suns gravity, causing the stars
apparent position in the sky to be shifted. The detection of this
eect was one of the rst experimental tests of general relativity.
Ordinarily the bright light from the sun would make it impossible
to accurately measure a stars location on the celestial sphere, but
this problem was sidestepped by Arthur Eddington during an eclipse
of the sun in 1919.
Lets estimate the size of this eect. Weve already seen that
the Riemann tensor is essentially just a tensorial way of writing
the Gaussian curvature K = d/dA. Suppose, for the sake of this
rough estimate, that the sun, earth, and star form a non-Euclidean
triangle with a right angle at the sun. Then the angular deection
is the same as the angular defect of this triangle, and equals the
integral of the curvature over the interior of the triangle. Ignoring
unitless constants, this ends up being exactly the same calculation
as in section 5.5.1, and the result is GM/c
2
r, where r is the
light rays distance of closest approach to the sun. The value of r
cant be less than the radius of the sun, so the maximum size of the
eect is on the order of GM/c
2
r, where M is the suns mass, and r
c / One of the photos from Ed-
dingtons observations of the
1919 eclipse. This is a pho-
tographic negative, so the cir-
cle that appears bright is actually
the dark face of the moon, and
the dark area is really the bright
corona of the sun. The stars,
marked by lines above and be-
low them, appeared at positions
slightly different than their nor-
mal ones, indicating that their light
had been bent by the suns gravity
on its way to our planet.
is its radius. We nd 10
5
radians, or about a second of arc. To
measure a stars position to within an arc second was well within
the state of the art in 1919, under good conditions in a comfortable
observatory. This observation, however, required that Eddingtons
team travel to the island of Principe, o the coast of West Africa.
The weather was cloudy, and only during the last 10 seconds of the
seven-minute eclipse did the sky clear enough to allow photographic
plates to be taken of the Hyades star cluster against the background
of the eclipse-darkened sky. The observed deection was 1.6 seconds
of arc, in agreement with the relativistic prediction. The relativistic
prediction is derived on page 182.
5.6 The covariant derivative
In the preceding section we were able to estimate a nontrivial general
relativistic eect, the geodetic precession of the gyroscopes aboard
Gravity Probe B, up to a unitless constant 3. Lets think about
what additional machinery would be needed in order to carry out
the calculation in detail, including the 3.
First we would need to know the Einstein eld equation, but in a
vacuum this is fairly straightforward: R
ab
= 0. Einstein posited this
equation based essentially on the considerations laid out in section
5.1.
But just knowing that a certain tensor vanishes identically in the
space surrounding the earth clearly doesnt tell us anything explicit
about the structure of the spacetime in that region. We want to
know the metric. As suggested at the beginning of the chapter, we
Section 5.6 The covariant derivative 141
a / A double-slit experiment
with electrons. If we add an
arbitrary constant to the potential,
no observable changes result.
The wavelength is shortened, but
the relative phase of the two parts
of the waves stays the same.
b / Two wavefunctions with
constant wavelengths, and a
third with a varying wavelength.
None of these are physically
distinguishable, provided that the
same variation in wavelength is
applied to all electrons in the
universe at any given point in
spacetime. There is not even
any unambiguous way to pick out
the third one as the one with a
varying wavelength. We could
choose a different gauge in which
the third wave was the only one
with a constant wavelength.
expect that the rst derivatives of the metric will give a quantity
analogous to the gravitational eld of Newtonian mechanics, but this
quantity will not be directly observable, and will not be a tensor.
The second derivatives of the metric are the ones that we expect to
relate to the Ricci tensor R
ab
.
5.6.1 The covariant derivative in electromagnetism
Were talking blithely about derivatives, but its not obvious how
to dene a derivative in the context of general relativity in such a
way that taking a derivative results in well-behaved tensor.
To see how this issue arises, lets retreat to the more familiar
terrain of electromagnetism. In quantum mechanics, the phase of a
charged particles wavefunction is unobservable, so that for example
the transformation does not change the results of experi-
ments. As a less trivial example, we can redene the ground of our
electrical potential, +, and this will add a constant onto
the energy of every electron in the universe, causing their phases to
oscillate at a greater rate due to the quantum-mechanical relation
E = hf. There are no observable consequences, however, because
what is observable is the phase of one electron relative to another,
as in a double-slit interference experiment. Since every electron has
been made to oscillate faster, the eect is simply like letting the con-
ductor of an orchestra wave her baton more quickly; every musician
is still in step with every other musician. The rate of change of the
wavefunction, i.e., its derivative, has some built-in ambiguity.
For simplicity, lets now restrict ourselves to spin-zero parti-
cles, since details of electrons polarization clearly wont tell us
anything useful when we make the analogy with relativity. For a
spin-zero particle, the wavefunction is simply a complex number,
and there are no observable consequences arising from the transfor-
mation
= e
i
, where is a constant. The transformation
is also allowed, and it gives (t) = (q/)t, so that
the phase factor e
i(t)
is a function of time t. Now from the point
of view of electromagnetism in the age of Maxwell, with the elec-
tric and magnetic elds imagined as playing their roles against a
background of Euclidean space and absolute time, the form of this
time-dependent phase factor is very special and symmetrical; it de-
pends only on the absolute time variable. But to a relativist, there is
nothing very nice about this function at all, because there is nothing
special about a time coordinate. If were going to allow a function
of this form, then based on the coordinate-invariance of relativity, it
seems that we should probably allow to be any function at all of
the spacetime coordinates. The proper generalization of
is now A
b
A
b

b
, where A
b
is the electromagnetic potential
four-vector (section 4.2.5, page 113).
Self-check: Suppose we said we would allow to be a function
of t, but forbid it to depend on the spatial coordinates. Prove that
c / These three rulers represent
three choices of coordinates. As
in gure b on page 142, switching
from one set of coordinates to
another has no effect on any
experimental observables. It is
merely a choice of gauge.
this would violate Lorentz invariance.
The transformation has no eect on the electromagnetic elds,
which are the direct observables. We can also verify that the change
of gauge will have no eect on observable behavior of charged par-
ticles. This is because the phase of a wavefunction can only be
determined relative to the phase of another particles wavefunction,
when they occupy the same point in space and, for example, inter-
fere. Since the phase shift depends only on the location in spacetime,
there is no change in the relative phase.
But bad things will happen if we dont make a corresponding
adjustment to the derivatives appearing in the Schrodinger equation.
These derivatives are essentially the momentum operators, and they
give dierent results when applied to
than when applied to :
b

b
_
e
i
_
= e
i
b
+i
b
_
e
i
_
=
_
b
+A
b
A
b
_
To avoid getting incorrect results, we have to do the substitution
b

b
+ ieA
b
, where the correction term compensates for the
change of gauge. We call the operator dened as
b
=
b
+ieA
b
the covariant derivative. It gives the right answer regardless of a
change of gauge.
5.6.2 The covariant derivative in general relativity
Now consider how all of this plays out in the context of gen-
eral relativity. The gauge transformations of general relativity are
arbitrary smooth changes of coordinates. One of the most basic
properties we could require of a derivative operator is that it must
give zero on a constant function. A constant scalar function remains
constant when expressed in a new coordinate system, but the same
is not true for a constant vector function, or for any tensor of higher
rank. This is because the change of coordinates changes the units
in which the vector is measured, and if the change of coordinates is
nonlinear, the units vary from point to point.
Consider the one-dimensional case, in which a vector v
a
has only
one component, and the metric is also a single number, so that we
can omit the indices and simply write v and g. (We just have to
remember that v is really a covariant vector, even though were
leaving out the upper index.) If v is constant, its derivative dv/dx,
computed in the ordinary way without any correction term, is zero.
If we further assume that the coordinate x is a normal coordinate, so
that the metric is simply the constant g = 1, then zero is not just the
answer but the right answer. (The existence of a preferred, global
set of normal coordinates is a special feature of a one-dimensional
space, because there is no curvature in one dimension. In more than
one dimension, there will typically be no possible set of coordinates
in which the metric is constant, and normal coordinates only give a
metric that is approximately constant in the neighborhood around
a certain point. See gure g pn page 134 for an example of normal
coordinates on a sphere, which do not have a constant metric.)
Now suppose we transform into a new coordinate system X,
which is not normal. The metric G, expressed in this coordinate
system, is not constant. Applying the tensor transformation law,
we have V = v dX/dx, and dierentiation with respect to X will
not give zero, because the factor dX/dx isnt constant. This is the
wrong answer: V isnt really varying, it just appears to vary because
G does.
We want to add a correction term onto the derivative operator
d/dX, forming a covariant derivative operator
X
that gives the
right answer. This correction term is easy to nd if we consider
what the result ought to be when dierentiating the metric itself.
In general, if a tensor appears to vary, it could vary either because
it really does vary or because the metric varies. If the metric itself
varies, it could be either because the metric really does vary or
. . . because the metric varies. In other words, there is no sensible
way to assign a nonzero covariant derivative to the metric itself, so
we must have
X
G = 0. The required correction therefore consists
of replacing d/dX with
X
=
d
dX
G
1
dG
dX
.
Applying this to G gives zero. G is a second-rank contravariant
tensor. If we apply the same correction to the derivatives of other
second-rank contravariant tensors, we will get nonzero results, and
they will be the right nonzero results. For example, the covariant
derivative of the stress-energy tensor T (assuming such a thing could
have some physical signicance in one dimension!) will be
X
T =
dT/dX G
1
(dG/dX)T.
Physically, the correction term is a derivative of the metric, and
weve already seen that the derivatives of the metric (1) are the clos-
est thing we get in general relativity to the gravitational eld, and
(2) are not tensors. In 1+1 dimensions, suppose we observe that a
free-falling rock has dV/dT = 9.8 m/s
2
. This acceleration cannot be
a tensor, because we could make it vanish by changing from Earth-
xed coordinates X to free-falling (normal, locally Lorentzian) co-
ordinates x, and a tensor cannot be made to vanish by a change of
coordinates. According to a free-falling observer, the vector v isnt
changing at all; it is only the variation in the Earth-xed observers
metric G that makes it appear to change.
Mathematically, the form of the derivative is (1/y)dy/dx, which
is known as a logarithmic derivative, since it equals d(ln y)/dx. It
d / Example 8.
measures the multiplicative rate of change of y. For example, if
y scales up by a factor of k when x increases by 1 unit, then the
logarithmic derivative of y is ln k. The logarithmic derivative of
e
cx
is c. The logarithmic nature of the correction term to
X
is a
good thing, because it lets us take changes of scale, which are mul-
tiplicative changes, and convert them to additive corrections to the
derivative operator. The additivity of the corrections is necessary if
the result of a covariant derivative is to be a tensor, since tensors
are additive creatures.
What about quantities that are not second-rank covariant ten-
sors? Under a rescaling of contravariant coordinates by a factor of
k, covariant vectors scale by k
1
, and second-rank covariant tensors
by k
2
. The correction term should therefore be half as much for
covariant vectors,
X
=
d
dX

1
2
G
1
dG
dX
.
and should have an opposite sign for contravariant vectors.
Generalizing the correction term to derivatives of vectors in more
than one dimension, we should have something of this form:
a
v
b
=
a
v
b
+
b
ac
v
c
a
v
b
=
a
v
b
c
ba
v
c
,
where
b
ac
, called the Christoel symbol, does not transform like
a tensor, and involves derivatives of the metric. (Christoel is
pronounced Krist-AWful, with the accent on the middle syllable.)
The explicit computation of the Christoel symbols from the metric
is deferred until section 5.9, but the intervening sections 5.7 and 5.8
can be omitted on a rst reading without loss of continuity.
Christoffel symbols on the globe Example: 8
As a qualitative example, consider the geodesic airplane trajec-
tory shown in gure d, from London to Mexico City. In physics
it is customary to work with the colatitude, , measured down
from the north pole, rather then the latitude, measured from the
equator. At P, over the North Atlantic, the planes colatitude has
a minimum. (We can see, without having to take it on faith from
the gure, that such a minimum must occur. The easiest way to
convince oneself of this is to consider a path that goes directly
over the pole, at = 0.)
At P, the planes velocity vector points directly west. At Q, over
New England, its velocity has a large component to the south.
Since the path is a geodesic and the plane has constant speed,
the velocity vector is simply being parallel-transported; the vec-
tors covariant derivative is zero. Since we have v
= 0 at P, the
only way to explain the nonzero and positive value of
is that
we have a nonzero and negative value of
.
a / The geodesic, 1, preserves
tangency under parallel trans-
port. The non-geodesic curve,
2, doesnt have this property;
a vector initially tangent to the
curve is no longer tangent to it
when parallel-transported along
it.
By symmetry, we can infer that
must have a positive value

in the southern hemisphere, and must vanish at the equator.
is computed in example 9 on page 157.

Symmetry also requires that this Christoffel symbol be indepen-
dent of , and it must also be independent of the radius of the
sphere.
Example 8 is in two spatial dimensions. In spacetime, is es-
sentially the gravitational eld (see problem 5, p. 162), and early
papers in relativity essentially refer to it that way.
6
This may feel
like a joyous reunion with our old friend from freshman mechanics,
g = 9.8 m/s. But our old friend has changed. In Newtonian me-
chanics, accelerations like g are frame-invariant (considering only
inertial frames, which are the only legitimate ones in that theory).
In general relativity they are frame-dependent, and as we saw on
page 144, the acceleration of gravity can be made to equal anything
we like, based on our choice of a frame of reference.
For higher-rank tensors, we just add more correction terms, e.g.,
a
U
bc
=
a
U
bc
d
ba
U
dc
d
ca
U
bd
or
a
U
c
b
=
a
U
c
b

d
ba
U
c
d
+
c
ad
U
d
b
.
With the partial derivative
a
, it does not make sense to use the
metric to raise the index and form
a
. It does make sense to do so
with covariant derivatives, so
a
= g
ab
b
is a correct identity.
Comma and semicolon notation
Some authors use superscripts with commas and semicolons to
indicate partial and covariant derivatives. The following equations
give equivalent notations for the same derivatives:
a
X
b
= X
b,a
a
X
b
= X
b;a
a
X
b
= X
;a
b
5.7 The geodesic equation
In this section, which can be skipped at a rst reading, we show how
the Christoel symbols can be used to nd dierential equations that
describe geodesics.
6
On the gravitational eld of a point mass according to Einsteins the-
ory, Sitzungsberichte der K oniglich Preussischen Akademie der Wissenschaften
1 (1916) 189, translated in arxiv.org/abs/physics/9905030v1.
5.7.1 Characterization of the geodesic
A geodesic can be dened as a world-line that preserves tangency
under parallel transport, a. This is essentially a mathematical way
of expressing the notion that we have previously expressed more
informally in terms of staying on course or moving inertially.
A curve can be specied by giving functions x
i
() for its coor-
dinates, where is a real parameter. A vector lying tangent to the
curve can then be calculated using partial derivatives, T
i
= x
i
/.
There are three ways in which a vector function of could change:
(1) it could change for the trivial reason that the metric is changing,
so that its components changed when expressed in the new metric;
(2) it could change its components perpendicular to the curve; or
(3) it could change its component parallel to the curve. Possibility
1 should not really be considered a change at all, and the denition
of the covariant derivative is specically designed to be insensitive
to this kind of thing. 2 cannot apply to T
i
, which is tangent by
construction. It would therefore be convenient if T
i
happened to
be always the same length. If so, then 3 would not happen either,
and we could reexpress the denition of a geodesic by saying that
the covariant derivative of T
i
was zero. For this reason, we will
assume for the remainder of this section that the parametrization
of the curve has this property. In a Newtonian context, we could
imagine the x
i
to be purely spatial coordinates, and to be a uni-
versal time coordinate. We would then interpret T
i
as the velocity,
and the restriction would be to a parametrization describing motion
with constant speed. In relativity, the restriction is that must be
an ane parameter. For example, it could be the proper time of a
particle, if the curve in question is timelike.
5.7.2 Covariant derivative with respect to a parameter
The notation of section 5.6 is not quite adapted to our present
purposes, since it allows us to express a covariant derivative with
respect to one of the coordinates, but not with respect to a param-
eter such as . We would like to notate the covariant derivative of
T
i
with respect to as
T
i
, even though isnt a coordinate. To
connect the two types of derivatives, we can use a total derivative.
To make the idea clear, here is how we calculate a total derivative
for a scalar function f(x, y), without tensor notation:
df
d
=
f
x
x
+
f
y
y
.
This is just the generalization of the chain rule to a function of two
variables. For example, if represents time and f temperature,
then this would tell us the rate of change of the temperature as
a thermometer was carried through space. Applying this to the
Section 5.7 The geodesic equation 147
present problem, we express the total covariant derivative as
T
i
= (
b
T
i
)
dx
b
d
=
_
b
T
i
+
i
bc
T
c
_
dx
b
d
.
5.7.3 The geodesic equation
Recognizing
b
T
i
dx
b
/d as a total non-covariant derivative, we
nd
T
i
=
dT
i
d
+
i
bc
T
c
dx
b
d
.
Substituting x
i
/ for T
i
, and setting the covariant derivative
equal to zero, we obtain
d
2
x
i
d
2
+
i
bc
dx
c
d
dx
b
d
= 0.
This is known as the geodesic equation.
If this dierential equation is satised for one ane parameter
, then it is also satised for any other ane parameter
= a+b,
where a and b are constants (problem 3). Recall that ane param-
eters are only dened along geodesics, not along arbitrary curves.
We cant start by dening an ane parameter and then use it to
nd geodesics using this equation, because we cant dene an ane
parameter without rst specifying a geodesic. Likewise, we cant
do the geodesic rst and then the ane parameter, because if we
already had a geodesic in hand, we wouldnt need the dierential
equation in order to nd a geodesic. The solution to this chicken-
and-egg conundrum is to write down the dierential equations and
try to nd a solution, without trying to specify either the ane pa-
rameter or the geodesic in advance. We will seldom have occasion
to resort to this technique, an exception being example 3 on page
215.
5.7.4 Uniqueness
The geodesic equation is useful in establishing one of the neces-
sary theoretical foundations of relativity, which is the uniqueness of
geodesics for a given set of initial conditions. This is related to ax-
iom O1 of ordered geometry, that two points determine a line, and
is necessary physically for the reasons discussed on page 21; briey,
if the geodesic were not uniquely determined, then particles would
have no way of deciding how to move. The form of the geodesic
equation guarantees uniqueness. To see this, consider the following
algorithm for determining a numerical approximation to a geodesic:
1. Initialize , the x
i
and their derivatives dx
i
/d. Also, set a
small step-size by which to increment at each step below.
2. For each i, calculate d
2
x
i
/d
2
using the geodesic equation.
3. Add (d
2
x
i
/d
2
) to the currently stored value of dx
i
/d.
4. Add (dx
i
/d) to x
i
.
5. Add to .
6. Repeat steps 2-5 until the the geodesic has been extended to
the desired ane distance.
Since the result of the calculation depends only on the inputs at
step 1, we nd that the geodesic is uniquely determined.
To see that this is really a valid way of proving uniqueness, it
may be helpful to consider how the proof could have failed. Omitting
some of the details of the tensors and the multidimensionality of the
space, the form of the geodesic equation is essentially x + f x
2
= 0,
where dots indicate derivatives with respect to . Suppose that it
had instead had the form x
2
+ f x = 0. Then at step 2 we would
have had to pick either a positive or a negative square root for x.
Although continuity would usually suce to maintain a consistent
sign from one iteration to the next, that would not work if we ever
came to a point where x vanished momentarily. An equation of this
form therefore would not have a unique solution for a given set of
initial conditions.
The practical use of this algorithm to compute geodesics numer-
ically is demonstrated in section 5.9.2 on page 157.
5.8 Torsion
This section describes the concept of gravitational torsion. It can
be skipped without loss of continuity, provided that you accept the
symmetry property
a
[bc]
= 0 without worrying about what it means
physically or what empirical evidence supports it.
Self-check: Interpret the mathematical meaning of the equation
a
[bc]
= 0, which is expressed in the notation introduced on page 84.
5.8.1 Are scalars path-dependent?
It seems clear that something like the covariant derivative is
needed for vectors, since they have a direction in spacetime, and
thus their measures vary when the measure of spacetime itself varies.
Since scalars dont have a direction in spacetime, the same reasoning
doesnt apply to them, and this is reected in our rules for covariant
derivatives. The covariant derivative has one term for every index
of the tensor being dierentiated, so for a scalar there should be no
terms at all, i.e.,
a
is the same as
a
.
But just because derivatives of scalars dont require special treat-
ment for this particular reason, that doesnt mean they are guaran-
Section 5.8 Torsion 149
a / Measuring
2
T/xy for
a scalar T.
teed to behave as we intuitively expect, in the strange world of
coordinate-invariant relativity.
One possible way for scalars to behave counterintuitively would
be by analogy with parallel transport of vectors. If we stick a vector
in a box (as with, e.g., the gyroscopes aboard Gravity Probe B) and
carry it around a closed loop, it changes. Could the same happen
with a scalar? This is extremely counterintuitive, since there is no
reason to imagine such an eect in any of the models weve con-
structed of curved spaces. In fact, it is not just counterintuitive but
mathematically impossible, according to the following argument.
The only reason we can interpret the vector-in-a-box eect as aris-
ing from the geometry of spacetime is that it applies equally to all
vectors. If, for example, it only applied to the magnetic polariza-
tion vectors of ferromagnetic substances, then we would interpret
it as a magnetic eld living in spacetime, not a property of space-
time itself. If the value of a scalar-in-a-box was path-dependent,
and this path-dependence was a geometric property of spacetime,
then it would have to apply to all scalars, including, say, masses
and charges of particles. Thus if an electrons mass increased by 1%
when transported in a box along a certain path, its charge would
have to increase by 1% as well. But then its charge-to-mass ra-
tio would remain invariant, and this is a contradiction, since the
charge-to-mass ratio is also a scalar, and should have felt the same
1% eect. Since the varying scalar-in-a-box idea leads to a contra-
diction, it wasnt a coincidence that we couldnt nd a model that
produced such an eect; a theory that lacks self-consistency doesnt
have any models.
Self-check: Explain why parallel transporting a vector can only
rotate it, not change its magnitude.
There is, however, a dierent way in which scalars could behave
counterintuitively, and this one is mathematically self-consistent.
Suppose that Helen lives in two spatial dimensions and owns a ther-
mometer. She wants to measure the spatial variation of tempera-
ture, in particular its mixed second derivative
2
T/xy. At home
in the morning at point A, she prepares by calibrating her gyro-
compass to point north and measuring the temperature. Then she
travels = 1 km east along a geodesic to B, consults her gyro-
compass, and turns north. She continues one kilometer north to C,
samples the change in temperature T
1
relative to her home, and
then retraces her steps to come home for lunch. In the afternoon,
she checks her work by carrying out the same process, but this time
she interchanges the roles of north and east, traveling along ADE.
If she were living in a at space, this would form the other two sides
of a square, and her afternoon temperature sample T
2
would be
at the same point in space C as her morning sample. She actually
doesnt recognize the landscape, so the sample points C and E are
dierent, but this just conrms what she already knew: the space
b / The gyroscopes both ro-
tate when transported from A
to B, causing Helen to navigate
along BC, which does not form
a right angle with AB. The angle
between the two gyroscopes
axes is always the same, so the
rotation is not locally observable,
but it does produce an observable
gap between C and E.
isnt at.
7
None of this seems surprising yet, but there are now two quali-
tatively dierent ways that her analysis of her data could turn out,
indicating qualitatively dierent things about the laws of physics
in her universe. The denition of the derivative as a limit requires
that she repeat the experiment at smaller scales. As 0, the
result for
2
T/xy should approach a denite limit, and the er-
ror should diminish in proportion to . In particular the dierence
between the results inferred from T
1
and T
2
indicate an error,
and the discrepancy between the second derivatives inferred from
them should shrink appropriately as shrinks. Suppose this doesnt
happen. Since partial derivatives commute, we conclude that her
measuring procedure is not the same as a partial derivative. Lets
call her measuring procedure , so that she is observing a discrep-
ancy between
x
y
and
y
x
. The fact that the commutator
x
doesnt vanish cannot be explained by the Christof-
fel symbols, because what shes dierentiating is a scalar. Since the
discrepancy arises entirely from the failure of T
1
T
2
to scale
down appropriately, the conclusion is that the distance between
the two sampling points is not scaling down as quickly as we ex-
pect. In our familiar models of two-dimensional spaces as surfaces
embedded in three-space, we always have
3
for small , but she
has found that it only shrinks as quickly as
2
.
For a clue as to what is going on, note that the commutator
y

y
x
has a particular handedness to it. For example,
it ips its sign under a reection across the line y = x. When we
parallel-transport vectors, they arent actually staying parallel. In
this hypothetical universe, a vector in a box transported by a small
distance rotates by an angle proportional to . This eect is called
torsion. Although no torsion eect shows up in our familiar models,
that is not because torsion lacks self-consistency. Models of spaces
with torsion do exist. In particular, we can see that torsion doesnt
lead to the same kind of logical contradiction as the varying-scalar-
in-a-box idea. Since all vectors twist by the same amount when
transported, inner products are preserved, so it is not possible to
put two vectors in one box and get the scalar-in-a-box paradox by
watching their inner product change when the box is transported.
Note that the elbows ABC and ADE are not right angles. If
Helen had brought a pair of gyrocompasses with her, one for x and
one for y, she would have found that the right angle between the
gyrocompasses was preserved under parallel transport, but that a
gyrocompass initially tangent to a geodesic did not remain so. There
are in fact two inequivalent denitions of a geodesic in a space with
torsion. The shortest path between two points is not necessarily
the same as the straightest possible path, i.e., the one that parallel-
7
This point was mentioned on page 137, in connection with the denition of
the Riemann tensor.
transports its own tangent vector.
5.8.2 The torsion tensor
Since torsion is odd under parity, it must be represented by an
odd-rank tensor, which we call
c
ab
and dene according to
(
a
a
)f =
c
ab
c
f ,
where f is any scalar eld, such as the temperature in the preced-
ing section. There are two dierent ways in which a space can be
non-Euclidean: it can have curvature, or it can have torsion. For
a full discussion of how to handle the mathematics of a spacetime
with both curvature and torsion, see the article by Steuard Jensen at
http://www.slimy.com/
~
steuard/teaching/tutorials/GRtorsion.
pdf. For our present purposes, the main mathematical fact worth
noting is that vanishing torsion is equivalent to the symmetry
a
bc
=
a
cb
of the Christoel symbols. Using the notation introduced on
page 84,
a
[bc]
= 0 if = 0.
Self-check: Use an argument similar to the one in example 5
on page 136 to prove that no model of a two-space embedded in a
three-space can have torsion.
Generalizing to more dimensions, the torsion tensor is odd under
the full spacetime reection x
a
x
a
, i.e., a parity inversion plus
a time-reversal, PT.
In the story above, we had a torsion that didnt preserve tan-
gent vectors. In three or more dimensions, however, it is possible
to have torsion that does preserve tangent vectors. For example,
transporting a vector along the x axis could cause only a rotation in
the y-z plane. This relates to the symmetries of the torsion tensor,
which for convenience well write in an x-y-z coordinate system and
in the fully covariant form
. The denition of the torsion tensor

implies
()
= 0, i.e., that the torsion tensor is antisymmetric in
its two nal indices. Torsion that does not preserve tangent vectors
will have nonvanishing elements such as
xxy
, meaning that parallel-
transporting a vector along the x axis can change its x component.
Torsion that preserves tangent vectors will have vanishing
un-
less , , and are all distinct. This is an example of the type
of antisymmetry that is familiar from the vector cross product, in
which the cross products of the basis vectors behave as x y = z,
y z = x, y z = x. Generalizing the notation for symmetrization
and antisymmetrization of tensors from page 84, we have
T
(abc)
=
1
3!
T
abc
T
[abc]
=
1
3!
abc
T
abc
,
where the sums are over all permutations of the indices, and in the
second line the totally antisymmetric Levi-Civita tensor is dened
c / Three gyroscopes are ini-
tially aligned with the x, y, and
z axes. After parallel transport
along the geodesic x axis, the
x gyro is still aligned with the x
axis, but the y and z gyros have
rotated.
by
abc
=
bca
=
cab
= +1,
bac
=
acb
=
cba
= 1, and
= 0
for all other combinations of indices. In other words,
is +1 if
represent an even permutation of abc (i.e., they can be obtained
from abc by an even number of pairwise swaps), 1 for odd permu-
tations, and 0 otherwise (i.e., if is not a permutation of abc).
In this notation, a totally antisymmetric torsion tensor is one with
[]
, and torsion of this type preserves tangent vectors under
translation.
In two dimensions, there are no totally antisymmetric rank-3
tensors, because we cant write three indices without repeating one.
In three dimensions, an antisymmetric rank-3 tensor is simply a mul-
tiple of the Levi-Civita tensor, so a totally antisymmetric torsion, if
it exists, is represented by a single number; under translation, vec-
tors rotate like either right-handed or left-handed screws, and this
number tells us the rate of rotation. In four dimensions, we have
four independently variable quantities,
xyz
,
tyz
,
txz
, and
txy
. In
other words, an antisymmetric torsion of 3+1 spacetime can be rep-
resented by a four-vector,
a
=
abcd
bcd
.
5.8.3 Experimental searches for torsion
One way of stating the equivalence principle (see p. 118) is that
it forbids spacetime from coming equipped with a vector eld that
could be measured by free-falling observers, i.e., observers in local
Lorentz frames. A variety of high-precision tests of the equivalence
principle have been carried out. From the point of view of an ex-
perimenter doing this kind of test, it is important to distinguish
between elds that are built in to spacetime and those that live
in spacetime. For example, the existence of the earths magnetic
eld does not violate the equivalence principle, but if an experi-
ment was sensitive to the earths eld, and the experimenter didnt
know about it, there would appear to be a violation. Antisymmet-
ric torsion in four dimensions acts like a vector. If it constitutes
a universal background eect built into spacetime, then it violates
the equivalence principle. If it instead arises from specic material
sources, then it may still show up as a measurable eect in exper-
imental tests designed to detect Lorentz-invariance. Lets consider
the latter possibility.
Since curvature in general relativity comes from mass and en-
ergy, as represented by the stress-energy tensor T
ab
, we could ask
what would be the sources of torsion, if it exists in our universe.
The source cant be the rank-2 stress-energy tensor. It would have
to be an odd-rank tensor, i.e., a quantity that is odd under PT, and
in theories that include torsion it is commonly assumed that the
source is the quantum-mechanical angular momentum of subatomic
particles. If this is the case, then torsion eects are expected to be
proportional to G, the product of Plancks constant and the gravi-
tational constant, and they should therefore be extremely small and
d / The University of Wash-
ington torsion pendulum used to
search for torsion. The light gray
wedges are Alnico, the darker
ones SmCo
5
. The arrows with
the lled heads represent the
directions of the electron spins,
with denser arrows indicating
higher polarization. The arrows
with the open heads show the
direction of the B eld. University
of Washington Eot-Wash group,
http://www.npl.washington.
edu/eotwash/publications/
pdf/lowfrontier2.pdf
hard to measure. String theory, for example, includes torsion, but
nobody has found a way to test string theory empirically because it
essentially makes predictions about phenomena at the Planck scale,
_
G/c
3
10
35
m, where both gravity and quantum mechanics
are strong eects.
There are, however, some high-precision experiments that have
a reasonable chance of detecting whether our universe has torsion.
Torsion violates the equivalence principle, and by the turn of the
century tests of the equivalence principle had reached a level of
precision sucient to rule out some models that include torsion.
Figure d shows a torsion pendulum used in an experiment by the
Eot-Wash group at the University of Washington.
8
If torsion exists,
then the intrinsic spin of an electron should have an energy ,
where is the spacelike part of the torsion vector. The torsion
could be generated by the earth, the sun, or some other object at a
greater distance. The interaction will modify the behavior of a
torsion pendulum if the spins of the electrons in the pendulum are
polarized nonrandomly, as in a magnetic material. The pendulum
will tend to precess around the axis dened by .
This type of experiment is extremely dicult, because the pen-
dulum tends to act as an ultra-sensitive magnetic compass, resulting
in a measurement of the ambient magnetic eld rather than the hy-
pothetical torsion eld . To eliminate this source of systematic
error, the UW group rst eliminated the ambient magnetic eld
as well as possible, using mu-metal shielding and Helmholtz coils.
They also constructed the pendulum out of a combination of two
magnetic materials, Alnico 5 and SmCo
5
, in such a way that the
magnetic dipole moment vanished, but the spin dipole moment did
not; Alnico 5s magnetic eld is due almost entirely to electron spin,
whereas the magnetic eld of SmCo
5
contains signicant contribu-
tions from orbital motion. The result was a nonmagnetic object
whose spins were polarized. After four years of data collection, they
found [[ 10
21
eV. Models that include torsion typically predict
such eects to be of the order of m
2
e
/m
P
10
17
eV, where m
e
is
the mass of the electron and m
P
=
_
c/G 10
19
GeV 20 g is
the Planck mass. A wide class of these models is therefore ruled out
by these experiments.
Since there appears to be no experimental evidence for the exis-
tence of gravitational torsion in our universe, we will assume from
now on that it vanishes identically. Einstein made the same as-
sumption when he originally created general relativity, although he
and Cartan later tinkered with non-torsion-free theories in a failed
attempt to unify gravity with electromagnetism. Some models that
include torsion remain viable. For example, it has been argued that
the torsion tensor should fall o quickly with distance from the
8
http://www.npl.washington.edu/eotwash/publications/pdf/lowfrontier2.pdf
source.
9
9
Carroll and Field, http://arxiv.org/abs/gr-qc/9403058
5.9 From metric to curvature
5.9.1 Finding given g
Weve already found the Christoel symbol in terms of the metric
in one dimension. Expressing it in tensor notation, we have
d
ba
=
1
2
g
cd
(
?
g
??
) ,
where inversion of the one-component matrix G has been replaced
by matrix inversion, and, more importantly, the question marks indi-
cate that there would be more than one way to place the subscripts
so that the result would be a grammatical tensor equation. The
most general form for the Christoel symbol would be
b
ac
=
1
2
g
db
(L
c
g
ab
+M
a
g
cb
+N
b
g
ca
) ,
where L, M, and N are constants. Consistency with the one-
dimensional expression requires L + M + N = 1, and vanishing
torsion gives L = M. The L and M terms have a dierent physical
signicance than the N term.
Suppose an observer uses coordinates such that all objects are
described as lengthening over time, and the change of scale accu-
mulated over one day is a factor of k > 1. This is described by the
derivative
t
g
xx
< 1, which aects the M term. Since the metric is
used to calculate squared distances, the g
xx
matrix element scales
down by 1/
k. To compensate for
t
v
x
< 0, so we need to add a
positive correction term, M > 0, to the covariant derivative. When
the same observer measures the rate of change of a vector v
t
with
respect to space, the rate of change comes out to be too small, be-
cause the variable she dierentiates with respect to is too big. This
requires N < 0, and the correction is of the same size as the M
correction, so [M[ = [N[. We nd L = M = N = 1.
Self-check: Does the above argument depend on the use of space
for one coordinate and time for the other?
The resulting general expression for the Christoel symbol in
terms of the metric is
c
ab
=
1
2
g
cd
(
a
g
bd
+
b
g
ad
d
g
ab
) .
One can readily go back and check that this gives
c
g
ab
= 0. In fact,
the calculation is a bit tedious. For that matter, tensor calculations
in general can be infamously time-consuming and error-prone. Any
reasonable person living in the 21st century will therefore resort to
a computer algebra system. The most widely used computer alge-
bra system is Mathematica, but its expensive and proprietary, and
it doesnt have extensive built-in facilities for handling tensors. It
turns out that there is quite a bit of free and open-source tensor soft-
ware, and it falls into two classes: coordinate-based and coordinate-
independent. The best open-source coordinate-independent facil-
ity available appears to be Cadabra, and in fact the verication of
c
g
ab
= 0 is the rst example given in the Leo Brewins handy guide
to applications of Cadabra to general relativity.
10
Self-check: In the case of 1 dimension, show that this reduces to
the earlier result of (1/2)dG/dX.
Since is not a tensor, it is not obvious that the covariant deriva-
tive, which is constructed from it, is a tensor. But if it isnt obvious,
neither is it surprising the goal of the above derivation was to get
results that would be coordinate-independent.
Christoffel symbols on the globe, quantitatively Example: 9
In example 8 on page 145, we inferred the following properties
for the Christoffel symbol
on a sphere of radius R:
is
independent of and R,
< 0 in the northern hemisphere

(colatitude less than /2),
= 0 on the equator, and
>
0 in the southern hemisphere.
The metric on a sphere is ds
2
= R
2
d
2
+ R
2
sin
2
d
2
. The only
nonvanishing term in the expression for
is the one involving
= 2R
2
sin cos . The result is
= sin cos , which

can be veried to have the properties claimed above.
5.9.2 Numerical solution of the geodesic equation
On page 148 I gave an algorithm that demonstrated the unique-
ness of the solutions to the geodesic equation. This algorithm can
also be used to nd geodesics in cases where the metric is known.
The following program, written in the computer language Python,
carries out a very simple calculation of this kind, in a case where
we know what the answer should be; even without any previous
familiarity with Python, it shouldnt be dicult to see the corre-
spondence between the abstract algorithm presented on page 148
and its concrete realization below. For polar coordinates in a Eu-
clidean plane, one can compute
r
= r and
r
= 1/r (problem
1, page 162). Here we compute the geodesic that starts out tangent
to the unit circle at = 0.
1 import math
2
3 l = 0 # affine parameter lambda
4 dl = .001 # change in l with each iteration
5 l_max = 100.
6
7 # initial position:
10
Section 5.9 From metric to curvature 157
8 r=1
9 phi=0
10 # initial derivatives of coordinates w.r.t. lambda
11 vr = 0
12 vphi = 1
13
14 k = 0 # keep track of how often to print out updates
15 while l<l_max:
16 l = l+dl
17 # Christoffel symbols:
18 Grphiphi = -r
19 Gphirphi = 1/r
20 # second derivatives:
21 ar = -Grphiphi*vphi*vphi
22 aphi = -2.*Gphirphi*vr*vphi
23 # ... factor of 2 because Gâ_{bc}=Gâ_{cb} and b
24 # is not the same as c
25 # update velocity:
26 vr = vr + dl*ar
27 vphi = vphi + dl*aphi
28 # update position:
29 r = r + vr*dl
30 phi = phi + vphi*dl
31 if k%10000==0: # k is divisible by 10000
32 phi_deg = phi*180./math.pi
33 print "lambda=%6.2f r=%6.2f phi=%6.2f deg." % (l,r,phi_deg)
34 k = k+1
It is not necessary to worry about all the technical details of the
language (e.g., line 1, which makes available such conveniences as
math.pi for ). Comments are set o by pound signs. Lines 16-34
are indented because they are all to be executed repeatedly, until it
is no longer true that <
max
(line 15).
Self-check: By inspecting lines 18-22, nd the signs of r and

at = 0. Convince yourself that these signs are what we expect

geometrically.
The output is as follows:
1 lambda= 0.00 r= 1.00 phi= 0.06 deg.
2 lambda= 10.00 r= 10.06 phi= 84.23 deg.
3 lambda= 20.00 r= 20.04 phi= 87.07 deg.
4 lambda= 30.00 r= 30.04 phi= 88.02 deg.
5 lambda= 40.00 r= 40.04 phi= 88.50 deg.
6 lambda= 50.00 r= 50.04 phi= 88.78 deg.
7 lambda= 60.00 r= 60.05 phi= 88.98 deg.
8 lambda= 70.00 r= 70.05 phi= 89.11 deg.
9 lambda= 80.00 r= 80.06 phi= 89.21 deg.
10 lambda= 90.00 r= 90.06 phi= 89.29 deg.
We can see that 90 deg. as , which makes sense,
because the geodesic is a straight line parallel to the y axis.
A less trivial use of the technique is demonstrated on page 6.2.7,
where we calculate the deection of light rays in a gravitational eld,
one of the classic observational tests of general relativity.
5.9.3 The Riemann tensor in terms of the Christoffel symbols
The covariant derivative of a vector can be interpreted as the rate
of change of a vector in a certain direction, relative to the result of
parallel-transporting the original vector in the same direction. We
can therefore see that the denition of the Riemann curvature tensor
on page 137 is a measure of the failure of covariant derivatives to
commute:
(
a
a
)A
c
= A
d
R
c
dab
A tedious calculation now gives R in terms of the s:
R
a
bcd
=
c
a
db
a
cb
+
a
ce
e
db
a
de
e
cb
This is given as another example later in Brewins manual for apply-
ing Cadabra to general relativity.
11
(Brewin writes the upper index
in the second slot of R.)
5.9.4 Some general ideas about gauge
Lets step back now for a moment and try to gain some physi-
cal insight by looking at the features that the electromagnetic and
relativistic gauge transformations have in common. We have the
following analogies:
11
a / The Aharonov-Bohm ef-
fect. An electron enters a beam
splitter at P, and is sent out in
two different directions. The two
parts of the wave are reected so
that they reunite at Q. The arrows
represent the vector potential A.
The observable magnetic eld
B is zero everywhere outside
the solenoid, and yet the inter-
ference observed at Q depends
on whether the eld is turned
on. See page 114 for further
discussion of the A and B elds
of a solenoid.
electromagnetism dierential
geometry
global symmetry A constant phase
shift has no ob-
servable eects.
Adding a con-
stant onto a
coordinate has no
observable eects.
local symmetry A phase shift
that varies from
point to point has
no observable ef-
fects.
An arbitrary co-
ordinate transfor-
mation has no ob-
servable eects.
The gauge is de-
scribed by . . . g
. . . and dierentia-
tion of this gives the
gauge eld. . . A
b

c
ab
A second dieren-
tiation gives the
directly observable
eld(s) . . .
E and B R
c
dab
The interesting thing here is that the directly observable elds
do not carry all of the necessary information, but the gauge elds are
not directly observable. In electromagnetism, we can see this from
the Aharonov-Bohm eect, shown in gure a.
12
The solenoid has
B = 0 externally, and the electron beams only ever move through
the external region, so they never experience any magnetic eld. Ex-
periments show, however, that turning the solenoid on and o does
change the interference between the two beams. This is because the
vector potential does not vanish outside the solenoid, and as weve
seen on page 114, the phase of the beams varies according to the
path integral of the A
b
. We are therefore left with an uncomfort-
able, but unavoidable, situation. The concept of a eld is supposed
to eliminate the need for instantaneous action at a distance, which
is forbidden by relativity; that is, (1) we want our elds to have only
local eects. On the other hand, (2) we would like our elds to be
directly observable quantities. We cannot have both 1 and 2. The
gauge eld satises 1 but not 2, and the electromagnetic elds give
2 but not 1.
12
We describe the eect here in terms of an idealized, impractical experiment.
For the actual empirical status of the Aharonov-Bohm eect, see Batelaan and
Tonomura, Physics Today 62 (2009) 38.
b / The cone has zero intrin-
sic curvature everywhere except
at its tip. An observer who never
visits the tip can nevertheless
detect its existence, because
parallel transport around a path
that encloses the tip causes a
vector to change its direction.
Figure b shows an analog of the Aharonov-Bohm experiment in
dierential geometry. Everywhere but at the tip, the cone has zero
curvature, as we can see by cutting it and laying it out at. But even
an observer who never visits the tightly curved region at the tip can
detect its existence, because parallel-transporting a vector around
a closed loop can change the vectors direction, provided that the
loop surrounds the tip.
In the electromagnetic example, integrating A around a closed
loop reveals, via Stokes theorem, the existence of a magnetic ux
through the loop, even though the magnetic eld is zero at every
location where A has to be sampled. In the relativistic example,
integrating around a closed loop shows that there is curvature
inside the loop, even though the curvature is zero at all the places
where has to be sampled.
The fact that is a gauge eld, and therefore not locally ob-
servable, is simply a fancy way of expressing the ideas introduced
on pp. 144 and 146, that due to the equivalence principle, the gravi-
tational eld in general relativity is not locally observable. This non-
observability is local because the equivalence principle is a statement
about local Lorentz frames. The example in gure b is non-local.
Geodetic effect and structure of the source Example: 10
In section 5.5.1 on page 139, we estimated the geodetic effect
on Gravity Probe B and found a result that was only off by a factor
of 3. The mathematically pure form of the 3 suggests that the
geodetic effect is insensitive to the distribution of mass inside the
earth. Why should this be so?
The change in a vector upon parallel transporting it around a
closed loop can be expressed in terms of either (1) the area inte-
gral of the curvature within the loop or (2) the line integral of the
Christoffel symbol (essentially the gravitational eld) on the loop
itself. Although I expressed the estimate as 1, it would have been
equally valid to use 2. By Newtons shell theorem, the gravita-
tional eld is not sensitive to anything about its mass distribution
other than its near spherical symmetry. The earth spins, and this
does affect the stress-energy tensor, but since the velocity with
which it spins is everywhere much smaller than c, the resulting
effect, called frame-dragging, is much smaller.
Problems
1 Show, as claimed on page 157, that for polar coordinates in
a Euclidean plane,
r
= r and
r
= 1/r.
2 Partial derivatives commute with partial derivatives. Co-
variant derivatives dont commute with covariant derivatives. Do
covariant derivatives commute with partial derivatives?
3 Show that if the dierential equation for geodesics on page
146 is satised for one ane parameter , then it is also satised for
any other ane parameter
= a+b, where a and b are constants.

4 Equation [2] on page 91 gives a at-spacetime metric in
rotating polar coordinates. (a) Verify by explicit computation that
this metric represents a at spacetime. (b) Reexpress the metric in
rotating Cartesian coordinates, and check your answer by verifying
that the Riemann tensor vanishes.
5 The purpose of this problem is to explore the diculties
inherent in nding anything in general relativity that represents a
uniform gravitational eld g. In example 7 on page 46, we found,
based on elementary arguments about the equivalence principle and
photons in elevators, that gravitational time dilation must be given
by e
, where = gz is the gravitational potential. This results in

a metric
[1] ds
2
= e
2gz
dt
2
dz
2
.
On the other hand, example 14 on page 116 derived the metric
[2] ds
2
= (1 +gz)
2
dt
2
dz
2
.
by transforming from a Lorentz frame to a frame whose origin moves
with constant proper acceleration g. (These are known as Rindler
coordinates.) Prove the following facts. None of the calculations
are so complex as to require symbolic math software, so you might
want to perform them by hand rst, and then check yourself on a
computer.
(a) The metrics [1] and [2] are approximately consistent with one
another for z near 0.
(b) When a test particle is released from rest in either of these met-
rics, its initial proper acceleration is g.
(c) The two metrics are not exactly equivalent to one another under
any change of coordinates.
(d) Both spacetimes are uniform in the sense that the curvature is
constant. (In both cases, this can be proved without an explicit
computation of the Riemann tensor.)
Remark: The incompatibility between [1] and [2] can be interpreted as showing
that general relativity does not admit any spacetime that has all the global
properties we would like for a uniform gravitational eld. This is related to Bells
spaceship paradox (example 11, p. 51). Some further properties of the metric [1]
are analyzed in subsection 7.1.3 on page 195. Solution, p. 264
Problems 163
a / A Swiss commemorative
coin shows the vacuum eld
equation.
Chapter 6
Vacuum Solutions
In this chapter we investigate general relativity in regions of space
that have no matter to act as sources of the gravitational eld.
We will not, however, limit ourselves to calculating spacetimes in
cases in which the entire universe has no matter. For example,
we will be able to calculate general-relativistic eects in the region
surrounding the earth, including a full calculation of the geodetic
eect, which was estimated in section 5.5.1 only to within an order
of magnitude. We can have sources, but we just wont describe the
metric in the regions where the sources exist, e.g., inside the earth.
The advantage of accepting this limitation is that in regions of empty
space, we dont have to worry about the details of the stress-energy
tensor or how it relates to curvature. As should be plausible based
on the physical motivation given in section 5.1, page 130, the eld
equations in a vacuum are simply R
ab
= 0.
6.1 Event horizons
One seemingly trivial way to generate solutions to the eld equations
in vacuum is simply to start with a at Lorentzian spacetime and do
a change of coordinates. This might seem pointless, since it would
simply give a new description (and probably a less convenient and
descriptive one) of the same old, boring, at spacetime. It turns
out, however, that some very interesting things can happen when
we do this.
6.1.1 The event horizon of an accelerated observer
Consider the uniformly accelerated observer described in exam-
ples 3 on page 105 and 14 on page 116. Recalling these earlier results,
we have for the ships equation of motion in an inertial frame
x =
1
a
_
_
1 +a
2
t
2
1
_
,
and for the metric in the ships frame
g
t
= (1 +ax
)
2
g
x
= 1 .
Since this metric was derived by a change of coordinates from a at-
space metric, and the Ricci curvature is an intrinsic property, we
165
a / A spaceship (curved world-
line) moves with an acceleration
perceived as constant by its
passengers. The photon (straight
world-line) come closer and
closer to the ship, but will never
quite catch up.
expect that this one also has zero Ricci curvature. This is straight-
forward to verify. The nonvanishing Christoel symbols are
t
=
a
1 +ax
and
x
t
= a(1 +ax
) .
The only elements of the Riemann tensor that look like they might
be nonzero are R
t
and R
x
, but both of these in fact vanish.

Self-check: Verify these facts.
This seemingly routine exercise now leads us into some very in-
teresting territory. Way back on page 12, we conjectured that not all
events could be time-ordered: that is, that there might exists events
in spacetime 1 and 2 such that 1 cannot cause 2, but neither can 2
cause 1. We now have enough mathematical tools at our disposal
to see that this is indeed the case.
We observe that x(t) approaches the asymptote x = t1/a. This
asymptote has a slope of 1, so it can be interpreted as the world-line
of a photon that chases the ship but never quite catches up to it.
Any event to the left of this line can never have a causal relationship
with any event on the ships world-line. Spacetime, as seen by an
observer on the ship, has been divided by a curtain into two causally
disconnected parts. This boundary is called an event horizon. Its
existence is relative to the world-line of a particular observer. An
observer who is not accelerating along with the ship does consider
an event horizon to exist. Although this particular example of the
indenitely accelerating spaceship has some physically implausible
features (e.g., the ship would have to run out of fuel someday), event
horizons are real things. In particular, we will see in section 6.3.2
that black holes have event horizons.
Interpreting everything in the (t
, x
) coordinates tied to the ship,

the metrics component g
vanishes at x
= 1/a. An observer
aboard the ship reasons as follows. If I start out with a head-start
of 1/a relative to some event, then the timelike part of the metric at
that event vanishes. If the event marks the emission of a material
particle, then there is no possible way for that particles world-line
to have ds
2
> 0. If I were to detect a particle emitted at that event,
it would violate the laws of physics, since material particles must
have ds
2
> 0, so I conclude that I will never observe such a particle.
Since all of this applies to any material particle, regardless of its
mass m, it must also apply in the limit m 0, i.e., to photons and
other massless particles. Therefore I can never receive a particle
emitted from this event, and in fact it appears that there is no way
for that event, or any other event behind the event horizon, to have
any eect on me. In my frame of reference, it appears that light
cones near the horizon are tipped over so far that their future light-
cones lie entirely in the direction away from me.
Weve already seen in example 10 on page 50 that a naive New-
tonian argument suggests the existence of black holes; if a body is
166 Chapter 6 Vacuum Solutions
suciently compact, light cannot escape from it. In a relativistic
treatment, this should be described as an event horizon.
6.1.2 Information paradox
The existence of event horizons in general relativity has deep
implications, and in particular it helps to explain why it is so di-
cult to reconcile general relativity with quantum mechanics, despite
nearly a century of valiant attempts. Quantum mechanics has a
property called unitarity. Mathematically, this says that of the state
of a quantum mechanical system is given, at a certain time, in the
form of a vector, then its state at some point in the future can be
predicted by applying a unitary matrix to that vector. A unitary
matrix is the generalization to complex numbers of the ordinary
concept of an orthogonal matrix, and essentially it just represents a
change of basis, in which the basis vectors have unit length and are
perpendicular to one another.
To see what this means physically, consider the following nonex-
amples. The matrix
_
1 0
0 0
_
is not unitary, because its rows and columns are not orthogonal vec-
tors with unit lengths. If this matrix represented the time-evolution
of a quantum mechanical system, then its meaning would be that
any particle in state number 1 would be left alone, but any particle
in state 2 would disappear. Any information carried by particles in
state 2 is lost forever and can never be retrieved. This also violates
the time-reversal symmetry of quantum mechanics.
Another nonunitary matrix is:
_
1 0
0

2
_
Here, any particle in state 2 is increased in amplitude by a factor of
2, meaning that it is doubled in probability. That is, the particle

is cloned. This is the opposite problem compared to the one posed
by the rst matrix, and it is equally problematic in terms of time-
reversal symmetry and conservation of information. Actually, if we
could clone a particle in this way, it would violate the Heisenberg
uncertainty principle. We could make two copies of the particle,
and then measure the position of one copy and the momentum of
the other, each with unlimited precision. This would violate the
uncertainty principle, so we believe that it cannot be done. This is
known as the no-cloning theorem.
The existence of event horizons in general relativity violates uni-
tarity, because it allows information to be destroyed. If a particle is
thrown behind an event horizon, it can never be retrieved.
Section 6.1 Event horizons 167
b / Bill Unruh (1945-).
6.1.3 Radiation from event horizons
In interesting twist on the situation was introduced by Bill Un-
ruh in 1976. Observer B aboard the accelerating spaceship believes
in the equivalence principle, so she knows that the local proper-
ties of space at the event horizon would seem entirely normal and
Lorentzian to a local observer A. (The same applies to a black holes
horizon.) In particular, B knows that A would see pairs of virtual
particles being spontaneously created and destroyed in the local vac-
uum. This is simply a manifestation of the time-energy form of the
uncertainty principle, Et h. Now suppose that a pair of par-
ticles is created, but one is created in front of the horizon and one
behind it. To A these are virtual particles that will have to be an-
nihilated within the time t, but according to B the one created in
front of the horizon will eventually catch up with the spaceship, and
can be observed there, although it will be red-shifted. The amount
of redshift is given by
_
g
=
_
(1 +ax
)
2
. Say the pair is created
right near the horizon, at x
= 1/a. By the uncertainty princi-

ple, each of the two particles is spread out over a region of space
of size x
. Since these are photons, which travel at the speed of

light, the uncertainty in position is essentially the same as the un-
certainty in time. The forward-going photons redshift comes out
to be ax
= at
, which by the uncertainty principle should be at

least ha/E, so that when the photon is observed by B, its energy is
E(ha/E) = ha.
Now B sees a uniform background of photons, with energies of
around ha, being emitted randomly from the horizon. They are
being emitted from empty space, so it seems plausible to believe
that they dont encode any information at all; they are completely
random. A surface emitting a completely random (i.e., maximum-
entropy) hail of photons is a black-body radiator, so we expect that
the photons will have a black-body spectrum, with its peak at an
energy of about ha. This peak is related to the temperature of
the black body by E kT, where k is Boltzmanns constant. We
conclude that the horizon acts like a black-body radiator with a
temperature T ha/k. The more careful treatment by Unruh shows
that the exact relation is T = ha/4
2
k, or ha/4
2
kc in SI units.
An important observation here is that not only do dierent ob-
servers disagree about the number of quanta that are present (which
is true in the case of ordinary Doppler shifts), but about the number
of quanta in the vacuum as well. B sees photons that according to
A do not exist.
Lets consider some real-world examples of large accelerations:
a / The eld equations of general
relativity are nonlinear.
acceleration
(m/s
2
)
temperature of
horizon (K)
bullet red from a gun 10
3
10
17
electron in a CRT 10
7
10
13
plasmas produced by intense
laser pulses
10
21
10
proton in a helium nucleus 10
27
10
8
To detect Unruh radiation experimentally, we would ideally like to
be able to accelerate a detector and let it detect the radiation. This
is clearly impractical. The third line shows that it is possible to
impart very large linear accelerations to subatomic particles, but
then one can only hope to infer the eect of the Unruh radiation
indirectly by its eect on the particles. As shown on the nal line,
examples of extremely large nonlinear accelerations are not hard to
nd, but the interpretation of Unruh radiation for nonlinear motion
is unclear. A summary of the prospects for direct experimental de-
tection of this eect is given by Rosu.
1
This type of experiment is
clearly extremely dicult, but it is one of the few ways in which one
could hope to get direct empirical insight, under controlled condi-
tions, into the interface between gravity and quantum mechanics.
6.2 The Schwarzschild metric
We now set ourselves the goal of nding the metric describing the
static spacetime outside a spherically symmetric, nonrotating, body
of mass m. This problem was rst solved by Karl Schwarzschild
in 1915.
2
One byproduct of nding this metric will be the ability
to calculate the geodetic eect exactly, but it will have more far-
reaching consequences, including the existence of black holes.
The problem we are solving is similar to calculating the spher-
ically symmetric solution to Gausss law in a vacuum. The solu-
tion to the electrical problem is of the form r/r
2
, with an arbitrary
constant of proportionality that turns out to be proportional to
the charge creating the eld. One big dierence, however, is that
whereas Gausss law is linear, the equation R
ab
= 0 is highly non-
linear, so that the solution cannot simply be scaled up and down in
proportion to m.
The reason for this nonlinearity is fundamental to general rela-
tivity. For example, when the earth condensed out of the primordial
solar nebula, large amounts of heat were produced, and this energy
was then gradually radiated into outer space, decreasing the total
mass of the earth. If we pretend, as in gure a, that this process
involved the merging of only two bodies, each with mass m, then
1
http://xxx.lanl.gov/abs/gr-qc/9605032
2
On the gravitational eld of a point mass according to Einsteins the-
ory, Sitzungsberichte der K oniglich Preussischen Akademie der Wissenschaften
1 (1916) 189. An English translation is available at http://arxiv.org/abs/
physics/9905030v1.
Section 6.2 The Schwarzschild metric 169
the net result was essentially to take separated masses m and m at
rest, and bring them close together to form close-neighbor masses
m and m, again at rest. The amount of energy radiated away was
proportional to m
2
, so the gravitational mass of the combined sys-
tem has been reduced from 2m to 2m(. . .)m
2
, where . . . is roughly
G/c
2
r. There is a nonlinear dependence of the gravitational eld on
the masses.
Self-check: Verify that any constant metric (including a metric
with the wrong signature, e.g., 2+2 dimensions rather than 3+1)
is a solution to the Einstein eld equation in vacuum.
The correspondence principle tells us that our result must have
a Newtonian limit, but the only variables involved are m and r, so
this limit must be the one in which r/m is large. Large compared
to what? There is nothing else available with which to compare,
so it can only be large compared to some expression composed of
the unitless constants G and c. We have already chosen units such
that c = 1, and we will now set G = 1 as well. Mass and distance
are now comparable, with the conversion factor being G/c
2
= 7
10
28
m/kg, or about a mile per solar mass. Since the earths radius
is thousands of times more than a mile, and its mass hundreds of
thousands of times less than the suns, its r/m is very large, and
the Newtonian approximation is good enough for all but the most
precise applications, such as the GPS network or the Gravity Probe
B experiment.
6.2.1 The zero-mass case
First lets demonstrate the trivial solution with at spacetime.
In spherical coordinates, we have
ds
2
= dt
2
dr
2
r
2
d
2
r
2
sin
2
d
2
.
The nonvanishing Christoel symbols (ignoring swaps of the lower
indices) are:
r
=
1
r
r
=
1
r
= r
= r sin
2
= sin cos
= cot
Self-check: If wed been using the ( + ++) metric instead of
(+), what would have been the eect on the Christoel sym-
bols? What if wed expressed the metric in dierent units, rescaling
all the coordinates by a factor k?
Use of ctensor
In fact, when I calculated the Christoel symbols above by hand,
I got one of them wrong, and missed calculating one other because I
thought it was zero. I only found my mistake by comparing against
a result in a textbook. The computation of the Riemann tensor is
an even bigger mess. Its clearly a good idea to resort to a com-
puter algebra system here. Cadabra, which was discussed earlier, is
specically designed for coordinate-independent calculations, so it
wont help us here. A good free and open-source choice is ctensor,
which is one of the standard packages distributed along with the
computer algebra system Maxima, introduced on page 61.
The following Maxima program calculates the Christoel sym-
bols found in section 6.2.1.
1 load(ctensor);
2 ct_coords:[t,r,theta,phi];
3 lg:matrix([1,0,0,0],
4 [0,-1,0,0],
5 [0,0,-r^2,0],
6 [0,0,0,-r^2*sin(theta)^2]);
7 cmetric();
8 christof(mcs);
Line 1 loads the ctensor package. Line 2 sets up the names of the
coordinates. Line 3 denes the g
ab
, with lg meaning the version of
g with lower indices. Line 7 tells Maxima to do some setup work
with g
ab
, including the calculation of the inverse matrix g
ab
, which
is stored in ug. Line 8 says to calculate the Christoel symbols.
The notation mcs refers to the tensor
a
bc
with the indices swapped
around a little compared to the convention
a
bc
followed in this
book. On a Linux system, we put the program in a le flat.mac
and run it using the command maxima -b flat.mac. The relevant
part of the output is:
1 1
2 (%t6) mcs = -
3 2, 3, 3 r
4
5 1
6 (%t7) mcs = -
7 2, 4, 4 r
8
9 (%t8) mcs = - r
10 3, 3, 2
11
12 cos(theta)
13 (%t9) mcs = ----------
14 3, 4, 4 sin(theta)
15
16 2
17 (%t10) mcs = - r sin (theta)
18 4, 4, 2
19
20 (%t11) mcs = - cos(theta) sin(theta)
21 4, 4, 3
Adding the command ricci(true); at the end of the program re-
sults in the output THIS SPACETIME IS EMPTY AND/OR FLAT, which
saves us hours of tedious computation. The tensor ric (which here
happens to be zero) is computed, and all its nonzero elements are
printed out. There is a similar command riemann(true); to com-
pute the Riemann rensor riem. This is stored so that riem[i,j,k,l]
is what we would call R
l
ikj
. Note that l is moved to the end, and j
and k are also swapped.
6.2.2 Geometrized units
If the mass creating the gravitational eld isnt zero, then we
need to decide what units to measure it in. It has already proved
very convenient to adopt units with c = 1, and we will now also set
the gravitational constant G = 1. Previously, with only c set to 1,
the units of time and length were the same, [T] = [L], and so were
the units of mass and energy, [M] = [E]. With G = 1, all of these
become the same units, [T] = [L] = [M] = [E].
Self-check: Verify this statement by combining Newtons law of
gravity with Newtons second law of motion.
The resulting system is referred to as geometrized, because units
like mass that had formerly belonged to the province of mechanics
are now measured using the same units we would use to do geometry.
6.2.3 A large-r limit
Now lets think about how to tackle the real problem of nding
the non-at metric. Although general relativity lets us pick any co-
ordinates we like, the spherical symmetry of the problem suggests
using coordinates that exploit that symmetry. The at-space coor-
dinates and can stil be dened in the same way, and they have
the same interpretation. For example, if we drop a test particle
toward the mass from some point in space, its world-line will have
constant and . The r coordinate is a little dierent. In curved
spacetime, the circumference of a circle is not equal to 2 times the
distance from the center to the circle; in fact, the discrepancy be-
tween these two is essentially the denition of the Ricci curvature.
This gives us a choice of two logical ways to dene r. Well de-
ne it as the circumference divided by 2, which has the advantage
that the last two terms of the metric are the same as in at space:
r
2
d
2
r
2
sin
2
d
2
. Since were looking for static solutions, none
of the elements of the metric can depend on t. Also, the solution
is going to be symmetric under t t, , and , so
we cant have any o-diagonal elements. The result is that we have
narrowed the metric down to something of the form
ds
2
= h(r)dt
2
k(r)dr
2
r
2
d
2
r
2
sin
2
d
2
,
where both h and k approach 1 for r , where spacetime is at.
For guidance in how to construct h and k, lets consider the
acceleration of a test particle at r m, which we know to be
m/r
2
, since nonrelativistic physics applies there. We have
t
v
r
=
t
v
r
+
r
tc
v
c
.
An observer free-falling along with the particle observes its acceler-
ation to be zero, and a tensor that is zero in one coordinate system
is zero in all others. Since the covariant derivative is a tensor, we
conclude that
t
v
r
= 0 in all coordinate systems, including the
(t, r, . . .) system were using. If the particle is released from rest,
then initially its velocity four-vector is (1, 0, 0, 0), so we nd that its
acceleration in (t, r) coordinates is
r
tt
=
1
2
g
rr
r
g
tt
=
1
2
h
/k.
Setting this equal to m/r
2
, we nd h
/k = 2m/r
2
for r m.
Since k 1 for large r, we have
h

2m
r
2
for r m .
The interpretation of this calculation is as follows. We assert the
equivalence principle, by which the acceleration of a free-falling par-
ticle can be said to be zero. After some calculations, we nd that
the rate at which time ows (encoded in h) is not constant. It is
dierent for observers at dierent heights in a gravitational poten-
tial well. But this is something we had already deduced, without
the tensor gymnastics, in example 4 on page 106.
Integrating, we nd that for large r, h = 1 2m/r.
6.2.4 The complete solution
A series solution
Weve learned some interesting things, but we still have an ex-
tremely nasty nonlinear dierential equation to solve. One way to
attack a dierential equation, when you have no idea how to pro-
ceed, is to try a series solution. We have a small parameter m/r to
expand around, so lets try to write h and k as series of the form
h =
n=0
a
k
_
m
r
_
n
k =
n=0
b
k
_
m
r
_
n
We already know a
0
, a
1
, and b
0
. Lets try to nd b
1
. In the
following Maxima code I omit the factor of m in h
1
for convenience.
In other words, were looking for the solution for m = 1.
1 load(ctensor);
3 lg:matrix([(1-2/r),0,0,0],
4 [0,-(1+b1/r),0,0],
5 [0,0,-r^2,0],
6 [0,0,0,-r^2*sin(theta)^2]);
7 cmetric();
8 ricci(true);
I wont reproduce the entire output of the Ricci tensor, which
is voluminous. We want all four of its nonvanishing components to
vanish as quickly as possible for large values of r, so I decided to
ddle with R
t
t, which looked as simple as any of them. It appears
to vary as r
4
for large r, so lets evaluate lim
r
_
r
4
R
tt
_
:
9 limit(r^4*ric[1,1],r,inf);
The result is (b
1
2)/2, so lets set b
1
= 2. The approximate solution
weve found so far (reinserting the ms),
ds
2
_
1
2m
r
_
dt
2
_
1 +
2m
r
_
dr
2
r
2
d
2
r
2
sin
2
d
2
,
was rst derived by Einstein in 1915, and he used it to solve the
problem of the non-Keplerian relativistic correction to the orbit of
Mercury, which was one of the rst empirical tests of general rela-
tivity.
Continuing in this fashion, the results are as follows:
a
0
= 1 b
0
= 1
a
1
= 2 b
1
= 2
a
2
= 0 b
2
= 4
a
3
= 0 b
2
= 8
The closed-form solution
The solution is unexpectedly simple, and can be put into closed
form. The approximate result we found for h was in fact exact.
For k we have a geometric series 1/(1 2/r), and when we reinsert
the factor of m in the only way that makes the units work, we get
1/(1 2m/r). The result for the metric is
ds
2
=
_
1
2m
r
_
dt
2
_
1
1 2m/r
_
dr
2
r
2
d
2
r
2
sin
2
d
2
.
A quick calculation in Maxima demonstrates that this is an exact
solution for all r, i.e., the Ricci tensor vanishes everywhere, even at
r < 2m, which is outside the radius of convergence of the geometric
series.
6.2.5 Geodetic effect
As promised in section 5.5.1, we now calculate the geodetic eect
on Gravity Probe B, including all the niggling factors of 3 and . To
make the physics clear, we approach the actual calculation through
a series of warmups.
Flat space
As a rst warmup, consider two spatial dimensions, represented
by Euclidean polar coordinates (r, ). Parallel-transport of a gyro-
scopes angular momentum around a circle of constant r gives
= 0
L
r
= 0 .
Computing the covariant derivatives, and we have
0 =
r
L
r
0 =
L
r
+
r
.
The Christoel symbols are
r
= 1/r and
r
= r. This is
all made to look needlessly complicated because L
and L
r
are ex-
pressed in dierent units. Essentially the vector is staying the same,
but were expressing it in terms of basis vectors in the r and di-
rections that are rotating. To see this more transparently, let r = 1,
and write P for L
and Q for L
r
, so that
P
= Q
Q
= P ,
which have solutions such as P = sin , Q = cos . For each orbit
(2 change in ), the basis vectors rotate by 2, so the angular
momentum vector once again has the same components. In other
words, it hasnt really changed at all.
Spatial curvature only
The at-space calculation above diers in two ways from the
actual result for an orbiting gyroscope: (1) it uses a at spatial
geometry, and (2) it is purely spatial. The purely spatial nature of
the calculation is manifested in the fact that there is nothing in the
result relating to how quickly weve moved the vector around the
circle. We know that if we whip a gyroscope around in a circle on
the end of a rope, there will be a Thomas precession (section 2.5.4),
which depends on the speed.
As our next warmup, lets curve the spatial geometry, but con-
tinue to omit the time dimension. Using the Schwarzschild metric,
we replace the at-space Christoel symbol
r
= r with r+2m.
The dierential equations for the components of the L vector, again
evaluated at r = 1 for convenience, are now
P
= Q
Q
= (1 )P ,
where = 2m. The solutions rotate with frequency
=

1 .
The result is that when the basis vectors rotate by 2, the compo-
nents no longer return to their original values; they lag by a factor
of
1 1 m. Putting the factors of r back in, this is 1 m/r.

The deviation from unity shows that after one full revolution, the L
vector no longer has quite the same components expressed in terms
of the (r, ) basis vectors.
To understand the sign of the eect, lets imagine a counter-
clockwise rotation. The (r, ) rotate counterclockwise, so relative
to them, the L vector rotates clockwise. After one revolution, it has
not rotated clockwise by a full 2, so its orientation is now slightly
counterclockwise compared to what it was. Thus the contribution
to the geodetic eect arising from spatial curvature is in the same
direction as the orbit.
Comparing with the actual results from Gravity Probe B, we see
that the direction of the eect is correct. The magnitude, however,
is o. The precession accumulated over n periods is 2nm/r, or,
in SI units, 2nGm/c
2
r. Using the data from section 2.5.4, we nd
= 2 10
5
radians, which is too small compared to the data
shown in gure b on page 140.
2+1 dimensions
To reproduce the experimental results correctly, we need to in-
clude the time dimension. The angular momentum vector now has
components (L
, L
r
, L
t
). The physical interpretation of the L
t
com-
ponent is obscure at this point; well return to this question later.
Writing down the total derivatives of the three components, and
notating dt/d as
1
, we have
dL
d
=
+
1
t
L
dL
r
d
=
L
r
+
1
t
L
r
dL
t
d
=
L
t
+
1
t
L
t
Setting the covariant derivatives equal to zero gives
0 =
r
L
r
0 =
L
r
+
r
0 =
t
L
r
+
r
tt
L
t
0 =
t
L
t
+
t
tr
L
r
.
Self-check: There are not just four but six covariant derivatives
that could in principle have occurred, and in these six covariant
derivatives we could have had a total of 18 Christoel symbols. Of
these 18, only four are nonvanishing. Explain based on symmetry
arguments why the following Christoel symbols must vanish:
t
,
t
tt
.
Putting all this together in matrix form, we have L
= ML,
where
M =
_
_
0 1 0
1 0 (1 )/2
0 /2(1 ) 0
_
_
.
The solutions of this dierential equation oscillate like e
it
, where
i is an eigenvalue of the matrix.
Self-check: The frequency in the purely spatial calculation was
found by inspection. Verify the result by applying the eigenvalue
technique to the relevant 2 2 submatrix.
To lowest order, we can use the Newtonian relation
2
r = Gm/r
and neglect terms of order
2
, so that the two new o-diagonal ma-
trix elements are both approximated as
_
/2. The three resulting
eigenfrequencies are zero and = [1 (3/2)m/r].
The presence of the mysterious zero-frequency solution can now
be understood by recalling the earlier mystery of the physical inter-
pretation of the angular momentums L
t
component. Our results
come from calculating parallel transport, and parallel transport is
a purely geometric process, so it gives the same result regardless
of the physical nature of the four-vector. Suppose that we had
instead chosen the velocity four-vector as our guinea pig. The def-
inition of a geodesic is that it parallel-transports its own tangent
vector, so the velocity vector has to stay constant. If we inspect
the eigenvector corresponding to the zero-frequency eigenfrequency,
we nd a timelike vector that is parallel to the velocity four-vector.
In our 2+1-dimensional space, the other two eigenvectors, which
are spacelike, span the subspace of spacelike vectors, which are the
ones that can physically be realized as the angular momentum of
a gyroscope. These two eigenvectors, which vary as e
i
, can be
superposed to make real-valued spacelike solutions that match the
initial conditions, and these lag the rotation of the basis vectors
by = (3/2)mr. This is greater than the purely spatial result
by a factor of 3/2. The resulting precession angle, over n orbits of
Gravity Probe B, is 3nGm/c
2
r = 3 10
5
radians, in excellent
agreement with experiment.
One will see apparently contradictory statements in the litera-
ture about whether Thomas precession occurs for a satellite: The
Thomas precession comes into play for a gyroscope on the surface
of the Earth . . . , but not for a gyroscope in a freely moving satel-
lite.
3
But: The total eect, geometrical and Thomas, gives the
well-known Fokker-de Sitter precession of 3m/r, in the same sense
as the orbit.
4
The second statement arises from subtracting the
purely spatial result from the 2+1-dimensional result, and noting
that the absolute value of this dierence is the same as the Thomas
precession that would have been obtained if the gyroscope had been
whirled at the end of a rope. In my opinion this is an unnatural
way of looking at the physics, for two reasons. (1) The signs dont
match, so one is forced to say that the Thomas precession has a dif-
ferent sign depending on whether the rotation is the result of grav-
itational or nongravitational forces. (2) Referring to observation, it
is clearly articial to treat the spatial curvature and Thomas eects
separately, since neither one can be disentangled from the other by
varying the quantities n, m, and r.
6.2.6 Orbits
The main event of Newtons Principia Mathematica is his proof
of Keplers laws. Similarly, Einsteins rst important application in
general relativity, which he began before he even had the exact form
of the Schwarzschild metric in hand, was to nd the non-Newtonian
behavior of the planet Mercury. The planets deviate from Keplerian
behavior for a variety of Newtonian reasons, and in particular there
is a long list of reasons why the major axis of a planets elliptical
orbit is expected to gradually rotate. When all of these were taken
into account, however, there was a remaining discrepancy of about
40 seconds of arc per century, or 6.6 10
7
radians per orbit. The
direction of the eect was in the forward direction, in the sense
that if we view Mercurys orbit from above the ecliptic, so that it
orbits in the counterclockwise direction, then the gradual rotation
of the major axis is also counterclockwise. In other words, Mercury
spends more time near perihelion than it should classically. During
this time, it sweeps out a greater angle than classically expected,
so that when it ies back out and away from the sun, its orbit has
rotated counterclockwise.
We can easily understand why there should be such an eect
according to general relativity, by passing to the limit in which the
relativistic eect is very large. Consider the orbits of material ob-
jects in the vicinity of a black hole. An object that passes through
the event horizon spends innite time at its perihelion and never
emerges again. Applying this reasoning to the case of Mercurys
orbit, we nd that an eect in the observed direction is expected.
Based on the examples in section 5.5, we also expect that the
eect will be of order m/r, where m is the mass of the sun and r
is the radius of Mercurys orbit. This works out to be 2.5 10
8
,
which is smaller than the observed precession by a factor of about
3
Misner, Thorne, and Wheeler, Gravitation, p. 1118
4
Rindler, Essential Relativity, 1969, p. 141
b / Proof that if the metrics
components are independent of
t , the geodesic of a test particle
conserves p
t
.
26.
Conserved quantities
If Einstein had had a computer on his desk, he probably would
simply have integrated the motion numerically using the geodesic
equation. But it is possible to simplify the problem enough to at-
tack it with pencil and paper, if we can nd the relevant conserved
quantities of the motion. Classically, these are energy and angular
momentum.
Consider a rock falling directly toward the sun. The Schwarzschild
metric is of the special form
ds
2
= h(r)dt
2
k(r)dr
2
. . . .
The rocks trajectory is a geodesic, so it extremizes the proper time
s between any two events xed in spacetime, just as a piece of string
stretched across a curved surface extremizes its length. Let the rock
pass through distance r
1
in coordinate time t
1
, and then through r
2
in t
2
. (These should really be notated as r
1
, . . . or dr
1
, . . . , but we
avoid the s or ds for convenience.) Approximating the geodesic
using two line segments, the proper time is
s = s
1
+s
2
=
_
h
1
t
2
1
k
1
r
2
1
+
_
h
2
t
2
2
k
2
r
2
2
=
_
h
1
t
2
1
k
1
r
2
1
+
_
h
2
(T t
1
)
2
k
2
r
2
2
,
where T = t
1
+ t
2
is xed. If this is to be extremized with respect
to t
1
, then ds/dt
1
= 0, which leads to
0 =
h
1
t
1
s
1
h
2
t
2
s
2
,
which means that
h
dt
ds
= g
tt
dx
t
ds
=
dx
t
ds
is a constant of the motion. Except for an irrelevant factor of m,
this is the same as p
t
, the timelike component of the covariant mo-
mentum vector. Weve already seen that in special relativity, the
timelike component of the momentum four-vector is interpreted as
the mass-energy E, and the quantity p
t
has a similar interpretation
here. Note that no special assumption was made about the form of
the functions h and k. In addition, it turns out that the assumption
of purely radial motion was unnecessary. All that really mattered
was that h and k were independent of t. Therefore we will have
a similar conserved quantity p
any time the metrics components,

expressed in a particular coordinate system, are independent of x
.
In particular, the Schwarzschild metrics components are indepen-
dent of as well as t, so we have a second conserved quantity p
,
which is interpreted as angular momentum.
Writing these two quantities out explicitly in terms of the con-
travariant coordinates, in the case of the Schwarzschild spacetime,
we have
E =
_
1
2m
r
_
dt
ds
and
L = r
2
d
ds
for the conserved energy per unit mass and angular momentum per
unit mass. In interpreting the energy E, it is important to under-
stand that in the general-relativistic context, there is no useful way
of separating the rest mass, kinetic energy, and potential energy into
separate terms; E includes all of these, and turns out to be less than
the rest mass (i.e., less than 1) for a planet orbiting the sun.
Perihelion advance
For convenience, let the mass of the orbiting rock be 1, while m
stands for the mass of the gravitating body.
The unit mass of the rock is a third conserved quantity, and
since the magnitude of the momentum vector equals the square of
the mass, we have for an orbit in the plane = /2,
1 = g
tt
p
2
t
g
rr
p
2
r
g
p
2
= g
tt
p
2
t
g
rr
(p
r
)
2
g
p
2
=
1
1 2m/r
E
2
1
1 2m/r
_
dr
ds
_
2
1
r
2
L
2
.
Rearranging terms and writing r for dr/ds, this becomes
r
2
= E
2
(1 2m/r)(1 +L
2
/r
2
)
or
r
2
= E
2
U
2
where
U
2
= (1 2m/r)(1 +L
2
/r
2
) .
There is a varied and strange family of orbits in the Schwarzschild
eld, including bizarre knife-edge trajectories that take several nearly
circular turns before suddenly ying o. We turn our attention in-
stead to the case of an orbit such as Mercurys which is nearly
classical and nearly circular.
Classically, a circular orbit has radius r = L
2
/m and period
T = 2L
3
/m
2
.
Relativistically, a circular orbit occurs when there is only one
turning point at which r = 0. This requires that E
2
equal the
minimum value of U
2
, which occurs at
r =
L
2
2m
_
1 +
_
1 12m
2
/L
2
_
L
2
m
(1 ) ,
where = 3(m/L)
2
. A planet in a nearly circular orbit oscillates
between perihelion and aphelion with a period that depends on the
curvature of U
2
at its minimum. We have
k =
d
2
(U
2
)
dr
2
=
d
2
dr
2
_
1
2m
r
+
L
2
r
2

2mL
2
r
3
_
=
4m
r
3
+
6L
2
r
4

24mL
2
r
5
= 2L
6
m
4
(1 + 2)
The period of the oscillations is
s
osc
= 2
_
2/k
= 2L
3
m
2
(1 ) .
The period of the azimuthal motion is
s
az
= 2r
2
/L
= 2L
3
m
2
(1 2) .
The periods are slightly mismatched because of the relativistic cor-
rection terms. The period of the radial oscillations is longer, so
that, as expected, the perihelion shift is in the forward direction.
The mismatch is s, and because of it each orbit rotates the ma-
jor axis by an angle 2 = 6(m/L)
2
= 6m/r. Plugging in the
data for Mercury, we obtain 5.8 10
7
radians per orbit, which
agrees with the observed value to within about 10%. Eliminating
some of the approximations weve made brings the results in agree-
ment to within the experimental error bars, and Einstein recalled
that when the calculation came out right, for a few days, I was
beside myself with joyous excitement.
Further attempts were made to improve on the precision of this
historically crucial test of general relativity. Radar now gives the
most precise orbital data for Mercury. At the level of about one part
per thousand, however, an eect creeps in due to the oblateness of
the sun, which is dicult to measure precisely.
In 1974, astronomers J.H. Taylor and R.A. Hulse of Princeton,
working at the Arecibo radio telescope, discovered a binary star
system whose members are both neutron stars. The detection of
the system was made possible because one of the neutron stars is
a pulsar: a neutron star that emits a strong radio pulse in the
direction of the earth once per rotational period. The orbit is highly
elliptical, and the minimum separation between the two stars is very
small, about the same as the radius of our sun. Both because the
r is small and because the period is short (about 8 hours), the rate
of perihelion advance per unit time is very large, about 4.2 degrees
per year. The system has been compared in great detail with the
predictions of general relativity,
5
giving extremely good agreement,
and as a result astronomers have been condent enough to reason in
the opposite direction and infer properties of the system, such as its
total mass, from the general-relativistic analysis. The systems orbit
is decaying due to the radiation of energy in the form of gravitational
waves, which are predicted to exist by relativity.
6.2.7 Deection of light
As discussed on page 140, one of the rst tests of general rel-
ativity was Eddingtons measurement of the deection of rays of
light by the suns gravitational eld. The deection measured by
Eddington was 1.6 seconds of arc. For a light ray that grazes the
suns surface, the only physically relevant parameters are the suns
mass m and radius r. Since the deection is unitless, it can only
depend on m/r, the unitless ratio of the suns mass to its radius.
Expressed in SI units, this is Gm/c
2
r, which comes out to be about
10
6
. Roughly speaking, then, we expect the order of magnitude of
the eect to be about this big, and indeed 10
6
radians comes out
to be in the same ball-park as a second of arc. We get a similar
estimate in Newtonian physics by treating a photon as a (massive)
particle moving at speed c.
It is possible to calculate a precise value for the deection us-
ing methods very much like those used to determine the perihelion
advance in section 6.2.6. However, some of the details would have
to be changed. For example, it is no longer possible to parametrize
the trajectory using the proper time s, since a light ray has ds = 0;
we must use an ane parameter. Let us instead use this an an
example of the numerical technique for solving the geodesic equa-
tion, rst demonstrated in section 5.9.2 on page 157. Modifying our
earlier program, we have the following:
1 import math
2
3 # constants, in SI units:
4 G = 6.67e-11 # gravitational constant
5
http://arxiv.org/abs/astro-ph/0407149
5 c = 3.00e8 # speed of light
6 m_kg = 1.99e30 # mass of sun
7 r_m = 6.96e8 # radius of sun
8
9 # From now on, all calculations are in units of the
10 # radius of the sun.
11
12 # mass of sun, in units of the radius of the sun:
13 m_sun = (G/c**2)*(m_kg/r_m)
14 m = 1000.*m_sun
15 print "m/r=",m
16
17 # Start at point of closest approach.
18 # initial position:
19 t=0
20 r=1 # closest approach, grazing the suns surface
21 phi=-math.pi/2
22 # initial derivatives of coordinates w.r.t. lambda
23 vr = 0
24 vt = 1
25 vphi = math.sqrt((1.-2.*m/r)/r**2)*vt # gives ds=0, lightlike
26
27 l = 0 # affine parameter lambda
28 l_max = 20000.
29 epsilon = 1e-6 # controls how fast lambda varies
30 while l<l_max:
31 dl = epsilon*(1.+r**2) # giant steps when farther out
32 l = l+dl
33 # Christoffel symbols:
34 Gttr = m/(r**2-2*m*r)
35 Grtt = m/r**2-2*m**2/r**3
36 Grrr = -m/(r**2-2*m*r)
37 Grphiphi = -r+2*m
38 Gphirphi = 1/r
39 # second derivatives:
40 # The factors of 2 are because we have, e.g., Gâ_{bc}=Gâ_{cb}
41 at = -2.*Gttr*vt*vr
42 ar = -(Grtt*vt*vt + Grrr*vr*vr + Grphiphi*vphi*vphi)
43 aphi = -2.*Gphirphi*vr*vphi
44 # update velocity:
45 vt = vt + dl*at
46 vr = vr + dl*ar
47 vphi = vphi + dl*aphi
48 # update position:
49 r = r + vr*dl
50 t = t + vt*dl
51 phi = phi + vphi*dl
52
53 # Direction of propagation, approximated in asymptotically flat coords.
54 # First, differentiate (x,y)=(r cos phi,r sin phi) to get vx and vy:
55 vx = vr*math.cos(phi)-r*math.sin(phi)*vphi
56 vy = vr*math.sin(phi)+r*math.cos(phi)*vphi
57 prop = math.atan2(vy,vx) # inverse tan of vy/vx, in the proper quadrant
58 prop_sec = prop*180.*3600/math.pi
59 print "final direction of propagation = %6.2f arc-seconds" % prop_sec
At line 14, we take the mass to be 1000 times greater than the
mass of the sun. This helps to make the deection easier to calcu-
late accurately without running into problems with rounding errors.
Lines 17-25 set up the initial conditions to be at the point of closest
approach, as the photon is grazing the sun. This is easier to set
up than initial conditions in which the photon approaches from far
away. Because of this, the deection angle calculated by the pro-
gram is cut in half. Combining the factors of 1000 and one half, the
nal result from the program is to be interpreted as 500 times the
actual deection angle.
The result is that the deection angle is predicted to be 870
seconds of arc. As a check, we can run the program again with
m = 0; the result is a deection of 8 seconds, which is a measure
of the accumulated error due to rounding and the nite increment
used for .
Dividing by 500, we nd that the predicted deection angle is
1.74 seconds, which, expressed in radians, is exactly 4Gm/c
2
r. The
unitless factor of 4 is in fact the correct result in the case of small
deections, i.e., for m/r 1.
Although the numerical technique has the disadvantage that it
doesnt let us directly prove a nice formula, it has some advantages
as well. For one thing, we can use it to investigate cases for which
the approximation m/r 1 fails. For m/r = 0.3, the numerical
techique gives a deection of 222 degrees, whereas the weak-eld
approximation 4Gm/c
2
r gives only 69 degrees. What is happening
here is that were getting closer and closer to the event horizon of a
black hole. Black holes are the topic of section 6.3, but it should be
intuitively reasonable that something wildly nonlinear has to happen
as we get close to the point where the light wouldnt even be able
to escape.
The precision of Eddingtons original test was only about 30%,
and has never been improved on signicantly with visible-light as-
tronomy. A better technique is radio astronomy, which allows mea-
surements to be carried out without waiting for an eclipse. One
merely has to wait for the sun to pass in front of a strong, compact
radio source such as a quasar. These techniques have now veried
the deection of light predicted by general relativity to a relative
precision of about 10
5
.
6
6.3 Black holes
6.3.1 Singularities
A provocative feature of the Schwarzschild metric is that it has
elements that blow up at r = 0 and at r = 2m. If this is a description
of the sun, for example, then these singularities are of no physical
signicance, since we only solved the Einstein eld equation for the
vacuum region outside the sun, whereas r = 2m would lie about 3
km from the suns center. Furthermore, it is possible that one or
both of these singularities is nothing more than a spot where our
coordinate system misbehaves. This would be known as a coordinate
singularity. For example, the metric of ordinary polar coordinates
in a Euclidean plane has g
as r 0.
One way to test whether a singularity is a coordinate singularity
is to calculate a scalar measure of curvature, whose value is inde-
pendent of the coordinate system. We can take the trace of the
Ricci tensor, R
a
a
, but since the Ricci tensor is zero, its not sur-
prising that that is zero. A dierent scalar we can construct is the
product R
abcd
R
abcd
of the Riemann tensor with itself. The Maxima
command lriemann(true) displays the nonvanishing components
of R
abcd
. The component that misbehaves the most severely at
r = 0 is R
trrt
= 2m/r
3
. Because of this, R
abcd
R
abcd
blows up like
r
6
as r 0. This shows that the singularity at r = 0 is a real,
physical singularity.
The singularity at r = 2m, on the other hand, turns out to
be only a coordinate singularity. To prove this, we have to use
some technique other than constructing scalar measures of curva-
ture. Even if every such scalar we construct is nite at r = 2m, that
doesnt prove that every such scalar we could construct is also well
behaved. We can instead search for some other coordinate system
in which to express the solution to the eld equations, one in which
no such singularity appears. A partially successful change of coordi-
nates for the Schwarzschild metric, found by Eddington in 1924, is
t t
= t 2mln(r 2m) (see problem 5 on page 185). This makes

the covariant metric nite at r = 2m, although the contravariant
metric still blows up there. A more complicated change of coor-
dinates that completely eliminates the singularity at r = 2m was
found by Eddington and Finkelstein in 1958, establishing that the
singularity was only a coordinate singularity. Thus, if an observer
is so unlucky as to fall into a black hole, he will not be subjected
to innite tidal stresses or innite anything at r = 2m. He
6
For a review article on this topic, see Cliord Will, The Con-
frontation between General Relativity and Experiment, http://relativity.
livingreviews.org/Articles/lrr-2006-3/.
Section 6.3 Black holes 185
may not notice anything special at all about his local environment.
(Or he may already be dead because the tidal stresses at r > 2m,
although nite, were nevertheless great enough to kill him.)
6.3.2 Event horizon
Even though r = 2m isnt a real singularity, interesting things
do happen there. For r < 2m, the sign of g
tt
becomes negative,
while g
rr
is positive. In our + signature, this has the fol-
lowing interpretation. For the world-line of a material particle, ds
2
is supposed to be the square of the particles proper time, and it
must always be positive. If a particle had a constant value of r, for
r < 2m, it would have ds
2
< 0, which is impossible.
The timelike and spacelike characters of the r and t coordinates
have been swapped, so r acts like a time coordinate.
Thus for an object compact enough that r = 2m is exterior,
r = 2m is an event horizon: future light cones tip over so far that
they do not allow causal relationships to connect with the spacetime
outside. In relativity, event horizons do not occur only in the context
of black holes; their properties, and some of the implications for
black holes, have already been discussed in section 6.1.
6.3.3 Expected formation
Einstein and Schwarzschild did not believe, however, that any of
these features of the Schwarzschild metric were more than a mathe-
matical curiosity, and the term black hole was not invented until
the 1967, by John Wheeler. Although there is quite a bit of evi-
dence these days that black holes do exist, there is also the related
question of what sizes they come in.
We might expect naively that since gravity is an attractive force,
there would be a tendency for any primordial cloud of gas or dust
to spontaneously collapse into a black hole. But clouds of less than
about 0.1M
(0.1 solar masses) form planets, which achieve a per-

manent equilibrium between gravity and internal pressure. Heavier
objects initiate nuclear fusion, but those with masses above about
100M
are immediately torn apart by their own solar winds. In the

range from 0.1 to 100M
, stars form. As discussed in section 4.4.3,

those with masses greater than about a few solar are expected to
form black holes when they die. We therefore expect, on theoretical
grounds, that the universe should contain black holes with masses
ranging from a few solar masses to a few tens of solar masses.
6.3.4 Observational evidence
A black hole is expected to be a very compact object, with a
strong gravitational eld, that does not emit any of its own light. A
bare, isolated black hole would be dicult to detect, except perhaps
via its lensing of light rays that happen to pass by it. But if a black
hole occurs in a binary star system, it is possible for mass to be
a / A black hole accretes matter
from a companion star.
transferred onto the black hole from its companion, if the compan-
ions evolution causes it to expand into a giant and intrude upon the
black holes gravity well. The object known as Cygnus X-1 is the
best-studied example. This X-ray-emitting object was discovered
by a rocket-based experiment in 1964. It is part of a double-star
system, the other member being a blue supergiant. They orbit their
common center of mass with a period of 5.6 days. The orbit is
nearly circular, and has a semi-major axis of about 0.2 times the
distance from the earth to the sun. Applying Keplers law of peri-
ods to these data constrains the sum of the masses, and knowledge
of stellar structure xes the mass of the supergiant. The result is
that the mass of Cygnus X-1 is greater than about 10 solar masses,
and this is conrmed by multiple methods. Since this is far above
the Tolman-Oppenheimer-Volko limit, Cygnus X-1 is believed to
be a black hole, and its X-ray emissions are interpreted as the radi-
ation from the disk of superheated material accreting onto it from
its companion.
Around the turn of the 21st century, new evidence was found
for the prevalence of supermassive black holes near the centers of
nearly all galaxies, including our own. Near our galaxys center is
an object called Sagittarius A*, detected because nearby stars orbit
around it. The orbital data show that Sagittarius A* has a mass
of about four million solar masses, conned within a sphere with
a radius less than 2.2 10
7
km. There is no known astrophysical
model that could prevent the collapse of such a compact object into
a black hole, nor is there any plausible model that would allow this
much mass to exist in equilibrium in such a small space, without
emitting enough light to be observable.
The existence of supermassive black holes is surprising. Gas
clouds with masses greater than about 100 solar masses cannot form
stable stars, so supermassive black holes cannot be the end-point of
the evolution of heavy stars. Mergers of multiple stars to form more
massive objects are generally statistically unlikely, since a star is
such a small target in relation to the distance between the stars.
Once astronomers were confronted with the empirical fact of their
existence, a variety of mechanisms was proposed for their formation.
Little is known about which of these mechanisms is correct, although
the existence of quasars in the early universe is interpreted as evi-
dence that mass accreted rapidly onto supermassive black holes in
the early stages of the evolution of the galaxies.
6.3.5 Singularities and cosmic censorship
Since we observe that black holes really do exist, maybe we
should take the singularity at r = 0 seriously. Physically, it says
that the mass density and tidal forces blow up to innity there.
Generally when a physical theory says that observable quantities
blow up to innity at a particular point, it means that the theory has
reached the point at which it can no longer make physical predic-
tions. For instance, Maxwells theory of electromagnetism predicts
that the electric eld blows up like r
2
near a point charge, and
this implies that innite energy is stored in the eld within a nite
radius around the charge. Physically, this cant be right, because
we know it only takes 511 keV of energy to create an electron out
of nothing, e.g., in nuclear beta decay. The paradox is resolved by
quantum electrodynamics, which modies the description of the vac-
uum around the electron to include a sea of virtual particles popping
into and out of existence.
In the case of the singularity at the center of a black hole, it is
possible that quantum mechanical eects at the Planck scale prevent
the formation of a singularity. Unfortunately, we are unlikely to nd
any empirical evidence about this, since black holes always seem to
come clothed in event horizons, so we cannot extract any data about
a singularity inside. In a way, this is a good thing. If a singularity
exists, it is a point at which all the known laws of physics broke
down, and physicists therefore have no way of predicting anything
about its behavior. As John Earman of the University of Pittsburgh
puts it, anything could pop out of a singularity, including green slime
or your lost socks. In more technical language, a naked singularity
would constitute an extreme violation of unitarity and an acute
instance of the information paradox (see page 167).
As long as singularities are hidden behind event horizons, this
has no eect on our ability to make predictions about the physi-
cal behavior of the universe. There is no obvious built-in reason
that general relativity should not allow naked singularities, but nei-
ther do we know of any real-world process by which one could be
formed. Roger Penroses 1969 cosmic censorship hypothesis states
that our universe contains no naked singularities other than the Big
Bang. A reasonably readable treatment of these issues is given in
Earman, Bangs, Crunches, Whimpers, and Shrieks: Singularities
and Acausalities in Relativistic Spacetimes, Oxford, 1995; he clearly
marks the sections that are highly technical and suggests how the
non-specialist reader can navigate the book and absorb the main
points.
6.3.6 Black hole radiation
Since event horizons are expected to emit blackbody radiation,
a black hole should not be entirely black; it should radiate. Sup-
pose observer B just outside the event horizon blasts the engines of
her rocket ship, producing enough acceleration to keep from being
sucked in. By the equivalence principle, what she observes cannot
depend on whether the acceleration she experiences is actually due
to a gravitational eld. She therefore detects radiation, which she
interprets as coming from the event horizon below her. As she gets
closer and closer to the horizon, the acceleration approaches innity,
so the intensity and frequency of the radiation grows without limit.
A distant observer A, however, sees a dierent picture. Accord-
ing to A, Bs time is extremely dilated. A sees Bs acceleration
as being only 1/m, where m is the mass of the black hole; A
does not perceive this acceleration as blowing up to innity as B
approaches the horizon. When A detects the radiation, it is ex-
tremely red-shifted, and it has the spectrum that one would expect
for a horizon characterized by an acceleration a 1/m. The result
for a 10-solar-mass black hole is T 10
8
K, which is so low that
the black hole is actually absorbing more energy from the cosmic
microwave background radiation than it emits.
Direct observation of black-hole radiation is therefore probably
only possible for black holes of very small masses. These may have
been produced soon after the big bang, or it is conceivable that
they could be created articially, by advanced technology. If black-
hole radiation does exist, it may help to resolve the information
paradox, since it is possible that information that goes into a black
hole is eventually released via subtle correlations in the black-body
radiation it emits.
A very dicult question about the relationship between quan-
tum mechanics and general relativity occurs as follows. In our ex-
ample above, observer A detects an extremely red-shifted spectrum
of light from the black hole. A interprets this as evidence that
the space near the event horizon is actually an intense maelstrom
of radiation, with the temperature approaching innity as one gets
closer and closer to the horizon. If B returns from the region near
the horizon, B will agree with this description. But suppose that
observer C simply drops straight through the horizon. C does not
feel any acceleration, so by the equivalence principle C does not
detect any radiation at all. Passing down through the event hori-
zon, C says, A and B are liars! Theres no radiation at all. A
and B, however, C see as having entered a region of innitely in-
tense radiation. Ah, says A, too bad. C should have turned
back before it got too hot, just as I did. This is an example of a
principle weve encountered before, that when gravity and quantum
mechanics are combined, dierent observers disagree on the number
of quanta present in the vacuum. We are presented with a paradox,
because A and B believe in an entirely dierent version of reality
that C. A and B say C was fricasseed, but C knows that that didnt
happen. One suggestion is that this contradiction shows that the
proper logic for describing quantum gravity is nonaristotelian, as
described on page 53. This idea, suggested by Susskind et al., goes
by the name of black-hole complementarity, by analogy with Niels
Bohrs philosophical description of wave-particle duality as being
complementary rather than contradictory. In this interpretation,
we have to accept the fact that C experiences a qualitatively dier-
ent reality than A and B, and we comfort ourselves by recognizing
that the contradiction can never become too acute, since C is lost
behind the event horizon and can never send information back out.
Problems
1 Show that in geometrized units, power is unitless. Find the
equivalent in watts of a power that equals 1 in geometrized units.
2 The metric of coordinates (, ) on the unit sphere is ds
2
=
d
2
+ sin
2
d
2
. (a) Show that there is a singular point at which
g
ab
. (b) Verify directly that the scalar curvature R = R
a
a
constructed from the trace of the Ricci tensor is never innite. (c)
Prove that the singularity is a coordinate singularity.
3 (a) Section 3.4.4 (p. 89) gave a at-spacetime metric in
rotating polar coordinates,
ds
2
= (1
2
r
2
)dt
2
dr
2
r
2
d
2
2r
2
d
dt .
Identify the two values of r at which singularities occur, and classify
them as coordinate or non-coordinate singularities.
(b) The corresponding spatial metric was found to be
ds
2
= dr
2
r
2
1
2
r
2
d
2
.
Identify the two values of r at which singularities occur, and classify
them as coordinate or non-coordinate singularities.
(c) Consider the following argument, which is intended to provide
an answer to part b without any computation. In two dimensions,
there is only one measure of curvature, which is equivalent (up to a
constant of proportionality) to the Gaussian curvature. The Gaus-
sian curvature is proportional to the angular decit of a triangle.
Since the angular decit of a triangle in a space with negative cur-
vature satises the inequality < < 0, we conclude that the
Gaussian curvature can never be innite. Since there is only one
measure of curvature in a two-dimensional space, this means that
there is no non-coordinate singularity. Is this argument correct,
and is the claimed result consistent with your answers to part b?
Solution, p. 265
4 The rst experimental verication of gravitational redshifts
was a measurement in 1925 by W.S. Adams of the spectrum of light
emitted from the surface of the white dwarf star Sirius B. Sirius B
has a mass of 0.98M
and a radius of 5.910

6
m. Find the redshift.
5 Show that, as claimed on page 185, applying the change of
coordinates t
= t2mln(r2m) to the Schwarzschild metric results

in a metric for which g
rr
and g
t
t
never blow up, but that g
t
does
blow up.
Problems 191
Chapter 7
Sources
7.1 Sources in general relativity
7.1.1 Point sources in a background-independent theory
Schrodinger equation and Maxwells equations treat spacetime
as a stage on which particles and elds act out their roles. General
relativity, however, is essentially a theory of spacetime itself. The
role played by atoms or rays of light is so peripheral that by the time
Einstein had derived an approximate version of the Schwarzschild
metric, and used it to nd the precession of Mercurys perihelion, he
still had only vague ideas of how light and matter would t into the
picture. In his calculation, Mercury played the role of a test particle:
a lump of mass so tiny that it can be tossed into spacetime in order
to measure spacetimes curvature, without worrying about its eect
on the spacetime, which is assumed to be negligible. Likewise the
sun was treated as in one of those orchestral pieces in which some
of the brass play from o-stage, so as to produce the eect of a
second band heard from a distance. Its mass appears simply as an
adjustable parameter m in the metric, and if we had never heard of
the Newtonian theory we would have had no way of knowing how
to interpret m.
When Schwarzschild published his exact solution to the vacuum
eld equations, Einstein suered from philosophical indigestion. His
strong belief in Machs principle led him to believe that there was a
paradox implicit in an exact spacetime with only one mass in it. If
Einsteins eld equations were to mean anything, he believed that
they had to be interpreted in terms of the motion of one body rela-
tive to another. In a universe with only one massive particle, there
would be no relative motion, and so, it seemed to him, no motion
of any kind, and no meaningful interpretation for the surrounding
spacetime.
Not only that, but Schwarzschilds solution had a singularity
at its center. When a classical eld theory contains singularities,
Einstein believed, it contains the seeds of its own destruction. As
weve seen on page 188, this issue is still far from being resolved, a
century later.
However much he might have liked to disown it, Einstein was
now in possession of a solution to his eld equations for a point
source. In a linear, background-dependent theory like electromag-
193
netism, knowledge of such a solution leads directly to the ability to
write down the eld equations with sources included. If Coulombs
law tells us the 1/r
2
variation of the electric eld of a point charge,
then we can infer Gausss law. The situation in general relativity
is not this simple. The eld equations of general relativity, unlike
the Gausss law, are nonlinear, so we cant simply say that a planet
or a star is a solution to be found by adding up a large number of
point-source solutions. Its also not clear how one could represent a
moving source, since the singularity is a point that isnt even part
of the continuous structure of spacetime (and its location is also
hidden behind an event horizon, so it cant be observed from the
outside).
7.1.2 The Einstein eld equation
The Einstein tensor
Given these diculties, its not surprising that Einsteins rst
attempt at incorporating sources into his eld equation was a dead
end. He postulated that the eld equation would have the Ricci
tensor on one side, and the energy-momentum tensor T
ab
(page
131) on the other,
R
ab
= 8T
ab
,
where a factor of G/c
4
on the right is suppressed by our choice
of units, and the 8 is determined on the basis of consistency with
Newtonian gravity in the limit of weak elds and low velocities. The
problem with this version of the eld equations can be demonstrated
by counting variables. R and T are symmetric tensors, so the eld
equation contains 10 constraints on the metric: 4 from the diagonal
elements and 6 from the o-diagonal ones. In addition, conservation
of mass-energy requires the divergence-free property
b
T
ab
= 0,
because otherwise, for example, we could have a mass-energy tensor
that varied as T
00
= kt, describing a region of space in which mass
was uniformly appearing or disappearing at a constant rate. But
this adds 4 more constraints on the metric, for a total of 14. The
metric, however, is a symmetric rank-2 tensor itself, so it only has
10 independent components. This overdetermination of the metric
suggests that the proposed eld equation will not in general allow a
solution to be evolved forward in time from a set of initial conditions
given on a spacelike surface, and this turns out to be true. It can in
fact be shown that the only possible solutions are those in which the
traces R = R
a
a
and T = T
a
a
are constant throughout spacetime.
The solution is to replace R
ab
in the eld equations with the a
dierent tensor G
ab
, called the Einstein tensor, dened by G
ab
=
R
ab
(1/2)Rg
ab
,
G
ab
= 8T
ab
.
The Einstein tensor is constructed exactly so that it is divergence-
free,
b
G
ab
= 0. (This is not obvious, but can be proved by direct
194 Chapter 7 Sources
computation.) Therefore any energy-momentum tensor that satis-
es the eld equation is automatically divergenceless, and thus no
additional constraints need to be applied in order to guarantee con-
servation of mass-energy.
Self-check: Does replacing R
ab
with G
ab
invalidate the Schwarzschild
metric?
Further interpretation of the energy-momentum tensor
The energy-momentum tensor was briey introduced in section
5.2 on page 131. By applying the Newtonian limit of the eld equa-
tion to the Schwarzschild metric, we nd that T
tt
is to be identied
as the mass density. The Schwarzschild metric describes a spacetime
using coordinates in which the mass is at rest. In the cosmological
applications well be considering shortly, it also makes sense to adopt
a frame of reference in which the local mass-energy is, on average, at
rest, so we can continue to think of T
tt
as the (average) mass density.
By symmetry, T must be diagonal in such a frame. For example, if
we had T
tx
,= 0, then the positive x direction would be distinguished
from the negative x direction, but there is nothing that would allow
such a distinction. The spacelike components are associated with
the pressure, P. The form of the tensor with mixed upper and lower
indices has the simple form T
= diag(, P, P, P).
7.1.3 The uniform gravitational eld revisited
This subsection gives a somewhat exotic example that provides
a little practice in interpreting the energy-momentum tensor. It is
not necessary to read it in order to understand the later material.
In problem 5 on page 162, we made a wish list of desired proper-
ties for a uniform gravitational eld, and found that they could not
all be satised at once. That is, there is no global solution to the
Einstein eld equations that uniquely and satisfactorily embodies
all of our Newtonian ideas about a uniform eld. We now revisit
this question in the light of our new knowledge about sources and
the stress-energy tensor.
The 1+1-dimensional metric
ds
2
= e
2gz
dt
2
dz
2
is the one that uniquely satises our expectations based on the
equivalence principle (example 7, p. 46), and it is a vacuum so-
lution. We might logically try to generalize this to 3+1 dimensions
as follows:
ds
2
= e
2gz
dt
2
dx
2
dy
2
dz
2
.
But a funny thing happens now simply by slapping on the two
new Cartesian axes x and y, it turns out that we have made our
vacuum solution into a non-vacuum solution, and not only that,
but the resulting energy-momentum tensor is unphysical (problem
1, p. 227).
Section 7.1 Sources in general relativity 195
One way to proceed would be to relax our insistence on making
the spacetime one that exactly embodies the equivalence principles
requirements for a uniform eld.
1
This can be done by taking g
tt
=
e
2
, where is not necessarily equal to 2gz. By requiring that the
metric be a 3+1 vacuum solution, we arrive at a dierential equation
whose solution is = ln(z +k
1
) +k
2
, which recovers the at-space
metric that we found in example 14 on page 116 by applying a
change of coordinates to the Lorentz metric.
What if we want to carry out the generalization from 1+1 to
3+1 without violating the equivalence principle? For physical mo-
tivation in how to get past this obstacle, consider the following ar-
gument made by Born in 1920.
2
Take a frame of reference tied to a
rotating disk, as in the example from which Einstein originally took
much of the motivation for creating a geometrical theory of gravity
(subsection 3.4.4, p. 89). Clocks near the edge of the disk run slowly,
and by the equivalence principle, an observer on the disk interprets
this as a gravitational time dilation. But this is not the only rela-
tivistic eect seen by such an observer. Her rulers are also Lorentz
contracted as seen by a non-rotating observer, and she interprets
this as evidence of a non-Euclidean spatial geometry. There are
some physical dierences between the rotating disk and our default
conception of a uniform eld, specically in the question of whether
the metric should be static (i.e., lacking in cross-terms between the
space and time variables). But even so, these considerations make
it natural to hypothesize that the correct 3+1-dimensional metric
should have transverse spatial coecients that decrease with height.
With this motivation, lets consider a metric of the form
ds
2
= e
2z
dt
2
e
2jz
dx
2
e
2kz
dy
2
dz
2
,
where j and k are constants, and Ive taken g = 1 for convenience.
3
The following Maxima code calculates the scalar curvature and the
Einstein tensor:
1 load(ctensor);
2 ct_coords:[t,x,y,z];
3 lg:matrix([exp(2*z),0,0,0],
4 [0,-exp(-2*j*z),0,0],
5 [0,0,-exp(-2*k*z),0],
6 [0,0,0,-1]
7 );
1
Thanks to physicsforums.com user Mentz114 for suggesting this approach
and demonstrating the following calculation.
2
Max Born, Einsteins Theory of Relativity, 1920. In the 1962 Dover edition,
the relevant passage is on p. 320
3
A metric of this general form is referred to as a Kasner metric. One usually
sees it written with a logarithmic change of variables, so that z appears in the
base rather than in the exponent.
8 cmetric();
9 scurvature();
10 leinstein(true);
The output from line 9 shows that the scalar curvature is constant,
which is a necessary condition for any spacetime that we want to
think of as representing a uniform eld. Inspecting the Einstein
tensor output by line 10, we nd that in order to get G
xx
and G
yy
to vanish, we need j and k to be (1
3i)/2. By trial and error, we

nd that assigning the complex-conjugate values to j and k makes
G
tt
and G
zz
vanish as well, so that we have a vacuum solution.
This solution is, unfortunately, complex, so it is not of any obvious
value as a physically meaningful result. Since the eld equations
are nonlinear, we cant use the usual trick of forming real-valued
superpositions of the complex solutions. We could try simply tak-
ing the real part of the metric. This gives g
xx
= e
z
cos
3z and
g
yy
= e
z
sin
3z, and is unsatisfactory because the metric becomes

degenerate (has a zero determinant) at z = n/2
3, where n is an
integer.
It turns out, however, that there is a very similar solution, found
by Petrov in 1962,
4
that is real-valued. The Petrov metric, which
describes a spacetime with cylindrical symmetry, is:
ds
2
= dr
2
e
2r
dz
2
+e
r
[2 sin
3rddt cos
3r(d
2
dt
2
)]
Note that it has many features in common with the complex oscilla-
tory solution we found above. There are transverse length contrac-
tions that decay and oscillate in exactly the same way. The presence
of the ddt term tells us that this is a non-static, rotating solution
exactly like the one that Einstein and Born had in mind in their
prototypical example! We typically obtain this type of eect due
to frame dragging by some rotating massive body (see p. 123), and
the Petrov solution can indeed be interpreted as the spacetime that
exists in the vacuum on the exterior of an innite, rigidly rotating
cylinder of dust dust in the sense of material particles with
an energy-momentum tensor of the form T
= diag(, 0, 0, 0).
The complicated Petrov metric might seem like the furthest pos-
sible thing from a uniform gravitational eld, but in fact it is about
the closest thing general relativity provides to such a eld. We rst
note that the metric displays symmetry under z z +c, +c,
and t t + c, so it has at least three out of the four translation
symmetries we expect from a uniform eld. By analogy with elec-
tromagnetism, we would expect this symmetry to be absent in the
radial direction, since by Gausss law the electric eld of a line of
4
Petrov, in Recent Developments in General Relativity, 1962, Pergamon, p.
383. For a presentation that is freely accessible online, see Gibbons and Gielen,
The Petrov and Kaigorodov-Ozsv ath Solutions: Spacetime as a Group Mani-
fold, arxiv.org/abs/0802.4082.
charge falls o like 1/r. Adding in a centrifugal force, we dont get
anything like a constant. But surprisingly, the Petrov metric has
a constant scalar curvature (problem 2, p. 227). This is what we
want in a uniform eld, and it also suggests that this spacetime has
an even higher degree of symmetry than the threefold translational
symmetry one would guess by looking at the metric.
For insight into this surprising result, recall that in our attempt
at constructing the Cartesian version of this metric, we ran into the
problem that the metric became degenerate at z = n/2
3. The
presence of the ddt term prevents this from happening in Petrovs
cylindrical version; two of the metrics diagonal components can
vanish at certain values of r, but the presence of the o-diagonal
component prevents the determinant from going to zero. (The de-
terminant is in fact equal to 1 everywhere.) What is happening
physically is that although the labeling of the and t coordinates
suggests a time and an azimuthal angle, these two coordinates are
in fact treated completely symmetrically. At values of r where the
cosine factor equals 1, the metric is diagonal, and has signature
(t, , r, z) = (+, , , ), but when the cosine equals 1, this be-
comes (, +, , ), so that is now the timelike coordinate. This
perfect symmetry between and t is an extreme example of frame-
dragging, and is produced because of the specially chosen rate of
rotation of the dust cylinder, such that the velocity of the dust at
the outer surface is exactly c (or approaches it).
Classically, we would expect that a test particle released close
enough to the cylinder would be pulled in by the gravitational at-
traction and destroyed on impact, while a particle released farther
away would y o due to the centrifugal force, escaping and even-
tually approaching a constant velocity. Neither of these would be
anything like the experience of a test particle released in a uniform
eld. But consider a particle released at rest in the rotating frame
at a radius r
1
for which cos
3r
1
= 1, so that t is the timelike co-
ordinate. The particle accelerates (lets say outward), but at some
point it arrives at an r
2
where the cosine equals zero, and the t
part of the metric is purely of the form ddt. At this location, we
can dene local coordinates u = t and v = + t, so that the
metric depends only on du
2
dv
2
. One of the coordinates, say u, is
now the timelike one. Since our particle is material, its world-line
must be timelike, so it is swept along in the direction. Gibbons
and Gielen show that the particle will now come back inward, and
continue forever by oscillating back and forth between two radii at
which the cosine vanishes.
This oscillation still doesnt sound like the motion of a particle
in a uniform eld, but another strange thing happens, as we can
see by taking another look at the values of r at which the cosine
vanishes. At such a value of r, construct a curve of the form (t =
constant, r = constant, , z = constant). This is a closed curve, and
its proper length is zero, i.e., it is lightlike. This violates causality.
A photon could travel around this path and arrive at its starting
point at the same time when it was emitted. Something similarly
weird hapens to the test particle described above: whereas it seems
to fall sometimes up and sometimes down, in fact it is always falling
down but sometimes it achieves this by falling up while moving
backward in time!
Although the Petov metric violates causality, Gibbons and Gie-
len have shown that it satises the chronology protection conjecture:
In the context of causality violation we have shown that one cannot
create CTCs [closed timelike curves] by spinning up a cylinder be-
yond its critical angular velocity by shooting in particles on timelike
or null curves.
7.1.4 Energy conditions
The eld equations are agnostic on the question of what kinds
of matter elds contribute to the energy-momentum tensor. In fact,
any spacetime at all is a solution to the Einstein eld equations, pro-
vided were willing to admit the corresponding energy-momentum
tensor. Suppose that we want to prove all-encompassing theorems
about general relativity. For instance, we may want to reassure our-
selves that general relativity will never lead to violations of causality
such as the closed timelike curves exhibited by the Petrov metric
(p. 197). We can never make this type of general proof without
assuming something about the energy-momentum tensor.
In example 10 on page 110, we saw that radiation has P = /3
and dust has P = 0. Both have 0. If the universe is made out
of nothing but dust and radiation, then we can obtain the following
four constraints on the energy-momentum tensor:
trace energy condition 3P 0
strong energy condition + 3P 0 and +P 0
dominant energy condition 0 and [P[
weak energy condition 0 and +P 0
null energy condition +P 0
These are arranged roughly in order from strongest to weakest. They
all have to do with the idea that negative mass-energy doesnt seem
to exist in our universe, i.e., that gravity is always attractive rather
than repulsive. With this motivation, it would seem that there
should only be one way to state an energy condition: > 0. But
the symbols and P refer to the form of the energy-momentum
tensor in a special frame of reference, interpreted as the one that
is at rest relative to the average motion of the ambient matter. In
this frame, the tensor is diagonal. Switching to some other frame of
reference, the and P parts of the tensor would mix, and it might
be possible to end up with a negative energy density. The weak
energy condition is the constraint we need in order to make sure
that the energy density is never negative in any frame.
The dominant energy condition is like the weak energy condition,
but it also guarantees that no observer will see a ux of energy
owing at speeds greater than c.
As an example of the use of the energy conditions, Hawking and
Ellis have proved that under the assumption of the strong energy
condition, any body that becomes suciently compact will end up
forming a singularity. We might imagine that the formation of a
black hole would be a delicate thing, requiring perfectly symmetric
initial conditions in order to end up with the perfectly symmetric
Schwarzschild metric. Many early relativists thought so, for good
reasons. If we look around the universe at various scales, we nd
that collisions between astronomical bodies are extremely rare. This
is partly because the distances are vast compared to the sizes of the
objects, but also because conservation of angular momentum has
a tendency to make objects swing past one another rather than
colliding head-on. Starting with a cloud of objects, e.g., a globular
cluster, Newtons laws make it extremely dicult, regardless of the
attractive nature of gravity, to pick initial conditions that will make
them all collide the future. For one thing, they would have to have
exactly zero total angular momentum.
Most relativists now believe that this is not the case. General
relativity describes gravity in terms of the tipping of light cones.
When the eld is strong enough, there is a tendency for the light
cones to tip over so far that the entire future light-cone points at the
source of the eld. If this occurs on an entire surface surrounding
the source, it is referred to as a trapped surface. Any lumpiness or
rotation in the initial conditions becomes irrelevant, because every
particles entire future world-line lies inward rather than outward.
A possible loophole in this argument is the question of whether the
light cones will really tip over far enough. We could imagine that
under extreme conditions of high density and temperature, matter
might demonstrate unusual behavior, perhaps including a negative
energy density, which would then give rise to a gravitational repul-
sion. Gravitational repulsion would tend to make the light cones tip
outward rather than inward, possibly preventing the collapse to a
singularity. We can close this loophole by assuming an appropriate
energy condition. Penrose and Hawking have formalized the above
argument in the form of a series of theorems, known as the singular-
ity theorems. One of these applies to the formation of black holes,
and another one to cosmological singularities such as the Big Bang.
The current status of the energy conditions is shaky. Although
it is clear that all of them hold in a variety of situations, there are
strong reasons to believe that they are violated at both microscopic
and cosmological scales, for reasons both classical and quantum-
mechanical.
5
We will see such a violation in the following section.
7.1.5 The cosmological constant
Having included the source term in the Einstein eld equations,
our most important application will be to cosmology. Some of the
relevant ideas originate long before Einstein. Once Newton had
formulated a theory of gravity as a universal attractive force, he
realized that there would be a tendency for the universe to collapse.
He resolved this diculty by assuming that the universe was innite
in spatial extent, so that it would have no center of symmetry, and
therefore no preferred point to collapse toward. The trouble with
this argument is that the equilibrium it describes is unstable. Any
perturbation of the uniform density of matter breaks the symmetry,
leading to the collapse of some pocket of the universe. If the radius
of such a collapsing region is r, then its gravitational is proportional
to r
3
, and its gravitational eld is proportional to r
3
/r
2
= r. Since
its acceleration is proportional to its own size, the time it takes to
collapse is independent of its size. The prediction is that the uni-
verse will have a self-similar structure, in which the clumping on
small scales behaves in the same way as clumping on large scales;
zooming in or out in such a picture gives a landscape that appears
the same. With modern hindsight, this is actually not in bad agree-
ment with reality. We observe that the universe has a hierarchical
structure consisting of solar systems, galaxies, clusters of galaxies,
superclusters, and so on. Once such a structure starts to condense,
the collapse tends to stop at some point because of conservation of
angular momentum. This is what happened, for example, when our
own solar system formed out of a cloud of gas and dust.
Einstein confronted similar issues, but in a more acute form.
Newtons symmetry argument, which failed only because of its in-
stability, fails even more badly in relativity: the entire spacetime
can simply contract uniformly over time, without singling out any
particular point as a center. Furthermore, it is not obvious that
angular momentum prevents total collapse in relativity in the same
way that it does classically, and even if it did, how would that apply
to the universe as a whole? Einsteins Machian orientation would
have led him to reject the idea that the universe as a whole could
be in a state of rotation, and in any case it was sensible to start the
study of relativistic cosmology with the simplest and most symmet-
ric possible models, which would have no preferred axis of rotations.
Because of these issues, Einstein decided to try to patch up his
eld equation so that it would allow a static universe. Looking back
over the considerations that led us to this form of the equation,
we see that it is very nearly uniquely determined by the following
criteria:
5
Barcelo and Visser, Twilight for the energy conditions?, http://arxiv.
org/abs/gr-qc/0205066v1.
The equivalence principle is satised.
It should be coordinate-independent.
It should be equivalent to Newtonian gravity in the appropri-
ate limit.
It should not be overdetermined.
This is not meant to be a rigorous proof, just a general observation
that its not easy to tinker with the theory without breaking it.
A failed attempt at tinkering Example: 1
As an example of the lack of wiggle room in the structure of the
eld equations, suppose we construct the scalar T
a
a
, the trace of
the energy-momentum tensor, and try to insert it into the eld
equations as a further source term. The rst problem is that
the eld equation involves rank-2 tensors, so we cant just add
a scalar. To get around this, suppose we multiply by the metric.
We then have something like G
ab
= c
1
T
ab
+ c
2
g
ab
T
c
c
, where the
two constants c
1
and c
2
would be constrained by the requirement
that the theory agree with Newtonian gravity in the classical limit.
This particular attempt fails, because it violates the equivalence
principle. Consider a beam of light directed along the x axis.
Its momentum is equal to its energy (see page 106), so its con-
tributions to the local energy density and pressure are equal.
Thus its contribution to the energy-momentum tensor is of the
form T
= (constant) di ag(1, 1, 0, 0). The trace vanishes,

so its coupling to gravity in the c
2
term is zero. But this violates
the equivalence principle, which requires that all forms of mass-
energy contribute equally to gravitational mass.
One way in which we can change the eld equation without
violating any of these is to add a term g
ab
, giving
G
ab
= 8T
ab
+ g
ab
,
which is what we will refer to as the Einstein eld equation.
6
The
universal constant is called the cosmological constant. Einstein
originally introduced a positive cosmological constant because he
wanted relativity to be able to describe a static universe. To see
why it would have this eect, compare its behavior with that of
an ordinary uid. When an ordinary uid, such as the exploding
air-gas mixture in a cars cylinder, expands, it does work on its en-
vironment, and therefore by conservation of energy its own internal
energy is reduced. A positive cosmological constant, however, acts
like a certain amount of mass-energy built into every cubic meter of
vacuum. Thus when it expands, it releases energy. Its pressure is
negative.
6
In books that use a +++ metric rather then our +, the sign of the
cosmological constant term is reversed relative to ours.
Now consider the following pseudo-classical argument. Although
weve already seen (page 180) that there is no useful way to sepa-
rate the roles of kinetic and potential energy in general relativity,
suppose that there are some quantities analogous to them in the
description of the universe as a whole. (Well see below that the
universes contraction and expansion is indeed described by a set of
dierential equations that can be interpreted in essentially this way.)
If the universe contracts, a cubic meter of space becomes less than
a cubic meter. The cosmological-constant energy associated with
that volume is reduced, so some energy has been consumed. The
kinetic energy of the collapsing matter goes down, and the collapse
is decelerated.
The addition of the term constitutes a change to the vacuum
eld equations, and the good agreement between theory and exper-
iment in the case of, e.g., Mercurys orbit puts an upper limit on
then implies that must be small. For an order-of-magnitude
estimate, consider that has units of mass density, and the only
parameters with units that appear in the description of Mercurys
orbit are the mass of the sun, m, and the radius of Mercurys orbit,
r. The relativistic corrections to Mercurys orbit are on the order
of v
2
, or about 10
8
, and they come out right. Therefore we can es-
timate that the cosmological constant could not have been greater
than about (10
8
)m/r
3
10
10
kg/m
3
, or it would have caused
noticeable discrepancies. This is a very poor bound; if was this
big, we might even be able to detect its eects in laboratory ex-
periments. Looking at the role played by r in the estimate, we see
that the upper bound could have been made tighter by increasing
r. Observations on galactic scales, for example, constrain it much
more tightly. This justies the description of as cosmological: the
larger the scale, the more signicant the eect of a nonzero would
be.
Since the right-hand side of the eld equation is 8T
ab
+g
ab
, it
is possible to consider the cosmological constant as a type of matter
contributing to the energy-momentum tensor. Its negative pressure
causes a violation of the strong energy condition. If the cosmolog-
ical constant is a product of the quantum-mechanical structure of
the vacuum, then this is not too surprising, because quantum elds
are known to display negative energy. For example, the energy den-
sity between two parallel conducting plates is negative due to the
Casimir eect.
7.2 Cosmological solutions
We are thus led to pose two interrelated questions. First, what
can empirical observations about the universe tell us about the laws
of physics, such as the zero or nonzero value of the cosmological
constant? Second, what can the laws of physics tell us about the
Section 7.2 Cosmological solutions 203
large-scale structure of the universe, its origin, and its fate?
Surveys of distant quasars show that the universe has very little
structure at scales greater than a few times 10
25
m. (This can be
seen on a remarkable logarithmic map constructed by Gott et al.,
astro.princeton.edu/universe.) This suggests that we can, to
a good approximation, model the universe as being isotropic (the
same in all spatial directions) and homogeneous (the same at all
locations in space).
Motivated by Hubbles observation that the universe is expand-
ing, we hypothesize the existence of solutions of the eld equation
in which the properties of space are homogeneous and isotropic, but
the over-all scale of space is increasing as described by some scale
function a(t). Because of coordinate invariance, the metric can still
be written in a variety of forms. One such form is
ds
2
= dt
2
a(t)d
2
,
where the spatial part is
d
2
= f(r)dr
2
+r
2
d
2
+r
2
sin
2
d
2
.
In these coordinates, the time t is interpreted as the proper time of a
particle that has always been at rest. Events that are simultaneous
according to this t are events at which the local properties of the
universe i.e., its curvature are the same. These coordinates
are referred as the standard cosmological coordinates; one will
also encounter other choices, such as the comoving and conformal
coordinates, which are more convenient for certain purposes. His-
torically, the solution for the functions a and f was found by de
Sitter in 1917.
The unknown function f(r) has to make the 3-space metric d
2
have a constant Einstein curvature tensor. The following Maxima
program computes the curvature.
1 load(ctensor);
2 dim:3;
3 ct_coords:[r,theta,phi];
4 depends(f,t);
5 lg:matrix([f,0,0],
6 [0,r^2,0],
7 [0,0,r^2*sin(theta)^2]);
8 cmetric();
9 einstein(true);
Line 2 tells Maxima that were working in a space with three di-
mensions rather than its default of four. Line 4 tells it that f is a
function of time. Line 9 uses its built-in function for computing the
Einstein tensor G
a
b
. The result has only one nonvanishing compo-
nent, G
t
t
= (1 1/f)/r
2
. This has to be constant, and since scaling
can be absorbed in the factor a(t) in the 3+1-dimensional metric,
we can just set the value of G
tt
more or less arbitrarily, except for its
sign. The result is f = 1/(1kr
2
), where k = 1, 0, or 1. The form
of d
2
shows us that k can be interpreted in terms of the sign of the
spatial curvature. The k = 0 case gives a at space. For negative
k, a circle of radius r centered on the origin has a circumference
2rf(r) that is less than its Euclidean value of 2r. The opposite
occurs for k > 0. The resulting metric, called the Robertson-Walker
metric, is
ds
2
= dt
2
a
2
_
dr
2
1 kr
2
+r
2
d
2
+r
2
sin
2
d
2
_
.
Having xed f(r), we can now see what the eld equation tells
us about a(t). The next program computes the Einstein tensor for
the full four-dimensional spacetime:
1 load(ctensor);
3 depends(a,t);
4 lg:matrix([1,0,0,0],
5 [0,-a^2/(1-k*r^2),0,0],
6 [0,0,-a^2*r^2,0],
7 [0,0,0,-a^2*r^2*sin(theta)^2]);
8 cmetric();
9 einstein(true);
The result is
G
t
t
= 3
_
a
a
_
2
+ 3ka
2
G
r
r
= G
= G
= 2
a
a
+
_
a
a
_
2
+ka
2
,
where dots indicate dierentiation with respect to time.
Since we have G
a
b
with mixed upper and lower indices, we either
have to convert it into G
ab
, or write out the eld equations in this
mixed form. The latter turns out to be simpler. In terms of mixed
indices, g
a
b
is always simply diag(1, 1, 1, 1). Arbitrarily singling out
r = 0 for simplicity, we have g = diag(1, a
2
, 0, 0). The energy-
momentum tensor is T
= diag(, P, P, P). Substituting into

G
a
b
= 8T
a
b
+ g
a
b
, we nd
3
_
a
a
_
2
+ 3ka
2
= 8
2
a
a
+
_
a
a
_
2
+ka
2
= 8P .
Rearranging a little, we have a set of dierential equations known
as the Friedmann equations,
a
a
=
1
3

4
3
( + 3P)
_
a
a
_
2
=
1
3
+
8
3
ka
2
.
The cosmology that results from a solution of these dierential
equations is known as the Friedmann-Robertson-Walker (FRW) or
Friedmann-Lematre-Robertson-Walker (FLRW) cosmology.
7.2.1 Evidence for expansion of the universe
By 1929, Edwin Hubble at Mount Wilson had determined that
the universe was expanding rather than static, so that Einsteins
original goal of allowing a static cosmology became pointless. The
universe, it seemed, had originated in a Big Bang (a concept that
originated with the Belgian Roman Catholic priest Georges Le-
matre). This now appears natural, since the Friedmann equations
would only allow a constant a in the case where was perfectly
tuned relative to the other parameters. Einstein later referred to
the cosmological constant as the greatest blunder of my life, and
for the next 70 years it was commonly assumed that was exactly
zero.
Self-check: Why is it not correct to think of the Big Bang as an
explosion that occurred at a specic point in space?
The existence of the Big Bang is conrmed directly by looking up
in the sky and seeing it. In 1964, Penzias and Wilson at Bell Labora-
tories in New Jersey detected a mysterious background of microwave
radiation using a directional horn antenna. As with many acciden-
tal discoveries in science, the important thing was to pay attention
to the surprising observation rather than giving up and moving on
when it confounded attempts to understand it. They pointed the
antenna at New York City, but the signal didnt increase. The ra-
diation didnt show a 24-hour periodicity, so it couldnt be from a
source in a certain direction in the sky. They even went so far as to
sweep out the pigeon droppings inside. It was eventually established
that the radiation was coming uniformly from all directions in the
sky and had a black-body spectrum with a temperature of about 3
K.
This is now interpreted as follows. Soon after the Big Bang, the
universe was hot enough to ionize matter. An ionized gas is opaque
to light, since the oscillating elds of an electromagnetic wave ac-
celerate the charged particles, depositing kinetic energy into them.
Once the universe became cool enough, however, matter became
electrically neutral, and the universe became transparent. Light
from this time is the most long-traveling light that we can detect
now. The latest data show that transparency set in around 4 10
5
years after the big bang, when the temperature was about 3000 K.
The surface we see, dating back to this time, is known as the surface
of last scattering. Since then, the universe has expanded by about
a factor of 1000, causing the wavelengths of photons to be stretched
by the same amount due to the expansion of the underlying space.
This is equivalent to a Doppler shift due to the sources motion
away from us; the two explanations are equivalent. We therefore see
the 3000 K optical black-body radiation red-shifted to 3 K, in the
microwave region.
7.2.2 A singularity at the Big Bang
In section 6.3.1, we saw that a black hole contains a singularity.
However, it appears that black hole singularities are always hidden
behind event horizons, so that we can never observe them from the
outside. Now if we extrapolate the Friedmann equations backward
in time, we nd that they always have a = 0 at some point in the
past, and this occurs regardless of the details of what we assume
about the matter and radiation that lls the universe. To see this,
note that, as discussed in example 10 on page 110, radiation is
expected to dominate the early universe, for generic reasons that are
not sensitive to the (substantial) observational uncertainties about
the universes present-day mixture of ingredients. Under radiation-
dominated conditions, we can approximate = 0 and P = 0 in the
rst Friedmann equation, nding
a
a
=
4
3

where is the density of mass-energy due to radiation. Since a/a
is always negative, the graph of a(t) is always concave down, and
since a is currently increasing, there must be some time in the past
when a = 0. One can readily verify that this is not just a coordinate
singularity; the Ricci scalar curvature R
a
a
diverges, and the singu-
larity occurs at a nite proper time in the past. If this singularity
in the model corresponds to a singularity in the real universe, then
it is not a singularity that is hidden behind an event horizon. It lies
in our past light-cone, and our own world-lines emerged from it.
One of the Penrose-Hawking singularity theorems proves that
the Big Bang singularity is not just a product of the unrealistic
assumption of perfect symmetry that went into the FRW cosmol-
ogy. The only loophole is that if the cosmological constant is large
enough, it violates the strong energy equation, and we can have a
Big Bounce rather than a Big Bang (see page 217).
7.2.3 Observability of expansion
Brooklyn is not expanding!
The proper interpretation of the expansion of the universe, as
described by the Friedmann equations, can be tricky. It might seem
as though the expansion would be undetectable, in the sense that
general relativity is coordinate-independent, and therefore does not
pick out any preferred distance scale. That is, if all our meter-sticks
expand, and the rest of the universe expands as well, we would
have no way to detect the expansion. The aw in this reasoning is
that the Friedmann equations only describe the average behavior of
spacetime. As dramatized in the classic Woody Allen movie Annie
Hall: Well, the universe is everything, and if its expanding, some-
day it will break apart and that would be the end of everything!
What has the universe got to do with it? Youre here in Brooklyn!
Brooklyn is not expanding!
To organize our thoughts, lets consider the following hypotheses:
1. The distance between one galaxy and another increases at the
rate given by a(t) (assuming the galaxies are suciently dis-
tant from one another that they are not gravitationally bound
within the same galactic cluster, supercluster, etc.).
2. The wavelength of a photon increases according to a(t) as it
travels cosmological distances.
3. The size of the solar system increases at this rate as well (i.e.,
gravitationally bound systems get bigger, including the earth
and the Milky Way).
4. The size of Brooklyn increases at this rate (i.e., electromag-
netically bound systems get bigger).
5. The size of a helium nucleus increases at this rate (i.e., systems
bound by the strong nuclear force get bigger).
We can imagine that:
All the above hypotheses are true.
All the above hypotheses are false, and in fact none of these
sizes increases at all.
Some are true and some false.
If all ve hypotheses were true, the expansion would be unde-
tectable, because all available meter-sticks would be expanding to-
gether. Likewise if no sizes were increasing, there would be nothing
to detect. These two possibilities are really the same cosmology,
described in two dierent coordinate systems. But the Ricci and
Einstein tensors were carefully constructed so as to be intrinsic.
The fact that the expansion aects the Einstein tensor shows that
it cannot interpreted as a mere coordinate expansion. Specically,
suppose someone tells you that the FRW metric can be made into
a at-space metric by a change of coordinates. (I have come across
this claim on internet forums.) The linear structure of the ten-
sor transformation equations guarantees that a nonzero tensor can
never be made into a zero tensor by a change of coordinates. Since
the Einstein tensor is nonzero for an FRW metric, and zero for a
at-space metric, the claim is false.
We can now see some of the limitations of a common metaphor
used to explain cosmic expansion, in which the universe is visual-
ized as the surface of an expanding balloon. The metaphor correctly
gets across several ideas: that the Big Bang is not an explosion that
occurred at a preexisting point in empty space; that hypothesis 1
above holds; and that the rate of recession of one galaxy relative
to another is proportional to the distance between them. Neverthe-
less the metaphor may be misleading, because if we take a laundry
marker and draw any structure on the balloon, that structure will
expand at the same rate. But this implies that hypotheses 1-5 all
hold, which cannot be true.
Since some of the ve hypotheses must be true and some false,
and we would like to sort out which are which. It should also be
clear by now that these are not ve independent hypotheses. For
example, we can test empirically whether the ratio of Brooklyns
size to the distances between galaxies changes like a(t), remains
constant, or changes with some other time dependence, but it is
only the ratio that is actually observable.
Empirically, we nd that hypotheses 1 and 2 are true (i.e., the
photons wavelength maintains a constant ratio with the intergalac-
tic distance scale), while 3, 4, and 5 are false. For example, the
orbits of the planets in our solar system have been measured ex-
tremely accurately by radar reection and by signal propagation
times to space probes, and no expanding trend is detected.
General-relativistic predictions
Does general relativity correctly reproduce these observations?
General relativity is mainly a theory of gravity, so it should be well
within its domain to explain why the solar system does not expand
while intergalactic distances do. It is impractical to solve the Ein-
stein eld equations exactly so as to describe the internal structure
of all the bodies that occupy the universe: galaxies, superclusters,
etc. We can, however, handle simple cases, as in example 4 on page
216, where we display an exact solution for the case of a universe
containing only two things: an isolated black hole, and an energy
density described by a cosmological constant. We nd that the char-
acteristic scale of the black hole, e.g., the radius of its event horizon,
is still set by the constant mass m, so we can see that cosmological
expansion does not aect the size of this gravitationally bound sys-
tem. We can also imagine putting a test particle in a circular orbit
around the black hole. Since the metric near the black hole is very
nearly the same as an ordinary Schwarzschild metric, we nd that
the test particles orbit does not expand by any signicant amount.
Estimates have also been carried out for more realistic cosmologies
and for actual systems of interest such as the solar system.
7
For ex-
ample, the predicted general-relativistic eect on the radius of the
earths orbit since the time of the dinosaurs is calculated to be about
as big as the diameter of an atomic nucleus; if the earths orbit had
expanded according to a(t), the increase would have been millions
of kilometers.
It is more dicult to demonstrate by explicit calculation that
atoms and nuclei do not expand, since we do not have a theory of
quantum gravity at our disposal. It is, however, easy to see that
such an expansion would violate either the equivalence principle or
the basic properties of quantum mechanics. One way of stating
the equivalence principle is that the local geometry of spacetime is
always approximately Lorentzian, so that the the laws of physics do
not depend on ones position or state of motion. Among these laws
of physics are the principles of quantum mechanics, which imply
that an atom or a nucleus has a well-dened ground state, with a
certain size that depends only on fundamental constants such as
Plancks constant and the masses of the particles involved.
This is dierent from the case of a photon traveling across the
universe. The argument given above fails, because the photon does
not have a ground state. The photon does expand, and this is
required by the correspondence principle. If the photon did not ex-
pand, then its wavelength would remain constant, and this would
be inconsistent with the classical theory of electromagnetism, which
predicts a Doppler shift due to the relative motion of the source
and the observer. One can choose to describe cosmological redshifts
either as Doppler shifts or as expansions of wavelength due to cos-
mological expansion.
More than one dimension required
Another good way of understanding why a photon expands,
while an atom does not, is to recall that a one-dimensional space
can never have any intrinsic curvature. If the expansion of atoms
were to be detectable, we would need to detect it by comparing
against some other meter-stick. Lets suppose that a hydrogen atom
expands more, while a more tightly bound uranium atom expands
less, so that over time, we can detect a change in the ratio of the two
atoms sizes. The world-lines of the two atoms are one-dimensional
curves in spacetime. They are housed in a laboratory, and although
the laboratory does have some spatial extent, the equivalence prin-
ciple guarantees that to a good approximation, this small spatial
extent doesnt matter. This implies an intrinsic curvature in a one-
dimensional space, which is mathematically impossible, so we have
a proof by contradiction that atoms do not expand.
7
http://arxiv.org/abs/astro-ph/9803097v1
Now why does this one-dimensionality argument fail for photons
and galaxies? For a pair of galaxies, it fails because the galaxies are
not suciently close together to allow them both to be covered by
a single Lorentz frame, and therefore the set of world-lines com-
prising the observation cannot be approximated well as lying within
a one-dimensional space. Similar reasoning applies for cosmologi-
cal redshifts of photons received from distant galaxies. One could
instead propose ying along in a spaceship next to an electromag-
netic wave, and monitoring the change in its wavelength while it is
in ight. All the world-lines involved in such an experiment would
indeed be conned to a one-dimensional space. The experiment is
impossible, however, because the measuring apparatus cannot be
accelerated to the speed of light. In reality, the speed of the light
wave relative to the measuring apparatus will always equal c, so the
two world-lines involved in the experiment will diverge, and will not
be conned to a one-dimensional region of spacetime.
stvangs quasi-metric relativity Example: 2
Over the years, a variety of theories of gravity have been pro-
posed as alternatives to general relativity. Some of these, such
as the Brans-Dicke theory, remain viable, i.e., they are consis-
tent with all the available experimental data that have been used
to test general relativity. One of the most important reasons for
trying to construct such theories is that it can be impossible to
interpret tests of general relativitys predictions unless one also
possesses a theory that predicts something different. This issue,
for example, has made it impossible to test Einsteins century-old
prediction that gravitational effects propagate at c, since there is
no viable theory available that predicts any other speed for them
(see section 8.1).
stvang (arxiv.org/abs/gr-qc/0112025v6) has proposed an al-
ternative theory of gravity, called quasi-metric relativity, which,
unlike general relativity, predicts a signicant cosmological ex-
pansion of the solar system, and which is claimed to be able
to explain the observation of small, unexplained accelerations of
the Pioneer space probes that remain after all accelerations due
to known effects have been subtracted (the Pioneer anomaly).
Weve seen above that there are a variety of arguments against
such an expansion of the solar system, and that many of these
arguments do not require detailed technical calculations but only
knowledge of certain fundamental principles, such as the struc-
ture of differential geometry (no intrinsic curvature in one dimen-
sion), the equivalence principle, and the existence of ground states
in quantum mechanics. We therefore expect that stvangs the-
ory, if it is logically self-consistent, will probably violate these as-
sumptions, but that the violations must be relatively small if the
theory is claimed to be consistent with existing observations. This
is in fact the case. The theory violates the strictest form of the
equivalence principle.
Over the years, a variety of explanations have been proposed
for the Pioneer anomaly, including both glamorous ones (a mod-
ication of the 1/r
2
law of gravitational forces) and others more
pedestrian (effects due to outgassing of fuel or radiation pressure
from sunlight). Calculations by Iorio
8
in 2006-2009 show that if
the force law for gravity is modied in order to explain the Pio-
neer anomalies, and if gravity obeys the equivalence principle,
then the results are inconsistent with the observed orbital motion
of the satellites of Neptune. This makes gravitational explana-
tions unlikely, but does not obviously rule out stvangs theory,
since the theory is not supposed to obey the equivalence prin-
ciple. stvang says
9
that his theory predicts an expansion of
1m/yr in the orbit of Tritons moon Nereid, which is consistent
with observation.
Does space expand?
Finally, the balloon metaphor encourages us to interpret cosmo-
logical expansion as a phenomenon in which space itself expands,
or perhaps one in which new space is produced. Does space really
expand? Without posing the question in terms of more rigorously
dened, empirically observable quantities, we cant say yes or no.
It is merely a matter of which denitions one chooses and which
conceptual framework one nds easier and more natural to work
within. Bunn and Hogg have stated the minority view against ex-
pansion of space
10
, while the opposite opinion is given by Francis
et al.
11
As an example of a self-consistent set of denitions that
lead to the conclusion that space does expand, Francis et al. give
the following. Dene eight observers positioned at the corners of a
cube, at cosmological distances from one another. Let each observer
be at rest relative to the local matter and radiation that were used
as ingredients in the FRW cosmology. (For example, we know that
our own solar system is not at rest in this sense, because we observe
that the cosmic microwave background radiation is slightly Doppler
shifted in our frame of reference.) Then these eight observers will
observe that, over time, the volume of the cube grows according to
the function a(t) in the FRW model.
7.2.4 The vacuum-dominated solution
For 70 years after Hubbles discovery of cosmological expansion,
the standard picture was one in which the universe expanded, but
the expansion must be decelerating. The deceleration is predicted
by the special cases of the FRW cosmology that were believed to
be applicable, and even if we didnt know anything about general
8
http://arxiv.org/abs/0912.2947v1
9
private communication, Jan. 4, 2010
10
11
relativity, it would be reasonable to expect a deceleration due to the
mutual Newtonian gravitational attraction of all the mass in the
universe.
But observations of distant supernovae starting around 1998 in-
troduced a further twist in the plot. In a binary star system con-
sisting of a white dwarf and a non-degenerate star, as the non-
degenerate star evolves into a red giant, its size increases, and it
can begin dumping mass onto the white dwarf. This can cause the
white dwarf to exceed the Chandrasekhar limit (page 120), resulting
in an explosion known as a type Ia supernova. Because the Chan-
drasekhar limit provides a uniform set of initial conditions, the be-
havior of type Ia supernovae is fairly predictable, and in particular
their luminosities are approximately equal. They therefore provide
a kind of standard candle: since the intrinsic brightness is known,
the distance can be inferred from the apparent brightness. Given
the distance, we can infer the time that was spent in transit by the
light on its way to us, i.e. the look-back time. From measurements
of Doppler shifts of spectral lines, we can also nd the velocity at
which the supernova was receding from us. The result is that we
can measure the universes rate of expansion as a function of time.
Observations show that this rate of expansion has been accelerating.
The Friedmann equations show that this can only occur for 4.
This picture has been independently veried by measurements of the
cosmic microwave background (CMB) radiation. A more detailed
discussion of the supernova and CMB data is given in section 7.2.6
on page 219.
With hindsight, we can see that in a quantum-mechanical con-
text, it is natural to expect that uctuations of the vacuum, required
by the Heisenberg uncertainty principle, would contribute to the cos-
mological constant, and in fact models tend to overpredict by a
factor of about 10
120
! From this point of view, the mystery is why
these eects cancel out so precisely. A correct understanding of the
cosmological constant presumably requires a full theory of quantum
gravity, which is presently far out of our reach.
The latest data show that our universe, in the present epoch, is
dominated by the cosmological constant, so as an approximation we
can write the Friedmann equations as
a
a
=
1
3
_
a
a
_
2
=
1
3
.
This is referred to as a vacuum-dominated universe. The solution is
a = exp
_
_
3
t
_
.
The implications for the fate of the universe are depressing. All parts
of the universe will accelerate away from one another faster and
faster as time goes on. The relative separation between two objects,
say galaxy A and galaxy B, will eventually be increasing faster than
the speed of light. (The Lorentzian character of spacetime is local,
so relative motion faster than c is only forbidden between objects
that are passing right by one another.) At this point, an observer
in either galaxy will say that the other one has passed behind an
event horizon. If intelligent observers do actually exist in the far
future, they may have no way to tell that the cosmos even exists.
They will perceive themselves as living in island universes, such as
we believed our own galaxy to be a hundred years ago.
When I introduced the standard cosmological coordinates on
page 204, I described them as coordinates in which events that
are simultaneous according to this t are events at which the local
properties of the universe are the same. In the case of a perfectly
vacuum-dominated universe, however, this notion loses its meaning.
The only observable local property of such a universe is the vacuum
energy described by the cosmological constant, and its density is al-
ways the same, because it is built into the structure of the vacuum.
Thus the vacuum-dominated cosmology is a special one that maxi-
mally symmetric, in the sense that it has not only the symmetries of
homogeneity and isotropy that weve been assuming all along, but
also a symmetry with respect to time: it is a cosmology without
history, in which all times appear identical to a local observer. In
the special case of this cosmology, the time variation of the scaling
factor a(t) is unobservable, and may be thought of as the unfortu-
nate result of choosing an inappropriate set of coordinates, which
obscure the underlying symmetry. When I argued in section 7.2.3
for the observability of the universes expansion, note that all my
arguments assumed the presence of matter or radiation. These are
completely absent in a perfectly vacuum-dominated cosmology.
For these reasons de Sitter originally proposed this solution as a
static universe in 1927. But by 1920 it was realized that this was an
oversimplication. The argument above only shows that the time
variation of a(t) does not allow us to distinguish one epoch of the
universe from another. That is, we cant look out the window and
infer the date (e.g., from the temperature of the cosmic microwave
background radiation). It does not, however, imply that the uni-
verse is static in the sense that had been assumed until Hubbles
observations. The r-t part of the metric is
ds
2
= dt
2
a
2
dr
2
,
where a blows up exponentially with time, and the k-dependence
has been neglected, as it was in the approximation to the Friedmann
equations used to derive a(t).
12
Let a test particle travel in the radial
direction, starting at event A = (0, 0) and ending at B = (t
, r
). In
12
A computation of the Einstein tensor with ds
2
= dt
2
a
2
(1 kr
2
)
1
dr
2
shows that k enters only via a factor the form (. . .)e
(...)t
+(. . .)k. For large t, the
at space, a world-line of the linear form r = vt would be a geodesic
connecting A and B; it would maximize the particles proper time.
But in the this metric, it cannot be a geodesic. The curvature of
geodesics relative to a line on an r-t plot is most easily understood
in the limit where t
is fairly long compared to the time-scale T =

_
3/ of the exponential, so that a(t
) is huge. The particles best

strategy for maximizing its proper time is to make sure that its dr
is extremely small when a is extremely large. The geodesic must
therefore have nearly constant r at the end. This makes it sound as
though the particle was decelerating, but in fact the opposite is true.
If r is constant, then the particles spacelike distance from the origin
is just ra(t), which blows up exponentially. The near-constancy of
the coordinate r at large t actually means that the particles motion
at large t isnt really due to the particles inertial memory of its
original motion, as in Newtons rst law. What happens instead
is that the particles initial motion allows it to move some distance
away from the origin during a time on the order of T, but after that,
the expansion of the universe has become so rapid that the particles
motion simply streams outward because of the expansion of space
itself. Its initial motion only mattered because it determined how
far out the particle got before being swept away by the exponential
expansion.
Geodesics in a vacuum-dominated universe Example: 3
In this example we conrm the above interpretation in the special
case where the particle, rather than being released in motion at
the origin, is released at some nonzero radius r , with dr /dt = 0
initially. First we recall the geodesic equation
d
2
x
i
d
2
=
i
j k
dx
j
d
dx
k
d
.
from page 146. The nonvanishing Christoffel symbols for the 1+1-
dimensional metric ds
2
= dt
2
a
2
dr
2
are
r
t r
= a/a and
t
r r
=
aa. Setting T = 1 for convenience, we have
r
t r
= 1 and
t
r r
=
e
2t
.
We conjecture that the particle remains at the same value of r .
Given this conjecture, the particles proper time
_
ds is simply the
same as its time coordinate t , and we can therefore use t as an
k term becomes negligible, and the Einstein tensor becomes G
a
b
= g
a
b
, This is
consistent with the approximation we used in deriving the solution, which was
to ignore both the source terms and the k term in the Friedmann equations.
The exact solutions with > 0 and k = 1, 0, and 1 turn out in fact to be
equivalent except for a change of coordinates.
afne coordinate. Letting = t , we have
d
2
t
dt
2

t
r r
_
dr
dt
_
2
= 0
0
t
r r
r
2
= 0
r = 0
r = constant
This conrms the self-consistency of the conjecture that r = constant
is a geodesic.
Note that we never actually had to use the actual expressions for
the Christoffel symbols; we only needed to know which of them
vanished and which didnt. The conclusion depended only on
the fact that the metric had the form ds
2
= dt
2
a
2
dr
2
for some
function a(t ). This provides a rigorous justication for the inter-
pretation of the cosmological scale factor a as giving a universal
time-variation on all distance scales.
The calculation also conrms that there is nothing special about
r = 0. A particle released with r = 0 and r = 0 initially stays at
r = 0, but a particle released at any other value of r also stays
at that r . This cosmology is homogeneous, so any point could
have been chosen as r = 0. If we sprinkle test particles, all at
rest, across the surface of a sphere centered on this arbitrarily
chosen point, then they will all accelerate outward relative to one
another, and the volume of the sphere will increase. This is ex-
actly what we expect. The Ricci curvature is interpreted as the
second derivative of the volume of a region of space dened by
test particles in this way. The fact that the second derivative is
positive rather than negative tells us that we are observing the
kind of repulsion provided by the cosmological constant, not the
attraction that results from the existence of material sources.
Schwarzschild-de Sitter space Example: 4
The metric
ds
2
=
_
1
2m
r

1
3
r
2
_
dt
2
dr
2
1
2m
r

1
3
r
2
r
2
d
2
r
2
sin
2
d
2
is an exact solution to the Einstein eld equations with cosmo-
logical constant , and can be interpreted as a universe in which
the only mass is a black hole of mass m located at r = 0. Near
the black hole, the terms become negligible, and this is simply
the Schwarzschild metric. As argued in section 7.2.3, page 207,
this is a simple example of how cosmological expansion does not
cause all structures in the universe to grow at the same rate.
The Big Bang singularity in a universe with a cosmological
constant
On page 207 we discussed the possibility that the Big Bang
singularity was an artifact of the unrealistically perfect symmetry
assumed by our cosmological models, and we found that this was
not the case: the Penrose-Hawking singularity theorems demon-
strate that the singularity is real, provided that the cosmological
constant is zero. The cosmological constant is not zero, however.
Models with a very large positive cosmological constant can also
display a Big Bounce rather than a Big Bang. If we imagine us-
ing the Friedmann equations to evolve the universe backward in
time from its present state, the scaling arguments of example 10 on
page 110 suggest that at early enough times, radiation and mat-
ter should dominate over the cosmological constant. For a large
enough value of the cosmological constant, however, it can happen
that this switch-over never happens. In such a model, the universe
is and always has been dominated by the cosmological constant,
and we get a Big Bounce in the past because of the cosmological
constants repulsion. In this book I will only develop simple cos-
mological models in which the universe is dominated by a single
component; for a discussion of bouncing models with both matter
and a cosmological constant, see Carroll, The Cosmological Con-
stant, http://www.livingreviews.org/lrr-2001-1. By 2008, a
variety of observational data had pinned down the cosmological con-
stant well enough to rule out the possibility of a bounce caused by
a very strong cosmological constant.
7.2.5 The matter-dominated solution
Our universe is not perfectly vacuum-dominated, and in the past
it was even less so. Let us consider the matter-dominated epoch,
in which the cosmological constant was negligible compared to the
material sources. The equations of state for nonrelativistic matter
(p. 110) are
P = 0
a
3
so the Friedmann equations become
a
a
=
4
3

_
a
a
_
2
=
8
3
ka
2
,
where for compactness s dependence on a, with some constant
of proportionality, is not shown explicitly. A static solution, with
constant a, is impossible, and a is negative, which we can inter-
pret semiclassically in terms of the deceleration of the matter in the
universe due to gravitational attraction. There are three cases to
consider, according to the value of k.
The closed universe
Weve seen that k = +1 describes a universe in which the spatial
curvature is positive, i.e., the circumference of a circle is less than
its Euclidean value. By analogy with a sphere, which is the two-
dimensional surface of constant positive curvature, we expect that
the total volume of this universe is nite.
The second Friedmann equation also shows us that at some value
of a, we will have a = 0. The universe will expand, stop, and then
recollapse, eventually coming back together in a Big Crunch which
is the time-reversed version of the Big Bang.
Suppose we were to describe an initial-value problem in this
cosmology, in which the initial conditions are given for all points in
the universe on some spacelike surface, say t = constant. Since the
universe is assumed to be homogeneous at all times, there are really
only three numbers to specify, a, a, and : how big is the universe,
how fast is it expanding, and how much matter is in it? But these
three pieces of data may or may not be consistent with the second
Friedmann equation. That is, the problem is overdetermined. In
particular, we can see that for small enough values of , we do
not have a valid solution, since the square of a/a would have to be
negative. Thus a closed universe requires a certain amount of matter
in it. The present observational evidence (from supernovae and the
cosmic microwave background, as described above) is sucient to
show that our universe does not contain this much matter.
The at universe
The case of k = 0 describes a universe that is spatially at.
It represents a knife-edge case lying between the closed and open
universes. In a semiclassical analogy, it represents the case in which
the universe is moving exactly at escape velocity; as t approaches
innity, we have a , 0, and a 0. This case, unlike the
others, allows an easy closed-form solution to the motion. Let the
constant of proportionality in the equation of state a
3
be xed
by setting 4/3 = ca
3
. The Friedmann equations are
a = ca
2
a =
2ca
1/2
.
Looking for a solution of the form a t
p
, we nd that by choosing
p = 2/3 we can simultaneously satisfy both equations. The constant
c is also xed, and we can investigate this most transparently by
recognizing that a/a is interpreted as the Hubble constant, H, which
is the constant of proportionality relating a far-o galaxys velocity
to its distance. Note that H is a constant in the sense that it is
the same for all galaxies, in this particular model with a vanishing
cosmological constant; it does not stay constant with the passage
of cosmological time. Plugging back into the original form of the
a / The angular scale of uc-
tuations in the cosmic microwave
background can be used to infer
the curvature of the universe.
Friedmann equations, we nd that the at universe can only exist if
the density of matter satises =
crit
= 2H
2
/8 = 2H
2
/8G. The
observed value of the Hubble constant is about 1/(14 10
9
years),
which is roughly interpreted as the age of the universe, i.e., the
proper time experienced by a test particle since the Big Bang. This
gives
crit
10
26
kg/m
3
.
The open universe
The k = 1 case represents a universe that has negative spatial
curvature, is spatially innite, and is also innite in time, i.e., even
if the cosmological constant had been zero, the expansion of the uni-
verse would have had too little matter in it to cause it to recontract
and end in a Big Crunch.
7.2.6 Observation
Historically, it was very dicult to determine the universes av-
erage density, even to within an order of magnitude. Most of the
matter in the universe probably doesnt emit light, making it dif-
cult to detect. Astronomical distance scales are also very poorly
calibrated against absolute units such as the SI.
A strong constraint on the models comes from accurate mea-
surements of the cosmic microwave background, especially by the
1989-1993 COBE probe, and its 2001-2009 successor, the Wilkin-
son Microwave Anisotropy Probe, positioned at the L2 Lagrange
point of the earth-sun system, beyond the Earth on the line con-
necting sun and earth. The temperature of the cosmic microwave
background radiation is not the same in all directions, and its can
be measured at dierent angles. In a universe with negative spatial
curvature, the sum of the interior angles of a triangle is less than
the Euclidean value of 180 degrees. Therefore if we observe a varia-
tion in the CMB over some angle, the distance between two points
on the surface of last scattering is actually greater than would have
been inferred from Euclidean geometry. The distance scale of such
variations is limited by the speed of sound in the early universe, so
one can work backward and infer the universes spatial curvature
based on the angular scale of the anisotropies. The measurements
of spatial curvature are usually stated in terms of the parameter ,
dened as the total average density of all source terms in the Ein-
stein eld equations, divided by the critical density that results in
a at universe. includes contributions from matter,
M
, the cos-
mological constant,
, and radiation (negligible in the present-day

unverse). The results from WMAP, combined with other data from
other methods, gives = 1.005 .006. In other words, the universe
is very nearly spatially at.
The supernova data described on page 212 complement the CMB
data because they are mainly sensitive to the dierence

M
,
rather than their sum =
+
M
. This is because these data
c / The cosmological param-
eters of our universe, after
Perlmutter et al., arxiv.org/
abs/astro-ph/9812133.
b / A Hubble plot for distant
supernovae. Each data point
represents an average over
several different supernovae with
nearly the same z.
measure the acceleration or deceleration of the universes expansion.
Matter produces deceleration, while the cosmological constant gives
acceleration. Figure b shows some recent supernova data.
13
The
horizontal axis gives the redshift factor z = (
)/, where
is
the wavelength observed on earth and the wavelength originally
emitted. It measures how fast the supernovas galaxy is receding
from us. The vertical axis is (mM) = (mM) (mM)
empty
,
where m is the apparent magnitude, M is the absolute magnitude,
and (m M)
empty
is the value expected in a model of an empty
universe, with = 0. The dierence mM is a measure of distance,
so essentially this is a graph of distance versus recessional velocity, of
the same general type used by Hubble in his original discovery of the
expansion of the universe. Subtracting (mM)
empty
on the vertical
axis makes it easier to see small dierences. Since the WMAP data
require = 1, we need to t the supernova data with values of
M
and
that add up to one. Attempting to do so with

M
= 1 and
= 0 is clearly inconsistent with the data, so we can conclude

that the cosmological constant is denitely positive.
Figure c summarizes what we can conclude about our universe,
parametrized in terms of a model with both
M
and
nonzero.
14
We can tell that it originated in a Big Bang singularity, that it
will go on expanding forever, and that it is very nearly at. Note
that in a cosmology with nonzero values for both
M
and
, there
is no strict linkage between the spatial curvature and the question
of recollapse, as there is in a model with only matter and no cosmo-
logical constant; therefore even though we know that the universe
will not recollapse, we do not know whether its spatial curvature is
slightly positive (closed) or negative (open).
Astrophysical considerations provide further constraints and con-
sistency checks. In the era before the advent of high-precision cos-
mology, estimates of the age of the universe ranged from 10 billion
to 20 billion years, and the low end was inconsistent with the age
of the oldest globular clusters. This was believed to be a problem
either for observational cosmology or for the astrophysical models
used to estimate the age of the clusters: You cant be older than
your ma. Current data have shown that the low estimates of the
age were incorrect, so consistency is restored.
Another constraint comes from models of nucleosynthesis dur-
ing the era shortly after the Big Bang (before the formation of the
rst stars). The observed relative abundances of hydrogen, helium,
and deuterium cannot be reconciled with the density of dust (i.e.,
nonrelativistic matter) inferred from the observational data. If the
inferred mass density were entirely due to normal baryonic matter
(i.e., matter whose mass consisted mostly of protons and neutrons),
13
Riess et al., 2007, arxiv.org/abs/astro-ph/0611572
14
See Carroll, The Cosmological Constant, http://www.livingreviews.
org/lrr-2001-1 for a full mathematical treatment of such models.
then nuclear reactions in the dense early universe should have pro-
ceeded relatively eciently, leading to a much higher ratio of helium
to hydrogen, and a much lower abundance of deuterium. The con-
clusion is that most of the matter in the universe must be made of
an unknown type of exotic non-baryonic matter, known generically
as dark matter.
7.3 Machs principle revisited
7.3.1 The Brans-Dicke theory
Mach himself never succeeded in stating his ideas in the form of
a precisely testable physical theory, and weve seen that to the ex-
tent that Einsteins hopes and intuition had been formed by Machs
ideas, he often felt that his own theory of gravity came up short.
The reader has so far encountered Machs principle in the context of
certain thought experiments that are obviously impossible to realize,
involving a hypothetical universe that is empty except for certain
apparatus (e.g., section 3.5.2, p. 96). It would be easy, then, to get
an impression of Machs principle as one of those theories that is
not even wrong, i.e., so ill-dened that it cannot even be falsied
by experiment, any more than Christianity can be.
But in 1961, Robert Dicke and his student Carl Brans came up
with a theory of gravity that made testable predictions, and that
was specically designed to be more Machian than general relativity.
Their paper
15
is extremely readable, even for the non-specialist. On
the rst page, they propose one of those seemingly foolish thought
experiments about a nearly empty universe:
The imperfect expression of [Machs ideas] in general
relativity can be seen by considering the case of a space
empty except for a lone experimenter in his laboratory.
[...] The observer would, according to general relativity,
observe normal behavior of his apparatus in accordance
with the usual laws of physics. However, also accord-
ing to general relativity, the experimenter could set his
laboratory rotating by leaning out a window and ring
his 22-caliber rie tangentially. Thereafter the delicate
gyroscope in the laboratory would continue to point in a
direction nearly xed relative to the direction of motion
of the rapidly receding bullet. The gyroscope would ro-
tate relative to the walls of the laboratory. Thus, from
the point of view of Mach, the tiny, almost massless,
very distant bullet seems to be more important than the
massive, nearby walls of the laboratory in determining
15
C. Brans and R. H. Dicke, Machs Principle and a Relativistic Theory of
Gravitation, Physical Review 124 (1961) 925
Section 7.3 Machs principle revisited 221
inertial coordinate frames and the orientation of the gy-
roscope.
They then proceed to construct a mathematical and more Machian
theory of gravity. From the Machian point of view, the correct lo-
cal denition of an inertial frame must be determined relative to the
bulk of the matter in the universe. We want to retain the Lorentzian
local character of spacetime, so this inuence cant be transmitted
via instantaneous action at a distance. It must propagate via some
physical eld, at a speed less than or equal to c. It is implausible that
this eld would be the gravitational eld as described by general rel-
ativity. Suppose we divide the cosmos up into a series of concentric
spherical shells centered on our galaxy. In Newtonian mechanics,
the gravitational eld obeys Gausss law, so the eld of such a shell
vanishes identically on the interior. In relativity, the corresponding
statement is Birkhos theorem, which states that the Schwarzschild
metric is the unique spherically symmetric solution to the vacuum
eld equations. Given this solution in the exterior universe, we can
set a boundary condition at the outside surface of the shell, use the
Einstein eld equations to extend the solution through it, and nd
a unique solution on the interior, which is simply a at space.
Since the Machian eect cant be carried by the gravitational
eld, Brans and Dicke took up an idea earlier proposed by Pascual
Jordan
16
of hypothesizing an auxiliary eld . The fact that such a
eld has never been detected directly suggests that it has no mass or
charge. If it is massless, it must propagate at exactly c, and this also
makes sense because if it were to propagate at speeds less than c,
there would be no obvious physical parameter that would determine
that speed. How many tensor indices should it have? Since Machs
principle tries to give an account of inertia, and inertial mass is a
scalar, should presumably be a scalar (quantized by a spin-zero
particle). Theories of this type are called tensor-scalar theories,
because they use a scalar eld in addition to the metric tensor.
The wave equation for a massless scalar eld, in the absence of
sources, is simply
i
i
= 0. The solutions of this wave equa-
tion fall o as 1/r. This is gentler than the 1/r
2
variation of
the gravitational eld, so results like Newtons shell theorem and
Birkhos theorem no longer apply. If a spherical shell of mass acts
as a source of , then can be nonzero and varying inside the shell.
The that you experience right now as you read this book should be
a sum of wavelets originating from all the masses whose world-lines
intersected the surface of your past light-cone. In a static universe,
this sum would diverge linearly, so a self-consistency requirement for
Brans-Dicke gravity is that it should produce cosmological solutions
16
Jordan was a member of the Nazi Sturmabteilung or brown shirts who
nevertheless ran afoul of the Nazis for his close professional relationships with
Jews.
that avoid such a divergence, e.g., ones that begin with Big Bangs.
Masses are the sources of the eld . How should they cou-
ple to it? Since is a scalar, we need to construct a scalar as its
source, and the only reasonable scalar that can play this role is the
trace of the stress-energy tensor, T
i
i
. As discussed in example 1
on page 202, this vanishes for light, so only material particles are
sources of . Even so, the Brans-Dicke theory retains a form of the
equivalence principle. As discussed on pp. 34 and 29, the equiva-
lence principle is a statement about the results of local experiments,
and at any given location in the universe is dominated by con-
tributions from matter lying at cosmological distances. Objects of
dierent composition will have diering fractions of their mass that
arise from internal electromagnetic elds. Two such objects will still
follow identical geodesics, since their own eect on the local value
of is negligible. This is unlike the behavior of electrically charged
objects, which experience signicant back-reaction eects in curved
space (p. 34). However, the strongest form of the equivalence princi-
ple requires that all experiments in free-falling laboratories produce
identical results, no matter where and when they are carried out.
Brans-Dicke gravity violates this, because such experiments could
detect dierences between the value of at dierent locations
but of course this is part and parcel of the purpose of the theory.
We now need to see how to connect to the local notion of
inertia so as to produce an eect of the kind that would tend to
fulll Machs principle. In Machs original formulation, this would
entail some kind of local rescaling of all inertial masses, but Brans
and Dicke point out that in a theory of gravity, this is equivalent to
scaling the Newtonian gravitational constant G down by the same
factor. The latter turns out to be a better approach. For one thing,
it has a natural interpretation in terms of units. Since s amplitude
falls o as 1/r, we can write m
i
/r, where the sum is over the
past light cone. If we then make the identication of with 1/G
(or c
2
/G in a system wher c ,= 1), the units work out properly, and
the coupling constant between matter and can be unitless. If this
coupling constant, notated 1/, were not unitless, then the theorys
predictive value would be weakened, because there would be no way
to know what value to pick for it. For a unitless constant, however,
there is a reasonable way to guess what it should be: in any sensible
theory, Brans and Dicke write, must be of the general order of
magnitude of unity. This is, of course, assuming that the Brans-
Dicke theory was correct. In general, there are other reasonable
values to pick for a unitless number, including zero and innity. The
limit of recovers the special case of general relativity. Thus
Machs principle, which once seemed too vague to be empirically
falsiable, comes down to measuring a specic number, , which
quanties how non-Machian our universe is.
17
17
There are also good technical reasons for thinking of as relating to the
7.3.2 Predictions of the Brans-Dicke theory
Returning to the example of the spherical shell of mass, we can
see based on considerations of units that the value of inside should
be m/r, where m is the total mass of the shell and r is its radius.
There may be a unitless factor out in front, which will depend on ,
but for 1 we expect this constant to be of order 1. Solving the
nasty set of eld equations that result from their Lagrangian, Brans
and Dicke indeed found [2/(3 + 2)](m/r), where the constant
in square brackets is of order unity if is of order unity. In the
limit of , = 0, and the shell has no physical eect on its
interior, as predicted by general relativity.
Brans and Dicke were also able to calculate cosmological models,
and in a typical model with a nearly spatially at universe, they
found would vary according to
= 8
4 + 3
6 + 4
o
t
2
o
_
t
t
o
_
2/(4+3)
,
where
o
is the density of matter in the universe at time t = t
o
.
When the density of matter is small, G is large, which has the same
observational consequences as the disappearance of inertia; this is
exactly what one expects according to Machs principle. For ,
the gravitational constant G = 1/ really is constant.
Returning to the thought experiment involving the 22-caliber ri-
e red out the window, we nd that in this imaginary universe, with
a very small density of matter, G should be very large. This causes
a frame-dragging eect from the laboratory on the gyroscope, one
much stronger than we would see in our universe. Brans and Dicke
calculated this eect for a laboratory consisting of a spherical shell,
and although technical diculties prevented the reliable extrapo-
lation of their result to
o
0, the trend was that as
o
became
small, the frame-dragging eect would get stronger and stronger,
gravitational constant is that general relativity has a standard prescription for
describing elds on a background of curved spacetime. The vacuum eld equa-
tions of general relativity can be derived from the principle of least action, and
although the details are beyond the scope of this book (see, e.g., Wald, General
Relativity, appendix E), the general idea is that we dene a Lagrangian density
LG that depends on the Ricci scalar curvature, and then extremize its integral
over all possible histories of the evolution of the gravitational eld. If we want to
describe some other eld, such as matter, light, or , we simply take the special-
relativistic Lagrangian LM for that eld, change all the derivatives to covariant
derivatives, and form the sum (1/G)LG + LM. In the Brans-Dicke theory, we
have three pieces, (1/G)LG + LM + L
, where LM is for matter and L
for .
If we were to interpret as a rescaling of inertia, then we would have to have
appearing as a fudge factor modifying all the inner workings of LM. If, on the
other hand, we think of as changing the value of the gravitational constant G,
then the necessary modication is extremely simple. Brans and Dicke introduce
one further modication to L
so that the coupling constant between matter

and can be unitless. This modication has no eect on the wave equation of
in at spacetime.
a / The apparatus used by
Dicke and Goldenberg to mea-
sure the oblateness of the sun
was essentially a telescope with
a disk inserted in order to black
out most of the light from the sun.
presumably eventually forcing the gyroscope to precess in lock-step
with the laboratory. There would thus be no way to determine, once
the bullet was far away, that the laboratory was rotating at all
in perfect agreement with Machs principle.
7.3.3 Hints of empirical support
Only six years after the publication of the Brans-Dicke theory,
Dicke himself, along with H.M. Goldenberg
18
carried out a measure-
ment that seemed to support the theory empirically. Fifty years
before, one of the rst empirical tests of general relativity, which it
had seemed to pass with ying colors, was the anomalous perihelion
precession of Mercury. The word anomalous, which is often left
out in descriptions of this test, is required because there are many
nonrelativistic reasons why Mercurys orbit precesses, including in-
teractions with the other planets and the suns oblate shape. It is
only when these other eects are subtracted out that one sees the
general-relativistic eect calculated on page 178. The suns oblate-
ness is dicult to measure optically, so the original analysis of the
data had proceeded by determining the suns rotational period by
observing sunspots, and then assuming that the suns bulge was the
one found for a rotating uid in static equilibrium. The result was an
assumed oblateness of about 1 10
5
. But we know that the suns
dynamics are more complicated than this, since it has convection
currents and magnetic elds. Dicke, who was already a renowned
experimentalist, set out to determine the oblateness by direct op-
tical measurements, and the result was (5.0 0.7) 10
5
, which,
although still very small, was enough to put the observed perihelion
precession out of agreement with general relativity by about 8%.
The perihelion precession predicted by Brans-Dicke gravity diers
from the general relativistic result by a factor of (4 +3)/(6 +3).
The data therefore appeared to require 6 1, which would be
inconsistent with general relativity.
7.3.4 Machs principle is false.
The trouble with the solar oblateness measurements was that
they were subject to a large number of possible systematic errors,
and for this reason it was desirable to nd a more reliable test of
Brans-Dicke gravity. Not until about 1990 did a consensus arise,
based on measurements of oscillations of the solar surface, that the
pre-Dicke value was correct. In the interim, the confusion had the
salutary eect of stimulating a renaissance of theoretical and ex-
perimental work in general relativity. Often if one doesnt have an
alternative theory, one has no reasonable basis on which to design
and interpret experiments to test the original theory.
Currently, the best bound on is based on measurements
19
18
Dicke and Goldenberg, Solar Oblateness and General Relativity, Physical
Review Letters 18 (1967) 313
19
Bertotti, Iess, and Tortora, A test of general relativity using radio links
of the propagation of radio signals between earth and the Cassini-
Huygens space probe in 2003, which require > 4 10
4
. This is
so much greater than unity that it is reasonable to take Brans and
Dicke at their word that in any sensible theory, must be of the
general order of magnitude of unity. Brans-Dicke fails this test, and
is no longer a sensible candidate for a theory of gravity. We can
now see that Machs principle, far from being a fuzzy piece of philo-
sophical navel-gazing, is a testable hypothesis. It has been tested
and found to be false, in the following sense. Brans-Dicke gravity
is about as natural a formal implementation of Machs principle as
could be hoped for, and it gives us a number that parametrizes
how Machian the universe is. The empirical value of is so large
that it shows our universe to be essentially as non-Machian as gen-
eral relativity.
with the Cassini spacecraft, Nature 425 (2003) 374
Problems
1 In problem 5 on page 162, we analyzed the properties of the
metric
ds
2
= e
2gz
dt
2
dz
2
.
(a) In that problem we found that this metric had the same proper-
ties at all points in space. Verify in particular that it has the same
scalar curvature R at all points in space.
(b) Show that this is a vacuum solution in the two-dimensional (t, z)
space.
(c) Suppose we try to generalize this metric to four dimensions as
ds
2
= e
2gz
dt
2
dx
2
dy
2
dz
2
.
Show that this requires an Einstein tensor with unphysical proper-
ties.
Solution, p. 265
2 Subsection 7.1.3 on page ?? presented the Petrov metric.
Use symbolic math software such as Maxima to verify that it is
a vacuum solution, and that its scalar curvature is constant (as we
expect for a spacetime whose properties are supposed to be uniform).
Solution, p. 266
Problems 227
Chapter 8
Gravitational Waves
8.1 The speed of gravity
In Newtonian gravity, gravitational eects are assumed to propagate
at innite speed, so that for example the lunar tides correspond at
any time to the position of the moon at the same instant. This
clearly cant be true in relativity, since simultaneity isnt something
that dierent observers even agree on. Not only should the speed
of gravity be nite, but it seems implausible that that it would be
greater than c; in section 2.2 (p. 40), we argued based on empirically
well established principles that there must be a maximum speed of
cause and eect. Although the argument was only applicable to
special relativity, i.e., to a at, Lorentzian space, it seems likely to
apply to general relativity as well, at least for ripples in spacetime
that are relatively weak, so that space is approximately Lorentzian.
As early as 1913, before Einstein had even developed the full theory
of general relativity, he had carried out calculations in the weak-eld
limit that showed that gravitational eects should propagate at c.
This seems eminently reasonable, since (a) it is likely to be consistent
with causality, and (b) G and c are the only constants with units
that appear in the eld equations (obscured by our choice of units,
in which G = 1 and c = 1), and the only velocity-scale that can be
constructed from these two constants is c itself.
Although extremely well founded theoretically, this turns out to
be extremely dicult to test empirically. In a 2003 experiment,
1
Fo-
malont and Kopeikin used a world-wide array of radio telescopes to
observe a conjunction in which Jupiter passed within 3.7
of a quasar,
so that the quasars radio waves came within about 3 light-seconds
of the planet on their way to the earth. Since Jupiter moves with
v = 410
5
, one expects naively that the radio waves passing by it
should be deected by the eld produced by Jupiter at the position it
had 3 seconds earlier. This position diers from its present position
by about 10
4
light-seconds, and the result should be a dierence
in propagation time, which should be dierent when observed from
dierent locations on earth. Fomalont and Kopeikin measured these
phase dierences with picosecond precision, and found them to be
in good agreement with the predictions of general relativity. The
real excitement started when they published their result with the
interpretation that they had measured, for the rst time, the speed
1
229
of gravity, and found it to be within 20% error bars of c. Samuel
2
and Will
3
published refutations, arguing that Kopeikins calcula-
tions contained mistakes, and that what had really been measured
was the speed of light, not the speed of gravity.
The reason that the interpretation of this type of experiment
is likely to be controversial is that although we do have theories of
gravity that are viable alternatives to general relativity (e.g., the
Brans-Dicke theory, in which the gravitational constant is a dynam-
ically changing variable), such theories have generally been carefully
designed to agree with general relativity in the weak-eld limit, and
in particular every such theory (or at least every theory that remains
viable given current experimental data) predicts that gravitational
eects propagate at c in the weak-eld limit. Without an alternative
theory to act as a framework one that disagrees with relativity
about the speed of gravity it is dicult to know whether an ob-
servation that agrees with relativity is a test of this specic aspect
of relativity.
8.2 Gravitational radiation
8.2.1 Empirical evidence
So we still dont know, a century after Einstein found the eld
equations, whether gravitational ripples travel at c. Nevertheless,
we do have strong empirical evidence that such ripples exist. The
Hulse-Taylor system (page 182) contains two neutron stars orbiting
around their common center of mass, and the period of the orbit
is observed to be lengthening gradually over time, a. This is inter-
preted as evidence that the stars are losing energy to radiation of
gravitational waves.
4
More dramatic, if less clearcut, evidence is provided by Komossa,
Zhou, and Lus observationhttp://arxiv.org/abs/0804.4585 of a
supermassive black hole that appears to be recoiling from its parent
galaxy at a velocity of 2650 km/s (projected along the line of sight).
They interpret this as evidence for the following scenario. In the
early universe, galaxies form with supermassive black holes at their
centers. When two such galaxies collide, the black holes can merge.
The merger is a violent process in which intense gravitational waves
are emitted, and these waves carry a large amount of momentum,
causing the black holes to recoil at a velocity greater than the escape
velocity of the merged galaxy.
2
3
4
Stairs, Testing General Relativity with Pulsar Timing, http://
relativity.livingreviews.org/Articles/lrr-2003-5/
230 Chapter 8 Gravitational Waves
a / The Hulse-Taylor pulsars or-
bital motion is gradually losing
energy due to the emission of
gravitational waves. The linear
decrease of the period is inte-
grated on this plot, resulting in
a parabola. From Weisberg and
Taylor, http://arxiv.org/abs/
astro-ph/0211217.
8.2.2 Expected properties
To see what properties we should expect for such radiation, rst
consider the reasoning that led to the construction of the Ricci and
Einstein tensors. If a certain volume of space is lled with test
particles, then the Ricci and Einstein tensors measure the tendency
for this volume to accelerate; i.e., d
2
V/dt
2
is a measure of the
attraction of any mass lying inside the volume. A distant mass,
however, will exert only tidal forces, which distort a region without
changing its volume. This suggests that as a gravitational wave
passes through a certain region of space, it should distort the shape
of a given region, without changing its volume.
When the idea of gravitational waves was rst discussed, there
was some skepticism about whether they represented an eect that
was observable, even in principle. The most naive such doubt is of
the same avor as the one discussed in section 7.2.3 about the ob-
servability of the universes expansion: if everything distorts, then
Section 8.2 Gravitational radiation 231
b / As the gravitational wave
propagates in the z direction, the
metric oscillates in the x and y
directions, preserving volume.
dont our meter-sticks distort as well, making it impossible to mea-
sure the eect? The answer is the same as before in section 7.2.3;
systems that are gravitationally or electromagnetically bound do
not have their scales distorted by an amount equal to the change in
the elements of the metric.
A less naive reason to be skeptical about gravitational waves is
that just because a metric looks oscillatory, that doesnt mean its
oscillatory behavior is observable. Consider the following example.
ds
2
= dt
2
_
1 +
1
10
sin x
_
dx
2
dy
2
dz
2
The Christoel symbols depend on derivatives of the form
a
g
bc
, so
here the only nonvanishing Christoel symbol is
x
xx
. It is then
straightforward to check that the Riemann tensor R
a
bcd
=
c
a
db
a
cb
+
a
ce
e
db

a
de
e
cb
vanishes by symmetry. Therefore this
metric must really just be a at-spacetime metric that has been
subjected to a silly change of coordinates.
Self-check: Do the Christoel symbols vanish for this metric?
Is there a reason for paying more attention to one or the other, R
or ?
To keep the curvature from vanishing, it looks like we need a
metric in which the oscillation is not restricted to a single variable.
For example, the metric
ds
2
= dt
2
_
1 +
1
10
sin y
_
dx
2
dy
2
dz
2
does have nonvanishing curvature. In other words, it seems like
we should be looking for transverse waves rather than longitudinal
ones.
5
On the other hand, this metric cannot be a solution to the
vacuum eld equations, since it doesnt preserve volume. It also
stands still, whereas we expect that solutions to the eld equations
should propagate at the velocity of light, at least for small ampli-
tudes. These conclusions are self-consistent, because a waves po-
larization can only be constrained if it propagates at c (see p. 107).
Based on what weve found out, the following seems like a metric
that might have a ghting chance of representing a real gravitational
wave:
ds
2
= dt
2
(1 +Asin(z t)) dx
2
dy
2
1 +Asin(z t)
dz
2
It is transverse, it propagates at c(= 1), and the fact that g
xx
is the
reciprocal of g
yy
makes it volume-conserving. The following Maxima
program calculates its Einstein tensor:
5
A more careful treatment shows that longitudinal waves can always be inter-
preted as physically unobservable coordinate waves, in the limit of large distances
from the source. Longitudinal waves near the source are referred to as Type III
solutions in a classication scheme due to Petrov. Transverse waves, which are
what we could actually observe in practical experiments, are type N.
1 load(ctensor);
3 lg:matrix([1,0,0,0],
4 [0,-(1+A*sin(z-t)),0,0],
5 [0,0,-1/(1+A*sin(z-t)),0],
6 [0,0,0,-1]);
7 cmetric();
8 einstein(true);
For a representative component of the Einstein tensor, we nd
G
tt
=
A
2
cos
2
(z t)
2 + 4Asin(z t) + 2A
2
sin
2
(z t)
For small values of A, we have [G
tt
[ A
2
/2. The vacuum eld
equations require G
tt
= 0, so this isnt an exact solution. But all
the components of G, not just G
tt
, are of order A
2
, so this is an
approximate solution to the equations.
It is also straightforward to check that propagation at approx-
imately c was a necessary feature. For example, if we replace the
factors of sin(z t) in the metric with sin(z 2t), we get a G
xx
that
is of order unity, not of order A
2
.
To prove that gravitational waves are an observable eect, we
would like to be able to display a metric that (1) is an exact solution
of the vacuum eld equations; (2) is not merely a coordinate wave;
and (3) carries momentum and energy. As late as 1936, Einstein and
Rosen published a paper claiming that gravitational waves were a
mathematical artifact, and did not actually exist.
6
8.2.3 Some exact solutions
In this section we study several examples of exact solutions to
the eld equations. Each of these can readily be shown not to be
a mere coordinate wave, since in each case the Riemann tensor has
nonzero elements.
An exact solution Example: 1
Weve already seen, e.g., in the derivation of the Schwarzschild
metric in section 6.2.4, that once we have an approximate solu-
tion to the equations of general relativity, we may be able to nd a
series solution. Historically this approach was only used as a last
resort, because the lack of computers made the calculations too
complex to handle, and the tendency was to look for tricks that
would make a closed-form solution possible. But today the series
method has the advantage that any mere mortal can have some
reasonable hope of success with it and there is nothing more
boring (or demoralizing) than laboriously learning someone elses
6
Some of the history is related at http://en.wikipedia.org/wiki/Sticky_
bead_argument.
special trick that only works for a specic problem. In this exam-
ple, well see that such an approach comes tantalizingly close
to providing an exact, oscillatory plane wave solution to the eld
equations.
Our best solution so far was of the form
ds
2
= dt
2
(1 + f ) dx
2
dy
2
1 + f
dz
2
,
where f = Asin(z t ). This doesnt seem likely to be an exact
solution for large amplitudes, since the x and y coordinates are
treated asymmetrically. In the extreme case of [A[ 1, there
would be singularities in g
yy
, but not in g
xx
. Clearly the metric will
have to have some kind of nonlinear dependence on f , but we just
havent found quite the right nonlinear dependence. Suppose we
try something of this form:
ds
2
= dt
2
_
1 + f + cf
2
_
dx
2
_
1 f + df
2
_
dy
2
dz
2
This approximately conserves volume, since (1+f +. . .)(1f +. . .)
equals unity, up to terms of order f
2
. The following program tests
this form.
1 load(ctensor);
3 f : A*exp(%i*k*(z-t));
4 lg:matrix([1,0,0,0],
5 [0,-(1+f+c*f^2),0,0],
6 [0,0,-(1-f+d*f^2),0],
7 [0,0,0,-1]);
8 cmetric();
9 einstein(true);
In line 3, the motivation for using the complex exponential rather
than a sine wave in f is the usual one of obtaining simpler ex-
pressions; as well see, this ends up causing problems. In lines 5
and 6, the symbols c and d have not been dened, and have not
been declared as depending on other variables, so Maxima treats
them as unknown constants. The result is G
t t
(4d + 4c 3)A
2
for small A, so we can make the A
2
term disappear by an appro-
priate choice of d and c. For symmetry, we choose c = d = 3/8.
With these values of the constants, the result for G
t t
is of order
A
4
. This technique can be extended to higher and higher orders
of approximation, resulting in an exact series solution to the eld
equations.
Unfortunately, the whole story ends up being too good to be true.
The resulting metric has complex-valued elements. If general rel-
ativity were a linear eld theory, then we could apply the usual
c / The sticky bead argument
for the reality of gravitational
waves. As a gravitational wave
with the appropriate polarization
passes by, the bead vibrates back
and forth on the rod. Friction cre-
ates heat. This demonstrates that
gravitational waves carry energy,
and are thus real, observable
phenomena.
technique of forming linear combinations of expressions of the
form e
+i ...
and e
i ...
, so as to give a real result. Unfortunately the
eld equations of general relativity are nonlinear, so the resulting
linear combination is no longer a solution. The best we can do is
to make a non-oscillatory real exponential solution (problem 2).
An exact, oscillatory, non-monochromatic solution Example: 2
Assume a metric of the form
ds
2
= dt
2
p(z t )
2
dx
2
q(z t )
2
dy
2
dz
2
,
where p and q are arbitrary functions. Such a metric would clearly
represent some kind of transverse-polarized plane wave traveling
at velocity c(= 1) in the z direction. The following Maxima code
calculates its Einstein tensor.
1 load(ctensor);
3 depends(p,[z,t]);
4 depends(q,[z,t]);
5 lg:matrix([1,0,0,0],
6 [0,-p^2,0,0],
7 [0,0,-q^2,0],
8 [0,0,0,-1]);
9 cmetric();
10 einstein(true);
The result is proportional to q/q + p/p, so any functions p and
q that satisfy the differential equation q/q + p/p = 0 will result
in a solution to the eld equations. Setting p(u) = 1 + Acos u, for
example, we nd that q is oscillatory, but with a period longer than
2 (problem 3).
An exact, plane, monochromatic wave Example: 3
Any metric of the form
ds
2
= (1 h)dt
2
dx
2
dy
2
(1 + h)dz
2
+ 2hdzdt ,
where h = f (z t )xy, and f is any function, is an exact solution of
the eld equations (problem 4).
Because h is proportional to xy, this does not appear at rst
glance to be a uniform plane wave. One can verify, however, that
all the components of the Riemann tensor depend only on z t ,
not on x or y. Therefore there is no measurable property of this
metric that varies with x and y.
8.2.4 Energy content
To show that these waves carry momentum and energy, we can
use the nonmathematical sticky bead argument (gure c), which
was originated by Feynman in 1957 and later popularized by Bondi.
d / The power emitted by a
multipole source of order m is
proportional to
2(m+1)
, when
the size of the source is small
compared to the wavelength. The
main reason for the depen-
dence is that at low frequencies,
the wavelength is long, so the
number of wavelengths traveled
to a particular point in space is
nearly the same from any point
in the source; we therefore get
strong cancellation.
We would next like to nd an expression for the energy of the
wave in terms of its amplitude. Alas, this was not meant to be.
It seems like it ought to be straightforward. We have such ex-
pressions in other classical eld theories. In electromagnetism, we
have energy densities +(1/8k)[E[
2
and +(1/2
o
)[B[
2
associated
with the electric and magnetic elds. In Newtonian gravity, we can
assign an energy density (1/8G)[g[
2
to the gravitational eld g;
the minus sign indicates that when masses glom onto each other,
they produce a greater eld, and energy is released.
In general relativity, however, the equivalence principle tells us
that for any gravitational eld measured by one observer, we can
nd another observer, one who is free-falling, who says that the local
eld is zero. It follows that we cannot associate an energy with the
curvature of a particular region of spacetime in any exact way. The
best we can do is to nd expressions that give the energy density (1)
in the limit of weak elds, and (2) when averaged over a region of
space that is large compared to the wavelength. These expressions
are not unique. There are a number of ways to write them in terms
of the metric and its derivatives, and they all give the same result
in the appropriate limit. The reader who is interested in seeing the
subject developed in detail is referred to Carrolls Lecture Notes on
General Relativity, http://arxiv.org/abs/gr-qc/?9712019.
8.2.5 Rate of radiation
How can we nd the rate of gravitational radiation from a system
such as the Hulse-Taylor pulsar?
Lets proceed by analogy. The simplest source of sound waves
is something like the cone of a stereo speaker. Since typical sound
waves have wavelengths measured in meters, the entire speaker is
generally small compared to the wavelength. The speaker cone is
a surface of oscillating displacement x = x
o
sin t. Idealizing such
a source to a radially pulsating spherical surface, we have an oscil-
lating monopole that radiates sound waves uniformly in all direc-
tions. To nd the power radiated, we note that the velocity of the
source-surface is proportional to x
o
, so the kinetic energy of the air
immediately in contact with it is proportional to
2
x
2
o
. The power
radiated is therefore proportional to
2
x
2
o
.
In electromagnetism, conservation of charge forbids the existence
of an oscillating electric monopole. The simplest radiating source is
therefore an oscillating electric dipole q = q
o
sin t. If the dipoles
physical size is small compared to a wavelength of the radiation,
then the radiation is an inecient process; at any point in space,
there is only a small dierence in path length between the positive
and negative portions of the dipole, so there tends to be strong
cancellation of their contributions, which were emitted with opposite
phases. The result is that the waves electromagnetic potential four-
vector (section 4.2.5) is proportional to q
o
, the elds to q
o
2
, and
the radiated power to q
2
o
4
. The factor of
4
can be broken down
into (
2
)(
2
), where the rst factor of
2
occurs for reasons similar
to the ones that explain the
2
factor for the monopole radiation
of sound, while the second
2
arises because the smaller is, the
longer the wavelength, and the greater the ineciency in radiation
caused by the small size of the source compared to the wavelength.
Since our universe doesnt seem to have particles with negative
mass, we cant form a gravitational dipole by putting positive and
negative masses on opposite ends of a stick and furthermore,
such a stick will not spin freely about its center, because its center
of mass does not lie at its center! In a more realistic system, such as
the Hulse-Taylor pulsar, we have two unequal masses orbiting about
their common center of mass. By conservation of momentum, the
mass dipole moment of such a system is constant, so we cannot have
an oscillating mass dipole. The simplest source of gravitational ra-
diation is therefore an oscillating mass quadrupole, Q = Q
o
sin t.
As in the case of the oscillating electric dipole, the radiation is sup-
pressed if, as is usually the case, the source is small compared to
the wavelength. The suppression is even stronger in the case of a
quadrupole, and the result is that the radiated power is proportional
to Q
2
o
6
.
The general pattern we have observed is that for multipole radi-
ation of order m (0=monopole, 1=dipole, 2=quadrupole), the radi-
ated power depends on
2(m+1)
. Since gravitational radiation must
always have m = 2 or higher, we have the very steep
6
depen-
dence of power on frequency. This demonstrates that if we want
to see strong gravitational radiation, we need to look at systems
that are oscillating extremely rapidly. For a binary system with un-
equal masses of order m, with orbits having radii of order r, we have
Q
o
mr
2
. Newtons laws give m
1/2
r
3/2
, which is essentially
Keplers law of periods. The result is that the radiated power should
depend on (m/r)
5
. In units with G = 1 and c = 1, power is unit-
less, so the units of this expression check out. Reinserting the proper
constants to give an equation that allows practical calculation in SI
units, we have
P = k
G
4
c
5
_
m
r
_
5
,
where k is a unitless constant of order unity.
For the Hulse-Taylor pulsar,
7
we have m 3 10
30
kg (about
one and a half solar masses) and r 10
9
m. The binary pulsar is
made to order our purposes, since m/r is extremely large compared
to what one sees in almost any other astronomical system. The
resulting estimate for the power is about 10
24
watts.
The pulsars period is observed to be steadily lengthening at a
7
rate of = 2.418 10
12
seconds per second. To compare this with
our crude theoretical estimate, we take the Newtonian energy of the
system Gm
2
/r and multiply by , giving 10
25
W, which checks to
within an order of magnitude. A full general-relativistic calculation
reproduces the observed value of to within the 0.1% error bars of
the data.
Problems
1 (a) Starting on page 21, we have associated geodesics with
the world-lines of low-mass objects (test particles). Use the Hulse-
Taylor pulsar as an example to show that the assumption of low
mass was a necessary one. How is this similar to the issues encoun-
tered on pp. 34 involving charged particles?
(b) Show that if low-mass, uncharged particles did not follow geodesics
(in a spacetime with no ambient electromagnetic elds), it would vi-
olate Lorentz invariance. Make sure that your argument explicitly
invokes the low mass and the lack of charge, because otherwise your
argument is wrong. Solution, p. 266
2 Show that the metric ds
2
= dt
2
Adx
2
Bdy
2
dz
2
with
A = 1 f +
3
8
f
2
25
416
f
3
+
15211
10729472
f
5
B = 1 +f +
3
8
f
2
+
25
416
f
3
15211
10729472
f
5
f = Ae
k(tz)
is an approximate solution to the vacuum eld equations, provided
that k is real which prevents this from being a physically realistic,
oscillating wave. Find the next nonvanishing term in each series.
3 Verify the claims made in example 2. Characterize the (some-
what complex) behavior of the function q obtained when p(u) =
1 +Acos u.
4 Verify the claims made in example 3 using Maxima. Although
the result holds for any function f, you may nd it more convenient
to use some specic form of f, such as a sine wave, so that Maxima
will be able to simplify the result to zero at the end. Note that when
the metric is expressed in terms of the line element, there is a factor
of 2 in the 2hdzdt term, but when expressing it as a matrix, the 2 is
not present in the matrix elements, because there are two elements
in the matrix that each contribute an equal amount.
Problems 239
Appendix 1: Excerpts from three papers by
Einstein
The following English translations of excerpts from three papers by Einstein were originally
published in The Principle of Relativity, Methuen and Co., 1923. The translation was by
W. Perrett and G.B. Jeery, and notes were provided by A. Sommerfeld. John Walker (www.
fourmilab.ch) has provided machine-readable versions of the rst two and placed them in the
public domain. Some notation has been modernized, British spelling has been Americanized,
etc. Footnotes by Sommerfeld, Walker, and B. Crowell are marked with initials. B. Crowells
modications to the present version are also in the public domain.
The paper On the electrodynamics of moving bodies contains two parts, the rst dealing
with kinematics and the second with electrodynamics. Ive given only the rst part here, since
the second one is lengthy, and painful to read because of the cumbersome old-fashioned notation.
The second section can be obtained from John Walkers web site.
The paper Does the inertia of a body depend upon its energy content?, which begins on
page 253, is very short and readable. A shorter and less general version of its main argument is
given on p. 111.
The foundation of the general theory of relativity is a long review article in which Einstein
systematically laid out the general theory, which he had previously published in a series of
shorter papers. The rst three sections of the paper give the general physical reasoning behind
coordinate independence, referred to as general covariance. It begins on page 255.
The reader who is interested in seeing these papers in their entirety can obtain them inex-
pensively in a Dover reprint of the original Methuen anthology.
On the electrodynamics of moving bodies
A. Einstein, Annalen der Physik 17 (1905) 891.
It is known that Maxwells electrodynamicsas usually understood at the present time
when applied to moving bodies, leads to asymmetries which do not appear to be inherent in
the phenomena.
8
Take, for example, the reciprocal electrodynamic action of a magnet and
a conductor. The observable phenomenon here depends only on the relative motion of the
conductor and the magnet, whereas the customary view draws a sharp distinction between the
two cases in which either the one or the other of these bodies is in motion. For if the magnet
is in motion and the conductor at rest, there arises in the neighbourhood of the magnet an
electric eld with a certain denite energy, producing a current at the places where parts of
the conductor are situated. But if the magnet is stationary and the conductor in motion, no
electric eld arises in the neighbourhood of the magnet. In the conductor, however, we nd an
electromotive force, to which in itself there is no corresponding energy, but which gives rise
assuming equality of relative motion in the two cases discussedto electric currents of the same
path and intensity as those produced by the electric forces in the former case.
Examples of this sort, together with the unsuccessful attempts to discover any motion of the
earth relative to the light medium, suggest that the phenomena of electrodynamics as well as
8
Einstein begins by giving an example involving electromagnetic induction, considered in two dierent frames
of reference. With modern hindsight, we would describe this by saying that a Lorentz boost mixes the electric
and magnetic elds, as described in section 4.2.4, p. 112. BC
of mechanics possess no properties corresponding to the idea of absolute rest.
9
They suggest
rather that, as has already been shown to the rst order of small quantities,
10
the same laws
of electrodynamics and optics will be valid for all frames of reference for which the equations
of mechanics hold good.
11
We will raise this conjecture (the purport of which will hereafter be
called the Principle of Relativity) to the status of a postulate, and also introduce another
postulate, which is only apparently irreconcilable with the former, namely, that light is always
propagated in empty space with a denite velocity c which is independent of the state of motion
of the emitting body.
12
These two postulates suce for the attainment of a simple and consistent
theory of the electrodynamics of moving bodies based on Maxwells theory for stationary bodies.
The introduction of a luminiferous ether will prove to be superuous inasmuch as the view
here to be developed will not require an absolutely stationary space provided with special
properties, nor assign a velocity-vector to a point of the empty space in which electromagnetic
processes take place.
The theory to be developed is basedlike all electrodynamicson the kinematics of the
rigid body, since the assertions of any such theory have to do with the relationships between
rigid bodies (systems of coordinates), clocks, and electromagnetic processes.
13
Insucient con-
sideration of this circumstance lies at the root of the diculties which the electrodynamics of
moving bodies at present encounters.
I. KINEMATICAL PART
1. Denition of Simultaneity
Let us take a system of coordinates in which the equations of Newtonian mechanics hold
good.
14
In order to render our presentation more precise and to distinguish this system of
coordinates verbally from others which will be introduced hereafter, we call it the stationary
9
Einstein knew about the Michelson-Morley experiment by 1905 (J. van Dongen, arxiv.org/abs/0908.1545),
but it isnt cited specically here. The 1881 and 1887 Michelson-Morley papers are available online at en.
wikisource.org. BC
10
I.e., to rst order in v/c. Experimenters as early as Fresnel (1788-1827) had shown that there were no eects
of order v/c due to the earths motion through the aether, but they were able to interpret this without jettisoning
the aether, my contriving models in which solid substances dragged the aether along with them. The negative
result of the Michelson-Morley experiment showed a lack of an eect of order (v/c)
2
. BC
11
The preceding memoir by Lorentz was not at this time known to the author. AS
12
The second postulate is redundant if we take the laws of electrodynamics and optics to refer to Maxwells
equations. Maxwells equations require that light move at c in any frame of reference in which they are valid, and
the rst postulate has already claimed that they are valid in all inertial frames of reference. Einstein probably
states constancy of c as a separate postulate because his audience is accustomed to thinking of Maxwells equations
as a partial mathematical representation of certain aspects of an underlying aether theory. Throughout part I of
the paper, Einstein is able to derive all his results without assuming anything from Maxwells equations other than
the constancy of c. The use of the term postulate suggests the construction of a formal axiomatic system like
Euclidean geometry, but Einsteins real intention here is to lay out a set of philosophical criteria for evaluating
candidate theories; he freely brings in other, less central, assumptions later in the paper, as when he invokes
homogeneity of spacetime on page 245. BC
13
Essentially what Einstein means here is that you cant have Maxwells equations without establishing position
and time coordinates, and you cant have position and time coordinates without clocks and rulers. Therefore even
the description of a purely electromagnetic phenomenon such as a light wave depends on the existence of material
objects. He doesnt spell out exactly what he means by rigid, and we now know that relativity doesnt actually
allow the existence of perfectly rigid solids (see p. 90). Essentially he wants to be able to talk about rulers that
behave like solids rather than liquids, in the sense that if they are accelerated suciently gently from rest and
later brought gently back to rest, their properties will be unchanged. When he derives the length contraction
later, he wants it to be clear that this isnt a dynamical phenomenon caused by an eect such as the drag of the
aether.
14
i.e., to the rst approximation.AS
Problems 241
system.
If a material point is at rest relative to this system of coordinates, its position can be dened
relative thereto by the employment of rigid standards of measurement and the methods of
Euclidean geometry, and can be expressed in Cartesian coordinates.
If we wish to describe the motion of a material point, we give the values of its coordinates
as functions of the time. Now we must bear carefully in mind that a mathematical description
of this kind has no physical meaning unless we are quite clear as to what we understand by
time. We have to take into account that all our judgments in which time plays a part are
always judgments of simultaneous events. If, for instance, I say, That train arrives here at 7
oclock, I mean something like this: The pointing of the small hand of my watch to 7 and the
arrival of the train are simultaneous events.
15
It might appear possible to overcome all the diculties attending the denition of time
by substituting the position of the small hand of my watch for time. And in fact such a
denition is satisfactory when we are concerned with dening a time exclusively for the place
where the watch is located; but it is no longer satisfactory when we have to connect in time
series of events occurring at dierent places, orwhat comes to the same thingto evaluate
the times of events occurring at places remote from the watch.
We might, of course, content ourselves with time values determined by an observer stationed
together with the watch at the origin of the coordinates, and coordinating the corresponding
positions of the hands with light signals, given out by every event to be timed, and reaching him
through empty space. But this coordination has the disadvantage that it is not independent of
the standpoint of the observer with the watch or clock, as we know from experience. We arrive
at a much more practical determination along the following line of thought.
If at the point A of space there is a clock, an observer at A can determine the time values
of events in the immediate proximity of A by nding the positions of the hands which are
simultaneous with these events. If there is at the point B of space another clock in all respects
resembling the one at A, it is possible for an observer at B to determine the time values of
events in the immediate neighbourhood of B. But it is not possible without further assumption
to compare, in respect of time, an event at A with an event at B. We have so far dened only
an A time and a B time. We have not dened a common time for A and B, for the latter
cannot be dened at all unless we establish by denition that the time required by light to
travel from A to B equals the time it requires to travel from B to A. Let a ray of light start
at the A time t
A
from A towards B, let it at the B time t
B
be reected at B in the direction
of A, and arrive again at A at the A time t
A
.
In accordance with denition the two clocks synchronize if
t
B
t
A
= t
A
t
B
.
We assume that this denition of synchronism is free from contradictions, and possible for
any number of points; and that the following relations are universally valid:
1. If the clock at B synchronizes with the clock at A, the clock at A synchronizes with the
clock at B.
15
We shall not here discuss the inexactitude which lurks in the concept of simultaneity of two events at approx-
imately the same place, which can only be removed by an abstraction.AS
2. If the clock at A synchronizes with the clock at B and also with the clock at C, the clocks
at B and C also synchronize with each other.
16
Thus with the help of certain imaginary physical experiments we have settled what is to
be understood by synchronous stationary clocks located at dierent places, and have evidently
obtained a denition of simultaneous, or synchronous, and of time. The time of an event
is that which is given simultaneously with the event by a stationary clock located at the place
of the event, this clock being synchronous, and indeed synchronous for all time determinations,
with a specied stationary clock.
In agreement with experience we further assume the quantity
2AB
t
A
t
A
= c,
to be a universal constantthe velocity of light in empty space.
It is essential to have time dened by means of stationary clocks in the stationary system,
and the time now dened being appropriate to the stationary system we call it the time of the
stationary system.
2. On the Relativity of Lengths and Times
The following reections are based on the principle of relativity and on the principle of the
constancy of the velocity of light. These two principles we dene as follows:
1. The laws by which the states of physical systems undergo change are not aected, whether
these changes of state be referred to the one or the other of two systems of coordinates in uniform
translatory motion.
2. Any ray of light moves in the stationary system of coordinates with the determined
velocity c, whether the ray be emitted by a stationary or by a moving body. Hence
velocity =
light path
time interval
where time interval is to be taken in the sense of the denition in 1.
Let there be given a stationary rigid rod; and let its length be l as measured by a measuring-
rod which is also stationary. We now imagine the axis of the rod lying along the axis of x of
the stationary system of coordinates, and that a uniform motion of parallel translation with
velocity v along the axis of x in the direction of increasing x is then imparted to the rod. We
now inquire as to the length of the moving rod, and imagine its length to be ascertained by the
following two operations:
(a) The observer moves together with the given measuring-rod and the rod to be measured,
and measures the length of the rod directly by superposing the measuring-rod, in just the same
way as if all three were at rest.
(b) By means of stationary clocks set up in the stationary system and synchronizing in
accordance with 1, the observer ascertains at what points of the stationary system the two
16
This assumption fails in a rotating frame (see p. 92), but Einstein has restricted himself here to an approxi-
mately inertial frame of reference.BC
Problems 243
ends of the rod to be measured are located at a denite time. The distance between these two
points, measured by the measuring-rod already employed, which in this case is at rest, is also a
length which may be designated the length of the rod.
In accordance with the principle of relativity the length to be discovered by the operation
(a)we will call it the length of the rod in the moving systemmust be equal to the length
l of the stationary rod.
The length to be discovered by the operation (b) we will call the length of the (moving)
rod in the stationary system. This we shall determine on the basis of our two principles, and
we shall nd that it diers from l.
Current kinematics tacitly assumes that the lengths determined by these two operations are
precisely equal, or in other words, that a moving rigid body at the epoch t may in geometrical
respects be perfectly represented by the same body at rest in a denite position.
We imagine further that at the two ends A and B of the rod, clocks are placed which syn-
chronize with the clocks of the stationary system, that is to say that their indications correspond
at any instant to the time of the stationary system at the places where they happen to be.
These clocks are therefore synchronous in the stationary system.
We imagine further that with each clock there is a moving observer, and that these observers
apply to both clocks the criterion established in 1 for the synchronization of two clocks. Let
a ray of light depart from A at the time
17
t
A
, let it be reected at B at the time t
B
, and reach
A again at the time t
A
. Taking into consideration the principle of the constancy of the velocity
of light we nd that
t
B
t
A
=
r
AB
c v
and t
A
t
B
=
r
AB
c +v
where r
AB
denotes the length of the moving rodmeasured in the stationary system. Observers
moving with the moving rod would thus nd that the two clocks were not synchronous, while
observers in the stationary system would declare the clocks to be synchronous.
So we see that we cannot attach any absolute signication to the concept of simultaneity, but
that two events which, viewed from a system of coordinates, are simultaneous, can no longer be
looked upon as simultaneous events when envisaged from a system which is in motion relative
to that system.
3. Theory of the Transformation of coordinates and Times from a Stationary System to
another System in Uniform Motion of Translation Relative to the Former
Let us in stationary space take two systems of coordinates, i.e., two systems, each of three
rigid material lines, perpendicular to one another, and issuing from a point. Let the axes of X
of the two systems coincide, and their axes of Y and Z respectively be parallel. Let each system
be provided with a rigid measuring-rod and a number of clocks, and let the two measuring-rods,
and likewise all the clocks of the two systems, be in all respects alike.
Now to the origin of one of the two systems (k) let a constant velocity v be imparted in
the direction of the increasing x of the other stationary system (K), and let this velocity be
communicated to the axes of the coordinates, the relevant measuring-rod, and the clocks. To
17
Time here denotes time of the stationary system and also position of hands of the moving clock situated
at the place under discussion.AS
any time of the stationary system K there then will correspond a denite position of the axes
of the moving system, and from reasons of symmetry we are entitled to assume that the motion
of k may be such that the axes of the moving system are at the time t (this t always denotes
a time of the stationary system) parallel to the axes of the stationary system.
We now imagine space to be measured from the stationary system K by means of the
stationary measuring-rod, and also from the moving system k by means of the measuring-rod
moving with it; and that we thus obtain the coordinates x, y, z, and , , respectively. Further,
let the time t of the stationary system be determined for all points thereof at which there are
clocks by means of light signals in the manner indicated in 1; similarly let the time of the
moving system be determined for all points of the moving system at which there are clocks at
rest relative to that system by applying the method, given in 1, of light signals between the
points at which the latter clocks are located.
To any system of values x, y, z, t, which completely denes the place and time of an event
in the stationary system, there belongs a system of values , , , , determining that event
relative to the system k, and our task is now to nd the system of equations connecting these
quantities.
In the rst place it is clear that the equations must be linear on account of the properties of
homogeneity which we attribute to space and time.
If we place x
= x vt, it is clear that a point at rest in the system k must have a system of
values x
, y, z, independent of time. We rst dene as a function of x
, y, z, and t. To do this
we have to express in equations that is nothing else than the summary of the data of clocks
at rest in system k, which have been synchronized according to the rule given in 1.
From the origin of system k let a ray be emitted at the time
0
along the X-axis to x
, and at
the time
1
be reected thence to the origin of the coordinates, arriving there at the time
2
; we
then must have
1
2
(
0
+
2
) =
1
, or, by inserting the arguments of the function and applying
the principle of the constancy of the velocity of light in the stationary system:
1
2
_
(0, 0, 0, t) +
_
0, 0, 0, t +
x
c v
+
x
c +v
__
=
_
x
, 0, 0, t +
x
c v
_
.
Hence, if x
be chosen innitesimally small,

1
2
_
1
c v
+
1
c +v
_
t
=

x
+
1
c v
t
,
or
+
v
c
2
v
2
t
= 0.
It is to be noted that instead of the origin of the coordinates we might have chosen any other
point for the point of origin of the ray, and the equation just obtained is therefore valid for all
values of x
, y, z.
An analogous considerationapplied to the axes of Y and Zit being borne in mind that
light is always propagated along these axes, when viewed from the stationary system, with the
velocity

c
2
v
2
gives us
Problems 245
y
= 0,

z
= 0.
Since is a linear function, it follows from these equations that
= a
_
t
v
c
2
v
2
x
_
where a is a function (v) at present unknown, and where for brevity it is assumed that at the
origin of k, = 0, when t = 0.
With the help of this result we easily determine the quantities , , by expressing in
equations that light (as required by the principle of the constancy of the velocity of light, in
combination with the principle of relativity) is also propagated with velocity c when measured
in the moving system. For a ray of light emitted at the time = 0 in the direction of the
increasing
= c or = ac
_
t
v
c
2
v
2
x
_
.
But the ray moves relative to the initial point of k, when measured in the stationary system,
with the velocity c v, so that
x
c v
= t.
If we insert this value of t in the equation for , we obtain
= a
c
2
c
2
v
2
x
.
In an analogous manner we nd, by considering rays moving along the two other axes, that
= c = ac
_
t
v
c
2
v
2
x
_
when
y
c
2
v
2
= t, x
= 0.
Thus
= a
c
c
2
v
2
y and = a
c
c
2
v
2
z.
Substituting for x
its value, we obtain

= (v)(t vx/c
2
),
= (v)(x vt),
= (v)y,
= (v)z,
where
=
1
_
1 v
2
/c
2
,
and is an as yet unknown function of v. If no assumption whatever be made as to the initial
position of the moving system and as to the zero point of , an additive constant is to be placed
on the right side of each of these equations.
We now have to prove that any ray of light, measured in the moving system, is propagated
with the velocity c, if, as we have assumed, this is the case in the stationary system; for we
have not as yet furnished the proof that the principle of the constancy of the velocity of light is
compatible with the principle of relativity.
At the time t = = 0, when the origin of the coordinates is common to the two systems,
let a spherical wave be emitted therefrom, and be propagated with the velocity c in system K.
If (x, y, z) be a point just attained by this wave, then
x
2
+y
2
+z
2
= c
2
t
2
.
Transforming this equation with the aid of our equations of transformation we obtain after
a simple calculation
2
+
2
+
2
= c
2
2
.
The wave under consideration is therefore no less a spherical wave with velocity of propaga-
tion c when viewed in the moving system. This shows that our two fundamental principles are
compatible.
18
In the equations of transformation which have been developed there enters an unknown
function of v, which we will now determine.
For this purpose we introduce a third system of coordinates K
, which relative to the system

k is in a state of parallel translatory motion parallel to the axis of ,
19
such that the origin of
18
The equations of the Lorentz transformation may be more simply deduced directly from the condition that
in virtue of those equations the relation x
2
+ y
2
+ z
2
= c
2
t
2
shall have as its consequence the second relation
2
+
2
+
2
= c
2
2
.AS
19
In Einsteins original paper, the symbols (, H, Z) for the coordinates of the moving system k were introduced
without explicitly dening them. In the 1923 English translation, (X, Y, Z) were used, creating an ambiguity
between X coordinates in the xed system K and the parallel axis in moving system k. Here and in subsequent
references we use when referring to the axis of system k along which the system is translating with respect to
K. In addition, the reference to system K
later in this sentence was incorrectly given as k in the 1923 English

translation.JW
Problems 247
coordinates of system K
moves with velocity v on the axis of . At the time t = 0 let all

three origins coincide, and when t = x = y = z = 0 let the time t
of the system K
be zero. We
call the coordinates, measured in the system K
, x
, y
, z
, and by a twofold application of our

equations of transformation we obtain
t
= (v)(v)( +v/c
2
) = (v)(v)t,
x
= (v)(v)( +v) = (v)(v)x,

y
= (v) = (v)(v)y,
z
= (v) = (v)(v)z.
Since the relations between x
, y
, z
and x, y, z do not contain the time t, the systems K

and K
are at rest with respect to one another, and it is clear that the transformation from K
to K
must be the identical transformation. Thus

(v)(v) = 1.
We now inquire into the signication of (v). We give our attention to that part of the axis of
Y of system k which lies between = 0, = 0, = 0 and = 0, = l, = 0. This part of the
axis of Y is a rod moving perpendicularly to its axis with velocity v relative to system K. Its
ends possess in K the coordinates
x
1
= vt, y
1
=
l
(v)
, z
1
= 0
and
x
2
= vt, y
2
= 0, z
2
= 0.
The length of the rod measured in K is therefore l/(v); and this gives us the meaning of the
function (v). From reasons of symmetry it is now evident that the length of a given rod moving
perpendicularly to its axis, measured in the stationary system, must depend only on the velocity
and not on the direction and the sense of the motion. The length of the moving rod measured in
the stationary system does not change, therefore, if v and v are interchanged. Hence follows
that l/(v) = l/(v), or
(v) = (v).
It follows from this relation and the one previously found that (v) = 1, so that the transfor-
mation equations which have been found become
= (t vx/c
2
),
= (x vt),
= y,
= z,
where
= 1/
_
1 v
2
/c
2
.
4. Physical Meaning of the Equations Obtained in Respect to Moving Rigid Bodies and
Moving Clocks
We envisage a rigid sphere
20
of radius R, at rest relative to the moving system k, and with
its centre at the origin of coordinates of k. The equation of the surface of this sphere moving
relative to the system K with velocity v is
2
+
2
+
2
= R
2
.
The equation of this surface expressed in x, y, z at the time t = 0 is
x
2
(
_
1 v
2
/c
2
)
2
+y
2
+z
2
= R
2
.
A rigid body which, measured in a state of rest, has the form of a sphere, therefore has in a
state of motionviewed from the stationary systemthe form of an ellipsoid of revolution with
the axes
R
_
1 v
2
/c
2
, R, R.
Thus, whereas the Y and Z dimensions of the sphere (and therefore of every rigid body of no
matter what form) do not appear modied by the motion, the X dimension appears shortened
in the ratio 1 :
_
1 v
2
/c
2
, i.e., the greater the value of v, the greater the shortening. For v = c
all moving objectsviewed from the stationary systemshrivel up into plane gures.
21
For
velocities greater than that of light our deliberations become meaningless; we shall, however,
nd in what follows, that the velocity of light in our theory plays the part, physically, of an
innitely great velocity.
It is clear that the same results hold good of bodies at rest in the stationary system,
viewed from a system in uniform motion.
Further, we imagine one of the clocks which are qualied to mark the time t when at rest
relative to the stationary system, and the time when at rest relative to the moving system,
to be located at the origin of the coordinates of k, and so adjusted that it marks the time .
What is the rate of this clock, when viewed from the stationary system?
Between the quantities x, t, and , which refer to the position of the clock, we have, evidently,
x = vt and
=
1
_
1 v
2
/c
2
(t vx/c
2
).
Therefore,
20
That is, a body possessing spherical form when examined at rest.AS
21
In the 1923 English translation, this phrase was erroneously translated as plain gures. I have used the
correct plane gures in this edition.JW
Problems 249
= t
_
1 v
2
/c
2
= t (1
_
1 v
2
/c
2
)t
whence it follows that the time marked by the clock (viewed in the stationary system) is slow by
1
_
1 v
2
/c
2
seconds per second, orneglecting magnitudes of fourth and higher orderby
1
2
v
2
/c
2
.
From this there ensues the following peculiar consequence. If at the points A and B of K
there are stationary clocks which, viewed in the stationary system, are synchronous; and if the
clock at A is moved with the velocity v along the line AB to B, then on its arrival at B the
two clocks no longer synchronize, but the clock moved from A to B lags behind the other which
has remained at B by
1
2
tv
2
/c
2
(up to magnitudes of fourth and higher order), t being the time
occupied in the journey from A to B.
It is at once apparent that this result still holds good if the clock moves from A to B in any
polygonal line, and also when the points A and B coincide.
If we assume that the result proved for a polygonal line is also valid for a continuously curved
line, we arrive at this result: If one of two synchronous clocks at A is moved in a closed curve
with constant velocity until it returns to A, the journey lasting t seconds, then by the clock
which has remained at rest the travelled clock on its arrival at A will be
1
2
tv
2
/c
2
second slow.
Thence we conclude that a spring-clock at the equator must go more slowly, by a very small
amount, than a precisely similar clock situated at one of the poles under otherwise identical
conditions.
22
5. The Composition of Velocities
In the system k moving along the axis of X of the system K with velocity v, let a point move
in accordance with the equations
= w
, = w
, = 0,
where w
and w
denote constants.
Required: the motion of the point relative to the system K. If with the help of the equations
of transformation developed in 3 we introduce the quantities x, y, z, t into the equations of
motion of the point, we obtain
22
Einstein species a spring-clock (unruhuhr) because the eective gravitational eld is weaker at the equator
than at the poles, so a pendulum clock at the equator would run more slowly by about two parts per thousand
than one at the north pole, for nonrelativistic reasons. This would completely mask any relativistic eect, which
he expected to be on the order of v
2
/c
2
, or about 10
13
. In any case, Einstein later realized that he was mistaken
about this example. There is also a gravitational time dilation, which produces an eect in the opposite direction.
The two eects cancel out, as we can see by going into the frame that rotates with the earth. In this frame,
both clocks are at rest, so the kinematic time dilation vanishes. The earths surface is a surface of constant
gravitational potential , and the gtt component of the metric depends only on , so the gravitational time
dilation also vanishes. The two clocks in this example would actually agree.
x =
w
+v
1 +vw
/c
2
t,
y =
_
1 v
2
/c
2
1 +vw
/c
2
w
t,
z = 0.
Thus the law of the parallelogram of velocities is valid according to our theory only to a rst
approximation. We set
23
V
2
=
_
dx
dt
_
2
+
_
dy
dt
_
2
,
w
2
= w
2
+w
2
,
a = tan
1
w
/w
,
a is then to be looked upon as the angle between the velocities v and w. After a simple calculation
we obtain
V =
_
(v
2
+w
2
+ 2vwcos a) (vwsin a/c)
2
1 +vwcos a/c
2
.
It is worthy of remark that v and w enter into the expression for the resultant velocity in a
symmetrical manner. If w also has the direction of the axis of X, we get
V =
v +w
1 +vw/c
2
.
It follows from this equation that from a composition of two velocities which are less than c,
there always results a velocity less than c. For if we set v = c , w = c , and being
positive and less than c, then
V = c
2c
2c +/c
< c.
It follows, further, that the velocity of light c cannot be altered by composition with a
velocity less than that of light. For this case we obtain
V =
c +w
1 +w/c
= c.
We might also have obtained the formula for V, for the case when v and w have the same
direction, by compounding two transformations in accordance with 3. If in addition to the
systems K and k guring in 3 we introduce still another system of coordinates k
moving
parallel to k, its initial point moving on the axis of
24
with the velocity w, we obtain equations
23
This equation was incorrectly given in Einsteins original paper and the 1923 English translation as a =
tan
1
wy/wx.JW
24
X in the 1923 English translation.JW
Problems 251
between the quantities x, y, z, t and the corresponding quantities of k
, which dier from the

equations found in 3 only in that the place of v is taken by the quantity
v +w
1 +vw/c
2
;
from which we see that such parallel transformationsnecessarilyform a group.
We have now deduced the requisite laws of the theory of kinematics corresponding to our
two principles, and we proceed to show their application to electrodynamics.
25
25
The remainder of the paper is not given here, but can be obtained from John Walkers web site at www.
fourmilab.ch.
Does the inertia of a body depend upon its energy content?
A. Einstein, Annalen der Physik. 18 (1905) 639.
The results of the previous investigation lead to a very interesting conclusion, which is here
to be deduced.
I based that investigation on the Maxwell-Hertz equations for empty space, together with
the Maxwellian expression for the electromagnetic energy of space, and in addition the principle
that:
The laws by which the states of physical systems alter are independent of the alternative, to
which of two systems of coordinates, in uniform motion of parallel translation relative to each
other, these alterations of state are referred (principle of relativity).
With these principles
26
as my basis I deduced inter alia the following result ( 8):
Let a system of plane waves of light, referred to the system of coordinates (x, y, z), possess
the energy l; let the direction of the ray (the wave-normal) make an angle with the axis of x
of the system. If we introduce a new system of coordinates (, , ) moving in uniform parallel
translation with respect to the system (x, y, z), and having its origin of coordinates in motion
along the axis of x with the velocity v, then this quantity of lightmeasured in the system
(, , )possesses the energy
27
l
= l
1
v
c
cos
_
1 v
2
/c
2
where c denotes the velocity of light. We shall make use of this result in what follows.
Let there be a stationary body in the system (x, y, z), and let its energyreferred to the
system (x, y, z) be E
0
. Let the energy of the body relative to the system (, , ) moving as
above with the velocity v, be H
0
.
Let this body send out, in a direction making an angle with the axis of x, plane waves
of light, of energy
1
2
L measured relative to (x, y, z), and simultaneously an equal quantity of
light in the opposite direction. Meanwhile the body remains at rest with respect to the system
(x, y, z). The principle of energy must apply to this process, and in fact (by the principle of
relativity) with respect to both systems of coordinates. If we call the energy of the body after
the emission of light E
1
or H
1
respectively, measured relative to the system (x, y, z) or (, , )
respectively, then by employing the relation given above we obtain
E
0
= E
1
+
1
2
L +
1
2
L,
H
0
= H
1
+
1
2
L
1
v
c
cos
_
1 v
2
/c
2
+
1
2
L
1 +
v
c
cos
_
1 v
2
/c
2
= H
1
+
L
_
1 v
2
/c
2
.
By subtraction we obtain from these equations
26
The principle of the constancy of the velocity of light is of course contained in Maxwells equations.AS
27
See homework problem 10, p. 126.
Problems 253
H
0
E
0
(H
1
E
1
) = L
_
1
_
1 v
2
/c
2
1
_
.
The two dierences of the form HE occurring in this expression have simple physical signica-
tions. H and E are energy values of the same body referred to two systems of coordinates which
are in motion relative to each other, the body being at rest in one of the two systems (system
(x, y, z)). Thus it is clear that the dierence H E can dier from the kinetic energy K of the
body, with respect to the other system (, , ), only by an additive constant C, which depends
on the choice of the arbitrary additive constants of the energies H and E. Thus we may place
H
0
E
0
= K
0
+ C,
H
1
E
1
= K
1
+ C,
since C does not change during the emission of light. So we have
K
0
K
1
= L
_
1
_
1 v
2
/c
2
1
_
.
The kinetic energy of the body with respect to (, , ) diminishes as a result of the emission
of light, and the amount of diminution is independent of the properties of the body. Moreover,
the dierence K
0
K
1
, like the kinetic energy of the electron ( 10), depends on the velocity.
Neglecting magnitudes of fourth and higher orders we may place
K
0
K
1
=
1
2
L
c
2
v
2
.
From this equation it directly follows that:
If a body gives o the energy L in the form of radiation, its mass diminishes by L/c
2
. The
fact that the energy withdrawn from the body becomes energy of radiation evidently makes no
dierence, so that we are led to the more general conclusion that
The mass of a body is a measure of its energy-content; if the energy changes by L, the mass
changes in the same sense by L/9 10
20
, the energy being measured in ergs, and the mass in
grammes.
It is not impossible that with bodies whose energy-content is variable to a high degree (e.g.
with radium salts) the theory may be successfully put to the test.
If the theory corresponds to the facts, radiation conveys inertia between the emitting and
absorbing bodies.
The foundation of the general theory of relativity
A. Einstein, Annalen der Physik 49 (1916) 769.
[A one-page introduction relating to history and personalities is omitted.BC]
A. FUNDAMENTAL CONSIDERATIONS ON THE POSTULATE OF RELATIVITY
1. Observations on the Special Theory of Relativity
The special theory of relativity is based on the following postulate, which is also satised by
the mechanics of Galileo and Newton. If a system of coordinates K is chosen so that, in relation
to it, physical laws hold good in their simplest form, the same laws also hold good in relation to
any other system of coordinates K
moving in uniform translation relative to K. This postulate

we call the special principle of relativity. The word special is meant to intimate that the
principle is restricted to the case when K
has a motion of uniform translation

28
relative to K,
but that the equivalence of K
and K does not extend to the case of non-uniform motion of K
relative to K.
Thus the special theory of relativity does not depart from classical mechanics through the
postulate of relativity, but through the postulate of the constancy of the velocity of light in
vacuo, from which, in combination with the special principle of relativity, there follow, in the
well-known way, the relativity of simultaneity, the Lorentzian transformation and the related
laws for the behaviour of moving bodies and clocks.
The modication to which the special theory of relativity has subjected the theory of space
and time is indeed far-reaching, but one important point has remained unaected. For the laws of
geometry, even according to the special theory of relativity, are to be interpreted directly as laws
relating to the possible relative positions of solid bodies at rest; and, in a more general way, the
laws of kinematics are to be interpreted as laws which describe the relations of measuring bodies
and clocks. To two selected material points of a stationary rigid body there always corresponds
a distance of quite denite length, which is independent of the locality and orientation of the
body, and is also independent of the time. To two selected positions of the hands of a clock at
rest relative to the privileged system of reference there always corresponds an interval of time
of a denite length, which is independent of place and time. We shall soon see that the general
theory of relativity cannot adhere to this simple physical interpretation of space and time.
29
2. The Need for an Extension of the Postulate of Relativity
In classical mechanics, and no less in the special theory of relativity, there is an inherent
epistemological defect which was, perhaps for the rst time, clearly pointed out by Ernst Mach.
We will elucidate it by the following example:
30
Two uid bodies of the same size and nature
28
Here Einstein denes the distinction between special and general relativity according to whether accelerated
frames of reference are allowed. The modern tendency is to pose this distinction in terms of at versus curved
spacetime, so that accelerated frames of reference in at spacetime are considered to be part of special relativity.
None of this has anything to do with the ability to describe accelerated objects. For example, special relativity is
perfectly capable of describing the twin paradox.
29
Einstein is just starting to lay out his argument, and has not yet made clear in what sense these statements
about location-independence of clocks and rulers could be empirically tested. It becomes more clear later that
he means something like this. We could try to ll spacetime with a lattice of clocks and rulers, to synchronize
the clocks, and to construct the lattice so that it consisted of right angles and equal-length line segments. This
succeeds in special relativity, so that the geometry of spacetime is compatible with frames of reference that split
up spacetime into 3+1 dimensions, where the three dimensions are Euclidean. The same prescription fails in
general relativity.
30
This example was described on p. 96.
Problems 255
hover freely in space at so great a distance from each other and from all other masses that only
those gravitational forces need be taken into account which arise from the interaction of dierent
parts of the same body. Let the distance between the two bodies be invariable, and in neither
of the bodies let there be any relative movements of the parts with respect to one another.
But let either mass, as judged by an observer at rest relative to the other mass, rotate
with constant angular velocity about the line joining the masses. This is a veriable relative
motion of the two bodies. Now let us imagine that each of the bodies has been surveyed by
means of measuring instruments at rest relative to itself, and let the surface of S
1
prove to be a
sphere, and that of S
2
an ellipsoid of revolution. Thereupon we put the question What is the
reason for this dierence in the two bodies? No answer can be admitted as epistemologically
satisfactory,
31
unless the reason given is an observable fact of experience. The law of causality
has not the signicance of a statement as to the world of experience, except when observable
facts ultimately appear as causes and eects.
Newtonian mechanics does not give a satisfactory answer to this question. It pronounces as
follows: The laws of mechanics apply to the space R
1
, in respect to which the body S
1
is at
rest, but not to the space R
2
, in respect to which the body S
2
is at rest. But the privileged
space R
1
of Galileo, thus introduced, is a merely factitious
32
cause, and not a thing that can be
observed. It is therefore clear that Newtons mechanics does not really satisfy the requirement
of causality in the case under consideration but only apparently does so, since it makes the
factitious cause R
1
responsible for the observable dierence in the bodies S
1
and S
2
.
The only satisfactory answer must be that the physical system consisting of S
1
and S
2
reveals within itself no imaginable cause to which the diering behaviour of S
1
and S
2
can be
referred. The cause must therefore lie outside this system. We have to take it that the general
laws of motion, which in particular determine the shapes of S
1
and S
2
, must be such that the
mechanical behaviour of S
1
and S
2
is partly conditioned in quite essential respects, by distant
masses which we have not included in the system under consideration. These distant masses
and their motions relative to S
1
and S
2
must then be regarded as the seat of the causes (which
must be susceptible to observation) of the dierent behaviour of our two bodies S
1
and S
2
. They
take over the role of the factitious cause R
1
. Of all imaginable spaces R
1
, R
2
, etc., in any kind
of motion relative to one another there is none which we may look upon as privileged a priori
without reviving the above-mentioned epistemological objection. The laws of physics must be
of such a nature that they apply to systems reference in any kind of motion.
33
Along this road
we arrive at an extension at the postulate of relativity.
In addition to this weighty argument from the theory of knowledge, there is a well-known
physical fact which favours an extension of the theory of relativity. Let K be a Galilean system
of reference, i.e., a system relative to which (at least in the four-dimensional region under
consideration) a mass, suciently distant from other masses, is moving with uniform motion
in a straight line. Let K
be a second system of reference which is moving relative to K in

uniformly accelerated translation. Then, relative to K
, a mass suciently distant from other

31
Of course an answer may be satisfactory from the point of view of epistemology, and yet be unsound hysically,
if it is in conict with other experiences. AS
32
i.e., articial BC
33
At this time, Einstein had high hopes that his theory would be fully Machian. He was already aware of the
Schwarzschild solution (he refers to it near the end of the paper), which oended his Machian sensibilities because
it imputed properties to spacetime in a universe containing only a single point-mass. In the present example of
the bodies S1 and S2, general relativity actually turns out to give the non-Machian result which Einstein here
says would be unsatisfactory.BC
masses would have an accelerated motion such that its acceleration and direction of acceleration
are independent of the material composition and physical state of the mass.
Does this permit an observer at rest relative to K
to infer that he is on a really accelerated

system of reference? The answer is in the negative; for the above-mentioned relation of freely
movable masses to K
may be interpreted equally well in the following way. The system of

reference K
is unaccelerated, but the space-time territory in question is under the sway of a

gravitational eld, which generates the accelerated motion of the bodies relative to K
.
This view is made possible for us by the teaching of experience as to the existence of a eld
of force, namely, the gravitational eld, which possesses the remarkable property of imparting
the same acceleration to all bodies.
34
The mechanical behaviour of bodies relative to K
is
the same as presents itself to experience in the case of systems which we are wont to regard
as stationary or as privileged. Therefore, from the physical standpoint, the assumption
readily suggests itself that the systems K and K
may both with equal right be looked upon

as stationary that is to say, they have an equal title as systems of reference for the physical
description of phenomena.
It will be seen from these reections that in pursuing the general theory of relativity we shall
be led to a theory of gravitation, since we are able to produce a gravitational eld merely by
changing the system of coordinates. It will also be obvious that the principle of the constancy
of the velocity of light in vacuo must be modied, since we easily recognize that the path of
a ray of light with respect to K
must in general be curvilinear, if with respect to K light is

propagated in a straight line with a denite constant velocity.
3. The Space-Time Continuum. Requirement of General Covariance for the Equations
Expressing General Laws of Nature
In classical mechanics, as well as in the special theory of relativity, the coordinates of space
and time have a direct physical meaning. To say that a point-event has the X
1
coordinate x
1
means that the projection of the point-event on the axis of X
1
, determined by rigid rods and
in accordance with the rules of Euclidean geometry, is obtained by measuring o a given rod
(the unit of length) x
1
times from the origin of coordinates along the axis of X
1
. To say that
a point-event has the X
4
coordinate x
4
= t, means that a standard clock, made to measure
time in a denite unit period, and which is stationary relative to the system of coordinates and
practically coincident in space with the point-event,
35
will have measured o x
4
= t periods at
the occurrence of the event.
This view of space and time has always been in the minds of physicists, even if, as a rule, they
have been unconscious of it. This is clear from the part which these concepts play in physical
measurements; it must also have underlain the readers reections on the preceding paragraph
for him to connect any meaning with what he there read. But we shall now show that we must
put it aside and replace it by a more general view, in order to be able to carry through the
postulate of general relativity, if the special theory of relativity applies to the special case of the
absence of a gravitational eld.
In a space which is free of gravitational elds we introduce a Galilean system of reference
34
E otv os has proved experimentally that the gravitational eld has this property in great accuracy.AS
35
We assume the possibility of verifying simultaneity for events immediately proximate in space, or to
speak more precisely for immediate proximity or coincidence in space-time, without giving a denition of this
fundamental concept.AS
Problems 257
K (x, y, z, t), and also a system of coordinates K
(x
, y
, z
, t
) in uniform rotation
36
relative to
K. Let the origins of both systems, as well as their axes of Z, permanently coincide. We shall
show that for a space-time measurement in the system K
the above denition of the physical

meaning of lengths and times cannot be maintained. For reasons of symmetry it is clear that a
circle around the origin in the X, Y plane of K may at the same time be regarded as a circle
in the X
, Y
plane of K
. We suppose that the circumference and diameter of this circle have

been measured with a unit measure innitely small compared with the radius, and that we have
the quotient of the two results. If this experiment were performed with a measuring-rod
37
at
rest relative to the Galilean system K, the quotient would be . With a measuring-rod at rest
relative to K
, the quotient would be greater than . This is readily understood if we envisage

the whole process of measuring from the stationary system K, and take into consideration
that the measuring-rod applied to the periphery undergoes a Lorentzian contraction, while the
one applied along the radius does not.
38
Hence Euclidean geometry does not apply to K
.
The notion of coordinates dened above, which presupposes the validity of Euclidean geometry,
therefore breaks down in relation to the system K
. So, too, we are unable to introduce a

time corresponding to physical requirements in K
, indicated by clocks at rest relative to K
.
To convince ourselves of this impossibility, let us imagine two clocks of identical constitution
placed, one at the origin of coordinates, and the other at the circumference of the circle, and
both envisaged from the stationary system K. By a familiar result of the special theory of
relativity, the clock at the circumference judged from K goes more slowly than the other,
because the former is in motion and the latter at rest. An observer at the common origin
of coordinates, capable of observing the clock at the circumference by means of light, would
therefore see it lagging behind the clock beside him. As he will not make up his mind to let the
velocity of light along the path in question depend explicitly on the time, he will interpret his
observations as showing that the clock at the circumference really goes more slowly than the
clock at the origin. So he will be obliged to dene time in such a way that the rate of a clock
depends upon where the clock may be.
We therefore reach this result: In the general theory of relativity, space and time cannot
be dened in such a way that dierences of the spatial coordinates can be directly measured by
the unit measuring-rod, or dierences in the time coordinate by a standard clock.
The method hitherto employed for laying coordinates into the space-time continuum in a
denite manner thus breaks down, and there seems to be no other way which would allow us
to adapt systems of coordinates to the four-dimensional universe so that we might expect from
their application a particularly simple formulation of the laws of nature. So there is nothing for
it but to regard all imaginable systems of coordinates, on principle, as equally suitable for the
description of nature.
39
This comes to requiring that:
36
This example of a rotating frame of reference was discussed on p. 89.BC
37
Einstein implicitly assumes that the measuring rods are perfectly rigid, but it is not obvious that this is
possible. This issue is discussed on p. 94.
38
As described on p. 90, Ehrenfest originally imagined that the circumference of the disk would be reduced
by its rotation. His argument was incorrect, because it assumed the ability to start the disk rotating when it
had originally been at rest. The present paper marks the rst time that Einstein asserted the opposite, that the
circumference is increased.
39
This is a conceptual leap, not a direct inference from the argument about the rotating frame. Einstein started
thinking about this argument in 1912, and concluded from it that he should base a theory of gravity on non-
Euclidean geometry. Inuenced by Levi-Civita, he tried to carry out this project in a coordinate-independent way,
but he failed at rst, and for a while explored a theory that was not coordinate-independent. Only later did he
return to coordinate-independence. It should be clear, then, that the link between the rotating-frame argument
and coordinate-independence was not as clearcut as Einstein makes out here, since he himself lost faith in it for
The general laws of nature are to be expressed by equations which hold good for all the
systems of coordinates, that is, are covariant with respect to any substitutions whatever (generally
covariant).
40
It is clear that a physical theory which satises this postulate will also be suitable for the
general postulate of relativity.
41
For the sum of all substitutions in any case includes those
which correspond to all relative motions of three-dimensional systems of coordinates. That
this requirement of general covariance, which takes away from space and time the last remnant
of physical objectivity,
42
is a natural one, will be seen from the following reection. All our
space-time verications invariably amount to a determination of space-time coincidences.
43
If,
for example, events consisted merely in the motion of material points, then ultimately nothing
would be observable but the meetings of two or more of these points. Moreover, the results
of our measurings are nothing but verications of such meetings of the material points of our
measuring instruments with other material points, coincidences between the hands of a clock
and points on the clock-dial, and observed point-events happening at the same place at the same
time.
The introduction of a system of reference serves no other purpose than to facilitate the
description of the totality of such coincidences. We allot to the universe four space-time variables
x
1
, x
2
, x
3
, x
4
in such a way that for every point-event there is a corresponding system of values
of the variables x
1
. . . x
4
. To two coincident point-events there corresponds one system of values
of the variables x
1
. . . x
4
, i.e., coincidence is characterized by the identity of the coordinates.
If, in place of the variables x
1
. . . x
4
, we introduce functions of them, x
1
, x
2
, x
3
, x
4
, as a new
system of coordinates, so that the systems of values are made to correspond to one another
without ambiguity, the equality of all four coordinates in the new system will also serve as an
expression for the space-time coincidence of the two point-events. As all our physical experience
can be ultimately reduced to such coincidences, there is no immediate reason for preferring
certain systems of coordinates to others, that is to say, we arrive at the requirement of general
covariance.
a while.BC
40
In this book Ive used the more transparent terminology coordinate independence rather than general
covariance.BC
41
For more on this point, see p. 97.
42
This is an extreme interpretation of general covariance, and one that Einstein himself didnt hew closely to
later on. He presented an almost diametrically opposed interpretation in a philosophical paper, On the aether,
Schweizerische naturforschende Gesellschaft 105 (1924) 85.BC
43
i.e., what this book refers to as incidence measurements (p. 80)BC
Problems 259
Appendix 2: Hints and solutions
Hints
Hints for Chapter 1
Page 33, problem 5: Apply the equivalence principle.
Solutions to Selected Homework Problems
Solutions for Chapter 1
Page 33, problem 3:
Pick two points P1 and P2. By O2, there is another point P3 that is distinct from P1 and
P2. (Recall that the notation [ABC] was dened so that all three points must be distinct.)
Applying O2 again, there must be a further point P4 out beyond P3, and by O3 this cant be
the same as P1. Continuing in this way, we can produce as many points as there are integers.
Page 33, problem 4:
(a) If the violation of (1) is tiny, then of course Kip wont really have any practical way to
violate (2), but the idea here is just to illustrate the idea, so to make things easy, lets imagine an
unrealistically large violation of (1). Suppose that neutrons have about the same inertial mass as
protons, but zero gravitational mass, in extreme violation of (1). This implies that neutron-rich
elements like uranium would have a much lower gravitational acceleration on earth than ones
like oxygen that are roughly 50-50 mixtures of neutrons and protons. Lets also simplify by
making a second unrealistically extreme assumption: lets say that Kip has a keychain in his
pocket made of neutronium, a substance composed of pure neutrons. On earth, the keychain
hovers in mid-air. Now he can release his keychain in the prison cell. If hes on a planet, it
will hover. If hes in an accelerating spaceship, then the keychain will follow Newtons rst law
(its tendency to do so being measured by its nonzero inertial mass), while the deck of the ship
accelerates up to hit it.
(b) It violates O1. O1 says that objects prepared in identical inertial states (as dened by
two successive events in their motion) are predicted to have identical motion in the future. This
fails in the case where Kip releases the neutronium keychain side by side with a penny.
Page 33, problem 5: By the equivalence principle, we can adopt a frame tied to the tossed
clock, B, and in this frame there is no gravitational eld. We see a desk and clock A go by.
The desk applies a force to clock A, decelerating it and then reaccelerating it so that it comes
back. Weve already established that the eect of motion is to slow down time, so clock A reads
a smaller time interval.
Page 33, problem 6: (a) In case 1 there is no source of energy, so the particle cannot radiate.
In case 2-4, the particle radiates, because there are sources of energy (loss of gravitational energy
in 2 and 3, the rocket fuel in 4).
(b) In 1, Newton says the object is subject to zero net force, so its motion is inertial. In
2-4, he says the object is subject to a nonvanishing net force, so its motion is noninertial. This
matches up with the results of the energy analysis.
(c) The equivalence principle, as discussed on page 34, is vague, and is particularly dicult
to apply successful and unambiguously to situations involving electrically charged objects, due
to the diculty of dening locality. Applying the equivalence principle in the most naive way,
we predict that there can be no radiation in cases 2 and 3 (because the object is following a
geodesic, minding its own business). In case 4, everyone agrees that there will be radiation
observable back on earth (although its possible that it would not be observable to an observer
momentarily matching velocities with the rocket). The naive equivalence principle says that 1
and 4 must give the same result, so we should have radiation in 1 as well. These predictions are
wrong in two out of the four equations, which tells us that we had better either not apply the
equivalence principle to charged objects, or not apply it in such a naive way.
Page 34, problem 7:
(a) The dominant form of radiation from the orbiting charge will be the lowest-order non-
vanishing multipole, which in this case is a dipole. The power radiated from a dipole scales like
d
2
4
, where d is the dipole moment. For an orbit of radius r, this becomes q
2
r
2
4
. To nd
the reaction force on the charged particle, we can use the relation p = E/c for electromagnetic
waves (section 1.5.7), which tells us that the force is equal to the power, up to a proportionality
constant c. Therefore a
r
q
2
r
2
4
/m. The gravitational acceleration is a
g
=
2
r, so we have
a
r
/a
g
(q
2
/m)
2
r, or a
r
/a
g
(q
2
/m)a
g
, where the a
g
on the right can be taken as an orbital
parameter, and for a low-earth orbit is very nearly equal to the usual acceleration of gravity at
the earths surface.
(b) In SI units, a
r
/a
g
(k/c
4
)(q
2
/m)a
g
, where k is the Coulomb constant.
(c) The result is 10
34
. If one tried to do this experiment in reality, the eect would be
impossible to detect, because the proton would be aected much more strongly by ambient
electric and magnetic elds than by the eect weve calculated.
Remark: It is odd that the result depends on q
2
/m, rather than on the charge-to-mass ratio
q/m, as is usually the case for a test particles trajectory. This means that we get a dierent
answer if we take two identical objects, place them side by side, and consider them as one big
object! This is not as unphysical as it sounds. The two side-by-side objects radiate coherently,
so the eld they radiate is doubled, and the radiated power is quadrupled. Each objects rate
of orbital decay is doubled, with the extra eect coming from electromagnetic interactions with
the other objects elds.
Page 68, problem 1:
(a) Let t be the time taken in the lab frame for the light to go from one mirror to the other,
and t
the corresponding interval in the clocks frame. Then t
= L, and (vt)
2
+L
2
= t
2
, where the
use of the same L in both equations makes use of our prior knowledge that there is no transverse
length contraction. Eliminating L, we nd the expected expression for , which is independent
of L (b) If the result of a were independent of L, then the relativistic time dilation would depend
on the details of the construction of the clock measuring the time dilation. We would be forced
to abandon the geometrical interpretation of special relativity. (c) The eect is to replace vt with
vt+at
2
/2 as the quantity inside the parentheses in the expression (. . .)
2
+L
2
= t
2
. The resulting
correction terms are of higher order in t than the ones appearing in the original expression, and
can therefore be made as small in relative size as desired by shortening the time t. But this is
Problems 261
exactly what happens when we make the clock suciently small.
Page 68, problem 2:
(a) Reinterpret gure i on p. 66 as a picture of a Sagnac ring interferometer. Let light waves
1 and 2 move around the loop in opposite senses. Wave 1 takes time t
1i
to move inward along
the crack, and time t
1o
to come back out. Wave 2 takes times t
2i
and t
2o
. But t
1i
= t
2i
(since
the two world-lines are identical), and similarly t
1o
= t
2o
. Therefore creating the crack has no
eect on the interference between 1 and 2, and splitting the big loop into two smaller loops
merely splits the total phase shift between them. (b) For a circular loop of radius r, the time
of ight of each wave is proportional to r, and in this time, each point on the circumference
of the rotating interferometer travels a distance v(time) = (r)(time) r
2
. (c) The eect is
proportional to area, and the area is zero. (d) The light clock in c has its two ends synchronized
according to the Einstein prescription, and the success of this synchronization veries Einsteins
assumption of commutativity in this particular case. If we make a Sagnac interferometer in the
shape of a triangle, then the Sagnac eect measures the failure of Einsteins assumption that all
three corners can be synchronized with one another.
Page 69, problem 4:
Here is the program:
1 L1:matrix([cosh(h1),sinh(h1)],[sinh(h1),cosh(h1)]);
2 L2:matrix([cosh(h2),sinh(h2)],[sinh(h2),cosh(h2)]);
3 T:L1.L2;
4 taylor(taylor(T,h1,0,2),h2,0,2);
The diagonal components of the result are both 1 +
2
1
/2 +
2
2
/2 +
1
2
+ . . . Everything after
the 1 is nonclassical. The o-diagonal components are
1
+
2
+
1
2
2
/2 +
2
2
1
/2 +. . ., with the
third-order terms being nonclassical.
Page 100, problem 1:
(a) As discussed in example 4 on page 79, a cylinder has local, intrinsic properties identical
to those of at space. The cylindrical model therefore has the same properties L1-L5 as our
standard model of Lorentzian space, provided that L1-L5 are taken as purely local statements.
(b) The cylindrical model does violate L3. In this model, the doubly-intersecting world-lines
described by property G will not occur if the world-lines are oriented exactly parallel to the
cylinder. This picks out a preferred direction in space, violating L3 if L3 is interpreted globally.
Frames moving parallel to the axis have dierent properties from frames moving perpendicular
to the axis.
But just because this particular model violates the global interpretation of L3, that doesnt
mean that all models of G violate it. We could instead construct a model in which space wraps
around in every direction. In the 2+1-dimensional case, we can visualize the spatial part of such
a model as the surface of a doughnut embedded in three-space, with the caveat that we dont
want to think of the doughnut holes circumference as being shorter than the doughnuts outer
radius. Giving up the idea of a visualizable model embedded in a higher-dimensional space,
we can simply take a three-dimensional cube and identify its opposite faces. Does this model
violate L3? Its not quite as obvious, but actually it does. The spacelike great-circle geodesics of
this model come in dierent circumferences, with the shortest being those parallel to the cubes
axes.
We cant prove by constructing a nite number of models that every possible model of G
violates L3. The two models weve found, however, can make us suspect that this is true, and
can give us insight into how to prove it. For any pair of world-lines that provide an example of
G, we can x a coordinate system K in which the two particles started out at A by ying o
back-to-back. In this coordinate system, we can measure the sum of the distances traversed by
the two particles from A to B. (If homegeneity, L1, holds, then they make equal contributions
to this sum.) The fact that the world-lines were traversed by material particles means that we
can, at least in principle, visit every point on them and measure the total distance using rigid
rulers. We call this the circumference of the great circle AB, as measured in a particular frame.
The set of all such circumferences has some greatest lower bound. If this bound is zero, then
such geodesics can exist locally, and this would violate even the local interpretation of L1-L5.
If the bound is nonzero, then lets x a circle that has this minimum circumference. Mark the
spatial points this circle passes through, in the frame of reference dened above. This set of
points is a spacelike circle of minimum radius. Near a given point on the circle, the circle looks
like a perfectly straight axis, whose orientation is presumably random. Now let some observer
K
travel around this circle at a velocity v relative to K, measuring the circumference with a
Lorentz-contracted ruler. The circumference is greater than the minimal one measured by K.
Therefore for any axis with a randomly chosen orientation, we have a preferred rest frame in
which the corresponding great circle has minimum circumference. This violates L3. Thanks to
physicsforums user atyy for suggesting this argument.
More detailed discussions of these issues are given in Bansal et al., arxiv.org/abs/gr-qc/
0503070v1, and Barrow and Levin, arxiv.org/abs/gr-qc/0101014v1.
(a) No. The track is straight in the lab frame, but curved in the rotating frame. Since the
spatial metric in the rotating frame is symmetric with respect to clockwise and counterclockwise,
the metric can never result in geodesics with a specic handedness. (b) The d
2
term of the
metric blows up here. A geodesic connecting point A, at r = 1/, with point B, at r < 1/,
must have minimum length. This requires that the geodesic be directly radial at A, so that
d
= 0; for if not, then we could vary the curve slightly so as to reduce [d
[, and the resulting

increase in the dr
2
term would be negligible compared to the decrease in the d
2
term. (c) No.
As we found in part a, laser beams cant be used to form geodesics.
The process that led from the Euclidean metric of example 6 on page 85 to the non-Euclidean
one of equation [3] on page 92 was not just a series of coordinate transformations. At the nal
step, we got rid of the variable t, reducing the number of dimensions by one. Similarly, we could
take a Euclidean three-dimensional space and eliminate all the points except for the ones on the
surface of the unit sphere; the geometry of the embedded sphere is non-Euclidean, because weve
redened geodesics to be lines that are as straight as they can be (i.e., have minimum length)
while restricted to the sphere. In the example of the carousel, the nal step eectively redenes
geodesics so that they have minimal length as determined by a chain of radar measurements.
Problems 263
The coordinate T would have a discontinuity of 2r
2
/(1
2
r
2
). Reinserting factors of c
to make it work out in SI units, we have 2r
2
c
2
/(1
2
r
2
c
2
) 207 ns. The exact error in
position that would result is dependent on the geometry of the current position of the satellites,
but it would be on the order of cT, which is 100 m. This is considerably worse than civilian
GPSs 20-meter error bars.
To avoid loss of precision in numerical operations like subtracting v from 1, its better
to derive an ultrarelativistic approximation. The velocity corresponding to a given is v =
_
1
2
1 1/2
2
, so 1 v 1/2
2
= (m/E)
2
/2. Reinserting factors of c so as to make
the units come out right in the SI system, this becomes (mc
2
/E)
2
/2 = 9 10
9
.
The time on the clock is given by s =
_
ds, where the integral is over the clocks world-line.
The quantity ds is our prototypical Lorentz scalar, so its frame-independent. An integral is
just a sum, and the tensor transformation laws are linear, so the integral of a Lorentz scalar is
still a Lorentz scalar. Therefore s is frame-independent. There is no requirement that we use an
inertial frame. It would also work ne, for example, in a frame rotating with the earth. We dont
even need to have a frame of reference. All of the above applies equally well to any coordinate
system at all, even one that doesnt have any sensible interpretation as some observers frame
of reference.
Such a transformation would take an energy-momentum four-vector (E, p), with E > 0, to
a dierent four-vector (E
, p
), with E
< 0. That transformation would also have the eect of

transforming a timelike displacement vector from the future light cone to the past light cone. But
the Lorentz transformations were specically constructed so as to preserve causality (property
L5 on p. 40), so this cant happen.
A spatial plane is determined by the lights direction of propagation and the relative ve-
locity of the source and observer, so the 3+1 case reduces without loss of generality to 2+1
dimensions. The frequency four-vector must be lightlike, so its most general possible form
is (f, f cos , f sin ), where is interpreted as the angle between the direction of propaga-
tion and the relative velocity. Putting this through a Lorentz boost along the x axis, we nd
f
= f(1+v cos ), which agrees with Einsteins equation on page 253, except for the arbitrary
convention involved in dening the sign of v.
(a) Expanding in a Taylor series, they both have g
tt
= 1 + 2gz +. . .
(b) This property holds for [2] automatically because of the way it was constructed. In [1],
the nonvanishing Christoel symbols (ignoring permutations of the lower indices) are
t
zt
= g
and
z
tt
= ge
2gz
. We can apply the geodesic equation with the ane parameter taken to be the
proper time, and this gives z = ge
2gz
t
2
, where dots represent dierentiation with respect to
proper time. For a particle instantaneously at rest,

t = 1/
g
tt
= e
2gz
, so z = g.
(c) [2] was constructed by performing a change of coordinates on a at-space metric, so it
is at. The Riemann tensor of [1] has R
t
ztz
= g
2
, so [1] isnt at. Therefore the two cant be
the same under a change of coordinates.
(d) [2] is at, so its curvature is constant. [1] has the property that under the transformation
z z + c, where c is a constant, the only change is a rescaling of the time coordinate; by
coordinate invariance, such a rescaling is unobservable.
Page 191, problem 3: (a) There are singularities at r = 0, where g
= 0, and r = 1/,
where g
tt
= 0. These are considered singularities because the inverse of the metric blows up.
Theyre coordinate singularities, because they can be removed by a change of coordinates back
to the original non-rotating frame.
(b) This one has singularities in the same places. The one at r = 0 is a coordinate singularity,
because at small r the dependence is negligible, and the metric is simply that of ordinary
plane polar coordinates in at space. The one at r = 1/ is not a coordinate singularity. The
following Maxima code calculates its scalar curvature R = R
a
a
, which is esentially just the
Gaussian curvature, since this is a two-dimensional space.
1 load(ctensor);
2 dim:2;
3 ct_coords:[r,theta];
4 lg:matrix([-1,0],
5 [0,-r^2/(1-w^2*r^2)]);
6 cmetric();
7 ricci(true);
8 scurvature();
The result is R = 6
2
/(1 2
2
r
2
+
4
r
4
). This blows up at r = 1/, which shows that this is
not a coordinate singularity. The fact that R does not blow up at r = 0 is consistent with our
earlier conclusion that r = 0 is a coordinate singularity, but would not have been sucient to
prove that conclusion.
(c) The argument is incorrect. The Gaussian curvature is not just proportional to the angular
decit , it is proportional to the limit of /A, where A is the area of the triangle. The area of
the triangle can be small, so there is no upper bound on the ratio /A. Debunking the argument
restores consistency with the answer to part b.
(a) The Ricci tensor is R
tt
= g
2
e
2gz
, R
zz
= g
2
. The scalar curvature is 2g
2
, which is
constant, as expected.
(b) Both G
tt
and G
zz
vanish by a straightforward computation.
(c) The Einstein tensor is G
tt
= 0, G
xx
= G
yy
= g
2
, G
zz
= 0. It is unphysical because it has
Problems 265
a zero mass-energy density, but a nonvanishing pressure.
The following Maxima code does the calculation:
1 load(ctensor);
2 ct_coords:[t,r,phi,z];
3 c: cos(sqrt(3)*r);
4 s: sin(sqrt(3)*r);
5 tt : exp(r)*c;
6 phit : exp(r)*s;
7 lg:matrix([tt,0,phit,0],
8 [0,-1,0,0],
9 [phit,0,-tt,0],
10 [0,0,0,-exp(-2*r)]
11 );
12 cmetric();
13 scurvature(); /* scalar curvature */
14 leinstein(true); /* Einstein tensor with lower indices, display=true */
Note that our denition of phit omits the factor of 2 from the line element, because were using
it to set two dierent o-diagonal elements of the metric, and we dont want to double-count.
The output of leinstein is silent, despite our setting the display ag to true; this is because
G
ab
s elements are all zero, and the function only outputs any nonzero elements. We can verify
this by explicitly asking for output of the components lein[1,1], etc.
Maxima computes the scalar curvature R and outputs it, but isnt smart enough to simplify
it completely. It is smart enough, however, to show us that it does not depend on t, , or z. A
good way to get around this problem is to expand R in a Taylor series as a function of r. For
example, a call to taylor(scurvature(),r,0,12) shows that the rst 12 terms of the Taylor
series all vanish.
Page 239, problem 1: (a) The members of the Hulse-Taylor system are spiraling toward one
another as they lose energy to gravitational radiation. If one of them were replaced with a
low-mass test particle, there would be negligible radiation, and the motion would no longer be
a spiral. This is similar to the issues encountered on pp. 34 because the neutron stars in the
Hulse-Taylor system suer a back-reaction from their own gravitational radiation.
(b) If this occurred, then the particles world-line would be displaced in space relative to a
geodesic of the spacetime that would have existed without the presence of the particle. What
would determine the direction of that displacement? It cant be determined by properties of
this preexisting, ambient spacetime, because the Riemann tensor is that spacetimes only local,
intrinsic, observable property. At a xed point in spacetime, the Riemann tensor is even under
spatial reection, so theres no way it can distinguish a certain direction in space from the
opposite direction.
What else could determine this mysterious displacement? By assumption, its not deter-
mined by a preexisting, ambient electromagnetic eld. If the particle had charge, the direction
could be one imposed by the back-reaction from the electromagnetic radiation it had emitted
in the past. If the particle had a lot of mass, then we could have something similar with gravi-
tational radiation, or some other nonlinear interaction of the particles gravitational eld with
the ambient eld. But these nonlinear or back-reaction eects are proportional to q
2
and m
2
,
so they vanish when q = 0 and m 0.
The only remaining possibility is that the result violates the symmetry of space expressed by
L1 on p. 40; the Lorentzian geometry is the result of L1-L5, so violating L1 should be considered
a violation of Lorentz invariance.
Problems 267
Photo Credits
Cover Galactic center: NASA, ESA, SSC, CXC, and STScI 15 Atomic clock: USNO
ocial photograph, public domain. 17 Gravity Probe A: I believe this diagram to be pub-
lic domain, due to its age and the improbability of its copyright having been renewed. 20
Stephen Hawking: unknown NASA photographer, 1999, public-domain product of NASA. 22
Eotvos: Unknown source. Since Eotvos died in 1919, the painting itself would be public do-
main if done from life. Under U.S. law, this makes photographic reproductions of the painting
public domain. 24 Earth: NASA, Apollo 17. Public domain. 24 Orion: Wikipedia user
Mouser, GFDL. 24 M100: European Southern Observatory, CC-BY-SA. 24 Supercluster:
Wikipedia user Azcolvin429, CC-BY-SA. 32 Pound and Rebka photo: Harvard University. I
presume this photo to be in the public domain, since it is unlikely to have had its copyright re-
newed. 35 Lorentz: Jan Veth (1864-1925), public domain. 47 Muon storage ring at CERN:
(c) 1974 by CERN; used here under the U.S. fair use doctrine. 50 Galaxies: Hubble Space
Telescope. Hubble material is copyright-free and may be freely used as in the public domain
without fee, on the condition that NASA and ESA is credited as the source of the material. The
material was created for NASA by STScI under Contract NAS5-26555 and for ESA by the Hub-
ble European Space Agency Information Centre. 54 Gamma-Ray burst: NASA/Swift/Mary
Pat Hrybyk-Keith and John Jones. 75 Levi-Civita: Believed to be public domain. Source:
http://www-history.mcs.st-and.ac.uk/PictDisplay/Levi-Civita.html. 78 Einsteins
ring: I have lost the information about the source of this image. I would be grateful to anyone
who could put me in touch with the copyright owners. 99 SU Aurigaes eld lines: P. Petit,
GFDL 1.2. 95 Galaxies: Hubble Space Telescope. Hubble material is copyright-free and may
be freely used as in the public domain without fee, on the condition that NASA and ESA is cred-
ited as the source of the material. The material was created for NASA by STScI under Contract
NAS5-26555 and for ESA by the Hubble European Space Agency Information Centre. 120
Chandrasekhar: University of Chicago. I believe the use of this photo in this book falls under the
fair use exception to copyright in the U.S. 123 Relativistic jet: Biretta et al., NASA/ESA, pub-
lic domain. 129 Rocks: Siim Sepp, CC-BY-SA 3.0. 130 Jupiter and comet: Hubble Space
Telescope, NASA, public domain. 131 Earth: NASA, Apollo 17. Public domain. 131
Moon: Luc Viatour, CC-BY-SA 3.0. 132 Heliotrope: ca. 1878, public domain. 132 Tri-
angulation survey: Otto Lueger, 1904, public domain. 136 Triangle in a space with negative
curvature: Wikipedia user Kie, public domain. 141 Eclipse: Eddingtons original 1919 photo,
public domain. 154 Torsion pendulum: University of Washington Eot-Wash group, http:
//www.npl.washington.edu/eotwash/publications/pdf/lowfrontier2.pdf. 165 Coin:
Kurt Wirth, public-domain product of the Swiss government. 168 Bill Unruh: Wikipedia user
Childrenofthedragon, public domain. 187 Accretion disk: Public-domain product of NASA
and ESA. 219 Cosmic microwave background image: NASA/WMAP Science Team, public
domain. 225 Dickes apparatus: Dicke, 1967. Used under the US fair-use doctrine.. ??
Graph of pulsars period: Weisberg and Taylor, http://arxiv.org/abs/astro-ph/0211217.
Index
absolute geometry, 18
abstract index notation, 85
Adams, W.S., 16
ane geometry, 37
ane parameter, 38
Aharonov-Bohm eect, 160
angular defect, 133
antisymmetrization, 84
Aristotelian logic, 53
atomic clocks, 14, 59
background independence, 98
Bell, John, 51
spaceship paradox, 51, 163
Big Bang, 206
Big Crunch, 78
Birkhos theorem, 222
black body spectrum, 168
Bohr model, 67
boost, 41
Brans-Dicke theory, 28, 221, 230
cadabra, 157, 159
Cartan, 154
curved-spacetime theory of Newtonian grav-
ity, 35, 97, 131
Casimir eect, 203
Cerenkov radiation, 118
Chandrasekhar limit, 120
Chiaos paradox, 26, 29
Christoel symbol, 145
chronology protection conjecture, 20, 199
clock postulate, 68, 104
cloning of particles, 167
closed timelike curve, 199
comoving cosmological coordinates, 204
Compton scattering, 127
conformal cosmological coordinates, 204
connection, 73
coordinate independence, 81
coordinate singularity, 185
correspondence principle, 28, 30, 44, 170
cosmic censorship, 187
cosmic microwave background, 213
cosmic rays, 16
cosmological constant, 50, 202
cosmological coordinates
comoving, 204
conformal, 204
standard, 204
covariant derivative, 142, 143
in electromagnetism, 142, 143
in relativity, 143
ctensor, 171
Cygnus X-1, 187
dark energy, 25
dark matter, 221
de Sitter, 204
Willem, 67
deection of light, 140, 182
derivative
covariant, 142, 143
in electromagnetism, 142
in relativity, 143
Dirac sea, 107, 126
dual, 82
dust, 110, 197
Eotvos experiments, 22
Eddington, 140
Ehrenfests paradox, 90
Einstein eld equation, 202
Einstein tensor, 194
Einstein-Cartan theory, 154
electromagnetic potential four-vector, 113, 117
electromagnetic tensor, 117
electron capture, 120
elliptic geometry, 75
energy conditions, 199
violated by cosmological constant, 203
energy-momentum tensor, 132, 195
equiconsistency, 76
equivalence principle
accelerations and elds equivalent, 23
application to charged particles, 26, 29
no preferred eld, 118
spacetime locally Lorentzian, 25
Erlangen program, 89
ether, 55
event horizon, 165
extrinsic quantity, 79
Fermats principle, 113
ne structure constant, 67
frame dragging, 125, 197
frame of reference
inertial, 23
frame-dragging, 161
frequency four-vector, 111
Friedmann equations, 206
Friedmann-Robertson-Walker cosmology, 206
Godels theorem, 76
Godel, Kurt, 76
gauge transformation, 99, 142
Gaussian curvature, 134
Gell-Mann, Murray, 107
general covariance, 81
geodesic, 21
as world-line of a test particle, 21, 239
dierential equation for, 146
geodesic equation, 146
geodetic eect, 139, 175
geometrized units, 172
geometry
elliptic, 75
hyperbolic, 136
spherical, 76
Goudsmit, 67
GPS
frames of reference used in, 93
timing signals, 117
gravitational constant, 172
gravitational eld
uniform, 162, 195, 227
gravitational mass, 21
gravitational redshift, 29
gravitational waves
transverse nature, 232
Gravity Probe A, 16
Gravity Probe B, 59, 117
frame dragging, 125
geodetic eect calculated, 175
geodetic eect estimated, 139
group, 88
Hafele-Keating experiment, 14, 59
Hawking
Stephen, 20
hole argument, 95
Hubble constant, 218
Hubble, Edwin, 206
Hulse, R.A., 182
Hulse-Taylor pulsar, 182, 237
hyperbolic geometry, 136
inertial frame, 23
inertial mass, 21
information paradox, 167, 188
inner product, 88
intrinsic quantity, 79
isometry, 88
Kasner metric, 196
Lense-Thirring eect, 125, 197
Levi-Civita
Tullio, 97
Levi-Civita tensor, 152
light
deection by sun, 140, 182
light clock, 68
light cone, 49
lightlike, 49
logic
Aristotelian, 53
loop quantum gravity, 54
Lorentz boost, 41
lune, 78
Mossbauer eect, 31
Machs principle, 97, 193, 221
mass
gravitational, 21
inertial, 21
mass-energy, 106
Maxima, 61, 171
Mercury
orbit of, 174
metric, 83
Michelson-Morley experiment, 55
Minkowski, 35
model
mathematical, 76
muon, 16
270 Index
naked singularity, 188
neutrino, 108
neutron star, 120, 182
no-cloning theorem, 167
normal coordinates, 134
parallel transport, 73, 74
Pasch
Moritz, 19
Penrose, Roger, 85, 188
Penrose-Hawking singularity theorems, 200, 207
Penzias, Arno, 206
Petrov classication, 232
Petrov metric, 197, 227
photon
mass, 108
Pioneer anomaly, 211
Planck mass, 154
Planck scale, 154
Poincare group, 88
polarization
of gravitational waves, 232
of light, 107
Pound-Rebka experiment, 16, 29
principal group, 89
prior geometry, 98
projective geometry, 80
proper time, 103
pulsar, 121, 182
rank of a tensor, 84
rapidity, 51
redshift
gravitational, 16, 29
Ricci curvature, 131
dened, 138
Riemann tensor
dened, 137
Rindler coordinates, 162
ring laser, 58
Robinson
Abraham, 72, 77
rotating frame of reference, 89, 196
Sagnac eect, 92
dened, 58
in GPS, 46
proportional to area, 68
Schwarzschild
Karl, 169
singularity, 20, 187
naked, 188
singularity theorems, 200
Sirius B, 16
spacelike, 49
spaceship paradox, 51, 163
spherical geometry, 76
standard cosmological coordinates, 204
stress-energy tensor, 132
string theory, 154
surface of last scattering, 207
Susskind, Leonard, 189
symmetrization, 84
Tarski, Alfred, 76
Taylor, J.H., 182
tensor, 84, 115
antisymmetric, 84
rank, 84
symmetric, 84
transformation law, 115
Thomas
Llewellyn, 67
Thomas precession, 139, 175
timelike, 49
Tolman-Oppenheimer-Volko limit, 121
torsion, 149
tensor, 152
transverse polarization
of gravitational waves, 232
of light, 107
trapped surface, 200
triangle inequality, 88
Type III solution, 232
Type N solution, 232
Uhlenbeck, 67
uniform gravitational eld, 162, 195, 227
unitarity, 167, 188
units
geometrized, 172
velocity vector, 104
wavenumber, 111
white dwarf, 119
Wilson, Robert, 206
world-line, 21
Index 271
Euclidean geometry (page 18):
E1 Two points determine a line.
E2 Line segments can be extended.
E3 A unique circle can be constructed given any point as its center and any line segment as
its radius.
E4 All right angles are equal to one another.
E5 Parallel postulate: Given a line and a point not on the line, exactly one line can be drawn
through the point and parallel to the given line.
44
Ordered geometry (page 19):
O1 Two events determine a line.
O2 Line segments can be extended: given A and B, there is at least one event such that [ABC]
is true.
O3 Lines dont wrap around: if [ABC] is true, then [BCA] is false.
O4 Betweenness: For any three distinct events A, B, and C lying on the same line, we can
determine whether or not B is between A and C (and by statement 3, this ordering is
unique except for a possible over-all reversal to form [CBA]).
Ane geometry (page 37):
In addition to O1-O4, postulate the following axioms:
A1 Constructibility of parallelograms: Given any P, Q, and R, there exists S such that [PQRS],
and if P, Q, and R are distinct then S is unique.
A2 Symmetric treatment of the sides of a parallelogram: If [PQRS], then [QRSP], [QPSR],
and [PRQS].
A3 Lines parallel to the same line are parallel to one another: If [ABCD] and [ABEF], then
[CDEF].
Experimentally motivated statements about Lorentzian geometry (page 272):
L1 Spacetime is homogeneous and isotropic. No point has special properties that make it
distinguishable from other points, nor is one direction distinguishable from another.
L2 Inertial frames of reference exist. These are frames in which particles move at constant
velocity if not subject to any forces. We can construct such a frame by using a particular
particle, which is not subject to any forces, as a reference point.
44
This is a form known as Playfairs axiom, rather than the version of the postulate originally given by Euclid.
272 Index
L3 Equivalence of inertial frames: If a frame is in constant-velocity translational motion
relative to an inertial frame, then it is also an inertial frame. No experiment can distinguish
one inertial frame from another.
L4 Causality: Observers in dierent inertial frames agree on the time-ordering of events.
L5 No simultaneity: The experimental evidence in section 1.2 shows that observers in dierent
inertial frames do not agree on the simultaneity of events.
Statements of the equivalence principle:
Accelerations and gravitational elds are equivalent. There is no experiment that can
distinguish one from the other (page 23).
It is always possible to dene a local Lorentz frame in a particular neighborhood of space-
time (page 25).
There is no way to associate a preferred tensor eld with spacetime (page 118).
Index 273

General Relativity - Benjamin Crowell

Uploaded by

Copyright:

Available Formats

General Relativity - Benjamin Crowell

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

General Relativity - Benjamin Crowell

Uploaded by

Copyright:

Available Formats

What are the main topics covered in the book?

What are the main topics covered in the book?

What are some of the experimental tests discussed regarding relativity?

What are some of the experimental tests discussed regarding relativity?

2

C. The total mass of the

count rate y. Since the eect

in the new coordinate system (x

be , interpreted as the ratio between the

and the corresponding

has coordinates (v, ), and the other geo-

symmetrically on the other

with respect to reection across the

, we could just as easily have used the

, so represents a factor of both time dila-

in the Lorentz-boosted frame. An

, the component of the emitting atoms velocity along the line

, which states that parallels dont exist: given a line and

consisting of E1-E4 plus E5

, we can prove a variety of theorems,

lacks self-consistency. Oops! As your

has a limit on the lengths of line segments.

that elliptic geometry, un-

, that it is possible in elliptic geometry for two lines to enclose a

= f(x), where f is a smooth one-to-one function. For ex-

are duals of each other.

dx dx, since the vectors are the wrong length. We therefore

, and the ones above it are the covariant vectors,

. Its not particularly important to keep track of which is which,

, how do we nd its dual dx

, and vice versa? In

are no longer synonyms, so their product is no longer simply

, but soon began writing down facts like

= 0. The area of the sphere is

= 0, i.e., g = diag(+1, +1, . . . , +1). This is

when the disk is spinning. B can

. The outer edge of the disk, however, is everywhere tan-

< C. The resolution of the paradox is that it rests on

) do not correspond physically to the

dt cross-term. Suppose that A sends two cars driv-

dt term is opposite for the

. They then synchronize

, then A can say

has increased by 2, so it is no longer possible for the

, ) coordinates. The frame of

, ), as in equation [2]. These are known as the

) is also a solution. But further-

(x) will still be a solution.

doesnt change the observable fact that they collided.

, = 1, 2, 3, is the ordinary three-

, in the absence of a charged particle to interact with. To see

[, where is a spacelike index. In thermal equilibrium, each of

= f/(1 v) for a moving source.

= (1 v)f . The effect on the energies of the beams can

. Generalizing to more than one dimension, and to

= v, since all the mixed partial derivatives x

with ,= are zero, and all the derivatives for = equal 1.

Section 4.3 The tensor transformation laws 115

= 1, when the relative

cannot correspond to Bs proper time,