General Relativity, Black Holes, and Cosmology

General Relativity, Black Holes, and Cosmology
Andrew J. S. Hamilton
Contents
Preface page 1
Notation 6
PART ONE SPECIAL RELATIVITY 9
Concept Questions 11
Whats important in Special Relativity 13
1 Special Relativity 14
1.1 The postulates of special relativity 14
1.2 The paradox of the constancy of the speed of light 16
1.3 Paradoxes and simultaneity 17
1.4 Time dilation 18
1.5 Lorentz transformation 20
1.6 Paradoxes: Time dilation, Lorentz contraction, and the Twin paradox 22
1.7 The spacetime wheel 23
1.8 Scalar spacetime distance 26
1.9 4-vectors 27
1.10 Energy-momentum 4-vector 28
1.11 Photon energy-momentum 30
1.12 Abstract 4-vectors 31
1.13 What things look like at relativistic speeds 32
1.14 How to programme Lorentz transformations on a computer 33
PART TWO COORDINATE APPROACH TO GENERAL RELATIVITY 35
Whats important? 39
iv Contents
2 Fundamentals of General Relativity 40
2.1 The postulates of General Relativity 40
2.2 Existence of locally inertial frames 41
2.3 Metric 41
2.4 Basis g
of tangent vectors 42
2.5 4-vectors and tensors 43
2.6 Covariant derivatives 45
2.7 Coordinate 4-velocity 51
2.8 Geodesic equation 51
2.9 Coordinate 4-momentum 52
2.10 Ane parameter 52
2.11 Ane distance 53
2.12 Riemann curvature tensor 53
2.13 Symmetries of the Riemann tensor 54
2.14 Ricci tensor, Ricci scalar 55
2.15 Einstein tensor 55
2.16 Bianchi identities 55
2.17 Covariant conservation of the Einstein tensor 56
2.18 Einstein equations 56
2.19 Summary of the path from metric to the energy-momentum tensor 57
2.20 Energy-momentum tensor of an ideal uid 57
2.21 Newtonian limit 58
3

More on the coordinate approach 59
3.1 Weyl tensor 59
3.2 Evolution equations for the Weyl tensor 59
3.3 Geodesic deviation 61
3.4 Commutator of the covariant derivative revisited 62
4

Action principle 65
4.1 Principle of least action for point particles 66
4.2 Action for a test particle 67
4.3 Action for a charged test particle in an electromagnetic eld 68
4.4 Generalized momentum 69
4.5 Hamiltonian 69
4.6 Derivatives of the action 70
PART THREE IDEAL BLACK HOLES 71
Whats important? 75
Contents v
5 Observational Evidence for Black Holes 76
6 Ideal Black Holes 78
6.1 Denition of a black hole 78
6.2 Ideal black hole 78
6.3 No-hair theorem 79
7 Schwarzschild Black Hole 80
7.1 Schwarzschild metric 80
7.2 Birkhos theorem 81
7.3 Stationary, static 81
7.4 Spherically symmetric 82
7.5 Horizon 83
7.6 Proper time 84
7.7 Redshift 84
7.8 Proper distance 85
7.9 Schwarzschild singularity 85
7.10 Embedding diagram 85
7.11 Energy-momentum tensor 86
7.12 Weyl tensor 86
7.13 Gullstrand-Painleve coordinates 86
7.14 Eddington-Finkelstein coordinates 87
7.15 Kruskal-Szekeres coordinates 88
7.16 Penrose diagrams 89
7.17 Schwarzschild white hole, wormhole 90
7.18 Collapse to a black hole 91
7.19 Killing vectors 92
7.20 Time translation symmetry 92
7.21 Spherical symmetry 92
7.22 Killing equation 93
8 Reissner-Nordstr om Black Hole 95
8.1 Reissner-Nordstr om metric 95
8.3 Weyl tensor 96
8.4 Horizons 97
8.5 Gullstrand-Painleve metric 97
8.6 Complete Reissner-Nordstr om geometry 98
8.7 Antiverse: Reissner-Nordstr om geometry with negative mass 100
8.8 Ingoing, outgoing 100
8.9 Mass ination instability 101
8.10 Inevitability of mass ination 103
vi Contents
8.11 The black hole particle accelerator 104
8.12 The X point 104
8.13 Extremal Reissner-Nordstr om geometry 105
8.14 Reissner-Nordstr om geometry with charge exceeding mass 106
8.15 Reissner-Nordstr om geometry with imaginary charge 106
9 Kerr-Newman Black Hole 109
9.1 Boyer-Lindquist metric 109
9.2 Oblate spheroidal coordinates 110
9.3 Time and rotation symmetries 110
9.4 Ring singularity 111
9.5 Horizons 111
9.6 Angular velocity of the horizon 113
9.7 Ergospheres 113
9.8 Antiverse 114
9.9 Closed timelike curves 114
9.11 Weyl tensor 116
9.12 Electromagnetic eld 116
9.13 Doran coordinates 117
9.14 Extremal Kerr-Newman geometry 117
9.15 Trajectories of test particles in the Kerr-Newman geometry 118
9.16 Penrose process 122
9.17 Constant latitude trajectories in the Kerr-Newman geometry 123
9.18 Principal null congruence 123
9.19 Circular orbits in the Kerr-Newman geometry 124
PART FOUR HOMOGENEOUS, ISOTROPIC COSMOLOGY 133
Whats important? 137
10 Homogeneous, Isotropic Cosmology 138
10.1 Observational basis 138
10.2 Cosmological Principle 139
10.3 Friedmann-Robertson-Walker metric 140
10.4 Spatial part of the FRW metric: informal approach 140
10.5 Comoving coordinates 142
10.6 Spatial part of the FRW metric: more formal approach 143
10.7 FRW metric 144
10.8 Einstein equations for FRW metric 144
Contents vii
10.9 Newtonian derivation of Friedmann equations 145
10.10 Hubble parameter 146
10.11 Critical density 147
10.12 Omega 147
10.13 Redshifting 148
10.14 Types of mass-energy 148
10.15 Evolution of the cosmic scale factor 149
10.16 Conformal time 151
10.17 Looking back along the lightcone 151
10.18 Horizon 152
PART FIVE TETRAD APPROACH TO GENERAL RELATIVITY 157
11 The tetrad formalism 162
11.1 Tetrad 162
11.2 Vierbein 162
11.3 The metric encodes the vierbein 163
11.4 Tetrad transformations 164
11.5 Tetrad Tensor 165
11.6 Raising and lowering indices 165
11.7 Gauge transformations 165
11.8 Directed derivatives 166
11.9 Tetrad covariant derivative 166
11.10 Relation between tetrad and coordinate connections 168
11.11 Torsion tensor 168
11.12 No-torsion condition 168
11.13 Antisymmetry of the connection coecients 169
11.14 Connection coecients in terms of the vierbein 169
11.15 Riemann curvature tensor 170
11.16 Ricci, Einstein, Bianchi 171
11.17 Electromagnetism 171
12

More on the tetrad formalism 174
12.1 Spinor tetrad formalism 174
12.2 Newman-Penrose tetrad formalism 177
12.3 Electromagnetic eld tensor 179
12.5 Petrov classication of the Weyl tensor 186
viii Contents
12.6 Raychaudhuri equations and the Sachs optical scalars 187
12.7 Focussing theorem 189
13

The 3+1 (ADM) formalism 191
13.1 ADM tetrad 192
13.2 Traditional ADM approach 193
13.3 Spatial tetrad vectors and tensors 194
13.4 ADM connections, gravity, and extrinsic curvature 194
13.5 ADM Riemann, Ricci, and Einstein tensors 195
13.6 ADM action 196
13.7 ADM equations of motion 199
13.8 Constraints and energy-momentum conservation 200
14

The geometric algebra 201
14.1 Products of vectors 202
14.2 Geometric product 203
14.3 Reverse 204
14.4 The pseudoscalar and the Hodge dual 205
14.5 Reection 206
14.6 Rotation 207
14.7 A rotor is a spin-
1
2
object 209
14.8 A multivector rotation is an active rotation 210
14.9 2D rotations and complex numbers 210
14.10 Quaternions 212
14.11 3D rotations and quaternions 213
14.12 Pauli matrices 215
14.13 Pauli spinors 216
14.14 Pauli spinors as scaled 3D rotors, or quaternions 218
14.15 Spacetime algebra 219
14.16 Complex quaternions 221
14.17 Lorentz transformations and complex quaternions 223
14.18 Spatial Inversion (P) and Time Inversion (T) 224
14.19 Electromagnetic eld bivector 225
14.20 How to implement Lorentz transformations on a computer 225
14.21 Dirac matrices 229
14.22 Dirac spinors 231
14.23 Dirac spinors as complex quaternions 232
14.24 Non-null Dirac spinor particle and antiparticle 235
14.25 Null Dirac Spinor 236
14.26 Chiral decomposition of a Dirac spinor 237
14.27 Dirac equation 238
Contents ix
14.28 Antiparticles are negative mass particles moving backwards in time 239
14.29 Dirac equation with electromagnetism 240
14.30 CPT 240
14.31 Charge conjugation C 241
14.32 Parity reversal P 242
14.33 Time reversal T 243
14.34 Majorana spinor 243
14.35 Covariant derivatives revisited 243
14.36 General relativistic Dirac equation 244
14.37 3D Vectors as rank-2 spinors 244
PART SIX BLACK HOLE INTERIORS 247
15 Black hole waterfalls 251
15.1 Tetrads move through coordinates 251
15.2 Gullstrand-Painleve waterfall 252
15.3 Boyer-Lindquist tetrad 258
15.4 Doran waterfall 259
16 General spherically symmetric spacetime 265
16.1 Spherical spacetime 265
16.2 Spherical electromagnetic eld 276
16.3 General relativistic stellar structure 277
16.4 Self-similar spherically symmetric spacetime 278
17 The interiors of spherical black holes 290
17.1 The mechanism of mass ination 290
17.2 The far future? 293
17.3 Self-similar models of the interior structure of black holes 295
17.4 Instability at outer horizon? 310
PART SEVEN GENERAL RELATIVISTIC PERTURBATION THEORY 311
18 Perturbations and gauge transformations 316
18.1 Notation for perturbations 316
18.2 Vierbein perturbation 316
x Contents
18.4 Tetrad metric assumed constant 317
18.5 Perturbed coordinate metric 317
18.6 Tetrad gauge transformations 318
18.7 Coordinate gauge transformations 319
18.8 Coordinate gauge transformation of a coordinate scalar 319
18.9 Coordinate gauge transformation of a coordinate vector or tensor 320
18.10 Coordinate gauge transformation of a tetrad vector 320
18.11 Coordinate gauge transformation of the vierbein 321
18.12 Coordinate gauge transformation of the metric 321
18.13 Lie derivative 322
19 Scalar, vector, tensor decomposition 323
19.1 Decomposition of a vector in at 3D space 323
19.2 Fourier version of the decomposition of a vector in at 3D space 324
19.3 Decomposition of a tensor in at 3D space 325
20 Flat space background 326
20.1 Classication of vierbein perturbations 326
20.2 Metric, tetrad connections, and Einstein and Weyl tensors 328
20.3 Spinor components of the Einstein tensor 330
20.4 Too many Einstein equations? 331
20.5 Action at a distance? 332
20.6 Comparison to electromagnetism 333
20.7 Harmonic gauge 337
20.8 What is the gravitational eld? 338
20.9 Newtonian (Copernican) gauge 338
20.10 Synchronous gauge 339
20.11 Newtonian potential 341
20.12 Dragging of inertial frames 342
20.13 Quadrupole pressure 343
20.14 Gravitational waves 344
20.15 Energy-momentum carried by gravitational waves 347
PART EIGHT COSMOLOGICAL PERTURBATIONS 349
21 An overview of cosmological perturbations 352
22

Cosmological perturbations in a at Friedmann-Robertson-Walker background 357
22.1 Unperturbed line-element 357
22.2 Comoving Fourier modes 358
Contents xi
22.4 Metric, tetrad connections, and Einstein tensor 360
22.5 ADM gauge choices 362
22.6 Conformal Newtonian gauge 362
23 Cosmological perturbations: a simplest set of assumptions 364
23.1 Perturbed FRW line-element 364
23.2 Energy-momenta of ideal uids 364
23.3 Diusive damping 367
23.4 Equations for the simplest set of assumptions 368
23.5 Unperturbed background 370
23.6 Generic behaviour of non-baryonic cold dark matter 371
23.7 Generic behaviour of radiation 372
23.8 Regimes 373
23.9 Superhorizon scales 373
23.10 Radiation-dominated, adiabatic initial conditions 376
23.11 Radiation-dominated, isocurvature initial conditions 379
23.12 Subhorizon scales 380
23.13 Matter-dominated 381
23.14 Recombination 382
23.15 Post-recombination 383
23.16 Matter with dark energy 384
23.17 Matter with dark energy and curvature 385
24

Cosmological perturbations: a more careful treatment of photons and baryons 387
24.1 Lorentz-invariant spatial and momentum volume elements 388
24.2 Occupation numbers 388
24.3 Occupation numbers in thermodynamic equilibrium 389
24.4 Boltzmann equation 390
24.5 Non-baryonic cold dark matter 392
24.6 The left hand side of the Boltzmann equation for photons 394
24.7 Spherical harmonics of the photon distribution 395
24.8 Energy-momentum tensor for photons 396
24.9 Collisions 396
24.10 Electron-photon scattering 398
24.11 The photon collision term for electron-photon scattering 399
24.12 Boltzmann equation for photons 402
24.13 Diusive (Silk) damping 403
24.14 Baryons 403
24.15 Viscous baryon drag damping 405
24.16 Photon-baryon wave equation 405
xii Contents
24.17 Damping of photon-baryon sound waves 407
24.18 Ionization and recombination 409
24.19 Neutrinos 409
24.20 Summary of equations 409
24.21 Legendre polynomials 410
25 Fluctuations in the Cosmic Microwave Background 411
25.1 Primordial power spectrum 411
25.2 Normalization of the power spectrum 412
25.3 CMB power spectrum 412
25.4 Matter power spectrum 413
25.5 Radiative transfer of CMB photons 413
25.6 Integrals over spherical Bessel functions 415
25.7 Large-scale CMB uctuations 417
25.8 Monopole, dipole, and quadrupole contributions to C
418
25.9 Integrated Sachs-Wolfe (ISW) eect 418
Preface
Illusory preface
As of writing (May 2010), this book is incomplete. If you happen to discover this draft on the internet, you
are welcome to it (and especially welcome to send me helpful advice and criticism). This book has been
written during two semesters of teaching graduate general relativity at the University of Colorado, Boulder,
during Spring 2008 and 2010. I hope to complete the book the next time I teach the course, which could
possibly be Spring 2012.
Meanwhile, I am vividly aware of the books shortcomings. The book is incomplete in many parts, and
needs pruning in others. If the early chapters read more like notes than a book, that is true; I was some
way into writing before I realised that a book was taking shape. Especially, the book is missing many
planned gures. Many of the anticipated gures can be found at three websites: Special Relativity
(http://casa.colorado.edu/
~
ajsh/sr/sr.shtml), Falling into a Black Hole (http://casa.colorado.
edu/
~
ajsh/schw.shtml), and Inside Black Holes (http://jila.colorado.edu/
~
ajsh/insidebh/index.
html).
Although the book is incomplete, I have tried hard to keep mathematical errors from creeping in. If you
nd an error especially in a minus sign or a factor please let me know.
True preface
This book is driven by one overriding question: What do students want?
A fundamental premise of this book is that, in the eld of general relativity, the number one thing, by
far, that students want to learn about is black holes. And the second thing they want to learn about is the
cosmic microwave background.
This book is born in part out of frustration with the relentlessly left-brained character of so many texts
on general relativity. Im probably one of those left-brained characters myself. However, my experience in
general relativistic visualization has convinced me that those interested in black holes from a mathematical
perspective are vastly outnumbered by those fascinated by black holes for other reasons. Among those
2 Preface
reasons are that black holes are, like dinosaurs, awesomely powerful, and supremely mysterious. I worry that
general relativity is (as I hear from students) often taught as if it were little more than tensor calculus. I
fear that the abstract approach repels and culls our right-brained students until only the most left-leaning
of our students remain to trasmit left-brained relativity to the next generation.
Although I was fascinated by general relativity already as a graduate student, my active involvement with
relativity was stimulated by students who insisted that I teach it. From the beginning it seemed obvious
that the way to teach relativity was through visualization. Thus, through teaching, I began to do general
relativistic visualizations of black holes. Initially the visualizations were simple animations, which I put
together into a website Falling into a Black Hole in 1997 and 1998. The visualizations touched a chord
with the outside world. In 2001/2 I had the privelege of spending a years sabbatical with the Denver
Museum of Nature and Science, where I began developing the Black Hole Flight Simulator (BHFS). That
sabbatical eventually led to a large-format immersive digital dome show Black Holes: The Other Side of
Innity, produced at the DMNS and directed by Tom Lucas. Premiering in 2006, that dome show has
been distributed to some 40 digital domes worldwide. Since that time visualizations with the BHFS have
appeared in several TV documentaries and in a number of exhibits. The experience of working with non-
science professionals has been, and continues to be, intensely enjoyable, and has sensitized me to the insidious
cultural chasms that divide us in an increasingly specialized society.
These experiences have left a prominent dent in my thinking. For example, I think that the highlight of
special relativity is the question of what you see and experience when you pass through a scene at near the
speed of light. Yet most texts scarcely mention the subject, if at all. Similarly, I think that the highlight
of black holes is what (general relativity predicts) you see and experience when you fall into a black hole.
Again, most texts hardly address the issue. Texts often do mention Penrose-Hawking singularity theorems.
Yet few texts mention the all-important mass ination instability discovered by E. Poisson & W. Israel
(1990). The inationary instability probably plays the central role in determining the interior structure of
astronomically realistic black holes, and in particular in cutting o the wormhole and white hole connections
to other universes that exist in the ideal Kerr geometry of a rotating black hole. Even E. Poisson (1994) A
Relativists Toolkit: The Mathematics of Black-Hole Mechanics mentions the inationary instability only
in the problems at the end of the last chapter. An important goal of this book is to redress this hole in the
teaching of general relativity.
The second focus of this book, after black holes, is the Cosmic Microwave Background (CMB). The CMB
oers a profound window on the genesis of our Universe. Observations of uctuations in the CMB are in
astonishing agreement with the predictions of general relativistic perturbation theory coupled with some
well-understood physics and some less well-understood but neverthess successful ideas about ination. For
this book, the goal I set myself was to attempt the simplest possible treatment of CMB uctuations that
would yield a result that could be compared to observation. This is not an easy goal, since calculation of
CMB uctuations presents many technical challenges. I applaud recent texts such as M. P. Hobson, G.
P. Efstathiou, & A. N. Lasenby (2006) General Relativity: An Introduction for Physicists, and T. Pad-
manabhan (2010) Gravitation: Foundations and Frontiers, which include chapters on the CMB power
spectrum. General texts like Hobson et al., Padmanabhan, and the present book by no means replace spe-
cialized books on the Cosmology and the CMB, such as S. Dodelson (2002) Modern Cosmology, R. Durrer
Preface 3
(2008) The Cosmic Microwave Background, or D. H. Lyth & A. R. Liddle (2009) The Primordial Density
Perturbation. However, for many students a course on general relativity may be the only opportunity they
get to learn about the CMB, and I think that a modern course on general relativity should include a basic
introduction to the CMB.
Notwithstanding its intended focus on applications rather than mathematics, this is not an easy book.
It is a serious graduate-level text. R. M. Wald (2006 Teaching the mathematics of general relativity,
Am. J. Phys. 74, 471477 http://arxiv.org/abs/gr-qc/0511073) describes the challenges of teaching the
necessary mathematics in a course on general relativity. This book will not make climbing the mathematical
mountain of general relativity any easier. But the intention is that this book will help you get a clear view
from the top, and not abandon you in fog. While the book does not shirk mathematics, I have tried hard to
make the logic and derivations as clear and tight as possible.
So much for the overall goals of this book. What about the strategy to achieve those goals? Firstly, this
book is intended as a book from which a student can learn, and a lecturer can teach. It is not intended as
a reference book. In a learning/teaching book, one must choose carefully not only what to include, but also
what not to include, because the latter distracts and dilutes.
The grand strategy is to go through general relativity in two passes. In the rst pass, the aim is to
run through the foundations of general relativity, and to get to ideal black holes as quickly as possible.
In the second pass, the book essentially starts all over again, using a tetrad-based approach rather than a
coordinate-based approach. The tetrad approach provides the basis for the subsequent treatment of non-ideal
black holes, and of the cosmic microwave background.
The emphasis of (the second half of) this book on tetrads is unusual for a textbook, but consistent with
its right-brained emphasis. If you want to see whats happening in a spacetime, then you need to look
at it with respect to the frame of an observer, which means working in a locally inertial (orthonormal)
frame. The problem with coordinate frames is that they prescribe that the axes of the spacetime are the
tangent vectors to the coordinates. These tangent vectors are skewed, not orthonormal. Looking at things
in a coordinate frame is like looking at a scene with eyes crossed. Even with geometries as simple as the
Friedmann-Robertson-Walker geometry of homogeneous, isotropic cosmology, it is necessary to play games
to see plainly what its energy-momentum is (for FRW, raise one of the indices on the energy-momentum
tensor but that trick fails in more complicated spacetimes). Tetrads obviate the need to waste time
attempting to conceptualize the distinction between vectors and covectors (one-forms). My own suspicion
is that the locally at structure of general relativity may be more fundamental than its globally geometric
character, which could be an emergent phenomenon.
A virtue of the two-pass approach is that the student gets to revisit the fundamentals of general relativity
from two similar but not identical perspectives. This rearmation of fundamentals is especially important
given the fast pace and stripped-down coverage.
The course that I teach to senior physics undergraduates and beginning graduate students at the University
of Colorado covers the following 8 topics, each topic taking about 2 weeks during the 16-week semester:
1. Pass 1.
a. Chapter 1: Special relativity.
4 Preface
b. Chapter 2: Coordinate approach to general relativity.
c. Chapters 59: Ideal black holes, namely Schwarzschild, Reissner-Nordstr om, and Kerr-Newman.
d. Chapter 10: Homogeneous, isotropic cosmology.
2. Pass 2.
a. Chapter 11: Tetrad approach to general relativity.
b. Chapters 1517: Black hole interiors.
c. Chapters 1820: General relativistic perturbation theory.
d. Chapters 21, 23, and 25: Cosmological perturbations.
The rst of the eight topics is special relativity. Special relativity is an essential precursor to general
relativity, since a fundamental postulate of general relativity is the Principle of Equivalence, which asserts
that at any point there exist frames, called locally inertial, or free-fall, with respect to which special relativity
operates locally. The strategy of Chapter 1 on Special Relativity is rst to confront the paradox of the
constancy of the speed of light, and from there to proceed rapidly to the highlight of special relativity, the
question of what you see and experience when you pass through a scene at near the speed of light. I choose
not to pause to discuss electromagnetism, actions, or other important topics in special relativity, since that
would get in the way of the driving goal, to head to a black hole at the fastest possible pace
1
.
The second of the eight topics is what I call the coordinate approach to general relativity. This is a lightning
introduction to the fundamental ingredients of general relativity, from the metric through to the Einstein
tensor, using the traditional coordinate-based approach, where components of tensors are expressed relative
to a basis of coordinate tangent vectors. To make the material more accessible, and to lay the groundwork
for tetrads, the book builds on concepts of vectors familiar from high school, and avoids unncecessary
mathematical distractions, such as emphasizing the distinction between vectors and 1-forms. Typically,
texts go through this material at a more leisurely pace, taking time to convey challenging conceptual issues.
Here however I choose not to linger, for two reasons. The rst is the obvious one: the goal is to get to black
holes post-haste. The second reason is that, as mentioned earlier in this preface, looking at tensors in a
coordinate basis is like looking at the world with eyes crossed. As my mother used to say, If you do that
and the wind changes, youll be stuck like that forever.
The third topic is ideal black holes, and here the pace slows. An ideal black hole is one that is stationary
(time translation invariant), and empty outside its singularity, except for the contribution of a static electric
eld. In the 4 dimensions of the spacetime we live in, ideal black holes come in just a few varieties: the
Schwarzschild geometry for a spherical, uncharged black hole, the Reissner-Nordstr omfor a spherical, charged
black hole, and the Kerr-Newman geometry for a rotating, charged black hole.
The fourth topic is homogeneous, isotropic cosmology, the Friedmann-Robertson-Walker (FRW) geometry.
The FRW geometry forms the essential background spacetime for the cosmological perturbation theory to
be encountered later.
The book now enters the second pass. Tetrads systems of locally inertial (or other) frames attached to
1
A strategy that might fail for students like myself. I learned special relativity from L. D. Landau & E. M. Lifshitzs
incomparable The Classical Theory of Fields. I recall vividly the extraordinary delight in discovering a text that, in
contrast to those dreadful books that conveyed the idea that electromagnetism was something to do with resistors and
capacitors, put relativity, Maxwells equations, and actions up front.
Preface 5
each point of spacetime are well known to, and widely used by, general relativists. The tetrad approach
to general relatiivity is more complicated than the coordinate approach in that it requires an additional
superstructure. However, the advantage of being able to see straight, because you are working in an or-
thonormal frame, outweighs the disadvantage of the additional overhead. While the coordinate approach is
adequate for simple spacetimes ideal black holes, and the FRW geometry its defects are a barrier to
understanding more complicated spacetimes. Most texts do not cover tetrads, or cover them as an aside. In
this book, tetrads are developed systematically, in one self-contained chapter.
The sixth topic is black hole interiors. Cool, but needs more work.
The seventh topic is general relativistic perturbation theory, some understanding of which is prerequisite
for dealing with cosmological perturbations and the CMB. The approach starts in a thankfully short
chapter with one of the most dicult aspects of general relativistic perturbation theory, namely the
problem of coordinate and tetrad gauge ambiguities. This might seem a peculiar starting point. A more
typical starting point is to vary the metric, pick a gauge, and lo there are waves. However, I think that it
is important to show how, at least in at or FRW background spacetimes, the real physical perturbations
emerge naturally from the formalism, without having to pick a gauge. In at spacetime, the formalism
picks out one particular gauge, the Newtonian gauge (though I think it should be called the Copernican
gauge, because in the solar system it would pick out a Sun-centered almost-Cartesian frame), in which the
perturbations retained are precisely the physical perturbations and no others. The Newtonian/Copernican
gauge provides the natural arena for elucidating general relativistic phenomena such as the dragging of
inertial frames, and gravitational waves.
The eighth and nal topic is cosmological perturbation theory, emphasizing the calculation of uctuations
in the CMB. This was one of the most challenging parts of the book to write, because of the shear volume of
physics that goes into the calculation. I did my best to condense the core of the calculation into a simplest
set of assumptions, Chapter 23. However, if you want to calculate a CMB power spectrum that you can
actually compare to observations, then youll have to go beyond the simple chapter. Subsequent chapters
will help you do that.
Beyond the eight topics described above, there are several chapters, all starred
2
, that contain relevant,
but lower priority, material. The material contains some fun stu, my favourite being How to implement
Lorentz transformations on a computer, 14.20.
2
The idea of starred chapters comes from Steven L. Weinbergs classic 1972 text Gravitation and Cosmology, from which I
learned general relativity. Curiously, I found his starred chapters often more interesting than the unstarred ones.
Notation
Except where actual units are needed, units are such that the speed of light is one, c = 1, and Newtons
gravitational constant is one, G = 1.
The metric signature is +++.
Greek (brown) letters , , ..., denote dummy 4D coordinate indices. Latin (black) letters a, b, ..., denote
dummy 4D tetrad indices. Mid-alphabet Latin letters i, j, ... denote 3D indices, either coordinate (brown)
or tetrad (black). To avoid distraction, colouring is applied only to coordinate indices, not to the coordinates
themselves.
Specic (non-dummy) components of a vector are labelled by the corresponding coordinate (brown) or
tetrad (black) direction, for example A
= A
t
, A
x
, A
y
, A
z
or A
m
= A
t
, A
x
, A
y
, A
z
. Allowing the same
label to denote either a coordinate or a tetrad index risks ambiguity, but it should apparent from the context
what is meant. Some texts distinguish coordinate and tetrad indices for example by a caret on the latter,
but this produces notational overload.
Boldface denotes abstract vectors, in either 3D or 4D. In 4D, A = A
= A
m
m
, where g
denote
coordinate tangent axes, and
m
denote tetrad axes.
Repeated paired dummy indices are summed over, the implicit summation convention. In special and
general relativity, one index of a pair must be up (contravariant), while the other must be down (covariant).
If the space being considered is Euclidean, then both indices may be down.
/x
denotes coordinate partial derivatives, which commute.

m
denotes tetrad directed derivatives,
which do not commute. D
and D
m
denote respectively coordinate-frame and tetrad-frame covariant deriva-
tives.
Choice of metric signature
There is a tendency, by no means unanimous, for general relativists to prefer the +++ metric signature,
while particle physicists prefer +.
For someone like me who does general relativistic visualization, there is no contest: the choice has to be
+++, so that signs remain consistent between 3D spatial vectors and 4D spacetime vectors. For example,
Notation 7
the 3D industry knows well that quaternions provide the most ecient and powerful way to implement
spatial rotations. As shown in Chapter 14, complex quaternions provide the best way to implement Lorentz
transformations, with the subgroup of real quaternions continuing to provide spatial rotations. Compatibility
requires +++. Actually, OpenGL and other graphics languages put spatial coordinates in the rst three
indices, leaving time to occupy the fourth index; but in these notes I stick to the physics convention of
putting time in the zeroth index.
In practical calculations it is convenient to be able to switch transparently between boldface and index
notation in both 3D and 4D contexts. This is where the + signature poses greater potential for
misinterpretation in 3D. For example, with this signature, what is the sign of the 3D scalar product
a b ? (0.1)
Is it a b =
3
i=1
a
i
b
i
or a b =
3
i=1
a
i
b
i
? To be consistent with common 3D usage, it must be the
latter. With the + signature, it must be that a b = a
i
b
i
, where the repeated indices signify implicit
summation over spatial indices. So you have to remember to introduce a minus sign in switching between
boldface and index notation.
As another example, what is the sign of the 3D vector product
a b ? (0.2)
Is it ab =
3
jk=1

ijk
a
j
b
k
or ab =
3
jk=1

i
jk
a
j
b
k
or ab =
3
jk=1

ijk
a
j
b
k
? Well, if you want to switch
transparently between boldface and index notation, and you decide that you want boldface consistently to
signify a vector with a raised index, then maybe youd choose the middle option. To be consistent with
standard 3D convention for the sign of the vector product, maybe youd choose
i
jk
to have positive sign for
ijk an even permutation of xyz.
Finally, what is the sign of the 3D spatial gradient operator

x
? (0.3)
Is it = /x
i
or = /x
i
? Convention dictates the former, in which case it must be that some boldface
3D vectors must signify a vector with a raised index, and others a vector with a lowered index. Oh dear.
PART ONE
SPECIAL RELATIVITY
Concept Questions
1. What does c = universal constant mean? What is speed? What is distance? What is time?
2. c +c = c. How can that be possible?
3. The rst postulate of special relativity asserts that spacetime forms a 4-dimensional continuum. The fourth
postulate of special relativity asserts that spacetime has no absolute existence. Isnt that a contradiction?
4. The principle of special relativity says that there is no absolute spacetime, no absolute frame of reference
with respect to which position and velocity are dened. Yet does not the cosmic microwave background
dene such a frame of reference?
5. How can two people moving relative to each other at near c both think each others clock runs slow?
6. How can two people moving relative to each other at near c both think the other is Lorentz-contracted?
7. All paradoxes in special relativity have the same solution. In one word, what is that solution?
8. All conceptual paradoxes in special relativity can be understood by drawing what kind of diagram?
9. Your twin takes a trip to Cen at near c, then returns to Earth at near c. Meeting your twin, you see
that the twin has aged less than you. But from your twins perspective, it was you that receded at near
c, then returned at near c, so your twin thinks you aged less. Is it true?
10. Blobs in the jet of the galaxy M87 have been tracked by the Hubble Space Telescope to be moving at
about 6c. Does this violate special relativity?
11. If you watch an object move at near c, does it actually appear Lorentz-contracted? Explain.
12. You speed towards the center of our Galaxy, the Milky Way, at near c. Does the center appear to you
closer or farther away?
13. You go on a trip to the center of the Milky Way, 30,000 lightyears distant, at near c. How long does the
trip take you?
14. You surf a light ray from a distant quasar to Earth. How much time does the trip take, from your
perspective?
15. If light is a wave, what is waving?
16. As you surf the light ray, how fast does it appear to vibrate?
17. How does the phase of a light ray vary along the light ray? Draw surfaces of constant phase on a spacetime
diagram.
12 Concept Questions
18. You see a distant galaxy at a redshift of z = 1. If you could see a clock on the galaxy, how fast would the
clock appear to tick? Could this be tested observationally?
19. You take a trip to Cen at near c, then instantaneously accelerate to return at near c. If you are looking
through a telescope at a clock on the Earth while you instantaneously accelerate, what do you see happen
to the clock?
20. In what sense is time an imaginary spatial dimension?
21. In what sense is a Lorentz boost a rotation by an imaginary angle?
22. You know what it means for an object to be rotating at constant angular velocity. What does it mean for
an object to be boosting at a constant rate?
23. A wheel is spinning so that its rim is moving at near c. The rim is Lorentz-contracted, but the spokes are
not. How can that be?
24. You watch a wheel rotate at near the speed of light. The spokes appear bent. How can that be?
25. Does a sunbeam appear straight or bent when you pass by it at near the speed of light?
26. Energy and momentum are unied in special relativity. Explain.
27. In what sense is mass equivalent to energy in special relativity? In what sense is mass dierent from
energy?
28. Why is the Minkowski metric unchanged by a Lorentz transformation?
29. What is the best way to program Lorentz transformations on a computer?
Whats important in Special Relativity
See
http://casa.colorado.edu/
~
ajsh/sr/
1. Postulates of special relativity.
2. Understanding conceptually the unication of space and time implied by special relativity.
a. Spacetime diagrams.
b. Simultaneity.
c. Understanding the paradoxes of relativity time dilation, Lorentz contraction, the twin paradox.
3. The mathematics of spacetime transformations
a. Lorentz transformations.
b. Invariant spacetime distance.
c. Minkowski metric.
d. 4-vectors.
e. Energy-momentum 4-vector. E = mc
2
.
f. The energy-momentum 4-vector of massless particles, such as photons.
4. What things look like at relativistic speeds.
1
Special Relativity
1.1 The postulates of special relativity
The theory of special relativity can be derived formally from a small number of postulates:
1. Space and time form a 4-dimensional continuum:
2. The existence of locally inertial frames;
3. The speed of light is constant;
4. The principle of special relativity.
The rst two postulates are assertions about the structure of spacetime, while the last two postulates form
the heart of special relativity. Most books mention just the last two postulates, but I think it is important
to know that special (and general) relativity simply postulate the 4-dimensional character of spacetime, and
that special relativity postulates moreover that spacetime is at.
1. Space and time form a 4-dimensional continuum. The correct mathematical word for continuum
is manifold. A 4-dimensional manifold is dened mathematically to be a topological space that is locally
homeomorphic to Euclidean 4-space R
4
.
The postulate that spacetime forms a 4-dimensional continuum is a generalization of the classical Galilean
concept that space and time form separate 3 and 1 dimensional continua. The postulate of a 4-dimensional
spacetime continuum is retained in general relativity.
Physicists widely believe that this postulate must ultimately break down, that space and time are quantized
over extremely small intervals of space and time, the Planck length
_
G/c
3
10
35
m, and the Planck time
_
G/c
5
10
43
s, where G is Newtons gravitational constant, h/(2) is Plancks constant divided by
2, and c is the speed of light.
2. The existence of globally inertial frames. Statement: There exist global spacetime frames with
respect to which unaccelerated objects move in straight lines at constant velocity.
A spacetime frame is a system of coordinates for labelling space and time. Four coordinates are needed,
because spacetime is 4-dimensional. A frame in which unaccelerated objects move in straight lines at constant
1.1 The postulates of special relativity 15
velocity is called an inertial frame. One can easily think of non-inertial frames: a rotating frame, an
accelerating frame, or simply a frame with some bizarre Dahlian labelling of coordinates.
A globally inertial frame is an inertial frame that covers all of space and time. The postulate that
globally inertial frames exist is carried over from classical mechanics (Newtons rst law of motion).
Notice the subtle shift from the Newtonian perspective. The postulate is not that particles move in straight
lines, but rather that there exist spacetime frames with respect to which particles move in straight lines.
Implicit in the assumption of the existence of globally inertial frames is the assumption that the geometry
of spacetime is at, the geometry of Euclid, where parallel lines remain parallel to innity. In general
relativity, this postulate is replaced by the weaker postulate that local (not global) inertial frames exist. A
locally inertial frame is one which is inertial in a small neighbourhood of a spacetime point. In general
relativity, spacetime can be curved.
3. The speed of light is constant. Statement: The speed of light c is a universal constant, the same in
any inertial frame.
This postulate is the nub of special relativity. The immediate challenge of this chapter, 1.2, is to confront
its paradoxical implications, and to resolve them.
Measuring speed requires being able to measure intervals of both space and time: speed is distance travelled
divided by time elapsed. Inertial frames constitute a special class of spacetime coordinate systems; it is with
respect to distance and time intervals in these special frames that the speed of light is asserted to be constant.
In general relativity, arbitrarily weird coordinate systems are allowed, and light need move neither in
straight lines nor at constant velocity with respect to bizarre coordinates (why should it, if the labelling
of space and time is totally arbitrary?). However, general relativity asserts the existence of locally inertial
frames, and the speed of light is a universal constant in those frames.
In 1983, the General Conference on Weights and Measures ocially dened the speed of light to be
c 299,792,458 ms
1
, (1.1)
and the meter, instead of being a primary measure, became a secondary quantity, dened in terms of the
second and the speed of light.
4. The principle of special relativity. Statement: The laws of physics are the same in any inertial
frame, regardless of position or velocity.
Physically, this means that there is no absolute spacetime, no absolute frame of reference with respect to
which position and velocity are dened. Only relative positions and velocities between objects are meaningful.
It is to be noted that the principle of special relativity does not imply the constancy of the speed of light,
although the postulates are consistent with each other. Moreover the constancy of the speed of light does
not imply the Principle of Special Relativity, although for Einstein the former appears to have been the
inspiration for the latter.
An example of the application of the principle of special relativity is the construction of the energy-
momentum 4-vector of a particle, which should have the same form in any inertial frame (1.10).
16 Special Relativity
1.2 The paradox of the constancy of the speed of light
The postulate that the speed of light is the same in any inertial frame leads immediately to a paradox.
Resolution of this paradox compels a revolution in which space and time are united from separate 3 and
1-dimensional continua into a single 4-dimensional continuum.
Here, Figure ??, is Vermilion. She emits a ash of light. Vermilion thinks that the light moves outward
at the same speed in all directions. So Vermilion thinks that she is at the centre of the expanding sphere of
light.
But here also, Figure ??, is Cerulean, moving away from Vermilion, at about
1
2
the speed of light. Vermilion
thinks that she is at the centre of the expanding sphere of light, as before. But, says special relativity,
Cerulean also thinks that the light moves outward at the same speed in all directions from him. So Cerulean
should be at the centre of the expanding light sphere too. But hes not, is he. Paradox!
Concept question 1.1 Would the light have expanded dierently if Cerulean had emitted the light?
1.2.1 Challenge
Can you gure out Einsteins solution to the paradox? Somehow you have to arrange that both Vermilion
and Cerulean regard themselves as being in the centre of the expanding sphere of light.
1.2.2 Spacetime diagram
A spacetime diagram suggests a way of thinking which leads to the solution of the paradox of the constancy
of the speed of light. Indeed, spacetime diagrams provide the way to resolve all conceptual paradoxes in
special relativity, so it is thoroughly worthwhile to understand them.
A spacetime diagram, Figure ??, is a diagram in which the vertical axis represents time, while the
horizontal axis represents space. Really there are three dimensions of space, which can be thought of as
lling additional horizontal dimensions. But for simplicity a spacetime diagram usually shows just one
spatial dimension.
In a spacetime diagram, the units of space and time are chosen so that light goes one unit of distance
in one unit of time, i.e. the units are such that the speed of light is one, c = 1. Thus light always moves
upward at 45
from vertical in a spacetime diagram. Each point in 4-dimensional spacetime is called an

event. Light signals converging to or expanding from an event follow a 3-dimensional hypersurface called
the lightcone. Light converging on to an event in on the past lightcone, while light emerging from an
event is on the future lightcone.
Here is a spacetime diagram of Vermilion emitting a ash of light, and Cerulean moving relative to
Vermilion at about
1
2
the speed of light. This is a spacetime diagram version of the situation illustrated
in Figure ??. The lines along which Vermilion and Cerulean move through spacetime are called their
worldlines.
Consider again the challenge problem. The problem is to arrange that both Vermilion and Cerulean are
at the centre of the lightcone, from their own points of view.
1.3 Paradoxes and simultaneity 17
Heres a clue. Ceruleans concept of space and time may not be the same as Vermilions.
1.2.3 Centre of the lightcone
Einsteins solution to the paradox is that Ceruleans spacetime is skewed compared to Vermilions, as illus-
trated by Figure ??. The thing to notice in the diagram is that Cerulean is in the centre of the lightcone,
according to the way Cerulean perceives space and time. Vermilion remains at the centre of the lightcone
according to the way Vermilion perceives space and time. In the diagram Vermilion and her space are drawn
at one tick of her clock past the point of emission, and likewise Cerulean and his space are drawn at one
tick of his identical clock past the point of emission. Of course, from Ceruleans point of view his spacetime
is quite normal, and its Vermilions spacetime that is skewed.
In special relativity, the transformation between the spacetime frames of two inertial observers is called a
Lorentz transformation. In general, a Lorentz transformation consists of a spatial rotation about some
spatial axis, combined with a Lorentz boost by some velocity in some direction.
Only space along the direction of motion gets skewed with time. Distances perpendicular to the direction
of motion remain unchanged. Why must this be so? Consider two hoops which have the same size when at
rest relative to each other. Now set the hoops moving towards each other. Which hoop passes inside the
other? Neither! For suppose Vermilion thinks Ceruleans hoop passed inside hers; by symmetry, Cerulean
must think Vermilions hoop passed inside his; but both cannot be true; the only possibility is that the hoops
remain the same size in directions perpendicular to the direction of motion.
Cottoned on? Then you have understood the crux of special relativity, and you can now go away and
gure out all the mathematics of Lorentz transformations. Just like Einstein. The mathematical problem is:
what is the relation between the spacetime coordinates t, x, y, z and t
, x
, y
, z
of a spacetime interval,
a 4-vector, in Vermilions versus Ceruleans frames, if Cerulean is moving relative to Vermilion at velocity v
in, say, the x direction? The solution follows from requiring
1. that both observers consider themselves to be at the centre of the lightcone, and
2. that distances perpendicular to the direction of motion remain unchanged,
as illustrated by Figure ??. (An alternative version of the second condition is that a Lorentz transformation
at velocity v followed by a Lorentz transformation at velocity v should yield the unit transformation.)
Note that the postulate of the existence of globally inertial frames implies that Lorentz transformations
are linear, that straight lines (4-vectors) in one inertial spacetime frame transform into straight lines in other
inertial frames.
You will solve this problem in the next section but two, 1.5. As a prelude, the next two sections, 1.3
and 1.4 discuss simultaneity and time dilation.
1.3 Paradoxes and simultaneity
Most (all?) of the apparent paradoxes of special relativity arise because observers moving at dierent veloc-
ities relative to each other have dierent notions of simultaneity.
1.3.1 Operational denition of simultaneity
How can simultaneity, the notion of events ocurring at the same time at dierent places, be dened opera-
tionally?
One way is illustrated in Figure ??. Vermilion surrounds herself with a set of mirrors, equidistant from
Vermilion. She sends out a ash of light, which reects o the mirrors back to Vermilion. How does Vermilion
know that the mirrors are all the same distance from her? Because the relected ash returns from the mirrors
to Vermilion all at the same instant.
Vermilion asserts that the light ash must have hit all the mirrors simultaneously. Vermilion also asserts
that the instant when the light hit the mirrors must have been the instant, as registered by her wristwatch,
precisely half way between the moment she emitted the ash and the moment she received it back again. If it
takes, say, 2 seconds between ash and receipt, then Vermilion concludes that the mirrors are 1 lightsecond
away from her.
Figure ?? shows a spacetime diagram of Vermilions mirror experiment above. According to Vermilion, the
light hits the mirrors everywhere at the same instant, and the spatial hyperplane passing through these events
is a hypersurface of simultaneity. More generally, from Vermilions perspective, each horizontal hyperplane
in the spacetime diagram is a hypersurface of simultaneity.
Cerulean denes surfaces of simultaneity using the same operational setup: he encompasses himself with
mirrors, arranging them so that a ash of light returns from them to him all at the same instant. But
whereas Cerulean concludes that his mirrors are all equidistant from him and that the light bounces o
them all at the same instant, Vermilion thinks otherwise. From Vermilions point of view, the light bounces
o Ceruleans mirrors at dierent times and moreover at dierent distances from Cerulean. Only so can the
speed of light be constant, as Vermilion sees it, and yet the light return to Cerulean all at the same instant.
Of course from Ceruleans point of view all is ne: he thinks his mirrors are equidistant from him, and
that the light bounces o them all at the same instant. The inevitable conclusion is that Cerulean must
measure space and time along axes that are skewed relative to Vermilions. Events that happen at the same
time according to Cerulean happen at dierent times according to Vermilion; and vice versa. Ceruleans
hypersurfaces of simultaneity are not the same as Vermilions.
From Ceruleans point of view, Cerulean remains always at the centre of the lightcone. Thus for Cerulean,
as for Vermilion, the speed of light is constant, the same in all directions.
1.4 Time dilation
Vermilion and Cerulean construct identical clocks, consisting of a light beam which bounces o a mirror.
Tick, the light beam hits the mirror, tock, the beam returns to its owner. As long as Vermilion and Cerulean
remain at rest relative to each other, both agree that each others clock tick-tocks at the same rate as their
own.
But now suppose Cerulean goes o at velocity v relative to Vermilion, in a direction perpendicular to the
direction of the mirror. A far as Cerulean is concerned, his clock tick-tocks at the same rate as before, a tick
1.4 Time dilation 19
at the mirror, a tock on return. But from Vermilions point of view, although the distance between Cerulean
and his mirror at any instant remains the same as before, the light has further to go. And since the speed
of light is constant, Vermilion thinks it takes longer for Ceruleans clock to tick-tock than her own. Thus
Vermilion thinks Ceruleans clock runs slow relative to her own.
1.4.1 Lorentz gamma factor
How much slower does Ceruleans clock run, from Vermilions point of view? In special relativity the factor
is called the Lorentz gamma factor , introduced by the Dutch physicist Hendrik A. Lorentz in 1904, one
year before Einstein proposed his theory of special relativity. Let us see how the Lorentz gamma factor is
related to Ceruleans velocity v.
In units where the speed of light is one, c = 1, Vermilions mirror is one tick away from her, and from her
point of view the vertical distance between Cerulean and his mirror is the same, one tick. But Vermilion
thinks that the distance travelled by the light beam between Cerulean and his mirror is ticks. Cerulean is
moving at speed v, so Vermilion thinks he moves a distance of v ticks during the ticks of time taken by
the light to travel from Cerulean to his mirror. Thus, from Vermilions point of view, the vertical line from
Cerulean to his mirror, Ceruleans light beam, and Ceruleans path form a triangle with sides 1, , and v,
as illustrated. Pythogoras theorem implies that
1
2
+ (v)
2
=
2
. (1.2)
From this it follows that the Lorentz gamma factor is related to Ceruleans velocity v by
=
1
1 v
2
, (1.3)
which is Lorentzs famous formula.
1.4.2 Paradox
Vermilion thinks Ceruleans clock runs slow, by the Lorentz factor . But of course from Ceruleans perspec-
tive it is Vermilion who is moving, and Vermilion whose clock runs slow. How can both think the others
clock runs slow? Paradox!
The resolution of the paradox, as usual in special relativity, involves simultaneity, and as usual it helps to
draw a spacetime diagram, such as the one in Figure ??.
While Vermilion thinks events happen simultaneously along horizontal planes in this diagram, Cerulean
thinks events occur simultaneously along skewed planes. Thus Vermilion thinks her clock ticks when Cerulean
is at the point before Ceruleans clock ticks. Conversely, Cerulean thinks his clock ticks when Vermilion is
at the point before Vermilions clock ticks.
Where do the two light beams in Vermilions and Ceruleans clocks go in this spacetime diagram? Figure ??
shows a 3D spacetime diagram.
Concept question 1.2 Figure ?? shows a picture of a 3D cube. Is one edge shorter than the other?
Projected on to the page, it appears so, but in reality all the edges have equal length. In what ways is this
situation similar or disimilar to time dilation in 4D relativity?
1.5 Lorentz transformation
A Lorentz transformation is a rotation of space and time. Lorentz transformations form a 6-dimensional
group, with 3 dimensions from spatial rotations, and 3 dimensions from Lorentz boosts.
If you wish to understand special relativity mathematically, then it is essential for you to go through the
exercise of deriving the form of Lorentz transformations for yourself. Indeed, this problem is the challenge
problem posed in 1.2, recast as a mathematical exercise. For simplicity, it is enough to consider the case of
a Lorentz boost by velocity v along the x-axis.
You can derive the form of a Lorentz transformation either pictorially (geometrically), or algebraically.
Ideally you should do both.
Exercise 1.3 Pictorial derivation of the Lorentz transformation. Construct, with ruler and com-
pass, a spacetime diagram that looks like the one in Figure ??. You should recognize that the square
represents the paths of lightrays that Vermilion uses to dene a hypersurface of simultaneity, while the
rectangle represents the same thing for Cerulean. Notice that Ceruleans worldline and line of simultaneity
are diagonals along his light rectangle, so the angles between those lines and the lightcone are equal. Notice
also that the areas of the square and the rectangle are the same, which expresses the fact that the area is
multiplied by the determinant of the Lorentz transformation matrix, which must be one (why?). Use your
geometric construction to derive the mathematical form of the Lorentz transformation.
Exercise 1.4 3D model of the Lorentz transformation. Make a 3D spacetime diagram of the Lorentz
transformation, with not only an x-dimension, as in the previous problem, but also a y-dimension. Resist the
temptation to use a 3D computer modelling program. Believe me, you will learn much more from hands-on
model-making. Make the lightcone from exible paperboard, the spatial hypersurface of simultaneity from
sti paperboard, and the worldline from wooden dowel.
Exercise 1.5 Mathematical derivation of the Lorentz transformation. Relative to person A (Ver-
milion, unprimed frame), person B (Cerulean, primed frame) moves at velocity v along the x-axis. Derive
the form of the Lorentz transformation between the coordinates (t, x, y, z) of a 4-vector in As frame and the
corresponding coordinates (t
, x
, y
, z
) in Bs frame from the assumptions:

1. that the transformation is linear;
2. that the spatial coordinates in the directions orthogonal to the direction of motion are unchanged;
3. that the speed of light c is the same for both A and B, so that x = t in As frame transforms to x
= t
in
Bs frame, and likewise x = t in As frame transforms to x
= t
in Bs frame;
4. the denition of speed; if B is moving at speed v relative to A, then x = vt in As frame transforms to
x
= 0 in Bs frame;
1.5 Lorentz transformation 21
5. spatial isotropy; specically, show that if A thinks B is moving at velocity v, then B must think that A
is moving at velocity v, and symmetry (spatial isotropy) between these two situations then xes the
Lorentz factor.
Your logic should be precise, and explained in clear, concise English.
You should nd that the Lorentz transformation for a Lorentz boost by velocity v along the x-axis is
t
= t vx
x
= vt +x
y
= y
z
= z
,
t = t
+vx
x = vt
+x
y = y
z = z
. (1.4)
The transformation can be written elegantly in matrix notation:
_
_
_
_
t
_
_
_
_
=
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
t
x
y
z
_
_
_
_
, (1.5)
with inverse
_
_
_
_
t
x
y
z
_
_
_
_
=
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
t
_
_
_
_
. (1.6)
A Lorentz transformation at velocity v followed by a Lorentz transformation at velocity v in the opposite
direction, i.e. at velocity v, yields the unit transformation, as it should:
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
=
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
. (1.7)
The determinant of the Lorentz transformation is one, as it should be:
v 0 0
v 0 0
0 0 1 0
0 0 0 1
=
2
(1 v
2
) = 1 . (1.8)
Indeed, requiring that the determinant be one provides another derivation of the formula (1.3) for the Lorentz
gamma factor.
Concept question 1.6 Why must the determinant of a Lorentz transformation be one?
1.6 Paradoxes: Time dilation, Lorentz contraction, and the Twin paradox
There are several classic paradoxes in special relativity. Two of them have already been met above: the
paradox of the constancy of the speed of light in 1.2, and the paradox of time dilation in 1.4. This section
collects three famous paradoxes: time dilation (reiterating 1.4), Lorentz contraction, and the Twin paradox.
If you wish to understand special relativity conceptually, then you should work through all these paradoxes
yourself. As remarked in 1.3, most (all?) paradoxes in special relativity arise because dierent observers
have dierent notions of simultaneity, and most (all?) paradoxes can be solved using spacetime diagrams.
The Twin paradox is particularly helpful because it illustrates several dierent facets of special relativity,
not only time dilation, but also how light travel time modies what an observer actually sees.
1.6.1 Time dilation
If a timelike interval t, r corresponds to motion at velocity v, then r = vt. The proper time along the
interval is
=
_
t
2
r
2
= t
_
1 v
2
=
t
. (1.9)
This is Lorentz time dilation: the proper time interval experienced by a moving person is a factor less
than the time interval t according to an onlooker.
Exercise 1.7 On a spacetime diagram, show how two observers moving relative to each other can both
consider the others clock to run slow compared to their own.
1.6.2 Fitzgerald-Lorentz contraction
Consider a rocket of proper length l, so that in the rockets own rest frame (primed) the back and front ends
of the rocket move through time t
with coordinates
t
, x
= t
, 0 and t
, l . (1.10)
From the perspective of an observer who sees the rocket move at velocity v in the x-direction, the worldlines
of the back and front ends of the rocket are at
t, x = t
, vt
and t
+vl, vt
+l . (1.11)
However, the observer measures the length of the rocket simultaneously in their own frame, not the rocket
frame. Solving for t
= t at the back and t
+vl = t at the front gives

t, x = t, vt and
_
t, vt +
l
_
(1.12)
which says that the observer measures the front end of the rocket to be a distance l/ ahead of the back
end. This is Lorentz contraction: an object of proper length l is measured by a moving person to be shorter
by a factor .
Exercise 1.8 On a spacetime diagram, show how two observers moving relative to each other can both
consider the other to be contracted along the direction of motion.
1.6.3 Twin paradox
See Exercise 1.11 at the end of the chapter.
1.7 The spacetime wheel
1.7.1 3D wheel
Figure ?? shows an ordinary 3D wheel. As the wheel rotates, a point on the wheel describes an invariant
circle. The coordinates x, y of a point on the wheel relative to its centre change, but the distance r between
the point and the centre remains constant
r
2
= x
2
+y
2
= constant . (1.13)
More generally, the coordinates x, y, z of the interval between any two points in 3-dimensional space (a
vector) change when the coordinate system is rotated in 3 dimensions, but the separation r of the two points
remains constant
r
2
= x
2
+y
2
+z
2
= constant . (1.14)
1.7.2 4D spacetime wheel
Figure ?? shows a 4D spacetime wheel. The diagram here is a spacetime diagram, with time t vertical and
space x horizontal. A rotation between time t and space x is a Lorentz boost in the x-direction. As the
spacetime wheel boosts, a point on the wheel describes an invariant hyperbola. The spacetime coordinates
t, x of a point on the wheel relative to its centre change, but the spacetime separation s between the point
and the centre remains constant
s
2
= t
2
+x
2
= constant . (1.15)
More generally, the coordinates t, x, y, z of the interval between any two events in 4-dimensional spacetime
(a 4-vector) change when the coordinate system is boosted or rotated, but the spacetime separation s of the
two events remains constant
s
2
= t
2
+x
2
+y
2
+z
2
= constant . (1.16)
1.7.3 Lorentz boost as a rotation by an imaginary angle
The sign instead of a + sign in front of the t
2
in the spacetime separation formula (1.16) means that time
t can often be treated mathematically as if it were an imaginary spatial dimension. That is, t = iw where
i
1 and w is a fourth spatial coordinate.

A Lorentz boost by a velocity v can likewise be treated as a rotation by an imaginary angle. Consider a
normal spatial rotation in which a primed frame is rotated in the wx-plane clockwise by an angle a about
the origin, relative to the unprimed frame. The relation between the coordinates w
, x
and w, x of a
point in the two frames is
_
w
_
=
_
cos a sina
sin a cos a
__
w
x
_
. (1.17)
Now set t = iw and = ia with t and both real. In other words, take the spatial coordinate w to be
imaginary, and the rotation angle a likewise to be imaginary. Then the rotation formula above becomes
_
t
_
=
_
cosh sinh
sinh cosh
__
t
x
_
(1.18)
This agrees with the usual Lorentz transformation formula (??) if the boost velocity v and boost angle
are related by
v = tanh , (1.19)
so that
= cosh , v = sinh . (1.20)
This provides a convenient way to add velocities in special relativity: the boost angles simply add (for boosts
along the same direction), just as spatial rotation angles add (for rotations about the same axis). Thus a
boost by velocity v
1
= tanh
1
followed by a boost by velocity v
2
= tanh
2
in the same direction gives a
net velocity boost of v = tanh where
=
1
+
2
. (1.21)
The equivalent formula for the velocities themselves is
v =
v
1
+v
2
1 +v
1
v
2
, (1.22)
the special relativistic velocity addition formula.
1.7.4 Trip across the Universe at constant acceleration
Suppose that you took a trip across the Universe in a spaceship, accelerating all the time at one Earth
gravity g. How far would you travel in how much time?
The spacetime wheel oers a cute way to solve this problem, since the rotating spacetime wheel can
be regarded as representing spacetime frames undergoing constant acceleration. Specically, points on the
right quadrant of the rotating spacetime wheel represent worldlines of persons who accelerate with constant
acceleration in their own frame.
If the units of space and time are chosen so that the speed of light and the gravitational acceleration are
both one, c = g = 1, then the proper time experienced by the accelerating person is the boost angle , and
Table 1.1 Trip across the Universe.
Time elapsed Time elapsed
on spaceship on Earth Distance travelled To
in years in years in lightyears
sinh cosh 1
0 0 0 Earth (starting point)
1 1.175 .5431
2 3.627 2.762
2.337 5.127 4.223 Proxima Cen
3.962 26.3 25.3 Vega
6.60 368 367 Pleiades
10.9 2.7 10
4
2.7 10
4
Centre of Milky Way
15.4 2.44 10
6
2.44 10
6
Andromeda galaxy
18.4 4.9 10
7
4.9 10
7
Virgo cluster
19.2 1.1 10
8
1.1 10
8
Coma cluster
25.3 5 10
10
5 10
10
Edge of observable Universe
the time and space coordinates of the accelerating person, relative to a person who remains at rest, are those
of a point on the spacetime wheel, namely
t, x = sinh, cosh . (1.23)
In the case where the acceleration is one Earth gravity, g = 9.80665 ms
2
, the unit of time is
c
g
=
299,792,458 ms
1
9.80665 ms
2
= 0.97 yr , (1.24)
just short of one year. For simplicity, Table 1.1, which tabulates some milestones along the way, takes the
unit of time to be exactly one year, which would be the case if one were accelerating at 0.97 g = 9.5 ms
2
.
After a slow start, you cover ground at an ever increasing rate, crossing 50 billion lightyears, the distance
to the edge of the currently observable Universe, in just over 25 years of your own time.
Does this mean you go faster than the speed of light? No. From the point of view of a person at rest
on Earth, you never go faster than the speed of light. From your own point of view, distances along your
direction of motion are Lorentz-contracted, so distances that are vast from Earths point of view appear
much shorter to you. Fast as the Universe rushes by, it never goes faster than the speed of light.
This rosy picture of being able to it around the Universe has drawbacks. Firstly, it would take a huge
amount of energy to keep you accelerating at g. Secondly, you would use up a huge amount of Earth time
travelling around at relativistic speeds. If you took a trip to the edge of the Universe, then by the time
you got back not only would all your friends and relations be dead, but the Earth would probably be gone,
swallowed by the Sun in its red giant phase, the Sun would have exhausted its fuel and shrivelled into a
cold white dwarf star, and the Solar System, having orbited the Galaxy a thousand times, would be lost
somewhere in its milky ways.
Technical point. The Universe is expanding, so the distance to the edge of the currently observable Universe
is increasing. Thus it would actually take longer than indicated in the table to reach the edge of the currently
observable Universe. Moreover if the Universe is accelerating, as recent evidence from the Hubble diagram
of Type Ia Supernovae suggests, then you will never be able to reach the edge of the currently observable
Universe, however fast you go.
1.8 Scalar spacetime distance
One of the most fundamental features of a Lorentz transformation is that its leaves invariant a certain
distance. the scalar spacetime distance, between any two events in spacetime. The scalar spacetime distance
s between two events separated by t, x, y, z is given by
s
2
= t
2
+ r
2
= t
2
+ x
2
+ y
2
+ z
2
. (1.25)
A quantity such as s
2
that remains unchanged under any Lorentz transformation is called a scalar. It is
left to you in exercise 1.9 to show explicitly that s
2
is unchanged under Lorentz transformations. Lorentz
transformations can be dened as linear spacetime transformations that leave s
2
invariant.
The single scalar spacetime squared interval s
2
replaces the two scalar quantities
time interval t
distance interval r =
_
x
2
+ y
2
+ z
2
(1.26)
of classical Galilean spacetime.
Exercise 1.9 Invariant spacetime interval. Show that the squared spacetime interval s
2
dened by
equation (1.25) is unchanged by a Lorentz transformation, that is, show that
t
2
+ x
2
+ y
2
+ z
2
= t
2
+ x
2
+ y
2
+ z
2
. (1.27)
You may assume without proof the familiar results that the 3D scalar product r
2
= x
2
+ y
2
+ z
2
is
unchanged by a spatial rotation, so it suces to consider a Lorentz boost, say in the x direction.
1.8.1 Proper time, proper distance
The scalar spacetime squared interval s
2
has a physical meaning.
If an interval t, r is timelike, t > r, then the square root of minus the spacetime interval squared
is the proper time along it
=
_
s
2
=
_
t
2
r
2
. (1.28)
This is the time experienced by an observer moving along that interval.
1.9 4-vectors 27
If an interval t, r is spacelike, t < r, then the spacetime interval equals the proper distance
l along it
l =
s
2
=
_
r
2
t
2
. (1.29)
This is the distance between two events measured by an observer for whom those events are simultaneous.
Concept question 1.10 Justify the assertions (1.28) and (1.29).
1.8.2 Timelike, lightlike, spacelike
A spacetime interval x
m
is called
timelike if s
2
< 0 ,
null or lightlike if s
2
= 0 ,
spacelike if s
2
> 0 .
(1.30)
1.8.3 Minkowski metric
The scalar spacetime squared interval s
2
associated with an interval x
m
= t, r = t, x, y, z
can be written
s
2
= x
m
x
m
=
mn
x
m
x
n
(1.31)
where
mn
is the Minkowski metric
mn

_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
. (1.32)
Equation (1.31) uses the implicit summation convention, according to which paired indices are explicitly
summed over. Invariably, one of a pair of repeated indices is raised, the other lowered.
1.9 4-vectors
A 4-vector in special relativity is a quantity a
m
= a
t
, a
x
, a
y
, a
z
that transforms under Lorentz transfor-
mations like an interval x
m
= t, x, y, z of spacetime
a
m
= L
m
n
a
n
(1.33)
where L
m
n
denotes a Lorentz transformation.
1.9.1 Index notation
In special and general relativity it is convenient to introduce two versions of the same 4-vector quantity, one
with raised indices, called the contravariant components of the 4-vector,
a
m
a
t
, a
x
, a
y
, a
z
, (1.34)
and one with lowered indices called the covariant components of the 4-vector,
a
m
a
t
, a
x
, a
y
, a
z
(1.35)
(the naming is crazy, and you do not need to remember it).
The indices run over m = t, x, y, z, or sometimes m = 0, 1, 2, 3.
Why introduce raised and lowered indices? Because
a
m
a
m
m
a
m
a
m
= a
t
a
t
+a
x
a
x
+a
y
a
y
+a
z
a
z
= (a
t
)
2
+ (a
x
)
2
+ (a
y
)
2
+ (a
z
)
2
(1.36)
is a Lorentz scalar.
1.10 Energy-momentum 4-vector
Symmetry argument:
Symmetry Conservation law
Time translation Energy
Space translation Momentum
suggests
energy = time component
momentum = space component
_
of 4-vector. (1.37)
The Principle of Special Relativity requires that the equation of energy-momentum conservation
energy
momentum
= constant (1.38)
should take the same form in any inertial frame. The equation should be Lorentz covariant, that is, the
equation should transform like a Lorentz 4-vector.
1.10 Energy-momentum 4-vector 29
1.10.1 Construction of the energy-momentum 4-vector
Require:
1. its a 4-vector
2. goes over to the Newtonian limit as v 0.
Newtonian limit:
Momentum p is mass m times velocity v
p = mv = m
dr
dt
. (1.39)
4D version:
Need to do two things to Newtonian momentum:
replace r by a 4-vector x
m
= t, r
replace dt by a scalar the only obvious choice is the proper time interval .
Result:
p
k
= m
dx
k
d
= m
_
dt
d
,
dr
d
_
= m, v (1.40)
which are special relativistic versions of energy E and momentum p
p
k
= E, p = m, mv . (1.41)
1.10.2 Special relativistic energy
E = m (units c = 1) (1.42)
or, restoring standard units
E = mc
2
. (1.43)
Taylor expand for small velocity v:
=
1
_
1 v
2
/c
2
= 1 +
1
2
v
2
c
2
+... (1.44)
so
E = mc
2
_
1 +
1
2
v
2
c
2
+...
_
= mc
2
+
1
2
mv
2
+... . (1.45)
The rst term, mc
2
, is the rest-mass energy. The second term,
1
2
mv
2
, is the non-relativistic kinetic energy.
Higher-order terms give relativistic corrections to the kinetic energy.
1.10.3 Rest mass is a scalar
The scalar quantity constructed from the energy-momentum 4-vector p
k
= E, p is
p
k
p
k
= E
2
+p
2
= m
2
(
2
2
v
2
)
= m
2
(1.46)
minus the square of the rest mass.
1.11 Photon energy-momentum
Photons have zero rest mass
m = 0 . (1.47)
Thus
p
k
p
k
= E
2
+p
2
= m
2
= 0 (1.48)
whence
p [p[ = E . (1.49)
Hence
p
k
= E, p
= E1, n
= h1, n (1.50)
where is the photon frequency.
The photon velocity is n, a unit vector. The photon speed is one (the speed of light).
1.11.1 Lorentz transformation of photon energy-momentum 4-vector
Follows the usual rules for 4-vectors.
In the case that the Lorentz transformation is a Lorentz boost along the x-axis, the transformation is
_
_
_
_
p
t
p
x
p
y
p
z
_
_
_
_
=
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
p
t
p
x
p
y
p
z
_
_
_
_
=
_
_
_
_
(p
t
vp
x
)
(p
x
vp
t
)
p
y
p
z
_
_
_
_
. (1.51)
1.12 Abstract 4-vectors 31
Equivalently
h
_
_
_
_
1
n
x
n
y
n
z
_
_
_
_
=
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
h
_
_
_
_
1
n
x
n
y
n
z
_
_
_
_
= h
_
_
_
_
(1 n
x
v)
(n
x
v)
n
y
n
z
_
_
_
_
.
These mathematical relations imply the rules of 4-dimensional perspective, 1.13.1.
1.11.2 Redshift
Astronomers dene the redshift z of a photon by
z

obs
emit
emit
. (1.52)
In relativity, it is often more convenient to use the redshift factor 1 +z
1 +z

obs
emit
=

emit
obs
. (1.53)
1.11.3 Special relativistic Doppler shift
If the emitter frame (primed) is moving with velocity v in the x-direction relative to the observer frame
(unprimed) then
h
emit
= h
obs
(1 n
x
v) (1.54)
so
1 +z =

emit
obs
= (1 n
x
v)
= (1 n.v) . (1.55)
This is the general formula for the special relativistic Doppler shift.
1.12 Abstract 4-vectors
A = A
m
m
(1.56)
1.13 What things look like at relativistic speeds
1.13.1 The rules of 4-dimensional perspective
The diagram below illustrates the rules of 4-dimensional perspective, also called special relativistic beam-
ing, which describe how a scene appears when you move through it at near light speed.
1
1
v
On the left, you are at rest relative to the scene. Imagine painting the scene on a celestial sphere around
you. The arrows represent the directions of light rays (photons) from the scene on the celestial sphere to
you at the center.
On the right, you are moving to the right through the scene, at some fraction of the speed of light. The
celestial sphere is stretched along the direction of your motion into a celestial ellipsoid. You, the observer,
are not at the center of the ellipsoid, but rather at one of its foci (the left one, if you are moving to the
right). The scene appears relativistically aberrated, which is to say concentrated ahead of you, and expanded
behind you.
The lengths of the arrows are proportional to the energies, or frequencies, of the photons that you see.
When you are moving through the scene at near light speed, the arrows ahead of you, in your direction
of motion, are longer than at rest, so you see the photons blue-shifted, increased in energy, increased in
frequency. Conversely, the arrows behind you are shorter than at rest, so you see the photons red-shifted,
decreased in energy, decreased in frequency. Since photons are good clocks, the change in photon frequency
also tells you how fast or slow clocks attached to the scene appear to you to run.
Numbers? On the right, you are moving through the scene at v = 0.6 c. The celestial ellipsoid is stretched
along the direction of your motion by the Lorentz gamma factor, which here is = 1/
1 0.6
2
= 1.25. The
focus of the celestial ellipsoid, where you the observer are, is displaced from center by v = 1.250.6 = 0.75.
1.14 How to programme Lorentz transformations on a computer 33
1.14 How to programme Lorentz transformations on a computer
3D gaming programmers are familiar with the fact that the best way to program spatial rotations on a
computer is with quaternions. Compared to standard rotation matrices, quaternions oer increased speed
and require less storage, and their algebraic properties simplify interpolation and splining.
Section 1.7 showed that a Lorentz boost is mathematically equivalent to a rotation by an imaginary
angle. Thus suggests that Lorentz transformations might be treated as complexied spatial rotations, which
proves to be true. Indeed, the best way to program Lorentz transformation on a computer is with complex
quaternions, as will be demonstrated in Chapter 14.
Exercises
Exercise 1.11 Twin paradox. Your twin leaves you on Earth and travels to the spacestation Alpha,
= 3 lyr away, at a good fraction of the speed of light, then immediately returns to Earth at the same speed.
The accompanying spacetime diagram shows the corresponding worldlines of both you and your twin. Aside
from part (a) and the rst part of (b), I want you to derive your answers mathematically, using logic and
Lorentz transformations. However, the diagram is accurately drawn, and you should be able to check your
answers by measuring.
1. Label the worldlines of you and your twin. Draw the worldline of a light signal which travels from you on
Earth, hits Alpha just when your twin arrives, and immediately returns to Earth. Draw the twins now
when just arriving at Alpha, and the twins now just departing from Alpha (in the rst case the twin is
moving toward Alpha, while in the second case the twin is moving back toward Earth).
2. From the diagram, measure the twins speed v relative to you, in units where the speed of light is unity,
c = 1. Deduce the Lorentz gamma factor , and the redshift factor 1 + z = [(1 + v)/(1 v)]
1/2
, in the
cases (i) where the twin is receding, and (ii) where the twin is approaching.
3. Choose the spacetime origin to be the event where the twin leaves Earth. Argue that the position 4-vector
of the twin on arrival at Alpha is
t, x, y, z = /v, , 0, 0 . (1.57)
Lorentz transform this 4-vector to determine the position 4-vector of the twin on arrival at Alpha, in the
twins frame. Express your answer rst in terms of , v, and , and then in (light)years. State in words
what this position 4-vector means.
4. How much do you and your twin age respectively during the round trip to Alpha and back? What is the
ratio of these ages? Express your answers rst in terms of , v, and , and then in years.
5. What is the distance between the Earth and Alpha from the twins point of view? What is the ratio of
this distance to the distance between Earth and Alpha from your point of view? Explain how your arrived
at your result. Express your answer rst in terms of , v, and , and then in lightyears.
6. You watch your twin through a telescope. How much time do you see (through the telescope) elapse
on your twins wristwatch between launch and arrival on Alpha? How much time passes on your own
wristwatch during this time? What is the ratio of these two times? Express your answers rst in terms of
, v, and , and then in years.
7. On arrival at Alpha, your twin looks back through a telescope at your wristwatch. How much time does
your twin see (through the telescope) has elapsed since launch on your watch? How much time has elapsed
on the twins own wristwatch during this time? What is the ratio of these two times? Express your answers
rst in terms of , v, and , and then in years.
8. You continue to watch your twin through a telescope. How much time elapses on your twins wristwatch,
as seen by you through the telescope, during the twins journey back from Alpha to Earth? How much
time passes on your own watch as you watch (through the telescope) the twin journey back from Alpha
to Earth? What is the ratio of these two times? Express your answers rst in terms of , v, and , and
then in years.
9. During the journey back from Alpha to Earth, your twin likewise continues to look through a telescope
at the time registered on your watch. How much time passes on your wristwatch, as seen by your twin
through the telescope, during the journey back? How much time passes on the twins wristwatch from the
twins point of view during the journey back? What is the ratio of these two times? Express your answers
rst in terms of , v, and , and then in years.
Exercise 1.12 Lines intersecting at right angles. Prove that if two lines appear to intersect at right-
angles projected on the sky in one frame, then they appear to intersect at right-angles in another frame
Lorentz-transformed with respect to the rst.
PART TWO
COORDINATE APPROACH TO GENERAL RELATIVITY
Concept Questions
1. What assumption of general relativity makes it possible to introduce a coordinate system?
2. Is the speed of light a universal constant in general relativity? If so, in what sense?
3. What does locally inertial mean? How local is local?
4. Why is spacetime locally inertial?
5. What assumption of general relativity makes it possible to introduce clocks and rulers?
6. Consider two observers at the same point and with the same instantaneous velocity, but one is accelerating
and the other is in free-fall. What is the relation between the proper time or proper distance along an
innitesimal interval measured by the two observers? What assumption of general relativity implies this?
7. Does the (Strong) Principle of Equivalence imply that two unequal masses will fall at the same rate in a
gravitational eld? Explain.
8. In what respects is the Strong Principle of Equivalence (gravity is equivalent to acceleration) stronger
than the Weak Principle of Equivalence (gravitating mass equals inertial mass)?
9. Standing on the surface of the Earth, you hold an object of negative mass in your hand, and drop it.
According to the Principle of Equivalence, does the negative mass fall up or down?
10. Same as the previous question, but what does Newtonian gravity predict?
11. You have a box of negative mass particles, and you remove energy from it. Do the particles move faster
or slower? Does the entropy of the box increase or decrease? Does the pressure exerted by the particles
on the walls of the box increase or decrease?
12. You shine two light beams along identical directions in a gravitational eld. The two light beams are
identical in every way except that they have two dierent frequencies. Does the Equivalence Principle
imply that the interference pattern produced by each of the beams individually is the same?
13. What is a straight line, according to the Principle of Equivalence?
14. If all objects move on straight lines, how is it that when, standing on the surface of the Earth, you throw
two objects in the same direction but with dierent velocities, they follow two dierent trajectories?
15. In relativity, what is the generalization of the shortest distance between two points?
16. What kinds of general coordinate transformations are allowed in general relativity?
17. In general relativity, what is a scalar? A 4-vector? A tensor? Which of the following is a scalar/vector/
tensor/none-of-the-above? (a) a set of coordinates x
; (b) a coordinate interval dx
; (c) proper time

?
18. What does general covariance mean?
19. What does parallel transport mean?
20. Why is it important to dene covariant derivatives that behave like tensors?
21. Is covariant dierentiation a derivation? That is, is covariant dierentiation a linear operation, and does
it obey the Leibniz rule for the derivative of a product?
22. What is the covariant derivative of the metric tensor? Explain.
23. What does a connection coecient
mean physically? Is it a tensor? Why, or why not?

24. An astronaut is in free-fall in orbit around the Earth. Can the astronaut detect that there is a gravitational
eld?
25. Can a gravitational eld exist in at space?
26. How can you tell whether a given metric is equivalent to the Minkowski metric of at space?
27. How many degrees of freedom does the metric have? How many of these degrees of freedom can be removed
by arbitrary transformations of the spacetime coordinates, and therefore how many physical degrees of
freedom are there in spacetime?
28. If you insist that the spacetime is spherical, how many physical degrees of freedom are there in the
spacetime?
29. If you insist that the spacetime is spatially homogeneous and isotropic (the cosmological principle), how
many physical degrees of freedom are there in the spacetime?
30. In general relativity, you are free to prescribe any spacetime (any metric) you like, including metrics with
wormholes and metrics that connect the future to the past so as to violate causality. True or false?
31. If it is true that in general relativity you can prescribe any metric you like, then why arent you bumping
into wormholes and causality violations all the time?
32. How much mass does it take to curve space signicantly (signicantly meaning by of order unity)?
33. What is the relation between the energy-momentum 4-vector of a particle and the energy-momentum
tensor?
34. It is straightforward to go from a prescribed metric to the energy-momentum tensor. True or false?
35. It is straightforward to go from a prescribed energy-momentum tensor to the metric. True or false?
36. Does the Principle of Equivalence imply Einsteins equations?
37. What do Einsteins equations mean physically?
38. What does the Riemann curvature tensor R
mean physically? Is it a tensor?

39. The Riemann tensor splits into compressive (Ricci) and tidal (Weyl) parts. What do these parts mean,
physically?
40. Einsteins equations imply conservation of energy-momentum, but what does that mean?
41. Do Einsteins equations describe gravitational waves?
42. Do photons (massless particles) gravitate?
43. How do dierent forms of mass-energy gravitate?
44. How does negative mass gravitate?
Whats important?
This part of the notes adopts the traditional coordinate-based approach to general relativity. The approach
is neither the most insightful nor the most powerful, but it is the fastest route to connecting the metric to
the energy-momentum content of spacetime.
1. Postulates of general relativity. How do the various postulates imply the mathematical structure of general
relativity?
2. The road from spacetime curvature to energy-momentum:
metric g
connection coecients
Riemann curvature tensor R
Ricci tensor R
and scalar R
Einstein tensor G
= R
1
2
g
R
energy-momentum tensor T
3. 4-velocity and 4-momentum. Geodesic equation.

4. Bianchi identities guarantee conservation of energy-momentum.
2
Fundamentals of General Relativity
2.1 The postulates of General Relativity
General relativity follows from three postulates:
1. Spacetime is a 4-dimensional manifold;
2. The (Strong) Principle of Equivalence;
3. Einsteins Equations.
2.1.1 Spacetime is a 4-dimensional manifold
A 4-dimensional manifold is dened mathematically to be a topological space that is locally homeomorphic
to Euclidean 4-space R
4
.
This postulate implies that it is possible to set up a coordinate system (possibly in patches)
x
x
0
, x
1
, x
2
, x
3
(2.1)
such that each point of (the patch of) spacetime has a unique coordinate.
Andrews convention:
Greek (brown) dummy indices label curved spacetime coordinates.
Latin (black) dummy indices label locally inertial (more generally, tetrad) coordinates.
2.1.2 (Strong) Principle of Equivalence (PE)
The laws of physics in a gravitating frame are equivalent to those in an accelerating frame.
The Weak Principle of Equivalence is Gravitating mass = inertial mass.
PE spacetime is locally inertial (see 2.2).
2.2 Existence of locally inertial frames 41
2.1.3 Einsteins equations
Einsteins equations comprise a 4 4 symmetric matrix of equations
G
= 8GT
. (2.2)
Here G is the Newtonian gravitational constant, G
is the Einstein tensor, and T
is the energy-
momentum tensor.
Physically, Einsteins equations signify
(compressive part of) curvature = energy-momentum content . (2.3)
Einsteins equations generalize Poissons equation
2
= 4G (2.4)
where is the Newtonian gravitational potential, and the mass-energy density. Poissons equation is the
time-time component of Einsteins equations in the limit of a weak gravitational eld and slowly moving
matter.
2.2 Existence of locally inertial frames
The Principle of Equivalence implies that at each point of spacetime it is possible to choose a locally inertial,
or free-fall, frame, such that the laws of special relativity apply within an innitesimal neighbourhood of
that point. By this is meant that at each point of spacetime it is possible to choose coordinates such that
(a) the metric at that point is Minkowski, and (b) the rst derivatives of the metric are all zero.
It is built into the Principle of Equivalence that general relativity is, like special relativity, a metric
theory. Notably, the proper times and distances measured by an accelerating observer are the same as
those measured by a freely-falling observer at the same point and with the same instantaneous velocity.
2.3 Metric
The metric is the essential mathematical object that converts an innitesimal coordinate interval
dx
dx
0
, dx
1
, dx
2
, dx
3
(2.5)
to a proper measurement of an interval of time or space.
Postulate (1) of general relativity means that it is possible to choose coordinates
x
x
0
, x
1
, x
2
, x
3
(2.6)
covering (a patch of) spacetime.
42 Fundamentals of General Relativity
Postulate (2) of general relativity implies that at each point of spacetime it is possible to choose locally
inertial coordinates
m

0
,
1
,
2
,
3
(2.7)
such that the metric is Minkowski,
ds
2
=
mn
d
m
d
n
, (2.8)
in an innitesimal neighborhood of the point. The spacetime distance squared ds
2
is a scalar, a quantity
that is unchanged by the choice of coordinates.
Since
d
m
=

m
x
dx
(2.9)
it follows that
ds
2
=
mn
m
x
n
x
dx
dx
(2.10)
so the scalar spacetime distance squared is
ds
2
= g
dx
dx
(2.11)
where g
is the metric, a 4 4 symmetric matrix

g
=
mn
m
x
n
x
. (2.12)
2.4 Basis g
of tangent vectors
You are familiar with the idea that in ordinary 3D Euclidean geometry it is often convenient to treat vectors
in an abstract coordinate-independent formalism. Thus for example a 3-vector is commonly written as an
abstract quantity r. The coordinates of the vector r may be x, y, z in some particular coordinate system,
but one recognizes that the vector r has a meaning, a magnitude and a direction, that is independent of the
coordinate system adopted. In an arbitrary Cartesian coordinate system, the Euclidean 3-vector r can be
expressed
r =
i
x
i
x
i
= xx + y y + z z (2.13)
where x
i
x, y, z are unit vectors along each of the coordinate axes.
The same kind of abstract notation is useful in general relativity. Dene g
g
0
, g
1
, g
2
, g
3
(2.14)
to be the basis of axes tangent to the coordinates x
. Each axis g
is a 4D vector object, with both magnitude

and direction in spacetime. (Some texts represent the tangent vectors g
with the notation
, but this
notation is not used here, to avoid the potential confusion between
as a derivative and
as a vector.)
2.5 4-vectors and tensors 43
An interval dx
of spacetime can be expressed in coordinate-independent fashion as the abstract vector

interval dx
dx g
dx
= g
0
dx
0
+g
1
dx
1
+g
2
dx
2
+g
3
dx
3
. (2.15)
The scalar length squared of the abstract vector interval dx is
ds
2
= dx dx = g
dx
dx
(2.16)
whence
g
= g
(2.17)
the metric is the (4D) scalar product of tangent vectors.
The tangent vectors g
form a basis for a 4D tangent space that has three important mathematical
properties. First, the tangent space is a vector space, that is, it has the properties of linearity that dene
a vector space. Second, the tangent space has an inner (or scalar) product, dened by the metric (2.17).
Third, vectors in the tangent space can be dierentiated with respect to coordinates, as will be elucidated
in 2.6.3.
2.5 4-vectors and tensors
2.5.1 Contravariant coordinate 4-vector
Under a general coordinate transformation
x
(2.18)
a coordinate interval dx
transforms as
dx
=
x
dx
. (2.19)
In general relativity, a coordinate 4-vector is dened to be a quantity A
= A
0
, A
1
, A
2
, A
3
that trans-
forms under a coordinate transformation (2.18) like a coordinate interval
A
=
x
. (2.20)
2.5.2 Abstract 4-vector
A 4-vector may be written in coordinate-independent fashion as
A = g
. (2.21)
The quantity A is an abstract 4-vector. Although A is a 4-vector, it is by construction unchanged by a
coordinate transformation, and is therefore a coordinate scalar. See 2.5.6 for commentary on the distinction
between abstract and coordinate vectors.
2.5.3 Lowering and raising indices
Dene g
to be the inverse metric, satisfying

g
=
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
. (2.22)
The metric g
and its inverse g
provide the means of lowering and raising coordinate indices. The

components of a coordinate 4-vector A
with raised index are called its contravariant components, while

those A
with lowered indices are called its covariant components,

A
= g
, (2.23)
A
= g
. (2.24)
2.5.4 Covariant coordinate 4-vector
Under a general coordinate transformation (2.18), the covariant components A
of a coordinate 4-vector
transform as
A
=
x
. (2.25)
You can check that the transformation law (2.25) for the covariant components A
is consisistent with the

transformation law (2.20) for the contravariant components A
.
You can check that the tangent vectors g
transform as a covariant coordinate 4-vector.

2.5.5 Scalar product
If A
and B
are coordinate 4-vectors, then their scalar product is

A
= A
= g
. (2.26)
This is a coordinate scalar, a quantity that remains invariant under general coordinate transformations.
In abstract vector formalism, the scalar product of two 4-vectors A = g
and B = g
is
A B = g
= g
. (2.27)
2.5.6 Comment on vector naming and notation
Dierent texts follow dierent conventions for naming and notating vectors and tensors.
In this book I follow the convention of calling both A
(with a dummy index ) and A A
vectors.
Although A
and A are both vectors, they are mathematically dierent objects.

If the index on a vector indicates a specic coordinate, then the indexed vector is the component of the
vector; for example A
0
(or A
t
) is the x
0
(or time t) component of the coordinate 4-vector A
.
In this book, the dierent species of vector are distinguished by an adjective:
1. A coordinate vector A
, identied by Greek (brown) indices , is one that changes in a prescribed way

under coordinate transformations. A coordinate transformation is one that changes the coordinates of the
spacetime without actually changing the spacetime or whatever lies in it.
2. An abstract vector A, identied by boldface, is the thing itself, and is unchanged by the choice of
coordinates. Since the abstract vector is unchanged by a coordinate transformation, it is a coordinate
scalar.
All the types of vector have the properties of linearity (additivity, multiplication by scalars) that identify
them mathematically as belonging to vector spaces. The important distinction between the types of vector
is how they behave under transformations.
In referring to both A
and Aas vectors, I am following the standard physics practice of mentally regarding
A
and A as equivalent objects. You are familiar with the advantages of treating a vector in 3D Euclidean
space either as an abstract vector A, or as a coordinate vector A
i
. Depending on the problem, sometimes
the abstract notation A is more convenient, and sometimes the coordinate notation A
i
is more convenient.
Sometimes its convenient to switch between the two in the middle of a calculation. Likewise in general
relativity it is convenient to have the exibility to work in either coordinate or abstract notation, whatever
suits the problem of the moment.
2.5.7 Coordinate tensor
In general, a coordinate tensor A
...
...
is an object that transforms under general coordinate transforma-
tions (2.18) as
A
...
...
=
x
...
x
... A
...
...
. (2.28)
You can check that the metric tensor g
and its inverse g
are indeed coordinate tensors, transforming

like (2.28).
The rank of a tensor is the number of indices. A scalar is a tensor of rank 0. A 4-vector is a tensor of
rank 1.
2.6 Covariant derivatives
2.6.1 Derivative of a coordinate scalar
Suppose that is a coordinate scalar. Then the coordinate derivative of is a coordinate 4-vector
is a coordinate tensor (2.29)

transforming like equation (2.25).
As a shorthand, the ordinary partial derivative is often denoted in the literature with a comma
=
,
. (2.30)
For the most part this book does not use the comma notation.
2.6.2 Derivative of a coordinate 4-vector
The ordinary partial derivative of a covariant coordinate 4-vector A
is not a tensor
A
is not a coordinate tensor (2.31)

because it does not transform like a coordinate tensor.
However, the 4-vector A = g
, being by construction invariant under coordinate transformations, is a

coordinate scalar, and its partial derivative is a coordinate 4-vector
A
x
=
g
= g
+
g
is a coordinate tensor . (2.32)

The last line of equation (2.32) assumes that it is legitimate to dierentiate the tangent vectors g
, but
what does this mean? The partial derivatives of basis vectors g
are dened in the usual way by

g
lim
x
0
g
(x
0
, ..., x
+x
, ..., x
3
) g
(x
0
, ..., x
, ..., x
3
)
x
. (2.33)
This denition relies on being able to compare the vectors g
(x) at some point x with the vectors g
(x+x)
at another point x+x a small distance away. The comparison between two vectors a small distance apart
is made possible by the existence of locally inertial frames. In a locally inertial frame, two vectors a small
distance apart can be compared by parallel-transporting one vector to the location of the other along
the small interval between them, that is, by transporting the vector without accelerating or precessing with
respect to the locally inertial frame. Thus g
(x+x) in the denition (2.33) should be interpreted as its

value parallel-transported from position x+x to position x along the small interval x between them.
2.6.3 Coordinate connection coecients (Christoel symbols)
The partial derivatives of the basis vectors g
that appear on the right hand side of equation (2.32) dene

the coordinate connection coecients
, also known as Christoel symbols,

g
is not a coordinate tensor . (2.34)

The denition (2.34) shows that the connection coecients express how each tangent vector g
changes,
relative to parallel-transport, when shifted along an interval x
.
2.6.4 Covariant derivative of a contravariant 4-vector
Expression (2.32) along with the denition (2.34) of the connection coecients implies that
A
x
= g
= g
_
A
_
The expression in parentheses is a coordinate tensor, and denes the covariant derivative D
of the
contravariant coordinate 4-vector A

As a shorthand, the covariant derivative is often denoted in the literature with a semi-colon
D
= A
;
. (2.37)
For the most part this book does not use the semi-colon notation.
2.6.5 Covariant derivative of a covariant coordinate 4-vector
Similarly,
A
x
= g

where D
is the covariant derivative of the covariant coordinate 4-vector A

2.6.6 Covariant derivative of a coordinate tensor
In general, the covariant derivative of a coordinate tensor is
D
A
...
...
=
A
...
...
x
A
...
...
+
A
...
...
+...
A
...
...

A
...
...
... (2.40)
with a positive term for each contravariant index, and a negative term for each covariant index.
2.6.7 No-torsion condition
The existence of locally inertial frames requires that it must be possible to arrange not only that the tangent
axes g
are orthonormal at a point, but also that they remain orthonormal to rst order in a Taylor expansion
about the point. That is, it must be possible to choose the coordinates such that the tangent axes g
are
orthonormal, and unchanged to linear order:
g
, (2.41)
g
= 0 . (2.42)
In view of the denition (2.34) of the connection coecients, the second condition (2.42) is equivalent to the
vanishing of all the connection coecients:
= 0 . (2.43)
Under a general coordinate transformation x
, the tangent axes transformas g
= x
/x
.
The 44 matrix x
/x
of partial derivatives provides 16 degrees of freedom in choosing the tangent axes

at a point. The 16 degrees of freedom are enough more than enough to accomplish the orthonormality
condition (2.41), which is a symmetric 4 4 matrix equation with 10 degrees of freedom. The additional
16 10 = 6 degrees of freedom are Lorentz transformations, which rotate the tangent axes g
, but leave the

metric
unchanged.
Just as it is possible to reorient the tangent axes g
at a point by adjusting the matrix x
/x
of rst
partial derivatives of the coordinate transformation x
, so also it is possible to reorient the derivatives

g
/x
of the tangent axes by adjusting the matrix

2
x
/x
of second partial derivatives. The second

partial derivatives comprise a set of 4 symmetric 44 matrices, for a total of 410 = 40 degrees of freedom.
However, there are 4 4 4 = 64 connection coecients
, all of which the condition (2.43) requires to

vanish. The matrix of second derivatives is thus 64 40 = 24 degrees of freedom short of being able to
make all the connections vanish. The resolution of the problem is that, as shown below, equation (2.51),
there are 24 combinations of the connections that form a tensor, the torsion tensor. If a tensor is zero in one
frame, then it is automatically zero in any other frame. Thus the requirement that all the connections vanish
requires that the torsion tensor vanish. This requires, from the expression (2.51) for the torsion tensor, the
no-torsion condition that the connection coecients are symmetric in their last two indices
. (2.44)
It should be emphasized that the condition of vanishing torsion is an assumption of general relativity, not
a mathematical necessity. It has been shown in this section that torsion vanishes if and only if spacetime is
locally at, meaning that at any point coordinates can be found such that conditions (2.41) are true. The
assumption of local atness is central to the idea of the principle of equivalence. But it is an assumption,
not a consequence, of the theory.
Concept question 2.1 If torsion does not vanish, then there is no locally inertial frame. What does
parallel-transport mean in such a case?
2.6.8 Aside: torsion and the integrability of the position vector
From the denitions (2.34) of the connection coecients, the no-torsion condition (2.44) is equivalent to
g
= 0 . (2.45)
According to Frobenius theorem, this condition (2.45) is precisely the condition for the system g
to be
integrable, that is, there exists a position vector X whose partial derivatives are
g
=
X
x
. (2.46)
Equivalently, the total dierential of the position vector X is
dX = g
dx
. (2.47)
The abstract vector interval dx g
dx
was dened by equation (2.15) as the coordinate-independent

version of a spacetime interval dx
. The notation dx was merely symbolic: dx was not necessarily a total

dierential of something. However, the no-torsion condition (2.47) implies that dx is in fact the total
dierential dX of the position vector X
dx = dX . (2.48)
The no-torsion condition (2.44) is equivalent to the commutation of partial derivatives of X:
=

2
X
x
=
g
. (2.49)
The physical meaning of torsion is discussed further in 3.4.
2.6.9 Torsion tensor
General relativity assumes no torsion, but it is possible to consider generalizations to theories with torsion.
The torsion tensor S
is dened by the commutator of the covariant derivative acting on a scalar

[D
, D
] = S

Note that the covariant derivative of a scalar is just the ordinary derivative, D
= /x
. The expres-
sion (2.39) for the covariant derivatives shows that the torsion tensor is
S

which is evidently antisymmetric in the indices .
In Einstein-Cartan theory, the torsion tensor is related to the spin content of spacetime. Since this vanishes
in empty space, Einstein-Cartan theory is indistinguishable from general relativity in experiments carried
out in vacuum.
Exercise 2.2 Show that
D
=
A
+S
. (2.52)
Conclude that, if torsion vanishes as general relativity assumes, S
= 0, then
D
=
A
. (2.53)
2.6.10 Connection coecients in terms of the metric

The connection coecients have been dened, equation (2.34), as derivatives of the tangent basis vectors g
.
However, the connection coecients can be expressed purely in terms of the (rst derivatives of the) metric,
without reference to the individual basis vectors. The partial derivatives of the metric are
g
=
g
= g
+g
= g
+g
= g
+g
, (2.54)
which is a sum of two connection coecients. Here
with all indices lowered is dened to be
with
the rst index lowered by the metric,
. (2.55)
Combining the metric derivatives in the following fashion yields an expression for a single connection:
g
+
g
= 2
= 2
, (2.56)
the last line of which follows from the no-torsion condition S
= 0. Thus
=
1
2
_
g
+
g
_
is not a coordinate tensor . (2.57)
This is the formula that allows connection coecients to be calculated from the metric.
2.7 Coordinate 4-velocity 51
2.6.11 Mathematical aside
General relativity is a metric theory. Many of the structures introduced above can be dened mathematically
without a metric. For example, it is possible to dene the tangent space of vectors with basis g
, and to
dene a dual vector space with basis g
such that g
. Elements of the dual vector space are

commonly called one-forms. Similarly it is possible to dene connections and covariant derivatives without
a metric. However, this book follows general relativity in assuming that spacetime has a metric.
2.7 Coordinate 4-velocity
Consider a particle following a worldline
x
() , (2.58)
where is the particles proper time. The proper time along any interval of the worldline is d
ds
2
.
Dene the coordinate 4-velocity u
by
u
dx
d
is a coordinate 4-vector . (2.59)
The magnitude squared of the 4-velocity is constant
u
= g
dx
d
dx
d
=
ds
2
d
2
= 1 . (2.60)
The negative sign arises from the choice of metric signature: with the signature +++ adopted here, there
is a sign between ds
2
and d
2
. Equation (2.60) can be regarded as an integral of motion associated with
conservation of particle rest mass.
2.8 Geodesic equation
Let u g
be the 4-velocity in coordinate-independent notation. The principle of equivalence implies

that the geodesic equation, the equation of motion of a freely-falling particle, is
du
d
= 0 . (2.61)
Why? Because du/d = 0 in the particles own free-fall frame, and the equation is coordinate-independent.
In the particles own free-fall frame, the particles 4-velocity is u
= 1, 0, 0, 0, and the particles locally

inertial axes g
= g
0
, g
1
, g
2
, g
3
are constant.
What does the equation of motion look like in coordinate notation? The acceleration is
du
d
=
dx
d
u
x
= u
= u
_
u
_
= g
_
du
d
+
_
. (2.62)
The geodesic equation is then
du
d
+
= 0 . (2.63)
Another way of writing the geodesic equation is
Du
D
= 0 , (2.64)
where D/D is the covariant proper time derivative
D
D
u
. (2.65)
2.9 Coordinate 4-momentum
The coordinate 4-momentum of a particle of rest mass m is dened to be
p
mu
= m
dx
d
is a coordinate 4-vector . (2.66)
The momentum squared is
p
= m
2
u
= m
2
(2.67)
minus the square of the rest mass. Again, the minus sign arises from the choice +++ of metric signature.
2.10 Ane parameter
For photons, the rest mass is zero, m = 0, but the 4-momentum p
remains nite. Dene the ane

parameter by

m
is a coordinate scalar (2.68)
2.11 Ane distance 53
which remains nite in the limit m 0. The ane parameter is unique up to an overall linear transfor-
mation (that is, + is also an ane parameter, for constant and ), because of the freedom in the
choice of mass m and the zero point of proper time . In terms of the ane parameter, the 4-momentum is
p
=
dx
d
. (2.69)
The geodesic equation is then in coordinate-independent notation
dp
d
= 0 , (2.70)
or in component form
dp
d
+
= 0 , (2.71)
which works for massless as well as massive particles.
Another way of writing this is
Dp
D
= 0 , (2.72)
where D/D is the covariant ane derivative
D
D
p
. (2.73)
2.11 Ane distance
The ane parameter is also called the ane distance, because it provides a measure of distance along null
geodesics. When you look at a scene with your eyes, you are looking along null geodesics, and the natural
measure of distance to objects that you see is the ane distance. The freedom in the overall scaling of the
ane distance is xed by setting it equal to the proper distance near the observer in the observers locally
inertial rest frame.
In special relativity, the ane distance coincides with the perceived (e.g. binocular) distance to objects.
2.12 Riemann curvature tensor
The Riemann curvature tensor R
is dened by the commutator of the covariant derivative acting

on a 4-vector
[D
, D
] A
= R

The expression (2.74) assumes vanishing torsion; the more general expression with non-zero torsion is (3.20).
The expression (2.39) for the covariant derivative yields the following formula for the Riemann tensor in
terms of connection coecients
R

This is the formula that allows the Riemann tensor to be calculated from the connection coecients.
In at (Minkowski) space, covariant derivatives reduce to partial derivatives, D
/x
, and
[D
, D
]
_

x
,

x
_
= 0 in at space (2.76)
so that R
= 0 in at space.
Comment: In quantum eld theories (QED, QCD), the commutator of the gauge-covariant derivative is
taken to be the eld. In conventional general relativity, by contrast, the metric is taken to be the fundamental
eld, rather than the curvature. Another dierence between quantum eld theories and general relativity is
that the Lagrangian of quantum eld theories is taken to be quadratic in the eld, whereas the Lagrangian
of general relativity is taken to be linear in the curvature (specically, the general relativity Lagrangian is
the Ricci scalar R).
2.13 Symmetries of the Riemann tensor
In a locally inertial frame, the connection coecients all vanish,
= 0, but their partial derivatives,

which are proportional to second derivatives of the metric tensor, do not vanish. Thus in a locally inertial
frame the Riemann tensor is
R
=
1
2
_

2
g
+

2
g

2
g

2
g

2
g
+

2
g
_
=
1
2
_

2
g

2
g

2
g
+

2
g
_
. (2.77)
You can check that the bottom line of equation (2.77):
1. is antisymmetric in ,
2. is antisymmetric in ,
3. is symmetric in ,
4. has the property that the sum of the cyclic permutations of the last three indices vanishes
R
+R
+R
= 0 . (2.78)
The rst three of these four symmetries can be summarized by the shorthand notation
R
([][])
(2.79)
2.14 Ricci tensor, Ricci scalar 55
in which [ ] denotes anti-symmetrization and ( ) symmetrization. These symmetries imply that the Riemann
tensor is a symmetric matrix of antisymmetric matrices. An antisymmetric matrix has 6 degrees of freedom.
A symmetric matrix of these things is a 6 6 symmetric matrix, which has 21 degrees of freedom. The nal,
cyclic symmetry of the Riemann tensor, equation (2.78), removes 1 degree of freedom. Thus the Riemann
tensor has a net 20 degrees of freedom.
Although the above symmetries were derived in a locally inertial frame, the fact that the Riemann tensor
is a tensor means that the symmetries hold in any frame. If you prefer, you can add back the products of
connection coecients in equation (2.75), and check that the claimed symmetries remain.
2.14 Ricci tensor, Ricci scalar
The Ricci tensor R
and Ricci scalar R are the essentially unique contractions of the Riemann curvature
tensor. The Ricci tensor, the compressive part of the Riemann tensor, is
R

The symmetries of the Riemann tensor imply that the Ricci tensor is symmetric
R
= R
(2.81)
and therefore has 10 independent components.
The Ricci scalar is
R g
is a coordinate tensor (a scalar) . (2.82)

2.15 Einstein tensor
The Einstein tensor G
is dened by
G
1
2
g
R is a coordinate tensor . (2.83)

The symmetry of the Ricci and metric tensors imply that the Einstein tensor is likewise symmetric
G
= G
. (2.84)
The Einstein tensor has 10 independent components.
2.16 Bianchi identities
The Jacobi identity
[D
, [D
, D
]] + [D
, [D
, D
]] + [D
, [D
, D
]] = 0 (2.85)
implies the Bianchi identities
D
+ D
+D
= 0 (2.86)
which can be written in shorthand
D
[
R
]
= 0 . (2.87)
The Bianchi identities constitute a set of dierential relations between the components of the Riemann
tensor, which are distinct from the algebraic symmetries of the Riemann tensor.
There are 20 independent Bianchi identities. If just the symmetries (2.79) of the Riemann tensor are
taken into account, then there are 24 identities; but the cyclic symmetry (2.78) eliminates 4, leaving 20
independent identities.
2.17 Covariant conservation of the Einstein tensor
The most important consequence of the Bianchi identities (2.87) is obtained from the double contraction
g
(D
+D
+D
) = D
+D
R = 0 (2.88)
which implies that
D
= 0 . (2.89)
This equation is a primary motivation for the form of the Einstein equations, since it implies energy-
momentum conservation, equation (2.91).
2.18 Einstein equations
Einsteins equations are
G
= 8GT
is a coordinate tensor equation . (2.90)

What motivates the form of Einsteins equations?
1. The equation is generally covariant;
2. The Bianchi identities guarantee conservation of energy-momentum;
3. The Einstein tensor depends on the lowest (second) order derivatives of the metric tensor that do not
vanish in a locally inertial frame;
The covariant conservation of the Einstein tensor, equation (2.89), implies the conservation of energy-
momentum
D
= 0 . (2.91)
Einsteins equations (2.90) constitute a complete set of gravitational equations, generalizing Poissons
2.19 Summary of the path from metric to the energy-momentum tensor 57
equation of Newtonian gravity. However, Einsteins equations by themselves do not constitute a closed set
of equations: in general, other equations, such as Maxwells equations of electromagnetism, and equations
describing the microphysics of the energy-momentum, must be adjoined to form a closed set.
2.19 Summary of the path from metric to the energy-momentum tensor
1. Start by dening the metric g
.
2. Compute the connection coecients
from equation (2.57).

3. Compute the Riemann tensor R

4. Compute the Ricci tensor R
from equation (2.80), the Ricci scalar R from equation (2.82), and the
Einstein tensor G

5. The Einstein equations (2.90) then imply the energy-momentum tensor T
.
The path from metric to energy-momentum tensor is straightforward to program on a computer, but the
results are typically messy and complicated, even for fairly simple spacetimes. Inverting the path to recover
the metric from a given energy-momentum content is typically highly non-trivial, the subject of a huge
literature.
The great majority of metrics g
yield an energy-momentum tensor T
that cannot be achieved with

normal matter.
2.20 Energy-momentum tensor of an ideal uid
The simplest non-trival energy-momentum tensor is that of an ideal uid. In this case T
is isotropic in
the locally inertial rest frame of the uid, taking the form
T
=
_
_
_
_
0 0 0
0 p 0 0
0 0 p 0
0 0 0 p
_
_
_
_
(2.92)
where
is the proper mass-energy density ,
p is the proper pressure .
(2.93)
The expression (2.92) is valid only in the locally inertial rest frame of the uid. An expression that is valid
in any frame is
T
= ( +p)u
+p g
, (2.94)
where u
is the 4-velocity of the uid. Equation (2.94) is valid because it is a tensor equation, and it is true
in the locally inertial rest frame, where u
= 1, 0, 0, 0.
2.21 Newtonian limit
The Newtonian limit is obtained in the limit of a weak gravitational eld and non-relativistic (pressureless)
matter. In Cartesian coordinates, the metric in the Newtonian limit is
ds
2
= (1 + 2)dt
2
+ (1 2)(dx
2
+dy
2
+dz
2
) , (2.95)
in which
(x, y, z) = Newtonian potential (2.96)
is a function only of the spatial coordinates x, y, z, not of time t.
For this metric, to rst order in the potential the only non-vanishing component of the Einstein tensor
is the time-time component
G
tt
= 2
2
, (2.97)
where
2
=
2
/x
2
+
2
/y
2
+
2
/z
2
is the usual 3D Laplacian operator. This component (2.97) of the
Einstein tensor plugged into Einsteins equations (2.90) implies Poissons equation (2.4).
3
More on the coordinate approach

3.1 Weyl tensor
The trace-free, or tidal, part of the Riemann curvature tensor denes the Weyl tensor C

1
2
(g
+g
) +
1
6
(g
) R is a coordinate tensor .
(3.1)
The Weyl tensor has 10 independent components. These 10 components together with the 10 components
of the Ricci tensor account for the 20 components of the Riemann tensor. The Weyl tensor inherits the
symmetries (2.79) of the Riemann tensor
C
([][])
. (3.2)
Whereas the Einstein tensor G
, necessarily vanishes in a region of spacetime where there is no energy-

momentum, T
= 0, the Weyl tensor does not. The Weyl tensor expresses the presence of tidal gravitational
forces, and of gravitational waves.
3.2 Evolution equations for the Weyl tensor
This section is included because (a) the comparison to Maxwells equations is neat and insightful, (b) it helps
to account for the degrees of freedom of the gravitational eld, (c) it shows how the Weyl tensor encodes
gravitational waves.
Contracted on one index, the Bianchi identities (2.86) are
D
[
R
]
= 0 . (3.3)
There are 20 such independent contracted identities. Since this is the same as the number of independent
Bianchi identities, it follows that the contracted Bianchi identities (3.3) are equivalent to the full set of
Bianchi identities (2.87). If the Riemann tensor is separated into its trace (Ricci) and traceless (Weyl) parts,
60

equation (3.1), then the contracted Bianchi identities (3.3) become the Weyl evolution equations
D
= J
, (3.4)
where J
is the Weyl current

J

1
2
(D
)
1
6
(g
Gg
G) . (3.5)
The Weyl evolution equations (3.4) can be regarded as the gravitational analogue of Maxwells equations of
electromagnetism.
The Weyl current J
is a vector of bivectors, which would suggest that it has 4 6 = 24 components,

but it loses 4 of those components because of the cyclic identity (2.78), which implies the cyclic symmetry
J
[]
= 0 . (3.6)
Thus J
has 20 independent components, in agreement with the above assertion that there are 20 inde-
pendent contracted Bianchi identities. Since the Weyl tensor is traceless, contracting the Weyl evolution
equations (3.4) on yields zero on the left hand side, so that the contracted Weyl current satises
J
= 0 . (3.7)
This doubly-contracted Bianchi identity, which is the same as equation (2.89), enforces conservation of
energy-momentum. Unlike the cyclic symmetries (3.6), which are automatically satised, equations (3.7)
constitute a non-trivial set of 4 conditions on the Einstein tensor. Besides the algebraic relations (3.6) and
(3.7), the Weyl current satises 6 dierential identities comprising the conservation law
D
= 0 (3.8)
in view of equation (3.4) and the antisymmetry of C
with respect to the indices . The Weyl current

conservation law (3.8) follows automatically from the form (3.5) of the Weyl current, coupled with energy-
momentum conservation (2.89), so does not impose any additional non-trivial conditions on the Riemann
tensor.
The 4 relations (3.7) and the 6 identities (3.8) account for 10 of the 20 contracted Bianchi identities (3.3).
The remaining 10 equations comprise Maxwell-like equations (3.4) for the evolution of the 10 components
of the Weyl tensor.
Whereas the Einstein equations relating the Einstein tensor to the energy-momentum tensor are postulated
equations of general relativity, the 10 evolution equations for the Weyl tensor, and the 4 equations enforcing
covariant conservation of the Einstein tensor, follow mathematically from the Bianchi identities, and do not
represent additional assumptions of the theory.
Exercise 3.1 Conrm the counting of degrees of freedom.
Exercise 3.2 From the Bianchi identities, show that the Riemann tensor satises the covariant wave
equation
R
= D
+D
, (3.9)
3.3 Geodesic deviation 61
where is the DAlembertian operator, the 4-dimensional wave operator
D
. (3.10)
Show that contracting equation (3.9) with g
yields the identity R
= R
. Conclude that the wave

equation (3.9) is non-trivial only for the trace-free part of the Riemann tensor, the Weyl tensor C
. Show
that the wave equation for the Weyl tensor is
C
= (D
1
2
g
)R
(D
1
2
g
)R
+(D

1
2
g
)R
(D
1
2
g
)R
+
1
6
(g
)R . (3.11)
Conclude that in a vacuum, where R
= 0,
C
= 0 . (3.12)
3.3 Geodesic deviation

This section on geodesic deviation is included not because the equation of geodesic deviation is crucial to
everyday calculations in general relativity, but rather for two reasons. First, the equation oers insight into
the physical meaning of the Riemann tensor. Second, the derivation of the equation oers a ne illustration
of the fact that in general relativity, whenever you take dierences at innitesimally separated points in
space or time, you should always take covariant dierences.
Consider two objects that are free-falling along two innitesimally separated geodesics. In at space the
acceleration between the two objects would be zero, but in curved space the curvature induces a nite
acceleration between the two objects. This is how an observer can measure curvature, at least in principle:
set up an ensemble of objects initially at rest a small distance away from the observer in the observers
locally inertial frame, and watch how the objects begin to move. The equation (3.18) that describes this
acceleration between objects an innitesimal distance apart is called the equation of geodesic deviation.
The covariant dierence in the velocities of two objects an innitesimal distance x
apart is
Dx
D
= u
. (3.13)
In general relativity, the ordinary dierence between vectors at two points a small interval apart is not
a physically meaningful thing, because the frames of reference at the two points are dierent. The only
physically meaningful dierence is the covariant dierence, which is the dierence in the two vectors parallel-
transported across the gap between them. It is only this covariant dierence that is independent of the frame
of reference. On the left hand side of equation (3.13), the proper time derivative must be the covariant proper
time derivative, D/D = u
. On the right hand side of equation (3.13), the dierence in the 4-velocity
62

at two points x
apart must be the covariant dierence = x
. Thus equation (3.13) means explicitly

the covariant equation
u
= x
. (3.14)
To derive the equation of geodesic deviation, rst vary the geodesic equation Du
/D = 0 (Ive put the

index downstairs so that the nal equation (3.18) looks cosmetically better, but of course since everything
is covariant the index could just as well be put upstairs everywhere):
0 =
Du
D
= x
_
u
_
= u
+x
. (3.15)
On the second line, the covariant dience between quantities a small distance x
apart has been set equal

to x
, while D/D has been set equal to the covariant time derivative u
along the geodesic. On the

last line, x
has been replaced by u
. Next, consider the covariant acceleration of the interval x
,
which is the covariant proper time derivative of the covariant velocity dierence u
:
D
2
x
D
2
=
Du
D
= u
(x
)
= u
+x
. (3.16)
As in the previous equation (3.15), on the second line D/D has been set equal to u
, while has been

set equal to x
. On the last line, u
has been set equal to u
, equation (3.14). Subtracting (3.15)

from (3.16) gives
D
2
x
D
2
= x
[D
, D
]u
, (3.17)
or equivalently
D
2
x
D
2
+R
= 0 , (3.18)
which is the desired equation of geodesic deviation.
3.4 Commutator of the covariant derivative revisited
The commutator of the covariant derivative is of fundamental importance because it denes what is meant
by the eld in gauge theories. It was seen above that the commutator of the covariant derivative acting
on a scalar dened the torsion tensor, equation (2.50), which general relativity assumes vanishes, while the
commutator of the covariant derivative acting on a vector dened the Riemann tensor, equation (2.74). Does
the commutator of the covariant derivative acting on a general tensor introduce any other distinct tensor? No:
3.4 Commutator of the covariant derivative revisited 63
the torsion and Riemann tensors completely dene the action of the commutator of the covariant derivative
on any tensor, equation (3.22).
The general expression (3.22) for the commutator of the covariant derivative reveals the meaning of the
torsion and Riemann tensors. The torsion and Riemann tensors describe respectively the displacement and
the Lorentz transformation experienced by an object when parallel-transported around a curve. Displacement
and Lorentz transformations together constitute the Poincare group, the complete group of symmetries of
at spacetime.
How can an object detect a displacement when parallel-transported around a curve? If you go around
a curve back to the same coordinate in spacetime where you began, wont you necessarily be at the same
position? This is a question that goes to heart of the meaning of spacetime. To answer the question, you
have to consider how fundamental particles are able to detect position, orientation, and velocity. Classically,
particles may be structureless points, but quantum mechanically, particles possess frequency, wavelength,
spin, and (in the relativistic theory) boost, and presumably it is these properties that allow particles to
measure the properties of the spacetime in which they live. Specically, a Dirac spinor (relativistic spin-
1
2
particle) has 8 degrees of freedom, of which 6 dene a Lorentz transformation (a Lorentz rotor, a member of
the group of spin-
1
2
Lorentz transformations), and the remaining 2 comprise a complex number REALLY?
THE COMPLEX NUMBER IS WITH RESPECT TO THE PSEUDOSCALAR I e
ipx
whose phase
encodes the displacement. Thus a Dirac spinor could potentially detect a displacement through a change in
its phase when parallel-transported around a curve back to the same point in spacetime. General relativity,
which assumes that torsion vanishes, asserts that there is no such change of phase.
In the presence of torsion, the expression for the connection coecients
is, from equation (2.57),
=
1
2
_
g
+
g
+S
+S
+S
_
. (3.19)
The rst part
1
2
(g
,
+g
,
g
,
) of this expression is called the Christoel symbol of the rst kind [the
same thing with the rst index raised,
1
2
g
(g
,
+g
,
g
,
), is called the Christoel symbol of the
second kind], while the second part
1
2
(S
+S
+S
) is called the contortion (not contorsion!) tensor.

Theres no need to remember the crazy jargon, but in case you meet it, thats what it means.
If torsion does not vanish, then the commutator of the covariant derivative acting on a contravariant
4-vector is
[D
, D
] A
= S
+R

where the Riemann tensor R
is given in terms of the connection coecients by the same formula (2.75)

as before, but the connection coecients
themselves are given by (3.19). The Riemann tensor is still

antisymmetric in each of and , but with torsion it is no longer symmetric in . In other
words, the symmetries of the Riemann tensor with torsion are
R
[][]
. (3.21)
As a matrix of antisymmetric tensors, the Riemann tensor with torsion has 6 6 = 36 degrees of freedom.
64

Because the Riemann tensor R
is no longer symmetric in , the Ricci tensor R
is
no longer symmetric, and likewise the Einstein tensor G
1
2
Rg
is no longer symmetric. Evidently

the antisymmetric part of the Einstein tensor depends on torsion.
Acting on a general tensor, the commutator of the covariant derivative is
[D
, D
] A
...
...
= S
A
...
...
+R
A
...
...
+R
A
...
...
R
A
...
...
R
A
...
...
. (3.22)
In more abstract notation, the commutator of the covariant derivative is the operator
[D
, D
] = S
D +R
(3.23)
where S
and D g
, and the Riemann curvature operator R
is an operator whose action

on any tensor is specied by equation (3.22). The action of the operator R
is analogous to that of the

covariant derivative (2.40): theres a positive R term for each covariant index, and a negative R term for each
contravariant index. The action of R
on a scalar is zero, which reects the fact that a scalar is unchanged

by a Lorentz transformation.
4
Action principle
Hamiltons principle of least action postulates that any dynamical system is characterized by a scalar action
S, which has the property that when the system evolves from one specied state to another, the path by
which it gets between the two states is such as to minimize the action. The action need not be a global
minimum, just a local minimum with respect to small variations in the path between xed initial and nal
states.
That nature appears to respect the principle of least action is of the profoundest signicance.
x
0
x
1
Figure 4.1 The action principle considers various paths through spacetime between xed initial and nal
conditions, and chooses that path that minimizes the action.
66

Action principle
4.1 Principle of least action for point particles
The path of a point particle through spacetime is specied by its coordinates x
() as a function of some
arbitrary parameter . In non-relativistic mechanics it is usual to take the parameter to be the time t,
and the path of a particle through space is then specied by three spatial coordinates x
i
(t). In relativity
however it is more natural to treat the time and space coordinates on an equal footing, and regard the path
of a particle as being specied by four spacetime coordinates x
() as a function of an arbitrary parameter

. The parameter is simply a continuous parameter that labels points along the path, and has no physical
signicance (for example, it is not necessarily an ane parameter).
The path of a system of N point particles through spacetime is specied by 4N coordinates x
(). The
action principle postulates that, for a system of N point particles, the action S is an integral of a Lagrangian
L(x
, dx
/d) which is a function of the 4N coordinates x
() together with the 4N velocities dx
/d with
respect to the arbitrary parameter . The action from an initial state at
i
to a nal state at
f
is thus
S =
_

f
i
L
_
x
,
dx
d
_
d . (4.1)
The principle of least action demands that the actual path taken by the system between given initial and
nal coordinates x
i
and x
f
is such as to minimize the action. Thus the variation S of the action must be
zero under any change x
in the path, subject to the constraint that the coordinates at the endpoints are
xed, x
i
= 0 and x
f
= 0,
S =
_

f
i
_
L
x
+
L
(dx
/d)
(dx
/d)
_
d = 0 . (4.2)
The change in the velocity along the path is just the velocity of the change, (dx
/d) = d(x
)/d.
Integrating the second term in the integrand of equation (4.2) by parts yields
S =
_
L
(dx
/d)
x
f
i
+
_

f
i
_
L
x

d
d
L
(dx
/d)
_
x
d = 0 . (4.3)
The surface term in equation (4.3) vanishes, since by hypothesis the coordinates are held xed at the end
points, so x
= 0 at the end points. Therefore the integral in equation (4.3) must vanish. Indeed least
action requires the integral to vanish for all possible variations x
in the path. The only way this can

happen is that the integrand must be identically zero. The result is the Euler-Lagrange equations of
motion
d
d
L
(dx
/d)

L
x
= 0 . (4.4)
It might seem that the Euler-Lagrange equations (4.4) are inadequately specied, since they depend on
some arbitrary unknown parameter . But in fact the Euler-Lagrange equations are the same regardless of
the choice of . An example of the irrelevance of will be seen in the next section, 4.2. Since can be
chosen arbitrarily, it is usual to choose it in some convenient fashion. For a massive particle, can be taken
4.2 Action for a test particle 67
equal to the proper time of the particle. For a massless particle, whose proper time never progresses,
can be taken equal to an ane parameter.
4.2 Action for a test particle
According to the principle of equivalence, a test particle in a gravitating system moves along a geodesic, a
straight line relative to local free-falling frames. A geodesic is the shortest distance between two points. In
relativity this translates, for a massive particle, into the longest proper time between two points. The proper
time along any path is d =
ds
2
=
_
g
dx
dx
. Thus the action S

m
of a test particle of rest mass m
in a gravitating system is
S
m
= m
_

f
i
d = m
_

f
i
_
g
dx
d
dx
d
d . (4.5)
The factor of rest mass m brings the action, which has units of angular momentum, to standard normalization.
The overall minus sign comes from the fact that the action is a minimum whereas the proper time is a
maximum along the path. The action principle requires that the Lagrangian be written as a function
of the coordinates x
and velocities dx
/d, and it is seen that the integrand in the last expression of

equation (4.5) has the desired form, the metric g
being considered a given function of the coordinates.

Thus the Lagrangian L
m
of a test particle of mass m is
L
m
= m
_
g
dx
d
dx
d
. (4.6)
The partial derivatives that go in the Euler-Lagrange equations (4.4) are then
L
m
(dx
/d)
= m
g
dx
d
_
g
(dx
/d)(dx
/d)
, (4.7a)
L
m
x
= m
1
2
g
dx
dx
d
dx
d
_
g
(dx
/d)(dx
/d)
. (4.7b)
The denominators in the expressions (4.7) for the partial derivatives of the Lagrangian are
_
g
(dx
/d)(dx
/d) = d/d. It was not legitimate to make this substitution before taking the partial
derivatives, since the Euler-Lagrange equations require that the Lagrangian be expressed in terms of x
and
dx
/d, but it is ne to make the substitution now that the partial derivatives have been obtained. The
partial derivatives (4.7) thus simplify to
L
m
(dx
/d)
= mg
dx
d
d
d
= mu
, (4.8)
L
m
x
=
1
2
m
g
dx
dx
d
dx
d
d
d
= m
d
d
, (4.9)
68

Action principle
in which u
dx
/d is the usual 4-velocity, and the derivative of the metric has replaced by connections
in accordance with equation (2.54). The resulting Euler-Lagrange equations of motion (4.4) are
dmu
d
= m
d
d
. (4.10)
As remarked in 4.1, the choice of the arbitrary parameter has no eect on the equations of motion.
With a factor of md/d cancelled, and with the derivative converted to a covariant derivative by (2.39),
equation (4.10) becomes
Du
D
= S
, (4.11)
where S
is the torsion, equation (2.51). If torsion vanishes, as general relativity assumes, then the result
is the usual equation of geodesic motion
Du
D
= 0 . (4.12)
The fact that motion is geodesic only if torsion vanishes is to be expected, since, as argued in 2.6.7, space
is locally inertial only if torsion vanishes.
4.3 Action for a charged test particle in an electromagnetic eld
Aim is to reproduce the Lorentz force law.
S = S
m
+S
q
(4.13)
where
S
q
= q
_

f
i
A
dx
= q
_

f
i
A
dx
d
d . (4.14)
The Lagrangian is therefore
L
q
= qA
dx
d
. (4.15)
Partial derivatives are
L
q
(dx
/d)
= qA
,
L
q
x
= q
A
dx
d
= q
A
d
d
. (4.16a)
Applied to the Lagrangian L = L
m
+L
q
, the Euler-Lagrange equations (4.4) are
d
d
(mu
+qA
) =
_
m
+q
A
_
d
d
. (4.17)
If torsion vanishes, as general relativity assumes, then the result is the Lorentz force law for a test particle
of mass m and charge q moving in a prescribed gravitational and electromagnetic eld
Dmu
D
= qF
. (4.18)
4.4 Generalized momentum 69
Exercise 4.1 Show that if torsion does not vanish, then the Lorentz force law becomes
Dmu
D
= [qF
+S
(mu
+qA
)] u
. (4.19)
[Hint: Recall the relations (2.51) and (2.52).]
4.4 Generalized momentum
The generalized momentum

L
(dx
/d)
. (4.20)
The generalized momentum
of a test particle coincides with the ordinary momentum p
= p
= mu
. (4.21)
The generalized momentum of a test particle of charge q in an electromagnetic eld of potential A
= p
+qA
. (4.22)
4.5 Hamiltonian
Work with coordinates and generalized momenta instead of coordinates and velocities. Dene Hamiltonian
H by
H
dx
d
L (4.23)
S =
_ _
dx
d
H
_
d (4.24)
S = [
] +
_ _
_
d
d
+
H
x
_
x
+
_
dx
d

H
_
d (4.25)
d
d
=
H
x
,
dx
d
=
H
. (4.26)
What is this useful for?
70

Action principle
4.6 Derivatives of the action
Besides being a scalar whose minimum value between xed endpoints denes the path between those points,
the action can also be treated as a function of its endpoints along the actual path. Along the actual path,
the equations of motion are satised, so the integral in the variation (4.3) of the action vanishes identically.
The surface term in the variation (4.3) then implies that S =
, so that the partial derivatives of the

action with respect to the coordinates are equal to the generalized momenta,
S
x
. (4.27)
This is the basis of the Hamilton-Jacobi method.
PART THREE
IDEAL BLACK HOLES
Concept Questions
1. What evidence do astronomers currently accept as indicating the presence of a black hole in a system?
2. Why can astronomers measure the masses of supermassive black holes only in relatively nearby galaxies?
3. To what extent (with what accuracy) are real black holes in our Universe described by the no-hair theorem?
4. Does the no-hair theorem apply inside a black hole?
5. Black holes lose their hair on a light-crossing time. How long is a light-crossing time for a typical stellar-
sized or supermassive astronomical black hole?
6. Relativists say that the metric is g
, but they also say that the metric is ds

2
= g
dx
dx
. How can
both statements be correct?
7. The Schwarzschild geometry is said to describe the geometry of spacetime outside the surface of the Sun
or Earth. But the Schwarzschild geometry supposedly describes non-rotating masses, whereas the Sun
and Earth are rotating. If the Sun or Earth collapsed to a black hole conserving their mass M and angular
momentum L, roughly what would the spin a/M = L/M
2
of the black hole be relative to the maximal
spin a/M = 1 of a Kerr black hole?
8. What happens at the horizon of a black hole?
9. As cold matter becomes denser, it goes through the stages of being solid/liquid like a planet, then electron
degenerate like a white dwarf, then neutron degenerate like a neutron star, then nally it collapses to a
black hole. Why could there not be a denser state of matter, denser than a neutron star, that brings a
star to rest inside its horizon?
10. How can an observer determine whether they are at rest in the Schwarzschild geometry?
11. An observer outside the horizon of a black hole never sees anything pass through the horizon, even to the
end of the Universe. Does the black hole then ever actually collapse, if no one ever sees it do so?
12. If nothing can ever get out of a black hole, how does its gravity get out?
13. Why did Einstein believe that black holes could not exist in nature?
14. In what sense is a rotating black hole stationary, but not static?
15. What is a white hole? Do they exist?
16. Could the expanding Universe be a white hole?
17. Could the Universe be the interior of a black hole?
18. You know the Schwarzschild metric for a black hole. What is the corresponding metric for a white hole?
19. What is the best kind of black hole to fall into if you want to avoid being tidally torn apart?
20. Why do astronomers often assume that the inner edge of an accretion disk around a black hole occurs at
the innermost stable orbit?
21. A collapsing star of uniform density has the geometry of a collapsing Friedmann-Robertson-Walker cos-
mology. If a spatially at FRW cosmology corresponds to a star that starts from zero velocity at innity,
then to what do open or closed FRW cosmologies correspond?
22. Is the singularity of a Reissner-Nordstr om black hole gravitationally attractive or repulsive?
23. If you are a charged particle, which dominates near the singularity of the Reissner-Nordstr om geometry,
the electrical attraction/repulsion or the gravitational attraction/repulsion?
24. Is a white hole gravitationally attractive or repulsive?
25. What happens if you fall into a white hole?
26. Which way does time go in Parallel Universes in the Reissner-Nordstr om geometry?
27. What does it mean that geodesics inside a black hole can have negative energy?
28. Can geodesics have negative energy outside a black hole? How about inside the ergosphere?
29. Physically, what causes mass ination?
30. Is mass ination likely to occur inside real astronomical black holes?
31. What happens at the X point, where the ingoing and outgoing inner horizons of the Reissner-Nordstr om
geometry intersect?
32. Can a particle like an electron or proton, whose charge far exceeds its mass (in geometric units), be
modeled as Reissner-Nordstr om black hole?
33. Does it makes sense that a person might be at rest in the Kerr-Newman geometry? How would the
Boyer-Linquist coordinates of such a person vary along their worldline?
34. In identifying M as the mass and a the angular momentum per unit mass of the black hole in the Boyer-
Linquist metric, why is it sucient to consider the behaviour of the metric at r ?
35. Does space move faster than light inside the ergosphere?
36. If space moves faster than light inside the ergosphere, why is the outer boundary of the ergosphere not a
horizon?
37. Do closed timelike curves make sense?
38. What does Carters fourth integral of motion Q signify physically?
39. What is special about a principal null congruence?
40. Evaluated in the locally inertial frame of a principal null congruence, the spin-0 component of the Weyl
scalar of the Kerr geometry is C = M/(r ia cos )
3
, which looks like the Weyl scalar C = M/r
3
of the
Schwarzschild geometry but with radius r replaced by the complex radius r ia cos . Is there something
deep here? Can the Kerr geometry be constructed from the Schwarzschild geometry by complexifying the
radial coordinate r?
Whats important?
1. Astronomical evidence suggests that stellar-sized and supermassive black holes exist ubiquitously in nature.
2. The no-hair theorem, and when and why it applies.
3. The physical picture of black holes as regions of spacetime where space is falling faster than light.
4. A physical understanding of how the metric of a black hole relates to its physical properties.
5. Penrose (conformal) diagrams. In particular, the Penrose diagrams of the various kinds of vacuum black
hole: Schwarzschild, Reissner-Nordstr om, Kerr-Newman.
6. What really happens inside black holes. Collapse of a star. Mass ination instability.
5
Observational Evidence for Black Holes
It is beyond the scope of this course to discuss the observational evidence for black holes in any detail.
However, it is useful to summarize a few facts.
1. Observational evidence supports the idea that black holes occur ubiquitously in nature. They are not
observed directly, but reveal themselves through their eects on their surroundings. Two kinds of black
hole are observed: stellar-sized black holes in x-ray binary systems, mostly in our own Milky Way galaxy,
and supermassive black holes in Active Galactic Nuclei (AGN) found at the centers of our own and other
galaxies.
2. The primary evidence that astronomers accept as indicating the presence of a black hole is a lot of mass
compacted into a tiny space.
a. In an x-ray binary system, if the mass of the compact object exceeds 3 M
, the maximum theoretical

mass of a neutron star, then the object is considered to be a black hole. Many hundreds of x-ray binary
systems are known in our Milky Way galaxy, but only 10s of these have measured masses, and in about
20 the measured mass indicates a black hole.
b. Several tens of thousands of AGN have been cataloged, identied either in the radio, optical, or x-rays.
But only in nearby galaxies can the mass of a supermassive black hole be measured directly. This is
because it is only in nearby galaxies that the velocities of gas or stars can be measured suciently close
to the nuclear center to distinguish a regime where the velocity becomes constant, so that the mass
can be attribute to an unresolved central point as opposed to a continuous distribution of stars. The
masses of about 40 supermassive black holes have been measured in this way. The masses range from
the 4 10
6
M
mass of the black hole at the center of the Milky Way to the 3 10
9
M
mass of the
black hole at the center of the M87 galaxy at the center of the Virgo cluster at the center of the Local
Supercluster of galaxies.
3. Secondary evidences for the presence of a black hole are:
a. high luminosity;
b. non-stellar spectrum, extending from radio to gamma-rays;
c. rapid variability.
d. relativistic jets.
Observational Evidence for Black Holes 77
Jets in AGN are often one-sided, and a few that are bright enough to be resolved at high angular resolution
show superluminal motion. Both evidences indicate that jets are commonly relativistic, moving at close
to the speed of light. There are a few cases of jets in x-ray binary systems.
4. Stellar-sized black holes are thought to be created in supernovae as the result of the core-collapse of
stars more massive than about 25 M
(this number depends in part on uncertain computer simulations).

Supermassive black holes are probably created initially in the same way, but they then grow by accretion of
gas funnelled to the center of the galaxy. The growth rates inferred from AGN luminosities are consistent
with this picture.
5. Long gamma-ray bursts (lasting more than about 2 seconds) are associated observationally with super-
novae. It is thought that in such bursts we are seeing the formation of a black hole. As the black hole gulps
down the huge quantity of material needed to make it, it regurgitates a relativistic jet that punches through
the envelope of the star. If the jet happens to be pointed in our direction, then we see it relativistically
beamed as a gamma-ray burst.
6. Astronomical black holes present the only realistic prospect for testing general relativity in the strong eld
regime, since such elds cannot be reproduced in the laboratory. At the present time the observational
tests of general relativity from astronomical black holes are at best tentative. One test is the redshifting
of 7 keV iron lines in a small number of AGN, notably MCG-6-30-15, which can be interpreted as being
emitted by matter falling on to a rotating (Kerr) black hole.
7. At present, no gravitational waves have been denitely detected from anything. In the future, gravitational
wave astronomy should eventually detect the merger of two black holes. If the waveforms of merging black
holes are consistent with the predictions of general relativity, it will provide a far more stringent test of
strong eld general relativity than has been possible to date.
8. Although gravitational waves have yet to be detected directly, their existence has been inferred from the
gradual speeding up of the orbit of the Hulse-Taylor binary, which consists of two neutron stars, one of
which, PSR1913+16, is a pulsar. The parameters of the orbit have been measured with exquisite precision,
and the rate of orbital speed-up is in good agreement with the energy loss by quadrupole gravitational
wave emission predicted by general relativity.
6
Ideal Black Holes
6.1 Denition of a black hole
What is a black hole? Doubtless you have heard the standard denition many times: It is a region whose
gravity is so strong that not even light can escape.
But why can light not escape from a black hole? A standard answer, which John Michell (1784, Phil.
Trans. Roy. Soc. London 74, 35) would have found familiar, is that the escape velocity exceeds the speed of
light. But that answer brings to mind a Newtonian picture of light going up, turning around, and coming
back down, that is altogether dierent from what general relativity actually predicts.
A better denition of a black hole is that it is a
region where space is falling faster than light.
Inside the horizon, light emitted outwards is carried inward by the faster-than-light inow of space, like a
sh trying but failing to swim up a waterfall.
The denition may seem jarring. If space has no substance, how can it fall faster than light? It means
that inside the horizon any locally inertial frame is compelled to fall to smaller radius as its proper time goes
by. This fundamental fact is true regardless of the choice of coordinates.
A similar concept of space moving arises in cosmology. Astronomers observe that the Universe is expand-
ing. Cosmologists nd it convenient to conceptualize the expansion by saying that space itself is expanding.
For example, the picture that space expands makes it more straightforward, both conceptually and math-
ematically, to deal with regions of spacetime beyond the horizon, the surface of innite redshift, of an
observer.
6.2 Ideal black hole
The simplest kind of black hole, an ideal black hole, is one that is stationary, electrovac outside its singularity,
and extends to asymptotically at empty space at innity. Electrovac means that the energy-momentum
tensor T
is zero except for the contribution from a stationary electromagnetic eld.

6.3 No-hair theorem 79
The next several chapters deal with ideal black holes. The importance of ideal black holes stems from
the no-hair theorem, discussed in the next section. The no-hair theorem has the consequence that, except
during their initial collapse, or during a merger, real astronomical black holes are accurately described as
ideal outside their horizons.
6.3 No-hair theorem
I will state and justify the no-hair theorem, but I will not prove it mathematically, since the proof is technical.
The no-hair theorem states that a stationary black hole in asymptotically at space is characterized by
just three quantities:
1. Mass M;
2. Electric charge Q;
3. Spin, usually parameterized by the angular momentum a per unit mass.
The mechanism by which a black hole loses its hair is gravitational radiation. When initially formed,
whether from the collapse of a massive star or from the merger of two black holes, a black hole will form
a complicated, oscillating region of spacetime. But over the course of several light crossing times, the
oscillations lose energy by gravitational radiation, and damp out, leaving a stationary black hole.
Real astronomical black holes are not isolated, and continue to accrete (cosmic microwave background
photons, if nothing else). However, the timescale (a light crossing time) for oscillations to damp out by
gravitational radiation is usually far shorter than the timescale for accretion, so in practice real black holes
are extremely well described by no-hair solutions almost all of their lives.
The physical reason that the no-hair theorem applies is that space is falling faster than light inside the
horizon. Consequently, unlike a star, no energy can bubble up from below to replace the energy lost by
gravitational radiation, so that the black hole tends to the lowest energy state characterized by conserved
quantities.
As a corollary, the no-hair theorem does not apply from the inner horizon of a black hole inward, because
there space ceases to fall superluminally.
If there exist other absolutely conserved quantities, such as magnetic charge (magnetic monopoles), or
various supersymmetric charges in theories where supersymmetry is not broken, then the black hole will also
be characterized by those quantities.
Black holes are expected not to conserve quantities such as baryon or lepton number that are thought not
to be absolutely conserved, even though they appear to be conserved in low energy physics.
Other stationary solutions exist that describe black holes in spacetimes that are not asymptotically at,
such as spacetimes with a cosmological constant, or with a uniform electromagnetic eld.
It is legitimate to think of the process of reaching a stationary state as analogous to reaching a condition
of thermodynamic equilibrium, in which a macroscopic system is described by a small number of parameters
associated with the conserved quantities of the system.
7
Schwarzschild Black Hole
The Schwarzschild geometry was discovered by Karl Schwarzschild in late 1915 at essentially the same time
that Einstein was arriving at his nal version of the General Theory of Relativity.
7.1 Schwarzschild metric
The Schwarzschild metric is, in a polar coordinate system t, r, , , and in geometric units c = G = 1,
ds
2
=
_
1
2M
r
_
dt
2
+
_
1
2M
r
_
1
dr
2
+r
2
do
2
, (7.1)
where do
2
(this is the Landau & Lifshitz notation) is the metric of a unit 2-sphere
do
2
= d
2
+ sin
2
d
2
. (7.2)
The Schwarzschild geometry describes the simplest kind of black hole: a black hole with mass M, but no
electric charge, and no spin.
The geometry describes not only a black hole, but also any empty space surrounding a spherically sym-
metric mass. Thus the Schwarzschild geometry describes to a good approximation the spacetime outside the
surfaces of the Sun and the Earth.
Comparison with the spherically symmetric Newtonian metric
ds
2
= (1 + 2)dt
2
+ (1 2)(dr
2
+r
2
do
2
) (7.3)
with Newtonian potential
(r) =
M
r
(7.4)
establishes that the M in the Schwarzschild metric is to be interpreted as the mass of the black hole.
The Schwarzschild geometry is asymptotically at, because the metric tends to the Minkowski metric in
7.2 Birkhos theorem 81
polar coordinates at large radius
ds
2
dt
2
+dr
2
+r
2
do
2
as r . (7.5)
Exercise 7.1 The Schwarschild metric (7.1) does not have the same form as the spherically symmetric
Newtonian metric (7.3). By a suitable transformation of the radial coordinate r, bring the Schwarschild
metric (7.1) to the isotropic form
ds
2
=
_
1 M/2R
1 +M/2R
_
2
dt
2
+ (1 +M/2R)
4
(dR
2
+R
2
do
2
) . (7.6)
What is the relation between R and r? Hence conclude that the identication (7.4) is correct, and therefore
that M is indeed the mass of the black hole. Is the isotropic form (7.6) of the Schwarzschild metric valid
inside the horizon?
7.2 Birkhos theorem
Birkhos theorem states that the geometry of empty space surrounding a spherically symmetric matter
distribution is the Schwarzschild geometry. That is, if the metric is of the form
ds
2
= A(t, r) dt
2
+B(t, r) dt dr +C(t, r) dr
2
+D(t, r) do
2
, (7.7)
where the metric coecients A, B, C, and D are allowed to be arbitrary functions of t and r, and if the
energy momentum tensor vanishes, T
= 0, outside some value of the circumferential radius r
dened by
r
2
= D, then the geometry is necessarily Schwarzschild outside that radius.
This means that if a mass undergoes spherically symmetric pulsations, then those pulsations do not aect
the geometry of the surrounding spacetime. This reects the fact that there are no spherically symmetric
gravitational waves.
7.3 Stationary, static
The Schwarzschild geometry is stationary. A spacetime is said to be stationary if and only if there exists
a timelike coordinate t such that the metric is independent of t. In other words, the spacetime possesses
time translation symmetry: the metric is unchanged by a time translation t t + t
0
where t
0
is some
constant. Evidently the Schwarzschild metric (7.1) is independent of the timelike coordinate t, and is
therefore stationary, time translation symmetric.
The Schwarzschild geometry is also static. A spacetime is static if and only if the coordinates can be
chosen so that, in addition to being stationary with respect to a time coordinate t, the spatial coordinates
82 Schwarzschild Black Hole
do not change along the direction of the tangent vector g
t
. This requires that the tangent vector g
t
be
orthogonal to all the spatial tangent vectors
g
t
g
= g
t
= 0 for ,= t . (7.8)
The Gullstrand-Painleve metric for the Schwarzschild geometry, discussed in section 7.13, is an example of a
metric that is stationary but not static (although the underlying spacetime, being Schwarzschild, is static).
The Gullstrand-Painleve metric is independent of the free-fall time t
, so is stationary, but observers who

follow the tangent vector g
t
ff
fall into the black hole, so the metric is not manifestly static.
The Schwarzschild time coordinate t is thus identied as a special one: it is the unique time coordinate
with respect to which the Schwarzschild geometry is manifestly static.
7.4 Spherically symmetric
The Schwarzschild geometry is also spherically symmetric. This is evident from the fact that the angular
part r
2
do
2
of the metric is the metric of a 2-sphere of radius r. This can be see as follows. Consider the
metric of ordinary at 3-dimensional Euclidean space in Cartesian coordinates x, y, z:
ds
2
= dx
2
+dy
2
+dz
2
. (7.9)
Convert to polar coordinates r, , , dened so that
x = r sin cos ,
y = r sin sin ,
z = r cos .
(7.10)
Substituting equations (7.10) into the Euclidean metric (7.9) gives
ds
2
= dr
2
+r
2
(d
2
+ sin
2
d
2
) . (7.11)
Restricting to a surface r = constant of constant radius then gives the metric of a 2-sphere of radius r
ds
2
= r
2
(d
2
+ sin
2
d
2
) (7.12)
as claimed.
The radius r in Schwarzschild coordinates is the circumferential radius, dened such that the proper
circumference of the 2-sphere measured by observers at rest in Schwarschild coordinates is 2r. This is a
coordinate-invariant denition of the meaning of r, which implies that r is a scalar.
7.5 Horizon 83
7.5 Horizon
The horizon of the Schwarzschild geometry lies at the Schwarzschild radius r = r
s
r
s
=
2GM
c
2
. (7.13)
where units of c and G have been restored. Where does this come from? The Schwarzschild metric shows
that the scalar spacetime distance squared ds
2
along an interval at rest in Schwarzschild coordinates, dr =
d = d = 0, is timelike, lightlike, or spacelike depending on whether the radius is greater than, equal to, or
less than r = 2M:
ds
2
=
_
1
2M
r
_
dt
2
_
_
_
< 0 if r > 2M ,
= 0 if r = 2M ,
> 0 if r < 2M .
(7.14)
Since the worldline of a massive observer must be timelike, it follows that a massive observer can remain at
rest only outside the horizon, r > 2M. An object at rest at the horizon, r = 2M, follows a null geodesic,
which is to say it is a possible worldline of a massless particle, a photon. Inside the horizon, r < 2M, neither
massive nor massless objects can remain at rest.
A full treatment of what is going on requires solving the geodesic equation in the Schwarzschild geometry,
but the results may be anticipated already at this point. In eect, space is falling into the black hole. Outside
the horizon, space is falling less than the speed of light; at the horizon space is falling at the speed of light;
and inside the horizon, space is falling faster than light, carrying everything with it. This is why light cannot
escape from a black hole: inside the horizon, space falls inward faster than light, carrying light inward even if
that light is pointed radially outward. The statement that space is falling superluminally inside the horizon
of a black hole is a coordinate-invariant statement: massive or massless particles are carried inward whatever
their state of motion and whatever the coordinate system.
Whereas an interval of coordinate time t switches from timelike outside the horizon to spacelike inside the
horizon, an interval of coordinate radius r does the opposite: it switches from spacelike to timelike:
ds
2
=
_
1
2M
r
_
1
dr
2
_
_
_
> 0 if r > 2M ,
= 0 if r = 2M ,
< 0 if r < 2M .
(7.15)
It appears then that the Schwarzschild time and radial coordinates swap roles inside the horizon. Inside the
horizon, the radial coordinate becomes timelike, meaning that it becomes a possible worldline of a massive
observer. That is, a trajectory at xed t and decreasing r is a possible wordline. Again this reects the fact
that space is falling faster than light inside the horizon. A person inside the horizon is inevitably compelled
as time goes by to move to smaller radial coordinate r.
7.6 Proper time
The proper time experienced by an observer at rest in Schwarzschild coordinates, dr = d = d = 0, is
d =
_
ds
2
=
_
1
2M
r
_
1/2
dt . (7.16)
For an observer at rest at innity, r , the proper time is the same as the coordinate time,
d dt as r . (7.17)
Among other things, this implies that the Schwarzschild time coordinate t is a scalar: not only is it the
unique coordinate with respect to which the metric is manifestly static, but it coincides with the proper time
of observers at rest at innity. This coordinate-invariant denition of time t implies that it is a scalar.
At nite radii outside the horizon, r > 2M, the proper time d is less than the Schwarzchild time dt, so
the clocks of observers at rest run slower at smaller than at larger radii.
At the horizon, r = 2M, the proper time d of an observer at rest goes to zero,
d 0 as r 2M . (7.18)
This reects the fact that an object at rest at the horizon is following a null geodesic, and as such experiences
zero proper time.
7.7 Redshift
An observer at rest at innity looking through a telesope at an emitter at rest at radius r sees the emitter
redshifted by a factor
1 +z

obs
emit
=

emit
obs
=
d
obs
d
emit
=
_
1
2M
r
_
1/2
. (7.19)
This is an example of the universally valid statement that photons are good clocks: the redshift factor is
given by the rate at which the emitters clock appears to tick relative to the observers own clock.
It should be emphasized that the redshift factor (7.19) is valid only for an observer and an emitter at rest
in the Schwarzschild geometry. If the observer and emitter are not at rest, then additional special relativistic
factors will fold into the redshift.
The redshift goes to innity for an emitter at the horizon
1 +z as r 2M . (7.20)
Here the redshift tends to innity regardless of the motion of the observer or emitter. An observer watching
an emitter fall through the horizon will see the emitter appear to freeze at the horizon, becoming ever slower
and more redshifted. Physically, photons emitted vertically upward at the horizon by an emitter falling
through it remain at the horizon for ever, taking an innite time to get out to the outside observer.
7.8 Proper distance 85
7.8 Proper distance
The proper radial distance measure by observers at rest in Schwarzschild coordinates, dr = d = d = 0, is
dl =
ds
2
=
_
1
2M
r
_
1/2
dr . (7.21)
For an observer at rest at innity, r , an interval of proper radial distance equals an interval of
circumferential radial distance, as you might expect for asymptotically at space
dl dr as r . (7.22)
At the horizon, r = 2M, a proper radial interval dl measured by an observer at rest goes to innity
dl as r 2M . (7.23)
7.9 Schwarzschild singularity
The apparent singularity in the Schwarzschild metric at the horizon r = 2M is not a real singularity, because
it can be removed by a change of coordinates, such as to Gullstrand-Painleve coordinates (7.26). Prior to
as late as the 1950s, people, including Einstein, thought that the Schwarzschild singularity at r = 2M
marked the physical boundary of the Schwarzschild spacetime. After all, an outside observer watching stu
fall in never sees anything beyond that boundary.
Schwarzschilds choice of coordinates was certainly a natural one. It was natural to search for static
solutions, and his time coordinate t is the only one with respect to which the metric is manifestly static.
The problem is that physically there can be no static observers inside the horizon: they must necessarily fall
inward as time passes. The fact that Schwarzschilds coordinate system shows an apparent singularity at the
horizon reects the fact that the assumption of a static spacetime necessarily breaks down at the horizon,
where space is falling at the speed of light.
Does stu actually fall in, even though no outside observer ever sees it happen? Classically, the answer
is yes: when a black hole forms, it does actually collapse, and when an observer falls through the horizon,
they really do fall through the horizon. The reason that an outside observer sees everything freeze at the
horizon is simply a light travel time eect: it takes an innite time for light to lift o the horizon and make
it to the outside world.
7.10 Embedding diagram
An embedding diagram is a visual aid to understanding geometry. It is a depiction of a lower dimensional
geometry in a higher dimension. A classic example is the illustration of the geometry of a 2-sphere embedded
in 3-dimensional space. The 2-sphere has a meaning independent of any embedding in 3 dimensions because
the geometry of the 2-sphere can be measured by 2-dimensional inhabitants of its surface without reference
to any encompassing 3-dimensional space. Nevertheless, the pictorial representation aids imagination.
Textbooks sometimes illustrate the Schwarzschild geometry with an embedding diagram that shows the
spatial geometry at a xed instant of Schwarzschild time t. The diagram illustrates the stretching of proper
distances in the radial direction. Ill let you gure out how to construct this embedding diagram.
It should be emphasized that the embedding diagram of the Schwarzshild geometry at xed Schwarzschild
time t has a limited physical meaning. Fixing the time t means choosing a certain hypersurface through the
geometry. Other choices of hypersurface will yield dierent diagrams. For example, the Gullstrand-Painleve
metric is spatially at at xed free-fall time t
, so in that case the embedding diagram would simply illustrate

at space, with no funny business at the horizon.
7.11 Energy-momentum tensor
The energy-momentum tensor of the Schwarzschild geometry is zero, by construction.
7.12 Weyl tensor
It turns out that the 10 components of the Weyl tensor, the tidal part of the Riemann tensor, can be decom-
posed in any locally inertial frame into 5 complex components of spin 0, 1, and 2. In the Schwarzschild
metric, all components vanish except the real spin 0 component. This component is a coordinate-invariant
scalar, the Weyl scalar C
C =
M
r
3
. (7.24)
The Weyl scalar, which expresses the presence of tidal forces, goes to innity at zero radius,
C as r 0 , (7.25)
signalling the presence of a real singularity at zero radius.
7.13 Gullstrand-Painleve coordinates
The Gullstrand-Painleve metric is an alternative metric for the Schwarzschild geometry, discovered indepen-
dently by Allvar Gullstrand and Paul Painleve in (1921). When we have done tetrads, we will recognize
that the standard way in which metrics are written encodes not only metric but also a complete tetrad. The
Gullstrand-Painleve line-element (7.26) encodes a tetrad that represents locally inertial frames free-falling
radially into the black hole at the Newtonian escape velocity. Unlike Schwarzschild coordinates, there is no
singularity at the horizon in Gullstrand-Painleve coordinates. It is striking that the mathematics was known
long before physical understanding emerged.
7.14 Eddington-Finkelstein coordinates 87
The Gullstrand-Painleve metric is
ds
2
= dt
2
+ (dr dt
)
2
+r
2
do
2
. (7.26)
Here is the Newtonian escape velocity (with a minus sign because space is falling inward)
=
_
2M
r
_
1/2
(7.27)
and t
is the proper time experienced by an object that free falls radially inward from zero velocity at innity.
The free fall time t
is related to the Schwarzschild time coordinate t by

dt
= dt

1
2
dr , (7.28)
which integrates to
t
= t + 2M
_
2
_
r
2M
_
1/2
+ ln
(r/2M)
1/2
1
(r/2M)
1/2
+ 1
_
. (7.29)
The time axis g
t
ff
in Gullstrand-Painleve coordinates is not orthogonal to the radial axis g
r
, but rather is
tilted along the radial axis, g
t
ff
g
r
= g
t
ff
r
= .
The proper time of a person at rest in Gullstrand-Painleve coordinates, dr = d = d = 0, is
d = dt
_
1
2
. (7.30)
The horizon occurs where this proper time vanishes, which happens when the infall velocity is the speed
of light
[[ = 1 . (7.31)
According to equation (7.27), this happens at r = 2M, which is the Schwarzschild radius, as it should be.
7.14 Eddington-Finkelstein coordinates
In Schwarzschild coordinates, radially infalling or outfalling light rays appear never to cross the horizon
of the Schwarzschild black hole. This feature of Schwarzschild coordinates contributed to the historical
misconception that black holes stopped at their horizons. In 1958, David Finkelstein carried out a trivial
transformation of the time coordinate which seeemed to show that infalling light rays could indeed pass
through the horizon. It turned out that Eddington had already discovered the transformation in 1924,
though at that time the physical implications were not grasped. Again, it is striking that the mathematics
was in place long before physical understanding.
In Schwarzschild coordinates, light rays that fall radially (d = d = 0) inward or outward follow null
geodesics
ds
2
=
_
1
2M
r
_
dt
2
+
_
1
2M
r
_
1
dr
2
= 0 . (7.32)
Radial null geodesics thus follow
dr
dt
=
_
1
2M
r
_
(7.33)
in which the sign is + for outfalling, for infalling rays. Equation (7.33) shows that dr/dt 0 as
r 2M, suggesting that null rays, whether infalling or outfalling, never cross the horizon. The solution to
equation (7.33) is
t = (r + 2M ln[r 2M[) , (7.34)
which shows that Schwarzschild time t approaches logarithmically as null rays approach the horizon.
Finkelstein dened his time coordinate t
F
by
t
F
t + 2M ln [r 2M[ , (7.35)
which has the property that infalling null rays follow
t
F
+r = 0 . (7.36)
In other words, on a spacetime diagram in Finkelstein coordinates, radially infalling light rays move at 45
,
the same as in special relativistic spacetime diagrams.
7.15 Kruskal-Szekeres coordinates
Since Finkelstein transformed coordinates so that radially infalling light rays moved at 45
in a spacetime
diagram, it is natural to look for coordinates in which outfalling as well as infalling light rays are at 45
.
Kruskal and Szekeres independently provided such a transformation, in 1960.
Dene the tortoise (or Regge-Wheeler 1959) coordinate r
by
r

_
dr
1 2M/r
= r + 2M ln [r 2M[ . (7.37)
Then radially infalling and outfalling null rays follow
r
+t = 0 infalling ,
r
t = 0 outfalling .
(7.38)
In a spacetime diagram in coordinates t and r
, infalling and outfalling light rays are indeed at 45
. Unfor-
tunately the metric in these coordinates is still singular at the horizon r = 2M:
ds
2
=
_
1
2M
r
_
_
dt
2
+dr
2
_
+r
2
do
2
. (7.39)
The singularity at the horizon can be eliminated by the following transformation into Kruskal-Szekeres
coordinates t
K
and r
K
:
r
K
+t
K
= e
(r
+t)/2
,
r
K
t
K
= e
(r
t)/2
,
(7.40)
7.16 Penrose diagrams 89
r
=
r
=
H
o
r
i
z
o
n
A
n
t
i
h
o
r
i
z
o
n
Universe
Singularity (r = 0)
Black Hole
T
i
m
e
Space
L
i
g
h
t
L
i
g
h
t
Figure 7.1 Penrose diagram of the Schwarzschild geometry.
where the sign in the last equation is + outside the horizon, inside the horizon. The Kruskal-Szekeres
metric is
ds
2
= r
1
e
r
_
dt
2
K
+dr
2
K
_
+r
2
do
2
, (7.41)
which is non-singular at the horizon. The Schwarzschild radial coordinate r, which appears in the factors
r
1
e
r
and r
2
in the Kruskal metric, is to be understood as an implicit function of the Kruskal coordinates
t
K
and r
K
.
7.16 Penrose diagrams
Roger Penrose, as so often, had a novel take on the business of spacetime diagrams. Penrose conceived that
the primary purpose of a spacetime diagram should be to portray the causal structure of the spacetime, and
that the specic choice of coordinates was largely irrelevant. After all, general relativity allows arbitrary
choices of coordinates.
In addition to requiring that light rays be at 45
, Penrose wanted to bring regions at innity (in time or

space) to a nite position on the spacetime diagram, so that the entire spacetime could be seen at once. He
calls these thing conformal diagrams, but the rest of us commonly call them Penrose diagrams.
Penrose diagrams are bona-de spacetime diagrams. For example, a coordinate transformation from
Kruskal to Penrose coordinates (the following transformation is not analytic, but Penrose does not care)
r
P
+t
P
=
r
K
+t
K
2 +[r
K
+t
K
[
,
r
P
t
P
=
r
K
t
K
2 [r
K
t
K
[
,
(7.42)
brings spatial and temporal innity to nite values of the coordinates, while keeping infalling and outfalling
light rays at 45
in the spacetime diagram. However, there are many such transformations, and Penrose
would be the last person to advocate any one of them in particular.
r = 0
r
=
r
=
r
=
r
=
H
o
r
i
z
o
n
Universe
Black Hole
r = 0
P
a
r
a
l
l
e
l
H
o
r
i
z
o
n
A
n
t
i
h
o
r
i
z
o
n
P
a
r
a
l
l
e
l
A
n
t
i
h
o
r
i
z
o
n
Parallel Universe
White Hole
Figure 7.2 Penrose diagram of the complete, analytically extended Schwarzschild geometry.
7.17 Schwarzschild white hole, wormhole
The Kruskal-Szekeres spacetime diagram reveals a new feature that was not apparent in Schwarzschild or
Finkelstein coordinates. Dredged from the depths of t = appears a null line r
K
+t
K
= 0. The null line
is at radius r = 2M, but it does not correspond to the horizon that a person might fall into. The null line is
called the antihorizon. The horizon is sometimes called the true horizon, and the antihorizon the illusory
horizon. In a real black hole, only the true horizon is real. The antihorizon is replaced by an exponentially
dimming and redshifting image of the star that collapsed to form the black hole.
The Kruskal-Szekeres (= Schwarzschild) geometry is analytic, and there is a unique analytic continuation
of the geometry through the antihorizon. The analytic continuation is a time-reversed copy of the original
Schwarzschild geometry, glued at the antihorizon. Whereas the original Schwarzschild geometry showed an
asymptotically at region and a black hole region separated by a horizon, the complete analytically extended
Schwarzschild geometry shows two asymptotically at regions, together with a black hole and a white hole.
Relativists label the regions I, II, III, and IV, but I like to call them by name: Universe, Black Hole,
Parallel Universe, and White Hole.
The white hole is a time-reversed version of the black hole. Whereas space falls inward faster than light
inside the black hole, space falls outward faster than light inside the white hole. In the Gullstrand-Painleve
metric (7.26), the velocity = (2M/r)
1/2
is negative for the black hole, positive for the white hole.
The Kruskal or Penrose diagrams show that the universe and the parallel universe are connected, but
7.18 Collapse to a black hole 91
only by spacelike lines. This spacelike connection is called the Einstein-Rosen bridge, and constitutes a
wormhole connecting the two universes. Because the connection is spacelike, it is impossible for a traveler
to pass through this wormhole.
Although two travelers, one from the universe and one from the parallel universe, cannot travel to each
others universe, they can meet, but only inside the black hole. Inside the black hole, they can talk to each
other, and they can see light from each others universe. Sadly, the enlightenment is only temporary, because
they are doomed soon to hit the central singularity.
It should be emphasized that the white hole and the wormhole in the Schwarzschild geometry are a
mathematical construction with as far as anyone knows no relevance to reality. Nevertheless it is intriguing
that such bizarre objects emerge already in the simplest general relativistic solution for a black hole.
7.18 Collapse to a black hole
Realistic collapse of a star to a black hole is not expected to produce a white hole or parallel universe.
The simplest model of a collapsing star is a spherical ball of uniform density and zero pressure which free
falls from zero velocity at innity. In this simple model, the interior of the star is described by a collapsing
Friedmann-Robertson-Walker metric (the canonical cosmological metric), while the exterior is described by
the Schwarzschild solution. The assumption that the star collapses from zero velocity at innity implies
that the FRW metric is spatially at, the simplest case. To continue the geometry between Schwarzchild
and FRW metrics, it is neatest to use the Gullstrand-Painleve metric, with the Gullstrand-Painleve infall
velocity at the edge of the star set equal to minus r times the Hubble parameter rH r d ln a/dt of
the collapsing FRW metric.
The simple model shows that the antihorizon of the complete Schwarzschild geometry is replaced by the
surface of the collapsing star, and that beyond the antihorizon is not a parallel universe and a white hole,
but merely the interior of the star (and the distant Universe glimpsed through the stars interior).
Since light can escape from the collapsing star system as long as it is even slightly larger than its Schwarz-
schild radius, it is possible to take the view that the horizon comes instantaneously into being at the moment
the star collapses through its Schwarzschild radius. This denition of the horizon is called the apparent hori-
zon.
Hawking has advocated that a better denition of the horizon is to take it to be the boundary between
outgoing null rays that fall into the black hole versus those that go to innity. In any evolving situation,
this denition of the horizon, which is called the absolute horizon, depends formally on what happens in the
innite future, though in slowly evolving systems the absolute horizon can be located with some precision
without knowing the future. The absolute horizon of the collapsing star forms before the star has collapsed,
and grows to meet the apparent horizon as the star falls through its Schwarzchild radius.
In this simple model, the central singularity forms slightly before the star has collapsed to zero radius.
The formation of the singularity is marked by the fact that light rays emitted at zero radius cease to be able
to move outward. In other words, the singularity forms when space starts to fall into it faster than light.
7.19 Killing vectors
The Schwarzschild metric presents an opportunity to introduce the concept of Killing vectors (after Wil-
helm Killing, not because the vectors kill things, though the latter is true), which are associated with
symmetries of the spacetime.
7.20 Time translation symmetry
The time translation invariance of the Schwarzschild geometry is evident from the fact that the metric is
independent of the time coordinate t. Equivalently, the partial time derivative /t of the Schwarzschild
metric is zero. The associated Killing vector
is then dened by
=

t
(7.43)
so that in Schwarzschild coordinates t, r, ,
= 1, 0, 0, 0 . (7.44)
In coordinate-independent notation, the Killing vector is
= g
= g
t
. (7.45)
This may seem like overkill couldnt we just say that the metric is independent of time t and be done
with it? The answer is that symmetries are not always evident from the metric, as will be seen in the next
section 7.21.
Because the Killing vector g
t
is the unique timelike Killing vector of the Schwarzschild geometry, it has
a denite meaning independent of the coordinate system. It follows that its scalar product with itself is a
coordinate-independent scalar
= g
t
g
t
= g
tt
=
_
1
2M
r
_
. (7.46)
In curved spacetimes, it is hugely important to be able to identify scalars, which have a physical meaning
independent of the choice of coordinates.
7.21 Spherical symmetry
The rotational symmetry of the Schwarzschild metric about the azimuthal axis is evident from the fact that
the metric is independent of the azimuthal coordinate . The associated Killing vector is
g
(7.47)
with components 0, 0, 0, 1 in Schwarzschild coordinates t, r, , .
7.22 Killing equation 93
The Schwarzschild metric is fully spherically symmetric, not just azimuthally symmetric. Since the 3D
rotation group O(3) is 3-dimensional, it is to be expected that there are three Killing vectors. You may
recognize from quantum mechanics that / is (modulo factors of i and ) the z-component of the angular
momentum operator L = L
x
, L
y
, L
z
in a coordinate system where the azimuthal axis is the z-axis. The 3
components of the angular momentum operator are given by:
iL
x
= y

z
z

y
= sin

cot cos

,
iL
y
= z

x
x

z
= cos

cot sin

,
iL
z
= x

y
y

x
=

.
(7.48)
The 3 rotational Killing vectors are correspondingly:
rotation about x-axis: sing
cot cos g
,
rotation about y-axis: cos g
cot sing
,
rotation about z-axis: g
.
(7.49)
You can check that the action of the x and y rotational Killing vectors on the metric does not kill the
metric. For example, iL
x
g
= 2r
2
cos sin cos does not vanish. This example shows that a more powerful
and general condition, described in the next section 7.22, is needed to establish whether a quantity is or is
not a Killing vector.
Because spherical symmetry does not dene a unique azimuthal axis g
, its scalar product with itself

g
= g
= r
2
sin
2
is not a coordinate-invariant scalar. However, the sum of the scalar products of
the 3 rotational Killing vectors is rotationally invariant, and is therefore a coordinate-invariant scalar
( sing
cot cos g
)
2
+ (cos g
cot sing
)
2
+g
2
= g
+ (cot
2
+ 1)g
= 2r
2
. (7.50)
This shows that the circumferential radius r is a scalar, as you would expect.
7.22 Killing equation
As seen in the previous section, a Killing vector does not always kill the metric in a given coordinate system.
This is not really surprising given the arbitrariness of coordinates in GR. What is true is that a quantity is
a Killing vector if and only if there exists a coordinate system such that the Killing vector kills the metric
in that system.
Suppose that in some coordinate system the metric is independent of the coordinate . In problem set 2
you showed that in such a case the covariant component u
of the 4-velocity along a geodesic is constant

u
= constant . (7.51)
Equivalently
= constant (7.52)
where
is the associated Killing vector, whose only non-zero component is
= 1 in this particular
coordinate system. The converse is also true: if
= constant along all geodesics, then
is a Killing
vector. The constancy of
along all geodesics is equivalent to the condition that its proper time derivative
vanish along all geodesics
d
d
= 0 . (7.53)
But this is equivalent to
0 = u
) = u
=
1
2
u
(D
+D
) (7.54)
where the second equality follows from the geodesic equation, u
= 0, and the last equality is true

because of the symmetry of u
in . A necessary and sucient condition for equation (7.54) to be

true for all geodesics is that
D
+D
= 0 (7.55)
which is Killings equation. This equation is the desired necessary and sucient condition for
to be a
Killing vector. It is a generally covariant equation, valid in any coordinate system.
8
Reissner-Nordstrom Black Hole
The Reissner-Nordstr om geometry, discovered independently by Hans Reissner in 1916, Hermann Weyl in
1917, and Gunnar Nordstr om in 1918, describes the unique spherically symmetric static solution for a black
hole with mass and electric charge in asymptotically at spacetime.
8.1 Reissner-Nordstrom metric
The Reissner-Nordstr om metric for a black hole of mass M and electric charge Q is, in geometric units
c = G = 1,
ds
2
=
_
1
2M
r
+
Q
2
r
2
_
dt
2
+
_
1
2M
r
+
Q
2
r
2
_
1
dr
2
+r
2
do
2
(8.1)
which looks like the Schwarzschild metric with the replacement
M M(r) = M
Q
2
2r
. (8.2)
In fact equation (8.2) has a coordinate independent interpretation as the mass M(r) interior to radius r,
which here is the mass M at innity, minus the mass in the electric eld E = Q/r
2
outside r
_

r
E
2
8
4r
2
dr =
_

r
Q
2
8r
4
4r
2
dr =
Q
2
2r
. (8.3)
This seems like a Newtonian calculation of the energy in the electric eld, but it turns out to be valid also
in general relativity.
Real astronomical black holes probably have very little electric charge, because the Universe as a whole
appears almost electrically neutral (and Maxwells equations in fact require that the Universe in its entirety
should be exactly electrically neutral), and a charged black hole would quickly neutralize itself. It would
probably not neutralize itself completely, but have some small residual positive charge, because protons
(positive charge) are more massive than electrons (negative charge), so it is slightly easier for a black hole
to accrete protons than electrons.
96 Reissner-Nordstr om Black Hole
Nevertheless, the Reissner-Nordstr om solution is of more than passing interest because its internal geom-
etry resembles that of the Kerr solution for a rotating black hole.
Concept question 8.1 What is the charge Q in standard (gaussian) units?
The Einstein tensor of the Reissner-Nordstr om metric (8.1) is diagonal, with elements given by
G
=
_
_
_
_
_
G
t
t
0 0 0
0 G
r
r
0 0
0 0 G
0
0 0 0 G
_
_
_
_
_
= 8
_
_
_
_
0 0 0
0 p
r
0 0
0 0 p
0
0 0 0 p
_
_
_
_
=
Q
2
r
4
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
. (8.4)
The trick of writing one index up and the other down on the Einstein tensor G
partially cancels the

distorting eect of the metric, yielding the proper energy density , the proper radial pressure p
r
, and
transverse pressure p
, up to factors of 1. A more systematic way to extract proper quantities is to work

in the tetrad formalism, but this will do for now.
The energy-momentum tensor is that of a radial electric eld
E =
Q
r
2
. (8.5)
Notice that the radial pressure p
r
is negative, while the transverse pressure p
is positive. It is no coincidence
that the sum of the energy density and pressures is twice the energy density, +p
r
+ 2p
= 2.
The negative pressure, or tension, of the radial electric eld produces a gravitational repulsion that domi-
nates at small radii, and that is responsible for much of the strange phenomenology of the Reissner-Nordstr om
geometry. The gravitational repulsion mimics the centrifugal repulsion inside a rotating black hole, for which
reason the Reissner-Nordstr om geometry is often used a surrogate for the rotating Kerr-Newman geometry.
At this point, the statements that the energy-momentum tensor is that of a radial electric eld, and that
the radial tension produces a gravitational repulsion that dominates at small radii, are true but unjustied
assertions.
8.3 Weyl tensor
As with the Schwarzschild geometry (indeed, any spherically symmetric geometry), only 1 of the 10 inde-
pendent spin components of the Weyl tensor is non-vanishing, the real spin-0 component, the Weyl scalar
8.4 Horizons 97
C. The Weyl scalar for the Reissner-Nordstr om geometry is
C =
M
r
3
+
Q
2
r
4
. (8.6)
The Weyl scalar goes to innity at zero radius
C as r 0 (8.7)
signalling the presence of a real singularity at zero radius.
8.4 Horizons
The Reissner-Nordstr om geometry has not one but two horizons. The horizons occur where an object at
rest in the geometry, dr = d = d = 0, follows a null geodesic, ds
2
= 0, which occurs where
1
2M
r
+
Q
2
r
2
= 0 . (8.8)
This is a quadratic equation in r, and it has two solutions, an outer horizon r
+
and an inner horizon r
= M
_
M
2
Q
2
. (8.9)
It is straightforward to check that the Reissner-Nordstr om time coordinate t is timelike outside the outer
horizon, r > r
+
, spacelike between the horizons r
< r < r
+
, and again timelike inside the inner horizon
r < r
. Conversely, the radial coordinate r is spacelike outside the outer horizon, r > r
+
, timelike between
the horizons r
< r < r
+
, and spacelike inside the inner horizon r < r
.
The physical meaning of this strange behaviour is akin to that of the Schwarzschild geometry. As in the
Schwarzschild geometry, outside the outer horizon space is falling at less than the speed of light; at the outer
horizon space hits the speed of light; and inside the outer horizon space is falling faster than light. But
a new ingredient appears. The gravitational repulsion caused by the negative pressure of the electric eld
slows down the ow of space, so that it slows back down to the speed of light at the inner horizon. Inside
the inner horizon space is falling at less than the speed of light.
8.5 Gullstrand-Painleve metric
Deeper insight into the Reissner-Nordstr om geometry comes from examining its Gullstrand-Painleve metric.
The Gullstrand-Painleve metric for the Reissner-Nordstr om geometry is the same as that for the Schwarz-
schild geometry
ds
2
= dt
2
+ (dr dt
)
2
+r
2
do
2
. (8.10)
The velocity is again the escape velocity, but this is now
=
_
2M(r)
r
(8.11)
where M(r) = M Q
2
/2r is the interior mass already given as equation (8.2). Horizons occur where the
magnitude of the velocity equals the speed of light
[[ = 1 (8.12)
which happens at the outer and inner horizons r = r
+
and r = r
, equation (8.9).
The Gullstrand-Painleve metric once again paints the picture of space falling into the black hole. Outside
the outer horizon r
+
space falls at less than the speed of light, at the horizon space falls at the speed of
light, and inside the horizon space falls faster than light. But the gravitational repulsion produced by the
tension of the radial electric eld starts to slow down the inow of space, so that the infall velocity reaches
a maximum at r = Q
2
/M. The infall slows back down to the speed of light at the inner horizon r
. Inside
the inner horizon, the ow of space slows all the way to zero velocity, = 0, at the turnaround radius
r
0
=
Q
2
2M
. (8.13)
Space then turns around, the velocity becoming positive, and accelerates back up to the speed of light.
Space is now accelerating outward, to larger radii r. The outfall velocity reaches the speed of light at the
inner horizon r
, but now the motion is outward, not inward. Passing back out through the inner horizon,
space is falling outward faster than light. This is not the black hole, but an altogether new piece of spacetime,
a white hole. The white hole looks like a time-reversed black hole. As space falls outward, the gravitational
repulsion produced by the tension of the radial electric eld declines, and the outow slows. The outow
slows back to the speed of light at the outer horizon r
+
of the white hole. Outside the outer horizon of the
white hole is a new universe, where once again space is owing at less than the speed of light.
What happens inward of the turnaround radius r
0
, equation (8.13)? Inside this radius the interior mass
M(r), equation (8.2), is negative, and the velocity is imaginary. The interior mass M(r) diverges to negative
innity towards the central singularity at r 0. The singularity is timelike, and innitely gravitationally
repulsive, unlike the central singularity of the Schwarzschild geometry. Is it physically realistic to have a
singularity that has innite negative mass and is innitely gravitationally repulsive? Undoubtedly not.
8.6 Complete Reissner-Nordstrom geometry
As with the Schwarzschild geometry, it is possible to go through the steps: Reissner-Nordstr omcoordinates
Eddington-Finkelstein coordinates Kruskal-Szekeres coordinates Penrose coordinates. The conclusion
of these constructions is that the Reissner-Nordstr om geometry can be analytically continued, and the
complete analytic continuation consists of an innite ladder of universes and parallel universes connected to
each other by black hole wormhole white hole tunnels. I like to call the various pieces of spacetime
8.6 Complete Reissner-Nordstr om geometry 99
r
=
r
=
r
=
r
=
r
=
r
=
r
=
r
=
P
a
r
a
l
l
e
l
A
n
t
i
h
o
r
i
z
o
n
A
n
t
i
h
o
r
i
z
o
n
Parallel Universe Universe
P
a
r
a
l
l
e
l
H
o
r
i
z
o
n
H
o
r
i
z
o
n
Black Hole
I
n
n
e
r
H
o
r
i
z
o
n
P
a
r
a
l
l
e
l
I
n
n
e
r
H
o
r
i
z
o
n
W
o
r
m
h
o
l
e
P
a
r
a
l
l
e
l
W
o
r
m
h
o
l
e
Antiverse
Parallel
Antiverse
White Hole
I
n
n
e
r
A
n
t
i
h
o
r
i
z
o
n
P
a
r
a
l
l
e
l
I
n
n
e
r
A
n
t
i
h
o
r
i
z
o
n
New Parallel Universe New Universe
Figure 8.1 Penrose diagram of the complete Reissner-Nordstrom geometry.
Universe, Parallel Universe, Black Hole, Wormhole, Parallel Wormhole, and White Hole. These
pieces repeat in an innite ladder.
The Wormhole and Parallel Wormhole contain separate central singularities, the Singularity and the
Parallel Singularity, which are oppositely charged. If the black hole is positively charged as measured by
observers in the Universe, then it is negatively charged as measured by observers in the Parallel Universe,
and the Wormhole contains a positive charge singularity while the Parallel Wormhole contains a negative
charge singularity.
Where does the electric charge of the Reissner-Nordstr om geometry actually reside? This comes down
to the question of how observers detect the presence of charge. Observers detect charge by the electric eld
that it produces. Equip all (radially moving) observers with a gyroscope that they orient consistently in
the same radial direction, which can be taken to be towards the black hole as measured by observers in
the Universe. Observers in the Parallel Universe nd that their gyroscope is pointed away from the black
hole. Inside the black hole, observers from either Universe agree that the gyroscope is pointed towards the
Wormhole, and away from the Parallel Wormhole. All observes agree that the electric eld is pointed in the
same radial direction. Observers who end up inside the Wormhole measure an electric eld that appears to
emanate from the Singularity, and which they therefore attribute to charge in the Singularity. Observers
who end up inside the Parallel Wormhole measure an electric eld that appears to emanate in the opposite
direction from the Parallel Singularity, and which they therefore attribute to charge of opposite sign in the
Parallel Singularity. Strange, but all consistent.
8.7 Antiverse: Reissner-Nordstrom geometry with negative mass
It is also possible to consider the Reissner-Nordstr om geometry for negative values of the radius r. I call the
extension to negative r the Antiverse. There is also a Parallel Antiverse.
Changing the sign of r in the Reissner-Nordstr om metric (8.1) is equivalent to changing the sign of the
mass M. Thus the Reissner-Nordstr om metric with negative r describes a charged black hole of negative
mass
M < 0 . (8.14)
The negative mass black hole is gravitationally repulsive at all radii, and it has no horizons.
8.8 Ingoing, outgoing
The black hole in the Reissner-Nordstr om geometry is bounded at its inner edge by not one but two inner
horizons. The two distinct horizons play a crucial role in the mass ination instability described in 8.9
below.
The inner horizons can be called ingoing and outgoing. Persons freely falling in the Black Hole region are
8.9 Mass ination instability 101
all moving inward in coordinate radius r, but they may be moving either forward or backward in Reissner-
Nordstr om coordinate time t. In the Black Hole region, the conserved energy along a geodesic is positive if
the time coordinate t is decreasing, negative if the time coordinate t is increasing
1
. Persons with positive
energy are ingoing, while persons with negative energy are outgoing. Both ingoing and outgoing persons
fall inward, to smaller radii, but ingoing persons think that the inward direction is towards the Wormhole,
while outgoing persons think that the inward direction is in the opposite direction, towards the Parallel
Wormhole. Ingoing persons fall through the ingoing inner horizon, while outgoing persons fall through the
outgoing inner horizon.
Coordinate time t moves forwards in the Universe and Wormhole regions, and geodesics have positive
energy in these regions. Conversely, coordinate time t moves backwards in the Parallel Universe and Parallel
Wormhole regions, and geodesics have negative energy in these regions. Of course, all observers, whereever
they may be, always perceive their own proper time to be moving forward in the usual fashion, at the rate
of one second per second.
8.9 Mass ination instability
Roger Penrose (1968) rst pointed out that a person passing through the outgoing inner horizon (also
called the Cauchy horizon) of the Reissner-Nordstr om geometry would see the outside Universe innitely
blueshifted, and he suggested that this would destabilize the geometry. Perturbation theory calculations,
starting with Simpson & Penrose (1973) and culminating with Chandrasekhar and Hartle (1982), conrmed
that waves become innitely blueshifted as they approach the outgoing inner horizon, and that their energy
density diverges. The perturbation theory calculations were widely construed as indicating that the Reissner-
Nordstr om geometry was unstable, although the precise nature of this instability remained obscure.
It was not until a seminal paper by Poisson & Israel (1990) that the true nonlinear nature of the instability
at the inner horizon was claried. Poisson & Israel showed that the Reissner-Nordstr om geometry is subject
to an exponentially growing instability which they dubbed mass ination. The term refers to the fact that
the interior mass M(r) grows exponentially during mass ination. The interior mass M(r) has the property
of being a gauge-invariant, scalar quantity, so it has a physical meaning independent of the coordinate system.
What causes mass ination? Actually it has nothing to do with mass: the inating mass is just a symptom
of the underlying cause. What causes mass ination is relativistic counter-streaming between ingoing and
outgoing streams. As the Penrose diagram of the Reissner-Nordstr om geometry shows, ingoing and outgoing
streams must drop through separate ingoing and outgoing inner horizons into separate pieces of spacetime,
the Wormhole and the Parallel Wormhole. The regions of spacetime must be separate because coordinate
time t is timelike in both regions, but going in opposite directions in the two regions, forward in the Wormhole,
backward in the Parallel Wormhole. In other words, ingoing and outgoing streams cannot co-exist in the
1
The apparently wrong direction of time comes from requiring that t for ingoing trajectories near the horizon, whether
just above or just below the horizon, while t for outgoing trajectories near the horizon, whether just above or just
below the horizon. It would be unconventional, but consistent, to ip the sign of time t between the horizons so that
positive energy increased with time, and negative energy decreased with time.
Universe
H
o
r
i
z
o
n
Black Hole
I
n
g
o
i
n
g
i
n
n
e
r
h
o
r
i
z
o
n
O
u
t
g
o
i
n
g
i
n
n
e
r
h
o
r
i
z
o
n
I
n
g
o
i
n
g
O
u
t
g
o
i
n
g
t
t
t t
Figure 8.2 Penrose diagram illustrating why the Reissner-Nordstrom geometry is subject to the mass ina-
tion instability. Ingoing and outgoing streams just outside the inner horizon must pass through separate
ingoing and outgoing inner horizons into causally separated pieces of spacetime where the timelike time
coordinate t goes in opposite directions. To accomplish this, the ingoing and outgoing streams must exceed
the speed of light through each other, which physically they cannot do. The mass ination instability is
driven by the pressure of the relativistic counter-streaming between ingoing and outgoing streams. The
inset shows the direction of coordinate time t in the various regions. Proper time of course always increases
upward in a Penrose diagram.
same subluminal region of spacetime because they would have to be moving in opposite directions in time,
which cannot be.
In the Reissner-Nordstr om geometry, ingoing and outgoing streams resolve their dierences by exceeding
the speed of light relative to each other, and passing into causally separated regions. As the ingoing and
outgoing streams drop through their respective inner horizons, they each see the other stream innitely
blueshifted.
In reality however, this cannot occur: ingoing and outgoing streams cannot exceed the speed of light
relative to each other. Instead, as the ingoing and outgoing streams move ever faster through each other
in their eort to drop through the inner horizon, their counter-streaming generates a radial pressure. The
pressure, which is positive, exerts an inward gravitational force. As the counter-streaming approaches the
8.10 Inevitability of mass ination 103
speed of light, the gravitational force produced by the counter-streaming pressure eventually exceeds the
gravitational force produced by the background Reissner-Nordstr om geometry. At this point, mass ination
begins.
The gravitational force produced by the counter-streaming is inwards, but, in the strange way that general
relativity operates, the inward direction is in opposite directions for the ingoing streams, towards the black
hole for the ingoing stream, and away from the black hole for the outgoing stream. Consequently the counter-
streaming pressure simply accelerates the ingoing and outgoing streams ever faster through each other. The
result is an exponential feedback instability. The increasing pressure accelerates the streams faster through
each other, which increases the pressure, which increases the acceleration.
The interior mass is not the only thing that increases exponentially during mass ination. The proper
density and pressure, and the Weyl scalar (all gauge-invariant scalars) exponentiate together.
Exercise 8.2 Show that, in the Reissner-Nordstr om geometry, the blueshift of a photon with energy
v
t
= 1 and angular momentum per unit energy v
= J observed by observer on a geodesic with energy

per unit mass u
t
= E and angular momentum per unit mass u
= L is
u
=
something
B
. (8.15)
Argue that the blueshift diverges at the horizon for ingoing observers observing outgoing photons, and for
outgoing observers observing ingoing photons.
8.10 Inevitability of mass ination
Mass ination requires the simultaneous presence of both ingoing and outgoing streams near the inner
horizon. Will that happen in real black holes? Any real black hole will of course accrete matter from its
surroundings, so certainly there will be a stream of one kind or another (ingoing or outgoing) inside the
black hole. But is it guaranteed that there will also be a stream of the other kind? The answer is probably.
One of the remarkable features of the mass ination instability is that, as long as ingoing and outgoing
streams are both present, the smaller the perturbation the more violent the instability. That is, if say the
outgoing stream is reduced to a tiny trickle compared to the ingoing stream (or vice versa), then the length
scale (and time scale) over which mass ination occurs gets shorter. During mass ination, as the counter-
streaming streams drop through an interval r of circumferential radius, the interior mass M(r) increases
exponentially with length scale l
M(r) e
r/l
. (8.16)
It turns out that the inationary length scale l is proportional to the accretion rate
l

M , (8.17)
so that smaller accretion rates produce more violent ination. Physically, the smaller accretion rate, the
closer the streams must approach the inner horizon before the pressure of their counter-streaming begins to
dominate the gravitational force. The distance between the inner horizon and where mass ination begins
eectively sets the length scale l of ination.
Given this feature of mass ination, that the tinier the perturbation the more rapid the growth, it seems
almost inevitable that mass ination must occur inside real black holes. Even the tiniest piece of stu going
the wrong way is apparently enough to trigger the mass ination instability.
One way to avoid mass ination inside a real black hole is to have a large level of dissipation inside the
black hole, sucient to reduce the charge (or spin) to zero near the singularity. In that case the central
singularity reverts to being spacelike, like the Schwarzschild singularity. While the electrical conductivity of
a realistic plasma is more than adequate to neutralize a charged black hole, angular momentum transport
is intrinsically a much weaker process, and it is not clear whether the dissipation of angular momentum
might be large enough to eliminate the spin near the singularity of a rotating black hole. There has been no
research on the latter subject.
8.11 The black hole particle accelerator
A good way to think conceptually about mass ination is that it acts like a particle accelerator. The counter-
streaming pressure accelerates ingoing and outgoing streams through each other at an exponential rate, so
that a Lagrangian gas element spends equal amounts of proper time accelerating through equal decades of
counter-streaming velocity. The center of mass energy easily exceeds the Planck energy, where quantum
mechanics presumably comes into play.
Mass ination is expected to occur just above the inner horizon of a black hole. In a realistic rotating
astronomical black hole, the inner horizon is likely to be at a considerable fraction of the radius of the
outer horizon. Thus the black hole accelerator operates not near a central singularity, but rather at a
macroscopically huge scale. This machine is truly monstrous.
Undoubtedly much fascinating physics occurs in the black hole particle accelerator. The situation is far
more extreme than anywhere else in our Universe today. Who knows what Nature does there? To my
knowledge, there has been no research on the subject.
8.12 The X point
The point in the Reissner-Nordstr om geometry where the ingoing and outgoing inner horizons intersect, the
X point, is a special one. This is the point through which geodesics of zero energy must pass. Persons with
zero energy who reach the X point see both ingoing and outgoing streams, coming from opposite directions,
innitely blueshifted.
8.13 Extremal Reissner-Nordstr om geometry 105
r
=
r
=
r
=
r
=
r
=
r
=
r
=
r
=
r
=
r
=
A
n
t
i
h
o
r
i
z
o
n
Universe
H
o
r
i
z
o
n
W
o
r
m
h
o
l
e
Antiverse
New Universe
Figure 8.3 Penrose diagram of the extremal Reissner-Nordstrom geometry.
8.13 Extremal Reissner-Nordstrom geometry
So far the discussion of the Reissner-Nordstr om geometry has centered on the case Q < M (or more generally,
[Q[ < [M[) where there are separate outer and inner horizons. In the special case that the charge and mass
are equal,
Q = M , (8.18)
the inner and outer horizons merge into one, r
+
= r
, equation (8.9). This special case describes the

extremal Reissner-Nordstr om geometry.
The extremal Reissner-Nordstr om geometry proves to be of particular interest in quantum gravity because
its Hawking temperature is zero, and in string theory because extremal black holes arise as solutions under
certain duality transformations.
The Penrose diagram of the extremal Reissner-Nordstr om geometry is dierent from that of the standard
Reissner-Nordstr om geometry.
8.14 Reissner-Nordstrom geometry with charge exceeding mass
The Reissner-Nordstr om geometry with charge greater than mass,
Q > M , (8.19)
has no horizons. The change in geometry from an extremal black hole, with horizon at nite radius r
+
=
r
= M, to one without horizons is discontinuous. This suggests that there is no way to pack a black hole
with more charge than its mass. Indeed, if you try to force additional charge into an extremal black hole,
then the work needed to do so increases its mass so that the charge Q does not exceed its mass M.
Real fundamental particles nevertheless have charge far exceeding their mass. For example, the charge-
to-mass ratio of a proton is
e
m
p
10
18
(8.20)
where e is the square root of the ne-structure constant e
2
/c 1/137, and m
p
10
19
is the mass of
the proton in Planck units. However, the Schwarzschild radius of such a fundamental particle is far tinier
than its Compton wavelength /m (or its classical radius e
2
/m = /m), so quantum mechanics, not
general relativity, governs the structure of these fundamental particles.
8.15 Reissner-Nordstrom geometry with imaginary charge
It is possible formally to consider the Reissner-Nordstr om geometry with imaginary charge Q
Q
2
< 0 . (8.21)
This is completely unphysical. If charge were imaginary, then electromagnetic energy would be negative.
However, the Reissner-Nordstr om metric with Q
2
< 0 is well-dened, and it is possible to calculate
geodesics in that geometry. What makes the geometry interesting is that the singularity, instead of being
gravitationally repulsive, becomes gravitationally attractive. Thus particles, instead of bouncing o the
singularity, are attracted to it, and it turns out to be possible to continue geodesics through the singularity.
Mathematically, the geometry can be considered as the Kerr-Newman geometry in the limit of zero spin. In
8.15 Reissner-Nordstr om geometry with imaginary charge 107
r
=
r
=
r
=
r
=
r
=
r
=
r
=
r
=
Singularity
P
a
r
a
l
l
e
l
A
n
t
i
h
o
r
i
z
o
n
A
n
t
i
h
o
r
i
z
o
n
P
a
r
a
l
l
e
l
H
o
r
i
z
o
n
H
o
r
i
z
o
n
Black Hole
Singularity (r = 0)
Black Hole
I
n
n
e
r
H
o
r
i
z
o
n
P
a
r
a
l
l
e
l
I
n
n
e
r
H
o
r
i
z
o
n
Antiverse
Parallel
Antiverse
White Hole
Singularity
White Hole
I
n
n
e
r
A
n
t
i
h
o
r
i
z
o
n
P
a
r
a
l
l
e
l
I
n
n
e
r
A
n
t
i
h
o
r
i
z
o
n
i
i
i
Figure 8.4 Penrose diagram of the Reissner-Nordstrom geometry with imaginary charge Q. If charge were
imaginary, then electromagnetic energy would be negative, which is completely unphysical. But the metric
is well-dened, and the spacetime is fun.
the Kerr-Newman geometry, geodesics can pass from positive to negative radius r, and the passage through
the singularity of the Reissner-Nordstr om geometry can be regarded as this process in the limit of zero spin.
Suce to say that it is intriguing to see what it looks like to pass through the singularity of a charged
black hole of imaginary charge, however unrealistic. The Penrose diagram is even more eventful than that
for the usual Reissner-Nordstr om geometry.
9
Kerr-Newman Black Hole
The geometry of a stationary, rotating, uncharged black hole in asymptotically at empty space was dis-
covered unexpectedly by Roy Kerr in 1963. Kerrs (2007) own account of the history of the discovery is at
http://arxiv.org/abs/0706.1109. You can read in that paper that the discovery was not mere chance:
Kerr used sophisticated mathematical methods to make it. The extension to a rotating electrically charged
black hole was made shortly thereafter by Ted Newman (Newman et al. 1965). Newman told me (private
communication 2009) that, after seeing Kerrs work, he quickly realized that the extension to a charged black
hole was straightforward. He set the problem to the graduate students in his relativity class, who became
coauthors of Newman et al. (1965).
The importance of the Kerr-Newman geometry stems in part from the no-hair theorem, which states
that this geometry is the unique end state of spacetime outside the horizon of an undisturbed black hole in
asymptotically at space.
9.1 Boyer-Lindquist metric
The Boyer-Linquist metric of the Kerr-Newman geometry is
ds
2
=

2
_
dt a sin
2
d
_
2
+

2
dr
2
+
2
d
2
+
R
4
sin
2
2
_
d
a
R
2
dt
_
2
(9.1)
where R and are dened by
R
_
r
2
+a
2
,
_
r
2
+a
2
cos
2
, (9.2)
and is the horizon function dened by
R
2
2Mr +Q
2
. (9.3)
At large radius r, the Boyer-Linquist metric is
ds
2
_
1
2M
r
_
dt
2
+
_
1 +
2M
r
_
dr
2
+r
2
_
d
2
+ sin
2
d
2
_
4aM sin
2
r
dtd . (9.4)
110 Kerr-Newman Black Hole
Comparison of this metric to the metric of a weak eld establishes that M is the mass of the black hole and
a is its angular momentum per unit mass. For positive a, the black hole rotates right-handedly about its
polar axis = 0.
The Boyer-Linquist line-element (9.1) denes not only a metric but also a tetrad. The Boyer-Linquist
coordinates and tetrad are carefully chosen to exhibit the symmetries of the geometry. In the locally inertial
frame dened by the Boyer-Linquist tetrad, the energy-momentum tensor (which is non-vanishing for charged
Kerr-Newman) and the Weyl tensor are both diagonal. These assertions becomes apparent only in the tetrad
frame, and are obscure in the coordinate frame.
9.2 Oblate spheroidal coordinates
Boyer-Linquist coordinates r, , are oblate spheroidal coordinates (not polar coordinates). Corresponding
Cartesian coordinates are
x = Rsin cos ,
y = Rsin sin ,
z = r cos .
(9.5)
Surfaces of constant r are confocal oblate spheroids, satisfying
x
2
+y
2
r
2
+a
2
+
z
2
r
2
= 1 . (9.6)
Equation (9.6) implies that the spheroidal coordinate r is given in terms of x, y, z by the quadratic equation
r
4
r
2
(x
2
+y
2
+z
2
a
2
) a
2
z
2
= 0 . (9.7)
9.3 Time and rotation symmetries
The Boyer-Linquist metric coecients are independent of the time coordinate t and of the azimuthal angle
. This shows that the Kerr-Newman geometry has time translation symmetry, and rotational symmetry
about its azimuthal axis. The time and rotation symmetries means that the tangent vectors g
t
and g
in
Boyer-Linquist coordinates are Killing vectors. It follows that their scalar products
g
t
g
t
= g
tt
=
1
2
_
a
2
sin
2
_
,
g
t
g
= g
t
=
a sin
2
2
_
R
2
_
,
g
= g
=
sin
2
2
_
R
4
a
2
sin
2

_
, (9.8)
9.4 Ring singularity 111
are all gauge-invariant scalar quantities. As will be seen below, g
tt
= 0 denes the boundary of ergospheres,
g
t
= 0 denes the turnaround radius, and g
= 0 denes the boundary of the toroidal region containing

closed timelike curves.
The Boyer-Linquist time t and azimuthal angle are arranged further to satisfy the condition that g
t
and
g
are each orthogonal to both g

r
and g
.
9.4 Ring singularity
The Kerr-Newman geometry contains a ring singularity where the Weyl tensor (9.21) diverges, = 0, or
equivalently at
r = 0 and = /2 . (9.9)
The ring singularity is at the focus of the confocal ellipsoids of the Boyer-Linquist metric. Physically, the
singularity is kept open by the centrifugal force.
9.5 Horizons
The horizon of a Kerr-Newmman black hole rotates, as observed by a distant observer, so it is incorrect to
try to solve for the location of the horizon by assuming that the horizon is at rest. The worldline of a photon
that sits on the horizon, battling against the inow of space, remains at xed radius r and polar angle , but
it moves in time t and azimuthal angle . The photons 4-velocity is v
= v
t
, 0, 0, v
, and the condition

that it is on a null geodesic is
0 = v
= g
= g
tt
(v
t
)
2
+ 2 g
t
v
t
v
+ g
(v
)
2
. (9.10)
This equation has solutions provided that the determinant of the 2 2 matrix of metric coecients in t and
is less than or equal to zero (why?). The determinant is
g
tt
g
g
2
t
= sin
2
(9.11)
where is the horizon function dened above, equation (9.3). Thus if 0, then there exist null geodesics
such that a photon can be instantaneously at rest in r and , whereas if < 0, then no such geodesics exist.
The boundary
= 0 (9.12)
denes the location of horizons. With given by equation (9.3), equation (9.12) gives outer and inner
horizons at
r
= M
_
M
2
Q
2
a
2
. (9.13)
Between the horizons is negative, and photons cannot be at rest. This is consistent with the picture that
space is falling faster than light between the horizons.
E
r
g
o
s
p
h
e
r
e
CTCs
R
o
t
a
t
i
o
n
a
x
i
s
In
ner horizo
n
O
u
ter horizo
n
r = 0 Ring
singularity
E
r
g
o
s
p
h
e
r
e
CTCs
R
o
t
a
t
i
o
n
a
x
i
s
I
n
n
er horiz
o
n
O
u
ter horiz
o
n
T
u
rnaroun
d
r = 0 Ring
singularity
Figure 9.1 Geometry of (upper) a Kerr black hole with spin parameter a = 0.96M, and (lower) a Kerr-
Newman black hole with charge Q = 0.8M and spin parameter a = 0.56M. The upper half of each diagram
shows r 0, while the lower half shows r 0, the Antiverse. The outer and inner horizons are confocal
oblate spheroids whose focus is the ring singularity. For the Kerr geometry, the turnaround radius is at
r = 0. CTCs are closed timelike curves.
9.6 Angular velocity of the horizon 113
Figure 9.2 Not a mouses eye view of a snake coming down its mousehole, uhoh. Contours of constant ,
and their normals, in Boyer-Linquist coordinates, in a Kerr black hole of spin parameter a = 0.96M. The
thicker contours are the outer and inner horizons, which are confocal spheroids with the ring singularity at
their focus.
9.6 Angular velocity of the horizon
The Boyer-Linquist metric (9.1) has been cunningly written so that you can read o the angular velocity of
the horizon as observed by observers at rest at innity. The horizon is at dr = d = 0 and = 0, and then
the null condition ds
2
= 0 implies that the angular velocity is
d
dt
=
a
R
2
. (9.14)
The derivative is with respect to the proper time t of observers at rest at innity, so this is the angular
velocity observed by such observers.
9.7 Ergospheres
There are nite regions, just outside the outer horizon and just inside the inner horizon, within which the
worldline of an object at rest, dr = d = d = 0, is spacelike. These regions, called ergospheres, are places
where nothing can remain at rest (the place where little children come from). Objects can escape from within
the outer ergosphere (whereas they cannot escape from within the outer horizon), but they cannot remain
at rest there. A distant observer will see any object within the outer ergosphere being dragged around by
the rotation of the black hole. The direction of dragging is the same as the rotation direction of the black
hole in both outer and inner ergospheres.
The boundary of the ergosphere is at
g
tt
= 0 (9.15)
which occurs where
= a
2
sin
2
. (9.16)
Equation (9.16) has two solutions, the outer and inner ergospheres. The outer and inner ergospheres touch
respectively the outer and inner horizons at the poles, = 0 and .
9.8 Antiverse
The surface at zero radius, r = 0, forms a disk bounded by the ring singularity. Objects can pass through
this disk into the region at negative radius, r < 0, the Antiverse.
The Boyer-Lindquist metric (9.1) is unchanged by a symmetry transformation that simultaneously ips
the sign both of the radius and mass, r r and M M. Thus the Boyer-Linquist geometry at
negative r with positive mass is equivalent to the geometry at positive r with negative mass. In eect, the
Boyer-Linquist metric with negative r describes a rotating black hole of negative mass
M < 0 . (9.17)
9.9 Closed timelike curves
Inside the inner horizon there is a toroidal region around the ring singularity within which the light cone in
t- coordinates opens up to the point that as well as t are timelike coordinates. The direction of increasing
proper time along t is t increasing, and along is decreasing, which is retrograde. Within the toroidal
region, there exist timelike trajectories that go either forwards or backwards in coordinate time t as they wind
retrograde around the toroidal tunnel. Because the coordinate is periodic, these timelike curves connect
not only the past to the future (the usual case), but also the future to the past, which violates causality. In
particular, as rst pointed out by Carter (1968), there exist closed timelike curves (CTCs), trajectories
that connect to themselves, connecting their own future to their own past, and repeating interminably, like
Sisyphus pushing his rock up the mountain.
The boundary of this toroidal region is at
g
= 0 (9.18)
which occurs where
R
4
= a
2
sin
2
. (9.19)
9.9 Closed timelike curves 115
r
=
r
=
r
=
r
=
r
=
r
=
r
=
r
=
White Hole
P
a
r
a
l
l
e
l
A
n
t
i
h
o
r
i
z
o
n
A
n
t
i
h
o
r
i
z
o
n
P
a
r
a
l
l
e
l
H
o
r
i
z
o
n
H
o
r
i
z
o
n
Black Hole
I
n
n
e
r
H
o
r
i
z
o
n
P
a
r
a
l
l
e
l
I
n
n
e
r
H
o
r
i
z
o
n
W
o
r
m
h
o
l
e
P
a
r
a
l
l
e
l
W
o
r
m
h
o
l
e
Antiverse
Parallel
Antiverse
White Hole
I
n
n
e
r
A
n
t
i
h
o
r
i
z
o
n
P
a
r
a
l
l
e
l
I
n
n
e
r
A
n
t
i
h
o
r
i
z
o
n
Figure 9.3 Penrose diagram of the Kerr-Newman geometry. The diagram is similar to that of the Reissner-
Nordstrom geometry, except that it is possible to pass through the disk at r = 0 from the Wormhole region
into the Antiverse region. This Penrose diagram, which represents a slice at xed and , does not capture
the full richness of the geometry, which contains closed timelike curves in a torus around the ring singularity.
In the uncharged Kerr geometry the CTC torus is entirely at negative radius, r < 0, but in the Kerr-Newman
geometry the CTC torus extends to positive radius.
The Einstein tensor of the Kerr-Newman geometry in Boyer-Linquist coordinates is a bit of a mess, so I wont
write it down. The trick of raising one index, which for the Reissner-Nordstr om metric brought the Einstein
tensor to diagonal form, equation (8.4), fails for Boyer-Linquist (because the Boyer-Linquist metric is not
diagonal). The problem is endemic to the coordinate approach to general relativity. When we have done
tetrads we will nd that, in the Boyler-Linquist tetrad, the Einstein tensor is diagonal, and that the proper
density , the proper radial pressure p
r
, and the proper transverse pressure p
in that frame are (do not

confuse the notation for proper density with the radial parameter , equation (9.2), of the Boyer-Linquist
metric)
= p
r
= p
=
Q
2
8
4
. (9.20)
This looks like the energy-momentum tensor (8.4) of the Reissner-Nordstr om geometry with the replacement
r . The energy-momentum is that of an electric eld produced by a charge Q seemingly located at the
singularity.
9.11 Weyl tensor
The Weyl tensor of the Kerr-Newman geometry in Boyer-Linquist coordinates is likewise a mess. After
tetrads, we will nd that the 10 components of the Weyl tensor can be decomposed into 5 complex components
of spin 0, 1, and 2. In the Boyer-Linquist tetrad, the only non-vanishing component is the spin-0
component, the Weyl scalar C, but in contrast to the Schwarzschild and Reissner-Nordstr om geometries the
spin-0 component is complex, not real:
C =
1
(r ia cos )
3
_
M
Q
2
r +ia cos
_
. (9.21)
9.12 Electromagnetic eld
The expression for the electromagnetic eld in Boyer-Linquist coordinates is again a mess. After tetrads, we
will discover that, in the Boyer-Linquist tetrad, the electromagnetic eld is purely radial, and the electro-
magnetic potential has only a time component. For subsequent reference, the electromagnetic potential A
in the Boyer-Linquist coordinate (not tetrad) frame is

A
=
Qr
2
_
1, 0, 0, a sin
2
_
. (9.22)
9.13 Doran coordinates 117
9.13 Doran coordinates
For the Kerr-Newman geometry, the analog of the Gullstrand-Painleve metric is the Doran (2000) metric
ds
2
= dt
2
+
_
R
dr
R
_
dt
a sin
2
d
_
_
2
+
2
d
2
+R
2
sin
2
d
2
(9.23)
where the free-fall time t
and azimuthal angle
are related to the Boyer-Linquist time t and azimuthal

angle by
dt
= dt

1
2
dr , d
=
a
R
2
(1
2
)
dr . (9.24)
The free-fall time t
is the proper time experienced by persons who free-fall from rest at innity, with zero
angular momentum. They follow trajectories of xed and
, with radial velocity dr/dt
= . In other
words, the 4-velocity u
dx
/d of such free-falling observers is

u
t
ff
= 1 , u
r
= , u
= 0 , u
ff
= 0 . (9.25)
For the Kerr-Newman geometry, the velocity is
=
_
2Mr Q
2
R
(9.26)
where the sign is (infalling) for black hole solutions, and + (outfalling) for white hole solutions.
Horizons occur where the magnitude of the velocity equals the speed of light
= 1 . (9.27)
The boundaries of ergospheres occur where the velocity is
=
R
. (9.28)
The turnaround radius is where the velocity is zero
= 0 . (9.29)
The region containing closed timelike curves is bounded by the imaginary velocity
= i

a sin
. (9.30)
9.14 Extremal Kerr-Newman geometry
The Kerr-Newman geometry is called extremal when the outer and inner horizons coincide, r
+
= r
, which
occurs where
M
2
= Q
2
+a
2
. (9.31)
R
o
t
a
t
i
o
n
a
x
i
s
In
ner horizo
n
O
u
ter horizo
n
Figure 9.4 Geometry of a Kerr black hole with spin parameter a = 0.96M. The arrows show the velocity
in the Doran metric. The ow follows lines of constant , which form nested hyperboloids orthogonal to
and confocal with the nested spheroids of constant r.
Figure 9.5 illustrates the structure of an extremal Kerr (uncharged) black hole, and an extremal Kerr-Newman
(charged) black hole.
9.15 Trajectories of test particles in the Kerr-Newman geometry
Geodesics of test particles in the Kerr-Newman geometry have the expected three constants of motion
associated with conservation of energy, conservation of azimuthal angular momentum, and conservation of
rest mass. Remarkably, Carter (1968) was able to show by separation of variables in the Hamilton-Jacobi
equation that a fourth integral of motion exists, the Carter integral, so that there is a complete set of
four integrals of motion. Moreover, the complete set of integrals exists not only for uncharged particles
(geodesics), but also for charged particles.
The Hamilton-Jacobi method is the most powerful known method for solving equations of motion. For a
test particle it starts with conservation of rest mass m,
g
= m
2
, (9.32)
where p
mdx
/d is the usual coordinate 4-momentum of the particle. Hamilton-Jacobi replaces the

usual momentum p
in favour of the generalized momentum
= p
+qA
, so that the equation (9.32) of

E
r
g
o
s
p
h
e
r
e
CTCs
R
o
t
a
t
i
o
n
a
x
i
s
Horizon
r = 0 Ring
singularity
E
r
g
o
s
p
h
e
r
e
CTCs
R
o
t
a
t
i
o
n
a
x
i
s
H
orizon
T
u
rnaroun
d
r = 0 Ring
singularity
Figure 9.5 Geometry of (upper) an extremal (a = M) Kerr black hole, and (lower) an extremal Kerr-Newman
black hole with charge Q = 0.8M and spin parameter a = 0.6M.
conservation of rest mass becomes
g
qA
) (
qA
) = m
2
. (9.33)
Finally, Hamilton-Jacobi replaces the generalized momentum with its derivative with respect to the action,
= S/x
,
g
_
S
x
qA
__
S
x
qA
_
= m
2
. (9.34)
Equation (9.34) is the Hamilton-Jacobi equation for a test particle of rest mass m and charge q moving in a
prescribed background with metric g
and electromagnetic potential A
.
For the Boyer-Linqust geometry, two integrals of motion follow immediately from the fact that the metric
is independent of time t and azimuthal angle . These imply conservation of energy E and azimuthal angular
momentum L
z
,
t
=
S
t
= E ,
=
S
= L
z
. (9.35)
Thus for Boyer-Linquist the Hamiton-Jacobi equation (9.34) becomes
g
tt
(E qA
t
)
2
+2g
t
(E qA
t
)(L
z
qA
) +g
(L
z
qA
)
2
+g
rr
_
S
r
_
2
+g
_
S
_
2
= m
2
. (9.36)
Substituting in the Boyer-Linquist metric and the electromagnetic potential (9.22) brings equation (9.36) to
P
2
+
_
aE sin
L
z
sin
_
2
+
_
S
r
_
2
+
_
S
_
2
= m
2
2
, (9.37)
where P, a function of radius r, is
P ER
2
aL
z
qQr . (9.38)
Separating variables in equation (9.37) gives
_
S
r
_
2
+
P
2
m
2
r
2
=
_
S
_
2
+
_
aE sin
L
z
sin
_
2
+m
2
a
2
cos
2
= / , (9.39)
with / a separation constant. The separation constant / provides the desired fourth integral of motion, and
the solution is now essentially complete. The separated equation (9.39) implies
r
=
S
r
=
, (9.40a)
=
S
, (9.40b)
where 1 and are radial and angular potentials
1 P
2
_
/ +m
2
r
2
_
, (9.41a)
/
_
aE sin
L
z
sin
_
2
m
2
a
2
cos
2
. (9.41b)
The middle expression of equation (9.39) shows that the constant / is necessarily positive, hitting zero for
a massless particle, m = 0, moving along the polar axis, L
z
= 0 and = 0 or . However, it is common to
replace the constant / by the Carter integral Q
/ = Q+ (aE L
z
)
2
, (9.42)
which has the property that Q = 0 for orbits in the equatorial plane, = /2. In terms of the Carter integral
Q, the radial and angular potentials 1 and are
1 = P
2
_
Q+ (aE L
z
)
2
+m
2
r
2
, (9.43a)
= Qcos
2
_
a
2
(m
2
E
2
) +
L
2
z
sin
2
_
. (9.43b)
Converting the generalized covariant momenta, equations (9.35) and (9.40), to ordinary contravariant mo-
menta, p
= g
qA
), yields
p
t
=
1
2
_
PR
2
a
_
aE sin
2
L
z
_
_
, (9.44a)
p
r
=
1
1 , (9.44b)
p
=
1
, (9.44c)
p
=
1
2
_
aP
aE +
L
z
sin
2
_
. (9.44d)
Equations (9.44) give the general solution for the 4-momentum p
of a test particle of mass m and charge q

moving in the Kerr-Newman geometry, in terms of its integrals of motion E, L
z
, and Q.
The following exercise shows that among ideal black holes, the Schwarzschild geometry is exceptional, not
typical, in having a gravitationally attractive singularity.
Exercise 9.1 Near the Kerr-Newman ring singularity. Explore the behaviour of trajectories of test
particles in the vicinity of the Kerr-Newman singularity, where 0. Under what conditions does a test
particle reach the singularity?
1. Argue that for a particle to reach the singularity at = /2, positivity of the angular potential requires
that
Q 0 . (9.45)
2. Argue that for a particle to reach the singularity at r = 0, positivity of the radial potential 1 requires
that
Q
2
(aE L
z
)
2
+ (Q
2
+a
2
)Q 0 . (9.46)
3. Schwarzschild case: show that if Q = 0 and a = 0, then a particle reaches the singularity provided that
the mass of the black hole is positive, M > 0.
4. Reissner-Nordstr om case: show that if Q
2
> 0 and a = 0, then a particle can reach the singularity only if
it has zero angular momentum, Q = L
z
= 0, and if the particles charge-to-mass exceeds unity,
q
2
m
2
1 . (9.47)
5. Kerr case: show that if Q = 0 but a
2
> 0, then a particle can reach the singularity only if Q = 0, and
provided that the mass of the black hole is positive, M > 0.
6. Kerr-Newman case: show that if Q
2
> 0 and a
2
> 0, then a particle can reach the singularity only if
L
z
= aE and Q = 0, and if the particles charge-to-mass is large enough,
q
2
m
2

Q
2
+a
2
Q
2
, (9.48)
which generalizes the Reissner-Nordstr om condition (9.47).
9.16 Penrose process
Trajectories in the Kerr-Newman geometry can have negative energy E outside the horizon. It is possible to
reduce the mass M of the black hole by dropping negative energy particles into the black hole. This process
of extracting mass-energy from the black hole is called the Penrose process.
Exercise 9.2 Negative energy trajectories outside the horizon. Under what conditions can test
particles have negative energy trajectories, E < 0, outside the horizon?
1. Argue that outside the horizon, the positivity of the horizon function , and of the radial and angular
potentials 1 and , equations (9.41) implies that P, equation (9.38), satises
P
2
_
/ +m
2
r
2
_

_
_
aE sin
L
z
sin
_
2
+m
2
2
_
. (9.49)
2. Argue that the condition (9.49) implies by continuity that for a massive particle P must be strictly positive
outside the horizon. Extend your argument to a massless particle by taking a massless particle as a massive
particle in the limit of large energy.
3. Argue that the positivity of P implies that aL
z
+ qQr must be negative for the energy E to be negative.
Show that, more stringently, negative E requires that
aL
z
+qQr
_
L
2
z
sin
2
+m
2
2
_
. (9.50)
4. Argue that for an uncharged particle, q = 0, negative energy trajectories exist only inside the ergosphere.
5. Do negative energy trajectories exist outside the ergosphere for a charged particle?
9.17 Constant latitude trajectories in the Kerr-Newman geometry 123
6. For the Penrose process to work, the negative energy particle must fall through the horizon, where = 0.
Does this happen?
Exercise 9.3 When can objects go forwards or backwards in time t?
9.17 Constant latitude trajectories in the Kerr-Newman geometry
A trajectory is at constant latitude if it is at constant polar angle ,
= constant . (9.51)
Constant latitude orbits occur where the angular potential , equation (9.43b), not only vanishes, but is an
extremum,
=
d
d
= 0 , (9.52)
the derivative being taken with the constants of motion E, L
z
, and Q of the orbit being held xed. The
condition = 0 simply sets the value of the Carter integral Q. Solving d/d = 0 yields the condition
between energy E and angular momentum L
z
E =
_
1 +
L
2
z
a
2
sin
4
. (9.53)
Solutions at any polar angle and any angular momentum L
z
exist, ranging from E = 1 at L
z
= 0, to
E = L
z
/(a sin
2
) at L
z
. The solutions with L
z
= 0 are those of the freely-falling observers that
dene the Doran coordinate system, 9.13. The solutions with L
z
dene the principal null congruences
discussed in 9.18.
9.18 Principal null congruence
A congruence is a space-lling, non-overlapping set of geodesics. In the Kerr-Newman geometry there is
a special set of null geodesics, the ingoing and outgoing principal null congruences, with respect to
which the symmetries of the geometry are especially apparent. Photons that hold steady on the horizon are
members of the outgoing principal null congruence. The energy-momentum tensor is diagonal in a locally
inertial frame aligned with the ingoing or outgoing principal null congruence. The Weyl tensor, decomposed
into spin components in the locally inertial frame of the principal null congruences, contains only spin-0
components.
The Boyer-Linquist metric (9.1) is specically constructed so that the Boyer Linquist tetrad is aligned with
the principal null tetrad. Along the principal null congruences, the nal two terms of the Boyer-Linquist
line element (9.1) vanish
d = d
a
R
2
dt = 0 . (9.54)
Solving the null condition ds
2
= 0 on the rest of the metric yields the photon 4-velocity v
dx
/d on the
principal null congruences
v
t
=
R
2
, v
r
= 1 , v
= 0 , v
=
a
. (9.55)
In the regions outside the outer horizon or inside the inner horizon, the sign in front of v
r
is + for outgoing,
for ingoing geodesics. Between the outer and inner horizons, v
r
is negative in the Black Hole region, and
positive in the White Hole region, while v
t
and v
are negative for ingoing, positive for outgoing geodesics.

The angular momentum per unit energy J
z
L
z
/[E[ of photons along the principal null congruences is not
zero, but is
J
z
= a sin
2
(9.56)
with the same sign for both ingoing and outgoing geodesics.
9.19 Circular orbits in the Kerr-Newman geometry
An orbit can be termed circular if it is at constant radius r,
r = constant . (9.57)
It is convenient to call such an orbit circular even if the orbit is at nite inclination (not conned to the
equatorial plane) about a rotating black hole, and therefore follows the surface of a spheroid (in Boyer-
Lindquist coordinates).
Orbits turn around in r, reaching periapsis or apoapsis, where the radial potential 1, equation (9.43a),
vanishes. Circular orbits occur where the radial potential 1 not only vanishes, but is an extremum,
1 =
d1
dr
= 0 , (9.58)
the derivative being taken with the constants of motion E, L
z
, and Q of the orbit being held xed. Circular
orbits may be either stable or unstable. The stability of a circular orbit is determined by the sign of the
second derivative of the potential
d
2
1
dr
2
, (9.59)
with + for stable, for unstable circular orbits. Marginally stable orbits occur where d
2
1/dr
2
= 0.
Circular orbits occur not only in the equatorial plane, but at general inclinations. The inclination of an
orbit can be characterized by the minimum polar angle
min
to which it extends. An astronomer would call
/2
min
the inclination angle of the orbit. It is convenient to dene an inclination parameter by
cos
2
min
, (9.60)
which lies in the interval [0, 1]. Equatorial orbits, at = /2, correspond to = 0, while polar orbits, those
that go over the poles at = 0 and , correspond to = 1.
9.19.1 General solution for circular orbits
The general solution for circular orbits of a test particle of arbitrary electric charge q in the Kerr-Newman
geometry is as follows.
The rest mass m of the test particle can be set equal to unity, m = 1, without loss of generality. Circular
orbits of particles with zero rest mass, m = 0, discussed later in this section, occur in cases where the circular
orbits for massive particles attain innite energy and angular momentum.
In the radial potential 1, equation (9.43a), eliminate the Carter integral Q in favour of the inclination
parameter , equation (9.60), using equation (9.43b)
Q =
_
a
2
(1 E
2
) +
L
2
z
1
_
. (9.61)
Furthermore, eliminate the energy E in favour of P, equation (9.38). The radial derivatives d
n
1/dr
n
must
be taken before E is replaced by P, since E is a constant of motion, whereas P varies with r. The physical
motivation for replacing E with P lies in the sign of P. Solutions with positive P correspond to orbits in the
Universe, Wormhole, or Antiverse parts of the Kerr-Newman geometry in the Penrose diagram of Figure 9.3,
while solutions with negative P correspond to orbits in their Parallel counterparts. If only the Universe
region is considered, then P is necessarily positive. By contrast, the energy E can be either positive or
negative in the same region of the Kerr-Newman geometry (the energy E is negative for orbits of suciently
large negative angular momentum L
z
inside the ergosphere of the Universe).
The condition 1 = 0 is a quadratic equation in L
z
, whose solutions are
L
z
=
1
r
2
+a
2
_
a(1 )(P +qQr) R
2
_
(1 ) [P
2
/(r
2
+a
2
)]
_
. (9.62)
Substituting the two () expressions (9.62) for L
z
into d1/dr, and setting the product of the resulting two
expressions for d1/dr equal to zero, yields a quartic equation for P/:
p
0
+p
1
(P/) +p
2
(P/)
2
+p
3
(P/)
3
+p
4
(P/)
4
= 0 , (9.63)
with coecients
p
0
r
2
(r
2
+a
2
)
2
, (9.64a)
p
1
2qQr(r
2
a
2
)(r
2
+a
2
) , (9.64b)
p
2
2r
2
(r
2
+a
2
)(r
2
3Mr + 2Q
2
+a
2
+a
2
M/r) +q
2
Q
2
(r
2
a
2
)
2
, (9.64c)
p
3
2qQr(r
2
a
2
)(r
2
3Mr + 2Q
2
+ 2a
2
a
2
+a
2
M/r) , (9.64d)
p
4

_
r
6
6Mr
5
+ (9M
2
+4Q
2
+2a
2
)r
4
4M(3Q
2
+a
2
)r
3
+(4Q
4
6a
2
M+4a
2
Q
2
+a
4
2
)r
2
+ 2a
2
(2Q
2
+2a
2
a
2
)Mr +a
4
2
M
2
. (9.64e)
The quartic (9.63) is the condition for an orbit at radius r to be circular. Physical solutions must be real.
The quartic (9.63) has either zero, two, or four real solutions at any one radius r. Numerically, it is better
to solve the quartic (9.63) for the reciprocal /P rather than P/, since the vanishing of 1/P denes the
location of circular orbits of massless particles.
3 2 1 0 1 2 3 4 5 6 7 8 9
3
2
1
0
1
2
3
Radius r/M
/P
O
u
t
e
r
h
o
r
i
z
o
n
I
n
n
e
r
h
o
r
i
z
o
n
Figure 9.6 Values of /P for circular orbits at radius r of a charged particle about a Kerr-Newman black
hole. The values /P are real roots of the quartic (9.63); there are either zero, two, or four real roots at
any one radius. The parameters are representative: a particle of charge-to-mass q/m = 2.4 on an orbit of
inclination parameter = 0.5 about a black hole of charge Q = 0.5M and spin parameter a = 0.5M. Solid
(green) lines indicate stable orbits; dashed (brown) lines indicate unstable orbits. Positive /P orbits occur
in Universe, Wormhole, and Antiverse regions; negative /P orbits occur in their Parallel counterparts;
zero /P orbits are null. The fact that the particle is charged breaks the symmetry between positive and
negative /P. If the charge of the particle were ipped, q/m = 2.4, then the diagram would be reected
about the horizontal axis (the sign of /P would ip).
The angular momentum L
z
, energy E, and stability d
2
1/dr
2
of a circular orbit are, in terms of a solution
P/ of the quartic (9.63),
L
z
=
1
r
2
+a
2
_
(1 ) [l
1
(/P) +l
0
+l
1
(P/) +l
2
(P/)
2
] , (9.65a)
E =
1
2
[(/P) +qQ/r + (1 M/r)(P/)] , (9.65b)
d
2
1
dr
2
=
2
(r
2
+a
2
)
2
_
q
1
(/P) +q
0
+q
1
(P/) +q
2
(P/)
2
, (9.65c)
where the coecients l
i
and q
i
are
l
1
= qQrR
2
(r
2
+a
2
) , (9.66a)
l
0
= R
2
(r
2
+a
2
)(2Mr Q
2
) q
2
Q
2
(r
4
a
4
) , (9.66b)
l
1
= qQr
_
2r
4
5Mr
3
+ 3(Q
2
+a
2
)r
2
a
2
(1+)Mr +a
2
(Q
2
+Q
2
+a
2
a
2
)
+3a
4
M/r a
4
(Q
2
+a
2
)/r
2
, (9.66c)
l
2
=
_
3Mr
3
2Q
2
r
2
+a
2
(1+)Mr a
2
(1+)Q
2
a
4
M/r
, (9.66d)
and
q
1
= 2qQr(r
2
a
2
)(r
2
+a
2
) , (9.67a)
q
0
= 4(r
2
+a
2
)(Mr
3
Q
2
r
2
a
2
Mr) q
2
Q
2
(r
2
a
2
)
2
, (9.67b)
q
1
= qQr
_
r
4
4Mr
3
+ 3(Q
2
+a
2
2a
2
)r
2
+ 12a
2
Mr a
2
(6Q
2
+6a
2
a
2
)
a
4
2
(Q
2
+a
2
)/r
2
, (9.67c)
q
2
=
_
3Mr
3
4Q
2
r
2
6a
2
Mr a
4
2
M/r
_
. (9.67d)
The sign of the angular momentum L
z
in equation (9.65a) should be chosen such that the relations (9.38)
for P and (9.65b) for E hold. This choice of sign becomes ambiguous for a = 0; but this is as it should be,
since either sign of L
z
is valid for a = 0, where the black hole is spherically symmetric, and therefore denes
no preferred direction.
The expressions (9.62) and (9.65a) for L
z
are equal on a circular orbit. The advantage of the latter
expression (9.65a) will become apparent below, where it is found that for particles of zero electric charge,
q = 0, one circular orbit is always prograde, aL
z
> 0, while the other is always retrograde, aL
z
< 0.
For non-zero a, the reality of a solution P/ of the quartic (9.63) is a necessary and sucient condition for
a corresponding circular orbit to exist. In particular, the argument of the square root in the expression (9.65a)
for L
z
is guaranteed to be positive. For zero a, however, the quartic (9.65a), which reduces in this case to
the square of a quadratic, admits real solutions that do not correspond to a circular orbit. For these invalid
solutions, the argument of the square root in the expression (9.65a) for L
z
is negative. Thus for zero a, a
necessary and sucient condition for a circular orbit to exist is that the solutions for both P/ and L
z
be
real.
9.19.2 Circular orbits for massless particles
Circular orbits for massless particles, m = 0, or null circular orbits, follow from the solutions for massive
particles in the case where the energy and angular momentum on the circular orbit become innite, which
occurs when P . Except at horizons, where = 0, the solution for P from the quartic (9.63) diverges
when the ratio p
4
/p
0
of the highest to lowest order coecients vanishes. The ratio p
4
/p
0
, equations (9.64),
factors as
p
4
/p
0
=
F
+
F
(r
2
+a
2
)
2
, (9.68)
where
F
r
2
3Mr + 2Q
2
+a
2
(1 +M/r) 2a
_
(1 )(Mr Q
2
a
2
M/r) . (9.69)
A null circular orbit thus occurs at a radius r such that
F
+
= 0 or F
= 0 , (9.70)
with + for prograde (aL
z
> 0) orbits, for retrograde (aL
z
< 0) orbits. The location of null circular
orbits are independent of the charge q of the particle, since F
are independent of charge q. The angular

momentum J
z
per unit energy on the null circular orbit is, from equations (9.65a) and (9.65b) in the limit
P ,
J
z
L
z
/[E[ =
2
_
(1 )l
2
(r
2
+a
2
)
2
(1 M/r)
. (9.71)
The case where F
+
or F
vanishes at a horizon is special. This occurs when the black hole is extremal,
M
2
= Q
2
+ a
2
. A circular orbit exists at the horizon of an extremal black hole provided that the charge
squared Q
2
and inclination parameter are not too large, the precise condition being
a
4
2
+ 6(Q
2
+a
2
) (Q
2
+a
2
)(Q
2
3a
2
) 0 . (9.72)
The circular orbit is non-null, since the vanishing of /P no longer implies that P diverges if = 0, as is
true on the horizon. A careful analysis shows that the limiting value of P/
is nite for a circular orbit

at the horizon of an extremal black hole, so in fact P = 0 for such an orbit.
Since there are null geodesics, the ingoing or outgoing principal null geodesics, that hold steady on the
horizon, one might have expected that there would always be solutions for null circular orbits on the horizon,
but this is false. The resolution of the paradox is that massless particles experience no proper time along
their geodesics. If a massive particle is put on the horizon on a relativistic geodesic, then the massive particle
necessarily falls o the horizon in a nite proper time: it is impossible for the geodesic to hold steady on
the horizon. The only exception is that, as discussed in the previous paragraph, an extremal black hole may
have circular orbits at its horizon; but these orbits have P = 0, and are not null.
9.19.3 Circular orbits for particles with zero electric charge
For a particle with zero electric charge, q = 0, the quartic condition (9.63) for a circular orbit reduces to a
quadratic in (P/)
2
. Solving the quadratic for the reciprocal (/P)
2
yields two possible solutions
(/P)
2
=
F
r
2
+a
2
, (9.73)
where F
are dened by equation (9.69), with + for prograde (aL

z
z
< 0)
orbits. The sign of P is positive in the Universe, Wormhole, and Antiverse of Figure 9.3, negative in their
Parallel counterparts. For zero electric charge, the expressions (9.65) for the angular momentum L
z
, energy
E, and stability d
2
1/dr
2
of a circular orbit simplify to
L
z
=
1
r
2
+a
2
_
(1 ) [l
0
+l
2
(P/)
2
] , (9.74a)
E =
1
2
[(/P) + (1 M/r)(P/)] , (9.74b)
d
2
1
dr
2
=
2
(r
2
+a
2
)
2
_
q
0
+q
2
(P/)
2
. (9.74c)
The coecients l
i
and q
i
in equations (9.74) reduce from the expressions (9.66) and (9.67) to
l
0
= R
2
(r
2
+a
2
)(2Mr Q
2
) , (9.75a)
l
2
=
_
3Mr
3
2Q
2
r
2
+ a
2
(1+)Mr a
2
(1+)Q
2
a
4
M/r
, (9.75b)
and
q
0
= 4(r
2
+a
2
)(Mr
3
Q
2
r
2
a
2
Mr) , (9.76a)
q
2
=
_
3Mr
3
4Q
2
r
2
6a
2
Mr a
4
2
M/r
_
. (9.76b)
.0 .5 1.0 1.5 2.0
2.0
1.5
1.0
.5
.0
.5
1.0
1.5
2.0
Radius r/M
P
O
u
t
e
r
h
o
r
i
z
o
n
I
n
n
e
r
h
o
r
i
z
o
n
Figure 9.7 Values of P for circular orbits at radius r in the equatorial plane of a near-extremal Kerr black
hole, with black hole spin parameter a = 0.999M. The diagram illustrates that as the orbital radius r
approaches the horizon, P rst approaches zero, but then increases sharply to innity, corresponding to null
circular orbits. In the case of an exactly extremal black hole, P goes as to zero at the horizon, there is no
increase of P to innity, and no null circular orbit. Solid (green) lines indicate stable orbits; dashed (brown)
lines indicate unstable orbits.
9.19.4 Equatorial circular orbits in the Kerr geometry
The case of greatest practical interest to astrophysicists is that of circular orbits in the equatorial plane of
an uncharged black hole, the Kerr geometry.
For circular orbits in the equatorial plane, = 0, of an uncharged black hole, Q = 0, the solution (9.73)
simplies to
(/P)
2
=
F
r
2
(9.77)
where F
, equation (9.69), reduce to

F
r
2
3Mr 2a
Mr , (9.78)
with + for prograde (aL
z
z
< 0) orbits.
As discussed above, null circular orbits occur where F
= 0, except in the special case that the circular

orbit is at the horizon, which occurs when the black hole is extremal. In the limit where the Kerr black hole
is near but not exactly extremal, a [M[, null circular orbits occur at r M (prograde) and r 4M
(retrograde). For an exactly extremal Kerr black hole, a = [M[, the (prograde) circular orbit at the horizon
is no longer null. The situation of a near extremal Kerr black hole is illustrated by Figure 9.7.
It is generally argued that the inner edge of an accretion disk is likely to occur at the innermost stable
equatorial circular orbit. An orbit at this point has marginal stability, d
2
1/dr
2
= 0. Simplifying the stability
d
2
1/dr
2
from equation (9.74c) to the case of equatorial orbits, = 0, and zero black hole charge, Q = 0,
yields the condition of marginal stability
r
2
6Mr 3a
2
8a
Mr = 0 . (9.79)
The + (prograde) orbit has the smaller radius, and so denes the innermost stable circular orbit. For an
extremal Kerr black hole, a = [M[, marginally stable circular equatorial orbits are at r = M (prograde) and
r = 9M (retrograde).
9.19.5 Circular orbits in the Reissner-Nordstr om geometry
Circular orbits of particles in the Reissner-Nordstr om geometry follow from those in the Kerr-Newman
geometry in the limit of a non-rotating black hole, a = 0. For a non-rotating black hole, an orbit can be
taken without loss of generality to circulate right-handedly in the equatorial plane, = /2, so that = 0
and the azimuthal angular momentum L
z
equals the positive total angular momentum L. For non-equatorial
orbits, the relation between azimuthal and total angular momentum is L
z
=
1 L.
For a non-rotating black hole, a = 0, the quartic condition (9.63) for a circular orbit of a particle of rest
mass m = 1 and electric charge q reduces to the square of a quadratic,
r
2
qQr(P/)
_
r
2
3Mr + 2Q
2
_
(P/)
2
= 0 . (9.80)
Solving the quadratic (9.80) for the reciprocal /P yields two solutions
/P =
qQ
2r

_
1
3M
r
+
2Q
2
r
2
+
q
2
Q
2
4r
2
. (9.81)
The sign of P is positive in the Universe, Wormhole, and Antiverse parts of the Reissner-Nordstr om geometry
in the Penrose diagram of Figure 8.1, negative in their Parallel counterparts. The angular momentum L,
energy E, and stability d
2
1/dr
2
of a circular orbit are, in terms of a solution /P of the quadratic (9.80),
L =
_
P
2
/r
2
, (9.82)
E =
P
r
2
+
qQ
r
, (9.83)
d
2
1
dr
2
= 2
_
r
2
6Mr + 5Q
2
+q
2
Q
2
_
2
_
1
6M
r
+
6Q
2
r
2
_
P
2
. (9.84)
For massless particles, circular orbits occur where the solution (9.81) for /P vanishes, which occurs when
r
2
3Mr + 2Q
2
= 0 , (9.85)
independent of the charge q of the particle. The condition (9.85) is consistent with the Kerr-Newman
condition for a null circular orbit, the vanishing of F
given by equation (9.69). However, for Kerr-Newman,

the argument of the square root on the right hand side of equation (9.69) for F
must be positive, even in

the limit of innitesimal a. In the limit of small a, this requires that Mr Q
2
0. If the charge Q of the
Reissner-Nordstr om black hole lies in the standard range 0 Q
2
M
2
, then one of the solutions of the
quadratic (9.85) lies outside the outer horizon, while the other lies between the outer and inner horizons.
As one might hope, the additional condition Mr Q
2
0 eliminates the undesirable solution between the
horizons, leaving only the solution outside the horizon, which is
r =
3M
2
_
1 +
_
1
8Q
2
9
_
for 0 Q
2
M
2
. (9.86)
In (unphysical) cases Q
2
< 0 or M
2
< Q
2
(9/8)M
2
, both solutions of equation (9.85) are valid.
PART FOUR
HOMOGENEOUS, ISOTROPIC COSMOLOGY
Concept Questions
1. What does it mean that the Universe is expanding?
2. Does the expansion aect the solar system or the Milky Way?
3. How far out do you have to go before the expansion is evident?
4. What is the Universe expanding into?
5. In what sense is the Hubble constant constant?
6. Does our Universe have a center, and if so where is it?
7. What evidence suggests that the Universe at large is homogeneous and isotropic?
8. How can the CMB be construed as evidence for homogeneity and isotropy given that it provides information
only about a 2D surface on the sky?
9. What is thermodynamic equilibrium? What evidence suggests that the early Universe was in thermody-
namic equilibrium?
10. What are cosmological parameters?
11. What cosmological parameters can or cannot be measured from the power spectrum of uctuations of the
CMB?
12. FRW Universes are characterized as closed, at, or open. Does at here mean the same as at Minkowski
space?
13. What is it that astronomers call dark matter?
14. What is the primary evidence for the existence of non-baryonic cold dark matter?
15. How can astronomers detect dark matter in galaxies or clusters of galaxies?
16. How can cosmologists claim that the Universe is dominated by not one but two distinct kinds of mysterious
mass-energy, dark matter and dark energy, neither of which has been observed in the laboratory?
17. What key property or properties distinguish dark energy from dark matter?
18. Does the Universe conserve entropy?
19. Does the annihilation of electron-positron pairs into photons generate entropy in the early Universe, as its
temperature cools through 1 MeV?
20. How does the wavelength of light change with the expansion of the Universe?
21. How does the temperature of the CMB change with the expansion of the Universe?
22. How does a blackbody (Planck) distribution change with the expansion of the Universe? What about a
non-relativistic distribution? What about a semi-relativistic distribution?
23. What is the horizon of our Universe?
24. What happens beyond the horizon of our Universe?
25. What caused the Big Bang?
26. What happened before the Big Bang?
27. What will be the fate of the Universe?
Whats important?
1. The CMB indicates that the early ( 400,000 year old) Universe was (a) uniform to a few 10
5
, and (b)
in thermodynamic equilibrium. This indicates that
the Universe was once very simple .
It is this simplicity that makes it possible to model the early Universe with some degree of condence.
2. The power spectrum of uctuations of the CMB has enabled precise measurements of cosmological pa-
rameters, excepting the Hubble constant.
3. There is a remarkable concordance of evidence from a broad range of astronomical observations su-
pernovae, big bang nucleosynthesis, the clustering of galaxies, the abundances of clusters of galaxies,
measurements of the Hubble constant from Cepheid variables, the ages of the oldest stars.
4. Observational evidence is consistent with the predictions of the theory of ination in its simplest form
the expansion of the Universe, the spatial atness of the Universe, the near uniformity of temperature
uctuations of the CMB (the horizon problem), the presence of acoustic peaks and troughs in the power
spectrum of uctuations of the CMB, the near power law shape of the power spectrum at large scales, its
spectral index (tilt), the gaussian distribution of uctuations at large scales.
5. What is non-baryonic dark matter?
6. What is dark energy? What is its equation of state w p/, and how does w evolve with time?
10
Homogeneous, Isotropic Cosmology
10.1 Observational basis
Since 1998, observations have converged on a Standard Model of Cosmology, a spatially at universe domi-
nated by dark energy and by non-baryonic dark matter.
1. The Hubble diagram (distance versus redshift) of galaxies indicates that the Universe is expanding (Hubble
1929).
2. The Cosmic Microwave Background (CMB).
Near black body spectrum, with T
0
= 2.725 0.001 K (Fixsen & Mather 2002).
Dipole the solar system is moving at 365 kms
1
through the CMB.
After dipole subtraction, the temperature of the CMB over the sky is uniform to a few parts in 10
5
.
The power spectrum of temperature T uctuations shows a scale-invariant spectrum at large scales,
and prominent acoustic peaks at smaller scales. Allows measurement of the amplitude A
s
and tilt n
s
of
primordial uctuations, the curvature density
k
, and the proper densities
c
h
2
of non-baryonic cold
dark matter and
b
h
2
of baryons. Does not measure Hubble constant h H
0
/(100 kms
1
Mpc
1
).
The power spectra of E and B polarization uctuations, and the various cross power spectra (only T-E
should be non-vanishing).
3. The Hubble diagram of Type Ia (thermonuclear) supernovae indicates that the Universe is accelerating.
This points to the dominance of gravitationally repulsive dark energy, with
0.75. The amount of

dark energy is consistent with observations from the CMB indicating that the Universe is spatially at,
1, and observations from CMB, galaxy clustering, and clusters of galaxies indicating that the density
in gravitationally attractive matter is only
m
0.25.
4. Observed abundances of light elements H, D,
3
He, He, and Li are consistent with the predictions of big
bang nucleosynthesis (BBN) provideed that
b
0.04, in good agreement with measurements from the
CMB.
5. The clustering of matter (dark and bright) shows a power spectrum in good agreement with the Standard
Model:
galaxies;
the Lyman alpha forest;
10.2 Cosmological Principle 139
gravitational lensing.
Historically, the principle evidence for non-baryonic cold dark matter is comparison between the power
spectra of galaxies versus CMB. How can tiny uctuations in the CMB grow into the observed uctuations
in matter today in only the age of the Universe? Answer: non-baryonic dark matter that begins to cluster
before Recombination, when the CMB was released.
6. The abundance of galaxy clusters as a function of redshift.
7. The ages of the oldest stars, in globular clusters. The Hubble constant yields an estimate of the age of
the Universe that is older with dark energy than without. The ages of the oldest stars agree with the age
of the Universe with dark enery, but are older than the Universe without dark energy.
8. Ubiquitous evidence for dark matter, deduced from sizes and velocities (or in the case of gravitational
lensing, the gravitational potential) of various objects.
The Local Group of galaxies.
Rotation curves of spiral galaxies.
The temperature and distribution of x-ray gas in elliptical galaxies.
The temperature and distribution of x-ray gas in clusters of galaxies.
Gravitational lensing by clusters of galaxies.
9. The Bullet cluster is a rare example that supports the notion that the dark matter is non-baryonic. In the
Bullet cluster, two clusters recently passed through each other. The baryonic matter, as measured from
x-ray emission of hot gas, appears displaced from the dark matter, as measured from weak gravitational
lensing.
10.2 Cosmological Principle
The cosmological principle states that the Universe at large is
homogeneous (has spatial translation symmetry),
isotropic (has spatial rotation symmetry).
The primary evidence for this is the uniformity of the temperature of the CMB, which, after subtraction
of the dipole produced by the motion of the solar system through the CMB, is constant over the sky to a
few parts in 10
5
. Conrming evidence is the statistical uniformity of the distribution of galaxies over large
scales.
The cosmological principle allows that the Universe evolves in time, as observations surely indicate the
Universe is expanding, galaxies, quasars, and galaxy clusters evolve with redshift, and the temperature of
the CMB is undoubtedly decreasing as the Universe expands.
140 Homogeneous, Isotropic Cosmology
10.3 Friedmann-Robertson-Walker metric
Universes satisfying the cosmological principle are described by the Friedmann-Robertson-Walker (FRW)
metric, equation (10.25) below. The metric, and the associated Einstein equations, which are known as the
Friedmann equations, are set forward in the next several sections, 10.410.9.
10.4 Spatial part of the FRW metric: informal approach
The cosmological principle implies that
the spatial part of the FRW metric is a 3D hypersphere (10.1)
where in this context the term hypersphere is to be construed as including not only cases of positive curvature,
which have nite positive radius of curvature, but also cases of zero and negative curvature, which have
innite and imaginary radius of curvature.
w
r
=
R
s
in
R
x
y
r
/
/
=
R
R
d
R
sin
Figure 10.1 Embedding diagram of the FRW geometry.

Figure 10.1 shows an embedding diagram of a 3D hypersphere in 4D Euclidean space. The horizontal
directions in the diagram represent the normal 3 spatial x, y, z dimensions, with one dimension z suppressed,
10.4 Spatial part of the FRW metric: informal approach 141
while the vertical dimension represents the 4th spatial dimension w. The 3D hypersphere is a set of points
x, y, z, w satisfying
_
x
2
+y
2
+z
2
+w
2
_
1/2
= R = constant . (10.2)
An observer is sitting at the north pole of the diagram, at 0, 0, 0, 1. A 2D sphere (which forms a 1D circle
in the embedding diagram of Figure 10.1) at xed distance surrounding the observer has geodesic distance
r
dened by
r
proper distance to sphere measured along a radial geodesic , (10.3)

and circumferential radius r dened by
r
_
x
2
+y
2
+z
2
_
1/2
, (10.4)
which has the property that the proper circumference of the sphere is 2r. In terms of r
and r, the spatial

metric is
dl
2
= dr
2
+r
2
do
2
(10.5)
where do
2
d
2
+ sin
2
d
2
is the metric of a unit 2-sphere.
Introduce the angle illustrated in the diagram. Evidently
r
= R ,
r = Rsin . (10.6)
In terms of the angle , the spatial metric is
dl
2
= R
2
_
d
2
+ sin
2
do
2
_
(10.7)
which is one version of the spatial FRW metric. The metric resembles the metric of a 2-sphere of radius R,
which is not surprising since the same construction, with Figure 10.1 interpreted as the embedding diagram
of a 2D sphere in 3D, yields the metric of a 2-sphere. Indeed, the construction iterates to give the metric of
an n-dimensional sphere of arbitrarily many dimensions n.
Instead of the angle , the metric can be expressed in terms of the circumferential radius r. It follows
from equations (10.6) that
r
= Rsin
1
(r/R) (10.8)
whence
dr
=
dr
_
1 r
2
/R
2
=
dr
1 Kr
2
(10.9)
where K is the curvature
K
1
R
2
. (10.10)
In terms of r, the spatial FRW metric is then
dl
2
=
dr
2
1 Kr
2
+r
2
do
2
. (10.11)
The embedding diagram Figure 10.1 is a nice prop for the imagination, but it is not the whole story. The
curvature K in the metric (10.11) may be not only positive, corresponding to real nite radius R, but also
zero or negative, corresponding to innite or imaginary radius R. The possibilities are called closed, at,
and open:
K
_
_
_
> 0 closed R real ,
= 0 at R ,
< 0 open R imaginary .
(10.12)
10.5 Comoving coordinates
The metric (10.11) is valid at any single instant of cosmic time t. As the Universe expands, the 3D spatial
hypersphere (whether closed, at, or open) expands. In cosmology it is highly advantageous to work in
comoving coordinates that expand with the Universe. Why? First, it is helpful conceptually and math-
ematically to think of the Universe as at rest in comoving coordinates. Second, linear perturbations, such
as those in the CMB, have wavelengths that expand with the Universe, and are therefore xed in comoving
coordinates.
In practice, cosmologists introduce the cosmic scale factor a(t)
a(t) measure of the size of the Universe, expanding with the Universe (10.13)
which is proportional to but not necessarily equal to the radius R of the Universe. The cosmic scale factor
a can be normalized in any arbitrary way. The most common convention adopted by cosmologists is to
normalize it to unity at the present time,
a
0
= 1 , (10.14)
where the 0 subscript conventionally signies the present time.
Comoving geodesic and circumferential radial distances x
and x are dened in terms of the proper geodesic

and circumferential radial distances r
and r by
ax
, ax r . (10.15)
Objects expanding with the Universe remain at xed comoving positions x
and x. In terms of the comoving

circumferential radius x, the spatial FRW metric is
dl
2
= a
2
_
dx
2
1 x
2
+x
2
do
2
_
, (10.16)
10.6 Spatial part of the FRW metric: more formal approach 143
where the curvature constant , a constant in time and space, is related to the curvature K, equation (10.10),
by
a
2
K . (10.17)
Alternatively, in terms of the geodesic comoving radius x
, the spatial FRW metric is

dl
2
= a
2
_
dx
2
+x
2
do
2
_
, (10.18)
where
x =
_
_
sin(
1/2
x
1/2
> 0 closed ,
x
= 0 at ,
sinh([[
1/2
x
)
[[
1/2
< 0 open .
(10.19)
For some purposes it is convenient to normalize the cosmic scale factor a so that = 1, 0, or 1. In this
case the spatial FRW metric may be written
dl
2
= a
2
_
d
2
+x
2
do
2
_
, (10.20)
where
x =
_
_
_
sin() = 1 closed ,
= 0 at ,
sinh() = 1 open .
(10.21)
Exercise 10.1 By a suitable transformation of the comoving radial coordinate x, bring the spatial FRW
metric (10.16) to the isotropic form
dl
2
=
a
2
_
1 +
1
4
X
2
_
2
_
dX
2
+X
2
do
2
_
. (10.22)
What is the relation between X and x?
10.6 Spatial part of the FRW metric: more formal approach
A more formal approach to the derivation of the spatial FRW metric from the cosmological principle starts
with the proposition that the spatial components G
ij
of the Einstein tensor at xed scale factor a (all time
derivatives of a set to zero) should be proportional to the metric tensor
G
ij
= K g
ij
(i, j = 1, 2, 3) . (10.23)
Without loss of generality, the spatial metric can be taken to be of the form
dl
2
= f(r) dr
2
+r
2
do
2
. (10.24)
Imposing the condition (10.23) on the metric (10.24) recovers the spatial FRW metric (10.11).
10.7 FRW metric
The full Friedmann-Robertson-Walker spacetime metric is
ds
2
= dt
2
+a(t)
2
_
dx
2
1 x
2
+x
2
do
2
_
(10.25)
where t is cosmic time, which is the proper time experienced by comoving observers, who remain at rest
in comoving coordinates dx = d = d = 0. Any of the alternative versions of the comoving spatial FRW
metric, equations (10.16), (10.18), (10.20), or (10.22). may be used as the spatial part of the FRW spacetime
metric (10.25).
10.8 Einstein equations for FRW metric
The Einstein equations for the FRW metric (10.25) are
G
t
t
= 3
_

a
2
+
a
2
a
2
_
= 8G ,
G
x
x
= G
= G
=

a
2

a
2
a
2

2 a
a
= 8Gp , (10.26)
where overdots represent dierentiation with respect to cosmic time t, so that for example a da/dt. Note
the trick of one index up, one down, to remove, modulo signs, the distorting eect of the metric on the
Einstein tensor. The Einstein equations (10.26) rearrange to give Friedmanns equations
a
2
a
2
=
8G
3

a
2
,
a
a
=
4G
3
( + 3p) .
(10.27)
Friedmanns two equations (10.27) are fundamental to cosmology. The rst one relates the curvature of
the Universe to the expansion rate a/a and the density . The second one relates the acceleration a/a to
the density plus 3 times the pressure p.
10.9 Newtonian derivation of Friedmann equations 145
10.9 Newtonian derivation of Friedmann equations
10.9.1 Energy equation
Model a piece of the Universe as a ball of radius a and mass M =
4
3
a
3
. Consider a small mass m attracted
by this ball. Conservation of the kinetic plus potential energy of the small mass m implies
1
2
m a
2
GMm
a
=
mc
2
2
, (10.28)
where the quantity on the right is some constant whose value is not determined by this Newtonian treatment,
but which GR implies is as given. The energy equation (10.28) rearranges to
a
2
a
2
=
8G
3

c
2
a
2
, (10.29)
which reproduces the rst Friedmann equation.
10.9.2 First law of thermodynamics
For adiabatic expansion, the rst law of thermodynamics is
dE +p dV = 0 . (10.30)
With E = V and V =
4
3
a
3
, the rst law (10.30) becomes
d(a
3
) +p da
3
= 0 , (10.31)
or, with the derivative taken with respect to cosmic time t,
+ 3( +p)
a
a
= 0 . (10.32)
Dierentiating the rst Friedmann equation in the form
a
2
=
8Ga
2
3
c
2
(10.33)
gives
2 a a =
8G
3
_
a
2
+ 2a a
_
, (10.34)
and substituting from the rst law (10.32) reduces this to
2 a a =
8G
3
a a ( 3p) . (10.35)
Hence
a
a
=
4G
3
( + 3p) , (10.36)
which reproduces the second Friedmann equation.
10.9.3 Comment on the Newtonian derivation
The above Newtonian derivation of Friedmanns equations is only heuristic. A dierent result could have
been obtained if dierent assumptions had been made. If for example the Newtonian gravitational force law
m a = GMm/a
2
were taken as correct, then it would follow that a/a =
4
3
G, which is missing the all
important 3p contribution (without which there would be no ination or dark energy) to Friedmanns second
equation.
It is notable that the rst law of thermodynamics is built in to the Friedmann equations. This implies that
entropy is conserved in FRW Universes. This remains true even when the mix of particles changes, as happens
for example during the epoch of electron-positron annihilation, or during big bang nucleosynthesis. How
then does entropy increase in the real Universe? Through uctuations away from the perfect homogeneity
and isotropy assumed by the FRW metric.
10.10 Hubble parameter
The Hubble parameter H(t) is dened by
H
a
a
. (10.37)
The Hubble parameter H varies in cosmic time t, but is constant in space at xed cosmic time t.
The value of the Hubble parameter today is called the Hubble constant H
0
(the subscript 0 signies
the present time). The Hubble constant is measured from Cepheids and Type Ia supernova to be (Riess et
al. 2005, astro-ph/0503159)
H
0
= 73 4(stat) 5(sys) kms
1
Mpc
1
. (10.38)
The distance d to an object that is receding with the expansion of the universe is proportional to the
cosmic scale factor, d a, and its recession velocity v is consequently proportional to a. The result is
Hubbles law relating the recession velocity v and distance d of distant objects
v = H
0
d . (10.39)
Since it takes light time to travel from a distant object, and the Hubble parameter varies in time, the linear
relation (10.39) breaks down at cosmological distances.
We, in the Milky Way, reside in an overdense region of the Universe that has collapsed out of the general
Hubble expansion of the Universe. The local overdense region of the Universe that has just turned around
from the general expansion and is beginning to collapse for the rst time is called the Local Group of
galaxies. The Local Group consists of about 40 or so galaxies, mostly dwarf and irregular galaxies. It
contains two major spiral galaxies, Andromeda (M31) and the Milky Way, and one mid-sized spiral galaxy
Triangulum (M33). The Local Group is about 1 Mpc in radius.
10.11 Critical density 147
Because of the ubiquity of the Hubble constant in cosmological studies, cosmologists often parameterize
it by the quantity h dened by
h
H
0
100 kms
1
Mpc
1
. (10.40)
10.11 Critical density
The critical density
crit
is dened to be the density required for the Universe to be at, = 0. According
to the rst of Friedmann equations (10.27), this sets
crit

3H
2
8G
. (10.41)
The critical density
crit
, like the Hubble parameter H, evolves with time.
10.12 Omega
Cosmologists designate the ratio of the actual density of the Universe to the critical density
crit
by the
fateful letter , the nal letter of the Greek alphabet,

crit
. (10.42)
With no subscript, denotes the total mass-energy density in all forms. A subscript x on
x
denotes
mass-energy density of type x.
The curvature density
k
, which is not really a form of mass-energy but it is sometimes convenient to treat
Table 10.1 Cosmic inventory
Species (2008)
Dark energy () 0.72 0.02
Non-baryonic cold dark matter (CDM) c 0.234 0.02
Baryonic matter
b
0.046 0.002
Neutrinos < 0.014
Photons (CMB) 5 10
5
Total 1.005 0.006
Curvature
k
0.005 0.006
it as though it were, is dened by
k

3c
2
8Ga
2
(10.43)
and correspondingly
k

k
/
crit
. According to the rst of Friedmanns equations (10.27), the curvature
density
k
satises
k
= 1 . (10.44)
Table 10.1 gives 2008 measurements of in various species, obtained by combining 5-year WMAP CMB
measurements with a variety of other astronomical evidence, including supernovae, big bang nucleosynthesis,
galaxy clustering, weak lensing, and local measurements of the Hubble constant H
0
.
10.13 Redshifting
The spatial translation symmetry of the FRW metric implies conservation of generalized momentum. As you
will show in a problem set, a particle that moves along a geodesic in the radial direction, so that d = d = 0,
has 4-velocity u
satisfying
u
x
= constant . (10.45)
This conservation law implies that the proper momentum p
of a radially moving particle decays as

p
ma
dx
d

1
a
, (10.46)
which is true for both massive and massless particles.
It follows from equation (10.46) that light observed on Earth from a distant object will be redshifted by
a factor
1 +z =
a
0
a
, (10.47)
where a
0
is the present day cosmic scale factor. Cosmologists often refer to the redshift of an epoch, since
the cosmological redshift is an observationally accessible quantity that uniquely determines the cosmic time
of emission.
10.14 Types of mass-energy
The energy-momentum tensor T
of an FRW Universe is necessarily homogeneous and isotropic, by as-

sumption of the cosmological principle, taking the form (note yet again the trick of one index up and one
10.15 Evolution of the cosmic scale factor 149
down to remove the distorting eect of the metric)
T
=
_
_
_
_
_
T
t
t
0 0 0
0 T
r
r
0 0
0 0 T
0
0 0 0 T
_
_
_
_
_
=
_
_
_
_
0 0 0
0 p 0 0
0 0 p 0
0 0 0 p
_
_
_
_
. (10.48)
Table 10.2 gives equations of state p/ for generic species of mass-energy, along with ( + 3p)/, which
determines the gravitational attraction per unit energy, and how the mass-energy varies with cosmic scale
factor, a
n
.
Table 10.2 Properties of universes dominated by various species
Species p/ ( + 3p)/
Radiation 1/3 2 a
4
Matter 0 1 a
3
Curvature 1/3 0 a
2
Vaccum 1 2 a
0
As commented in 23.16 above, the rst law of thermodynamics for adiabatic expansion is built into Fried-
manns equations. In fact the law represents covariant conservation of energy-momentum for the system as
a whole
D
= 0 . (10.49)
As long as species do not convert into each other (for example, no annihilation), covariant energy-momentum
conservation holds individually for each species, so the rst law applies to each species individually, deter-
mining how its energy density varies with cosmic scale factor a. Figure 10.2 illustrates how the energy
densities of various species evolve as a function of scale factor a.
10.15 Evolution of the cosmic scale factor
Given how the energy density of each species evolves with cosmic scale factor a, the rst Friedmann
equation then determines how the cosmic scale factor a(t) itself evolves with cosmic time t. The evolution
equation for a(t) can be cast as an equation for the Hubble parameter H a/a, which in view of the
denition (10.41) of the critical density can be written
H(t)
H
0
=
_

crit
(t)
crit
(t
0
)
_
1/2
. (10.50)
Cosmic scale factor a
M
a
s
s
-
e
n
e
r
g
y
d
e
n
s
i
t
y
a
4
m
a
3
k
a
2
= constant
Figure 10.2 Behavior of the mass-energy density of various species as a function of cosmic time t.
Given the denition (10.43) of the curvature density as the critical density minus the total density, the
critical density
crit
is itself the sum of the densities of all species including the curvature density
crit
=
k
+
species x
x
. (10.51)
Integrating equation (10.50) gives cosmic time t as a function of cosmic scale factor a
t =
_
da
aH
. (10.52)
For example, in the case that the density is comprised of radiation, matter, and vacuum, the critical density
is
crit
=
+
m
+
k
+
, (10.53)
and equation (10.50) is
H(t)
H
0
=
_
a
4
+
m
a
3
+
k
a
2
+
_
1/2
, (10.54)
where
x
represents its value at the present time. The time t, equation (10.52), is then
t =
1
H
0
_
da
a (
a
4
+
m
a
3
+
k
a
2
+
)
1/2
, (10.55)
10.16 Conformal time 151
which is an elliptical integral of the 3rd kind.
If one single species in particular dominates the mass-energy density, then equation (10.55) integrates
easily to give the results in the following table.
Table 10.3 Evolution of cosmic scale factor in universes dominated by various species
Dominant Species a
Radiation t
1/2
Matter t
2/3
Curvature t
Vaccum e
Ht
10.16 Conformal time
Especially when doing cosmological perturbation theory, it is convenient to use conformal time dened
by (with units c temporarily restored)
a d c dt (10.56)
with respect to which the FRW metric is
ds
2
= a()
2
_
d
2
+
dx
2
1 x
2
+x
2
do
2
_
. (10.57)
The term conformal refers to a metric that is multiplied by an overall factor, the conformal factor. In the
FRW metric (10.57), the cosmic scale factor a is the conformal factor.
Conformal time has the property that the speed of light is one in conformal coordinates: light moves
unit comoving distance per unit conformal time. In particular, light moving radially towards an observer at
x
= 0, with d = d = 0, satises
dx
d
= 1 . (10.58)
10.17 Looking back along the lightcone
Since light moves at unit velocity in conformal coordinates, an object at geodesic distance x
that emits
light at conformal time
emit
is observed at conformal time
obs
given by
x
=
obs
emit
. (10.59)
The comoving geodesic distance x
to an object is
x
=
_

obs
emit
d =
_
t
obs
temit
c dt
a
=
_
a
obs
aemit
c da
a
2
H
=
_
z
0
c dz
H
, (10.60)
where the last equation assumes the relation 1 + z = 1/a, valid as long as a is normalized to unity at the
observer (us) at the present time a
obs
= a
0
= 1. In the case that the density is comprised of (curvature and)
radiation, matter, and vacuum, equation (10.60) gives
x
=
c
H
0
_
1
1/(1+z)
da
a
2
(
a
4
+
m
a
3
+
k
a
2
+
)
1/2
, (10.61)
which is an elliptical integral of the 1st kind. Given the geodesic comoving distance x
, the circumferential
comoving distance x then follows as
x =
sinh(
1/2
k
H
0
x
/c)
1/2
k
H
0
/c
. (10.62)
To second order in redshift z,
x x

c
H
0
_
z z
2
_
+
3
4
m
+
1
2
k
_
+...
. (10.63)
The geodesic and circumferential distances x
and x dier at order z

3
.
10.18 Horizon
Light can come from no more distant point than the Big Bang. This distant point denes the horizon of
our Universe, which is located at innite redshift, z = . Equation (10.60) gives the geodesic distance from
us at redshift zero to the horizon as
x
(horizon) =
_

0
c dz
H
(10.64)
where again the cosmic scale factor has been normalized to unity at the present time, a
0
= 1.
Equation (10.64) formally denes the event horizon of the Universe, but the cosmological scale over which
objects can continue to aect each other causally is typically smaller than this (much smaller, post-ination).
It is thus common to dene the cosmological horizon distance at any time as
cosmological horizon distance
c
H
(10.65)
which is roughly the scale over which objects can remain in causal contact.
Exercise 10.2 Then versus now.
10.18 Horizon 153
10
40
10
30
10
20
10
10
10
0
10
10
10
20
10
40
10
30
10
20
10
10
10
0
10
10
10
20
10
30
10
50
10
40
10
30
10
20
10
10
10
0
10
10
Age of the Universe (seconds)
Age of the Universe (years)
S
i
z
e
o
f
t
h
e
U
n
i
v
e
r
s
e
(
m
e
t
e
r
s
)
Figure 10.3 Cosmic scale factor a and cosmological horizon distance c/H as a function of cosmic time t.
1. Prove that
_

0
x
n1
dx
e
x
+ 1
=
_
1 2
1n
_
_

0
x
n1
dx
e
x
1
. (10.66)
[Hint: Use the fact that (e
x
+ 1)(e
x
1) = (e
2x
1).] Hence argue that the ratios of energy, entropy,
and number densities of relativistic fermionic (f) to relativistic bosonic (b) species in thermodynamic
equilibrium at the same temperature are
b
=
s
f
s
b
=
7
8
,
n
f
n
b
=
3
4
. (10.67)
[Hint: The proper entropy density of each relativistic species is s = ( +p)/T = (4/3)/T.]
2. Weak interactions were fast enough to keep neutrinos in thermodynamic equilibrium with photons, elec-
trons, and positrons up to just before e e annihilation, but then neutrinos decoupled. Argue that conser-
vation of comoving entropy implies
a
3
T
3
_
g
+
7
8
g
e
_
= T
3
, (10.68a)
a
3
T
3
g
= T
3
, (10.68b)
10
40
10
30
10
20
10
10
10
0
10
10
10
20
10
5
10
0
10
5
10
10
10
15
10
20
10
25
10
30
10
35
10
50
10
40
10
30
10
20
10
10
10
0
10
10
R
a
d
i
a
t
i
o
n
T
e
m
p
e
r
a
t
u
r
e
o
f
t
h
e
U
n
i
v
e
r
s
e
(
K
e
l
v
i
n
)
10
40
10
30
10
20
10
10
10
0
10
10
10
20
10
30
10
20
10
10
10
0
10
10
10
20
10
30
10
40
10
50
10
60
10
70
10
80
10
90
10
100
10
50
10
40
10
30
10
20
10
10
10
0
10
10
M
a
s
s
-
E
n
e
r
g
y
D
e
n
s
i
t
y
o
f
t
h
e
U
n
i
v
e
r
s
e
(
k
g
/
m
3
)
Figure 10.4 (Top) Temperature T, and (bottom) mass-energy density , of the Universe as a function of
cosmic time t.
10.18 Horizon 155
where the left hand sides refer to quantities before e e annihilation, which happened at T 1 MeV 10
10
K,
and the right hand sides to quantities after e e annihilation (including today). Deduce the ratio of neutrino
to photon temperatures today,
T
. (10.69)
Does the temperature ratio (10.69) depend on the number of neutrino types? What is the neutrino
temperature today in K, if the photon temperature today is 2.725 K?
3. The energy, entropy, and number densities of relativistic particles today are, with units restored (energy
density in units energy/volume; entropy and number density s and n in units 1/volume),
= g
,0
2
(kT
0
)
4
30c
3
3
, s = g
s,0
2
2
(kT
0
)
3
45c
3
3
, n = g
n,0
(3)(kT
0
)
3
2
c
3
3
, (10.70)
where T
0
= 2.725 K is the CMB temperature today, (3) = 1.2020569 is a Riemann zeta function, and g
,0
,
g
s,0
, and g
n,0
denote the energy-, entropy-, and number-weighted eective number of relativistic species
today, normalized to 1 per bosonic degree of freedom. What are the arithmetic values of g
,0
, g
s,0
, and
g
n,0
if the relativistic species consist of photons and three species of neutrino? What is the energy density
r
of relativistic particles today relative to the critical density? [Hint: Dont forget to take into account
the fact that the neutrino temperature today diers from the photon temperature.]
4. Evidence for neutrino oscillations from the MINOS experiment (2008, http://www-numi.fnal.gov/
PublicInfo/forscientists.html) indicates that at least one neutrino type has mass m
>
0.05 eV. At
what redshift z
would such a neutrino become non-relativistic? If neutrinos are non-relativistic, what is

the neutrino density
relative to the critical density, in terms of the sum of the neutrino masses
?
Which of the eective number of relativistic species today g
,0
, g
s,0
, and g
n,0
is changed if some neutrinos
are non-relativistic today?
5. Use entropy conservation to argue that the ratio of the photon temperature T at redshift z in the early
Universe to the photon temperature T
0
today is
T
T
0
= (1 +z)
_
g
s,0
g
s
_
1/3
. (10.71)
What is g
s
in terms of the numbers g
b
and g
f
of relativistic boson and fermion types, if all species were
at the same temperature T?
Solution. The ratio of neutrino to photon temperatures post e e annihilation is
T
=
_
g
+
7
8
g
e
_
1/3
=
_
4
11
_
1/3
. (10.72)
The Cosmic Neutrino Background temperature is
T
=
_
4
11
_
1/3
2.725 K = 1.945 K . (10.73)
With 2 bosonic degrees of freedom from photons, and 6 fermionic degrees of freedom from 3 relativistic
neutrino types, the eective energy-, entropy-, and number-weighted number of relativistic degrees of freedom
is
g
,0
= g
+
_
T
_
4
7
8
g
= 2 +
_
4
11
_
4/3
7
8
6 = 3.36 , (10.74a)
g
s,0
= g
+
_
T
_
3
7
8
g
= 2 +
4
11
7
8
6 =
43
11
= 3.91 , (10.74b)
g
n,0
= g
+
_
T
_
3
3
4
g
= 2 +
4
11
3
4
6 =
40
11
= 3.64 . (10.74c)
The redshift at which a neutrino of mass m
becomes non-relativistic is
1 +z
=
m
= 300
_
m
0.05 eV
_
. (10.75)
If some neutrinos are non-relativistic, then the neutrino density
today is related to the sum
of
neutrino masses by
=
8G
3H
2
0
= 5.3 10
4
_
m
0.05 eV
__
h
0.71
_
2
. (10.76)
g
,0
is changed if some neutrinos are non-relativistic today, but g
s,0
and g
n,0
remain unchanged. If just one
of the neutrino types is massive, and the other two are relativistic, then
g
,0
= 2 +
_
4
11
_
4/3
7
8
4 = 2.91 . (10.77)
The radiation density
r
today, including photons and neutrinos, is
r
=
8G
r
3c
2
H
2
0
= 1.236 10
5
g
,0
h
2
= 8.2 10
5
_
g
,0
3.36
_
_
h
0.71
_
2
. (10.78)
If the temperatures of all species are equal, then the entropy-weighted eective number of relativistic species
is
g
s
= g
b
+
7
8
g
f
. (10.79)
PART FIVE
TETRAD APPROACH TO GENERAL RELATIVITY
Concept Questions
1. The vierbein has 16 degrees of freedom instead of the 10 degrees of freedom of the metric. What do the
extra 6 degrees of freedom correspond to?
2. Tetrad transformations are dened to be Lorentz transformations. Dont general coordinate transfor-
mations already include Lorentz transformations as a particular case, so arent tetrad transformations
redundant?
3. What does coordinate gauge-invariant mean? What does tetrad gauge-invariant mean?
4. Is the coordinate metric g
tetrad gauge-invariant?
5. What does a directed derivative
m
mean physically?
6. Is the directed derivative
m
coordinate gauge-invariant?
7. Is the tetrad metric
mn
coordinate gauge-invariant? Is it tetrad gauge-invariant?
8. What is the tetrad-frame 4-velocity u
m
of a person at rest in an orthonormal tetrad frame?
9. If the tetrad frame is accelerating (not in free-fall), which of the following is true/false?
a. Does the tetrad-frame 4-velocity u
m
of a person continuously at rest in the tetrad frame change with
time?
0
u
m
= 0? D
0
u
m
= 0?
b. Do the tetrad axes
m
change with time?
0
m
= 0? D
0
m
= 0?
c. Does the tetrad metric
mn
change with time?
0
mn
= 0? D
0
mn
= 0?
d. Do the covariant components u
m
of the 4-velocity of a person continuously at rest in the tetrad frame
change with time?
0
u
m
= 0? D
0
u
m
= 0?
10. Suppose that p =
m
p
m
is a 4-vector. Is the proper rate of change of the proper components p
m
measured
by an observer equal to the directed time derivative
0
p
m
or to the covariant time derivative D
0
p
m
? What
about the covariant components p
m
of the 4-vector? [Hint: The proper contravariant components of the
4-vector measured by an observer are p
m

m
p where
m
are the contravariant locally inertial rest
axes of the observer. Similarly the proper covariant components are p
m

m
p.]
11. A person with two eyes separated by proper distance
n
observes an object. The observer observes the
photon 4-vector from the object to be p
m
. The observer uses the dierence p
m
in the two 4-vectors
detected by the two eyes to infer the binocular distance to the object. Is the dierence p
m
in photon
4-vectors detected by the two eyes equal to the directed derivative
n
n
p
m
or to the covariant derivative
n
D
n
p
m
?
12. Suppose that p
m
is a tetrad 4-vector. Parallel-transport the 4-vector by an innitesimal proper distance
n
. Is the change in p
m
measured by an ensemble of observers at rest in the tetrad frame equal to the
directed derivative
n
n
p
m
or to the covariant derivative
n
D
n
p
m
? [Hint: What if rest means that the
observer at each point is separately at rest in the tetrad frame at that point? What if rest means that
the observers are mutually at rest relative to each other in the rest frame of the tetrad at one particular
point?]
13. What is the physical signicance of the fact that directed derivatives fail to commute?
14. Physically, what do the tetrad connection coecients
kmn
mean?
15. What is the physical signicance of the fact that
kmn
is antisymmetric in its rst two indices (if the
tetrad metric
mn
is constant)?
16. Are the tetrad connections
kmn
coordinate gauge-invariant?
Whats important?
This part of the notes describes the tetrad formalism of GR.
1. Why tetrads? Because physics is clearer in a locally inertial frame than in a coordinate frame.
2. The primitive object in the tetrad formalism is the vierbein e
m
, in place of the metric in the coordinate

formalism.
3. Written suitably, for example as equation (11.9), a metric ds
2
encodes not only the metric coecients g
,
but a full (inverse) vierbein e
m
, through ds
2
=
mn
e
m
dx
e
n
dx
.
4. The tetrad road from vierbein to energy-momentum is similar to the coordinate road from metric to
energy-momentum, albeit a little more complicated.
5. In the tetrad formalism, the directed derivative
m
is the analog of the coordinate partial derivative /x
of the coordinate formalism. Directed derivatives

m
do not commute, whereas coordinate derivatives
/x
do commute.
11
The tetrad formalism
11.1 Tetrad
A tetrad (greek foursome)
m
(x) is a set of axes
m

0
,
1
,
2
,
3
(11.1)
attached to each point x
of spacetime. The common case is that of an orthonormal tetrad, where the axes
form a locally inertial frame at each point, so that the scalar products of the axes constitute the Minkowski
metric
mn
m

n
=
mn
. (11.2)
However, other tetrads prove useful in appropriate circumstances. There are spinor tetrads, null tetrads
(notably the Newman-Penrose double null tetrad), and others (indeed, the basis of coordinate tangent
vectors g
is itself a tetrad). In general, the tetrad metric is some symmetric matrix

mn
m

n

mn
. (11.3)
Andrews convention:
latin (black) dummy indices label tetrad frames.
greek (brown) dummy indices label coordinate frames.
Why introduce tetrads?
1. The physics is more transparent when expressed in a locally inertial frame (or some other frame adapted
to the physics), as opposed to the coordinate frame, where Salvador Dali rules.
2. If you want to consider spin-
1
2
particles and quantum physics, you better work with tetrads.
3. For good reason, much of the GR literature works with tetrads, so its useful to understand them.
11.2 Vierbein
The vierbein (German four-legs, or colloquially, critter) e
m
is dened to be the matrix that transforms

between the tetrad frame and the coordinate frame (note the placement of indices: the tetrad index m comes
11.3 The metric encodes the vierbein 163
rst, then the coordinate index )
m
= e
m
. (11.4)
The vierbein is a 4 4 matrix, with 16 independent components. The inverse vierbein e
m
is dened to be
the matrix inverse of the vierbein e
m
, so that
e
m
e
m
, e
m
e
n
=
n
m
. (11.5)
Thus equation (11.4) inverts to
g
= e
m
m
. (11.6)
11.3 The metric encodes the vierbein
The scalar spacetime distance is
ds
2
= g
dx
dx
= g
dx
dx
=
mn
e
m
e
n
dx
dx
(11.7)
from which it follows that the coordinate metric g
is
g
=
mn
e
m
e
n
. (11.8)
The shorthand way in which metrics are commonly written encodes not only a metric but also an inverse
vierbein, hence a tetrad. For example, the Schwarzschild metric
ds
2
=
_
1
2M
r
_
dt
2
+
_
1
2M
r
_
1
dr
2
+r
2
d
2
+r
2
sin
2
d
2
(11.9)
takes the form (11.7) with an orthonormal (Minkowski) tetrad metric
mn
=
mn
, and an inverse vierbein
encoded in the dierentials
e
0
dx
=
_
1
2M
r
_
1/2
dt , (11.10a)
e
1
dx
=
_
1
2M
r
_
1/2
dr , (11.10b)
e
2
dx
= r d , (11.10c)
e
3
dx
= r sin d , (11.10d)
Explicitly, the inverse vierbein of the Schwarzschild metric is the diagonal matrix
e
m
=
_
_
_
_
(1 2M/r)
1/2
0 0 0
0 (1 2M/r)
1/2
0 0
0 0 r 0
0 0 0 r sin
_
_
_
_
, (11.11)
164 The tetrad formalism
and the corresponding vierbein is (note that, because the tetrad index is always in the rst place and the
coordinate index is always in the second place, the matrices as written are actually inverse transposes of
each other, not just inverses)
e
m
=
_
_
_
_
(1 2M/r)
1/2
0 0 0
0 (1 2M/r)
1/2
0 0
0 0 1/r 0
0 0 0 1/(r sin)
_
_
_
_
. (11.12)
Concept question 11.1 Schwarzschild vierbein. The components e
0
t
and e
1
r
of the Schwarzschild
vierbein (11.12) are imaginary inside the horizon. What does this mean? Is the vierbein still valid inside
the horizon?
11.4 Tetrad transformations
Tetrad transformations are transformations that preserve the fundamental property of interest, for example
the orthonormality, of the tetrad. For all tetrads of interest in these notes, which includes not only orthonor-
mal tetrads, but also spinor tetrads and null tetrads (but not coordinate-based tetrads), tetrad transfor-
mations are Lorentz transformations. Hereafter these notes will presume that a tetrad transformation is a
Lorentz transformation. The Lorentz transformation may be, and usually is, a dierent transformation at
each point. Tetrad transformations rotate the tetrad axes
k
at each point by a Lorentz transformation
L
k
m
, while keeping the background coordinates x
unchanged:
k
= L
k
m
m
. (11.13)
In the case that the tetrad axes
k
are orthonormal, with a Minkowski metric, the Lorentz transformation
matrices L
k
m
in equation (11.13) take the familiar special relativistic form, but the linear matrices L
k
m
in
equation (11.13) signify a Lorentz transformation in any case.
In all the cases of interest, including orthonormal, spinor, and null tetrads, the tetrad metric
mn
is
constant. Lorentz transformations are precisely those transformations that leave the tetrad metric unchanged
kl
=
l
= L
k
m
L
l
n
m

n
= L
k
m
L
l
n
mn
=
kl
. (11.14)
Exercise 11.2 Generators of Lorentz transformations are antisymmetric. From the condition that
the tetrad metric
mn
is unchanged by a Lorentz transformation, show that the generator of an innitesimal
Lorentz transformation is an antisymmetric matrix. Is this true only for an orthonormal tetrad, or is it true
more generally?
Solution. An innitesimal Lorentz transformation is the sum of the unit matrix and an innitesimal piece
L
k
m
, the generator of the innitesimal Lorentz transformation,
L
k
m
=
m
k
+ L
k
m
. (11.15)
11.5 Tetrad Tensor 165
Under such an innitesimal Lorentz transformation, the tetrad metric transforms to
kl
= (
m
k
+ L
k
m
)(
n
l
+ L
l
n
)
mn

kl
+ L
kl
+ L
lk
, (11.16)
which by proposition equals the original tetrad metric
kl
, equation (11.14). It follows that
L
kl
+ L
lk
= 0 , (11.17)
that is, the generator L
kl
is antisymmetric, as claimed.
11.5 Tetrad Tensor
In general, a tetrad-frame tensor A
kl...
mn...
is an object that transforms under tetrad (Lorentz) transforma-
tions (11.13) as
A
kl...
mn...
= L
k
a
L
l
b
... L
m
c
L
n
d
... A
ab...
cd...
. (11.18)
11.6 Raising and lowering indices
In the coordinate approach to GR, coordinate indices were lowered and raised with the coordinate metric
g
and its inverse g
. In the tetrad formalism there are two kinds of indices, tetrad indices and coordinate
indices, and they ip around as follows:
1. Lower and raise coordinate indices with the coordinate metric g
and its inverse g
;
2. Lower and raise tetrad indices with the tetrad metric
mn
and its inverse
mn
;
3. Switch between coordinate and tetrad frames with the vierbein e
m
and its inverse e

m
.
The kinds of objects for which this ippery is valid are called tensors. Tensors with only tetrad indices,
such as the tetrad axes
m
or the tetrad metric
mn
, are called tetrad tensors, and they remain unchanged
under coordinate transformations. Tensors with only coordinate indices, such as the coordinate tangent
axes g
or the coordinate metric g
, are called coordinate tensors, and they remain unchanged under tetrad
transformations. Tensors may also be mixed, such as the vierbein e
m
. And of course just because something

has an index, greek or latin, does not make it a tensor: a tensor is a tensor if any only if it transforms like
a tensor.
11.7 Gauge transformations
Gauge transformations are transformations of the coordinates or tetrad. Such transformations do not
change the underlying spacetime.
Quantities that are unchanged by a coordinate transformation are coordinate gauge-invariant. Quan-
tities that are unchanged under a tetrad transformation are tetrad gauge-invariant. For example, tetrad
tensors are coordinate gauge-invariant, while coordinate tensors are tetrad gauge-invariant.
Tetrad transformations have the 6 degrees of freedom of Lorentz transformations, with 3 degrees of freedom
in spatial rotations, and 3 more in Lorentz boosts. General coordinate transformations have 4 degrees of
freedom. Thus there are 10 degrees of freedom in the choice of tetrad and coordinate system. The 16 degrees
of freedom of the vierbein, minus the 10 degrees of freedom from the transformations of the tetrad and
coordinates, leave 6 physical degrees of freedom in spacetime, the same as in the coordinate approach to GR,
which is as it should be.
11.8 Directed derivatives
Directed derivatives
m
are dened to be the directional derivatives along the axes
m
m

m
=
m
g
= e
m
is a tetrad 4-vector . (11.19)

The directed derivative
m
is independent of the choice of coordinates, as signalled by the fact that it has
only a tetrad index, no coordinate index.
Unlike coordinate derivatives /x
, directed derivatives
m
do not commute. Their commutator is
[
m
,
n
] =
_
e
m
, e
n
_
= e
m
e
n
e
n
e
m
= (d
k
nm
d
k
mn
)
k
is not a tetrad tensor (11.20)
where d
lmn

lk
d
k
mn
is the vierbein derivative
d
lmn

lk
e
k
e
n
e
m
is not a tetrad tensor . (11.21)

Since the vierbein and inverse vierbein are inverse to each other, an equivalent denition of d
lmn
in terms of
the inverse vierbein is
d
lmn

lk
e
m
e
n
e
k

11.9 Tetrad covariant derivative
The derivation of tetrad covariant derivatives D
m
follows precisely the analogous derivation of coordinate
covariant derivatives D
. The tetrad-frame formulae look entirely similar to the coordinate-frame formu-

lae, with the replacement of coordinate partial derivatives by directed derivatives, /x

m
, and the
replacement of coordinate-frame connections by tetrad-frame connections

k
mn
. There are two things
to be careful about: rst, unlike coordinate partial derivatives, directed derivatives
m
do not commute;
11.9 Tetrad covariant derivative 167
and second, neither tetrad-frame nor coordinate-frame connections are tensors, and therefore it should be
no surprise that the tetrad-frame connections
lmn
are not related to the coordinate-frame connections
by the usual vierbein transformations. Rather, the tetrad and coordinate connections are related by
equation (11.32).
If is a scalar, then
m
is a tetrad 4-vector. The tetrad covariant derivative of a scalar is just the
directed derivative
D
m
=
m
is a tetrad 4-vector . (11.23)
If A
m
is a tetrad 4-vector, then
n
A
m
is not a tetrad tensor, and
n
A
m
is not a tetrad tensor. But the
4-vector A =
m
A
m
, being by construction invariant under both tetrad and coordinate transformations, is
a scalar, and its directed derivative is therefore a 4-vector
n
A =
n
(
m
A
m
) is a tetrad 4-vector
=
m
n
A
m
+ (
n
m
)A
m
=
m
n
A
m
+
k
mn
k
A
m
(11.24)
where the tetrad-frame connection coecients,
k
mn
, also known as Ricci rotation coecients (or, in
the context of Newman-Penrose tetrads, spin coecients) are dened by
m

k
mn
k
Equation (11.24) shows that
n
A =
k
(D
n
A
k
) is a tetrad tensor (11.26)
where D
n
A
k
is the covariant derivative of the contravariant 4-vector A
k
D
n
A
k

n
A
k
+
k
mn
A
m
is a tetrad tensor . (11.27)
Similarly,
n
A =
k
(D
n
A
k
) (11.28)
where D
n
A
k
is the covariant derivative of the covariant 4-vector A
k
D
n
A
k

n
A
k
m
kn
A
m
In general, the covariant derivative of a tensor is
D
a
A
kl...
mn...
=
a
A
kl...
mn...
+
k
ba
A
bl...
mn...
+
l
ba
A
kb...
mn...
+...
b
ma
A
kl...
bn...
b
na
A
kl...
mb...
... (11.30)
with a positive term for each contravariant index, and a negative term for each covariant index.
11.10 Relation between tetrad and coordinate connections
The relation between the tetrad connections
k
mn
and their coordinate counterparts
follows from
k
mn
k
=
n
m
= e
n
e
m
is not a tetrad tensor

= e
n
e
m
+e
n
e
m
= d
kmn
e
k
g
+e
n
e
m
. (11.31)
Thus the relation is
lmn
d
lmn
= e
l
e
m
e
n
is not a tetrad tensor (11.32)

where
lmn

lk

k
mn
. (11.33)
11.11 Torsion tensor
The torsion tensor S
m
kl
, which GR assumes to vanish, is dened in the usual way by the commutator of
the covariant derivative acting on a scalar
[D
k
, D
l
] = S
m
kl

m
The expression (11.29) for the covariant derivatives coupled with the commutator (11.20) of directed deriva-
tives shows that the torsion tensor is
S
m
kl
=
m
kl
m
lk
d
m
kl
+d
m
lk
is a tetrad tensor (11.35)
where d
m
kl
are the vierbein derivatives dened by equation (11.21). The torsion tensor S
m
kl
is antisymmetric
in k l, as is evident from its denition (11.34).
11.12 No-torsion condition
GR assumes vanishing torsion. Then equation (11.35) implies the no-torsion condition
mkl
d
mkl
=
mlk
d
mlk
In view of the relation (11.32) between tetrad and coordinate connections, the no-torsion condition (11.36) is
equivalent to the usual symmetry condition
on the coordinate frame connections, as it should

be.
11.13 Antisymmetry of the connection coecients 169
11.13 Antisymmetry of the connection coecients
The directed derivative of the tetrad metric is
lm
=
n
(
l

m
)
=
l

n
m
+
m

n
l
=
lmn
+
mln
. (11.37)
In most cases of interest, including orthonormal, spinor, and null tetrads, the tetrad metric is chosen to be a
constant. For example, if the tetrad is orthonormal, then the tetrad metric is the Minkowski metric, which
is constant, the same everywhere. If the tetrad metric is constant, then all derivatives of the tetrad metric
vanish, and then equation (11.37) shows that the tetrad connections are antisymmetric in their rst two
indices
lmn
=
mln
. (11.38)
This antisymmetry reects the fact that
lmn
is the generator of a Lorentz transformation for each n.
11.14 Connection coecients in terms of the vierbein
In the general case of non-constant tetrad metric, and non-vanishing torsion, the following manipulation
lm
+
m
ln
mn
=
lmn
+
mln
+
lnm
+
nlm
mnl
nml
(11.39)
= 2
lmn
S
lmn
S
mnl
S
nml
d
lmn
+d
lnm
d
mnl
+d
mln
d
nml
+d
nlm
implies that the tetrad connections
lmn
are given in terms of the derivatives
n
lm
of the tetrad metric,
the torsion S
lmn
, and the vierbein derivatives d
lmn
by
lmn
=
1
2
(
n
lm
+
m
ln
mn
+S
lmn
+S
mnl
+S
nml
+d
lmn
d
lnm
+d
mnl
d
mln
+d
nml
d
nlm
) is not a tetrad tensor . (11.40)
If torsion vanishes, as GR assumes, and if furthermore the tetrad metric is constant, then equation (11.40)
simplies to the following expression for the tetrad connections in terms of the vierbein derivatives d
lmn
dened by (11.21)
lmn
=
1
2
(d
lmn
d
lnm
+d
mnl
d
mln
+d
nml
d
nlm
) is not a tetrad tensor . (11.41)
This is the formula that allows connection coecients to be calculated from the vierbein.
11.15 Riemann curvature tensor
The Riemann curvature tensor R
klmn
is dened in the usual way by the commutator of the covariant
derivative acting on a contravariant 4-vector. In the presence of torsion,
[D
k
, D
l
] A
m
S
n
kl
D
n
A
m
+R
klmn
A
n
If torsion vanishes, as GR assumes, then the denition (11.42) reduces to
[D
k
, D
l
] A
m
R
klmn
A
n
The expression (11.29) for the covariant derivative coupled with the torsion equation (11.34) yields the
following formula for the Riemann tensor in terms of connection coecients, for the general case of non-
vanishing torsion:
R
klmn
=
k
mnl
mnk
+
a
ml
ank
a
mk
anl
+ (
a
kl
a
lk
S
a
kl
)
mna
The formula has extra terms (
a
kl
a
lk
S
a
kl
)
mna
compared to the usual formula for the coordinate-frame
Riemann tensor R
. If torsion vanishes, as GR assumes, then

R
klmn
=
k
mnl
mnk
+
a
ml
ank
a
mk
anl
+ (
a
kl
a
lk
)
mna
The symmetries of the tetrad-frame Riemann tensor are the same as those of the coordinate-frame Riemann
tensor. For vanishing torsion, these are
R
([kl][mn])
, (11.46)
R
klmn
+R
knlm
+R
kmnl
= 0 . (11.47)
Exercise 11.3 Riemann tensor. From the denition (11.42), derive the expression (11.44) for the
Riemann tensor. Show that, in addition to the antisymmetry in kl which follows immediately from the
denition (11.42), the Riemann tensor R
klmn
is antisymmetric in the indices mn. [Hint: Start by expanding
out the denition (11.42) using the denition (11.30) of the covariant derivative. You will nd it easier to
derive an expression for the Riemann tensor with one index raised, such as R
klm
n
, but you should resist the
temptation to leave it there, because the symmetries of the Riemann tensor are obscured when one index is
raised. To switch to all lowered indices, you will need to convert terms such as
k
n
ml
by
n
ml
=
k
(
np
pml
) =
np
pml
+
pml

k
np
. (11.48)
You should show that the directed derivative
k
np
in this expression is related to tetrad connections through
a formula similar to equation (11.37)
np
=
np
k
pn
k
, (11.49)
which you should recognize as equivalent to D
k
np
= 0. To complete the derivation, show that
k
(
mnl
+
nml
)
l
(
mnk
+
nmk
) = [
k
,
l
]
mn
= (
a
lk
a
kl
+S
a
kl
)(
mna
+
nma
) . (11.50)
The antisymmetry of R
klmn
in mn follows from equation (11.50).]
11.16 Ricci, Einstein, Bianchi 171
11.16 Ricci, Einstein, Bianchi
The usual suite of formulae leading to Einsteins equations apply. Since all the quantities are tensors, and
all the equations are tensor equations, their form follows immediately from their coordinate counterparts.
Ricci tensor:
R
km

ln
R
klmn
. (11.51)
Ricci scalar:
R
km
R
km
. (11.52)
Einstein tensor:
G
km
R
km
1
2

km
R . (11.53)
Einsteins equations:
G
km
= 8GT
km
. (11.54)
Bianchi identities:
D
k
R
lmnp
+D
l
R
mknp
+D
m
R
klnp
= 0 , (11.55)
which most importantly imply covariant conservation of the Einstein tensor, hence conservation of energy-
momentum
D
k
T
km
= 0 . (11.56)
11.17 Electromagnetism
11.17.1 Electromagnetic potential and eld
The electromagnetic eld is derivable from an electromagnetic 4-potential A
m
,
A
m
= (, A) . (11.57)
The electromagnetic eld is a bivector eld (an antisymmetric tensor) F
mn
,
F
mn
D
n
A
m
D
m
A
n
. (11.58)
If torsion vanishes, as general relativity assumes, then the covariant derivatives in the denition (11.58) can
be replaced by directed derivatives,
F
mn
=
n
A
m
m
A
n
. (11.59)
If the tetrad is orthonormal, then the traditional description of the electromagnetic eld in terms of electric
and magnetic elds E and B is valid:
E =
t
A , B = A , (11.60)
where
x
,
y
,
z
denotes the spatial tetrad directed derivative 3-vector. In an orthonormal tetrad
t
,
x
,
y
,
z
, the 6 components of the electromagnetic eld F
mn
are related to the electric and magnetic
elds E = E
x
, E
y
, E
z
and B = B
x
, B
y
, B
z
by
F
mn
=
_
_
_
_
0 E
x
E
y
E
z
E
x
0 B
z
B
y
E
y
B
z
0 B
x
E
z
B
y
B
x
0
_
_
_
_
, F
mn
=
_
_
_
_
0 E
x
E
y
E
z
E
x
0 B
z
B
y
E
y
B
z
0 B
x
E
z
B
y
B
x
0
_
_
_
_
. (11.61)
11.17.2 Lorentz force law
In the presence of an electromagnetic eld F
mn
, the general relativistic equation of motion for the 4-velocity
u
m
dx
m
/d of a particle of mass m and charge q is modied by the addition of a Lorentz force qF
n
m
u
n
m
Du
m
D
= qF
n
m
u
n
. (11.62)
In the absence of gravitational elds, so D/D = d/d, and with u
m
= u
t
1, v where v is the 3-velocity,
the spatial components of equation (11.62) reduce to [note that d/dt = (1/u
t
) d/d]
m
du
t
v
dt
= q (E +v B) (11.63)
which is the classical special relativistic Lorentz force law. The signs in the expression (11.61) for F
mn
in
terms of E = E
x
, E
y
, E
z
and B = B
x
, B
y
, B
z
are arranged to agree with the classical law (11.63).
11.17.3 Maxwells equations
The source-free Maxwells equations are
D
l
F
mn
+D
m
F
nl
+D
n
F
lm
= 0 , (11.64)
while the sourced Maxwells equations are
D
m
F
mn
= 4j
n
, (11.65)
where j
n
is the electric 4-current. The sourced Maxwells equations (11.65) coupled with the antisymmetry
of the electromagnetic eld tensor F
mn
ensure conservation of electric charge
D
n
j
n
= 0 . (11.66)
In at space with a Minkowski metric, the covariant derivatives simplify to ordinary derivatives, D
m

m
/x
m
, and the source-free Maxwells equations (11.64) reduce to the traditional form
B = 0 , E +
B
t
= 0 , (11.67)
11.17 Electromagnetism 173
while the sourced Maxwells equations reduce to
E = 4q , B
E
t
= 4j , (11.68)
where the electric charge density q and the electric current density j are the time and space components of
the electric 4-current
j
n
= q, j . (11.69)
11.17.4 Electromagnetic energy-momentum tensor
mn
e
of an electromagnetic eld F
mn
is
T
mn
e
=
1
4
_
F
m
k
F
nk
1
4
mn
F
kl
F
kl
_
. (11.70)
12
More on the tetrad formalism

This chapter presents some more advanced aspects of the tetrad formalism. It discusses spinor tetrads
(12.1) and Newman-Penrose tetrads (12.2). The chapter goes on to show how the elds that describe
electromagnetic (12.3) and gravitational (12.4) waves have a natural and insightful complex structure that
is brought out in a Newman-Penrose tetrad. The Newman-Penrose formalism provides a natural context for
the Petrov classication of the Weyl tensor (12.5), and for the derivation of the Raychaudhuri equations
(12.6) which imply the focussing theorem (12.7) that is a key ingredient of the Penrose-Hawking singularity
theorems.
12.1 Spinor tetrad formalism
In quantum mechanics, fundamental particles have spin. The 3 generations of leptons (electrons, muons,
tauons, and their respective neutrino partners) and quarks (up, strange, top, and their down, charm, and
bottom partners) have spin
1
2
(in units = 1). The carrier particles of the electromagnetic force (photons),
the weak force (the W
and Z bosons), and the colour force (the 8 gluons), have spin 1. The carrier of the
gravitational force, the graviton, is expected to have spin 2, though as of 2010 no gravitational wave, let
alone its quantum, the graviton, has been detected.
General relativity is a classical, not quantum, theory. Nevertheless the spin properties of classical waves,
such as electromagnetic or gravitational waves, are already apparent classically.
12.1.1 Spinor tetrad
A systematic way to project objects into spin components is to work in a spinor tetrad. As will become
apparent below, equation (12.5), spin describes how an object transforms under rotation about some preferred
axis. In the case of an electromagnetic or gravitational wave, the natural preferred axis is the direction of
propagation of the wave. With respect to the direction of propagation, electromagnetic waves prove to have
two possible spins, or helicities, 1, while gravitational waves have two possible spins, or helicities, 2. A
preferred axis might also be set by an experimenter who chooses to measure spin along some particular
12.1 Spinor tetrad formalism 175
direction. The following treatment takes the preferred direction to lie along the z-axis
z
, but there is no
loss of generality in making this choice.
Start with an orthonormal tetrad
t
,
x
,
y
,
z
. If the preferred tetrad axis is the z-axis
z
, then the
spinor tetrad axes
+
,
are dened to be complex combinations of the transverse axes

x
,
y
,
+

1
2
(
x
+i
y
) , (12.1a)
2
(
x
i
y
) . (12.1b)
The tetrad metric of the spinor tetrad
t
,
z
,
+
,
is
mn
=
_
_
_
_
1 0 0 0
0 1 0 0
0 0 0 1
0 0 1 0
_
_
_
_
. (12.2)
Notice that the spinor axes
+
,
are themselves null,

+

+
=
= 0, whereas their scalar product

with each other is non-zero
+

= 1. The null character of the spinor axes is what makes spin especially
well-suited to describing elds, such as electromagnetism and gravity, that propagate at the speed of light.
An even better trick in dealing with elds that propagate at the speed of light is to work in a Newman-Penrose
tetrad, 12.2, in which all 4 tetrad axes are taken to be null.
12.1.2 Transformation of spin under rotation about the preferred axis
Under a right-handed rotation by angle about the preferred axis
z
, the transverse axes
x
and
y
transform as SIGN?!
x
cos
x
sin
y
,
y
sin
x
+ cos
y
. (12.3)
It follows that the spinor axes
+
and
transform under a right-handed rotation by angle about

z
as
e
i
. (12.4)
The transformation (12.4) identies the spinor axes
+
and
as having spin +1 and 1 respectively.

12.1.3 Spin
More generally, an object can be dened as having spin s if it varies by
e
si
(12.5)
under a right-handed rotation by angle about the preferred axis
z
. Thus an object of spin s is unchanged
by a rotation of 2/s about the preferred axis. A spin-0 object is symmetric about the
z
axis, unchanged
by a rotation of any angle about the axis. The
z
axis itself is spin-0, as is the time axis
t
.
176

The components of a tensor in a spinor tetrad inherit spin properties from that of the spinor basis. The
general rule is that the spin s of any tensor component is equal to the number of + covariant indices minus
the number of covariant indices:
spin s = number of + minus covariant indices . (12.6)
12.1.4 Spin ip
Under a reection through the y-axis, the spinor axes swap:
, (12.7)
which may also be accomplished by complex conjugation. Reection through the y-axis, or equivalently
complex conjugation, changes the sign of all spinor indices of a tensor component
+ . (12.8)
In short, complex conjugation ips spin, a pretty feature of the spinor formalism.
12.1.5 Spin versus spherical harmonics
In physical problems, such as in cosmological perturbations, or in perturbations of spherical black holes, or
in the hydrogen atom, spin often appears in conjunction with an expansion in spherical harmonics. Spin
should not be confused with spherical harmonics.
Spin and spherical harmonics appear together whenever the problem at hand has a symmetry under the 3D
special orthogonal group SO(3) of spatial rotations (special means of unit determinant; the full orthogonal
group O(3) contains in addition the discrete transformation corresponding to reection of one of the axes,
which ips the sign of the determinant). Rotations in SO(3) are described by 3 Euler angles , , . Spin
is associated with the Euler angle . The usual spherical harmonics Y
m
(, ) are the spin-0 eigenfunctions
of SO(3). The eigenfunctions of the full SO(3) group are the spin harmonics SIGN?
s
Y
m
(, , ) =
ms
(, , )e
im
e
is
. (12.9)
12.1.6 Spinor components of the Einstein tensor
With respect to a spinor tetrad, the components of the Einstein tensor G
mn
are
G
mn
=
_
_
_
_
G
tt
G
tz
G
t+
G
t
G
tz
G
zz
G
z+
G
z
G
t+
G
z+
G
++
G
+
G
t
G
z
G
+
G
_
_
_
_
. (12.10)
12.2 Newman-Penrose tetrad formalism 177
From this it is apparent that the 10 components of the Einstein tensor decompose into 4 spin-0 components,
4 spin-1 components, and 2 spin-2 components:
2 : G
,
1 : G
t
, G
z
,
0 : G
tt
, G
tz
, G
zz
, G
+
,
+1 : G
t+
, G
z+
,
+2 : G
++
.
(12.11)
The 4 spin-0 components are all real; in particular G
+
is real since G
+
= G
+
= G
+
. The 4 spin-1 and
2 spin-2 components comprise 3 complex components
G
++
= G
, G
t+
= G
t
, G
z+
= G
z
. (12.12)
In some contexts, for example in cosmological perturbation theory, REALLY? the various spin components
are commonly referred to as scalar (spin-0), vector (spin-1), and tensor (spin-2).
12.2 Newman-Penrose tetrad formalism
The Newman-Penrose formalism (E. T. Newman & R. Penrose, 1962, An Approach to Gravitational Ra-
diation by a Method of Spin Coecients, J. Math. Phys. 3, 566579; E. (Ted) Newman & R. Penrose,
2009, Spin-coecient formalism, Scholarpedia, 4(6), 7445, http://www.scholarpedia.org/article/
Newman-Penrose_formalism) provides a particularly powerful way to deal with elds that propagate at the
speed of light. The Newman-Penrose formalism adopts a tetrad in which the two axes
v
(outgoing) and
u
(ingoing) along the direction of propagation are chosen to be lightlike, while the two axes
+
and
transverse to the direction of propagation are chosen to be spinor axes.

Sadly, the literature on the Newman-Penrose formalism is characterized by an arcane and random notation
whose principal purpose seems to be to perpetuate exclusivity for an old-boys club of people who understand
it. This is unfortunate given the intrinsic power of the formalism. A. Held (1974, A formalism for the
investigation of algebraically special metrics. I, Commun. Math. Phys. 37, 311326) comments that the
Newman-Penrose formalism presents a formidable notational barrier to the uninitiate. For example, the
tetrad connections
kmn
are called spin coecients, and assigned individual greek letters that obscure
their transformation properties. Do not be fooled: all the standard tetrad formalism presented in Chapter 11
carries through unaltered. One ill-born child of the notation that persists in widespread use is
2s
for the
spin s component of the Weyl tensor, equations (12.49).
Gravitational waves are commonly characterized by the Newman-Penrose (NP) components of the Weyl
tensor. The NP components of the Weyl tensor are sometimes referred to as the NP scalars. The designation
as NP scalars is potentially misleading, because the NP components of the Weyl tensor form a tetrad-frame
tensor, not a set of scalars (though of course the tetrad-frame Weyl tensor is, like any tetrad-frame quantity, a
coordinate scalar). The NP components do become proper quantities, and in that sense scalars, when referred
178

to the frame of a particular observer, such as a gravitational wave telescope, observing along a particular
direction. However, the use of the word scalar to describe the components of a tensor is unfortunate.
12.2.1 Newman-Penrose tetrad
A Newman-Penrose tetrad
v
,
u
,
+
,
is dened in terms of an orthonormal tetrad

t
,
x
,
y
,
z
by
v

1
2
(
t
+
z
) , (12.13a)
u

1
2
(
t

z
) , (12.13b)
+

1
2
(
x
+i
y
) , (12.13c)
2
(
x
i
y
) , (12.13d)
or in matrix form
_
_
_
_
_
_
_
_
=
1
2
_
_
_
_
1 0 0 1
1 0 0 1
0 1 i 0
0 1 i 0
_
_
_
_
=
_
_
_
_
z
_
_
_
_
. (12.14)
Just as each of the spinor axes
+
and
individually species not one but two distinct directions

x
and
y
, so also each of the null axes
v
and
v
individually species not one but two distinct directions
t
and
z
. The ingoing null axis
u
may be obtained from the outgoing null axis
v
, and vice versa, by the parity
operation of inverting all the spatial axes.
All four tetrad axes are null
v

v
=
u

u
=
+

+
=
= 0 . (12.15)
In a profound sense, the null, or lightlike, character of each the four NP axes explains why the NP formalism is
well adapted to treating elds that propagate at the speed of light. The tetrad metric of the Newman-Penrose
tetrad
v
,
u
,
+
,
is
mn
=
_
_
_
_
0 1 0 0
1 0 0 0
0 0 0 1
0 0 1 0
_
_
_
_
. (12.16)
12.2.2 Boost weight
A boost along the
z
axis multiplies the outgoing and ingoing axes
v
and
u
by a blueshift factor and its
reciprocal
v

v
,
u
(1/)
u
. (12.17)
If the observer boosts by velocity v in the
z
direction away from the source, then the blueshift factor is the
special relativistic Doppler shift factor
=
_
1 v
1 +v
_
1/2
. (12.18)
The exponent n of the power
n
by which an object changes under a boost along the
z
axis is called its
boost weight. Thus
v
has boost weight +1, and
u
has boost weight 1. The spinor axes
both have
boost weight 0. The NP components of a tensor inherit their boost weight properties from those of the NP
basis. The general rule is that the boost weight n of any tensor component is equal to the number of v
covariant indices minus the number of u covariant indices:
boost weight n = number of v minus u covariant indices . (12.19)
12.3 Electromagnetic eld tensor
12.3.1 Complexied electromagnetic eld tensor
The electromagnetic eld F
mn
is a bivector, and as such has a natural complex structure. The real part
of the electromagnetic bivector eld is the electric eld E, which changes sign under spatial inversion,
while the imaginary part is the magnetic eld B, which remains unchanged under spatial inversion. In an
orthonormal tetrad
t
,
x
,
y
,
z
, the electromagnetic bivector can be written as though it were a vector
with 6 components:
F =
_
E B
_
=
_
E
x
E
y
E
z
B
x
B
y
B
z
_
=
_
F
tx
F
ty
F
tz
F
zy
F
xz
F
yx
_
. (12.20)
The bivector has 3 electric and 3 magnetic components:
electric bivector indices: tx, ty, tz ,
magnetic bivector indices: zy, xz, yx .
(12.21)
The natural complex structure motivates dening the complexied electromagnetic eld tensor

F
kl
to be
the complex combination
F
kl

1
2
(F
kl
+

F
kl
) =
1
2
_
m
k

n
l
+
i
2

kl
mn
_
F
mn
is a tetrad tensor , (12.22)
180

where

F
kl
denotes the Hodge dual of F
kl
F
kl

i
2

kl
mn
F
mn
. (12.23)
Here
klmn
is the totally antisymmetric tensor (see Exercise 12.1). The overall factor of
1
2
on the right hand
sides of equations (12.22) is introduced so that the complexication operator P
mn
kl

1
2
(
m
k

n
l
+
i
2

kl
mn
)
is a projection operator, satisfying P
2
= P. The denitions (12.22) and (12.23) of the complexied and
dual electromagnetic eld tensors

F
kl
and

F
kl
are valid in any frame, not just an orthonormal frame or
a Newman-Penrose frame. In an orthonormal frame, the dual

F has the structure, in the same notation
as (12.20),
F = i
_
B E
_
. (12.24)
In an orthonormal frame the complexied electromagnetic eld

F, equation (12.22), then has the structure
F =
1
2
_
1 i
_
(E +iB) . (12.25)
Thus the complexied electromagnetic eld eectively embodies the electric and magnetic elds in the
complex 3-vector combination E +iB. The complexied electromagnetic eld is self-dual
F =

F . (12.26)
One advantage of this approach is that Maxwells equations (11.64) and (11.65) combine into a single
complex equation
D
m

F
mn
= 4j
n
(12.27)
whose real and imaginary parts represent respectively the source and source-free Maxwells equations.
In some cases, such as the Kerr-Newman geometry expressed in the Boyer-Lindquist orthonormal tetrad,
equation (15.33), the complexied electromagnetic eld makes manifest the inner elegance of the geometry.
Exercise 12.1 Totally antisymmetric tensor. In an orthonormal tetrad
m
where
0
points to the
future and
1
,
2
,
3
are right-handed, the contravariant totally antisymmetric tensor
klmn
is dened by
(this is the opposite sign from MTWs notation)
klmn
[klmn] (12.28)
and hence
klmn
= [klmn] (12.29)
where [klmn] is the totally antisymmetric symbol
[klmn]
_
_
_
+1 if klmn is an even permutation of 0123 ,
1 if klmn is an odd permutation of 0123 ,
0 if klmn are not all dierent .
(12.30)
Argue that in a general basis g
the contravariant totally antisymmetric tensor
is
= e
k
e
l
e
m
e
n
klmn
= e [] (12.31)
while its covariant counterpart is
= (1/e) [] (12.32)
where e [e
m
[ is the determinant of the vierbein.

12.3.2 Newman-Penrose components of the electromagnetic eld
With respect to a NP null tetrad
v
,
u
,
+
,
, equation (12.13), the electromagnetic eld F

mn
has 3
distinct complex components, here denoted
s
, of spins respectively s = 1, 0, and +1 in accordance with
the rule (12.6):
1 :
1
F
u
,
0 :
0

1
2
(F
uv
+F
+
) ,
+1 :
1
F
v+
.
(12.33)
The complex conjugates
s
of the 3 NP components of the electromagnetic eld are
1
= F
u+
,
0
=
1
2
(F
uv
F
+
) ,
1
= F
v
,
(12.34)
whose spins have the opposite sign, in accordance with the rule (12.8) that complex conjugation ips spin.
The above convention that the index s on the NP component
s
labels its spin diers from the standard
convention, where the spin s component is capriciously denoted
1s
(e.g. S. Chandrasekhar, 1983, The
Mathematical Theory of Black Holes, Clarendon Press, Oxford, 1983):
1 :
2
,
0 :
1
,
+1 :
0
.
(standard convention, not followed here) (12.35)
In terms of the electric and magnetic elds E and B in the parent orthonormal tetrad of the NP tetrad,
the 3 complex NP components
s
of the electromagnetic eld are
1
=
1
2
[E
x
+iB
x
i(E
y
+iB
y
)] ,
0
=
1
2
(E
z
+iB
z
) ,
1
=
1
2
[E
x
+iB
x
+i(E
y
+iB
y
)] .
(12.36)
Equations (12.36) show that the NP components of the electromagnetic eld contain the electric and mag-
netic elds in the complex combination E + iB, just like the complexied electromagnetic eld

F
kl
, equa-
182

tion (12.25). Explicitly, the NP components are related to the components

F
kl
of the complexied electro-
magnetic eld by
1
=

F
tx
i
F
ty
,
0
=

F
tz
,
1
=

F
tx
+i
F
ty
.
(12.37)
Part of the power of the NP formalism arises from the fact that it exploits the natural complex structure of
the electromagnetic bivector eld.
12.3.3 Newman-Penrose components of the complexied electromagnetic eld
The non-vanishing NP components of the complexied electromagnetic eld

F
kl
dened by equation (12.22)
are
F
u
=
1
,
F
uv
=

F
+
=
0
,
F
v+
=
1
,
(12.38)
whereas components with bivector indices v or u+ vanish,
F
v
=

F
u+
= 0 . (12.39)
The rule that complex conjugation ips spin fails here because the complexication operator in equa-
tion (12.22) breaks the rule. Equations (12.38) and (12.39) show that the complexied electromagnetic
eld in an NP tetrad contains just 3 distinct non-vanishing complex components, and those components are
precisely equal to the complex spin components
s
.
12.3.4 Propagating components of electromagnetic waves
An oscillating electric charge emits electromagnetic waves. Similarly, an electromagnetic wave incident on an
electric charge causes it to oscillate. An electromagnetic wave moving away from a source is called outgoing,
while a wave moving towards a source is called ingoing.
It can be shown that only the spin 1 NP component
1
of an outgoing electromagnetic wave propagates,
carrying electromagnetic energy to innity:
1
: propagating, outgoing . (12.40)
This propagating, outgoing 1 component has spin 1, but its complex conjugate has spin +1, so eec-
tively both spin components, or helicities, or circular polarizations, of an outgoing electromagnetic wave are
embodied in the single complex component
1
. The remaining 2 complex NP components
0
and
1
of an
outgoing wave are short range, describing the electromagnetic eld near the source.
Similarly, only the spin +1 component
1
of an ingoing electromagnetic wave propagates, carrying energy
from innity:
1
: propagating, ingoing . (12.41)
The isolation of each propagating mode into a single complex NP mode, incorporating both helicities, is
simpler than the standard picture of oscillating orthogonal electric and magnetic elds.
12.4 Weyl tensor
The Weyl tensor is the trace-free part of the Riemann tensor,
C
klmn
R
klmn
1
2
(
km
R
ln
kn
R
lm
+
ln
R
km
lm
R
kn
) +
1
6
(
km
ln
kn
lm
) R . (12.42)
By construction, the Weyl tensor vanishes when contracted on any pair of indices. Whereas the Ricci and
Einstein tensors vanish identically in any region of spacetime containing no energy-momentum, T
mn
= 0,
the Weyl tensor can be non-vanishing. Physically, the Weyl tensor describes tidal forces and gravitational
waves.
12.4.1 Complexied Weyl tensor
The Weyl tensor is is, like the Riemann tensor, a symmetric matrix of bivectors. Just as the electromagnetic
bivector F
kl
has a natural complex structure, so also the Weyl tensor C
klmn
has a natural complex structure.
The properties of the Weyl tensor emerge most plainly when that complex structure is made manifest.
In an orthonormal tetrad
t
,
x
,
y
,
z
, the Weyl tensor C
klmn
can be written as a 6 6 symmetric
bivector matrix, organized as a 2 2 matrix of 3 3 blocks, with the structure
C =
_
C
EE
C
EB
C
BE
C
BB
_
=
_
_
_
_
_
_
_
_
_
C
txtx
C
txty
C
txtz
C
txzy
C
txxz
C
txyx
C
tytx
... ... ... ... ...
C
tztx
... ... ... ... ...
C
zytx
... ... ... ... ...
C
xztx
... ... ... ... ...
C
yxtx
... ... ... ... ...
_
_
_
_
_
_
_
_
_
, (12.43)
where E denotes electric indices, B magnetic indices, per the designation (12.21). The condition of being
symmetric implies that the 3 3 blocks C
EE
and C
BB
are symmetric, while C
BE
= C
EB
. The cyclic
symmetry (11.47) of the Riemann, hence Weyl, tensor implies that the o-diagonal 3 3 block C
EB
(and
likewise C
BE
) is traceless.
The natural complex structure motivates dening a complexied Weyl tensor

C
klmn
by
C
klmn

1
4
_
p
k
q
l
+
i
2

kl
pq
__
r
m
s
n
+
i
2

mn
rs
_
C
pqrs
is a tetrad tensor (12.44)
analogously to the denition (12.22) of the complexied electromagnetic eld. The denition (12.44) of the
complexied Weyl tensor

C
klmn
is valid in any frame, not just an orthonormal frame. In an orthonormal
184

frame, if the Weyl tensor C
klmn
is organized according to the structure (12.43), then the complexied Weyl
tensor

C
klmn
dened by equation (12.44) has the structure
C =
1
4
_
1 i
i 1
_
(C
EE
C
BB
+i C
EB
+i C
BE
) . (12.45)
Thus the independent components of the complexied Weyl tensor

C
klmn
constitute a 3 3 complex sym-
metric traceless matrix C
EE
C
BB
+ i(C
EB
+ C
BE
), with 5 complex degrees of freedom. Although the
complexied Weyl tensor

C
klmn
is dened, equation (12.44), as a projection of the Weyl tensor, it nevertheless
retains all the 10 degrees of freedom of the original Weyl tensor C
klmn
.
The same complexication projection operator applied to the trace (Ricci) parts of the Riemann tensor
yields only the Ricci scalar multiplied by that unique combination of the tetrad metric that has the symme-
tries of the Riemann tensor. Thus complexifying the trace parts of the Riemann tensor produces nothing
useful.
12.4.2 Newman-Penrose components of the Weyl tensor
With respect to a NP null tetrad
v
,
u
,
+
,
, equation (12.13), the Weyl tensor C

klmn
has 5 distinct
complex components, here denoted
s
, of spins respectively s = 2, 1, 0, +1, and +2:
2 :
2
C
uu
,
1 :
1
C
uvu
= C
+u
,
0 :
0

1
2
(C
uvuv
+C
uv+
) =
1
2
(C
++
+C
uv+
) = C
v+u
,
+1 :
1
C
vuv+
= C
+v+
,
+2 :
2
C
v+v+
.
(12.46)
The complex conjugates
s
of the 5 NP components of the Weyl tensor are:
2
= C
u+u+
,
1
= C
uvu+
= C
+u+
,
0
=
1
2
(C
uvuv
+C
uv+
) =
1
2
(C
++
+C
uv+
) = C
v+u
,
1
= C
vuv
= C
+v
,
2
= C
vv
.
(12.47)
whose spins have the opposite sign, in accordance with the rule (12.8) that complex conjugation ips spin.
The above expressions (12.46) and (12.47) account for all the NP components C
klmn
of the Weyl tensor but
four, which vanish identically:
C
v+v
= C
u+u
= C
v+u+
= C
vu
= 0 . (12.48)
The above convention that the index s on the NP component
s
labels its spin diers from the standard con-
vention, where the spin s component of the Weyl tensor is impenetrably denoted
2s
(e.g. S. Chandrasekhar
1983):
2 :
4
,
1 :
3
,
0 :
2
,
+1 :
1
,
+2 :
0
.
(standard convention, not followed here) (12.49)
With respect to a triple of bivector indices ordered as u, uv, +v, the NP components of the Weyl tensor
constitute the 3 3 complex symmetric matrix IS THIS NP OR COMPLEX NP?
C
klmn
=
_
_
2

1

0
1

0

1
0

1

2
_
_
. (12.50)
12.4.3 Newman-Penrose components of the complexied Weyl tensor
The non-vanishing NP components of the complexied Weyl tensor

C
klmn
dened by equation (12.44) are
C
uu
=
2
,
C
uvu
=

C
+u
=
1
,
C
uvuv
=

C
++
=

C
uv+
=

C
v+u
=
0
,
C
vuv+
=

C
+v+
=
1
,
C
v+v+
=
2
.
(12.51)
whereas any component with either of its two bivector indices equal to v or u+ vanishes. As with the
complexied electromagnetic eld, the rule that complex conjugation ips spin fails here because the com-
plexication operator breaks the rule. Equations (12.51) show that the complexied Weyl tensor in an NP
tetrad contains just 5 distinct non-vanishing complex components, and those components are precisely equal
to the complex spin components
s
.
12.4.4 Components of the complexied Weyl tensor in an orthonormal tetrad
The complexied Weyl tensor forms a 3 3 complex symmetric traceless matrix in any frame, not just an
NP frame. In an orthonormal frame, with respect to a triple of bivector indices tx, ty, tz, the complexied
Weyl tensor

C
klmn
can be expressed in terms of the NP spin components
s
as
C
klmn
=
_
_
0
1
2
(
1
1
)
i
2
(
1
+
1
)
1
2
(
1
1
)
1
2
0
+
1
4
(
2
+
2
)
i
4
(
2
2
)
i
2
(
1
+
1
)
i
4
(
2
2
)
1
2
1
4
(
2
+
2
)
_
_
. (12.52)
186

12.4.5 Propagating components of gravitational waves
For outgoing gravitational waves, only the spin 2 component
2
(the one conventionally called
4
) prop-
agates, carrying gravitational waves from a source to innity:
2
: propagating, outgoing . (12.53)
This propagating, outgoing 2 component has spin 2, but its complex conjugate has spin +2, so eectively
both spin components, or helicities, or circular polarizations, of an outgoing gravitational wave are embodied
in the single complex component. The remaining 4 complex NP components (spins 1 to 2) of an outgoing
gravitational wave are short range, describing the gravitational eld near the source.
Similarly, only the spin +2 component
2
of an ingoing gravitational wave propagates, carrying energy
from innity:
2
: propagating, ingoing . (12.54)
12.5 Petrov classication of the Weyl tensor
As seen above, the complexied Weyl tensor is a complex symmetric traceless 3 3 matrix. If the matrix
were real symmetric (or complex Hermitian), then standard mathematical theorems would guarantee that
it would be diagonalizable, with a complete set of eigenvalues and eigenvectors. But the Weyl matrix is
complex symmetric, and there is no such theorem.
The mathematical theorems state that a matrix is diagonalizable if and only if it has a complete set of
linearly independent eigenvectors. Since there is always at least one distinct linearly independent eigenvector
associated with each distinct eigenvalue, if all eigenvalues are distinct, then necessarily there is a complete
set of eigenvectors, and the Weyl tensor is diagonalizable. However, if some of the eigenvalues coincide, then
there may not be a complete set of linearly independent eigenvectors, in which case the Weyl tensor is not
diagonalizable.
The Petrov classication, tabulated in Table 12.1, classies the Weyl tensor in accordance with the number
of distinct eigenvalues and eigenvectors. The normal form is with respect to an orthonormal frame aligned
with the eigenvectors to the extent possible. The tetrad with respect to which the complexied Weyl tensor
takes its normal form is called the Weyl principal tetrad. The Weyl principal tetrad is unique except in
cases D, O, and N. For Types D and N, the Weyl principal tetrad is unique up to Lorentz transformations
that leave the eigen-bivector
tz
unchanged, which is to say, transformations generated by the Lorentz rotor
exp(
tz
) where is complex.
The Kerr-Newman geometry is Type D. General spherically symmetric geometries are Type D. The
Friedmann-Robertson-Walker geometry is Type O. Plane gravitational waves are Type N.
12.6 Raychaudhuri equations and the Sachs optical scalars 187
expansion rotation shear
Figure 12.1 Illustrating how the Sachs optical scalars, the expansion , the rotation , and the shear ,
dened by equations (12.60), characterize the rate at which a bundle of light rays changes shape as it
propagates. The bundle of light is coming vertically upward out of the paper.
12.6 Raychaudhuri equations and the Sachs optical scalars
Consider a light ray. Let the
v
null axis lie along the worldline of the light ray. Choose the Newman-
Penrose tetrad so that the tetrad axes
m
are parallel-transported along the path of the light ray. You can
think of a bunch of observers arrayed along the ray each observing the same image, unprecessed, unboosted,
Table 12.1 Petrov classication of the Weyl tensor
Petrov Distinct Distinct Normal form
type eigenvalues eigenvectors of the complexied Weyl tensor
I 3 3
0
@
0 0 0
0
1
2
0 +
1
2
2 0
0 0
1
2
0
1
2
2
1
A
D 2 3
0
@
0 0 0
0
1
2
0 0
0 0
1
2
0
1
A
II 2 2
0
@
0 0 0
0
1
2
0 +
1
4
2
i
4
2
0
i
4
2
1
2
0
1
4
2
1
A
O 1 3
0
@
0 0 0
0 0 0
0 0 0
1
A
N 1 2
0
@
0 0 0
0
1
4
2
i
4
2
0
i
4
2
1
4
2
1
A
III 1 1
0
@
0
1
2
1
i
2
1
1
2
1 0 0
i
2
1 0 0
1
A
188

unredshifted. Mathematically, this means that the tetrad axes along the worldline of the light ray satisfy
m
= 0 . (12.55)
By denition of the tetrad connections, this is equivalent to the conditions that the Newman-Penrose tetrad
connections with nal index v all vanish
kmv
= 0 . (12.56)
The conditions (12.56) constitute a set of 6 conditions which dene the Lorentz transformation of the tetrad
axes along the worldline of the light ray. Given the conditions (12.56), the usual expression (11.45) for the
Riemann tensor implies that the rate of change
v
mnl
of each of the 18 remaining tetrad connections
mnl
,
those with nal index l ,= v, along the worldline of the light ray satises
mnl
+
k
vl
mnk
= R
vlmn
. (12.57)
As commented after the denition (12.13) of the NP tetrad, the null axis
v
(
t
+
z
)/
2 denes not a
single direction, but rather a 2D surface spanned by the two directions
t
and
z
. Orthogonal to this surface
is the 2D surface spanned by the transverse axes
x
and
y
, or equivalently by the spinor axes
+
and
.
Of the 18 tetrad connections
mnl
with l ,= v, four are embodied in the extrinsic curvature K
ab
dened by
K
ab

a

b
v
=
avb
for a, b = +, . (12.58)
The extrinsic curvature (12.58) describes how the null axis
v
varies over the 2D surface spanned by the
transverse axes
+
and
. For the extrinsic curvature K

ab
, the evolution equations (12.57) become
v
K
ab
+K
+
b
K
a+
+K
b
K
a
= R
vbav
. (12.59)
The Sachs optical scalars constitute the components of the extrinsic curvature K
ab
dened by equation (12.58).
Conventionally, the Sachs optical scalars consist of the expansion , the rotation , and the complex shear
, dened in terms of the extrinsic curvature (12.58) by
+i K
+
, (12.60a)
K
++
, (12.60b)
whose complex conjugates are i K
+
and
= K
. Resolved into real and imaginary parts, the

denitions (12.60) of the Sachs optical scalars are

1
2
(K
+
+K
+
) , (12.61a)

1
2i
(K
+
K
+
) , (12.61b)
Re
1
2
(K
++
+K
) , (12.61c)
Im
1
2i
(K
++
K
) . (12.61d)
Physically, the Sachs scalars characterize how the shape of a bundle of light rays evolves as it propagates,
as illustrated in Figure 12.1. The expansion represents how fast the bundle expands, the rotation how fast
12.7 Focussing theorem 189
it rotates, and the shear how fast its ellipticity is changing. The amplitude and phase of the complex shear
represent the amplitude and phase of the major axis of the shear ellipse.
The derivation from equations (12.59) of the evolutionary equations governing the Sachs scalars , , and
is left as Exercise 12.2. The equation (12.62b) for the shear shows that the shear changes only in the
presence of a non-vanishing Weyl tensor, that is, in the presence of tidal forces. The equation (12.63b) for
the rotation shows that if the rotation is initially zero, then it will remain zero along the path of the light
bundle. Thus geodesic motion cannot by itself generate rotation from nothing; but non-geodesic processes,
such the electromagnetic scattering of light, can generate rotation. The equation (12.63a) for the expansion
is commonly called the Raychaudhuri equation (A. K. Raychaudhuri, 1955, Relativistic cosmology,
I, Phys. Rev. 98, 1123). The Raychaudhuri equation is the basis of the focussing theorem, 12.7, which is
a central ingredient of the Penrose-Hawking singularity theorems.
Exercise 12.2 Raychaudhuri equations
Show that equations (12.59) imply the following evolutionary equations for the Sachs scalars dened by (12.60):
(
v
+ +i)( +i) +
+
1
2
G
vv
= 0 , (12.62a)
(
v
+ 2) +C
v+v+
= 0 , (12.62b)
where G
mn
and C
klmn
denote the Einstein and Weyl tensors as usual. Show that the rst (12.62a) of these
equations is equivalent to the two equations
v
+ (
2
2
) +
+
1
2
G
vv
= 0 , (12.63a)
(
v
+ 2) = 0 . (12.63b)
12.7 Focussing theorem

The Raychaudhuri equation (12.63a) provides the basis for the focussing theorem, which is a key ingredient
of the singularity theorems introduced by Penrose (R. Penrose, 1965, Gravitational collapse and spacetime
singularities, Phys. Rev. Lett. 14, 5759) and elaborated extensively by Hawking (S. W. Hawking and G.
F. R. Ellis, 1975, The large scale structure of space-time, Cambridge University Press).
If the rotation vanishes, then equation (12.63a) for the expansion simplies to
v
+
2
+
+
1
2
G
vv
= 0 . (12.64)
The terms
2
and
are necessarily positive. The NP component G

vv
of the Einstein tensor is related to
the components in the parent orthonormal tetrad by
G
vv
=
1
2
G
tt
+G
tz
+
1
2
G
zz
. (12.65)
190

Boosted along the z-direction into the center-of-mass frame, where G
tz
= 0, equation (12.65) reduces to
G
vv
= 4( +p
z
) (12.66)
where is the energy density and p
z
the pressure along the z-direction. If it is true that
+p
z
0 , (12.67)
then the rotation-free Raychaudhuri equation (12.64) shows that the expansion must always decrease.
13
The 3+1 (ADM) formalism

Einsteins equations constitute a set of 10 coupled second-order partial dierential equations. Solving these
equations in a general fashion presents a formidable challenge. The 3+1, or ADM, formalism devised by
R. Arnowitt, S. Deser, & C. W. Misner (1959, Phys. Rev. 116, 13221330; 1962, The dynamics of general
relativity, in Gravitation: an introduction to current research, ed. L. Witten, 227265) oers an insightful
and systematic way to proceed. The formalism is widely used in numerical general relativity. For reviews,
see L. Lehner (2001, Numerical relativity: a review, CQG 18, R2586, gr-qc/0106072), and H. Shinkai
(2008, Formulations of the Einstein equations for numerical simulations, APCTP winter school on black
hole astrophysics, arXiv:0805.0068).
The central idea of the ADM formalism is to recast the Einstein equations into Hamiltonian form. The
Hamiltonian approach identies canonical momenta conjugate to the coordinates, and converts the
equations of motion from second order partial dierential equations in the coordinates into coupled rst
order partial dierential equations in the coordinates and momenta. The Hamiltonian H of a system is
its energy expressed in terms of the coordinates and momenta. In quantum mechanics, equating the
time translation operator to the Hamiltonian operator, i /t = H, determines the evolution in time t of
the system. To implement the Hamiltonian approach in general relativity, it is necessary to identify one
coordinate, the time coordinate t, as having a special status. The system of Einstein (and other) equations
is evolved by integrating from one spacelike hypersurface of constant time, t = constant, to the next.
The 3+1 formalism provides answers to several basic questions about the dynamical structure of Einsteins
equations:
1. Are there natural coordinates for the gravitational eld, and what are they? Answer: Yes. The natural
coordinates are the 6 components of the spatial metric g
on the hypersurfaces of constant time. The

3+1 formalism shows that only these 6 of the 10 components of the 4D metric g
are governed by time

evolution equations. The remaining 4 degrees of freedom in the metric represent gauge freedoms associated
with general coordinate transformations.
2. What happens to the 4 degrees of freedom associated with general coordinate transformations? Answer:
They are accomodated into the lapse and shift
, which express the rate at which the unit timelike

normal
0
to the spatial hypersurfaces of constant time marches through the coordinates, equations (13.9).
192

In the 3+1 formalism, the lapse and shift are speciable arbitrarily, and are not governed by time evolution
equations.
3. What are the 6 momenta conjugate to the 6 coordinates g
? Answer: The momenta are, up to a factor,

the components of the trace-modied extrinsic curvature K
K, equation (13.36). The extrinsic

curvature K
, whose tetrad-frame expression is dened by equation (13.12b), is a symmetric 33 matrix

that describes how the unit timelike normal
0
varies over the spatial hypersurfaces of constant time.
4. What is the structure of the 10 Einstein equations in the 3+1 formalism? Answer: The 6 spatial com-
ponents of the Einstein equations provide dynamical time evolution equations for the 6 momenta. The
4 remaining components of the Einstein equations, the time-time and time-space components, prove to
be constraint equations, called the Hamiltonian (or scalar) and momentum (or vector) constraints. The
constraint equations specify conditions that must be arranged to be satised on the initial hypersurface
of constant time, but thereafter the constraints are automatically satised (modulo numerical error and
instabilities).
5. How is covariant conservation of energy-momentum expressed? Answer: The fact that the Hamiltonian
and momentum constraints continue to be satised as time advances expresses covariant conservation of
energy-momentum, as guaranteed by the contracted Bianchi identities.
In this chapter, the coordinate time index is t, while the tetrad time index is 0. Early-alphabet greek
(brown) letters , , ..., denote 3D spatial coordinate indices, while mid-alphabet greek letters , , ...,
denote 4D spacetime coordinate indices. Early-alphabet latin (black) letters a, b, ..., denote 3D spatial
tetrad indices, while mid-alphabet latin (black) letters k, l, ..., denote 4D spacetime tetrad indices.
13.1 ADM tetrad
The ADM formalism splits the spacetime coordinates x
into a time coordinate t and spatial coordinates

x
, = 1, 2, 3,
x
t, x
. (13.1)
At each point of spacetime, the spacelike hypersurface of constant time t has a unique future-pointing unit
normal
0
, dened to have unit length and to be orthogonal to the spatial tangent axes g
0

0
= 1 ,
0
g
= 0 = 1, 2, 3 . (13.2)
The central element of the ADM approach is to work in a tetrad frame
m
consisting of this time axis
0
,
together with three spatial tetrad axes
a
that are orthogonal to the tetrad time axis
0
, and therefore lie
in the 3D spatial hypersurface of constant time,
0

a
= 0 a = 1, 2, 3 . (13.3)
The tetrad metric
mn
in the ADM formalism is thus
mn
=
_
1 0
0
ab
_
, (13.4)
13.2 Traditional ADM approach 193
and the inverse tetrad metric
mn
is correspondingly
mn
=
_
1 0
0
ab
_
, (13.5)
whose spatial part
ab
is the inverse of
ab
. Given the conditions (13.2) and (13.3), the vierbein e
m
and
inverse vierbein e
m
take the form

e
m
=
_
1/
/
0 e
a
_
, e
m
=
_
0
e
a
e
a
_
, (13.6)
where and
are the lapse and shift (see next paragraph), and e

a
and e
a
represent the spatial vierbein

and inverse vierbein, which are inverse to each other, e
a
e
b
=
c
a
. The ADM metric is
ds
2
=
2
dt
2
+g
(dx
dt)
_
dx
dt
_
, (13.7)
where g
is the spatial coordinate metric

g
=
ab
e
a
e
b
. (13.8)
Essentially all the tetrad formalism developed in Chapter 11 carries through, subject only to the condi-
tions (13.2) and (13.3).
The vierbein coecient is called the lapse, while
is called the shift. Physically, the lapse is the

rate at which the proper time of the tetrad rest frame elapses per unit coordinate time t, while the shift
is the velocity at which the tetrad rest frame moves through the spatial coordinates x
per unit coordinate

time t,
=
d
dt
,
=
dx
dt
. (13.9)
These relations (13.9) follow from the fact that the 4-velocity in the tetrad rest frame is by denition
u
m
1, 0, 0, 0, so the coordinate 4-velocity u
e
m
u
m
of the tetrad rest frame is
dx
d
u
= e
t
=
1
1,
. (13.10)
13.2 Traditional ADM approach
The traditional ADM approach sets the spatial tetrad axes
a
equal to the spatial coordinate tangent axes
g
a
= g
(traditional ADM) , (13.11)

equivalent to choosing the spatial vierbein to be the unit matrix, e
a
a
. The traditional ADM approach
may be termed semi-tetrad, since it works with a tetrad time axis
0
together with coordinate spatial axes
g
. It is natural however to extend the ADM approach into a full tetrad approach, allowing the spatial
tetrad axes
a
to be chosen more generally, subject only to the condition (13.3) that they be orthogonal to
194

the tetrad time axis, and therefore lie in the hypersurface of constant time t. For example, the spatial tetrad
a
can be chosen to form 3D orthonormal axes,
ab

a

b
=
ab
, so that the full 4D tetrad metric
mn
is
Minkowski.
This chapter follows the full tetrad approach to the ADM formalism, but all the results hold for traditional
case where the spatial tetrad axes are set equal to the coordinate spatial axes, equation (13.11).
13.3 Spatial tetrad vectors and tensors
Since the tetrad time axis
0
in the ADM formalism is dened uniquely by the choice of hypersurfaces
of constant time t, there is no freedom of tetrad transformations of the time axis distinct from temporal
coordinate transformations (no distinct freedom of Lorentz boosts). However, there is still freedom of tetrad
transformations of the spatial tetrad axes (spatial rotations).
A covariant spatial tetrad vector A
a
is dened in the usual way as a vector that transforms like the
spatial tetrad axes
a
. Similarly a covariant spatial tetrad tensor is a tensor that transforms like products
of the spatial tetrad axes
a
. Indices on spatial tetrad vectors and tensors are raised with the inverse spatial
tetrad metric
ab
, and lowered with the spatial tetrad metric
ab
.
A temporal coordinate transformation changes the hypersurface of constant time t, and therefore changes
its unit normal, the tetrad time axis
0
, and correspondingly all the spatial tetrad axes
a
.
13.4 ADM connections, gravity, and extrinsic curvature
Since the tetrad time axis
0
is a spatial tetrad scalar, its directed time derivative
0
0
is a spatial tetrad
scalar, while its directed spatial derivatives
b
0
form a spatial tetrad vector. It follows that the connections
m0n
dened by the directed derivatives of
0
are spatial tetrad tensors. These connections play an important
role in the ADM formalism, and they are given special names and symbols, the gravity
a
, and the extrinsic
curvature K
ab
(the remaining components of the directed derivatives of
0
vanish,
000
=
00a
= 0):
a

a

0
0
=
a00
is a spatial tetrad vector , (13.12a)
K
ab

a

b
0
=
a0b
is a spatial tetrad tensor . (13.12b)
The gravity
a
is justly named because the geodesic equation shows that it is minus the acceleration expe-
rienced in the tetrad rest frame, where u
m
= 1, 0, 0, 0,
du
a
d
=
a
. (13.13)
The extrinsic curvature K
ab
describes how the unit normal
0
to the 3-dimensional spatial hypersurface
changes over the hypersurface, and can therefore be regarded as embodying the curvature of the 3-dimensional
spatial hypersurface embedded in the 4-dimensional spacetime. From equation (11.40) with vanishing torsion,
13.5 ADM Riemann, Ricci, and Einstein tensors 195
it follows that the gravity and the extrinsic curvature are
a
= d
00a
, (13.14a)
K
ab
=
1
2
(
0
ab
d
ab0
d
ba0
+d
a0b
+d
b0a
) , (13.14b)
where the relevant vierbein derivatives d
lmn
are
d
00a
=
1

a
, d
a0b
=
1
e
a
, d
ab0
=
ac
e
b
0
e
c
. (13.15)
Equation (13.14b) shows that the extrinsic curvature is symmetric,
K
ab
= K
ba
. (13.16)
The non-vanishing tetrad connections are, from the general formula (11.40) with vanishing torsion,
a00
=
0a0
=
a
, (13.17a)
a0b
=
0ab
= K
ab
, (13.17b)
ab0
= K
ab
+d
ab0
d
a0b
, (13.17c)
abc
= same as eq. (11.40) . (13.17d)
The connections (13.17a) and (13.17b) form, as commented above, a spatial tetrad vector
a
and tensor
K
ab
, but the remaining connections (13.17c) and (13.17d) are not spatial tetrad tensors. Note that the
purely spatial tetrad connections
abc
, like the spatial tetrad axes
a
, transform under temporal coordinate
transformations despite the absence of temporal indices.
13.5 ADM Riemann, Ricci, and Einstein tensors
The ADM Riemann tensor R
klmn
inherits from the standard tetrad formalism the property of being a full
tetrad tensor, its components transforming like products of the tetrad axes
m
. Of course, since the ADM
tetrad time axis
0
is tied to the coordinates, a tetrad transformation of the time axis requires a simultaneous
coordinate transformation consistent with it. The usual expression (11.45) for the tetrad-frame Riemann
tensor with vanishing torsion yields the ADM Riemann tensor
R
0a0b
= D
0
K
ab
K
ac
K
c
b
+
1
D
a
D
b
, (13.18a)
R
0abc
= D
c
K
ab
D
b
K
ac
, (13.18b)
R
abcd
= K
ac
K
bd
K
ad
K
bc
+R
(3)
abcd
. (13.18c)
Here R
(3)
abcd
is the spatial tetrad Riemann tensor considered conned to the 3D spatial hypersurface (given by
equation (11.45) with all time components discarded), and D
m
denotes the usual 4D tetrad-frame covariant
derivative, with the understanding that when acting on a spatial tensor such as K
ab
, all time components
196

of the spatial tensor are to be considered equal to zero. Thus the 4D tetrad covariant derivative of a spatial
tetrad vector A
a
is
D
m
A
a
=
m
A
a
b
am
A
b
, (13.19)
in which the possible
0
am
A
0
term is considered to vanish. When acting on a spatial tetrad tensor, the spatial
part D
a
of the 4D tetrad covariant derivative D
m
involves only spatial connections, and is identical to the
3D spatial tetrad covariant derivative D
(3)
considered conned to the 3D spatial hypersurface,
D
a
D
(3)
a
when acting on a spatial tetrad tensor . (13.20)
The covariant tetrad time derivative D
0
acting on a spatial tetrad tensor yields a 3D spatial tetrad tensor,
but the full 4D tetrad covariant derivative D
m
acting on a spatial tetrad tensor does not yield a 4D tetrad
tensor. It is important to bear these ne distinctions in mind when integrating by parts, as done in 13.6.
The Ricci tensor R
km

ln
R
klmn
is, like the Riemann tensor, a tetrad tensor. Its components are
R
00
=
0
K K
ab
K
ab
+
1
D
a
D
a
, (13.21a)
R
0a
= D
b
K
ab
a
K , (13.21b)
R
ab
= D
0
K
ab
+KK
ab
D
a
D
b
+R
(3)
ab
, (13.21c)
where K
ab
K
ab
is the trace of the extrinsic curvature, and R
(3)
ac

bd
R
(3)
abcd
is the Ricci tensor conned
to the 3D spatial hypersurface. The Ricci scalar R
km
R
km
is
R = 2
0
K +K
2
+K
ab
K
ab
D
a
D
a
+R
(3)
, (13.22)
where R
(3)

ab
R
(3)
ab
is the Ricci scalar conned to the 3D spatial hypersurface.
The Einstein tensor G
km
R
km
1
2
km
R is
G
00
=
1
2
_
K
2
K
ab
K
ab
+R
(3)
_
, (13.23a)
G
0a
= D
b
K
ab
D
a
K , (13.23b)
G
ab
= D
0
K
ab
+KK
ab
D
a
D
b

ab
_
0
K +
1
2
K
2
+
1
2
K
cd
K
cd
D
c
D
c
_
+G
(3)
ab
, (13.23c)
where G
(3)
ab
R
(3)
ab

1
2
ab
R
(3)
is the Einstein tensor conned to the 3D spatial hypersurface.
13.6 ADM action
In this chapter up to this point, the Einstein tensor and other quantities have been expressed in terms of
other things, but no equation of motion has been invoked. The ADM philosophy is to derive the gravitational
equations of motion Einsteins equations in Hamiltonian form. The starting point is to write down the
gravitational action, extremization of which will yield equations of motion.
13.6 ADM action 197
The Hilbert gravitational action S
g
is
S
g
=
_
t
f
ti
L
g
dx
4
=
1
16
_
t
f
ti
Rdx
4
, (13.24)
with scalar Lagrangian L
g
equal to a normalization factor times the Ricci scalar R,
L
g
=
1
16
R . (13.25)
The integration in the action (13.24) is over a 4-volume from an initial hypersurface of constant time t
i
to
a nal hypersurface of constant time t
f
. The integration measure dx
4
in the integral (13.24) denotes the
scalar 4-volume element
1
. With respect to coordinates t, x
, the scalar 4-volume element is

dx
4
= dt dx
3
0
= (/e) dt d
3
x , (13.26)
where dx
3
0
is the tetrad time component (the component along the timelike normal
0
to the spatial hy-
persurface) of the vector 3-volume element dx
3
k
, and e = [e
a
[ is the determinant of the spatial vierbein, so

that 1/e = [e
a
[ is the determinant of the inverse spatial vierbein. Instead of the scalar Lagrangian L
g
, it is
equally possible to work with the Lagrangian density L
g
,
S
g
=
_
t
f
ti
L
g
dt d
3
x with L
g
L
g
/e . (13.27)
Either way, the important thing is to be careful to get factors in the integration measure right when inte-
grating by parts. The following development works with the scalar Lagrangian L
g
, in which case a term
integrates over a scalar 4-volume V to a 3D surface integral over the 3-boundary V of the volume provided
that the term is a covariant 4-divergence,
_
V
D A dx
4
=
_
V
A dx
3
. (13.28)
The least action principle demands that the conditions on the initial and nal hypersurfaces t
i
and t
f
be
considered xed, and asserts that the path followed by the system between the xed initial and nal conditions
is such that the action is minimized. The equations of motion that result from extremizing the action are
unaected by terms in the scalar Lagrangian that are covariant 4-divergences, since these integrate to surface
terms that are asserted to be xed, and therefore unchanged by variation. The ADM expression (13.22) for
the Ricci scalar involves two terms, 2
0
K and (2/)D
a
D
a
, that contain second derivatives of the vierbein
coecients, and therefore demand to be integrated by parts to bring the Lagrangian into Hamiltonian form,
depending only on rst derivatives. To integrate the
0
K term by parts, it is necessary to express it in terms
of a covariant 4-divergence, which is accomplished by
0
K = D
m
K
m
K
2
, (13.29)
1
Technically, the scalar volume element dx
4
is the quadvector, or pseudoscalar, 4-volume element, the dierential 4-form
1
4!
klmn
dx
k
dx
l
dx
m
dx
n
. Likewise the vector 3-volume element dx
3
k
is the trivector, or pseudovector, 3-volume
element, the dierential 3-form
1
3!
klmn
dx
l
dx
m
dx
n
.
198

where K
m
K, 0, 0, 0. Similarly, the (1/)D
a
D
a
term is converted to a 4-divergence by
1
D
a
D
a
= D
m
m
, (13.30)
where
m
0,
a
with
a
dened by equation (13.12a).
Inserting the ADM expression (13.22) for the Ricci scalar into the Hilbert action (13.24), and integrating
the
0
K and (1/)D
a
D
a
terms by parts, yields the ADM gravitational action
S
g
=
1
8
__
Kdx
3
0
_
t
f
ti
+
1
8
_
V
a
dx
3
a
+
1
16
_
t
f
ti
_
K
ab
K
ab
K
2
+R
(3)
_
dx
4
. (13.31)
For the purpose of extremizing the action, the surface terms can be discarded. Thus the action to be
extremized is the one with the ADM Lagrangian
L
ADM
=
1
16
_
K
ab
K
ab
K
2
+R
(3)
_
. (13.32)
According to the usual procedure, conjugate momenta are obtained as partial derivatives of the Lagrangian
with respect to velocities. In the present instance, the velocities are time derivatives of the coordinates. Now
the only things containing time derivatives in the ADM Lagrangian (13.32) are those contained in the terms
involving the extrinsic curvature K
ab
(the spatial Ricci scalar R
(3)
contains no time derivatives). From the
expression (13.14b) for the extrinsic curvature, together with equations (13.15), the time derivatives in the
extrinsic curvature are the directed time derivatives
0
ab
of the spatial tetrad metric, and the directed time
derivatives
0
e
c
of the spatial inverse vierbein. However, these time derivatives appear only the combination
ab
d
ab0
d
ba0
= e
a
e
b
0
(
cd
e
c
e
d
) = e
a
e
b
0
g
. (13.33)
Thus the ADM Lagrangian picks out the natural coordinates as being the spatial components g
of the
coordinate metric, since only time derivatives of these appear in the ADM Lagrangian. An expression for
the extrinsic curvature that demonstrates explicitly its dependence on the time derivatives of the spatial
coordinate metric g
is, from manipulating equation (13.14b),

K
ab
=
1
2
_
e
a
e
b
t
D
a
e
bt
D
b
e
at
_
. (13.34)
Conjugate momenta are obtained by dierentiating the Lagrangian with respect to the velocities. The
derivatives of the extrinsic curvature with respect to the velocities g
/t are
K
ab
g
=
1
2
e
a
e
b
. (13.35)
The conjugate momenta
are therefore

L
ADM
g
=
1
16
_
K
K
_
, (13.36)
13.7 ADM equations of motion 199
in which the factor of is introduced to convert the conjugate momenta into a tensor, as opposed to a tensor
density. Projected into the tetrad frame, the conjugate momenta are
ab
= e
a
e
b
=
1
16
_
K
ab
ab
K
_
. (13.37)
The ADM Lagrangian (13.32) can be rewritten in terms of the conjugate momenta (13.37) as
L
ADM
= 2K
ab
ab
G
0
0
8
, (13.38)
in which the 3D Ricci scalar R
(3)
has been eliminated in favour of the time-time component G
0
0
of the tetrad-
frame Einstein tensor by equation (13.23a). Substituting the expression (13.34) for the extrinsic curvature
K
ab
brings the ADM Lagrangian to
L
ADM
=
1
_
e
a
e
b
t
D
a
e
bt
D
b
e
at
_
ab
G
0
0
8
. (13.39)
The two terms D
a
e
bt
and D
b
e
at
can be combined into one because of the symmetry of
ab
, and integrated
by parts to give the ADM action
S
ADM

_
t
f
ti
L
ADM
dx
4
= 2
_
V
e
bt
ab
dx
3
a
+
_
t
f
ti
_
e
a
e
b
t

ab
+ 2 e
at
D
b
ab
G
0
0
8
_
dt dx
3
0
.
(13.40)
Once again, for the purposes of extremizing the action, the surface term can be discarded. Eliminating the
D
b
ab
term in favour of the time-space part G
0
a
of the tetrad-frame Einstein tensor by equation (13.23b)
produces
S
ADM
=
_
t
f
ti
_
e
a
e
b
t

ab
e
a
t
G
0
a
8
e
0
t
G
0
0
8
_
dt dx
3
0
. (13.41)
This is the ADM gravitational action in desired Hamiltonian form. Compactly,
S
ADM
=
_
t
f
ti
_
g
H
_
dt dx
3
0
(13.42)
where H is the Hamiltonian
H
e
m
t
G
0
m
8
=
G
0
t
8
. (13.43)
13.7 ADM equations of motion
Equations of motion follow from extremizing the ADM action (13.42) in Hamiltonian form. To accomplish
this, the Hamiltonian H must be expressed in terms of the coordinates and momenta. The expression (13.42)
for the ADM action shows that the natural dynamical coordinates and momenta of the system are the
200

components of the spatial coordinate-frame metric g
and their conjugate momenta
. In terms of these,
the Hamiltonian is
H =
1
8
_
G
0
0
+
G
0
_
=
_
16
_
)
2
R
(3)
16
_
_
2 D
_
, (13.44)
where the expressions for G
0
0
and G
0
are from equations (13.23a) and (13.23b) recast into coordinate-frame

quantities, the extrinsic curvature K
being eliminated in favour of the momenta
by equation (13.36).
The expression (13.44) for the Hamiltonian depends not only on the coordinates g
and momenta
, but
also on the lapse and shift
, which are independent of g
and
. Consequently the lapse and shift must

also be treated as additional coordinates. The Hamiltonian (13.44) depends on the lapse and shift linearly,
the quantities in braces in equation (13.44) being independent of the lapse and shift. The Hamiltonian (13.44)
also depends on spatial derivatives g
/x
of the coordinates through its dependence on the Riemann

3-scalar R
(3)
and on the spatial connections
associated with the covariant spatial coordinate derivative

D
in the term D
. However, the spatial derivatives g
/x
are to be considered as determined by

the coordinates g
as a function of the spatial coordinates x
on a spatial hypersurface of constant time t,

not as independent coordinates to be varied separately.
Variation of the ADM action (13.42) with respect to the lapse , the shift
, the coordinates g
, and
the momenta
, gives
S
ADM
=
__

dx
3
0
_
t
f
ti
(13.45)
+
_
t
f
ti
_
H
+
H
t
+
H
g
_
g
+
_
g
t

H
_
dt dx
3
0
.
The least action principle asserts that the action is minimized along the actual path taken by the system
between xed initial and nal conditions at t
i
and t
f
. Setting the variation (13.45) of the action equal to
zero with respect to arbitrary variations and
of the lapse and shift yields the constraint equations

H
= 0 ,
H
= 0 . (13.46)
Setting the variation (13.45) of the action equal to zero with respect to arbitrary variations g
and
of the coordinates and momenta yields Hamiltons equations
t
=
H
g
,
g
t
=
H
. (13.47)
13.8 Constraints and energy-momentum conservation
14
The geometric algebra

The geometric algebra is an intuitively appealing formalism that draws together several mathematical threads
relevant to special and general relativity:
1. What is the best way to conceptualize Lorentz transformations (14.6), and to implement them on a
computer (14.20)?
2. How is it that bivectors in 4D spacetime have a natural complex structure, which has been seen to be the
heart of the Newman-Penrose formalism? See 14.17.
3. How can spin-
1
2
objects be incorporated into general relativity? What is a Dirac spinor? See 14.23.
4. How and why do dierential forms work?
This chapter starts by setting up the geometric algebra in n-dimensional Euclidean space R
n
, then gener-
alizes to Minkowski space, where the geometric algebra is called the spacetime algebra. The 4D spacetime
algebra proves to be identical to the Cliord algebra of the Dirac -matrices (which explains the adoption
of the symbol
m
to denote the basis vectors of a tetrad). Although the formalism is presented initially
in Euclidean or Minkowski space, everything generalizes immediately to general relativity, where the basis
vectors
m
form the basis of an orthonormal tetrad at each point of spacetime.
One convention adopted here, which agrees with the convention adopted by OpenGL and the computer
graphics industry, but is opposite to the standard physics convention, is that a rotor R rotates a multivector
a as a

RaR, equation (14.35). This, along with the standard denition (14.16) for the pseudoscalar, has
the consequence that a right-handed rotation corresponds to R = e
i/2
with increasing, and that rotations
accumulate to the right, that is, a rotation R followed by a rotation S is the product RS. By contrast, in the
standard physics convention a Ra
R, a right-handed rotation corresponds to R = e

i/2
, and rotations
accumulate to the left, that is, R followed by S is SR. The convention adopted here also means that a Weyl
or Dirac spinor is isomorphic to a scaled reverse rotor

R, not to R as in the standard physics convention.
In this chapter, boldface denotes a multivector. A rotor is written in normal (not bold) face as a reminder
that, even though a rotor is an even member of the geometric algebra, it can also be regarded as a spin-
1
2
object with a transformation law (14.37) dierent from that (14.35) of multivectors. Later Latin indices
m, n, ... run over both time and space indices 0, 1, 2, 3, while earlier Latin indices i, j, k run over spatial
indices 1, 2, 3 only.
202

a
a
b
a
b
c
Figure 14.1 Multivectors of grade 1, 2, and 3: a vector a (left), a bivector a b (middle), and a trivector
a b c (right).
14.1 Products of vectors
In 3-dimensional Euclidean space R
3
, there are two familiar ways of taking the product of two vectors, the
scalar product and the vector product.
1. The scalar product a b, also known as the dot product or inner product, of two vectors a and b is
a scalar of magnitude [a[ [b[ cos , where [a[ and [b[ are the lengths of the two vectors, and the angle
between them. The scalar product is commutative, a b = b a.
2. The vector product, ab, also known as the cross product, is a vector of magnitude [a[ [b[ sin , directed
perpendicular to both a and b, such that a, b, and a b form a right-handed set. The vector product is
anticommutative, a b = b a.
The denition of the scalar product continues to work ne in a Euclidean space of any dimension, but the
denition of the vector product works only in three dimensions, because in two dimensions there is no vector
perpendicular to two vectors, and in four or more dimensions there are many vectors perpendicular to two
vectors. It is therefore useful to dene a more general version, the outer product (H. Grassmann, 1862 Die
Ausdehnungslehre, Berlin) that works in Euclidean space R
n
of any dimension.
3. The outer product ab, also known as the wedge product, of two vectors a and b is a bivector, a
multivector of dimension 2, or grade 2. The bivector ab is the directed 2-dimensional area, of magnitude
[a[ [b[ sin, of the parallelogram formed by the vectors a and b, as illustrated in Figure 14.1. The bivector
has an orientation, or handedness, dened by circulating the parallelogram rst along a, then along b.
The outer product is anticommutative, ab = b a, like its forebear the vector product.
The outer product can be repeated, so that (ab) c is a trivector, a directed volume, a multivector
of grade 3. The magnitude of the trivector is the volume of the parallelepiped dened by the vectors a, b,
and c, illustrated in Figure 14.1. The outer product is by construction associative, (ab) c = a(b c).
Associativity, together with anticommutativity of bivectors, implies that the trivector ab c is totally
antisymmetric under permutations of the three vectors, that is, it is unchanged under even permutations, and
changes sign under odd permutations. The ordering of an outer product thus denes one of two handednesses.
14.2 Geometric product 203
It is a familiar concept that a vector a can be regarded as a geometric object, a directed length, independent
of the coordinates used to describe it. The components of a vector change when the reference frame changes,
but the vector itself remains the same physical thing. In the same way, a bivector ab is a directed area,
and a trivector ab c is a directed volume, both geometric objects with a physical meaning independent
of the coordinate system.
In two dimensions the triple outer product of any three vectors is zero, ab c = 0, because the volume
of a parallelepiped conned to a plane is zero. More generally, in n-dimensional space R
n
, the outer product
of n + 1 vectors is zero
a
1
a
2
a
n+1
= 0 (n dimensions) . (14.1)
14.2 Geometric product
The inner and outer products oer two dierent ways of multiplying vectors. However, by itself neither
product conforms to the usual desideratum of multiplication, that the product of two elements of a set be
an element of the set. Taking the inner product of a vector with another vector lowers the dimension by
one, while taking the outer product raises the dimension by one.
H. Grassmann H. (1877, Der ort der Hamiltonschen quaternionen in der audehnungslehre, Math. Ann.
12, 375) and W. K. Cliord (1878, Applications of Grassmanns extensive algebra, Am. J. Math. 1, 350)
resolved the problem by dening a multivector as any linear combination of scalars, vectors, bivectors, and
objects of higher grade. Let
1
,
2
, ...,
n
form an orthonormal basis for n-dimensional Euclidean space R
n
.
A multivector in n = 2 dimensions is then a linear combination of
1 ,
1 scalar
1
,
2
,
2 vectors
2
,
1 bivector
(14.2)
forming a linear space of dimension 1 + 2 + 1 = 4 = 2
2
. Similarly, a multivector in n = 3 dimensions is a
linear combination of
1 ,
1 scalar
1
,
2
,
3
,
3 vectors
2
,
2
3
,
3
1
,
3 bivectors
3
,
1 trivector
(14.3)
forming a linear space of dimension 1 + 3 + 3 + 1 = 8 = 2
3
. In general, multivectors in n dimensions form a
linear space of dimension 2
n
, with n!/[m!(nm)!] distinct basis elements of grade m.
A multivector a in n-dimensional Euclidean space R
n
can thus be written as a linear combination of basis
elements
a =
distinct {i,j,...,m} {1,2,...,n}

a
ij...m
j
...
m
(14.4)
the sum being over all 2
n
distinct subsets of 1, 2, ..., n. The index on each component a
ij...m
is a totally
antisymmetric quantity, reecting the total antisymmetry of
i
j
...
m
.
The point of introducing multivectors is to allow multiplication to be dened so that the product of two
204

multivectors is a multivector. The key trick is to dene the geometric product ab of two vectors a and b
to be the sum of their inner and outer products:
ab = a b +ab . (14.5)
That is a seriously big trick, and if you buy a ticket to it, you are in for a seriously big ride. As a particular
example of (14.5), the geometric product of any element
i
of the orthonormal basis with itself is a scalar,
and with any other element of the basis is a bivector:
j
=
_
1 (i = j)
j
(i ,= j) .
(14.6)
Conversely, the rules (14.6), plus distributivity, imply the multiplication rule (14.5). A generalization of the
rule (14.6) completes the denition of the geometric product:
j
...
m
=
i
j
...
m
(i, j, ..., m all distinct) . (14.7)
The rules (14.6) and (14.7), along with the usual requirements of associativity and distributivity, combined
with commutativity of scalars and anticommutativity of pairs of
i
, uniquely dene multiplication over the
space of multivectors. For example, the product of the bivector
1
2
with the vector
1
is
(
1
2
)
1
=
1
1
=
2
1
=
2
. (14.8)
Sometimes it is convenient to denote the wedge product (14.7) of distinct basis elements by the abbreviated
symbol
ij...m
ij...m

i
j
...
m
(i, j, ..., m all distinct) . (14.9)
By construction,
ij...m
is antisymmetric in its indices. The product of two general multivectors a = a
and b = b
, with paired indices implicitly summed over distinct subsets of 1, ..., n, is

ab = a
. (14.10)
Does the geometric algebra form a group under multiplication? No. One of the dening properties of a
group is that every element should have an inverse. But, for example,
(1 +
1
)(1
1
) = 0 (14.11)
shows that neither 1 +
1
nor 1
1
has an inverse.
14.3 Reverse
The reverse of any basis element is dened to be the reversed product
j
...
m

m
...
j
i
. (14.12)
14.4 The pseudoscalar and the Hodge dual 205
The reverse a of any multivector a is the multivector obtained by reversing each of its components. Reversion
leaves unchanged all multivectors whose grade is 0 or 1, modulo 4, and changes the sign of all multivectors
whose grade is 2 or 3, modulo 4. For example, scalars and vectors are unchanged by reversion, but bivectors
and trivectors change sign. Reversion satises
a +b = a +

b , (14.13)
ab =

b a . (14.14)
Among other things, it follows that the reverse of any product of multivectors is the reversed product, as
you would hope:
ab ... c = c ...
b a . (14.15)
14.4 The pseudoscalar and the Hodge dual
Orthogonal to any m-dimensional subspace of n-dimensional space is an (nm)-dimensional space, called
the Hodge dual space. For example, the Hodge dual of a bivector in 2 dimensions is a 0-dimensional
object, a pseudoscalar. Similarly, the Hodge dual of a bivector in 3 dimensions is a 1-dimensional object, a
pseudovector.
Dene the pseudoscalar i
n
in n dimensions to be
i
n

1
2
...
n
(14.16)
with reverse
i
n
= ()
[n/2]
2
...
n
. (14.17)
The quantity [n/2] in equation (14.17) signies the largest integer less than or equal to n/2. The square of
the pseudoscalar is
i
2
n
= ()
[n/2]
=
_
1 if n = (0 or 1) modulo 4
1 if n = (2 or 3) modulo 4 .
(14.18)
The pseudoscalar anticommutes (commutes) with vectors a, that is, with multivectors of grade 1, if n is
even (odd):
i
n
a = ai
n
if n is even
i
n
a = ai
n
if n is odd .
(14.19)
This implies that the pseudoscalar i
n
commutes with all even grade elements of the geometric algebra, and
that it anticommutes (commutes) with all odd elements of the algebra if n is even (odd).
Exercise 14.1 Prove that the only multivectors that commute with all elements of the algebra are linear
combinations of the scalar 1 and, if n is odd, the pseudoscalar i
n
.
206

The Hodge dual

a of a multivector a in n dimensions is dened by premultiplication by the pseudoscalar
i
n
,
a i
n
a . (14.20)
In 3 dimensions, the Hodge duals of the basis vectors
i
are the bivectors
i
3
1
=
2
3
, i
3
2
=
3
1
, i
3
3
=
1
2
. (14.21)
Thus in 3 dimensions the bivector ab is seen to be the pseudovector Hodge dual to the familiar vector
product a b:
ab = i
3
a b . (14.22)
14.5 Reection
Multiplying a vector (a multivector of grade 1) by a vector shifts the grade (dimension) of the vector by
1. Thus, if one wants to transform a vector into another vector (with the same grade, one), at least two
multiplications by a vector are required.
The simplest non-trivial transformation of a vector a is
n : a nan (14.23)
in which the vector a is multiplied on both left and right with a unit vector n. If a is resolved into components
a
and a
respectively parallel and perpendicular to n, then the transformation (14.23) is

n : a
+a
(14.24)
which represents a reection of the vector a through the axis n, a reversal of all components of the vector
n
a
nan
nan
Figure 14.2 Reection of a vector a through axis n.
14.6 Rotation 207
perpendicular to n, as illustrated by Figure 14.2. Note that nan is the reection of a through the
hypersurface normal to n, a reversal of the component of the vector parallel to n.
The operation of left- and right-multiplying by a unit vector n reects not only vectors, but multivectors
a in general:
n : a nan . (14.25)
For example, the product ab of two vectors transforms as
n : ab n(ab)n = (nan)(nbn) (14.26)
which works because n
2
= 1.
A reection leaves any scalar unchanged, n : nn = n
2
= . Geometrically, a reection preserves
the lengths of, and angles between, all vectors.
14.6 Rotation
Two successive reections yield a rotation. Consider reecting a vector a (a multivector of grade 1) rst
through the unit vector m, then through the unit vector n:
mn : a nmamn . (14.27)
Any component a
of a simultaneously orthogonal to both m and n (i.e. m a
= n a
= 0) is unchanged
by the transformation (14.27), since each reection ips the sign of a
:
mn : a
nma
mn = na
n = a
. (14.28)
Rotations inherit from reections the property of preserving the lengths of, and angles between, all vectors.
Thus the transformation (14.27) must represent a rotation of those components a
of a lying in the 2-dim-

ensional plane spanned by m and n, as illustrated by Figure 14.3. To determine the angle by which the
plane is rotated, it suces to consider the case where the vector a
is equal to m (or n, as a check). It is

not too hard to gure out that, if the angle from m to n is /2, then the rotation angle is in the same
sense, from m to n.
m
a
mam
n
nmamn
Figure 14.3 Rotation of a vector a by the bivector mn. Baed? Hey, draw your own picture.
208

For example, if m and n are parallel, so that m = n, then the angle between m and n is /2 = 0 or ,
and the transformation (14.27) rotates the vector a
by = 0 or 2, that is, it leaves a
unchanged. This
makes sense: two reections through the same plane leave everything unchanged. If on the other hand m
and n are orthogonal, then the angle between them is /2 = /2, and the transformation (14.27) rotates
a
by = , that is, it maps a
to a
.
The rotation (14.27) can be abbreviated
R : a

RaR (14.29)
where R = mn is called a rotor, and

R = nm is its reverse. Rotors are unimodular, satisfying

RR =
R

R = 1. According to the discussion above, the transformation (14.29) corresponds to a rotation by angle
in the mn plane if the angle from m to n is /2. Then m n = cos /2 and mn = (
1
2
) sin /2,
where
1
and
2
are two orthonormal vectors spanning the mn plane, oriented so that the angle from
1
to
2
is positive /2 (i.e.
1
is the x-axis and
2
the y-axis). Note that the outer product
1
2
is invariant
under rotations in the mn plane, hence independent of the choice of orthonormal basis vectors
1
and
2
.
It follows that the rotor R corresponding to a right-handed rotation by in the
1
2
plane is given by
R = cos

2
+ (
1
2
) sin

2
. (14.30)
It is straightforward to check that the rotor (14.30) rotates the basis vectors
i
as
R :
1

R
1
R =
1
cos +
2
sin , (14.31a)
R :
2

R
2
R =
2
cos
1
sin , (14.31b)
R :
i

R
i
R =
i
(i ,= 1, 2) , (14.31c)
which indeed corresponds to a right-handed rotation by angle in the
1
2
plane. The inverse rotation is
R : a Ra
R (14.32)
with
R = cos

2
(
1
2
) sin

2
. (14.33)
A rotation of the form (14.30), a rotation in a single plane, is called a simple rotation.
A rotation rst by R and then by S transforms a vector a as
RS : a

S

RaRS = RS aRS . (14.34)
Thus the composition of two rotations, rst R and then S, is given by their geometric product RS. In three
dimensions or less, all rotations are simple, but in four dimensions or higher, compositions of simple rotations
can yield rotations that are not simple. For example, a rotation in the
1
2
plane followed by a rotation in
the
3
4
plane is not equivalent to any simple rotation. However, it will be seen in 14.17 that bivectors
in the 4D spacetime algebra have a natural complex structure, which allows 4D spacetime rotations to take
1
2
object 209
a simple form similar to (14.30), but with complex angle and two orthogonal planes of rotation combined
into a complex pair of planes.
Simple rotors are both even and unimodular, and composition preserves those properties. A rotor R is
dened in general to be any even, unimodular (
RR = 1) element of the geometric algebra. The set of rotors

denes a group the rotor group, also referred to here as the rotation group. A rotor R rotates not only
vectors, but multivectors a in general:
R : a

RaR . (14.35)
For example, the product ab of two vectors transforms as
R : ab

R(ab)R = (

RaR)(

RbR) (14.36)
which works because R
R = 1.
Concept question 14.2 If vectors rotate twice as fast as rotors, do bivectors rotate twice as fast as
vectors? What happens to a bivector when you rotate it by radians? Construct a mental picture of a
rotating bivector.
To summarize, the characterization of rotations by rotors has considerable advantages. Firstly, the trans-
formation (14.35) applies to multivectors a of arbitrary grade in arbitrarily many dimensions. Secondly,
the composition law is particularly simple, the composition of two rotations being given by their geometric
product. A third advantage is that rotors rotate not only vectors and multivectors, but also spin-
1
2
objects
indeed rotors are themselves spin-
1
2
objects as might be suspected from the intriguing factor of
1
2
in
front of the angle in equation (14.30).
1
2
object
A rotor was dened in the previous section, 14.6, as an even, unimodular element of the geometric algebra.
As a multivector, a rotor R would transform under a rotation by the rotor S as R

SRS. As a rotor,
however, the rotor R transforms under a rotation by the rotor S as
S : R RS , (14.37)
according to the transformation law (14.34). That is, composition in the rotor group is dened by the
transformation (14.37): R rotated by S is RS.
The expression (14.30) for a simple rotation in the
1
2
plane shows that the rotor corresponding to a
rotation by 2 is 1. Thus under a rotation (14.37) by 2, a rotor R changes sign:
2 : R R . (14.38)
A rotation by 4 is necessary to bring the rotor R back to its original value:
4 : R R . (14.39)
210

Thus a rotor R behaves like a spin-
1
2
object, requiring 2 full rotations to restore it to its original state.
The two dierent transformation laws for a rotor as a multivector, and as a rotor describe two
dierent physical situations. The transformation of a rotor as a multivector answers the question, what is
the form of a rotor R rotated into another, primed, frame? In the unprimed frame, the rotor R transforms
a multivector a to

RaR. In the primed frame rotated by rotor S from the unprimed frame, a
=

SaS, the
transformed rotor is

SRS, since
a
=

SaS

S

RaRS =

S

RSa
SRS . (14.40)
By contrast, the transformation (14.37) of a rotor as a rotor answers the question, what is the rotor corre-
sponding to a rotation R followed by a rotation S?
14.8 A multivector rotation is an active rotation
In most of the rest of this book, indices indicate how an object transforms, so that the notation
a
m
m
(14.41)
indicates a scalar, an object that is unchanged by a transformation, because the transformation of the con-
travariant vector a
m
cancels against the corresponding transformation of the covariant vector
m
. However,
the transformation (14.35) of a multivector is to be understood as an active transformation that rotates the
basis vectors
while keeping the coecients a
xed, as opposed to a passive transformation that rotates

the tetrad while keeping the thing itself unchanged. Thus a multivector a a
(implicit summation over

1, ..., n) is not a scalar under the transformation (14.35), but rather transforms to the multivector
a
given by
R : a
R = a
. (14.42)
An explicit example is the transformation (14.31) of the tetrad axes
i
under a right-handed rotation by
angle .
14.9 2D rotations and complex numbers
Section 14.6 identied the rotation group in n dimensions with the geometric subalgebra of even, unimodular
multivectors. In two dimensions, the even grade multivectors are linear combinations of the basis set
1 ,
1 scalar
i
2
,
1 bivector (pseudoscalar)
(14.43)
forming a linear space of dimension 2. The sole bivector is the pseudoscalar i
2

1
2
, equation (14.16),
the highest grade element in 2 dimensions. The rotor R that generates a right-handed rotation by angle
14.9 2D rotations and complex numbers 211
is, according to equation (14.30),
R = e
/2
= e
i2 /2
= cos

2
+i
2
sin

2
, (14.44)
where = i
2
is the bivector of magnitude .
Since the square of the pseudoscalar i
2
is minus one, the pseudoscalar resembles the pure imaginary i, the
square root of 1. Sure enough, the mapping
i
2
i (14.45)
denes an isomorphism between the algebra of even grade multivectors in 2 dimensions and the eld of
complex numbers
a +i
2
b a +i b . (14.46)
With the isomorphism (14.46), the rotor R that generates a right-handed rotation by angle is equivalent
to the complex number
R = e
i/2
. (14.47)
The associated reverse rotor

R is
R = e
i/2
, (14.48)
the complex conjugate of R. The group of 2D rotors is isomorphic to the group of complex numbers of unit
magnitude, the unitary group U(1),
2D rotors U(1) . (14.49)
Let z denote an even multivector, equivalent to some complex number by the isomorphism (14.46). Ac-
cording to the transformation formula (14.35), under the rotation R = e
i/2
, the even multivector, or complex
number, z transforms as
R : z e
i/2
z e
i/2
= e
i/2
e
i/2
z = z (14.50)
which is true because even multivectors in 2 dimensions commute, as complex numbers should. Equa-
tion (14.50) shows that the even multivector, or complex number, z is unchanged by a rotation. This might
seem strange: shouldnt the rotation rotate the complex number z by in the Argand plane? The an-
swer is that the rotation R : a

RaR rotates vectors
1
and
2
(Exercise 14.3), as already seen in the
transformation (14.31). The same rotation leaves the scalar 1 and the bivector i
2

1
2
unchanged. If
temporarily you permit yourself to think in 3 dimensions, you see that the bivector
1
2
is Hodge dual to
the pseudovector
1
2
, which is the axis of rotation and is itself unchanged by the rotation, even though
the individual vectors
1
and
2
are rotated.
Exercise 14.3 Conrm that a right-handed rotation by angle rotates the axes
i
by
R :
1
e
i/2
1
e
i/2
=
1
cos +
2
sin , (14.51a)
R :
2
e
i/2
2
e
i/2
=
2
cos
1
sin , (14.51b)
212

in agreement with (14.31). The important thing to notice is that the pseudoscalar i
2
, hence i, anticommutes
with the vectors
i
.
14.10 Quaternions
A quaternion can be regarded as a kind of souped-up complex number
q = a +b
1
+b
2
+kb
3
, (14.52)
where a and b
i
(i = 1, 2, 3) are real numbers, and the three imaginary numbers , , k, also denoted
1
,
2
,
3
here for convenience and brevity, are dened to satisfy
1
2
=
2
= k
2
= k = 1 . (14.53)
Remark the dotless , to distinguish these quaternionic imaginaries from other possible imaginaries. A
consequence of equations (14.53) is that each pair of imaginary numbers anticommutes:
= = k , k = k = , k = k = . (14.54)
A quaternion (14.52) can be expressed compactly as a sum of its scalar, a, and vector (actually pseudovector,
as will become apparent below from the isomorphism (14.68)), b, parts
q = a + b , (14.55)
where is shorthand for the triple of quaternionic imaginaries,
, , k
1
,
2
,
3
, (14.56)
and where b b
1
, b
2
, b
3
, and b
i
b
i
(implicit summation over i = 1, 2, 3) is the usual Euclidean dot
product. A fundamentally useful formula, which follows from the dening equations (14.53), is
( a)( b) = a b (a b) (14.57)
where a b is the usual 3D vector product. The product of two quaternions p a + b and q c + d
can thus be written
pq = (a + b)(c + d) = ac b d + (ad +cb b d) . (14.58)
The quaternionic conjugate q of a quaternion q a + b is (the overbar symbol for quaternionic
conjugation distinguishes it from the asterisk symbol

for complex conjugation)
q = a b . (14.59)
1
The choice k = 1 in the denition (14.53) is the opposite of the conventional denition ijk = 1 famously carved by W.
R. Hamilton in the stone of Brougham Bridge while walking with his wife along the Royal Canal to Dublin on 16 October
1843 (S. ODonnell, 1983, William Rowan Hamilton: Portrait of a Prodigy, Boole Press, Dublin). To map to Hamiltons
denition, you can take = i, = j, k = k, or alternatively = i, = j, k = k, or = k, = j, k = i. The adopted
choice k = 1 has the merit that it avoids a treacherous minus sign in the isomorphism (14.68) between 3-dimensional
pseudovectors and quaternions. The present choice also conforms to the convention used by OpenGL and other computer
graphics programs.
14.11 3D rotations and quaternions 213
The quaternionic conjugate of a product is the reversed product of quaternionic conjugates
pq = q p (14.60)
just like reversion in the geometric algebra, equation (14.14) (the choice of the same symbol, an overbar,
to represent both reversion and quaternionic conjugation is not coincidental). The magnitude [q[ of the
quaternion q a + b is
[q[ = ( qq)
1/2
= (q q)
1/2
= (a
2
+b b)
1/2
= (a
2
+b
2
1
+b
2
2
+b
2
3
)
1/2
. (14.61)
The inverse q
1
of the quaternion, satisfying qq
1
= q
1
q = 1, is
q
1
= q/( qq) = (a b)/(a
2
+b b) = (a
1
b
1
2
b
2
3
b
3
)/(a
2
+b
2
1
+b
2
2
+b
2
3
) . (14.62)
14.11 3D rotations and quaternions
As before, the rotation group is the group of even, unimodular multivectors of the geometric algebra. In
three dimensions, the even grade multivectors are linear combinations of the basis set
1 ,
1 scalar
i
3
1
, i
3
2
, i
3
3
,
3 bivectors (pseudovectors)
(14.63)
forming a linear space of dimension 4. The three bivectors are pseudovectors, equation (14.21). The squares
of the pseudovector basis elements are all minus one,
(i
3
1
)
2
= (i
3
2
)
2
= (i
3
3
)
2
= 1 , (14.64)
and they anticommute with each other,
(i
3
1
)(i
3
2
) = (i
3
2
)(i
3
1
) = i
3
3
,
(i
3
2
)(i
3
3
) = (i
3
3
)(i
3
2
) = i
3
1
, (14.65)
(i
3
3
)(i
3
1
) = (i
3
1
)(i
3
3
) = i
3
2
.
The rotor R that generates a rotation by angle right-handedly about unit vector n in 3 dimensions is,
according to equation (14.30),
R = e
/2
= e
i3 n/2
= cos

2
+i
3
n sin

2
. (14.66)
where is the bivector
i
3
n (14.67)
of magnitude and unit vector direction n
i
n
i
.
Comparison of equations (14.64) and (14.65) to equations (14.53) and (14.54), shows that the mapping
i
3
i

i
(i = 1, 2, 3) (14.68)
214

denes an isomorphism between the space of even multivectors in 3 dimensions and the non-commutative
division algebra of quaternions
a +i
3
i
b
i
a +
i
b
i
. (14.69)
With the equivalence (14.69), the rotor R that generates a rotation by angle right-handedly about unit
vector n in 3 dimensions is equivalent to the quaternion
R = e
/2
= e
n/2
= cos

2
+ n sin

2
, (14.70)
where is the pseudovector quaternion
n (
1
n
1
+
2
n
2
+
3
n
3
) (14.71)
whose magnitude is [[ = and whose unit direction is

/ = n. The associated reverse rotor

R is
R = e
/2
= e
n/2
= cos

2
n sin

2
, (14.72)
the quaternionic conjugate of R.
The group of rotors is isomorphic to the group of unit quaternions, quaternions q = a +
1
b
1
+
2
b
2
+
3
b
3
satisfying q q = a
2
+b
2
1
+b
2
2
+b
2
3
= 1. Unit quaternions evidently dene a unit 3-sphere in the 4-dimensional
space of coordinates a, b
1
, b
2
, b
3
. From this it is apparent that the rotor group in 3 dimensions has the
geometry of a 3-sphere S
3
.
Exercise 14.4 This exercise is a precursor to Exercise 14.15. Let b
i
b
i
be a 3D vector, a multivector of
grade 1 in the 3D geometric algebra. Use the quaternionic composition rule (14.57) to show that the vector
b transforms under a right-handed rotation by angle about unit direction n =
i
n
i
as
R : b

Rb R = b + 2 sin

2
n
_
cos

2
b + sin

2
n b
_
. (14.73)
Here the cross-product n b denotes the usual vector product, which is dual to the bivector product
nb, equation (14.22). Suppose that the quaternionic components of the rotor R are w, x, y, z, that is,
R = e
n/2
= w+
1
x+
2
y+
3
z. Show that the transformation (14.73) is (note that the 33 rotation matrix
is written to the right of the vector, in accordance with the computer graphics convention that rotations
accumulate to the right opposite to the physics convention; to recover the physics convention, take the
transpose):
R :
_
b
1
b
2
b
2
_
_
b
1
b
2
b
3
_
_
_
w
2
+x
2
y
2
z
2
2(xy+wz) 2(zxwy)
2(xywz) w
2
x
2
+y
2
z
2
2(yz+wx)
2(zx+wy) 2(yzwx) w
2
x
2
y
2
+z
2
_
_
. (14.74)
Conrm that the 3 3 rotation matrix on the right hand side of the transformation (14.74) is an orthogonal
matrix (its inverse is its transverse) provided that the rotor is unimodular, R

R = 1, so that w
2
+x
2
+y
2
+z
2
=
14.12 Pauli matrices 215
1. As a simple example, show that the transformation (14.74) in the case of a right-handed rotation by angle
about the 3-axis (the 12 plane), where w = cos

2
and z = sin

2
, is
R :
_
b
1
b
2
b
2
_
_
b
1
b
2
b
3
_
_
_
cos sin 0
sin cos 0
0 0 1
_
_
. (14.75)
14.12 Pauli matrices

The Pauli matrices
i

1
,
2
,
3
form a vector of 2 2 complex (with respect to a quantum-
mechanical imaginary i) matrices whose three components are each traceless (Tr
i
= 0), Hermitian (
i
=
i
),
and unitary (
i
= 1, no implicit summation):
1

_
0 1
1 0
_
,
2

_
0 i
i 0
_
,
3

_
1 0
0 1
_
. (14.76)
The Pauli matrices anticommute with each other
2
=
1
2
= i
3
,
2
3
=
3
2
= i
1
,
3
1
=
1
3
= i
2
. (14.77)
The particular choice (14.76) of Pauli matrices is conventional but not unique: any three traceless, Hermitian,
unitary, anticommuting 2 2 complex matrices will do. The product of the 3 Pauli matrices is i times the
unit matrix,
3
= i
_
1 0
0 1
_
. (14.78)
The multiplication rules of the Pauli matrices
i
are identical to those of the basis vectors
i
of the 3D
geometric algebra. If the scalar 1 in the geometric algebra is identied with the unit 2 2 matrix, and the
pseudoscalar i
3
is identied with the imaginary i times the unit matrix, then the 3D geometric algebra is
isomorphic to the algebra generated by the Pauli matrices, the Pauli algebra, through the mapping
1
_
1 0
0 1
_
,
i

i
, i
3
i
_
1 0
0 1
_
. (14.79)
The 3D pseudoscalar i
3
commutes with all elements of the 3D geometric algebra.
Concept question 14.5 The Pauli matrices are traceless, Hermitian, unitary, and anticommuting. What
do these properties correspond to in the geometric algebra? Are all these properties necessary for the Pauli
algebra to be isomorphic to the 3D geometric algebra? Are the properties sucient?
The rotation group is the group of even, unimodular multivectors of the geometric algebra. The isomor-
phism (14.79) establishes that the rotation group is isomorphic to the group of complex 2 2 matrices of
216

the form
a +i b , (14.80)
with a, b
i
(i = 1, 3) real, and with the unimodular condition requiring that a
2
+b b = 1. It is straightforward
to check (Exercise 14.6) that the group of such matrices constitutes the group of unitary complex 2 2
matrices of unit determinant, the special unitary group SU(2). The isomorphisms
a +i
3
i
b
i
a +
i
b
i
a +i
i
b
i
(14.81)
have thus established isomorphisms between the group of 3D rotors, the group of unit quaternions, and the
special unitary group of complex 2 2 matrices
3D rotors unit quaternions SU(2) . (14.82)
An isomorphism that maps a group into a set of matrices, such that group multiplication corresponds to
ordinary matrix multiplication, is called a representation of the group. The representation of the rotation
group as 22 complex matrices may be termed the Pauli representation. The Pauli representation is the
lowest dimensional representation of the 3D rotation group.
Exercise 14.6 Translate a rotor into an element of SU(2). Show that the rotor R = e
i3 n
corre-
sponding to a right-handed rotation by angle about unit axis n n
1
, n
2
, n
3
is equivalent to the special
unitary 2 2 matrix
R
_
cos

2
+in
3
sin

2
(n
2
+in
1
) sin

2
(n
2
+in
1
) sin

2
cos

2
in
3
sin

2
_
. (14.83)
Show that the reverse rotor

R is equivalent to the Hermitian conjugate R
of the corresponding 22 matrix.

Show that the determinant of the matrix equals

RR, which is 1.
14.13 Pauli spinors
In the Pauli representation, spin-
1
2
objects are Pauli spinors, 2-dimensional complex (with respect to i)
vectors
=
_

_
(14.84)
that are rotated by pre-multiplying by elements of the special unitary group SU(2). According to the
equivalence (14.83), a rotation by 2 is represented by minus the unit matrix,
_
1 0
0 1
_
. (14.85)
Consequently a rotation by 2 changes the sign of a Pauli spinor . A rotation by 4 is required to rotate
a Pauli spinor back to its original value. Thus a Pauli spinor indeed behaves like a spin-
1
2
object.
14.13 Pauli spinors 217
In quantum mechanics, the Pauli matrices
i
, equations (14.76), provide a representation of the spin
operator s s
i
s
1
, s
2
, s
3
(equation (14.86) is in units = 1; in standard units, s =

2
)
s =
1
2
. (14.86)
The eigenvectors of the spin operator s projected along any axis dene objects of denite spin
1
2
measured along that axis. Each Pauli matrix
i
has two eigenvalues 1, and thus each spin operator
component s
i
has two eigenvalues
1
2
. The eigenvectors of s
i
are spin-up (eigenvalue +
1
2
) and spin-down
(eigenvalue
1
2
), as measured along the i-axis. In particular, the normalized eigenvectors of
3
are

_
1
0
_
,
_
0
1
_
, (14.87)
satisfying
s
3
=
1
2
, s
3
=
1
2
. (14.88)
The Pauli spinor , equation (14.84), can thus be expressed
=
(14.89)
where
and
are the complex amplitudes along the up and down directions of the 3-axis (the z-axis).
Essential to quantum mechanics is the existence of an inner product. The inner product of two Pauli
spinors and is the product
of the Hermitian conjugate of with . The Hermitian conjugate
of
the Pauli spinor (14.84) is
=
_

_
. (14.90)
The magnitude squared of the spinor is the real number
[[
2
=
= [
[
2
+ [
[
2
. (14.91)
In quantum mechanics, the magnitude squared of the spinor [[
2
is interpreted as the total probability (or
probability density) of the particle. The two parts [
[
2
and [
[
2
are the probabilities of the particle being
in the up and down states. The probabilities in the up and down states depend on the direction along which
the spin is measured, but the total probability [[
2
is independent on the choice of direction.
Exercise 14.7 Orthonormal eigenvectors of the spin operator. Show that the orthonormal eigen-
vectors
and
of the spin operator s projected along the unit direction

1
,
2
,
3
are
=
1
_
2(1 +
3
)
_
1 +
3
1
+i
2
_
,
=
1
_
2(1
3
)
_
1 +
3
1
+i
2
_
. (14.92)
218

14.14 Pauli spinors as scaled 3D rotors, or quaternions
Rotors and Pauli spinors both behave like spin-
1
2
objects, requiring a rotation of 4 to bring them full circle.
In fact a Pauli spinor is equivalent (14.96) to a (reverse) 3D rotor

R scaled by a positive real scalar .
Consequently a Pauli spinor is equivalent (14.97) to a quaternion. The 2 complex degrees of freedom of the
Pauli spinor are equivalent to the 4 real degrees of freedom of the quaternion.
Start with the eigenequation (14.88) for the unit spin-up eigenvector in the 3-direction (z-direction),
s
3
=
1
2
. (14.93)
If the spin operator s
3
is rotated by rotor R, and the spin-up eigenvector is pre-multiplied by

R, then the
eigenequation (14.93) transforms into an eigenequation for the unit Pauli spinor

R:
(

Rs
3
R) (

R) =
1
2
(

R) (14.94)
(notice that the isomorphism (14.79) between the geometric algebra and the Pauli algebra guarantees that
the rule s
3

Rs
3
R for rotating the spin operator s
3
is valid regardless of whether s
3
and R are considered
as elements of the geometric algebra or as 2 2 complex matrices in the Pauli algebra). The Pauli spinor
, equation (14.87), is normalized to unit magnitude, and the rotated Pauli spinor

R in equation (14.94)
is likewise of unit magnitude. A general Pauli spinor is the product of a real scalar and a rotated unit
spinor

R,
=
R . (14.95)
The real scalar can be taken without loss of generality to be positive, since any minus sign can be absorbed
into a rotation by 2 of the rotor

R. It is straightforward to check (Exercise 14.8) that any Pauli spinor
can be expressed in the form (14.95). Equation (14.95) establishes an equivalence between Pauli spinors and
reversed 3D rotors

R scaled by a positive real scalar

R . (14.96)
Given the equivalence (14.82) between 3D rotors and unit quaternions, it follows that Pauli spinors are
equivalent to reverse quaternions (see Exercises 14.8 and 14.9 for the precise translation)
Pauli spinors reverse quaternions . (14.97)
The equivalence means that there is a one-to-one correspondence between Pauli spinors and reverse quater-
nions, and that they transform in the same way under 3D rotations.
The Hermitian conjugate
of the Pauli spinor is
R , (14.98)
where
=
_
1 0
_
is the Hermitian conjugate of the spin-up eigenvector . The squared amplitude of the
Pauli spinor
=
2
(14.99)
is the probability (or probability density) of the particle, which is unchanged by a rotation.
14.15 Spacetime algebra 219
One is used to thinking of a Pauli spinor as an instrinsically quantum mechanical object. The equiva-
lence (14.96) between Pauli spinors and scaled reverse rotors shows that Pauli spinors also have a classical
interpretation: they encode a real amplitude , and a rotation

R. This provides a mathematical basis for
the idea that, through their spin, fundamental particles know about the rotational structure of space.
The spin axis

of a Pauli spinor is the direction along which the Pauli spinor is pure up. For example,
the spin axis of the of the spin-up eigenvector is the positive 3-axis (the z-axis), while the spin axis of the
of the spin-down eigenvector is the negative 3-axis. In general, the spin axis of a Pauli spinor (14.95) is
the unit direction

of the rotated 3-axis,
Rs
3
R = s

. (14.100)
The rotor S corresponding to a right-handed rotation by angle about the spin axis

is S = e
i3/2
where
=

. Such a rotation transforms

R

S

R, hence transforms the Pauli spinor (14.95) as

S.
Rotating the Pauli spinor right-handedly by angle about its spin axis leaves the spin axis unchanged,
but multiplies the spinor by a phase e
i/2
,
e
i3/2
= e
i/2
. (14.101)
Exercise 14.8 Translate a Pauli spinor into a quaternion. Given any Pauli spinor
_

_
,
show that the corresponding scaled reverse rotor
R in the Pauli representation (14.76) is the unitary 2 2

matrix
R =
_

_
. (14.102)
Show that the corresponding real quaternion is
q =
R = Re
, Im
, Re
, Im
. (14.103)
Exercise 14.9 Translate a quaternion into a Pauli spinor. Show that if the rotor R corresponds
to a right-handed rotation by angle about unit axis n n
1
, n
2
, n
3
, then the corresponding scaled Pauli
spinor
R is, from (14.83),

R =
_
cos

2
in
3
sin

2
(n
2
in
1
) sin

2
_
. (14.104)
14.15 Spacetime algebra

So far this chapter has concerned itself with ordinary n-dimensional Euclidean space, in which the length
squared of a vector is the sum of the squares of its components. In special relativity, however, the scalar
220

length s of a spacetime interval t, x, y, x is given by s
2
= t
2
+ x
2
+ y
2
+ z
2
. Happily, all the results of
previous sections hold with scarcely a change of stride.
Let
m
(m = 0, 1, 2, 3) denote an orthonormal basis of spacetime, with
0
representing the time axis, and
i
(i = 1, 2, 3) the spatial axes. Geometric multiplication in the spacetime algebra is dened by
n
=
m

n
+
m
n
(14.105)
in the usual way. The key dierence between the spacetime basis
m
and Euclidean bases is that scalar
products of the basis vectors
m
form the Minkowski metric
mn
,
m

n
=
mn
(14.106)
whereas scalar products of Euclidean basis elements
i
formed the unit matrix,
i

j
=
ij
, equation (14.6).
In less abbreviated form, equations (14.105) state that the geometric product of each basis element with
itself is
2
0
=
2
1
=
2
2
=
2
3
= 1 , (14.107)
while geometric products of dierent basis elements
m
anticommute
n
=
n
m
=
m
n
(m ,= n) . (14.108)
In the Dirac theory of relativistic spin-
1
2
particles, the Dirac -matrices are required to satisfy
m
,
n
= 2
mn
(14.109)
where denotes the anticommutator,
m
,
n

m
n
+
n
m
. The multiplication rules (14.109) for
the Dirac -matrices are the same as those for geometric multiplication in the spacetime algebra, equa-
tions (14.107) and (14.108).
A 4-vector a, a multivector of grade 1 in the geometric algebra of spacetime, is
a =
m
a
m
=
0
a
0
+
1
a
1
+
2
a
2
+
3
a
3
. (14.110)
Such a 4-vector a would be denoted ,a in the Dirac slash notation. The product of two 4-vectors a and b is
ab = a b +ab = a
m
b
n
m

n
+a
m
b
n
n
= a
m
b
n
mn
+
1
2
a
m
b
n
[
m
,
n
] . (14.111)
It is convenient to denote three of the six bivectors of the spacetime algebra by
i
,
i

0
i
(i = 1, 2, 3) . (14.112)
The symbol
i
is used because the algebra of bivectors
i
is isomorphic to the algebra of Pauli matrices
i
.
The triple of bivectors
i
will often be denoted shorthandedly by the symbol

1
,
2
,
3
. (14.113)
The pseudoscalar, the highest grade basis element of the spacetime algebra, is denoted I
3
=
1
3
= I . (14.114)
14.16 Complex quaternions 221
The pseudoscalar I satises
I
2
= 1 , I
m
=
m
I , I
i
=
i
I . (14.115)
The basis elements of the 4-dimensional spacetime algebra are then
1 ,
1 scalar
m
,
4 vectors
i
, I
i
,
6 bivectors
I
m
,
4 pseudovectors
I ,
1 pseudoscalar
(14.116)
forming a linear space of dimension 1 + 4 + 6 + 4 + 1 = 16 = 2
4
. The reverse is dened in the usual
way, equation (14.12), leaving unchanged multivectors of grade 0 or 1, modulo 4, and changing the sign of
multivectors of grade 2 or 3, modulo 4:
1 = 1 ,
m
=
m
,
i
=
i
, I
i
= I
i
, I
m
= I
m
,

I = I . (14.117)
The mapping
(3)
i

i
(i = 1, 2, 3) (14.118)
(the superscript
(3)
distinguishes the 3D basis vectors from the 4D spacetime basis vectors) denes an iso-
morphism between the 8-dimensional geometric algebra (14.3) of 3 spatial dimensions and the 8-dimensional
even spacetime subalgebra. Among other things, the isomorphism (14.118) implies the equivalence of the
3D spatial pseudoscalar i
3
and the 4D spacetime pseudoscalar I
i
3
I (14.119)
since i
3
=
1
3
and I =
1
3
.
14.16 Complex quaternions
A complex quaternion (also called a biquaternion by W. R. Hamilton) is a quaternion
q = a +
i
b
i
= a + b (14.120)
in which the four coecients a, b
i
(i = 1, 2, 3) are each complex numbers
a = a
R
+Ia
I
, b
i
= b
i,R
+ Ib
i,I
. (14.121)
The imaginary I is taken to commute with each of the quaternionic imaginaries
i
. The choice of symbol
I is deliberate: in the isomorphism (14.133) between the even spacetime algebra and complex quaternions,
the commuting imaginary I is isomorphic to the spacetime pseudoscalar I.
All of the equations in 14.10 on real quaternions remain valid without change, including the multiplication,
conjugation, and inversion formulae (14.57)(14.62). In the quaternionic conjugate q of a complex quaternion
q a + b,
q = a b , (14.122)
222

the complex coecients a and b are not conjugated with respect to the complex imaginary I. The magnitude
[q[ of a complex quaternion q a + b,
[q[ = ( qq)
1/2
= (q q)
1/2
= (a
2
+b b)
1/2
= (a
2
+b
2
1
+b
2
2
+b
2
3
)
1/2
, (14.123)
is a complex number, not a real number. The complex conjugate q
of the complex quaternion is

q
= a
+ b
, (14.124)
in which the complex coecients a and b are conjugated with respect to the imaginary I, but the quaternionic
imaginaries are not conjugated.
A non-zero complex quaternion can have zero magnitude (unlike a real quaternion), in which case it is
null. The null condition qq = a
2
+b
2
1
+b
2
2
+b
2
3
= 0 is a complex condition. The product of two null complex
quaternions is a null quaternion. Under multiplication, null quaternions form a 6-dimensional subsemigroup
(not a subgroup, because null quaternions do not have inverses) of the 8-dimensional semigroup of complex
quaternions.
Exercise 14.10 Show that any non-trivial null complex quaternion q can be written uniquely in the form
q = (1 +I n)p , (14.125)
where p is a real quaternion, and n is a unit real 3-vector. Equivalently,
q = p(1 +I n
) , (14.126)
where n
is the unit real 3-vector

n
=
pnp
[p[
2
. (14.127)
Solution. Write the null quaternion q as
q = p +Ir (14.128)
where p and r are real quaternions, both of which must be non-zero if q is non-trivial. Then equation (14.125)
is true with
n =
r p
[p[
2
. (14.129)
The null condition is q q = 0. The vanishing of the real part, Re (q q) = p p r r = 0, shows that [p[
2
= [r[
2
.
The vanishing of the imaginary (I) part, Im(q q) = r p +p r = r p +r p = 0 shows that the r p must be a pure
quaternionic imaginary, since the quaternionic conjugate of r p is minus itself, so r p/ [p[
2
must be of the form
n. Its squared magnitude n n = r p p r/ [p[
4
= [r[
2
/ [p[
2
= 1 is unity, so n must be a unit 3-vector. It
follows immediately from the manner of construction that the expression (14.125) is unique, as long as q is
non-trivial.
14.17 Lorentz transformations and complex quaternions 223
14.17 Lorentz transformations and complex quaternions
Lorentz transformations are rotations of spacetime. Such rotations correspond, in the usual way, to even,
unimodular elements of the geometric algebra of spacetime. The basis elements of the even spacetime algebra
are
1 ,
1 scalar
i
, I
i
,
6 bivectors
I ,
1 pseudoscalar
(14.130)
forming a linear space of dimension 1 + 6 + 1 = 8 over the real numbers. However, it is more elegant to
treat the even spacetime algebra as a linear space of dimension 8 2 = 4 over complex scalars of the form
=
R
+ I
I
. The pseudoscalar I qualies as a scalar because it commutes with all elements of the even
spacetime algebra, and it qualies as an imaginary because I
2
= 1. It is convenient to take the basis
elements of the even spacetime algebra over the complex numbers to be
1 ,
1 scalar
I
i
,
3 bivectors
(14.131)
forming a linear space of dimension 1 +3 = 4. The reason for choosing I
i
rather than
i
as the elements of
the basis (14.131) is that the basis is 1, I
i
is equivalent to the basis (14.63) of the even algebra of 3-dim-
ensional Euclidean space through the isomorphism (14.118) and (14.119). This basis in turn is equivalent to
the quaternionic basis 1,
i
through the isomorphism (14.68):
I
i
i
3
(3)
i

i
(i = 1, 2, 3) . (14.132)
In other words, the even spacetime algebra is isomorphic to the algebra of quaternions with complex coe-
cients:
a +I b a + b (14.133)
where a = a
R
+ Ia
I
is a complex number, b = b
R
+ Ib
I
, is a triple of complex numbers, is the triple of
bivectors
i
, and is the triple of quaternionic imaginaries.
The isomorphism (14.133) between even elements of the spacetime algebra and complex quaternions implies
that the group of Lorentz rotors, which are unimodular elements of the even spacetime algebra, is isomorphic
to the group of unimodular complex quaternions
spacetime rotors unit complex quaternions . (14.134)
In 14.11 it was found that the group of 3D spatial rotors is isomorphic to the group of unimodular real
quaternions. Thus Lorentz transformations are mathematically equivalent to complexied spatial rotations.
A Lorentz rotor can be written as a complex quaternion in what looks like the same form as the expres-
sion (14.70) for a 3D spatial rotor, with the dierence that the rotation angle is complex, and the axis n
of rotation is likewise complex. Thus
R = e
/2
= e
n/2
= cos

2
+ n sin

2
(14.135)
224

where is the bivector complex quaternion
n (
1
n
1
+
2
n
2
+
3
n
3
) (14.136)
whose complex magnitude is [[ (
)
1/2
= and whose complex unit direction is

/ n. The angle
=
R
+I
I
is a complex angle, and n = n
R
+In
I
is a complex-valued unit 3-vector, satisfying n n = 1.
The condition n n = 1 of unit normalization is equivalent to the two conditions n
R
n
R
n
I
n
I
= 1 and
2n
R
n
I
= 0 on the real and imaginary parts of n n. The complex angle has 2 degrees of freedom, while
the complex unit vector has 4 degrees of freedom, so the Lorentz rotor R has 6 degrees of freedom, which
is the correct number of degrees of freedom of the group of Lorentz transformations. The associated reverse
rotor

R is
R = e
/2
= e
n/2
= cos

2
n sin

2
(14.137)
the quaternionic conjugate of R. Note that and n in equation (14.137) are not conjugated with respect to
the imaginary I. The sine and cosine of the complex angle appearing in equations (14.135) and (14.137)
are related to its real and imaginary parts in the usual way,
cos

2
= cos

R
2
cosh

I
2
I sin

R
2
sinh

I
2
, sin

2
= sin

R
2
cosh

I
2
+I cos

R
2
sinh

I
2
. (14.138)
In the case of a pure spatial rotation, the angle =
R
and axis n = n
R
in the rotor (14.135) are both
real. The rotor corresponding to a pure spatial rotation by angle
R
right-handedly about unit real axis n
R
is
R = e
nR R/2
= cos

R
2
+ n
R
sin

R
2
. (14.139)
A Lorentz boost is a change of velocity in some direction, without any spatial rotation, and represents
a rotation of spacetime about some time-space plane. For example, a Lorentz boost along the 1-axis (the
x-axis) is a rotation of spacetime in the 01 plane (the tx plane). In the case of a pure Lorentz boost, the
angle = I
I
is pure imaginary, but the axis n = n
R
remains pure real. The rotor corresponding to a boost
by velocity v = tanh
I
in unit real direction n
R
is
R = e
nR II/2
= cosh

I
2
+ n
R
I sinh

I
2
. (14.140)
Exercise 14.11 Factor a general Lorentz rotor R = e
n/2
into the product UL of a pure spatial rotation
U followed by a pure Lorentz boost L. Do the two factors commute?
Exercise 14.12 Show that the geometry of the group of Lorentz rotors is the product of the geometries
of the spatial rotation group and the boost group, which is a 3-sphere times Euclidean 3-space, S
3
R
3
.
14.18 Spatial Inversion (P) and Time Inversion (T)
Spatial inversion, or P for parity, is the operation of reecting all spatial coordinates while keeping the time
coordinate unchanged. Spatial inversion may be accomplished by reecting the spatial vector basis elements
14.19 Electromagnetic eld bivector 225
i

i
, while keeping the time vector basis element
0
unchanged. This results in and I I.
The equivalence I means that the quaternionic imaginary is unchanged. Thus, if multivectors in
the geometric spacetime algebra are written as linear combinations of products of
0
, , and I, then spatial
inversion P corresponds to the transformation
P :
0

0
, , I I . (14.141)
In other words spatial inversion may be accomplished by the rule, take the complex conjugate (with respect
to I) of a multivector.
Time inversion, or T, is the operation of reversing time while keeping all spatial coordinates unchanged.
Time inversion may be accomplished by reecting the time vector basis element
0

0
, while keeping the
spatial vector basis elements
i
unchanged. As with spatial inversion, this results in and I I,
which keeps I hence unchanged. If multivectors in the geometric spacetime algebra are written as linear
combinations of products of
0
, , and I, then time inversion T corresponds to the transformation
T :
0

0
, , I I . (14.142)
For any multivector, time inversion corresponds to the instruction to ip
0
and take the complex conjugate
(with respect to I).
The combined operation PT of inverting both space and time corresponds to
PT :
0

0
, , I I . (14.143)
For any multivector, spacetime inversion corresponds to the instruction to ip
0
, while keeping and I
unchanged.
14.19 Electromagnetic eld bivector
The electromagnetic eld tensor F
mn
can be expressed as the bivector
F =
1
2
F
mn
n
, (14.144)
the factor of
1
2
compensating for the double-counting over indices m and n (the
1
2
could be omitted if the
counting were over distinct bivector indices only). In terms of the electric and magnetic elds E and B, and
the bivector basis elements
1
,
2
,
3
dened by equation (14.112), the electromagnetic eld bivector
F is
F = (E +IB) . (14.145)
14.20 How to implement Lorentz transformations on a computer
The advantages of quaternions for implementing spatial rotations are well-known to 3D game programmers.
Compared to standard rotation matrices, quaternions oer increased speed and require less storage, and
226

their algebraic properties simplify interpolation and splining. Complex quaternions retain similar advantages
for implementing Lorentz transformations. They are fast, compact, and straightforward to interpolate or
spline (Exercises 14.13 and 14.14). Moreover, since complex quaternions contain real quaternions, Lorentz
transformations can be implemented simply as an extension of spatial rotations in 3D programs that use
quaternions to implement spatial rotations.
Lorentz rotors, 4-vectors, spacetime bivectors, and spinors (spin-
1
2
objects) can all be implemented as
complex quaternions. A complex quaternion
q = w +
1
x +
2
y +
3
z (14.146)
with complex coecients w, x, y, z can be stored as the 8-component object
q =
_
w
R
x
R
y
R
z
R
w
I
x
I
y
I
z
I
_
. (14.147)
Actually, OpenGL and other computer software store the scalar (w) component of a quaternion in the last
(fourth) place, but here the scalar components are put in the zeroth position to conform to standard physics
convention. The quaternion conjugate q of the quaternion (14.147) is
q =
_
w
R
x
R
y
R
z
R
w
I
x
I
y
I
z
I
_
, (14.148)
while its complex conjugate q
is
q
=
_
w
R
x
R
y
R
z
R
w
I
x
I
y
I
z
I
_
. (14.149)
A Lorentz rotor R corresponds to a complex quaternion of unit modulus. The unimodular condition

RR =
1, a complex condition, removes 2 degrees of freedom from the 8 degrees of freedom of complex quaternions,
leaving the Lorentz group with 6 degrees of freedom, which is as it should be. Spatial rotations correspond
to real unimodular quaternions, and account for 3 of the 6 degrees of freedom of Lorentz transformations.
A spatial rotation by angle right-handedly about the 1-axis (the x-axis) is the real Lorentz rotor
R = cos(/2) +
1
sin(/2) , (14.150)
or, stored as a complex quaternion,
R =
_
cos(/2) sin(/2) 0 0
0 0 0 0
_
. (14.151)
Lorentz boosts account for the remaining 3 of the 6 degrees of freedom of Lorentz transformations. A Lorentz
boost by velocity v, or equivalently by boost angle = atanh(v), along the 1-axis (the x-axis) is the complex
Lorentz rotor
R = cosh(/2) +I
1
sinh(/2) , (14.152)
14.20 How to implement Lorentz transformations on a computer 227
or, stored as a complex quaternion,
R =
_
cosh(/2) 0 0 0
0 sinh(/2) 0 0
_
. (14.153)
The rule for composing Lorentz transformations is simple: a Lorentz transformation R followed by a Lorentz
transformation S is just the product RS of the corresponding complex quaternions.
The inverse of a Lorentz rotor R is its quaternionic conjugate

R.
Any even multivector q is equivalent to a complex quaternion by the isomorphism (14.133). According to
the usual transformation law (14.35) for multivectors, the rule for Lorentz transforming an even multivector
q is
R : q

RqR (even multivector) . (14.154)
The transformation (14.154) instructs to multiply three complex quaternions

R, q, and R, a one-line expres-
sion in a c++ program.
As an example of an even multivector, the electromagnetic eld F, equation (14.145), is a bivector in the
spacetime algebra. In view of the isomorphism (14.133), the electromagnetic eld bivector F can be written
as the complex quaternion
F =
_
0 B
1
B
2
B
3
0 E
1
E
2
E
3
_
. (14.155)
Under the parity transformation P (14.141), the electric eld E changes sign, whereas the magnetic eld B
does not, which is as it should be:
P : E E , B B . (14.156)
According to the rule (14.154), the electromagnetic eld bivector F Lorentz transforms as F

RFR, which
is a powerful and elegant way to Lorentz transform the electromagnetic eld.
A 4-vector a
m
a
m
is a multivector of grade 1 in the spacetime algebra. A general odd multivector in
the spacetime algebra is the sum of a vector (grade 1) part a and a pseudovector (grade 3) part Ib = I
m
b
m
.
The odd multivector can be written as the product of the time basis vector
0
and an even multivector q
a +Ib =
0
q =
0
_
a
0
+I
i
a
i
Ib
0
+
i
b
i
_
. (14.157)
By the isomorphism (14.133), the even multivector q is equivalent to the complex quaternion
q =
_
a
0
b
1
b
2
b
3
b
0
a
1
a
2
a
3
_
. (14.158)
According to the usual transformation law (14.35) for multivectors, the rule for Lorentz transforming the
odd multivector
0
q is
R :
0
q

R
0
qR =
0

R
qR . (14.159)
In the last expression of (14.159), the factor
0
has been brought to the left, to be consistent with the
convention (14.157) that an odd multivector is
0
on the left times an even multivector on the right. Notice
228

that commuting
0
through

R converts the latter to its complex conjugate (with respect to I)

R
, which
is true because
0
commutes with the quaternionic imaginary , but anticommutes with the pseudoscalar
I. Thus if the components of an odd multivector are stored as a complex quaternion (14.158), then that
complex quaternion q Lorentz transforms as
R : q

R
qR (odd multivector) . (14.160)

The rule (14.160) again instructs to multiply three complex quaternions

R
, q, and R, a one-line expression in

a c++ program. The transformation rule (14.160) for an odd multivector encoded as a complex quaternion
diers from that (14.154) for an even multivector in that the rst factor

R is complex conjugated (with
respect to I).
A vector a diers from a pseudovector Ib in that the vector a changes sign under a parity transformation
P whereas the pseudovector Ib does not. However, the behaviour of a pseudovector under a normal Lorentz
transformation (which preserves parity) is identical to that of a vector. Thus in practical situations two
4-vectors a and b can be encoded into a single complex quaternion (14.158), and Lorentz transformed
simultaneously, enabling two transformations to be done for the price of one.
Finally, a Dirac spinor is equivalent to a complex quaternion q (14.23). It Lorentz transforms as
R : q

Rq (spinor) . (14.161)
Exercise 14.13 Interpolate a Lorentz transformation. Argue that the interpolating Lorentz rotor
R(x) that corresponds to uniform rotation and acceleration between initial and nal Lorentz rotors R
0
and
R
1
as the parameter x varies uniformly from 0 to 1 is
R(x) = R
0
exp [xln(R
1
/R
0
)] . (14.162)
What are the exponential and logarithm of a complex quaternion in terms of its components? Address the
issue of the multi-valued character of the logarithm.
Exercise 14.14 Spline a Lorentz transformation. A spline is a polynomial that interpolates between
two points with given values and derivatives at the two points. Conrm that the cubic spline of a real
function f(x) with given initial and nal values f
0
and f
1
and given initial and nal derivatives f
0
and f
1
at x = 0 and x = 1 is
f(x) = f
0
+f
0
x + [3(f
1
f
0
) 2f
0
f
1
] x
2
+ [2(f
0
f
1
) +f
0
+f
1
] x
3
. (14.163)
The case in which the derivatives at the endpoints are set to zero, f
0
= f
1
= 0, is called the natural spline.
Argue that a Lorentz rotor can be splined by splining the quaternionic components of the logarithm of the
Lorentz rotor.
Exercise 14.15 The wrong way to implement a Lorentz transformation. The purpose of this
exercise is to persuade you that Lorentz transforming a 4-vector by the rule (14.160) is a much better idea
than Lorentz transforming by multiplying by an explicit 4 4 matrix. Suppose that the Lorentz rotor R is
14.21 Dirac matrices 229
the complex quaternion
R =
_
w
R
x
R
y
R
z
R
w
I
x
I
y
I
z
I
_
. (14.164)
Show that the Lorentz transformation (14.160) transforms the 4-vector components a
m
= a
0
, a
1
, a
2
, a
3
as
(note that the 4 4 rotation matrix is written to the right of the 4-vector in accordance with the computer
graphics convention that rotations accumulate to the right opposite to the physics convention; to recover
the physics convention, take the transpose):
R :
_
a
0
a
1
a
2
a
3
_
_
a
0
a
1
a
2
a
3
_
_
_
_
_
[w[
2
+[x[
2
+[y[
2
+[z[
2
2 (w
R
x
I
w
I
x
R
+y
R
z
I
y
I
z
R
)
2 (w
R
x
I
w
I
x
R
y
R
z
I
+y
I
z
R
) [w[
2
+[x[
2
[y[
2
[z[
2
2 (w
R
y
I
w
I
y
R
z
R
x
I
+z
I
x
R
) 2 (x
R
y
R
+x
I
y
I
w
R
z
R
w
I
z
I
)
2 (w
R
z
I
w
I
z
R
x
R
y
I
+x
I
y
R
) 2 (z
R
x
R
+z
I
x
I
+w
R
y
R
+w
I
y
I
)
2 (w
R
y
I
w
I
y
R
+z
R
x
I
z
I
x
R
) 2 (w
R
z
I
w
I
z
R
+x
R
y
I
x
I
y
R
)
2 (x
R
y
R
+x
I
y
I
+w
R
z
R
+w
I
z
I
) 2 (z
R
x
R
+z
I
x
I
w
R
y
R
w
I
y
I
)
[w[
2
[x[
2
+[y[
2
[z[
2
2 (y
R
z
R
+y
I
z
I
+w
R
x
R
+w
I
x
I
)
2 (y
R
z
R
+y
I
z
I
w
R
x
R
w
I
x
I
) [w[
2
[x[
2
[y[
2
+[z[
2
_
_
_
_
, (14.165)
where [ [ signies the absolute value of a complex number, as in [w[
2
= w
2
R
+w
2
I
. As a simple example, show
that the transformation (14.165) in the case of a Lorentz boost by velocity v along the 1-axis, where the
rotor R takes the form (14.153), is
R :
_
a
0
a
1
a
2
a
3
_
_
a
0
a
1
a
2
a
3
_
_
_
_
_
v 0 0
v 0 0
0 0 1 0
0 0 0 1
_
_
_
_
, (14.166)
with the familiar Lorentz gamma factor
= cosh =
1
(1 v
2
)
1/2
, v = sinh =
v
(1 v
2
)
1/2
. (14.167)
14.21 Dirac matrices

The multiplication rules (14.105) for the basis vectors
m
of the spacetime algebra are identical to the
rules (14.109) governing the Cliord algebra of the Dirac -matrices used in the Dirac theory of relativistic
spin-
1
2
particles.
The Dirac -matrices are conventionally represented by 4 4 complex matrices. To ensure consistency
230

between the relativistic and quantum mechanical ways of taking the scalar (inner) product, it is desirable to
require that taking the Hermitian conjugate of any of the basis vectors
m
be equivalent to raising its index,
m
=
m
. (14.168)
Given the requirement (14.168), the matrices representing the basis vectors
m
must be traceless (because
a trace is a scalar, and the basis vectors cannot contain any scalar part), Hermitian or anti-Hermitian as
the self-product of the matrix is 1 (so that
m
=
m
=
mn
n
), and unitary and anticommuting (so that
m

n
=
m

n
=
m
n
). The precise choice of matrices is not fundamental: any set of 4 matrices satisfying
these conditions will do.
The high-energy physics community conventionally adopts the + metric signature, which is opposite
to the convention adopted here. With the high-energy + signature, the standard convention for the
Dirac -matrices is
0
=
_
1 0
0 1
_
,
i
=
_
0
i
i
0
_
, (14.169)
where 1 denotes the unit 2 2 matrix, and
i
denote the three 2 2 Pauli matrices (14.76). The choice of
0
as a diagonal matrix is motivated by Diracs discovery that eigenvectors of the time basis vector
0
with
eigenvalues of opposite sign dene particles and antiparticles in their rest frames (see 14.22). To convert to
the +++ metric signature adopted here while retaining the conventional set of eigenvectors, an additional
factor of i must be inserted into the -matrices:
0
= i
_
1 0
0 1
_
,
i
= i
_
0
i
i
0
_
, (14.170)
In the representation of equations (14.169) or (14.170), the bivectors
i
and I
i
and the pseudoscalar I of
the spacetime algebra are
i
=
_
0
i
i
0
_
, I
i
= i
_

i
0
0
i
_
, I = i
_
0 1
1 0
_
, (14.171)
whose representation as matrices is the same for either signature +++ or +. The Hermitian conjugates
of the bivector and pseudoscalar basis elements are
i
=
i
, (I
i
)
= I
i
, I
= I . (14.172)
The conventional chiral matrix
5
of Dirac theory is dened by
5
i
0
3
= iI =
_
0 1
1 0
_
, (14.173)
whose representation is again the same for either signature +++ or +. The chiral matrix is Hermitian
5
=
5
. (14.174)
14.22 Dirac spinors 231
14.22 Dirac spinors
In the Dirac theory of relativistic spin-
1
2
particles, a Dirac spinor is represented as a 2-component column
vector of Pauli spinors
and
, comprising 4 complex (with respect to i) components and hence 8 degrees

of freedom,
=
_

_
=
_
_
_
_
_
_
_
_
. (14.175)
The Dirac -matrices operate by pre-multiplication on Dirac spinors , yielding other Dirac spinors.
In the Dirac representation (14.170), the four unit Dirac spinors
=
_
_
_
_
1
0
0
0
_
_
_
_
, =
_
_
_
_
0
1
0
0
_
_
_
_
, =
_
_
_
_
0
0
1
0
_
_
_
_
, =
_
_
_
_
0
0
0
1
_
_
_
_
, (14.176)
are eigenvectors of the time basis vector
0
and of the bivector I
3
, with and denoting eigenvectors of
0
, and and eigenvectors of I
3
,
0
= i ,
0
= i , I
3
= i , I
3
= i . (14.177)
The bivector I
3
is the generator of a spatial rotation about the 3-axis (z-axis), equation (14.132). The four
eigenvectors (14.177) form an orthonormal basis
A pure spin-up state can be rotated into a pure spin-down state , or vice versa, by a spatial rotation
about the 1-axis or 2-axis. By contrast, a pure time-up state cannot be rotated into a pure time-down
state , or vice versa, by any Lorentz transformation. Consider for example trying to rotate the pure time-up
spin-up state into any combination of pure time-down states. According to the expression (14.192), the
Dirac spinor obtained by Lorentz transforming the state is pure only if the corresponding complex
quaternion q is pure imaginary. But a pure imaginary quaternion has negative squared magnitude qq, so
cannot be equivalent to any rotor of unit magnitude.
Thus the pure time-up and pure time-down states and are distinct states that cannot be transformed
into each other by any Lorentz transformation. The two states represent distinct species, particles and
antiparticles.
Although a pure time-up state cannot be transformed into a pure time-down state or vice versa by any
Lorentz transformation, the time-up and time-down eigenstates and do mix under Lorentz transforma-
tions. The manner in which Dirac spinors transform is described in 14.23.
The choice of time-axis
0
and spin-axis
3
with respect to which the eigenvectors are dened can of course
be adjusted arbitrarily by a Lorentz boost and a spatial rotation. The eigenvectors of a particular time-axis
0
correspond to particles and antiparticles that are at rest in that frame. The eigenvectors associated with
a particular spin-axis
3
correspond to particles or antiparticles that are pure spin-up or pure spin-down in
that frame.
232

14.23 Dirac spinors as complex quaternions
In 14.14 it was found that a spin-
1
2
object in 3D space, a Pauli spinor, is isomorphic to a scaled 3D reverse
rotor, or real quaternion. In the relativistic theory, the corresponding spin-
1
2
object, a Dirac spinor , is
isomorphic (14.181) to a complex quaternion. The 4 complex degrees of freedom of the Dirac spinor are
equivalent to the 8 degrees of freedom of a complex quaternion. A physically interesting complication arises
in the relativistic case because a non-trivial Dirac spinor can be null, with zero magnitude, whereas any non-
trivial Pauli spinor is necessarily non-null. The case of non-null (massive) and null (massless) Dirac spinors
are considered respectively in 14.24 and 14.25. The present section establishes an isomorphism (14.181)
between Dirac spinors and complex quaternions that is valid in general, regardless of whether the Dirac
spinor is null or not.
If a is a spacetime multivector, equivalent to an element of the Cliord algebra of Dirac -matrices, then
under rotation by Lorentz rotor R, the multivector a operating on the Dirac spinor transforms as
R : a (

RaR)(

R) =

Ra . (14.178)
This shows that a Dirac spinor Lorentz transforms, by construction, as
R :

R . (14.179)
The rule (14.179) is precisely the transformation rule for reverse spacetime rotors under Lorentz trans-
formations: under a rotation by rotor R, a reverse rotor

S transforms as

S

R
S. More generally, the

transformation law (14.179) holds for any linear combination of Dirac spinors . The isomorphism (14.134)
between spacetime rotors and unit quaternions shows that unit Dirac spinors are isomorphic to unit (reverse)
complex quaternions. The algebra of linear combinations of unit complex quaternions is just the algebra of
complex quaternions. Thus the algebra of Dirac spinors is isomorphic to the algebra of (reverse) complex
quaternions. Specically, any Dirac spinor can be expressed uniquely in the form of a 4 4 matrix q, the
Dirac representation of a reverse complex quaternion q, acting on the time-up spin-up eigenvector (the
precise translation between Dirac spinors and complex quaternions is left as Exercises 14.16 and 14.17):
= q . (14.180)
In this section (including the Exercises) the 4 4 matrix q is written in boldface to distinguish it from the
quaternion q that it represents; but the distinction is not fundamental, so the temporary boldface notation
is dropped in subsequent sections. The equivalence (14.180) establishes that Dirac spinors are isomorphic to
reverse complex quaternions
q . (14.181)
The isomorphism means that there is a one-to-one correspondence between Dirac spinors and reverse
complex quaternions q, and that they transform in the same way under Lorentz transformations.
Notwithstanding the isomorphism (14.181), Dirac spinors dier from complex quaternions in that they
have an additional structure that is essential to quantum mechanics, an inner product
2
of two Dirac
14.23 Dirac spinors as complex quaternions 233
spinors
1
and
2
. The inner product
2
is a complex (with respect to i) number. The Hermitian
conjugate
of a Dirac spinor (14.180) is dened to be
= ()
, (14.182)
where ()
=
_
1 0 0 0
_
is the Hermitian conjugate of the time-up spin-up eigenvector , and q
is
the Hermitian conjugate of the matrix q. A related spinor is the the reverse, or adjoint, spinor , dened to
be
= ()
q , (14.183)
where q is the reverse of the matrix q. The Hermitian conjugate spinor
is related to the adjoint spinor

by (Exercise 14.18)
= i
0
. (14.184)
The product of the adjoint spinor with is a Lorentz-invariant scalar, as follows from the Lorentz
invariance of q q. On the other hand, the product
of the Hermitian conjugate spinor
with is the
time component of a 4-vector
i
m
. (14.185)
Exercise 14.16 Translate a Dirac spinor into a complex quaternion. Given any Dirac spinor
=
_
_
_
_
_
_
_
_
, (14.186)
show that the corresponding reverse complex quaternion q, and the equivalent 4 4 matrix q in the Dirac
representation (14.170), such that = q, are (the complex conjugates
a
of the components
a
of the
spinor are with respect to the quantum mechanical imaginary i)
q =
_
Re
Im
Re
Im
Im
Re
Im
Re
_
q =
_
_
_
_
_
_
_
_
. (14.187)
Show that the complex quaternion q (the reverse of q), and the equivalent 4 4 matrix q (the reverse of q)
in the Dirac representation (14.170), are
q =
_
Re
Im
Re
Im
Im
Re
Im
Re
_
q =
_
_
_
_
_
_
_
_
. (14.188)
Conclude that the reverse spinor ()
q is
()
q =
_

_
. (14.189)
234

Exercise 14.17 Translate a complex quaternion into a Dirac spinor. Show that the complex
quaternion q w +x +y +kz is equivalent in the Dirac representation (14.170) to the 4 4 matrix q
q =
_
w
R
x
R
y
R
z
R
w
I
x
I
y
I
z
I
_
q =
_
_
_
_
w
R
+iz
R
ix
R
+y
R
iw
I
+z
I
x
I
iy
I
ix
R
y
R
w
R
iz
R
x
I
+iy
I
iw
I
z
I
iw
I
+z
I
x
I
iy
I
w
R
+iz
R
ix
R
+y
R
x
I
+iy
I
iw
I
z
I
ix
R
y
R
w
R
iz
R
_
_
_
_
. (14.190)
Show that the reverse quaternion q, the complex conjugate (with respect to I) quaternion q
, and the reverse

complex conjugate (with respect to I) quaternion q
are respectively equivalent to the 4 4 matrices

q q
0
q
0
, (14.191a)
q
=
0
q
0
, (14.191b)
q
=
0
q
0
. (14.191c)
Conclude that the Dirac spinor q corresponding to the reverse complex quaternion q is
q =
_
_
_
_
w
R
iz
R
ix
R
+y
R
iw
I
z
I
x
I
iy
I
_
_
_
_
, (14.192)
that the reverse spinor ()
q is
()
q =
_
w
R
+iz
R
ix
R
+y
R
iw
I
+z
I
x
I
iy
I
_
, (14.193)
and that the Hermitian conjugate spinor
()
is
()
=
_
w
R
+iz
R
ix
R
+y
R
iw
I
z
I
x
I
+iy
I
_
. (14.194)
Hence conclude that and
are respectively the real part of, and the absolute value of, the complex
magnitude squared qq
2
of the complex quaternion q,
=
2
R
2
I
, (14.195a)
=
2
R
+
2
I
, (14.195b)
with
2
R
= w
2
R
+x
2
R
+y
2
R
+z
2
R
, (14.196a)
2
I
= w
2
I
+x
2
I
+y
2
I
+z
2
I
, (14.196b)
14.24 Non-null Dirac spinor particle and antiparticle 235

Exercise 14.18 Relation between
and . Conrm equation (14.184) by showing from equa-

tion (14.191b) that
= i()
q
0
. (14.197)
14.24 Non-null Dirac spinor particle and antiparticle

A non-null, or massive, Dirac spinor , one for which ,= 0, is isomorphic (14.181) to a non-null reverse
complex quaternion q, which can be factored as a non-zero complex scalar times a unit quaternion, a rotor.
Thus a non-null Dirac spinor can be expressed as the product of a complex scalar =
R
+I
I
and a reverse
Lorentz rotor

R, acting on the time-up spin-up eigenvector ,
=
R . (14.198)
The complex scalar can be taken without loss of generality to lie in the right hemisphere of the complex
plane (positive real part), since a minus sign can be absorbed into a spatial rotation by 2 of the rotor
R. There is no further ambiguity in the decomposition (14.198) into scalar and rotor, because the squared
magnitude
R =
2
of the scaled rotor
R is the same for any decomposition.

The fact that a non-null Dirac spinor encodes a Lorentz rotor shows that a non-null Dirac spinor in
some sense knows about the Lorentz structure of spacetime. It is intriguing that the Lorentz structure of
spacetime is built in to a non-null Dirac particle.
As discussed in 14.22, a pure time-up eigenvector represents a particle in its own rest frame, while a pure
time-down eigenvector represents an antiparticle in its own rest frame. The time-up spin-up eigenvector
is by denition (14.198) equivalent to the unit scaled rotor,
R = 1, so in this case the scalar is pure real.

Lorentz transforming the eigenvector multiplies it by a rotor, but leaves the scalar unchanged, therefore
pure real. Conversely, if the time-up spin-up eigenvector is multiplied by the imaginary I, then according
to the expression (14.192) the resulting spinor can be Lorentz transformed into a pure state, corresponding
to a pure antiparticle. Thus one may conclude that the real and imaginary parts (with respect to I)
of the complex scalar =
R
+ I
I
correspond respectively to particles and antiparticles. The
magnitude squared
of the Dirac spinor, equation (14.195b),
= [[
2
=
2
R
+
2
I
, (14.199)
is the sum of the probabilities
2
R
of particles and
2
I
of antiparticles.
Among other things, the decomposition of a Dirac spinor into its particle and antiparticle parts shows
that multiplying a non-null Dirac spinor by the pseudoscalar I converts a particle to an antiparticle, and
vice versa.
236

14.25 Null Dirac Spinor
A null spinor is a spinor whose magnitude is zero,
= 0 . (14.200)
Such a spinor is equal to a null complex quaternion q acting on the time-up spin-up eigenvector ,
= q . (14.201)
Physically, a null spinor represents a spin-
1
2
particle moving at the speed of light. A non-trivial null spinor
must be moving at the speed of light because if it were not, then there would be a rest frame where the rotor
part of the spinor =
R would be unity,

R = 1, and the spinor, being non-trivial, ,= 0, would not be
null. The null condition (14.200) is a complex constraint, which eliminates 2 of the 8 degrees of freedom of
a complex quaternion, so that a null spinor has 6 degrees of freedom.
Any non-trivial null complex quaternion q can be written uniquely as the product of a null factor
(1 +I n)/
2 and a real quaternion U (Exercise 14.10):

q =
(1 +I n)
2
U . (14.202)
Here n is a unit real 3-vector, is a positive real scalar, and U is a purely spatial (i.e. real, with no I part)
rotor. The factor of 1/
2 is inserted for normalization purposes. Physically, equation (14.202) contains the

instruction to boost to light speed in the direction n, then scale by the real scalar and rotate spatially by
U. The 2 +1 +3 = 6 degrees of freedom from the real unit vector n, the real scalar , and the spatial rotor
U in the expression (14.202) are precisely the number needed to specify a null quaternion. One might have
thought that the boost factor 1 +I n in equation (14.202) would change under a Lorentz transformation,
but in fact it is Lorentz-invariant. For if the boost factor 1+I n is transformed (multiplied) by any complex
quaternion p +Ir, then the result
(1 +I n)(p +Ir) = (1 +I n)(p nr) (14.203)
is the same unchanged boost factor 1+In multiplied by a purely spatial transformation, the real quaternion
pnr. Equation (14.203) is true because (n)
2
= 1. Since the boost factor 1+In is Lorentz-invariant,
Lorentz transforming the null quaternion q (14.202) probes only 4 of the 6 degrees of freedom of the group
of null quaternions.
The null Dirac spinor corresponding to the reverse q of the null complex quaternion q, equation (14.202),
is
q =
U
(1 I n)
2
. (14.204)
It is natural to choose basis vectors of the representation to be eigenvectors of the Lorentz-invariant boost
factor 1I n. In the Dirac representation (14.170), the basis spinors (14.176) are eigenvectors of I
3
, and
it natural to choose the 3-direction to be in either the positive or negative n direction, in which case
(1 I n) = (1 I
3
) = (1
3
) = ( ) . (14.205)
14.26 Chiral decomposition of a Dirac spinor 237
The null basis vectors are left- and right-handed chiral eigenvectors, eigenvectors of the chiral operator
5
with eigenvalues respectively,
5
()
2
=
()
2
. (14.206)
The 1/
2 factor ensures unit normalization

_
()
2
_
()
2
= 1 . (14.207)
A general null Dirac spinor , equation (14.204), is the real scaled spatial reverse rotor
U acting on one of
the two null chiral basis spinors, either left-handed () /
2, or right-handed (+) /
2,
L
=
U
()
2
,
R
=
U
(+)
2
. (14.208)
A left- or right-handed null Dirac spinor is called a Weyl spinor.
Concept question 14.19 The null boost factor (1 + I n) in a null quaternion, equation (14.202), is
Lorentz-invariant, as shown by equation (14.203) (which you should conrm). Consequently a null Dirac
spinor has a Lorentz-invariant boost axis n. Does a null 4-vector have a Lorentz-invariant axis? What does
it mean physically that a null Dirac spinor has a Lorentz-invariant boost axis n?
14.26 Chiral decomposition of a Dirac spinor
A general (non-null or null) Dirac spinor can be decomposed into a sum of left- and right-handed chiral
components
=
L
+
R
, (14.209)
that are eigenvectors of the chiral operator
5
,
L
R
=
L
R
. (14.210)
The left- and right-handed chiral components can be projected out by applying the chiral projection operators
1
2
(1
5
) (which are projection operators because their squares are themselves):
1
2
(1
5
) =
L
R
. (14.211)
The decomposition into chiral components is Lorentz-invariant because the pseudoscalar I, hence the chiral
operator
5
iI, is Lorentz-invariant, which is true because the pseudoscalar I commutes with any Lorentz
rotor. The chiral projection operators are null, because the reverse of the chiral operator is
5
= iI = iI =
5
, and its square is one,
2
5
= 1, so
_
1 +
5
_
(1 +
5
) = (1
5
)
_
1
5
_
= (1
5
) (1 +
5
) = 0 . (14.212)
238

Consequently each of the chiral components is null, hence massless,

L
L
=
R
R
= 0 . (14.213)
Since a pure left- or right-handed spinor must be null, a non-null Dirac particle cannot be purely left- or
right-handed.
14.27 Dirac equation
The Dirac equation is the relativistic quantum mechanical wave equation for spin-
1
2
particles. By itself, the
Dirac equation does not provide a consistent theory of relativistic quantum mechanics, because in relativistic
quantum mechanics there is no such thing as a single particle that evolves in isolation. Rather, a funda-
mental particle such as an electron is dressed in a sea of particle-antiparticle pairs polarized out of the
vacuum by the presence of the electron. Nevertheless, the Dirac equation is a fundamental building block
for the quantum eld theory of spin-
1
2
particles.
The Dirac theory starts with the momentum 4-vector in the form p =
m
p
m
, where
m
are not only the
basis vectors of an orthonormal tetrad, but also the basis vectors of the spacetime (Cliord) algebra. In the
Dirac slash notation p =,p, but the notation is superuous here. For a particle of rest mass m, the geometric
square of the momentum is
pp =
m
n
p
m
p
n
= p
n
p
n
= m
2
. (14.214)
The vanishing sum pp +m
2
factors as
pp +m
2
= (p +im)(p im) = 0 . (14.215)
The factorization provides the motivation for the Dirac wave equation for a free relativistic spin-
1
2
particle
or antiparticle,
(p im) = 0 (particle) , (14.216a)
(p +im) = 0 (antiparticle) , (14.216b)
in which is a Dirac spinor, and the momentum operator p
m
p
m
should be replaced, according to the
usual rules of quantum mechanics, by minus i times the gradient operator =
m
m
,
p = i . (14.217)
With respect to locally inertial coordinates x
m
t, x
i
,
p
0
= p
0
= i

t
, p
i
= p
i
= i

x
i
. (14.218)
With the replacement (14.217), the Dirac equations (14.216) are
( +m) = 0 (particle) , (14.219a)
( m) = 0 (antiparticle) . (14.219b)
14.28 Antiparticles are negative mass particles moving backwards in time 239
The reason for choosing opposite signs for the mass m in the particle and antiparticle equations is discussed
further in 14.28. For now, two comments can be made about the dierent choice of sign. Firstly, as seen in
14.24, particles and antiparticles belong to distinct representations that do not mix under Lorentz transfor-
mations, so it is consistent to allow them to satisfy dierent equations. Secondly, the factorization (14.215)
involves factors with both signs of m, so it is reasonable one could say demanded that both equations
would occur.
In at (Minkowski) space, the Dirac wave equations (14.219) for a free particle or antiparticle are most
easily solved by Fourier transforming with respect to space and time. The dierential wave equations (14.219)
then revert to being algebraic equations (14.216), with p being the momentum of the corresponding Fourier
mode. If the Dirac spinor is a particle as opposed to an antiparticle, so that is (up to an irrelevant real
scale factor)

R where is any rest-frame particle eigenvector, then the following calculation
p = (

Rm
0
R)(

R) = m
R
0
= im
R = im (particle) , (14.220)
conrms that the particle Dirac equation (14.216a) recovers the expected momentum p =

Rm
0
R, which
is the rest frame momentum p p
m
m
= m
0
Lorentz transformed into the tetrad frame. Likewise, if the
Dirac spinor is an antiparticle as opposed to a particle, so that is (up to an irrelevant real scale factor)
R where is any rest-frame antiparticle eigenvector, then

p = (

Rm
0
R)(

R) = m
R
0
= im
R = im (antiparticle) (14.221)
conrms that the antiparticle Dirac equation (14.216b) again recovers the expected momentum p =

Rm
0
R.
The sign ip between the particle and antiparticle equations (14.220) and (14.221) occurs because the
time basis vector
0
yields opposite signs when acting on a rest-frame particle eigenvector versus rest-
frame antiparticle eigenvector , equation (14.177). The same sign ip occurs if the rest-frame antiparticle
eigenvector is taken to be I (per 14.24) instead of , since
0
anticommutes with I.
Fourier-transformed back into real space, the free Dirac spinor wavefunctions in at space are, for either
particles or antiparticles, FREQUENCY HAS WRONG SIGN
=
0
e
ipmx
m
, (14.222)
where p
m
is the momentum, satisfying p =
m
p
m
, and
0
is the value of the Dirac spinor at the origin
x
m
= 0. Regarded as an element of the spacetime algebra, the exponential factor e
ipmx
m
is a scalar, so it
commutes with
0
, so it does not matter on which side of
0
the exponential is placed.
14.28 Antiparticles are negative mass particles moving backwards in time
The original Dirac treatment took the particle Dirac equation (14.219a) as describing both particles and
antiparticles. This led to solutions in which the free-wave factor contained not only a positive frequency
component, as in equation (14.222), but also a negative frequency component e
ipmx
m
. These negative
frequency components were interpreted as indicating an antiparticle, with negative mass m.
In the original Dirac theory, the prediction of particles with negative rest mass was problematic: where are
240

such particles? And if they existed, why wouldnt pairs of positive and negative mass particles spontaneously
pop out of the vacuum, causing a catastrophic breakdown of the vacuum? To solve the problem, Dirac
proposed that all negative energy states of the vacuum are already occupied, and that antiparticles correspond
to holes in the negative energy sea.
Diracs conundrum was eventually solved by Feynman, who realised that anti-particles are equivalent
to negative mass particles moving backwards in time. A negative mass particle moving backwards in
time looks like a positive mass particle moving forwards in time. Feynmans solution obviates the need for
any negative energy sea of antiparticles. Feynmans dictum corresponds mathematically to choosing particle
and antiparticle spinors not only to yield opposite signs when acted on by the time basis vector
0
in their
rest frames, as in Dirac theory, but also to have the opposite sign of mass m in the Dirac equations (14.219a).
This is the approach adopted in the previous section, 14.27.
14.29 Dirac equation with electromagnetism
The Dirac equation for a spin-
1
2
particle of charge e moving in an external electromagnetic eld with potential
A A
m
m
is given by the same Dirac equations (14.216), but now the momentum p is rewritten in terms
of the canonical momentum ,
p = +eA , (14.223)
and it is the canonical momentum that is replaced by minus i times the gradient operator
m
m
,
= i . (14.224)
Consequently the Dirac equation for charged particle or antiparticle is
( +ieA+m) = 0 (particle) , (14.225a)
( +ieAm) = 0 (antiparticle) . (14.225b)
Equation (14.225b) appears to describe an antiparticle as having mass m opposite to that of a particle,
and charge e the same as that of a particle. If the antiparticle is interpreted as having negative mass
moving backwards in time, then the antiparticle has positive mass m moving forwards in time, and charge
e opposite to that of a particle.
Equations (14.225) are not easy to solve in general. Analytic solutions exist in some cases, such as when
the electromagnetic eld consists of a uniform magnetic eld B.
14.30 CPT
It was seen in 14.24 that multiplying a non-null Dirac spinor by the pseudoscalar I converts a particle
spinor into an antiparticle spinor, and vice versa. This operation is conventionally called CPT,
CPT : I . (14.226)
14.31 Charge conjugation C 241
The operation is called CPT because it is conventionally parsed into 3 distinct discrete transformations C,
P, and T, discussed in turn below. In the spacetime algebra, the CPT operation ips the sign of all the
spacetime axes
m
,
CPT :
m
I
m
I
1
=
m
, (14.227)
which is true because the pseudoscalar I anti-commutes with each of the basis vectors
m
. The CPT
operation leaves all even multivectors unchanged since I commutes with all even multivectors. In particular,
CPT is Lorentz invariant, since I commutes with Lorentz rotors.
The CPT operation converts the particle Dirac equation (14.225a) into the antiparticle Dirac equa-
tion (14.225b):
CPT : I( +ieA+m) = ( +ieAm)I . (14.228)
Equation (14.228) shows that the spinor I satises the Dirac equation (14.225b) for an antiparticle, con-
sistent with conclusion of 14.24 that multiplying a non-null Dirac spinor by I converts a particle into an
antiparticle.
14.31 Charge conjugation C
A Dirac spinor, non-null or null, contains two distinct components that remain separate under Lorentz
transformations. For a non-null spinor the two components are particles and antiparticles. For a null spinor
the two components are the left- and right-handed chiralities. For a non-null spinor, the CPT operation
of multiplying the spinor by I converts a particle into an antiparticle and versa. But for a null spinor,
multiplying by I = i
5
leaves left- and right-handed particles as they are: it does not transform opposite
chiralities into each other.
A charge conjugation operation C can be dened with the property that it converts particles of one type
into the opposite type for both non-null and null spinors: particles into antiparticles and vice versa, and
left-handed into right-handed chiralities and vice versa.
In quantum mechanics, it is natural to regard particles and antiparticles as belonging to complex (with
respect to i) conjugate representations. The charge conjugation operator C is dened by the requirement
that it transforms the spacetime basis vectors
m
to their complex conjugates (with respect to i)
C :
m
C
m
C
1
=
m
. (14.229)
In the Dirac representation (14.170), the condition (14.229) requires that C commute with
2
, but anticom-
mute with
0
,
1
, and
3
. A suitable matrix is
2
itself,
C =
2
. (14.230)
Charge conjugation of a Dirac spinor is accomplished by taking the complex conjugate (with respect to i)
of the spinor, and multiplying by the charge conjugation operator C:
C : C
. (14.231)
242

The charge-conjugates C
of the non-null basis eigenvectors (14.176) in the Dirac representation are

C()
= , C()
= , C()
= , C()
= , (14.232)
while the charge-conjugates of the null chiral basis eigenvectors (14.206) are
C
_
( )
2
_
=
( )
2
. (14.233)
Equations (14.232) and (14.233) show that charge conjugation not only converts rest-frame particle eigen-
vectors into rest-frame antiparticle eigenvectors , and vice versa, but also converts a left-handed null
eigenvector into a right-handed null eigenvector and vice versa. Since the operation of complex conjuga-
tion commutes with Lorentz transformation (because Lorentz rotors R are independent of the quantum
mechanical imaginary i), it is true in general that charge conjugation switches particles into antiparticles,
and left-handed into right-handed chiralities.
The complex conjugate (with respect to i) of the charged particle Dirac equation (14.225a) is
(
ieA
+m)
= 0 . (14.234)
Complex conjugation leaves the components
m
and A
m
of the gradient operator and electromagnetic po-
tential unchanged, but conjugates the basis vectors
m
. Left-multiplying equation (14.234) by C, and
commuting C through the wave operator, yields
( ieA+m)C
= 0 . (14.235)
Thus the charge-conjugated spinor C
satises the Dirac equation (14.225a) for a particle with the same
mass m but opposite charge e.
14.32 Parity reversal P
The parity operation P is the operation of reversing all the spatial axes, while keeping the time axis un-
changed,
P :
m
P
m
P
1
=
_

m
m = 0 ,
m
m = 1, 2, 3 .
(14.236)
A suitable matrix is the time axis
0
P =
0
. (14.237)
Parity reversal transforms a Dirac spinor as
P : P . (14.238)
Parity reversal commutes with spatial rotations, but not with Lorentz boosts. Parity reversal transforms the
left- and right-handed chiral components of a Dirac spinor into each other:
P :
L

R
, (14.239)
14.33 Time reversal T 243
which is true because parity reversal ips the sign of the pseudoscalar, PIP
1
= I, hence also the sign of
the chiral operator
5
iI.
14.33 Time reversal T
The Wigner time reversal operation T is conventionally dened so that CPT is the product of the three
operations C, P, and T. This requires that time-reversal transforms a Dirac spinor as
T : T
(14.240)
with
T =
1
3
, (14.241)
so that
CPT =
2
3
= I . (14.242)
14.34 Majorana spinor
The charge conjugation operation considered in 14.31 switched left- and right-handed null spinors into each
other. However, it is possible for a null spinor to be its own antiparticle. In this case the complex (with
respect to i) conjugate representation is itself, rather than being a distinct representation. A null spinor
which is its own antiparticle is a Majorana spinor.
For a Majorana spinor, complex conjugation should leave the representation unchanged; that is, the -
matrices should be real, in contrast to the Dirac representation (14.170), where the -matrices are complex.
A suitable real representation of the -matrices is
0
=
_
0 i
2
i
2
0
_
,
1
=
_

1
0
0
1
_
,
2
=
_
0 i
2
i
2
0
_
,
3
=
_

3
0
0
3
_
. (14.243)
The chiral matrix
5
iI is
5
=
_

2
0
0
2
_
. (14.244)
14.35 Covariant derivatives revisited
Under a Lorentz transformation by rotor R, any multivector a transforms as a

RaR. The covariant
derivative D
m
D
m
must transform likewise as
R : D

RDR . (14.245)
244

14.36 General relativistic Dirac equation
To convert the Dirac equations (14.219) or (14.225) into general relativistic equations, the derivative
m
must be converted to a covariant derivative.
14.37 3D Vectors as rank-2 spinors
THIS NEEDS TO BE CONVERTED FROM 3D TO SPACETIME.
Concept question 14.20 One is used to thinking of a spin-
1
2
particle as in some sense the square root of
a spin-1 particle, a vector. How is this concept compatible with the idea that a spin-
1
2
object is a (scaled)
rotor?
A Pauli spinor contains two complex components
a
, where the index a runs over the two indices and
. The Pauli spinor is a spinor of rank 1, having one spinor index. Under a rotation by rotor R, the spinor
transforms as

R. Under a rotation, the spinor components
a
of the Pauli spinor transform as
a
R
b
a
b
(14.246)
where R
b
a
is the special unitary 2 2 matrix representing the reverse rotor

R. The rotor R itself has
components R
a
b
. Note the placement of indices: for the rotor R, the rst index is down and the second up,
while for the reverse rotor

R, the rst index is up and the second down. The rotor and its reverse are inverse
to each other, satisfying

RR = R
c
a
R
c
b
=
b
a
.
The Hermitian conjugate spinor
transforms in the opposite way
R, and can therefore be written

as a spinor with contravariant (raised) index,
=
a
. The contravariant spinor components
a
transform
as
b
R
b
a
. (14.247)
The product of a Hermitial conjugate spinor
with another spinor denes their scalar product, which is

unchanged by a rotation,
R =
=
a
a
. (14.248)
Explicitly, the scalar product
a
a
is the complex number
a
=
. (14.249)
An element a of the 3D geometric algebra transforms under rotation as a

RaR. The behaviour under
rotation shows that a 3D multivector is a rank-2 spinor a = a
a
b
, a complex 2 2 matrix, with the rst index
being covariant (lowered), and the second being contravariant (raised). The spinor components a
a
b
of the
multivector transform as
a
a
b
R
c
a
a
c
d
R
d
b
. (14.250)
14.37 3D Vectors as rank-2 spinors 245
A 3D multivector a contains 8 real components, comprising a scalar, a vector, a pseudovector, and a pseu-
doscalar. All 8 complex components are represented by the complex 2 2 spinor a
a
b
. The scalar and
pseudoscalar components are contained in the complex trace a
a
a
of the spinor. The rotational transforma-
tion a

RaR preserves the grade of the multivector, so scalars transform to scalars, vectors to vectors,
pseudovectors to pseudovectors, and pseudoscalars to pseudoscalars.
A reversed multivector a transforms in the same way a

R aR as the parent multivector a. Consquently,
like the multivector a, the reversed multivector a transforms like a rank-2 spinor with one covariant (lowered)
and one contravariant (raised) index. It is consistent to write the reversed spinor as the original spinor with
the rst index raised and the second lowered, a = a
a
b
.
The above has shown that a 3D multivector a is represented naturally as a rank-2 spinor a
a
b
with one
covariant and one contravariant index. This might seem strange: one might have expected that a vector a
spin-1 particle might be represented as a rank-2 spinor a
ab
with two covariant indices a sum of tensor
products =
a
b
of two spin-
1
2
particles. Under a rotation, the components a
ab
of a covariant rank-2
spinor transform as
a
ab
R
c
a
R
d
b
a
cd
. (14.251)
One aspect of this transformation is straightforward: the following combination of tensor products of spinors
is invariant under rotation, and is therefore a scalar:
. (14.252)
The remaining tensor combinations , , and + provide a basis for vectors, but under
rotation they transform into complex, not real, linear combinations of each other. This contrasts with the
representation of multivectors a by spinors a
a
b
with one covariant and one contravariant index, where the
grade-preserving property of the rotational transformation a

RaR ensures that vectors rotate into real,
not complex, linear combinations of each other.
PART SIX
BLACK HOLE INTERIORS
Concept Questions
1. Explain how the equation for the Gullstrand-Painleve metric (15.16) encodes not merely a metric but a
full vierbein.
2. In what sense does the Gullstrand-Painleve metric (15.16) depict a ow of space? [Are the coordinates
moving? If not, then what is moving?]
3. If space has no substance, what does it mean that space falls into a black hole?
4. Would there be any gravitational eld in a spacetime where space fell at constant velocity instead of
accelerating?
5. In spherically symmetric spacetimes, what is the most important Einstein equation, the one that causes
Reissner-Nordstr om black holes to be repulsive in their interiors, and causes mass ination in non-empty
(non Reissner-Nordstr om) charged black holes?
Whats important?
1. The tetrad formalism provides a rm mathematical foundation for the concept that space falls faster than
light inside a black hole.
2. Whereas the Kerr-Newman geometry of an ideal rotating black hole contains inside its horizon wormhole
and white hole connections to other universes, real black holes are subject to the mass ination stability
discovered by Eric Poisson & Werner Israel (1990, Internal structure of black holes, Phys. Rev. D 41,
1796-1809).
15
Black hole waterfalls
15.1 Tetrads move through coordinates
As already discussed in 11.3, the way in which metrics are commonly written, as a (weighted) sum of squares
of dierentials,
ds
2
=
mn
e
m
e
n
dx
dx
, (15.1)
encodes not only a metric g
=
mn
e
m
e
n
, but also an inverse vierbein e

m
, and consequently a vierbein

e
m
, and associated tetrad

m
. Most commonly the tetrad metric is orthonormal (Minkowski),
mn
=
mn
,
but other tetrad metrics, such as Newman-Penrose, occur. Usually it is self-evident from the form of the
line-element what the tetrad metric
mn
is in any particular case.
If the tetrad is orthonormal,
mn
=
mn
, then the 4-velocity u
m
of an object at rest in the tetrad, or
equivalently the 4-velocity of the tetrad rest frame itself, is
u
m
= 1, 0, 0, 0 . (15.2)
The tetrad-frame 4-velocity (15.2) of the tetrad rest frame is transformed to a coordinate-frame 4-velocity
u
in the usual way, by applying the vierbein,

dx
d
u
= e
m
u
m
= e
0
. (15.3)
Equation (15.3) says that the tetrad rest frame moves through the coordinates at coordinate 4-velocity given
by the zeroth row of the vierbein, dx
/d = e
0
. The coordinate 4-velocity u
is related to the lapse and

shift
in the ADM formalism by u
= 1,
/, equation (13.10).
The idea that locally inertial frames move through the coordinates provides the simplest way to conceptu-
alize black holes. The motion of locally inertial frames through coordinates is what is meant by the dragging
of inertial frames around rotating masses.
252 Black hole waterfalls
Figure 15.1 The sh upstream can make way against the current, but the sh downstream is swept to the
bottom of the waterfall. Art by Wildrose Hamilton.
15.2 Gullstrand-Painleve waterfall
The Gullstrand-Painleve metric is a version of the metric for a spherical (Schwarzschild or Reissner-Nordstr om)
black hole discovered in 1921 independently by Allvar Gullstrand (1922, Allgemeine L osung des statischen
Einkorperproblems in der Einsteinschen Gravitationstheorie, translated by http://babelfish.altavista.
com/tr as General solution of the static body problem in Einsteins Gravitation Theory, Arkiv. Mat. As-
tron. Fys. 16(8), 115) and Paul Painleve (1921, La mecanique classique et la theorie de la relativite, C.
R. Acad. Sci. (Paris) 173, 677680). Although Gullstrands paper was published in 1922, after Painleves, it
appears that Gullstrands work has priority. Gullstrands paper was dated 25 May 1921, whereas Painleves
is a write up of a presentation to the Academie des Sciences in Paris on 24 October 1921. Moreover, Gull-
strand seems to have had a better grasp of what he had discovered than Painleve, for Gullstrand recognized
that observables such as the redshift of light from the Sun are unaected by the choice of coordinates in
the Schwarzschild geometry, whereas Painleve, noting that the spatial metric was at at constant free-fall
time, dt
= 0, concluded in his nal sentence that, as regards the redshift of light and such, cest pure
imagination de pretendre tirer du ds
2
des consequences de cette nature.
Although neither Gullstrand nor Painleve understood it, their metric paints a picture of space falling like
a river, or waterfall, into a spherical black hole, Figure 15.1. The river has two key features: rst, the river
ows in Galilean fashion through a at Galilean background, equation (15.19); and second, as a freely-falling
shy swims through the river, its 4-velocity, or more generally any 4-vector attached to it, evolves by a
series of innitesimal Lorentz boosts induced by the change in the velocity of the river from place to place,
equation (15.24). Because the river moves in Galilean fashion, it can, and inside the horizon does, move
faster than light through the background coordinates. However, objects moving in the river move according
to the rules of special relativity, and so cannot move faster than light through the river.
15.2.1 Gullstrand-Painleve tetrad
The Gullstrand-Painleve metric (7.26) is
ds
2
= dt
2
+ (dr dt
)
2
+r
2
(d
2
+ sin
2
d
2
) , (15.4)
where is dened to be the radial velocity of a person who free-falls radially from rest at innity,
=
dr
d
=
dr
dt
, (15.5)
and t
is the free-fall time, the proper time experienced by a person who free-falls from rest at innity. The
radial velocity is the (apparently) Newtonian escape velocity
=
_
2M(r)
r
, (15.6)
where M(r) is the interior mass within radius r, and the sign is (infalling) for a black hole, + (outfalling)
for a white hole. For the Schwarzschild or Reissner-Nordstr om geometry the interior mass M(r) is the mass
M at innity minus the mass Q
2
/2r in the electric eld outside r,
M(r) = M
Q
2
2r
. (15.7)
Figure 15.2 illustrates the velocity elds in Schwarzschild and Reissner-Nordstr om black holes. Horizons
occur where the radial velocity equals the speed of light
= 1 , (15.8)
with for black hole solutions, + for white hole solutions. The phenomenology of Schwarzschild and
Reissner-Nordstr om black holes has already been explored in Chapters 7 and 8.
Exercise 15.1 Schwarzschild to Gullstrand-Painleve. Show that the Schwarzschild metric transforms
into the Gullstrand-Painleve metric under the coordinate transformation of the time coordinate
dt
= dt

1
2
dr . (15.9)
Exercise 15.2 Radial free-fall from rest. Conrm that given by equation (15.6) is indeed the
velocity (15.5) of a person who free-falls radially from rest at innity in the Reissner-Nordstr om geometry.
Horizon
I
n
n
e
r horiz
o
n
O
u
ter horiz
o
n
T
u
r
n
aro
u
n
d
Figure 15.2 Radial velocity in (upper panel) a Schwarzschild black hole, and (lower panel) a Reissner-
Nordstrom black hole with electric charge Q = 0.96.
The Gullstrand-Painleve line-element (15.4) encodes an inverse vierbein with an orthonormal tetrad metric
mn
=
mn
through
e
0
dx
= dt
, (15.10a)
e
1
dx
= dr dt
, (15.10b)
e
2
dx
= r d , (15.10c)
e
3
dx
= r sin d . (15.10d)
Explicitly, the inverse vierbein e
m
of the Gullstrand-Painleve line-element (15.4), and the corresponding

vierbein e
m
, are the matrices

e
m
=
_
_
_
_
1 0 0 0
1 0 0
0 0 r 0
0 0 0 r sin
_
_
_
_
, e
m
=
_
_
_
_
1 0 0
0 1 0 0
0 0 1/r 0
0 0 0 1/(r sin )
_
_
_
_
. (15.11)
According to equation (15.3), the coordinate 4-velocity of the tetrad frame through the coordinates is
_
dt
d
,
dr
d
,
d
d
,
d
d
_
= u
= e
0
= 1, , 0, 0 , (15.12)
consistent with the claim (15.5) that represents a radial velocity, while t
coincides with the proper time

in the tetrad frame.
The tetrad and coordinate axes
m
and g
are related to each other by the vierbein and inverse vierbein

in the usual way,
m
= e
m
and g
= e
m
m
. The Gullstrand-Painleve orthonormal tetrad axes
m
are
thus related to the coordinate axes g
by
0
= g
t
ff
+g
r
,
1
= g
r
,
2
= g
/r ,
3
= g
/(r sin ) . (15.13)

Physically, the Gullstrand-Painleve-Cartesian tetrad (15.13) are the axes of locally inertial orthonormal
frames (with spatial axes
i
oriented in the polar directions r, , ) attached to observers who free-fall
radially, without rotating, starting from zero velocity and zero angular momentum at innity. The fact
that the tetrad axes
m
are parallel-transported, without precessing, along the worldlines of the radially
free-falling observers can be conrmed by checking that the tetrad connections
nm0
with nal index 0 all
vanish, which implies that
d
m
d
=
0
m

n
m0
n
= 0 . (15.14)
That the proper time derivative d/d in equation (15.14) of a person at rest in the tetrad frame, with
4-velocity (15.2), is equal to the directed time derivative
0
follows from
d
d
= u
= u
m
m
=
0
. (15.15)
15.2.2 Gullstrand-Painleve-Cartesian tetrad
The manner in which the Gullstrand-Painleve line-element depicts a ow of space into a black hole is eluci-
dated further if the line-element is written in Cartesian rather than spherical polar coordinates. Introduce a
Cartesian coordinate system x
, x
i
t
, x, y, z. The Gullstrand-Painleve metric in these Cartesian

coordinates is
ds
2
= dt
2
+
ij
(dx
i
i
dt
)(dx
j
j
dt
) , (15.16)
with implicit summation over spatial indices i, j = 1, 2, 3. The
i
in the metric (15.16) are the components
of the radial velocity expressed in Cartesian coordinates
i
=
_
x
r
,
y
r
,
z
r
_
. (15.17)
The inverse vierbein e
m
and vierbein e
m
encoded in the Gullstrand-Painleve-Cartesian line-element (15.16)

are
e
m
=
_
_
_
_
1 0 0 0
1
1 0 0
2
0 1 0
3
0 0 1
_
_
_
_
, e
m
=
_
_
_
_
1
1
3
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
. (15.18)
The tetrad axes
m
of the Gullstrand-Painleve-Cartesian line-element (15.16) are related to the coordinate
tangent axes g
by
0
= g
t
ff
+
i
g
i
,
i
= g
i
, (15.19)
and conversely the coordinate tangent axes g
are related to the tetrad axes

m
by
g
t
ff
=
0
i
, g
i
=
i
. (15.20)
Note that the tetrad-frame contravariant components
i
of the radial velocity coincide with the coordinate-
frame contravariant components
i
; for clarication of this point see the more general equation (15.48)
for a rotating black hole. The Gullstrand-Painleve-Cartesian tetrad axes (15.19) are the same as the tetrad
axes (15.13), but rotated to point in Cartesian directions x, y, z rather than in polar directions r, , . Like the
polar tetrad, the Cartesian tetrad axes
m
are parallel-transported, without precessing, along the worldlines
of radially free-falling observers, as can be conrmed by checking once again that the tetrad connections
nm0
with nal index 0 all vanish.
Remarkably, the transformation (15.19) from coordinate to tetrad axes is just a Galilean transformation
of space and time, which shifts the time axis by velocity along the direction of motion, but which leaves
unchanged both the time component of the time axis and all the spatial axes. In other words, the black
hole behaves as if it were a river of space that ows radially inward through Galilean space and time at the
Newtonian escape velocity.
15.2.3 Gullstrand-Painleve shies
The Gullstrand-Painleve line element paints a picture of locally inertial frames falling like a river of space into
a spherical black hole. What happens to shies swimming in that river? Of course general relativity supplies
a mathematical answer in the form of the geodesic equation of motion (15.21). Does that mathematical
answer lead to further conceptual insight?
Consider a shy swimming in the Gullstrand-Painleve river, with some arbitrary tetrad-frame 4-velocity
u
m
, and consider a tetrad-frame 4-vector p
k
attached to the shy. If the shy is in free-fall, then the geodesic
equation of motion for p
k
is as usual
dp
k
d
+
k
mn
u
n
p
m
= 0 . (15.21)
As remarked in 11.13, for a constant (for example Minkowski) tetrad metric, as here, the tetrad connections
k
mn
constitute a set of four generators of Lorentz transformations, one in each of the directions n. In
particular
k
mn
u
n
is the generator of a Lorentz transformation along the path of a shy moving with 4-
velocity u
n
. In a small (innitesimal) time , the shy moves a proper distance
n
u
n
relative to the
infalling river. This proper distance
n
= e
n
=
n
(x
) = x
n

n
equals the distance
x
n
moved relative to the background Gullstrand-Painleve-Cartesian coordinates, minus the distance
n
moved by the river. The geodesic equation (15.21) says that the change p
k
in the tetrad 4-vector p
k
in the
time is
p
k
=
k
mn
n
p
m
. (15.22)
Equation (15.22) describes an innitesimal Lorentz transformation
k
mn
n
of the 4-vector p
k
.
Equation (15.22) is quite general in general relativity: it says that as a 4-vector p
k
free-falls through a
system of locally inertial tetrads, it nds itself Lorentz-transformed relative those tetrads. What is special
about the Gullstrand-Painleve-Cartesian tetrad is that the tetrad-frame connections, computed by the usual
formula (11.41), are given by the coordinate gradient of the radial velocity (the following equation is valid
component-by-component despite the non-matching up-down placement of indices)
0
ij
=
i
0j
=

i
x
j
(i, j = 1, 2, 3) . (15.23)
The same property, that the tetrad connections are a pure coordinate gradient, holds also for the Doran-
Cartesian tetrad for a rotating black hole, equation (15.51). With the connections (15.23), the change
p
k
(15.22) in the tetrad 4-vector is
p
0
=
i
p
i
, p
i
=
i
p
0
, (15.24)
where
i
is the change in the velocity of the river as seen in the tetrad frame,
i
=
j

i
x
j
. (15.25)
But equation (15.24) is nothing more than an innitesimal Lorentz boost by a velocity change
i
. This
shows that a shy swimming in the river follows the rules of special relativity, being Lorentz boosted by tidal
changes
i
in the river velocity from place to place.
Is it correct to interpret equation (15.25) as giving the change
i
in the river velocity seen by a shy? Of
course general relativity demands that equation (15.25) be mathematically correct; the issue is merely one
of interpretation. Shouldnt the change in the river velocity really be
i
?
= x

i
x
, (15.26)
where x
is the full change in the coordinate position of the shy? No. Part of the change (15.26) in the
river velocity can be attributed to the change in the velocity of the river itself over the time , which is
x
river
i
/x
with x
river
=
. The change in the velocity relative to the owing river is
i
= (x
river
)
i
x
= (x
i
x
(15.27)
which reproduces the earlier expression (15.25). Indeed, in the picture of shies being carried by the river,
it is essential to subtract the change in velocity of the river itself, as in equation (15.27), because otherwise
shies at rest in the river (going with the ow) would not continue to remain at rest in the river.
15.3 Boyer-Lindquist tetrad
The Boyer-Lindquist metric for an ideal rotating black hole was explored already in Chapter 9. With the
tetrad formalism in hand, the advantages of the Boyer-Lindquist tetrad for portraying the Kerr-Newman
geometry become manifest. With respect to the orthonormal Boyer-Lindquist tetrad, the electromagnetic
eld is purely radial, and the energy-momentum and Weyl tensors are diagonal. The Boyer-Lindquist tetrad
is aligned with the principal (ingoing or outgoing) null congruences.
The Boyer-Lindquist orthonormal tetrad is encoded in the Boyer-Lindquist metric
ds
2
=

2
_
dt a sin
2
d
_
2
+

2
dr
2
+
2
d
2
+
R
4
sin
2
2
_
d
a
R
2
dt
_
2
(15.28)
where
R
_
r
2
+a
2
,
_
r
2
+a
2
cos
2
, R
2
2Mr +Q
2
= R
2
(1
2
) . (15.29)
Explicitly, the vierbein e
m
of the Boyer-Lindquist orthonormal tetrad is

e
m
=
1
_
_
_
_
R
2
/
0 0 a/
0 0
0 0 1 0
a sin 0 0 1/ sin
_
_
_
_
, (15.30)
with inverse vierbein e
m
e
m
=
_
_
_
_
/ 0 0 a sin
2
/
0 /
0 0
0 0 0
a sin/ 0 0 R
2
sin /
_
_
_
_
. (15.31)
With respect to the Boyer-Lindquist tetrad, only the time component A
t
of the electromagnetic potential
A
m
is non-vanishing,
A
m
=
_
Qr
, 0, 0, 0
_
. (15.32)
Only the radial components E
r
and B
r
of the electric and magnetic elds are non-vanishing, and they are
given by the complex combination
E
r
+i B
r
=
Q
(r ia cos )
2
, (15.33)
or explicitly
E
r
=
Q
_
r
2
a
2
cos
2
4
, B
r
=
2Qar cos
4
. (15.34)
The electrogmagnetic eld (15.33) satises Maxwells equations (11.64) and (11.65) with zero electric charge
and current, j
n
= 0, except at the singularity = 0.
The non-vanishing components of the tetrad-frame Einstein tensor G
mn
are
G
mn
=
Q
2
4
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
, (15.35)
which is the energy-momentum tensor of the electromagnetic eld. The non-vanishing components of the
tetrad-frame Weyl tensor C
klmn
are
1
2
C
trtr
=
1
2
C
= C
tt
= C
tt
= C
rr
= C
rr
= Re C , (15.36a)
1
2
C
tr
= C
tr
= C
tr
= ImC , (15.36b)
where C is the complex Weyl scalar
C =
1
(r ia cos )
3
_
M
Q
2
r +ia cos
_
. (15.37)
In the Boyer-Lindquist tetrad, the photon 4-velocity v
m
on the principal null congruences is radial,
v
t
=

, v
r
=

, v
= 0 , v
= 0 . (15.38)
Exercise 15.3 Dragging of inertial frames around a Kerr-Newman black hole. What is the
coordinate-frame 4-velocity u
of the Boyer-Lindquist tetrad through the Boyer-Lindquist coordinates?

15.4 Doran waterfall
The picture of space falling into a black hole like a river or waterfall works also for rotating black holes. For
Kerr-Newman rotating black holes, the counterpart of the Gullstrand-Painleve metric is the Doran (2000)
metric.
The space river that falls into a rotating black hole has a twist. One might have expected that the rotation
of the black hole would be manifested by a velocity that spirals inward, but this is not the case. Instead,
the river is characterized not merely by a velocity but also by a twist. The velocity and the twist together
comprise a 6-dimensional river bivector
km
, equation (15.52) below, whose electric part is the velocity,
and whose magnetic part is the twist. Recall that the 6-dimensional group of Lorentz transformations is
generated by a combination of 3-dimensional Lorentz boosts and 3-dimensional spatial rotations. A shy
that swims through the river is Lorentz boosted by tidal changes in the velocity, and rotated by tidal changes
in the twist, equation (15.61).
Thanks to the twist, unlike the Gullstrand-Painleve metric, the Doran metric is not spatially at at
constant free-fall time t
. Rather, the spatial metric is sheared in the azimuthal direction. Just as the
velocity produces a Lorentz boost that makes the metric non-at with respect to the time components, so
also the twist produces a rotation that makes the metric non-at with respect to the spatial components.
15.4.1 Doran-Cartesian coordinates
In place of the polar coordinates r, ,
of the Doran metric, introduce corresponding Doran-Cartesian

coordinates x, y, z with z taken along the rotation axis of the black hole (the black hole rotates right-
handedly about z, for positive spin parameter a)
x Rsin cos
, y Rsin sin
, z r cos . (15.39)
The metric in Doran-Cartesian coordinates x
, x
i
t
, x, y, z, is
ds
2
= dt
2
+
ij
_
dx
i
dx
_ _
dx
j
dx
_
(15.40)
where
is the rotational velocity vector
=
_
1,
ay
R
2
,
ax
R
2
, 0
_
, (15.41)
and
is the velocity vector
=
R
_
0,
xr
R
,
yr
R
,
zR
r
_
. (15.42)
The rotational velocity and radial velocity vectors are orthogonal
= 0 . (15.43)
For the Kerr-Newman metric, the radial velocity is
=
_
2Mr Q
2
R
(15.44)
with for black hole (infalling), + for white hole (outfalling) solutions. Horizons occur where
= 1 , (15.45)
with = 1 for black hole horizons, and = 1 for white hole horizons. Note that the squared magnitude
of the velocity vector is not

2
, but rather diers from
2
by a factor of R
2
/
2
:
=
m
m
=

2
R
2
2
. (15.46)
The point of the convention adopted here is that (r) is any and only a function of r, rather than depending
also on through . Moreover, with the convention here, is 1 at horizons, equation (15.45). Finally, the
4-velocity
is simply related to by
= (/r) r/x
.
The Doran-Cartesian metric (15.40) encodes a vierbein e
m
and inverse vierbein e

m
e
m
m
+
m
, e
m
=
m
m
. (15.47)
Here the tetrad-frame components
m
of the rotational velocity vector and
m
of the radial velocity vector
are
m
= e
m
,
m
= e
m
=
m
, (15.48)
which works thanks to the orthogonality (15.43) of
and
. Equation (15.48) says that the covariant

tetrad-frame components of the rotational velocity vector are the same as its covariant coordinate-frame
components in the Doran-Cartesian coordinate system,
m
=
, and likewise the contravariant tetrad-frame

components of the radial velocity vector are the same as its contravariant coordinate-frame components,
m
=
.
15.4.2 Doran-Cartesian tetrad
Like the Gullstrand-Painleve tetrad, the Doran-Cartesian tetrad
m

0
,
1
,
2
,
3
is aligned with the
Cartesian rest frame g
g
t
ff
, g
x
, g
y
, g
z
at innity, and is parallel-transported, without precessing, by
observers who free-fall from zero velocity and zero angular momentum at innity, as can be conrmed by
checking that the tetrad connections with nal index 0 all vanish,
nm0
= 0, equation (15.14).
Let | and subscripts denote horizontal radial and azimuthal directions respectively, so that
cos

1
+ sin

2
,
sin

1
+ cos

2
,
g
cos
g
x
+ sin
g
y
, g
sin
g
x
+ cos
g
y
.
(15.49)
Then the relation between Doran-Cartesian tetrad axes
m
and the tangent axes g
of the Doran-Cartesian
metric (15.40) is
0
= g
t
ff
+
i
g
i
, (15.50a)
= g
, (15.50b)
= g
a sin
R

i
g
i
, (15.50c)
3
= g
z
. (15.50d)
The relations (15.50) resemble those (15.19) of the Gullstrand-Painleve tetrad, except that the azimuthal
tetrad axis
is shifted radially relative to the azimuthal tangent axis g
. This shift reects the fact that,

unlike the Gullstrand-Painleve metric, the Doran metric is not spatially at at constant free-fall time, but
rather is sheared azimuthally.
15.4.3 Doran shies
The tetrad-frame connections equal the ordinary coordinate partial derivatives in Doran-Cartesian coordi-
nates of a bivector (antisymmetric tensor)
km
kmn
=
km
x
n
, (15.51)
which I call the river eld because it encapsulates all the properties of the infalling river of space. The
bivector river eld
km
is
km
=
k
0kmi
i
, (15.52)
where
m
=
mn
m
, the totally antisymmetric tensor
klmn
is normalized so that
0123
= 1, and the vector
i
points vertically upward along the rotation axis of the black hole
i
0, 0, 0, , a
_
r
dr
R
2
. (15.53)
The electric part of
km
, where one of the indices is time 0, constitutes the velocity vector
i
0i
=
i
(15.54)
while the magnetic part of
km
, where both indices are spatial, constitutes the twist vector
i
dened by
1
2

0ikm
km
=
0ikm
m
+
i
. (15.55)
The sense of the twist is that induces a right-handed rotation about an axis equal to the direction of
i
by
an angle equal to the magnitude of
i
. In 3-vector notation, with
i
,
i
,
i
,
i
,
+ . (15.56)
In terms of the velocity and twist vectors, the river eld
km
is
km
=
_
_
_
_
0
x
x
0
z
z
0
x
x
0
_
_
_
_
. (15.57)
Note that the sign of the magnetic part of
km
is opposite to the sign of the analogous magnetic eld
B associated with an electromagnetic eld F
km
; but the adopted signs are natural in that the river eld
R
o
t
a
t
i
o
n
a
x
i
s
In
ner horizon
O
uter horizon
3
6
0
R
o
t
a
t
i
o
n
a
x
i
s
In
ner horizon
O
uter horizon
Figure 15.3 (Upper panel) velocity
i
and (lower panel) twist
i
vector elds for a Kerr black hole with spin
parameter a = 0.96. Both vectors lie, as shown, in the plane of constant free-fall azimuthal angle
. The
vertical bar in the lower panel shows the length of a twist vector corresponding to a full rotation of 360
.
induces boosts in the direction of the velocity
i
, and right-handed rotations about the twist
i
. Like a
static electric eld, the velocity vector
i
is the gradient of a potential
i
=

x
i
_
r
dr , (15.58)
but unlike a magnetic eld the twist vector
i
is not pure curl: rather, it is
i
+
i
that is pure curl.
Figure 15.3 illustrates the velocity and twist elds in a Kerr black hole.
With the tetrad connection coecients given by equation (15.51), the equation of motion (15.21) for a
4-vector p
k
attached to a shy following a geodesic in the Doran river translates to
dp
k
d
=

k
m
x
n
u
n
p
m
. (15.59)
In a proper time , the shy moves a proper distance
m
u
m
relative to the background Doran-
Cartesian coordinates. As a result, the shy sees a tidal change
k
m
in the river eld
k
m
=
n
k
m
x
n
. (15.60)
Consequently the 4-vector p
k
is changed by
p
k
p
k
+
k
m
p
m
. (15.61)
But equation (15.61) corresponds to an innitesimal Lorentz transformation by
k
m
, equivalent to a Lorentz
boost by
i
and a rotation by
i
.
As discussed previously with regard to the Gullstrand-Painleve river, 15.2.3, the tidal change
k
m
,
equation (15.60), in the river eld seen by a shy is not the full change x
k
m
/x
relative to the
background coordinates, but rather the change relative to the river
k
m
= (x
river
)
k
m
x
=
_
x
(t
a sin
2

k
m
x
, (15.62)
with the change in the velocity and twist of the river itself subtracted o.
That there exists a tetrad (the Doran-Cartesian tetrad) where the tetrad-frame connections are a coor-
dinate gradient of a bivector, equation (15.51), is a peculiar feature of ideal black holes. It is an intriguing
thought that perhaps the 6 physical degrees of freedom of a general spacetime might always be encoded in
the 6 degrees of freedom of a bivector, but I suspect that that is not true.
Exercise 15.4 River model of the Friedmann-Robertson-Walker metric. Show that the at FRW
line-element
ds
2
= dt
2
+a
2
(dx
2
+x
2
do
2
) (15.63)
can be re-expressed as
ds
2
= dt
2
+ (dr Hr dt)
2
+r
2
do
2
, (15.64)
where r ax is the proper radial distance, and H a/a is the Hubble parameter. Interpret the line-
element (15.64). What is the generalization to a non-at FRW universe?
Exercise 15.5 Program geodesics in a rotating black hole. Write a graphics (Java?) program
that uses the prescription (15.60) to draw geodesics of test particles in an ideal (Kerr-Newman) black
hole, expressed in Doran-Cartesian coordinates. Attach 3D bodies to your test particles, and use the same
prescription (15.60) to rotate the bodies. Implement an option to translate to Boyer-Lindquist coordinates.
16
General spherically symmetric spacetime
16.1 Spherical spacetime
The most important equations in this chapter are the two Einstein equations (16.52). Spherical spacetimes
have 2 physical degrees of freedom. Spherical symmetry eliminates any angular degrees of freedom, leaving
4 adjustable metric coecients g
tt
, g
tr
, g
rr
, and g
. But coordinate transformations of the time t and radial

r coordinates remove 2 degrees of freedom, leaving a spherical spacetime with a net 2 physical degrees of
freedom. Spherical spacetimes have 4 distinct Einstein equations (16.30). But 2 of the Einstein equations
serve to enforce energy-momentum conservation, so the evolution of the spacetime is governed by 2 Einstein
equations, in agreement with the number of physical degrees of freedom of spherical spacetime.
The 2 degrees of freedom mean that spherical spacetimes in general relativity have a richer structure than
in Newtonian gravity, which has only degree of freedom, the Newtonian potential . The richer structure
is most striking in the case of the mass ination instability, Chapter 17, which is an intrinsically general
relativistic instability, with no Newtonian analogue.
16.1.1 Spherical line-element
The spherical line-element adopted in this chapter is, in spherical polar coordinates x
t, r, , ,
ds
2
=
dt
2
2
+
1
2
r
_
dr
t
dt
_
2
+r
2
do
2
. (16.1)
Here r is the circumferential radius, dened such that the circumference around any great circle is 2r. The
line-element (16.1) is somewhat unconventional in that it is not diagonal: g
tr
does not vanish. There are
two good reasons to consider a non-diagonal metric. Firstly, as discussed in 16.1.11, Einsteins equations
take a more insightful form when expressed in a non-diagonal frame where
t
does not vanish. Secondly, if a
horizon is present, as in the case of black holes, and if the radial coordinate is taken to be the circumferential
radius r, then a diagonal metric will have a coordinate singularity at the horizon, which is not ideal.
266 General spherically symmetric spacetime
The vierbein e
m

m
corresponding to the spherical line-element (16.1) are

e
m
=
_
_
_
_

t
0 0
0
r
0 0
0 0 1/r 0
0 0 0 1/(r sin )
_
_
_
_
, e
m
=
_
_
_
_
1/ 0 0 0
t
/(
r
) 1/
r
0 0
0 0 r 0
0 0 0 r sin
_
_
_
_
. (16.2)
As in the ADM formalism, Chapter 13, the tetrad time axis
t
is chosen to be orthogonal to hypersurfaces of
constant time t. However, the convention here for the vierbein coecients diers from the ADM convention:
here 1/ is the ADM lapse, while
t
/ is the ADM shift. The directed derivatives
t
and
r
along the time
and radial tetrad axes
t
and
r
are
t
= e
t
=

t
+
t
r
,
r
= e
r
=
r
r
. (16.3)
The tetrad-frame 4-velocity u
m
of a person at rest in the tetrad frame is by denition u
m
= 1, 0, 0, 0. It
follows that the coordinate 4-velocity u
of such a person is
u
= e
m
u
m
= e
t
= ,
t
, 0, 0 . (16.4)
A person instantaneously at rest in the tetrad frame satises dr/dt =
t
/ according to equation (16.4),
so it follows from the line-element (16.1) that the proper time of a person at rest in the tetrad frame is
related to the coordinate time t by
d =
dt
in tetrad rest frame . (16.5)

The directed time derivative
t
is just the proper time derivative along the worldline of a person continuously
at rest in the tetrad frame (and who is therefore not in free-fall, but accelerating with the tetrad frame),
which follows from
d
d
=
dx
= u
= u
m
m
=
t
. (16.6)
By contrast, the proper time derivative measured by a person who is instantaneously at rest in the tetrad
frame, but is in free-fall, is the covariant time derivative
D
D
=
dx
d
D
= u
= u
m
D
m
= D
t
. (16.7)
Since the coordinate radius r has been dened to be the circumferential radius, a gauge-invariant denition,
it follows that the tetrad-frame gradient
m
of the coordinate radius r is a tetrad-frame 4-vector (a coordinate
gauge-invariant object)
m
r = e
m
r
x
= e
m
r
=
m
=
t
,
r
, 0, 0 is a tetrad 4-vector . (16.8)
This accounts for the notation
t
and
r
introduced above. Since
m
is a tetrad 4-vector, its scalar product
with itself must be a scalar. This scalar denes the interior mass M(t, r), also called the Misner-Sharp
mass, by
1
2M
r

m
m
=
2
t
+
2
r
is a coordinate and tetrad scalar . (16.9)
The interpretation of M as the interior mass will become evident below, 16.1.8.
16.1.2 Rest diagonal line-element
Although this is not the choice adopted here, the line-element (16.1) can always be brought to diagonal form
by a coordinate transformation t t
(subscripted for diagonal) of the time coordinate. The tr part of

the metric is
g
tt
dt
2
+ 2 g
tr
dt dr +g
rr
dr
2
=
1
g
tt
_
(g
tt
dt +g
tr
dr)
2
+ (g
tt
g
rr
g
2
tr
) dr
2
. (16.10)
This can be diagonalized by choosing the time coordinate t
such that
f dt
= g
tt
dt +g
tr
dr (16.11)
for some integrating factor f(t, r). Equation (16.11) can be solved by choosing t
to be constant along
integral curves
dr
dt
=
g
tt
g
tr
. (16.12)
The resulting diagonal rest line-element is
ds
2
=
dt
2
+
dr
2
1 2M/r
+r
2
do
2
. (16.13)
The line-element (16.13) corresponds physically to the case where the tetrad frame is taken to be at rest in
the spatial coordinates,
t
= 0, as can be seen by comparing it to the earlier line-element (16.1). In changing
the tetrad frame from one moving at dr/dt =
t
/ to one that is at rest (at constant circumferential radius
r), a tetrad transformation has in eect been done at the same time as the coordinate transformation (16.11),
the tetrad transformation being precisely that needed to make the line-element (16.13) diagonal. The metric
coecient g
rr
in the line-element (16.13) follows from the fact that
2
r
= 1 2M/r when
t
= 0, equa-
tion (16.9). The transformed time coordinate t
is unspecied up to a transformation t
f(t
). If the
spacetime is asymptotically at at innity, then a natural way to x the transformation is to choose t
to
be the proper time at rest at innity.
16.1.3 Comoving diagonal line-element
Although once again this is not the path followed here, the line-element (16.1) can also be brought to diagonal
form by a coordinate transformation r r
, where, analogously to equation (16.11), r
is chosen to satisfy
f dr
= g
tr
dt +g
rr
dr
1
r
_
dr
t
dt
_
(16.14)
for some integrating factor f(t, r). The new coordinate r
is constant along the worldline of an object at

rest in the tetrad frame, with dr/dt =
t
/, equation (16.4), so r
can be regarded as a comoving radial

coordinate. The comoving radial coordinate r
could for example be chosen to equal the circumferential

radius r at some xed instant of coordinate time t (say t = 0). The diagonal comoving line-element in
this comoving coordinate system takes the form
ds
2
=
dt
2
2
+
dr
2
2
+r
2
do
2
, (16.15)
where the circumferential radius r(t, r
) is considered to be an implicit function of time t and the comoving

radial coordinate r
. Whereas in the rest line-element (16.13) the tetrad was changed from one that was
moving at dr/dt =
t
/ to one that was at rest, here the transformation keeps the tetrad unchanged. In
both the rest and comoving diagonal line-elements (16.13) and (16.15) the tetrad is at rest relative to the
respective radial coordinate r or r
; but whereas in the rest line-element (16.13) the radial coordinate was
xed to be the circumferential radius r, in the comoving line-element (16.15) the comoving radial coordinate
r
is a label that follows the tetrad. Because the tetrad is unchanged by the transformation to the comoving
radial coordinate r
, the directed time and radial derivatives are unchanged:
t
=

t
=

t
r
+
t
t
,
r
=

r
t
=
r
t
. (16.16)
16.1.4 Tetrad connections
Now turn the handle to proceed towards the Einstein equations. The tetrad connections coecients
kmn
corresponding to the spherical line-element (16.1) are
rtt
= h
t
, (16.17a)
rtr
= h
r
, (16.17b)
t
=
t
=

t
r
, (16.17c)
r
=
r
=

r
r
, (16.17d)
=
cot
r
, (16.17e)
where h
t
is the proper radial acceleration (minus the gravitational force) experienced by a person at rest in
the tetrad frame
h
t

r
ln , (16.18)
and h
r
is the Hubble parameter of the radial ow, as measured in the tetrad rest frame, dened by
h
r

t
ln
r
+

t
r

t
ln
r
. (16.19)
The interpretation of h
t
as a proper acceleration and h
r
as a radial Hubble parameter goes as follows. The
tetrad-frame 4-velocity u
m
of a person at rest in the tetrad frame is by denition u
m
= 1, 0, 0, 0. If the
person at rest were in free fall, then the proper acceleration would be zero, but because this is a general
spherical spacetime, the tetrad frame is not necessarily in free fall. The proper acceleration experienced by
a person continuously at rest in the tetrad frame is the proper time derivative Du
m
/D of the 4-velocity,
which is
Du
m
D
= D
t
u
m
=
t
u
m
+
m
tt
u
t
=
m
tt
= 0,
r
tt
, 0, 0 = 0, h
t
, 0, 0 , (16.20)
the rst step of which follows from equation (16.7). Similarly, a person at rest in the tetrad frame will
measure the 4-velocity of an adjacent person at rest in the tetrad frame a small proper radial distance
r
away to dier by
r
D
r
u
m
. The Hubble parameter of the radial ow is thus the covariant radial derivative
D
r
u
m
, which is
D
r
u
m
=
r
u
m
+
m
tr
u
t
=
m
tr
= 0,
r
tr
, 0, 0 = 0, h
r
, 0, 0 . (16.21)
Conned to the tr-plane (that is, considering only Lorentz transformations in the tr-plane, which is to
say radial Lorentz boosts), the acceleration h
t
and Hubble parameter h
r
constitute the components of a
tetrad-frame 2-vector h
n
= h
t
, h
r
:
h
n
=
rtn
. (16.22)
The Riemann tensor, equations (16.24) below, involves covariant derivatives D
m
h
n
of h
n
, which coincide
with the covariant derivatives D
(2)
h
n
conned to tr-plane.
Since h
r
is a kind of radial Hubble parameter, it can be useful to dene a corresponding radial scale factor
by
h
r

t
ln . (16.23)
The scale factor is the same as the in the comoving line-element of equation (16.15). This is true because
h
r
is a tetrad connection and therefore coordinate gauge-invariant, and the line-element (16.15) is related
to the line-element (16.1) being considered by a coordinate transformation r r
that leaves the tetrad

unchanged.
16.1.5 Riemann, Einstein, and Weyl tensors
The non-vanishing components of the tetrad-frame Riemann tensor R
klmn
corresponding to the spherical
line-element (16.1) are
R
trtr
= D
r
h
t
D
t
h
r
, (16.24a)
R
tt
= R
tt
=
1
r
D
t
t
, (16.24b)
R
rr
= R
rr
=
1
r
D
r
r
, (16.24c)
R
tr
= R
tr
=
1
r
D
t
r
=
1
r
D
r
t
, (16.24d)
R
=
2M
r
3
, (16.24e)
where D
m
denotes the covariant derivative as usual.
The non-vanishing components of the tetrad-frame Einstein tensor G
km
are
G
tt
= 2 R
rr
+R
, (16.25a)
G
rr
= 2 R
tt
R
, (16.25b)
G
tr
= 2 R
tr
, (16.25c)
G
= G
= R
trtr
+R
tt
R
rr
, (16.25d)
whence
G
tt
=
2
r
_
D
r
r
+
M
r
2
_
, (16.26a)
G
rr
=
2
r
_
D
t
M
r
2
_
, (16.26b)
G
tr
=
2
r
D
t
r
=
2
r
D
r
t
, (16.26c)
G
= G
= D
r
h
t
D
t
h
r
+
1
r
(D
r
r
D
t
t
) . (16.26d)
The non-vanishing components of the tetrad-frame Weyl tensor C
klmn
are
1
2
C
trtr
= C
tt
= C
tt
= C
rr
= C
rr
=
1
2
C
= C , (16.27)
where C is the Weyl scalar (the spin-0 component of the Weyl tensor),
C
1
6
(R
trtr
R
tt
+R
rr
R
) =
1
6
_
G
tt
G
rr
+G
M
r
3
. (16.28)
16.1.6 Einstein equations
The tetrad-frame Einstein equations
G
km
= 8T
km
(16.29)
imply that
_
_
_
_
G
tt
G
tr
0 0
G
tr
G
rr
0 0
0 0 G
0
0 0 0 G
_
_
_
_
= 8T
km
= 8
_
_
_
_
f 0 0
f p 0 0
0 0 p
0
0 0 0 p
_
_
_
_
(16.30)
where T
tt
is the proper energy density, f T
tr
is the proper radial energy ux, p T
rr
is the proper
radial pressure, and p
= T
is the proper transverse pressure.

16.1.7 Choose your frame
So far the radial motion of the tetrad frame has been left unspecied. Any arbitrary choice can be made.
For example, the tetrad frame could be chosen to be at rest,
t
= 0 , (16.31)
as in the Schwarzschild or Reissner-Nordstr om line-elements. Alternatively, the tetrad frame could be chosen
to be in free-fall,
h
t
= 0 , (16.32)
as in the Gullstrand-Painleve line-element. For situations where the spacetime contains matter, one natural
choice is the center-of-mass frame, dened to be the frame in which the energy ux f is zero
G
tr
= 8f = 0 . (16.33)
Whatever the choice of radial tetrad frame, tetrad-frame quantities in dierent radial tetrad frames are
related to each other by a radial Lorentz boost.
16.1.8 Interior mass
Equations (16.26a) with the middle expression of (16.26c), and (16.26b) with the nal expression of (16.26c),
respectively, along with the denition (16.9) of the interior mass M, and the Einstein equations (16.30),
imply
p =
1
t
_
1
4r
2

t
M
r
f
_
, (16.34a)
=
1
r
_
1
4r
2

r
M
t
f
_
. (16.34b)
In the center-of-mass frame, f = 0, these equations reduce to
t
M = 4r
2
t
p , (16.35a)
r
M = 4r
2
r
. (16.35b)
Equations (16.35) amply justify the interpretation of M as the interior mass. The rst equation (16.35a)
can be written
t
M +p 4r
2
t
r = 0 , (16.36)
which can be recognized as an expression of the rst law of thermodynamics,
dE +p dV = 0 , (16.37)
with mass-energy E equal to M. The second equation (16.35b) can be written, since
r
=
r
/r, equa-
tion (16.3),
M
r
= 4r
2
, (16.38)
which looks exactly like the Newtonian relation between interior mass M and density . Actually, this
apparently Newtonian equation (16.38) is deceiving. The proper 3-volume element d
3
r in the center-of-mass
tetrad frame is given by
d
3
r
r

= g
r
dr g
d g
d =
r
2
sin dr d d
, (16.39)
so that the proper 3-volume element dV d
3
r of a radial shell of width dr is
dV =
4r
2
dr
r
. (16.40)
Thus the true mass-energy dM
m
associated with the proper density in a proper radial volume element
dV might be expected to be
dM
m
= dV =
4r
2
dr
r
, (16.41)
whereas equation (16.38) indicates that the actual mass-energy is
dM = 4r
2
dr =
r
dV . (16.42)
A person in the center-of-mass frame might perhaps, although there is really no formal justication for doing
so, interpret the balance of the mass-energy as gravitational mass-energy M
g
dM
g
= (
r
1) dV . (16.43)
Whatever the case, the moral of this is that you should beware of interpreting the interior mass M too
literally as palpable mass-energy.
16.1.9 Energy-momentum conservation
Covariant conservation of the Einstein tensor D
m
G
mn
= 0 implies conservation of energy-momentum
D
m
T
mn
= 0. The two non-vanishing equations represent conservation of energy and of radial momentum,
and are
D
m
T
mt
=
t
+
2
t
r
( +p
) +h
r
( +p) +
_
r
+
2
r
r
+ 2 h
t
_
f = 0 , (16.44a)
D
m
T
mr
=
r
p +
2
r
r
(p p
) +h
t
( +p) +
_
t
+
2
t
r
+ 2 h
r
_
f = 0 . (16.44b)
In the center-of-mass frame, f = 0, these energy-momentum conservation equations reduce to
t
+
2
t
r
( +p
) +h
r
( +p) = 0 , (16.45a)
r
p +
2
r
r
(p p
) +h
t
( +p) = 0 . (16.45b)
In a general situation where the mass-energy is the sum over several individual components a,
T
mn
=
species a
T
mn
a
, (16.46)
the individual mass-energy components a of the system each satisfy an energy-momentum conservation
equation of the form
D
m
T
mn
a
= F
n
a
, (16.47)
where F
n
a
is the ux of energy into component a. Einsteins equations enforce energy-momentum conservation
of the system as a whole, so the sum of the energy uxes must be zero
species a
F
n
a
= 0 . (16.48)
16.1.10 First law of thermodynamics
For an individual species a, the energy conservation equation (16.44a) in the center-of-mass frame of the
species, f
a
= 0, can be written
D
m
T
mt
a
=
t
a
+ (
a
+p
a
)
t
ln r
2
+ (
a
+p
a
)
t
ln
a
= F
t
a
, (16.49)
where
a
is the radial scale factor, equation (16.23), in the center-of-mass frame of the species (the scale
factor is dierent in dierent frames). Equation (16.49) can be recognized as an expression of the rst law
of thermodynamics for a volume element V of species a, in the form
V
1
_
t
(
a
V ) +p
a
V
r

t
V
+p
a
V
t
V
r
_
= F
t
a
, (16.50)
with transverse volume (area) V
r
2
, radial volume (width) V
r

a
, and total volume V V
V
r
. The
ux F
t
a
on the right hand side is the heat per unit volume per unit time going into species a. If the pressure
of species a is isotropic, p
a
= p
a
, then equation (16.50) simplies to
V
1
_
t
(
a
V ) +p
a
t
V
_
= F
t
a
, (16.51)
with volume V r
2
a
.
16.1.11 Structure of the Einstein equations
The spherically symmetric spacetime under consideration is described by 3 vierbein coecients, ,
t
, and
r
.
However, some combination of the 3 coecients represents a gauge freedom, since the spherically symmetric
spacetime has only two physical degrees of freedom. As commented in 16.1.7, various gauge-xing choices
can be made, such as choosing to work in the center-of-mass frame, f = 0.
Equations (16.26) give 4 equations for the 4 non-vanishing components of the Einstein tensor. The two
expressions for G
tr
are identical when expressed in terms of the vierbein and vierbein derivatives, so are not
distinct equations. Conservation of energy-momentum of the system as a whole is built in to the Einstein
equations, a consequence of the Bianchi identities, so 2 of the Einstein equations are eectively equivalent to
the energy-momentum conservation equations (16.44). In the general case where the matter contains multiple
components, it is usually a good idea to include the equations describing the conservation or exchange of
energy-momentum separately for each component, so that global conservation of energy-momentum is then
satised as a consequence of the matter equations.
This leaves 2 independent Einstein equations to describe the 2 physical degrees of freedom of the spacetime.
The 2 equations may be taken to be the evolution equations (16.26c) and (16.26b) for
t
and
r
,
D
t
t
=
M
r
2
4rp , (16.52a)
D
t
r
= 4rf , (16.52b)
which are valid for any choice of tetrad frame, not just the center-of-mass frame.
Equations (16.52) are the most important of the general relativistic equations governing spherically sym-
metric spacetimes. It is these equations that are responsible (to the extent that equations may be con-
sidered responsible) for the strange internal structure of Reissner-Nordstr om black holes, and for mass
ination. The coecient
t
equals the coordinate radial 4-velocity dr/d =
t
r =
t
of the tetrad
frame, equation (16.4), and thus equation (16.52a) can be regarded as giving the proper radial acceleration
D
2
r/D
2
= D
t
/D = D
t
t
of the tetrad frame as measured by a person who is in free-fall and instanta-
neously at rest in the tetrad frame. If the acceleration is measured by an observer who is continuously at
rest in the tetrad frame (as opposed to being in free-fall), then the proper acceleration is
t
t
= D
t
t
+
r
h
t
.
The presence of the extra term
r
h
t
, proportional to the proper acceleration h
t
actually experienced by
the observer continuously at rest in the tetrad frame, reects the principle of equivalence of gravity and
acceleration.
The right hand side of equation (16.52a) can be interpreted as the radial gravitational force, which consists
of two terms. The rst term, M/r
2
, looks like the familiar Newtonian gravitational force, which is attractive
(negative, inward) in the usual case of positive mass M. The second term, 4rp, proportional to the radial
pressure p, is what makes spherical spacetimes in general relativity interesting. In a Reissner-Nordstr omblack
hole, the negative radial pressure produced by the radial electric eld produces a radial gravitational repulsion
(positive, outward), according to equation (16.52a), and this repulsion dominates the gravitational force at
small radii, producing an inner horizon. In mass ination, the (positive) radial pressure of relativistically
counter-streaming ingoing and outgoing streams just above the inner horizon dominates the gravitational
force (inward), and it is this that drives mass ination.
Like the second half of a vaudeville act, the second Einstein equation (16.52b) also plays an indispensible
role. The quantity
r

r
r on the left hand side is the proper radial gradient of the circumferential radius
r measured by a person at rest in the tetrad frame. The sign of
r
determines which way an observer at
rest in the tetrad frame thinks is outwards, the direction of larger circumferential radius r. A positive
r
means that the observer thinks the outward direction points away from the black hole, while a negative
r
means that the observer thinks the outward direction points towards from the black hole. Outside the
outer horizon
r
is necessarily positive, because
m
must be spacelike there. But inside the horizon
r
may
be either positive or negative. A tetrad frame can be dened as ingoing if the proper radial gradient
r
is positive, and outgoing if
r
is negative. In the Reissner-Nordstr om geometry, ingoing geodesics have
positive energy, and outgoing geodesics have negative energy. However, the present denition of ingoing or
outgoing based on the sign of
r
is general there is no need for a timelike Killing vector such as would be
necessary to dene the (conserved) energy of a geodesic.
Equation (16.52b) shows that the proper rate of change D
t
r
in the radial gradient
r
measured by an
observer who is in free-fall and instantaneously at rest in the tetrad frame is proportional to the radial energy
ux f in that frame. But ingoing observers tend to see energy ux pointing away from the black hole, while
outgoing observers tend to see energy ux pointing towards the black hole. Thus the change in
r
tends to
be in the same direction as
r
, amplifying
r
whatever its sign.
Exercise 16.1 Birkhos theorem. Prove Birkhos theorem from equations (16.52).
16.1.12 Comment on the vierbein coecient
Whereas the Einstein equations (16.52) give evolution equations for the vierbein coecients
t
and
r
, there
is no evolution equation for the vierbein coecient . Indeed, the Einstein equations involve the vierbein
coecient only in the combination h
t

r
ln . This reects the fact that, even after the tetrad frame
is xed, there is still a coordinate freedom t t
(t) in the choice of coordinate time t. Under such a gauge

transformation, transforms as
= f(t) where f(t) = t
/t is an arbitrary function of coordinate

time t. Only h
t

r
ln is independent of this coordinate gauge freedom, and thus only h
t
, not itself,
appears in the tetrad-frame Einstein equations.
Since is needed to propagate the equations from one coordinate time to the next (because
t
= /t +
t
/r), it is necessary to construct by integrating h
t

r
ln /r along the radial direction r at each
time step. The arbitrary normalization of at each step might be xed by choosing to be unity at innity,
which corresponds to xing the time coordinate t to equal the proper time at innity.
In the particular case that the tetrad frame is taken to be in free-fall everywhere, h
t
= 0, as in the
Gullstrand-Painleve line-element, then is constant at xed t, and without loss of generality it can be xed
equal to unity everywhere, = 1. I like to think of a free-fall frame as being realized physically by tracer
dark matter particles that free-fall radially (from zero velocity, typically) at innity, and stream freely,
without interacting, through any actual matter that may be present.
16.2 Spherical electromagnetic eld
The internal structure of a charged black hole resembles that of a rotating black hole because the negative
pressure (tension) of the radial electric eld produces a gravitational repulsion analogous to the centrifugal
repulsion in a rotating black hole. Since it is much easier to deal with spherical than rotating black holes, it
is common to use charge as a surrogate for rotation in exploring black holes.
16.2.1 Electromagnetic eld
The assumption of spherical symmetry means that any electromagnetic eld can consist only of a radial elec-
tric eld (in the absence of magnetic monopoles). The only non-vanishing components of the electromagnetic
eld F
mn
are then
F
tr
= F
rt
= E =
Q
r
2
, (16.53)
where E is the radial electric eld, and Q(t, r) is the interior electric charge. Equation (16.53) can be
regarded as dening what is meant by the electric charge Q interior to radius r at time t.
16.2.2 Maxwells equations
A radial electric eld automatically satises two of Maxwells equations, the source-free ones (11.64). For
the radial electric eld (16.53), the other two Maxwells equations, the sourced ones (11.65), are
r
Q = 4r
2
q , (16.54a)
t
Q = 4r
2
j , (16.54b)
where q j
t
is the proper electric charge density and j j
r
is the proper radial electric current density in
the tetrad frame.
16.2.3 Electromagnetic energy-momentum tensor
For the radial electric eld (16.53), the electromagnetic energy-momentum tensor (11.70) in the tetrad frame
is the diagonal tensor
T
mn
e
=
Q
2
8r
4
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
. (16.55)
16.3 General relativistic stellar structure 277
The radial electric energy-momentum tensor is independent of the radial motion of the tetrad frame, which
reects the fact that the electric eld is invariant under a radial Lorentz boost. The energy density
e
and
radial and transverse pressures p
e
and p
e
of the electromagnetic eld are the same as those from a spherical
charge distribution with interior electric charge Q in at space
e
= p
e
= p
e
=
Q
2
8r
4
=
E
2
8
. (16.56)
The non-vanishing components of the covariant derivative D
m
T
mn
e
of the electromagnetic energy-mom-
entum (16.55) are
D
m
T
mt
e
=
t
e
+
4
t
r

e
=
Q
4r
4

t
Q =
jQ
r
2
= jE , (16.57a)
D
m
T
mr
e
=
r
p
e
+
4
r
r
p
e
=
Q
4r
4

r
Q =
qQ
r
2
= qE . (16.57b)
The rst expression (16.57a), which gives the rate of energy transfer out of the electromagnetic eld as the
current density j times the electric eld E, is the same as in at space. The second expression (16.57b),
which gives the rate of transfer of radial momentum out of the electromagnetic eld as the charge density q
times the electric eld E, is the Lorentz force on a charge density q, and again is the same as in at space.
16.3 General relativistic stellar structure
A star can be well approximated as static as well as spherically symmetric. In this case all time derivatives
can be taken to vanish, /t = 0, and, since the center-of-mass frame coincides with the rest frame, it is
natural to choose the tetrad frame to be at rest,
t
= 0. The Einstein equation (16.52b) then vanishes
identically, while the Einstein equation (16.52a) becomes
r
h
t
=
M
r
2
+ 4rp , (16.58)
which expresses the proper acceleration h
t
in the rest frame in terms of the familiar Newtonian gravitational
force M/r
2
plus a term 4rp proportional to the radial pressure. The radial pressure p, if positive as is the
usual case for a star, enhances the inward gravitational force, helping to destabilize the star. Because
t
is
zero, the interior mass M given by equation (16.9) reduces to
1 2M/r =
2
r
. (16.59)
When equations (16.58) and (16.59) are substituted into the momentum equation (16.44b), and if the pressure
is taken to be isotropic, so p
= p, the result is the Oppenheimer-Volkov equation for general relativistic

hydrostatic equilibrium
p
r
=
( +p)(M + 4r
3
p)
r
2
(1 2M/r)
. (16.60)
In the Newtonian limit p and M r this goes over to (with units restored)
p
r
=
GM
r
2
, (16.61)
which is the usual Newtonian equation of spherically symmetric hydrostatic equilibrium.
16.4 Self-similar spherically symmetric spacetime
Even with the assumption of spherical symmetry, it is by no means easy to solve the system of partial
dierential equations that comprise the Einstein equations coupled to mass-energy of various kinds. One
way to simplify the system of equations, transforming them into ordinary dierential equations, is to consider
self-similar solutions.
16.4.1 Self-similarity
The assumption of self-similarity (also known as homothety, if you can pronounce it) is the assumption
that the system possesses conformal time translation invariance. This implies that there exists a conformal
time coordinate such that the geometry at any one time is conformally related to the geometry at any
other time
ds
2
= a()
2
_
g
(c)
(x) d
2
+ 2 g
(c)
x
(x) d dx +g
(c)
xx
(x) dx
2
+e
2x
do
2
_
. (16.62)
Here the conformal metric coecients g
(c)
(x) are functions only of conformal radius x, not of conformal time

. The choice e
2x
of coecient of do
2
is a gauge choice of the conformal radius x, carefully chosen here so as
to bring the self-similar line-element into a form (16.66) below that resembles as far as possible the spherical
line-element (16.1). In place of the conformal factor a() it is convenient to work with the circumferential
radius r
r a()e
x
(16.63)
which is to be considered as a function r(, x) of the coordinates and x. The circumferential radius r has
a gauge-invariant meaning, whereas neither a() nor x are independently gauge-invariant. The conformal
factor r has the dimensions of length. In self-similar solutions, all quantities are proportional to some power of
r, and that power can be determined by dimensional analysis. Quantites that depend only on the conformal
radial coordinate x, independent of the circumferential radius r, are called dimensionless.
The fact that dimensionless quantities such as the conformal metric coecients g
(c)
(x) are independent of

conformal time implies that the tangent vector g
, which by denition satises
= g
, (16.64)
is a conformal Killing vector, also known as the homothetic vector. The tetrad-frame components of the
conformal Killing vector g
denes the tetrad-frame conformal Killing 4-vector

m
r
m
m
, (16.65)
in which the factor r is introduced so as to make
m
dimensionless. The conformal Killing vector g
is
the generator of the conformal time translation symmetry, and as such it is gauge-invariant (up to a global
rescaling of conformal time, b for some constant b). It follows that its dimensionless tetrad-frame
components
m
constitute a tetrad 4-vector (again, up to global rescaling of conformal time).
16.4.2 Self-similar line-element
The self-similar line-element can be taken to have the same form as the spherical line-element (16.1), but
with the dependence on the dimensionless conformal Killing vector
m
made manifest:
ds
2
= r
2
_
(
d)
2
+
1
2
x
(dx +
x
x
d)
2
+do
2
_
. (16.66)
The vierbein e
m

m
corresponding to the self-similar line-element (16.66) are

e
m
=
1
r
_
_
_
_
1/
x
/
0 0
0
x
0 0
0 0 1 0
0 0 0 1/ sin
_
_
_
_
, e
m
= r
_
_
_
_
0 0 0
x
1/
x
0 0
0 0 1 0
0 0 0 sin
_
_
_
_
. (16.67)
It is straightforward to see that the coordinate time components of the inverse vierbein must be e
m
= r
m
,
since / = e
m

m
equals r
m
m
, equation (16.65).
16.4.3 Tetrad-frame scalars and vectors
Since the conformal factor r is gauge-invariant, the directed gradient
m
r constitutes a tetrad-frame 4-vector
m
(which unlike
m
is independent of any global rescaling of conformal time)
m

m
r . (16.68)
It is straightforward to check that
x
dened by equation (16.68) is consistent with its appearance in the
vierbein (16.67) provided that r e
x
as earlier assumed, equation (16.63).
With two distinct dimensionless tetrad 4-vectors in hand,
m
and the conformal Killing vector
m
, three
gauge-invariant dimensionless scalars can be constructed,
m
m
,
m
m
, and
m
m
,
1
2M
r
=
m
m
=
2
+
2
x
, (16.69)
v
m
m
=
1
r
r
=
1
a
a
, (16.70)

m
m
= (
)
2
(
x
)
2
. (16.71)
Equation (16.69) is essentially the same as equation (16.9). The dimensionless quantity v, equation (16.70),
may be interpreted as a measure of the expansion velocity of the self-similar spacetime. Equation (16.70)
shows that v is a function only of (since a() is a function only of ), and it therefore follows that v must
be constant (since being dimensionless means that v must be a function only of x). Equation (16.70) then
also implies that the conformal factor a() must take the form
a() = e
v
. (16.72)
Because of the freedom of a global rescaling of conformal time, it is possible to set v = 1 without loss of
generality, but in practice it is convenient to keep v, because it is then transparent how to take the static
limit v 0. Equation (16.72) along with (16.63) shows that the circumferential radius r is related to the
conformal coordinates and x by
r = e
v+x
. (16.73)
The dimensionsless quantity , equation (16.71), is the dimensionless horizon function: horizons occur where
the horizon function vanishes
= 0 at horizons . (16.74)
16.4.4 Self-similar diagonal line-element
The self-similar line-element (16.66) can be brought to diagonal form by a coordinate transformation to
diagonal conformal coordinates
, x
(subscripted for diagonal)

= +f(x) , x x
= x vf(x) , (16.75)
which leaves unchanged the conformal factor r, equation (16.73). The resulting diagonal metric is
ds
2
= r
2
_
d
2
+
dx
2
1 2M/r + v
2
/
+do
2
_
. (16.76)
The diagonal line-element (16.76) corresponds physically to the case where the tetrad frame is at rest in
the similarity frame,
x
= 0, as can be seen by comparing it to the line-element (16.66). The frame can be
called the similarity frame. The form of the metric coecients follows from the line-element (16.66) and
the gauge-invariant scalars (16.69)(16.71).
The conformal Killing vector in the similarity frame is
m
=
1/2
, 0, 0, 0, and the 4-velocity of the
similarity frame in its own frame is u
m
= 1, 0, 0, 0. Since both are tetrad 4-vectors, it follows that with
respect to a general tetrad frame
m
= u
m
1/2
(16.77)
where u
m
is the 4-velocity of the similarity frame with respect to the general frame. This shows that the
conformal Killing vector
m
in a general tetrad frame is proportional to the 4-velocity of the similarity frame
through the tetrad frame. In particular, the proper 3-velocity of the similarity frame through the tetrad
frame is
proper 3-velocity of similarity frame through tetrad frame =

x
. (16.78)
16.4.5 Ray-tracing line-element
It proves useful to introduce a ray-tracing conformal radial coordinate X related to the coordinate x
of
the diagonal line-element (16.76) by
dX
dx
[(1 2M/r) + v
2
]
1/2
. (16.79)
In terms of the ray-tracing coordinate X, the diagonal metric is
ds
2
= r
2
_
d
2
+
dX
2
+do
2
_
. (16.80)
For the Reissner-Nordstr om geometry, = (1 2M/r)/r
2
,
= t, and X = 1/r.
16.4.6 Geodesics
Spherical symmetry and conformal time translation symmetry imply that geodesic motion in spherically
symmetric self-similar spacetimes is described by a complete set of integrals of motion.
The integral of motion associated with conformal time translation symmetry can be obtained from La-
granges equations of motion
d
d
L
u
=
L
, (16.81)
with eective Lagrangian L = g
for a particle with 4-velocity u
. The self-similar metric depends on

the conformal time only through the overall conformal factor g
a()
2
. The derivative of the conformal
factor is given by ln a/ = v, equation (16.70), so it follows that L/ = 2 v L. For a massive particle,
for which conservation of rest mass implies g
= 1, Lagranges equations (16.81) thus yield

du
d
= v . (16.82)
In the limit of zero accretion rate, v 0, equation (16.82) would integrate to give u
as a constant, the
energy per unit mass of the geodesic. But here there is conformal time translation symmetry in place of
time translation symmetry, and equation (16.82) integrates to
u
= v , (16.83)
in which an arbitrary constant of integration has been absorbed into a shift in the zero point of the proper
time . Although the above derivation was for a massive particle, it holds also for a massless particle, with the
understanding that the proper time is constant along a null geodesic. The quantity u
in equation (16.83)
is the covariant time component of the coordinate-frame 4-velocity u
of the particle; it is related to the

covariant components u
m
of the tetrad-frame 4-velocity of the particle by
u
= e
m
u
m
= r
m
u
m
. (16.84)
Without loss of generality, geodesic motion can be taken to lie in the equatorial plane = /2 of the
spherical spacetime. The integrals of motion associated with conformal time translation symmetry, rotational
symmetry about the polar axis, and conservation of rest mass, are, for a massive particle
u
= v , u
= L
z
, u
= 1 , (16.85)
where L
z
is the orbital angular momentum per unit rest mass of the particle. The coordinate 4-velocity
u
dx
/d that follows from equations (16.85) takes its simplest form in the conformal coordinates
, X, , of the ray-tracing metric (16.80)

u
=
v
r
2
, u
X
=
1
r
2
_
v
2
2
(r
2
+L
2
z
)
1/2
, u
=
L
z
r
2
. (16.86)
16.4.7 Null geodesics
The important case of a massless particle follows from taking the limit of a massive particle with innite
energy and angular momentum, v and L
z
. To obtain nite results, dene an ane parameter
by CHECK d v d, and a 4-velocity in terms of it by v
dx
/d. The integrals of motion (16.85)

then become, for a null geodesic,
v
= 1 , v
= J
z
, v
= 0 , (16.87)
where J
z
L
z
/(v ) is the (dimensionless) conformal angular momentum of the particle. The 4-velocity v
along the null geodesic is then, in terms of the coordinates of the ray-tracing metric (16.80),
v
=
1
r
2
, v
X
=
1
r
2
_
1 J
2
z
_
1/2
, v
=
J
z
r
2
. (16.88)
Equations (16.88) yield the shape of a null geodesic by quadrature
=
_
J
z
dX
(1 J
2
z
)
1/2
. (16.89)
Equation (16.89) shows that the shape of null geodesics in spherically symmetric self-similar spacetimes
hinges on the behavior of the dimensionless horizon function (X) as a function of the dimensionless ray-
tracing variable X. Null geodesics go through periapsis or apoapsis in the self-similar frame where the
denominator of the integrand of (16.89) is zero, corresponding to v
X
= 0.
In the Reissner-Nordstr om geometry there is a radius, the photon sphere, where photons can orbit in
circles for ever. In non-stationary self-similar solutions there is no conformal radius where photons can orbit
for ever (to remain at xed conformal radius, the photon angular momentum would have to increase in
proportion to the conformal factor r). There is however a separatrix between null geodesics that do or do
not fall into the black hole, and the conformal radius where this occurs can be called the photon sphere
equivalent. The photon sphere equivalent occurs where the denominator of the integrand of equation (16.89)
not only vanishes, v
X
= 0, but is an extremum, which happens where the horizon function is an extremum,
d
dX
= 0 at photon sphere equivalent . (16.90)
16.4.8 Dimensional analysis
Dimensional analysis shows that the conformal coordinates x
, x, , and the tetrad metric

mn
are
dimensionless, while the coordinate metric g
scales as r
2
,
x
r
0
,
mn
r
0
, g
r
2
. (16.91)
The vierbein e
m

m
, equations (16.67), scale as

e
m
r
1
, e
m
r . (16.92)
Coordinate derivatives /x
are dimensionless, while directed derivatives

m
scale as 1/r,
r
0
,
m
r
1
. (16.93)
The tetrad connections
kmn
and the tetrad-frame Riemann tensor R
klmn
scale as
kmn
r
1
, R
klmn
r
2
. (16.94)
16.4.9 Variety of self-similar solutions
Self-similar solutions exist provided that the properties of the energy-momentum introduce no additional
dimensional parameters. For example, the pressure-to-density ratio w p/ of any species is dimensionless,
and since the ratio can depend only on the nature of the species itself, not for example on where it happens
to be located in the spacetime, it follows that the ratio w must be a constant. It is legitimate for the
pressure-to-density ratio to be dierent in the radial and transverse directions (as it is for a radial electric
eld), but otherwise self-similarity requires that
w p/ , w
/ , (16.95)
be constants for each species. For example, w = 1 for an ultrahard uid (which can mimic the behaviour
of a massless scalar eld: E. Babichev, S. Chernov, V. Dokuchaev, Yu. Eroshenko, 2008, Ultra-hard uid
and scalar eld in the Kerr-Newman metric, Phys. Rev. D 78, 104027, arXiv:0807.0449), w = 1/3 for a
relativistic uid, w = 0 for pressureless cold dark matter, w = 1 for vacuum energy, and w = 1 with
w
= 1 for a radial electric eld.

Self-similarity allows that the energy-momentum may consist of several distinct components, such as
a relativistic uid, plus dark matter, plus an electric eld. The components may interact with each other
provided that the properties of the interaction introduce no additional dimensional parameters. For example,
the relativistic uid (and the dark matter) may be charged, and if so then the charged uid will experience
a Lorentz force from the electric eld, and will therefore exchange momentum with the electric eld. If
the uid is non-conducting, then there is no dissipation, and the interaction between the charged uid and
electric eld automatically introduces no additional dimensional parameters.
However, if the charged uid is electrically conducting, then the electrical conductivity could potentially
introduce an additional dimensional parameter, and this must not be allowed if self-similarity is to be
maintained. In diusive electrical conduction in a uid of conductivity , an electric eld E gives rise to a
current
j = E , (16.96)
which is just Ohms law. Dimensional analysis shows that j r
2
and E r
1
, so the conductivity must
scale as r
1
. The conductivity can depend only on the intrinsic properties of the conducting uid, and
the only intrinsic property available is its density, which scales as r
2
. If follows that the conductivity
must be proportional to the square root of the density of the conducting uid
=
1/2
, (16.97)
where is a dimensionless conductivity constant. The form (16.97) is required by self-similarity, and is
not necessarily realistic (although it is realistic that the conductivity increases with density). However, the
conductivity (16.97) is adequate for the purpose of exploring the consequences of dissipation in simple models
of black holes.
16.4.10 Tetrad connections
The expressions for the tetrad connections for the self-similar spacetime are the same as those (16.17) for
a general spherically symmetric spacetime, with just a relabelling of the time and radial coordinates into
conformal coordinates
t , r x . (16.98)
Specically, equations (16.17) for the tetrad connections become become
x
= h
,
xx
= h
x
,
r
,
x
=
x
=

x
r
,
=
cot
r
, (16.99)
in which h
and h
x
have the same physical interpretation discussed in 16.1.4 for the general spherically sym-
metric case: h
is the proper radial acceleration, and h

x
is the radial Hubble parameter. Expressions (16.18)
and (16.19) for h
and h
x
translate in the self-similar spacetime to
h

x
ln(r
) , h
x

ln(r
x
) . (16.100)
Comparing equations (16.100) to equations (16.18) and (16.23) shows that the vierbein coecient and
scale factor translate in the self-similar spacetime to
=
1
r
, =
1
r
x
. (16.101)
16.4.11 Spherical equations carry over to the self-similar case
The tetrad-frame Riemann, Weyl, and Einstein tensors in the self-similar spacetime take the same form
as in the general spherical case, equations (16.24)(16.26), with just a relabelling (16.98) into conformal
coordinates.
Likewise, the equations for the interior mass in 16.1.8, for energy-momentum conservation in 16.1.9, for
the rst law in 16.1.10, and the various equations for the electromagnetic eld in 16.2, all carry through
unchanged except for a relabelling (16.98) of coordinates.
16.4.12 From partial to ordinary dierential equations
The central simplifying feature of self-similar solutions is that they turn a system of partial dierential
equations into a system of ordinary dierential equations.
By denition, a dimensionless quantity F(x) is independent of conformal time . It follows that the partial
derivative of any dimensionless quantity F(x) with respect to conformal time vanishes
0 =
F(x)
=
m
m
F(x) = (
+
x
x
) F(x) . (16.102)
Consequently the directed radial derivative
x
F of a dimensionless quantity F(x) is related to its directed
time derivative
F by
x
F(x) =
F(x) . (16.103)
Equation (16.103) allows radial derivatives to be converted to time derivatives.
16.4.13 Integrals of motion
As remarked above, equation (16.102), in self-similar solutions
m
m
F(x) = 0 for any dimensionless function
F(x). If both the directed derivatives
F(x) and
x
F(x) are known from the Einstein equations or elsewhere,
then the result will be an integral of motion.
The spherically symmetric, self-similar Einstein equations admit two integrals of motion
0 = r
m
=r
x
(
+
x
h
x
)
_
M
r
+ 4r
2
p
_
+
x
4r
2
f , (16.104a)
0 = r
m
x
=r
+
x
h
x
) +
x
_
M
r
4r
2
_
+
4r
2
f . (16.104b)
Taking
x
times (16.104a) plus
times (16.104b), and then
times (16.104a) minus

x
times (16.104b),
gives
0 = r v (
+
x
h
x
) 4r
2
_
x
( +p)
_
(
)
2
+ (
x
)
2
_
f
, (16.105a)
0 = r
m
m
M
r
= v
M
r
+ 4r
2
[
x
p + (
)f] . (16.105b)
The quantities in square brackets on the right hand sides of equations (16.105) are scalars for each species
a, so equations (16.105) can also be written
r v (
+
x
h
x
) = 4r
2
species a
x
a
(
a
+p
a
) , (16.106a)
v
M
r
= 4r
2
species a
(
a,x
x
a
a,
a
p
a
) , (16.106b)
where the sum is over all species a, and
a,m
and
m
a
are the 4-vectors
m
and
m
expressed in the rest frame
of species a.
For electrically charged solutions, a third integral of motion comes from
0 = r
m
m
Q
r
= v
Q
r
+ 4r
2
(
x
q
j) (16.107)
which is valid in any radial tetrad frame, not just the center-of-mass frame.
For a uid with equation of state p = w, a fourth integral comes from considering
0 = r
m
m
(r
2
p) = r
_
w
(r
2
) +
x
x
(r
2
p)
(16.108)
and simplifying using the energy conservation equation for
and the momentum conservation equation

for
x
p.
16.4.14 Integration variable
It is desirable to choose an integration variable that varies monotonically. A natural choice is the proper
time in the tetrad frame, since this is guaranteed to increase monotonically. Since the 4-velocity at rest
in the tetrad frame is by denition u
m
= 1, 0, 0, 0, the proper time derivative is related to the directed
conformal time derivative in the tetrad frame by d/d = u
m
m
=
.
However, there is another choice of integration variable, the ray-tracing variable X dened by equa-
tion (16.79), that is not specically tied to any tetrad frame, and that has a desirable (tetrad and coordinate)
gauge-invariant meaning. The proper time derivative of any dimensionless function F(x) in the tetrad frame
is related to its derivative dF/dX with respect to the ray-tracing variable X by
F = u
m
m
F = u
X
X
F =
x
r
dF
dX
. (16.109)
In the third expression, u
X
X
F is u
m
m
F expressed in the similarity frame of 16.4.4, the time contribu-
tion u
F vanishing in the similarity frame because it is proportional to the conformal time derivative
F/
= 0. In the last expression of (16.109), u

X
has been replaced by u
x
=
x
/
1/2
in view of equa-
tion (16.77), the minus sign coming from the fact that u
X
is the radial component of the tetrad 4-velocity of
the tetrad frame relative to the similarity frame, while u
x
in equation (16.77) is the radial component of the
tetrad 4-velocity of the similarity frame relative to the tetrad frame. Also in the last expression of (16.109),
the directed derivative
X
with respect to the ray-tracing variable X has been translated into its coordinate
partial derivative,
X
= (
1/2
/r) /X, which follows from the metric (16.80).
In summary, the chosen integration variable is the dimensionless ray-tracing variable X (with a minus
because X is monotonically increasing), the derivative with respect to which, acting on any dimensionless
function, is related to the proper time derivative in any tetrad frame (not just the baryonic frame) by
d
dX
=
r
. (16.110)
Equation (16.110) involves
x
, which is proportional to the proper velocity of the tetrad frame through the
similarity frame, equation (16.78), and which therefore, being initially positive, must always remain positive
in any tetrad frame attached to a uid, as long as the uid does not turn back on itself, as must be true for
the self-similar solution to be consistent.
16.4.15 Summary of equations for a single charged uid
For reference, it is helpful to collect here the full set of equations governing self-similar spherically symmetric
evolution in the case of a single charged uid with isotropic equation of state
p = p
= w , (16.111)
and conductivity
=
1/2
. (16.112)
In accordance with the arguments in 16.4.9, equations (16.95) and (16.97), self-similarity requires that the
pressure-to-density ratio w and the conductivity coecient both be (dimensionless) constants.
It is natural to work in the center-of-mass frame of the uid, which also coincides with the center-of-mass
frame of the uid plus electric eld (the electric eld, being invariant under Lorentz boosts, does not pick
out any particular radial frame).
The proper time in the uid frame evolves as
d
dX
=
r
x
, (16.113)
which follows from equation (16.110) and the fact that
= 1. The circumferential radius r evolves along

the path of the uid as
d ln r
dX
=

x
. (16.114)
Although it is straightforward to write down the equations governing how the tetrad frame moves through the
conformal coordinates and x, there is not much to be gained from this because the conformal coordinates
have no fundamental physical signicance.
Next, the dening equations (16.100) for the proper acceleration h
and Hubble parameter h

x
yield
equations for the evolution of the time and radial components of the conformal Killing vector
m
dX
=
x
rh
, (16.115a)
d
x
dX
=
+rh
x
, (16.115b)
in which, in the formula for h
, equation (16.103) has been used to convert the conformal radial derivative
x
to the conformal time derivative
, and thence to d/dX by equation (16.110).

Next, the Einstein equations (16.26c) and (16.26c) (with coordinates relabeled per (16.98) in the center-of-
mass frame (16.33) yield evolution equations for the time and radial components of the vierbein coecients
dX
=
rh
x
, (16.116a)
d
x
dX
=

x
rh
, (16.116b)
where again, in the formula for
, equation (16.103) has been used to convert the conformal radial derivative
x
to the conformal time derivative
. The 4 evolution equations (16.115) and (16.116) for

m
and
m
are
not independent: they are related by
m
m
= v, a constant, equation (16.70). To maintain numerical
precision, it is important to avoid expressing small quantities as dierences of large quantities. In practice,
a suitable choice of variables to integrate proves to be
+
x
,

x
, and
x
, each of which can be
tiny in some circumstances. Starting from these variables, the following equations yield
x
, along with
the interior mass M and the horizon function , equations (16.69) and (16.71), in a fashion that ensures
numerical stability:
x
=
2v (
+
x
)(
+
x
)
x
, (16.117a)
2M
r
= 1 + (
+
x
)(
x
) , (16.117b)
= (
+
x
)(
x
) . (16.117c)
The evolution equations (16.115) and (16.116) involve h
and h
x
. The integrals of motion considered in
16.4.13 yield explicit expressions for h
and h
x
not involving any derivatives. For the Hubble parameter
h
x
, equation (16.105a) gives
rh
x
=
x
rh
v
4 , (16.118)
where is the dimensionless enthalpy
r
2
(1 +w) . (16.119)
For the proper acceleration h
, a somewhat lengthy calculation starting from the integral of motion (16.108),

and simplifying using the integral of motion (16.107) for Q, the expression (16.118) for h
x
, Maxwells equa-
tion (16.54b) [with the relabelling (16.98)], and the conductivity (16.112) in Ohms law (16.96), gives
rh
x
_
8w
(
x
x
w
)r
2
+ [v + (1 +w)4r
] Q
2
/r
2
w(4
)
2
/v
_
4 [(
x
)
2
w(
)
2
]
. (16.120)
Two more equations complete the suite. The rst, which represents energy conservation for the uid, can
be written as an equation governing the entropy S of the uid
d ln S
dX
=
Q
2
r(1 +w)
x
, (16.121)
in which the S is (up to an arbitrary constant) the entropy of a comoving volume element V r
3
x
of the
uid
S r
3
1/(1+w)
. (16.122)
That equation (16.121) really is an entropy equation can be conrmed by rewriting it as
1
V
_
dV
d
+p
dV
d
_
= jE =
Q
2
r
4
, (16.123)
in which jE is recognized as the Ohmic dissipation, the rate per unit volume at which the volume element
V is being heated.
The nal equation represents electromagnetic energy conservation, equation (16.57a), which can be written
d ln Q
dX
=
4r
x
. (16.124)
The (heat) energy going into the uid is balanced by the (free) energy coming out of the electromagnetic
eld.
17
The interiors of spherical black holes
As discussed in Chapter 8, the Reissner-Nordstr om geometry for an ideal charged spherical black hole
contains mathematical wormhole and white hole extensions to other universes. In reality, these extensions
are not expected to occur, thanks to the mass ination instability discovered by Poisson & Israel (1990).
17.1 The mechanism of mass ination
17.1.1 Reissner-Nordstr om phase
Figure 17.1 illustrates how the two Einstein equations (16.52) produce the three phases of mass ination
inside a charged spherical black hole.
During the initial phase, illustrated in the top panel of Figure 17.1, the spacetime geometry is well-
approximated by the vacuum, Reissner-Nordstr om geometry. During this phase the radial energy ux f is
eectively zero, so
r
remains constant, according to equation (16.52b). The change in the radial velocity
t
,
equation (16.52a), depends on the competition between the Newtonian gravitational force M/r
2
, which is
always attractive (tending to make the radial velocity
t
more negative), and the gravitational force 4rp
sourced by the radial pressure p. In the Reissner-Nordstr om geometry, the static electric eld produces a
negative radial pressure, or tension, p = Q
2
/(8r
4
), which produces a gravitational repulsion 4rp =
Q
2
/(2r
3
). At some point (depending on the charge-to-mass ratio) inside the outer horizon, the gravitational
repulsion produced by the tension of the electric eld exceeds the attraction produced by the interior mass
M, so that the radial velocity
t
slows down. This regime, where the (negative) radial velocity
t
is slowing
down (becoming less negative), while
r
remains constant, is illustrated in the top panel of Figure 17.1.
If the initial Reissner-Nordstr om phase were to continue, then the radial 4-gradient
m
would become
lightlike. In the Reissner-Nordstr om geometry this does in fact happen, and where it happens denes the
inner horizon. The problem with this is that the lightlike 4-vector
m
points in one direction for ingoing
frames, and in the opposite direction for outgoing frames. If
m
becomes lightlike, then ingoing and outgoing
frames are streaming through each other at the speed of light. This is the innite blueshift at the inner
horizon rst pointed out by Penrose (1968).
17.1 The mechanism of mass ination 291
r
o
u
t
g
o
i
n
g
i
n
g
o
i
n
g
1. RN
r
o
u
t
g
o
i
n
g
i
n
g
o
i
n
g
2. Ination
r
o
u
t
g
o
i
n
g
i
n
g
o
i
n
g
3. Collapse
Figure 17.1 Spacetime diagrams of the tetrad-frame 4-vector m, equation (16.8), illustrating qualitatively
the three successive phases of mass ination: 1. (top) the Reissner-Nordstrom phase, where ination ignites;
2. (middle) the inationary phase itself; and 3. (bottom) the collapse phase, where ination comes to an end.
In each diagram, the arrowed lines labeled ingoing and outgoing illustrate two representative examples of
the 4-vector {t, r}, while the double-arrowed lines illustrate the rate of change of these 4-vectors implied
by Einsteins equations (16.52). Inside the horizon of a black hole, all locally inertial frames necessarily fall
inward, so the radial velocity t tr is always negative. A locally inertial frame is ingoing or outgoing
depending on whether the proper radial gradient r rr measured in that frame is positive or negative.
292 The interiors of spherical black holes
If there were no matter present, or if there were only one stream of matter, either ingoing or outgoing but
not both, then
m
could indeed become lightlike. But if both ingoing and outgoing matter are present, even
in the tiniest amount, then it is physically impossible for the ingoing and outgoing frames to stream through
each other at the speed of light.
If both ingoing and outgoing streams are present, then as they race through each other ever faster, they
generate a radial pressure p, and an energy ux f, which begin to take over as the main source on the right
hand side of the Einstein equations (16.52). This is how mass ination is ignited.
17.1.2 Inationary phase
The infalling matter now enters the second, mass inationary phase, illustrated in the middle panel of
Figure 17.1.
During this phase, the gravitational force on the right hand side of the Einstein equation (16.52a) is
dominated by the pressure p produced by the counter-streaming ingoing and outgoing matter. The mass M
is completely sub-dominant during this phase (in this respect, the designation mass ination is misleading,
since although the mass inates, it does not drive ination). The counter-streaming pressure p is positive,
and so accelerates the radial velocity
t
(makes it more negative). At the same time, the radial gradient
r
is being driven by the energy ux f, equation (16.52b). For typically low accretion rates, the streams
are cold, in the sense that the streaming energy density greatly exceeds the thermal energy density, even if
the accreted material is at relativistic temperatures. This follows from the fact that for mass ination to
begin, the gravitational force produced by the counter-streaming pressure p must become comparable to that
produced by the mass M, which for streams of low proper density requires a hyper-relativistic streaming
velocity. For a cold stream of proper density moving at 4-velocity u
m
u
t
, u
r
, 0, 0, the streaming energy
ux would be f u
t
u
r
, while the streaming pressure would be p (u
r
)
2
. Thus their ratio f/p u
t
/u
r
is slightly greater than one. It follows that, as illustrated in the middle panel of Figure 17.1, the change in
r
slightly exceeds the change in
t
, which drives the 4-vector
m
, already nearly lightlike, to be even more
nearly lightlike. This is mass ination.
Ination feeds on itself. The radial pressure p and energy ux f generated by the counter-streaming ingoing
and outgoing streams increase the gravitational force. But, as illustrated in the middle panel of Figure 17.1,
the gravitational force acts in opposite directions for ingoing and outgoing streams, tending to accelerate
the streams faster through each other. An intuitive way to understand this is that the gravitational force
is always inwards, meaning in the direction of smaller radius, but the inward direction is towards the black
hole for ingoing streams, and away from the black hole for outgoing streams.
The feedback loop in which the streaming pressure and ux increase the gravitational force, which ac-
celerates the streams faster through each other, which increases the streaming pressure and ux, is what
drives mass ination. Ination produces an exponential growth in the streaming energy, and along with it
the interior mass, and the Weyl curvature.
17.2 The far future? 293
17.1.3 Collapse phase
It might seem that ination is locked into an exponential growth from which there is no exit. But the
Einstein equations (16.52) have one more trick up their sleave.
For the counter-streaming velocity to continue to increase requires that the change in
r
from equa-
tion (16.52b) continues to exceed the change in
t
from equation (16.52a). This remains true as long as the
counter-streaming pressure p and energy ux f continue to dominate the source on the right hand side of
the equations. But the mass term M/r
2
also makes a contribution to the change in
t
, equation (16.52a).
As will be seen in the examples of the next two sections, ?? and ??, at least in the case of pressureless
streams the mass term exponentiates slightly faster than the pressure term. At a certain point, the additional
acceleration produced by the mass means that the combined gravitational force M/r
2
+4rp exceeds 4rf.
Once this happens, the 4-vector
m
, instead of being driven to becoming more lightlike, starts to become
less lightlike. That is, the counter-streaming velocity starts to slow. At that point ination ceases, and the
streams quickly collapse to zero radius.
It is ironic that it is the increase of mass that brings mass ination to an end. Not only does mass not
drive mass ination, but as soon as mass begins to contribute signicantly to the gravitational force, it brings
mass ination to an end.
17.2 The far future?
The Penrose diagram of a Reissner-Nordstr om or Kerr-Newman black hole indicates that an observer who
passes through the outgoing inner horizon sees the entire future of the outside universe go by. In a sense,
this is why the outside universe appears innitely blueshifted.
This raises the question of whether what happens at the outgoing inner horizon of a real black hole indeed
depends on what happens in the far future. If it did, then the conclusions of previous sections, which are
based in part on the proposition that the accretion rate is approximately constant, would be suspect. A
lot can happen in the far future, such as black hole mergers, the Universe ending in a big crunch, Hawking
evaporation, or something else beyond our current ken.
Ingoing and outgoing observers both see each other highly blueshifted near the inner horizon. An outgoing
observer sees ingoing observers from the future, while an ingoing observer sees outgoing observers from the
past. If the streaming 4-velocity between ingoing and outgoing streams is u
m
ba
, equation (??), then the proper
time d
b
that elapses on stream b observed by stream a equals the blueshift factor u
t
ba
times the proper time
d
a
experienced by stream a,
d
b
= u
t
ba
d
a
. (17.1)
17.2.1 Inationary phase
A physically relevant timescale for the observing stream a is how long it takes for the blueshift to increase
by one e-fold, which is d
a
/d ln u
t
ba
. During the inationary phase, stream a sees the amount of time elapsed
on stream b through one e-fold of blueshift to be
u
t
ba
d
a
d ln u
t
ba
=
2C
b
r
. (17.2)
The right hand side of equation (17.2) is derived from u
t
ba
2[u
a
u
b
[, [dr
a
/d
a
[ = [
a,t
[ u
a
, r
a
r
,
and the approximations (??) and (??) valid during the inationary phase. The constants C
b
and on the
right hand side of equation (17.2) are typically of order unity, while r
is the radius of the inner horizon

where mass ination takes place. Thus the right hand side of equation (17.2) is of the order of one black hole
crossing time. In other words, stream a sees approximately one black hole crossing time elapse on stream b
for each e-fold of blueshift.
For astronomically realistic black holes, exponentiating the Weyl curvature up to the Planck scale will
take typically a few hundred e-folds of blueshift, as illustrated for example in Figure 17.10. Thus what
happens at the inner horizon of a realistic black hole before quantum gravity intervenes depends only on the
immediate past and future of the black hole a few hundred black hole crossing times not on the distant
future or past. This conclusion holds even if the accretion rate of one of the ingoing or outgoing streams is
tiny compared to the other, as considered in ??.
From a streams own point of view, the entire inationary episode goes by in a ash. At the onset of
ination, where goes through its minimum, at
a
u
2
a
=
b
u
2
b
= according to equation (??), the blueshift
u
t
ba
is already large
u
t
ba
=
2
b
, (17.3)
thanks to the small accretion rates
a
and
b
. During the rst e-fold of blueshift, each stream experiences a
proper time of order
b
times the black hole crossing time, which is tiny. Subsequent e-folds of blueshift
race by in proportionately shorter proper times.
17.2.2 Collapse phase
The time to reach the collapse phase is another matter. According to the estimates in ?? and ??, reaching
the collapse phase takes of order 1/
a
e-folds of blueshift, where
a
is the larger of the accretion rates
of the ingoing and outgoing streams, equation (??). Thus in reaching the collapse phase, each stream has
seen approximately 1/
a
black hole crossing times elapse on the other stream. But 1/
a
black hole crossing
times is just the accretion time essentially, how long the black hole has existed. This timescale, the age of
the black hole, is not innite, but it can hardly be expected that the accretion rate of a black hole would be
constant over its lifetime.
If the accretion rate were in fact constant, and if quantum gravity did not intervene and the streams
remained non-interacting, then indeed streams inside the black hole would reach the collapse phase, where-
upon they would plunge to a spacelike singularity at zero radius. For example, in the self-similar models
illustrated in Figures ?? and 17.10, outgoing baryons hitting the central singularity see ingoing dark matter
accreted a factor of two into the future (specically, for ingoing baryons and outgoing dark matter hitting the
singularity, the radius of the outer horizon when the dark matter is accreted is twice that when the baryons
were accreted; the numbers are 2.11 for

M
= 0.03, 1.97 for

M
= 0.01, and unknown for

M
= 0.003 because
in that case the numerics overow before the central singularity is reached). The same conclusion applies to
the ultra-hard uid models illustrated in Figure 17.8: if the accretion rate is constant, then outgoing streams
see only a factor of order unity or a few into the future before hitting the central singularity.
If on the other hard the accretion rate decreases suciently rapidly with time, then it is possible that
an outgoing stream never reaches the collapse phase, because the number of e-folds to reach the collapse
phase just keeps increasing as the accretion rate decreases. By contrast, an ingoing stream is always liable to
reach the collapse phase, if quantum gravity does not intervene, because an ingoing stream sees the outgoing
stream from the past, when the accretion rate was liable to have been larger.
However, as already remarked in ??, in astronomically realistic black holes, it is only for large accretion
rates, such as may occur when the black hole rst collapses (

M
0.01 for the models illustrated in

Fig. 17.10), that the collapse phase has a chance of being reached before the Weyl curvature exceeds the
Planck scale.
To summarize, it is only streams accreted during the rst few hundred or so black hole crossing times
since a black holes formation that have a chance of hitting a central spacelike singularity. Streams accreted
at later times, whether ingoing or outgoing, are likely to meet their fate in the inationary zone at the inner
horizon, where the Weyl curvature exponentiates to the Planck scale and beyond.
17.3 Self-similar models of the interior structure of black holes
The apparatus is now in hand actually to do some real calculations of the interior structure of black holes.
All the models presented in this section are spherical and self-similar. See Hamilton & Pollack (2005, PRD
71, 084031 & 084031) and Wallace, Hamilton & Polhemus (2008, arXiv:0801.4415) for more.
17.3.1 Boundary conditions at an outer sonic point
Because information can propagate only inward inside the horizon of a black hole, it is natural to set the
boundary conditions outside the horizon. The policy adopted here is to set them at a sonic point, where
the infalling uid accelerates from subsonic to supersonic. The proper 3-velocity of the uid through the
self-similar frame is
x
/
, equation (16.78) (the velocity

x
/
is positive falling inward), and the sound

speed is
sound speed =
_
p
b
b
=
w
b
, (17.4)
and sonic points occur where the velocity equals the sound speed
w
b
at sonic points . (17.5)
The denominator of the expression (16.120) for the proper acceleration g is zero at sonic points, indicating
that the acceleration will diverge unless the numerator is also zero. What happens at a sonic point depends
on whether the uid transitions from subsonic upstream to supersonic downstream (as here) or vice versa.
If (as here) the uid transitions from subsonic to supersonic, then sound waves generated by discontinuities
near the sonic point can propagate upstream, plausibly modifying the ow so as to ensure a smooth transition
through the sonic point, eectively forcing the numerator, like the denominator, of the expression (16.120)
to pass through zero at the sonic point. Conversely, if the uid transitions from supersonic to subsonic, then
sound waves cannot propagate upstream to warn the incoming uid that a divergent acceleration is coming,
and the result is a shock wave, where the uid accelerates discontinuously, is heated, and thereby passes
from supersonic to subsonic.
The solutions considered here assume that the acceleration g at the sonic point is not only continuous [so
the numerator of (16.120) is zero] but also dierentiable. Such a sonic point is said to be regular, and the
assumption imposes two boundary conditions at the sonic point.
The accretion in real black holes is likely to be much more complicated, but the assumption of a regular
sonic point is the simplest physically reasonable one.
17.3.2 Mass and charge of the black hole
The mass M
and charge Q
of the black hole at any instant can be dened to be those that would be
measured by a distant observer if there were no mass or charge outside the sonic point
M
= M +
Q
2
2r
, Q
= Q at the sonic point . (17.6)

The mass M
in equation (17.6) includes the mass-energy Q

2
/2r that would be in the electric eld outside
the sonic point if there were no charge outside the sonic point, but it does not include mass-energy from any
additional mass or charge that might be outside the sonic point.
In self-similar evolution, the black hole mass increases linearly with time, M
t, where t is the proper

time at rest far from the black hole. As discussed in ??, this time t equals the proper time
d
= r
d
/v
recorded by dark matter clocks that free-fall radially from zero velocity at innity. Thus the mass accretion
rate

M
is

dM
dt
=
M
d
=
vM
d
at the sonic point . (17.7)
If there is no mass outside the sonic point (apart from the mass-energy in the electric eld), then a freely-
falling dark matter particle will have
d,x
= 1 at the sonic point , (17.8)
which can be taken as the boundary condition on the dark matter at the sonic point, for either massive or
massless dark matter. Equation (17.8) follows from the facts that the geodesic equations in empty space
around a charged black hole (Reissner-Nordstrom metric) imply that
d,x
= constant for a radially free-
falling particle (the same conclusion can drawn from the Einstein equation (16.26c)), and that a particle at
rest at innity satises
d,
=
d,
r = 0, and consequently
d,x
= 1 from equation (16.69) with r .
As remarked following equation (16.72), the residual gauge freedom in the global rescaling of conformal
time allows the expansion velocity v to be adjusted at will. One choice suggested by equation (17.7) is to
set (but one could equally well set v to the expansion velocity of the horizon, v = r
+
, for example)
v =

M
, (17.9)
which is equivalent to setting
d
=
M
r
at the sonic point . (17.10)
Equation (17.10) is not a boundary condition: it is just a choice of units of conformal time . Equation (17.10)
and the boundary condition (17.8) coupled with the scalar relations (16.69) and (16.70) fully determine the
dark matter 4-vectors
d,m
and
m
d
at the sonic point.
17.3.3 Equation of state
The density
b
and temperature T
b
of an ideal relativistic baryonic uid in thermodynamic equilibrium are
related by
b
=

2
g
30
T
4
b
, (17.11)
where
g = g
B
+
7
8
g
F
(17.12)
is the eective number of relativistic particle species, with g
B
and g
F
being the number of bosonic and
fermionic species. If the expected increase in g with temperature T is modeled (so as not to spoil self-
similarity) as a weak power law g/g
p
= T
, with g
p
the eective number of relativistic species at the Planck
temperature, then the relation between density
b
and temperature T
b
is
b
=

2
g
p
30
T
(1+w)/w
b
, (17.13)
with equation of state parameter w
b
= 1/(3 + ) slightly less than the standard relativistic value w = 1/3.
In the models considered here, the baryonic equation of state is taken to be
w
b
= 0.32 . (17.14)
The eective number g
p
is xed by setting the number of relativistic particles species to g = 5.5 at T =
10 MeV, corresponding to a plasma of relativistic photons, electrons, and positrons. This corresponds to
choosing the eective number of relativistic species at the Planck temperature to be g
p
2,400, which is
not unreasonable.
The chemical potential of the relativistic baryonic uid is likely to be close to zero, corresponding to equal
numbers of particles and anti-particles. The entropy S
b
of a proper Lagrangian volume element V of the
uid is then
S
b
=
(
b
+p
b
)V
T
b
, (17.15)
which agrees with the earlier expression (16.122), but now has the correct normalization.
17.3.4 Entropy creation
One fundamentally interesting question about black hole interiors is how much entropy might be created
inside the horizon. Bekenstein rst argued that a black hole should have a quantum entropy proportional to
its horizon area A, and Hawking (1974) supplied the constant of proportionality 1/4 in Planck units. The
Bekenstein-Hawking entropy S
BH
is, in Planck units c = G = = 1,
S
BH
=
A
4
. (17.16)
For a spherical black hole of horizon radius r
+
, the area is A = 4r
2
+
. Hawking showed that a black hole
has a temperature T
H
equal to 1/(2) times the surface gravity g
+
at its horizon, again in Planck units,
T
H
=
g
+
2
. (17.17)
The surface gravity is dened to be the proper radial acceleration measured by a person in free-fall at the
horizon. For a spherical black hole, the surface gravity is g
+
= D
t
t
= M/r
2
+ 4rp evaluated at the
horizon, equation (16.52a).
The proper velocity of the baryonic uid through the sonic point equals
x
/
, equation (16.78). Thus

the entropy S
b
accreted through the sonic point per unit proper time of the uid is
dS
b
d
=
(1 +w
b
)
b
T
b
4r
2
. (17.18)
The horizon radius r
+
, which is at xed conformal radius x, expands in proportion to the conformal factor,
r
+
a, and the conformal factor a increases as d ln a/d =
ln a = v/(r
), so the Bekenstein-Hawking
entropy S
BH
= r
2
+
increases as
dS
BH
d
=
2r
2
+
v
r
. (17.19)
Thus the entropy S
b
accreted through the sonic point per unit increase of the Bekenstein-Hawking entropy
S
BH
is
dS
b
dS
BH
=
(1 +w
b
)
b
4r
3
x
2r
2
+
v T
b
r=rs
. (17.20)
Inside the sonic point, dissipation increases the entropy according to equation (16.121). Since the entropy
can diverge at a central singularity where the density diverges, and quantum gravity presumably intervenes
at some point, it makes sense to truncate the production of entropy at a splat point where the density
b
hits a prescribed splat density
#
b
=
#
. (17.21)
Integrating equation (16.121) from the sonic point to the splat point yields the ratio of the entropies at the
sonic and splat points. Multiplying the accreted entropy, equation (17.20), by this ratio yields the rate of
increase of the entropy of the black hole, truncated at the splat point, per unit increase of its Bekenstein-
Hawking entropy
dS
b
dS
BH
=
(1 + w
b
)
b
4r
3
x
2r
2
+
v T
b
b
=
#
. (17.22)
17.3.5 Holography
The idea that the entropy of a black hole cannot exceed its Bekenstein-Hawking entropy has motivated
holographic conjectures that the degrees of freedom of a volume are somehow encoded on its boundary,
and consequently that the entropy of a volume is bounded by those degrees of freedom. Various counter-
examples dispose of most simple-minded versions of holographic entropy bounds. The most successful entropy
bound, with no known counter-examples, is Boussos covariant entropy bound (Bousso 2002, Rev. Mod.
Phys. 74, 825). The covariant entropy bound concerns not just any old 3-dimensional volume, but rather the
3-dimensional volume formed by a null hypersurface, a lightsheet. For example, the horizon of a black hole is
a null hypersurface, a lightsheet. The covariant entropy bound asserts that the entropy that passes (inward
or outward) through a lightsheet that is everywhere converging cannot exceed 1/4 of the 2-dimensional area
of the boundary of the lightsheet.
In the self-similar black holes under consideration, the horizon is expanding, and outgoing lightrays that
sit on the horizon do not constitute a converging lightsheet. However, a spherical shell of ingoing lightrays
that starts on the horizon falls inwards and therefore does form a converging lightsheet, and a spherical shell
of outgoing lightrays that starts just slightly inside the horizon also falls inward and forms a converging
lightsheet. The rate at which entropy S
b
passes through such ingoing or outgoing spherical lightsheets per
unit decrease in the area S
cov
r
2
of the lightsheet is
dS
b
dS
cov
=
dS
b
dS
BH
r
2
+
r
2
v
x
[
x
[
, (17.23)
in which the sign is for ingoing, + for outgoing lightsheets. A sucient condition for Boussos covariant
entropy bound to be satised is
[dS
b
/dS
cov
[ 1 . (17.24)
17.3.6 Black hole accreting a neutral relativistic plasma
The simplest case to consider is that of a black hole accreting a neutral relativistic plasma. In the self-similar
solutions, the charge of the black hole is produced self-consistently by the accreted charge of the baryonic
uid, so a neutral uid produces an uncharged black hole.
Figure 17.2 shows the baryonic density
b
and Weyl curvature C inside the uncharged black hole. The
mass and accretion rate have been taken to be
M
= 4 10
6
M
,

M
= 10
16
, (17.25)
h
o
r
i
z
o
n
P
l
a
n
c
k
s
c
a
l
e
b
C
dS
b
/
dS
BH
10
10
10
20
10
30
10
40
10
50
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
Radius r (Planck units)
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.2 An uncharged baryonic plasma falls into an uncharged spherical black hole. The plot shows in
Planck units, as a function of radius, the plasma density
b
, the Weyl curvature scalar C (which is negative),
and the rate dS
b
/dSBH of increase of the plasma entropy per unit increase in the Bekenstein-Hawking entropy
of the black hole. The mass is M = 410
6
M, the accretion rate is

M = 10
16
, and the equation of state
is w
b
= 0.32.
which are motivated by the fact that the mass of the supermassive black hole at the center of the Milky Way
is 4 10
6
M
, and its accretion rate is

Mass of MW black hole
age of Universe

4 10
6
M
10
10
yr

6 10
60
Planck units
4 10
44
Planck units
10
16
. (17.26)
Figure 17.2 shows that the baryonic plasma plunges uneventfully to a central singularity, just as in the
Schwarzschild solution. The Weyl curvature scalar hits the Planck scale, [C[ = 1, while the baryonic proper
density
b
is still well below the Planck density, so this singularity is curvature-dominated.
17.3.7 Black hole accreting a non-conducting charged relativistic plasma
The next simplest case is that of a black hole accreting a charged but non-conducting relativistic plasma.
Figure 17.3 shows a black hole with charge-to-mass Q
/M
= 10
5
, but otherwise the same parameters
as in the uncharged black hole of 17.3.6: M
= 4 10
6
M
,

M
= 10
16
, and w
b
= 0.32. Inside the outer
horizon, the baryonic plasma, repelled by the electric charge of the black hole self-consistently generated by
the accretion of the charged baryons, becomes outgoing. Like the Reissner-Nordstr om geometry, the black
hole has an (outgoing) inner horizon. The baryons drop through the inner horizon, shortly after which the
self-similar solution terminates at an irregular sonic point, where the proper acceleration diverges. Normally
h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
e
|C|
dS
b
/
dS
BH
10
10
10
20
10
30
10
40
10
50
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.3 A plasma that is charged but non-conducting. The black hole has an inner horizon like the
Reissner-Nordstrom geometry. The self-similar solution terminates at an irregular sonic point just beneath
the inner horizon. The mass is M = 4 10
6
M, accretion rate

M = 10
16
, equation of state w
b
= 0.32,
and black hole charge-to-mass Q/M = 10
5
.
this is a signal that a shock must form, but even if a shock is introduced, the plasma still terminates at an
irregular sonic point shortly downstream of the shock. The failure of the self-similar to continue does not
invalidate the solution, because the failure is hidden beneath the inner horizon, and cannot be communicated
to infalling matter above it.
The solution is nevertheless not realistic, because it assumes that there is no ingoing matter, such as
would inevitably be produced for example by infalling neutral dark matter. Such ingoing matter would
appear innitely blueshifted to the outgoing baryons falling through the inner horizon, which would produce
mass ination, as in 17.3.10.
17.3.8 Black hole accreting a conducting relativistic plasma
What happens if the baryonic plasma is not only electrically charged but also electrically conducting? If the
conductivity is small, then the solutions resemble the non-conducting solutions of the previous subsection,
17.3.7. But if the conductivity is large enough eectively to neutralize the plasma as it approaches the
center, then the plasma can plunge all the way to the central singularity, as in the uncharged case in 17.3.6.
Figure 17.4 shows a case in which the conductivity has been tuned to equal, within numerical accuracy, the
critical conductivity
b
= 0.35 above which the plasma collapses to a central singularity. The parameters are
h
o
r
i
z
o
n
P
l
a
n
c
k
s
c
a
l
e
e
|C|
|dS
b
/
dS
cov
|
dS
b
/
dS
BH
10
10
10
20
10
30
10
40
10
50
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.4 Here the baryonic plasma is charged, and electrically conducting. The conductivity is at (within
numerical accuracy) the threshold above which the plasma plunges to a central singularity. The mass is
M = 4 10
6
M, the accretion rate

M = 10
16
, the equation of state w
b
= 0.32, the charge-to-mass
Q/M = 10
5
, and the conductivity parameter
b
= 0.35. Arrows show how quantities vary a factor of 10
into the past and future.
otherwise the same as in previous subsections: a mass of M
= 4 10
6
M
, an accretion rate

M
= 10
16
,
an equation of state w
b
= 0.32, and a black hole charge-to-mass of Q
/M
= 10
5
.
The solution at the critical conductivity exhibits the periodic self-similar behavior rst discovered in
numerical simulations by Choptuik (1993, PRL 70, 9), and known as critical collapse because it happens
at the borderline between solutions that do and do not collapse to a black hole. The ringing of curves in
Figure 17.4 is a manifestation of the self-similar periodicity, not a numerical error.
These solutions are not subject to the mass ination instability, and they could therefore be prototypical
of the behavior inside realistic rotating black holes. For this to work, the outward transport of angular
momentum inside a rotating black hole must be large enough eectively to produce zero angular momentum
at the center. My instinct is that angular momentum transport is probably not strong enough, but I do not
know this for sure. If angular momentum transport is not strong enough, then mass ination will take place.
Figure 17.4 shows that the entropy produced by Ohmic dissipation inside the black hole can potentially
exceed the Bekenstein-Hawking entropy of the black hole by a large factor. The Figure shows the rate
dS
b
/dS
BH
of increase of entropy per unit increase in its Bekenstein-Hawking entropy, as a function of the
hypothetical splat point above which entropy production is truncated. The rate is almost independent of
the black hole mass M
at xed splat density

#
, so it is legitimate to interpret dS
b
/dS
BH
as the cumulative
h
o
r
i
z
o
n
P
l
a
n
c
k
s
c
a
l
e
e
|C|
|dS
b
/
dS
cov
|
dS
b
/
dS
BH
10
10
10
20
10
30
10
40
10
50
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.5 This black hole creates a lot of entropy by having a large charge-to-mass Q/M = 0.8 and a
low accretion rate

M = 10
28
. The conductivity parameter
b
= 0.35 is again at the threshold above which
the plasma plunges to a central singularity. The equation of state is w
b
= 0.32.
entropy created inside the black hole relative to the Bekenstein-Hawking entropy. Truncated at the Planck
scale, [C[ = 1, the entropy relative to Bekenstein-Hawking is dS
b
/dS
BH
10
10
.
Generally, the smaller the accretion rate

M
, the more entropy is produced. If moreover the charge-to-

mass Q
/M
is large, then the entropy can be produced closer to the outer horizon. Figure 17.5 shows a
model with a relatively large charge-to-mass Q
/M
= 0.8, and a low accretion rate

M
= 10
28
. The large
charge-to-mass ratio in spite of the relatively high conductivity requires force-feeding the black hole: the
sonic point must be pushed to just above the horizon. The large charge and high conductivity leads to a
burst of entropy production just beneath the horizon.
If the entropy created inside a black hole exceeds the Bekenstein-Hawking entropy, and the black hole later
evaporates radiating only the Bekenstein-Hawking entropy, then entropy is destroyed, violating the second
law of thermodynamics.
This startling conclusion is premised on the assumption that entropy created inside a black hole accumu-
lates additively, which in turn derives from the assumption that the Hilbert space of states is multiplicative
over spacelike-separated regions. This assumption, called locality, derives from the fundamental proposi-
tion of quantum eld theory in at space that eld operators at spacelike-separated points commute. This
reasoning is essentially the same as originally led Hawking (1976) to conclude that black holes must destroy
information.
The same ideas that motivate holography also rescue the second law. If the future lightcones of spacelike-
separated points do not intersect, then the points are permanently out of communication, and can behave
matter
Infalling
S
b
>> S
BH
S
b
= S
BH
S
b
<
S
B
H
Figure 17.6 Partial Penrose diagram of the black hole. The entropy passing through the spacelike slice before
the black hole evaporates exceeds that passing through the spacelike slice after the black hole evaporates,
apparently violating the second law of thermodynamics. However, the entropy passing through any null slice
respects the second law.
like alternate quantum realities, like Schrodingers dead-and-alive quantum cat. Just as it is not legitimate
to the add the entropies of the dead cat and the live cat, so also it is not legitimate to add the entropies of
regions inside a black hole whose future lightcones do not intersect. The states of such separated regions,
instead of being distinct, are quantum entangled with each other.
Figures 17.4 and 17.5 show that the rate [dS
b
/dS
cov
[ at which entropy passes through ingoing or outgoing
spherical lightsheets is less than one at all scales below the Planck scale. This shows not only that the black
holes obey Boussos covariant entropy bound, but also that no individual observer inside the black hole sees
more than the Bekenstein-Hawking entropy on their lightcone. No observer actually witnesses a violation of
the second law.
17.3.9 Black hole accreting a charged massless scalar eld
The charged, non-conducting plasma considered in 17.3.7 fell through an (outgoing) inner horizon without
undergoing mass ination. This can be attributed to the fact that relativistic counter-streaming could not
occur: there was only a single (outgoing) uid, and the speed of sound in the uid was less than the speed
of light.
In reality, unless dissipation destroys the inner horizon as in 17.3.8, then relativistic counter-streaming
between ingoing and outgoing uids will undoubtedly take place, through gravitational waves if nothing else.
One way to allow relativistic counter-streaming is to let the speed of sound be the speed of light. This is
true in a massless scalar (= spin-0) eld , which has an equation of state w
= 1. Figure 17.7 shows a black

hole that accretes a charged, non-conducting uid with this equation of state. The parameters are otherwise
the same as as in previous subsections: a mass of M
= 4 10
6
M
, an accretion rate of

M
= 10
16
, and
a black hole charge-to-mass of Q
/M
= 10
5
. As the Figure shows, mass ination takes place just above
h
o
r
i
z
o
n m
a
s
s
i
n
a
t
i
o
n
e
|C|
10
10
10
20
10
30
10
40
10
50
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.7 Instead of a relativistic plasma, this shows a charged scalar eld whose equation of state
w
= 1 means that the speed of sound equals the speed of light. The scalar eld therefore supports relativistic
counter-streaming, as a result of which mass ination occurs just above the erstwhile inner horizon. The
mass is M = 4 10
6

M = 10
16
, the charge-to-mass Q/M = 10
5
, and the
conductivity is zero.
the place where the inner horizon would be. During mass ination, the density
and the Weyl scalar C

rapidly exponentiate up to the Planck scale and beyond.
One of the remarkable features of the mass ination instability is that the smaller the accretion rate, the
more violent the instability. Figure 17.8 shows mass ination in a black hole of charge-to-mass Q
/M
= 0.8
accreting a massless scalar eld at rates

M
= 0.01, 0.003, and 0.001. The charge-to-mass has been chosen

to be largish so that the inner horizon is not too far below the outer horizon, and the accretion rates have
been chosen to be large because otherwise the inationary growth rate is too rapid to be discerned easily on
the graph. The density
and Weyl scalar C exponentiate along with, and in proportion to, the interior
mass M, which increases as the radius r decreases as
M exp(ln r/

M
) . (17.27)
Physically, the scale of length of ination is set by how close to the inner horizon infalling material approaches
before mass ination begins. The smaller the accretion rate, the closer the approach, and consequently the
shorter the length scale of ination.
h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
0
.
0
1 h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
0
.
0
0
3
h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
0
.
0
0
1
e
|C|
.1 .2 .5 1 2
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
Radius r (geometric units)
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.8 The density
and Weyl curvature scalar |C| inside a black hole accreting a massless scalar eld.
The graph shows three cases, with mass accretion rates

M = 0.01, 0.003, and 0.001. The graph illustrates
that the smaller the accretion rate, the faster the density and curvature inate. Mass ination destroys the
inner horizon: the dashed vertical line labeled inner horizon shows the position that the inner horizon
would have if mass ination did not occur. The black hole mass is M = 4 10
6
M, the charge-to-mass is
Q/M = 0.8, and the conductivity is zero.
17.3.10 Black hole accreting charged baryons and dark matter
No scalar eld (massless or otherwise) has yet been observed in nature, although it is supposed that the
Higgs eld is a scalar eld, and it is likely that cosmological ination was driven by a scalar eld. Another
way to allow mass ination in simple models is to admit not one but two uids that can counter-stream
relativistically through each other. A natural possibility is to feed the black hole not only with a charged
relativistic uid of baryons but also with neutral pressureless dark matter that streams freely through the
baryons. The charged baryons, being repelled by the electric charge of the black hole, become outgoing,
while the neutral dark matter remains ingoing.
Figure 17.9 shows that relativistic counter-streaming between the baryons and the dark matter causes
the center-of-mass density and the Weyl curvature scalar C to inate quickly up to the Planck scale and
beyond. The ratio of dark matter to baryonic density at the sonic point is
d
/
b
= 0.1, but otherwise the
parameters are the generic parameters of previous subsections: M
= 4 10
6
M
,

M
= 10
16
, w
b
= 0.32,
Q
/M
= 10
5
, and zero conductivity. Almost all the center-of-mass energy is in the counter-streaming
energy between the outgoing baryonic and ingoing dark matter. The individual densities
b
of baryons and
d
of dark matter (and
e
of electromagnetic energy) increase only modestly.
As in the case of the massless scalar eld considered in the previous subsection, 17.3.9, the smaller the
h
o
r
i
z
o
n m
a
s
s
i
n
a
t
i
o
n
e
|C|
dS
b
/
dS
BH
10
10
10
20
10
30
10
40
10
50
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.9 Back to the relativistic charged baryonic plasma, but now in addition the black hole accretes
neutral pressureless uncharged dark matter, which streams freely through the baryonic plasma. The rel-
ativistic counter-streaming produces mass ination just above the erstwhile inner horizon. The mass is
M = 4 10
6

M = 10
16
, the baryonic equation of state w
b
= 0.32, the charge-to-
mass Q/M = 10
5
, the conductivity is zero, and the ratio of dark matter to baryonic density at the outer
sonic point is
d
/
b
= 0.1.
accretion rate, the shorter the length scale of ination. Not only that, but the smaller one of the ingoing or
outgoing streams is relative to the other, the shorter the length scale of ination. Figure 17.10 shows a black
hole with three dierent ratios of the dark-matter-to-baryon density ratio at the sonic point,
d
/
b
= 0.3,
0.1, and 0.03, all with the same total accretion rate

M
= 10
2
. The smaller the dark matter stream, the
faster is ination. The accretion rate

M
and the dark-matter-to-baryon ratio

d
/
b
have been chosen to be
relatively large so that the inationary growth rate is discernable easily on the graph.
Figure 17.10 shows that, as in Figure 17.9, almost all the center-of-mass energy is in the streaming energy
between the baryons and the dark matter. For one case,
d
/
b
= 0.3, Figure 17.10 shows the individual
densities
b
of baryons,
d
of dark matter, and
e
of electromagnetic energy, all of which remain tiny compared
to the streaming energy.
17.3.11 The black hole particle accelerator
The previous subsection, 17.3.10, showed that almost all the center-of-mass energy during mass ination
is in the energy of counter-streaming. Thus the black hole acts like an extravagantly powerful particle
accelerator.
Mass ination is an exponential instability. The nature of the black hole particle accelerator is that an
h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
d

b
e |C|
0
.
3
h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
e |C|
0
.
1
h
o
r
i
z
o
n
i
n
n
e
r
h
o
r
i
z
o
n
e |C|
0
.
0
3
.1 .2 .5 1 2
10
120
10
110
10
100
10
90
10
80
10
70
10
60
10
50
10
40
10
30
10
20
10
10
10
0
10
10
10
20
Radius r (geometric units)
(
P
l
a
n
c
k
u
n
i
t
s
)
Figure 17.10 The center-of-mass density and Weyl curvature |C| inside a black hole accreting baryons
and dark matter at rate

M = 0.01. The graph shows three cases, with dark-matter-to-baryon ratio at the
sonic point of
d
/
b
= 0.3, 0.1, and 0.03. The smaller the ratio of dark matter to baryons, the faster the
center-of-mass density and curvature C inate. For the largest ratio,
d
/
b
= 0.3 (to avoid confusion,
only this case is plotted), the graph also shows the individual proper densities
b
of baryons,
d
of dark
matter, and e of electromagnetic energy. During mass ination, almost all the center-of-mass energy is
in the streaming energy: the proper densities of individual components remain small. The black hole mass
is M = 4 10
6
M, the baryonic equation of state is w
b
= 0.32, the charge-to-mass is Q/M = 0.8, and
the conductivity is zero.
individual particle spends approximately an equal interval of proper time being accelerated through each
decade of collision energy.
Each baryon in the black hole collider sees a ux n
d
u
r
of dark matter particles per unit area per unit
time, where n
d
=
d
/m
d
is the proper number density of dark matter particles in their own frame, and u
r
is
the radial component of the proper 4-velocity, the v, of the dark matter through the baryons. The factor
in u
r
is the relavistic beaming factor: all frequencies, including the collision frequency, are speeded up by
the relativistic beaming factor . As the baryons accelerate through the collider, they spend a proper time
interval d/d ln u
r
in each e-fold of Lorentz factor u
r
. The number of collisions per baryon per e-fold of u
r
is
the dark matter ux (
d
/m
d
)u
r
, multiplied by the time d/d ln u
r
, multiplied by the collision cross-section
. The total cumulative number of collisions that have happened in the black hole particle collider equals
this multiplied by the total number of baryons that have fallen into the black hole, which is approximately
equal to the black hole mass M
divided by the mass m

b
per baryon. Thus the total cumulative number of
10
0
10
50
10
100
10
2
10
1
1
10
Velocity u
C
o
l
l
i
s
i
o
n
r
a
t
e
d
u
d
/
d
l
n
u
(
M
/
M
)
0.03
0.01
0.003
10
16
Figure 17.11 Collision rate of the black hole particle accelerator per e-fold of velocity u (meaning v),
expressed in units of the inverse black hole accretion time

M/M. The models illustrated are the same as
those in Figure 17.10. The curves are labeled with their mass accretion rates:

M = 0.03, 0.01, 0.003, and
10
16
. Stars mark where the center-of-mass energy of colliding baryons and dark matter particles exceeds
the Planck energy, while disks show where the Weyl curvature scalar C exceeds the Planck scale.
collisions in the black hole collider is
number of collisions
e-fold of u
r
=
M
m
b
d
m
d
u
r
d
d ln u
r
. (17.28)
Figure 17.11 shows, for several dierent accretion rates

M
, the collision rate M
d
u
r
d/d ln u
r
of the black
hole collider, expressed in units of the black hole accretion rate

M
. This collision rate, multiplied by
/(m
d
m
b
), gives the number of collisions (17.28) in the black hole. In the units c = G = 1 being used
here, the mass of a baryon (proton) is 1 GeV 10
54
m. If the cross-section is expressed in canonical
accelerator units of femtobarns (1 fb = 10
43
m
2
) then the number of collisions (17.28) is
number of collisions
e-fold of u
r
= 10
45
_

1 fb
_
_
300 GeV
2
m
b
m
d
_
_

M
10
16
_
_
M
d
u
r
d/d ln u
r
0.03

M
_
. (17.29)
Particle accelerators measure their cumulative luminosities in inverse femtobarns. Equation (17.29) shows
that the black hole accelerator delivers about 10
45
femtobarns, and it does so in each e-fold of collision energy
up to the Planck energy and beyond.
17.4 Instability at outer horizon?
A number of papers have suggested that a magical phase transition at, or just outside, the outer horizon
prevents any horizon from forming. Is it true? For example, is there a mass ination instability at the outer
horizon?
If the were a White Hole on the other side of the outer horizon, then indeed an object entering the outer
horizon would encounter an inationary instability. But otherwise, no.
PART SEVEN
GENERAL RELATIVISTIC PERTURBATION THEORY
Concept Questions
1. Why do general relativistic perturbation theory using the tetrad formalism as opposed to the coordinate
approach?
2. Why is the tetrad metric
mn
assumed xed in the presence of perturbations?
3. Are the tetrad axes
m
xed under a perturbation?
4. Is it true that the tetrad components
mn
of a perturbation are (anti-)symmetric in m n if and only if
its coordinate components
are (anti-)symmetric in ?
5. Does an unperturbed quantity, such as the unperturbed metric
0
g
, change under an innitesimal coordi-

nate gauge transformation?
6. How can the vierbein perturbation
mn
be considered a tetrad tensor eld if it changes under an innites-
imal coordinate gauge transformations?
7. What properties of the unperturbed spacetime allow decomposition of perturbations into independently
evolving Fourier modes?
8. What properties of the unperturbed spacetime allow decomposition of perturbations into independently
evolving scalar, vector, and tensor modes?
9. In what sense do scalar, vector, and tensor modes have spin 0, 1, and 2 respectively?
10. Tensor modes represent gravitational waves that, in vacuo, propagate at the speed of light. Do scalar
and vector modes also propagate at the speed of light in vacuo? If so, do scalar and vector modes also
constitute gravitational waves?
11. If scalar, vector, and tensor modes evolve independently, does that mean that scalar modes can exist and
evolve in the complete absence of tensor modes? If so, does it mean that scalar modes can propagate
causally, in vacuo at the speed of light, without any tensor modes being present?
12. Equation (20.74) denes the mass M of a body as what a distant observer would measure from its
gravitational potential. Similarly equation (20.82) denes the angular momentum L of a body as what a
distant observer would measure from the dragging of inertial frames. In what sense are these denitions
legitimate?
13. Can an observer far from a body detect the dierence between the scalar potentials and produced by
the body?
14. If a gravitational wave is a wave of spacetime itself, distorting the very rulers and clocks that measure
spacetime, how is it possible to measure gravitational waves at all?
15. Have gravitational waves been detected?
16. If gravitational waves carry energy-momentum, then can gravitational waves be present in a region of
spacetime with vanishing energy-momentum tensor, T
mn
= 0?
Whats important?
1. Getting your brain around coordinate and tetrad gauge transformations.
2. A central aim of general relativistic perturbation theory is to identify the coordinate and tetrad gauge-
invariant perturbations, since only these have physical meaning.
3. A second central aim is to classify perturbations into independently evolving modes, to the extent that
this is possible.
4. In background spacetimes with spatial translation and rotation symmetry, which includes Minkowski space
and the Friedmann-Roberston-Walker metric of cosmology, modes decompose into independently evolving
scalar (spin-0), vector (spin-1), and tensor (spin-2) modes. In background spacetimes without spatial
translation and rotation symmetry, such as black holes, scalar, vector, and tensor modes scatter o the
curvature of space, and therefore mix with each other.
5. In background spacetimes with spatial translation and rotation symmetry, there are 6 algebraic combina-
tions of metric coecients that are coordinate and tetrad gauge-invariant, and therefore represent physical
perturbations. There are 2 scalar modes, 2 vector modes, and 2 tensor modes. A spin-m mode varies as
e
im
where is the rotational angle about the spatial wavevector k of the mode.
6. In background spacetimes without spatial translation and rotation symmetry, the coordinate and tetrad
gauge-invariant perturbations are not algebraic combinations of the metric coecients, but rather com-
binations that involve rst and second derivatives of the metric coecients. Gravitational waves are
described by the Weyl tensor, which can be decomposed into 5 complex components, with spin 0, 1,
and 2. The spin-2 components describe propagating gravitational waves, while the spin-0 and spin-1
components describe the non-propagating gravitational eld near a source.
7. The preeminent application of general relativistic perturbation theory is to cosmology. Coupled with
physics that is either well understood (such as photon-electron scattering) or straightforward to model
even without a deep understanding (such as the dynamical behavior of non-baryonic dark matter and
dark energy), the theory has yielded predictions that are in spectacular agreement with observations of
uctuations in the CMB and in the large scale distribution of galaxies and other tracers of the distribution
of matter in the Universe.
18
Perturbations and gauge transformations
This chapter sets up the basics equations that dene perturbations to an arbitrary spacetime in the tetrad
formalism of general relativity, and it examines the eect of tetrad and coordinate gauge transformations
on those perturbations. The perturbations are supposed to be small, in the sense that quantities quadratic
in the perturbations can be neglected. The formalism set up in this chapter provides a foundation used in
subsequent chapters.
18.1 Notation for perturbations
A 0 (zero) overscript signies an unperturbed quantity, while a 1 (one) overscript signies a perturbation.
No overscript means the full quantity, including both unperturbed and perturbed parts. An overscript is
attached only where necessary. Thus if the unperturbed part of a quantity is zero, then no overscript is
needed, and none is attached.
18.2 Vierbein perturbation
Let the vierbein perturbation
mn
be dened so that the perturbed vierbein is
e
m
= (
n
m
+
m
n
)
0
e
n
, (18.1)
with corresponding inverse
e
m
= (
m
n

n
m
)
0
e
n
. (18.2)
Since the perturbation
m
n
is already of linear order, to linear order its indices can be raised and lowered
with the unperturbed metric, and transformed between tetrad and coordinate frames with the unperturbed
vierbein. In practice it proves convenient to work with the covariant tetrad-frame components
mn
of the
vierbein pertubation
mn
=
nl
m
l
. (18.3)
The perturbation
mn
can be regarded as a tetrad tensor eld dened on the unperturbed background.
18.3 Gauge transformations
The vierbein perturbation
mn
has 16 degrees of freedom, but only 6 of these degrees of freedom correspond
to real physical perturbations, since 6 degrees of freedom are associated with arbitrary innitesimal changes
in the choice of tetrad, which is to say arbitrary innitesimal Lorentz transformations, and a further 4 degrees
of freedom are associated with arbitrary innitesimal changes in the coordinates.
In the context of perturbation theory, these innitesimal tetrad and coordinate transformations are called
gauge transformations. Real physical perturbations are perturbations that are gauge-invariant under
both tetrad and coordinate gauge transformations.
18.4 Tetrad metric assumed constant
In the tetrad formalism, tetrad axes
m
are introduced as locally inertial (or other physically motivated)
axes attached to an observer. The axes enable quantities to be projected into the frame of the observer.
In a spacetime bueted by perturbations, it is natural for an observer to cling to the rock provided by the
locally inertial (or other) axes, as opposed to allowing the axes to bend with the wind. For example, when
a gravitational wave goes by, the tidal compression and rarefaction causes the proper distance between two
freely falling test masses to oscillate. It is natural to choose the tetrad so that it continues to measure proper
times and distances in the perturbed spacetime.
In these notes on general relativistic perturbation theory, the tetrad metric will be taken to be constant
everywhere, and unchanged by a perturbation
mn
=
0
mn
= constant . (18.4)
For example, if the tetrad is orthonormal, then the tetrad metric is constant, the Minkowski metric
mn
.
However, the tetrad could also be some other tetrad for which the tetrad metric is constant, such as a spinor
tetrad (12.1.1), or a Newman-Penrose tetrad (12.2.1).
18.5 Perturbed coordinate metric
The perturbed coordinate metric is
g
=
mn
e
m
e
n
=
kl
(
k
m
m
k
)
0
e
m
(
l
n

n
l
)
0
e
n
=
0
g
) . (18.5)
318 Perturbations and gauge transformations
Thus the perturbation of the coordinate metric depends only on the symmetric part of the vierbein pertur-
bation
mn
, not the antisymmetric part
1
g
= (
) . (18.6)
18.6 Tetrad gauge transformations
Under an innitesimal tetrad transformation, the covariant vierbein perturbations
mn
transform as
mn

mn
+
mn
, (18.7)
where
mn
is the generator of a Lorentz transformation, which is to say an arbitrary antisymmetric tensor
(Exercise 11.2). Thus the antisymmetric part
mn

nm
of the covariant perturbation
mn
is arbitrarily
adjustable through an innitesimal tetrad transformation, while the symmetric part
mn
+
nm
is tetrad
gauge-invariant.
It is easy to see when a quantity is tetrad gauge-invariant: it is tetrad gauge-invariant if and only if it
depends only on the symmetric part of the vierbein perturbation, not on the antisymmetric part. Evidently
the perturbation (18.6) to the coordinate metric g
is tetrad gauge-invariant. This is as it should be, since

the coordinate metric g
is a coordinate-frame quantity, independent of the choice of tetrad frame.

If only tetrad gauge-invariant perturbations are physical, why not just discard tetrad perturbations (the
antisymmetric part of
mn
) altogether, and work only with the tetrad gauge-invariant part (the symmetric
part of
mn
)? The answer is that tetrad-frame quantities such as the tetrad-frame Einstein tensor do change
under tetrad gauge transformations (innitesimal Lorentz transformations of the tetrad). It is true that
the only physical perturbations of the Einstein tensor are those combinations of it that are tetrad gauge-
invariant. But in order to identify these tetrad gauge-invariant combinations, it is necessary to carry through
the dependence on the non-tetrad-gauge-invariant part, the antisymmetric part of
mn
.
Much of the professional literature on general relativistic perturbation theory works with the traditional
coordinate formalism, as opposed to the tetrad formalism. The term gauge-invariant then means coor-
dinate gauge-invariant, as opposed to both coordinate and tetrad gauge-invariant. This is ne as far as it
goes: the coordinate approach is perfectly able to identify physical perturbations versus gauge perturbations.
However, there still remains the problem of projecting the perturbations into the frame of an observer, so
ultimately the issue of perturbations of the observers frame, tetrad perturbations, must be faced.
Concept question 18.1 In perturbation theory, can tetrad gauge transformations be non-innitesimal?
18.7 Coordinate gauge transformations 319
18.7 Coordinate gauge transformations
A coordinate gauge transformation is a transformation of the coordinates x
by an innitesimal shift
= x
. (18.8)
You should not think of this as shifting the underlying spacetime around; rather, it is just a change of the
coordinate system, which leaves the underlying spacetime unchanged. Because the shift
is, like the vierbein

perturbations
mn
, already of linear order, its indices can be raised and lowered with the unperturbed metric,
and transformed between coordinate and tetrad frames with the unperturbed vierbein. Thus the shift
can be regarded as a vector eld dened on the unperturbed background. The tetrad components
m
of the
shift
are
m
=
0
e
m
. (18.9)
Physically, the tetrad-frame shift
m
is the shift measured in locally inertial coordinates
m
=
m
+
m
. (18.10)
18.8 Coordinate gauge transformation of a coordinate scalar
Under a coordinate transformation (18.8), a coordinate-frame scalar (x) remains unchanged
(x)
(x
) = (x) . (18.11)
Here the scalar
(x
) is evaluated at position x
, which is the same as the original physical position x since

all that has changed is the coordinates, not the physical position. However, in perturbation theory, quantities
are evaluated at coordinate position x, not x
. The value of at x is related to that at x
by
(x) =
(x
) =
(x
. (18.12)
Since
is a small quantity, and
diers from by a small quantity, the last term
/x
in equa-
tion (18.16) can be replaced by
/x
to linear order. Putting equations (18.11) and (18.12) together

shows that the scalar changes under a coordinate gauge transformation (18.8) as
(x)
(x) = (x)
. (18.13)
The transformation (18.13) can also be written
(x)
(x) = (x) +L
, (18.14)
where L
is the Lie derivative, 18.13.

18.9 Coordinate gauge transformation of a coordinate vector or tensor
A similar argument applies to coordinate vectors and tensors. Under a coordinate transformation (18.8), a
coordinate-frame 4-vector A
(x) transforms in the usual way as

A
(x) A
(x
) = A
(x)
x
= A
(x) +A
(x)
. (18.15)
As in the scalar case, the vector A
(x
) is evaluated at position x
, which is the same as the original physical

position x since all that has changed is the coordinates, not the physical position. Again, in perturbation
theory, quantities are evaluated at coordinate position x, not x
. The value of A
at x is related to that at
x
by
A
(x) = A
(x
) = A
(x
. (18.16)
The last term
/x
in equation (18.16) can be replaced by
/x
to linear order. Putting

equations (18.15) and (18.16) together shows that the 4-vector A
changes under a coordinate gauge trans-

formation (18.8) as
A
(x) A
(x) = A
(x) +A
. (18.17)
The transformation (18.17) can also be written
A
(x) A
(x) = A
(x) +L
, (18.18)
where L
is the Lie derivative, 18.13.

More generally, under a coordinate gauge transformation (18.8), a coordinate tensor A
...
...
transforms as,
equation (18.32),
A
...
...
(x) A
...
...
(x) = A
...
...
(x) +L
A
...
...
. (18.19)
18.10 Coordinate gauge transformation of a tetrad vector
A tetrad-frame 4-vector A
m
is a coordinate-invariant quantity, and therefore acts like a coordinate scalar,
equation (18.13), under a coordinate gauge transformation (18.8)
A
m
(x) A
m
(x) = A
m
(x)
A
m
x
= A
m
(x)
k
k
A
m
. (18.20)
The change
k
k
A
m
is a coordinate tensor (specically, a coordinate scalar), but it is not a tetrad tensor.
More generally, a tetrad-frame tensor A
kl...
mn...
transforms under a coordinate gauge transformation (18.8)
as
A
kl...
mn...
(x) A
kl...
mn...
(x) = A
kl...
mn...
(x)
a
a
A
kl...
mn...
. (18.21)
Again, the change
a
a
A
kl...
mn...
is a coordinate tensor (a coordinate scalar), but not a tetrad tensor.
18.11 Coordinate gauge transformation of the vierbein 321
18.11 Coordinate gauge transformation of the vierbein
The inverse vierbein e
m
equals the scalar product of the tetrad and coordinate axes, e

m
=
m
g
.
Therefore the transformation of the vierbein under a coordinate gauge transformation (18.8) follows from
the transformations of
m
and g
. The tetrad axes

m
transform in accordance with (18.20) as
m
=
m
m
=
m
+
k
m
nk
n
. (18.22)
The coordinate axes g
transform in accordance with (18.32) and (18.34) as (including torsion S
)
g
= g
+L
= g
= g
(g
) g
= g
n
_
D
m
n
+S
n
mk
k
_
0
e
m
, (18.23)
where the third line follows from the second because the axes g
are by denition covariantly constant,

D
= 0. It follows from (18.22) and (18.23) that the inverse vierbein e

m
transforms under an innitesimal

coordinate gauge transformation (18.8) as
e
m
e
m
=
m
g
= e
m
+
_
D
n
m
S
m
nk
k
+
m
nk
k
_
0
e
n
. (18.24)
From equation (18.24) and the denition (18.2) of the vierbein perturbations
mn
, it follows that the vierbein
perturbations transform under a coordinate gauge transformation (18.8) as
mn

mn
=
mn
+
m
n
(
knm
+
nmk
S
nmk
)
k
, (18.25)
in which
k
are the tetrad components of the coordinate shift, and
kmn
are tetrad connection coecients.
If torsion vanishes, S
nmk
= 0, as general relativity assumes, then the transfomation (18.25) of the vierbein
perturbations under a coordinate gauge transformation reduces to
mn

mn
=
mn
+
m
n
(
knm
+
nmk
)
k
. (18.26)
18.12 Coordinate gauge transformation of the metric
The tetrad metric
mn
transforms under an innitesimal coordinate gauge transformation (18.8) as
mn

mn
=
mn
(
mnk
+
nmk
)
k
=
mn
, (18.27)
where the last expression is true because the tetrad metric
mn
is being assumed constant, equation (18.4),
in which case
mnk
+
nmk
=
k
nm
= 0.
The coordinate metric g
transforms under an innitesimal coordinate gauge transformation (18.8) as

g
= g
+L
= g
(D
+D
) (S
+S
. (18.28)
18.13 Lie derivative
The change in the coordinate 4-vector A
on the right hand side of equation (18.17) is called the Lie

derivative of A
along the direction
, and it is designated by the operator L
. (18.29)
The Lie derivative has the important property of being a tensor, which is one reason that it merits a special
name. As its name suggests, the Lie derivative acts like a derivative: it is linear, and it satises the Leibniz
rule. Translating from ordinary partial derivatives to covariant derivatives yields the following expression
for the Lie derivative in covariant form
L
= A
+A
is a coordinate vector , (18.30)

where S
is the torsion. If torsion vanishes, as GR assumes, then the Lie derivative of a 4-vector is
L
= A
is a coordinate vector . (18.31)

More generally, under a coordinate gauge transformation (18.8), a coordinate tensor A
...
...
transforms as
A
...
...
(x) A
...
...
(x) = A
...
...
(x) +L
A
...
...
(18.32)
where the Lie derivative is dened by
L
A
...
...
A
...
...
+A
...
...
... A
...
...
A
...
...
A
...
...
x
. (18.33)
In covariant form, equation (18.33) is
L
A
...
...
= A
...
...
D
+A
...
...
D
... A
...
...
D
A
...
...
D
...
A
...
...
(18.34)
+
_
A
...
...
S
+A
...
...
S
... A
...
...
S
A
...
...
S
is a coordinate tensor ,
which in the case of vanishing torsion, as GR assumes, reduces to
L
A
...
...
= A
...
...
D
+A
...
...
D
... A
...
...
D
A
...
...
D
...
A
...
...
is a coordinate tensor .
(18.35)
19
Scalar, vector, tensor decomposition
In the particular case that the unperturbed spacetime is spatially homogeneous and isotropic, which includes
not only Minkoswki space but also the important case of the cosmological Friedmann-Robertson-Walker
metric, perturbations decompose into independently evolving scalar (spin-0), vector (spin-1), and tensor
(spin-2) modes.
Similarly to Fourier decomposition, decomposition into scalar, vector, and tensor modes is non-local, in
principle requiring knowledge of perturbation amplitudes simultaneously throughout all of space. In practical
problems however, an adequate decomposition is possible as long as the scales probed are suciently larger
than the wavelengths of the modes probed. Ultimately, the fact that an adequate decomposition is possible
is a consequence of the fact that gravitational uctuations in the real Universe appear to converge at the
cosmological horizon, so that what happens locally is largely independent of what is happening far away.
19.1 Decomposition of a vector in at 3D space
Theorem: In at 3-dimensional space, a 3-vector eld w(x) can be decomposed uniquely (subject to the
boundary condition that w vanishes suciently rapidly at innity) into a sum of scalar and vector parts
w = w
scalar
+ w
vector
. (19.1)
In this context, the term vector signies a 3-vector w
that is transverse, that is to say, it has vanishing

divergence,
w
= 0 . (19.2)
Here /x
i
/x
i
is the gradient in at 3D space. The scalar and vector parts are also known
as spin-0 and spin-1, or gradient and curl, or longitudinal and transverse. The scalar part w
contains 1
degree of freedom, while the vector part w
contains 2 degrees of freedom. Together they account for the 3

degrees of freedom of the vector w.
324 Scalar, vector, tensor decomposition
Proof: Take the divergence of equation (19.1)
w =
2
w
. (19.3)
The operator
2
on the right hand side of equation (19.3) is the 3D Laplacian. The solution of equation (19.3)
is
w
(x) =
_

w(x
)
[x
x[
d
3
x
4
. (19.4)
The solution (19.4) is valid subject to boundary conditions that the vector w vanish suciently rapidly at
innity. In cosmology, the required boundary conditions, which are set at the Big Bang, are apparently
satised because uctuations at the Big Bang were small. Equation (19.1) then immediately implies that
the vector part is w
= ww
.
19.2 Fourier version of the decomposition of a vector in at 3D space
When the background has some symmetry, it is natural to expand perturbations in eigenmodes of the
symmetry. If the background space is at, then it is translation symmetric. Eigenmodes of the translation
operator are Fourier modes.
A function a(x) in at 3D space and its Fourier transform a(k) are related by (the disposition of factors
of 2 in the following denition follows the convention most commonly adopted by cosmologists)
a(k) =
_
a(x)e
ikx
d
3
x , a(x) =
_
a(k)e
ikx
d
3
k
(2)
3
. (19.5)
You may not be familiar with the practice of using the same symbol a in both real and Fourier space; but
a is the same vector in Hilbert space, with components a
x
= a(x) in real space, and a
k
= a(k) in Fourier
space.
Taking the gradient in real space is equivalent to multiplying by ik in Fourier space
ik . (19.6)
Thus the decomposition (19.1) of the 3D vector w translates into Fourier space as
w = i kw
scalar
+ w
vector
, (19.7)
where the vector part w
satises
k w
= 0 . (19.8)
In other words, in Fourier space the scalar part w
of the vector w is the part parallel (longitudinal) to

the wavevector k, while the vector part w
is the part perpendicular (transverse) to the wavevector k.

19.3 Decomposition of a tensor in at 3D space 325
19.3 Decomposition of a tensor in at 3D space
Similarly, the 9 components of a 3 3 spatial matrix h
ij
can be decomposed into 3 scalars, 2 vectors, and 1
tensor:
h
ij
=
ij

scalar
+
i
j
h
scalar
+
ijk
h
scalar
+
i
h
j
vector
+
j
h
i
vector
+ h
T
ij
tensor
. (19.9)
In this context, the term tensor signies a 3 3 matrix h
T
ij
that is traceless, symmetric, and transverse:
h
T i
i
= 0 , h
T
ij
= h
T
ji
,
i
h
T
ij
= 0 . (19.10)
The transverse-traceless-symmetric matrix h
T
ij
has two degrees of freedom. The vector components h
i
and
h
i
are by denition transverse,
i
h
i
=
i
h
i
= 0 . (19.11)
The tildes on

h and

h
i
simply distinguish those symbols; the tildes have no other signicance. The trace of
the 3 3 matrix h
ij
is
h
i
i
= 3 +
2
h . (19.12)
Spinor decomposition.
20
Flat space background
General relativistic perturbation theory is simplest in the case that the unperturbed background space
is Minkowski space. In Cartesian coordinates x
t, x, y, z, the unperturbed coordinate metric is the

Minkowski metric
0
g
. (20.1)
In this chapter the tetrad is taken to be orthonormal, and aligned with the unperturbed coordinate axes, so
that the unperturbed vierbein is the unit matrix
0
e
m
m
. (20.2)
Let overdot denote partial dierentiation with respect to time t,
overdot

t
, (20.3)
and let denote the spatial gradient

x

i

x
i
. (20.4)
Sometimes it will also be convenient to use
m
to denote the 4-dimensional spacetime derivative
m

_

t
,
_
. (20.5)
20.1 Classication of vierbein perturbations
The aims of this section are two-fold. First, decompose perturbations into scalar, vector, and tensor parts.
Second, identify the coordinate and tetrad gauge-invariant perturbations. It will be found, equations (20.13),
that there are 6 coordinate and tetrad gauge-invariant perturbations, comprising 2 scalars and , 1 vector
W
i
containing 2 degrees of freedom, and 1 tensor h
ij
containing 2 degrees of freedom.
The vierbein perturbations
mn
decompose into 6 scalars, 4 vectors, and 1 tensor
tt
=
scalar
, (20.6a)
ti
=
i
w
scalar
+ w
i
vector
, (20.6b)
it
=
i
w
scalar
+ w
i
vector
, (20.6c)
ij
=
ij

scalar
+
i
j
h
scalar
+
ijk
h
scalar
+
i
h
j
vector
+
j
h
i
vector
+ h
ij
tensor
. (20.6d)
The tildes on w and

h simply distinguish those symbols; the tildes have no other signicance. The vector
components are by denition transverse (have vanishing divergence), while the tensor component h
ij
is by
denition traceless, symmetric, and transverse. For a single Fourier mode whose wavevector k is taken
without loss of generality to lie in the z-direction, equations (20.6) are
mn
=
_
_
_
_
w
x
w
y

z
w
w
x
+h
xx
h
xy
+
z
h
z
h
x
w
y
h
xy
h h
xx

z
h
y
z
w
z
h
x

z
h
y
+
2
z
h
_
_
_
_
. (20.7)
To identify coordinate gauge-invariant quantities, it is necessary to consider innitesimal coordinate gauge
transformations (18.8). The tetrad-frame components
m
of the coordinate shift of the coordinate gauge
transformation decompose into 2 scalars and 1 vector
m
=
t
scalar
,
i
scalar
+
i
vector
. (20.8)
In the at space background space being considered the coordinate gauge transformation (18.26) of the
vierbein perturbation simplies to
mn

mn
=
mn
+
m
n
. (20.9)
In terms of the scalar, vector, and tensor potentials introduced in equations (20.6), the gauge transforma-
tions (20.9) are
tt
+
t
scalar
, (20.10a)
ti

i
(w + )
scalar
+ (w
i
+
i
)
vector
, (20.10b)
it

i
( w +
t
)
scalar
+ w
i
vector
, (20.10c)
ij

ij

scalar
+
i
j
(h +)
scalar
+
ijk
h
scalar
+
i
(h
j
+
j
)
vector
+
j
h
i
vector
+ h
ij
tensor
. (20.10d)
328 Flat space background
Equations (20.10a) imply that under an innitesimal coordinate gauge transformation
+
t
, (20.11a)
w w + , w
i
w
i
+
i
, (20.11b)
w w +
t
, w
i
w
i
, (20.11c)
, h h + ,

h
h , h
i
h
i
+
i
,

h
i

h
i
, h
ij
h
ij
. (20.11d)
Eliminating the coordinate shift
m
from the transformations (20.11) yields 12 coordinate gauge-invariant
combinations of the potentials

w , w

h , w
i

h
i
, w
i
, ,

h ,

h
i
, h
ij
. (20.12)
Physical perturbations are not only coordinate but also tetrad gauge-invariant. A quantity is tetrad gauge-
invariant if and only if it depends only on the symmetric part of the vierbein pertubations, not on the
antisymmetric part, 18.6. There are 6 combinations of the coordinate gauge-invariant perturbations (20.12)
that are symmetric, and therefore not only coordinate but also tetrad gauge-invariant. These 6 coordinate
and tetrad gauge-invariant perturbations comprise 2 scalars, 1 vector, and 1 tensor
scalar
w

w +

h , (20.13a)
scalar
, (20.13b)
W
i
vector
w
i
+ w
i

h
i
h
i
, (20.13c)
h
ij
tensor
. (20.13d)
Since only the 6 tetrad and coordinate gauge-invariant potentials , , W
i
, and h
ij
have physical signi-
cance, it is legitimate to choose a particular gauge, a set of conditions on the non-gauge-invariant potentials,
arranged to simplify the equations, or to bring out some physical aspect. Three gauges considered later are
harmonic gauge (20.7), Newtonian gauge (20.9), and synchronous gauge (20.10). However, for the next
several sections, no gauge will be chosen: the exposition will continue to be completely general.
20.2 Metric, tetrad connections, and Einstein and Weyl tensors
This section gives expressions in a completely general gauge for perturbed quantities in at background
Minkowski space.
20.2 Metric, tetrad connections, and Einstein and Weyl tensors 329
The perturbed coordinate metric g
, equation (18.5), is
g
tt
= (1 + 2 ) , (20.14a)
g
ti
=
i
(w + w) (w
i
+ w
i
) , (20.14b)
g
ij
=
ij
(1 2 ) 2
i
j
h
i
(h
j
+

h
j
)
j
(h
i
+

h
i
) 2 h
ij
. (20.14c)
The coordinate metric is tetrad gauge-invariant, but not coordinate gauge-invariant.
The perturbed tetrad connections
kmn
are
tit
=
i
(

w) +

w
i
, (20.15a)
tij
=
ij

i
j
(w

h)
1
2
(
i
W
j
+
j
W
i
) +
j
w
i
+

h
ij
, (20.15b)
ijt
=
1
2
(
i
W
j

j
W
i
)

t
(
ijl
h
i
h
j
+
j
h
i
) , (20.15c)
ijk
= (
jk
ik
j
)
k
(
ijl
h
i
h
j
+
j
h
i
) +
i
h
jk
j
h
ik
. (20.15d)
Being purely tetrad-frame quantities, the tetrad connections are automatically coordinate gauge-invariant.
However, they are not tetrad gauge-invariant, as is evident from the fact that they (all) depend on antisym-
metric parts of the vierbein perturbations
mn
.
One of the advantages of working with tetrads is that tetrad-frame quantities such as the tetrad connec-
tions
kmn
and the tetrad-frame Riemann tensor R
klmn
are by construction independent of the choice of
coordinates, and are therefore automatically coordinate gauge-invariant. In the tetrad formalism, you do
not have to work too hard (is that really ever true?) to construct coordinate gauge-invariant combinations
of the vierbein perturbations
mn
: the tetrad-frame connections and Riemann tensor will automatically
give you the coordinate gauge-invariant combinations. You can check that in the present case the tetrad
connections (20.15) depend only on, and on all 12 of, the coordinate gauge-invariant combinations (20.12).
The perturbed tetrad-frame Einstein tensor G
mn
is
G
tt
= 2
2
scalar
, (20.16a)
G
ti
= 2
i

scalar
+
1
2
2
W
i
vector
, (20.16b)
G
ij
= 2
ij

scalar
(
i
j

ij
2
)()
scalar
+
1
2
(
i

W
j
+
j

W
i
)
vector
h
ij
tensor
, (20.16c)
where is the dAlembertian, the 4-dimensional wave operator

m
m
=

2
t
2

2
. (20.17)
Being a tetrad-frame quantity, the tetrad-frame Einstein tensor is automatically coordinate gauge-invariant.
Equations (20.16) show that the tetrad-frame Einstein tensor G
mn
is also tetrad gauge-invariant, since it
depends only on the tetrad-gauge invariant combinations (20.13) of the vierbein perturbations. The property
that the Einstein tensor is tetrad as well as coordinate gauge-invariant is a feature of empty background
space, and does not persist to more general spacetimes, such as the Friedmann-Robertson-Walker spacetime.
In a frame with the wavector k taken along the z-axis, the perturbed Einstein tensor is
G
mn
=
_
_
_
_
_
_
_
2
2
z
1
2
2
z
W
x
1
2
2
z
W
y
2
z

1
2
2
z
W
x
2

+
2
z
() h
+
h
1
2
z

W
x
1
2
2
z
W
y
h
2

+
2
z
() +h
+
1
2
z

W
y
2
z

1
2
z

W
x
1
2
z

W
y
2

_
_
_
_
_
_
_
(20.18)
where h
+
and h
are the two polarizations of gravitational waves, discussed further in 20.14,

h
+
h
xx
= h
yy
, h
h
xy
= h
yx
. (20.19)
The tetrad-frame complexied Weyl tensor is
C
titj
=
1
4
(
i
j

1
3

ij
2
)( + )
scalar
+
1
8
_
(
i

W
j
+
j

W
i
) +i(
ikl
j
+
jkl
i
)
k
W
l
vector
+
1
4
_
h
ij

ikl
jmn
m
h
ln
i(
ikl
h
jl
+
jkl
h
il
)
tensor
. (20.20)
Like the tetrad-frame Einstein tensor, the tetrad-frame Weyl tensor is both coordinate and tetrad gauge-
invariant, depending only on the coordinate and tetrad gauge-invariant potentials , , W
i
, and h
ij
.
20.3 Spinor components of the Einstein tensor
Scalar, vector, and tensor perturbations correspond respectively to perturbations of spin 0, 1, and 2. An
object has spin m if it is unchanged by a rotation of 2/m about a prescribed direction. In perturbed
Minkowski space, the prescribed direction is the direction of the wavevector k in the Fourier decomposition
of the modes. The spin components may be projected out by working in a spinor tetrad, 12.1.1.
In a frame where the wavevector k is taken along the z-axis, the spinor components of the perturbed
Einstein tensor G
mn
are (compare equations (20.16))
G
tt
= 2
2
z
spin-0
, G
tz
= 2
z

spin-0
, G
zz
= 2

spin-0
, (20.21a)
G
+
G
zz
=
2
z
()
spin-0
, (20.21b)
G
t
=
1
2
2
z
W
spin-1
, G
z
=
1
2
z

W
spin-1
, (20.21c)
G
= h
spin-2
, (20.21d)
20.4 Too many Einstein equations? 331
where W
are the spin-1 components of the vector perturbation W

i
W
=
1
2
(W
x
i W
y
) , (20.22)
and h
are the spin-2 components of the tensor perturbation h

ij
h
= h
xx
i h
xy
= h
+
i h
. (20.23)
The spin +2 and 2 components h
++
and h
of the tensor perturbation are called the right- and left-handed

circular polarizations. The spin +2 and 2 circular polarizations h
++
and h
have respective shapes e

i2
and e
i2
, under a right-handed rotation by angle about the z-axis, which may be compared to the cos 2
and sin 2 shapes of the linear polarizations h
+
and h
.
20.4 Too many Einstein equations?
The Einstein equations are as usual (units c = G = 1)
G
mn
= 8T
mn
. (20.24)
There are 10 Einstein equations, but the Einstein tensor (20.16) depends on only 6 independent potentials:
the two scalars and , the vector W
i
, and the tensor h
ij
. The system of Einstein equations is thus over-
complete. Why? The answer is that 4 of the Einstein equations enforce conservation of energy-momentum,
and can therefore be considered as governing the evolution of the energy-momentum as opposed to being
equations for the gravitational potentials. For example, the form of equations (20.16a) and (20.16b) for G
tt
and G
ti
enforces conservation of energy
D
m
G
mt
= 0 , (20.25)
while the form of equations (20.16b) and (20.16c) for G
ti
and G
ij
enforces conservation of momentum
D
m
G
mi
= 0 . (20.26)
Normally, the equations governing the evolution of the energy-momentum T
mn
a
of each species a of mass-
energy would be set up so as to ensure overall conservation of energy-momentum. If this is done, then
the conservation equations (20.25) and (20.26) can be regarded as redundant. Since equations (20.25) and
(20.26) are equations for the time evolution of G
tt
and G
ti
, one might think that the Einstein equations
for G
tt
and G
ti
would become redundant, but this is not quite true. In fact the Einstein equations for
G
tt
and G
ti
impose constraints that must be satised on the initial spatial hypersurface. Conservation
of energy-momentum guarantees that those constraints will continue to be satised on subsequent spatial
hypersurfaces, but still the initial conditions must be arranged to satisfy the constraints. Because the Einstein
equations for G
tt
and G
ti
must be satised as constraints on the initial conditions, but thereafter can be
ignored, the equations are called constraint equations. The Einstein equation for G
tt
is called the energy
constraint, or Hamiltonian constraint. The Einstein equations for G
ti
are called the momentum constraints.
Exercise 20.1 Energy and momentum constraints. Conrm the argument of this section. Suppose
that the spatial Einstein equations are true, G
ij
= 8T
ij
. Show that if the time-time and time-space
Einstein equations G
tm
= 8T
tm
are initially true, then conservation of energy-momentum implies that
these equations must necessarily remain true at all times. [Hint: Conservation of energy-momentum requires
that D
m
T
mn
= 0, and the Bianchi identities require that the Einstein tensor satises D
m
G
mn
= 0, so
D
m
(G
mn
8T
mn
) = 0 . (20.27)
By expanding out these equations in full, or otherwise, show that the solution satisfying G
ij
8T
ij
= 0 at
all times, and G
tm
8T
tm
= 0 initially, is G
tm
8T
tm
= 0 at all times.]
Concept question 20.2 Which Einstein equations are redundant? It has been argued in this
section that, if the energy-momentum tensor T
mn
is arranged to satisfy energy-conservation D
m
T
mn
= 0 as
it should, then the time-time and time-space Einstein equations must be satised by the initial conditions,
but thereafter become redundant. Question: Can any 4 of the 10 Einstein equations be dropped, or just the
time-time and time-space Einstein equations?
20.5 Action at a distance?
The tensor component of the Einstein equations shows that, in a vacuum T
mn
= 0, the tensor perturbations
h
ij
propagate at the speed of light, satisfying the wave equation
h
ij
= 0 . (20.28)
The tensor perturbations represent propagating gravitational waves.
It is to be expected that scalar and vector perturbations would also propagate at the speed of light, yet
this is not obvious from the form of the Einstein tensor (20.16). Specically, there are 4 components of the
Einstein tensor (20.16) that apparently depend only on spatial derivatives, not on time derivatives. The 4
corresponding Einstein equations are
2
= 4T
tt
scalar
, (20.29a)
2
W
i
= 16T
ti
vector
, (20.29b)
2
() = 8Q
ij
T
ij
scalar
, (20.29c)
where Q
ij
in equation (20.29c) is the quadrupole operator dened below, equation (20.87). These condi-
tions must be satised everywhere at every instant of time, giving the impression that signals are traveling
instantaneously from place to place.
20.6 Comparison to electromagnetism
The previous two sections 20.4, 20.5 brought up two issues:
1. There are 10 Einstein equations, but only 6 independent gauge-invariant potentials , , W
i
, and h
ij
.
The additional 4 Einstein equations serve to enforce conservation of energy-momentum.
2. Only 2 of the gauge-invariant potentials, the tensor potentials h
ij
, satisfy causal wave equations. The
remaining 4 gauge-invariant potentials , , and W
i
, satisfy equations (20.29) that depend on the instan-
taneous distribution of energy-momentum throughout space, on the face of it violating causality.
These facts may seem surprising, but in fact the equations of electromagnetism have a similar structure, as
will now be shown. In this section, the spacetime is assumed to be at Minkowski space. The discussion
in this section is based in part on the exposition by Bertschinger (1995 Cosmological Dynamics, 1993 Les
Houches Lectures, arXiv:astro-ph/9503125).
In accordance with the usual procedure, the electromagnetic eld may be dened in terms of an elec-
tromagnetic 4-potential A
m
, whose time and spatial parts constitute the scalar potential and the vector
potential A:
A
m
, A . (20.30)
The electric and magnetic elds E and B may be dened in terms of the potentials and A by
E
A
t
, (20.31a)
B A . (20.31b)
Given their denition (20.31), the electric and magnetic elds automatically satisfy the two source-free
Maxwells equations
B = 0 , (20.32a)
E +
B
t
= 0 . (20.32b)
The remaining two Maxwells equations, the sourced ones, are
E = 4q , (20.33a)
B
E
t
= 4j , (20.33b)
where q and j are the electric charge and current density, the time and space components of the electric
4-current density j
m
j
m
q, j . (20.34)
The electromagnetic potentials and A are not unique, but rather are dened only up to a gauge transfor-
mation by some arbitrary gauge eld
+

t
, A A . (20.35)
The gauge transformation (20.35) evidently leaves the electric and magnetic elds E and B, equations (20.31),
invariant.
Following the path of previous sections, 20.1 and thereafter, decompose the vector potential A into its
scalar and vector parts
A = A
scalar
+ A
vector
, (20.36)
in which the vector part by denition satises the transversality condition A
= 0. Under a gauge
transformation (20.35), the potentials transform as
+

t
, (20.37a)
A
, (20.37b)
A
. (20.37c)
Eliminating the gauge eld yields 3 gauge-invariant potentials, comprising 1 scalar , and 1 vector A
containing 2 degrees of freedom:
scalar
+
A
t
, (20.38a)
A
vector
. (20.38b)
This shows that the electromagnetic eld contains 3 independent degrees of freedom, consisting of 1 scalar
and 1 vector.
Concept question 20.3 Are gauge-invariant potentials Lorentz invariant? The potentials and
A
, equations (20.38), are by construction gauge-invariant, but is this construction Lorentz invariant? Do
and A
constitute the components of a 4-vector?

In terms of the gauge-invariant potentials and A
, equations (20.38), the electric and magnetic elds

are
E =
A
t
, (20.39a)
B = A
. (20.39b)
The sourced Maxwells equations (20.33) thus become, in terms of and A
scalar
= 4q
scalar
, (20.40a)
scalar
+A
vector
= 4j
scalar
+ 4j
vector
, (20.40b)
where j
and j
are the scalar and vector parts of the current density j. Equations (20.40) bear a striking
similarity to the Einstein equations (20.16). Only the vector part A
satises a wave equation,

A
= 4j
, (20.41)
while the scalar part satises an instantaneous equation (20.40a) that seemingly violates causality. And just
as Einsteins equations (20.16) enforce conservation of energy-momentum, so also Maxwells equations (20.40)
enforce conservation of electric charge
q
t
+ j = 0 , (20.42)
or in 4-dimensional form
m
j
m
= 0 . (20.43)
The fact that only the vector part A
satises a wave equation (20.41) reects physically the fact that

electromagnetic waves are transverse, and they contain only two propagating degrees of freedom, the vector,
or spin 1, components.
Why do Maxwells equations (20.40) have this structure? Although equation (20.41) appears to be a local
wave equation for the vector part A
of the potential sourced by the vector part j
of the current, in fact

the wave equation is non-local because the decomposition of the potential and current into scalar and vector
parts is non-local (it involves the solution of a Laplacian equation, eq. (19.3)). It is only the sumj = j
+j
of the scalar and vector parts of the current density that is local. Therefore, the Maxwells equation (20.40b)
must have a scalar part to go along with the vector part, such that the source on the right hand side, the
current density j, is local. Given this Maxwell equation (20.40b), the Maxwell equation (20.40a) then serves
precisely to enforce conservation of electric charge, equation (20.42).
Just as it is possible to regard the Einstein equations (20.16a) and (20.16b) as constraint equations
whose continued satisfaction is guaranteed by conservation of energy-momentum, so also the Maxwell equa-
tion (20.40a) for can be regarded as a constraint equation whose continued satisfaction is guaranteed
by conservation of electric charge. For charge conservation (20.42) coupled with the spatial Maxwell equa-
tion (20.40b) ensures that
t
_
4q +
2
_
= 0 , (20.44)
the solution of which, subject to the condition that 4q +
2
= 0 initially, is 4q +
2
= 0 at all times,
which is precisely the Maxwell equation (20.40a).
Exercise 20.4 Is it possible to discard the scalar part of the spatial Maxwell equation (20.40b),
rather than equation (20.40a) for ? Project out the scalar part of equation (20.40b) by taking its
divergence,
2
_
4j
_
= 0 . (20.45)
Argue that the Maxwell equation (20.40a), coupled with charge conservation (20.42), ensures that equa-
tion (20.45) is true, subject to boundary condition that the current j vanish suciently rapidly at spatial
innity, in accordance with the decomposition theorem of 19.1.
Since only gauge-invariant quantities have physical signicance, it is legitimate to impose any condition
on the gauge eld . A gauge in which the potentials and A individually satisfy wave equations is Lorenz
(not Lorentz!) gauge, which consists of the Lorentz-invariant condition
m
A
m
= 0 . (20.46)
Under a gauge transformation (20.35), the left hand side of equation (20.46) transforms as
m
A
m
m
A
m
+ , (20.47)
and the Lorenz gauge condition (20.46) can be accomplished as a particular solution of the wave equation
for the gauge eld . In terms of the potentials and A
, the Lorenz gauge condition (20.46) is
t
+
2
A
= 0 . (20.48)
In Lorenz gauge, Maxwells equations (20.40) become
= 4q , (20.49a)
A = 4j , (20.49b)
which are manifestly wave equations for the potentials and A.
Does the fact that the potentials and A in one particular gauge, Lorenz gauge, satisfy wave equations
necessarily guarantee that the electric and magnetic elds E and B satisfy wave equations? Yes, because it
follows from the denitions (20.31) of E and B that if the potentials and A satisfy wave equations, then
so also must the elds E and B themselves; but the elds E and B are gauge-invariant, so if they satisfy
wave equations in one gauge, then they must satisfy the same wave equations in any gauge.
In electromagnetism, the most physical choice of gauge is one in which the potentials and A coincide
with the gauge-invariant potentials and A
, equations (20.38). This gauge, known as Coulomb gauge,

is accomplished by setting
A
= 0 , (20.50)
or equivalently
A = 0 . (20.51)
The gravitational analogue of this gauge is the Newtonian gauge discussed in the next section but one, 20.9.
Does the fact that in Lorenz gauge the potentials and A propagate at the speed of light (in the absence
of sources, j
m
= 0) imply that the gauge-invariant potentials and A
propagate at the speed of light?

No. The gauge-invariant potentials and A
, equations (20.38), are related to the Lorenz gauge potentials

and A by a non-local decomposition.
20.7 Harmonic gauge 337
20.7 Harmonic gauge
The fact that all locally measurable gravitational perturbations do propagate causally, at the speed of light
in the absence of sources, can be demonstrated by choosing a particular gauge, harmonic gauge, equa-
tion (20.52), which can be considered an analogue of the Lorenz gauge of electromagnetism, equation (20.46).
In harmonic gauge, all 10 of the tetrad gauge-variant (i.e. symmetric) combinations
mn
+
nm
of the vierbein
perturbations satisfy wave equations (20.56), and therefore propagate causally. This does not imply that
the scalar, vector, and tensor components of the vierbein perturbations individually propagate causally, be-
cause the decomposition into scalar, vector, and tensor modes is non-local. In particular, the coordinate and
tetrad-gauge invariant potentials , , W
i
, and h
ij
dened by equations (20.13) do not propagate causally.
The situation is entirely analogous to that of electromagnetism, 20.6, where in Lorenz gauge the potentials
and A propagate causally, equations (20.49), yet the gauge-invariant potentials and A
dened by
equations (20.38) do not.
Harmonic gauge is the set of 4 coordinate conditions
m
(
mn
+
nm
)
n
m
m
= 0 . (20.52)
The conditions (20.52) are arranged in a form that is tetrad gauge-invariant (the conditions depend only on
the symmetric part of
mn
). The quantities on the left hand side of equations (20.52) transform under a
coordinate gauge transformation, in accordance with (20.9), as
m
(
mn
+
nm
)
n
m
m
m
(
mn
+
nm
)
n
m
m
+
n
. (20.53)
The change
n
resulting from the coordinate gauge transformation is the 4-dimensional wave operator
acting on the coordinate shift
n
. Indeed, the harmonic gauge conditions (20.52) follow uniquely from the
requirements (a) that the change produced by a coordinate gauge transformation be
n
, as suggested by
the analogous electromagnetic transformation (20.47), and (b) that the conditions be tetrad gauge-invariant.
The harmonic gauge conditions (20.52) can be accomplished as a particular solution of the wave equation for
the coordinate shift
n
. In terms of the potentials dened by equations (20.6) and (20.13), the 4 harmonic
gauge conditions (20.52) are
+ 3
+(w + w

h) = 0 , (20.54a)
W
i
+(h
i
+

h
i
) = 0 , (20.54b)
+ +h = 0 , (20.54c)
or equivalently
4

= (w + w) , (20.55a)

W
i
= (h
i
+

h
i
) , (20.55b)
= h . (20.55c)
Substituting equations (20.55) into the Einstein tensor G
mn
leads, after some calculation, to the result that
in harmonic gauge
1
2
(
mn
+
nm
) = R
mn
, (20.56)
where R
mn
= G
mn
1
2

mn
G is the Ricci tensor. Equation (20.56) shows that in harmonic gauge, all tetrad
gauge-invariant (i.e. symmetric) combinations
mn
+
nm
of the vierbein potentials propagate causally, at
the speed of light in vacuo, R
mn
= 0. Although the result (20.56) is true only in a particular gauge,
harmonic gauge, it follows that all quantities that are (coordinate and tetrad) gauge-invariant, and that can
be constructed from the vierbein potentials
mn
and their derivatives (and are therefore local), must also
propagate at the speed of light.
20.8 What is the gravitational eld?
In electromagnetism, the electromagnetic elds are the electric eld E and the magnetic eld B. These elds
have the property that they are gauge-invariant, and measurable locally. The electromagnetic potentials
and A
, equations (20.38), are gauge-invariant, but they are not measurable locally.
What are the analogous gauge-invariant and locally measurable quantities for the gravitational eld in
perturbed Minkowski space? The answer is, the Weyl tensor C
klmn
, the trace-free or tidal part of the
Riemann tensor, the expression (20.20) for which depends only on the coordinate and tetrad gauge-invariant
potentials.
20.9 Newtonian (Copernican) gauge
If the unperturbed background is Minkowski space, then the most physical gauge is one in which the 6
perturbations retained coincide with the 6 coordinate and tetrad gauge-invariant perturbations (20.13).
This gauge is called Newtonian gauge. Because in Newtonian gauge the perturbations are precisely the
physical perturbations, if the perturbations are physically weak (small), then the perturbations in Newtonian
gauge will necessarily be small.
I think Newtonian gauge should be called Copernican gauge. Even though the solar system is a highly non-
linear system, from the perspective of general relativity it is a weakly perturbed gravitating system. Applied
to the solar system, Newtonian gauge eectively keeps the coordinates aligned with the classical Sun-centred
Copernican coordinate frame. By contrast, the coordinates of synchronous gauge (20.10), which are chosen
to follow freely-falling bodies, would quickly collapse or get wound up by orbital motions if applied to the
solar system, and would cease to provide a useful description.
Newtonian gauge sets
w = w = w
i
= h =

h = h
i
=

h
i
= 0 , (20.57)
so that the retained perturbations are the 6 coordinate and tetrad gauge-invariant perturbations (20.13)
scalar
= , (20.58a)
scalar
, (20.58b)
W
i
vector
= w
i
, (20.58c)
h
ij
tensor
. (20.58d)
The Newtonian line-element is, in a form that keeps the Newtonian tetrad manifest,
ds
2
=
_
(1 + ) dt
2
+
ij
_
(1 )dx
i
h
i
k
dx
k
W
i
dt
_
(1 )dx
j
h
j
l
dx
l
W
j
dt
, (20.59)
which reduces to the Newtonian metric
ds
2
= (1 + 2 ) dt
2
2 W
i
dt dx
i
+
_
ij
(1 2 ) 2 h
ij
dx
i
dx
j
. (20.60)
Since scalar, vector, and tensor perturbations evolve independently, it is legitimate to consider each in
isolation. For example, if one is interested only in scalar perturbations, then it is ne to keep only the
scalar potentials and non-zero. Furthermore, as discussed in 20.13, since the dierence in
scalar potentials is sourced by anisotropic relativistic pressure, which is typically small, it is often a good
approximation to set = .
The tetrad-frame 4-velocity of a person at rest in the tetrad frame is by denition u
m
= 1, 0, 0, 0, and
the corresponding coordinate 4-velocity u
is, in Newtonian gauge,

u
= e
t
= 1 , W
i
. (20.61)
This shows that W
i
can be interpreted as a 3-velocity at which the tetrad frame is moving through the coor-
dinates. This is the dragging of inertial frames discussed in 20.12. The proper acceleration experienced
by a person at rest in the tetrad frame, with tetrad 4-velocity u
m
= 1, 0, 0, 0, is
Du
i
D
= u
t
D
t
u
i
= u
t
_
t
u
i
+
i
tt
u
t
_
=
i
tt
=
i
. (20.62)
This shows that the gravity, or minus the proper acceleration, experienced by a person at rest in the tetrad
frame is minus the gradient of the potential .
Concept question 20.5 If the decomposition into scalar, vector, and tensor modes is non-local, how can
it be legimate to consider the evolution of the modes in isolation from each other?
20.10 Synchronous gauge
One of the earliest gauges used in general relativistic perturbation theory, and still (in its conformal version)
widely used in cosmology, is synchronous gauge. As will be seen below, equations (20.69) and (20.70),
synchronous gauge eectively chooses a coordinate system and tetrad that is attached to the locally inertial
frames of freely falling observers. This is ne as long as the observers move only slightly from their initial
positions, but the coordinate system will fail when the system evolves too far, even if, as in the solar system,
the gravitational perturbations remain weak and therefore treatable in principle with perturbation theory.
Synchronous gauge sets the time components
mn
with m = t or n = t of the vierbein perturbations to
zero
= w = w = w
i
= w
i
= 0 , (20.63)
and makes the additional tetrad gauge choices
h =

h
i
= 0 , (20.64)
with the result that the retained perturbations are the spatial perturbations
scalar
, h
scalar
, h
i
vector
, h
ij
tensor
. (20.65)
In terms of these spatial perturbations, the gauge-invariant perturbations (20.13) are
scalar
=

h , (20.66a)
scalar
, (20.66b)
W
i
vector
=
h
i
, (20.66c)
h
ij
tensor
. (20.66d)
The synchronous line-element is, in a form that keeps the synchronous tetrad manifest,
ds
2
= dt
2
+
ij
_
(1 )dx
i
(
k
i
h +
k
h
i
+h
i
k
)dx
k
_
(1 )dx
j
(
l
j
h +
l
h
j
+h
j
l
)dx
l
, (20.67)
which reduces to the synchronous metric
ds
2
= dt
2
+ [(1 2 )
ij
2
i
j
h
i
h
j
j
h
i
2 h
ij
] dx
i
dx
j
. (20.68)
In synchronous gauge, a person at rest in the tetrad frame has coordinate 4-velocity
u
= e
t
= 1, 0, 0, 0 , (20.69)
so that the tetrad rest frame coincides with the coordinate rest frame. Moreover a person at rest in the
tetrad frame is freely falling, which follows from the fact that the acceleration experienced by a person at
rest in the tetrad frame is zero
Du
k
D
= u
t
_
t
u
k
+
k
tt
u
t
_
=
k
tt
= 0 , (20.70)
in which
t
u
k
= 0 because the 4-velocity at rest in the tetrad frame is constant, u
k
= 1, 0, 0, 0, and
k
tt
= 0
from equations (20.15a) with the synchronous gauge choices (20.63) and (20.64).
20.11 Newtonian potential 341
20.11 Newtonian potential
The next few sections examine the physical meaning of each of the gauge-invariant potentials , , W
i
, and
h
ij
by looking at the potentials at large distances produced by a nite body containing energy-momentum,
such as the Sun.
Einsteins equations G
mn
= 8T
mn
applied to the time-time component G
tt
of the Einstein tensor, equa-
tion (20.16a), imply Poissons equation
2
= 4 , (20.71)
where is the mass-energy density
T
tt
. (20.72)
The solution of Poissons equation (20.71) is
(x) =
_
(x
) d
3
x
[x
x[
. (20.73)
Consider a nite body, for example the Sun, whose energy-momentum is conned within a certain region.
Dene the mass M of the body to be the integral of the mass-energy density ,
M
_
(x
) d
3
x
. (20.74)
Equation (20.74) agrees with what the denition of the mass M would be in the non-relativistic limit, and
as seen below, equation (20.77), it is what a distant observer would infer the mass of the body to be based
on its gravitational potential far away. Thus equation (20.74) can be taken as the denition of the mass
of the body even when the energy-momentum is relativistic. Choose the origin of the coordinates to be at
the centre of mass, meaning that
_
x
(x
) d
3
x
= 0 . (20.75)
Consider the potential at a point x far outside the body. Expand the denominator of the integral on the
right hand side of equation (20.73) as a Taylor series in 1/x where x [x[
1
[x
x[
=
1
x
=0
_
x
x
_
( x x
) =
1
x
+
x x
x
2
+... (20.76)
where P
() are Legendre polynomials. Then

(x) =
1
x
_
(x
) d
3
x
1
x
2
x
_
x
(x
) d
3
x
O(x
3
)
=
M
x
O(x
3
) . (20.77)
Equation (20.77) shows that the potential far from a body goes as = M/x, reproducing the usual
Newtonian formula.
20.12 Dragging of inertial frames
In Newtonian gauge, the vector potential W W
i
is the velocity at which the locally inertial tetrad frame
moves through the coordinates, equation (20.61). This is called the dragging of inertial frames. As shown
below, a body of angular momentum L drags frames around it with an angular velocity that goes to 2L/x
3
at large distances x.
Einsteins equations applied to the vector part of the time-space component G
ti
of the Einstein tensor,
equation (20.16b), imply
2
W = 16f , (20.78)
where W W
i
is the gauge-invariant vector potenial, and f is the vector part of the energy ux T
ti
f f
i
= f
i
T
ti
vector
= T
ti
vector
. (20.79)
The solution of equation (20.78) is
W(x) = 4
_
f(x
) d
3
x
[x
x[
. (20.80)
As in the previous section, 20.11, consider a nite body, such as the Sun, whose energy-momentum is
conned within a certain region. Work in the rest frame of the body, dened to be the frame where the
energy ux f integrated over the body is zero,
_
f(x
) d
3
x
= 0 . (20.81)
Dene the angular momentum L of the body to be
L
_
x
f(x
) d
3
x
. (20.82)
Equation (20.82) agrees with what the denition of angular momentum would be in the non-relativistic limit,
where the mass-energy ux of a mass density moving at velocity v is f = v. As will be seen below, the
angular momentum (20.82) is what a distant observer would infer the angular momentum of the body to be
based on the potential W far away, and equation (20.82) can be taken to be the denition of the angular
momentum of the body even when the energy-momentum is relativistic. As will be proven momentarily,
equation (20.83), the integral
_
x
i
f
j
(x
) d
3
x
is antisymmetric in ij. To show this, write f

j
=
jkl
l
for
some potential
l
, which is valid because f
j
is the vector (curl) part of the energy ux. Then
_
x
i
f
j
(x
) d
3
x
=
_
x
jkl
l
(x
) d
3
x
=
_

jkl
l
(x
k
x
i
d
3
x
=
_

ijl
l
(x
) d
3
x
, (20.83)
where the third expression follows from the second by integration by parts, the surface term vanishing
because of the assumption that the energy-momentum of the body is conned within a certain region.
20.13 Quadrupole pressure 343
Taylor expanding equation (20.80) using equation (20.76) gives
W(x) =
4
x
_
f(x
) d
3
x
+
4
x
2
_
( x x
)f(x
) d
3
x +O(x
3
)
=
2
x
2
_
[( x x
)f(x
) ( x f(x
))x
] d
3
x +O(x
3
)
=
2
x
2
L x +O(x
3
) , (20.84)
where the rst integral on the right hand side of the rst line of equation (20.84) vanishes because the frame
is the rest frame of the body, equation (20.81), and the second integral on the right hand side of the rst line
equals the rst integral on the second line thanks to the antisymmetry of
_
x
f(x
) d
3
x, equation (20.83).
The vector potential W W
i
points in the direction of rotation, right-handedly about the axis of angular
momentum L. Equation (20.84) says that a body of angular momemtumL drags frames around it at angular
velocity at large distances x
W = x , =
2L
x
3
. (20.85)
20.13 Quadrupole pressure
Einsteins equations applied to the part of the Einstein tensor (20.16c) involving imply
2
() = 8Q
ij
T
ij
, (20.86)
where Q
ij
is the quadrupole operator (an integro-dierential operator) dened by
Q
ij

3
2
1
2

ij
, (20.87)
with
2
the inverse spatial Laplacian operator. In Fourier space, the quadrupole operator is
Q
ij
=
3
2
k
i
k
j

1
2

ij
. (20.88)
The quadrupole operator Q
ij
yields zero when acting on
ij
, and the Laplacian operator
2
when acting on
j
Q
ij
ij
= 0 , Q
ij
j
=
2
. (20.89)
The solution of equation (20.86) is
=
_ _
3
2
(x
i
x
i
)(x
j
x
j
)
[x x
[
2

1
2

ij
_
T
ij
(x
) d
3
x
[x x
[
. (20.90)
At large distance in the z-direction from a nite body
=
1
x
_
_
T
zz
1
2
(T
xx
+T
yy
)
d
3
x
+O(x
2
) . (20.91)
Equation (20.86) shows that the source of the dierence between the two scalar potentials is the
quadrupole pressure. Since the quadrupole pressure is small if either there are no relativistic sources, or any
relativistic sources are isotropic, it is often a good approximation to set = . An exception is where there
is a signicant anisotropic relativistic component. For example, the energy-momentum tensor of a static
electric eld is relativistic and anisotropic. However, this is still not enough to ensure that diers from :
as found in Exercise 20.6, if the energy-momentum of a body is spherically symmetric, then vanishes
outside (but not inside) the body.
One situation where the dierence between and is appreciable is the case of freely-streaming neutrinos
at around the time of recombination in cosmology. The 2008 analysis of the CMB by the WMAP team claims
to detect a non-zero value of from a slight shift in the third acoustic peak.
Exercise 20.6 Argue that the traceless part of the energy-momentum tensor of a spherically symmetric
distribution must take the form
T
ij
(r) =
_
r
i
r
j

1
3

ij
__
p(r) p
(r)
_
, (20.92)
where p(r) and p
(r) are the radial and transverse pressures at radius r. From equation (20.90), show that
at radial distance x from the centre of a spherically symmetric distribution is
(x) (x) =
_

x
(r
2
x
2
)
_
p(r) p
(r)
_
4dr
r
. (20.93)
Notice that the integral is over r > x, that is, only energy-momentum outside radius x produces non-
vanishing . In particular, if the body has nite extent, then vanishes outside the body.
20.14 Gravitational waves
The tensor perturbations h
ij
describe propagating gravitational waves. The two independent components of
the tensor perturbations describe two polarizations. The two components are commonly designated h
+
and
h
, equations (20.19). Gravitational waves induce a quadrupole tidal oscillation transverse to the direction
of propagation, and the subscripts + and represent the shape of the quadrupole oscillation, as illustrated
by Figure 20.1. The h
+
polarization has a cos 2 shape, while the h
polarization has a sin 2 shape, where

is the azimuthal angle with respect to the y-axis about the direction x of propagation.
Einsteins equations applied to the tensor component of the spatial Einstein tensor (20.16c) imply that
gravitational waves are sourced by the tensor component of the energy-momentum
h
ij
= 8 T
ij
tensor
. (20.94)
The solution of the wave equation (20.94) can be obtained from the Greens function of the dAlembertian
wave operator . The Greens function is by denition the solution of the wave equation with a delta-
function source. There are retarded solutions, which propagate into the future along the future light cone,
20.14 Gravitational waves 345
Figure 20.1 The two polarizations of gravitational waves. The (top) polarization h+ has a cos 2 shape
about the direction of propagation (into the paper), while the (bottom) polarization h has a sin 2 shape.
A gravitational wave causes a system of freely falling test masses to oscillate relative to a grid of points a
xed proper distance apart.
and advanced solutions, which propagate into the past along the past light cone. In the present case, the
solutions of interest are the retarded solutions, since these represent gravitational waves emitted by a source.
Because of the time and space translation symmetry of the dAlembertian, the delta-function source of the
Greens function can without loss of generality be taken at the origin t = x = 0. Thus the Greens function
F is the solution of
F =
4
(x) , (20.95)
where
4
(x) (t)
3
(x) is the 4-dimensional Dirac delta-function. The solution of equation (20.95) subject
to retarded boundary conditions is (a standard exercise in mathematics) the retarded Greens function
F =
(x t)(t)
4x
, (20.96)
where x [x[ and (t) is the Heaviside function, (t) = 0 for t < 0 and (t) = 1 for t 0. The solution of
the sourced gravitational wave equation (20.94) is thus
h
ij
(t, x) = 2
_ T
ij
(t
, x
)
tensor
d
3
x
[x
x[
, (20.97)
where t
is the retarded time

t
t [x
x[ , (20.98)
which lies on the past light cone of the observer, and is the time at which the source emitted the signal. The
solution (20.97) resembles the solution of Poissons equation, except that the source is evaluated along the
past light cone of the observer.
As in 20.11 and 20.12, consider a nite body, whose energy-momentum is conned within a certain region,
and which is a source of gravitational waves. The Hulse-Taylor binary pulsar is a ne example. Far from the
body, the leading order contribution to the tensor potential h
ij
is, from the multipole expansion (20.76),
h
ij
(t, x) =
2
x
_
T
ij
(t
, x
)
tensor
d
3
x
. (20.99)
The integral (20.99) is hard to solve in general, but there is a simple solution for gravitational waves whose
wavelengths are large compared to the size of the body. To obtain this solution, rst consider that conser-
vation of energy-momentum implies that
2
T
tt
t
2

i
j
T
ji
=

t
_
T
tt
t
+
i
T
ti
_
i
_
T
ti
t
+
j
T
ji
_
= 0 . (20.100)
Multiply by x
i
x
j
and integrate
_
x
i
x
j

2
T
tt
t
2
d
3
x =
_
x
i
x
j
l
T
kl
d
3
x =
_
T
kl
l
(x
i
x
j
) d
3
x = 2
_
T
ij
d
3
x , (20.101)
where the third expression follows from the second by a double integration by parts. For wavelengths that
are long compared to the size of the body, the rst expression of equations (20.101) is
_
x
i
x
j
2
T
tt
t
2
d
3
x

2
t
2
_
x
i
x
j
T
tt
d
3
x =

2
I
ij
t
2
(20.102)
where I
ij
is the second moment of the mass
I
ij

_
x
i
x
j
T
tt
d
3
x . (20.103)
The tensor (spin-2) part of the energy-momentum is trace-free. The trace-free part I
ij
of the second moment
I
ij
is the quadrupole moment of the mass distribution (this denition is conventional, but diers by a factor
of 2/3 from what is called the quadrupole moment in spherical harmonics)
I
ij
I
ij

1
3

ij
I
k
k
=
_
(x
i
x
j

1
3

ij
x
2
) T
tt
d
3
x . (20.104)
Substituting the last expression of equations (20.101) into equation (20.99) gives the quadrupole formula for
gravitational radiation at wavelengths long compared to the size of the emitting body
h
ij
(t, x) =
1
x
I
ij
tensor
. (20.105)
20.15 Energy-momentum carried by gravitational waves 347
If the gravitational wave is moving in the z-direction, then the tensor components of the quadrupole moment
I
ij
are
I
+
=
1
2
(I
xx
I
yy
) , I
=
1
2
(I
xy
+I
yx
) . (20.106)
20.15 Energy-momentum carried by gravitational waves
The gravitational wave equation (20.28) in empty space appears to describe gravitational waves propagating
in a region where the energy-momentum tensor T
mn
is zero. However, gravitational waves do carry energy-
momentum, just as do other kinds of waves, such as electromagnetic waves. The energy-momentum is
quadratic in the tensor perturbation h
ij
, and so vanishes to linear order.
To determine the energy-momentum in gravitational waves, calculate the Einstein tensor G
mn
to second
order, imposing the vacuum conditions that the unperturbed and linear parts of the Einstein tensor vanish
0
G
mn
=
1
G
mn
= 0 . (20.107)
The parts of the second-order perturbation that depend on the tensor perturbation h
ij
are, in a frame where
the wavevector k is along the z-axis,
2
G
tt
= (
h
ij
)(
h
ij
) +
1
4
_

2
t
2
+
2
z
_
h
2
, (20.108a)
2
G
tz
= (
h
ij
)(
z
h
ij
) +
1
2
z
h
2
, (20.108b)
2
G
zz
= (
z
h
ij
)(
z
h
ij
) +
1
4
_

2
t
2
+
2
z
_
h
2
, (20.108c)
where
h
2
h
ij
h
ij
= 2(h
2
+
+h
2
) = 2h
++
h
. (20.109)
Being tetrad-frame quantities, the expressions (20.108) are automatically coordinate gauge-invariant, and
they are also tetrad gauge-invariant since they depend only on the (coordinate and) tetrad gauge-invariant
perturbation h
ij
. The rightmost set of terms on the right hand side of each of equations (20.108) are total
derivatives (with respect to time t or space z). These terms yield surface terms when integrated over a
region, and tend to average to zero when integrated over a region much larger than a wavelength. On the
other hand, the leftmost set of terms on the right hand side of each of equations (20.108) do not average
to zero; for example, the terms for G
tt
and G
zz
are negative everywhere, being minus a sum of squares. A
negative energy density? The interpretation is that these terms are to be taken over to the right hand side
of the Einstein equations, and re-interpreted as the energy-momentum T
gw
mn
in gravitational waves
T
gw
tt

1
8
_
(
h
ij
)(
h
ij
)
1
4
_

2
t
2
+
2
z
_
h
2
_
, (20.110a)
T
gw
tz

1
8
_
(
h
ij
)(
z
h
ij
)
1
2
z
h
2
_
, (20.110b)
T
gw
zz

1
8
_
(
z
h
ij
)(
z
h
ij
)
1
4
_
2
t
2
+
2
z
_
h
2
_
. (20.110c)
The terms involving total derivatives, although they vanish when averaged over a region larger than many
wavelengths, ensure that the energy-momentum T
gw
mn
in gravitational waves satises conservation of energy-
momentum in the at background space
m
T
gw
mn
= 0 . (20.111)
Averaged over a region larger than many wavelengths, the energy-momentum in gravitational waves is
T
gw
mn
) =
1
8
(
m
h
ij
)(
n
h
ij
) . (20.112)
Equation (20.112) may also be written explicitly as a sum over the two linear or circular polarizations
T
gw
mn
) =
1
4
[(
m
h
+
)(
n
h
+
) + (
m
h
)(
n
h
)]
=
1
8
[(
m
h
++
)(
n
h
) + (
n
h
++
)(
m
h
)] . (20.113)
PART EIGHT
COSMOLOGICAL PERTURBATIONS
Concept Questions
1. Why do the wavelengths of perturbations in cosmology expand with the Universe, whereas perturbations
in Minkowski space do not expand?
2. What does power spectrum mean?
3. Why is the power spectrum a good way to characterize the amplitude of uctuations?
4. Why is the power spectrum of uctuations of the Cosmic Microwave Background (CMB) plotted as a
function of harmonic number?
5. What causes the acoustic peaks in the power spectrum of uctuations of the CMB?
6. Are there acoustic peaks in the power spectrum of matter (galaxies) today?
7. What sets the scale of the rst peak in the power spectrum of the CMB? [What sets the physical scale?
Then what sets the angular scale?]
8. The odd peaks (including the rst peak) in the CMB power spectrum are compression peaks, while the
even peaks are rarefaction peaks. Why does a rarefaction produce a peak, not a trough?
9. Why is the rst peak the most prominent? Why do higher peaks generally get progressively weaker?
10. The third peak is about as strong as the second peak? Why?
11. The matter power spectrum reaches a maximum at a scale that is slightly larger than the scale of the rst
baryonic acoustic peak. Why?
12. The physical density of species x at the time of recombination is proportional to
x
h
2
where
x
is the
ratio of the actual to critical density of species x at the present time, and h H
0
/100 kms
1
Mpc
1
is
the present-day Hubble constant. Explain.
13. How does changing the baryon density
b
h
2
aect the CMB power spectrum?
14. How does changing the non-baryonic cold dark matter density
c
h
2
, without changing the baryon density
b
h
2
, aect the CMB power spectrum?
15. What eects do neutrinos have on perturbations?
16. How does changing the curvature
k
17. How does changing the dark energy

21
An overview of cosmological perturbations
Undoubtedly the preeminent application of general relativistic perturbation theory is to cosmology. Flucta-
tions in the temperature and polarization of the Cosmic Microwave Background (CMB) provide an observa-
tional window on the Universe at 400,000 years old that, coupled with other astronomical observations, has
yielded impressively precise measurements of cosmological parameters.
The theory of cosmological perturbations is based principally on general relativistic perturbation theory
coupled to the physics of 5 species of energy-momentum: photons, baryons, non-baryonic cold dark matter,
neutrinos, and dark energy.
Dark energy was not important at the time of recombination, where the CMB that we see comes from,
but it is important today. If dark energy has a vacuum equation of state, p = , then dark energy does
not cluster (vacuum energy density is a constant), but it aects the evolution of the cosmic scale factor,
and thereby does aect the clustering of baryons and dark matter today. Moreover the evolution of the
gravitational potential along the line-of-sight to the CMB does aect the observed power spectrum of the
CMB, the so-called integrated Sachs-Wolfe eect.
Unfortunately, it is beyond the scope of these notes to treat cosmological perturbations in full. For that,
consult Scott Dodelsons incomparable text Modern Cosmology.
1. Inationary initial conditions. The theory of ination has been remarkably successful in accounting for
many aspects of observational cosmology, even though a fundamental understanding of the inaton scalar
eld that supposedly drove ination is missing. The current paradigm holds that primordial uctuations
were generated by vacuum quantum uctuations in the inaton eld at the time of ination. The theory
makes the generic predictions that the gravitational potentials generated by vacuum uctuations were (a)
Gaussian, (b) adiabatic (meaning that all species of mass-energy uctuated together, as opposed to in
opposition to each other), and (c) scale-free, or rather almost scale-free (the fact that ination came to
an end modies slightly the scale-free character). The three predictions t the observed power spectrum
of the CMB astonishingly well.
2. Comoving Fourier modes. The spatial homogeneity of the Friedmann-Robertson-Walker background
spacetime means that its perturbations are characterized by Fourier modes of constant comoving wavevec-
An overview of cosmological perturbations 353
tor. Each Fourier mode generated by ination evolved independently, and its wavelength expanded with
the Universe.
3. Scalar, vector, tensor modes. Spatial isotropy on top of spatial homogenity means that the pertur-
bations comprised independently evolving scalar, vector, and tensor modes. Scalar modes dominate the
uctuations of the CMB, and caused the clustering of matter today. Vector modes are usually assumed
to vanish, because there is no mechanism to generate the vorticity that sources vector modes, and the
expansion of the Universe tends to redshift away any vector modes that might have been present. Ination
generates gravitational waves, which then propagate essentially freely to the present time. Gravitational
waves leave an observational imprint in the B (curl) mode of polarization of the CMB, whereas scalar
modes produce only an E (gradient) mode of polarization.
4. Power spectrum. The primary quantity measurable from observations is the power spectrum, which
is the variance of uctuations of the CMB or of matter (as traced by galaxies, galaxy clusters, the Lyman
alpha forest, peculiar velocities, weak lensing, or 21 centimeter observations at high redshift). The statis-
tics of a Gaussian eld are completely characterized by its mean and variance. The mean characterizes
the unperturbed background, while the variance characterizes the uctuations. For a 3-dimensional sta-
tistically homogeneous and isotropic eld, the variance of Fourier modes
k
denes the power spectrum
P(k)
k
) = 1
kk
P(k) , (21.1)
where 1
kk
is the unit matrix in the Hilbert space of Fourier modes
1
kk
(2)
3
3
D
(k +k
) . (21.2)
The momentum-conserving Dirac delta-function in equation (21.2) is a consequence of spatial translation
symmetry. Isotropy implies that the power spectrum P(k) is a function only of the absolute value k [k[
of the wavevector. For a statistically rotation invariant eld projected on the sky, such as the CMB, the
variance of spherical harmonic modes
m
T
m
/T denes the power spectrum C
m
) = 1
m,
m
C
(21.3)
where 1
m,
m
is the unit matrix in the Hilbert space of spherical harmonics (distinguish the three usages
of in this paragraph: meaning uctuation,
D
meaning Dirac delta-function, and meaning Kronecker
delta, as in the following equation)
1
m,

m,m
. (21.4)
Again, the angular momentum-preserving condition (21.4) that =
and m+m
= 0 is a consequence
of rotational symmetry. The same rotational symmetry implies that the power spectrum C
is a function
only of the harmonic number , not of the directional harmonic number m.
5. Reheating. Early Universe ination evidently came to an end. It is presumed that the vacuum energy
released by the decay of the inaton eld, an event called reheating, somehow eciently produced the
matter and radiation elds that we see today. After reheating, the Universe was dominated by relativistic
354 An overview of cosmological perturbations
elds, collectively called radiation. Reheating changed the evolution of the cosmic scale factor from
acceleration to deceleration, but is presumed not to have generated additional uctuations.
6. Photon-baryon uid and the sound horizon. Photon-electron (Thomson) scattering kept photons
and baryons tightly coupled to each other, so that they behaved like a relativistic uid. As long as the
radiation density exceeded the baryon density, which remained true up to near the time of recombination,
the speed of sound in the photon-baryon uid was
_
p/
_
1
3
of the speed of light. Fluctuations with
wavelengths outside the sound horizon grew by gravity. As time went by, the sound horizon expanded
in comoving radius, and uctuations thereby came inside the sound horizon. Once inside the sound
horizon, sound waves could propagate, which tended to decrease the gravitational potential. However,
each individual sound wave itself continued to oscillate, its oscillation amplitude T/T relative to the
background temperature T remaining approximately constant. The relativistic suppression of the potential
at small scales is responsible for the fact that the power spectrum of matter declines at small scales.
7. Acoustic peaks in the power spectrum. The oscillations of the photon-baryon uid produced the
characteristic pattern of peaks and troughs in the CMB power spectrum observed today. The same
peaks and troughs occur in the matter power spectrum, but are much less prominent, at a level of about
10% as opposed to the order unity oscillations observed in the CMB power spectrum. For adiabatic
uctuations, the amplitude of the temperature uctuations follows a pattern cos(k
s
) where
s
is the
comoving sound horizon. The nth peak occurs at a wavenumber k where k
s
n. In the observed
CMB power spectrum, the relevant value of the sound horizon
s
is its value
s,
at recombination.
Thus the wavenumber k of the rst peak of the observed CMB power spectrum occurs where k
s,
.
Two competing forces cause a mode to evolve: a gravitational force that amplies compression, and
a restoring pressure force that counteracts compression. When a mode enters the sound horizon for
the rst time, the compressing gravitational force beats the restoring pressure force, so the rst thing
that happens is that the mode compresses further. Consequently the rst peak is a compression peak.
This sets the subsequent pattern: odd peaks are compression peaks, while even peaks are rarefaction
peaks. The observed temperature uctuations of the CMB are produced by a combination of intrinsic
temperature uctuations, Doppler shifts, and gravitational redshifting out of potential wells. The Doppler
shift produced by the velocity of a perturbation is 90
out of phase with the temperature uctuation, and

so tends to ll in the troughs in the power spectrum of the temperature uctuation. This is the main
reason that the observed CMB power spectrum remains above zero at all scales.
8. Logarithmic growth of matter uctuations. Non-baryonic cold dark matter interacts weakly except
by gravity, and is needed to explain the observed clustering of matter in the Universe today in spite of the
small amplitude of temperature uctuations in the CMB. The adjective cold refers to the requirement
that the dark matter became non-relativistic (p = 0) at some early time. If the dark matter is both
non-baryonic and cold, then it did not participate in the oscillations of the photon-baryon uid. During
the radiation-dominated phase prior to matter-radiation equality, dark matter matter uctuations inside
the sound horizon grow logarithmically. The logarithmic growth translates into a logarithmic increase
in the amplitude of matter uctuations at small scales, and is a characteristic signature of non-baryonic
An overview of cosmological perturbations 355
cold dark matter. Unfortunately this signature is not readily discernible in the power spectrum of matter
today, because of nonlinear clustering.
9. Epoch of matter-radiation equality. The density of non-relativistic matter decreases more slowly than
the density of relativistic radiation. There came a point where the matter density equaled the radiation
density, an epoch called matter-radiation equality, after which the matter density exceeded the radiation
density. The observed ratio of the density of matter and radiation (CMB) today require that matter-
radiation equality occurred at a redshift of z
eq
3200, a factor of 3 higher in redshift than recombination
at z
1100. After matter-radiation equality, dark matter perturbations grew more rapidly, linearly
instead of just logarithmically with cosmic scale factor. A larger dark matter density causes matter-
radiation equality to occur earlier. The sound horizon at matter-radiation equality corresponds to a scale
roughly around the 2.5th peak in the CMB power spectrum. For adiabatic uctuations, the way that the
temperature and gravitational perturbations interact when a mode rst enters the sound horizon means
that the temperature oscillation is 5 times larger for modes that enter the horizon well into the radiation-
dominated epoch versus well into the matter-dominated epoch. The eect enhances the amplitude of
observed CMB peaks higher than 2.5 relative to those lower than 2.5. The observed relative strengths
of the 3rd versus the 2nd peak of the CMB power spectrum provides a measurement of the redshift of
matter-radiation equality, and direct evidence for the presence of non-baryonic cold dark matter.
10. Sound speed. The density of baryons decreased more slowly than the density of radiation, so that at
around recombination the baryon density was becoming comparable to the radiation density. The sound
speed
_
p/ depends on the ratio of pressure p, which was essentially entirely that of the photons, to the
density , which was produced by both photons and baryons. The sound speed consequently decreased
below
_
1
3
. Increasing the baryon-to-photon ratio at recombination has several observational eects on
the acoustic peaks of the CMB power spectrum, making it a prime measurable parameter from the CMB.
First, an increased baryon fraction increases the gravitational forcing (baryon loading), which enhances
the compression (odd) peaks while reducing the rarefaction (even) peaks. Second, increasing the baryon
fraction reduces the sound speed, which: (a) decreases the amplitude of the radiation dipole relative to the
radiation monopole, so increasing the prominence of the peaks; and (b) reduces the oscillation frequency
of the photon-baryon uid, which shifts the peaks to larger scales. The reduced sound speed also causes
an adiabatic reduction of the amplitudes of all modes by the square root of the sound speed, but this eect
is degenerate with an overall reduction in the initial amplitudes of modes produced by ination.
11. Recombination. As the temperature cooled below about 3,000 K, electrons combined with hydrogen
and helium nuclei into neutral atoms. This drastically reduced the amount of photon-electron scattering,
releasing the CMB to propagate almost freely. At the same time, the baryons were released from the
photons. Without radiation pressure to support them, uctuations in the baryons began to grow like the
dark matter uctuations.
12. Neutrinos. Probably all three species of neutrino have mass less than 0.3 eV and were therefore relativistic
up to and at the time of recombination. Each of the 3 species of neutrino had an abundance comparable
to that of photons, and therefore made an important contribution to the relativistic background and its
356 An overview of cosmological perturbations
uctuations. Unlike photons, neutrinos streamed freely, without scattering. The relativistic free-streaming
of neutrinos provided the main source of the quadrupole pressure that produces a non-vanishing dierence
between the scalar potentials. However, the neutrino quadrupole pressure was still only 10% of
the neutrino monopole pressure. To the extent that the neutrino quadrupole pressure can be approximated
as negligible, the neutrinos and their uctuations can be treated the same as photons.
13. CMB uctuations. The CMB uctuations seen on the sky today represent a projection of uctuations
on a thin but nite shell at a redshift of about 1100, corresponding to an age of the Universe of about
400,000 yr. The temperature, and the degrees of polarization in two dierent directions, provide 3 inde-
pendent observables at each point on the sky. The isotropy of the unperturbed radiation means that it is
most natural to measure the uctuations in spherical harmonics, which are the eigenmodes of the rotation
operator. Similarly, it is natural to measure the CMB polarization in spin harmonics.
14. Matter uctuations. After recombination, perturbations in the non-baryonic and baryonic matter grew
by gravity, essentially unaected any longer by photon pressure. If one or more of the neutrino types
had a mass small enough to be relativistic but large enough to contribute appreciable density, then its
relativistic streaming could have suppressed power in matter uctuations at small scales, but observations
show no evidence of such suppression, which places an upper limit of about an eV on the mass of the
most massive neutrino. The matter power spectrum measured from the clustering of galaxies contains
acoustic oscillations like the CMB power spectrum, but because the non-baryonic dark matter dominates
the baryons, the oscillations are much smaller.
15. Integrated Sachs-Wolfe eect. Variations in the gravitational potential along the line-of-sight to the
CMB aect the CMB power spectrum at large scales. This is called the integrated Sachs-Wolfe (ISW)
eect. If matter dominates the background, then the gravitational potential has the property that it
remains constant in time for (subhorizon) linear uctuations, and there is no ISW eect. In practice,
ISW eects are produced by at least three distinct causes. First, an early-time ISW eect is produced
by the fact that the Universe at recombination still has an appreciable component of radiation, and is
not yet wholly matter-dominated. Second, a late-time ISW eect is produced either by curvature or by a
cosmological constant. Third, a non-linear ISW eect is produced by non-linear evolution of the potential.
22
Cosmological perturbations in a at
Friedmann-Robertson-Walker background
For simplicity, this book considers only a at (not closed or open) Friedmann-Robertson-Walker background.
The comoving cosmological horizon size at recombination was much smaller than today, and consequently the
cosmological density was much closer to 1 at recombination than it is today. Since observations indicate
that the Universe today is within 1% of being spatially at, it is an excellent approximation to treat the
Universe at the time of recombination as being spatially at.
With some modications arising from cosmological expansion, perturbation theory on a at FRW back-
ground is quite similar to perturbation theory in at (Minkowski) space, Chapter 20.
The strategy is to start in a completely general gauge, and to discover how the conformal Newtonian
gauge, which is used in subsequent Chapters, emerges naturally as that gauge in which the perturbations
are precisely the physical perturbations.
22.1 Unperturbed line-element
It is convenient to choose the coordinate system x
= , x
i
to consist of conformal time together with
Cartesian comoving coordinates x x
i
x, y, z. The coordinate metric of the unperturbed background
FRW geometry is then
ds
2
= a()
2
_
d
2
+dx
2
+dy
2
+dz
2
_
, (22.1)
where a() is the cosmic scale factor. The unperturbed coordinate metric is thus the conformal Minkowski
metric
0
g
= a()
2
. (22.2)
The tetrad is taken to be orthonormal, with the unperturbed tetrad axes
m

0
.
1
,
2
,
3
being aligned
with the unperturbed coordinate axes g
, g
x
, g
y
, g
z
so that the unperturbed vierbein and inverse
vierbein are respectively 1/a and a times the unit matrix
0
e
m
=
1
a

m
,
0
e
m
= a
m
. (22.3)
358

Cosmological perturbations in a at Friedmann-Robertson-Walker background
Let overdot denote partial dierentiation with respect to conformal time ,
overdot

, (22.4)
so that for example a da/d. The coordinate time derivative / is to be distinguished from the directed
time derivative
0
e
0
/x
. Let
i
denote the comoving gradient
i

x
i
, (22.5)
which should be distinguished from the directed derivative
i
e
i
/x
.
22.2 Comoving Fourier modes
Since the unperturbed Friedmann-Robertson-Walker spacetime is spatially homogeneous and isotropic, it
is natural to work in comoving Fourier modes. Comoving Fourier modes have the key property that they
evolve independently of each other, as long as perturbations remain linear. Equations in Fourier space are
obtained by replacing the comoving spatial gradient
i
by i times the comoving wavevector k
i
(the choice
of sign is the standard convention in cosmology)
i
ik
i
. (22.6)
By this means, the spatial derivatives become algebraic, so that the partial dierential equations governing
the evolution of perturbations become ordinary dierential equations.
In what follows, the comoving gradient
i
will be used interchangeably with ik
i
, whichever is most
convenient.
22.3 Classication of vierbein perturbations
The tetrad-frame components
mn
of the vierbein perturbation of the FRW geometry decompose in much
the same way as in at Minkowski case into 6 scalars, 4 vectors (8 degrees of freedom), and 1 tensor (2
degrees of freedom) (the following equations are essentially the same as those (20.6) for the at Minkowski
background),
00
=
scalar
, (22.7a)
0i
=
i
w
scalar
+ w
i
vector
, (22.7b)
i0
=
i
w
scalar
+ w
i
vector
, (22.7c)
ij
=
ij

scalar
+
i
j
h
scalar
+
ijk
h
scalar
+
i
h
j
vector
+
j
h
i
vector
+ h
ij
tensor
. (22.7d)
The tetrad-frame components
m
of the coordinate shift of the coordinate gauge transformation (18.8)
similarly decompose into 2 scalars and 1 vector (the following equation is essentially the same as that (20.8)
for the at Minkowski background)
m
=
0
scalar
,
i
scalar
+
i
vector
. (22.8)
The vierbein perturbations
mn
transformunder a coordinate gauge transformation (18.8) as, equation (18.25),
00
+
1
a
scalar
, (22.9a)
0i

i
_
w +
1
a
_

a
a
_
_
scalar
+
_
w
i
+
1
a
_

a
a
_
i
_
vector
, (22.9b)
i0

i
_
w +
1
a
0
_
scalar
+ w
i
vector
, (22.9c)
ij

ij
_

a
a
2
0
_
scalar
+
i
j
_
h +
1
a
_
scalar
+
ijk
h
scalar
+
i
_
h
j
+
1
a
j
_
vector
+
j
h
i
vector
+ h
ij
tensor
, (22.9d)
or equivalently
+
1
a
, (22.10a)
w w +
1
a
_

a
a
_
, w
i
w
i
+
1
a
_

a
a
_
i
, (22.10b)
w w +
1
a
0
, w
i
w
i
, (22.10c)

a
a
2
0
, h h +
1
a
,

h
h , h
i
h
i
+
1
a
i
,

h
i

h
i
, h
ij
h
ij
. (22.10d)
Eliminating the coordinate shift
m
from the transformations (22.10) yields 12 coordinate gauge-invariant
combinations of the perturbations

_

+
a
a
_
w , w

h , w
i

h
i
, w
i
, +
a
a
w ,

h ,

h
i
, h
ij
. (22.11)
Six combinations of these coordinate gauge-invariant perturbations depend only on the symmetric part
mn
+
nm
of the vierbein perturbations, and are therefore tetrad gauge-invariant as well as coordinate
gauge-invariant. These 6 coordinate and tetrad gauge-invariant perturbations comprise 2 scalars, 1 vector,
360

and 1 tensor
scalar

_

+
a
a
_
(w + w

h) , (22.12a)
scalar
+
a
a
(w + w

h) , (22.12b)
W
i
vector
w
i
+ w
i

h
i
h
i
, (22.12c)
h
ij
tensor
. (22.12d)
22.4 Metric, tetrad connections, and Einstein tensor
This section gives expressions in a completely general gauge for perturbed quantities in the at Friedmann-
Robertson-Walker background geometry.
The perturbed coordinate metric g
is
g
= a
2
(1 + 2 ) , (22.13a)
g
i
= a
2
_
i
(w + w) + (w
i
+ w
i
)
, (22.13b)
g
ij
= a
2
_
(1 2 )
ij
2
i
j
h
i
(h
j
+

h
j
)
j
(h
i
+

h
i
) 2 h
ij
. (22.13c)
The coordinate metric is tetrad gauge-invariant, but not coordinate gauge-invariant.
The perturbed tetrad connections
kmn
are
0i0
=
1
a
_
i
_

_

+
a
a
_
w
_
+
_

+
a
a
_
w
i
_
, (22.14a)
0ij
=
1
a
_
_
a
a
+F
_
ij

i
j
(w

h)
1
2
(
i
W
j
+
j
W
i
) +
j
w
i
+

h
ij
_
, (22.14b)
ij0
=
1
a
_
1
2
(
i
W
j
j
W
i
)

(
ijl
h
i
h
j
+
j
h
i
)
_
, (22.14c)
ijk
=
1
a
_
(
jk
ik
j
)
_
+
a
a
w
_
a
a
(
ik
jl
jk
il
) w
l
k
(
ijl
h
i
h
j
+
j
h
i
) +
i
h
jk
j
h
ik
_
, (22.14d)
where F is dened by
F
a
a
+

. (22.15)
Being purely tetrad-frame quantities, the tetrad connections are automatically coordinate gauge-invariant,
22.4 Metric, tetrad connections, and Einstein tensor 361
but they are not tetrad gauge-invariant. The quantity F dened by equation (22.15) is not coordinate gauge-
invariant, but the combination a/a
2
+ F/a that appears in the expression (22.14b) for
0ij
is coordinate
and tetrad gauge-invariant.
Exercise 22.1 Coordinate gauge invariance of a/a
2
+ F/a. Argue that under a coordinate gauge
transformation of the conformal time,
= +
, the cosmic scale factor a() and its derivative

a da/d transforms as (see 18.8)
a a +L
a = a a
= a +
a
a
0
, a a +L
a = a a
= a +
a
a
0
. (22.16)
Check that this behaviour is consistent with the gauge transformation (18.28) of g
, equation (22.13a).
Hence show that a/a
2
+F/a is coordinate gauge invariant. THIS IS NOT WORKING.
The perturbed tetrad-frame Einstein tensor G
mn
is
G
00
=
1
a
2
_
_
3
a
2
a
2
6
a
a
F + 2
2
scalar
_
_
, (22.17a)
G
0i
=
1
a
2
_
_
2
i
_
F +
_
a
a
2
a
2
a
2
_
w
_
scalar
+
1
2
2
W
i
+ 2
_
a
a
2
a
2
a
2
_
w
i
vector
_
_
, (22.17b)
G
ij
=
1
a
2
_
_
_
2
a
a
+
a
2
a
2
+ 2
_

+ 2
a
a
_
F + 2
_
a
a
2
a
2
a
2
_
ij
scalar
(
i
ij
2
)( )
scalar
+
1
2
_

+ 2
a
a
_
(
i
W
j
+
j
W
i
)
vector
_

2
2
+ 2
a
a

2
_
h
ij
tensor
_
_
. (22.17c)
Being tetrad-frame quantities, all components of the tetrad-frame Einstein tensor are automatically coordi-
nate gauge-invariant. The time-time G
00
and space-space G
ij
components are not only coordinate but also
tetrad gauge-invariant, as follows from the fact that these components depend only on symmetric combi-
nations of the vierbein potentials. Specically, the quantities 3( a
2
/a
4
) 6( a/a
3
)F on the right hand side
of equation (22.17a) for G
00
, and the coecient of
ij
(including the overall 1/a
2
factor) on the right hand
side of equation (22.17c) for G
ij
, are coordinate and tetrad gauge-invariant. However, the time-space com-
ponents G
0i
are not tetrad gauge-invariant, as is evident from the fact that equation (22.17b) involves the
non-tetrad-gauge-invariant perturbations w and w
i
. Physically, under a tetrad boost by a velocity v of linear
order, the time-space components G
0i
change by rst order v, but G
00
and G
ij
change only to second order
v
2
. Thus to linear order, only G
0i
changes under a tetrad boost. Note that G
0i
changes under a tetrad boost
( w and w
i
), but not under a tetrad rotation (
h and

h
i
).
362

22.5 ADM gauge choices
The ADM (3+1) formalism, Chapter 13, chooses the tetrad time axis
0
to be orthogonal to hypersurfaces
of constant time, = constant, equivalent to requiring that the tetrad time axis be orthogonal to each of
the spatial coordinate axes,
0
g
i
= 0, equation (13.2). The ADM choice is equivalent to setting
w = w
i
= 0 . (22.18)
The ADM choice simplies the tetrad-frame connections (22.14) and the time-space component G
0i
of the
tetrad-frame frame Einstein tensor, equation (22.17b).
Another gauge choice that signicantly simplies the tetrad connections (22.14), though does not aect
the Einstein tensor (22.17), is
h =

h
i
= 0 . (22.19)
If the wavevector k is taken along the coordinate z-direction, then the gauge choice

h
i
= 0 is equivalent to
choosing the tetrad 3-axis (z-axis)
3
to be orthogonal to the coordinate x and y-axes,
3
g
x
=
3
g
y
= 0.
The gauge choice

h = 0 is equivalent to rotating the tetrad axes about the 3-axis (z-axis) so that
1
g
y
=
2
g
x
.
22.6 Conformal Newtonian gauge
Conformal Newtonian gauge sets
w = w = w
i
= h =

h = h
i
=

h
i
= 0 , (22.20)
so that the retained perturbations are the 6 coordinate and tetrad gauge-invariant perturbations (22.12)
scalar
= , (22.21a)
scalar
= , (22.21b)
W
i
vector
= w
i
, (22.21c)
h
ij
tensor
. (22.21d)
In conformal Newtonian gauge, the scalar perturbations of the Einstein equations are the energy density,
energy ux, monopole pressure, and quadrupole pressure equations
3
a
a
F k
2
= 4Ga
2
1
T
00
, (22.22a)
ikF = 4Ga
2

k
i
1
T
0i
, (22.22b)
F + 2
a
a
F +
_
a
a
2
a
2
a
2
_
k
2
3
( ) =
4
3
Ga
2
ij
1
T
ij
, (22.22c)
k
2
( ) = 8Ga
2
_
3
2
k
i
k
j

1
2

ij
_
1
T
ij
, (22.22d)
where F is the coordinate and tetrad gauge-invariant quantity
F
a
a
+

. (22.23)
All 4 of the scalar Einstein equations (22.22) are expressed in terms of gauge-invariant variables, and are
therefore fully gauge-invariant.
The energy and momentum equations (22.22a) and (22.22b) can be combined to eliminate F, yielding an
equation for alone
k
2
= 4a
2
_
1
T
00
+ 3
a
a
k
i
ik
1
T
0i
_
. (22.24)
The quantity in parentheses on the right hand side of equation (22.24) is the source for the scalar potential
, and can be interpreted as a measure of the true energy uctuation. This is however just a matter of
interpretation: the individual perturbations
1
T
00
and
1
T
01
are both individually gauge-invariant, and therefore
have physical meaning.
If the energy-momentum tensors of the various matter components are arranged so as to conserve overall
energy-momentum, as they should, then 2 of the 4 equations (22.22a)(22.22d) are redundant, since they
serve simply to enforce conservation of energy and scalar momentum. Thus it suces to take, as the equations
governing the potentials and , any 2 of the equations (22.22a)(22.22d). One is free to retain whichever 2
of the equations is convenient. Usually the 1st equation, the energy equation (22.22a), and the 4th equation,
the quadrupole pressure equation (22.22d), are most convenient to retain. But sometimes the 2nd equation,
the scalar momentum equation (22.22b), is more convenient in place of the energy equation (22.22a).
22.7 Synchronous gauge
One gauge that remains in common use in cosmology, but is not used here, is synchronous gauge, discussed
in the case of Minkowski background space in 20.10. The cosmological synchronous gauge choices are the
same as for the Minkowski background, equations (20.63) and (20.64):
= w = w = w
i
= w
i
=

h =

h
i
= 0 . (22.25)
The gauge-invariant perturbations (22.12) in synchronous gauge are
scalar
=

h +
a
a
h , (22.26a)
scalar
=
a
a
h , (22.26b)
W
i
vector
=
h
i
, (22.26c)
h
ij
tensor
. (22.26d)
23
Cosmological perturbations: a simplest set of
assumptions
1. Consider only scalar modes.
2. Consider explicitly only two species: non-baryonic cold dark matter, and radiation consisting of photons
and neutrinos lumped together. Neglect the contribution of baryons to the mass density.
3. Treat the radiation as almost isotropic, so it is dominated by its rst two moments, the monopole and
dipole. In practice, electron-photon scattering keeps photons almost isotropic. Unlike photons, neutrinos
stream freely, but they inherit an approximately isotropic distribution from an early time when they were
in thermodynamic equilibrium.
4. Include damping from photon-electron (Thomson) scattering by allowing the radiation a small quadrupole
moment, the diusion approximation.
23.1 Perturbed FRW line-element
Perturbed FRW line-element in conformal Newtonian gauge
ds
2
= a
2
_
(1 + 2)d
2
+
ij
(1 2)dx
i
dx
j
, (23.1)
where a() is the cosmic scale factor, a function only of conformal time .
23.2 Energy-momenta of ideal uids
In the simplest approximation, matter, radiation, and dark energy can each be treated as ideal uids. The
energy-momentum tensor of an ideal uid with proper density and isotropic pressure p in its own rest
frame, and moving with bulk 4-velocity u
m
relative to the conformal Newtonian tetrad frame, is
T
mn
= ( +p)u
m
u
n
+p
mn
. (23.2)
In the situation under consideration, the uids satisfy equations of state
p = w (23.3)
23.2 Energy-momenta of ideal uids 365
with w constant. Specically, w = 0 for non-relativistic matter, w = 1/3 for relativistic radiation, and
w = 1 for dark energy with constant density. Furthermore, all the uids are moving with non-relativistic
bulk velocities, including the radiation, which is almost isotropic, and therefore has a small bulk velocity
even though the individual particles of radiation move at the speed of light. The bulk tetrad-frame 4-velocity
u
m
is thus, to linear order
u
m
= 1, v
i
, (23.4)
where v
i
is the non-relativistic spatial bulk 3-velocity (the spatial tetrad metric is Euclidean, so v
i
= v
i
).
The density can be written in terms of the unperturbed density and a uctuation dened by
= [1 + (1 +w)] . (23.5)
The factor 1+w is included in the denition of the uctuation because it simplies the resulting perturbation
equations (23.12). As you will discover in Exercise 23.1, the uctuation can be interpreted physically as
the entropy uctuation,
=
1
1 +w

=
s
s
. (23.6)
For matter, w = 0, the entropy uctuation coincides with the density uctuation, = / . For dark
energy, w = 1, the density uctuation is necessarily zero, / = 0. To linear order in the velocity v
i
, the
tetrad-frame energy-momentum tensor (23.2) of the ideal uid is then
T
00
[1 + (1 +w)] , (23.7a)
T
0i
(1 +w) v
i
, (23.7b)
T
ij
= w [1 + (1 +w)]
ij
. (23.7c)
To linear order in the uctuation , velocity v
i
, and potentials and , and in conformal Newtonian gauge,
conservation of energy and momentum requires
D
m
T
m0
=
1
a
_
+ (1 +w)
i
(v
i
) + 3(1 +w)
_
a
a

_
_
= 0 , (23.8a)
D
m
T
mi
=
1
a
_
(1 +w)
v
i
+ 4(1 +w)
a
a
v
i
+w
i
+ (1 +w)
i
_
= 0 . (23.8b)
The energy conservation equation (23.8a) has an unperturbed part,
D
m
0
T
m0
=
1
a
_

+ 3(1 +w)
a
a
_
= 0 , (23.9)
which implies the usual result that the mean density evolves as a power law with cosmic scale factor,
a
3(1+w)
. Subtracting appropriate amounts of the unperturbed energy conservation equation (23.9)
from the perturbed energy-momentum conservation equations (23.8) yields equations for the uctuation
366 Cosmological perturbations: a simplest set of assumptions
and velocity v
i
:
+
i
v
i
= 3

, (23.10a)
v
i
+ (1 3w)
a
a
v
i
+w
i
=
i
. (23.10b)
Now decompose the 3-velocity v
i
into its scalar v and vector v
,i
parts. Up to this point, the scalar part of
a vector has been taken to be the gradient of a potential. But here it is advantageous to absorb a factor of k
into the denition of the scalar part v of the velocity, so that instead of v
i
= ik
i
v + v
,i
in Fourier space,
the velocity is given in Fourier space by
v
i
= i
k
i
v + v
,i
. (23.11)
The advantage of this choice is that v is dimensionless, as are and and . The scalar parts of the
perturbation equations (23.10) are then
kv = 3

, (23.12a)
v + (1 3w)
a
a
v +wk = k . (23.12b)
Combining the two equations (23.12) for the uctuation and velocity v yields a second-order dierential
equation for 3,
_
d
2
d
2
+ (1 3w)
a
a
d
d
+k
2
w
_
( 3) = k
2
( + 3w) . (23.13)
For positive w, equation (23.13) is a wave equation for a damped, forced oscillator with sound speed
w. The
resulting generic behaviour for the particular cases of matter (w = 0) and radiation (w =
1
3
) is considered
in 23.6 and 23.7 below.
A more careful treatment, deferred to Chapter 24, accounts for the complete momentum distribution of
radiation by expanding the temperature perturbation T/
T in multipole moments, equation (24.48).

The radiation uctuation
r
and scalar bulk velocity v
r
are related to the rst two multipole moments of
the temperature perturbation, the monopole
0
and the dipole
1
, by
r
= 3
0
, (23.14a)
v
r
= 3
1
. (23.14b)
The factor of 3 arises because the unperturbed radiation distribution is in thermodynamic equilibrium, for
which the entropy density is s T
3
, so
r
= 3T/
T.
The energy-momentum perturbation
1
T
mn
that goes into the Einstein equations (22.22) are, from equa-
tions (23.7) with the unperturbed part subtracted,
1
T
00
(1 +w) , (23.15a)
1
T
0i
(1 +w) v
i
, (23.15b)
1
T
ij
= w(1 +w)
ij
. (23.15c)
23.3 Diusive damping 367
Exercise 23.1 Entropy uctuation. The purpose of this exercise is to discover that the uctuation
dened by equation (23.5) can be interpreted as the entropy uctuation. According to the rst law of
thermodynamics, the entropy density s of a uid of energy density , pressure p, and temperature T in a
volume V satises
d(V ) +pdV = Td(sV ) . (23.16)
If the uid is ideal, so that , p, T, and s are independent of volume V , then integrating the rst law (23.16)
implies that
V +pV = TsV . (23.17)
This implies that the entropy density s is related to the other variables by
s =
+p
T
. (23.18)
Show that, for ideal uid with equation of state p/ = w = constant, the rst law (23.16) together with the
expression (23.18) for entropy implies that
T
w/(1+w)
, (23.19)
and hence
s
1/(1+w)
. (23.20)
Conclude that small variations of the entropy and density are related by
s
s
=
1
1 +w
, (23.21)
conrming equation (23.6). [Hint: Do not confuse what is being asked here with adiabatic expansion. The
results (23.19) are properties of the uid, independent of whether the uid is changing adiabatically. For
adiabatic expansion, the uid satises the additional condition sV = constant.]
23.3 Diusive damping
The treatment of matter and radiation as ideal uids misses a feature that has a major impact on observed
uctuations in the CMB, namely the diusive damping of sound waves that results from the nite mean
free path to electron-photon scattering. As recombination approaches, the scattering mean free path length-
ens, until at recombination photons are able to travel freely across the Universe, ready to be observed by
astronomers. The damping is greater at smaller scales, and is responsible for the systematic decrease in the
CMB power spectrum to smaller scales.
As expounded in Chapter 24, in 24.13 and following, the damping can be taken into account to lowest
order, the diusion approximation, by admitting a small quadrupole moment
2
to the photon distribu-
tion. A detailed analysis of the collisional Boltzmann equation for photons, 24.12, reveals that the photon
quadrupole
2
is given in the diusion approximation by equation (24.94). There is an additional source of
damping that arises from viscous baryon drag, 24.15, but this eect vanishes in the limit of small baryon
density, and is neglected in the present Chapter.
The diusive damping resulting from a small photon quadrupole conserves the energy and momentum
of the photon uid, so that covariant momentum conservation D
m
T
mn
= 0 continues to hold true within
the photon uid. By contrast, viscous baryon drag, 24.15, neglected in this Chapter, transfers momentum
between photons and baryons.
Dene the dimensionless quadrupole q by
T
ij
quadrupole
= (1 +w) q
_
3
2
k
i
k
j

1
2

ij
_
, (1 +w) q
_
k
i
k
j

1
3

ij
_
T
ij
. (23.22)
For photons, the dimensionless quadrupole q is related to the photon quadrupole harmonic
2
by, equa-
tion (24.55d),
q = 2
2
. (23.23)
In the presence of a quadrupole pressure, the energy conservation equation (23.8a) is unchanged, but the
momentum conservation equation (23.8b) is modied by the change +q:
D
m
T
mi
=
1
a
_
(1 +w)
v
i
+ 4(1 +w)
a
a
v
i
+w
i
+ (1 +w)
i
( +q)
_
= 0 . (23.24)
The consequent equations (23.10b), (23.12b), (23.13) are similarly modied by + q. In particular,
the velocity equation (23.12b) is modied to
v + (1 3w)
a
a
v +wk = k( + q) . (23.25)
23.4 Equations for the simplest set of assumptions
Non-baryonic cold dark matter, subscripted c:
c
k v
c
= 3

, (23.26a)
v
c
+
a
a
v
c
= k . (23.26b)
Radiation, which includes both photons and neutrinos:
0
k
1
=

, (23.27a)
1
+
k
3

0
=
k
3
(2
2
) , (23.27b)
2
=
4k
9 n
e
T
a
1
. (23.27c)
23.4 Equations for the simplest set of assumptions 369
Einstein energy and quadrupole pressure equations:
k
2
3
a
a
F = 4Ga
2
(
c
c
+ 4
r
0
) , (23.28a)
k
2
() = 32Ga
2

r
2
, (23.28b)
where
F
a
a
+

. (23.29)
In place of the Einstein energy equation (23.28a) it is sometimes convenient to use the Einstein momentum
equation
kF = 4Ga
2
(
c
v
c
+ 4
r
1
) , (23.30)
which, because the matter and radiation equations (23.26) and (23.27) already satisfy covariant energy-
momentum conservation, is not an independent equation.
The radiation quadrupole
2
, equation (23.27c), derived in Chapter 24, equation (24.94), is proportional
to the comoving mean free path l
T
to electron-photon (Thomson) scattering,
l
T

1
n
e
T
a
, (23.31)
where
T
is the Thomson cross-section. The quadrupole becomes important only near recombination, where
the increasing mean free path to electron-photon scattering leads to dissipation of photon-baryon sound
waves. In the simple treatment of this Chapter, neutrinos are being lumped with photons, and of course
neutrinos do not scatter, but rather stream freely. However, radiation is gravitationally sub-dominant near
recombination, so not much error arises from treating neutrinos as gravitationally the same as photons near
recombination. To make comparison with observed CMB uctuations, the important thing is to follow
the evolution of the photon multipoles, and for this purpose the radiation quadrupole dened by equa-
tion (23.27c), without any correction for neutrino-to-photon ratio, is the appropriate choice.
In much of the remainder of this Chapter, that is, excepting in 23.7 and 23.14, the radiation quadrupole
will be set to zero,
2
= 0 , (23.32)
which is equivalent to neglecting the eect of damping. If the radiation quadrupole vanishes, then the
Einstein quadrupole pressure equation (23.28b) implies that the scalar potentials and are equal,
= . (23.33)
In any case, the radiation quadrupole is always small, so that to a good approximation.
23.5 Unperturbed background
In the unperturbed background, the unperturbed dark matter density
c
and radiation density
r
evolve
with cosmic scale factor as

c
a
3
,
r
a
4
. (23.34)
The Hubble parameter H is dened in the usual way to be
H
1
a
da
dt
=
a
a
2
, (23.35)
in which overdot represents dierentiation with respect to conformal time, a da/d. The Friedmann
equations for the background imply that the Hubble parameter for a universe dominated by dark matter
and radiation is
H
2
=
8G
3
(
c
+
r
) =
H
2
eq
2
_
a
3
eq
a
3
+
a
4
eq
a
4
_
(23.36)
where a
eq
and H
eq
are the cosmic scale factor and the Hubble parameter at the time of matter-radiation
equality,
c
=
r
.
The comoving horizon distance is dened to be the comoving distance that light travels starting from
zero expansion. This is
=
_
a
0
da
a
2
H
=
2
2
a
eq
H
eq
__
1 +
a
a
eq
1
_
=
2
2
a
eq
H
eq
_
a/a
eq
1 +
_
1 +a/a
eq
_
. (23.37)
In the radiation- and matter-dominated epochs respectively, the comoving horizon distance is
=
_
2
a
eq
H
eq
_
a
a
eq
_
a radiation-dominated ,
2
2
a
eq
H
eq
_
a
a
eq
_
1/2
a
1/2
matter-dominated .
(23.38)
The ratio of the comoving horizon distance to the comoving cosmological horizon distance 1/(aH) is
aH =
2
_
1 +a/a
eq
1 +
_
1 +a/a
eq
, (23.39)
which is evidently a number of order unity, varying between 1 in the radiation-dominated epoch a a
eq
,
and 2 in the matter-dominated epoch a a
eq
.
Exercise 23.2 Matter-radiation equality.
1. Argue that the redshift z
eq
of matter-radiation equality is given by
1 +z
eq
=
a
0
a
eq
= ?
m
h
2
, (23.40)
23.6 Generic behaviour of non-baryonic cold dark matter 371
where
m
is the matter density today relative to critical. What is the factor, and what is its numerical
value? The factor depends on the energy-weighted eective number of relativistic species g
, Exercise 10.18.
Should this g
be that now, or that at matter-radiation equality?

2. Show that the ratio H
eq
/H
0
of the Hubble parameter at matter-radiation equality to that today is
H
eq
H
0
=
_
2
m
(1 +z
eq
)
3/2
. (23.41)
Solution. The redshift z
eq
of matter-radiation equality is given by
1 +z
eq
=

m
r
=
45c
5
m
H
2
0
4
3
Gg
(kT
0
)
4
= 8.093 10
4

m
h
2
g
= 3200
_
g
3.36
_
1
_
m
h
2
0.133
_
, (23.42)
where T
0
= 2.725 K is the present-day CMB temperature, and g
= 2 + 6
7
8
_
4
11
_
4/3
= 3.36 is the energy-
weighted eective number of relativistic species at matter-radiation equality.
23.6 Generic behaviour of non-baryonic cold dark matter
Combining equations (23.26) for the dark matter overdensity and velocity gives
_
d
2
d
2
+
a
a
d
d
_
(
c
3) = k
2
= k
2
, (23.43)
where the last expression follows because = to a good approximation. In the absence of a driving
potential, = 0, the dark matter velocity would redshift as v
c
1/a, and the dark matter density equa-
tion (23.26a) would then imply that

c
= kv
c
a
1
. In the radiation-dominated epoch, where a,
this leads to a logarithmic growth in the overdensity
c
, even though there is no driving potential, and the
velocity is redshifting to a halt. In the matter-dominated epoch, where a
1/2
, the dark matter overdensity
c
would freeze out at a constant value, in the absence of a driving potential.
Exercise 23.3 Generic behaviour of dark matter. Find the homogeneous solutions of equation (23.43).
Hence nd the retarded Greens function of the equation. Write down the general solution of equation (23.43)
as an integral over the Greens function.
Solution. The general solution of equation (23.43) is
c
(a)3(a) = A
0
+A
1
ln
_
1 +a + 1
1 +a 1
_
+2k
2
_
a
0
ln
_
(
1 +a + 1)
(
1 +a 1)
(
1 +a
1)
(
1 +a
1)
_
(a
)
a
da
1 +a
, (23.44)
where A
0
and A
1
are constants.
23.7 Generic behaviour of radiation
Combining equations (23.27) for the radiation monopole, dipole, and quadrupole gives
_
3
d
2
d
2
+ 2
3
d
d
+k
2
_
(
0
) = k
2
( + ) = 2 k
2
, (23.45)
where in the last expression the approximation = has again been invoked. The coecient of the linear
derivative term in equation (23.45) is a damping coecient

4k
2
l
T
9
3
, (23.46)
where l
T
is the comoving electron-photon scattering (Thomson) mean free path, equation (23.31), The mean
free path is small, Exercise 23.6, except near recombination. In the absence of a driving potential, = 0,
and in the absence of damping, = 0, the radiation oscillates as
0
e
i
with frequency =
_
1
3
k. In
other words, the solutions are sound waves, moving at the sound speed
c
s
=

k
=
_
1
3
. (23.47)
Dene the conformal sound time by
s
c
s
=

3
. (23.48)
In terms of the conformal sound time
s
, the dierential equation (23.45) becomes
_
d
2
d
2
s
+ 2
d
d
s
+k
2
_
(
0
) = 2k
2
. (23.49)
As you will discover in Exercises 23.7 and 23.5, equation (23.49) describes damped oscillations forced by the
potential on the right hand side.
Exercise 23.4 Generic behaviour of radiation. Find the homogeneous solutions of equation (23.49)
in the case of zero damping, = 0. Hence nd the retarded Greens function of the equation. Write down
the general solution of equation (23.49) as an integral over the Greens function. Convince yourself that
0
oscillates about 2.
Solution. The general solution of equation (23.49) is, with k
s
,
0
() () = B
0
cos +B
1
sin 2
_

0
sin(
)(
) d
, (23.50)
where B
0
and B
1
are constants.
Exercise 23.5 Behaviour of radiation in the presence of damping. Now suppose that there is a
small damping coecient, k. Try a solution of the form
0
e
R
ds
in equation (23.49). Suppose
that the frequency changes slowly over a period,

2
, so that
can be set to zero. Show that

23.8 Regimes 373
the homogeneous solutions of equation (23.49) are approximately
0
e
R
ds iks
. Hence nd the
retarded Greens function, and write down the general solution to equation (23.49).
Solution. See 24.17. The general solution of equation (23.49) is, with k
s
and
_
d
s
,
0
() () = e
(B
0
cos +B
1
sin) 2
_

0
e
(
)
sin(
)(
) d
, (23.51)
where B
0
and B
1
are constants.
23.8 Regimes
In the remainder of this Chapter, approximate analytic solutions are developed that describe the evolution
of perturbations in the matter and radiation in various regimes. The regimes are:
1. Superhorizon scales, 23.9.
2. Radiation-dominated:
a. adiabatic initial conditions, 23.10;
b. isocurvature initial conditions, 23.11.
3. Subhorizon scales, 23.12.
4. Matter-dominated, 23.13.
5. Recombination 23.14.
6. Post-recombination 23.15.
7. Matter with dark energy 23.16.
8. Matter with dark energy and curvature 23.17.
23.9 Superhorizon scales
At suciently early times, any mode is outside the horizon, k < 1. In the superhorizon limit k 1, the
evolution equations (23.27)(23.28) reduce to
c
= 3

, (23.52a)
0
=

, (23.52b)
3
a
a
F = 4Ga
2
(
c
c
+ 4
r
0
) . (23.52c)
The rst two of these equations evidently imply that the dark matter overdensity
c
and radiation monopole
0
are related to the potential by
c
= 3 + constant , (23.53a)
0
= + constant . (23.53b)
a
eq
a
rec
Cosmic scale factor
C
o
m
o
v
i
n
g
d
i
s
t
a
n
c
e
superhorizon
radiation
r
a
d
i
a
t
i
o
n
b
a
c
k
g
r
o
u
n
d
m
a
t
t
e
r
f
l
u
c
t
u
a
t
i
o
n
matter
h
o
r
i
z
o
n
Figure 23.1 Various regimes in the evolution of uctuations. The line increasing diagonally from bottom
left to top right is the comoving horizon distance . Above this line are superhorizon uctuations, whose
comoving wavelengths exceed the horizon distance, while below the line are subhorizon uctuations, whose
comoving wavelengths are less than the horizon distance. The vertical line at cosmic scale factor aeq
a0/3200 marks the moment of matter-radiation equality. Before matter-radiation equality (to the left), the
background mass-energy is dominated by radiation, while after matter-radiation equality (to the right), the
background mass-energy is dominated by matter. Once a uctuation enters the horizon, the non-baryonic
matter uctuation tends to grow, whereas the radiation uctuation tend to decay, so there is an epoch
prior to matter-radiation equality where gravitational perturbations are dominated by matter rather than
radiation uctuations, even though radiation dominates the background energy density. The vertical line
at a a0/1100 marks recombination, where the temperature has cooled to the point that baryons change
from being mostly ionized to mostly neutral, and the Universe changes from being opaque to transparent.
The observed CMB comes from the time of recombination.
In eect, the dark matter velocity v
c
and radiation dipole
1
are negligibly small at superhorizon scales,
v
c
=
1
= 0 . (23.54)
Plugging the solutions (23.53) into the Einstein energy equation (23.52c), and replacing derivatives with
respect to conformal time with derivatives with respect to cosmic scale factor a,
= a

a
= a
2
H

a
, (23.55)
23.9 Superhorizon scales 375
10
3
10
2
10
1
1 10
10
2
10
3
.0
.5
1.0
Cosmic scale factor a/ a
eq
(
l
a
t
e
)
isocurvature
adiabatic
Figure 23.2 Evolution of the scalar potential at superhorizon scales, from radiation-dominated to matter-
dominated. The scale for the potential is normalized to its value (late) at late times a aeq.
with the Hubble parameter H from equation (23.36) gives the rst order dierential equation, in units
a
eq
= H
eq
= 1,
2a(1 +a)
+ (6 + 5a) + 4C
0
+C
1
a = 0 , (23.56)
where prime

denotes dierentiation with respect to cosmic scale factor, d/da, and the constants C
0
and C
1
are
C
0
=
0
(0) (0) , C
1
=
c
(0) 3(0) . (23.57)
The constants C
0
and C
1
are set by initial conditions. There are adiabatic and isocurvature initial conditions.
Ination generically produces adiabatic uctuations, in which matter and radiation uctuate together
c
(0) = 3
0
(0) =
3
2
(0) adiabiatic . (23.58)
Notice that a positive energy uctuation corresponds to a negative potential, consistent with Newtonian
intuition. Isocurvature initial conditions are dened by the vanishing of the initial potential, (0) = 0. This,
together with equations (23.56) and (23.57), implies the isocurvature initial conditions
(0) =
0
(0) = 0 ,
c
(0) = 8
(0) = 8
0
(0) isocurvature . (23.59)
Subject to the condition that remains nite at a 0, the adiabatic and isocurvature solutions to equa-
tion (23.56) are, in units a
eq
= 1,
ad
=
(0)
10
_
9 +
2
a

8
a
2

16
a
3
+
16
1 + a
a
3
_
=
(0)
10
_
3(14 + 9a) + (38 + 9a)
1 +a
_
1 +
1 +a
_
3
, (23.60a)
iso
=
8
(0)
5
_
1
2
a
+
8
a
2

16
a
3
+
16
1 +a
a
3
_
=
8
(0)
5
a
_
6 +a + 4
1 +a
_
_
1 +
1 +a
_
4
, (23.60b)
in which the last expressions in each case are written in a form that is numerically well-behaved for all a.
Figure 23.2 shows the evolution of the potential from equations (23.60), normalized to the value (late)
at late times a a
eq
. For adiabatic uctuations, the potential changes by a factor of 9/10 from initial to
nal value, while for isocurvature uctuations the potential evolves from zero to a nal value of
8
5
(0):
ad
(late) =
9
10
(0) =
3
5
c
(0) adiabatic , (23.61a)
iso
(late) =
8
5
(0) =
1
5
c
(0) isocurvature . (23.61b)
The superhorizon solutions (23.53) for the dark matter overdensity
c
and radiation monopole
0
are,
c
= 3
0
= 3
_
ad
3
2
(0)
_
adiabatic , (23.62a)
c
= 3
iso
8
(0) ,
0
=
iso
isocurvature . (23.62b)
23.10 Radiation-dominated, adiabatic initial conditions
For adiabatic initial conditions, uctuations that enter the horizon before matter-radiation equality, k
eq

1, are dominated by radiation. In the regime where radiation dominates both the unperturbed energy and
its uctuations, the relevant equations are, from equations (23.27), (23.28), and (23.30),
0
k
1
=

, (23.63a)
k
2
3
a
a
F = 16Ga
2

r
0
, (23.63b)
kF = 16Ga
2

r
1
, (23.63c)
in which, because it simplies the mathematics, the Einstein momentum equation is used as a substitute
for the radiation dipole equation. In the radiation-dominated epoch, the horizon is proportional to the
cosmic scale factor, a, equation (23.38). Inserting
0
and
1
from the Einstein energy and momentum
equations (23.63b) and (23.63c) into the radiation monopole equation (23.63a) gives a second order dierential
equation for the potential
+
4
+
k
2
3
= 0 . (23.64)
23.10 Radiation-dominated, adiabatic initial conditions 377
0 1 2 3 4 5 6 7 8
2.0
1.5
1.0
.5
.0
.5
1.0
1.5
2.0
2.5
3.0
k
s
/
a
n
d
adiabatic initial conditions

large scales (k
s,eq
<< 1)
small scales (k
s,eq
>> 1)
0 1 2 3 4 5 6 7 8
.5
.0
.5
1.0
k
s
/
a
n
d
isocurvature initial conditions

large scales (k
s,eq
<< 1)
small scales (k
s,eq
>> 1)
Figure 23.3 The dierence 0 between the radiation monopole and the Newtonian scalar potential
oscillates about 2, in accordance with equation (23.45). The dierence (0 ) (2) = 0 + ,
which is the temperature 0 redshifted by the potential , is the monopole contribution to temperature
uctuation of the CMB. The top panel is for adiabatic initial conditions, equations (23.58), while the bottom
panel is for isocurvature initial conditions, equation (23.59). The units of and 0 are such that (0) = 1
for adiabatic uctuations, and c(0) = 1 for isocurvature uctuations. In each case, the thin lines show
the evolution of small scale uctuations, which enter the horizon during the radiation-dominated epoch well
before matter-radiation equality, while the thick lines show the evolution of large-scale uctuations, which
enter the horizon during the matter-dominated epoch well after matter-radiation equality.
Equation (23.64) describes damped sound waves moving at sound speed
_
1
3
times the speed of light. The
sound horizon, the comoving distance that sound can travel, is /
3, the horizon distance multiplied by

the sound speed. The growing and decaying solutions to equation (23.64) are
grow
=
3j
1
()
=
3(sin cos )
3
,
decay
=
j
2
()
=
cos +sin
3
, (23.65)
where the dimensionless parameter is the wavevector k multiplied by the sound horizon /
3,

k
3
=
_
2
3
k
a
eq
H
eq
a
a
eq
, (23.66)
and j
l
()
_
/(2)J
l+1/2
() are spherical Bessel functions. The physically relevant solution that satises
adiabatic initial conditions, remaining nite as 0, is the growing solution
= (0)
grow
. (23.67)
The solution (23.65) shows that, after a mode enters the sound horizon the scalar potential oscillates with
an envelope that decays as
2
. Physically, relativistically propagating sound waves tend to suppress the
gravitational potential .
For the growing solution (23.67), the radiation monopole
0
is
0
= (0)
3
3
_
(1
2
) sin
_
1
1
2
2
_
cos
. (23.68)
The thin lines in the top panel of Figure 23.45 show the growing mode potential and the radiation monopole
0
, equation (23.68). The Figure plots these two quantities in the form 2 and
0
, to bring out the
fact that
0
oscillates about 2, in accordance with equation (23.45). After a mode is well inside the
sound horizon, 1, the radiation monopole oscillates with constant amplitude,
0
=
3(0)
2
cos for 1 . (23.69)
The dark matter uctuations are driven by the gravitational potential of the radiation. The solution
of the dark matter equation (23.43) driven by the potential (23.67) and satisfying adiabatic initial condi-
tions (23.62a) is
c
= 3 9(0)
_

1
2
+ ln Ci +
sin
_
, (23.70)
where the potential is the growing mode solution (23.67), 0.5772... is Eulers constant, and Ci()
_
cos xdx/x is the cosine integral. Once the mode is well inside the sound horizon, 1, the dark matter
density
c
, equation (23.70), evolves as
c
= 9(0)
_

1
2
+ ln
_
for 1 , (23.71)
which grows logarithmically. This logarithmic growth translates into a logarithmic increase in the amplitude
of matter uctuations at small scales, and is a characteristic signature of non-baryonic cold dark matter.
23.11 Radiation-dominated, isocurvature initial conditions 379
23.11 Radiation-dominated, isocurvature initial conditions
For isocurvature initial conditions, the matter uctuation contributes from the outset, [
c
c
[ > [
r
0
[ even
while radiation dominates the background density,
c

r
.
To develop an approximation adequate for isocurvature uctuations entering the horizon well before
matter-radiation equality, k
eq
1, regard the Einstein energy equation (23.28a) as giving the radiation
monopole
0
, and the Einstein momentum equation (23.30) as giving the radiation dipole
1
. Insert these
into the radiation monopole equation (23.27a), and eliminate the

c
terms using the dark matter density
equation (23.26a). The result is, in units a
eq
= H
eq
= 1,
2a(1 +a)
+ (8 + 9a)
+ 2
_
1 +
2k
2
a
3
_
+
c
= 0 , (23.72)
where prime

denotes dierentiation with respect to cosmic scale factor a. Equation (23.72) is valid in
all regimes, for any combination of matter and radiation. For isocurvature initial conditions, the radiation
monopole and potential vanish initially,
0
(0) = (0) = 0, whereas the dark matter overdensity is nite,
c
(0) ,= 0. For small scales that enter the horizon well before matter-radiation equality, k
eq
1, the
potential is small, while
c
has some approximately constant non-zero value, up to and through the time
when the mode enters the horizon, k ka 1. In the radiation-dominated epoch, a 1, and with 0
(but k large and ka 1, so k
2
a is not small) equation (23.72) simplies to
2a
+ 8
+
4k
2
a
3
+
c
= 0 . (23.73)
The solution of equation (23.73) for constant
c
=
c
(0) is, with given by equation (23.66),
=

c
(0)
_
2/3 k
1 +
1
2
2
cos sin
3
. (23.74)
The solution (23.51) for the radiation monopole
0
driven by the potential (23.74) is
0
=

c
(0)
_
2/3 k
(1 +
2
) cos + (1 +
1
2
2
)(1 +sin )
3
. (23.75)
Equations (23.74) and (23.75) are the solution for small scale modes with isocurvature initial conditions
that enter the horizon well before matter-radiation equality. After a mode is well inside the sound horizon,
1, the radiation monopole (23.75) oscillates with constant amplitude,
0
=

c
(0)
2
_
2/3 k
sin 1 . (23.76)
Whereas for adiabatic initial conditions the radiation monopole oscillated as cos well inside the hori-
zon, equation (23.69), for isocurvature initial conditions it oscillates as sin well inside the horizon, equa-
tion (23.76).
23.12 Subhorizon scales
After a mode enters the horizon, the radiation uctuation
0
oscillates, but the non-baryonic cold dark
matter uctuation
c
grows monotonically. In due course, the dark matter density uctuation
c
c
dominates
the radiation density uctuation
r
0
, and this necessarily occurs before matter-radiation equality; that is,
[
c
c
[ > [
r
0
[ even though
c
<
r
. This is true for both adiabatic and isocurvature initial conditions; of
course, for isocurvature initial conditions, the dark matter density uctuation dominates from the outset.
Even before the dark matter density uctuation dominates, the cumulative contribution of the dark matter
to the potential begins to be more important than that of the radiation, because the potential sourced by
the radiation oscillates, with an eect that tends to cancel when averaged over an oscillation.
Regard the Einstein energy equation (23.28a) as giving the dark matter overdensity
c
, and the Einstein
momentum equation (23.30) as giving the dark matter velocity v
c
. Insert these into the dark matter density
equation (23.26a) and eliminate the

0
terms using the radiation monopole equation (23.27a). The result
is, in units a
eq
= 1,
2a
2
(1 +a)
+a(6 + 7a)
2 4
0
= 0 , (23.77)
where prime

denotes dierentiation with respect to cosmic scale factor a. Equation (23.77) is valid in all
regimes, for any combination of matter and radiation. Once the mode is well inside the horizon, k 1,
the radiation monopole
0
oscillates about an average value of (since
0
oscillates about 2, as
noted in 23.7):
0
) = . (23.78)
Inserting this cycle-averaged value of
0
into equation (23.77) gives the Meszaros dierential equation
2(1 +a)a
2
+ (6 + 7a)a
+ 2 = 0 . (23.79)
The solutions of Meszaros dierential equation (23.79) are
=
3
4
_
a
eq
H
eq
k
_
2
c
a/a
eq
, (23.80)
where the dark matter overdensity
c
is a linear combination
c
= C
grow

c,grow
+C
decay
c,decay
(23.81)
of growing and decaying solutions, in units a
eq
= 1,
c,grow
= 1 +
3
2
a ,
c,decay
=
_
1 +
3
2
a
_
ln
_
1 +a + 1
1 +a 1
_
3
1 +a . (23.82)
For adiabatic initial conditions, the desired solution is the one that matches smoothly on to the the logarith-
mically growing solution for the dark matter overdensity
c
given by equation (23.70). For modes that enter
the horizon well before matter-radiation equality, the matching may be done in the radiation-dominated
epoch a 1, where the growing and decaying modes (23.82) simplify to
c,grow
= 1 ,
decay
= ln(a/4) 3 , for a 1 . (23.83)
23.13 Matter-dominated 381
Matching to the solution for
c
well inside the horizon, equation (23.71), determines the constants
C
grow
= 9 (0)
_

7
2
+ ln
_
4
_
2
3
k
a
eq
H
eq
_
_
, C
decay
= 9 (0) adiabatic . (23.84)
For isocurvature initial conditions, for modes that enter the horizon well before matter-radiation equality,
only the growing mode is present,
C
grow
=
c
(0) , C
decay
= 0 isocurvature . (23.85)
The dark matter overdensity
c
then evolves as the linear combination (23.81) of growing and decaying
modes (23.82). For modes that enter the horizon well before matter-radiation equality, the constants are
set by equation (23.84) for adiabatic intial conditions, or equation (23.85) for isocurvature intial conditions.
The solution remains valid from the radiation-dominated through into the matter-dominated epoch. At late
times well into the matter-dominated epoch, a 1, the growing mode of the Meszaros solution dominates,
c,grow
=
3
2
a ,
c,decay
=
4
15
a
3/2
for a 1 , (23.86)
so that the dark matter overdensity
c
at late times is
c
=
3
2
C
grow
a for a 1 . (23.87)
The potential , equation (23.80), at late times is constant,
=
9
8k
2
C
grow
for a 1 . (23.88)
For modes that enter the horizon well before matter-radiation equality, the radiation monopole
0
at late
times a 1 is, with k/
3,
0
= +
3
2
(0) cos adiabatic , (23.89a)
0
=

c
(0)
2
_
2/3 k
sin isocurvature . (23.89b)
23.13 Matter-dominated
After matter-radiation equality, but before curvature or dark energy become important, non-relativistic
matter dominates the mass-energy density of the Universe.
In the matter-dominated epoch, the relevant equations are, from equations (23.26), (23.28a), and (23.30),
c
k v
c
= 3

, (23.90a)
k
2
3
a
a
F = 4Ga
2

c
c
, (23.90b)
kF = 4Ga
2

c
v
c
, (23.90c)
in which, because it simplies the mathematics, the Einstein momentum equation is used as a substitute
for the matter velocity equation. In the matter-dominated epoch, the horizon is proportional to the square
root of the cosmic scale factor, a
1/2
, equation (23.38). Inserting
c
and v
c
from the Einstein energy and
momentum equations (23.90b) and (23.90c) into the matter density equation (23.90a) yields a second order
dierential equation for the potential
+
6
= 0 . (23.91)
The general solution of equation (23.91) is a linear combination
= C
grow

grow
+C
decay

decay
(23.92)
of growing and decaying solutions
grow
= 1 ,
decay
=
5
, (23.93)
where the dimensionless paramater is the wavevector k multiplied by the sound horizon /
3,

k
3
= 2
_
2
3
ka
1/2
a
3/2
eq
H
eq
. (23.94)
The constants C
grow
and C
decay
in the solution (23.92) depend on conditions established before the matter-
dominated epoch. The corresponding growing and decaying modes for the dark matter overdensity
c
are
c,grow
=
_
2 +

2
2
_
grow
,
c,decay
=
_
3

2
2
_
decay
. (23.95)
For modes well inside the horizon, 1, the behaviour of the growing and decaying modes agrees with
that (23.86) of the Meszaros solution, as it should. Any admixture of the decaying solution tends quickly to
decay away, leaving the growing solution. The solution (23.51) for the radiation monopole
0
driven by the
potential (23.92) is a sum of a homogeneous solution and a particular solution,
0
= B
0
cos +B
1
sin +C
grow

0,grow
+C
decay

0,decay
, (23.96)
with growing and decaying modes
0,grow
=
grow
,
0,decay
=

decay
12
_
12 2
2
+
4
+
5
[cos (Si /2) sin Ci ]
_
. (23.97)
23.14 Recombination
Exercise 23.6 Electron-scattering mean free path.
23.15 Post-recombination 383
1. Dene the neutron faction X
n
by
X
n

n
n
n
p
+n
n
, (23.98)
where the proton and neutron number densities n
p
and n
n
are taken to include protons and neutrons in
all nuclei. For a H plus
4
He composition, the proton and neutron number densities are
n
p
= n
H
+ 2n4
He
, n
n
= 2n4
He
. (23.99)
Show that the primordial
4
He mass fraction dened by Y4
He
4
He
/(
H
+4
He
) satises
Y4
He
= 2X
n
. (23.100)
The observed primordial
4
He abundance is Y4
He
= 0.24, implying
X
n
= 0.12 . (23.101)
2. Dene the ionization fraction X
e
by
X
e

n
e
n
p
(23.102)
where again the proton number density n
p
is taken to include protons in all nuclei, not just in hydrogen.
Show that
n
e
b
=
X
e
(1 X
n
)
m
p
(23.103)
where m
p
is the mass of a proton or neutron.
3. Show that the (dimensionless) ratio of the comoving electron-photon scattering mean free path l
T
to the
comoving cosmological horizon distance c/(a
eq
H
eq
) at matter-radiation equality is
a
eq
H
eq
l
T
c

a
eq
H
eq
c n
e
T
a
=
16Gm
p
3c
T
X
e
(1 X
n
)H
eq
b
_
a
a
eq
_
2
=
28.94 h
1
X
e
(1 X
n
)
H
0
H
eq
b
_
a
a
eq
_
2
=
0.002
X
e
_
a
a
eq
_
2
, (23.104)
the Hubble parameter H
eq
at matter-radiation equality being related to the present-day Hubble parameter
H
0
by equation (23.41).
23.15 Post-recombination
Recombination frees baryons and photons from each others grasp.
Exercise 23.7 Growth of baryon uctuations after recombination.
23.16 Matter with dark energy
Some time after recombination, dark energy becomes important. Observational evidence suggests that the
dominant energy-momentum component of the Universe today is dark energy, with an equation of state
consistent with that of a cosmological constant, p
. In what follows, dark energy is taken to have

constant density, and therefore to be synonymous with a cosmological constant. Since dark energy has a
constant energy density whereas matter density declines as a
3
, dark energy becomes important only well
after recombination.
Dark energy does not cluster gravitationally, so the Einstein equations for the perturbed energy-momentum
depend only on the matter uctuation. However, dark energy does aect the evolution of the cosmic scale
factor a. In fact, if matter is taken to be the only source of perturbation, then covariant energy-momentum
conservation, as enforced by the Einstein equations, implies that the only addition that can be made to the
unperturbed background is dark energy, with constant energy density. To see this, consider the equations
governing the matter overdensity
m
and scalar velocity v
m
(now subscripted m, since post-recombination
matter includes baryons as well as non-baryonic cold dark matter), together with the Einstein energy and
momentum equations:
m
k v
m
= 3

, (23.105a)
v
m
+
a
a
v
m
= k , (23.105b)
k
2
3
a
a
F = 4Ga
2

m
m
, (23.105c)
kF = 4Ga
2

m
v
m
. (23.105d)
The factor 4Ga
2

m
on the right hand side of the two Einstein equations can be written
4Ga
2

m
=
3a
3
0
H
2
0
m
2a
, (23.106)
where a
0
and H
0
are the present-day cosmic scale factor and Hubble parameter, and
m
is the present-
day matter density (a constant). Allow the Hubble parameter H(a) a/a
2
to be an arbitrary function
of cosmic scale factor a. Inserting
m
and velocity v
m
from the Einstein energy and momentum equa-
tions (23.105c) and (23.105d) into the matter equations (23.105a) and (23.105b), and taking the overdensity
equation (23.105a) minus 3 a/a times the velocity equation (23.105b), yields the condition
a
4
dH
2
da
+ 3a
3
0
H
2
0
m
= 0 , (23.107)
whose solution is
H
2
H
2
0
=

m
(a/a
0
)
3
+
(23.108)
for some constant
. This shows that, as claimed, if only matter perturbations are present, then the unper-
turbed background can contain, besides matter, only dark energy with constant density
= H
2
0
/(
8
3
G).
23.17 Matter with dark energy and curvature 385
The result is a consequence of the fact that the Einstein equations enforce covariant conservation of energy-
momentum.
With the Hubble parameter given by equation (23.108), the matter and Einstein equations (23.105) yield
a second order dierential equation for the potential , in units a
0
= 1:
2a(
m
+a
3
+ (7
m
+ 10a
3
+ 6a
2
= 0 . (23.109)
The growing and decaying solutions to equation (23.109) are, in units a
0
= 1,
grow
=
5
m
H
2
0
2
H(a)
a
_
a
0
da
a
3
H(a
)
3
,
decay
=
H
a
. (23.110)
The factor
5
2
m
H
2
0
in the growing solution is chosen so that
grow
1 as a 0. The growing solution
grow
can be expressed as an elliptic integral. The corresponding growing and decaying solutions for the
matter overdensity
m
are, again in units a
0
= 1,
m,grow
=
_
3
2k
2
a
3
m
H
2
0
_
grow
5 ,
m,decay
=
_
3
2k
2
a
3
m
H
2
0
_
decay
. (23.111)
For modes well inside the horizon, k ka
1/2
/H
0
1, the relation (23.111) agrees with that (23.117) below.
23.17 Matter with dark energy and curvature
Curvature may also play a role after recombination. Observational evidence as of 2010 is consistent with the
Universe having zero curvature, but it is possible that there may be some small curvature.
If the curvature is non-zero, then strictly the unperturbed metric should be an FRW metric with curvature.
However, a at background FRW metric remains a good approximation for modes whose scales are small
compared to the curvature, that is, for modes that are well inside the horizon, k 1. For modes well inside
the horizon, the time derivative of the potential can be neglected,

= 0. With matter, curvature, and dark
energy present, and for modes well inside the horizon, the equations go over to the Newtonian limit:
m
k v
m
= 0 , (23.112a)
v
m
+
a
a
v
m
= k , (23.112b)
k
2
= 4Ga
2

m
m
. (23.112c)
The factor 4Ga
2

m
in the Einstein equation can be written as equation (23.106). The matter and Einstein
equations (23.112) yield a second order equation for the matter overdensity
m
, in units a
0
= 1:
m
+
a
a
3
m
H
2
0
2
m
a
= 0 . (23.113)
Equation (23.113) can be recast as a dierential equation with respect to cosmic scale factor a:
m
+
_
H
H
+
3
a
_
3
m
H
2
0
2
m
a
5
H
2
= 0 , (23.114)
where H a/a
2
is the Hubble parameter, and prime

denotes dierentiation with respect to a. In the case
of matter plus curvature plus dark energy, the Hubble parameter H satises, again in units a
0
= 1,
H
2
H
2
0
=
m
a
3
+
k
a
2
+
, (23.115)
where
m
,
k
, and
are the (constant) present-day values of the matter, curvature, and dark energy
densities. The growing and decaying solutions to equation (23.114) are
m,grow
a g(a) =
5
m
H
2
0
2
H(a)
_
a
0
da
a
3
H(a
)
3
,
m,decay
=
H
H
0
. (23.116)
The potential is related to the matter overdensity
m
by, again in units a
0
= 1, equation (23.112c),
=
3
m
H
2
0
2k
2
m
a
. (23.117)
The observationally relevant solution is the growing mode. The growing mode is conventionally given a
special notation, the growth factor g(a), because of its importance to relating the amplitude of clustering at
various times, from recombination up to the present. For the growing mode,
a g(a) , g(a) . (23.118)
The normalization factor
5
2
m
H
2
0
in equation (23.116) is chosen so that in the matter-dominated phase
shortly after recombination (small a), the growth factor g(a) is
g(a) = 1 . (23.119)
Thus as long as the Universe remains matter-dominated, the potential remains constant. Curvature or
dark energy causes the potential to decrease.
It should be emphasized that the growing and decaying solutions (23.116) are valid only for the case of
matter plus curvature plus constant density dark energy, where the Hubble parameter takes the form (23.115).
If another kind of mass-energy is considered, such as dark energy with non-constant density, then equations
governing perturbations of the other kind must be adjoined, and the Einstein equations modied accordingly.
The growth factor g(a) may expressed analytically as an elliptic function. A good analytic approximation
is (Carroll, Press & Turner 1992, Ann. Rev. Astron. Astrophys. 30, 449)
g
5
m
2
_
4/7
m

+
_
1 +
1
2
m
_ _
1 +
1
70
_
_ , (23.120)
where
a
are densities at the epoch being considered (such as the present, a = a
0
).
24
Cosmological perturbations: a more careful

treatment of photons and baryons
The simple treatment of cosmological perturbations in the previous Chapter is sucient to reveal that the
photon-baryon uid at the time of recombination shows a characteristic pattern of oscillations. Translating
this pattern into something that can be compared to observations of the CMB requires a more careful
treatment that follows the evolution of photons using a collisional Boltzmann equation.
Cosmologists conventionally refer to atomic matter anything that acts by either strong or electromag-
netic forces as baryons (from the greek baryos, meaning heavy), even though they mean by that not only
baryons (protons, neutrons, and other nuclei), but also electrons, which are leptons (from the greek leptos,
meaning light). The designation baryons does not include relativistic species, such as photons and neutrinos,
nor does it include non-baryonic dark matter or dark energy. The designation baryons is nonsensical, but
has stuck.
Although baryons are gravitationally sub-dominant, having a mass density about 1/5 that of the non-
baryonic dark matter, they play an important role in CMB uctuations. Most importantly, before recombi-
nation atomic matter is ionized, and the free electrons scatter photons, preventing photons from travelling
far. Recombination occurs when the temperature drops to the point that electrons combine into neutral
atoms, releasing the photons to travel freely across the Universe, into astronomers telescopes.
Electron-photon scattering keeps photons and baryons tightly coupled, so that up to the time of recom-
bination they oscillate together as a photon-baryon uid. As recombination approaches, the baryon density
becomes increasingly important relative to the photon density. The baryons, which provide mass but no
pressure, decrease the sound speed of the photon-baryon uid, and their gravity enhances sound wave com-
pressions while weakening rarefactions. As recombination approaches, the mean free path of photons to
scattering increases, which tends to damp sound waves at short scales. All these eects a decreased sound
speed, an enhancement of compression over rarefaction, and damping at small scales produce observable
signatures in the power spectrum of temperature uctuations in the CMB.
388

Cosmological perturbations: a more careful treatment of photons and baryons
24.1 Lorentz-invariant spatial and momentum volume elements
DO THIS BETTER. To dene an occupation number in a Lorentz-invariant fashion, it is rst necessary
to dene Lorentz-invariant volume elements of space and momentum. With respect to an orthonormal
tetrad, volume elements transform as they do in special relativity. With respect to an orthonormal tetrad,
a Lorentz-invariant spatial 3-volume element can be constructed as
d
4
x
d
=
dt
d
d
3
x = E d
3
x . (24.1)
Lorentz invariance of Ed
3
x is evident because the tetrad-frame 4-volume element d
4
x is a scalar, and likewise
the interval d of proper time is a scalar. Similarly, with respect to an orthonormal tetrad, a Lorentz-invariant
momentum-space 3-volume element can be constructed as
_
E>0
2
D
(E
2
p
2
m
2
) d
4
p =
_
E>0
2
D
(E
2
p
2
m
2
) dE d
3
p =
d
3
p
E
. (24.2)
Lorentz-invariance of d
3
p/E is evident because d
4
p is a Lorentz-invariant momentum 4-volume, and the
argument E
2
p
2
m
2
of the delta-function is a scalar. From the Lorentz invariance of Ed
3
x and d
3
p/E it
follows that the product
d
3
xd
3
p (24.3)
of spatial and momentum 3-volumes is Lorentz-invariant.
24.2 Occupation numbers
WARNING: d
3
x IS TETRAD-FRAME VOLUME, BUT x IS A COORDINATE. Each species of energy-
momentum is described by a dimensionless occupation number, or phase-space distribution, a function
f(, x, p) of conformal time , comoving position x, and tetrad-frame momentum p, which describes the
number dN of particles in a tetrad-frame element d
3
xd
3
p/(2)
3
of phase-space
dN(, x, p) = f(, x, p)
g d
3
xd
3
p
(2)
3
, (24.4)
with g being the number of spin states of the particle. The tetrad-frame phase-space element d
3
xd
3
p/(2)
3
is dimensionless and Lorentz invariant, and the occupation number f is likewise dimensionless and Lorentz
invariant. The tetrad-frame energy-momentum 4-vector p
m
of a particle is
p
m
e
m
dx
d
= E, p = E, p
i
, (24.5)
where is the ane parameter, related to proper time along the worldline of the particle by d d/m,
which remains well-dened in the limit of massless particles, m = 0. The tetrad-frame energy E and
momentum p [p[ for a particle of rest mass m are related by
E
2
p
2
= m
2
. (24.6)
24.3 Occupation numbers in thermodynamic equilibrium 389
The tetrad-frame components of the energy-momentum tensor T
mn
of any species are integrals over its
occupation number f weighted by the product p
m
p
n
of 4-momenta:
T
mn
=
_
f p
m
p
n
g d
3
p
E(2)
3
. (24.7)
mn
dened by equation (24.7) is manifestly a tetrad-frame tensor, thanks
to the Lorentz-invariance of the momentum-space 3-volume element d
3
p/E.
24.3 Occupation numbers in thermodynamic equilibrium
Frequent collisions tend to drive a system towards thermodynamic equilibrium. Electron-photon scatter-
ing keeps photons in near equilibrium with electrons, while Coulomb scattering keeps electrons in near
equilibrium with ions, primarily hydrogen ions (protons) and helium nuclei. Thus photons and baryons
can be treated as having unperturbed distributions in mutual thermodynamic equilibrium, and perturbed
distributions that are small departures from thermodynamic equilibrium.
In thermodynamic equilibrium at temperature T, the occupation numbers of fermions, which obey an
exclusion principle, and of bosons, which obey an anti-exclusion principle, are
f =
_
_
1
e
(E)/T
+ 1
fermion ,
1
e
(E)/T
1
boson ,
(24.8)
where is the chemical potential of the species. In the limit of small occupation numbers, f 1, equivalent
to large negative chemical potential, large, both fermion and boson distributions go over to the
Boltzmann distribution
f = e
(E+)/T
Boltzmann . (24.9)
Chemical potential is the thermodynamic potential assocatiated with conservation of number. There is a
distinct potential for each conserved species. For example, photoionization and radiative recombination of
hydrogen,
H + p +e , (24.10)
separately preserves proton and electron number, hydrogen being composed of one proton and one electron.
In thermodynamic equilibrium, the chemical potential
H
of hydrogen is the sum of the chemical potentials
p
and
e
of protons and electrons
H
=
p
+
e
. (24.11)
Photon number is not conserved, so photons have zero chemical potential,
= 0 , (24.12)
which is closely associated with the fact that photons are their own antiparticles.
390

24.4 Boltzmann equation
The evolution of each species is described by the general relativistic Boltzmann equation
df
d
= C[f] , (24.13)
where C[f] is a collision term. The derivative with respect to ane parameter on the left hand side of
the Boltzmann equation (24.13) is a Lagrangian derivative along the (timelike or lightlike) worldline of a
particle in the uid. Since both the occupation number f and the ane parameter are Lorentz scalars,
the collision term C[f] is a Lorentz scalar. In the absence of collisions, the collisionless Boltzmann equation
df/d = 0 expresses conservation of particle number: a particle is neither created nor destroyed as it moves
along its wordline.
The left hand side of the Boltzmann equation (24.13) is
df
d
= p
m
m
f +
dp
i
d
f
p
i
= E
0
f +p
i
i
f +
d p
d

f
p
+
dp
d
f
p
. (24.14)
Both d p/d and f/ p vanish in the unperturbed background, so d p/d f/ p is of second order, and can
be neglected to linear order, so that
df
d
= E
0
f +p
i
i
f +
dp
d
f
p
. (24.15)
The expression (24.15) for the left hand side df/d of the Boltzmann equation involves dp/d, which in
free-fall is determined by the usual geodesic equation
dp
k
d
+
k
mn
p
m
p
n
= 0 . (24.16)
Since E
2
p
2
= m
2
, it follows that the equation of motion for the magnitude p of the tetrad-frame momentum
is related to the equation of motion for the tetrad-frame energy E by
p
dp
d
= E
dE
d
. (24.17)
The equation of motion for the tetrad-frame energy E p
0
is
dE
d
=
0
mn
p
m
p
n
=
0i0
p
i
E +
0ij
p
i
p
j
. (24.18)
From this it follows that
d ln p
d
=
E
p
2
dE
d
= E
_
E p
i
p

0i0
+ p
i
p
j
0ij
_
= E
_
a
a
2
+
E p
i
p

0i0
+ p
i
p
j
1
0ij
_
, (24.19)
where in the last expression the tetrad connection
0ij
, equation (22.14b), has been separated into its
unperturbed and perturbed parts ( a/a
2
)
ij
and
1
0ij
.
24.4 Boltzmann equation 391
In practice, the integration variable used to evolve equations is the conformal time , not the ane
parameter . The relation between conformal time and ane parameter is
d
d
= p
= e
m
p
m
= (
n
m
+
m
n
)
0
e
n
p
m
=
1
a
_
E(1
00
) p
i
i0
, (24.20)
whose reciprocal is to linear order
d
d
=
a
E
_
1 +
00
+
p
i
E
i0
_
. (24.21)
With conformal time as the integration variable, the equation of motion (24.19) for the magnitude p of
the tetrad-frame momentum becomes, to linear order,
d ln p
d
=
a
a
_
1 +
00
+
p
i
E
i0
_
+
E p
i
p
a
0i0
+ p
i
p
j
a
1
0ij
. (24.22)
With respect to conformal time , the Boltzmann equation (24.13) is
df
d
=
f
+v
i
i
f +
d ln p
d
f
ln p
=
d
d
C[f] , (24.23)
with d/d from equation (24.21), and d ln p/d from equation (24.22). Expressions for d/d and d ln p/d
in terms of the vierbein perturbations in a general gauge are left as Exercise 24.1. In conformal Newtonian
gauge, the factor d/d, equation (24.21), is
d
d
=
a
E
(1 + ) . (24.24)
In conformal Newtonian gauge, and including only scalar uctuations, the factor d ln p/d, equation (24.22),
is
d ln p
d
=
a
a
+

E p
i
p

i
. (24.25)
To unperturbed order, the Boltzmann equation (24.23) is
d
0
f
d
=

0
f

a
a
0
f
ln p
=
a
E
C[
0
f] , (24.26)
where C[
0
f] is the unperturbed collision term, the factor a/E coming from d/d = a/E to unperturbed
order, equation (24.21). The second term in the middle expression of equation (24.26) simply reects the
fact that the tetrad-frame momentum p redshifts as p 1/a as the Universe expands, a statement that is
true for both massive and massless particles.
Subtracting o the unperturbed part (24.26) of the Boltzmann equation (24.23) gives the perturbation of
the Boltzmann equation
d
1
f
d
=

1
f
+v
i
i
1
f
a
a
1
f
ln p
+
d ln(ap)
d
0
f
ln p
=
a
E
C[
1
f] +
1
d
d
C[
0
f] . (24.27)
392

In conformal Newtonian gauge, the perturbed part of d/d is
1
d
d
=
a
E
. (24.28)
In conformal Newtonian gauge, and including only scalar uctuations, d ln(ap)/d is
d ln(ap)
d
=

E p
i
p

i
. (24.29)
Exercise 24.1 Boltzmann equation factors in a general gauge. Show that in a general gauge, and
including not just scalar but also vector and tensor uctuations, equation (24.21) is
d
d
=
a
E
_
1 + +
p
i
E
(
i
w + w
i
)
_
, (24.30)
while equation (24.22) is, with only scalar uctuations included,
d ln p
d
=
a
a
+

+
E p
i
p
_
i
+
_

+
a
a
m
2
E
2
_
(
i
w + w
i
)
_
+ p
i
p
j
_
j
(w

h)
1
2
(
i
W
j
+
j
W
i
) +
j
w
i
+

h
ij
_
. (24.31)
24.5 Non-baryonic cold dark matter
Non-baryonic cold dark matter, subscripted c, is by assumption non-relativistic and collisionless. The un-
perturbed mean density is
c
, which evolves with cosmic scale factor a as

c
a
3
. (24.32)
Since dark matter particles are non-relativistic, the energy of a dark matter particle is its rest-mass energy,
E
c
= m
c
, and its momentum is the non-relativistic momentum p
i
c
= m
c
v
i
c
.
mn
c
of the dark matter is obtained from integrals over the dark matter
phase-space distribution f
c
, equation (24.7). The energy and momentum moments of the distribution dene
the dark matter overdensity
c
and bulk velocity vvv
c
, while the pressure is of order v
2
c
, and can be neglected
to linear order,
T
00
c

_
f
c
m
c
g
c
d
3
p
c
(2)
3

c
(1 +
c
) , (24.33a)
T
0i
c

_
f
c
m
c
v
i
c
g
c
d
3
p
c
(2)
3

c
v
i
c
, (24.33b)
T
ij
c

_
f
c
m
c
v
i
c
v
j
c
g
c
d
3
p
c
(2)
3
= 0 . (24.33c)
24.5 Non-baryonic cold dark matter 393
Non-baryonic cold dark matter is collisionless, so the collision term in the Boltzmann equation is zero,
C[f
c
] = 0, and the dark matter satises the collisionless Boltzmann equation
df
c
d
= 0 . (24.34)
The energy and momentum moments of the Boltzmann equation (24.23) yield equations for the overdensity
c
and bulk velocity v
c
, which in the conformal Newtonian gauge are
0 =
_
df
c
d
m
c
g
c
d
3
p
c
(2)
3
=

_
f
c
m
c
g
c
d
3
p
c
(2)
3
+
i
_
f
c
m
c
v
i
c
g
c
d
3
p
c
(2)
3

_ _
a
a

_
f
ln p
m
c
g
c
d
3
p
c
(2)
3
=

c
(1 +
c
)
+
i
(
c
v
i
c
) + 3
_
a
a

_

c
, (24.35a)
0 =
_
df
c
d
m
c
v
i
c
g
c
d
3
p
c
(2)
3
=

_
f
c
m
c
v
i
c
g
c
d
3
p
c
(2)
3
+
j
_
f
c
m
c
v
i
c
v
j
c
g
c
d
3
p
c
(2)
3
_ _
a
a

+
E p
j
p

j
_
f
ln p
m
c
v
i
g
c
d
3
p
c
(2)
3
=

c
v
i
c
+ 4
_
a
a

_

c
v
i
c
+
c
i
. (24.35b)
The

c
v
i
c
term on the last line of equation (24.35b) can be dropped, since the potential and the velocity
v
i
c
are both of rst order, so their product is of second order. Subtracting the unperturbed part from
equations (24.35a) and (24.35b) gives equations for the dark matter overdensity
c
and velocity vvv
c
,
c
+ vvv
c
3

= 0 , (24.36a)
vvv
c
+
a
a
vvv
c
+ = 0 . (24.36b)
Transform into Fourier space, and decompose the velocity 3-vector vvv
c
into scalar v
c
and vector vvv
c,
parts
vvv
c
= i
kv
c
+ vvv
c,
. (24.37)
For the scalar modes under consideration, only the scalar part of the dark matter equations (24.36) is
relevant:
c
kv
c
3

= 0 , (24.38a)
v
c
+
a
a
v
c
+k = 0 . (24.38b)
Equations (24.38) reproduce the equations (23.26) derived previously from conservation of energy and mo-
mentum.
Exercise 24.2 Moments of the non-baryonic cold dark matter Boltzmann equation. Conrm
equations (24.35).
394

24.6 The left hand side of the Boltzmann equation for photons
In the unperturbed background, the photons have a blackbody distribution with temperature T(). Dene
to be the photon temperature uctuation
(, x, p
)
T(, x, p
)
T()
. (24.39)
In the unperturbed background, the photon occupation number is
0
f
=
1
e
p/T
1
. (24.40)
Since p
T 1/a, the unperturbed occupation number is constant as a function of p
/T. The denition

T/T = ln T of the photon perturbation is to be interpreted as meaning that the perturbation to the
occupation number of photons is (the partial derivative with respect to temperature / ln T is at constant
photon momentum p
)
1
f
0
f
ln T
ln T =
0
f
ln T
, (24.41)
in which it follows from equation (24.40) that
0
f
ln T
=
0
f
(1 +
0
f
)
p
T
. (24.42)
The photon Boltzmann equation in terms of the occupation number f
can be recast as a Boltzmann

equation for the temperature uctuation through
d
1
f
d
=
d
d
_
_
0
f
ln T

_
_
=
d
d
_
_
0
f
ln T
_
_
+
0
f
ln T
d
d
=
0
f
ln T
d
d
, (24.43)
in which the rst term of the penultimate expression vanishes because
0
f
/ ln T is a function of p
/T only,
and p
/T is, to unperturbed order (which is all that is needed since the term is multiplied by , which is
already of rst order), independent of time, d(p
/T)/d = 0, since p
T a
1
:
d
d
0
f
ln T
=
d ln(p
/T)
d
2
0
f
ln T
2
= 0 . (24.44)
In terms of the temperature uctuation , the perturbed photon Boltzmann equation (24.27) is
d
d
=

+ p
i
a
a
ln p
d ln(ap
)
d
=
_
a
p
C[
1
f
] +
1
d
d
C[
0
f
]
__

0
f
ln T
, (24.45)
where the d ln(ap
)/d term gets a minus sign from

0
f/ ln p
=
0
f/ ln T.
The unperturbed photon distribution is in thermodynamic equilibrium, so the unperturbed collision term
24.7 Spherical harmonics of the photon distribution 395
in the photon Boltzmann equation (24.45) vanishes, C[
0
f
] = 0, as found in Exercise 24.3 below. The

photon distribution is modied by photon-electron (Thomson) scattering, 24.10. Since the electrons are
non-relativistic, to linear order collisions change the photon momentum but not the photon energy. As a
consequence, the temperature uctuation is a function (, x, p
) only of the direction p
of the photon
momentum p
, not of its magnitude, the energy p
. This is shown more carefully below, equation (24.78).

Hence the derivative / ln p
in the photon Boltzmann equation (24.45) vanishes to linear order. Thus

the photon Boltzmann equation (24.45) reduces to
d
d
=

+ p
i
d ln(ap
)
d
=
a
p
C[
1
f
]
_

0
f
ln T
. (24.46)
In conformal Newtonian gauge, and including only scalar uctuations, d ln(ap
)/d is given by equa-

tion (24.29), so the photon Boltzmann equation (24.45) becomes
d
d
=

+ p
i

+ p
i
i
=
a
p
C[
1
f
]
_

0
f
ln T
. (24.47)
24.7 Spherical harmonics of the photon distribution
It is natural to expand the temperature uctuation in spherical harmonics. As seen below, equa-
tions (24.55), the various components of the photon energy-momentum tensor T
mn
are determined by the

monopole, dipole, and quadrupole harmonics of the photon distribution. Scalar uctuations are those that
are rotationally symmetric about the wavevector direction

k, which correspond to spherical harmonics with
zero azimuthal quantum number, m = 0. Expanded in spherical harmonics, and with only scalar terms
retained, the temperature uctuation can be written
(, k, p
) =
=0
(i)
(2 + 1)
(, k)P
k p
) , (24.48)
where P
are Legendre polynomials, 24.21. The scalar harmonics
are angular integrals of the temperature

uctuation over photon directions p
(, k) = i
_
(, k, p
)P
k p
)
do
p
4
. (24.49)
Expanded into the scalar harmonics
(, k), the left hand side of the photon Boltzmann equation (24.46)
396

is, in conformal Newtonian gauge,
d
0
d
=

0
k
1

, (24.50)
d
1
d
=

1
+
k
3
_
0
2k
2
2
_
+
k
3
, (24.51)
d
d
=

+
k
2 + 1
[
1
( + 1)
+1
] ( 2) . (24.52)
24.8 Energy-momentum tensor for photons
Perturbations to the photon energy-momentum tensor involve integrals (24.7) over the perturbed occupation
number of the form, where F( p) is some arbitrary function of the momentum direction p,
_
1
f
p
2
F( p
)
2 d
3
p
(2)
3
=
_

0
f
ln T
p
2
2 4p
2
dp
(2)
3
_
F( p
)
do
p
4
= 4
_
F( p
)
do
p
4
, (24.53)
in which the last expression is true because
_

0
f
ln T
p
2
2 4p
2
dp
(2)
3
= 4
_
0
f
2 4p
2
dp
(2)
3
= 4
, (24.54)
which follows from
0
f
/ ln T =
0
f
/ ln p
and an integration by parts. The perturbation of the photon

energy density, energy ux, monopole pressure, and quadrupole pressure are
1
T
00
= 4

0
, (24.55a)
k
i
1
T
0i
= i 4

1
, (24.55b)
1
3

ij
1
T
ij
=
4
3

0
, (24.55c)
_
3
2
k
i
k
j
1
2

ij
_
1
T
ij
= 4

2
. (24.55d)
24.9 Collisions
For a 2-body collision of the form
1 + 2 3 + 4 , (24.56)
24.9 Collisions 397
the rate per unit time and volume at which particles of type 1 leave and enter an interval d
3
p
1
of momentum
space is, in units c = 1,
C[f
1
]
g
1
d
3
p
1
E
1
(2)
3
=
_
[/[
2
_
f
1
f
2
(1 f
3
)(1 f
4
) +f
3
f
4
(1 f
1
)(1 f
2
)
(2)
4
4
D
(p
1
+p
2
+p
3
+p
4
)
g
1
d
3
p
1
2E
1
(2)
3
g
2
d
3
p
2
2E
2
(2)
3
g
3
d
3
p
3
2E
3
(2)
3
g
4
d
3
p
4
2E
4
(2)
3
. (24.57)
All factors in equation (24.57) are Lorentz scalars. On the left hand side, the collision term C[f
1
] and
the momentum 3-volume element d
3
p
1
/E
1
are both Lorentz scalars. On the right hand side, the squared
amplitude [/[
2
, the various occupation numbers f
a
, the energy-momentum conserving 4-dimensional Dirac
delta-function
4
D
(p
1
+ p
2
+ p
3
+ p
4
), and each of the four momentum 3-volume elements d
3
p
a
/E
a
, are all
Lorentz scalars.
The rst ingredient in the integrand on the right hand side of the expression (24.57) is the Lorentz-invariant
scattering amplitude squared [/[
2
, calculated using quantum eld theory (F. Halzen & A. D. Martin 1984
Quarks and Leptons, Wiley, New York, p. 91).
The second ingredient in the integrand on the right hand side of expression (24.57) is the combination of
rate factors
rate(1 + 2 3 + 4) f
1
f
2
(1 f
3
)(1 f
4
) , (24.58a)
rate(1 + 2 3 + 4) f
3
f
4
(1 f
1
)(1 f
1
) , (24.58b)
where the 1 f factors are blocking or stimulation factors, the choice of sign depending on whether the
species in question is fermionic or bosonic:
1 f = Fermi-Dirac blocking factor , (24.59a)
1 +f = Bose-Einstein stimulation factor . (24.59b)
The rst rate factor (24.58a) expresses the fact that the rate to lose particles from 1 + 2 3 + 4 collisions
is proportional to the occupancy f
1
f
2
of the initial states, modulated by the blocking/stimulation factors
(1 f
3
)(1 f
4
) of the nal states. Likewise the second rate factor (24.58b) expresses the fact that the
rate to gain particles from 1 + 2 3 + 4 collisions is proportional to the occupancy f
3
f
4
of the initial
states, modulated by the blocking/stimulation factors (1 f
1
)(1 f
2
) of the nal states. In thermodynamic
equilibrium, the rates (24.58) balance, Exercise 24.3, a property that is called detailed balance, or microscopic
reversibility. Microscopic reversibility is a consequence of time reversal symmetry.
The nal ingredient in the integrand on the right hand side of expression (24.57) is the 4-dimensional
Dirac delta-function, which imposes energy-momentum conservation on the process 1 + 2 3 + 4. The
4-dimensional delta-function is a product of a 1-dimensional delta-function expressing energy conservation,
and a 3-dimensional delta-function expressing momentum conservation:
(2)
4
4
D
(p
1
+p
2
+p
3
+p
4
) = 2
D
(E
1
+E
2
+E
3
+E
4
) (2)
3
3
D
(p
1
+p
2
+p
3
+p
4
) . (24.60)
398

Exercise 24.3 Detailed balance. Show that the rates balance in thermodynamic equilibrium,
f
1
f
2
(1 f
3
)(1 f
4
) = f
3
f
4
(1 f
1
)(1 f
2
) . (24.61)
Solution. Equation (24.61) is true if and only if
f
1
1 f
1
f
2
1 f
2
=
f
3
1 f
3
f
4
1 f
4
. (24.62)
But
f
1 f
= e
(E+)/T
, (24.63)
so (24.62) is true if and only if
E
1
+
1
T
+
E
2
+
2
T
=
E
3
+
3
T
+
E
4
+
4
T
, (24.64)
which is true in thermodynamic equilibrium because
E
1
+E
2
= E
3
+E
4
,
1
+
2
=
3
+
4
. (24.65)
24.10 Electron-photon scattering
The dominant process that couples photons and baryons is electron-photon scattering
e + e
. (24.66)
The Lorentz-invariant transition probability for unpolarized non-relativistic electron-photon (Thomson) scat-
tering is
[/[
2
= 12m
2
e
2
_
1 + ( p
)
2
= 16m
2
e
2
_
1 +
1
2
P
2
( p
, (24.67)
where P
2
() is the quadrupole Legendre polynomial, 24.21, and
T
=
8
3
_
e
2
m
e
c
2
_
2
(24.68)
is the Thomson cross-section.
24.11 The photon collision term for electron-photon scattering
Electron-photon scattering keeps electrons and photons close to mutual thermodynamic equilibrium, and
their unperturbed distributions can be taken to be in thermodynamic equilibrium. The unperturbed photon
collision term for electron-photon scattering therefore vanishes, because of detailed balance, Exercise 24.3,
C[
0
f
] = 0 . (24.69)
Thanks to detailed balance, the combination of rates in the collision integral (24.57) almost cancels, so can
be treated as being of linear order in perturbation theory. This allows other factors in the collision integral
to be approximated by their unperturbed values.
The photon collision term for electron-photon scattering follows from the general expression (24.57). To
unperturbed order, the energies of the electrons, which are non-relativistic, may be set equal to their rest
masses, E
e
= m
e
. Since photons are massless, their energies are just equal to their momenta, E
= p
. The
electron occupation number is small, f
e
1, so the Fermi blocking factors for electrons may be neglected,
1 f
e
= 0. The number of spins of the incoming electron is two, g
e
= 2, because photons scatter o both
spins of electrons. On the other hand, the number of spins of the scattered electron and photon are one,
g
e
= g
= 1, because non-relativistic electron-photon scattering leaves the spins of the electron and photon
unchanged. These considerations bring the photon collision term for electron-photon scattering to, from the
general expression (24.57),
C[
1
f
] =
1
16
_
[/[
2
_
f
e
f
(1+f
)+f
e
f
(1+f
(2)
4
4
D
(p
e
+p
p
e
p
)
2 d
3
p
e
m
e
(2)
3
d
3
p
e
m
e
(2)
3
d
3
p
(2)
3
.
(24.70)
The various integrations over momenta are most conveniently carried out as follows. The energy-conserving
integral is best done over the energy of the scattered photon
, which is scattered into an interval do
of
solid angle:
_
2
D
(E
e
+E
E
e
E
)
d
3
p
(2)
3
= p
do
(2)
2
p
do
(2)
2
. (24.71)
The approximation in the last step of equation (24.71), replacing the energy p
of the scattered photon by

the energy p
of the incoming photon, is valid because, thanks to the smallness of the combination of rates
in the collision integral (24.70), it suces to treat the photon energy to unperturbed order. As seen below,
equation (24.76) the energy dierence p
between the incoming and scattered photons is of linear order

in electron velocities.
The momentum-conserving integral is best done over the momentum of the scattered electron, which is e
for outgoing scatterings e + e
, and e for incoming scatterings e + e
. In the former case

(e + e
),
_
(2)
3
3
D
(p
e
+p
p
e
p
)
d
3
p
e
E
e
(2)
3
=
1
m
e
, (24.72)
and the result is the same, 1/m
e
, in the latter case (e + e
). The energy- and momentum-

conserving integrals having been done, the electron e
in the latter case may be relabelled e. So relabelled,

400

the combination of rate factors in the collision integral (24.70) becomes
f
e
f
(1 +f
) +f
e
f
(1 +f
) = f
e
(f
+f
) . (24.73)
Notice that the stimulated terms cancel. The energy- and momentum-conserving integrations (24.71)
and (24.72) bring the photon collision term (24.70) to
C[
1
f
] =
p
16m
2
e
2
_
[/[
2
f
e
(f
+f
)
2 d
3
p
e
(2)
3
do
4
. (24.74)
The collision integral (24.74) involves the dierence f
+ f
between the occupancy of the initial and

nal photon states. To linear order, the dierence is
f
+f
=
0
f(p
) +
0
f(p
)
1
f(p
) +
1
f(p
) =
0
f
ln T
_
p
(p
) + (p
)
_
. (24.75)
The rst term (p
)/p
arises because the incoming and scattered photon energies dier slightly. The
dierence in photon energies is given by energy conservation:
p
= E
e
E
e
=
_
m
e
+
p
2
e
2m
e
_
_
m
e
+
p
2
e
2m
e
_
=
(p
e
+p
)
2
p
2
e
2m
e
=
(p
) (2p
e
+p
)
2m
e
(p
) v
e
, (24.76)
the last line of which follows because the photon momentum is small compared to the electron momentum,
p
T m
e
v
2
e
= p
e
v
e
p
e
. (24.77)
Because the photon energy dierence is of rst order, and the temperature uctuation is already of rst
order, it suces to regard the temperature uctuation as being a function only of the direction p
of the
photon momentum, not of its energy:
(p
) ( p
) . (24.78)
The linear approximations (24.76) and (24.78) bring the dierence (24.75) between the initial and nal
photon occupancies to
f
+f
0
f
ln T
[( p
) v
e
( p
) + ( p
)] . (24.79)
Inserting this dierence in occupancies into the collision integral (24.74) yields
C[
1
f
] =
p
16m
2
e
0
f
ln T
_
[/[
2
f
e
[( p
) v
e
( p
) + ( p
)]
2 d
3
p
e
(2)
3
do
4
. (24.80)
The transition probability [/[
2
, equation (24.67), is independent of electron momenta, so the integration
over electron momentum in the collision integral (24.80) is straightforward. The unperturbed electron density
n
e
and the electron bulk velocity vvv
e
are dened by
n
e

_
0
f
e
2 d
3
p
e
(2)
3
, n
e
vvv
e

_
0
f
e
v
e
2 d
3
p
e
(2)
3
. (24.81)
Coulomb scattering keeps electrons and ions tightly coupled, so the electron bulk velocity vvv
e
equals the
baryon bulk velocity vvv
b
,
vvv
e
= vvv
b
. (24.82)
Integration over the electron momemtum brings the collision integral (24.80) to
C[
1
f
] =
n
e
p
16m
2
e
0
f
ln T
_
[/[
2
[( p
) vvv
b
( p
) + ( p
)]
do
4
. (24.83)
Finally, the collision integral (24.83) must be integrated over the direction p
of the scattered photon. In-

serting the electron-photon scattering transition probability [/[
2
given by equation (24.67) into the collision
integral (24.83) brings it to
C[
1
f
] = n
e
T
p
0
f
ln T
_
_
1 +
1
2
P
2
( p
[( p
) vvv
b
( p
) + ( p
)]
do
4
. (24.84)
The p
vvv
b
term in the integrand of (24.84) is odd, and vanishes on angular integration:
_
_
1 +
1
2
P
2
( p
do
4
= 0 . (24.85)
Similarly, the angular integral over the quadrupole of quantities independent of p
vanishes:
_
P
2
( p
) [ p
vvv
b
( p
)]
do
4
= 0 . (24.86)
The collision integral (24.84) thus reduces to
C[
1
f
(x, p
)] = n
e
T
p
0
f
ln T
_
p
vvv
b
(x) (x, p
) +
_
_
1 +
1
2
P
2
( p
(x, p
)
do
4
_
, (24.87)
where the dependence of various quantities on comoving position x has been made explicit. Now transform
to Fourier space (in eect, replace comoving position x by comoving wavevector k). Replace the baryon bulk
velocity by its scalar part, vvv
b
= i
kv
b
. To perform the remaining angular integral over the photon direction
p
, expand the temperature uctuation (k, p
) in scalar multipole moments according to equation (24.48),

and invoke the orthogonality relation (24.134). With
dened by
k p
, (24.88)
402

these manipulations bring the photon collision integral (24.87) at last to
C[
1
f
(k, p
)] = n
e
T
p
0
f
ln T
_
i
v
b
(k) (k,
) +
0
(k)
1
2
2
(k)P
2
(
. (24.89)
24.12 Boltzmann equation for photons
Inserting the collision term (24.89) into equation (24.47) yields the photon Boltzmann equation for scalar
uctuations in conformal Newtonian gauge,
d
d
=

ik

ik
= n
e
T
a
_
i
v
b
+
0
1
2
2
P
2
(
. (24.90)
Expanded into the scalar harmonics
(, k), the photon Boltzmann equation (24.46) yields the hierarchy

of photon multipole equations
0
k
1

= 0 , (24.91a)
1
+
k
3
(
0
2
2
) +
k
3
=
1
3
n
e
T
a (v
b
3
1
) , (24.91b)
2
+
k
5
(2
1
3
3
) =
9
10
n
e
T
a
2
, (24.91c)
+
k
2 + 1
[
1
( + 1)
+1
] = n
e
T
a
( 3) . (24.91d)
The Boltzmann hierarchy (24.91) shows that all the photon multipoles are aected by electron-photon
scattering, but only the photon dipole
1
depends directly on one of the baryon variables, the baryon bulk
velocity v
b
. The dependence on the baryon velocity v
b
reects the fact that, to linear order, there is a
transfer of momentum between photons and baryons, but no transfer of number or of energy.
For the dipole, = 1, electron-photon scattering drives the electron and photon bulk velocities into near
equality,
v
b
3
1
, (24.92)
so that the right hand side side of the dipole equation (24.91b) is modest despite the large scattering factor
n
e
T
a. The approximation in which the bulk velocities of electrons and photons are exactly equal, v
b
= 3
1
,
and all the higher multipoles vanish,
= 0 for 2, is called the tight-coupling approximation. The tight-

coupling approximation was already invoked in the simple model of Chapter 23.
For multipoles 2, the electron-photon scattering term on the right hand side of the Boltzmann
hierarchy (24.91) acts as a damping term that tends to drive the multipole exponentially into equilibrium
(the solution to the homogeneous equation

+ n
e
T
a
= 0 is a decaying exponential). As seen in

Chapter 23, in the tight-coupling approximation the monopole and dipole oscillate with a natural frequency
of = c
s
k, where c
s
is the sound speed. These oscillations provide a source that propagates upward
to higher harmonic number . For scales much larger than a mean free path, k/( n
e
T
a) 1, the time
24.13 Diusive (Silk) damping 403
derivative is small compared to the scattering term, [
[ c
s
k[[ n
e
T
a[[, reecting the near-equilibrium
response of the higher harmonics. For multipoles 2, the dominant term on the left hand side of the
Boltzmann hierarchy (24.91) is the lowest order multipole, which acts as a driver. Solution of the Boltzmann
equations (24.91) then requires that
+1

k
n
e
T
a
for 2 . (24.93)
The relation (24.93) implies that higher order photon multipoles are successively smaller than lower orders,
[
+1
[ [
[, for scales much larger than a mean free path, k/( n

e
T
a) 1. This accords with the physical
expectation that electron-photon scattering drives the photon distribution to near isotropy.
24.13 Diusive (Silk) damping
The tight coupling between photons and baryons is not perfect, because the mean free path for electron-
photon scattering is nite, not zero. The imperfect coupling causes sound waves to dissipate at scales
comparable to the mean free path. To lowest order, the dissipation can be taken into account by including
the photon quadrupole
2
in the system (24.91) of photon multipole equations, but still neglecting the higher
multipoles,
= 0 for 3. According to the estimate (24.93), this approximation is valid for scales much
larger than a mean free path, k/( n
e
T
a) 1. The approximation is equivalent to a diusion approximation.
In the diusion approximation, the photon quadrupole equation (24.91c) reduces to
2
=
4k
9 n
e
T
a
1
. (24.94)
Substituted into the photon momentum equation (24.91b), the photon quadrupole
2
(24.94) acts as a source
of friction on the photon dipole
1
.
24.14 Baryons
The equations governing baryonic matter are similar to those governing non-baryonic cold dark matter,
24.5, except that baryons are collisional. Coulomb scattering between electrons and ions keep baryons
tightly coupled to each other. Electron-photon scattering couples baryons to the photons.
Since the unperturbed distribution of baryons is in thermodynamic equilibrium, the unperturbed collision
term vanishes for each species of baryonic matter, as it did for photons, equation (24.95),
C[
0
f
b
] = 0 . (24.95)
For the perturbed baryon distribution, only the rst and second moments of the phase-space distribution
are important, since these govern the baryon overdensity
b
and bulk velocity vvv
b
. The relevant collision term
404

is the electron collision term associated with electron-photon scattering. Since electron-photon scattering
neither creates nor destroys electrons, the zeroth moment of the electron collision term vanishes,
_
C[
1
f
e
]
2d
3
p
e
m
e
(2)
3
= 0 . (24.96)
The rst moment of the electron collision term is most easily determined from the fact that electron-photon
collisions must conserve the total momentum of electron and photons:
_
C[
1
f
e
] m
e
v
e
2 d
3
p
e
m
e
(2)
3
+
_
C[
1
f
] p
2 d
3
p
(2)
3
= 0 . (24.97)
Substituting the expression (24.87) for the photon collision integral into equation (24.97), separating out
factors depending on the absolute magnitude p
and direction p
of the photon momentum, and taking

into consideration that the integral terms in equation (24.87), when multiplied by p
, are odd in p
, and
therefore vanish on integration over directions p
, gives
_
C[
1
f
e
] m
e
v
e
2 d
3
p
e
m
e
(2)
3
= n
e
T
_

0
f
ln T
p
2
2 4p
2
dp
(2)
3
_
[ p
vvv
b
+ ( p
)] p
do
4
. (24.98)
The integral over the magnitude p
of the photon momentum in equation (24.98) yields 4 , in accordance

with equation (24.54). Transformed into Fourier space, and with only scalar terms retained, the collision
integral (24.98) becomes, with
k p
k
_
C[
1
f
e
] m
e
v
e
2 d
3
p
e
m
e
(2)
3
= 4
n
e
T
_
[i
v
b
+ ]
do
4
=
4
3
i
n
e
T
(v
b
3
1
) . (24.99)
The result is that the equations governing the baryon overdensity
b
and scalar bulk velocity v
b
look like
those (24.38) governing non-baryonic cold dark matter, except that the velocity equation has an additional
source (24.99) arising from momentum transfer with photons through electron-photon scattering:
b
kv
b
3

= 0 , (24.100a)
v
b
+
a
a
v
b
+k =
n
e
T
a
R
(v
b
3
1
) , (24.100b)
where R is
3
4
the baryon-to-photon density ratio,
R
3
b
4
= R
a
a
a
eq
, R
a
=
3g
b
8
m
0.21 , (24.101)
with g
= 2 + 6
7
8
_
4
11
_
4/3
= 3.36 being the energy-weighted eective number of relativistic particle species
at around the time of recombination (Exercise 10.18). The parameter R plays an important role in that it
modulates the sound speed in the photon-baryon uid, equation (24.110) below.
24.15 Viscous baryon drag damping 405
24.15 Viscous baryon drag damping
A second source of damping of sound waves, distinct from the diusive damping of 24.13, arises from the
viscous drag on photons that results from a small dierence v
b
3
1
between the baryon and photon bulk
velocities. This damping is associated with the nite mass density of baryons, and vanishes in the limit of
small baryon density. However, for realistic values of the baryon density, viscous baryon drag damping is
comparable to diusive damping.
Equation (24.100b) for the baryon bulk velocity may be written
v
b
3
1
=
R
n
e
T
a
_
v
b
+
a
a
v
b
+k
_
. (24.102)
The right hand side of this equation is small because the scattering rate n
e
T
is large, so to lowest order
v
b
= 3
1
, the tight-coupling approximation. To ascertain the eect of a small dierence in the baryon
and photon bulk velocities, take equation (24.102) to next order. In the circumstances where damping is
important, which is small scales well inside the sound horizon, k
s
1, the dominant term on the right hand
side of equation (24.100b) is the time derivative v
b
. With only this term kept, equation (24.102) reduces to
v
b
3
1

R
n
e
T
a
v
b

3R
n
e
T
a
1
. (24.103)
Substituting the approximation (24.103) into the left hand side of the baryon velocity equation (24.100b)
gives
v
b
+
a
a
v
b
+k = 3
_
1
+
a
a
1
+
k
3
3R
n
e
T
a
1
, (24.104)
in which the last term on the right hand side is the small correction from the non-vanishing baryon-photon
velocity dierence, equation (24.103). As in the approximation (24.103), only the dominant term, the one
arising from the time derivative v
b
, has been retained in the correction, and in dierentiating the right
hand side of equation (24.103), the time derivative of 3R/( n
e
T
a) has been neglected compared to the time
derivative of

1
. The nal simplication is to replace the second time derivative of the photon dipole in
the correction term by its unperturbed value,

1
c
2
s
k
2
1
, an approximation that is valid because the
correction term is already small. Here c
s
is the sound speed, found in the next section to be given by
equation (24.110). The result is
v
b
+
a
a
v
b
+k = 3
_
1
+
a
a
1
+
k
3
_
+
3Rc
2
s
k
2
n
e
T
a

1
. (24.105)
This equation (24.105) is used to develop the photon-baryon wave equation in the next section, 24.16.
24.16 Photon-baryon wave equation
Combining the photon monopole and dipole equations (24.91a) and (24.91b) with the baryon momentum
equation (24.100b), and making the diusion approximation (24.94) for the photon quadrupole, and the
406

approximation (24.105) for the baryon bulk velocity, yields a second order dierential equation (24.113) that
captures accurately the behaviour of the photon distribution. The equation is that for a damped harmonic
oscillator, forced by the gravitational potential terms on its right hand side.
Adding the momentum equations (24.91b) and (24.100b) for photons and baryons yields an equation that
expresses momentum conservation for the combined photon-baryon uid,
1
+
k
3
(
0
2
2
) +
k
3
+
R
3
_
v
b
+
a
a
v
b
+k
_
= 0 . (24.106)
Setting the photon quadrupole
2
equal to its diusive value (24.94), and substituting the baryon velocity
equation (24.105), brings the photon-baryon momentum conservation equation (24.106) to
1
+
k
3
0
+
8k
2
27 n
e
T
a
1
+
k
3
+R
_
1
+
a
a
1
+
k
3
_
+
R
2
c
2
s
k
2
n
e
T
a

1
= 0 . (24.107)
Equation (24.107) rearranges to
_
d
d
+
R
1 +R
a
a
+
k
2
3 n
e
T
a(1 +R)
_
8
9
+
R
2
1 +R
__
1
+
k
3(1 +R)
0
+
k
3
= 0 , (24.108)
where equation (24.110) has been used to replace the sound speed c
s
. Finally, eliminating the dipole
1
in
favour of the monopole
0
using the photon monopole equation (24.91a) yields a second order dierential
equation for
0
:
_
d
2
d
2
+
_
R
1 +R
a
a
+
k
2
3 n
e
T
a(1 +R)
_
8
9
+
R
2
1 +R
__
d
d
+
k
2
3(1 +R)
_
(
0
) =
k
2
3(1 +R)
[(1 +R) + ] .
(24.109)
Equation (24.109) is a wave equation for a damped, driven oscillator with sound speed
c
s
=
1
3(1 +R)
, (24.110)
which is the adiabatic sound speed c
s
for a uid in which photons provide all the pressure, but both
photons and baryons contribute to the mass density. The term proportional to a/a on the left hand side of
equation (24.109) can be expressed as, since R a,
R
1 +R
a
a
= 2
c
s
c
s
. (24.111)
Dene the conformal sound time
s
by
d
s
c
s
d , (24.112)
with respect to which sound waves move at unit velocity, unit comoving distance per unit conformal time,
dx/d
s
= 1. Recast in terms of the conformal sound time
s
, the wave equation (24.109) becomes
_
d
2
d
2
s
+
_
s
c
s
+
k
2
c
s
n
e
T
a
_
8
9
+
R
2
1 +R
__
d
d
s
+k
2
_
(
0
) = k
2
[(1 +R) + ] , (24.113)
24.17 Damping of photon-baryon sound waves 407
where prime

denotes derivative with respect to conformal sound time, c
s
= dc
s
/d
s
.
The simple photon wave equation (23.45) derived in Chapter 23 is obtained from the wave equa-
tion (24.113) in the limit of negligible baryon-to-photon density, R 0.
24.17 Damping of photon-baryon sound waves
The terms proportional to the linear derivative d/d
s
in the wave equation (24.113) are damping terms,
the rst being an adiabatic damping term associated with variation of the sound speed, and the others
being dissipative damping terms associated respectively with the nite mean free path of electron-photon
scattering, and with viscous baryon drag. Lump these terms into a damping parameter dened by

1
2
_
s
c
s
+
k
2
c
s
n
e
T
a
_
8
9
+
R
2
1 +R
__
. (24.114)
The damping parameter varies slowly compared to the frequency of the sound wave, so can be treated
as approximately constant. The homogeneous wave equation (equation (24.113) with zero on the right hand
side) can then be solved by introducing a frequency dened by
0
e
R
ds
. (24.115)
The homogeneous wave equation (24.113) is then equivalent to
+
2
+ 2 +k
2
= 0 . (24.116)
Since the damping parameter is approximately constant (and the comoving wavevector k is by denition
constant),
is small compared to the other terms in equation (24.116). With
set to zero, the solution of

equation (24.116) is
= i
_
k
2
2
ik , (24.117)
where the last approximation is valid since the damping rate is small compared to the frequency, k.
Thus the homogeneous solutions of the wave equation (24.113) are
0
e
R
ds iks
. (24.118)
In the present case, the rst of the sources (24.114) of damping is the adiabatic damping term
a

1
2
c
s
c
s
. (24.119)
The integral of the adiabatic damping term is
_
a
d
s
=
1
2
ln c
s
, whose exponential is
e
R
a ds
=
c
s
. (24.120)
This shows that, as the sound speed decreases thanks to the increasing baryon-to-photon density in the
expanding Universe, the amplitude of a sound wave decreases as the square root of the sound speed.
408

The remaining damping terms are the dissipative terms
d

k
2
c
s
2 n
e
T
a
_
8
9
+
R
2
1 +R
_
. (24.121)
The integral of the dissipative damping terms is
_

d
d
s
=
k
2
k
2
d
, (24.122)
where k
d
is the damping scale dened by
1
k
2
d
_
c
s
2 n
e
T
a
_
8
9
+
R
2
1 +R
_
d
s
=
_
1
6 n
e
T
a(1 +R)
_
8
9
+
R
2
1 +R
_
d . (24.123)
The resulting damping factor is
e
d
ds
= e
k
2
/k
2
d
. (24.124)
Thus the eect of dissipation is to damp temperature uctuations exponentially at scales smaller than the
diusion scale k
d
.
With adiabatic and diusion damping included, the homogeneous solutions to the wave equation (24.113)
are approximately
c
s
e
k
2
/k
2
d
e
iks
. (24.125)
The driving potential on the right hand side of the wave equation (24.113) causes
0
to oscillate not
around zero, but rather around the oset [(1 +R) + ]. At the high frequencies where damping is
important, this driving potential also varies slowly compared to the wave frequency. To the extent that the
driving potential is slowly varying, the complete solution of the inhomogeneous wave equation (24.113) is
0
+ (1 +R)
c
s
e
k
2
/k
2
d
e
iks
. (24.126)
As will be seen in Chapter 25, the monopole contribution to CMB uctuations is not the photon monopole
0
by itself, but rather
0
+, which is the monople redshifted by the potential . This redshifted monopole
is
0
+ = R +A
c
s
e
k
2
/k
2
d
e
iks
. (24.127)
Exercise 24.4 Diusion scale. Show that the damping scale k
d
dened by (24.123) is given by, with a
normalized to a
eq
= 1,
H
2
eq
k
2
d
=
8
2Gm
p
9c
T
(1 X
n
)H
eq
b
_
a
0
a
2
X
e
1 +a(1 +R)
_
8
9
+
R
2
1 +R
_
da , (24.128)
the Hubble parameter H
eq
at matter-radiation equality being related to the present-day Hubble parameter
H
0
by equation (23.41). If baryons are taken to be fully ionized, X
e
= 1, which ceases to be a good
24.18 Ionization and recombination 409
approximation near recombination, then the integral can be done analytically. At times well after matter-
radiation equality,
f(a)
_
a
0
a
2
1 +a(1 +R)
_
8
9
+
R
2
1 +R
_
da
2
5
a
5/2
for a 1 , (24.129)
independent of the value of the constant R
a
R/a. Conclude that, neglecting the eect of recombination
on the electron fraction X
e
,
H
2
eq
k
2
d
=
6.821 h
1
1 X
n
H
0
H
eq
b
f(a/a
eq
) = 4.9 10
4
f(a/a
eq
) 0.0016
_
a
a
_
5/2
, (24.130)
where a
is the cosmic scale factor at recombination.

24.18 Ionization and recombination
24.19 Neutrinos
Before electron-positron annihilation at temperature T 1 MeV, weak interactions were fast enough that
scattering between neutrinos, antineutrinos, electrons, and positrons kept neutrinos and antineutrinos in
thermodynamic equilibrium with baryons. After e e annihilation, neutrinos and antineutrinos decoupled,
rather like photons decoupled at recombination. After decoupling, neutrinos streamed freely.
24.20 Summary of equations
Non-baryonic cold dark matter, baryons, photons, neutrinos:
c
kv
c
= 3

, (24.131a)
v
c
+
a
a
v
c
= k , (24.131b)
b
kv
b
= 3

, (24.131c)
v
b
+
a
a
v
b
= k
n
e
T
a
R
(v
b
3
1
) , (24.131d)
0
k
1
=

, (24.131e)
1
+
k
3
(
0
2
2
) =
k
3
+
1
3
n
e
T
a (v
b
3
1
) , (24.131f)
2k
5

1
=
9
10
n
e
T
a
2
, (24.131g)
410

Einstein energy and quadrupole pressure equations:
k
2
3
a
a
F = 4Ga
2
(
c
c
+
b
b
+ 4
0
+ 4
^
0
) , (24.132a)
k
2
() = 32Ga
2
(
r
2
+
^
2
) , (24.132b)
24.21 Legendre polynomials
The Legendre polynomials P
() satisfy the orthogonality relations

_
1
1
P
()P
() d =
2
2 + 1

(24.133)
and
_
P
( z a)P
( z
b) do
z
=
4
2 + 1
P
( a
b)
, (24.134)
the recurrence relation
P
() =
1
2 + 1
[P
1
() + ( + 1)P
+1
()] , (24.135)
and the derivative relation
dP
()
d
=
+ 1
1
2
[P
1
() P
+1
()] . (24.136)
The rst few Legendre polynomials are
P
0
() = 1 , P
1
() = , P
2
() =
1
2
+
3
2
2
. (24.137)
25
Fluctuations in the Cosmic Microwave
Background
25.1 Primordial power spectrum
Ination generically predicts gaussian initial uctuations with a scale-free power spectrum, in which the
variance of the potential is the same on all scales,
(x
)(x))
([x
x[) = constant , (25.1)

independent of separation [x
x[. A scale-free primordial power spectrum was originally proposed as

a natural initial condition by Harrison and Zeldovich (Harrison, 1970, PRD 1, 2726; Zeldovich, 1972,
MNRAS, 160, 1P), before the idea of ination was conceived. During ination, vacuum uctuations generate
uctuations in the potential which become frozen in as they y over the horizon. The amplitude of these
uctuations remains constant as ination continues, producing a scale-free power spectrum.
The power spectrum P
(k) of potential uctuations is dened by

(k
)(k)) (2)
3
D
(k
+k)P
(k) , (25.2)
the power spectrum P
(k) being related to the correlation function
(x) by (with the standard convention

in cosmology for the choice of signs and factors of 2)
P
(k) =
_
e
ikx
(x) d
3
x ,
(x) =
_
e
ikx
P
(k)
d
3
k
(2)
3
. (25.3)
The scale-free character means that the dimensionless power spectrum
2
(k) dened by
(k) P
(k)
4k
3
(2)
3
(25.4)
is constant.
Actually, the power spectrum generated by ination is not precisely scale-free, because ination comes to
an end, which breaks scale-invariance. The departure from scale-invariance is conventionally characterized
by a scalar spectral index, the tilt n, such that
(k) k
n1
. (25.5)
412 Fluctuations in the Cosmic Microwave Background
Thus a scale-invariant power spectrum has
n = 1 (scale-invariant) . (25.6)
Dierent inationary models predict dierent tilts, mostly close to but slightly less than 1.
25.2 Normalization of the power spectrum
It is convenient to normalize the power spectrum of potential uctuations to its amplitude at the recombi-
nation distance today, k(
0
) = 1, and not to the initial potential (0) (which vanishes for isocurvature
initial conditions), but rather to the late-time matter-dominated potential (late) at superhorizon scales,
2
(late)
(k) = A
2
late
[k(
0
)]
n1
. (25.7)
The relation between the superhorizon late-time matter-dominated potential (late) and the primordial
potential (0) is, equations (23.61),
(late) =
_
9
10
(0) adiabatic ,
8
5
(0) isocurvature .
(25.8)
25.3 CMB power spectrum
The power spectrum of temperature uctuations in the CMB is dened by
C
(
0
) 4
_

0
[
(
0
, k)[
2
d
3
k
(2)
3
. (25.9)
As seen in Chapter 23, during linear evolution, scalar modes of given comoving wavevector k evolve with
amplitude proportional to the initial value (0, k) of the scalar potential (or of its derivative
(0, k), for

isocurvature initial conditions). The evolution of the amplitude may be encapsulated in a transfer function
T
(, k) dened by
T
(, k)

(, k)
(late, k)
, (25.10)
where (late) is the superhorizon late-time matter-dominated potential. By isotropy, the transfer function
T
(, k) is a function only of the magnitude k of the wavevector k. The power spectrum of the CMB observed
on the sky today is related to the primordial power spectrum P
(late)
(k) or
2
(late)
(k) by
C
(
0
) = 4
_

0
[T
(
0
, k)[
2
P
(late)
(k)
4k
2
dk
(2)
3
= 4
_

0
[T
(
0
, k)[
2
2
(late)
(k)
dk
k
. (25.11)
25.4 Matter power spectrum 413
25.4 Matter power spectrum
The matter power spectrum P
m
(k) is dened by
m
(k
)
m
(k)) (2)
3
D
(k
+k)P
m
(k) . (25.12)
At times well after recombination, the matter power spectrum P
m
(, k) at conformal time is related to the
potential power spectrum (25.2) by, equation (23.117), in units a
0
= 1,
P
m
(, k) =
_
2a
3
m
H
2
0
_
2
k
4
P
(, k) =
_
2a
3
m
H
2
0
_
2
(2)
3
4
k
2
(k, ) . (25.13)
At superhorizon scales, the potential (, k) at conformal time is related to the late-time matter-dominated
potential by
() = g(a) (late) for k 1 , (25.14)
where g(a) is the grown factor dened by equation (23.116). For a power-law primordial spectrum (25.5),
the matter power spectrum at the largest scales goes as
P
m
(, k) k
n
, (25.15)
which explains the origin of the scalar index n.
25.5 Radiative transfer of CMB photons
To determine the harmonics
(
0
, k) of the CMB photon distribution at the present time, return to the
Boltzmann equation (24.90) for the photon distribution (, k, ), where k p:
ik
+ n
e
T
a =

+ ik
+ n
e
T
a
_
i
v
b
+
0
1
2
2
P
2
(
. (25.16)
This equation is also called the radiation transfer equation. The terms on the right are the source terms.
Dene the electron-photon (Thomson) scattering optical depth by
d
d
n
e
T
a , (25.17)
starting from zero,
0
= 0, at zero redshift, and increasing going backwards in time to higher redshift. The
photon Boltzmann equation (25.16) can be written
e
ik+
d
d
_
e
ik
_
= S
0
i
S
1
+ (i
)
2
S
2
, (25.18)
where S
i
are source terms
S
0

_
0
+
1
4
2
_
, (25.19a)
S
1
k v
b
= k3
1
, (25.19b)
S
2

3
4
2
, (25.19c)
where in S
1
the tight-coupling approximation v
b
= 3
1
has been used to replace the baryon bulk velocity
v
b
with the photon dipole
1
. Thus a solution for the photon distribution (
0
, k,
) today is, at least

formally, an integral over the line of sight from the Big Bang to the present time,
(
0
, k,
) =
_
0
0
_
S
0
i
S
1
+ (i
)
2
S
2
e
ik(0)
d . (25.20)
The i
dependence of the source terms inside the integral can be accomodated through
(i
)
n
e
ik(0)
=
_
1
k
_
n
e
ik(0)
, (25.21)
which brings the formal solution (25.20) for (
0
, k,
) to
(
0
, k,
) =
_
0
0
e
_
S
0
+S
1
1
k
+S
2
1
k
2
2
_
e
ik(0)
d . (25.22)
The dipole source term S
1
contains a term which it is helpful to integrate by parts:
_
0
0
e
e
ik(0)
d =
_
e
e
ik(0)
_
0
0
+
_
0
0
_
e
_
e
ik(0)
d
= (
0
) +
_
0
0
e

_
e
ik(0)
d . (25.23)
With the source terms written out explicitly, the present-day photon distribution (
0
, k,
), equation (25.20),
is
(
0
, k,
) + (
0
, k)
=
_
0
0
_
e
_
e
0
+ +
1
4
2
+ 3
1
1
k
+
3
4
2
1
k
2
2
__
e
ik(0)
d . (25.24)
The spherical harmonics
(, k) of the photon distribution have been dened by equation (24.48). The

e
ik(0)
factor in the integral in equation (25.24) can be expanded in spherical harmonics through the
general formula
e
ikx
=
=0
i
(2 + 1)j
(kx)P
k x) , (25.25)
where j
(z)
_
/(2z)J
+1/2
(z) are spherical Bessel functions. Resolved into harmonics, equation (25.24)
becomes
[
(
0
, k) +
0
(
0
, k)]
=
_
0
0
_
e
_
e
0
+ +
1
4
2
+ 3
1
1
k
+
3
4
2
1
k
2
2
__
j
[k(
0
)] d . (25.26)
Introduce a visibility function g() dened by
g() e
, (25.27)
25.6 Integrals over spherical Bessel functions 415
whose integral is one,
_
0
0
g() d =
_
0
d =
_
e
= 1 . (25.28)
The visibility function is fairly narrowly peaked around recombination at =
. In this approximation of
instantaneous recombination,
[
(
0
, k) +
0
(
0
, k)] =
_
0
0
e
(, k) +

(, k)
_
j
[k(
0
)] d ISW
+ [
0
(
, k) + (
, k)] j
[k(
0
)] monopole
3
1
(
, k) j
[k(
0
)] dipole
+
2
(
, k)
_
1
4
j
[k(
0
)] +
3
4
j
[k(
0
)]
_
quadrupole .
(25.29)
Here prime

on j
denotes a total derivative, j
(z) = dj
(z)/dz. The rst and second derivatives of the

spherical Bessel functions are
j
(z) =

z
j
(z) j
+1
(z) , j
(z) =
( 1) z
2
z
2
j
(z) +
2
z
j
+1
(z) . (25.30)
Notice that the monopole term (on both sides of equation (25.29)) is not
0
but rather
0
+ , which
is the temperature uctuation redshifted by the potential . On the left hand side, the
0
term arises
from the redshift at our position today, but this monopole perturbation just adds to the mean unperturbed
termperature, and is not observable.
25.6 Integrals over spherical Bessel functions
Computing the photon harmonics
l
by integration of equation (25.26), or the CMB power spectrum C
in the instantaneous recombination approximation by integration of equation (25.11) with (25.29), involves
evaluating integrals of the form
_

0
f(z)g(qz)
dz
z
, (25.31)
with
g(z) j
(z)z
n1
or g(z) j
(z)j
(z)z
n1
. (25.32)
Such integrals present a challenge because of the oscillatory character of the functions g(z). This section
presents a method to evaluate such integrals reliably. Integrals with dierent are related by recursion
relations (25.38) that permit rapid evaluation over many . Details are left to Exercise 25.2.
The approach is to recast the integral (25.31) into Fourier space with respect to ln z, and to apply a
Fast Fourier Transform of f(z) over a logarithmic interval. This involves replacing the true f(z) with a
function that is periodic in lnz over a logarithmic interval [L/2, L/2] of width L centred on ln z = 0. The
approximation works because the functions g(z) given by equation (25.32) tend to zero at z 0 and z ,
so spurious periodic duplications at small and large z contribute negligibly to the integral (25.31) provided
that the logarithmic interval L is chosen suciently broad. The logarithmic interval L can be broadened to
whatever extent is necessary by extrapolating the function f(z) to smaller and larger z. If f(z) is periodic
in ln z over a logarithmic interval L, then f(z) is a sum of discrete Fourier modes e
2imln(z)/L
in which m is
integral. If f(z) is a smooth function, then f(z) may be adequately approximated by a nite number N of
discrete modes m = [(N 1)/2], ..., [N/2], where [N/2] denotes the largest integer greater than or equal to
N/2. Under these circumstances, the function f(z) is given by a discrete Fourier expansion whose Fourier
components f
m
are related to the values f(z
n
) at N logarithmically spaced points z
n
= e
nL/N
,
f(z) =
[N/2]
m=[(N1)/2]
f
m
e
2imln(z)/L
, f
m
=
1
N
[N/2]
n=[(N1)/2]
f(z
n
)e
2imn/N
. (25.33)
Since f(z) is real, the negative frequency modes are the complex conjugates of the positive frequency modes,
f
m
= f
m
, (25.34)
and for even N (the usual choice) the outermost (Nyquist) frequency mode f
N/2
is real. For a function f(z)
given by the Fourier sum (25.33), the integral (25.31) is
_

0
f(z)g(qz)
dz
z
=
[N/2]
m=[(N1)/2]
f
m
q
2im/L
_

0
g(z)z
2im/L
dz
z
. (25.35)
For functions g(z) given by equation (25.32), the integrals over g(z) on the right hand side of equation (25.35)
can be done analytically:
_

0
f(z)g(qz)
dz
z
=
[N/2]
m=[(N1)/2]
f
m
q
2im/L
U
_
n 1 +
2im
L
_
, (25.36)
where for g(z) = j
(z)z
n1
or g(z) = j
(z)j
(z)z
n1
the function U(x) is respectively
U
(x)
_

0
j
(z)z
x
dz
z
=
2
x2
_
1
2
( +x)
_
1
2
( x + 3)
, (25.37a)
U
(x)
_

0
j
(z)j
(z)z
x
dz
z
=
2
x3
(2 x)
_
1
2
( +
+x)
_
1
2
( +
x + 4)
_
1
2
(
x + 3)
_
1
2
(
x + 3)
. (25.37b)
25.7 Large-scale CMB uctuations 417
Recurrence relations such as
U
(x) = ( +x 2) U
1
(x 1) (25.38a)
=
+x 2
x + 1
U
2
(x) , (25.38b)
U
(x) =
+
+x 2
+
x + 2
U
1,
1
(x) (25.38c)
=
( +
+x 2)(
+x 3)
2(x 2)
U
1,
(x 1) (25.38d)
=
( +
+x 2)(
+x 3)
( +
x + 2)(
+x 1)
U
2,
(x) , (25.38e)
permit rapid evaluation of the functions U
(x) or U
(x) as a function of and
, starting from small ,
.
As written, the right hand side of equation (25.36) has a small imaginary part arising from the contribution
of the outermost (Nyquist) mode, m = [N/2], but this imaginary part should be dropped since it cancels
when averaged with the contribution of its negative frequency partner m = [N/2].
25.7 Large-scale CMB uctuations
The behaviour of the CMB power spectrum at the largest angular scales was rst predicted by R. K. Sachs
& A. M. Wolfe (1967, Astrophys. J., 147, 73), and is therefore called the Sachs-Wolfe eect, though why
it should be called an eect is mysterious. The Sachs-Wolfe (SW) eect is distinct from, but modulated by,
the Integrated Sachs-Wolfe (ISW) eect. The ISW eect, ignored in this section, is considered in 25.9.
At scales much larger than the sound horizon at recombination, k
s,
1, the redshifted monopole
uctuation
0
(
, k) + (
, k) at recombination is much larger than the dipole

1
(
, k) or quadrupole
2
(
, k), so only the monopole contributes materially to the temperature multipoles
(
0
, k) today. The
redshifted monopole contribution to the temperature multipoles
(
0
, k) today is, from equation (25.29),
(
0
, k) = [
0
(
, k) + (
, k)] j
[k(
0
)] . (25.39)
At the very largest scales, k
eq
1, the solution for the redshifted radiation monopole
0
+ at the time
of recombination is, from equation (23.62),
0
(
, k) + (
, k) = 2
super
(
, k)
3
2
(0)
A
SW
(
)
A
late
super
(late, k) , (25.40)
where the last expression denes the Sachs-Wolfe amplitude A
SW
(
) at recombination. In the approximation

that recombination happens well into the matter-dominated regime, so that
super
(
, k)
super
(late, k),
from equations (23.61),
A
SW
(
)
A
late
A
SW
(late)
A
late
= 2
3(0)
2(late)
=
_
1
3
adiabatic ,
2 isocurvature .
(25.41)
In reality, recombination occurs only somewhat into the matter-dominated regime, and the solutions for
the potential
super
(
, k) from 23.9 should be used in place of the approximation (25.41). Putting equa-
tions (25.39) and (25.40) together shows that the transfer function T
(
0
, k), equation (25.10) that goes into
the present-day CMB angular power spectrum C
(
0
), equation (25.11), is
T
(
0
, k) =
A
SW
(
)
A
late
j
[k(
0
)] . (25.42)
If the primordial power spectrum is a power law with tilt n, equation (25.7), then the resulting CMB angular
power spectrum is, with z = k(
0
),
C
(
0
) = 4A
SW
(
)
2
_

0
j
(z)
2
z
n1
dz
z
= 4A
SW
(
)
2
U
,
(n 1) , (25.43)
where U
,
(x) is given by equation (25.37b). For the particular case of a scale-invariant primordial power
spectrum, n = 1, the CMB power spectrum C
at large scales today is given by

( + 1)C
(
0
) = 2A
SW
(
)
2
if n = 1 . (25.44)
Thus the characteristic feature of a scale-invariant primordial power spectrum, n = 1, is that ( + 1)C
should be approximately constant at the largest angular scales,

0
/
. This is a primary reason why

CMB folk routinely plot ( + 1)C
, rather than C
.
25.8 Monopole, dipole, and quadrupole contributions to C
At smaller scales, k
>
1, not only the photon monopole

0
(
, k), but also the dipole

1
(
, k), and to
a small extent the quadrupole
2
(
, k), contribute to the temperature multipoles
(
0
, k) today, equa-
tion (25.29). The dipole is related to the monopole by the evolution equation (??) for the monopole.
k
1
=

. (25.45)
25.9 Integrated Sachs-Wolfe (ISW) eect
Concept question 25.1 Cosmic Neutrino Background. Just as photons decoupled at recombina-
tion, so also neutrinos decoupled at electron-positron annihilation. Compare qualitatively the expected
uctuations in the CB to those in the CMB.
25.9 Integrated Sachs-Wolfe (ISW) eect 419
Exercise 25.2 Numerical integration of sequences of integrals over Bessel functions. Write code
that solves integrals (25.31) numerically for g(z) given by equation (25.32), using a Fast Fourier Transform,
equation (25.36), amd recurrence relations appropriate for the monopole, dipole, and quadrupole contribu-
tions to equations (25.26) or (25.29). To compute U
(x) or U
(x) for the initial ,
, you will need to nd

code that implements the complex Gamma function. Note that most FFT codes store input and output
periodic sequences shifted by [N/2] compared to the convention (25.33). That is, an FFT code typically
takes an input sequence ordered as f(z
0
), f(z
1
), ..., f(z
[N/2]
), ..., f(z
2
), f(z
1
), with periodic identication
f(z
n
) = f(z
n+N
), and evaluates Fourier coecients f
m
as
f
m
=
1
N
N1
m=0
f(z
n
)e
2imn/N
, (25.46)
which yields the same Fourier components f
m
as (25.33) but in the order f
0
, f
1
, ..., f
[N/2]
, ..., f
2
, f
1
, with
periodic identication f
m
= f
m+N
.
Various recurrence relations. For g(z) = j
(z)z
n1
, the dipole and quadrupole integrals U
(1)
(x) and
U
(2)
(x) are related to the monopole integral U
(x) (25.37a) by
U
(1)
(x)
_

0
j
(z)z
x
dz
z
= (1 x) U
(x 1) , (25.47a)
U
(2)
(x)
_

0
_
1
4
j
(z) +
3
4
j
(z)
z
x
dz
z
=
( + 1) + 2x(x 2)
( +x 2)( x + 3)
U
(x) . (25.47b)
The dipole, and quadrupole integrals satisfy the recurrence relations
U
(1)
(x) =
+x 3
x + 2
U
(1)
2
(x) , (25.48a)
U
(2)
(x) =
( +x 4) [l(l + 1) + 2x(x 2)]
( x + 3) [l(l 3) + 2(x 1)
2
]
U
(2)
2
(x) . (25.48b)
For g(z) = j
(z)j
(z)z
n1
, the relations get a bit ugly its a good idea to use Mathematica or a similar
program to generate the relations automatically. The various multipole integrals of interest are related to
the integral U
(x) (25.37b) by
U
(0,1)
(x)
_

0
j
(z)j
(z)z
x
dz
z
=
1 x
2
U
(x 1) , (25.49a)
U
(1,1)
(x)
_

0
[j
(z)]
2
z
x
dz
z
=
4( + 1) x(x 2)(x 3)
(3 x)(2 +x 2)(2 x + 4)
U
(x) , (25.49b)
U
(0,2)
(x)
_

0
j
(z)
_
1
4
j
(z) +
3
4
j
(z)
z
x
dz
z
=
x[2( + 1) + (x 1)(x 2)]
2(x 3)(2 +x 2)(2 x + 4)
U
(x) , (25.49c)
U
(1,2)
(x)
_

0
j
(z)
_
1
4
j
(z) +
3
4
j
(z)
z
x
dz
z
(25.49d)
=
(1 x) [2( + 1)(x 7) + (x + 1)(x 3)(x 4)]
4(x 4)(2 +x 3)(2 x + 5)
U
(x 1) ,
U
(2,2)
(x)
_

0
_
1
4
j
(z) +
3
4
j
(z)
2
z
x
dz
z
= (25.49e)
4( 1)( + 1)( + 2) [12 +x(x 2)] +x(x 2)(x 5) [4( + 1)(x 6) + (x + 2)(x 3)(x 4)]
4(x 3)(x 5)(2 +x 2)(2 +x 4)(2 x + 4)(2 x + 6)
U
(x) .
The various recurrence relations of interest are
U
(0,1)
(x) =
(2 +x 3)
(2 x + 3)
U
(0,1)
1,1
(x) , (25.50a)
U
(1,1)
(x) =
(2 +x 4) [4( + 1) x(x 2)(x 3)]
(2 x + 4) [4( 1) x(x 2)(x 3)]
U
(1,1)
1,1
(x) , (25.50b)
U
(0,2)
(x) =
(2 +x 4) [2( + 1) + (x 1)(x 2)]
(2 x + 4) [2( 1) + (x 1)(x 2)]
U
(0,2)
1,1
(x) , (25.50c)
U
(1,2)
(x) =
(2 +x 5) [2( + 1)(x 7) + (x + 1)(x 3)(x 4)]
(2 x + 5) [2( 1)(x 7) + (x + 1)(x 3)(x 4)]
U
(1,2)
1,1
(x) , (25.50d)
U
(2,2)
(x) =
(2 +x 6)
(2 x + 6)
U
(2,2)
1,1
(x) (25.50e)
4( 1)( + 1)( + 2) [12 +x(x 2)] +x(x 2)(x 5) [4( + 1)(x 6) + (x + 2)(x 3)(x 4)]
4( 1)( + 1)( 2) [12 +x(x 2)] +x(x 2)(x 5) [4( 1)(x 6) + (x + 2)(x 3)(x 4)]
.

General Relativity, Black Holes, and Cosmology

Uploaded by

Document Information

Original Description:

Original Title

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, Black Holes, and Cosmology

Uploaded by

Copyright:

Available Formats

General Relativity, Black Holes, and Cosmology

denotes coordinate partial derivatives, which commute.

from vertical in a spacetime diagram. Each point in 4-dimensional spacetime is called an

) in Bs frame from the assumptions:

= t at the back and t

+vl = t at the front gives

1 and w is a fourth spatial coordinate.

; (b) a coordinate interval dx

; (c) proper time

mean physically? Is it a tensor? Why, or why not?

mean physically? Is it a tensor?

Riemann curvature tensor R

3. 4-velocity and 4-momentum. Geodesic equation.

is the Einstein tensor, and T

is the metric, a 4 4 symmetric matrix

is a 4D vector object, with both magnitude

with the notation

of spacetime can be expressed in coordinate-independent fashion as the abstract vector

to be the inverse metric, satisfying

and its inverse g

provide the means of lowering and raising coordinate indices. The

with raised index are called its contravariant components, while

with lowered indices are called its covariant components,

is consisistent with the

transform as a covariant coordinate 4-vector.

are coordinate 4-vectors, then their scalar product is

(with a dummy index ) and A A

and A are both vectors, they are mathematically dierent objects.

, identied by Greek (brown) indices , is one that changes in a prescribed way

and its inverse g

are indeed coordinate tensors, transforming

is a coordinate tensor (2.29)

is not a coordinate tensor (2.31)

, being by construction invariant under coordinate transformations, is a

is a coordinate tensor . (2.32)

are dened in the usual way by

(x) at some point x with the vectors g

(x+x) in the denition (2.33) should be interpreted as its

that appear on the right hand side of equation (2.32) dene

, also known as Christoel symbols,

is not a coordinate tensor . (2.34)

is a coordinate tensor . (2.36)

is a coordinate tensor (2.38)

is the covariant derivative of the covariant coordinate 4-vector A

is a coordinate tensor . (2.39)

, the tangent axes transformas g

of partial derivatives provides 16 degrees of freedom in choosing the tangent axes

, but leave the

at a point by adjusting the matrix x

, so also it is possible to reorient the derivatives

of the tangent axes by adjusting the matrix

of second partial derivatives. The second

, all of which the condition (2.43) requires to

was dened by equation (2.15) as the coordinate-independent

. The notation dx was merely symbolic: dx was not necessarily a total

is dened by the commutator of the covariant derivative acting on a scalar

is a coordinate tensor . (2.50)

is a coordinate tensor (2.51)

2.6.10 Connection coecients in terms of the metric

with all indices lowered is dened to be

. Elements of the dual vector space are

be the 4-velocity in coordinate-independent notation. The principle of equivalence implies