BANGS, CRUNCHES,

WHIMPERS, AND SHRIEKS

This page intentionally left blank
 BANGS, CRUNCHES,
WHIMPERS, AND SHRIEKS
Singularities and Acausalities
in Relativistic Spacetimes

JOHN EARMAN

New York Oxford

OXFORD U N I V E R S I T Y PRESS
1995
 Oxford University Press
Oxford New York
Athens Auckland Bangkok Bombay
Calcutta Cape Town Dar es Salaam Delhi
Florence Hong Kong Istanbul Karachi
Kuala Lumpur Madras Madrid Melbourne
Mexico City Nairobi Paris Singapore
Taipei Tokyo Toronto
and associated companies in
Berlin Ibadan

Copyright © 1995 by Oxford University Press, Inc.

Published by Oxford University Press, Inc.
198 Madison Avenue, New York, New York 10016
Oxford is a registered trademark of Oxford University Press
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording, or otherwise,
without the prior permission of Oxford University Press

Library of Congress Cataloging-in-Publication Data

Barman, John.
Bangs, crunches, whimpers, and shrieks : singularities and
acausalities in relativistic spacetimes / John Earman.
p. cm.
Includes bibliographical references and index.
ISBN 0-19-509591-X
1. Space and time. 2. General relativity (Physics)
3. Singularities (Mathematics) I. Title.
QC173.59.S65E15 1995
530.1'1—dc20 95-49132

Printing: 1 3 5 7 9 8 6 4 2
Printed in the United States of America
on acid-free paper
For David, Bob , John2, Larry, Tim,
Roberto and all the others
who have sustained me in my labors
This page intentionally left blank
 PREFACE

The editor of Philosophy of Science, the official journal of the Philosophy of

Science Association, informs me that in recent years only a small fraction of
the submitted papers deal with the foundations of general relativity theory,
and of these only a few focus on issues connected with spacetime singularities.
This is not because the journal discourages technical papers or papers
concerned with foundational issues in physics; indeed, it routinely publishes
highly technical articles on the problems of measurement, hidden variables,
and non-locality in quantum mechanics. Part of the difference may lie in the
fact that while anyone who can understand the notion of a vector space can
be brought to appreciate the measurement problem in quantum mechanics
and while even undergraduates can derive versions of Bell's inequalities, the
analysis of spacetime singularities involves comparatively difficult mathe-
matics. But this cannot be the whole answer since philosophers of science are
not strangers to differential geometry. Rather I think that the relative neglect
of this area lies in the failure of philosophers to appreciate the seriousness of
the foundational issues posed by singularities in general relativity and the
importance of these issues for the philosophy of space and time and for the
philosophy of science in general.
The present book is dedicated to the goal of ending the neglect. It is
aimed primarily at philosophers who have some prior acquaintance with
relativity theory and secondarily at philosophically minded physicists. It
makes no pretense at being a comprehensive survey of the subject. I doubt
that any philosopher at work today is capable of coping with all of the
history, philosophy, physics, and mathematics needed to produce such a
survey; and even if there were such a polymath, the resulting work would
likely be an indigestible tome that would gather dust on the shelves of the
few libraries that could afford to buy it. Rather than strive for comprehensive-
ness, my strategy is to cover some core topics and sufficiently many other
topics so that the reader will not be overwhelmed but can nevertheless get a
good feel for the topography. The level of presentation is not mathematically
rigorous, but I hope it is rigorous enough to illuminate the relevant technical
issues in general relativity and to reveal their connection with philosophical
issues about the nature of space, time, causality, and the laws of nature. The
chapters are interlinked but are designed to be self-contained enough that
the reader who tires of one topic, without too much effort, will be able to
turn to sampling another. Whether or not readers agree with my analyses
and positions, I will be content if they put down this book believing that if
viii Preface

philosophers were doing their jobs in trying to fathom the implications of

spacetime singularities, the result would be one of the more exciting areas in
the philosophy of science.
For those readers who want to consult background references in the
relavant branches of mathematics and physics, I have several suggestions.
There are many good textbooks on general relativity. For present purposes,
Robert Wald's General Relativity (1984a) is especially recommended. Hawking
and Ellis's classic, The Large Scale Structure of Space-Time (1973), though now
out of date, is an indispensable source of key results on singularities. A
semipopular but very insightful treatment of several of the topics studied here
is to be found in Kip Thome's Black Holes and Time Warps (1994). The most
recent technical treatises on singularities are P. S. Joshi's Global Aspects in
Gravitation and Cosmology (1993) and C. J. S. Clarke's Analysis of Spacetime
Singularities (1993).
Chapter 3 was based on two of my articles: "The Cosmic Censorship
Hypothesis," inj. Earman, A. I. Janis, G. J. Massey, and N. Rescher (eds.),
Philosophical Problems of the Internal and External Worlds: Essays on the Philosophy
of Adolf Grunbaum (Pittsburgh: University of Pittsburgh Press, 1993), and
"Cosmic Censorship," PSA 1992, Vol. II (East Lansing: Philosophy of Science
Association, 1993). Most of the material in Chapter 4 was coauthored by
John Norton and appeared as "Forever Is a Day: Super-tasks in Pitowsky
and Malament-Hogarth Spacetimes," Philosophy of Science, 60 (1993) 22-42.
Chapter 6 was based on "Recent Work on Time Travel," in Steven Savitt
(ed.), Time's Arrow Today (Cambridge: Cambridge University Press, 1994).
I am grateful to the editors of these books and to the presses and journals for
their kind permission to reprint this material.
Parts of this book were written while I was a participant in the research
group on semantical aspects of spacetime theories at the Zentrum fur
Interdisciplinare Forschung (ZiF), Universitat Bielefeld. I am grateful to
Ulrich Majer and Heinz-Jiirgen Schmidt for organizing the group and to the
ZiF for its financial support.
I am indebted to a number of people for ideas and helpful suggestions.
At the risk of failing to mention some of them, I would like to thank Chris
Clarke, Jean Eisenstaedt, John Norton, and John Stachel. Special thanks are
due to David Malament, Tim Maudlin, Roberto Torretti, and Robert Wald
for detailed comments on an earlier draft. Andrew Backe provided invaluable
help in preparing the bibliography

J.E.
Pittsburgh, Pennsylvania
August 1994
 CONTENTS

1. Introducing Spacetime Singularities and Acausalities, 3

1.1 Introduction, 3
1.2 Spacetime singularities: In the beginning, 5
1.3 Einstein's intolerance of singularities, 11
1.4 Acausality and time travel, 21
1.5 Singularities and acausalities together, 22

2. Defining, Characterizing, and Proving the Existence of

Spacetime Singularities, 27
2.1 Introduction, 27
2.2 What is a spacetime singularity? 28
2.3 Extensions of spacetimes, 31
2.4 The received definition of singularities, 33
2.5 The missing missing points, 40
2.6 Naked singularities, 44
2.7 What is a spacetime singularity (again)? 46
2.8 Singularity theorems, 50
2.9 Singularities and quantum effects, 56
2.10 Conclusion, 58

2. Cosmic Censorship, 64
3.1 Introduction, 64
3.2 Cozying up to singularities, 65
3.3 Naked singularities and cosmic censorship, 67
3.4 The cosmic censorship hypothesis, 80
3.5 Is the cosmic censorship hypothesis true? 86
3.6 Black hole evaporation, 90
3.7 What if cosmic censorship should fail? 92
3.8 A dirty open secret, 97
3.9 Conclusion, 98

4. Supertasks, 103
4.1 Introduction, 103
4.2 Pitowsky spacetimes, 105
4.3 Malament-Hogarth spacetimes, 107
x Contents

4.4 Paradoxes regained? 108

4.5 Characterization of Malament-Hogarth spacetimes, 110
4.6 Supertasks in Malament—Hogarth spacetimes, 114
4.7 Malament-Hogarth spacetimes and unresolved mathematical
conjectures, 116
4.8 Can y x carry out the assigned task? 119
4.9 Conclusion, 119
Appendix: Proofs of Lemma 4.2 and Equation 4.4, 120

5. The Big Bang and the Horizon Problem, 124

5.1 Introduction, 124
5.2 Observability and light cones, 125
5.3 What can we predict about the future? 128
5.4 Event and particle horizons, 130
5.5 What is the horizon problem? 134
5.6 Reichenbach's principle of common cause, 135
5.7 Particle horizons and common causes, 137
5.8 Diagnosing the bellyache: Electromagnetism, 140
5.9 Diagnosing the bellyache: Cosmic background radiation, 142
5.10 Strategies for solving the horizon problem, 147
5.11 Horizons in standard and inflationary models, 150
5.12 Does inflation solve the horizon problem? 152
5.13 Conclusion, 155

6. Time Travel, 160

6.1 Introduction, 160
6.2 Types of time travel; backward causation, 161
6.3 The causal structure of relativistic spacetimes, 164
6.4 Why take Godelian time travel seriously? 167
6.5 The paradoxes of time travel, 170
6.6 Consistency constraints, 173
6.7 Therapies for time travel malaise, 175
6.8 Non self-interacting test fields, 179
6.9 Self-interacting test fields, 183
6.10 Can we build and operate a time machine? 188
6.11 Conclusion, 193
Appendix: Godel on the ideality of time, 194

7. Eternal Recurrence, Cyclic Time, and All That, 203

7.1 Introduction, 203
7.2 Tolman on eternal recurrence, 204
7.3 Extending through the big bang and the big crunch, 205
 Contents xi

7.4 Finding God in the big bang, 207

7.5 No recurrence theorems, 210
7.6 Cyclic time, 213
7.7 Conclusion, 218

8. Afterword, 223

References, 228

Index,247
This page intentionally left blank
 BANGS, CRUNCHES,
WHIMPERS, AND SHRIEKS
This page intentionally left blank
 1
Introducing Spacetime Singularities
and Acausalities

1.1 Introduction

This book is simultaneously an essay in the philosophy of science and the

foundations of physics. It seeks to assess the implications of spacetime
singularities and acausal features of relativistic spacetimes for the general
theory of relativity and the philosophy of space and time. To the extent that
it succeeds, it should succeed in convincing the reader that for the constellation
of issues to be studied, physics, foundations of physics, and the philosophy of
science are all of a piece. Before trying to describe the issues, it may be helpful
to begin by setting them in context.
The two most fundamental theories of twentieth century physics, quantum
mechanics (QM) and the general theory of relativity (GTR), stand in war
regard of one another. Each is spectacularly successful in its own domain of
application, both in making accurate predictions and in providing conceptual
understanding of otherwise puzzling phenomena. In the latter achievements
lie the roots of what is arguably the most important challenge of contemporary
physics. It is hard to see how any future theory that incorporates the successes
of both QM and GTR can dispense with the conceptual apparatus of either.
But combining them into a unified quantum theory of gravity has proved to
be formidable and is thus far unfulfilled.
Despite their successes, QM and GTR are beset by problems that raise
worries about the foundations of these theories—QM by the measurement
problem and associated problems of non-locality, GTR by the problems that
are the focus of this study. Some physicists harbor the hope that both sets of
problems will be resolved by the sought-after quantum theory of gravity. It
is difficult to assess this hope since we can now only dimly perceive the shape
that a successful marriage of QM and GTR will take. And even if the hope
is eventually realized, it is important to pursue these foundational problems
for the light they may shed on the correct path towards the marriage.
For many purposes, the measurement problem in QM can be ignored by
3
4 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

experimentalists and theoreticians alike. Not surprisingly, it was ignored by

a large segment of the physics community, and the opinion was once
widespread that this problem is merely a Scheinproblem, I vividly recall the
occasion of a lecture on the measurement problem given in the early 1970s
at The Rockefeller University by a Nobel laureate in physics. The reaction
of the audience, composed largely of theoretical physicists and mathematicians,
was distinctly cool if not unfriendly. The skepticism was directed not so much
at the proposed solution as to the notion that there was a problem to be
solved. After the lecture, the laureate remarked ruefully: "I suppose that I
will have to do something new to restore my reputation." Today his lecture
would likely get a different reception, at least judging from the fact that The
Physical Review, the most prestigious journal in theoretical physics, now
routinely publishes articles on this topic. The implied change in attitude
reflects a recognition that the measurement problem poses a fundamental
challenge for QM, although how to state the challenge is controversial. On
the one hand, the problem can be seen as revealing that there is something
rotten at the core of the theory because of its inability to give a satisfactory
description of what occurs in the interaction between an object system and a
measurement apparatus. Those who share this diagnosis see a need for a new
dynamics for QM, involving a non-linear, non-unitary, stochastic evolution
to replace the linear, unitary, and deterministic evolution implied by the
Schrodinger equation. On the other hand, the challenge can be seen not as
calling forth new physics but rather a new understanding of quantum
ontology and the ways in which the Hilbert space apparatus of the theory
relates to that ontology. Reactions to this way of reading the challenge range
from new interpretation rules for assigning values to quantum observables,
even when the state vector of the system is not an eigenstate of the operator
corresponding to the observable, to a world picture not unlike Borges' garden
of forking paths (many worlds interpretation), to a model of reality that
replaces many worlds with many minds.1
My own conviction is that neither clever semantical rules nor extravagant
metaphysics will suffice and that eventually new physical principles will have
to be recognized. But this is not the place to argue for that opinion.
The history of the problem of spacetime singularities in GTR has followed
a rather different trajectory. As will be seen in the next section, the recognition
of the problem came not long after Einstein completed the theory in 1915—16.
Over the next four decades the discussion of the problem was somewhat
desultory, but no more so than the discussion of relativistic gravitation in
general, which came to be regarded as a backwater of physics research. The
renaissance of GTR in the 1960s made singularities a focus of attention. The
new mathematical techniques developed to analyze solutions of the Einstein
gravitational field equations permitted the proof of theorems which showed
unequivocally that spacetime singularities could not be ignored but are
generic features of general relativistic spacetimes.2 But this is getting ahead
of the story.
 Introducing Spacetime Singularities and Acausalities 5

1.2 Spacetime singularities: In the beginning

In this section and the following section I will provide a brief sketch of the
early struggles with the problem of spacetime singularities, concentrating
mainly on Einstein's evolving attitude towards the problem. The selection of
topics is guided by the unabashedly Whiggish aim of motivating later
chapters, where modern attempts to understand the nature and implications
of spacetime singularities in GTR are discussed. A careful history of the subject
will have to await another occasion,3 but nevertheless, I hope that the present
account will convey some sense of just how difficult and at times baffling the
subject was for the pioneers of GTR. This last remark is not a piece of
Whiggish condescension since, as the subsequent chapters will demonstrate,
the subject continues to challenge and baffle.
The problem of spacetime singularities in GTR arose even before the
theory was put in its final form. As of 18 November 1915, Einstein had not
arrived at the final form of his gravitational field equations, although the
equations he presented in his communication of that date (Einstein 1915) to
the Berlin Academy did reduce to his final equations, hereafter referred to as
the Einstein field equations (EFE), in the case of the exterior field of a mass
distribution.4 Einstein produced an approximate solution to these equations
for a static spherically symmetric field. Rewritten in spherical coordinates,
the line element of this approximate solution is

where a = 2MG/c2, M being the mass of the source, G the gravitational

constant, and c the velocity of light.5 (From here on I will work in units where
c — 1. A frequent choice of units sets a = 2M. Thus, the " Schwarzschild
radius" is often designated as r = 2M.) One naturally wonders about the
meaning of the non-regular points r = 0 and r = a where the line element
(1.1) ceases to make sense. In November of 1915 Einstein did not have time
to ponder such questions, for he was fully occupied in using (1.1) to resolve
the long-standing anomaly of the perihelion motion of Mercury.6 The
following year Karl Schwarzschild (1916) produced the exact solution of
which (1.1) can be regarded as a first-order approximation. In the coordinates
later introduced by Droste (1917), the Schwarzschild line element is

Because (1.2) is an exact solution, the question of singularities became more

pointed, and was to become still more pointed when Birkhoff (1923) showed
that the Schwarzschild solution is the only spherically symmetric solution to
the vacuum EFE (with vanishing cosmological constant).
The question was broached by David Hilbert (1917) in Part II of his
seminal article "Grundlagen der Physik." Hilbert supplied a general definition
D BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

of what it means for the spacetime metric to be non-singular or, to use

Hilbert's own terminology, "regular," and he applied his definition to (1.2).
For a 7^ 0, it turns out that r = 0 and (for positive a) also r = a are points
at which the line element is not regular. By that I mean that a line element
or a gravitational field g^v is regular at a point if it is possible to introduce
by a reversible, one-one transformation a coordinate system, such that in
this coordinate system the corresponding functions j^v are regular at that
point, i.e., they are continuous and arbitrarily differentiable at the point and
in a neighborhood of the point, and the determinant g' is different from 0.
(Hilbert 1917, p. 70-71)7
Given Hilbert's definition, his conclusion that the Schwarzschild metric
is singular at both r = 0 and r = a is perfectly correct. But by modern lights
the definition is defective in failing to capture the distinction between genuine
singularities—as in the case of r = 0—and mere coordinate singularities—as
in the case ofr — a. The Droste coordinates in (1.2) give the false appearance
of a singularity at r = a because of the peculiar way in which they fail to
cover the spacetime manifold. The failure can be made transparent by
suppressing the angular coordinates 6 and </> and by transforming from the
r, t coordinates to new coordinates X, T in which the metric components are
well behaved everywhere r > 0:

The r-t part of the Droste-Schwarzschild line element is replaced by

which is evidently well behaved at r = a. The Schwarzschild radius r = a

corresponds to the null lines X = + T. At these locations t = + oo, indicating
the failure of the t coordinate. Hilbert's demand that regular metric com-
ponents g'^ be produced from g^v by means of a transformation that is smooth
and invertible at r = a is thus inappropriate. 8 This little piece of hindsight
wisdom did not find its final fruition until forty-three years later when Kruskal
(1960) and Szekeres (1960) constructed the maximal analytic extension of the
Schwarzschild metric.
Einstein's initial worry about the Schwarzschild solution was not so much
with its apparently singular nature as with its anti-Machian character.
Rightly or wrongly, Einstein interpreted Mach's principle as implying that
the metrical structure of spacetime should be determined by its matter
content, and prima facie the Schwarzschild solution is incompatible with this
reading of the principle. As John Stachel has written, it was for Einstein a
 Introducing Spacetime Singularities and Acausalities 7

"scandal that a solution to his field equations should exist which corresponded
to the presence of a single body in an otherwise 'empty' universe" (Stachel
1979, p. 440). The scandal erupted anew the following year with the de Sitter
solution to EFE. In attempting to quash this second form of the scandal,
Einstein was forced to confront the issue of spacetime singularities.
In the coordinates de Sitter (1917a,b) used, the line element of his
spacetime takes the form

where R is a positive constant. The metric in (1.5) can beconsidered to be a

solution to the vacuum EFE with positive cosmological constant A = 3/R2.
In a postcard dated 24 March 1917, Einstein complained to de Sitter that
"your solution corresponds to no physical reality."9 A lengthy correspondence
ensued,10 and after several changes of position Einstein published his
complaint in the Proceedings of the Berlin Academy (Einstein 1918). The
published objection was based on the idea that the de Sitter metric is singular.
We can see in Einstein's objection some of the key elements that figure in the
modern definition of that term. For the Einstein of 1918, a spacetime is
non-singular if "in the finite realm" the covariant and contravariant
components of the metric are continuous and differentiable and, consequently,
the determinate g —det(gab) never vanishes. A point p is said to lie in the
finite realm just in case it can be joined to an arbitrarily chosen origin point
p0 of spacetime by a curve whose proper length J*Jo ds is finite.11 For the de
Sitter solution (1.5), g vanishes at r = 0 and i/f = 0. According to Einstein,
however, this represents only an apparent violation of continuity since g ^ 0
can be restored at these points by a change of coordinates. But the
discontinuity at r — nR/2 was another matter. "It seems," Einstein wrote,
"that this discontinuity cannot be removed by any choice of coordinates"
(Einstein 1918, p. 271). Thus, it seemed to Einstein that "until proof to the
contrary" one should conclude that a singularity occurs at these locations
since they lie at finite distances—starting at r = t = 0 and tracing to r — nR/2
along a curve with constant \[i, 9, t, the integral jo*'2 ds is finite.
Although Einstein's ideas here are true to the spirit of the definitions of
spacetime singularities that would be developed half a century later, the
attempted implementation of the ideas is seen in the harsh glare of hindsight
wisdom to be doubly defective.
In the first place, the appearance of a singularity at r = nR/2 is only an
illusion, for the de Sitter coordinates cover only a portion of a larger spacetime
that, by any reasonable standards, is singularity free. An understanding of
this point emerged from the papers of Felix Klein (1918a, b) and Cornelius
Lanczos (1922b). 12 A detailed discussion of de Sitter spacetime was also given
in Arthur Stanley Eddington's influential book The Mathematical Theory of
8 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Relativity (1923). Although Eddington clarified the nature of the de Sitter

"singularity," the way he presented the matter revealed an imperfect
understanding of the Schwarzschild solution.
Eddington rewrote the line element of the de Sitter metric in the form 13

He noted that in (1.6) there is a "singularity" at r — R "similar to the

singularity at r = 2Af[a] in the solution for a particle of matter," a reference
to the Droste-Schwarzschild line element (1.2). Comparison of (1.6) with
(1.2) raised the question of whether the de Sitter spacetime is really empty
of matter.

Must we not suppose that the former singularity [in (1.6)3 also indicates
matter—a "mass-horizon" or ring of peripheral matter necessary in order to
distend the empty region within. If so, it would seem that de Sitter's world
cannot exist without large quantities of matter . . . he has merely swept the
dust away into unobserved corners. (Eddington 1923, p. 165)

This is undoubtedly a reference to Einstein's critique of the de Sitter solution,

which had ended with the assertion that in the de Sitter world matter is
"entirely concentrated in the [singular] surface r = nR/2" (Einstein 1918, p.
272). However, Eddington's answer to his rhetorical question was an
emphatic no. "A singularity in ds2 does not necessarily indicate material
particles, for we can introduce or remove such singularities by making
transformations of coordinates" (ibid., p. 165). He continued:

The whole of de Sitter's world can be reached by a process of continuation;

that is to say the finite experiences of an observer A extends only over a
certain lune; he must then hand over the description to B whose experience
is partly overlapping and partly new; and so on by overlapping lunes. . . .
[In this way] we arrive at de Sitter's complete world without encountering
any barrier or mass-horizon, (ibid., p. 166)

Here then we have a codification of the first small success in understanding

the difference between genuine singularities and coordinate singularities.
Knowledge of the nature of the r — a Schwarzschild singularity was
harder won. Lanczos (1922a) hinted at the pseudo-nature of this singularity.
Two years later in a letter to Nature, Eddington (1924) employed the
transformation

which recast (1.2) in the form

 Introducing Spacetime Singularities and Acausalities 9

(Actually, there is a missing factor of 2 in Eddington's formula which,

in the units he was using, contained the term 2Mln(r — M) rather than
2Mln(r - 2M).) The blowup of grr = (1 - tx/r)'1 in (1.2) at r = a. has been
replaced by tame behavior of g'rr = (1 + 00") in (1.8). In addition, in the new
coordinates the determinant of the metric potentials is non-vanishing at r = a
and, therefore, the contravariant metric components have finite values. It is
on this basis that Eddington is sometimes credited with showing that the
Schwarzschild metric is non-singular at r = a (see Misner, Thome, and
Wheeler 1973, Box 31.1). However, in Eddington's coordinates £44 = £44,
which vanishes at r = a. Hindsight wisdom is needed to see that the vanishing
of £44 and £44 is an indication not of a breakdown in the metric but of the
fact that d/dt' and d/dt turn from timelike to null at the Schwarzschild radius.
In any case, the purpose of Eddington's communication was to compare
Einstein's and Whitehead's theories of gravity, and Eddington himself made
no claim to have clarified the status of the Schwarzschild radius.14
The first explicit and self-conscious demonstration that the r = <x
Schwarzschild singularity was merely a coordinate singularity was due to
Georges Lemaitre (1932). Here was a golden opportunity to illustrate within
the same solution the difference between genuine and coordinate singularities
and to provide a rough criterion for telling the two apart. The Kretschmann
curvature invariant K = RabcdRal":d and other curvature scalars as well blow
up as r = 0 is approached (K -~ 1/V6), whereas they remain well behaved as
r = a is approached. As we will see in chapter 2, trying to separate coordinate
singularities from genuine singularities on this basis is too crude, since genuine
singularities are not always signaled by the ill behavior of curvature invariants.
But even such an imperfect criterion would have had a salutary effect on the
discussions of these matters.
In the actual case, however, even this rudimentary understanding
continued to elude some of the most able researchers. An especially glaring
example is Einstein's 1939 treatment of the Schwarzschild radius. Einstein
chose to work in isotropic coordinates in which the Schwarzschild line element
(1.2) takes the form

where n = 2a. Einstein noted that at r = /i/2 a clock would "go at the rate
zero" and that light rays and material particles starting from r > fi/2 would
"take an infinitely long time (as measured in 'coordinate time') in order to
reach the point r = n/1." He concluded that "In this sense the sphere r = [1/2
constitutes a place where the field is singular" (Einstein 1939, p. 922).
The aim of Einstein's article was to show that the "'Schwarzschild
singularities' do not exist in physical reality" (ibid., p. 936). The basis of this
conclusion was the claim that a gravitational field with such singularities
10 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

cannot arise as the exterior field of a realistic system of masses. Other research
workers had put forward a weaker version of this claim by noting that the
Schwarzschild radius for any known astronomical body or elementary particle
is well within the body itself where, of course, the exterior Einstein field
equations do not apply. This "pragmatic attitude," as Eisenstaedt (1989) has
called it, towards the Schwarzschild singularity was maintained at least as
late as 1960 (see, for example, Synge 1960, p. 183). Towards the end of
demonstrating a stronger, in-principle version of this claim, Einstein con-
sidered a spherical cluster of particles, each of which moves in a circular orbit.
He found that, in a condition of static equilibrium, as the radius of the cluster
approaches the Schwarzschild radius, the particles would eventually have to
move faster than light. In hindsight, what Einstein's calculations showed was
not that a massive body cannot collapse inside its Schwarzschild radius but
that if it contracts sufficiently near to this radius it cannot remain in static
equilibrium. And we now know that the further contraction which the body
cannot resist may very well eventuate in a genuine curvature singularity (see
Thorne 1994, Ch. 3).
It is obvious that in 1939 Einstein was working in ignorance of Lemaitre's
1932 article, which was first published in an obscure Belgian journal and then
republished in a less obscure but not widely read Belgian journal. Nevertheless,
it is surprising that in 1939 Einstein failed to address the question of whether
the r = jU/2 "singularity" could be removed by a coordinate transformation
and simply concluded that "the field is singular" there. Could it be that
Einstein was applying his 1918 definition of singularity and making the same
mistake that Hilbert made in 1917? This is unlikely since in his critique of
de Sitter's solution Einstein qualified his conclusion that a singularity exists
at r = nR/2 with an "until further notice" clause, an unnecessary concession
if Hilbert's restriction on the nature of the coordinate transformation needed
to remove the singularity had been in effect. A much more likely explanation
of how Einstein misled himself is contained in the above quoted passage about
the need for an infinite amount of coordinate time for a particle to reach
r = n/% from r > [A/2. The irony, of course, is that the fact Einstein cites is
an indication of the breakdown of the t coordinate rather than a proof of the
physical inaccessibility of r = /z/2. A second irony derives from the observation
that in the original dispute which gave rise to Einstein's attempt to define
the notion of spacetime singularity, de Sitter defended his metric by appealing
to the physical inaccessibility of the r = nR/2 singularity. Here we witness a
potentially confusing entanglement of the problem of singularities and the
problem of horizons in cosmology. I will return to the latter topic in the next
section.
To return to Einstein's attempt to characterize spacetime singularities,
the second facet of his definition, which sought to capture the notion of "at
a finite distance," is also defective. In the de Sitter spacetime the supposed
singular region can be reached by spacelike curves of finite proper length.
But in other cases—for example, the standard big bang models—the singularity
 Introducing Spacetime Singularities and Acausalities 11

can only be reached by timelike or lightlike curves. The proper length of a

lightlike curve is, of course, 0, and if arbitrary tirnelike curves are permitted,
then what would be regarded as future or past infinity can be reached by a
timelike curve of finite proper length. Current attempts to capture the notion
of "at a finite distance" still use Einstein's idea of a curve which, in some
appropriate sense, has finite length (see chapter 2). But there is a subtle
difference. Modern constructions use a half-curve, a curve which starts at
some point p0 (Einstein's origin point) and which is extended as far as possible
in some direction from that point. For such a half-curve, there is no endpoint
corresponding to Einstein's p—the singularity does not have a location in
spacetime. Although this may seem to be nitpicking, it is of crucial importance
for the interpretation of singularities and, as will be seen in the following
section, the misunderstanding of this point had an adverse affect on Einstein's
attempt to deal with the problem of motion of a particle in a gravitational
field.
Leaving aside the technical niceties of Einstein's attempt to define
spacetime singularities, it is noteworthy that most of the subsequent attempts
have shared Einstein's philosophy on one crucial point; namely, if something
nasty happens to the spacetime metric, but the nastiness happens only "at
infinity," then no singularity in the spacetime is indicated. However, as Roger
Penrose's discussion of "naked singularities" and "cosmic censorship" has
revealed, nastiness at infinity can still be disruptive to physics, and if it is
disruptive enough it deserves the name singularity (see chapters 2 and 3).
Since the reader who is new to these issues may be reeling from trying
to follow the twists and turns of the discussion, it may be well to step back
from the complexities in order to gain more perspective on Einstein's attitude
towards singularities.

1.3 Einstein's intolerance of singularities

The visceral dislike of singularities which Einstein displayed in 1917—18

remained with him until the end of his life. Here is Peter Bergmann, who
served as a research assistant to Einstein at the Institute for Advanced Study,
speaking of Einstein's attitude towards singularities:
It seems that Einstein always was of the opinion that singularities in classical
field theory are intolerable. They are intolerable from the point of view of
classical field theory because a singular region represents a breakdown of the
postulated laws of nature. I think one can turn this argument around and
say that a theory that involves singularities and involves them unavoidably,
moreover, carries within itself the seeds of its own destruction. . . . (Bergmann
1980, p. 186)

In the first years following the formulation of the GTR, Einstein's attitude
towards singularities may have wavered, but Einstein's published papers and
12 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

the correspondence of which I am aware indicate that Bergmann has captured

Einstein's mature attitude. This conclusion raises three questions.
First, how is Einstein's rejection of singularities to be reconciled with his
willingness, in connection with the problem of motion, to treat massive
particles as singularities in the gravitational field? Second, how did Einstein
propose to resolve the tension between his intolerance of singularities and the
existence of singularities in solutions to his gravitational field equations?
Third, what exactly was the basis of his rejection of singularities? And apart
from understanding Einstein's attitude, this book is dedicated to understanding
whether his horror of singularities is justified.
To take first questions first, Einstein wanted to show that the postulate
that the world line of a massive neutral test particle is a timelike geodesic
need not be taken as a separate axiom of GTR but is in fact a consequence
of the gravitational field equations, and several publications stretching over
twenty years are devoted to this goal (see Einstein and Grommer 1927;
Einstein 1927; Einstein, Infeld, and Hoffman 1938; and Einstein and Infeld
1949). Towards this end he was willing to treat the particle as a singularity
in the gravitational field. His attitude towards this procedure is summarized
in a letter to Ludwig Silberstein dated 18 February 1941:
Of course, in a complete field theory the positing of singularities is altogether
forbidden. In the present case the introduction of singularities is justified
because it allows the treatment on the basis of the gravitational field alone
of a problem of which "matter" is a part, without having to use a theory of
the latter. (EA 21-090)15

It is irresistible to see a strong analogy between Einstein's procedure here and

his path to the special theory of relativity (STR). H.. A. Lorentz's version of
the special principle of relativity was based on what Einstein called a
"constructive theory" (Einstein 1954, p. 228)—in this instance, an elaborate
theory of the electromagnetic constitution of matter which made predictions
about how matter behaves when it moves through the ether. By contrast,
Einstein's approach employed a "principle-theory" which required no
commitments about the constitution of matter and only minimal commitments
about the behavior of light and electromagnetic waves. In parallel, Einstein's
strategy on the problem of motion in a gravitational field was to avoid
questions about the nature of matter by treating a test particle as a pole
singularity in a solution of the matter-free field equations Gab = 0 (or
equivalently, Rab = 0) for A = O.16
There are several perplexing features of Einstein's treatment of the
problem of motion, but the most perplexing of all lies in the fact that
singularities in the spacetime metric cannot be regarded as taking place at
points of the spacetime manifold M.17 Thus, to speak of singularities in gab
as geodesies of the spacetime is to speak in oxymorons. One might hope to
give some sense to this talk by representing singularities as ideal points
adjoined to M. But there are several different ways to accomplish the
 Introducing Spacetime Singularities and Acausalities 13

representation, and they yield inequivalent characterizations of singularities.

Moreover, all of the extant procedures yield unexpected and counterintuitive
results, e.g., the singular points may not be Hausdorff separated from the
regular points of M (see chapter 2).
Treating spacetime singularities as lying in the spacetime is not only
mathematically unjustified but is apparently also inconsistent with Einstein's
own attitude on the ontological status of spacetime.18 During 1913-14 when
he was searching for the relativistic gravitation field equations Einstein used
the notorious "hole argument" to justify rejecting the requirement of general
covariance (see Norton 1987). After reaffirming general covariance he
proposed to escape the hole argument by maintaining that

If we imagine the gravitational field, i.e., the functions gik, to be removed,

there does not remain a space of type (1) [Minkowski spacetime], but
absolutely nothing, and also no "topological space." For the functions gik
describe not only the field, but at the same time also the topological and
metrical properties of the manifold. . . . There is no such thing as empty
space, i.e., a space without field. Space-time does not claim an existence of
its own, but only as a structural quality of the field. (Einstein 1961, p. 155)

Einstein scholars have not provided an explanation of how Einstein was able
to reconcile the apparent conflict between the view quoted and his treatment
of the geodesic postulate in terms of singular regions of spacetime.
There might still seem to be room for compromise here. Start with a
non-singular background spacetime M, gab. Following Einstein, let us avoid
making any assumptions about the constitution of matter by treating a
massive particle as a point object with world line z" = £°(T) parameterized
by proper time T. The associated energy—momentum tensor can be postulated
to be Tab(x) = m J i"ibd(x - S(T)) dx where i" = dz'/dx and m is the mass of
the particle. The conservation law Va Tab = 0 can then be used to show that
£"(T) is a geodesic of M, gab,19 But such a demonstration does not constitute
a derivation of the geodesic postulate from the field equations Gab = 8ttTab.
These equations do entail the conservation law, but for the singular Tab
postulated here the field equations are not meaningful in even a distributional
sense. The widest class of spacetime metrics for which the Riemann tensor
makes sense as a distribution is arguably the regular metrics as defined by
Geroch and Traschen (1987) (see chapter 2); but this class is not wide enough
to encompass sources concentrated on one-dimensional submanifolds.
There are alternative approaches to deriving the geodesic postulate that
do not involve the problematics of singularities and are relatively neutral
about the constitution of matter. For example, Fock (1939) showed that under
mild assumptions about a non-singular Tab, the conservation law Va Tab = 0
entails the geodesic hypothesis for sufficiently small bodies whose self-gravity
is weak. 20 Einstein evinced no interest in such approaches that made any
assumptions about the nature of matter. His final official word on the subject
is found in a joint paper with Infeld:
14 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

All attempts to represent matter by an energy-momentum tensor are

unsatisfactory and we wish to free our theory from any particular choice of
such a tensor. Therefore we shall deal here only with gravitational equations
in empty space, and matter will be represented by singularities of the
gravitational field. (Einstein and Infeld 1949, p. 209)
I will return below to Einstein's struggles with the problem of motion and
singularities, but for the moment I turn to the tension between the view that
"in a complete theory the positing of singularities is altogether forbidden"
and the existence of singularities in solutions to EFE. There are at least three
ways to resolve the tension, which will be discussed in turn.
The first resolution is to argue that singularities are artifacts of the
idealizations of the models in which they occur. This is the tack Einstein took
in 1931 in reaction to the big bang singularity of the Friedmann model:
Here one can try to get out of the difficulty by pointing out that the
inhomogeneity of the stellar matter makes illusory our approximate treat-
ment. (Einstein 1931, p. 237)
Eight years later Einstein took a similar line with respect to what he regarded
as the singular aspect of the Schwarzschild solution. As we saw above in
section 1.2 he argued that it is not possible to build up a field containing such
"singularities" (as at r = fj,/2 in (1.9)) from physically realistic arrangements
of gravitating matter (Einstein 1939). The work of Oppenheimer and Volkof
(1939) and Oppenheimer and Snyder (1939) indicated that the gravitational
collapse of a star could produce a singularity. There is evidence that as late
as 1942 Einstein was either not aware of this work or else brushed it aside.21
In the latter case his justification would no doubt have been that the
singularity predicted was an artifact of the assumed perfect spherical symmetry.
In the end, however, this first resolution will not suffice. Einstein himself
produced evidence that singularities are not isolated features of solutions to
his gravitational field equations. In 1941 he argued that there are no solutions
to the vacuum field equations which are associated with a positive mass, which
are static and asymptotically Minkowskian, and which are singularity free
(Einstein 1941). Related results had already appeared in the literature (see
Lichnerowicz 1939). In 1943 Einstein and Pauli generalized Einstein's 1941
result to cover gravitational theories of the five-dimensional Kaluza type
(Einstein and Pauli 1943). However, the fact that GTR and related theories
seemed incapable of producing particle-like solutions without singularities 22
did not endear Einstein to singularities but served rather to make him
dissatisfied with these theories. Had Einstein lived until the end of the 1960s
he would no doubt have agreed that the theorems of Hawking and Penrose
show unequivocally that spacetime singularities in GTR are not artifacts of
specialized models. But these theorems would most likely have served to
increase his dissatisfaction with GTR.
The second resolution is to appeal to the cosmological constant term in
the EFE. A is not a completely effective holy cross against the vampire of
 Introducing Spacetime Singularities and Acausalities 15

singularities, but it can help. For example, a large enough positive value for
A will ward off the big bang singularities in the Friedmann models. This
would not have been an attractive alternative to the Einstein who reportedly
said that the introduction of A was the "biggest blunder he ever made in his
life" (see Gamov 1970, p. 44). However, Einstein's attitude towards A was
more complicated than this quotation indicates. In an appendix "On the
'Cosmologic Problem'" added to the second edition of The Meaning of
Relativity, Einstein opined that the "cosmologic member" (i.e., the A.gab term
in the gravitational field equations) "is to be rejected from the point of view
of logical economy" (Einstein 1955, fn pp. 120-121). His reasoning is
explained in a footnote:
If Hubble's expansion had been discovered at the time of the creation of the
general theory of relativity, the cosmologic member would never have been
introduced. It seems now so much less justified to introduce such a member
into the field equations, since its introduction loses its sole original justifica-
tion—that of leading to a natural solution of the cosmologic problem.
(Einstein 1955, fn p. 121)

The "cosmologic problem" to which Einstein refers is to reconcile a finite

average matter density with the gravitational field equations, a problem
Friedmann resolved by allowing for the expansion of the universe. The quoted
passage is consistent with the attitude that other cosmological problems could
justify the reintroduction of a A term. One such problem was already
mentioned in Einstein's appendix: the Friedmann model in conjunction with
the then existing astronomical measurements of the Hubble constant led to
an age for the universe that was too short.23 But Einstein showed no
inclination to use A to solve this problem. Nor is there any mention of A as
a solution to the problem of singularities.
The third resolution is to say that the field equations (without cosmological
constant term) break down under the conditions that singularities are
predicted to occur. In the same appendix "On the 'Cosmologic Problem,'"
Einstein expressed his doubts about the consequence of the spatially closed
Friedmann models that "for the time of the beginning of the expansion the
metric becomes singular and the density . . . becomes infinite."
For large densities of field and matter, the field equations and even the field
variables which enter into them have no real significance. One may not
therefore assume the validity of the equations for very high density of field
and matter, and one may not conclude that the "beginning of expansion"
must mean a singularity in the mathematical sense. (Einstein 1955, p. 123)

Out of context this quotation seems to represent an uncharacteristic loss of

nerve on Einstein's part: if we refuse to trust the most fundamental equations
of physics to work under extreme conditions, then we will be barred from
making many interesting predictions. And if the EFE do break down under
extreme conditions, what will take their place? Here Einstein would seem to
16 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

face a dilemma. The EFE can be replaced either by some other classical field
equations or by some system of quantum mechanical equations. In the former
case one can wonder why the new equations can be trusted anymore than
the EFE to work under extreme conditions. In the latter case one can imagine
that under extreme densities quantum effects emerge and somehow prevent
the formation of singularities. But Einstein would not have seen this escape
from singularities as a live option, for he never gave up the hope of deriving
quantum constraints from classical field theory.
Placed in proper context the quotation does not signal any loss of nerve,
for Einstein had long been dissatisfied that his field equation had a Tab term
on the right-hand side.24 And Einstein's work on unified field theory can be
seen as a seizing of the first horn of the apparent dilemma. His program of
unification focused on the gravitational and electromagnetic fields, with the
primary goal of treating both as aspects of the spacetime geometry. One
consequence of the unification he expected was that matter could be treated
as a derived rather than a basic concept in that the unified field equations,
with no matter source term, would admit singularity free, particle-like
solutions. And finally, quantum constraints were also to be derived, perhaps
by conditions overdetermining the classical field variables.25 With this as
background, we can understand the following optimistic passage from The
Meaning of Relativity:

The present relativistic theory of gravitation is based on a separation of the

concepts of "gravitational field" and of "matter". It may be plausible that
the theory is for this reason inadequate for very high density of matter. It may
well be the case that for a unified theory there would arise no singularity.
(Einstein 1955, fn p. 124)26
Had Einstein become as pessimistic as most of his contemporaries about
the prospects of a unified field theory, he would have been forced to conclude
that the third option for dealing with singularities, as well as the first two,
was closed off. This closing off would have necessitated a reassessment of his
rejection of spacetime singularities.27 Such a reassessment was suggested by
the man who spent the early part of his career trying to establish the
experimental basis of Einstein's GTR, Erwin Freundlich, now named Finlay-
Freundlich. In 1951 Finlay-Freundlich's monograph on Cosmology appeared
as Vol. 1, No. 8 of the Foundations of the Unity of Science series. In a section
entitled "General Remarks concerning Singularities in the Cosmological
Problem," he commented on the big bang singularity in the Friedmann
models. He noted that an initial singularity can be avoided by introducing a
large positive value for the cosmological constant, but termed the introduction
"a serious step" for which there "appears otherwise no necessity." He
continued:
We must ask therefore, since singularities arise when we get excessively large
values of the [mass] density . . . , whether our fear of excessive densities, the
opposite of horror vacui, is justified and not perhaps the last reminder of a
 Introducing Spacetime Singularities and Acausalities 17

subconscious yearning for a harmonious universe. (Finlay-Freundlich 1951,

p. 49).

There is no indication that Einstein had read Finlay-Freundlich's

Cosmology or paid attention to his suggestion any more than that he paid
much heed to the opinions of any of his contemporaries on these matters.28
Only from Einstein's own writings can we hope to discern the reasons for his
aversion to spacetime singularities. Early on (circa 1918) Einstein seems to
have thought that spacetime singularities involve singularities in the matter
sources and that the latter singularities are absurd. But it later became evident
that spacetime singularities need not hide mass concentrations, singular or
otherwise.
A good starting point for exploring the mature Einstein's aversion to
spacetime singularities is a joint paper by Einstein and Nathan Rosen
published in The Physical Review in 1935. They wrote:
A singularity brings so much arbitrariness into the theory . . . that it
actually nullifies its laws. . . . Every field theory, in our opinion, must
therefore adhere to the fundamental principle that singularities of the field
are to be excluded. (Einstein and Rosen 1935, p. 73)

Taken by itself this quotation is potentially misleading. What Einstein and

Rosen were objecting to was not so much spacetime singularities in general
as the idea that material particles can be treated as singularities of a
gravitational field, an idea that, as noted above, was explored by Einstein
himself from time to time. In the paper in question this idea was rejected by
Einstein and Rosen for the reason that such a singularity "nullifies" the laws
of the theory. They explained:
A pretty confirmation of this was imparted in a letter to one of the authors
by L. Silberstein. As is well known, Levi-Civita and Weyl have given a
general method for finding axially symmetric static solutions of the gravita-
tional equations. By this method one can readily obtain a solution which,
except for two point singularities lying on the axis of symmetry, is everywhere
regular and is Euclidean at infinity. Hence if one admitted singularities as
representing particles one would have here a case of two particles not
accelerated by their gravitational action, which would certainly be excluded
physically, (ibid., p. 73)
To understand the content of this passage requires a digression.
What Silberstein had communicated to Einstein was a rediscovery of a
version of the Curzon (1924a, b) bipolar solution, which belongs to the Weyl
class (1917, 1919) of static axially symmetric solutions. If the above quotation
accurately describes the solution and if singularities in the solution are taken
to represent massive particles, then as Silberstein went on to claim, not every
solution to the vacuum EFE Gab = 0 (or Rab = 0) would be physically
admissible. Einstein and Rosen seemed to agree, for the main point of their
paper was to contemplate a modification of the EFE which would admit of
18 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

non-singular solutions for the static spherically symmetric case. A particle

would then be represented not by a singularity but by the multiple connected-
ness of the space in the form of a "bridge" joining two identical sheets of space.
Encouraged by the remarks in the Einstein-Rosen paper, Silberstein went
on to publish his calculations in the February 1, 1936 issue of The Physical
Review.29 He concluded:
Thus our solution . . . corresponds to a complete absence of matter . . .
everywhere, except only at the points A, B themselves, the "mass centers."
There are thus . . . no stresses between A and B to keep them apart like a stiff
rod. And yet (the solution being stationary, invariable in time) the two points
are fixed relatively to each other instead of falling towards each other, which
flagrantly contradicts man's most ancient, primitive experience.
Here, then, is an example of a perfectly rigorous solution to Einstein's
field equations which does not correspond to reality. If, therefore, these
equations are to be retained, one cannot consider "material particles" (mass
points) as singularities of the field. (Silberstein 1936, p. 270)

Although Einstein had agreed to the gist of this conclusion, its appearance
in print prompted him to examine Silberstein's calculations more closely.
Together with Rosen he published a letter in The Physical Review rebutting
Silberstein's claims.
In a recent paper, Silberstein attempts to show the incorrectness of the
general theory of relativity. His reasoning is as follows:
(a) I set up a static solution of the gravitational field equations which
has two singular points and which is everywhere else free of
singularities.
(b) The two particles so represented are not accelerated in each other's
gravitational field, in contradiction with experience. Hence the
gravitational field equations of the general relativity theory are
incorrect.
We should like to point out the following. Even if (a) were the case the
conclusion (b) would not be justified. For in a field theory only a representa-
tion of masses which is free from singularities can be accepted, since at a
singularity the laws of the field are violated. However the assertion (a) is
not correct. We shall show that the solution given by Silberstein has
singularities outside the two points. (Einstein and Rosen 1936, p. 404)
Foisting (b) on Silberstein was a little disingenuous. Silberstein was not
claiming to have shown that EFE are incorrect but only that, assuming (a),
either the field equations are incorrect or else "one cannot consider 'material
particles' (mass points) as [pole] singularities in the field," a claim Einstein
would not have disputed. The real bone of contention was (a). Silberstein
assumed, and Einstein accepted, that the singularities in the Curzon-Silber-
stein solution are simple pole singularities. The first intimation that this
assumption was incorrect did not come until thirty years later when Gautreau
and Anderson (1967) showed that the singularities seem to have an angular
character—whether the Krctschmann curvature scalar blows up depends on
 Introducing Spacetime Singularities and Acausalities 19

the direction from which the singular 'points' are approached. Stachel (1968)
clarified the matter by showing that the singular 'points' are not pointlike at
all but have a non-trivial topological structure. The local and bizarrely
complicated global structure of the singularity in the Curzon monopolar
solution has only recently received its definitive elucidation by Scott and
Szekeres (1986a,b).
The main disagreement between Einstein and Silberstein was whether,
as Silberstein claimed, the axis joining the two singular 'points' is non-
singular. For the Curzon bipolar solution Silberstein's claim is incorrect. But
seeing how the claim is incorrect helps to reveal just how subtle is the business
of singularities. In Weyl's coordinates the line element takes the form

where U and k may be functions of p and z- (Here the £-axis is the axis of
symmetry, p is the radial coordinate, and (/) is the angular coordinate.) In
order that the axis joining the two singular 'points' be non-singular, it is
necessary that limp _ 0 k = 0, otherwise the ratio of the circumference to the
radius of an infinitesimal circle enclosing the axis would not be In. Einstein
and Rosen (1936) noted that this condition fails for the Curzon—Silberstein
bipolar metric.30 And yet there is also a sense in which the axis is non-singular.
If the components of the curvature tensor are computed in a suitable frame
along a geodesic approaching the axis, they remain well behaved—there is
no blowup or wild oscillation. What we have here is a non-curvature
singularity or in modern parlance a quasi-regular singularity (see chapter 2).
The intuitions shared by Einstein and Silberstein in discussing the implications
of spacetime singularities simply do not begin to do justice to the surprising
complexities that spacetime singularities can display.
It is evident from the controversy with Silberstein that Einstein's attitude
towards singularities was shaped in part by the problem of motion. But
Einstein's published response to Silberstein reveals a reason why he thought
that, independently of the problem of motion, singularities must be excluded:
"in a field theory only a representation of masses which is free of singularities
can be accepted, since at a singularity the laws of the field are violated."
Einstein is surely right that, whatever the technical details of a definition of
spacetime singularities, it should follow that physical laws, in so far as they
presuppose space and time, are violated or, perhaps more accurately, do not
make sense at singularities. This is a good reason for holding (as was urged
above) that singularities are not part of spacetime. But why should singularities
be seen as spelling such a disaster for physics that any spacetime theory
involving singular solutions must be deemed inadequate? Consider, for
example, the Friedmann—Robertson—Walker (FRW) models of the big bang.
There is a rigorous sense in which these models exhibit Laplaciari determinism,
by which the physical state of the universe at any time uniquely fixes the
state at any other time. Within these models the application of determinism
20 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

permits us to retrodict the existence of a big bang singularity and, in spatially

closed models with sufficient mass, to predict a big crunch singularity. Thus,
the application of the EFE implies that time is finite in the past, and, if there
is sufficient mass, also in the future; but within this finite stretch physics is
conducted as usual. There may be various psychological objections to the
finiteness of time, but it is far from obvious why physical laws that lead to this
conception thereby impugn themselves. Perhaps it will be said that these laws
are defective because they entail the existence of a big bang and a big crunch
without telling us what happened before the bang or after the crunch. But
this objection misses the point. As Einstein said, physical laws break down at
spacetime singularities, and for the big bang and the big crunch they break
down so strongly that it is physically meaningless to talk about the 'before'
and the 'after' (see chapters 2 and 7).
It remains to contemplate the possibility that singularities might occur
not at the beginning or end of time but, so to speak, in the interior of spacetime
where they could disrupt physics. An attempt to distinguish between disruptive
and non-disruptive singularities might be seen to be behind de Sitter's
response to Einstein's (1918) critique of the de Sitter solution. De Sitter
conceded—unnecessarily as we now know—that his solution violated Einstein's
stricture that in a physically admissible spacetime there are no discontinuities
at finite distances. He continued:

This postulate, however, in the form in which it is enounced by Einstein, is

a philosophical, or metaphysical, postulate. To make it a physical one, the
words "all points at finite distances" must be replaced by "all physically
accessible points". And if the postulate is thus formulated, my solution . . .
does fulfill it. For the discontinuity arises for r = rt = nR/2.
This is at a finite distance in space, but it is physically inaccessible. . . .
The time needed for a light ray, and a fortiori by a moving material point,
to travel from any point r, i/r, </> to a point rlt \l/l, (/>! (i/'j and </)1 being
arbitrary) is infinite. [And by symmetry, the time needed to travel in the
opposite direction is also infinite.] The singularity at ^ can thus never affect
any physical experiment . . . . (de Sitter 1918, p. 1309)

There are several things here that need to be disentangled. The class of
non-disruptive singularities in the sense of singularities that do not interfere
with determinism and predictability is not coextensive with the class of
singularities that are physically inaccessible—as already noted, the big bang
singularity is certainly accessible but it is not disruptive. The disruptive vs.
non-disruptive distinction is today pursued under the labels of naked vs.
non-naked singularities (see chapter 3). De Sitter's notion of physically
inaccessible singularities does call to mind one way in which a singularity can
be non-naked; namely, by being hidden inside a black hole. But the most
obvious application of the notion of physical inaccessibility to de Sitter
spacetime calls attention not to black holes but to two other features of general
relativistic cosmological models—particle horizons and event horizons (see
 Introducing Spacetime Singularities and Acausalities 21

chapter 6). Because de Sitter spacetime possesses both features there are
regions that have been and forever will be causally disconnected from one
another. The existence of particle horizons in the standard big bang models
makes it difficult for these models to satisfactorily explain the isotropy of the
cosmic microwave radiation (or so it has been claimed), and this "horizon
problem" has spawned inflationary cosmology and a number of other
competing models (see chapter 5). Once again we have an example where
the pioneers of GTR hit upon the problems that are now at the heart of
gravitational research but did not possess the analytical tools to properly
investigate the problems. Nor was there any call to develop those tools until
it became clear that singularities could not be ignored.

1.4 Acausality and time travel

By contrast with spacetime singularities, the recognition of the fact that

solutions to EFE can involve acausal spacetime structures did not come until
comparatively late. 31 Van Stockum (1937) found an exact solution to EFE
for a source consisting of an infinite rotating cylinder of dust; but it was not
realized until decades later that the exterior spacetime contains closed timelike
curves (see Tipler 1974). The first solution known to have this feature was
published by Godel (1949a). And while singularities have been a focal point
of research in relativistic gravitational theory for the past thirty years, it is
only quite recently that closed timelike curves and the like have received
much attention in the physics literature (see chapter 6).
Godel (1949b) contributed a brief account of his discovery to the Schilpp
(1949) volume, Albert Einstein: Philosopher-Scientist, in which he attempted
to use his technical results to support the idealistic conclusion that time is not
real (see the Appendix to chapter 6). In his reply to criticisms, Einstein
(1949b) brushed aside the relation of relativity theory to idealism in order to
concentrate on what he took to be the fundamental problem raised by Godel's
solution. He invited the reader to consider three points A, P, B lying on a
timelike world line, with P between A and B. Does it make sense, he asked,
to assign an arrow to the world line, indicating that, say, B is before P and
A after P? A positive answer is warranted, Einstein asserted, if it is possible
to send a signal from B through P to A but not vice versa. Einstein continued:
If, therefore, B and A are two, sufficiently neighboring, world-points, which
can be connected by a time-like line, then the assertion: "B is before A,"
makes physical sense. But does this assertion make sense, if the points, which
are connectable by the time-like line, are arbitrarily far separated from each
other? Certainly not, if there exist point-series connectable by time-like lines
in such a way that each point precedes temporally the preceding one, and if
the series is dosed in itselj. In that case the distinction "earlier—later" is
abandoned for world points which lie far apart in a cosmological sense, and
22 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

those paradoxes, regarding the direction of causal connection, arise, of which

Mr. Godel has spoken. (Einstein 1949b, p. 688)

I read Einstein as providing, if somewhat imperfectly, a distinction

between two features of time: time directionality and time order. The spacetime
of Godel's solution admits of a globally consistent assignment of time
directionality or distinction between past and future at each event, but
it does not admit of a consistent assignment of time order to events. The point
is not quite captured, as the quotation from Einstein would suggest, by saying
that the distinction 'earlier-later' makes sense for nearby events but has to
be abandoned for events that are far apart. But such niceties aside, what are
the paradoxes, "regarding the direction of causal connection," that arise from
Godel's solution? Again there is a slight mischaracterization of the problem
in the quotation: there is no ambiguity in the direction of causation in the
Godel model—causal propagation is assumed to always take place in the
future direction which, as already noted, is a globally well-defined notion in
this model. The problem is rather that the kinds of causal stories we are used
to telling threaten to degenerate into gibberish in the Godel universe. For
Godel's solution seems to allow for the physical possibility of a journey into
one's own past. Once one is there, why couldn't one then proceed to undo
what has already been done? Of course, it is nonsensical to suggest that one
can change what has already happened, but what is to prevent the deeds
that would lead to the absurdity?
Einstein's concern with such paradoxes was expressed in his closing
remarks.

Such cosmological solutions [that allow time travel] of the gravitational

equations (with non-vanishing A-constant) have been found by Mr. Godel.
It will be interesting to weigh whether these are not to be excluded on
physical grounds. (Einstein 1949b, p. 688)

In the first sentence of the quotation lies an implicit criticism of the Godel
solutions: they require a non-vanishing cosmological constant. Given Einstein's
rejection of this term in his field equations, there was reason for hope that
acausal solutions could be ignored on the grounds that they were confined
to the A ^ 0 cases. Today we know that this hope is not realized. Thus we
must face the question raised in the second sentence of the quotation: Are
such solutions to be "excluded on physical grounds"?

1.5 Singularities and acausalities together

To return to the analogy offered in section 1.1, spacetime singularities and

acausal spacetime structures pose challenges for GTR not unlike the challenge
the measurement problem poses for QM. Are the nastier sorts of singularities
and acausalities to be excluded on physical grounds? If so, can the exclusion
 Introducing Space time Singularities and Acausalities 23

be enforced on some non-ad hoc basis within GTR or some natural extension
of the theory, or must the theory be modified in some substantial way in order
to achieve the exclusion? If on the other hand we elect to try to find ways to
peacefully coexist with singularities and closed time loops, what price does
the coexistence exact of our concepts of physical laws, causality, and free will?
These are certainly questions worthy of study. But why study the
problems of singularities and acausality together in one work? The short
answer, which can only be justified by a lengthy analysis, is that they are not
separable problems. The theorems of Hawking and Penrose that prove the
existence of singularities in wide classes of solutions to EFE assume the absence
of closed time loops. This naturally raises the question of whether acausal
features of spacetime can prevent the occurrence of singularities. In the other
direction, attempts to prove that there are obstacles to operating a time
machine which would create closed time loops assume that spacetime
singularities are not formed through the operation of the machine. This raises
questions about what kinds of singularities might be tolerated by a time
machine operator. And third, Penrose's cosmic censorship hypothesis, which
seeks to show that the GTR does not permit the formation of naked
singularities under physically reasonable conditions, would count some forms
of acausality as violating cosmic censorship.
The first order of business is to try to say more precisely what we are
talking about. Chapter 2 discusses various analyses of the concept of spacetime
singularity and the means of classifying the types of singularities that can
occur in relativistic spacetimes. Because of the technical nature of the material,
some readers will find chapter 2 heavy going. If you are one of these, gentle
reader, do not lose heart, for the subsequent chapters are designed to stand
largely on their own. Thus, you may "want to skim chapter 2 and then turn
to the topics that interest you the most, while being prepared to occasionally
return to chapter 2 to clarify some point.
The topics of the subsequent chapters were chosen partly to illuminate
the issue of singularities in GTR and partly to emphasize the connection of
these issues to philosophical concerns about predictability and determinism,
paradoxes of infinity, the common cause principle and the nature of scientific
explanation, backwards causation and time travel, etc. Once the connections
are revealed, it is hard to see how philosophers of science can continue to
write on these topics without taking into account the implications of GTR.
Chapter 3 reviews various formulations of the cosmic censorship hy-
pothesis and weighs the evidence that has been amassed for and against
censorship. It also tackles the difficult question of how much a disaster for
physics it would be if cosmic censorship should fail. One startling consequence
of the failure is discussed in chapter 4. If singularities are naked enough, it
may be possible to perform the functional equivalent of a supertask in which
one observer, by making use of the labor of another observer, can gain within
a finite time a knowledge of the outcomes of an infinite number of operations.
This would, for example, seemingly undercut the moral of Church's proof of
24 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

the undecidability of arithmetic since it would now be possible to obtain

knowledge of the results of mechanically checking all potential proof sequences
for a given formula.
Chapter 5 deals with a problem pertaining not so much to the big bang
singularity itself as to the way in which the singularity is set in standard
cosmological models. The so-called horizon problem turns out to be as much
a problem about conditions for the explanatory adequacy of a scientific theory
as it is a problem of physics. Claims of the advocates of inflationary cosmology
to have solved the horizon problem are critically examined and, I hope,
somewhat deflated.
Chapter 6 examines recent work in physics on time travel and time
machines. It is argued that the philosophical literature on these topics, and
especially that part of the literature on the "grandfather paradox" and similar
paradoxes, has failed to diagnose what is genuinely problematic about closed
timelike curves in relativistic spacetimes. An attempt is made to reorient the
discussion in what I regard as a more fruitful direction.
Chapter 7 deals with a cluster of issues revolving around the notions of
eternal recurrence and circular time, notions that have exercised a peculiar
fascination on physicists and philosophers alike. It turns out to be more
difficult than might have been guessed to pin down what these notions mean
in terms of relativistic spacetime structure. Some of the ways in which eternal
recurrence have been imagined to be possible are shown to be physically
meaningless. Others are meaningful but are unrealizable according to GTR
because of the occurrence of singularities. And the supposed dichotomy
between open and closed times is shown to be not a dichotomy at all.
The closing chapter, chapter 8, does not attempt a "what does it all
mean" summary; indeed, any such summary would be an insult to an
enormously complex and fascinating subject. But it does address the antipathy
to spacetime singularities, an antipathy that started with Einstein and is still
found in some segments of the physics community.
Although I will from time to time have remarks about the implications
of quantum gravity, the focus of this book is squarely on classical GTR. This
is a severe limitation, but not an unnatural one if contentious speculation is
to be avoided. And the problems of singularities and acausalities as they occur
in classical GTR seem to me sufficiently interesting in their own right as to
justify treating them on their own terms. But readers will have to judge for
themselves.

Notes
1. An excellent survey of various approaches to the measurement problem is to
be found in Albert (1992).
2. Or so it would seem; but see chapter 2.
3. A comprehensive and historically accurate account of Einstein's struggles with
spacelime singularities, much less of singularities in general, has yet to be written.
 Introducing Spacetime Singularities and Acausalities 25

Havas (1989, 1993) has dealt with Einstein's attitude as it related to the problem of
motion. Eisenstaedt (1989, 1993) has provided an illuminating discussion of the
evolving understanding of the Schwarzschild singularity. A good overview of work on
singularities (up to 1980) is to be found in Tipler, Clarke, and Ellis (1980); but as
the authors of this seminal article would surely agree, for historians of science the
article is an invitation to do a careful history of this complex subject.
4. The final EFE can be written in two equivalent forms:

Here gab is the spacetime metric, Tab is the energy-momentum tensor, T =• Trace (Tab),
Rab is the Ricci tensor associated with gab, R =-Ra" is the curvature scalar, and A is the
cosmological constant. From (i) one infers that when A = 0 the vacuum fields equations
(Tab = 0) take the form Rab = 0. From (ii) one infers that they also take the
form Gab ='Rab — (\l2)Rgab = 0, where Gab is the Einstein tensor. By consistency, one
infers that R = 0 for the vacuum case. Einstein (1915) started from the equations
Rab = const, x Tab .
5. Here and below I have taken the liberty of changing sign conventions to
conform to a signature ( + + -i— ) for the spacetime metric.
6. See Earman and Janssen (1993) for an account of Einstein's derivation of the
perihelion shift for Mercury.
7. The translation is courtesy of John Norton.
8. I am indebted to Jean Eisenstaedt for pointing out to me how Hilbert's
definition goes astray; see Eisenstaedt (1993). There are some subtleties here that can
only be brought out with the help of the concept of an extension of one spacetime by
another; see sections 2.3 and 2.5 of chapter 2.
9. The postcard is in the archives of the Sterrewach at Leiden. The English
translation is from Kahn and Kahn (1975).
10. An account of this correspondence is given in Kerszberg (1989).
1 1 . Here the sign of ds2 is fiddled, if necessary, so that ds > 0.
12. John Stachel (1993) has noted that in addition to providing a non-singular
version of the de Sitter metric, Lanczos also showed that sense could be made of the
"mass-horizon" at r = nR/2 and in this way vindicated Einstein's claim that the de
Sitter solution was not anti-Machian (see below).
13. As the reader will no doubt have gathered, in order to save complications in
the notation, the same symbol r is being used in different line elements, although the
physical interpretation is different in the different cases.
14. Finkelstein (1958) independently discovered the same coordinate transforma-
tion. The coordinates are thus referred to as the Eddington- Finkelstein coordinates.
15. English translation from Havas (1993). EA xx-yyy stands for Control Index
item # xx-yyy in the Einstein Archive.
16. Einstein's attitude towards the cosmological constant is discussed below.
17. This point is emphasized by Torretti (1983, section 5.8).
18. I am indebted to John Norton for this point.
19. Here Va stands for the covariant derivative operator associated with the metric
gab- VaGab = 0 is a geometrical identity. Thus, Va Tab = 0 is a consequence of the EFE.
20. For a modern derivation of the geodesic postulate see Geroch and Jang
(1975).
26 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

21. Tipler et al. (1980) note that Einstein said that he had read the text of Peter
Bergmann's book Introduction to the Theory of Relativity (1942) and that in this
book there is no mention of the problem of singularities, save for a reference to
Einstein's (1939) way of dealing with the Schwarzschild r = 1M "singularity."
22. It was not until nearly four decades after Einstein's death that non-singular
particle-like solutions were discovered for the non-vacuum EFE. It has been shown
recently that the Einstein-Yang-Mills equations admit solutions that are static,
asymptotically Minkowskian, singularity free, and possessing positive ADM mass; see
Smoller, Wasserman, Yau, and McLeod (1991) and Smoller and Wasserman (1993).
Such solutions would not have satisfied Einstein since he hoped to be able to represent
particles without the help of a non-zero Tab (see below).
23. Recently this problem has arisen again, and some cosmologists want to solve
it by reintroducing A; see Crosswell (1993).
24. As John Stachel (private communication) has emphasized to me.
25. See Pais (1982) for an account of Einstein's work on unified field theories.
Pais (1982) and Stachel (1986) contain remarks on Einstein's attempt to derive
quantum constraints from classical field theory.
26. The same sentiment is expressed again a few pages later in the part of the
paragraph immediately preceding the previous quotation; see Einstein (1955, p. 123).
27. Or is it too far fetched to suggest that Einstein might have reconsidered his
opposition to QM if he had thought that the quantum might hold the key to getting
rid of spacetime singularities?
28. Havas (1989) documents how Einstein ignored much of the work done by
others on the problem of motion in the gravitational field. Einstein preferred to work
things out for himself from what he regarded as first principles.
29. I am skipping over many of the details of the Einstein—Silberstein dispute.
For the full story the reader will have to consult Havas (1993).
30. It is sometimes said that the singularity on the axis can be interpreted as
"rods or struts" that hold the particles apart; see Kramer et al. (1980, section 17.1).
As far as I know, this notion has never been made precise. The kind of singularity
here is similar to the cone singularity discussed in section 2.2 of chapter 2.
31. John Stachel (private communication) has noted that prior to the final
formulation of GTR Einstein raised the possibility of closed timelike curves in
relativistic spacetimes but pronounced the possibility repugnant to his physical
intuition: "Dies widerstrebt meinem physikalischen Gefiihl aufs lebhaftteste" (Einstein
1914, p. 1079).
 2
Defining, Characterizing, and
Proving the Existence of Spacetime
Singularities

2.1 Introduction

We saw in chapter 1 that shortly after GTR was put in its final form, attempts
were made by Hilbert (1917) and Einstein (1918) to specify the conditions
under which a spacetime is singular. Some small amount of progress was
made on the problem over the next four decades. But as late as 1960 Gyorgy
Szekeres could accurately state that "an exact definition of what should be
regarded as a true singularity of a [pseudo] Riemannian manifold has, to my
knowledge, never been proposed" (Szekeres 1960, p. 285).l Over the
following years rapid strides were made in proving the existence of singularities
in solutions to EFE and in understanding what these singularities involved
in terms of spacetime structure. But towards the end of the 1960s there were
still major uncertainties in how to pin down the elusive notion of spacetime
singularity. Some of these uncertainties were summarized in an important
paper by Robert Geroch (1968a), part of which was cast as a Galilean
dialogue, a form nicely chosen to reflect the unsettled state of the subject. In
the intervening years our knowledge about these matters has increased
tremendously, but uncertainties still remain.2 The account which follows does
not attempt to trace the historical development over recent decades but rather
is aimed at providing an accessible introduction to these issues, issues which
remain among the most exciting and difficult in the foundations of spacetime
theories.
It will emerge that there are at least four distinct though interrelated
concepts of spacetime singularities. The first can be seen as the natural heir
of Einstein's (1918) idea that a singular spacetime is one in which the metric
somehow breaks down at a finite distance. The second develops from an
attempt to make precise the notion of "at a finite distance"; it promotes
geodesic incompleteness or some generalization thereof as the essential
characteristic of singularities. The third, which appeals to the notion of
"missing points," is extensionally equivalent to the first two in many cases
27
28 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

but is nevertheless conceptually distinct. The fourth, which evolved from

Roger Penrose's cosmic censorship hypothesis, departs sharply from the path
set by Einstein's definition since it recognizes the presence of "naked
singularities" even when the singularities fail to lie "at a finite distance." It
will also emerge that there is a tension between the noun and adjective
conceptions of spacetime singularities. The former attempts to conceive of
singularities as entities that can be localized while the latter eschews
localization and is content to speak of singular spacetimes when these
spacetimes exhibit certain large-scale or global features.

2.2 What is a spacetime singularity?

In classical and special relativistic physics, the definition, if not the physical
interpretation, of a singularity is straightforward. So for instance, a singularity
in the electromagnetic field exists at a point p in Minkowski spacetime if the
electromagnetic energy density "blows up" at p (i.e., increases beyond all
bounds as one approaches p along any path). When we shift to the context
of GTR two related differences make themselves felt. First, there is no longer
a fixed spacetime background which by general agreement is non-singular
and against which the blowup of a physical quantity can be measured. And
secondly, we are now interested not in singularities in some physical field on
the spacetime but rather in singularities in the spacetime itself. Nevertheless,
we can try to carry over to GTR the idea that a singularity involves the
blowup of a physical quantity.

First try
A singular point in the spacetime is a point where, say, the curvature of the
spacetime blows up. Strictly speaking, this idea is incoherent since if the
spacetime metric is ill-behaved at a point, then that point is not part of the
spacetime. This attitude is codified in the usual definition of a general
relativistic spacetime as a pair M, gab where M is a connected differentiate
manifold (without boundary) and gab is a Lorentz metric which is defined
and Ck (k > 0) at every p e M3
Already at this early stage we have run into one of the perplexities that
will eventually force some fundamental choices in the treatment of singular-
ities. If we want to continue to speak of spacetime singularities in terms of
singular points, then we have to recognize that the kind of existence these
entities have (I will use 3xistence to refer to it) is a ghostly one, for it does
not conform to the wholly sensible slogan "To exist is to exist in space and
time." Perhaps this Bxistence can be made less ghostly by replacing it by
regular existence in an augmented spacetime which contains singular points
as ideal boundary points attached to the spacetime manifold. On the other
hand, a failure to find a sensible procedure for the replacement would support
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 29

the position that noun constructions about singularities should be dropped in

favor of adjectival constructions—talk of singularities as entities should be
replaced by the less ontologically loaded talk of singular spacetimes.
Whichever way one comes down on this issue of singularities as "missing
points," it seems sensible to try to develop a criterion for when a spacetime
is to be counted as singular. One can try to accomplish this with a more
sophisticated version of the blowup idea,

Second try

If the spacetime M, gab contains a suitable curve y c M along which the

curvature becomes unboundedly large, then the singular nature of the
spacetime is indicated. One difficulty here is that spacetime curvature is
characterized by the Riemann tensor Rabcd. It will not do to take the singular
nature of the spacetime to be indicated by the blowup of some component of
the curvature tensor in some coordinate system, for that ill behavior may be
the fault of the coordinate system and not of the Riemann tensor. We can
avoid this problem by looking at curvature scalars such as R ( = Raa), RabRab,
RabcdRabcd, and other scalar invariants formed by taking outer products of
Rabcd and its derivatives and contracting indices with the help of gab.4 The
blowup of such scalars is surely a good indication of the singular nature of
the spacetime (at least if suitable restrictions are placed on the curve j along
which the blowup occurs, as will be discussed below). The trouble is that
although we have a sufficient criterion, we do not have a necessary one. Wald
(1984a, p. 214) has noted that in some vacuum solutions to EFE describing
gravitational plane waves rippling through spacetime, all such curvature
scalars can vanish even though the curvature tensor itself is singular.5

Third try
It is too soon to give up on our basic idea. On our curve y, choose an
orthonormal tetrad field or frame6 e", i = 1, 2, 3, 4, and use this frame to
define the physical components R^j^km —Rabcdei'ej'ekeid of the curvature tensor.
These physical components may be badly behaved if, for example, the
frame is allowed to spin madly around. So let us require that the tetrad field
be parallely propagated (p.p.) along y.7 If the physical components of the
curvature tensor in such a p.p. frame blow up, we can say that the singular
nature of the spacetime has been revealed.8
The more precise our trial criterion of singularity becomes, the more
worries that arise. First, if the physical components of Rabcd in any p.p. frame
along y blow up but only as y goes off to spatial or temporal infinity, then
the singular nature of the spacetime has not been demonstrated in the sense
intended by Einstein since singular behavior does not lie "at a finite distance"
(see chapter 1). Sussmann (1988) has produced a family of spherically
symmetric, shear free, perfect fluid solutions to EFE, some of whose members
30 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

contain a kind of "asymptotically delayed big bang": curvature scalars

diverge along timelike and null geodesies that approach this delayed big bang,
but a geodesic of infinite affine length is needed to reach the bang.9 Clearly
then some appropriate restriction on the length of y is needed if curvature
divergence along 7 is to indicate a singularity in the sense intended by Einstein.
Second, what if the physical components of Rabcd do not blow up but do fail
to approach a limit? Should we recognize oscillatory as well as blowup
behavior as an indicator of singularities? Third, the necessity of the trial
criterion is also challenged by the following example of a cone singularity
(Ellis and Schmidt 1977, p. 921; Wald 1984a, p. 214). In a polar cylindrical
coordinate system, the Minkowski line element for R 4 is

Choose an angle (j)0 < 2n, remove the wedge consisting of the points such
that 0 < (j> < (f)0, and then identify the points (p, 0, z, t) with the correspond-
ing points (p, <p0, z, t). All the points on the resulting manifold, save for those
on p = 0, inherit a smooth metric from the Minkowski metric on R4.
If we exclude the points p = 0 we arrive at a spacetime with a flat (Rat,c^ = 0)
metric. But despite the fact that there is no ill behavior in the curvature, this
spacetime can plausibly be classified as singular. Consider an observer who
is born and then spends his life cruising along without acceleration. If his
world line is aimed at (the missing) p = 0, then even if he drinks of the
fountain of youth he will experience only a finite lifetime (as measured by
proper time along his world line). Certainly the ghost of such an observer
whose time has run out will feel justified in complaining that the spacetime
is pathological.

Fourth try

Combining the first and third points above suggests that we set aside for the
moment the idea that spacetime singularities involve curvature blowup and
explore instead the alternative idea that spacetime singularities are associated
with incomplete curves, curves that cannot be extended to an arbitrarily large
value of some suitable parameter. Implementing the latter idea requires some
care. A half-curve is a curve which has one endpoint and which is inextendible
in the direction away from the endpoint.10 A half-curve is said to be complete
with respect to some parameter if the values of the parameter are unbounded.
It will not do to count a spacetime as singular if it contains a timelike
half-curve which is incomplete with respect to proper time, for if no restriction
is placed on the acceleration of such a curve, then even Minkowski spacetime
will be counted as singular. 1J The simplest course to take at this juncture is
to focus on unaccelerated curves or geodesies. A geodesic half-curve is said to
be incomplete just in case it has finite affine length. (For a timelike or spacelike
geodesic this is equivalent to saying that its proper length is finite.)
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 31

It is easy to create artificial examples of spacetimes that are geodesically

incomplete. For example, start with Minkowski spacetime R4, rjab, remove a
compact ball K, and restrict qab to R 4 — K.12 The resulting spacetime contains
incomplete geodesies of all three kinds: timelike, spacelike, and null. In this
case there is a straightforward sense in which the incompleteness is due to the
Existence of missing points. But by the same token, the singular nature of the
surgically mutilated spacetime is non-intrinsic since it can be repaired by
extending the spacetime back to the full Minkowski spacetime. Here it is
appropriate to digress on the notion of extensions of spacetimes since the ideas
involved are crucial to much of what follows.

2.3 Extensions of spacetimes

M, gai, is said to be an extension of M, gab just in case there is an isometric

imbedding of the latter into the former, i.e., there is a diffeomorphism
(p: M -» M such that g^l^M) = <P*g-13 The extension is proper just in case
<p(M) is a proper subset of M. M,gab is properly extendible just in case there
is a proper extension of it. To illustrate some of the subtleties of these
definitions, it will be useful to consider an example adapted from Wald
(1984a, p. 149). Consider the submanifold M c R 2 covered by the coordinates
x, t such that — oo < x < +00 and 0 < t < +00. Define a metric gab on M
whose line element is ds^ = dx2 — dt2/t*. This metric is extendible to all of
R 2 by means of the imbedding (p that takes a point with coordinates (x, t) to
the point with coordinates (x, l / t ) . Introduce new coordinates x', t' such that
x' = x and t' = l/t, i.e., the new coordinates of the new point are numerically
the same as the old coordinates of the old point. So ((p*g)X'X-((p(x, t)) = 1 and
(<P*g)t'A<P(x, 0) = - J A' 4 - Thus, ds2.g = dx'2 - dt'2/t'4 = dx2 - dt2. This is
just the (two-dimensional) Minkowski metric, so of course there is agab defined
on all of R 2 such that gab\<i>(M) ~ <P*g- But now consider a second imbedding
of M into R 2 which is just the natural inclusion map. Then relative to this
imbedding there is no C° metric gab defined on all of R 2 such that gab\M — gab
since gtt = —l/t* blows up as t —» 0 + .
The example just given is relatively uninteresting from the point of view
of singularities since the metric in question would not be considered singular
(for the candidate singularity at t = 0 does not lie at a finite distance, e.g.,
all the geodesies that approach it are of infinite affine length). But an
analogous moral holds for the Schwarzschild metric. Let M c S2 x R 2 be the
open submanifold covered by the Droste coordinates for r > a (see section
1.2). The points at r = a do lie at a finite distance (e.g., they can be reached
by spacelike geodesies of finite length) and are candidate singularities since
the grr component of the Schwarzschild metric blows up as r approaches a.
And in fact under the imbedding that is given by the natural inclusion map
for points labeled by the Droste coordinates, there is no C° extension of the
metric to and beyond the Schwarzschild radius. (This is one way of expressing
32 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

what was right about Hilbert's claim as discussed in section 1.2) However,
there is another imbedding due to Kruskal (1960) which does implement a
C°° extension of the Schwarzschild metric to all of S2 x (R2. (This expresses
what is wrong about Hilbert's claim.) The details will be saved for section 2.5.
By contrast to the Schwarzschild radius (r = a), there is no C2 extension
of the Schwarzschild metric to and through r = 0 since the Kretschmann
curvature scalar RttbcdRabcd (~ l/ f6 ) blows up rapidly as r = 0 is approached.
But do not be misled by this example. Constructing extensions can be
prevented not only by ill behavior of the curvature but also by topological
obstructions. Consider again the case of the cone singularity introduced at
the end of section 2.2. The flat metric cannot be smoothly continued to p = 0.
For in the plane z. = constant, t — constant, a circle of radius p has circum-
ference (2n — (f>0) p < 2np for any p > 0, which is impossible if the origin
p = 0 is part of a Euclidean space (Ellis and Schmidt 1977, p. 921). 14
A spacetime is said to be maximal (with respect to some stipulated
continuity/differentiability condition on the metric, e.g., C2) just in case there
is no proper extension (satisfying the stipulated condition). In all of the above
examples of non-maximal spacetimes there is a maximal extension. Does this
feature hold in general? The answer might seem to be obviously positive, e.g.,
every Ck spacetime can be extended to a maximal Ck spacetime. However,
the "obvious" proof of this fact via a Zorn's lemma construction does not
work. 15 What can be shown directly from Zorn's lemma is something weaker.
An (w-dimensional) framed spacetime M,gab, F is a spacetime equipped with
a distinguished orthonormal n-a.d F of vectors at some point of M. Call
M, gab, Fan extension of M, gah, Fjust in case there is an isometric imbedding
(p of M into M such that <p*F = F. Then by a Zorn's lemma construction,
every framed spacetime can be extended to a (not necessarily unique)
maximal framed spacetime (Geroch 1970a). The desired result now follows
easily. Let M, gab be an arbitrary Ck spacetime. Frame it in some way to form
M,gab, F. Obtain a maximal framed extension M,gab, F. Then M,gab is a
(not necessarily unique) maximal extension of M, gab.
Metaphysical considerations suggest that to be a serious candidate for
describing actuality, a spacetime should be maximal. For example, for the
Creative Force to actualize a proper subpart of a larger spacetime would
seem to be a violation of Leibniz's principles of sufficient reason and plenitude.
If one adopts the image of spacetime as being generated or built up as time
passes then the dynamical version of the principle of sufficient reason would
ask why the Creative Force would stop building if it is possible to continue.
However, this image does not sit well with the four-dimensional way of
thinking, and in any case it runs into trouble in its own terms: since extensions
of spacetime are generally non-unique there may be many ways to continue
building and the Creative Force may be stymied by a Buridan's ass choice
(see Clarke 1993, pp. 8-9). Some readers may be shocked by the introduction
of metaphysical considerations in the hardest of the "hard sciences." But in
fact leading workers in relativistic gravitation, though they don't invoke the
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 33

name of Leibniz, are motivated by such principles (see, for example, Geroch
1970a, p. 262; Penrose 1969, p. 253). The intrusion of metaphysics will also
be evidenced in all of the chapters to follow.

2.4 The received definition of singularities

Any non-maximal spacetime is geodesically incomplete. But such a spacetime

is arguably not an acceptable model for actual spacetime. So suppose that
we focus on maximal spacetimes. If there are any singularities in such a
spacetime they are essential or irremovable. The task now is to find a criterion
that will signal when a spacetime is singular. Will geodesic incompleteness
suffice? An appeal to the analogy of the better understood case of a
Riemannian space would seem to encourage a positive answer.
A Riemannian space M, hab (where hab is a positive definite Riemannian
metric as opposed to a pseudo-Riemannian metric of relativistic spacetimes)16
can be made into a metric space by a standard construction. Choose an
arbitrary pair of points p, q e M and consider all piecewise differentiable
curves y(v) joining p and q, where the curve parameter v is arbitrary. The
length of y(v) between/; and q is jp9 (h(dy/dv, dy/dv))1*2 dv. Define the distance
d(p, q) between p and q to be the greatest lower bound on the length of such
curves. The following properties of this distance function can be proved: for
any p, q, r e M
(i) d ( p , q ) = d(q,p)
(ii) d(p, q} = 0 if and only if p = q

(iii) d(p, q} < d(p, r) + d(r, q}

The existence of a distance function with these three properties is what is
meant by saying that M, d( , ) is a metric space. Further, it can be shown
that the manifold topology is compatible with the metric topology in that
(iv) the metric balls B(p, e) =: {q E M: d(p, q} < e} for all p e M and all
8 > 0 form a basis for the topology of M.
For a metric space we can define the notion of a Cauchy sequence of
points pi, i = 1, 2, 3, . . . , by the condition that for any e > 0 there is an JV
such that for any m, n > M, d ( p m , p n ) < e. A metric space is said to be Cauchy
complete just in case every Cauchy sequence converges to some point in the
space. For metric spaces then, Cauchy incompleteness is the signal of the
3xistence of missing points. Furthermore, this peculiar sense of Bxistence can
be replaced by ordinary existence in a larger space. For it is a theorem that
any Cauchy incomplete metric space can be isometrically imbedded as an
open dense subset of a complete metric space. Roughly, the new points of the
larger space ( = original space plus the missing points of the incomplete space)
are constructed from equivalence classes of non-convergent Cauchy sequences,
34 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

where two Gauchy sequences^; and qi are taken to be equivalent just in case
d(Pt> ?i) -* 0 as i -> oo. (In just this way the real numbers can be constructed
by taking the Cauchy completion of the rationals.)
Returning now to the case of a relativistic spacetime M, gab with a
pseudo-Riemannian metric, we are unable to avail ourselves of the metric
space characterizations of completeness and incompleteness. For in M, gab any
pair of points p, q € M can be joined by a broken null curve, with the result
that with d(p, q) defined as before, the distance between the points is always
zero. However, in the Riemannian case the Hopf-Rinow theorem shows that
Cauchy completeness is equivalent to geodesic completeness. Thus, if the
analogy between Riemann spaces and relativistic spacetimes holds, it would
seem reasonable to use geodesic completeness as a criterion of completeness
in relativistic spacetimes.
The first problem we face in implementing the criterion is that in
relativistic spacetimes there are three different kinds of geodesies—timelike,
spacelike, and null—and it has been shown by explicit examples that the
various kinds of geodesic incompleteness are inequivalent. Indeed, there is
complete inequivalence in that one can have each form of incompleteness in
the presence of the other two forms of completeness, as well as any two forms
of incompleteness in the presence of the third form of completeness. To give
a feeling for how the various forms of incompleteness can come apart it will
be useful to sketch Geroch's (1968a) example of a spacetime that is timelike
geodesically incomplete while being spacelike and null geodesically complete.
Start with two-dimensional Minkowski spacetime IR2, tjab and generate the
conformally related spacetime IR2, Q2J?afc where Q2: IR2 -> (0, +00) is a C00
map with the following properties. For some fixed inertial coordinate system
(M):
(1) &(x, f) = 1 i f * < -1 o r * > 1
(2) Q.2(x, t)=Q2(~x, t) for all x, t
(3) fi2(0, i!) goes rapidly to 0 as / -» oo
The third property implies that the timelike half-geodesic that starts at (0, 0)
and then traces the t axis forward or backward in time is incomplete. On the
other hand, any inextendible null or spacelike geodesic that starts in or enters
the strip — 1 < x < 1 must eventually escape, and thus by property (1) is
complete.
Because we can imagine ourselves travelling along a timelike geodesic,
timelike geodesic incompleteness is the most psychologically disturbing of the
three forms of incompleteness. Nevertheless there are spacetimes which we
would want to count as singular but which are timelike geodesically complete.
Reissner-Nordstrom spacetime, which describes the exterior field of a spheric-
ally symmetric, electrically charged body, is one example (see Hawking and
Ellis 1973, pp. 156-161). Intuitively, the singularity, which can be reached
by a null or a spacelike geodesic, exerts a repulsive force that causes free-falling
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 35

massive particles to miss hitting it. Though less psychologically disturbing

than either timelike or null incompleteness, spacelike geodesic incompleteness
cannot be ignored either. So to be on the safe side, let us try on for size this
tentative definition: a spacetime is non-singular just in case it is geodesically
complete in all three senses.
An example due to Geroch (1968a) brings into question the sufficiency
of this definition. Consider a timelike half curve (not necessarily geodesic)
y(t) parameterized by its proper time r. At each point on the curve
construct the normalized tangent vector V(T:) (gatV"(T;)Vb(x) = — 1 ) . The
four-acceleration is then defined as Aa(i) =•• F ft (T)Vj, V(r). If y(t) is a geodesic,
then A"(T) = 0 for every T. But instead of considering only timelike geodesies,
let us consider the more general class of curves of bounded acceleration, i.e.,
there is a positive number B such that the magnitude of acceleration a(i) < B
for all T, where a(t] —• ( A b ( T ) A b ( x ) ) l / 2 . If in addition such a curve has finite
proper length, then one could in principle construct a rocket ship whose world
line would instantiate this curve using only a finite amount of fuel, and the
cosmonaut at the controls would biologically age only a finite number of
years. But by hypothesis, the curve cannot be extended, so the cosmonaut
does not enjoy the possibility of prolonging his life. The unfortunate fellow
would surely feel justified in regarding the spacetime he inhabits as being
singular. Now the point of Geroch's example is that such behavior can occur
even if the spacetime is geodesically complete. It seems prudent then to
strengthen the proposed definition of a non-singular spacetime to require not
only geodesic completeness but also bounded acceleration completeness, that
is, that every timelike half-curve of bounded acceleration has infinite proper
length.
But even this modification of our tentative definition may not suffice, for
there is a still more general sense of completeness that must be taken into
account. Consider a half-curve y: [0, v + ) -> M, v+ < +00. Choose an ortho-
normal basis «,•"(()) for the tangent space at y(0) and parallel transport this
basis along y(v) to give the frame field e " ( v ) , v € [0, »+ ). The tangent vector
V = (3/3»)v(o) to the curve can then be written as F" = £f =1 X'(v)«("(»).
And the generalized affine parameter (g.a.p.) of y(v) is defined by
A =: jo (Zf=i (xi(v)}2}112 dv- The choice of a different frame field e'ttt(v) will
lead to a different g.a.p. A' on y(v); but it can be shown that y(v) will have
finite A' length if and only if it has finite 1 length. Thus, the notion of finite
(or infinite) generalized affine length (g.a.l.) for a half-curve is a well-defined
notion.
It is easily seen that if y is a geodesic, then the g.a.p.s on y will be affine
parameters; but, of course, the g.a.p. is defined for all C1 curves, not just for
geodesies. For an arbitrary timelike curve (geodesic or not), if the curve has
infinite proper length, then it has infinite g.a.l. But a timelike curve of finite
proper length may not have finite g.a.l. if, for example, the curve experiences
unbounded acceleration. A sufficient condition for the g.a.l. incompleteness
of a timelike curve has been developed by Sussmann (1988). Consider a
36 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

timelike half curve y(r), T 6 [0, TJ), Tt < +00, where T is proper time along
the curve. Assuming that this curve has finite proper length, it will also have
finite g.a.l. if there exist an L and a T2 such that 0 < L < 1, 0 < T2 < Zj, and
( ) IT1 ~ T < L for all T2 < T < TJ. Clearly, if a(r) is bounded along y(r), this
a T

criterion is met.
A spacetime is said to be b-complete just in case every half-curve has
infinite g.a.l.17 From the above remarks it follows that ^-completeness
entails geodesic completeness and bounded acceleration completeness. But
^-completeness is stronger in that it is not entailed by the other senses of
completeness. We can now formulate the most widely accepted definition of
a singular spacetime, the semiofficial definition', a spacetime is singular if and
only if it is A-incomplete. (Hawking and Ellis (1973) take this criterion to
apply to all curves. However, one might reasonably want to restrict it to
timelike and null curves.) This semiofficial definition passes one quick reality
check: it counts Minkowski spacetime as non-singular. Also according to the
semiofficial view, the asymptotically delayed big bang in Sussmann's (1988)
solutions (see section 2.2 above) is not counted as a singularity since any curve
which reaches it is ^-complete.
Each ^-incomplete curve in a spacetime M, gab can be thought of as
defining an ideal point on a boundary dbM of M. To give some life to the
word 'boundary' here, we need a prescription that will tell us when two
^-incomplete curves define the same boundary point; and, more ambitiously,
we would like a prescription that will give some structure to the enlarged
manifold M =• M U db M—at a minimum, a topological structure, so that we
can say which points in the boundary dbM are in the neighborhoods of what
points in the interior M. We would then have succeeded in replacing the
3xistence of spacetime singularities with existence in dbM in a manner that
localizes the singularities. Schmidt's (1971) prescriptions use the fact that the
bundle of orthonormal frames 0(M) admits a natural positive definite
metric.18 Thus, we know from the above remarks that 0(M) can be made
into a metric space. It turns out that this metric space is Cauchy complete
just in case the spacetime is ^-complete. If M, gab is ^-incomplete, the Cauchy
completion can be taken in 0(M) and then projected down to give M u dbM.
Unfortunately, the resulting bundle boundary has some counterintuitive
properties. For instance, in the FRW spacetimes 8bM consists of a single point
that is not Hausdorff separated from points of M (see Johnson 1977).
Such results might be taken to call into doubt the wisdom of defining
singularities in terms of ^-completeness. For if one goes beyond geodesic
completeness and bounded acceleration completeness in order to guarantee
freedom from singularities, why settle on ^-completeness rather than on some
other stronger or weaker notion of completeness? Delivering an appealing
way to attach singular boundary points to the spacetime manifold would have
provided an answer, but this is just what the ^-boundary approach fails to
deliver. However, other results tend to absolve the 6-boundary approach for
the failure. For instance, Geroch, Liang, and Wald (1982) have argued that
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 37

Fig. 2.1 The Ellis-Schmidt classification of spacetime singularities in the i-boundary

approach

the failure of singular points to be Hausdorff separated from interior points

is an inescapable feature of any attempt which seeks to represent singular
points of a spacetime M, gab in terms of a topological space M which has M
as an open dense subset and which shares with the ^-boundary approach
certain natural conditions; namely, any incomplete geodesic in M, gab ter-
minates in a point of M — M, and (in a certain technical sense) M is
geodesically continuous.19 It appears then that the goal of localizing space-
time singularities is unattainable without paying the price of counterintuitive
features of the localization.
Another consideration in favor of the ^-boundary approach is the
interesting classification of singularities to which it gives rise. Following Ellis
and Schmidt (1977), spacetime singularities can be put into the categories
listed in Fig. 2.1. A regular point p e dbM in the Ellis-Schmidt classification
is what was called above an inessential singularity, i.e., M, gab can be extended
to a spacetime M, gab such that the image of p is a point of M. Singular points
are then defined as all of the non-regular points of dbM. These points are
then further subdivided into quasi-regular points and curvature singularities.
Let y be a ^-incomplete curve terminating in a point/* 6 dbM, i.e., there
is a generalized affine parameter v for y and a v+ < +00 such that y(v) c: M
for v € [0, v+ ) and y(v + ) = p. Suppose that in any p.p. frame on y some of the
physical components R(i)(j)(k)(i)(j(v)) do not approach limits as v -> v+ . Then
p is said to be a curvature singularity. On the other hand, suppose that for
any 6-incomplete j ( v ) , v € [0, v+ ) terminating in the point pedbM there
is some p.p. frame along y(v) such that all of the physical components of the
curvature tensor approach limits as v -> v + . Then p is said to be a quasi-
regular point. This terminology is justified by a pretty result of Clarke (1973).
Say that the 6-incomplete y is locally extendible just in case there is an open
U c M containing y([0, v+ )} arid an isometric imbedding (p of U into a U, gab
38 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 2.2 An example of a spacetime that is locally extendible but not extendible (after
Ellis and Schmidt 1977)

such that (p ° y is C° extendible in U beyond v = v+. Clarke showed that if 7 is

a ^-incomplete curve that corresponds to a quasi-regular singularity, then y
is locally extendible.
An example of a spacetime that is not extendible but is locally extendible
is the spacetime with the conical singularity discussed in section 2.2. The idea
is illustrated in Fig. 2.2. The image of the curve y can be extended into
U — cp(U). U can be enlarged to cover "almost all" of the cone—indeed, all
of the cone save a line extending down from the vertex. The boundary of the
image <p(U) of this U in U contains two lines which meet at the vertex,
whereas the pre-images of these lines on the cone consist of a single
line. Thus, adding the vertex p back to form a global extension of the cone is
prevented by there not being enough directions at/) (Ellis and Schmidt 1977,
p. 931). This example is admittedly artificial. But similar singularities occur
naturally, as in the Curzon bipolar solution to EFE (recall section 1.3).
Vickers (1987) has argued that two-dimensional cone singularities of this type
model cosmic strings.
Another and much more perplexing example of a quasi-regular singularity
is provided by Taub-NUT spacetime. Some of the causal features of this
spacetime are captured in the simpler two-dimensional spacetime of Misner
(1967); see Fig. 2.3. There is a null geodesic y that starts on E and that winds
round and round the Taub portion of the universe, approaching but never
crossing the null surface JV. Since in the future direction y uses up only a finite
amount of affine parameter it defines a point in dbM. However, a computation
shows that the curvature components in a p.p. frame on y remain well behaved
as N is approached; so by Clarke's result there is a local extension in which
y can be continued.
Curvature singularities can be further classified, the first main subcategory
being scalar polynomial (s.p.} singularities. There are in turn two main sub-
categories, (i) Blowup s.p. singularities. This is the case that is brought
most immediately to mind by the phrase "spacetime singularity." Such a
singularity occurs when there is a 6-incomplete curve y ( v ) , v e [ Q , v + ), defining
a p e dbM and a scalar curvature polynomial % such that %(y(v}) is not
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 39

Fig. 2.3 Misner's two-dimensional spacetime illustrating some of the features of

Taub-NUT spacetime

bounded as Siklos (1979) proved that an s.p. blowup occurs just in

case some of the physical components of the Riemann tensor are unbounded
in all frames along y, not just the p.p. ones. The FRW big bang models and
the Kruskal—Schwarzschild model are examples where there are such s.p.
blow up singularities, (ii) Oscillatory s.p. singularities. Here every s.p. remains
bounded along y(v) but some s.p. %(y(v)) fails to approach a limit as v —» » + .
A modified version of the Taub—NUT spacetime can be used to give an
example of this type of behavior. Consider an incomplete null geodesic y(v)
(with v an affine parameter) that winds its way up through the Taub region
and approaches the null surface N that separates the Taub region from the
NUT region. Since the curvature components in a p.p. frame along y(v)
approach finite limits as v —> » + , so a fortiori must any s.p. But one also has
an independent argument for the latter fact. Since the metric is smooth in a
compact neighborhood of JV, any s.p. must remain bounded in such a
neighborhood. Furthermore, every point/) 6 jVis a limit point of y. Choosing
on y an infinite subsequence of points that converges to p, it follows that the
values of any s.p. at these points must approach a finite limit as p is
approached. And since JV is a surface of homogeneity, the limit is the same
at every p e JV. Thus, any s.p. %(y(v)) must approach a limit as v -> v + . Now
modify the example by choosing a q e JV and multiplying the metric by a
conformal factor Q2 in a small neighborhood of q. The causal structure of (a
two-dimensional version of) the resulting spacetime is still as pictured in Fig.
2.3, and relative to the new affine parameter v of the conformal metric, y(v)
is still future incomplete. Furthermore, by the same argument as before, all
of the scalar curvature polynomials, calculated in the new conformal metric,
40 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

must remain bounded as y approaches N. But the homogeneity of JVhas been

destroyed by the conformal factor fi2, and with appropriate behavior of Q2
some s.p. can be made to oscillate without limit as v —> v+.20 Of course, this
is merely a mathematical example since the conformally modified Taub—
NUT spacetime is no longer a vacuum solution to EFE or even to any
reasonable field equation. In a more physical vein, King (1974) has found
oscillating s.p. singularities in cylindrically symmetric stationary dust solutions
to EFE.
The second main category of curvature singularities consists of the
non-scalar polynomial singularities. As with the first category, these can be
divided into blowup and oscillatory types. In the blowup case the trouble can
be attributed to something going wrong with parallel transport. For by
definition, some physical component of the curvature tensor is unbounded in
any p.p. frame along a ^-incomplete curve y while (by Siklos' result) all
components remain bounded in some non-p.p. frame. Although various
examples of non-s.p. (aka whimper] singularities can be given (see Siklos
1981) there are results that indicate that these singularities can occur only in
special and, perhaps, physically unrealistic circumstances (see Ellis and
Schmidt 1977).
Still further refinements of the classification of curvature singularities can
be contemplated. For example, the ill behavior of the curvature can be traced
to either the Ricci tensor or the Weyl tensor.21 Or the curvature singularities
can be ranked by their strength—for instance, as to whether they imply that
a volume element is squeezed to naught as the singularity is approached (see
chapter 3).
Although brief and incomplete, this glimpse of the varieties of spacetime
singularities should suffice to indicate that only in special cases can singularities
(qua incompleteness) in general relativistic spacetimes be made to conform
to our original crude idea of some quantity blowing up.

2.5 The missing missing points

The analysis of singularities in terms of ^-incompleteness leads to a pleasing

classification scheme. But there is a disturbing feature of the analysis; namely,
it is not true to an idea that is arguably a touchstone of singularities in
relativistic spacetimes: spacetime singularities correspond to missing points.
To see where contact is lost with this touchstone, return to the better
understood case of a Riemannian space M, hab. We saw that hab generates a
distance function d( , ) that makes M, d( , ) a metric space, and that Cauchy
completeness for M, d( , ) is equivalent to geodesic completeness for M, hab.
In turn each of these completeness properties is equivalent to the property
of finite compactness: every subset of M that is af-bounded (i.e.,
sup{«?(/>, q): f>, q e X} < co) has compact closure. It follows immediately that
any compact Riemann space is complete. This is essential to the idea of using
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 41

incompleteness as a criterion of missing points since a compact M is never a

proper submanifold of another (connected, Hausdorff) manifold.
In the relativistic spacetime setting the trouble is that a spacetime M, gab
can be ^-incomplete (indeed, geodesically incomplete) even though M is
compact. Misner (1963) provided an example of a two-torus T2 equipped
with a Lorentz metric that is null geodesically incomplete. There is also the
example of Taub-NUT spacetime studied in the preceding section. Here the
spacetime is not compact, but there are incomplete null and timelike geodesies
that are contained in a compact set. In the two-dimensional Misner spacetime
pictured in Fig. 2.3, the incomplete null geodesic y is confined to the closure
of a neighborhood of the surface N. As a result, the spacetime manifold M
cannot be extended to a larger M in which the incomplete y is continued (see
Hawking and Ellis 1973, p. 289). In this example, however, the incomplete
geodesic y defines a quasi-regular point and is, therefore, locally extendible.
But we also saw that a conformal modification of the metric turns this into
an example in which y is not even locally extendible.
Several reactions to these examples are possible. The first is to hold fast
to the idea of missing points as the touchstone of spacetime singularities and,
consequently, to modify the definition of singularities along the following
lines.22 Proposed modification of the semiofficial definition: M, gab is non-singular just
in case every half-curve is ^-complete or else is contained in a compact subset
of M. One possible objection to such a definition is that it contains an element
of arbitrariness in that one could construct related but different definitions
(see Geroch 1968a). For instance, there are presumably cases where a
^-incomplete curve is not contained in a compact set but does continually
reenter a compact set. Should we regard such a spacetime as being singular
or not? This particular challenge can-be met by the answer "Non-singular,
in order to be true to the 'missing points' touchstone." For the fact that the
curve continually reenters a compact set K c M means that it has a limit
point p e M, and this is enough to show that M cannot be extended to a
larger M into which the curve can be continued. But there may be other
challenges along these lines that do not have such a clear answer.
A second reaction would be to insist that ^-incompleteness, whether or
not it makes contact with the missing points touchstone, does correspond to
one good sense of spacetime singularity. Thus, speaking partly to examples
of the sort discussed immediately above, Hawking and Ellis acknowledge
"There is no possibility of the incompleteness having arisen from the cutting
out of singular points." But they add:

Nevertheless, it would be unpleasant to be moving on one of the incomplete

timelike geodesies for although one's world line never comes to an end and
would continue to wind round and round inside the compact set, one would
never get beyond a certain time in one's life. It would, therefore, seem
reasonable to say that such a spacetime is singular. (Hawking and Ellis 1973,
P. 261)
42 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

For those who share this attitude some subscripting of the word 'singularity'
is needed: singular j to refer to singularities associated with incompleteness,
singular2 to refer to singularities associated with missing points.
A middle way is also possible. It acknowledges that ideas of missing points
and incompleteness lead to different concepts of spacetime singularities, but
it claims that these concepts are nearly extensionally equivalent for physically
reasonable cases. For apparently the only cases where the concepts can come
apart are (1) for a ^-incomplete spacetime M, gab with compact M, or else
(2) for a spacetime M, gab where M is non-compact but a ^-incomplete curve
is contained in a compact K c M or else continually reenters K. In the first
case the spacetime contains a closed timelike curve. This follows from the fact
that the sets I+ (p), p 6 M, I+ (p) =•• {q e M: p « q}, where p « q means that
there is a non-trivial future directed timelike curve from p to q, form an open
cover of M. Since M is by assumption compact, there is a finite subcover
/ + (/>!), I + ( p 2 ) , . . . , / + (/>„)• Then either/)! e 7+ (/»,) for some i > 1 or not. If
so, then / + (/>i) can be omitted. If not, Pi^l+(p1) and there are closed
timelike curves. Continuing in this way produces the desired conclusion. In
the second case, if the incomplete curve is non-spacelike, the spacetime
contains almost closed causal curves. More precisely, there is a violation of
the strong causality condition which holds for a spacetime M, gab just in case
for all p e M, any open neighborhood N(p) of p contains a subneighborhood
JV'(/)) which no non-spacelike curve reenters after leaving. For a proof that
the strong causality condition is violated if a future inextendible causal curve
is totally or partially imprisoned in a compact set, see Hawking and Ellis
(1973, Prop. 6.4.7). There is a loophole here concerning a ^-incomplete
spacelike curve that is partially or totally imprisoned in a compact set, but
assuming that this loophole can be plugged, the ground is laid for the following
stance. Closed or almost closed causal curves (so the argument goes) make a
spacetime a priori unacceptable as an arena for physics. And within the
acceptable arenas, the two concepts of spacetime singularity—based respec-
tively on the ideas of missing points and incompleteness—are in agreement.
The a priori postulation of causal features was once a popular move. Of late,
however, its popularity has slipped badly among general relativists. This
matter will be discussed in section 2.6 and in more detail in chapter 6.
Even if in physically reasonable cases there is an extensional equivalence
of the concepts of singularities as involving incompleteness and singularities
as involving missing points, the two concepts are distinct, and the latter
deserves to be developed in its own right. The modification proposed above
for the semiofficial definition of singularities is ad hoc and does not do justice
to the missing points idea. More justice can be done by employing some of
the ideas of Susan Scott (1992), which I proceed to sketch below.23
If there are missing points for a spacetime M, gab they have to arise
from deleting points from an envelopment of M. Formally, an envelopment
of the (^-dimensional) M consists of a connected (w-dimensional) M and
an imbedding (p: M —» M. (p(M] is an open subset of M. Taking the
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 43

closure of (p(M) in M and subtracting off (p(M) gives the boundary

B(M, M, <p) =:<p(M) — (p(M) ofMin the envelopment M, (p. Ifp e B(M, M, (p)
is not the limit point 24 of some appropriately incomplete half-curve of M, gab
(e.g., an incomplete half-geodesic or a half-curve of finite g.a.l.) then/) is said
to be a point at infinity; otherwise p is said to be at a finite distance.
Focusing now on boundary points at a finite distance, we need to weed
out ones that do not correspond to genuine singularities. The weeding is done
in two steps. First, we have to pull out the regular points. Assume that the
metric gab on M is Ck (k > 1). A point p 6 B(M, M, (p) is said to be Cl regular
(k>l> 1) just in case there is an open submanifold M cr M with <p(M] cr M
and p 6 M, and a C1 metric gab on M such that M, gab is an extension of Af, gab
under (p. The second bit of weeding focuses on the non-regular boundary
points that lie at a finite distance— the candidate singularities— and is more
complicated because it takes into account the various ways the base manifold
M can be enveloped. Let M', <p' be a second envelopment, and let B(M, M', (p')
denote the boundary set for this second envelopment. B' cr B(M, M', (p')
is said to cover p e B(M, M, (p) just in case for every open neighborhood
U' cr M' of B', there is an open neighborhood U cr M of p such that
<p' o (p~l(U n <p(M)) £ U'. The candidate singular point p e B(M, M, (p) is
said to be removable just in case there is another envelopment M', <p' and a
C1 regular subset B' c B(M, M', (p') that covers p. A relevant example has
already been provided by the discussion of the Schwarzschild metric in section
2.3. Take M c S2 x [R2 to be the open subset covered by the Droste
coordinates for r> a (recall section 1.2.). Under the imbedding of M into
M = S2 x IR2 given by the identity map provided by the Droste coordinates,
the points (r = a, 8, <p, /) of M belong to B(M, M, id). These points are not
C1 regular for / > 0, but they lie at a finite distance since they are the endpoints
of incomplete geodesies in M. Thus, they are candidate singular points.
However, they are removable. A second envelopment is given with the help
of the Kruskal coordinates

The new imbedding (p' is given by sending (r, 6, (p, t) to (X, 0, (p, T). The
boundary points B(M, M', <p'), where M' = M = S2 x R 2 , are C00 regular
and cover the boundary points (r = a, 0, (p, t} ofB(M, M, id) (see Scott 1992,
p. 184). By contrast, r = 0 corresponds to a non-removable singularity.
The present approach counts the Misner compact spacetime and the
Taub— NUT spacetime as non-singular. It therefore reproduces in a motivated
way the results of the ad hoc modification of the semiofficial definition
considered above. Whether or not this approach will prove to be fruitful
depends upon how it can be used to classify the various types of singularities
44 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

it considers to be genuine (i.e., non-removable). This matter cannot be

pursued here (see Scott 1992 for some details).

2.6 Naked singularities

Penrose's (1969) cosmic censorship conjecture supposes that GTR has built
in prohibitions against "naked singularities." This conjecture will be discussed
in detail in chapter 3, but in advance it is worth pointing out that cosmic
censorship raises a set of concerns that, while not orthogonal to those of the
preceding sections, are at least oblique to them. The strictest form of cosmic
censorship rules out naked singularities by requiring that the spacetime be
globally hyperbolic, which is the technical way of saying that the spacetime
does not contain any pathologies that prevents the implementation of global
Laplacian determinism. A globally hyperbolic spacetime contains a Cauchy
surface, i.e., a spacelike hypersurface £ that intersects every causal curve
without endpoint exactly once. The specification of appropriate initial data
on S determines throughout the spacetime a unique solution to the coupled
Einstein-matter equations, as will be discussed in chapter 3.
What needs to be emphasized here is that a spacetime can be globally
hyperbolic and still singular in one or more of the senses studied above—the
FRW big bang models are prime examples. And in the other direction the
spacetime can fail to be globally hyperbolic (i.e., can be nakedly singular in
Penrose's sense) without being singular in any of the senses studied above.
To see this, note that global hyperbolicity is equivalent to the conjunction of
two conditions. The first is strong causality, which we met in the preceding
section. The reader can easily construct examples where strong causality
fails but there are no incomplete curves or missing points. The second
condition requires that J~ (p) n J+ (q) be compact for p, q 6 M, where
J~ (p) =: {s e M: s < p} and jf+ (q) =•• {s 6 M: q < s] and x <y means that
there is a future directed non-spacelike curve from x to y. Construct a
spacetime as follows. Pick two timelike related points q«p of Minkowski
spacetime IR4, rjab and remove a closed ball K in the interior of J~ (p) n J+ (q)
(see Fig. 2.4). The resulting spacetime has a trivial singularity which can be
doctored as followed. Choose a scalar function Q which blows up rapidly as
the (missing set) K is approached, and define a new metric by gab ='-Q2t]ab.
The resulting spacetime IR4 — K, gab is inextendible and strongly causal. It is
also non-singular in that there are no ^-incomplete curves, no curvature
blowup, no missing points, etc. But it fails to be globally hyperbolic. (This
example also illustrates how the failure of Penrose's cosmic censorship leads
to the possibility of supertasks discussed in chapter 4.) In sum, naked
singularities in Penrose's sense form a third category of singularities concep-
tually distinct from the singularities associated with incompleteness and
missing points.
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 45

Fig. 2.4 A nakedly singular spacetime without singularities

Fig. 2.5 A globally hyperbolic spacetime that is not maximal

The statement of cosmic censorship implicitly assumes that the spacetime

model is maximal. The issue of maximality here is most easily discussed by
using the fact that a globally hyperbolic spacetime contains a Cauchy surface.
If M, gab has a Cauchy surface Z, then there is a unique (up to isometry)
maximal extension M, gab of M, gab in which Z is still a Cauchy surface
(Choquet-Bruhat and Geroch 1969). The spacetime pictured in Fig. 2.5 is a
diamond-shaped cutout from two-dimensional Minkowski spacetime. It
cannot be extended in any way that keeps Z (or any other Cauchy surface)
Cauchy. But since it is extendible it does not satisfy the intended strong form
of cosmic censorship. I will return to these matters in chapter 3.
There is also a connection between cosmic censorship and the recently
discovered exotic differentiable structures IR4? for topological !R4. These are
structures which arc compatible with the topology of (R4 but are not
diffeomorphic to the standard atlas of charts. 25 Such exoticness cannot bloom
46 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

in the presence of Penrose's cosmic censorship hypothesis, at least if the

Poincare conjecture is correct.26 If M, gab is a four-dimensional globally
hyperbolic spacetime, then as follows from Dieckman (1988), M is diffeo-
morphically IR X £, where Z is a three-manifold. However, it follows from a
theorem of McMillan (1961) that if the Poincare conjecture is true, exotic
IR| cannot be diffeomorphically such a product.

2.7 What is a spacetime singularity (again)?

Haven't we already answered this question to the best of our ability? To be

honest, no. The main target of our quest is the set of essential singularities,
that is, singularities that cannot be removed by extending the spacetime.
(In the ^-incompleteness classification these are singularities that corre-
spond to non-regular points; in the Scott classification they are the non-
removable singularities.) But this is not a well-defined set until the continuity/
differentiability (c/d) conditions on the allowed extensions are specified.
There are twin dangers to be avoided in making the specification. On the
one hand we do not want to set the c/d conditions so high that they exclude
extensions that might be realized in some physically reasonable situation. On
the other hand, we do not want to set the c/d conditions so low that anything
goes. At the extreme, if no c/d conditions are imposed then no spacetime will
be counted as having an essential singularity (see Clarke 1993, pp. 115—116).
For example, in the case of an FRW universe with a big bang and a big
crunch we could "extend" both forward and backward in time and imagine
an oscillating universe that recycles itself in a never-ending series of expansions
and contractions. And indeed, such scenarios are to be found in the early
scientific literature (see chapter 7) and in current popular science writings.
But while it is amusing to construct such scenarios, one wonders whether they
belong to art rather than science. Are such constructions constrained by
physical principles, or is imagination the only guide?
At this juncture one might eschew the task of trying to specify a cut
between essential and inessential singularities in favor of classifying singular-
ities as to their strengths. Thus, for example, one could say that a spacetime
has a Ck (for <t = 0~, 0, 1 ~, 1, . . .) singularity in the sense of ^-incompleteness
if there is no Ck extension through the singularity; the lower the k, the stronger
the singularity. If, however, one tries to choose a cutoff point for essential
singularities, a plausible starting position is to set the c/d condition on
allowable extensions by what is necessary and sufficient to make sense of the
laws of GTR, principally EFE.
At a minimum, one wants EFE to be defined in the sense of distributions.27
Geroch and Traschen (1987) have argued that the appropriate class of metrics
for meeting this requirement are the regular metrics', a symmetric tensor field
gab is a regular metric just in case (i) the inverse gab exists and both gab and
gab are locally bounded, and (ii) the weak first derivative of gab exists and is
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 47

locally square integrable.28 A C2 metric is regular, but much less well-behaved

metrics are also regular. To understand the motivation behind this definition,
note that if gab is a smooth metric, the curvature tensor for gab can be written as

where

The philosophy of the Geroch-Traschen approach is that if R,^ is to be

interpreted as a distribution, then each term on the right-hand side of the
defining equation (2.1) should be independently interpretable as a distribu-
tion since one does not want to rely on the happenstance that the terms add
up in just the right way to produce a well-defined distribution. That the
metric is regular does guarantee that each term on the right-hand side of
(2.1) is meaningful as a distribution. To see this, note first that if the
derivatives of gab are interpreted as weak derivatives, then (2.2) is meaningful
for a regular metric, and T"bc will be locally square integrable. Thus, the first
term on the right-hand side of (2.1), which is the product of two F's, will be
a locally integrable function and thereby will define a distribution.29 By the
same token, Tabc defines a distribution, and since the ordinary derivative of
a distribution is a distribution, the second term on the right-hand side of (2.2)
is also meaningful as a distribution. Some further calculation shows that
taking the outer product of any number of gabs or g" 's for a regular metric
and the Rabcd from a regular metric results in a distribution. By taking
contractions of such products, it follows that for a regular metric the Einstein
tensor Gab —' Rab — ( l / 2 ) g a b R is meaningful as a distribution. So the vacuum
EFE Gab = 0 make distributional sense for a regular metric; and so do the full
EFE Gab = 8nTab, at least if Tab = (\/8n)Gab is used as a definition of the
energy— momentum tensor. In sum, the regularity of the metric is sufficient
to guarantee that EFE are distributionally meaningful. There is no airtight
argument to the effect that regularity is a necessary as well as sufficient
condition; for after all, there are distributions that do not arise from locally
integrable functions, the famous Dirac ^-function being a prime example.
But it is hard to see how to weaken the requirement of regularity and still
guarantee that Gab is a distribution.
Requiring that the metric of the extended spacetime be regular may be
necessary for a physically meaningful extension, but it arguably does not go
far enough. For the satisfaction of EFE in a distributional sense does not
guarantee that these equations attain their intended physical content. For
instance, the Bianchi identity ViaRbc}de = 0 need not make sense for a regular
metric. But this identity is crucial to two intended implications of EFE. The
first implication concerns conservation principles. It is the Bianchi identity
that entails that VaGab = 0. The application of EFE then produces the local
conservation law VaTab = 0. In the second place, the Bianchi identity is also
crucial to the initial value formulation of GTR. EFE can be divided into
48 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

constraint equations, which constrain the initial data on a spacelike hyper-

surface, and the evolution equations, which specify how the initial data is to
be evolved off the initial hypersurface. The Bianchi identity is used to prove
that if the constraints are satisfied at the initial moment, then they will be
preserved by the time evolution. Thus, from the point of view of physics, one
wants enough c/d to insure that the expression V[a/?bc]/ makes sense. This
seems to require that not only is the metric regular but also that the weak
second derivative of the metric exists and is locally square integrable.30
One may also want to make sure that the intended deterministic
development of the spacetime structure is secured. Thus, in the matter-free
case (Gab — 0) if the value of the metric and its time derivative are specified
on a spacelike hypersurface Z—technically, the first and second fundamental
forms of Z (see chapter 3)—then one wants to be able to prove the local
existence and uniqueness of solutions to Einstein's vacuum field equations
corresponding to the given initial data.31 Fischer and Marsden (1979) stated
the weakest known sufficient conditions for local existence and uniqueness in
terms of Sobolev spaces of non-integer dimension. A sufficient condition for
their sufficient condition is that the metric as restricted to the initial value
surface be C3 and the normal derivative of the metric be C2. It is apparently
not known what the minimally sufficient conditions are.
In sum, it is far from clear what c/d conditions should be required of
physically meaningful extensions. Presumably, the metric should be at least
regular, but what further conditions should be imposed remains up for grabs,
although there are a number of obvious factors that go into the decision.
Further discussion will have to be hypothetical, supposing one or another of
the decisions that could be made. For sake of discussion then, suppose first
that it has been decided that physically meaningful extensions require that
the metric must be at least C2. Then most of the familiar solutions to EFE
with singularities—FRW, Kruskal-Schwarzschild, etc.—will be counted as
essentially singular. Moreover, with the help of a little Leibnizian metaphysics
we can apply the Hawking—Penrose singularity theorems (to be discussed
in section 2.8) to conclude that essential singularities can be expected to occur
under generic conditions in metaphysically acceptable models of gravitational
collapse and cosmology. These theorems assume that (a) each p e M and each
direction V £ Mp together define a unique geodesic, and (b) the geodesic
depends differentiably on the initial data. Condition (a) requires that the
metric be C2~ (i.e., the second derivative exists and is locally bounded), and
(b) requires that the metric be C2. But since a C2~ metric can be approxi-
mated by C2 metrics, C2" suffices for both (a) and (b). If we are given a C2
spacetime M, gab to which the conditions of the Hawking—Penrose theorem
apply, we may conclude that the spacetime is singular in the sense of geodesic
incompleteness. Of course, this incompleteness may be due simply to the fact
that the spacetime is extendible. We know from section 2.3 that every C2
spacetime is extendible to a maximal C2 spacetime. So either the spacetime
in question is maximal or else not. In the former case the spacetime is
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 49

essentially singular. In the latter case the singularities may not be essential,
but the spacetime can be ignored on the grounds that it violates Leibniz's
principle of sufficient reason (see section 2.3).
It would be nice to be able to draw conclusions about essential
singularities from existing singularity theorems. But there is a potential
awkwardness to the situation we have envisioned. Impulse wave solutions
(metric C1", i.e., the first derivative exists and is locally bounded) will be
counted as essentially singular. More generally, solutions to EFE will be
counted as essentially singular even though there are well-defined geometrical
extensions of differentiability lower than C2 (or C 2 ~). In this sense there would
be nothing in the geometry of spacetime itself to prevent extending through
the singularity, and only the failure of the physical laws (that by supposition
require C2 metrics) to keep up with the geometry would lead us to say that
the singularity is essential. This parting of company of geometry and physics
in a theory that was supposed to marry them is simply the price that has to
be paid in the current approach to recognizing essential singularities.
Suppose next that we decide that the necessary and sufficient conditions
c/d for physically meaningful extensions are weaker than C2 (or C 2 "). One
could still talk about singularities in the sense of 6-incompleteness if the metric
were C1 ~. One result of the present supposition would be that fewer
spacetimes would be counted as essentially singular than on the previous
supposition. Another result would be that the Hawking-Penrose theorem
would no longer apply in its present form. It remains to be seen whether this
and related theorems can be reconstituted so as to predict the occurrence of
essential singularities under the hypothesized low c/d conditions. On the other
hand a potential benefit is the possibility of restoring a concordance between
geometry and physics by showing that for an essential singularity something
closer to a breakdown in the spacetime geometry prevents an extension.
It may be hard to believe that anything physically significant depends
on finicky details of the c/d conditions on the metric. But we have to be
prepared for the awful possibility that Nature does not share our beliefs. In
an optimistic vein, Hawking and Ellis wrote:
In fact, the order of differentiability of the metric is probably not physically
significant. Since one can never measure the metric exactly but only with
some margin of error, one could never determine that there was an actual
discontinuity in its derivatives of any order. Thus one can always represent
one's measurements by a C°° metric. (Hawking and Ellis 1973, p. 58)
The danger of this sort of reasoning has been emphasized by Chrusciel (1992).
The silent premise in the Hawking-Ellis argument is that any Ck (k > 0) field
on a C00 manifold can be approximated to arbitrarily great accuracy by a
C00 field. Their conclusion is that for purposes of physics one might as well
assume that the field is C00. By parallel reasoning one could argue that since
a C100 field on a manifold can be approximated to any desired accuracy by
an analytic field, we might as well assume that the field is analytic. But this
50 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

latter conclusion is misleading with respect to the problem of temporal

evolution in GTR; for while an analytic function is determined by its values in
any neighborhood, the non-analytic solutions of EFE do not have this property.
Moreover, as Chrusciel emphasizes, imposing differentiablity assumptions on
evolved solutions may actually be incompatible with EFE. Hyperbolic partial
differential equations are compatible with the loss of degrees of differentiability
through time. So while it may be justified to stipulate that the initial
conditions enjoy a certain smoothness, the stipulation that this smoothness
holds for all times may clash with EFE. More generally, if one takes a
non-instrumentalist view of spacetime structure, then continuity or differen-
tiability of the spacetime metric is either there or not, regardless of whether
we can directly detect it with our measuring instruments. And the differences
can have important consequences for the interpretation of singularities.
The upshot of this discussion is rather unsettling. Since we are left in a
state of uncertainty about what c/d conditions to require of extensions, we
are left with corresponding uncertainties about what an essential spacetime
singularity is and about how generic essential singularities are among the
solutions to EFE. Nevertheless, we do know some of the necessary conditions
that a physically meaningful extension should satisfy. This is enough to show
in some cases that the singularities involved are essential, but the demonstration
will be postponed until chapter 7.
Finally, a word should be said about the differentiable structure for the
spacetime manifold, which is often taken as an unproblematic given. An
illustration of how this structure may make a difference for singularities, at
least at the level of mathematics, has come to light in connection with the
exotic differentiable structures Kg for topological R 4 . It is known that flat
metrics (of any signature) which are smooth with respect to K| cannot be
geodesically complete (see Brans and Randall 1993). So either the exotic R|
do not admit flat Lorentzian metrics, which would be an interesting result
in itself; or else they do, and all these metrics are singular. In the latter case
there arises the question of the basis of the incompleteness. For some R|'s
there is no hope of completing such a metric by means of an extension into
the standard Rf since some R|'s cannot be smoothly imbedded into Rjj
nor, of course, can the completion be accomplished by extending to a flat
metric in R£. On the other hand, we know by the result of Clarke (1973)
that local extensions are possible. There is no topological obstruction to a
global extension, as with the quasi-regular singularity in Fig. 2.1 since we are
dealing with topological R 4 . So is the incompleteness due to acausal behavior,
as in Fig. 2.3, or is some new sort of differentiable obstruction involved?

2.8 Singularity theorems

How widespread are spacetime singularities among the solutions to EFE? One
approach to answering this question would be to take a headcount of the
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 51

known exact solutions to EFE. But such a count can hardly be expected to
give a reliable reflection of the situation in the class of all solutions since the
solutions we have in hand are perforce of a simple and, therefore, special
character. Another approach would be to attempt to establish theorems that
characterize general conditions under which singularities will occur. There
was no motivation to search for such theorems as long as it was believed that
spacetime singularities were artifacts of the idealizations involved in construct-
ing models of gravitational collapse and cosmology. Thus, it was not until
the 1960s that much progress was made on the second approach. At first there
were some false starts. For instance, in the second English edition of The
Classical Theory of Fields, Landau and Lifshitz reached the "fundamental
conclusion that the presence of a singularity in time is not a necessary property
of cosmological models of the general theory of relativity, and that the general
case of an arbitrary distribution of matter and gravitational field does not
lead to the appearance of a singularity" (Landau and Lifshitz 1962, p. 397;
see also Lifshitz and Khalatnikov 1963). Their argument was mistaken and
was later withdrawn. From the mid-1960s onward advances were rapidly
made, culminating in the main Hawking-Penrose theorem (1970). Since it
is one of the most important results in modern GTR and is central to our
concerns, it will be quoted and explained here.

Theorem (Hawking and Penrose). Let M, gab be a time-oriented spacetime

satisfying the following four conditions:
(1) RabVVb > 0 for any non-spacelike V.
(2) The timelike and null generic conditions are fulfilled.
(3) There is no closed timelike curve.
(4) At least one of the following holds:
(a) There exists a compact achronal set without edge.
(b) There exists a trapped surface.
(c) There is a p e M such that the expansion of the future (or past)
directed null geodesies through p becomes negative along each
of the geodesies.
Then M, gab contains at least one incomplete timelike or null geodesic.

Several explanatory comments are in order. The theorem is purely

geometrical—it uses no physical laws. Physics enters in justifying the various
hypotheses of the theorem. If EFE (without cosmological constant) are
satisfied and the energy—momentum tensor Tab satisfies the strong energy
condition: Tab^"C > -(1/2)T, T=:T"a, for any unit timelike £", then (1)
will hold. The strong energy condition is believed to be valid for any physically
reasonable classical source field. For a precise statement of the generic
conditions (2) the reader is referred to Hawking and Ellis (1973). These
conditions will be satisfied provided each timelike or null geodesic experiences
tidal force at some point in its history. So far then, the presumptions of the
52 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

theorem seem innocuous. The meat of the presumptions comes in condition

(4). Condition (4a) says that, at least at one time, the universe is closed and
that the compact slice corresponding to this time is not intersected more than
once by a future-directed timelike curve. Condition (4b) refers to a trapped
surface. This is a compact, two-dimensional submanifold Z c. M such that
the expansion of both ingoing and outgoing future-directed null geodesies
orthogonal to Z is everywhere negative. The physical significance of this
condition derives from the fact that trapped surfaces are expected to form in
gravitational collapse. Indeed, a result of Schoen and Yau (1983) shows that
when a sufficient amount of matter is concentrated in a small enough region,
a trapped surface results. Condition (4c) on the expansion of null geodesies
can be expected to hold in various scenarios when the universe is collapsing
in the past or future directions.
To convey some sense of how singularity theorems are proved, it will be
helpful to concentrate on a simpler and less powerful theorem. Consider a
spacetime that is globally hyperbolic (see section 2.5) and that, therefore,
possesses a Cauchy surface S. Cauchy surfaces have a number of nice
properties, but the only one needed here is that for any point p, say to the
future of S, there is a longest timelike curve from p to S. This curve will be
a geodesic y (because in relativistic spacetimes timelike geodesies maximize
rather than minimize proper length) and it will be orthogonal to Z (because
otherwise one could obtain a longer curve by veering offy near where it meets
E). Let us then focus on the congruence of timelike geodesies orthogonal to
Z. If V" is the unit timelike vector field tangent to the congruence, the expansion
9 of the congruence is defined by 0 =:VaVa. The Raychaudhuri equation (a
purely geometrical relation) shows that the rate of change 6 ='-d9/dx = VVa9
(where T is proper time) is given by

where aab and coab are respectively the shear and rotation matrixes for the
congruence. 32 Since the congruence is orthogonal to Z, coab is initially zero;
and since coab vanishes at all times if it vanishes at any time, (£>ab = 0. The
second term on the right hand side is obviously non-negative. And by
condition ( 1 ) the fourth term is also non-negative. Thus, we can conclude that

Integrating (2.4) yields

It follows that if the initial expansion 90 is negative, that is, the congruence
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 53

Fig. 2.6 An illustration of conjugate points

is converging, then 9 -» — oo (the geodesies begin to cross33) within a proper

time T < 3/|$0|. Now suppose for purposes of contradiction that there were a
timelike curve from E into the future whose proper length exceeded 3/|00|.
Choose a point p on the curve lying a proper length greater than 3/|00| to
the future of E. We know that there must be a longest timelike curve from p
to E and that it will be a timelike geodesic y orthogonal to E. But y cannot
be the longest curve. For a neighboring geodesic y' will intersect y at a point
r between p and E (see Fig. 2.6). By constructing qsp, a timelike curve of
proper length greater than 3/|00| can be obtained. To escape the contradiction
we must conclude that no timelike curve from E into the future has a length
greater than 3/|00l and, thus, that every timelike geodesic is incomplete.
This simple singularity theorem is limited to collapsing universes with
Cauchy surfaces, but proofs of the more far-reaching theorems, though much
more difficult, use many of the same concepts. The argument given above is
an application of the concept of conjugate points. A point r on a geodesic y
belonging to a congruence of geodesies orthogonal to a spacelike E is said to
be conjugate to E along y if (intuitively) geodesies that are infmitesimally
close to y at E begin to cross at r. This will happen just in case (as in the
illustration above) 0 —» — oo at r. More generally, consider any congruence
of timelike geodesies, hypersurface orthogonal or not. Points p and q on some
geodesic y of the congruence are conjugate just in case infinitesimally nearby
geodesies intersect 7 at p and q. The importance of this concept lies in the
fact that a timelike geodesic from p to q maximizes proper length between p
and q if and only if there is no point conjugate top between/? and q. Conditions
(1) and (2) of the Hawking-Penrose theorem are used to prove the existence
of conjugate points on complete geodesies. Thus, in a spacetime satisfying (1)
and (2), if all timelike geodesies are complete, they could not maximize proper
length. A contradiction is obtained if other hypotheses can be shown to entail
the existence of a longest timelike curve. The presence of a Cauchy surface
will suffice, but so will much weaker conditions, although the proof that
weaker conditions suffice is much more elaborate. Readers interested in the
details may consult Hawking and Ellis (1973) and Wald (1984a).
54 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

At this juncture two features of the Hawking—Penrose theorem should be

emphasized. The first is that, under the stated conditions, the theorem
demonstrates the occurrence of singularities in the form of geodesic incom-
pleteness and, a fortiori, 6-incompleteness. But it gives no information about
the type of singularity that can be expected to occur. I will have more to say
about this matter in section 2.9.
The second feature which calls for comment is the causality condition
(3). Such an assumption is harmless as long as "no time travel" is regarded
as an a priori necessary condition for physically reasonable models of the
universe. However, such an attitude is no longer as popular as it once was.
As R.P.A.C. Newman notes:
It has become customary to claim that closed timelike curves render a
spacetime physically unreasonable. Certainly, if the universe does contain
closed timelike curves, a revision of fundamental premises of physics, and
philosophy, may be necessary. However, to dismiss this, and other forms of
causality violation, out of hand is reminiscent of the dogmatism regarding
singularities prior to the singularity theorems. (Newman 1989, p. 982)

There is a singularity theorem due to Hawking (1967) that does not

explicitly mention causality.

Theorem (Hawking). Let M,gab be a time-oriented spacetime satisfying

the following three conditions:
(!') Rab VVb > 0 for every non-spacelike V.
(2') There exists a compact spacelike hypersurface £ c M without edges.
(3') The unit normals to E are everywhere converging (or diverging).
Then M, gab is timelike geodesically incomplete.

Condition (!') is the same as (1) in the Hawking-Penrose theorem. Condition

(3') says that at the time corresponding to £ the universe is everywhere
contracting (or expanding). Although the Hawking theorem does not explicitly
impose causality constraints, they are implicit in the hypotheses. For if the
spacetime is simply connected, the time slice 2 postulated by (2') cannot be
intersected more than once by a future-directed timelike curve. Thus, (2')
rules out various acausal spacetimes; in particular, viciously causal ones such
that a closed timelike curve passes through every point (at least assuming
simple connectedness).
One is thus led to wonder whether violations of causality can prevent the
occurrence of singularities. Newman (1989) has shown by explicit example
that the chronology condition in the Hawking-Penrose theorem cannot
be entirely removed. Starting from a conformal modification of a three-
dimensional Godel spacetime, he produced a four-dimensional spacetime that
satisfies conditions ( 1 ) , ( 2 ) , has a (non-achronal) trapped surface, but is
timelike and null geodesically complete. It is an open question whether
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 55

examples can be constructed with achronal trapped surfaces. If so, one can
envision cases of gravitational collapse which form black holes, the interiors
of which contain not singularities but causality violations.
Tipler (1976, 1977a) showed that in some restricted cases closed timelike
curves will be accompanied by singularities in the form of null geodesic
incompleteness. However, Tipler's results cannot be considered as a general-
ization of the Hawking-Penrose theorem since the crucial condition (4) of
this theorem plays no role in his proofs. Kriele (1990a) has proved a direct
generalization of the Hawking—Penrose theorem in which the chronology
condition (3) is replaced by a weaker but more complicated causality
condition which (very roughly) prohibits almost closed causal curves in the
surface generated by the null geodesies that are orthogonal to the trapped
surface. Kriele (1989, 1990b) also supplied a result that says in effect that
chronology violations which "do not extend to infinity" are accompanied by
singularities.

Theorem (Kriele). Let M,gah be a time-oriented spacetime such that

(1") RabK"Kb > 0 for all null vectors Ka.
(2") The null generic condition is satisfied.
(3") The chronology-violating region V a M has compact closure but
M- V+ 0.
Then if V ^ 0, the boundary of V is generated by almost closed incomplete
null geodesies.

If one imagines that the chronology violation is manufactured by a time

machine, one would like to strengthen this result to say that chronology
violations that do not start at infinity necessarily involve singularities. This
and other aspects of chronology violation will be taken up in chapter 6.
Summarizing the present situation, it seems fair to say that while
existing results are far from definitive, they strongly suggest that violation of
the chronology condition is not a promising way to avoid singularities.
Returning finally to the question with which this section began, are we
any wiser about how widespread singularities are among the solutions to EFE?
The question becomes more tractable if cases are divided. At a minimum we
should distinguish between solutions with A < 0, A = 0, and A > 0, the
expectation being that singularities are more prevalent for A = 0 than for
A > 0 and more prevalent still for A < 0. Cases can be further subdivided
between spatially open and spatially closed universes, the expectation being
that singularities are more prevalent in the latter cases. A further subdivision
would distinguish between vacuum solutions (7~a(> = 0) and non-vacuum
solutions (T ab ^ 0), the expectation being that singularities are more preval-
ent when matter-energy is present. For some sub-sub-sub- . . . cases we may
be able to prove unqualified results: every solution belonging to the said class
is singular (at least in the sense of geodesic completeness). In other sub-sub-
56 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

sub- . . . cases singularities may not be universal but nevertheless generic. To

make this precise one would have to define an appropriate topology on the
space of solutions to EFE and show that within the chosen subclass the singular
solutions form an open dense subset. I know of no such results in the literature;
indeed, how to define the appropriate topology seems to be an open question
(see Geroch 1970a). Nevertheless, intuition suggests that such results should
be forthcoming, for the existing singularity theorems encourage the notion
that when a solution to EFE is provably singular, a small perturbation of the
solution will produce a spacetime that is also singular.

2.9 Singularities and quantum effects

Faced with results indicating that singularities are a generic feature of general
relativistic spacetimes, two opposing attitudes are possible. One is that GTR
forces us to recognize spacetime singularities as a new feature of reality and
that we simply have to learn how to live with this new feature. The cosmic
censorship hypothesis, to be studied in detail in chapter 3, is an attempt to
make the coexistence a peaceful one. The opposing attitude is that no
cohabitation is possible since spacetime singularities are absurdities. This
second attitude promotes a reading of the Hawking-Penrose theorem and
allied results as threatening a reductio of GTR or at least as indicating that
GTR must break down under the conditions where the theory says that
singularities will develop. To be more than wishful thinking this second
attitude must be accompanied by some reason for believing that a more
complete and adequate theory will retain the successes of GTR while showing
how singularities are avoided. Some theoretical physicists espouse the hope
that a quantum theory of gravity will fit the bill. Since the outlines of such
a theory can only be dimly perceived, this hope cannot be assessed with any
accuracy at the present time. Nevertheless, it is a useful exercise to try to lay
out some of the thinking behind the hope.
To begin, why should one expect that quantum effects will come into
play in the regimes where the Hawking-Penrose singularity theorems and
allied results say that spacetime singularities develop? The answer would have
to come in two parts. First, one would have to cite considerations from various
approaches to quantum gravity to support the conclusion that when gravita-
tional fields become sufficiently strong on some quantum scale, quantum
effects will out. Second, one would have to find a way around the fact, noted
in the discussion in section 2.3, that it is mathematically possible to have
spacetime singularities (in the sense of geodesic incompleteness) that do not
involve arbitrarily strong gravitational fields in the form of unbounded
curvature. Thus, one would need to find support for what Clarke (1988) calls
the curvature conjecture', namely, that under physically reasonable conditions,
the singularities predicted by the Hawking-Penrose theorem and allied results
will involve unbounded curvature. As an example of a result that lends some
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 57

support to this conjecture, Clarke (1975a, b) showed that in a C°~ inextendible

globally hyperbolic spacetime that is not too specialized (in a technical sense),
a singularity p e 8bM that lies at the end of a A-incomplete timelike curve is
a curvature singularity. Combined with Penrose's form of cosmic censorship
the support lent to the curvature conjecture would be very strong; but if
cosmic censorship fails, the fate of the conjecture is left in doubt (see Clarke
and Schmidt 1977; Clarke 1988 for further discussion).
The next question to be addressed is why one should think that quantum
effects, assuming that they do kick in under the conditions that classical GTR
says that spacetime singularities form, should rescue us from singularities. The
first glimmering of hope has nothing to do with quantum gravity per se but
simply with the amazing ability of quantum mechanics to smooth away
classical singularities. As an example, compare Newtonian mechanics with
ordinary quantum mechanics. For the former, singularities for point mass
particles interacting via a l/r2 force can develop either because the particles
collide in a way that solutions to Newton's equations of motion cannot be
continued, or else because the particles, without colliding, all disappear to
spatial infinity in a finite amount of time (see Earman 1986). In the quantum
mechanical treatment of this problem the Hamiltonian operator is (essentially)
self-adjoint so that the time evolution operator, which is the exponentiation
of the Hamiltonian, is unitary and is defined for all time — 00 < t < + oo—the
solution never breaks down!
More substance is given to this glimmering by two further considerations.
The first concerns the treatment of the big bang singularity in the semiclassical
approach to quantum gravity where there is no attempt to quantize the metric
itself but where the effect of quantum fields on the spacetime geometry is
calculated by computing the expectation value (Taby of the (renormalized)
energy—momentum tensor of the fields and inserting the result back into EFE.
In a generic early universe one would expect the spacetime geometry to be
inhomogeneous and anisotropic. In some cases these lumpy conditions lead
to particle creation, and when the back reaction on the metric is calculated
the effect can be to remove the big bang as a genuine singularity. But
depending on the details, the effect can also be to leave the big bang
singularity while removing the particle horizons associated with it (see
Anderson 1983, 1984 and also chapter 5 below).
The second consideration starts from the observation that the applicability
of all of the aforementioned singularity theorems relies on various energy
conditions on classical fields and that even the weak energy condition can be
violated for quantum fields. (This condition requires that TabV"Vb > 0 for
all timelike V, which implies that the energy density as measured by any
observer is non-negative.) A more positive note is sounded by Parker and
Fulling (1973) who studied a closed FRW model filled with a neutral scalar
quantum field possessing mass. They showed that in this regime coherent
quantum states can give rise to negative pressures sufficiently large that (Taby
violates the weak energy condition. As a result, instead of collapsing to a
58 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

singularity, the universe 'bounces' and reexpands. However, it should be noted

that in some circumstances averaged or integrated versions of the classical
energy conditions do hold for quantum fields34 and that these averaged energy
conditions may suffice for proving singularity results (see Tipler 1978).
So far I have been speaking of spacetime singularities in the sense relevant
to the standard singularity theorems, that is, geodesic incompleteness or some
generalized kind of incompleteness such as g.a.p. incompleteness. Naked
singularities in the sense of violations of cosmic censorship (see section 2.6)
may or may not involve incompleteness and/or curvature blowup. This raises
the question of whether there is any hope that quantum gravity will help to
censor such naked singularities. In chapter 6 I will mention results that
indicate that quantum gravity may help to prevent the kind of naked
singularities involved in the manufacture of closed timelike curves. But the
mechanism of censorsip involves the divergence of < Tab) and an attendant
curvature blowup. On the other side of the ledger it should be noted that
quantum considerations can make things much worse with respect to naked
singularities. Suppose, in accord with one popular version of the cosmic
censorship conjecture, that the only singularities that develop in physically
reasonable cases of gravitational collapse in classical general relativity are
those that are safely hidden inside of black holes (see chapter 3 for details).
Hawking radiation, a quantum field theoretic effect, leads to the eventual
evaporation of the black hole. And insofar as classical spacetime geometry
can be used to describe the processes, a naked singularity will result from the
evaporation (see chapter 3). Of course, if quantum gravity manages to avoid
singularities, then what will be visible to observers when a black hole
evaporates is not literally a spacetime singularity but a region where quantum
effects dominate. The attendant quantum uncertainties associated with such
effects may be just as disruptive to predictability and determinism as are the
naked singularities of classical GTR.35
Such intimations about the relation between quantum effects and
spacetime singularities are solid in the sense that they are based on well-
established techniques of semiclassical quantum gravity. But precisely because
of their semiclassical character they may not provide reliable indications of
what happens if, as many theorists believe, the spacetime metric itself must
be quantized. There are various speculations about what will happen, the
most widely publicized being Hawking's (1988) proposal for a "no boundary
condition" for quantum cosmology, a condition that banishes the initial
singularity implied by classical general relativity for big bang universes. At
present, however, there are no clear experimental or theoretical guidelines
for evaluating such speculations.36

2.10 Conclusion

The above probing of the idea of spacetime singularities has revealed four
distinct root concepts: singularities as involving a breakdown of the metric
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 59

because, for example, of a blowup of the curvature at a finite distance;

singularities as involving geodesic incompleteness or some more general sense
of incompleteness such as generalized affine parameter incompleteness; singu-
larities as involving missing points; and singularities as involving violation of
cosmic censorship (naked singularities). The above discussion also helped to
reveal that within each of the first three categories there is a sizable family
ofsubcategories. The discussion in chapter 3 will show that 'naked singularity'
is a name for an extended family of notions. In short, spacetime singularities
exhibit a richness and complexity unimagined by the early pioneers of GTR.
The more-or-less official definition of singularities adopted in the physics
literature is in terms of incompleteness. This choice, quite frankly, seems to
have been guided by expediency: this is the sense that most easily lends itself
to proofs of the existence of singularities.
Our knowledge of the prevalence of singularities among the solutions of
EFE is rudimentary but growing. The received wisdom is that the Hawking-
Penrose theorem and allied results establish that singularities in the sense of
geodesic incompleteness are endemic in general relativistic spacetimes. That
conclusion is warranted if essential singularities are defined in terms of
extensions whose metrics are at least C2~. Whether the conclusion holds up
under substantially weaker differentiability requirements is unknown. There
are some results connecting geodesic incompleteness to singularities in some
more intuitive sense such as curvature blowup; but the connection is still not
well understood in general. And as chapter 3 will show, there are few
informative results about the occurrence and non-occurrence of naked
singularities.
Because of the absence of a technique that yields intuitively satisfactory
results for attaching singular points to the spacetime manifold, it is perhaps best
to drop talk of spacetime singularities—which suggests localizable objects—in
favor of talk about singular spacetimes—which does not carry any such sugges-
tion. However, talk about spacetime singularities is too well entrenched to fight.
In many ways we are still ignorant about spacetime singularities in
classical GTR. And yet compared to what was known only a few decades
ago, we know quite a lot about singularities. We certainly know enough to
begin investigating the implications of singularities for classical general
relativistic physics and spacetime philosophy. This will be the task undertaken
in the coming chapters. In advance let me lay some of my cards on the table.
If asked to say briefly what is wrong with a singular spacetime, my short
answer would be: nothing per se. Contrary to Einstein, I do not think the
fact that GTR predicts spacetime singularities is necessarily a cause for alarm,
and I certainly not think the prediction of singularities is a signal that the
theory self-destructs. But there are singularities and there are singularities.
Some types are associated with acausal features and/or a breakdown in
determinism. So while spacetime singularities per se may not be objectionable,
some of their attendant features do pose troublesome questions for physics.
These questions deserve detailed attention. And this book aims to provide it.
60 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Notes

1. Szekeres noted that Synge (1950) had proposed a definition of spacetime

singularities but that this definition could hardly be regarded as satisfactory since it
depended on the choice of coordinate system.
2. A survey of the state of the art circa 1970 is given in Geroch (1970a). Tipler
et al. (1980) summarized the situation to that point. I am not aware of any comparable
review article that brings things up to the present. But the book by Clarke (1993)
contains an authoritative statement of our present understanding of many technical
issues about singularities.
3. It will be assumed throughout that M is a C°° manifold. This means that the
transformations between two charts or coordinate systems is infinitely continuously
differentiable; see Wald (1984a, Ch. 2) for details. If one starts with a Ck (k > 1)
manifold, where the coordinate transformations are only Ck for some k < co, one can
always produce a C°° atlas by winnowing down the coordinate charts in the original
atlas. A Lorentz metric gab is a non-degenerate tensor field of type (0, 2) and signature
n — 1 where n — dim(Af). (The signature of gab is the number of positive eigenvalues
°f Sab-) I n tne main intended case of n = 4 (three space dimensions and one time
dimension), the signature convention (+ + -\—) will be used. The metric is C* just
in case partial derivatives of order k exist and are continuous. The metric is Ck~ just
in case partial derivatives of order k exist and are locally bounded. Later in the
discussion more complicated continuity/differentiability conditions on the metric will
be discussed.
4. A tensor of type (m, n) is one with m contravariant and n covariant indices.
The outer product of a tensor T of type (m, n) and a tensor T of type (m', n') is
a tensor T® T' of type (m + m',n + n') such that T® T'(wt, w2, • • • , wm +m,;
v1, v2,..., v"+"') = TK, . . . , wm; v\ . . . , / ) • 7>m+1, . . . , wm+m.; ,"+ 1, . . . , v"+"'),
where the v's are vectors (aka contravectors) and the w's are dual vectors (aka
covectors). In the text the outer product is indicated simply by juxtaposing the two
tensors, e.g., the outer product of Rab and Rcd is denoted by RabRcd.
5. In this case the metric will not be C2, but the curvature tensor will still make
sense as a distribution; see section 2.7.
6. At each point on j ( v ) , where v is some parametrization of y, the e?(v) form a
basis of the tangent space My(v). gab(y(v})ef(v)eja(v') = 0 iff J=j, and 1 or — 1 if i =j
according as «,•"(») is spacelike or timclike.
7. Given a metricg ab , there is a unique derivative operator Va which is compatible
with the metric in the sense that Vagbc = 0. The action of Va on a vector field Vb is
given by Va Vb = daVb + Ybac Vc where da is the ordinary derivative operator and the
Chnstoffel symbol T is defined by Y\c == (\/2}gbm{dagcm + dcgam - 8mgac}. If T" is
the tangent to a curve y, a vector field F* along y is said to be parallely transported
just in case TaVaVb = 0.
8. Either the physical components of the curvature blowup in all p.p. frames or
they blow up in none, for two such frames are related by a fixed Lorentz transformation.
9. A geodesic may be defined as a curve whose tangent T" is parallely propagated,
i.e., T"Va Tb = 0 (see Wald 1984a, p. 41). A parametrization /I of the geodesic is said
to be affine just in case T" = (5/3A)" satisfies this equation. Alternatively, a geodesic
could be defined as a curve y(v) such that T" = (d/dv)" satisfies T"VaTb = J'(v)Th,
i.e., the tangent vector remains proportional to itself under parallel transport. But it
is always possible to shift to a new parameterization which is affine. The affine
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 61

parameters for a geodesic are unique up to a positive linear transformation A -> aA + b,

a > 0. Thus, the notion of a geodesic of finite (or infinite) affine length is well defined.
For a timelike geodesic, proper time t = J ^/ — ds2 is an affine parameter.
10. This definition is redundant. Consider a curve y: [0, a), [0, a) c IR. A point
P t ?([0> a )) is counted as an endpoint of the curve if every neighborhood ofp intersects
the image y([0, a)) of the curve in M. Such an endpoint that is not on the curve exists
if and only if the curve is extendible in M in the direction away from 7(0). In what
follows I will abuse the distinction between a curve and its image in M.
1 1 . See chapter 4 for examples. If two points of a relativistic spacetime can be
joined by a timelike curve, then the longest such curve (if one exists) is a timelike
geodesic. There is no shortest timelike curve joining the two points since for any s > 0
one can construct a timelike curve which joins the points and which accelerates in a
frantic enough way that its proper length is less than £.
12. The symbol R4 will be used ambiguously to denote the topological space and
the differentiable manifold obtained by adding the standard differentiable structure.
Non-standard differentiable structures for R* will be discussed below in sections 2.6
and 2.7.
13. A diffeomorphism (p: M -» M between the G™ manifolds is a one-one C°°
map. <p* denotes the carry along by (p. For example, the carry along for tangent
vectors (p*: Mp -> M^(p) is defined by the condition (<p*V)(f) = V(f ° <p) for Ve Mp
and /: M -> IR a C°° function. The idea is easily generalized to other geometric objects
(see Wald 1984a, Appendix C). In the case where (p maps M into itself and the action
of (p can be represented as carrying a point p with coordinates x' to a new point (p(p)
with coordinates x"(<p(p)) = x ' ( p ) , the carry along of a geometric object is given by
the rule that in the new coordinates the components of the new object at the new
point are the same as the components of the old object in the old coordinates at the
old point.
14. We will see in the following section, however, that there is a sense in which
the cone singularity admits of local extensions.
15. The relation > is said to be a partial order on the set S iff for any a, b, ce S,
(i) a > a, (ii) a > b and b > c imply a > c, and (iii) a > b and b > a imply that a = b.
T c: S is totally ordered iff > holds between all pairs of elements of T. b e S is said
to be an upper bound for T iff b > a for all a 6 T. Finally, m e S is a maximal element
for S iff for any c € S, c > m implies m = c. ^orn ^s lemma, an equivalent of the axiom
of choice, asserts that if every totally ordered subset of S has an upper bound, then
there is a (not necessarily unique) maximal element. The obvious idea for proving the
existence of a maximal extension would be to take (M',g'ak) > (M,gab] to mean that
there is an isometric imbedding of the latter into the former. Start with a spacetime
M, gab for which we would like to prove the existence of a maximal extension. Take
S to be the set of all spacetimes M' , g'ab such that (M',g'ab) > (M,gab). To prove the
existence of an upper bound for any totally ordered subset 7 of S we can apply the
inductive limit construction in which the union of spacetimes in T is taken and each
spacetime in T is identified with its isometric image in the union (see Hawking and
Ellis 1973, p. 249). Then if Zorn's lemma could be applied, the existence of a (not
necessarily unique) maximal extension of M , gab would be immediate. Unfortunately
the lemma cannot be applied because the relation of isometric imbeddability as defined
above is not necessarily an order relation; in particular, property (iii) can fail: there
are spacetimes such that (M, gab) > (M',g'ab) and (M1 , g'ab) > (M,gab) even though
( M , g a b ] =£ ( M ' , g ' a b ) (i.e., the spacetimes arc not isometric). For example, let M,gab
62 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

be the result of truncating Minkowski spacetime by removing all the points on or

above a scallop-shaped time slice, and let M', g'ab be constructed similarly but using
a differently shaped time slice. The concept of a framed spacetime (see below)
overcomes this difficulty; for the imbedding of M, gal>, F into M',g'ab,F', if it exists,
is unique (see Geroch 1969).
16. That is, hab is of signature n = dim(Af).
17. The terms A-completeness and i-incompleteness derive from the fact that they
arise from Schmidt's (1971) bundle boundary construction; see below.
18. For details the reader is referred to Hawking and Ellis (1973) and Clarke
(1993).
19. Roughly, the natural extension of the exponential map in M to M is required
to be continuous; see Geroch et al. (1982) for details. (Assuming that the metric is
C2, for each p e M and each direction T" e Mp, there is a unique geodesic through p
in the direction T". The exponential map is the map from the tangent space Mp to
M which takes T" E Mp to the point lying at a unit affine distance along the geodesic
which passes through/) in the direction T".} A different characterization of spacetime
singularities, also subject to the difficulty under discussion, was given by Geroch
(1968b).
20. Compare this to the treatment of Hawking and Ellis (1973, pp. 291-292).
21. When dim (M} > 2, the Weyl tensor Cabcd is defined by

C.w = ROM - (2/(« - 2))(galcRd]b - gblcRd]a) - (2/((n - ! ) ( « - 2)))Rgaicgd]b .

22. The following definition is suggested by ideas of Misner (1963) and Ryan
and Shepley (1975), modified by using ^-completeness in place of geodesic incomplete-
ness.
23. Scott (1992) does not claim that her approach captures the missing points
idea; that is my interpretation. What follows is a simplified and somewhat distorted
version of Scott's analysis.
24. Let 7: / —> M, I = [a, b) c R, be a parameterized half-curve, and let M, tp
be an envelopment of M. p e M is said to be a limit point of y (A) just in case there is
an increasing sequence of numbers A; e [a, b), i = 1, 2, 3, . . . , such that A( -» b+ and
( < p ° y ) ( A j ) - » / > . / » is said to be an endpoint just in case (<j» ° y)(A() ->p for every
increasing sequence Aj such that A,- -» b+. I presume that boundary points which are
limit points of half-curves of finite g.a.l. will also be endpoints of such curves. But such
is not always the case for limit points of incomplete geodesies.
25. In other words, the global topological coordinates x,y, z, t for R4 are not
everywhere smooth with respect to the coordinate charts of (Rg.
26. I am grateful to Professor Robert Gompf for clarifying this point. The
Poincare conjecture, which remains unsettled, asserts that a three-manifold is topo-
logically S3 if it is compact and simply connected.
27. Einstein's field equations are partial differential equations involving second-
order derivatives of the metric tensor. The theory of distributions allows these
equations to make sense in some cases where the metric fails to be twice differentiable.
Technically, a distribution is a linear functional on a normed space of smooth
functions. See Choquet-Bruhat et al. (1982) for a treatment of distributions in physics.
28. Choose a volume element dV on M4, say, the usual d*x. The functions J<a are
the weak derivatives of a function f: R" —> R iff for any C°° function I// on R" with
compact support ("test function"), J/ 0 i/' dV' = — J f ( d \ l i j d x a ) dV. /: H." -+ R is locally
 Defining, Characterizing, and Proving the Existence of Spacetime Singularities 63

square integrable just in case J \f\2 dV < co for any compact domain of integration
in the support of the function. These ideas have natural generalizations to arbitrary
K-dimensional manifolds.
29. If /: F8" —> IR is a locally integrable function with respect to the volume
element dV, it defines a distribution Df whose action on a test function i/> is given by
Df(\l/} =\J\j/dV. This idea can be generalized to tensor fields on manifolds; see
Choquet-Bruhat et al. (1982).
30. V^Rj,^/ = 0 involves the ordinary derivative of Rabcd, which will be a
distribution if Rabcd is a distribution, and the product of T"bc and Rabce, which may fail
to be a distribution even if Y"bc and Rabcd are distributions. But if T"bc and Rabcd are
each locally square integrable the product will be locally integrable and, therefore,
will define a distribution.
31. Global existence and uniqueness is connected with the problem of cosmic
censorship; see chapter 3.
32. If V is the normalized (VVa— — 1) tangent field of a geodesic congruence,
then the associated spatial metric is hab=-gab + VaVb. The shear ffab and the rotation
or twist (oab of the congruence are defined by <rat =: V(i, V^ — (1/3) 9hab and a>ab ==• V[(, Va^.
Rotation will be discussed further in chapter 6 in connection with the Godel solution
to EFE.
33. Strictly speaking, 0 -» — oo does not mean that the geodesies actually cross.
The technically correct statement is that 6 -» — oo is the necessary and sufficient
condition for the vanishing of a Jacobi field vf along a geodesic y of the congruence.
v\a can be thought of as pointing from 7 to an "infinitesimally close" geodesic of the
congruence. See Wald (1984, Sec. 9.3).
34. The weak energy condition entails the null energy condition which requires
that TabKaK" > 0 for every null vector K". Wald and Yurtsever (1991) showed that
in two-dimensional curved spacetimes quantum fields obey an averaged version of the
null energy condition. But they also showed that in four-dimensional spacetimes even
this averaged version can fail.
35. As noted by Joshi (1993, p. 265). Just how naked singularities lead to
disruptions of predictability and determinism will be discussed in chapter 3.
36. Another approach to combining general relativity and QM is provided by
string theory. It has been claimed that this approach holds the promise of avoiding
or rendering harmless spacetime singularities; see Kostelecky and Perry (1994).
 3
Cosmic Censorship

It is one of the little ironies of our times that while the layman wasbeing
indoctrinated with the stereotypic image of black holes as the ultimate cookie
monsters, the professionals have been swinging round to the almost directly
opposing view that black holes, like growing old, are really not so bad when you
consider the alternative.
Werner Israel (1986)

3.1 Introduction

The idea of cosmic censorship was introduced over twenty years ago by
Roger Penrose (1969). About a decade later Penrose noted that it was not
then known "whether some form of cosmic censorship principle is actually
a consequence of general relativity." To which he added: "In fact, we may
regard this as possibly the most important unsolved problem of classical
general relativity theory" (Penrose 1978, p. 230). This sentiment has
been echoed by Stephen Hawking (1979, p. 1047), Werner Israel (1984,
p. 1049), Robert Wald (1984a, p. 303), Frank Tipler (1985, p. 499),
Douglas Eardley (1987, p. 229), Stuart Shapiro and Saul Teukolsky (1991b,
p. 330), Pankay Joshi (1993, p. 204), and many others. Thus, if an "important
problem" in physics is one which is deemed to be important by leading
research workers in the field, then the problem of cosmic censorship is
undoubtedly near the top of the list for classical GTR. One of my goals here
is to show why it is important in a more substantive sense. I also want to
indicate why it is that, despite the intense effort that has been devoted to this
problem, it remains unsolved. Indeed, the very statement of the problem
remains open to debate.
A study of this topic can lead to payoffs in several areas of philosophy
of science, two of which I will mention and one of which I will actually
pursue. In my Primer on Determinism (Earman 1986) I attempted to deflate
the popular image of determinism as unproblematically at work outside of
the non-quantum domain. My message fell largely on deaf ears. But the
64
 Cosmic Censorship 65

failure of cosmic censorship could well herald a breakdown in classical

predictability and determinism of such proportions that it cannot be ignored.1
Second, the growing band of philosophers of science who are turning towards
an increasingly sociological stance will find the history of the cosmic censorship
hypothesis a fascinating case study in the dynamics of a research program.
Particularly interesting to me is how, in a subject with few hard results, the
intuitions and pronouncements of a small number of people have shaped and
directed the research. I leave the investigation of such matters to more capable
hands.

3.2 Cozying up to singularities

As was seen in chapter 2, prior to the 1960s spacetime singularities were

regarded as a minor embarrassment for GTR. They constituted an embar-
rassment because it was thought by Einstein and others that a true singularity,
a singularity in the very fabric of spacetime itself, was an absurdity. But the
embarrassment seemed to be a minor one that could be swept under the rug;
for the then known models of GTR containing singularities all embodied very
special and physically unrealistic features. Two developments forced a major
shift in attitude. First, the observation of the cosmic low temperature
blackbody radiation lent credence to the notion that our universe originated
in a big bang singularity. Second, and even more importantly, a series of
theorems due principally to Stephen Hawking and Roger Penrose indicated
that, according to GTR, singularities cannot be relegated to the distant past
because under quite general conditions they can be expected to occur both
in cosmology and in the gravitational collapse of stars (see chapter 2). Thus,
singularities cannot be swept under the rug; they are, so to speak, woven into
the pattern of the rug. Of course, these theorems might have been taken as
turning what was initially only a minor embarrassment into a major scandal.
Instead, what occurred in some quarters was a 180° reorientation in point of
view: singularities were no longer to be relegated to obscurity; rather, they
were to be recognized as a central feature of the GTR, a feature which called
attention to a new aspect of reality that was neglected in all previous physical
theories, Newtonian and special relativistic alike. And thus we can hope to
get definitive confirmation of GTR by confirming the presence of these new
objects.
But before getting carried away with this newfound enthusiasm for
singularities, we should pause to contemplate a potential disaster. If the
singularities that occur in Nature are naked, then chaos would seem to
threaten. Since the spacetime structure breaks down at singularities and since
(pace Kant) physical laws presuppose space and time, it would seem that
these naked singularities are sources of lawlessness. The worry is illustrated
in Fig. 3.1 where all sorts of nasty things—TV sets showing Nixon's "Checkers
66 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 3.1 A naked singularity disgorges

speech," green slime, Japanese horror movie monsters, etc.—emerge helter-

skelter from the singularity.
The point can be put more formally in terms of the breakdown in
predictability and determinism. If E is an achronal spacelike surface of a
spacedme M,gab the future (respectively, past) domain of dependence Z) + (E)
(D~ (E)) of E is defined to be the collection of all points/) e M such that every
causal curve which passes through p and which has no past (respec-
tively, future) endpoint meets E. If p £ D+ (E) (respectively, p $ D~ (E)) then
it would seem that no amount of initial data on E will suffice for a sure
prediction (respectively, retrodiction) of events at p since there are possible
causal influences which can affect events at p but which do not register
onE. 2
To illustrate how naked singularities can lead to a breakdown in
predictability and determinism, start with Minkowski spacedme R4, t]ab and
consider the spacelike hypersurface E corresponding to the level surface
t = 0 of some inertial time coordinate t. D+ (E) encompasses the entire future
of S. Perform the now familiar trick of removing from R 4 a closed ball K on
the future side of E. The resulting spacedme has a naked singularity, the
presence of which excludes the shaded region of Fig. 3.2 from Z) + (E). The
future boundary of Z) + (E), called the future Cauchy horizon of E, is labeled
as // + (E). 3 This naked singularity is rather trivial since it can be removed
by extending the surgically mutilated spacedme back to full Minkowski
spacedme. To make the example less trivial, one can choose a scalar field Q
that goes rapidly to zero as the missing region K is approached. The new
conformally related spacedme Af, jaj,, where M — IR4 — K and gab = Q2r]ab,
is properly inextendible and so its naked singularity is irremovable.
Cosmic censorship is an attempt to have one's cake and eat it too. The
idea is that we can cozy up to singularities without fear of being infected by
the ghastly pathologies of naked singularities since GTR implies that, under
reasonable conditions, Nature exercises modesty and presents us only with
 Cosmic Censorship 67

Fig. 3.2 How to create a naked singularity

singularities that have been clothed in some appropriate sense. The task of
the following section is to try to understand in more precise terms what this
means.
Before turning to this task, I should emphasize that while the approaches
to cosmic censorship I will explore are motivated largely by concerns with
predictability and determinism, there are a number of other reasons why
physicists would like to believe that a cosmic censor is at work. I will mention
three. First, if a suitable form of cosmic censorship obtains, then one can
appeal to various "no hair" theorems for black holes to obtain a characteriza-
tion of the final state of a gravitationally collapsed body in terms of a member
of the two-parameter (mass, angular momentum) family of Kerr solutions.
Second, the now standard black hole thermodynamics makes use of Hawking's
area theorems, which in turn presuppose a form of cosmic censorship. And
third, the proof of the positivity of the total mass of an isolated gravitating
system (the Arnowitt-Deser-Misner (ADM) mass) presupposes the absence
of singularities on the initial time slice; such an absence can be viewed as
a kind of cosmic censorship. These matters will not be discussed here;
the interested reader may consult standard texts on GTR such as Wald
(1984a).

3.3 Naked singularities and cosmic censorship

When Penrose first raised the problem of cosmic censorship it was not clear
what to include under the notion of a naked singularity. For example, Penrose
said "In one sense, a 'cosmic censor' can be shown not to exist. For . . . the
'big bang' singularity is, in principle, observable" (Penrose 1969, p. 274).
Today standard big bang cosmologies would not be regarded as nakedly
singular and, thus, would not be regarded as being in conflict with cosmic
68 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

censorship. In what follows I will describe five attempts to pinpoint the

correlative notions of cosmic censorship and naked singularities. The first
approach seeks to supply necessary and sufficient conditions for cosmic
censorship to hold; a naked singularity is then characterized indirectly as a
condition that produces the violation of cosmic censorship. Other approaches
proceed the other way round; they first attempt to give a direct definition of
a naked singularity, and then subsequently define cosmic censorship in terms
of the absence of naked singularities. The reader who dips into the scientific
literature will find a seeming disrespect for the distinction between "What is
cosmic censorship?" and "Is cosmic censorship true?" Formulations of cosmic
censorship are often couched so as to maneuver around counterexamples.
This is not necessarily a dishonest practice. Nor is it a practice unfamiliar to
philosophers of science. Trying to provide a precise explication of a vague
concept calls for decisions about how to draw the line among borderline cases.
Since there are many plausible ways of scribing, it is not inappropriate for
ulterior motivations—such as the desire to establish some form of cosmic
censorship—to guide the pen. Cases of outright gerrymandering will be
flagged.

Approach 1

Since a key concern with the development of naked singularities is the

breakdown of predictability and determinism, cosmic censorship may be
formulated by imposing conditions that assure that no such unwanted
behavior occurs.
For future reference it is useful to begin with two definitions. A time slice
of the spacetime M, gab is a spacelike hypersurface E c: M without edges. Such
a E is the relativistic analogue of a Newtonian "constant time" slice. (Not
every relativistic spacetime admits time slices; see chapter 6.) A time slice is
said to be a partial Cauchy surface if it is achronal (i.e., not intersected more
than once by any future-directed timelike curve). 4 A partial Cauchy surface
is the appropriate platform on which to specify instantaneous initial data that,
one may hope, will allow the future to be predicted and the past retrodicted. To
help assure success in this regard, we can require that E is a Cauchy surface
for M,gab in that the total domain of dependence -D(E) — Z>~(E) u £ > + ( E )
of E is all of M. (This is equivalent to the definition given in section 2.6.)
Alternatively, one could say that E is a future (respectively, past) Cauchy
surface for M, gah just in case E partitions M and D+ (E) (respectively, D~ (E))
contains all of that part of M that lies to the future (respectively, past) of E.
E is a Cauchy surface just in case it is both past and future Cauchy.
The present approach leads to a strong and to a weak version of
censorship. Strong cosmic censorship (SCC) holds for M, gab just in case
M, gah possesses a Cauchy surface. Of course, it is assumed that M, gah is
maximal, otherwise it would not represent a physically reasonable model
(recall Fig. 2.5). The standard big bang cosmological models are maximal
 Cosmic Censorship 69

Fig. 3.3 (a) The behavior of light cones in anti-de Sitter spacetime; (b) Penrose
conformal diagram of anti-de Sitter

globally hyperbolic so they satisfy the present statement of SCG, and thus
the big bang singularity is not counted as naked.
Global hyperbolicity or the existence of a Cauchy surface is a very strong
condition, and one can wonder whether it is too heavy-handed a way to
capture the intuitive idea that there are no naked singularities. A relevant
example is provided by the universal covering of the anti-de Sitter spacetime,
a typical light cone of which is represented in Fig. 3.3a and whose Penrose
diagram is given in Fig. 3.3b.5 This spacetime violates the proposed formula-
tion of SCC but it is arguably singularity free, e.g., it is geodesically complete.
The proponent of the present approach could concede the point while
maintaining that since the key worry raised by naked singularities is the
breakdown in predictability and determinism, no harm is done by formulating
a version of cosmic censorship that is strong enough to assuage the worry
70 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

whatever the source, naked singularities or no. Even if we share this sentiment,
there is still an evident need for a separate definition of naked singularity.
The second approach described below takes up this challenge.
We also want to be able to distinguish breakdowns in predictability
that aren't too drastic. In particular, Nature may practice a form of modesty
by hiding singularities in "black holes," the exterior regions of which may be
predictable because they admit future Cauchy surfaces. If so, weak cosmic
censorship (WCC) is said to hold. In asymptotically flat spacetimes this idea
can be made precise by appealing to the notion of future null infinity J*', the
terminus of outgoing null geodesies that escape to spatial infinity. 6 The
interior B of a black hole is then defined to be the complement of the causal
past J~(-/ + ) of «/ + , i.e., the events that cannot be seen from future null
infinity. The boundary E of B is the absolute event horizon; this is the modesty
curtain that hides the goings-on in the interior of the black hole from external
observers. These concepts are illustrated in Kruskal spacetime (the maximal
analytic extension of Schwarzschild spacetime as shown in Fig. 3.4a) and its
Penrose diagram (Fig. 3.4b).7 Frank Tipler (1979) has proposed a more
general definition of a black hole that does not assume asymptotic flatness
and is supposed to apply to any stably causal spacetime.8 Until this or some
substitute definition is shown to be satisfactory, we cannot claim to have a
general formulation of WCC in the cosmological setting.
The Kruskal-Schwarzschild spacetime is not only an example of a black
hole and WCC but it also displays SCC since it possesses a Cauchy surface.
The difference between the strong and weak versions of cosmic censorship is
illustrated schematically in Fig. 3.5. In Fig. 3.5a the singularity that develops
in gravitational collapse is hidden from external observers but is visible to
observers within the black hole. In Fig. 3.5b the black hole is even blacker
because even those unfortunate observers who fall into the hole cannot 'see'
the singularity, though they may well feel and, indeed, may be torn apart by
the tidal forces as they are inevitably sucked into the singularity. Nevertheless,
they may take some cold comfort in the fact that SCC holds.
If SCC holds there is a sense in which, paradoxically, the singularity
never occurs. It follows from a result of Geroch (1970b) that if M, gab admits
a Cauchy surface then it also admits a global time function, a C1 map
t: M -» IR which has a timelike gradient and which therefore increases along
every future-directed timelike curve. And further, t can be chosen so that each
t = constant surface is Cauchy. Since no Cauchy surface can intersect the
singularity, we can conclude that there is no time t at which the singularity
exists. (Exercise for the reader: draw a foliation of Cauchy surfaces for the
spacetime pictured in Fig. 3.5b.) Of course, the statement that no singularity
exists will be hotly disputed by the ghosts of the observers who have been
sucked into the black hole and have been snuffed out after only a finite
existence (as measured by their respective proper times).
It is instructive to take the time reverses of the processes pictured in Fig.
3.5 to produce "white holes" where the singularities explode into expanding
 Cosmic Censorship 71

Fig. 3.4 (a) Kruskal-Schwarzschild spacetime; (b) conformal diagram of Kruskal-

Schwarschild spacetime

ordinary matter (Fig. 3.6). At first glance the spacetimes of Fig. 3.6 would
seem to possess naked singularities par excellence. But since the spacetime of
Fig. 3.5b possess a Cauchy surface, so does the spacetime of Fig. 3.6b. In this
sense the singularity in Fig. 3.6b is not naked even though it is highly visible
(as is the case with the big bang singularity of the Friedmann—Robertson-
Walker models).
But does predictability really hold in the situation pictured in Fig. 3.6b?
Penrose has argued for a negative answer:

The future behavior of such a white hole does not, in any sensible way, seem
to be determined by its past. In particular, the precise moment at which the
white hole explodes into ordinary matter seems to be entirely of its own
'choosing', being unpredicatble by the use of the normal laws of physics.
(Penrose 1979, p. 601)

Penrose's point seems to be this. The spacetime in Fig. 3.6b can be foliated
by Cauchy surfaces. But the singularity lies to the past of any such surface,
72 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 3.5 Black holes

Fig. 3.6 White holes

 Cosmic Censorship 73

which means that any such surface must intersect the ordinary matter. So
the explosion cannot properly be said to be predicted from any such surface.
There are other spacelike surfaces from which one can properly speak of
predicting the explosion since it lies to their future. But since these surfaces
do not have the Cauchy property, the prediction cannot be of the deterministic
variety. If Penrose's worry is to be taken seriously and if it is to be assuaged
by some form of cosmic censorship, then an even stronger version of SCC
than the one given above is needed. This way seems to end in disaster for
cosmic censorship, so I will not pursue the matter.
Figures 3.5 and 3.6 also raise another problem, not so much for the
statement of cosmic censorship as for the validity and proof of the hypothesis
that GTR contains a mechanism for enforcing cosmic censorship. Suppose
that in typical cases of gravitational collapse SCC fails while WCC holds, i.e.,
the process in Fig. 3.5a is what we should expect. Then since Einstein's field
equations are time reversal invariant, every solution of type 3.5a is matched
by a solution of type 3.6a. So if black holes that violate SCC but satisfy WCC
are a pervasive feature of general relativistic models, it would seem that white
holes that violate WCC are also a pervasive feature.
One can take the attitude that what is needed here is a division of labor.
The initial effort should be devoted to proving (or refuting) the conjecture
that naked singularities do not occur in reasonable models of gravitational
collapse. Then attention can be turned to the problem of white holes, which
may be regarded as an aspect of the more general problem of time's
asymmetries. In other branches of physics— electromagnetism and mechanics,
for example—the fundamental laws are also time reversal invariant. But we
find that while certain types of solutions are commonly encountered, their
time-reversed counterparts never or very seldomly occur (e.g., we often
encounter electromagnetic waves expanding from a center to spatial infinity
but we never encounter waves converging on a center). So it is hardly
unexpected that an analogous situation occurs in gravitational physics. The
fact that the origin of time's arrow remains an unsolved problem is nothing
to boast about, but it is not a special problem for gravitational physics and
it should not prevent work from going ahead on the issue of cosmic
censorship.9
Although the present approach to cosmic censorship has yielded some
valuable insights, it is subject to some serious shortcomings. Most obviously,
although it provides sufficient conditions (disregarding Penrose's worry) for
ruling out naked singularities, it hasn't told us directly what a naked
singularity is, nor has it told us how the violation of cosmic censorship leads
to singularities in some intuitive sense.
Work by Joshi and Saraykar (1987) provides some information on the
latter issue. We know from results discussed earlier that SCC in the guise of
the existence of a Cauchy surface Z implies that the topology of space does
not change with time in the sense that the spacetimc manifold is diffeo-
morphically Z x R. So it is natural to ask: If SCC is violated and there is a
74 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

change in topology, what can we expect about the existence of singularities?

To make this question more precise, define a partial Gauchy surface S of
M, gab to be maximal just in case /)(£) is maximal in the set of Z)(E') for all
partial Cauchy surfaces X' for M, gab. If Z and H' are both maximal,
£' c /+ (£), and 17 ^ Z then a topology change is said to take place.10 Since
£ cannot be a Cauchy surface, H+ (S) cannot be empty. Joshi and Saraykar
show that if in addition the weak energy condition is satisfied and if all timelike
trajectories encounter some non-zero matter or energy, then a singularity will
occur in that a null generator of// + (Z) will be past incomplete.11
Nevertheless, it would be nice to have a direct definition of 'naked
singularity' entailing that singularities cannot exist if there is a Cauchy
surface. This would confirm that the approach explored above is on the right
track.

Approach 2
The second approach seeks to define a naked singularity in terms of its
detectability. Cosmic censorship then becomes the statement that such
singularities do not occur.
Penrose's (1974, 1978, 1979) version of this approach emphasizes local
detectability.
It seems to me to be quite unreasonable to suppose that the physics in a
comparatively local region of spacetime should really 'care' whether a light
ray setting out from a singularity should ultimately escape to 'infinity' or
not. To put things another way, some observer . . . might intercept the light
ray and see the singularity as 'naked', though he be not actually situated at
infinity. . . . The unpredictability entailed by the presence of naked singular-
ities which is so abhorrent to many people would be present just as much
for this local observer . . . as for an observer at infinity.
It seems to me to be comparatively unimportant whether the observer
himself can escape to infinity. Classical general relativity is a scale-
independent theory, so if locally naked singularities can occur on a very tiny
scale, they should also, in principle, occur on a very large scale. . . .
It would seem, therefore, that if cosmic censorship is a principle
of Nature, it should be formulated in such a way as to preclude such locally
naked singularities. (Penrose 1979, pp. 618-619)
Penrose's technical explication of the notion of a locally naked singularity
uses the concepts of TIFs and TIPs which I do not want to introduce here,12
so I will follow a related idea used by Geroch and Horowitz (1979). The form
of a definition for the set 91 of points from which the spacetime M, gab can
be detected to be singular can be stated as follows:

DEFINITION 3.1
yt = {p 6 M: I~ (p) contains a timelike curve y which has no future endpoint
and which }
 Cosmic Censorship 75

The reason, intuitively speaking, that y has no future endpoint is that it runs
into a singularity. Since this fact is directly detectable from p, the spacetime
is nakedly singular as viewed from p. Thus, the statement that M, gab harbors
naked singularities becomes: 91 ^ 0; and conversely, cosmic censorship
becomes the statement: 91 = 0.13
The strongest version of cosmic censorship is obtained by putting no
further restrictions in the blank. Then 91 = 0 is equivalent to the existence
of a Cauchy surface, a result that dovetails nicely with Approach 1. If we are
somewhat more restrictive and fill the blank with "is a geodesic," then 91 = 0
no longer entails the existence of a Cauchy surface. Depending upon one's
point of view, it could be counted as a benefit of this latter version of cosmic
censorship that anti-de Sitter spacetime is no longer counted as nakedly
singular. (Referring back to the rendering of the null cone behavior in anti-de
Sitter spacetime in Fig. 3.3a, consider an arbitrary spacetime point p. A
timelike geodesic that starts at a point q 6 /~ (p) will eventually escape I~ (p)
if it is extended far enough into the future.)
If one wants the focus of cosmic censorship to be curvature singularities,
then the blank should be filled with restrictions that guarantee y terminates
on what one chooses to regard as a curvature singularity. Other fillings would
be appropriate depending upon what kinds of naked singularities one wants
to target. In this way we obtain several versions of cosmic censorship.
If one does not share Penrose's sentiments about local detectability, the
present approach can still be adapted to a weaker statement of cosmic
censorship by requiring only that 91 = 0 for the region exterior to black
holes.

Approach 3
As part of establishing cosmic censorship, one would like to prove that in
reasonable models of gravitational collapse naked singularities do not develop
from regular initial data. Thus, on this approach one would not regard the
negative mass Schwarzschild model14 as a counterexample to cosmic censorship
since, although it is nakedly singular, the singularity has been around for all
time.
Before we can start on this project we need a definition to isolate naked
singularities that can be said to develop from regular initial data, as illustrated
schematically in Fig. 3.7. The ideas of Geroch and Horowitz (1979) and
Horowitz (1979) suggest:

DEFINITION 3.2
A spacetime M,gab is future nakedly singular in the first sense (FNSj) with
respect to the partial Cauchy surface E c: M just in case there is ap e // + (X)
such that /~ (/>) n S is compact.

Newman (1984b) works with a somewhat stronger criterion:

76 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 3.7 The formation of a naked singularity from regular initial data

Fig. 3.8 A poor choice of initial value hypersurface

DEFINITION 3.3
A spacetime M, gab is future nakedly singular in the second sense (FNS 2 ) with
respect to the partial Gauchy surface 2 c M just in case there is a. p 6 // + (Z)
such that I" (p) n Z is compact and a null geodesic generator of H+ (Z)
through /> is past incomplete.

The implicit assumption of this approach is that all of the spacetimes

under discussion are maximally extended; for if M, gab were, say, the
maximal Cauchy development of initial data on Z, then it would auto-
matically be globally hyperbolic and H+ (Z) would be empty. The condition
that I~ (p) n £ is compact assures that Minkowski spacetime is not counted
as being FNS t or FNS 2 with respect to the Z illustrated in Fig. 3.8
(compare to Fig. 3.7). Flowever, as will be seen below, this condition has
some other not-so-nice consequences. Anti-de Sitter spacetime (Fig. 3.3) is
 Cosmic Censorship 77

not future nakedly singular in either sense FNSj or FNS2. The two-
dimensional Misner spacetime of Fig. 2.3, illustrating some of the causal
features of Taub—NUT spacetime, is obviously FNSj with respect to E. (The
surface labeled jV is in fact H+(L). For any p e I~ (E), E <= /"(/>); and since
E is compact, /~(E) n E is compact.) It is not FNS2; indeed, if// + (E) is
compact for a partial Cauchy surface E, then the null geodesic generators of
E are past complete; see Hawking and Ellis (1973, Lemma 8.5.5).
Definition 3.2 is more appropriate for capturing the notion of a breakdown
in predictability or determinism due to whatever cause, whereas Def. 3.3 is
more relevant to identifying the development of naked singularities of the
type that are indicated by geodesic incompleteness. But as noted in chapter 2,
geodesic incompleteness is not a reliable indicator of a curvature singularity.
So if it is curvature singularities that are to be ruled out by the statement of
cosmic censorship, a further definition is needed which strengthens Def. 3.3
by requiring that, in some appropriate sense, a generator of H+ (E) through/;
encounters a curvature singularity of some specified type (e.g., scalar poly-
nomial blowup) in the past direction.
Before proceeding further one should make sure that a future nakedly
singular spacetime as defined above really is singular in some minimal sense.
This would mean showing that if M, gab is future nakedly singular with respect
to E, then the region of M that is on the future side of E and that is causally
accessible from E is not globally hyperbolic. Suppose for purposes of
contradiction that J+ (E) is globally hyperbolic. Then on either Definition
3.2 or 3.3 there is a point />e// + (E) such that I~(p) n E is compact. So
J~(p) n J+(Tl]>) n E) is a compact set (Hawking and Ellis 1973, Cor. to
Prop. 6.6.1). Since E has no edge, the generator a of H+ (E) through p is past
endless. And since a e (J~ (p) n J+ (I~ (p) n E)) we have a past endless null
curve imprisoned in a compact set. Thus J+ (E) is not strongly causal and,
a fortiori, is not globally hyperbolic (Hawking and Ellis 1973, Prop. 6.4.7). 15
Conversely, can we be sure that if M, gab is not future nakedly singular
with respect to E then J+ (E) is globally hyperbolic? The answer is yes if E is
compact and "future nakedly singular" is taken as FNSj. Indeed, E itself
must be a Cauchy surface. If E were not a Cauchy surface // + (E) would be
non-empty. Since E is compact, /"(/>) n E is compact for any p e // + (E), so
that the spacetime is FNSj with respect to E. On the other hand, this result
does not hold if E is not compact, as we already know from the case of anti-de
Sitter spacetime. But an even worse counterexample is provided by Reissner—
Nordstrom spacetime (a piece of whose Penrose diagram is shown in Fig. 3.9)
since this model possesses a naked curvature singularity to the future of a
partial Cauchy surface. One could argue, however, that the very feature
which allows this spacetime to count as neither FNSj nor FNS2 makes it
physically unreasonable (see Wald 1984a, p. 304). Namely, for any partial
Cauchy surface E lying below the singularity and for any p e H+ (E), I~ (p)
contains an infinite (i.e., non-compact) portion of E. As a result a small per-
turbation on E can accumulate on // + (E) to produce an "infinite blue-shift"
78 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 3.9 A conformal diagram of part of Reissner-Nordstrom spacetime

singularity (Chandrasekar and Hartle 1982). This line of reasoning, however,

has the defect of failing to respect a division of labor between framing a defini-
tion of what it is for a spacetime to be future nakedly singular and distinguish-
ing between physically reasonable versus physically unreasonable violations of
cosmic censorship. I warned in the outset that the division can become blurry
in borderline cases; but the present case seems to lie far from the border.
If we do want to try to respect the division of labor, we are faced with
a version of the by now familiar tension. On the one hand we can try to
target various forms of singular behavior that in some sufficiently tight sense
are traceable to the development of initial data from 2; that is the course
taken in the above definitions. On the other hand we can try to frame the
definition of 'future nakedly singular with respect to 2' so that being not
future nakedly singular with respect to 2 entails that jf+ (2) is free of any
taint of singularity. But the only way to assure the latter is to require that
J + (2) is globally hyperbolic, and in turn the only sure way to guarantee that
the development of initial data from 2 makes J+ (2) globally hyperbolic is
to require that Z) + (2) includes J+(£), which is to say that 2 is a future
Cauchy surface and which is to collapse back to Approach 1.

Approach 4
Approach 3 was supposed to focus on how naked singularities might develop
from regular initial data. But in fact it ignored the details of the initial value
problem in GTR and implicitly assumed that the spacetimes at issue were
maximal without inquiring whether they got that way by evolution of initial
data or not. To remedy this defect, at least for an important subclass of cases,
 Cosmic Censorship 79

consider the initial value problem for Einstein's vacuum field equations. A
data set consists of a three-manifold Z without boundary, to be realized as a
spacelike hypersurface in the evolved spacetime; the first fundamental form
hab of Z, a Riemann metric that characterizes the intrinsic spatial geometry
of Z; and the second fundamental form Kab which is (roughly) the time
derivative of hab. To be consistent with EFE this data must satisfy a set of
four coupled partial differential equations, called the constraint equations.16
For any such data set, there exists a unique (up to isometry) spacetime M, gab
which is a development of the initial data set,17 which has Z as a Cauchy
surface, and which is inextendible in any way that keeps Z a Cauchy surface
(Choquet-Bruhat and Geroch 1969).
One might be tempted to say that if this M, gab is not maximal simpliciter
then cosmic censorship is violated because an observer crossing over H+ (Z)
into the region M — M of an extension M, gab ofM, gab will enter terra incognita
where the spacetime geometry is not determined by the initial data on Z.
However, the non-maximality of M, gab may not signal a lapse on the part
of the cosmic censor but only a poor choice of the initial value hypersurface.
For example, Z might be chosen to be an open spacelike disk of Minkowski
spacetime. This example can be ruled out by requiring that Z be "large
enough"; in particular, Z, hab should be a complete Riemann space (see
chapter 2). But the requirement of completeness is not strong enough
to guarantee that Z is a good choice in the appropriate sense. Consider the
spacelike hyperboloid Z of Minkowski spacetime as illustrated in Fig. 3.8.
With the space metric hab induced by the Minkowski metric i]ab, the Riemann
space Z, hab is complete, but the maximal development for which Z is a
Cauchy surface is extendible. This example can be ruled out by the further
requirement that when Z is non-compact, Z, hab should be asymptotically
flat. But this further requirement is too strong for a general formulation of
cosmic censorship since the censor may have to work in cosmologies that are
neither spatially closed nor asymptotically flat.
These difficulties do not arise in the spatially closed case. For if Z is
compact, then Z, hab must be complete. And, as already noted above, in a
globally hyperbolic spacetime admitting a compact Cauchy surface, any
compact time slice is a Cauchy surface (see Budic et al. 1978).

Approach 5

It is now time to question the obsession with Cauchy surfaces. To make the
point it suffices to continue with the special case of vacuum solutions. Let
Z, hab, Kab be an initial data set. And to sidestep the difficulties encountered
in Approach 4 we may further specialize to the case where Z is compact or
else Z, hab is complete and asymptotically flat. If the worry over cosmic
censorship is that deterministic evolution may break down, then the worry is
assuaged if it can be shown that there is a unique (up to diffeomorphism)
maximal spacetime M, gab which is a solution of the vacuum field equations
80 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

and which is a development of the given initial data. And that is so regardless
of whether M, gab contains a Cauchy surface or not. Of course, the expectation
is that in typical cases when the unique maximal development for which Z
is a Cauchy surface is not maximal, then extensions across H+ (Z) will not be
unique. But that expectation needs to be substantiated, otherwise formulations
of cosmic censorship in terms of global hyperbolicity will loose their interest
(see Chrusciel 1992). I will temporarily set aside this concern, but will return
to it in section 3.7.

3.4 The cosmic censorship hypothesis

The cosmic censorship hypothesis (CCH) is the claim that the only naked
singularities that occur in the models of GTR are harmless.18 Recall that a
model of GTR is a triple M, gab, Tab, where M,gab is a relativistic spacetime,
Tab is a symmetric second-rank tensor called the stress-energy-momentum
tensor, and gab and Tab together satisfy EFE. Examples of naked singularities
in such models are easily constructed, as we know from the above discussion.
But for one reason or another such examples may be brushed aside as
harmless. The principal reason for putting a singularity in the harmless
category is that the model in which it occurs has features that, apart from
the singularity itself, make it "physically unreasonable." The literature on
cosmic censorship can be confusing to the casual reader because it mixes
together a number of different senses in which a model can be physically
unreasonable, among them: the model is literally physically impossible; the
model involves unrealistic idealizations, and there is no reason to expect that
more realistic counterparts will also have naked singularities; the model is
physically possible but involves such rare features as to leave no reason to
think that anything like the model will be actually encountered.
Whatever specific content is given to the physically reasonable/physically
unreasonable distinction, there are two boundary conditions that should be
satisfied. First, 'physically unreasonable' should not be used as an elastic label
that can be stretched to include any ad hoc way of discrediting putative
counterexamples to the CCH. Second, the conception we settle on must
permit the CCH to be stated in a precise enough form that it lends itself to
proof (or refutation). Some aspects of the physically reasonable/unreasonable
distinction can be stated in advance. Others emerge only in the process of
assessing potential counterexamples to the CCH. This, of course, raises worry
about the first boundary condition. But as we shall see below, the real worry
is about satisfying the second boundary condition while at the same time
making the CCH not obviously false and also general enough to cover the
situations that can be expected to occur in the actual universe.
Among the constraints on a physically reasonable model of GTR there
are two which, at least in part, can be motivated independently of any concern
with cosmic censorship.
 Cosmic Censorship 81

Energy conditions
Start with a spacetime M, gab and compute the Einstein tensor Gab associated
with gab. Then define Tab ~(l/8n)Gab. The result is a model that satisfies EFE
with cosmological constant A = 0. In this way we create, at least formally,
innumerable models of GTR, among which will be many that violate cosmic
censorship.19 To return to an example from section 3.1, we can start with
empty Minkowski spacetime R4, qab, remove a compact set K c: R 4 , and
choose a conformal factor Q that goes rapidly to 0 as K is approached.20 The
resulting triple M = R 4 - K, gab = Cl2>lab, Tab = (l/8n)Gab(gab) is a nakedly
singular model in which the singularity cannot be removed by extensions
since the spacetime is maximal. Of course, we may not get a model in the
intended sense that the stress-energy-momentum tensor arises from normal
sources, such as massive particles, electromagnetic fields, and the like. This
intention is difficult to state in terms of precise formal conditions on Tab, but
at a minimum we can impose one or another energy condition that we expect
normal sources to satisfy.
In chapter 2 we encountered the weak and strong energy conditions
which require respectively that Tab{,at,b > 0 for any timelike £" and that
Tabt,"t,b > -(1/2)7", T- T"a, for any unit timelike (". The dominant energy
condition requires that for any timelike £ fl , T"b(,b is a future-directed timelike
or null vector, which means that the flow of energy-momentum as measured
by any observer does not exceed the speed of light. 21 To see what these
conditions mean for a concrete example, consider a perfect fluid whose
stress-energy-momentum tensor has the form Tab = (ji + p)Ua Ub + pgab,
where jU and p are respectively the energy density and pressure of the fluid
and U" is the unit tangent to the world lines of the fluid elements. The strong
energy condition says that jU + p > 0 and /I + 3p > 0. The weak energy
condition requires that n > 0 and [i+p>0. And the dominant energy
condition says that fj, > 0 and \i > p.
I conjecture that these energy conditions, which are thought to hold for
all physically reasonable classical fields, rule out all the artificial examples of
naked singularities constructed by the method of two paragraphs above, but
I know of no formal proof of this.
One can now appreciate the importance of the value assigned to the
cosmological constant. Anti-de Sitter spacetime has constant scalar curvature
R < 0 and Einstein tensor Gab = — (\/4)Rgab. With A = 0 we can interpret
this as a solution with a perfect fluid source of constant density ( — R/32n) > 0
and constant pressure (R/32n) < 0. It is ruled out by the strong energy
condition. If, however, the cosmological constant is allowed to be non-zero,
then anti-de Sitter spacetime can be interpreted as an empty-space solution
(Tab = 0) with A = (l/4:)R. Since this spacetime violates the strongest version
of cosmic censorship, it might seem that without the stipulation of a zero
cosmological constant it will be much more difficult to achieve strong cosmic
censorship. But recall from chapter 1 that a positive value for A can help to
82 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

prevent the occurrence of singularities. So it remains to be seen whether or

not, on balance, the cosmological constant helps or hinders the quest for
cosmic censorship.

Causality conditions

For both physical and philosophical reasons one might require that a
reasonable model not contain closed or almost closed causal curves.22 But
imposing causality conditions by fiat is a little awkward for present purposes
since it smacks of assuming what we want to prove, at least for versions of
cosmic censorship that seek to censure breakdowns of predictability and
determinism that occur because of the development of acausal features after
the specification of initial data. Thus, in Taub-NUT spacetime we can choose
a partial Cauchy surface Z lying in the Taub region such that in some
neighborhood of S things are causally as nice as we like; but further to the
future of S things turn causally nasty, and as a result a Cauchy horizon for
Z develops (recall Fig. 2.3). At the present stage of discussion, however, fiat
is needed to secure cosmic censorship since Taub-NUT spacetime is a vacuum
solution to EFE so that the energy conditions are trivially satisfied. (In chapter
6 I will take up the question of whether or not GTR allows the operation of
time machines which would manufacture closed timelike curves. A negative
answer would establish part of the cosmic censorship hypothesis. Conversely,
a positive answer would be very damaging to the hypothesis.)
With only energy and causality conditions in place, the CCH fails, as is
shown by the "shell-crossing" singularities that may arise in spherically
symmetric dust collapse (see Yqdzis et al. 1973). The collapse is arranged so
that the outer shells of dust fall inward faster than the inner shells. A black
hole eventually develops, but not before the crossing of shells produces an
infinite density singularity that is visible from both near and far. Hawking
(1979) and others have suggested that such naked singularities are harmless
because they are relatively mild; in fact, the solution can be continued through
the Cauchy horizon as a generalized distributional solution (Papapetrou and
Hamoui 1967). However, Seifert (1979) has cautioned that harmlessness in
the relevant sense has not been demonstrated if our concern with cosmic
censorship is over the potential breakdown in predictability and determinism,
for uniqueness theorems for such generalized solutions are not in place. In
any case this method of trying to render naked singularities harmless does
not have a very long reach since, as will be discussed below, stronger and
irremovable singularities threaten.

Further conditions on Tab

A second strategy for dealing with shell focusing singularities and similar
examples is to impose further conditions on Tab and the equations of state.
These further conditions are supposed to assure that the sources are sufficiently
 Cosmic Censorship 83

realistic. In particular, it could be demanded that pressure effects (neglected

in dust models) be taken into account, and further that if matter is treated
as a perfect fluid, the equation of state must specify that p = p(fj.) is an
increasing function of density and that the pressure becomes unbounded as
the density becomes unbounded. This would rule out some of the early
shell-focusing counterexamples. But energy, causality, and further conditions
are not sufficient to prevent violations of cosmic censorship in self-similar
gravitational collapse23 with an equation of state p = aft, with 0 < a < 1
constant (see Ori and Piran 1987, 1988, 1990). Perhaps reasons can be found
to deem this soft equation of state physically unreasonable. Or perhaps it
should be demanded that to be realistic the fluid description should incorporate
viscosity and, thus, shear stress into Tab (see Seifert 1983). Further violations
of cosmic censorship satisfying these additional demands would drain much
of the interest in this tack for trying to render potential counterexamples
harmless.

Eardley's conditions

A related but different tack is taken by Eardley (1987), who demands realistic
equations of motion for the sources. In analyzing the naked shell-focusing
singularities that emerge in the Tolman—Bondi models of spherical gravita-
tional collapse as matter piles up at the center of symmetry, Eardley found
that the tacit assumption underlying the model is that the dust shells cannot
pass through the origin. The objection is not to the idealization of matter as
a pressureless, viscosity-free dust but rather to the unreasonable assumption
that dust shells behave completely inelastically. He conjectures that if the
motion of the dust is treated more realistically, for example by specifying
elastic recoil when the shells collide at the origin, then naked singularities will
not develop. Settling this and related conjectures seems to me to be one of
the more important items on the agenda for evaluating the prospects of the
CCH. However, I worry that the present approach, while possibly effective
in dealing with putative counterexamples to cosmic censorship on a case-by-
case basis, may not lead to a neat formal statement of the GCH that can be
proved to hold for a wide class of models.

Fundamental fields

Yet another approach in the spirit of Further conditions and Eardley's

conditions would recognize counterexamples to cosmic censorship only in
models involving fundamental fields. The gravitational field itself is, of course,
a fundamental field. The electromagnetic field should also be counted as
fundamental. And presumably so should a scalar field obeying a Klein—
Gordon type equation. Two aspects of determinism unite these examples.
First, the combined gravitational—matter fields for these cases admit of a
locally well-posed initial value problem. (This is the motivation behind the
84 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

part of Wald's (1984a, p. 305) formulation of cosmic censorship that requires

that the coupled Einstein-matter equations can be put in the form of a
second-order, quasi-linear, diagonal hyperbolic system, for which local exist-
ence and uniqueness theorems are available.) Second, the non-gravitational
field should admit a globally well-posed initial value problem in Minkowski
spacetime. The imposition of fundamental fields would disqualify many
potential counterexamples to cosmic censorship, including those involving
perfect fluids.
The motivation for the restriction of fundamental fields on physically
reasonable models lies in the notion that if either of the above determinism
requirements fails, then the field description involves a false idealization or
otherwise fails to be detailed and precise enough. The underlying faith here
is that in classical GTR it is always possible to go down to a fundamental
enough level of description on which the determinism requirements can be
satisfied.24 Of course, this article of faith could turn out to be wrong. But it
is polemically buttressed by the fact that the question of cosmic censorship
would be much less interesting if determinism could fail in special and general
relativistic settings for reasons having nothing to do with the development of
naked singularities.

Strength of singularities
Tipler (1985) and Newman (1986) have suggested a way to avoid the delicate
and contentious question of what counts as a physically reasonable source by
concentrating instead on strength of singularities. The idea is that for a source
to create a physically realistic singularity, the singularity must be strong
enough to crush ordinary matter out of existence by squashing it to zero
volume. A formal definition of the shrinking to zero volume was provided by
Tipler (1977b). Clarke and Krolak (1985) showed that a sufficient condition
for such behavior to occur along an incomplete timelike or null geodesic is
that l i m A _ A t ^2RabVa()C) F b (A) + 0 where V(X) is the tangent to the geodesic,
1 is an affine parameter, and 1+ is the upper bound on A. The term strong
curvature singularity is sometimes used to denote a singularity such that at
least one timelike or null geodesic terminating in the singularity satisfies the
crushing condition. Tipler (1977b) originally reserved the term for cases where
all non-spacelike geodesies terminating in the singularity satisfy the crushing
condition. Presumably in the latter cases all strong curvature singularities are
scalar polynomial singularities (see section 2.4); but I know of no proof to
this effect.
On the present approach we can give the CCH a clean and precise
formulation: the strong (respectively, weak) CCH holds just in case strong
(weak) censorship holds in any model which satisfies the energy conditions
and the causality conditions and in which the only singularities that occur
are of the strong curvature type. Unfortunately, the virtues of simplicity and
precision are not rewarded by truth, for this version of the CCH is in fact
 Cosmic Censorship 85

false, as is shown by the presence of strong naked singularities in the models

of self-similar gravitational collapse (Lake 1988; Ori and Piran 1990).

Penrose 's stability constraint

The final constraint on physically reasonable models I will consider is

Penrose's (1974, 1978) idea of stability under perturbations of initial conditions
and equations of state. The idea is often illustrated by an example mentioned
in the preceding section, the Reissner-Nordstrom model, where the effects of
small perturbations on a partial Cauchy surface E in the initial data for the
Einstein-Maxwell equations can build until they become an infinite blueshift
singularity on // + (S). But exactly what does such a demonstration show?
Wald takes it to show "there is good reason to believe that in a physically
reasonable case where the shell is not exactly spherical, the Cauchy horizon
. . . will become a true, physical singularity, thereby producing an 'all
encompassing' singularity inside the black hole formed by the collapse"
(Wald 1984a, p. 318). But if this is the moral of the perturbational analysis
then one should be able to drop reference to instability under perturbations
and say directly (either in terms of the above ideas or in terms of some
altogether different ideas) what a physically reasonable model is and then
proceed to prove that physically reasonable models so characterized obey
cosmic censorship.
I take it, however, that what the talk about instability is supposed to
point to is the notion that naked singularities that develop from regular initial
data are relatively rare within the set of all models of GTR. To make this
precise one would need to define a topology on the set of solutions to Einstein's
field equations and show that the target set is the complement of an open
dense set. Given our present limited knowledge of generic features of solutions
to Einstein's field equations, the project of establishing this version of the
CCH seems rather grandiose. In the meantime we can lower our sights and
investigate particular families of models whose members can be parameterized
in a natural way and try to show that cosmic censorship holds for almost all
parameter values. Or failing this we can challenge potential counterexamples
by showing instability under small perturbations and take this as a sign that
in some sense yet to be made precise a measure-zero result should be
forthcoming. The bad news here is that various counterexamples have been
shown to be free of blueshift instabilities, and the formation of naked
singularities in some cases of spherical collapse have been shown to be stable
under spherically symmetric perturbations. If it can also be shown that the
formation of naked singularities is stable under generic perturbations, then
we will have a strong indication that solutions with naked singularities are
not rare beasts, and the CCH will have to avail itself of some other escape
route. If on the contrary it can be shown that solutions with naked singularities
are rare beasts, there still remains the nagging question of what we would
say if we found that the actual universe we inhabit is one of the rare ones.
86 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

The good news for philosophers of science from this review is that the
hunting ground for the CCH contains a rich array of examples for studying
the related concepts of 'physically possible', 'physically reasonable', and
'physically realistic'. The bad news for advocates of cosmic censorship is that
the CCH does not yield easily to a formulation that is not obviously false, is
reasonably precise, and is such that one could hope to demonstrate its truth.
There is the very real danger that the CCH will forever remain a hypothesis,
and a rather vague hypothesis at that.

3.5 Is the cosmic censorship hypothesis true?

Since the literature does not agree on a precise statement of the CCH, the
question is a murky one. Not surprisingly, the murk has provided fertile soil
from which have sprung a wide variety of clashing opinions. Rather than
review these opinions, it seems to me more productive to try to state the
strongest cases on either side of the issue.
On the negative side we can begin by noting that faith in the CCH
depends on the notion that the GTR has some built-in mechanism for
preserving modesty by clothing singularities. In this respect the original
Oppenheimer-Snyder-Volkoff model of gravitational collapse (Oppenheimer
and Snyder 1939; Oppenheimer and Volkoff 1939) was misleading in creating
a false sense of security. Indeed, subsequent analysis has revealed a number
of different mechanisms by means of which cosmic censorship can be violated.
Steinmuller, King, and Lasota (1975) showed that a momentarily naked
singularity (see Fig. 3.10) can be produced by having a collapsing star radiate
away its mass in such a way that the star never forms a black hole since it
remains outside of its Schwarzschild radius (see also Lake and Hellaby 1981).

Fig. 3.10 A naked singularity emerges from a collapsing, radiating star

 Cosmic Censorship 87

Fig. 3.11 Shell-focusing in self-similar gravitational collapse leads to a naked

singularity

Then there are the persistently naked strong curvature singularities that form
in self-similar gravitational collapse, either as a result of the gravitational
collapse of a spherical shell of radiation as in the Vaidya spacetime, or as a
result of the shell-focusing of dust in the Tolman-Bondi spacetimes (see Fig.
3.11), or as a result of the collapse of an adiabatic perfect fluid (see Joshi
1993, Ch. 6 for a review). Joshi and Dwivedi (1992) showed that non-self-
similar gravitational collapse of a shell of radiation can also form strong
curvature singularities that violate WCC. And Joshi (1993, Ch. 6) found that
shell-focusing singularities of dust in non-self-similar gravitational collapse can
also be strong and naked. To be sure, each of these examples involves special
features and idealizations. But Joshi (1993, Ch. 7) makes it plausible that for
self-similar collapse, the development of strong naked singularities is not due
to the idealizations of perfect fluids, pressureless dust, and pure radiation but
holds for any form of matter satisfying the weak energy condition. In any
case, the fact that naked singularities can arise in such a variety of ways
should shake one's faith that GTR does have a modesty mechanism at work.
This faith suffers another apparent blow from the computer simulation
studies of Shapiro and Teukolsky (199la). They interpret their results to
indicate that a naked singularity can emerge from the collapse of a prolate
spheroid of a collisionless gas.25 However, this interpretation is currently a
matter of dispute. In particular, Shapiro and Teukolsky take the absence of
an apparent horizon in their simulations to serve as evidence for the absence
of an event horizon clothing the singularity.26 Wald and Iyer (1991) showed
that such an inference is dangerous since in Schwarzschild-Kruskal spacetime
there is a family of time slices that passes as near to the black hole singularity
as you like and for which there is no apparent horizon although, of course,
88 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

there is an event horizon. Another reason for doubting the significance of the
Shapiro-Teukolsky result comes from their use of the Vlasov equation to
describe the distribution of dust particles in their model. Kendall (1992b)
argued that it is unreasonable to expect that more general solutions for this
equation will exhibit the singular behavior found in the Shapiro—Teukolsky
model. (But in any case the use of computer models provides a powerful tool
for the exploration of the cosmic censorship hypothesis, and given the difficulty
in establishing interesting mathematical results it may be expected that much
of the future progress in discovering counterexamples to various versions of
cosmic censorship will be achieved with the help of this tool.)
Now let us turn the coin to look at the evidence in favor of cosmic
censorship. Penrose (1973) noted that the CCH can be used to derive
inequalities, involving areas of trapped surfaces and available masses, on the
behavior of black holes. 27 He tried but failed to find physically plausible ways
to violate these inequalities. The failure constitutes only weak evidence for
cosmic censorship since, as Penrose himself notes, if cosmic censorship does
not hold, a naked singularity may form without a trapped surface also
forming. Wald's (1973) investigation starts with the inequality necessary for
the Kerr-Newman solutions to represent a black hole: M2 > Q2 + J2/M2,
where M, Q_, and J are respectively the mass, electric charge, and the angular
momentum. He found that various ways of trying to violate this inequality
by injecting charge and angular momentum into the black hole all fail. Again
this gives only weak support to the GCH since although it confirms the
stability of stationary black holes once they form, it gives no confidence that
a black hole rather than a naked singularity will form in gravitational collapse.
Krolak (1986, 1987a, 1987b) has offered various censorship theorems,
but they rely on the dubious assumption of the existence of marginally
outgoing null geodesies.28
Newman (1984b) established a censorship theorem for conformally flat
spacetimes. (M, gah is conformally flat just in case gab = Q2riab where r\ab has
vanishing Riemann curvature and Q2 is a C°° map from M to (0, +00).) In
a generic null convergent (RabK"Kb > 0 for any null vector K") conformally
flat spacetime all null geodesies are incomplete. Nevertheless, Newman
showed that if Q2 is a proper map (i.e., the inverse image of any compact
subset of (0, + oo) is a compact subset of M) then M, gab is not FNS 2 . Of more
potential significance is another result of Newman (1984a, 1986) that
establishes a weak form of cosmic censorship for weakly asymptotically simple
and empty spacetimes29 which satisfy a suitable causality condition and in
which every incomplete null geodesic experiences a persistent curvature of
sufficient strength. However, the applicability of the persistent curvature
condition to gravitational collapse and cosmology remains uncertain.
Chrusciel, Isenberg, and Moncrief (1990) have established a form of
strong cosmic censorship for a special class of Gowdy spacetimes. These are
vacuum solutions to Einstein's field equations with vanishing cosmological
constant. They are spatially closed with space sections E being topologically
 Cosmic Censorship 89

T3, S3, or S2 x 51. These sections are metrically inhomogeneous because,

intuitively, gravitational waves are rippling through space. But despite the
inhomogeneities, these models have a two-parameter spacelike symmetry
(technically, there are two commuting spacelike Killing vector fields).
Chrusciel et al. focus on polarized Gowdy spacetimes where the Killing fields
are everywhere orthogonal. We know that initial data on a slice E determines
a unique (up to isometry) maximal development for which Z is a Cauchy
surface. It is shown that for an open dense subset of this initial data, the
maximal development is not properly extendible. In the Z = T3 case a generic
solution is not extendible in the future because the solution evolves towards
I = + oo with diminishing curvature. In the Z = S3 or S2 x S1 cases a generic
solution is not future extendible because it ends in a big crunch.
The supporters of cosmic censorship can also note that there is a respect
in which the Oppenheirner-Synder-Volkoff model is suggestive of a feature
which may be generic to gravitational collapse; namely, an event horizon
forms whenever the collapse of matter (whose stress—energy—momentum
tensor obeys appropriate energy conditions) proceeds beyond some critical
stages such as the formation of a trapped surface. Even if true, this event
horizon conjecture (EHC), as Werner Israel dubs it, does not by itself suffice
to establish either the strong or the weak form of the CCH. Strong cosmic
censorship may still fail because within the event horizon a locally naked
singularity can develop. And weak cosmic censorship can fail because, as
illustrated in Fig. 3.11, the formation of the event horizon may be preceded
by the development of a globally naked singularity. Nevertheless, proving the
EHC would be a big step towards proving the CCH. Israel (1986) has
established a preliminary version of the EHC by assuming that a trapped
surface develops and that its cylindrical extension remains non-singular.
A crucial test case for the CCH concerns vacuum solutions to EFE. For
then worries about whether the coupled Einstein-matter equations admit a
well-posed initial value problem vanish, as do worries about what conditions
Tab and the equations of state must satisfy in order to qualify as physically
realistic. Some care is still needed in formulating the CCH in this setting. For
example, the Taub—NUT spacetime falsifies the most naive attempt to
formulate the CCH in this setting. Conjecture 0: Among vacuum solutions
to Einstein field equations, future naked singularities (on any reasonable
definition) do not develop from regular initial data.
One of two directions can be taken to maneuver around this counter-
example. The first is to exclude by fiat acausal solutions, leading to Conjecture
1: Let M, gab be a maximally extended vacuum solution to Einstein's field
equations; if Z c: M is a partial Cauchy surface, then the spacetime is not
future nakedly singular with respect to Z unless strong causality is violated
at some p 6 H+ (Z). If 'future nakedly singular' is taken as FNS1; then from
the point of view of evolution we can formulate this conjecture as it applies
to a compact initial data surface as follows: the maximal future Cauchy
development of the appropriate initial data for a vacuum solution to Einstein's
90 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

field equation prescribed for a compact Z (without boundary) is not

extendible as a solution of the field equations unless strong causality is violated
at some / > e / / + ( Z ) . (This follows from the same kind of argument used in
section 3.3. As noted, in a globally hyperbolic spacetime any compact partial
Cauchy surface is a Cauchy surface. So if cosmic censorship in the form of
global hyperbolicity is to hold, D+ (Z) for a compact partial Cauchy Z cannot
be extendible.) The alternative to dismissing acausal solutions is to eschew
spatially closed universes, leading to Conjecture 2: Let M, gab be a maximally
extended vacuum solution to Einstein's field equations; if M, gab does not
admit a compact partial Cauchy surface then it is not future nakedly singular
with respect to any partial Cauchy surface.
This retreat from Conjecture 0 to Conjectures 1 and 2 is less than satis-
fying. Conjecture 1 rules out by fiat one sort of breakdown in predictability
due to the emergence of acausal features from a past that may have been
causally pure, while Conjecture 2 refuses to consider spatially closed universes.
At present, however, more attractive alternatives do not seem to be available.
A failure of Conjecture 1 or 2 need not be seen as fatal to cosmic
censorship since one could retreat to the version of censorship that asserts
only that such failures are sparse among the vacuum field solutions. Genericity
considerations can also be used to fill the gap in the above conjectures. Moving
from Conjecture 0 to Conjecture 1 amounts to ignoring the violation of cosmic
censorship that arises in Taub—NUT spacetime. That such ignorance may
not be bliss was suggested by Moncrief's (1981) finding that there is an infinite
dimensional family of vacuum solutions with NUT-type extensions. But
Moncrief and Isenberg (1983) also established a result that suggests that such
behavior is nevertheless special. For a partial Cauchy surface Z in the Taub
region of Taub—NUT spacetime, H+ (Z) is a compact null surface ruled by
closed null curves. The result is that analytic vacuum solutions possessing
such a null surface must also possess a Killing field.
It is not easy to say whether the weight of evidence to date favors some
interesting form of cosmic censorship. Certainly there is enough evidence pro
and con to keep both the proponents and opponents at work.

3.6 Black hole evaporation

If quantum gravity doesn't banish singularities altogether, then a frightening

prospect opens up, even if cosmic censorship is true at the level of classical
GTR. For according to calculations from semiclassical quantum gravity,
black holes will eventually evaporate due to Hawking radiation (see Wald
1984a, Sec. 14.3). And assuming that classical relativistic spacetime concepts
can be applied to the result, either a naked singularity or a thunderbolt can
be expected to emerge. To discuss the reasons for that expectation, I will
review a technical result due to Gcroch and Wald (see Wald 1984b; see also
Kodarna 1979 for a related result).
 Cosmic Censorship 91

Fig. 3.12 Black hole evaporation

Theorem (Geroch and Wald). Let M,gab be a time-oriented spacetime,

and let Dj and Z 2 be closed achronal sets with £ t connected and £2 edgeless.
Suppose that (i) there is a pointp e D+ (Zj) such that/; £ (J~ (Z 2 ) u J + (Z 2 )),
and (ii) J+(K] n Sj has compact closure, where K =-1,i — (D~ (E2) <"> ^i)-
Then S2 £ £ + ( £ i ) .

The Penrose diagram of (one-half) of a generic black hole configuration of a

gravitationally collapsed body with center of symmetry r = 0 is shown in Fig.
3.12a. In this case D + (E 1 ) for the time slice Sj includes every point to the
future of Ej so that there is no slice Z2 to the future such that E 2 <t D+ (E^.
Here the conclusion of the theorem fails because neither of the conditions (i)
or (ii) is applicable. Figure 3.12b pictures the evaporation of a black hole by
means of a catastrophic burst of electromagnetic radiation (thunderbolt). No
naked singularity develops since the burst of radiation cuts off future
extensions in such a way that again D+ (Sj) includes everything to the future
of I^.30 The theorem fails to apply since although condition (i) holds, (ii)
fails. However, if Hawking radiation does not produce thunderbolt evapora-
tion, then something akin to Fig. 3.12c should result; since conditions (i) and
(ii) both apply, the theorem can then be invoked to conclude that there will
be a violation of cosmic censorship in the form of a breakdown in predict-
ability. The question then becomes whether the calculations of semiclassical
quantum gravity—which predict Hawking radiation—indicate that in black
hole evaporation a naked singularity is more likely to result than a thunderbolt
singularity. Hawking and Stewart (1993) and Lowe (1993) report somewhat
conflicting results.
What I would like to briefly explore is the prospect of proving that black
hole evaporation can produce a violation of cosmic censorship not only in
the sense of a breakdown in predictability but also in the stronger sense of a
singularity visible from J>+, as Fig. 3.12c would suggest. Towards this end I
will assume that black hole evaporation takes place in a spacetime where both
,f+ and .$> are defined. The theorem of Geroch and Wald already tells us
92 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

that when the hypotheses apply the future boundary // + (S 1 ) of D+(Z,i) is

non-empty. The generators ofH+ (Zj) are null geodesies. I will simply assume
that these generators extend to </ + '. It may also be assumed without loss of
generality that Z t is a partial Gauchy surface and also that Zj <£ J~(^ + ),
for Zj is supposed to correspond to a time before the black hole evaporates.
Because Z t is edgeless, the generators of H+(£1) are past endless and, thus,
past inextendible (Hawking and Ellis 1973, Prop. 6.5.3). There are now two
main possibilities to consider. (1) Some of the generators of H+ (Zj) are totally
or partially past imprisoned in a compact set of the spacetime. This possibility
can be ruled out by imposing the requirement of strong causality (Hawking
and Ellis 1973, Prop. 6.4.7). (2) With (1) ruled out, the generators of^ + (S 1 )
must in some sense "run off the edge" of spacetime in the past direction. There
are two main subcases to consider, (a) Some of the generators run into a
singularity, i.e., are past incomplete. Since, by assumption, the generators
extend to «/ + , we have our naked singularity. To rest content with this
subcase, we need to rule out the other, (b) The generators are all past
complete. Again we have two subcases to consider, (a) The generators run
into an "internal infinity," which can be illustrated by modifying the example
accompanying Fig. 3.2 by letting Q go to oo rather than to 0 as the region
K is approached.31 Imposing a suitable condition to rule out such pathologies
we should arrive at the second subcase. (/?) The generators of .// + (Zj) all run
to J~. But in this case we should be able to show that T,l c: J~( t / + ) ; which
is contrary to assumption. In sum let me emphasize that there is no pretense
of a proof here. But the considerations reviewed do seem to me to lend
credence to the notion that, excluding thunderbolt evaporation, physically
reasonable cases of black hole evaporation can be expected to produce
singularities, in the sense of geodesic incompleteness, that are visible from ,/ + ,
and thus are violations of WCC.
The upshot is potentially disturbing. If we believe the classical GTR, it is
likely that black holes have formed throughout spacetime. If semiclassical quan-
tum gravity has validity, then these black holes evaporate, leading eventually
either to thunderbolt singularities or to naked singularities. If we are not
always saved from naked singularities by thunderbolts, then perhaps we are
saved by the fact that evaporation time for black holes as massive as our sun
is very long indeed. But if mini-black holes formed in the early universe or
if, as has been recently suggested, mini-black holes form in supernovas, we
could be surrounded by naked singularities. Just how disturbing this prospect
is cannot be assessed until we are in possession of more detailed information
about the nature of the singularities that emerge from black hole evaporation.

3.7 What if cosmic censorship should fail?

How much of a disaster for physics would it be if the CCH should prove to
be wrong? Early in the investigation of the problem of cosmic censorship,
 Cosmic Censorship 93

Penrose posed this question and sketched a preliminary response:

It is sometimes said that if naked singularities do occur, then this would be

disastrous for physics. I do not share this view. We have already had the
example of the big-bang singularity in the remote past, which seems
not to be avoidable. The "disaster" to physics occurred right at the
beginning. Surely the presence of naked singularities arising occasionally in
collapse under more "controlled" circumstances would be the very reverse
of a disaster. The effects of such singular occurrences could then be accessible
to observation now. Theories of singularities would be open to observational
test. The initial mystery of creation, therefore, would no longer be able to
hide in the obscurity afforded by its supposed uniqueness. (Penrose 1973,
p. 133)

The reference to the big bang singularity shows that the visibility of the
singularity by itself portends no disaster; but this point has already been
codified in the various definitions of naked singularities, which exclude the
initial singularities in standard big bang cosmologies. Only those singularities
that entail a breakdown in predictability and determinism are counted as
naked. Now for whatever psychological reasons, we are less disturbed by an
inability to retrodict or determine the past than by an inability to predict or
determine the future. The effects of naked singularities in our past would
imply a breakdown in retrodiction and historical determinism; but, as Penrose
says, the effects of such singularities would be accessible to observation now,
and given a knowledge of these effects we can hope to determine what will
transpire in the future. This hope is undercut if our spacetime is future nakedly
singular. Again we must ask how much of a disaster it would be for physics
if we were exposed to such spacetime pathologies.
To make a start towards an answer it is necessary to sunder a potentially
misleading association made above between predictability and determinism.
Predictability in a generic relativistic spacetime is impossible since it is
generally impossible to acquire enough information for a sure forecast before
the occurrence of the event to be predicted (see chapter 5). However,
determinism does hold at least locally for the pure gravitational field, and
arguably it can be expected to hold locally for the coupled gravitational-
matter fields for any matter fields that are fundamental (see section 3.4). The
question of cosmic censorship is then whether this local determinism will break
down in the large due to the development of naked singularities.
At this juncture the devil's advocate may propose that we should not get
overly agitated by the prospect that a cosmic censor is not at work, for after
all, QM has accustomed us to breakdowns in determinism. While giving the
devil his due, it has to be noted that there are important differences between
the indeterminism of QM and the indeterminism associated with a failure of
cosmic censorship. In the quantum case the unitary evolution of the state
vector, of which the Schrodinger equation is simply the infinitesimal form, is
deterministic, and indeterminism enters only when the unitary evolution is
94 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

interrupted by a miracle of a "collapse of the state vector" when a

measurement is made. The nature of this miracle and even whether it occurs
are issues currently subject to intense debate. Whatever the outcome of the
debate, the kind of indeterminism at issue is at worst not of the anything-goes
variety since the quantum theory specifies the precise form for the statistics
of outcomes of quantum measurements. By contrast, the principles of classical
GTR do not tell us whether a naked singularity will passively absorb whatever
falls into it or will regurgitate helter-skelter TV sets, green slime, or God only
knows what. The fear here has been articulated by Shapiro and Teukolsky.

A counterexample to cosmic censorship—a configuration of matter that

evolves to produce a naked singularity—would be a catastrophe for the
theory [of general relativity]. Given the existence of just one singularity,
general relativity could not say anything precise about the future evolution
of any region of space in communication with the singularity. Determining
the validity of cosmic censorship is perhaps the most important outstanding
problem in the study of general relativity.32 (Shapiro and Teukolsky 1991b,
p. 330)

Such fears about the unbridled influences of naked singularities would

be somewhat assuaged if it could be shown that a naked singularity can have
only a minimal influence on external observers. For example, one might hope
that in the asymptotically flat regime such violations of WCC that do occur
are mild in the sense that the null geodesies that emerge from the singularity
and escape to «/+ are of "measure zero" in some appropriate sense. In fact,
however, this hope is not realized for the naked singularities that develop in
self-similar and non-self-similar gravitational collapse of dust, perfect fluids,
and pure radiation (see Joshi 1993, Ch. 6).
The further exploration of fears about the disruptive influence of naked
singularities requires attention to technical questions that are not only hard
to answer but also hard to formulate precisely. To give an indication of what
sorts of questions need to be addressed, let us pursue the considerations raised
in Approaches 4 and 5 of section 3.3. Choose a three-manifold X without
boundary and specify initial data appropriate to the system of gravitational
and matter fields one is interested in. Use the Einstein-matter equations to
evolve the initial data in time as far as compatible with X being a Cauchy
surface so that in the resulting spacetime M, gab, M is -D(X). (We know that
there is a unique (up to diffeomorphism) way of doing this.) Now ask whether
M, gab is extendible to a larger spacetime. We know that if £ is non-compact
the extendibility of M, gab may be due to an unfortunate choice of X and not
to the appearance of a naked singularity. To avoid such worries we can
concentrate on cases where we can be sure that the extendibility is not due
to a bad choice of £ (e.g., £ is compact or else X is non-compact and X, hab
is complete and asymptotically flat). A second hitch is that we are not
interested in just any mathematical extension of M, gah of M, gah but only in
those that satisfy EFE. But this is not a well-defined notion until the Tab is
 Cosmic Censorship 95

specified for the extension, and it is far from clear what conditions, beyond
the standard energy conditions, should be imposed on Tab in the extended
region. To avoid this difficulty for the moment, let us focus on the empty
space initial value problem and on extensions that are also solutions to the
vacuum EFE.
We are now in a position to formulate some more or less precise questions.
Are there examples where there are inequivalent (i.e., non-isometric) exten-
sions M, gab, M', gab, . . . , of M, gab? If so, what is the multiplicity of such
extensions? Can uniqueness of extensions be restored by requiring that the
extensions preserve some features of M, gab, such as symmetries of gab? On
the other hand, if there is a unique extension, is uniqueness undermined by
restricting the allowable isometrics, e.g., to ones that preserve orientation or
to ones that do not move the original Cauchy surface Z of the maximal
globally hyperbolic development M, gab?33 Of course, these questions are
moot if there are no physically realistic examples where cosmic censorship
fails. Not having any such examples in hand to investigate, we have to try
to get a feel for what answers are likely to emerge by investigating mathe-
matical examples where cosmic censorship fails.
It has been claimed that the maximal globally hyperbolic region of
Taub—NUT spacetime (the Taub region) admits different extensions. The
examples given in the literature do indeed exhibit the existence of different
extensions, but these extensions are not inequivalent, i.e., non-isometric (see
Chrusciel and Isenberg 1993). But if the isometry is required to preserve a
chosen Cauchy surface of the Taub region, then these extensions are
inequivalent (Chrusciel and Isenberg 1993). Furthermore, the existence of
non-isometric extensions of the Taub region (without any restriction on the
isometry) has been demonstrated by Chrusciel and Isenberg (1993), although
to achieve non-uniqueness the Taub region must be extended in the past as
well as in the future. They also show that among the polarized Gowdy
spacetimes that violate cosmic censorship there are some that admit of an
infinite multiplicity of inequivalent extensions. These results suggest that if
cosmic censorship fails for vacuum solutions then a variety of extensions across
H+ (£) will be available even if the extensions are restricted to vacuum
solutions.
If naked singularities are not quiescent but throw off matter, then even
if we start with a vacuum solution M, gab there is no a priori reason to limit
the extension across H+ (S) to vacuum solutions. So unless some limitations
are placed on what a naked singularity can spew out, the specter of a vast
additional range of underdetermination arises. Penrose has cautioned that
not just anything goes here.
If we envisage an isolated naked singularity as a source of new matter in
the universe, then we do not quite have unlimited freedom in this! For
although in the neighborhood of the singularity we have no equations, we
still have normal physics in the space-time surrounding the singularity. From
the mass energy flux theorems of Bondi et al. and Sachs, it follows that it
96 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

is not possible for more mass to be ejected from a singularity than the original
total mass of the system, unless we are allowed to be left with a singularity
of negative total mass. (Such a singularity would repel other bodies, but
would still be attracted by them!) (Penrose 1969, p. 274)

Assuming that there is no naked singularity to begin with, the proof of the
positive mass conjecture rules out the last awful possibility.34 But although
not just anything goes, the range of what a naked singularity can, consistently
with standard GTR, disgorge is vast.
Another way to get a handle on possible non-uniqueness due to a failure
of cosmic censorship is to suppose that M, gab has a unique extension across
// + (Z) or else that a reason has been found to single out one of the extensions.
Treat the extended spacetime M, §ab as a fixed background spacetime, and
ask whether test fields on the globally hyperbolic region of M, gab have unique
extensions across //+(S). Only a few relevant results are known, and they
are mixed. For elastic collisions of billiard balls in certain "wormhole
spacetimes" where closed timelike curves develop when one crosses H+(L) it
has been found that there is an infinite multiplicity of self-consistent but
inequivalent extensions of some initial trajectories of a billiard ball that take
the ball around a closed time loop and into collision with itself (see chapter 6).
On the other hand, Wald (1980) has argued that in static but non-
globally hyperbolic spacetimes there is a physically reasonable way to obtain
a global dynamics for a scalar field $ obeying the homogeneous wave equation
DO =: VaV<D = 0. If the value of <J> and its derivative normal to a spacelike £
are specified, then the wave equation determines $ only in /)(£), which in
the case of the hypothesized non-globally hyperbolic spacetime is only a
proper subset of the full spacetime. Wald's prescription fixes O throughout
the spacetime in such a way to secure agreement with the usual dynamics on
£)(S) and to make $ smooth everywhere for smooth initial data of compact
support. This is a nice result, but the prescription implicitly assumes boundary
conditions on the singularities. For instance, in the artificial case where the
singularities correspond to regions cut out of a larger singularity-free space-
time, Wald's prescription requires that <D vanishes on the boundaries of the
cutout regions. There are surely other prescriptions for obtaining global values
which also agree with the normal dynamics on D(Z) and which make $
smooth everywhere for smooth initial data. Adjudicating among the alternative
prescriptions will depend on which of the implicit boundary conditions on
singularities are regarded as reasonable. It is not clear what principles can
be used to guide these decisions.
Friedman and Morris (1991a, b) have demonstrated the uniqueness of
solutions of DO = 0 for a class of non-globally hyperbolic spacetimes that
contain closed timelike curves. However, these spacetimes do not conform to
the paradigm studied in this chapter since they contain no partial Cauchy
surfaces and initial data have to be specified on J>~.
Now for sake of discussion let us be pessimistic. Suppose that cosmic
 Cosmic Censorship 97

censorship fails and that as a consequence there is a vast range of alternative

extensions across the Cauchy horizon, all of which are compatible with
principles of classical GTR. Must we conclude that physics becomes hopeless?
Must we simply throw up our hands and wait for the weirdness to unfold?
No! We can try to discern what regularities naked singularities display. For
example, are the singularities that develop in certain situations quiescent? Do
those that develop in other situations all ooze green slime, and if so do they
ooze it at a regular rate? The attitude that physics is hopeless if naked
singularities occur stems from what may be termed GTR chauvinism—the
notion that Einstein and his followers discovered all of the laws relevant to
classical gravitation. If we acknowledge that laws of nature are simply
codifications of certain deep regularities,35 then we should be prepared to
discover through observation that naked singularities obey laws of their own.
If we are lucky these additional laws, when conjoined with the laws of
standard GTR, will restore predictability and determinism. Even if we are
not so lucky they may still give us some interesting physics. Of course, we
must be prepared for the eventuality that naked singularities exhibit no
interesting regularities at all, in which case they would indeed be a disaster
for physics. But at present only GTR chauvinism would lead us to fixate on
this worst-case scenario.

3.8 A dirty open secret

Cosmic censorship as it has been investigated here is concerned primarily with

determinism. But it is "well known" (to the handful of experts who think
about these matters) that determinism in general relativity fails without the
help of what looks very much like fiat.
For concreteness and simplicity, consider solutions to the source-free EFE.
Pick any two such solutions M, gab and M', g'ab that are maximal and choose
any spacelike hypersurfaces "L c. M and £' c M'. The metrics of the respective
spacetimes induce on the hypersurfaces their first and second fundamental
forms, resulting in the initial data sets E, hab, Kab and D', h'ab, K'ab. One would
like to be able to show that if these initial data sets agree (i.e., there is a
diffeomorphism d of £ onto "L' such that d*hab = h'ab d*Kab = K'ab) then so do
the respective domains of dependence D(L) £ M and D(L') £ M' (i.e., there
is an extension of d to an isometry of D(L) onto D(E')). Having shown this,
one can then go on to the question of whether for appropriate £ and £', Z3(£)
and D(Z') exhaust M and M' respectively and, if not, whether M — D(L)
and M' — /)(£') agree—i.e., the question of cosmic censorship as studied
under Approaches 4 and 5 of section 3.3. But the previous problem must be
attended to first, for if it should have a negative outcome, the failure of
determinism involved precedes any lapse by the cosmic censor.
The problem might seem to have an obviously positive outcome. The
theorem of Choquet-Bruhat and Geroch (1969) proves the existence of a
98 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

unique (up to isometry) maximal development for which the initial value
hypersurface is a Cauchy surface. So if each of the spacetimes considered is
such that for any spacelike hypersurface, the associated maximal development
is realized in the spacetime, then we are done. And it might seem that the
antecedent is discharged by the already adopted demand that the spacetimes
under consideration are maximal.
Unfortunately, the last step does not hold. That a spacetime M, gab is
maximal does not guarantee that it is determinism maximal in the sense that
there is no spacelike *L c. M such that an isometric imbedding (p: -D(S) -* M
into a spacetime M, gab can be found that makes <j0(Z)(2)) a proper subset of
Z)(<jj(Z)). 3 6 To create an example of a maximal but not determinism maximal
spacetime, start with a spacetime just as nice as you like—say, four-
dimensional Minkowski spacetime. Remove the two-plane {t = x = 1}. In the
resulting holey spacetime a time slice such as £ = {l = 0} is no longer a
Cauchy surface. Of course, the holey spacetime is properly extendible and is
thereby excluded from consideration by the initial stipulation. However, the
universal covering spacetime of the holey spacetime is maximal. But it is not
determinism maximal since if Z is any of the lifts of £ to the universal cover,
there is an isometric imbedding (p of -D(X) back into Minkowski spacetime
such that <p(D(%)) c Z)(<p(Z)) (see Geroch 1977). Such examples cannot be
brushed aside as being rare since for any normal spacetime one can create
innumerable mutilated versions that contain determinism holes.
Determinism can be restored by declaring any spacetime within which
maximal developments do not get realized as persona non grata. But on what
grounds that are not question begging? Certainly not on the grounds of failure
to satisfy the field equations and energy conditions. To rule out the above
example is to rule out one way Nature might, consistently with all of the
known laws of GTR, continue to evolve things across // + (£). What then is
to say that She cannot proceed this way? The most prevalent attitude among
general relativists seems to be that fiat is required (see Ellis and Schmidt
1977), otherwise questions about more interesting ways in which determinism
can fail are never reached. I implicitly adopted this attitude in the foregoing
sections. I am not proud of doing so, but I am no better than my brethren
in physics in seeing an alternative to fiat.

3.9 Conclusion

In chapter 2 I argued that questions about the existence of singularities may

turn on delicate questions about the continuity and differentiability of the
metric. Questions of cosmic censorship also involve delicacies of continuity
and differentiability. In the present chapter I ignored them for the most part,
for to have done otherwise would have turned an already complicated
discussion into an unreadable one. But ultimately these considerations must
be incorporated. Consider, for example, a theorem which purports to
 Cosmic Censorship 99

demonstrate some form of cosmic censorship by showing that the maximal

Cauchy development of some generic set of initial data is inextendible. One
must ask: What are the differentiability conditions assumed in the theorem?
Are they low enough to assure that all physically reasonable extensions have
been taken into account? One plausible attitude here is that, for purposes of
evaluating cosmic censorship, one may assume that the spacetime metric has
enough differentiability to assure the existence and uniqueness of Cauchy
developments, for otherwise determinism could fail even before the matter
gets submitted to a censor. Sufficient conditions to assure this existence and
uniqueness are given in Hawking and Ellis (1973, pp. 249-251), but
minimally sufficient conditions are apparently not known. An opposing
attitude would allow that once maximal Gauchy evolution stops, Nature
might continue in some lower differentiable way as long as EFE are satisfied
in some distributional sense.37 If the latter attitude is accepted, the censor
has to work much harder.
Even leaving aside such technicalities, the prospects seem dim for a quick
and clean resolution of the question of cosmic censorship. The above discussion
reveals that "cosmic censorship" is a label for a large and diverse class of
ideas and motivations. It is doubtful that these various ideas and motivations
can be accomodated in a few precise mathematical conjectures that can then
be either proved or refuted. For both the proponents and opponents of cosmic
censorship this means that there is no easy road to victory. The proponents
have to be content with proving limited censorship theorems for special cases;
the opponents have to pile up more and more counterexamples to various
forms of cosmic censorship; and each side has to hope that eventually it will
wear the other down. It is too soon to try to predict the outcome of this battle
of attrition, but two things seem certain. First, whatever the final resolution,
a good deal of interesting physics and interesting philosophy of physics will
be generated by the process. Second, the battle is not apt to fizzle out for
lack of interest, for the business of cosmic censorship is too important to ignore.

Notes
1. Although I will use predictability and determinism interchangeably in the
next few sections, it will eventually become important to distinguish them; see section
3.7 and Earman (1986).
2. It may be that initial conditions on 2 plus supplementary boundary conditions
may suffice for a determination; see section 3.7.
3. More precisely, // + (2) is denned as D+(E) - r(Z> + (2)). The past Cauchy
horizon H~ (2) of 2 is defined analogously.
4. It is assumed throughout that the spacetimes under discussion are temporally
orientable so that they admit a globally consistent time directionality. For a definition
of temporal orientability and a discussion of the hierarchy of causality conditions on
relativistic spacetimes, see chapter 6.
5. The Penrose diagrams bring infinity in to a finite distance by means of a
conformal transformation that preserves causal relations; see Hawking and Ellis (1973)
for details. In Fig. 3.3b i° denotes spatial infinity.
100 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

6. Future null infinity ^ + and past null infinity «/~ are defined in a non-physical
spacetime obtained by taking the conformal completion of physical spacetime (the
reader may consult Hawking and Ellis (1973) and Wald (1984a) for details). The
interior of a black hole B and the absolute event horizon E (see below) are defined
in terms of./", which lives in the non-physical spacetime. But B and E are part of
physical spacetime.
7. In Fig. 3.4b i+ and i~ denote respectively future and past timelike infinity.
They arc respectively the terminus and origin of timelike geodesies.
8. Roughly, a stably causal spacetime is one in which it is possible to widen the
light cones without permitting closed causal loops to form; see chapter 6 for a precise
definition.
9. Roger Penrose (1979) has put forward the "Weyl curvature hypothesis" which
rules out the type of white holes that violate cosmic censorship. He conjectures that
this asymmetry must be underwritten by a fundamental law of physics that is not
time reversal invariant; see chapter 5.
10. See Gowdy (1977). The reader is invited to construct an example to show
why the definition of topology change must be restricted to maximal partial Cauchy
surfaces.
11. H+ (2) for a spacelike hypersurface £ is a null surface and its generators are
null geodesies. The generators are either past endless or else have a past endpoint in
the edge of S. For a partial Cauchy surface, which by definition is edgeless, the
generators are past endless.
12. See Penrose (1978) for definitions of these concepts and for a discussion of
how these concepts can be used to formulate cosmic censorship.
13. If 5R ^ 0 because there is a/> e M such that /~ (p) contains a future-directed
timelike half-curve y of infinite proper length, then we have an example of Malament—
Hogarth spacetime. These spacetimes are investigated in chapter 4.
14. In Droste coordinates the line clement has the form

where a is proportional to the total mass M. If a < 0 the metric components are
regular down to r = 0 where a naked singularity resides.
15. Recall that, roughly speaking, strong causality rules out closed and almost
closed causal curves.
16. The constraint equations are

where 3R is the curvature scalar of the metric haly of 2 and Da is the derivative operator
associated with hab. The second fundamental form (aka extrinsic curvature) Kah of £
is defined as follows. Let £" be the normed tangent field of the congruence of timelike
geodesies orthogonal to 2. Then Kab = : V a ^ t . It can be shown that Kab = (I/2)£jA o 6
where £^ is the Lie derivative with respect to £ a (see Wald 1984a, Appendix C). In a
coordinate system x", t, where the x", cc = 1 , 2 , 3, label points on S and t is a parameter
along the geodesies orthogonal to I such that £" = (d/8t)a, Kall = (1/2) ohah/dt.
 Cosmic Censorship 101

17. M, gab is a development of £, /!„,,, A"o(, if there is an imbedding <p: X —> M such
that <p(Z) is a spacelike hypersurface whose first and second fundamental forms
induced by gab are respectively hab and A"^ .
18. Here I am using the terminology of Israel (1984).
19. Here I am following Geroch and Horowitz (1979).
20. Choosing Q to go to oo as K is approached would also produce a naked
singularity in Penrose's sense. But this naked singularity will not be a singularity in
the originally intended sense, e.g., involving geodesic incompleteness (see chapter 2).
21. The terms strong energy condition and weak energy condition are potentially
misleading since the former does not entail the latter. The dominant energy condition
does entail the weak energy condition. The weak energy condition entails the null
energy condition which requires that TabK"Kb > 0 for every null vector K".
22. See however the discussion of acausal spacetimes in chapter 6.
23. A spacetime is self-similar just in case there is a timelike vector field Va such
that V0 Vb — Vj,I£ = ®gab, ® = constant. For a spherically symmetric spacetime, this
means that under a coordinate transformation t -* t = at, r -> f = ar, 6 —> 8 = 6,
</> -> </> = (j), the metric components gtj (i,j = 1, 2, 3, 4) transform as gij(r, t) =
(V*2)gtj(f,i).
24. The description of a source as a perfect fluid involves idealizations. The fact
that it is difficult to obtain existence and uniqueness theorems for the coupled
Einstein-Euler equations when the fluid has compact spatial support (see Rendall
1992a) would, according to the present faith, not undermine determinism but would
only serve to indicate the limits of the usefulness of the perfect fluid description.
25. The findings of Shapiro and Teukolsky are in accord with Kip Thome's hoop
conjecture which posits that in the non-spherical collapse of a massive object, a black
hole forms if and only if the circumference of the object in all directions is less than
its critical circumference 4-nGM/c2. If correct, the hoop conjecture supports a limited
form of cosmic censorship but would also point to a class of counterexamples to the
general cosmic censorship hypothesis.
26. An apparent horizon is the outer boundary of a trapped region (see Hawking
and Ellis 1973, Sec. 9.2 for precise definitions). The relevant point here is that the
existence of an apparent horizon entails the existence of an event horizon but not
conversely.
27. Recall that a trapped surface 2 is a closed two-surface such that both ingoing
and outgoing null geodesies orthogonal to !X are converging.
28. A null geodesic of M, gab that forms an achronal set and has an endpoint in
^+ is called outgoing. It is marginally outgoing if it is a limit curve of all outgoing
null geodesies.
29. See Hawking and Ellis (1973, p. 252) for a definition of this class of
spacetimes.
30. The thunderbolt singularity could also spread to ^+ along a spacelike path
and cosmic censorship would be preserved.
31. The notion that M, gab contains an internal infinity can perhaps be captured
by the condition that there is a neighborhood U c: M which has a compact boundary
and which contains a geodesic half-curve of infinite affine length. I am indebted to
Al Janis for this suggestion.
32. Given Einstein's aversion to the notion that God plays dice with the world,
it is tempting to speculate that a nascent version of such concerns was behind his
visceral dislike of spacetime singularities. However, Einstein showed no inclination to
102 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

distinguish between "good singularities," with which we can peacefully coexist, and
"bad singularities," with which even detente is impossible. He thought that all
singularities must be excluded from a complete scientific theory (see chapter 1).
33. Let \ji: 2 —> M denote the imbedding of £ as a Cauchy surface of M, gab and
let Mr: M —» M and *P: M —> M denote the isometric imbeddings of M, gat> into M, gab
and M, gab respectively. Chrusciel and Isenberg (1993) note that if the Cauchy surface
(/'(S) has a privileged status, then one might not want to count M, gab and M,gab
as equivalent under an isometry (p: M —> M if (p moves 2 in the sense that
(p o T o ,/, j, <f- o i/,.
34. See Wald (1984a) for a discussion of the positive mass proof. Negative mass
Schwarzschild spacetime (see note 14) has negative ADM mass and a naked
singularity. It is not a counterexample to the positive mass theorem, which requires
that the initial hypersurface be singularity free. Nor, as noted above, would this
example be regarded as a violation of the form of cosmic censorship that excludes only
those naked singularities that develop from regular initial data, since the singularity
has been present for all times.
35. See chapter 6 for a discussion of the nature of physical laws.
36. A determinism maximal spacetime is also known in the literature as a
hole-free spacetime. Being hole free does not entail being globally hyperbolic (think
of Godel spacetime). Nor does the implication go in the opposite direction (think of
a truncated version of Minkowski spacetime with all of the points such that / > 1997
removed). However, a globally hyperbolic and inextendible spacetime is necessarily
hole free.
37. See chapter 2 for a discussion of this matter.
 4
Supertasks

4.1 Introduction

Is it possible to perform a supertask, that is, to carry out an infinite

number of operations in a finite span of time? In one sense the answer is
obviously yes since, for example, an ordinary walk from point a to point b
involves crossing an infinite number of finite (but rapidly shrinking) spatial
intervals in a finite time. Providing a criterion to separate such uninteresting
supertasks from the more interesting but controversial forms is in itself no
easy task,1 but there is no difficulty in providing exemplars of what
philosophers have in mind by the latter. There is, for instance, the Thomson
lamp (Thomson 1954-55). At t = 0 the lamp is on. Between t = 0 and t = 1/2
the switch at the base of the lamp is pressed, turning the lamp off. Between
t = 1/2 and / = 3/4 the switch is pressed again, turning the lamp on. Etcetera.
The upshot is that an infinite number of presses are completed by t = 1. Then
there is the super n machine. Between t = 0 and t = 1/2 it prints the first
digit of the decimal expansion of n. Between t = 1/2 and t = 3/4 it prints the
second digit. Etcetera. The result is that the complete expansion has been
printed at t = 1. More interestingly from the point of view of mathematical
knowledge there is the Plato machine which checks some unresolved existential
conjecture of number theory for T during the first half-second, for '2' during
the next quarter-second. Etcetera. The result is that the truth value of the
conjecture is determined at the end of one second.
Thomson thought that such devices are logically or conceptually impos-
sible. The operation of the Thomson lamp (a misnomer if Thomson were
correct) entails that (i) for any t such that 0 < t < 1, if the lamp is off
at t, then there is a t' such that t < t' < 1 and the lamp is on at t', and
(ii) for any t such that 0 < t < 1, if the lamp is on at t, then there is a t' such
that t < t' < 1 and the lamp is off at t'. Thomson thought that it followed
from (i) that the lamp cannot be off at t — 1 and from (ii) that the lamp
cannot be on at / = 1, a contradiction since it is assumed that the lamp must
be in one or the other of these states at any instant. The fallaciousness of the
argument was pointed out by Benacerraf (1962).
Others have held that though conceptually possible such devices are
103
104 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

physically impossible. Benacerraf and Putnam (1983, p. 20), for example,

seem to have thought that these devices are kinematically impossible due to
the fact that relativity theory sets c (the velocity of light) as the limit with
which the parts of the device can move. Again, however, the impossibility is
not as obvious as claimed. A demonstration is needed to rule out as a
kinematic possibility that the operation of the device is arranged so that with
each successive step the distance the parts have to move (as in an ordinary
stroll from a to b) shrinks sufficiently fast that the bound c is never violated.
Of course, even if the device can be shown to pass muster at the kinematic
level, it may still fail to satisfy necessary conditions for a dynamically possible
process (see Grunbaum 1968, 1969 for a discussion).
I have nothing new to add to this discussion here.2 Rather, my focus will
be on the ways that the relativistic nature of spacetime can be exploited so
as to finesse the accomplishment of a supertask. Very crudely, the strategy is
to use a division of labor. One observer has available to her an infinite amount
of proper time, thus allowing her to carry out an infinite task in an
unremarkable way. For example, she may check an unresolved conjecture in
number theory by checking it for '!' on day one, for '2' on day two, etc., ad
infinitum. (Or if, as the numerals increase, she needs increasing amounts of
time to complete the check, she can allow herself f ( n ) days to check the
conjecture for 'n', where f ( n ) is any increasing function of n as long as
f(ri) < oo for all n.) A second observer, who uses up only a finite amount of
his proper time, is so situated that his past light cone contains the entire world
line of the first observer. The second observer thus has direct causal access to
the infinite computation of the first observer, and in this way he obtains
knowledge of the truth value of the conjecture in a finite amount of time. If
this is genuinely possible in relativity theory, there is an irony involved. Prima
facie relativity might have been thought to make supertasks more difficult if
not impossible by imposing kinematic limitations on the workings of Thomson
lamps, Plato machines, and the like. But on further analysis relativity theory
seems to open up a royal road that leads to the functional equivalent of the
accomplishment of a supertask. The rough sketch just given contains an
unjustified optimism. We will see that relativistic spacetimes do provide
opportunities for carrying out the functional equivalents of supertasks, but
we will also see that they do so at a price. One approach is to set the supertask
in a well-behaved spacetime (see section 4.2). Here a double price has to be
paid; for the second observer who tries to take advantage of the infinite labor
of the first observer must submit himself to unbounded forces that end his
existence, and in any case he never observes the completion of the infinite
labor at any definite time in his existence.
Alternatively, both of these difficulties can be overcome by exploiting
spacetimes with unusual structures which I will dub Malament-Hogarth
spacetimes. A large part of this chapter will be devoted to articulating the
senses in which these spacetimes are physically problematic. As Hogarth has
already shown, they are not globally hyperbolic (Lemma 4.1, section 4.3), so
 Supertasks 105

that they violate strong cosmic censorship. And they may also violate other
requirements one would expect a physically realistic spacetime to fulfill
(section 4.6). It will turn out that the failure of global hyperbolicity occurs
in a way which necessarily defeats attempts to control disturbances to the
signaling between the first and second observer from singularities and other
sources. This signaling will prove to be problematic in other ways. It may
demand that the second observer pursue his own mini-supertask in his
neighborhood of spacetime, forfeiting the advantage that a Malament-
Hogarth spacetime was supposed to offer (section 4.7). Again, the signaling
will be associated with indefinite blueshifts (Lemma 4.2, section 4.5), so that
the energy of the signals can be indefinitely amplified, threatening to destroy
the second observer who receives them.

4.2 Pitowsky spacetimes

The first published attempt to make precise the vague ideas sketched in section
4.1 for using relativistic effects to finesse supertasks was that of Pitowsky
(1990). His approach is encapsulated in the following definition.

DEFINITION 4.1
M, gab is a Pitowsky spacetime just in case there are future-directed timelike
half-curves y l 5 y 2 e M- sucn tna t J yi di = oo, |V2 dx < oo, and yl cz I~(y2)-

The blandest relativistic spacetime of all, Minkowski spacetime, is Pitowskian,

as shown by Pitowsky's own example. (It seems a safe conjecture that this
example can be generalized to show that any relativistic spacetime that
possesses a timelike half-curve of infinite proper length is Pitowskian.) Choose
an inertial coordinate system (x, i). Let yl be the timelike half-geodesic
x(/) = constant, 0 < t < +00. Choose y 2 to be a timelike half-curve that
spirals around y1 in such a way that it keeps yt in its causal shadow and that
its tangential speed is u(t) — [1 — exp( — 2£)]1/2, c = 1. The proper time for
y 2 is rfi = exp( — t) dt, so that J y2 di = 1. Those familiar with the "twin
paradox" may wish to take this example as the extreme case of the paradox
with y2 as the ultimate traveling twin who ages biologically only a finite
amount while his stay-behind twin ages an infinite amount. But admittedly
this example does not conform to the usual twin paradox scenario where the
twins hold a final reunion.
Pitowsky tells the following story about this example.
While [the mathematician] M [y2] peacefully cruises in orbit, his graduate
students examine Fermat's conjecture one case after the other. . . . When
they grow old, or become professors, they transmit the holy task to
their own disciples, and so on. If a counterexample to Fermat's conjecture
is ever encountered, a message is sent to [M]. In this case M has a fraction
of a second to hit the brakes and return home. If no message arrives, M
106 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

disintegrates with a smile, knowing that Fermat was right after all. (Pitowsky
1990, p. 83)

(The example is now somewhat dated since a proof of Fermat's last theorem
has been offered. However, some lingering doubts may remain since the
purported proof is over 200 pages. In any case, the punch of Pitowsky's story
can be preserved by substituting for Fermat's theorem any unresolved
conjecture of number theory with a prenex normal form consisting either of
all universal quantifiers or else all existential quantifiers. Or the logician may
wish to contemplate the problem of deciding for a formal system strong
enough for arithmetic whether or not a given well-formed formula is a
theorem.)
There are two things wrong with this story. The first concerns the notion
that "M [y2] cruises peacefully in orbit." For ease of computation, assume
that the mathematician y2 undergoes linear acceleration with u(t) as before.
The magnitude of acceleration a(t) —• (Ab(t)Ab(t))1/2, where Ab is the four-
vector acceleration, is exp(£)/[l — exp( —2i)] 1/2 , which blows up rapidly. (To
stay within a linearly accelerating y 2 ' s causal shadow, y1 would also need to
accelerate. But y^'s acceleration can remain bounded. Indeed, y1 can
undergo constant ("Born") acceleration, which guarantees that y^'s velocity
approaches the speed of light sufficiently slowly that its proper length is
infinite.) Thus, any physically realistic embodiment of the mathematician will
be quickly crushed by ^-forces. The mathematician disintegrates with a
grimace, perhaps before learning the truth about Fermat's conjecture. What
is true in this example is true in general since any ultimate traveling twin in
Minkpwski spacetime must have unbounded acceleration. If the ultimate
traveling twin moves rectilinearly and has an upper bound to his acceleration,
then another traveler, Born accelerated at this upper bound, would achieve
equal or greater velocity at each instant and therefore age less. But this Born
accelerated traveler's world line has infinite proper length. Therefore the
rectilinearly accelerated traveller must have no upper bound to his acceleration
if he is to have finite total proper time. This result holds a fortiori for the
general case of a traveler in curvilinear motion, for part of his acceleration
will be transverse to the direction of motion, thus generating no velocity over
time and no resultant clock slowing.
The second and conceptually more important difficulty with Pitowsky's
story concerns the claim that the mathematician y2 can use the described
procedure to gain sure knowledge of the truth value of Fermat's conjecture.
If Fermat was wrong, J2 wm eventually receive a signal from y1 announcing
that a counterexample has been found, and at that moment y2 knows that
Fermat was wrong. On the other hand, if Fermat was right, y2 never receives
a signal from yt. But at no instant does y2 know whether the absence of a
signal is because Fermat was right or because y1 has not yet arrived at a
counterexample. Thus, at no definite moment in his existence does y 2 know
that Fermat was right. The fictitious mathematical sum of all of y2's stages
 Super tasks 107

knows the truth of the matter. But this is cold comfort to the actual,
non-mathematical y2. By way of analogy, if your world line y is a timelike
geodesic in Minkowski spacetime and you have drunk so deep from the
fountain of youth that you live forever, then 7~ (y) is the entirety of Minkowski
spacetime. So the fictitious sum of every stage of you can have direct causal
knowledge of every event in spacetime. But at no definite time does the actual
you possess such global knowledge.

4.3 Malament-Hogarth spacetimes

Malament (1988) and Hogarth (1992) sought to solve the conceptual problem
with Pitowsky's example by utilizing a different spacetime structure.

DEFINITION 4.2
M, gab is a Malament—Hogarth spacetime just in case there is a timelike
half-curve y1 c: M and a point p e M such that Jn dr = oo and y1 c I~ (p).

This definition contains no reference to a receiver y2. But if M, gab is a

Malament—Hogarth (hereafter, M—H) spacetime, then there will be a
future-directed timelike curve J2 from a point q e I~ (p} to p such that
In(«,p) dt < °°j where q can be chosen to lie in the causal future of the past
endpoint of yt. Thus, if yl proceeds as before to check Fermat's conjecture,
y2 can know for sure at event p that if he has received no signal from y±
announcing a counterexample, then Fermat was right.
Such arrangements can also be used to "effectively decide" membership
in a recursively enumerable but non-recursive set of integers.3 To decide
whether or not a given n is a member, y1 proceeds to enumerate the members
of the set. By assumption, this can be done effectively. As each new member
is generated, yi checks to see whether it is equal to n. This too can be done
effectively. y1 sends a signal to y2 just in case she gets a match. Consequently,
y2 knows that n is a member precisely if he has received a signal by the time
of the M-H event.
These scenarios cannot be carried out in Minkowski spacetime, as follows
from

LEMMA 4.1 An M-H spacetime is not globally hyperbolic.

A formal proof of Lemma 4.1 was given by Hogarth (1992). A simple informal
proof follows from the facts that a globally hyperbolic spacetime M, gab
contains a Cauchy surface and that a spacetime with a Cauchy surface can
be partitioned by a family of Cauchy surfaces. Suppose for purposes of
contradiction that M, gab is both globally hyperbolic and contains an M-H
point/; e M, i.e., that there is a future-directed timelike half-curve y such that
y c I~ (p) and jy di = oo. Choose a Cauchy surface Z through p, and extend
108 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 4.1 A toy Malament-Hogarth spacetime

y maximally in the past. This extended y' is also contained in / (p). Since y'
has no past or future endpoint, it must intersect Z. But then since there is a
timelike curve from the intersection point to p, S is not achronal and cannot,
contrary to assumption, be a Cauchy surface.4
What of the problem in Pitowsky's original example that the receiver y2
has to undergo unbounded acceleration? In principle, both y1 and y2 can be
timelike geodesies in at least some M-H spacetimes. The following toy
example illustrates the point and also serves as a useful concrete example of
an M—H spacetime. Start with Minkowski spacetime R4, t]ab and choose a
scalar field Q which has value 1 outside of a compact set C (see Fig. 4.1) and
which goes rapidly to + oo as the point r is approached. The M-H spacetime
is then M, gab where M = [R4 — r and gab = O 2 fj afc . Timelike geodesies of t]ab
in general do not remain geodesies in gab, but fi can be chosen so that y^ is
a geodesic of gab (e.g., if yt is a geodesic of r\ab, choose an £1 with y t as an
axis of symmetry).

4.4 Paradoxes regained?

Consider again the super n machine which is supposed to print all the digits
in the decimal expansion of n within a finite time span. Even leaving aside
worries about whether the movement of the parts of the machine can be made
to satisfy obvious kinematic and dynamic requirements, Chihara (1965)
averred that there is something unintelligible about this hypothetical machine.

The difficulty, as I see it, is not insufficiency of time, tape, ink, speed, strength
or material power, and the like, but rather the inconceivability of how the
machine could actually finish its supertask. The machine would supposedly
print the digits on tape, one after another, while the tape flows through the
machine, say from right to left. Hence, at each stage in the calculation, the
sequence of digits will extend to the left with the last digit printed being "at
 Supertasks 109

center." Now when the machine completes its task and shuts itself off, we
should be able to look at the tape to see what digit was printed last. But if
the machine finishes printing all the digits which constitute the decimal
expansion pi, no digit can be the last digit printed. And how are we to
understand this situation? (Chihara 1965, p. 80)

Note first that the baldest form of Chihara's worry does not apply to the
setup that has been imagined for M-H spacetimes; for the tape will not be
available for y 2 ' s inspection since yt goes crashing into a singularity or
disappears to infinity. However, it might seem that a more sophisticated
version of Chihara's conundrum can be mapped onto the M-H set up as
follows. y l 5 who has available to her an infinite amount of proper time, prints
the digits of 71, say, one per second. And at the end of each step she sends a
light signal to y2 announcing the result. y 2 nas a receiver equipped with an
indicator which displays 'even' or 'odd' according as the case may be. By
construction there is a p £ J2 at which y 2 nas received all of the signals from
}>!_ One can then ask: What does the indicator read at that moment?
Any attempt to consistently answer this query fails. How the failure is
reflected in any attempted physical instantiation will depend on the details
of the physics—in one instantiation the indicator device will burn out before
the crucial moment, in another the indicator will continue to display but the
display will not faithfully mirror the information sent from y 1; etc. But
independently of the details of the physics, we know in advance that the
functional description of the device is not self-consistent. Does this knowledge
constitute a general reductio of the possibility of using M-H spacetimes to
create the functional equivalents of Plato machines? No, for the inconsistency
here can be traced to the conditions imposed on one component of the n
machine—the receiver—indicator—and such conditions are not imposed in
mimicking Plato machines.
If the M-H analogue of the super n machine is to operate as intended,
then the receiver-indicator must satisfy three demands: (a) the indicator has
a definite state for all relevant values of its proper time T, (b) the indicator
is faithful in the sense that, if it receives an odd/even signal at T, then it
instantly adopts the corresponding odd/even indicator state, and (c) the
indicator does not change its state except in response to a received signal in
the sense that if Tns is a time at which no signal is received, then the indicator
state at Tns is the limit of indicator states as T approaches Tns from below. These
demands are supposed to guarantee that at the crucial moment the indicator
displays the parity of the "last digit" of n. That such a component is possible
by itself leads to contradictions if it is assumed that the receiver-indicator
device is subject to infinitely many alternating signals in a finite time. The
limit required by (c) does not always exist, contradicting (a). I take the
impossibility of such a component to be the lesson of forlorn attempts to
construct an M—H analogue of the super n machines.
Denying the use of such functionally inconsistent devices will not affect
110 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

attempts to construct M-H analogues of Plato machines and to use them to

gain new mathematical knowledge. The computer y t is an infinity machine
in the innocuous sense that it performs an infinite number of operations in
an infinite amount of proper time. I see no grounds for thinking that such
machines involve any conceptual difficulties unless they are required to
compute a non-existent quantity. The uses to which I will put them make no
such demand. Similarly, a conceptually non-problematic receiver—indicator
device can be coupled to the computer through M-H spacetime relations in
order to determine the truth values of mathematical conjectures. To flesh out
the suggestion already made above, imagine, as in Pitowsky's example, that
y1 is the world line of a computer which successively checks a conjecture of
number theory for '!', for '2', etc. Since it has available to it an infinite
amount of proper time the computer will in the fullness of time check the
conjecture for all the integers. It is arranged that ^ sends a signal to y 2 if
and only if a counterexample is found. J2 is equipped with a receiver and an
indicator device that is initially set to 'true' and which retains that state
unless the receiver detects a signal, when the indicator shifts to 'false' and the
receiver shuts off. By reading the display at the M—H point, J2 can learn
whether or not the conjecture is true. Although I can give no formal proof
of the consistency of this functional description, I see no basis for doubt.
However, I will show below that attempts to physically instantiate this
functional description run into various difficulties. But the difficulties have
nothing to do with the paradoxes and conundrums of Thomson lamps and
the like.

4.5 Characterization of Malament—Hogarth spacetimes

It was seen in section 4.3 that M—H spacetimes are not globally hyperbolic
and thus violate Penrose's version of strong cosmic censorship. The converse
is generally not true: some spacetimes that are not globally hyperbolic can
fail to be M-H spacetimes. (Trivial example: Minkowski spacetime with a
closed set of points removed does not contain a Cauchy surface but is not an
M—H spacetime.) Some M—H spacetimes are acausal. Godel spacetime is
causally vicious in that for every point p e M (= R 4 ) there is a closed
future-directed timelike curve through/; (see chapter 6). In fact, for any p e M,
I~ (p) = M. Since Godel spacetime contains timelike half-curves of infinite
proper length, every point is an M—H point. I will not discuss such acausal
spacetimes here. The reason is not because I think that the so-called paradoxes
of time travel show that such spacetimes are physically impossible; indeed, I
will argue just the opposite in chapter 6. But such paradoxes do raise a host
of difficulties which, though interesting in themselves, only serve to obscure
the issues about supertasks I wish to emphasize.
Therefore, in what follows I will restrict attention to causally well-
behaved spacetimes. In particular, all of the spacetimes I will discuss are
 Supertasks 111

stably causal, which entails the existence of a global time function (see chapter
6). I claim that among such spacetimes satisfying some subsidiary conditions
to be announced, the M~-H spacetimes are physically characterized by
divergent blueshifts. The intuitive argument for this assertion is straight-
forward. During her lifetime, yt measures an infinite number of vibrations of
her source, each vibration taking the same amount of her proper time. y2
must agree that an infinite number of vibrations take place. But within a finite
amount of his proper time, y2 receives an infinite number of light signals from
/
y 1 , each announcing the completion of a vibration. For this to happen y2
must receive the signals in ever decreasing intervals of his proper time. Thus,
y2 will perceive the frequency of y,/s source to increase without bound. (This
argument does not apply to acausal M-H spacetimes. The simplest example
to think about is the cylindrical spacetime formed from two-dimensional
Minkowski spacetime by identifying two points ( x l t ij) and (x2, t2) just in
case Xi = x2 and ti = t2 modulo n. J2 can be chosen to be some finite timelike
geodesic segment and y^ can be a timelike half-geodesic that spirals endlessly
around the cylinder. The light signals from y t may arrive at y2 all mixed up
and not blueshifted.)
The main difficulty with this informal argument, as with all of the early
literature on the redshift/blueshift effect (see Earman and Glymour 1980) is
that the concept of frequency it employs refers to the rate of vibration of the
source at yl and to the rates at which y^ sends and y2 receives signals. But
the effect actually measured by y2 depends on the frequency of the light signal
(photon) itself. Thus, we need to calculate the blueshift using the definition
of the emission frequency of a photon from a point/)! e y-i aso^ =: — (ka V 1 a ) \ p l
and the measured frequency of the photon as received at the point p2 € y2 as
eo2 =; — (ka V2") P2, where the timelike vectors J^° and V2" are respectively the
normed tangent vectors to the world lines yl and y2, and the null vector k"
is the tangent to the world line of the photon moving from the first to the
second observer (see Fig. 4.2). Then the redshift/blueshift effect is given by
the ratio

The following key fact is established in the appendix at the end of this
chapter.

LEMMA 4.2. Let M, gab be a Malament-Hogarth spacetime containing a timelike

half-curve yt and another timelike curve y2 from point q to point p such that Jn dx = oo,
f y 2 di < co, and yl c: I~ (p). Suppose that the family of null geodesies from y1 to y2
forms a two-dimensional integral submanifold in which the order of emission from y±
matches the order of reception aty2. If the photon frequency (O^ as measured by the sender
yl is constant, then the time-integrated photon frequency |P2 a>2 di as measured by the
receiver y2 diverges as p2 approaches p.
112 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 4.2 The redshift/blueshift effect

Parameterize y2 by a I such that y 2 ' s Past endpoint q corresponds to

t = 0 and p corresponds to t=\. Then it follows from Lemma 4.2 that
lim,_ <1 co2(t) = oo if the limit exists. If not, then lim,.^ colub(t) = oo, where
(Olub(t) =:lub{co 2 (0 : 0 < t' < t}. Thus, one can choose on y2 a countable
sequence of points approaching p such that the blueshift as measured by y2
at those points diverges. Typically this behavior will hold for any such
sequence of points on y2, but there are some mathematically possible M-H
spacetimes where y2 measures no red or blueshift at some sequence of points
approaching p.
The following example (due to R. Geroch and D. Malament) illustrates
this counterintuitive feature. As in the toy model in Fig. 4.1, start with parallel
timelike geodesies of Minkowski spacetime. Parameterize y^ by the proper
time T of the Minkowski metric and adjust the curve so that the past endpoint
corresponds to T = 0 and r corresponds to T = 1. At the points on y^
corresponding to T = TB = 1 — (3/4) (1/2") draw a sphere of radius rn = l/2" +3
(as measured in the natural Euclidean metric). On the nth sphere put a
conformal factor Qn which goes smoothly to 1 on the surface of the sphere
and which has its maximum value at the point on yt corresponding to TB.
Construct the Qn such that the proper time along y: in the conformal metric
O,2tjab is infinite. For instance, if y" is the part of y1 within the nth sphere,
set Qn so that J r j Qn di = 1. The result is an M-H spacetime. But at the
points on y2 that receive photons from the points on y1 corresponding to
T = 1/2, 3/4, 7/8, etc., there is no blue- or redshift.
While mathematically well defined, such examples are physically patho-
logical. In particular, I do not know of any examples of M-H spacetimes
which are solutions to Einstein's field equations for sources satisfying standard
 Supertasks 113

energy conditions (see section 4.6) and which have the curious feature that
the blueshift as measured by y2 diverges along some but not all sequences of
points approaching the M-H point. Thus, although the slogan that M-H
spacetimes involve divergent blueshifts is literally incorrect, it is essentially
correct in spirit.
It may help to fix intuitions by computing the blueshift in some concrete
examples. For the toy model pictured in Fig. 4.1 the result is

This ratio diverges as yt approaches the (missing point) r and y2 approaches

the M-H point p.
Another stably causal M-H spacetime is obtained by taking the universal
covering of anti-de Sitter spacetime (Hawking and Ellis 1973, pp. 131-134).
Suppressing two spatial dimensions, the line element can be written as
ds2 = dr2 — (cosh2 r) dt2. Following Hogarth (1992) we can take y2 to be given
by r = r2 = constant and y^ to be given by a solution to dr/dt = cosh r/^/2
(see Fig. 4.3). The blueshift is

which diverges as rt -* oo and p2 approaches the M-H point p.

We can also pose the converse question as to whether a divergent blueshift
behavior indicates that the spacetirne has the M—H property. The answer is
positive in the sense that the proof of Lemma 4.2 can be inverted.
The fact that an M—H spacetime gives an indefinitely large blueshift
for the photon frequency implies that the spacetime structure acts as
an arbitrarily powerful energy amplifier. This might seem to guarantee

Fig. 4.3 Anti-de Sitter spacetime is a Malament-Hogarth spacetime

114 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

unambiguous communication from y^ to y2. But this first impression neglects

the fact that a realistic instantiation of yl will have thermal properties. The
slightest amount of thermal radiation will be amplified indefinitely, which
will tend to make communication impossible. In order not to destroy the
receiver at y2, yt will have to progressively reduce the energy of the photons
she sends out. This means that there will be a point at which the energy of
the signal photons will be reduced below that of the thermal noise photons.
The indefinite amplification of the thermal noise will in any case destroy the
receiver. Perhaps this difficulty can be met by cooling down yt so as to
eliminate thermal noise or by devising a scheme for draining off the energy
of the signal photon while in transit. But even assuming a resolution of this
difficulty, still further problems dog the attempt to use M—H spacetimes to
accomplish supertasks.

4.6 Supertasks in Malament-Hogarth spacetimes

Are supertasks in Malament—Hogarth spacetimes to be taken seriously? The

question involves three aspects. The first concerns whether M—H spacetimes
are physically possible and physically realistic. As a necessary condition for
physical possibility, general relativists will want to demand that the spacetime
be part of a solution to Einstein's field equations for a stress—energy tensor
Tab satisfying some form of energy condition, weak, strong, or dominant (see
chapter 3). The toy model of Fig. 4.1 can be regarded as a solution to
Einstein's field equations with vanishing cosmological constant A by computing
the Einstein tensor Gab(g) and then defining Tab ~ (l/8n)Gab, But as conjec-
tured in chapter 3, such models may be ruled out by the energy conditions.
Anti-de Sitter spacetime, another M—H spacetime, can be regarded as a
vacuum solution to Einstein's field equations with A = R/4, R (<0) being
the curvature scalar; then the energy conditions are trivially satisfied.
However, if it is required that A = 0, anti-de Sitter spacetime is ruled out by
the strong energy condition if a perfect fluid source is assumed (see chapter 3).
None of these concerns touch Reissner—Nordstrom spacetime which is the
unique spherically symmetric electrovac solution of Einstein's field equations
with A = 0 (Hawking and Ellis 1973, pp. 156-161). Since this spacetime is
an M—H spacetime, at least some M—H spacetimes meet the minimal
requirements for physical possibility.
It is far from clear, however, that M-H spacetimes meet the (necessarily
vaguer) criteria for physically realistic spacetime arenas. For one thing, it was
seen in the preceding section that M-H spacetimes involve divergent
blueshifts, which may be taken as an indicator that these spacetimes involve
instabilities. Such is the case with Reissner—Nordstrom spacetime, where a
small perturbation on an initial value hypersurface £ (see Fig. 4.4) can
produce an infinite effect on the future Cauchy horizon //+ (X) of TL (see
Chandrasekhar and Hartle 1982).
 Supertasks 115

Fig. 4.4 Reissner-Nordstrom spacetime is a Malament-Hogarth spacetime

For another thing, various M-H spacetimes run afoul of one or other
versions of Penrose's cosmic censorship hypothesis. By Lemma 4.1 all M—H
spacetimes violate strong cosmic censorship, and many examples of M-H
spacetimes violate weaker versions as well. Thus, evidence that cosmic
censorship holds for physically reasonable spacetimes is ipso facto evidence
against the physical reasonableness of M—H spacetimes. Conversely, one
might take the rather bizarre scenarios that can be concocted in M-H
spacetimes as grounds for thinking chat a cosmic censor should be at work.
But then again, those with a taste for the bizarre may hope that cosmic
censorship fails just so that they can own the functional equivalent of a Plato
machine.
I now turn to the second aspect of the question of how seriously to
take the possibility of completing supertasks in M—H spacetimes. This aspect
concerns whether y t can be implemented by a physically possible/physically
realistic device which, over the infinite proper time available to it, carries out
the assigned infinite task. Once again the question is made difficult by the
fact that there is no agreed upon list of criteria that identify physically realistic
devices. I will make the task of tackling this question tractable by confining
attention to dynamical constraints that physically realistic yl should satisfy.
(One doesn't have to worry about dynamical constraints on y2 since typically
72 can be chosen to be a geodesic.) Minimally, the magnitude of acceleration
of ^ must remain bounded, otherwise any device that one could hope to
build would be crushed by ^-forces. This condition is satisfied in the anti-de
Sitter case (Fig. 4.3) where a(rt) = ^2 [exp^) - l]/[exp(2/i) + 1], which
approaches ^y2 as rt —> oo. However, we must also demand a finite bound on
the total acceleration ofy^. TA(y t ) =: J yi a dr. For even with perfectly efficient
rocket engines, the final mass m, of the rocket and the mass "fuel of
116 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

the fuel needed to accelerate the rocket must satisfy (see the appendix to this
chapter)

Thus if TA(y 1 ) = oo an infinite amount of fuel is needed for any finite

payload. In the anti-de Sitter case, drv = dx so that TA(y1) = oo, and the
demand fails. In the toy model of Fig. 4.1 the demand is met since
TA(7j) = 0, 7J being a geodesic; but the spacetime involved was ruled out as
not physically possible. In Reissner-Nordstrom spacetime a timelike geodesic
y1 can be chosen to start on the time slice S (see Fig. 4.4) and to go out to
future timelike infinity i+ .5 And y1 a I+ (p) for an appropriate pointy 6 H+ (Z).
But again there are reasons to regard this spacetime as not being physically
realistic.
Finally, since a physically realistic device must have some finite spatial
extent, we are really concerned not with a single world line y1 but with a
congruence T1 of world lines. Even if Tl is a geodesic congruence it cannot be
instantiated by a physically realistic computer (say) unless the tidal forces it
experiences remain bounded. Since the tidal forces are proportional to the
Riemann curvature tensor,6 one can satisfy this demand in Reissner—Nordstrom
spacetime, which is asymptotically flat in the relevant region. One simply
starts the geodesic congruence sufficiently far out towards spatial infinity and
has it terminate on future timelike infinity i+.
To summarize the discussion up to this point, it is not clear that any M-H
spacetime qualifies as physically possible and physically realistic. But to the
extent that M—H spacetirnes do clear this first hurdle, it seems that the role of
7i can be played by a world line or world tube satisfying realistic dynamical
constraints. However, Pitowsky (1990) feels that, for other reasons, yt cannot
be instantiated by a computer that will carry out the assigned infinite task.
I will take up his worry in section 4.8 below. Before doing so I turn to the
third aspect of whether M—H can be taken seriously. It concerns discrimina-
tions that the receiver y2 must make.

4.7 Malament-Hogarth spacetimes and

unresolved mathematical conjectures

Can Malament-Hogarth spacetimes be used to gain knowledge of the truth

values of unresolved mathematical conjectures? Suppose now for sake of
discussion that some M-H spacetimes are regarded as physically possible and
physically realistic and that in these arenas there are no barriers to a physically
possible and physically realistic instantiation ofyl by a computer which carries
out the task of checking Fermat's last theorem or some other unresolved
conjecture of number theory. Nevertheless there are reasons to doubt that y2
 Supertasks 117

can use y^s work to gain genuine knowledge of the truth value of the
conjecture. The pessimism is based on a strengthening of Lemma 4.1.

LEMMA 4.3. Suppose that p e M is a M-H point of the spacetime M, gab (that is,
there is a future-directed timelike half-curve yl c M such that §yi di = oo and
y! cr /"(/>)). Choose any connected spacelike hyper surf ace E cr M such that yl c I+ (£).
Then p is on or beyond H+ (E).

PROOF: Ifp e int[£> + (E)] then there is a q e D+ (E) which is chronologically

preceded by p. M' —•• (J~ (q) n / + (E)) c Z) + (E), and the smaller spacetime
M',gab\M. is globally hyperbolic. Choose a Cauchy surface E' for this smaller
spacetime which passes through p. Since y1 c M' we can proceed as in the
proof of Lemma 1 to obtain a contradiction.

Lemma 4.3 is illustrated by the Reissner-Nordstrom spacetime (Fig. 4.4).

Any M-H point involved with a y : starting in region I must lie on or beyond
// + (E).
Think of E as an initial value hypersurface on which one specifies initial
data that, along with the laws of physics, prescribes how the computer y1 is
to calculate and how it is to signal its results to y 2 . Since by Lemma 4.3 any
M—H point p € y 2 must lie on or beyond H+ (E) for any appropriate E, events
at p or at points arbitrarily close to p are subject to non-deterministic
influences. In typical cases such as the Reissner-Nordstrom spacetime
illustrated in Fig. 4.4 there are null rays which pass arbitrarily close to any
p e H+ (E) and which terminate in the past direction on the singularity. There
is nothing in the known laws of physics to prevent a false signal from emerging
from the singularity and conveying the misinformation to y 2 that a counter-
example to Fermat's conjecture has been found.7 (y 2 need not measure an
infinite blueshift for photons emerging from the singularity; at least there is
nothing in Lemma 4.2 or the known laws of physics that entails such a
divergent blueshift.) Of course, the receiver y 2 can ignore the signal if he
knows that it comes from the singularity rather than from y t . But to be able
to discriminate such a false signal from every possible true signal that might
come from y1, y2 must be able to make arbitrarily precise discriminations. In
the original situation it was the Plato machine that had to perform a supertask
by compressing an infinite computation into a finite time span. The trick
adopted here was to finesse the problems associated with such a supertask by
utilizing two observers in relativistic spacetime. But we have found that the
finesse also involves a kind of supertask—not on the part of the computer but
on the part of the receiver who tries to use the work of the computer to gain
new mathematical knowledge.
This verdict may seem unduly harsh. If y2 is to be sure beforehand
that, whatever y/s search procedure turns up, he will obtain knowledge of
the truth value of Fermat's conjecture, then y 2 must be capable of arbitrarily
precise discriminations. But, it may be urged, if y 2 is capable of only a finite
118 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

degree of precision in his signal discriminations, he may yet learn that

Fermat's conjecture is false (if indeed it is) if he receives a signal long enough
before the M—H point so that it lies within his discrimination range. This,
however, would be a matter of good fortune. One can pick at random a
quadruple of numbers (x,y, Z,n), n > 3, and check whether x" +j>" = z". If
one is lucky, a counterexample to Fermat's conjecture will have been found.
But the interest in Platonist computers and their M—H analogues lay in the
notion that they do not rely on luck.
Of course any observer faces the problem of filtering out spurious back-
gound signals from those genuinely sent from the system observed. But it is
usually assumed that sufficiently thorough attention to the experimental setup
could at least in principle control all such signals. What Lemma 4.3 shows,
however, is that no such efforts can succeed even in principle in our case. No
matter how carefully and expansively we set up our experiment—hat is, no
matter how large we choose our initial value hypersurface—we cannot
prevent spurious signals from reaching p or coming arbitrarily close to p.
The problem can be met by means of a somewhat more complicated
arrangement between y^ and y2 by which yi not only sends a signal to y2 to
announce the finding of a counterexample but also encodes the quadruple of
numbers that constitutes the counterexample. A false signal may emerge from
the singularity, but y 2 can discover the falsity by a mechanical check. With
the new arrangement J2 no longer has to discriminate where the signal came
from since a counterexample is a counterexample whatever its origin.
Unfortunately, y2 may still have to make arbitrarily fine discriminations since
the quadruple sent will be of arbitrarily great size (= number of bits) and
tnust be compressed into a correspondingly small time interval at J2.
The worry about whether 72 can gain knowledge of Fermat's conjecture
by using yt's efforts also involves the concern about y2's right to move from
'y1 has not sent me a signal' to 'Fermat's conjecture is true'. The correctness
of the inference is not secured by the agreement y1 and y2 have worked out,
for even with the best will in the world yt cannot carry out her part of the
agreement if events conspire against her. As suggested above, the most
straightforward way to underwrite the correctness of the inference is for there
to be a spacelike £ such that y^ c Z) + (E) and such that initial conditions on
£ together with the relevant laws of physics guarantee that y1 carries out her
search task. And if, as is compatible with at least some M-H spacetimes (e.g.,
Reissner—Nordstrom spacetime), the M—H point p can be chosen so that
S c I~(p), it would seem that y2 could in principle come to know that the
conditions which underwrite the inference do in fact obtain. But the rub is
that p or points arbitrarily close to p may receive a false signal from the
singularity, indicating that conditions are not conducive to yt's carrying out
her task. If so, y2 is not justified in making the inference unless he can
discriminate false signals as such. This, of course, is just another version
of the difficulty already discussed. But the present form docs not seem to have
an easy resolution.
 Super tasks 119

4.8 Can ^j carry out the assigned task?

y1 is supposed to check an unresolved conjecture of number theory for each

of the integers. By construction, yl has time enough. But Pitowsky feels that
}>! never has world enough.
The real reason why Platonist computers are physically impossible even in
theory has to do with the computation space. According to general relativity
the material universe is finite. Even if we use the state of every single
elementary particle in the universe, to code a digit of a natural number, we
shall very soon run out of hardware. (Pitowsky 1990, p. 84)

In response, I note that general relativity theory does not by itself imply a
spatially or materially finite universe. Further, it was seen that there are
spatially infinite M-H spacetimes, such as Reissner-Nordstrom spacetime,
that are live physical possibilities in the minimal sense that they satisfy
Einstein's field equations and the energy conditions. A y1 who wanders off
into the asymptotically flat region of this spacetime certainly has space enough
for any amount of hardware she needs to use. But she cannot avail herself of
an unlimited amount of hardware without violating the implicit assumption
of all of the foregoing; namely, that J1 and y2 have masses so small that they
do not significantly perturb the background metric. Here Pitowsky's objection
has some bite.
Perhaps there are solutions to Einstein's field equations where the
spacetime has the M—H property and there is both space enough and material
enough for a physically embodied computer with an unlimited amount of
computation space.8 Pending the exhibition of such models, however, one
must confine oneself to tasks that can be accomplished in an infinite amount
of time but with a finite amount of computation space. Whether there are
such tasks that deserve the appellation 'super' remains to be seen.

4.9 Conclusion

Thomson lamps, super it machines, and Platonist computers are playthings

of philosophers; they are able to survive only in the hothouse atmosphere of
philosophy journals. In the end, M—H spacetimes and the supertasks they
underwrite may similarly prove to be recreational fictions for general
relativists with nothing better to do. But in order to arrive at this latter
position requires that one first resolve some of the deepest foundation problems
in classical general relativity, including the nature of singularities and the fate
of cosmic censorship. It is this connection to real problems in physics that
makes them worthy of discussion.
There are also connections to the philosophy of mathematics and
to the theory of computability. Because of finitist scruples, some philosophers
have doubted that it is meaningful to assign a truth value to a formula of
120 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

arithmetic of the form (3;q) (3#2) . . • (3xn)F(xl, x2, • • • , xn). It seems to me

unattractive to make the truth of mathematical statements depend on the
contingencies of spacetime structure. The sorts of arrangements considered
above can be used to decide the truth value of assertions of arithmetic with
a prenex normal form that is purely existential or purely universal.9 (Fermat's
last theorem, for example, has a purely universal form.) For such an assertion
y1 is set to work to check through the (countably infinite) list of ra-tuples of
numbers in search of a falsifier or a verifier according as the assertion to be
tested is universal or existential, and y1 reaps from these labors a knowledge
of the truth value of the assertion. But as soon as mixed quantifiers are
involved, the method fails.10 However, Hogarth (1994) has shown how more
complicated arrangements in general relativistic spacetimes can in principle
be used to check the truth value of any arithmetic assertion of arbitrary
quantificational complexity. Within such a spacetime it is hard to see how
to maintain the attitude that we do not have a clear notion of truth in
arithmetic.11
The computational arrangements between yi and y2 envisioned might
also seem to bring into doubt Church's proposal that effective/mechanical
computability is to be equated with Turing computability or recursiveness,
for apparently y1 and y 2 can in concert obtain a resolution to recursively
unsolvable problems by means that certainly seem to merit the appellations
of 'effective' and 'mechanical'. But putting the matter this way is a little
unfair to Church since any account of effective/mechanical computability that
implies that there are subsets of numbers which can be effectively/mechanically
enumerated, but whose complements cannot be, will be subject to the
one-upmanship of bifurcated supertasks. Perhaps the most illuminating
way to state the moral to be drawn from bifurcated supertasks is that two
levels of computation need to be distinguished: the first corresponding to what
the slave computer yl can do, the second to what y2 can infer by having
causal access to all of y/s labors. Church's proposal is best construed as aimed
at the first level and as asserting that Turing computability is an upper bound
on what any physical instantiation o f y t can accomplish. Read in this way,
there is nothing in present concerns to raise doubts about Church's proposal.12

Appendix: Proofs of Lemma 4.2 and Equation 4.4

PROOF OF LEMMA 4.2. The null geodesies from y t to y 2 form a two-

dimensional submanifold. For each of the null geodesies select an affine
parameter A which varies from 0 at yt to 1 at y2. (This will always be possible
since an affine parameter can be rescaled by an arbitrary linear transforma-
tion.) The null propagation vector k" = (d/dk}a satisfies the geodesic equation
 Supertasks 121

By supposition, these null geodesies form a submanifold. By connecting points

of equal A values, form a family of curves indexed by A that covers the
submanifold and interpolates between y^ and y2. Select any parameterization
t of }>! and propagate this parameterization along the null geodesies to all the
interpolating curves so that each null geodesic passes through points of equal
I value. The indices A and t form a coordinate system for the two-manifold.
k" and C° = (S/St)a are its coordinate basis vector fields, which entails that
they satisfy the condition [£, K\" = 0 so that

It follows that (£„£") is a constant along the photon world lines. To show this
it needs to be demonstrated that

This is done by computing

The second term on the right-hand side of ( A4.4) vanishes in virtue of ( A4. 1 ) .
Equation (A4.2) can then be used to rewrite the first term on the right-hand
side as ^akbVakb = %aVtt(kbkb) = 0 since k" is a null vector.
Thus, for a photon sent from yt to y2, we have ka£,\ = kj,\, or
kaV\^2\ = kaVa\£"2\, where V =:C7IC" is the normed tangent vector to the
timelike world line. So from the definition (4.1) of photon frequency ratios
one can conclude that cOjICil = Oald which implies that

or

But J yi <h — co and J V2 rfz < oo. So if a>1 is constant along y^ (A4.6) can
hold only if J tu2 rfr = oo.

PROOF OF (4.4) (from Malament 1985). If V is the (normalized) four-

velocity of the rocket and m its mass, the rate at which its energy—momentum
changes is VVp(mV), which must balance the energy-momentum J" of its
exhaust (it being assumed that the rocket's motor is the only source of
122 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

propulsion). Thus,

where A" = VmVmV" is the four-acceleration. Since J" is not spacelike,

J"Jn < 0. Consequently,

which uses V"Vn = - 1, V"An = 0, and a =• (A"An)1/2. Furthermore, because

the rocket is using up fuel, VVpm < 0. Thus,

So if nil and m.f are the initial and final masses of the rocket, integration of
(A4.9) yields

And since mt = ms + nifuel,

Notes
1. This task is taken up in Earman and Norton (1994).
2. A few new wrinkles are added in Earman and Norton (1994) concerning Ross's
paradox (see Allis and Koctsier 1991; van Bendegem 1994) and some other paradoxes
of the infinite.
3. A subset of S c N is said to be recursively enumerable (r.e.) just in case it is
the range of a (partial) recursive function f : N — > N ; informally, this means that there
is an effective procedure for generating the members of S. S is said to be recursive
if both S and the complement of S are r.e.; informally, there is an effective
procedure for deciding membership in S. Key results on undecidability of formal
systems hinge on there being sets that are r.e. but not recursive.
4. David Malament has pointed out that a quick proof of Lemma 4.1 can be
obtained by using Prop. 6.7.1 of Hawking and Ellis (1973): For a globally hyperbolic
spacetime, if pej+(q), then there is a non-spacelike geodesic from q to p whose
length is greater than or equal to that of any other non-spacelike curve from
q to p. Suppose that y E 1~ (p) is a timelike half-curve with endpoint q and that
Jy di = oo. Since the endpoint q of y belongs to I~ (p), we could apply the proposition
to p and q if the spacetime were globally hyperbolic. But then a contradiction results,
since whatever the bound on the length of the timelike geodesic from q to p, we could
exceed it by going along y sufficiently far and then over to p. Robert Wald noted
 Supertasks 123

that an even quicker proof is obtained from the compactness of J~ (/>) n J+ (q) together
with strong causality, which are consequences of global hyperbolicity.
5. In Fig. 4.4 i° labels spatial infinity and J+ and J~ respectively label future
and past null infinity.
6. See Wald (1984, pp. 46-47) for a derivation of the formula for geodesic
deviation.
7. One might also worry that a burst of noise from the singularity could swamp
an authentic signal. But since any real signal arrives at y2 prior to the singularity
noise, the former is not masked by the latter as long as the receiver can discriminate
between a signal and noise.
8. The considerations raised here are similar to those discussed by Barrow and
Tipler (1986) under the heading of "omega points."
9. Assuming that the relation quantified over is effectively decidable.
10. Showing this requires a more careful specification of how bifurcated infinity
machines operate; see Earman and Norton (1994).
11. See Earman and Norton (1994) for more discussion of this and related
matters.
12. But there are independent reasons to doubt Church's proposal; see Earman
(1986) and Pitowsky (1990).
 5
The Big Bang and
the Horizon Problem

5.1 Introduction

Discussions from the recent astrophysics literature on observability, horizons,

and the like have trickled down to the level of philosophical consciousness,
at least in a folkloric way. For example, the folklore contains the wisdom that
the standard big bang cosmological models contain particle horizons, that
when this fact is coupled with the observed isotropy of the 3 K cosmic
background radiation there arises a "horizon problem," and that inflationary
cosmology solves this problem. Like most folklore, this example contains some
truth and a number of distortions. There is a need to set the record straight
and to correct some fundamental misimpressions about observability and
horizons in relativistic cosmological theories. But more is at stake than
correcting some misimpressions. The horizon problem provides an interesting
test case for accounts of scientific explanation, for the perception of a
"problem" in connection with particle horizons in standard big bang
cosmology depends on views about what features a good scientific explanation
should have. While this matter has received little attention in the philosophical
literature, it has played an important role in guiding research in relativistic
cosmology. The horizon problem is also a good test case for the principle of
common cause since cosmological models with particle horizons provide
examples where there is no common causal past for correlated events.
Sections 5.2 through 5.4 give a preliminary treatment of observability
and horizons in relativistic cosmological models and introduce the concepts
needed to assess the horizon problem. That problem is introduced in
section 5.5 and then elaborated and diagnosed in sections 5.6 through 5.9.
Section 5.10 reviews various strategies that astrophysicists have adopted in
attempts to reach a resolution. Section 5.11 treats in detail the currently
favored approach, inflationary cosmology. Section 5.12 confronts the issue of
whether or not inflationary scenarios resolve the horizon problem. Section
5.13 contains conclusions arid closing remarks.
124
 The Big Bang and the Horizon Problem 125

5.2 Observability and light cones

We receive most of our information about the large-scale structure of

spacetime on a series of light cones. How much does the information which,
in principle, we can gather on these light cones allow us to infer about the
structure of space and time? Remarkably, GTR yields the answer that the
light cone data in conjunction with EFE allow us in principle to determine
the spacetime geometry on and inside our past light cone (see Kristian and
Sachs 1966; Ellis, Maartens, and Nel 1978).
Folkloric understanding of relativity theory says that this is all that direct
observation and the laws of physics allow us to know about. As usual, idealize
an observer as a timelike world line y. Then two folklore dogmas can be
stated. The first dogma is that at p 6 y, an observer whose world line is y
cannot have direct observational knowledge of events at a spacetime location
q £ J~ (p).1 This dogma seems to be firmly rooted in good common sense; for
whatever else observation means, it must involve a causal link between
observer and the events observed, and the fact that q $ J~ (p) means that q
cannot be joined to p by a causal signal. (The possibility of faster-than-light
particles is ignored here.) The second dogma is that if secure knowledge is
limited to what can be deduced from direct observational knowledge plus the
laws of physics, then secure knowledge at p does not extend to events at
q £ J~ (p) because the state at q is not determined by the laws on the basis
of the state in J~' (p) and because, more generally, relativistic laws are
inconsistent with action at a distance and, thus, do not constrain events
happening in relatively spacelike regions of spacetime.
These folklore dogmas are, with some caveats and qualifications, basically
correct. Seeing where the caveats and qualifications are needed helps to
prepare the discussion of the horizon problem. The more sophisticated folklore
takes account of an exception to the second dogma that arises when J~ (p)
contains a Cauchy surface £. In that case the observed state on £ plus the
coupled Einstein—matter equations will determine the state throughout the
spacetime.2 A necessary but not sufficient condition for such a case is that
the universe be spatially finite in the sense that £ is compact; and, since £ is
a Cauchy surface, the spacetime manifold M must be diffeomorphically £ x R
(Dieckmann 1988) as illustrated by Fig. 5.1. The spacetime in this illustration
is created by artifice—specifically, by identifying points in a larger spacetime.
One would expect that examples of spacetimes which contain a (necessarily)
compact Cauchy surface contained in the causal past of a point and which
are not created by this artifice are relatively rare among the solutions to
Einstein's field equations.
The second dogma is mistaken in a more fundamental way: paradigm
examples of relativistic laws, such as Maxwell's equations of electromagnetism
and EFE for gravitation, do constrain relatively spacelike events. In both
cases the full set of equations can be divided into constraint equations, which
impose conditions on instantaneous initial data, and dynamical equations that
126 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 5.1 At p sure prediction of events at q is possible

govern the temporal evolution of the data. 3 And in both cases the dynamical
equations entail that if the constraint equations are satisfied for one instant,
then they are satisfied at all instants. The simplest illustration of how the
constraint equations work concerns electromagnetism in the special theory of
relativity (STR). The constraint equations are V'E = p and V'B = 0, where
E and B are respectively the electric and magnetic field strengths and p is
the electric charge density. Referring to Fig. 5.2, imagine that in the slice

Fig. 5.2 Charged particles interacting in Minkowski spacetime

t — tN (jVfor "now") of Minkowski spacetime a sphere centered on q is drawn.

(Here t is the time coordinate from some inertial coordinate system.) It follows
from the first constraint equation that the electric flux through the sphere
must be equal to the charge inside the sphere—in this case the charge on the
particle whose world line is d. Conversely, knowledge of E at the spacetime
points lying on the sphere allows one to infer the charge contained inside,
illustrating how relatively spacelike electromagnetic events are mutually
constrained by the laws of electromagnetism.
A modification of this example also serves to challenge the first dogma
of observation. Create a past-truncated Minkowski spacetime by deleting all
those spacetime points r such that t(r) < 2000 B.C. (see Fig. 5.3). Suppose that
y and § are the only particles in the universe and that both are electrically
charged. Then although at p the observer whose world line is y cannot see 6
 The Big Bang and the Horizon Problem 127

Fig. 5.3 Charged particles interacting in past-truncated Minkowski spacetime

since jf~ (p) n 8 = 0, this observer can feel d (see Ellis and Sciama 1972).
For by the above argument, the Coulomb field of d will have to be non-zero
•dtp (or at any rate at a set of points of positive measure on any closed surface
in t = tN containing q, so p might just as well be one of these points).
The first dogma can be defended in two ways against the notion that
this feeling is observing. It can be reiterated that observation requires
causal connection, and then it can be asserted that since by construction
J~ (p) n 3 — 0, the causal connection is missing. But this defense begs the
question at issue. Granted that in the present example there is no causal link
forged by a causal signal, there certainly is a lawlike connection between the
unseen d and the tug felt by y at p. Trying to decide whether this lawlike
connection amounts to a causal connection by appealing to existing philo-
sophical analyses of causation does not look promising. For instance, Lewis'
(1975) counterfactual analysis of causation would have us try to decide the
matter by asking (roughly) whether the nearest possible world to that of Fig.
5.3 in which there is no charge at q is also a world in which y feels no tug at
p. My intuition is that the answer is yes. But I do not put much store in such
intuitions, and if issues about observability really turn on such subjective and
context dependent notions as nearness or similarity of possible worlds, then
arguing over them does not seem worth the candle.
A second and better defense of the first dogma would start with the
position that genuine observational knowledge requires not only true belief
but also that the belief has been formed by a reliable process. But just from
the electric force felt by y at/), y cannot reliably infer the existence of a charged
particle outside of J~ (p)', for the force might be due to source-free electro-
magnetic radiation (which in any case must be present in the example of Fig.
5.3, as will be discussed in section 5.8). By exploring more of the electric field,
y can make a more reliable inference to the existence of an unseen charge;
but by the time the exploration is extensive enough to make the inference
completely secure, signals from <5 will have had time to reach y.
This defense fails in some spacetimes. Consider a modification of the
128 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Minkowski metric in which the line element becomes ds2 = *¥(x,j/, z) x

(dx2 + dy2 + dz2} — dt2.4 If the scale factor *¥(x,y, z) is chosen appropriately,
the sphere through q can have a small area even though p is a large distance
from q. Then before a signal from 6 can reach him, our observer may be able
to sample the electric field on the entire sphere and, thus, can obtain certain
knowledge of the existence of a charge within the sphere. However, because
of the special nature of this example, the proponents of the standard dogmas
may not be much moved. For most purposes then, I think we can agree with
the folklore dogma that the limits of the observable at p are set by J~ (p).
The past-truncated Minkowski model of Fig. 5.3 is, of course, very
artificial (at least for those who do not subscribe to the creationist line of
Protestant fundamentalism). But it does serve to illustrate some general
features of the particle horizons that occur naturally in general relativistic
cosmological models. And in the electromagnetic example the astute reader
can already detect the seeds of a problem that will be discussed in detail
below in later sections.

5.3 What can we predict about the future?

For present purposes take prediction to mean deterministic prediction from

the laws of physics. This implies that in a spacetime M, gab if events at/i 6 M
are to be predicted from the state in a region X c: M, then it is necessary that
p belong to the future domain of dependence D+ (X) of X; for itp$D + (X)
there will be possible causal influences that could effect the state at p without
registering on X. If £ is a Cauchy surface for M,gab then Z) + (E) includes
every point in the spacetime to the future of Z. So we may now put the
question: In the deterministic sense of prediction, what can an observer y
predict at some point pey from her knowledge of events in J~ (p)? In the
special case where there is a Cauchy surface £ c: J~ (p) the observer is in a
position to give a deterministic prediction of the entire future of the universe.
But in general the answer is: Nothing!5 The formal point is that in Minkowski
spacetime and in typical general relativistic spacetimes, D+ (J~(p)) = J~(p)
for any spacetime point p so that the only events which can be predicted from
p are those that, from the perspective ofp, have already happened—which is
to say that the prediction is not a genuine prediction. A concrete example to
illustrate the point for Minkowski spacetime is given in Fig. 5.4. In section
5.2 we saw that it is not true that there can be nothing in J~ (p} to alert y
to the presence of the photon a and the massive particle /? whose world lines
are past endless but never enter J ~ ( p ) . But the information furnished by
typical relativistic constraint equations is weak, and in general the information
will not be strong enough to tell y for sure whether a photon or a massive
particle will intersect her world line at some chosen point to the future of p.6
What then is the status of the forecasts we routinely make about the
future? While the success of our predictions is due to more than lucky guessing,
 The Big Bang and the Horizon Problem 129

Fig. 5.4 y's prediction at p of her future is foiled by the photon a and the massive
particle $

there is also a large element of wishful thinking that gets dignified by such
big names as the Copernican principle. According to this principle, we do
not occupy a privileged position in the universe. Thus, if no world lines of
photons or massive particles enter J~ (p), we may count ourselves as
reasonable in concluding at p that no such particles exist outside of J~ (p)
and that we will not be rudely surprised as is the observer in Fig. 5.4 who
relies on this principle at p. Put in the negative mode, this principle is
appealing—it modestly denies that we have some special status. But this
seeming modesty is belied by the immodest use to which the principle was
put in justifying an inductive extrapolation.
Can we do better than appeal to big name principles? Perhaps we can
have reasons to think that there are no source-free photons like a in Fig. 5.4.
If so, we are on the road to a more secure forecast. For the only way that a
(past endless) world line of a massive source of photons could fail to fall into
J~ (p) is for it to behave like /? in Fig. 5.4. But the total integrated acceleration,
from t = — oo to the present, of a world line like j9 is infinite. Thus, an infinite
amount of energy would be needed to produce such a trajectory; or, if /? were
the world line of a rocket ship, an infinite fuel-to-payload ratio would be
required. This makes promising the prospect of proving results to the effect
that prediction will not be undermined by well-understood physical mechan-
isms. For example, one could try to prove that a trajectory like that of ft
cannot result from electromagnetic forces produced by a system of charged
particles, at least not in a way that does not leave enough tracks on J~ (p]
to infer the existence of ft.
The example illustrated in Fig. 5.4 is based on Minkowski spacetime.
Things are much worse in a spacetime like the past-truncated spacetime of
Fig. 5.3, where there are past endless timelike geodesies that do not enter
J~(p). Similarly, in general relativistic spacetimes which are not artificially
past-truncated but which have particle horizons, prediction is much harder.
130 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

This is one reason to dislike particle horizons. But before the complaint can
be explored it is necessary to define particle horizons.

5.4 Event and particle horizons

The locus classicus of the modern treatment of horizons in cosmology is

Rindler (1956). According to Rindler's definition, a future event horizon
(FEH(y)) for an observer with world line y is a hypersurface in spacetime
"which divides all events into two non-empty classes: those which have been,
are, or will be observable by [y], and those that are forever outside [y's]
possible powers of observation" (p. 663). Leaving aside the caveats about
"observation" discussed in section 5.2, we can take FEH(y) to be given by
the boundary between those events lying inside of J~ (y) and those lying
outside. Some examples will help to illustrate the definition.

Example 1. If y has a future endpoint e, then one would like to think of

FEH(y) as the future boundary of J ~ ( e ) , i.e., J~ (e) — /""(«). When J ~ ( p ) ,
peM, is closed, this boundary will coincide with the past light cone L~ (p)
of p, i.e., the locus of the past-directed null geodesies through/;. 7 However,
it can happen that J~ (p) is not closed, in which case not all of the points on
the future boundary of J~ (p) can be joined to p by a causal curve and L~ (p)
will be a proper subset of this boundary. I will ignore such pathologies in the
discussion below.8
Example 2. For a future endless timelike geodesic y in Minkowski
spacetime, FEH(y) = 0 since y's past light cone sweeps out the entire
spacetime.
Example 3. For the future endless, hyperbolically accelerated t\ of Fig.
5.5 (depicting two-dimensional Minkowski spacetime), FEH(f/) = H so that
nothing on the opposite side of H is observable by r\ by means of causal signals.

Fig. 5.5 An illustration of future event horizons in (two-dimensional) Minkowski

spacetime
 The Big Bang and the Horizon Problem 131

Fig. 5.6 Event horizons for geodesic observers in (two-dimensional) future- and
past-truncated Minkowski spacetime

Fig. 5.7 Conformal diagram of de Sitter spacetime

Example 4. The timelike geodesic 9 of future- and past-truncated Min-

kowski spacetime depicted in Fig. 5.6 has a non-trivial future event horizon.

Example 5, De Sitter spacetime illustrates how a geodesically complete

spacetime can have future endless timelike geodesies with non-trivial event
horizons (see Fig. 5.7). In this example the existence of a non-trivial future
event horizon for a future endless geodesic observer is a consequence of the
fact that J^ + is spacelike. However, in asymptotically flat black hole
spacetimes, non-trivial event horizons exist even though ,/+ is null (recall
Fig. 3.4b).

The concept of a particle horizon is much murkier. Rindler applied the

notion to "fundamental observers" in Friedmann— Robertson—Walker (FRW)
cosmological models, that is, to timelike geodesies that represent the motions
of stars, galaxies, and the like during the matter-dominated portion of the
132 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

evolution.9 For such an observer y the particle horizon at a cosmic instant t0

was said to be "a surface in the instantaneous 3-space t — t0, which divides all
fundamental particles into two non-empty classes: those that already have
been observable by [y] at t = (0 and those that have not" (Rindler 1956, p.
663). Hawking and Ellis (1973) take the particle horizon PH(y,p) for y at
some point p e y to be "the division of particles into those seen by [y] at p
and those not seen by [7] at/)" (ibid., p. 128). Of course, by "seen" they
mean can be seen via a causal signal, and by "not seen" they mean cannot
be seen via a causal signal. Note that for any y and y' containing p,
PH(y,p) = PH(/,/>). Wald (1984a) speaks of the particle horizon of y at p
as the "boundary" between world lines of particles that can be seen by y at
p and those that cannot. The definitions of Rindler and Wald seem to associate
a particle horizon with a surface in spacetime, whereas the Hawking and Ellis
definition involves no such surface but only a division of particles. Some
concrete examples will help to illustrate some of the peculiarities and
ambiguities of usage. ;:

Example 6. In Fig. 5.4 particle /? cannot be seen by y at p, so that

following the definition of Hawking and Ellis one would say that /? is outside
of y's particle horizon at/). But in the standard usage, observers in Minkowski
spacetime would not be said to have particle horizons since for any point p,
J~ (p) contains the world line of any past endless timelike geodesic. This usage
reflects the original concern with particles that follow timelike geodesies.
Example 7. In the conformal representation of de Sitter spacetime (Fig.
5.7), the particle following the timelike geodesic K is beyond (/''s particle
horizon at x but not atjy. In this example, the existence of nontrivial particle
horizons, for particles following geodesies, is a consequence of the fact that
«/~ is spacelike. But as in the case of event horizons, particle horizons can
exist even when ^~ is not spacelike.
Example 8. A more physically relevant example of particle horizons is
provided by the FRW models. Because these models are the focus of the
current debates about the horizon problem, they will be treated in some detail
below. But first I want to make a few more general remarks about the
definitions of particle horizons.

The concentration on geodesies in the typical discussion of particle

horizons is a little difficult to understand. EFE entail that the energy-
momentum tensor Tah obeys the conservation law Va Tab = 0. If Tab repre-
sents a perfect fluid and the pressure p = 0—that is, we are dealing with a
dust source—then it follows directly from the conservation law that the
normed four-velocity V of the dust obeys VhVb V = 0, i.e., the world lines
of the dust particles are geodesies. But when p =£ 0 and, more generally, when
Tah is not a perfect fluid source, matter will not follow geodesies. And in any
case, observers equipped with rocket engines arc not confined to timelike
 The Big Bang and the Horizon Problem 133

geodesies. On the whole it might seem preferable to drop talk about particle
horizons, or else to reform usage so that a particle horizon at a point p marks
the division between those particles whose world lines fall into J~ (p) and those
whose world lines do not, regardless of whether the particle world lines are
geodesies. But the standard usage, whatever its shortcomings, does have the
virtue of focusing attention on a class of cases that generates the horizon
problem; namely, those cases like Figs. 5.3, 5.6, and 5.7 where there are
contemporaneous events that have disjoint causal pasts. When I speak of
particle horizons in the context of the "horizon problem" of big bang
cosmology in the sections below I will have this feature in mind.
Another potentially confusing aspect of the usual treatment of particle
horizons concerns the true slogan that once within a particle horizon, always
within a particle horizon; or as Rindler puts it, "particles that have at
sometime been visible to [y] remain so forever" (Rindler 1956, p. 666).
Trivially, if the world line of a particle 6 falls within J~ (p) for some pey,
then for any q e y such that q chronologically succeeds p, d falls within
J~ (q) :u J~ (p). But this truism is entirely compatible with there being a stage
in <5's existence after which <5 cannot causally signal to y. This will occur when
y has an event horizon and 6 has crossed it. In the Minkowski example of
Fig. 5.5, 77 and r\' are within each other's particle horizon at every moment.
But after q r\ can no longer signal to >/', and after p rj' can no longer signal
to r]. A similar behavior can occur for geodesic observers in spacetimes with
a spacelike future infinity. Thus, in a conformal diagram of de Sitter spacetime
(Fig. 5.7), the geodesic K is within the geodesic i/f's particle horizon after j;
but after z, K cannot successfully signal to I/'. Somewhat analogous behavior
can occur in inflationary scenarios, as will be discussed in sections 5.11 and
5.12.
For future reference it will be helpful to give a more detailed treatment
of particle horizons in the FRW models. Coordinates can be chosen so that
the line element of a homogeneous and isotropic metric takes the form

with/(r) = ! / ( ! — kr2), where k = + 1 (space sections t = constant of constant

positive curvature), k = Q (flat space sections), or k = — 1 (space sections
of constant negative curvature). Because of homogeneity there is no loss of
generality in focusing on a fundamental observer at r = 0. And because of
isotropy there is no loss of generality in focusing on radial null geodesies
(dO = d(j> = 0) in discussing what our chosen observer can see by optical
means. Since ds2 = 0 along a null geodesic, it follows from (5.1) that when
at a time / our observer looks backward in time to events occurring at t0 < t,
his optical observations will be able to extend to a coordinate distance of
134 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

As measured at time t the proper distance from the spatial location r(t, t0) to
the origin r = 0 is dr(t) = a(i)r(t, t0). Whether it is useful to choose t equal to
t, or to t0, or to something in between will depend on the application. As t0
approaches the time of the big bang (t = 0), the limiting value ofr(t, t0) marks
the boundary of the portion of the universe that at time t is accessible to our
observer by direct optical means:

Thus, a particle horizon is present for our observer at t just in case the horizon
coordinate distance rH(t) is finite, which will be the case if and only if the
integral in (5.3) converges.10
Obviously this matter cannot be decided until the temporal behavior of
the scale factor a(t) (aka the "radius of the universe") is specified. But to
help fix intuitions, consider the mathematical example where a(t) behaves as
t". Then particle horizons will be present just in case n < 1.
In GTR the behavior of the scale factor is determined by EFE in
conjunction with assumptions about the nature of the energy—momentum
tensor. In the FRW models the symmetries of the spacetime force the
energy—momentum tensor to have the form of a perfect fluid: Tab =
(H + p)VVb + pgab, where /i is the mass density, p is the pressure, and V is
the normalized four-velocity of the fluid. With cosmological constant A < 0
and with /u + 3jd > 0 (as will certainly be the case, for example, with
positive mass density and non-negative pressures), EFE entail that there is an
initial singularity. And for p > 0 the behavior of a(t) is such that the integral
in (5.3) converges (see section 5.11).
The FRW metric is conformally flat. In the case of a flat spatial geometry
(k = 0) this means that new coordinates can be chosen such that the line
element takes the form

The case of particle horizons then corresponds to the divergence of the

conformal factor SJ(?) as t —> 0 + .11 Since causal properties are not affected
by a conformal transformation, the causal features of the FRW models with
particle horizons can be represented as in Fig. 5.8, which gives a much better
intuitive feeling for particle horizons than do the convergence properties of
the integral in (5.3). The price to be paid is that Fig. 5.8 badly distorts
spatiotemporal distances.

5.5 What is the horizon problem?

Suppose that our universe is an FRW big bang universe with particle
horizons. We cannot see all the way back to the big bang with either an
optical or a radio telescope because the regime prior to the time td (x 1013 s)
 The Big Bang and the Horizon Problem 135

Fig. 5.8 The standard big bang model in conformal representation

of the decoupling of matter and radiation is opaque to such means of

observation. When at t = tN ("now" x 1018 s) we look backwards in time
towards the decoupling, we observe microwave radiation, a remnant of the
big bang, that is remarkably homogeneous and isotropic. (Recently, small-
scale fluctuations in the cosmic microwave background radiation (CMBR)
have been detected, but this does not affect what follows.) But the FRW
model implies that for directions in space with a sufficiently large angular
separation, the events at the time td of decoupling (e.g., p and q in Fig. 5.8)
have no common causal past.
The conjunction of this causal disjointness and the isotropy of the CMBR
is variously said to be "quite puzzling" (Hakim 1984), a "paradox" (Barrow
and Silk 1980), a "major mystery" (Rees 1972), and "philosophically
unsatisfactory" (MacCallum 1979). Several different research programs
heading off in different directions have been proposed to deal with this horizon
problem. Some of them will receive attention in due course. But before turning
to attempts to resolve the problem, it is important to get a fix on what the
problem is supposed to be.
Philosophers of science who are steeped in the literature on scientific
explanation and causality will be tempted to give a quick answer. In everyday
life and science, it is usual to search for a common cause explanation for a
correlation between distant events. But in the case at issue no such explanation
is possible since there is no common causal past for the events in question.
To explore this answer we need to know a bit more about Reichenbach's
principle of common cause.

5.6 Reichenbach's principle of common cause

The principle of common cause (PCC) was enunciated in Reichenbach's

posthumously published book, The Direction of Time (1971). In the succeeding
136 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

years the PCC gained an important niche in the philosophy of science: it

turns up in treatments of scientific explanation, causality, and realism, and
it also enters discussions of the foundations of biology and quantum mechanics
(see Salmon 1984; van Fraassen 1980; Sober 1988; Arntzenius 1992, and the
references given therein). And yet both the status and the content of the PCC
are the subject of vigorous debate.
Reichenbach's intentions in positing the PCC are not pellucid, but it
seems that he wanted to frame a principle that would give both relativity
theory and quantum mechanics their due. The former he took to imply no
action at a distance and the latter he took to dethrone determinism. In
keeping with the latter it is hopeless to look for deterministic causes for
events, while in keeping with the former (Reichenbach thought) we ought
to find a common cause, albeit probabilistic in nature, that explains the
correlation between distant events. More specifically, if A(p) and B(q)
are events that occur at relatively spacelike locations p and q and
Pr(A(p) & B(q)} =£ Pr(A(p)) x Pr(B(q)), then a probabilistic common cause
is an event C(w) such that w e J~ (p) n J~ (q) and Pr(A(p) & B(q)/C(w)) =
Pr(A(p)/C(w)) x Pr(B(q)/C(w)), or equivalently (assuming no divisions
by 0) C(w) screens off A(p) from B(q) and vice versa in that
Pr(A(p)/C(u>) & B(q)) = Pr(A(p)/C(w)) and Pr(B(q)/C(w) & A(p)} =
Pr(B(q}/C(w}}.
Quantum mechanics, which was one of the two prompters of the PCC,
implies two ironies. First, there are quantum states, such as the singlet state
for two spin 1/2 particles, that involve a strict correlation between events
(such as measurements of the spins of the particles along chosen axes) which
can be arranged to be relatively spacelike. For such strict correlations a
screener-off is possible only if the conditional probabilities on the common
cause are 0 or 1, which is a return to determinism or something very near it
(see van Fraassen 1980). Second, even when the correlations are not strict,
the PCC clashes with quantum mechanics. For the existence of a probabilistic
common cause implies a set of probabilistic inequalities—for instance, the
Bell-Clauser-Horne inequalities—that are provably violated by quantum
statistics (see Clauser and Home 1974).
One reaction to these facts would be to interpret the PCC as asserting
that either correlated relatively spacelike events have a common cause or else
they are connected by a direct causal link (see Arntzenius 1992). This reading
turns the PCC into a purely classificatory principle whose only function is to
distinguish between two ways relatively spacelike events can be causally
linked—either indirectly via a common cause or directly through some
non-local action. Parts of Reichenbach's The Direction of Time and his earlier
book Philosophy of Space and Time (1958) suggest that he had a stronger
interpretation in mind, one that gives the PCC a normative dimension, for
Reichenbach apparently thought that relativity theory prohibits non-local
action. Relativistic quantum field theory (QFT) provides a counterexample
to the normative reading that says that relativity theory is incompatible with
 The Big Bang and the Horizon Problem 137

any non-local action as defined by a failure of screening off. QFT is the proper
arena for discussing the PCC in connection with quantum phenomena; for
the spacetime background of this theory is Minkowski spacetime (as opposed
to Newtonian spacetime for ordinary quantum mechanics), and in addition
QFT (unlike ordinary quantum mechanics) allows operators to be associated
with spacetime regions. Thus, in QFT one can make precise sense of the
notion, which one can only indicate by hand-waving gestures in ordinary
non-relativistic quantum mechanics, of measurements made at relatively
spacelike locations. Violations of the Bell inequalities and, hence, of the PCC
can be arranged in this setting (see Landau 1987; Summers and Werner
1977a, b). But this failure of the PCC does not signal any sort of non-locality
that is in conflict with relativity theory, as is made manifest by the relativistic
invariance of QFT.12 Of course, one could still insist on the weaker,
classificatory reading of the PCC and conclude that quantum phenomena do
involve some sort of non-local features. And in seeming concert with this
reading there is a large literature that consists of hand-wringing and moaning
about quantum non-locality. There certainly are non-local features to QFT,
and some of them are surely puzzling. For example, even with the axiom that
operators associated with relatively spacelike regions commute (or anticom-
mute) one still has the Reeh-Schleider theorem which asserts that operat-
ing on the vacuum state with polynomials of operators associated with some
bounded region of spacetime generates a dense set of states in the Hilbert
space. One would like to understand better such non-local features. But in
carrying out this task the philosophy underlying the PCC is useless—indeed,
less than useless, for it suggests that these non-local features indicate a lurking
conflict with relativity theory, whereas relativistic invariance has been built
into the construction of QFT ab initio.
Even if we leave aside the mysteries of the quantum domain, the status
of the PCC is still open to challenge. A forceful illustration has been provided
by Arntzenius (1992), who shows that for generic, time homogeneous, Markov
processes a screener-off will generally not exist. In addition, Forster (1986)
and Arntzenius (1992) argue persuasively that there are a variety of
equilibrium correlations which do not call for a common cause explanation.13
Of course, even if the PCC cannot be defended as a general principle, it may
nonetheless help to pinpoint what is problematic about the horizon problem
in general relativistic cosmologies with particle horizons. It is to that matter
I now turn.

5.7 Particle horizons and common causes

It is not easy to bring Reichenbach's PCC to bear on relativistic cosmological

models. As a first stab, it might seem that the PCC is necessarily satisfied in
this setting since GTR is a deterministic theory, which means that the relevant
probabilities are always 0 or 1. This is too quick and too crude. It is too quick
138 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

because it is not evident that the probabilities can be made 0 or 1 by

conditioning on events in the common causal past (more on this anon). It is
also too crude because it does not address the nature of probabilities in GTR.
At the metalevel probabilities can enter in terms of measure theoretic
arguments concerning classes of solutions of EFE (see section 5.9); but this is
irrelevant to the PCC. At the object level, probabilities can enter in one of
two ways. First, they can be epistemic probabilities, representing our assess-
ments of how likely the available evidence makes the occurrence of the events
in question. Again, however, the PCC was not intended to apply to such
measures of our ignorance. Second, the probabilities may enter by introducing
some relativistic version of classical statistical mechanics, e.g., the relativistic
Boltzmann equation. But the central and unsolved mystery of non-quantum
statistical mechanics is what these probabilities mean and how they are to be
justified. The upshot is that any attempt to apply Reichenbach's PCC to
classical GTR is muddied by one of the most contentious foundation problems
in physics.
Can we not cut through some of these difficulties and say at least the
following? There is a perfectly good intuitive sense—even if we cannot make
it precise by the use of unproblematic probability assertions—in which
relatively spacelike events in general relativistic cosmological models can be
correlated. In models with particle horizons these events may have no
common causal past. Hence, there is a conflict with the spirit if not the letter
of the PCC, and this conflict helps to illuminate the widely expressed
queasiness about particle horizons. This answer is overly optimistic. The
conflict provides illumination only if one accepts the spirit of Reichenbach's
assumptions about common causal explanations; but the most straightforward
attempt to translate these assumptions into the setting of GTR leads to an
incoherent view of explanation. The reader who digested the discussion of
deterministic prediction in section 5.3 will have anticipated the point, but it
merits explicit elaboration here.
In the context of general relativistic theories, to construct a dynamical
explanation of the state in some region X a M spacetime M, gab requires
the choice of a spacelike hypersurface E such that X a D+(~L). This is so
whether X is a connected region R or the disjoint union of two relatively
spacelike regions R and R'. The specification of the relevant initial data on
~L together with the coupled Einstein—matter equations entail the state on
X. In the philosophical jargon, we have a paradigm case of Hempelian
deductive—nomological explanation. In some instances the existence of naked
singularities or other pathologies in the spacetime structure may prevent the
choice of such an D (see chapter 3). But in general the existence of particle
horizons is not such a pathology; and in particular, in the FRW models the
presence of particle horizons does not entail the non-existence of such a E
since the FRW models can be partitioned by Cauchy surfaces. Conversely,
even when there are no particle horizons, it will generally not suffice for a
dynamical explanation of the states on relatively spacelike regions R and R'
 The Big Bang and the Horizon Problem 139

Fig. 5.9 The common causal past of two relatively spacelike regions

to choose a £ that belongs to the common causal past J (R) n J (R'), for
as already noted above it is typically the case that D+(j~ (R) n J~ (R'}) =
J~ (R] n j ~ (/?'). It was also noted that there are exceptions (e.g., where
J~ (R) n J~ (R') contains a Cauchy surface); but these exceptions are truly
exceptional among general relativistic cosmological models.
In sum, the problem of the "horizon problem" cannot be that a paradigm
deductive—nomological explanation cannot be constructed in the presence
of particle horizons, for that complaint is patently false in typical cases. Nor
can the complaint be that in the presence of particle horizons such an
explanation fails to be a good explanation because it fails to be a common
cause explanation; for in typical cases—even without particle horizons—it
is impossible to give a dynamical explanation of the type under discussion
purely by reference to the common causal past.
By way of defense of common cause explanations let us suppose that a
spacelike hypersurface £ is chosen large enough with respect to the relatively
spacelike regions R and R' and that the spacetime lying between £ on the
one hand and R and R' on the other is non-pathological; that is just to say
that R <= D+(J~(R) n £) and R' <= D+ (J~(R'} n £) (see Fig. 5.9). Even
though (R u R'} <t D+(J~(R) n J~(R'} n £), it is reasonable to think that
in typical cases where J~ (R) n J~ (R'} n £ is, in some appropriate sense, a
large fraction of (jf~ (R) \J J~ (R'}} O £, the heart of the dynamical explana-
tion of the joint state on R and R' lies with the initial data on the common
causal past portion J" (R) n J~ (R'} n £ of £. In these circumstances there
is a deterministic deductive nomological explanation approximating a common
cause explanation. Particle horizons pose a problem because they tend to
block the satisfaction of the large fraction condition.
All of this is true enough but not really to the point. Suppose that because
of the presence of particle horizons J~ (R) n J' (R'} n £ = 0 for any
connected partial Cauchy surface £ but that there exists a connected partial
Cauchy surface £ such that (R u R'} c D+ ((J~ (R) u J~ (R'}} n £). So we
have an explanation, in terms of the dynamical evolution of the appropriate
initial data on (J~(R) u j ~ ( / ? ' ) ) n £, of the joint state on R and R', an
140 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

explanation which is not in any sense a common cause explanation. Our

problem is this: What precisely is it about such a dynamical explanation that
makes it deficient or unsatisfying as an explanation? Discussions surrounding
the PGC hint that something must be wrong with the explanation since by
hypothesis it cannot rely on common causes. But the hints are far from
articulate.
Knowing that we cannot rely on the PCC to solve our problem, there
appears to be no other recourse than to investigate in detail the physics
associated with particle horizons in order to try to pinpoint what it is that
makes models with a particle horizon problematic. What this investigation
uncovers can perhaps be identified with the valid core of the PCC as applied
to particle horizons.

5.8 Diagnosing the bellyache: Electromagnetism

In trying to diagnose the bellyache involved with particle horizons it will be

helpful to reconsider in more detail the remarks about electromagnetism made
in section 5.2.
In four-dimensional Minkowski spacetime Maxwell's equations entail
that the components of the vector and scalar potentials for the electromagnetic
field all obey the inhomogeneous wave equation

where D =:r]abVaVb, Va is the derivative operator associated with the

Minkowski metric qab, and f describes the source distribution. 14 Any solution
of (5.5) can be written in the Kirchoff retarded representation, which has the
form

The volume integral in (5.6) is I dV where F is a volume r containing

JF r
Qj r is the distance from Q, to the volume element, and [/^] means that f is
to be evaluated at the retarded time t' = t — r. In plain English, the volume
integral represents the contribution to the field 3>(Qj 0 coming from sources
at spacetime locations where the past light cone L~ (Q, t) cuts the world lines
of the sources. The second term on the right-hand side of (5.6) is a surface
integral. It gives a contribution only from points that lie at the intersection
of L~ (Qj t) with the past time slice corresponding to t' = t — r. (This is an
expression of Huygens' principle^ which means that the effects of $ propagate
cleanly at exactly the speed of light. This principle fails if the dimension of
the space is, say, two.) This surface integral gives contributions from (i)
 The Big Bang and the Horizon Problem 141

sources outside of V, and (ii) source-free radiation. If there are only a finite
number of sources (here, charged particles) then as Fis expanded eventually
all sources will be picked up and (i) will cease to contribute. If we further
postulate that there is no source-free radiation coming from infinity (Sommer-
feld radiation condition), then in the limit in which V expands without bound,
or correspondingly the retarded time t' is pushed back towards — oo, (ii) ceases
to contribute. We are left with the Linard-Wiechart potential

familiar from texts on electromagnetism.

Now suppose that particle horizons are introduced by truncating Min-
kowski spacetime in the past, as in Fig. 5.3. Then (5.7) cannot hold. This is
just a way of reexpressing the point about constraint equations made in section
5.2. Because of the presence of the charged particle d, the value of 3> at
P = (Qj 0 cannot be zero even though L~ (Q, t) never cuts 5. This means in
turn that the Sommerfeld radiation condition cannot be satisfied; no matter
how far the retarded time is pushed towards the beginning of time at 2000 B.C.,
there will still be source-free radiation entering the volume. (This is not to
say that the original solution to Maxwell's equations is no longer a solution;
indeed, any solution on full Minkowski spacetime remains a solution when
restricted to past-truncated Minkowski spacetime, as follows from the local
nature of Maxwell's equations. The point is that the restriction of a solution,
although still a solution, may not admit a representation with the desired
properties, such as no incoming radiation.)
Needless to say, past-truncated Minkowski spacetime is a highly artificial
illustration of particle horizons. So it is natural to wonder about the fate of
the Sommerfeld radiation condition in non-artificial cosmological models with
particle horizons. The matter is complicated, in part because in general
Huygens' principle does not hold for curved spacetimes and in part because
in this general setting it is not easy to give precise mathematical expression
to the idea that there is no incoming source-free radiation. However, the
investigations of Penrose (1964) and Ellis and Sciama (1972) indicate that
the Sommerfeld condition cannot be satisfied in generic cosmological models
with particle horizons arising from a spacelike ,/~. 15
The failure of the Sommerfeld radiation condition will be a source of
consternation to those who share the Machian intuition that all physical
effects must be explained in terms of sources in the form of ponderable bodies.
Einstein (1916) himself was of this persuasion when he wrote the paper that
put GTR in its final form. 16 Perhaps the wide appeal of this intuition helps
to explain why cosmological models with particle horizons are thought to be
objectionable. But the validity of the objection is another matter. The failure
of theories of modern physics to conform to Machian intuitions may indicate
142 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

that there is something wrong with these theories. But one should also be
prepared to consider the alternative that these intuitions need retraining.
Certainly Einstein's special and general theories of relativity promote the view
that fields are entities in their own right and are ontologically as basic as
particles. And in quantum field theory the now dominant point of view is
that all particle-like behavior is to be explained in purely field theoretic terms
and that there are many circumstances where the particle concept is not useful
or even applicable.
It should also be emphasized that implementing the condition of no
source-free radiation comes at a price; namely, it assumes an asymmetry of
time. To appreciate the point, return to the simpler context of STR and
Minkowski spacetime. The retarded representation (5.6) is just that—it is not
a particular solution of (5.5) but rather a representation of a general solution.
But there are many other representations as well. The advanced representa-
tion, for example, is obtained by evaluating the integrals in (5.6) at the
advanced time t + r. And in addition there are any number of linear
combinations of advanced and retarded representations. Sticking to the pure
retarded and advanced representations, we have

So if we require that j^ ret = 0, it follows that

And, as Sciama (1963) notes, J^ adv will not in general vanish. Perhaps this
temporal asymmetry will be welcomed, and perhaps it can be justified,
e.g., by showing that under certain cosmological conditions the advanced
representation cannot be valid (see Sciama 1963; Hawking 1965), thereby
providing at least part of the solution to the problem of the direction of time.
This is not the place to tackle this knot of contentious issues. The point I
want to emphasize is simply that the seemingly innocuous condition of no
source-free radiation carries with it some heavy baggage.
To summarize, if we subscribed to the prejudice that in electromagnetism
and gravitation all effects must be due to causal propagation from sources in
the form of ponderable bodies, then we would have at least a partial account
of what is objectionable about cosmological models with particle horizons.
But this prejudice is nothing more than a prejudice.

5.9 Diagnosing the bellyache: Cosmic background radiation

1 low are we to understand pronouncements such as " the existence of

[particle] horizons is a fundamental obstacle to any dynamical explanation"
 The Big Bang and the Horizon Problem 143

of the homogeneity of the CMBR (Hartle 1983, p. 79) or that because of the
presence of particle horizons in the standard big bang models it is "essentially
impossible" to account for the uniformity of the CMBR as having evolved
"due to a physical process operating in the early Universe" (Turner 1987,
p. 226)? In The Early Universe Borner at first seems to want to dismiss talk
about a horizon problem.
It does not seem fair to speak of "problems" in the context of the standard
big bang model. In any solution of a differential equation there are certain
specific properties of the initial data. If we compute backwards in time we
just find the initial data that are responsible for the state of affairs as we see
it now. (Borner 1988, pp. 274-275)

But he goes on to add

If we look at the present state of the universe as a consequence of certain
initial data we might feel a bit uneasy, if the initial data have to be extremely
specific. . . . As physicists we would feel rnore at ease if we could find an
understanding of such specific conditions in terms of physical processes.17
(ibid., p. 275)

Similarly, Turner (1987) notes that the uniformity of the CMBR "can be
accommodated by the standard model, but seemingly at the expense of highly
special initial data" (p. 227).
There are two contentions here. The first is that a physical theory that
postulates "special initial conditions" is somehow lacking or inadequate. The
second is that standard models of the big bang are forced to posit special
initial conditions. The first contention calls for careful evaluation since it gets
to the heart of issues about what makes a good scientific explanation. But
before turning to the evaluative task, I want to examine the second contention.
A crude argument for the second contention goes as follows. If the
universe starts in an inhomogeneous and anisotropic state in the presence of
particle horizons, then it cannot achieve uniformity in a reasonable time; for
distant parts of the universe cannot causally interact and, hence, there is no
physical mechanism for producing uniformity within the required time span.
From the previous sections we know that this argument is too quick. For in
inhomogeneous and anisotropic universes, particles beyond each other's
particle horizons do interact in the sense that each feels electrical and
gravitational forces due to the unseen presence of the other, and these forces
might conceivably provide a mechanism for helping to achieve uniformity.
What is true in classical relativistic physics is also true in relativistic quantum
physics. QFT entails the existence of correlations between relatively spacelike
regions, even if these regions have no common causal past. Let 01 and 02 be
open regions of spacetime and let j?/t and s^2 denote the local algebras of
observables (self-adjoint operators) associated with Ol and 02 respectively.
A1 6 J2/J and A2 6 ,-8/2 are said to be correlated with one another in the
quantum state *P just in case <VPM1^12|VP> ± < x Pi/l 1 | x F>< x PM 2 | l P>. When 01
144 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

and 02 are relatively spacelike, a standard causality axiom of relativistic QFT

asserts that A1 and A2 commute. Nevertheless, At and A2 may be correlated;
indeed, the Reeh-Schleider theorem18 entails that for any Oi and 02, no
matter how far apart, there are Av e j/t and A2 e j$2 such that Ai and A2
are correlated in the vacuum state (see Wald 1992).19 It is conceivable that
the correlations between regions that lie beyond each other's particle horizons
might be strong enough to have a discernible effect in these regions (see Wald
1992). A perhaps more effective mechanism for reducing anisotropy in the
CMBR is at work in compact k= —\ FRW models where the sensitive
dependence on initial conditions of the spacetime geodesies leads to a flow
on the surfaces of homogeneity that is strongly chaotic (see Ellis and Tavakol
1994).
It seems to me implausible that such trans-particle horizon mechanisms
could play a major role in smoothing out inhomogeneities and anisotropies.
But the point is that some non-negligible role cannot be ruled out a priori as
some of the thinking behind the PCC would suggest. In any case a more
sophisticated argument is needed to establish that in the presence of particle
horizons no mechanism can succeed in smoothing out inhomogeneities and
anisotropies in the initial state. The argument might take the form of detailed
calculations for various potential smoothing mechanisms. Or in the case of
classical GTR it might take the form of a measure theoretic examination of
solutions of EFE. According to Turner (1987), an argument of the latter type
has been provided by Collins and Hawking.
In 1973 Collins and Hawking "pointed out that the set of initial data
which evolve to a Universe such as ours is of measure zero providing that the
stress energy in the Universe has always satisfied the strong and dominant
energy conditions" (Turner 1987, p. 227). More specifically, what Collins
and Hawking (1973) showed is this: assuming that EFE are satisfied with
A = 0 and that matter obeys the dominant energy condition and the positive
pressure criterion, 20 then there is no open set of initial data which leads to
cosmological models that eventually approach isotropy and which intersects
the subspace of homogeneous initial data. As Collins and Hawking note, this
conclusion does not by itself rule out the possibility that there is an open set
of in homogeneous initial data that leads to homogeneity and isotropy at late
times, but it makes the possibility seem unlikely since presumably inhomogen-
eities in the initial data are more likely to lead to anisotropy than to isotropy.
What the powerful result by Collins and Hawking raises in the first
instance is not a horizon problem but a uniformity problem. Since initial data
and cosmological models satisfying EFE are in one—one correspondence, the
Collins—Hawking theorem shows that (in all plausibility) cosmological models
that achieve uniformity even at late times are very rare in the space of all
models satisfying EFE and the energy and pressure conditions. To lay the
blame for the uniformity problem at the doorstep of particle horizons requires
further results along the following lines. Since the big bang is the focus of
concern, one would like to single out a subspace of models that can be said
 The Big Bang and the Horizon Problem 145

to begin with a big bang. Next one would like to show that in the appropriate
subspace topology there is an open set of such big bang models which satisfy
EFE and energy conditions and which eventually achieve homogeneity and
isotropy. And finally one would try to show that within this open set there
is no open subset of models with particle horizons. In the light of such results
one could reasonably say that the uniformity of the universe is not an a priori
problem in big bang models but that it becomes a problem if particle horizons
are present. This is one more example of how tacit assumptions are needed
to make the horizon problem into a genuine problem. Nor is it obvious that
the tacit assumptions are warranted; for instance, we will see in section 5.12
that there is a natural way to assign a measure to initial conditions such that
uniformity is an a priori problem and not a problem of particle horizons.
For the sake of discussion I want to suppose now that we have in hand
results that indicate that particle horizons make uniformity at late times a
rare or improbable feature. It still remains to give a more detailed diagnosis
of this complaint against particle horizons. I will offer a twofold classification
for making the complaint more specific. The classification is based on two
different ways of faulting a putative explanation on the basis of improbability
considerations. The first way of faultfinding is the least searching, for it agrees
at the outset that if the hypotheses of the putative explanation were true, then
a satisfactory explanation would have been provided. But it goes on to claim
that in all probability the putative explanation involves a false assumption.
A neutral example may help to make the nature of the first complaint
more concrete. Suppose that the serious candidates for explaining the origin
of the earth's moon are: H^. condensation/accretion (i.e., the moon was
formed at the same time and by the same process by which the planets
condensed out of primordial matter); H2: fission (i.e., the moon was split off
from the earth by an impact or some other mechanism); and H3: capture
(i.e., the moon came from outside our solar system and was drawn into its
present orbit by the earth's gravitational field). It will be agreed, I think,
that if H3 were true, then spelling out the details would provide a satisfactory
explanation of the origin of the moon. The difficulty is that the set of initial
conditions that would lead to capture is of measure zero. Adopting Bayesian
terminology, it would seem plausible to set the prior probability of H3 very
low.21 But notice that one can easily imagine additional evidence that would
boost the posterior probability of H3 to a respectable value and thus remove
the objection at issue. For example, the first astronauts to land on the rnoon
might have brought back rock samples that strongly indicated an extrasolar
origin for the moon. In this case the probability of H± and H2 would be
dramatically lowered and the probability of H3 would be correspondingly
increased. Similarly, on the basis of the Collins-Hawking result (and
hypothesized supplementary results) one might be inclined to assign a low
prior probability to the standard big bang account of the uniformity of the
CMBR. But again one can easily imagine evidence that would substantially
shift the posterior probability in favor of the standard model. For instance,
146 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

neutrino and gravitational wave detection might improve to the extent that
we are able to probe conditions at times well before the decoupling time td,
and the probes might reveal that at these earlier times the conditions were
homogenous and isotropic. Then once again the first form of the improbability
objection would fall away. 22
The second way of faultfinding claims to locate a deeper flaw in the
explanation offered by the standard big bang model. The objection is not
that the Collins-Hawking result and hypothesized related results show that
the standard explanation has a low prior probability but rather that the
explanation only works for a negligible range of initial data. So even if
neutrino and gravitational wave astronomy were to indicate that the early
universe was in fact uniform, the objection would still stand that the
explanation lacks robustness. This interpretation is suggested by the above
quotation from Borner. It is much more explicit in Misner's (1968, 1969)
program of "chaotic cosmology" which had the goal of showing how neutrino
viscosity and other mechanisms would damp out inhomogeneities and
anisotropies of an arbitrary initial state.
Robustness of explanation has a nice ring to it. But the general claim
that the lack of robustness of a putative explanation (in the sense that it works
only for very circumscribed initial data) automatically discredits or diminishes
the value of the explanation seems to me to be unsustainable. The above
example of the origin of the earth's moon and other similar examples, where
the true explanation may rest on very special initial conditions, suffice to
make the point. It remains open that there is something specific to the lack
of robustness in the standard big bang explanation of the uniformity of the
CMBR that undermines its value as an explanation. But I have found nothing
in either the physics or the philosophy of science literature that provides
anything like an argument for—rather than a mere assertion of-—this more
cautious claim. Nor is it clear to what extent even this more cautious claim
is actually shared by the astrophysics community. It is certainly true, as will
be described in the next section, that the horizon problem has given rise to
an impressive variety of research programs. But this might not be an
indication that the advocates of these programs subscribe to the claim in
question but only that, in line with the first and weaker form of the
improbability complaint, they are laying their bets that the standard big bang
model is false but would be willing to withdraw their bets and their objections
to the standard model if it were revealed that the very early universe was
uniform. Here I will have to leave it to competent sociologists of science to
conduct the relevant interviews to decide this matter.
In closing, I note that a valid or at least defensible core to the PCC has
been located for the cosmological setting. When particle horizons are present
and a common cause explanation of the correlation between relatively
spacelike events is impossible because these events have no common causal
past, then the postulation of special initial conditions is required to achieve
an explanation.
 The Big Bang and the Horizon Problem 14-7

5.10 Strategies for solving the horizon problem

What follows is not supposed to be an exhaustive classification of attempts to

solve or dissolve the horizon problem but only serves the purpose of alerting
the reader to the kinds of reactions that have been explored in the literature.
The collection is motley, and some of the proposed solutions may strike one
as far-fetched. But these features serve to reinforce how seriously the horizon
problem is taken—for good reason or no—and to what lengths physicists are
prepared to go in order to resolve it. I begin with three strategies that I take
to lie outside the mainstream of current opinion as expressed in the
astrophysics literature.

Anthropic solution

The title of Collins and Hawking's 1973 paper is "Why Is the Universe
Isotropic?" In the last sentence of the article they state their answer: "Because
we are here." More specifically the idea is that only in universes that approach
isotropy can galaxies be expected to form. Since galaxies are a necessary
condition for life as we know it, we should not be surprised to find that we
live in a universe that is (on the appropriate scale) isotropic—if it weren't,
we wouldn't be here to ask the question. Whether such anthropic explanations
are genuine explanations or only soothing nostrums is part of a lively debate
that is not appropriate to enter here.

The Penrose conjecture

Penrose (1979, 1986, 1988, 1989a) has argued that there is a fundamental
misunderstanding involved in the hand-wringing about how special the initial
conditions in the standard big bang model would have to be in order to
accommodate the presently observed uniformity of the CMBR. In his opinion
the operation of the second law of thermodynamics requires that the initial
entropy of the universe be small, and this in turn requires that the big bang
be highly constrained. Penrose's working hypothesis for expressing the
constraint is the vanishing of the Weyl conformal curvature (or more precisely,
the smallness of the Weyl curvature in comparison with the Ricci curvature). 23
His conjecture is that such constraints must be grounded in some yet to be
discovered time asymmetric laws that lie at the juncture of quantum theory
and GTR.

A universe with handles

Hochberg and Kephart (1994) postulate that at the Planck time tp ( w 10"44 s)
the universe was riddled with wormholes. To make the wormholes traversable
and, thus, to permit causal connections among otherwise causally unconnected
148 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

regions, the weak energy condition must be violated at the wormhole

mouths.2*
I turn now to more mainstream attempts to solve the horizon problem.
The first has already been mentioned in section 5.8.

Misner 's chaotic cosmology

Type IX Bianchi models are standard general relativistic cosmological models

which are homogeneous but non-isotropic. In some of these models one spatial
direction is horizon free. The changing of this direction often enough as time
goes on (the so-called 'mixmaster universe') may explain the decay of
anisotropy (Misner 1969). One can then go on to study mechanisms that will
dissipate inhomogeneities, hopefully arriving at an explanation of how a
generic initial state can smooth itself out enough to fit current observations.
Two criticisms have effectively undercut interest in this program: the
probability of mixmaster behavior is low in Type IX models, and the amount
of dissipation of anisotropy is not sufficient to account for present observations
(see Stewart 1968; Collins and Stewart 1971; MacCallum 1971, 1979).

Getting rid of particle horizons

The most effective way to resolve the horizon problem is to get rid of the
feature that gives rise to the problem. The tactics which have been proposed
for implementing this strategy can be grouped into two categories, (a) New
physics before tp. (i) Zee (1980) proposed a symmetry-breaking mechanism
involving a scalar field whose presence effectively weakens the gravitational
"constant" as one goes backward in time. In terms of the discussion of section
5.3, the behavior of the scale factor in the Zee model near the big bang is
a(t) ~ t so that the integral (5.3) diverges. This proposal has been criticized
by Linde (1980) and Sato (1980), and subsequently defended by Pollock
(1981). (ii) Anderson (1983, 1984) and Hartle (1983) have explored quantum
gravity effects as a means of eliminating particle horizons. Here the mechanism
is the back-reaction of particle creation caused by the presence of inhomogen-
eities and anisotropies in the early universe, (iii) Akdeniz et al. (1991) studied
a string-dominated early universe model that evolves into a radiation-
dominated universe. As in (i), a(t) ~ t at early times, (b) No new physics but
topological identifications. Ellis (1971) and Ellis and Schreiber (1986) have
explored the possibility of partially eliminating particle horizons by identifying
points in such a way that after some chosen time observers can see all the
way round the universe. Such "small universes" are easily constructed from
the FRW models. Thus, for example, in the k — 0 case the points on the
spatially flat space sections can be identified modulo some triple of distances
</!, d2, d3 to give a torodial spacetime topology in which one returns to the
same spatial location by moving in a straight line a distance d1 (respectively,
</2, </3) in the x (respectively, y, z) direction. With two spatial dimensions
 The Big Bang and the Horizon Problem 149

suppressed, the causal properties of such a model are similar to those of a

version of Fig. 5.1 truncated in the past so as to simulate the big bang. In
contrast to (a) the elimination of particle horizons is partial rather than total;
that is, for any given t, the identification scale can be chosen small enough
so as to assure that at t an observer can see around the universe (many times,
if you like), but it is not true that the identification scale can be chosen so
that at no time t > 0 is there a particle horizon. The partial character of the
elimination is of no consequence if one is concerned only with explaining
observations made now. But as discussed in the next section, improvements
in technology may allow us to probe further and further back in time when
particle horizons are present. And if one is concerned with the strong form
of the horizon problem as distinguished in section 5.9, then it has to be shown
how the present observation of a uniform CMBR is compatible with a generic
non-homogeneous and non-isotropic initial state. Making small universes out
of generic universes by the sort of identification procedure used in the k = 0
FRW models is in general not possible—invariance under a discrete group
of isometrics is required. But lumpy or inhomogeneous small universes
certainly can be allowed, and one can study how uniformization of the CMBR
is achieved in such multiply connected models (see Ellis and Schreiber 1986).

Making the particle horizons effectively small

Even if particle horizons exist at the time td of decoupling of matter and

energy, something approaching a deterministic dynamical explanation in
terms of a common cause may be possible for two relatively spacelike events
that lie on t = td and that are now visible to us if the common causal past
J~ (p) n J~ (q) of the spacetime locations p and q of these events is in some
appropriate sense a large fraction of J~ (p) U J~ (<f) for a time near the big
bang. At least two schemes have been devised to accomplish this aim. (a)
The first involves the use of the Brans-Dicke theory (BDT). BDT is a
generalization of classical GTR to include an adjustable parameter a> and a
scalar field (p that serves as an additional source for the gravitational field.
With cp constant and co set large enough, BDT replicates the predictions of
GTR. But if in the early universe one tunes the available parameters of a
Brans-Dicke model, the desired condition on the common causal past can be
achieved (see Dominici et al. 1983). (b) The second and more popular
approach involves inflationary cosmology. Here the standard big bang model
is supposed to hold up until the Planck time tp. But at some later time tt a
new physical process not taken into account by the standard model is supposed
to enormously increase the expansion of the universe so that a region initially
the size of an atom at the beginning of inflation grows through inflation to
a size bigger than the presently observable portion of the universe. This
inflation is claimed by its proponents to solve the horizon problem. That
claim will be examined in due course. But first the inflationary model needs
to be examined in more detail.
150 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

5.11 Horizons in standard and inflationary models

Inflationary scenarios appeal to theories of elementary particles that unify the

strong, weak, and electromagnetic forces (so-called grand unified theories or
GUTs). 25 It is postulated that during the initial hot era immediately following
the big bang, the symmetry which unites these forces is unbroken and
consequently the predictions of the inflationary model for the expansion of
the universe agree with the standard big bang model. At a later stage,
however, the predictions of the two models radically diverge. To discuss these
predictions, EFE (without cosmological constant) are applied to the FRW
line element (5.1) 26 to obtain two ordinary differential equations for the scale
factor a(t):

where ' stands for d/dt. Together (5.10) and (5.11) entail the conservation law

Solving for a (I) requires some assumption about the equation of state linking
p, and p. Four epoch's need to be distinguished. 27

Epoch I: 0 < t < tt

In the inflationary model, as well as in the standard hot big bang model, the
universe is radiation dominated immediately after the big bang (t = 0). But
whereas in the standard model this era lasts until decoupling (t = t d ) , the
inflationary model posits that it ends at a time lt < td. Radiation dominance
amounts to positing the equation of state p = /i/3. For k = 0 (the case I will
concentrate on), (5.10) through (5.12) imply that a(t] ~t1/2. In keeping
with the notation of Ellis and Stoeger (1988), we have for Epoch I,
a(t] = a ; (2// ; ) 1/2 < 1/2 , where a; = : a(«;) and HI is the Hubble constant during
this era. Applying the procedure explained in section 5.4, we can now
calculate the coordinate horizon distance at the end of Epoch I, with the result
 The Big Bang and the Horizon Problem 151

Epoch 11: t,< t < tf

This is the inflationary era. The universe cools as it expands during Epoch
I. According to the inflationary scenario, when the temperature falls low
enough, one or more Higgs fields assume non-zero values, resulting in the
breaking of the symmetry that unites the strong, weak, and electromagnetic
forces. When the temperature drops below the phase transition temperature,
the universe enters a metastable state called the false vacuum, which has the
strange property that the pressure is negative; in fact, the prediction is that
p = — jJ.f, Hf > 0. It follows from (5.12) that [if is constant during this time.
The stress-energy tensor reduces to Tab = —Hfgab, which has the form of a
cosmological constant term Agai) with A = — /iy. Thus, the GUT mechanism
gives rise to an effective negative cosmological constant, which intuitively
means a repulsive force that drives rapid expansion. 28 Formally, using
p = — [if = constant and k = 0 in (5.10) through (5.11) leads to exponential
expansion (de Sitter universe). Assuming that a(t) and d(t) are continuous at
t = t{ (see Hiibner and Ehlers 1991), the time dependence of the scale factor
during this epoch is a(f) = ai exp[//;(i — £,-)]. The contribution to the horizon
coordinate distance is

where a^ ~ a ( l f ) . So although the volume of the universe expands enormously

during the inflationary phase, the horizon coordinate distance and, therefore,
the portion of the universe a light signal is able to traverse does not increase
as much as in the preinflationary phase (more on this below).

Epoch III: t
The effective cosmological constant now disappears, and we return to the
kind of radiation dominance of Epoch I. Thus, from the end of inflation to
the decoupling time, the scale factor behaves as a(i) = af[2Hf(t — */•) + 1]1/2-
The contribution to the horizon coordinate distance during this epoch is

where ad =a(td)

Epoch IV: t
From the time of decoupling to the present time tN the universe is assumed
to be matter dominated, i.e., the pressure terms in equations (5.10) through
152 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

(5.12) are negligible. EFE then imply that a(t) = ad[_lHd(l - td) + 1]2/3. The
contribution to the coordinate horizon distance during this epoch is

where aN =-a(tN) and rvh stands for the current visual horizon.

5.12 Does inflation solve the horizon problem?

The original inflationary scenario of Guth (1981) suffered from the defect
that the phase transition that creates exponential expansion also creates
inhomogeneities greater than allowed by current observational limits. The
new inflationary scenario ofLinde (1982) and Albrecht and Steinhardt (1982)
overcame this defect and provided for a "graceful exit" from inflation.
Although there is no direct experimental evidence to support the particle
theories used in the inflationary model, the model has gained a wide following
principally because (its proponents claim) it overcomes a number of short-
comings of the standard model, including not only the horizon problem but
also the flatness and monopole problems.29
The validity of the claim that inflation solves the horizon problem
depends, of course, on what the problem is supposed to be. In section 5.9 it
was suggested that the strong form of the problem poses the challenge of
presenting a robust explanation of the presently observed uniformity of the
universe. The proponents of inflationary cosmology seem to accept this
challenge. The model is then open to two related criticisms. First, it can be
charged that the fine tuning of initial conditions in the standard big bang
model is matched in the inflationary model by a fine tuning of parameters
needed to get the model to agree with observations (see Padmanabhan and
Seshadri 1987). Second, it can be charged that in at least one natural sense
of measure, the set of initial conditions that leads to inflation is of small
measure. Thus, Penrose (1986, 1989b) invites us to consider all possible
Cauchy data for the present stage of the universe. It would seem that space
slices on which the data are homogeneous and isotropic even at large scales
are rare as compared with slices with irregular data. We can then use the
backward determinism of EFE to trace backwards generic initial data from
the current time to find generic initial data for the big bang singularity.
Assuming that the inflationary mechanism is effective in smoothing out
irregularities, it follows that inflation does not occur in a generic universe
following the big bang.30 (A response to Penrose's argument is to be found
in Turner 1987, pp. 238—239; Penrose's rejoinder is in his 1989b, p. 267; sec
also Raychadhuri and Modak 1988.)
Setting aside these worries for sake of discussion, it still remains to
 The Big Bang and the Horizon Problem 153

understand how inflation solves the original puzzle about the homogeneity
and isotropy of the CMBR. Indeed, the more reflective reader may be puzzled
as to why inflation does not make a bad situation even worse. Consider again
the behavior of the particles t] and r\' in Fig. 5.5. The flying apart off/ and
r\' in Minkowski spacetime might reasonably be thought to mimic some
aspects of rapid inflation. If so, it would seem that inflation makes it harder
rather than easier for t] and f/' to interact; for although we have seen that it is a
truism that once within a particle horizon, always within a particle horizon,
the latter stages off; and f/' after hyperbolic acceleration begins («inflation)
cannot causally interact with one another. And as seen in the preceding
section, inflation gives rise effectively to a portion of a de Sitter universe whose
conformal structure is indicated in Fig. 5.7. Here we see the case of a particle
K which is in i/f's particle horizon but whose later stages cannot causally
interact with \JJ. In the inflationary scenario the de Sitter expansion lasts only
for a finite time and not into the infinite future; as a result the inflationary
and postinflationary stages of particles that are on the edges of each other's
particle horizons at the beginning of inflation do not lose forever the ability
to causally interact, but these particles do lose the ability to interact via signals
during inflation. The analysis of Epoch II shows how the inflationary stages
of particles at r = 0 and r = 1/fl;/^ cease to be able to causally interact by
means of causal signals (see Patzelt 1990 for a discussion of this and related
matters).
So how does inflation solve the original puzzle? It does not make particle
horizons disappear at the present epoch, contrary to what some enthusiastic
statements by proponents of inflation might lead the unwary reader to believe
(see Padmanabhan and Seshadri 1987). Rather, the virtue of inflation is to
make possible a robust explanation of the presently observed features of the
CMBR. From our present spacetime location s (Fig. 5.10) we look back to
events p and q at the time of decoupling and observe the CMBR. A robust
explanation of the observation requires a satisfaction of the large fraction
condition (section 5.7); that is, the intersection of the common causal past
J~' (p) n J~ (q) of p and q with a time slice t = constant (a measure of the
data that can affect both p and q} must become a large fraction of the volume
of the intersection of J~ (p) (or of J" (q)) with t = constant (a measure of the
data that can affect j> (or q) as t -» 0 + ). In computing this ratio the scale factor
a(t] divides out, so we can work in terms of coordinate distances. From Fig.
5.10, we see that what we want to compute is the cube of the ratio 2fccp/Vx.
But rx = 2rH(tN) - 2 r H ( t d ) , and 2rccp = 2rH(tN) - 2rx = 4rH(td) - 2rH(tN). Since
%(**) = rn(fd) + rvh, 2rccp/rx = rH(td)/rvh - 1. Of course, when rvh > r a ( t d ) the
last ratio does not make sense since then there is no common causal
past for p and q. From the calculations of section 5.11 and the relations
between the constants afHf = adHd(ad/af} and a^ — a f H f ( a , i / a f ) , 3 1 we have
154 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 5.10 The inflationary model of the big bang in conformal representation

In the standard big bang model, a^ = al (no inflation) and (5.17) has a value
less than 1, as expected. In the inflationary model flj/fl; is huge—in most
versions it is at least 1075—while a,j-/ad is of the order of 10~ 24 , with the
upshot that the numerator on the right-hand side of (5.17) is of the order of
1051 (Ellis and Stoeger 1988).
Four remarks about this calculation need to be made, and the first two
are closely related. At most, what has been accomplished is a demonstration
that the inflationary scenario makes the causal structure of spacetime friendly
to a robust dynamical explanation. That the similarity of the CMBR and
other conditions at p and q and other similarly situated points is in fact given
a robust explanation in terms of the conditions in the common causal past of
these points remains to be demonstrated. Only detailed physical calculations
can settle this matter (see Ellis and Stoeger 1988 for a discussion of what is
involved). The second remark is that for fields that are transmitted at
exactly the speed of light not even the minimal necessary condition for
a common cause explanation has been demonstrated. For the case in question
the appropriate ratio is not the one used above; rather, what counts
is the limit as t —> 0 + of the ratio of the area of the intersection
(t = constant) n L~ (p) r\ L~ (q) (i.e., the measure of the initial data that
can affect the field at both p and q) to the area of the intersection
(t = constant) n L~ (p) (or (t = constant) n L~ (q)} (i.e., the measure of the
initial data that can affect p or q as the case may be). This ratio is 0 unless
the past light cones of p and q merge, i.e., there are no particle horizons at
td, which is not the case in the k = 0 and k = — 1 inflationary models.
 The Big Bang and the Horizon Problem 155

The third remark concerns a seemingly curious feature of the ratio

rH(td)/rvh in the inflationary model. The denominator is the same for both the
standard model and the inflationary model, so the difference in the values of the
ratio for the two models must lie in the difference in the values assigned to the
numerator. In the inflationary model rH(td) = r a (t t ) + rH(tf — t{) + rH(td — t f } .
The third term on the right of the equality is small in comparison with the
first two, so the two leading terms must be responsible for the large value of
the ratio in (5.17). For strong inflation—a//a,- large—r H (t t ) ^ %('/ ~ */)> as
is seen by a glance at (5.13) and (5.14). So the consistency of the inflationary
model requires a large contribution to the horizon radius from the initial
non-inflationary, radiation-dominated epoch. To put it in teleological terms,
the initial epoch in the inflationary model gives a larger contribution than
in the standard model because in the inflationary model it is anticipating the
later inflation. However, this does not mean that in the inflationary model
the horizon problem is already solved by the end of the initial epoch. If we
could peer back to t = tt, events such as u and v of Fig. 5.10 have no common
causal past.
The fourth remark goes to the ultimate significance of the accomplishment
of the inflationary model in making rH(td)/rvh large at the current time. In
k = 0 and k = — 1 universes, if enough time elapses, we (or whatever creatures
are around at the time) will have a visual horizon that is comparable to or
larger than the horizon radius at the decoupling time, in which case the
possibility of robust dynamical explanation is lost. Similarly, as advances in
neutrino and gravitational wave astronomy allow us to probe further and
further backward in time we will eventually be able to "see" events which
have no common causal past (see Padmananbhan and Seshardhi 1987, 1988).
The proponents of the inflationary model may reply that the fact that their
model does not solve every problem does not diminish its accomplishment in
solving the problem that was initially posed. The rejoinder is that if problems
exactly similar to the original one arise because of advances in technology or
even merely waiting around, then the proposed solution to the original
problem is fragile and, thus, unsatisfying. So there is an argument to be made
that if the strong form of the horizon problem is accepted as a genuine
problem, then halfway houses such as the inflationary model, which do not
get rid of particle horizons, do not provide an adequate solution.

5.13 Conclusion

There is a popular view to the effect that an anomaly for a theoretical

framework is recognized as such only after a competing framework has
succeeded in resolving the problem (see Lightman and Gingerich 1991). The
horizon problem constitutes a counterexample in the sense that particle
horizons were perceived to be a problematic feature of standard big bang
models long before a satisfactory resolution was reached; indeed, the horizon
156 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

problem was one of the motivations for a large number of explorations of

alternative cosmological models, no one of which has yet achieved general
acceptance. In a deeper sense, however, the popular view is correct. For even
now it is far from clear what the horizon problem is and why the standard
big bang model should be deemed inadequate because of the presence of
particle horizons. Arguably, astrophysicists will have to settle on a model for
the big bang before the exact nature of the problem can be defined.
In the meantime, however, some attempt has to be made to delimit the
shape of the problem, if only in a rough and preliminary fashion. My attempt
here succeeded only in producing a blurry outline. (1) Sure prediction in a
generic general relativistic spacetime is impossible. The presence of particle
horizons makes the impossible even more so in that it increases the difficulty
in making reasonable inductive inferences about the data that may influence
the events to be predicted. But this fact does not seem to provide a sufficient
basis for consecrating a horizon problem. (2) Since the observed homogeneity
and isotropy of the universe within a model with particle horizons involves a
coordination between relatively spacelike events which have no common
causal past, a conflict with Reichenbach's PCG could be seen. But on closer
examination, it was not clear why the pronouncements of the PCC should be
heeded for general relativistic cosmological models or, indeed, even what these
pronouncements are. (3) Particle horizons are most definitely objectionable
from a Machian perspective that demands that all physical effects be tied to
ponderable sources, for in generic models with particle horizons there will be
source-free electromagnetic and gravitational radiation. But once again it is
not clear why such a perspective should be given any currency, especially
since from the modern point of view fields enjoy a status as fundamental as
particles. (4) The potentially most telling objection to particle horizons starts
from the charge that in standard cosmological models such horizons force the
use of very special initial conditions. (It was proposed that this insight is the
most plausible candidate for the valid core of the PCG as applied to
cosmological models with particle horizons.) 32 The objection can be continued
in two ways. First, it can be argued that this specialness shows that the prior
probability of the standard model is very low. This objection could be
overcome by observational evidence indicating that the postulated initial
conditions do in fact obtain in our universe. Second, a deeper objection, which
would not be overcome by such observational evidence, can be raised to the
effect that the standard big bang model lacks robustness and thereby fails to
provide a satisfactory explanation of the currently observed uniformity of the
CMBR. This complaint is to be taken seriously, but it rests on assumptions
about the nature of scientific explanation that require justification. Nor does
the currently most popular self-proclaimed solution to the horizon problem
provide even a might-makes-right justification. The inflationary model can
succeed only by fine-tuning its parameters, and even then, relative to some
natural measures on initial conditions, it may also have to fine-tune initial
conditions for inflation to work. And the inflationary explanation is fragile in
 The Big Bang and the Horizon Problem 157

other ways as well: the robustness of explanation promised by the inflationary

scenario may evaporate as we move into the future or probe into the past
with new means of detecting events in the early universe.
Perhaps all will be made clear by some new cosmological theory. Or
perhaps future generations will conclude that the horizon problem was a
tempest in a teapot. Perhaps what is needed is some deflation of the horizon
problem rather than inflation of the universe.

Notes
1. Throughout this chapter it is assumed that all spacetimes discussed are
temporally orientable (see chapter 6 for a definition of this concept) and that a
direction for time has been chosen.
2. At least if the coupled Einstein-matter equations are of the appropriate
second-order hyperbolic type; see Wald (1984a, Ch. 10) for a discussion of the exact
conditions needed for a well-posed initial value problem.
3. The constraint equations for GTR were stated in chapter 3. For details, see
Wald (1984a, p. 259).
4. This example was suggested by Robert Wald.
5. As noted by Geroch (1977). For a review of various senses of prediction in
GTR, see Hogarth (1993).
6. The failure of the implications of relativistic theories to count as genuine
predictions does not affect the testability or confirmability of these theories.
7. The terms null cone and light cone are used ambiguously in the literature.
Sometimes null cone at p e M is used to denote the object that lies in the tangent
space Mp. Other times it is used to denote what I have called the light cone at p.
8. Joshi (1993) says that M, gab is causally simple just in case J ± ( p ) are closed
for every/) e M. This condition is deserving of the name since it simplifies the hierarchy
of causality conditions for relativistic spacetimes; see chapter 6.
9. These models are described in more detail later in this section and in section
5.11.
10. For spatially open FRW universes, if there is a particle horizon for our
observer at any time, then there is one for all times.
11. The origin of the new time coordinate 7 can be chosen so that 7 = 0
corresponds to the big bang.
12. This is a little too facile, as Tim Maudlin has kindly told me. One cannot
speak of experiments that demonstrate the violation of the Bell inequalities without
engaging the measurement problem in QM. This problem takes a particularly nasty
form in the relativistic setting. For example, the orthodox treatment of quantum
measurement involves a collapse of the state vector. It is not clear whether or not such
a collapse can be given a relativistically invariant rendition. Gordon Fleming (1989)
argues that such a rendition requires relativizing the quantum state to spacelike
hyperplanes. I agree that these are real difficulties whose resolution may require
substantial revisions in current physical theory. But it seems to me that the root of
these difficulties has to do not with common causes but with much more fundamental
issues like the actualization of potentialities. And it is hardly surprising that problems
of non-locality and relativistic invariance arise in measurement collapse since the
collapse is literally a miracle in the sense of a violation of (what we take to
158 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

be) physical laws. Even in paradigms of local classical relativistic field theories (say,
electromagnetic field theory in Minkowski spacetime) the introduction of a local
miracle (say, the creation of an electrical charge) involves non-local instantaneous
effects.
13. For attempts to state a valid kernel of the PCC, see Sober and Barrett
(1992) and Arntzenius (1992).
14. Equation (5.5) requires that the electromagnetic potentials A, (p satisfy the
Lorentz gauge equation V - A + d<p/dt = 0.
15. For homogeneous and isotropic models the effects of charges at the location
p in Fig. 5.3 will cancel out so that there is no inconsistency in assuming that incoming
fields are zero.
16. In retrospect it seems that the extent to which GTR incorporates Mach type
principles is much less than Einstein originally thought; see Raine (1981) for a review
of various interpretations of Mach's principle in GTR.
17. Borner is here speaking of the "flatness problem" rather than the horizon
problem; but the sentiment is the same for both.
18. This theorem is proved for the context of Minkowski spacetime; but
presumably a suitable generalization will hold for curved spacetimes.
19. These correlations cannot be used to send causal signals between the relatively
spacelike regions of spacetime.
20. Recall that the dominant energy condition requires that for every timelike
vector V, TahVaVb > 0 (no negative energy densities) and that TabVa is non-spacelike
(local energy flow is non-spacelike). The positive pressure criterion requires that the
sum of the principal pressures of Tab is non-negative.
21. For an account of Bayesian reasoning in science, see Earman (1992).
22. This is a little oversimplified since the uniformity of conditions at earlier times
would raise problems for galaxy formation.
23. See chapter 2 for a definition of the Weyl curvature.
24. This matter is discussed further in chapter 6.
25. A semipopular exposition of inflationary cosmology is given in Guth and
Steinhardt (1989). A collection of original papers is Abbott (1986). For review articles,
see Barrow (1988), Gibbons, Hawking, and Siklos (1983), Blau and Guth (1987), and
Turner (1987).
26. Virtually all the standard treatments of horizons and inflation rely on the
FRW models. This may seem paradoxical since these models are homogeneous and
isotropic whereas inflationary scenarios are supposed to encompass non-homogeneous
and non-isotropic initial conditions. The explanation is expediency—calculating the
behavior of the scale factor is most easily done in the FRW models. It has to be hoped
that conclusions will not be qualitatively different in more complicated models.
27. The analysis of the four epochs given here is essentially a recapitulation of
the beautiful analysis of Ellis and Stoeger (1988).
28. The Higgs field is a scalar field <p. The associated stress-energy tensor
Tab = VacpVb<p - gab((l/2)g':dVc<pV<i(P - V(q>}}, where V(<p) is the potential energy. To
get Tab into the form A.gab, the first term must be zero and the spatial derivatives in
the second term must be constant.
29. Following Penrose (1989b) it is useful to distinguish internal from external
problems. The absence of monopoles is to be counted as an internal or self-consistency
problem for the inflationary model since it uses GUT theories that generate the
problem in the first place. By contrast the homogeneity and flatness problems are
 The Big Bang and the Horizon Problem 159

external problems. The flatness problem arises from the fact that the density of the
universe seems to be very near to the critical density that has to be exceeded if
expansion is to be halted.
30. Penrose's objection is a kind of jujitsu move against the inflationary scenario,
and it is independent of the details of the GUT mechanisms which are supposed to
produce inflation. Mazenko, Unruh, and Wald (1985) have examined in detail the
new inflationary scenario and claim that in general it does not lead to exponential
expansion; see also Wald (1986).
31. See equations (A4b) and (A4c) of the appendix to Ellis and Stoeger (1988).
32. This is not to say that, at bottom, the aims of the physicists who advocate
inflationary cosmology and the philosophers who advocate the PCC are the same.
The two groups are superficially joined in common cause in trying to explain
correlations between relatively spacelike events. But the driving force behind inflation-
ary cosmology—and several other attempts to solve the horizon problem—is the desire
for robust dynamical explanations. The fulfillment of this desire requires not only the
existence of a common causal past but also the satisfaction of the large fraction
condition (see sections 5.7 and 5.12). By contrast the advocates of the PCC look for
an event (or family of events) which lies in the common causal past of the correlated
events and which in some (perhaps probabilistic) sense can be said to be the cause of
the correlated events. In this search there is no commitment to robustness and no
need to satisfy the large fraction condition. But this very lack of commitment produces
a tension. For the lack of interfering causes that would prevent the "common cause"
event from setting up a correlation between relatively spacelike events, can be viewed
as another kind of correlation between distance events that is no less in need of
explanation than the original one. I assume that advocates of the PCC would respond
that in what we take to be normal background conditions, causal influences propagate
without undue interference and that no explanation of the background conditions is
called for—perhaps these things are simply part of the very meaning of 'causal
propagation' and 'background conditions'. I do not deny that this is so. But I submit
that what is going on here has much to do with common sense and very little to do
with fundamental physics. Right or wrong, the version of common cause reasoning
used in inflationary cosmology reflects one widely shared line of thinking about the
structure of good explanations in physics rather than commonsensical reasoning about
causes or some philosopher's image of what scientific explanation ought to be like.
 6
Time Travel

6.1 Introduction

Over the last few years leading physics journals, such as Physical Review,
Physical Review Letters, Journal of Mathematical Physics, and Classical and Quantum
Gravity, have been publishing articles dealing with time travel and time
machines.1 Why? Have physicists decided to set up in competition with
science fiction writers and Hollywood producers? More seriously, does this
research cast any light on the sorts of problems and puzzles that have featured
in the philosophical literature on time travel?
The last question is not easy to answer. The philosophical literature on
time travel is full of sound and fury, but the significance remains opaque.
Most of the literature focuses on two matters, backward causation and the
paradoxes of time travel.2 Properly understood, the first is irrelevant to the
type of time travel most deserving of serious attention; and the latter, while
always good for a chuckle, are a crude and unilluminating means of
approaching some delicate and deep issues about the nature of physical
possibility. The overarching goal of this chapter is to refocus attention on
what I take to be the important unresolved problems about time travel
and to use the recent work in physics to sharpen the formulation of these
issues.3
The plan of this chapter is as follows. Section 6.2 distinguishes two
main types of time travel—Wellsian and Godelian. The Wellsian type is
inextricably bound up with backward causation. By contrast, the Godelian
type does not involve backward causation, at least not in the form that
arises in Wellsian stories of time travel. This is not to say, however, that
Godelian time travel is unproblematic. This chapter is devoted largely to
attempts, first, to get a more accurate fix on what the problems are and,
second, to provide an assessment of the different means of dealing with
these problems. Section 6.3 provides a brief excursion into the hierarchy of
causality conditions on relativistic spacetimes and introduces the concepts
needed to assess the problems and prospects of Godelian time travel.
Section 6.4 reviews the known examples of general relativistic cosmological
models allowing Godelian time travel. Since Godel's (1949a) discovery, it
160
 Time Travel 161

has been found that closed timelike curves (CTCs) exist in a wide variety
of solutions to EFE. This suggests that if classical general relativity theory
is to be taken seriously, so must the possibility of Godelian time travel.
Section 6.5 introduces the infamous grandfather paradox of time travel. It
is argued that such paradoxes involve both less and more than initially
meets the eye. Such paradoxes cannot possibly show that time travel is
conceptually or physically impossible. Rather, the parading of the paradoxes
is a rather ham-handed way of making the point that local data in spacetimes
with CTCs are constrained in unfamiliar ways. The shape and status of these
constraints has to be discerned by other means. Section 6.6 poses the problem
of the status of the consistency constraints in terms of an apparent incongruence
between two concepts of physical possibility that diverge when CTCs are
present. Section 6.7 considers various therapies for the time travel malaise
caused by this incongruence. The preferred therapy would provide an account
of laws of nature on which the consistency constraints entailed by CTCs are
themselves laws. I offer an account of laws that holds out the hope of
implementing the preferred therapy. This approach is investigated by looking
at recent work in physics concerning the nature of consistency constraints for
both non-self-interacting systems (section 6.8) and self-interacting systems
(section 6.9) in spacetimes with CTCs. Section 6.10 investigates a question
that is related to but different from the question of whether time travel is
possible; namely, is it possible to build a time machine that will produce
CTCs where none existed before? Some concluding remarks are given in
section 6.11. An appendix reviews Godel's attempt to use his solution to EFE
to prove the ideality of time.

6.2 Types of time travel; backward causation

Two quite different types of time travel feature in the science fiction and the
philosophical literature, though the stories are often so vague that it is hard
to tell which is intended (or whether some altogether different mechanism is
supposed to be operating). In what I will call the Wellsian type4 the time
travel takes place in a garden variety spacetime—say, Newtonian spacetime
of classical physics or Minkowski spacetime of special relativistic physics. So
the funny business in this kind of time travel does not enter in terms of
spatiotemporal structure but in two other places: the structure of the world
lines of the time travellers and the causal relations among the events on these
world lines. Figure 6.1 illustrates two variants of the Wellsian theme. Figure
6.la shows the time traveller y,l cruising along in his time machine. At et he
sets the time travel dial to "minus 200 years," throws the switch, and presto
he and the machine disappear. Two hundred years prior to el (as measured
in Newtonian absolute time or the inertial time of the frame in which a t is
at rest) a person exactly resembling the time traveller both in terms of physical
appearance and mental states pops into existence at e2. Even if we swallow
162 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 6.1 Two forms of Wellsian time travel

these extraordinary occurrences, the description given so far does not justify
the appellation of "time travel." That appellation requires that although a.i
is discontinuous with « 2 , «2 is in some appropriate sense a continuation of X1.
Whatever else that sense involves, it seems to require that events on a1 cause
the events on « 2 . Thus enters backward causation in which causes are later
than their effects.
Figure 6.1b also involves funny world line structure, but now instead of
being discontinuous, the world line "bends backwards" on itself, the arrows
on the various segments indicating increasing biological time. Of course, as
with the previous case, this one also admits an alternative interpretation that
involves no time travel. As described in external time, the sequence of events
is as follows. At «4 a pair of middle-aged twins is spontaneously created; the
/?3 twin ages in the normal way while his fS2 brother gets progressively
younger; meanwhile, a third person /^ who undergoes normal biological
aging and who is the temporal mirror image of fJ2 is cruising for a fateful
meeting with /?2; when ^ and /?2 meet at e3 they annihilate one another.
Once again, the preference for the time travel description seems to require a
causal significance for the arrows on the world line segments so that, for
example, later events on /?t (as measured in external time) cause earlier events
on /?2 (again as measured in external time).
Much of the philosophical literature on Wellsian time travel revolves
around the question of whether backward causation is conceptually or
physically possible, with the discussion of this question often focusing on the
"paradoxes" to which backward causation would give rise. I will not treat
these paradoxes here except to say that they bear analogies to the paradoxes
of Godelian time travel that will receive detailed treatment below. But aside
from such paradoxes, there is the prior matter of whether the phenomena
represented in Fig. 6.1 are physically possible, even when shorn of their time
 Time Travel 163

travel/backward causation interpretations. In Fig. 6.la, for example, the

creation ex nihilo at «2 and the extinction ad nihilum at «j are at odds with
well entrenched conservation principles. Of course, the scenario can be
modified so that conservation of mass-energy is respected: at e1 the time
traveler and the time machine dematerialize as before but now their mass is
replaced by an equivalent amount of energy, while at e2 a non-material form
of energy is converted into an equivalent amount of ponderable matter. But
this emended scenario is much less receptive to a time travel/backward
causation reading. For the causal resultants of « t can be traced forwards in
time in the usual way while the causal antecedents of e2 can be traced
backwards in time, thus weakening the motivation for seeing a causal link
going from e1 to e2.
At first blush Godelian time travel would seem to have three advantages
over Wellsian time travel. First, on the most straightforward reading of
physical possibility—compatibility with accepted laws of physics—Godelian
time travel would seem to count as physically possible, at least as regards the
laws of the general theory of relativity (GTR). Second, unlike stories of
Wellsian time travel, Godelian stories are not open to a rereading on which
no time travel takes place. And third, no backward causation is involved. On
further analysis, however, the first advantage turns out to be something of a
mirage since (as discussed below in sections 6.5 through 6.8) Godelian time
travel produces a tension in the naive conception of physical possibility. And
the second and third advantages are gained in a manner that could lead one
to object that Godelian time travel so-called is not time travel after all.
To begin the explanation of the claims, I need to say in some detail what
is meant by Godelian time travel. This type of time travel does not involve
any funny business with discontinuous world lines or world lines that are
"bent backwards" on themselves. Rather, the funny business all derives'from
the structure of the spacetime which, of course, cannot be Newtonian or
Minkowskian. The funny spacetimes contain continuous and even infinitely
differentiable timelike curves such that if one traces along such a curve, always
moving in the future direction as defined by the globally defined external
time orientation, one eventually returns to the very same spacetime location
from whence one began. There is no room here for equivocation or alternative
descriptions; hence the second advantage. (More cautiously and more
precisely, there are some spacetimes admitting Godelian time travel in the
form of closed, future-directed timelike curves, and the curves cannot be
unrolled into open curves on which events are repeated over and over ad
infinitum—at least such a reinterpretation cannot be made without doing
damage to the local topological features of the spacetime; see section 6.3.) As
for the third advantage, consider a spacetime M, gab containing a CTC y that
is instantiated by, say, a massive particle. Pick a point pey and choose a
small neighborhood ,.;V of p. If" N is chosen wisely, all the causal relations in
the restricted spacetime JV, ga^ will be "normal." So if q e N is also on y
and is chronologically later than/), one would judge unequivocally that events
164 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

at p cause those at q and not vice versa. But in the encompassing spacetime
one might be tempted to say that backward causation is involved since,
although q is chronologically later than p, events at q causally influence those
atp because y emerges from jVand loops around to rejoin/*. But the situation
here is quite different from that in Wellsian time travel. In universes with
Godelian time travel it is consistent to assume—and, in fact, is implicitly
assumed in standard relativistic treatments—that all causal influences in the
form of energy-momentum transfers propagate forward in time with a speed
less than or equal to that of light. So in the case at issue, events at q causally
influence those at p because q chronologically precedes p and because there
is a continuous causal process linking q to p and involving always future-
directed causal propagation of energy-momentum. Of course, one could
posit that there is another kind of causal influence, not involving energy-
momentum transfer, by which events at q affect those at p backwards in time,
so that even if the future-directed segment of y from q to p were to disappear,
events at q would still influence those at p. But the point is that Godelian
time travel need not implicate such a backward causal influence.
We are now in a position to see why the second and third advantages
have been purchased at a price. One can object that Godelian time travel
does not deliver time travel in the sense wanted since Godelian time travel
so-called implies that there is no time in the usual sense in which to "go
back." In Godel's (1949a) universe, for example, there is no serial time order
for events, since every spacetime point p chronologically precedes itself; nor
is there a single time slice which would permit one to speak of the Godel
universe at a given time (see section 6.3). I feel that there is a good deal of
justice to this complaint. But I also feel that the phenomenon of "time travel"
in the Godel universe and in other general relativistic cosmologies is a worthy
object of investigation, whether under the label of "time travel" or under
another. The bulk of this chapter is devoted to that investigation.
Before starting on that task, it is worth mentioning for sake of completeness
other senses of time travel that appear in the literature. For example,
Chapman (1982) and Zemach (1968) devise various scenarios built around
the notion of "two times." One interpretation of such schemes would involve
the replacement of the usual relativistic conception of spacetime as a
four-dimensional manifold equipped with a Lorentz metric of signature
(+ + H ) (three space dimensions plus one time dimension) with a five-
dimensional manifold equipped with a metric of signature (+ + H )
(three space dimensions and two time dimensions). This scheme is worthy of
investigation in its own right, but I will confine attention here to standard
relativistic spacetimes.

6.3 The causal structure of relativistic spacetimes

There is an infinite hierarchy of causality conditions that can be imposed on

relativistic spacetimes (Carter 1971). I will mention only sufficiently many of
 Time Travel 165

these conditions to give some flavor of what the hierarchy is like. The review
also serves the purpose of introducing the concepts needed for an assessment
of Godelian time travel. To save the reader from having to refer back to
previous chapters, I will repeat some of the definitions of key concepts.
The basic presupposition of the causality hierarchy is that of temporal
orientability.

(CO) M, gab is temporally orienlable iff the null cones of gab admit of a
continuous division into two sets, 'past' and 'future'. 5

Which set is which is part of the problem of the direction of time, a problem
that for present purposes we may assume to have been resolved.
With a choice of temporal orientation in place, we can say that for
p, q 6 M, p chronologically precedes q (symbolically, p « q, or in previous notation
p e I~ (</)) just in case there is a smooth future-directed timelike curve
from p to q. Similarly, p causally precedes q (symbolically, p < q, or in previous
notation/) e J~ (q)) just in case there is a smooth future-directed non-spacelike
curve from p to q. It follows without any further restrictions on M, gab that
« and < are transitive relations. The first condition of the causality hierarchy
says that « has the other property we expect of an order relation, namely,
irreflexivity.

(Cl) M, gab exhibits chronology iff there is no p 6 M such that p « p.

Chronology is, of course, equivalent to saying that the spacetime does not
permit Godelian time travel.
The next condition up the hierarchy is simple causality.

(C2) M, gah exhibits simple causality iff there is no p e M such that

p<p.

The next step requires that distinct spacetime points have distinct
chronological pasts and futures. Formally,

(C3) M, gah is future (respectively, past) distinguishing iff for any p, q e M,

I + ( p ) = I+(q) =>p = q (respectively, I~ (p) = I~ (q) => p = q].

An equivalent definition states that M, gab is future (respectively, past)

distinguishing just in case for any p e M and any neighborhood N(p), there
is a neighborhood JV 1 '(p) £ N(p) such that no future (respectively, past)
directed causal curve from p intersects N'(p} more than once (Hawking and
Ellis 1973, p. 192).
Stronger than both simple causality and past and future distinguishing
is the condition of strong causality.
166 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

(C4) M, gab is strongly causal iff for any p € M and any neighborhood
JV(/>), there is a neighborhood jV'(/>) ^ N(p] such that no causal
curve intersects JV'(/») more than once.

Intuitively, strong causality not only rules out closed causal curves but also
"almost closed" causal curves. Carter (1971) showed that there is a countably
infinite hierarchy of conditions lying above (C4) which intuitively rule out
"almost almost closed" and "almost almost almost closed" etc. causal curves.
Strong causality is still not strong enough to guarantee the existence of a time
structure similar to that of familiar Newtonian or Minkowski spacetime, both
of which possess a time function. Recall that M,gab is said to possess a global
time function just in case there is a differentiate map t: M —> IR such that the
gradient of t is a past-directed timelike vector field. This implies that
t(p) < t(q) whenever p « q. The necessary and sufficient condition for such a
function is given in the next condition in the hierarchy.

(C5) M, gab is stably causal iff there exits on M a smooth non-vanishing

timelike vector field t" such that M, g'ab satisfies chronology, where
gab = gab ~ tatb and ta = gabt".

Intuitively, stable causality says that it is possible to widen out the null cones
of gab without allowing CTCs to appear. The proof that stable causality
implies strong causality uses the fact that a stably causal spacetime possesses
a global time function t. For any peM and any neighborhood N(p), a
judicious choice of a subneighborhood N'(p) can be made such that the value
of t on any causal curve leaving N'(p) is greater than its value on the curve
when entering N'(p). Since J must increase along a future-directed causal
curve, no such curve can intersect N'(p) more than once (Wald 1984a, p. 199).
None of the conditions given so far are enough to guarantee that causality
in the sense of determinism has a fighting chance on the global scale. That
guarantee is provided by a condition already encountered in connection with
the discussion of cosmic censorship—namely, global hyperbolicity. The
definition of this concept was given in chapter 2. Rather than repeat it here,
I will repeat the key fact about it:

(C6) M, gab is globally hyperbolic iff it possesses a Cauchy surface.

Recall that if M, gab admits one Cauchy surface, then it can be partitioned
by them. In fact, a global time function t can be chosen so that each of the
level surfaces t = constant is Cauchy.
The conditions ( C l ) through (C6) form a hierarchy in the sense that
each of (C2) through (C6) entails but is not entailed by the one below. There
are even stronger causality conditions above (C6), but they will play no role
in what follows. The philosophical literature has devoted most of its attention
to the ends of the hierarchy, principally to (CO) through (C2) and to (C6),
 Time Travel 167

and has largely neglected (C3) through (C5) and the infinity of other
intermediate conditions that have not been enumerated. There are both good
and dubious reasons for this selective attention. The intimate connection of
(CO) and (C6) respectively to the long-standing philosophical problems of
the direction of time and determinism is enough to explain and justify the
attention lavished on these conditions. Focusing on ( G l ) and (G2) to the
exclusion of (C3) through (C5) can be motivated by two considerations. First,
if one takes seriously the possibility that chronology can be violated, then one
must a fortiori take seriously the possibility that everything above can fail.
Second, Joshi (1985) showed that (C2) together with a continuity condition,
called causal simplicity, entail a good bit of the hierarchy above (C2). The
condition says that J± (p) are closed sets for all p £ M. If J + ( p ) were not
closed, there would have to be a situation where/) < qn, n = 1, 2, 3, . . . , with
qn —> q but ~1 (p < q), i.e., a causal signal can be sent from p to each of the
points qn but not to the limit point q.
Despite these good reasons for the selective focus, I suspect that most of the
philosophical attention lavished on (Cl) derives from the fascination with the
paradoxes of time travel, and that I take to be a dubious motivation. But
before taking up this matter in section 6.5, I turn to reasons for taking seriously
the possibility of chronology violation.

6.4 Why take Godelian time travel seriously?

Any relativistic spacetime M, gah based on a compact M contains CTCs (see

chapter 2). Stronger results are derivable for cosmological models M, gab, Tab
of GTR. Tipler (1977a) established that if the cosmological model M, gab, Tab
satisfies EFE without cosmological constant, the weak energy condition (recall
that this requires that Tab V" Vb > 0 for any timelike V"), and the generic
condition (which requires that every timelike and null geodesic experiences
a tidal force at some point in its history), then compactness of M entails that
the spacetime is totally vicious in that p « p for every p e M.
CTCs are not confined to compact spacetimes. In Godel's (1949a)
cosmological model, M = K4.6 This example is also important in how it
illustrates that the failure of chronology can be intrinsic in that chronology
cannot be restored by 'unwinding' the CTCs. More precisely, an intrinsic
violation of chronology occurs when the CTCs do not result (as in Fig. 6.2b)
by making identifications in a chronology-respecting covering spacetime
(Fig. 6.2a).
Godel spacetime is totally vicious. But there are other cosmological
models satisfying EFE and the energy conditions where chronology is violated
but not viciously. This raises the question of whether it is possible to have a
spacetime M, gab where the chronology-violating set V c: M is non-empty but
so small that it is unnoticeable in the sense of being measure zero. The answer
is negative since V is always an open set (see Hawking and Ellis 1973).
168 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 6.2 (a) Two-dimensional Minkowski spacetime; (b) rolled-up two-dimensional

Minkowski spacetime

In the Godel universe all CTCs are non-geodesic, necessitating the use
of a rocket ship to accomplish the time travel journey. Malament (1985, 1987)
provided estimates of the acceleration and the fuel/payload ratio needed to
make such a journey. These quantities are so large as to make the journey a
practical impossibility. It was on this basis that Godel (1949b) himself felt
justified in ignoring the paradoxes of time travel. Such complacency, however,
is not justified. Ozsvath (1967) produced a generalization of Godel's model
that accommodates electromagnetism. De (1969) showed that in such a
universe time travellers do not need to use a rocket ship; if they are electrically
charged they can use the Lorentz force to travel along CTCs. Even better
from the point of view of a lazy would-be time traveller would be a
cosmological model with intrinsic chronology violation where some timelike
geodesies are closed. An example is provided by the Taub-NUT model which
is a vacuum solution to EFE (energy conditions trivially satisfied).
As a result of being totally vicious and simply connected, Godel spacetime
does not contain a single time slice7 so that one cannot speak of the Godel
universe at a given time. But there are solutions to EFE which are intrinsically
chronology violating but which do contain time slices. Indeed, the time slices
can themselves be achronal, and thus partial Cauchy surfaces,8 even though
CTCs develop to the future. This raises the question of whether GTR allows
for the possibility of building a time machine whose operation in some sense
causes the development of CTCs where none existed before. This matter will
be taken up in section 6.10.
None of the chronology-violating models discussed so far (with the
exception of the trivial example of Fig. 6.2b) are asymptotically flat. But
CTCs can occur in such a setting. The Kerr solutions to EFE form a
two-parameter family described by the value of the mass M and the angular
momentum A. The case where M2 > A2 is thought to describe the unique
final exterior state of a non-charged stationary black hole. In this case
 Time Travel 169

chronology is satisfied. When M2 < A2 the violation of chronology is totally

vicious. Charlton and Clarke (1990) suggest that the latter case could arise
if a collapsing rotating star does not dissipate enough angular momentum to
form a black hole.
Van Stockum's (1937) solution to EFE, corresponding to a source
consisting of an infinite rotating cylinder of dust, contains CTCs. Tipler
(1974) suggests that chronology violation may also take place for a finite
cylinder source if the rotation rate is great enough.
The Godel, Ozsvath, Kerr-Newman, and van Stockum models all
involve rotating matter. But this is not an essential condition for the
appearance of CTCs in models of GTR—recall that the Taub-NUT model
is a vacuum solution. Also Morris, Thorne, and Yurtsever (1988) found that
generic relative motions of the mouths of traversable "wormholes" (multiply
connected surfaces) can produce CTCs, as can generic gravitational redshifts
at the wormhole mouths (Frolov and Novikov 1990). However, the main-
tenance of the traversable wormholes implies the use of exotic matter that
violates standard energy conditions. Such violations may or may not be
allowed by quantum field theory (see section 6.10).
Gott (1991) found that the relative motion of two infinitely long cosmic
strings can give rise to CTCs. This discovery has generated considerable
interest and controversy (see Carroll et al. 1992; Deser et al. 1992; Deser
1993; Ori 1991a; 't Hooft 1992). Part of the interest lies in the fact
that Gott's solution, unlike the wormhole solutions, does not violate the
standard energy conditions of classical GTR. The global structure of Gott's
solution has been elucidated by Cutler (1992).
The upshot of this discussion is that since the pioneering work of Godel
some forty years ago, it has been found that CTCs can appear in a wide
variety ofcircumstanc.es described by classical GTR and semiclassical quantum
gravity. And more broadly, there are many other known examples of
violations of causality principles higher up in the hierarchy. One reaction,
which is shared by a vocal if not large segment of the physics community,
holds that insofar as these theories are to be taken seriously, the possibility of
violations of various conditions in the causality hierarchy, including chronol-
ogy, must also be taken seriously. This is the attitude of the "Consortium"
led by Kip Thorne (see Friedman et al. 1990). Another vocal and influential
minority conjectures that GTR has within itself the resources to show that
chronology violations can be ignored because, for example, it can be proved
that if chronology violations are not present to begin with, they cannot arise
from physically reasonable initial data, or because such violations are of
"measure zero" in the space of solutions to EFE. This position is championed
by Hawking (1992). If such conjectures turn out to be false, one can still take
the attitude that in the short run classical GTR needs to be supplemented
by principles that rule out violations of the causality hierarchy, and one can
hope that in the long run the quantization of gravity will relieve the need for
such ad hoc supplementation. Which of these attitudes it is reasonable to
170 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

adopt will depend in large measure on whether it is possible to achieve a

peaceful coexistence with CTCs. It is to that matter I now turn.

6.5 The paradoxes of time travel

The darling of the philosophical literature on Godelian time travel is the

"grandfather paradox" and its variants. Example: Kurt travels into the past
and shoots his grandfather at a time before grandpa became a father, thus
preventing Kurt from being born, with the upshot that there is no Kurt to
travel into the past to kill his grandfather so that Kurt is born after all and
travels into the past . . . (Though the point is obvious, it is nevertheless worth
emphasizing that killing one's grandfather is overkill. If initially Kurt was
not present in the vicinity of some early segment of his grandfather's world
line, then traveling along a trajectory that will take him into that vicinity,
even if done with a heart innocent of any murderous intention, is enough to
produce an antinomy. This remark will be important for the eventual
unraveling of the real significance of the grandfather paradox.)
On one level it is easy to understand the fascination that such paradoxes
have exercised—they are cute and their formal elucidation calls for the sorts
of apparatus that is the stock-in-trade of philosophy. But at a deeper level
there is a meta-puzzle connected with the amount of attention lavished on
them. For what could such paradoxes possibly show? (1) Could the grandfather
paradox show that Godelian time travel is not logically or mathematically
possible?9 Certainly not, for we have mathematically consistent models in
which CTCs are instantiated by physical processes. (2) Could the grandfather
paradox show that Godelian time travel is not conceptually possible? Perhaps
so, but it is not evident what interest such a demonstration would have. The
grandfather paradox does bring out a clash between Godelian time travel
and what might be held to be conceptual truths about spatiotemporal/causal
order. But in a similar way the twin paradox of special relativity theory reveals
a clash between the structure of relativistic spacetimes and what were held
to be conceptual truths about time lapse. The special and general theories of
relativity have both produced conceptual revolutions. The twin paradox and
the grandfather paradox help to emphasize how radical these revolutions are,
but they do not show that these revolutions are not sustainable or contain
inherent contradictions. (3) Could the grandfather paradox show that Godelian
time travel is not physically possible? No, at least not if "physically possible"
means compatibility with EFE and the energy conditions, for we have models
which satisfy these laws and which contain CTCs. (4) Could the paradox
show that although Godelian time travel is physically possible it is not
physically realistic? This is not even a definite claim until the relevant sense
of "physically realistic" is specified. And in the abstract it is not easy to see
how the grandfather paradox would support that claim as opposed to the
claim that time travel is flatly impossible. Additional factors such as the need
 Time Travel 171

for high accelerations to complete a time travel journey or the instability of

Cauchy horizons connected with CTCs (see section 6.10) would seem to be
needed to support the charge that Godelian time travel is physically
unrealistic. If anything, such factors tend to mitigate the force of the paradoxes
(see sections 6.6, 6.8, and 6.10).
(5) Doesn't the grandfather paradox at least demonstrate that there is a
tension between time travel and free will? Of course Kurt cannot succeed in
killing his grandfather. But one might demand an explanation of why Kurt
doesn't succeed. He had the ability, the opportunity, and (let's assume) the
desire. What then prevented him from succeeding? Some authors pose this
question in the rhetorical mode, suggesting that there is no satisfactory answer
so that either time travel or free will must give way. But if the question is
intended non-rhetorically, it has an answer of exactly the same form as the
answer to analogous questions that arise when no CTCs exist and no time
travel is in the offing. Suppose, for instance, that in the time travel scenario
Kurt had his young grandfather in the sights of a .30-30 rifle but didn't pull
the trigger. The reason the trigger was not pulled is that laws of physics and
the relevant circumstances make pulling the trigger impossible at the relevant
spacetime location. With CTCs present, global Laplacian determinism (which
requires a Cauchy surface, as discussed in chapter 3) is inoperable. But local
determinism makes perfectly good sense. In any spacetime M, gab, chronology-
violating or not, and at any p € M one can always choose a small enough
neighborhood JV of p such that JV, gab\N possesses a Cauchy surface 2 with
/»eJ + (S). And the relevant initial data on E together with the coupled
Einstein—matter equations will uniquely determine the state at p. Taking p
to be the location of the fateful event of Kurt's pulling/not pulling the trigger
and carrying through the details of the deterministic physics for the case in
question shows why Kurt didn't pull the trigger. Of course, one can go on
to raise the usual puzzles about free will; namely, granting the validity of
what was just said, is there not a way of making room for Kurt to have
exercised free will in the sense that he could have done otherwise? At this
point all of the well-choreographed moves come into play. There are those
(the incompatibilists) who will respond with arguments intended to show that
determinism implies that Kurt couldn't have done otherwise, and there are
others (the compatibilists) Waiting to respond with equally well-rehearsed
counterarguments to show that determinism and free will can coexist in
harmony. But all of this has to do with the classic puzzles of determinism and
free will and not with CTCs and time travel per se.
(6) Perhaps we have missed something. Suppose that Kurt tries over and
over again to kill his grandfather. Of course, each time Kurt fails—sometimes
because his desire to pull the trigger evaporates before the opportune moment,
sometimes because although his murderous desire remains unabated his hand
cramps before he can pull the trigger, sometimes because although he pulls
the trigger the gun misfires, sometimes because although the gun fires the
bullet is deflected, etc. In each instance we can give a deterministic
172 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

explanation of the failure. But the obtainment of all the initial conditions that
result in the accumulated failures may seem to involve a coincidence that is
monstrously improbable (see Horwich 1989). Here we have reached a real
issue but one which is not easy to tackle.
A first clarificatory step can be taken by recognizing that the improbability
issue can be formulated using inanimate objects. (Consider, for example, the
behavior of the macroscopic objects in my study as I write: a radiator is
radiating heat, a light bulb is radiating electromagnetic waves, etc. If the
world lines of these objects are CTCs, it would seem to require an improbable
conspiracy to return these objects to their current states, as required by the
completion of the time loop.) Since free will is a murky and controversial
concept, it is best to set it aside in initial efforts at divining the implications
of the grandfather paradox. After some progress has been made it may then
be possible to draw some consequences for free will. As a second step we need
to formalize the intuition of improbability. One method would be to define
a measure on the space of solutions to EFE and to try to show that the
solutions corresponding to some kinds of time travel (those involving the
functional equivalent of Kurt trying over and over again to kill his grandfather)
have negligible or flatly zero measure. Even if such a demonstration is
forthcoming, we still have to face the question: So what? (After all, some types
of space travel will be measure zero, but this hardly shows that the concept
of space travel is suspect.) The answer will depend crucially on the justification
for and significance of the measure. This matter will receive some attention
in section 6.8. But for the moment I want to note that the impression of
improbability in connection with time travel stories may not be self-
reenforcing. In the above example the judgment of the improbability of the
failure of Kurt's repeated attempts to kill his grandfather was made relative
to our (presumably chronology respecting) world; but perhaps from the
perspective of the time travel world itself there is no improbability. By way
of analogy, suppose that the actual world is governed by all the familiar laws
of classical relativistic physics save for Maxwell's laws of electromagnetism.
If we peered into another world which was nomologically accessible from our
world but which was governed by Maxwell's laws we would see things that
from our perspective are improbable ("measure zero") coincidences. We
would find, for example, that the electric and magnetic fields on a time slice
cannot be freely specified but must satisfy a set of constraints; and we would
find that once these constraints are satisfied at any moment they are thereafter
maintained for all time (see chapter 5). Amazing! But, of course, from the
perspective of the new world there is no improbability at all; indeed, just the
opposite is true since the "amazing coincidences" are consequences of the
laws of that world. That this analogy may be opposite to the case of time
travel will be taken up in sections 6.6 and 6.7.
What then remains of the grandfather paradox? The paradox does point
to a seemingly awkward feature of spacetimes that contain CTCs: local data
are constrained in a way that is not present in spacetimes with a more normal
 Time Travel 173

causal structure. But the forms in which the paradox has been considered in
the philosophical literature are of little help in getting an accurate gauge of
the shape and extent of the constraints. And by itself the paradox is of no
help at all in assessing the status of these consistency constraints.

6.6 Consistency constraints

The laws of special and general relativistic physics that will be considered
here are all local in the following twofold sense. First, they deal with physical
situations that are characterized by local geometric object fields 0 (e.g., scalar,
vector, tensor fields) on a manifold M. Second, the laws governing these fields
are in the form of local ordinary or local partial differential equations. The
result is a global-to-local property: if M, gab, 0 satisfies the laws and U £ M
is an open neighborhood, then U,gab\N,0\v also satisfies the laws. (This
property holds whether or not CTCs are present.) Thus, it would seem at
first blush that the question of whether some local state of affairs is physically
possible can be answered by focusing exclusively on what is happening locally
and ignoring what is happening elsewhere.
In Minkowski spacetime and in general relativistic spacetimes with
nice causality properties we typically have the reverse local-to-global prop-
erty: any local solution can be extended to a global solution.10 Consider,
for example, the source-free wave equation for a massless scalar field
CD: gabVaVb<f> =• Dd> = 0, where Va is the derivative operator associated with
gab. On Minkowski spacetime (M = IR4 and gab = rjab (Minkowski metric)),
any Cm solution on an open U c: IR4 can be extended to a full solution on
1R4. But obviously this local-to-global property fails for the chronology-
violating spacetime of Fig. 6.2b. Figure 6.3a shows a local solution with a
single pencil of rays traversing U. This solution is obviously globally
inconsistent since the light rays from U will trip around the spacetime and
reintersect U.
The point is straightforward, but some attempts to elaborate it make it
sound mysterious. Thus, consider the presentation of the Consortium:

Fig. 6.3 (a) Light rays and (b) billiard balls in rolled-up Minkowski spacetime
174 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

The only type of causality violation the authors would find unacceptable is
that embodied in the science-fiction concept of going backward in time and
killing one's younger self ("changing the past" [grandfather paradox]).
Some years ago one of us (Novikov) briefly considered the possibility that
CTC's might exist and argued that they cannot entail this type of causality
violation: Events on a CTC are already guaranteed to be self-consistent,
Novikov argued; they influence each other around the closed curve in a
self-adjusted, cyclical, self-consistent way. The other authors have recently
arrived at the same viewpoint.
We shall embody this viewpoint in a principle of self-consistency, which
states that the only solutions to the laws of physics that can occur locally in
the real universe are those which are globally self-consistent. This principle allows
one to build a local solution to the equations of physics only if that
local solution can be extended to be part of a (not necessarily) unique global
solution, which is well defined throughout the nonsingular regions of
spacetime. (Friedman et al. 1990 pp. 1916-1917)

The first part of the quotation seems to invoke either a notion of preestablished
harmony or else a guiding hand that prevents the formation of an inconsistent
scenario. But once such connotations are removed, the "principle of self-
consistency" (PSC) threatens to deflate into a truism. Here is the Consortium's
comment:
That the principle of self-consistency is not totally tautological becomes clear
when one considers the following alternative: The laws of physics might
permit CTC's; and when CTC's occur, they might trigger new kinds of local
physics, which we have not previously met. . . . The principle of self-
consistency is intended to rule out such behavior. It insists that local physics
is governed by the same types of physical laws as we deal with in the absence
of CTC's. . . . If one is inclined from the outset to ignore or discount the
possibility of new physics, then one will regard self-consistency as a trivial
principle, (ibid., p. 1917)

What the Consortium means by discounting the possibility of "new physics"

is, for example, ignoring the possibility that propagation around a CTC can
lead to multivalued fields, calling for new types of laws that tolerate such
multivaluedness. I too will ignore this possibility. But I will argue shortly that
CTCs may call for "new physics" in another sense. For the moment, however,
all I want to insist upon is that taking the PSC at face value seems to force
a distinction between two senses of physical possibility.
In keeping with the global-to-local property introduced above, we can
say again more pedantically what was already said informally: a local
situation is physically possible-^ just in case it is a local solution of the laws.
But the PSC—which says that the only solutions that can occur locally are
globally self-consistent—seems to require a more demanding and relativized
sense of physical possibility; namely, a local situation is physically possible^
in a spacetime M, gab just in case the solution can be extended to a solution
of the laws on all of M, gah. If the conditions a local solution must fulfill in
order to be extendible to a global solution are called the consistency
 Time Travel 175

constraints, one can roughly paraphrase physical possibility2 as physical

possibility! plus satisfaction of the consistency constraints.11
This distinction might be regarded as desirable for its ability to let one
have one's cake and eat it too. On the one hand, we have the intuition that
it is physically possible to construct and launch a rocket probe in any direction
we like with any velocity less than that of light. This intuition is captured by
physical possibility!. But on the other hand it is not possible to realize all of
the physically possible! initial conditions for such a device in spacetimes with
certain kinds of CTCs since the traverse of a CTC may lead the probe to
interfere with itself in an inconsistent way. (Or so it would seem; but see the
discussion of section 6.8.) This impossibility is captured by the failure of
physical possibility2.
On reflection, however, having one's cake and eating it too is, as usual,
too good to be true. Thus, one might maintain with some justice that to be
physically impossible simply is to be incompatible with the laws of physics—as
codified in the definition of physical possibility!. So as it stands the notion of
physical impossibility2 seems misnamed when it does not reduce to physical
impossibility!, as it apparently does not when CTCs are present. To come at
the point from a slightly different angle, let us reconsider the grandfather
paradox. It was suggested in the preceding section that Kurt's failure to carry
out his murderous intentions could be explained in the usual way—by
reference to conditions that obtained before the crucial event and the (locally)
deterministic evolution of these conditions. But while not incorrect, such an
explanation deflects attention from a doubly puzzling aspect of spacetimes
with CTCs. First, it may not even be possible for Kurt to set out on his
murderous journey, much less to carry out his intentions. And second, the
ultimate root of this impossibility does not lie in prior contingent conditions
since there are no such conditions that can be realized in the spacetime at
issue and which eventuate in the commencement of the journey. The ultimate
root of this impossibility taps the fact that (as we are supposing) there is no
consistent way to continue Kurt's murderous journey in the spacetime. But,
one might complain, to call this impossibility "physical impossibility2" is to
give a label that is not backed by any explanatory power; for given the way
the story has been told so far the local conditions corresponding to the
commencement of Kurt's journey are compatible with all the postulated laws.
In such a complaint the reader will detect a way of trying to scratch the
residual itch of the grandfather paradox.
The itch must be dealt with once and for all. I see three main treatments,
the first two of which promise permanent cures while the third denies the
existence of the ailment.

6.7 Therapies for time travel malaise

(Tl) This treatment aims at resolving the tension between physical possibility!
and physical possibility2 by getting rid of the latter. The leading idea is to
176 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

argue that GTR shows that, strictly speaking, the notion of consistency
constraints is incoherent. For example, looking for consistency constraints in
a spacetime M, gab for a scalar field obeying DO = 0 makes sense if $ is
treated as a test field on a fixed spacetime background. But (the argument
continues) this is contrary to both the letter and spirit of GTR. For $ will
contribute to the total energy—momentum—the usual prescription being that
Tab(<&) = Va$Vs<I) - l/'2gabVc$Vc$>—that generates the gravitational field
cum metric. And (one could conjecture) if <E> and <!>' are interestingly different
(say, they differ by more than an additive constant), then the metrics gab and
g'ab solving EFE for the corresponding Tab(Q>) and T'ab($>') will be different
(i.e., non-isometric). This therapy is radical in that if it succeeds it does so
by the draconian measure of equating the physically possible! local states
with the actual states. This is intuitively unsatisfying. If we restrict attention
to <J>'s such that Tah(<&) is small in comparison with the total Tab and the
spacetime is stable under small perturbations of Tab, then <X> can to good
approximation be treated as a test field. Questions of stability will be
examined in section 6.10, but in the meanwhile I will assume that they can
be set aside. One could also object that (Tl) is inapplicable in cases where
there are GTGs and where the laws entail that the spacetime structure is
non-dynamical (in the sense of not responding to the matter content) and
that a variety of physically possible} states can be realized on a given local
region. However, the strength of this objection is hard to grasp since the laws
in question would have to be rather different from those of our world, at least
if something akin to GTR is true. And recall that the success of GTR is the
main reason for taking Godelian time travel seriously.
(T2) The second treatment strategy is to naturalize physical possibility2.
The idea is, first, to insist that physical possibility (relative to a world) just is
the compatibility with the laws (of that world) and, second, to go on to argue
that physical possibility2 can be brought into the fold by showing that in
chronology-violating environments the consistency constraints of physical possi-
bility2 have law status. Thus, (T2) insists that, contrary to the Consortium's ex-
planation of the PSC, there is a sense in which CTCs do call forth "new physics."12
(T2) can take two forms, (a) The naturalization of physical possibility2
would amount to a reduction to physical possibility}, understood as consistency
with the local laws of physics, if the consistency constraints/new laws were
purely local so that, even in the chronology-violating environments, what
is physically possible locally is exactly what is compatible with the (now
augmented) local laws of physics, (b) Unfortunately, the reduction of (a) can
be expected in only very special cases. In general, the consistency constraints
may have to refer to the global structure of spacetime. In these latter cases,
insofar as (T2) is correct, the concept of physical possibility! must be
understood to mean consistency of the local situation with all the laws, local
and non-local. The patient who demands a purely local explanation of the
difference between local conditions that are physically possible and those
which are not will continue to itch.
 Time Travel 177

(T3) If the first two therapies fail, the discomfort the patient feels can
be classified as psychosomatic. The therapist can urge that the patient is
getting overexcited about nothing or at least about nothing to do specifically
with time travel; for global features of spacetime other than CTCs can also
impose constraints on initial data. For example, particle horizons in standard
big bang cosmologies prevent the implementation of the Sommerfeld radiation
condition which says that no source-free electromagnetic radiation comes in
from infinity (see chapter 5). Here the patient may brighten for a moment
only to relapse into melancholy upon further reflection. For the constraints
entailed by the particle horizons are of quite a different character than those
entailed by the typical chronology-violating environment; the former, unlike
the latter, do not conflict with the local-to-global property and thus do not
drive a wedge between physical possibility! and physical possibility2. And in
any case the choice of the particle horizons example is not apt for therapeutic
purposes since these horizons are widely thought to be so problematic as to call
for new physics involving cosmic inflation or other non-standard scenarios.13
Clearly this line of therapy opens up a number of issues that require careful
investigation; but such an investigation is beyond the scope of this book.
My working hypothesis favors (T2). This is not because I think that (T2)
will succeed; indeed, I am somewhat pessimistic about the prospects of success.
Nevertheless, making (T2) the focus of attention seems justified on several
grounds: the success of (T2) would provide the most satisfying resolution of
the nagging worries about time travel, while its failure would have significant
negative implications for time travel; whether it succeeds or fails, (T2)
provides an illuminating perspective from which to read recent work on the
physics of time travel; and finally, (T2) forces us to confront issues about the
nature of the concept of physical law in chronology-violating spacetimes,
issues which most of the literature on time travel conveniently manages to
avoid.
It is well to note that my working hypothesis is incompatible with some
analyses of laws. For example, Carroll (1994) rejects the idea that laws
supervene on occurrent facts, 14 and adopts two principles which have the
effect that laws of the actual world W@ are transportable as laws to other
possible worlds which are nomologically accessible from H^g. The first
principle says that "if/" is physically possible and Q_is a law, then Q, would
(still) be a law if P were the case" (p. 59). The second says that "if P is
physically possible and Q,is not a law, then Q_ would (still) not be a law if P
were the case" (p. 59). Let P say that spacetime has the structure of Godel
spacetime or some other spacetime with CTCs. And let us agree that P is
physically possible because it is compatible with the laws of W^ (which for
sake of discussion we may take to be the laws of classical general relativistic
physics). It follows from Carroll's principles that if spacetime were Godelian,
the laws of W((il would still be laws and also that these would be the only laws
that would obtain.
I will not attempt to argue here for the supervenience of laws on occurrent
178 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

facts but will simply assume it. In exploring my working hypothesis, I will
rely on an account of laws that can be traced back to John Stuart Mill (1843);
its modern form is due to Frank Ramsey (1928/29) and David Lewis (1973).
The gist of the Mill-Ramsey-Lewis (MRL) account is that a law for a
logically possible world Wis an axiom or theorem of the best overall deductive
system for W (or what is common to the systems that tie for the best). A
deductive system for W is a deductively closed, axiomatizable, set of
(non-modal) sentences, each of which is true in W. Deductive systems are
ranked by how well they achieve a compromise between strength or
information content on the one hand and simplicity on the other. Simplicity
is a notoriously vague and slippery notion, but the hope is that, regardless of
how the details are settled, there will be for the actual world a clearly best
system or at least a non-trivial common core to the systems that tie for best.
If not, the MRL theorist is prepared to admit that there are no laws for our
world.
The MRL account of laws is naturalistic (all that exists is spacetime
and its/contents), actualistic (there is only one actual world); and empiricistic
(a world is a totality of occurrent facts; there are no irreducible modal facts).
In addition, I would claim that this account fits nicely with the actual
methodology used by scientists in search of laws. The reader should be
warned, however, that it is far from being universally accepted among
philosophers of science; see, for example, van Fraassen (1989) and my response
in Earman (1993). I will not attempt any defense of the MRL account here.
If you like, the ability to illuminate the problems of time travel can be
regarded as a test case for the MRL account.
Suppose for sake of discussion that the actual world has a spacetime
without CTGs; perhaps, for example, its global features are described more
or less by one of the FRW big bang models. And suppose that the MRL laws
of this world are just the things dubbed laws in textbooks on relativistic
physics, no more, no less. Now consider some other logically possible world
whose spacetime contains CTCs. But so as not to waste time on possibilities
that are too far removed from actuality, let us agree to restrict attention to
worlds that are nomologically accessible from the actual world in that the
laws of the actual world, taken as non-modal propositions, are all true of these
worlds. Nevertheless, one cannot safely assume that the MRL laws of our
world "govern" these time travel worlds in the sense that the set of laws of
our world coincides witTh the set of laws of time travel worlds.
One possibility is that the MRL laws of a time travel world W consist
of the MRL laws of this world plus the consistency constraints on the test
fields in question. If so, we have a naturalization of physical possibility2,
though it would remain to be seen whether the naturalization takes the
preferred (a) form or the less desirable (b) form. Additionally, time travel
would have implications for free will. In cases where an action is determined
by the laws plus contingent initial conditions, compatibilists and incompatibil-
ists split on whether the actor could be said to have the power to do otherwise
 Time Travel 179

and whether the action is free. But all parties to the free will debate agree
that if an action is precluded by the laws alone without the help of contingent
boundary or initial conditions, then the action is not in any interesting sense
open to the agent. Thus, if the possibility under discussion pans out, there
are various actions that, from a compatibilist perspective at least, we are free
to perform in this world that we are not free to perform in various time travel
scenarios.
Other possibilities also beg for consideration. For instance, it could turn
out that although (by construction) the MRL laws of this world are all true
of a time travel world W, they are not all laws of W, except in a very tenuous
sense. I will argue that this possibility is realized in cases where the consistency
constraints are so severe as to supplant the laws of this world. In such cases
the time travel involved is arguably such a remote possibility that it loses
much of its interest. But note that since the consistency constraints are still
subsumed under the laws of the time travel world, we retain the desirable
feature that physical possibility is naturalized.
Finally, these remarks point to the intriguing possibility that purely local
observations can give clues to the global structure of spacetime without the
help of a supplementary " cosmological principle"; namely, local observations
may reveal the absence of consistency constraints that would have to obtain
if we inhabited certain kinds of chronology-violating spacetimes.
What is needed as a first step in coming to grips with these matters is a
study of the nature of consistency conditions on test fields that arise for various
chronology-violating spacetimes. The recent physics literature has made some
progress on the project. In the next two sections I will report on some of the
results for self-interacting and non-self-interacting fields. On the basis of these
results I will advance some tentative conclusions to the series of questions
posed above concerning physical possibility and laws in chronology-violating
worlds.

6.8 Non-self-interacting test fields

When considering chronology-violating spacetimes, the simplest regime to

study mathematically is the case of a non-self-interacting field, e.g., solutions
to the source-free scalar wave equation D<B = 0. Of course, the grandfather
paradox and the related paradoxes of time travel that have been discussed
in the philosophical literature typically rely on self-interacting systems.
Even so, we shall see that non-trivial consistency conditions can emerge in
the non-self-interacting regime. But on the way to illustrating that point it is
worth emphasizing the complementary point that in small enough regions of
some chronology-violating spacetimes the consistency constraints for non-self-
interacting fields do not make themselves felt so that local observations in such
regions will not reveal the presence of CTCs.
Following Yurtsever (1990), call an open U c M causally regular for the
180 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

spacetime M, gab with respect to the scalar wave equation just in case for
every C100 solution O of D<P = 0 on U, there is a C100 extension to all
ofM, i.e., there is a C°° 2> on M such that D<D = 0 and &\v = <5. M, gab can
be said to be causally benign with respect to the scalar wave equation just
in case for every p e M and every open neighborhood U of p there is a
subneighborhood U' c U which is causally regular.
The two-cylinder of Fig. 6.2b is causally benign with respect to the scalar
wave equation. The following remarks, while not constituting a proof of this
fact, give an indication of why it holds in the optical limit. In that limit $
waves propagate at the speed of light (i.e., along null trajectories). At any
point on the cylinder a small enough neighborhood can be chosen such that
any null geodesic leaving this neighborhood in either the future or past
direction never returns. Consider then any solution on this neighborhood. To
extend this local solution to a global one, simply propagate the solution out
of the base neighborhood along null geodesies. If the propagated field does
not reach a point q 6 M, set O(^) = 0. If two null geodesies from the base
neighborhood cross at q, obtain O(^) by adding the propagated fields.
Consider next the toroidal spacetime 7^,., obtained from two-dimensional
Minkowski spacetime by identifying the points (x, t) and (x', t') when
x = x' mod 1 and t' — t mod r, where r > 0 is a real number. For r rational,
7(\ ,,) is benign with respect to the scalar wave equation, as shown by
Yurtsever (1990). Through any point on 7^ r) there is a time slice that lifts
to many t = constant surfaces in the Minkowski covering spacetime IR 2 , rjat>.
Consider one such surface, say, / = 0. Any solution ®(x, t) on 7(i ir) induces
on t = 0 initial data O 0 (*) ~®(x, 0) and 60(*) —-(d/di)®(x, f)\t =0. By con-
sidering solutions on R2, rjab of the wave equation that develop from this
initial data it follows that both O0 and d>0 must be periodic with periods 1
and r. Further, <I>0 must satisfy the integral constraint

When r is rational we can choose a small enough neighborhood of any point

on 7^ r) such that arbitrary initial data O0 and <J)0 can be extended so as to
meet the periodicity and integral constraints.
Friedman et al. (1990) argue that the benignity property with respect to
the scalar wave equation also holds for a class of chronology-violating
spacetimes that are asymptotically flat and globally Minkowskian except for
a single wormhole that is threaded by CTCs. In some of these spacetimes
there is a partial Cauchy surface E such that chronology violations lie entirely
to the future of E. It is argued that the formation of CTCs places no
consistency constraints on initial data specified on E.
In all the examples considered so far the chronology violations arc
non-intrinsic in that they result from making identifications of points in
 Time Travel 181

a chronology-preserving covering spacetime. Unfortunately, because of the

non-trivial mathematics involved, almost nothing is known about the benignity
properties of spacetimes with intrinsic chronology violations. I conjecture that
Godel spacetime is benign with respect to the scalar wave equation. The
conjecture is based on the fact that null geodesies in Godel spacetime are not
only open but are never almost closed, i.e., for any point of Godel spacetime
and any open neighborhood of that point there is a subneighborhood such that
every inextendible null geodesic starting in the neighborhood eventually
leaves and never returns. Thus, it may be possible to generalize to Godel
spacetime the construction indicated above for extending local solutions to
global solutions on the two-cylinder. 15
If correct, this conjecture would cast new light on a puzzling feature of
Godel's (1949b) own attitude towards the grandfather paradox in Godel
spacetime. Basically his attitude was one of "why worry" since the fuel
requirements on a rocket needed to realize a time travel journey in Godel
spacetime are so demanding as to be impossible to meet by any practical
scheme. But consistency constraints are constraints whether or not they can
be tested by practical means. So it would seem that whatever puzzles arise
with respect to the status of such constraints are unresolved by appeal to
practical considerations. However, the above conjecture, if correct, suggests
that for non-self-interacting systems the consistency constraints in Godel
spacetime are much milder than one might have thought. Similarly, Godel's
remarks can be interpreted as suggesting that the constraints will also be mild
for self-interacting systems. Some further information on this matter is
obtained in section 6.8.16
It should be emphasized at this juncture that in spite of the connotations
of the name, benignity does not necessarily imply physics as usual. For it does
not imply that there are no non-trivial consistency constraints nor that -the
constraints cannot be detected locally. Benignity implies only that the
constraints cannot be felt in sufficiently small neighborhoods, but this is
compatible with their being felt in regions of a size that we typically observe.
To give an example of a non-benign spacetime we can return to 7^ ^ and
choose r to be an irrational number. Now the periodicity constraints on the
initial data cannot be satisfied except for <J>0 and ^o constant. The integral
constraint then requires that $0 = 0, with the upshot that the only solutions
allowed are <t> = constant everywhere. No local solution that allows the tiniest
variation in <D can be extended to a global solution.
I turn now to the question of the status of the consistency constraints for
chronology-violating spacetimes. By way of introduction, I note the assertion
of the Consortium that a time traveler "who went through a wormhole [and
thus around a CTG] and tried to change the past would be prevented by
physical law from making the change" (Friedman et al. 1990, p. 1928; italics
added). One way of interpreting this assertion is in line with my working
hypothesis that the consistency constraints entailed by the presence of CTCs
in a nomologically accessible chronology-violating world arc laws of that
182 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

world. This position is arguably endorsed by the MRL account of laws. For
it is plausible that in each of the above examples the consistency constraints
would appear as axioms or theorems of (each of) the best overall true theories
of the world in question.
For the reasons discussed in sections 6.6 and 6.7, such a result is devoutly
to be desired. But to play the devil's advocate for a moment, one might charge
that the result is an artifact of the examples chosen. Each of these examples
involves a spacetime with a very high degree of symmetry, and it is this
symmetry one suspects of being responsible for the relative simplicity of the
consistency constraints. If this suspicion is correct, then the consistency
constraints that obtain in less symmetric spacetimes may be so complicated
that they will not appear as axioms or theorems of any theory that achieves
a good compromise between strength and simplicity. Due to the technical
difficulties involved in solving for the consistency constraints in non-symmetric
spacetimes, both the devil and his advocate may go blue in the face if they
each hold their breath while waiting for a confirmation of their suspicions. If
the devil's advocate should prove to be correct, the proponent of the
naturalization could still find comfort if the cases where the naturalization
fails could be deemed to be very remote possibilities.
An illustration of how time travel can be justly deemed to involve very
remote possibilities occurs when the chronology violating world W is so far
from actuality that, although the laws of the actual world are true of W7, they
are not laws of W except in a very attenuated sense. In the case of 7(1,,.) with
r irrational we saw that $ = constant are the only allowed solutions of the
scalar wave equation. I take it that DO = 0 will not appear as an axiom of
a best theory of the 7^ r) world so that the scalar wave equation is demoted
from fundamental law status. Presumably, however, <1> = constant will appear
as an axiom in any best theory. Of course, this axiom entails that Dfl> = 0;
but it also entails any number of other differential equations that are
incompatible with one another and with the scalar wave equation when <D
is not constant. Thus, in the T ( l t f ) world the scalar wave equation and its
rivals have much the same status as "All unicorns are red," "All unicorns are
blue," etc. in a world where it is a law that there are no unicorns. In this
sense DCD> = 0 has been supplanted as a law. This still is a case where physical
possibility2 is naturalized by reduction in the strong sense to physical
possibility! since the consistency constraint is stated in purely local terms. But
the more remote the possibility, the less interesting the reduction. And in this
case the possibility can be deemed to be very remote since the relation of
nomological accessibility has become non-symmetric—by construction the
toroidal world in question is nomologically accessible from the actual world,
but the converse is not true because the toroidal law <1> = constant is violated
here.
One might expect that such supplantation will take place in any
world with a spacetime structure that is not benign. What I take to be a
counterexample to this expectation is provided by the four-dimensional
 Time Travel 183

toroidal spacetime 7<i i i i) obtained from four-dimensional Minkowski space-

time by identifying the points (x,j, z, t) and (x',j>', z', t'} just in case the
corresponding coordinates are equal mod 1. This spacetime is not benign with
respect to the scalar wave equation. But solutions are not constrained to be
constant; in fact, the allowed solutions form an infinite-dimensional subspace
of the space of all solutions (Yurtsever 1990). The consistency constraint
imposes a high frequency cutoff on plane wave solutions propagating in
certain null directions. This constraint by my reckoning is simple and clean
enough to count as an MRL law, but it supplements rather than supplants

It is possible in principle to verify by means of local observations that we

do not inhabit a non-causally regular region of some non-benign spacetimes.
And if we indulge in an admittedly dangerous inductive extrapolation on the
above examples, we can conclude that we do not in fact inhabit a non-regular
region. For it follows from the (source free) Maxwell equations—which we
may assume are laws of our world— that the components of the electromagnetic
field obey the scalar wave equation. But the electromagnetic fields in our
portion of the universe do not satisfy the restrictions which (if the induction
is to be believed) are characteristic of non-benign spacetimes. Further
experience with other examples of non-benign spacetimes may serve to
strengthen or to refute this inference.
It is also possible to use local observations to rule out as models for our
world certain benign chronology-violating spacetimes, but only on the
assumption that we have looked at large enough neighborhoods to reveal the
consistency constraints indicative of these spacetimes. It is not easy to see how
we would come by a justification for this enabling assumption.

6.9 Self-interacting test fields

I begin with a reminder of two lessons from previous sections. First, the
grandfather paradox is in the first instance a way of pointing to the presence
of consistency constraints on local physics in a chronology-violating spacetime.
And second, while the usual discussion of the grandfather paradox assumes
a self-interacting system, we have found that non-trivial and indeed very
strong constraints can arise even for non-interacting systems. Of course, one
would expect that the constraints for self-interacting systems will be even more
severe. To test this expectation one could carry out an analysis that parallels
that of section 6.8 by considering a test field obeying an equation such as
n<t = k<&3 (where k is a constant with sign chosen so as to lead to positive
energy density), which implies that solutions do not superpose. Results are
not available in the literature. What are available are results for the simpler
and more artificial case of perfectly elastic billiard balls.
One assumes that between collisions the center of mass of such a ball
traces out a timelike geodesic in the background spacetime and that in a
184 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

collision the laws of elastic impact are obeyed. Under these assumptions the
initial trajectory <5 of the (two-dimensional) billiard ball in the spacetime of
Fig. 6.3b leads to an inconsistent time development. The ball trips around
the cylinder and participates in a grazing collision with its younger self,
knocking its former self from the trajectory that brought about the collision
(grandfather paradox for billiard balls). We saw in the previous section that
the cylinder can be benign for the scalar wave equation. But obviously the
corresponding property fails for billiard balls: for any point x on the cylinder
it is not the case that there is a sufficiently small neighborhood N(x) such
that any timelike geodesic segment on JV, representing the initial trajectory
of the ball, can be extended to a globally consistent trajectory. Nor are the
forbidden initial conditions of measure zero in any natural measure.
For a single sufficiently small billiard ball, Godel spacetime is benign,
not just because all timelike geodesies are open but also because they are not
almost closed. And it seems safe to conjecture that Godel spacetime is benign
with respect to any finite system of small billiard balls since it seems
implausible that collisions among a finite collection can be arranged so as to
achieve the sustained acceleration needed to instantiate a closed or almost
closed timelike curve.
Echeverria et al. (1991) have studied the behavior of billiard balls in two
types of wormhole spacetimes that violate chronology. The first type is called
the twin paradox spacetime because the relative motions of the wormhole
mouths give rise to a differential aging effect which in turn leads to GTCs
since the wormhole can be threaded by future-directed timelike curves. This
spacetime contains a partial Cauchy surface E. Chronology is violated only
to the future of E, indeed to the future of// + (E). The other type of wormhole
spacetime is called the eternal time machine spacetime because CTCs that
traverse the wormhole can reach arbitrarily far into the past. There are no
partial Gauchy surfaces in this arena; but because of asymptotic flatness the
notion of past null infinity ,/ ~ is well defined, and initial data can be posed
on J r ~.
For initial conditions (specified on E for the twin paradox spacetime or
on J~ for the eternal time travel spacetime) that would send the billiard ball
into the wormhole, one might expect to find that strong consistency constraints
are needed to avoid the grandfather paradox. But when Echeverria et al.
searched for forbidden initial conditions, they were unable to find any. Thus,
it is plausible, but not proven, that for each initial state of the billiard ball,
specified in the non-chronology-violating region of the spacetime, there exists
a globally consistent extension. Mikheeva and Novikov (1992) have argued
that a similar conclusion holds for an inelastic billiard ball.
Surprisingly, what Echeverria et al. did find was that each of many initial
trajectories had a countably infinite number of consistent extensions. The
consistency problem and the phenomenon of multiple extensions is illustrated
in Fig. 6.4. In Fig. 6.4a we have yet another instance of the grandfather
paradox. The initial trajectory £, if prolonged without interruption, takes the
 Time Travel 185

Fig. 6.4 (a) Self-inconsistent and (b, c) self-consistent billiard ball motions in a
wormhole spacetime (after Echeverria et al. 1991)

billiard ball into mouth 2 of the wormhole. The ball emerges from mouth 1
along tj, collides with its younger self, converting £ into £' and preventing
itself from entering the wormhole. Figures 6.4b and 6.4c show how £ admits
of self-consistent extensions. In Fig. 6.4b the ball suffers a grazing collision
which deflects it along trajectory £". It then reemerges from mouth 1 along
{// and suffers a glancing blow from its younger self and is deflected along \l/''.
The reader can provide the interpretation of Fig. 6.4c. The demonstration of
the existence of initial conditions that admit an infinite multiplicity of
consistent extensions involves the consideration of trajectories that make
multiple wormhole traversals; the details are too complicated to be considered
here.
These fascinating findings on the multiplicity of extensions are relevant
to the question of whether it is possible to operate a time machine; this matter
will be taken up in section 6.10. Of more direct relevance to present concerns
are the findings about consistency constraints for self-interacting systems. The
results of Echeverria et al. (1991) indicate that in the twin paradox spacetime,
for instance, the wow-chronology-violating portion of the spacetime is benign
with respect to all billiard ball trajectories, including those dangerous
trajectories that take the ball into situations where the grandfather paradox
might be expected. But the chronology-violating region of this spacetime is
most certainly not benign with respect to billiard ball motions. Perhaps it is
a feature of non-benign spacetimes that the failure of benignity only shows
up in the chronology-violating region, but one example does not give much
confidence.
The study of more complicated self-interacting systems quickly becomes
intractable at the level of fundamental physics. What one has to deal with is
a coupled set of equations describing the self- and cross-interactions of particles
and fields. Deriving properties of solutions of such a set of equations for
chronology-violating spacetimes is beyond present capabilities. Instead, one
studies the behavior of "devices" whose behavior is analyzed on the
macrolevel. The presumption is that if these devices were analyzed into
fundamental constituents and if the field equations and equations of motion
186 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 6.5 (a) Self-inconsistent and (b) self-consistent motions for Novikov's piston
device (after Novikov 1992)

for these constituents were solved for some relevant range of initial and
boundary conditions, then the solutions would display the behavior character-
istic of the type of device in question. From this point of view, a perfectly
elastic billiard ball might be considered to be one of the simplest devices.
A slightly more complicated device consists of a rigid Y-shaped tube and
a piston which moves frictionlessly in the tube. Imagine that the branches at
the top of the Y are hooked to the mouths of the wormhole in the twin
paradox spacetime. Because of the constraints the tube puts on the motion of
the piston, it is not true that every initial motion of the piston up the bottom
of the Y has a self-consistent extension. Figure 6.5a shows an initial motion
that takes the piston up along the sections /j,1 and [I1!, into the wormhole
mouth 2, through the wormhole, out of mouth 1 at an earlier time, down the
section [12 to the junction ^just in time to block its younger self from entering
the ii\ section.
Novikov (1992) argues that self-consistent solutions are possible if the
device is made slightly more realistic by allowing the older and younger
versions of the piston to experience friction as they rub against one another.
In Fig. 6.5b the piston starts with the same initial velocity as in Fig. 6.5. But
when the piston tries to pass through the junction ,£ it is slowed down by
rubbing against its time-traveling self. This slowing down means that when
the piston traverses the wormhole it will not emerge at an earlier enough time
to block the junction but only to slow its younger self down. Novikov gives
a semiquantitative argument to show that, with self-friction present, for any
initial motion of the piston that gives rise to an inconsistent/grandfather
paradox evolution as in Fig. 6.5a, there is a self-consistent extension as in Fig.
6.5b. For initial trajectories that have the time-traveling piston arriving at
the junction well before its younger self reaches that point, there is also
arguably a self-consistent continuation. Thus, it is plausible that there are no
non-trivial consistency constraints on the initial motion of the piston up the
fi1 section of the tube.
 Time Travel 187

The work of Wheeler and Feynman (1949) and Clarke (1977) suggests
that the absence of consistency constraints or at least the benignity of the
constraints can be demonstrated generally for a class of devices for which the
evolution is a continuous function of the parameters describing the initial
conditions and the self-interaction, the idea being that fixed point theorems
of topology can be invoked to yield the existence of a consistent evolution.
However, Maudlin (1990) showed that if the topology of the parameter space
is complicated enough, a fixed point/self-consistent solution may not exist for
some initial conditions. And one would suppose that in the general case the
problem of deciding whether the relevant state space topology admits a fixed
point theorem is as difficult as solving directly for the consistency constraints.
One might expect that with sufficiently complicated devices there may
be no (or only rare) initial conditions that admit a self-consistent continuation
in a chronology-violating environment that allows the device to follow CTCs.
Consider Novikov's (1992) device consisting of a radio transmitter, which
sends out a directed beam; a receiver, which listens for a signal; and a bomb.
The device is programmed to detonate the bomb if and only if it detects a
signal of a strength that would be experienced by being, say, 30 m from the
device's transmitter. A self-consistent traverse of the wormhole of the twin
paradox spacetime is possible if the device undergoes inelastic collisions; for
then such a collision between the older and younger versions can produce a
change in orientation of the transmitter such that the younger self does not
receive the signal from its older self and, consequently, no explosion takes
place. But one can think of any number of epicycles that do not admit of any
obvious self-consistent solution. For example, as Novikov himself suggests, the
device could be equipped with gyrostabilizers that maintain the direction of
the radio beam.
It is all too easy to get caught up in the fascinating details of such devices
and thereby to lose sight of the implications for what I take to be the important
issues about time travel. As a way of stepping back, let me reiterate the point
that came up in connection with the investigations of Echeverria et al. (1991)
of billiard ball motions in chronology-violating spacetimes. The absence of
consistency constraints or the benignity of these constraints with respect to
the initial conditions of a device, as specified in the non-chronology-violating
portion of the spacetime, does not establish the absence of consistency
constraints or their benignity simpliciter. For example, assuming some
self-friction of the piston of the Novikov cylinder-piston device, there may be
no non-trivial consistency constraints on the initial motion of the piston up
the bottom of the Y. But there most certainly are constraints on the motion
in the chronology-violating portion of the spacetime, and these constraints
are not benign. Consider any spacetime neighborhood that includes the
junction ^ at a time when the piston is passing ,£. Passing from /j t to ju't with
a speed v without rubbing against a piston coming down jj,2 is a physically
possible local state for every v. But for some values of v there is no self-consistent
extension. Thus, contrary to what some commentators have suggested, the
188 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

recent work on the physics of time travel does not dissolve the paradoxes of
time travel. Whatever exactly these paradoxes are, they rest on the existence
of consistency constraints entailed by field equations/laws of motion in the
presence of CTCs. Showing that these constraints are trivial would effectively
dissolve the paradoxes. But all of the recent work affirms the non-triviality of
such constraints, which are more or less severe depending on the case.
Does what we have learned about self-interacting systems give reason for
optimism about my working hypothesis; namely, that insofar as a chronology-
violating world admits a set of MRL laws for test fields, those laws will
subsume the consistency constraints forced out by the presence of CTCs? One
might see a basis for pessimism deriving from the fact that in the wormhole
spacetimes the constraints that obtain in different regions are different (e.g.,
no constraints on the initial conditions for the billiard ball in the non-
chronology-violating region of the twin paradox spacetime but non-trivial
constraints in the chronology-violating region). Since we want laws of nature
to be "universal" in the sense that they hold good for every region of
spacetime, it might seem that the wormhole spacetimes dash the hope that
the consistency constraints will have a lawlike status. But the hope is not to
be extinguished so easily. To be "universal," the constraint must be put in
a general form; namely, for any region R, constraint C(R) obtains iff
, where the blank is filled with conditions formulated in terms of
suitably general predicates. The blank will need to be filled not only with
features of R but also with features of the relation of R to the rest of spacetime.
So if the consistency constraints have law status, then the laws of a
chronology-violating wormhole spacetime cannot all be local. But that was
only to be expected. The real concern is the one that already surfaced in
section 6.7; namely, that as the spacetime gets more and more complicated,
the conditions that go into the blank may have to become so complicated
that the consistency constraints will not qualify as MRL laws. Remember,
however, that this concern is mitigated if the chronology-violating worlds in
question can be deemed to lie in the outer reaches of the space of worlds
nomologically accessible from the actual world. Here I think that intuition
pumping is useless until we have more concrete examples to serve as an
anchor.

6.10 Can we build and operate a time machine?

The question that forms the title of this section is not equivalent to the question
of whether time travel is possible, at least not if'time machine' is understood
in the strong sense of a device that manufactures or produces CTCs. A
sufficiently powerful rocket engine that allowed a person to trace out a CTC
in Godel spacetime might be counted as a time machine in the weak sense.
But clearly there is no time machine in the strong sense operating in this
context, since CTCs exist everywhere and everywhen. 17
 Time Travel 189

In trying to characterize a time machine we face something of a

conundrum. To make sure that the CTCs are due to the operation of the
time machine we could stipulate that there is a time slice E, corresponding
to a time before the machine is turned on, such that there are no CTCs in
J~ (E). Furthermore, by going to a covering spacetime, if necessary, we can
guarantee that E is achronal and, thus, a partial Cauchy surface. But then
by construction only time travel to the future of E is allowed, which
immediately eliminates the kind of time travel envisioned in the typical time
travel story of the science fiction literature. I do not see any easy way out of
this conundrum, and for present purposes I will assume that such a E exists.
Next, one would like a condition which says that only local manipulation
of matter and energy is involved in the operation of the machine. Requiring
that the spacetime be asymptotically flat would be one approach—intuitively,
the gravitational field of the machine falls off at large distances. But on the
one hand this requirement precludes many plausible cosmologies; and on the
other hand it is not evident that a condition on the space structure guarantees
that something funny with the causal structure was not already in progress
before the machine was switched on.
It is even more delicate to pin down what it means to say that switching
on the time machine produces CTCs. Programming the time machine
corresponds to setting initial conditions on the partial Cauchy surface E.
These conditions together with the coupled Einstein-matter equations deter-
mine a unique evolution for the portion of J + (E) contained in D + (E). But
D+ (L) is globally hyperbolic and therefore contains no CTCs. Moreover, the
future boundary H+ (E) of D + (L) is always achronal. So the notion that the
initial conditions on E are responsible for the formation of CTCs cannot be
cashed in terms of causal determinism. Perhaps this notion can be captured
by the requirement that some of the null geodesic generators of H+ (E) are
closed or almost closed, indicating that CTCs are on the verge of forming.
Perhaps it should also be required that any appropriate extension of the
spacetime across H+ (E) contains CTCs. Here an appropriate extension might
variously be taken to be one that is sufficiently smooth, that preserves various
symmetry properties of D + ( E ) , . . .
One need not be too fussy about the sufficient conditions for the operation
of a time machine if the goal is to prove negative results, for then one need
only fix on some precise necessary conditions. An example of such a negative
result was obtained recently by Hawking (1992). In concert with the above
discussion he assumes the existence of a partial Cauchy surface £ such that
H+ (E) separates the portion of spacetime with CTCs from the portion without
CTCs. In Hawking's terminology, H+ (E) is a chronology horizon. If all the
past-directed null geodesic generators of H (E) enter and remain within a
compact set, then H+ (E) is said to be compactly generated.

Theorem (Hawking). Let M,gab, Tah be a cosmological model satisfying

Kinstein's field equations (with or without cosmological constant). Suppose
190 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

that M, gab admits a partial Cauchy surface E and that Tab satisfies the weak
energy condition.18 If further H+ (E) is non-empty and compactly generated,
then (a) E cannot be non-compact, and (b) whatever the topology of E,
matter cannot cross // + (E).

How effective is this formal result as an argument against time machines?

At best, part (b) of the theorem shows that the operator of the time machine
cannot himself sample the fruits of his labor by crossing over the chronology
horizon to the region of spacetime containing CTCs. Part (a) has real bite if
it can be read as saying that in a spatially open universe a time machine
cannot operate without violating the weak energy condition.19 However, this
reading assumes that when CTCs are manufactured by a time machine, the
chronology horizon H+ (E) must be compactly generated. Physically this
requirement says that the generators of // + (E) do not emerge from a
curvature singularity, nor do they "come from infinity." These prohibitions
might seem well motivated by the idea that if the appearance of CTCs is to
be attributed to operation of a time machine, then they must result from the
manipulation of matter in a finite region of space. But it seems to me this
motivation is better served by requiring that (i) H+ (E) is compactly causally
generated in the sense that the topological closure of 7~(// + (E)) n E—which
is the portion of E from which events can influence H+ (E)—is compact, and
(ii) all appropriate extensions across H+ (E) contain CTCs. It is possible in
principle for H+ (E) to be compactly causally generated although not
compactly generated in Hawking's sense; in particular, the former does not
preclude that some generators of 7/ + (E) emerge from curvature singularities.
The would-be time machine operator may well be willing to create singularities
in order to satisfy his client's desire to experience the thrill of time travel.
Thus, it seems to me that an effective chronology protection theorem would
have to substitute the condition of compactly causally generated for Hawking's
condition of compactly generated. 20 But when the substitution is made,
Hawking's proof technique no longer works. How likely is it that some other
technique will yield an effective chronology protection theorem? The answer
depends on one's attitude towards the cosmic censorship hypothesis, for such
a theorem would constitute a proof of an important piece of cosmic censor-
ship. Given how hard it has been to prove censorship theorems (see chapter
3), one should not expect effective chronology protection theorems to sprout
like mushrooms.
A different approach to showing that the laws of physics are unfriendly
to the enterprise of building a time machine would be to try to show that the
operation of the machine involves physical instabilities. More specifically, in
terms of the setting suggested above, one would try to show that a chronology
horizon is necessarily unstable. This approach links back to the problem of
the behavior of test fields on chronology-violating spacetimes; indeed, the
stability property can be explicated in terms of the existence of extensions of
solutions of the test field. To return to the example of a scalar field $ obeying
 Time Travel 191

the wave equation D$ = 0, consider arbitrary initial data on the partial

Cauchy surface Z (the values on Z of <D and its normal derivative to Z) of
finite energy.21 Each such data set determines a unique solution in the region
Z) + (Z). // + (S) can then be said to be completely stable for $ iff every such
solution has a smooth extension across // + (Z). // + (Z) can be said to be
generically stable for <& iff a generic solution has a smooth extension across
// + (Z). 23
It has been argued by Morris et al. (1988) and by Friedman and Morris
(199la,b) that some asymptotically flat wormhole spacetimes with CTCs
have stable chronology horizons. However, these examples violate the weak
energy condition of classical GTR. This follows from the results of Tipler
(1976, 1977a) showing that the weak energy condition prevents the kind of
topology change which occurs with the development of wormholes. Such
violations are tolerated in quantum field theory. But the maintenance of
traversable wormholes leading to CTCs also requires the violation of averaged
or integrated versions of the energy conditions. What remains unresolved is
the extent to which quantum field theory tolerates violations of the averaged
energy conditions. Wald and Yurtsever (1991) demonstrated the satisfaction
of the averaged null energy condition24 in two-dimensional curved spacetimes;
but they also showed that this condition can fail in four-dimensional
spacetimes. The proponents of chronology protection can hope that quantum
fields can never violate the averaged energy conditions in such a way that
permits wormhole-based CTCs.
An example of a spacetime that contains a partial Cauchy surface Z with
CTCs to the future of Z and where the chronology horizon H+ (Z) is
generically unstable for $ is Misner's two-dimensional version of Taub-NUT
spacetime (see Fig. 2.3 and Hawking and Ellis 1973, pp. 170-178). Here the
chronology horizon is not only compactly generated but is itself compact;
indeed, it is generated by a smoothly closed null geodesic. Each time the
tangent vector of this geodesic is transported parallel to itself around the loop
it is expanded by a factor of eh, h > 0, indicating a blueshift. Now consider
a generic high frequency wave packet solution to D$ = 0 "propagating to
the right." As it nears // + (Z) it experiences a blueshift each time it makes a
circuit around the universe, and as an infinite number of circuits are needed
to reach H+(L), the blueshift diverges. This is already indicative of an
instability, but to demonstrate that the divergent blueshift involves the kind
of instability that prevents an extension across H+ (Z) it has to be checked
that the local energy density of the wave packet diverges as H+ (Z) is
approached.25 That is in fact the case in this example. One could then reason
that when <I> is not treated merely as a test field but as a source for the
gravitational field, spacetime singularities will develop on //+ (Z) thereby
stopping the spacetime evolution and preventing the formation of CTCs that
would otherwise have formed beyond the chronology horizon (see Morris
et al. 1988).
However, the classical instability of chronology horizons is certainly not
192 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

a generally effective mechanism for ensuring chronology protection. Hawking

(1992) showed that among compactly generated chronology horizons, a
non-negligible subset of the horizons are classically stable. Even in cases where
classical instability obtains, one can also wonder how this instability under-
mines the feasibility of operating a time machine. The worst case of instability
would be complete instability with respect to $, i.e., no solution of D<1> = 0
other than <E> = 0 is extendible across // + (H). Then insofar as the time
machine involves non-zero values of O, it cannot succeed. The next worst
case would involve instability that is not complete but is so generic that only
a set of solutions of "measure zero" admit extensions across H+ (L). Here one
could argue that if the time machine operator chose the parameter setting at
random (with respect to the preferred measure on initial conditions), she
would have a zero probability of hitting on a setting that would lead to
successful operation of the machine. This would not be a proof of the
impossibility of time travel via a time machine but only a demonstration that
initiating the journey requires luck. Perhaps some stronger conclusion can be
derived, but I do not see how. Measure zero arguments are commonly
assumed to have a good deal of force, but it is hardly ever explained why.
In sum, it seems fair to say that at present no mechanisms from classical
GTR have been shown to be effective enforcers of chronology protection. If
classical GTR does not offer chronology protection, then perhaps it can be
found in quantum effects. In particular, the quantum instability of chronology
horizons is currently under intensive investigation (see Boulware 1992;
Hawking 1992; Kim and Thorne 1991; Klinkhammer 1992). It seems that
in the wormhole spacetimes with CTCs, the expectation value of the
(renormalized) stress-energy tensor of a quantum field diverges as the
chronology horizon is approached, with the divergence being stronger for
cases of a compactly generated horizon than for a non-compactly generated
horizon. In the semiclassical approach to quantum gravity, the expectation
value of the stress-energy tensor is fed back into EFE to determine the effects
on the spacetime geometry. Whether or not the divergence of the stress—energy
tensor on the chronology horizon produces an alteration of the spacetime
sufficient to prevent the formation of CTCs is still controversial. Apparently
in Gott spacetime the stress-energy tensor for a scalar field remains regular
near the chronology horizon and therefore cannot prevent the formation of
CTCs (see Boulware 1992).
Now suppose for the sake of discussion that neither classical GTR nor
quantum mechanics prevents the construction of a time machine. The main
puzzle about its operation is not the grandfather paradox but something quite
different. The implicit assumption in the science fiction literature is that when
the time machine is switched on, some definite scenario will unfold, as
determined by the settings on the machine. But the still imperfectly understood
physics of time travel hints at something quite at variance with these
expectations. In the first place, there may be different extensions of the
spacetime across the chronology horizon H+(T,). In the second place, even
 Time Travel 193

when the spacetime extension is chosen and treated as a fixed background

for test fields, billiard balls, and other devices, the equations which govern
these systems may permit a multiplicity, perhaps even an infinite multiplicity,
of extensions across // + (Z) and into the time travel region. The point is not
simply that one does not know the upshot of turning on the time machine
but rather that the upshot is radically underdetermined on the ontological
level. And thus the new puzzle: How does the universe choose among these
ontologically distinct possibilities? Of course, it is unfair to demand a
mechanism for making the choice if "mechanism" implies determinism, for
that is what is expressly ruled out in this situation. But it is equally
unsatisfactory to respond with nothing more than the formula that "It is just
a matter of chance which option will unfold." If "just a matter of chance"
is to be more than an incantation or a recapitulation of the puzzle, then
"chance" must mean something like objective propensity. But the physics of
classical GTR provides no basis for saying that there are objective probabilities
of, say, .7 and .3 respectively for the scenarios of Fig. 6.4b and Fig. 6.4c. Here
quantum mechanics may come to the rescue of time travel by showing that
for any initial quantum state describing the motion of the billiard ball before
it enters the region of CTCs, there is a well-defined probability for each of
the subsequent classically consistent extensions.26
However, when CTCs are present it is to be expected that the time
evolution will not be unitary (see Friedman et al. 1992; Goldwirth et al. 1993;
Politzer 1992, 1994). The loss of unitarity is not necessarily fatal to a viable
quantum description; perhaps, for example, the path integral or sum-over-
histories approach will provide a means for consistently assigning probabilities
to measurement outcomes (see Friedman et al. 1992). Such an approach,
even if consistent, certainly exhibits some disturbing features, e.g., the
probability for the outcome of a measurement made in the pre-time travel
region can depend upon whether CTCs form in the future (see Friedman et
al. 1992; Politzer 1994). Clearly, how CTCs mesh or fail to mesh with
quantum mechanics will be an exciting area of investigation for some time to
come.

6.11 Conclusion

If nothing else, I hope this chapter has made it clear that progress on
understanding the problems and prospects of time travel is not going to come
from the sorts of contemplations of the grandfather paradox typical of past
philosophical writings. Using modal logic to symbolize the paradox, armchair
reflections on the concept of causation and the like are not going to yield new
insights. The grandfather paradox is simply a way of pointing to the fact that
if the familiar laws of classical relativistic physics are supposed to hold true
in a chronology-violating spacetime, then consistency constraints emerge.
The first step to understanding these constraints is to define their shape and
194 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

content. This involves solving problems in physics, not armchair philosophical

reflections.
But philosophy can help in understanding the status of the consistency
constraints. Indeed, the existence of consistency constraints is a strong
hint—but nevertheless a hint that most of the literature on time travel
has managed to ignore—that it is naive to expect the laws of a time travel
world which is nomologically accessible from our would will be identical with
the laws of our world. I explored this matter under the assumption that laws
of nature are to be constructed following the analysis of Mill—Ramsey—Lewis.
In some time travel worlds it is plausible that the MRL laws include
the consistency constraints; in these cases the grandfather paradox has a
satisfying resolution. In other cases the status of the consistency constraints
remains obscure; in these cases the grandfather paradox leaves a residual itch.
Those who wish to scratch the itch further may want to explore other analyses
of laws. Indeed, time travel would seem to provide a good testing ground
for competing analyses of laws.
I do not see any prospect for proving that time travel is impossible in
any interesting sense. It may be, however, that it is not physically possible to
operate a time machine that manufactures CTCs. But if so, no proof of
this impossibility has emerged in classical GTR. The prospects for getting
such a proof are intimately tied to the fate of cosmic censorship. If the
operation of the time machine is feasible there emerges a new puzzle: a setting
of the parameters on the time machine may correspond to many different
scenarios in the time travel region. The problem here is not that the operation
of the time machine is unpredictable or calls into play an element of
indeterminacy; rather, the problem lies in providing an objective content to
the notion of chance in this setting. Quantum mechanics is, of course, the
place to look for such content. But standard quantum mechanics is hard to
reconcile with CTCs. And it would be a little surprising and more than a little
disturbing if Godelian time machines, which seemed to be characterizable in
purely classical relativistic terms, turned out to be inherently quantum
mechanical. Is nothing safe from the clutches of the awful quantum?

Appendix: Godel on the ideality of time

As the reader of this chapter will no doubt have gathered, I think that too
much of the philosophical literature on time travel has been devoted to Godel
spacetime. It would be healthier if attention were directed to other solutions
to EFE which allow for time travel and which do not exhibit one or other of
the peculiarities of Godel spacetime (e.g., time travel in the Godel universe
requires a fantastically powerful rocket engine whereas in other solutions time
travel may be accomplished without the help of rocketry). By contrast, there
has been a relative neglect of the philosophical moral Godel (1949a) himself
 Time Travel 195

wanted to draw from his solution to EFE. 27 I will try to explain why the
neglect has been benign. Some explanation is called for, if for no other reason
than because a deeply held conviction of someone of GodePs stature deserves
serious consideration.
Unless otherwise indicated, page references are to Godel (1949b). The
dialectic of his argument goes as follows. He began with the idea that if STR
were true, then time would be ideal. He wrote:

Change becomes possible only through the lapse of time. The existence of
an objective lapse of time, however, means (or, at least, is equivalent to the
fact) that reality consists of an infinity of layers of "now" which come into
existence successively. But, if simultaneity is something relative [as is implied
by STR] . . . reality cannot be split up into such layers in an objectively
determined way. Each observer has his own set of "nows," and none of these
various systems of layers can claim the prerogative of representing the
objective lapse of time. (p. 558)

In a footnote, Godel acknowledges that a possible response to this

argument is that it shows "that time is something relative, which does not
exclude that it is something objective; whereas the idealists maintain that it
is something merely imagined" (p. 558, fn 5). GodePs rejoinder is emphatic.

A lapse of time, however, which is not a lapse in some definite way seems
to me as absurd as a coloured object which has no definite colours. But even if
such a thing were conceivable, it would again be something totally different
from the intuitive idea of the lapse of time, to which the idealistic assertion
refers, (p. 558, fn 5)

This is a pretty piece of ordinary language philosophizing. But like most of

its ilk, it leaves one up in the air: even if one shares the linguistic intuitions,
one can wonder how such intuitions can support weighty philosophical
morals. To thrash through these issues, however, would lead us astray, for it
is GTR and not STR that is true (or so we may suppose).
The relevance of GTR is it implies that the existence of matter causes the
curvature of spacetime and thereby destroys the equivalence of inertial
observers in the Minkowski spacetime presupposed in GodePs argument from
STR. Furthermore, in all the cosmological solutions to EFE known in the
1940s, there is a natural way to single out a distinguished time function. Godel
put it thus:

The existence of matter, however, as well as the particular kind of curvature

of space-time produced by it, largely destroy the equivalence of observers
and distinguish some of them conspicuously from the rest, namely those
which follow in their motion the mean motion of matter. Now in all
cosmological solutions of the gravitational equations (i.e., in all possible
universes) known at present the local times of all these observers fit together
into one world time, so that apparently it becomes possible to consider this
time as the "true" one, which lapses objectively, (p. 559)
196 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

From such considerations, Godel noted, James Jeans had concluded, in

Godel's words, that "there is no reason to abandon the intuitive idea of an
absolute time lapsing objectively" (p. 559).28
At this juncture Godel issues a demurrer. In a footnote he mentions that
the proposed method for picking out a preferred simultaneity can be
challenged. A successful challenge would open the way to parroting within
GTR the above argument from STR. Godel notes that making the notion of
the mean motion of matter into a precise concept may involve "introducing
more or less arbitrary elements (such as, for example, the size of the regions
or the weight function to be used in the computation of the mean motion of
matter)." And he goes on to assert "It is doubtful whether there exists
a precise definition which has so great merits, that there would be sufficient
reason to consider exactly the time thus obtained as the true one (p. 560, fn
9)." One can dispute this claim; for example, for the class of models Jeans
and Godel had in mind, one may be able to prove that there exists a unique
family of time slices with minimal intrinsic curvature. Such a family would
arguably be a good candidate for defining the true time. I will not pursue
this matter since Godel does not seem to put much weight on his demurrer
to Jeans and bases his case-in-chief on other considerations.
So the dialectical situation is now this. According to Godel, STR supports
the thesis of the ideality of time. However, Godel acknowledges that
proponents of an objective lapse of time can, with some justice, claim that
GTR supports their case. To break the stalemate Godel proposes to turn the
tables by showing how an expanded knowledge of the solutions to EFE
establishes his ideality thesis. His strategy is based on his discovery of a new
class of solutions to EFE. In these solutions "the aforementioned procedure
of defining an absolute time is not applicable, because the local time of the
special observers used above cannot be fitted together into one world time"
(p. 560). What Godel meant is that there are solutions to EFE where matter
is everywhere rotating so that the natural way of singling out a time function
by taking the spacelike hypersurfaces orthogonal to the world lines of matter
is not available. The technical point is this. If V" is the normed timelike
tangent field of the congruence of world lines of matter, the rotation or twist
of the congruence is defined by a>ab =• V[ffl Vb], By Froberiius' theorem, the
congruence is hypersurface orthogonal just in case a>ab = 0. The point can be
seen in a non-technical way by analogizing the world lines of matter to the
strands of a rope. If the rope is twisted, the strands will not be orthogonal
(in the Euclidean sense) to any plane slicing through the rope.29 Granting
for sake of argument that in a universe where matter is everywhere rotating
there is no natural way to single out a distinguished time function, how does
this conclusion bear on time in the actual universe where, presumably, matter
on the average has no twist? I will not press the point here since a similar
one will soon arise.
Godel does not rest his case with his twist argument but goes on to claim
that the idealistic viewpoint is strengthened by some of the other surprising
 Time Travel 197

features of his solution. The considerations that flow from these features add
up to something less than an argument. The task is to locate and assess the
missing premise(s) that would produce a valid argument.

(PI) In the Godel universe there is no global time function, nor does
there exist a single global time slice, nor is there a globally consistent
time order.
(P2) Therefore, in the Godel universe there is no objective lapse of time.
(P3) The Godel model satisfies EFE and other conditions, such as
non-negativity of energy density, that one wants for a physically
possible model.
(P4) Furthermore, the Godel model cannot be excluded a priori on the
grounds that time travel leads to the grandfather paradox.
(P5) However, the Godel universe can be excluded a posteriori as a
model for the actual universe since, for example, it gives no
cosmological redshift.
(P6) But our universe is different from the Godel universe only because
of contingent features—in particular, the distribution and motion
of matter.
(P7) ?
(C) Therefore, time in our universe is ideal.

Premises (PI), (P3), (P5), and (P6) are uncontroversial. And given
Godel's analysis of time lapse, (P2) is also unexceptionable. (P4), however,
is controversial. In its defense Godel writes:
This and similar contradictions [i.e., the grandfather paradox], however, in
order to prove the impossibility of the worlds under consideration, presuppose
the actual feasibility of the journey into one's own past. But the velocities
which would be necessary in order to complete the voyage in a reasonable
length of time are far beyond everything that can be expected ever to become
a practical possibility, (p. 561)
On the analysis of the grandfather paradox I have offered, it seems to me
that Godel's way of dismissing the grandfather paradox is too quick. On the
other hand, my analysis does support the contention that cosmological models
with CTCs and the other features listed in (PI) cannot be easily dismissed as
conceptually or physically impossible worlds. Thus, the evaluation of Godel's
argument devolves to the question of what has to go into (P7) in order to
make (C) follow from ( P I ) through (P7).
Here is one try.

(P7.1) The existence of an objective lapse is not a property that time can
possess contingently.

This way of filling in (P7) is supported by the following passage from Godel's
essay:
198 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

It might, however, be asked: Of what use is it if such conditions [i.e., those of

(PI)] prevail in certain possible worlds? Does that mean anything for the
question interesting us whether in our world there exists an objective lapse
of time? I think it does. . . . The mere compatibility with the laws of nature
of worlds in which there is no distinguished absolute time, and, therefore,
no objective time lapse can exist, throws some light on the meaning of time
also in those worlds in which absolute time can be defined. For, if someone
asserts that this absolute time is lapsing, he accepts the consequence that,
whether or not an objective lapse of time exists. . . depends on the particular
way in which matter and its motion are arranged in the world. This is not
a straightforward contradiction; nevertheless, a philosophical view leading
to such consequences can hardly be considered as satisfactory, (pp. 561—562)

The most direct and the crudest interpretation of the pattern of argument
would be: L -» N(L) (if time has the property of lapsing, then necessarily so),
~i N(L) (lapsing is not a necessary property of time), therefore ~\L (time does
not lapse). ~~\JV(L) is equivalent to P(~\L), where P ( - ) means that • is
possible. And P(~~\L] is established by showing there is a physically possible
world—the Godel universe—where ~~\L is true.
Godel's essentialist intuitions here are not easy to fathom. There seems
to be no lurking contradiction or anything philosophically unsatisfactory in
saying in the same breath: "Space in the actual world is open, but if the mass
density were a little greater, space would be closed," or "Time in the actual
universe goes on forever into the future, but if the mass density were greater
the universe would eventually recollapse and time would come to an end."
Why then is there a lurking contradiction or something philosophically
unsatisfactory in saying: "Time in our universe lapses, but if the distribution
and motion of matter were different, there would be no consistent time order
and so time would not lapse"? Godel seemed to have thought that one should
see the unsatisfactory character of this utterance just by reflecting on the
concept of time. This game of using an inner sense to perceive conceptual
truths is a dangerous one, for others claim to perceive the non-existence of
CTCs as essential to the concept of time and, therefore, that contrary to (P4)
the Godel model can be ruled out on a priori grounds. Godel gives us no
guidelines for judging superiority of conceptual insight.
But there is an even more puzzling feature of Godel's endorsement of
~\L. He concedes at this juncture of the dialectic that the actual universe has
all the geometrical properties necessary for an objective time lapse, namely,
the existence of an appropriately distinguished global time function. So in
affirming ~\L he must be claiming that time in the actual universe lacks some
non-geometrical feature necessary for time lapse. What is this missing
ingredient? Recall that Godel says "The existence of an objective time . . .
means (or, at least, is equivalent to the fact) that reality consists of an infinity
of layers o f ' n o w ' which come into existence successively." By hypothesis the
actual universe consists of an infinity of layers of'now'. So Godel must have
believed that these 'nows' fail to "come into existence successively." There
 Time Travel 199

are two mysteries here. First, how is the non-existence of a global time function
in some other possible world relevant to whether the 'nows' of the distinguished
time function of this world come into existence successively? Second, if what
GodePs argument for the ideality of time amounts to is that time lacks a
shifting 'newness', then there is no need to invoke GTR and the Godel
universe. For even if the actual universe and all physically possible universes
were fully Newtonian, it would be difficult to make any non-psychologistic
sense of shifting newness.
Another attempt to fill in (P7) comes from Yourgrau (1991).

(P7.2) "Since the actual world is lawlike compossible with the Godel
universe, it follows that our direct experience of time is compatible
with its ideality. . . . But if even direct experience is inadequate
to establish the existence of. . . genuine, successive time that lapses
or passes—then nothing further will suffice." (Yourgrau 1991,
p. 53)

In support of this reading Yourgrau cites the following passage from GodePs
essay:

If the experience of the lapse of time can exist without an objective lapse of
time, no reason can be given why an objective lapse of time should be
assumed at all. (p. 561)

Let it be granted for sake of argument that some observers in the Godel
universe are under an illusion—they experience a time lapse and in con-
sequence think that time objectively lapses even though in fact there is no
objective lapse of time. How is that fact about the Godel universe supposed
to impinge on us? Granted, it should make us cautious in drawing consequences
about the lapsing of time from our own experiences. But apart from our
experiences of time lapse we have all sorts of other experiences that lend strong
support to the inference that we do not inhabit a Godel type universe hut rather a universe
that fulfills all of the geometrical conditions necessary for an objective lapse of time.
To block this move, consider

(P7.3) There are cosmological models that (i) lack the features necessary
for an objective time lapse, but (ii) reproduce the redshift, etc.,
so that they are effectively observationally indistinguishable from
models that fit current astronomical data and have the spatio-
temporal structure needed to ground an objective lapse of time.

This tack is suggested by another passage from GodePs essay:

Our world, it is true, can hardly be represented by the particular kind

of rotating solutions referred to above (because these solutions are static and,
200 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

therefore, yield no red-shift for distant objects); there exist however also
expanding rotating solutions. In such universes an absolute time might fail
to exist, and it is not impossible that our world is a universe of this kind.
(p. 562)
And indeed Godel did go on to generalize his solutions to EFE in such a way
as to allow for a cosmological redshift (Godel 1952). However, I very much
doubt there are cosmological models which allow for time travel and which
are observationally indistinguishable from non-time travel models.30 But
even if there were, Godel would seem to need an additional premise
asserting something like a verifiability theory of meaning in order to reach
his conclusion (G). I take it that few people will be attracted by such a
premise.
I have been unable to locate any plausible argument which starts from
Godel's considerations and leads to the conclusion that time is ideal. Rather,
what I find is a collection of arguments each of which is intriguing but
ultimately unpersuasive. A bunch of unpersuasive arguments do not add up
to one persuasive one. Reading between the lines of his "Reply to Criticisms,"
one can infer that this was Einstein's view also. Einstein begins by praising
Godel's essay as "an important contribution to the general theory of relativity,
especially to the analysis of the concept of time" (Einstein 1949b, p. 687).
But then he immediately brushes aside the question of the relation of
GTR to idealistic philosophy and goes on to discuss issues of causation. This
seems to me to be the correct response to Godel.

Notes
1. Here is a sample: Boulware (1992); Carroll et al. (1992); Charlton and Clarke
(1990); Cutler (1992); Deser (1993); Deser et al. (1992); Deutch (1991); Echeverria
ct al. (1991); Friedman and Morris (199la, 1991b); Friedman et al. (1990); Friedman
ct al. (1992); Frolov (1991); Frolov and Novikov (1990); Gibbons and Hawking
(1992); Goldwirth et al. (1993); Gott (1991); Grant (1993); Hawking (1992); Kim
and Thorne (1991); Klinkhammer (1992); Menotti and Seminara (1993); Mikeeva
and Novikov (1992); Morris et al. (1988); Novikov (1989, 1992); Ori (1991a, 1993);
Ori and Soen (1994); Politzer (1992, 1994); 't Hooft (1992); Thorne (1991); Visser
(1993, 1994); Yurtsever (1990, 1991).
2. A representative sample of this literature follows: Brown (1992); Chapman
(1982); Dummett (1986); Dwyer (1975, 1977, 1978); Ehring (1987); Harrison (1971);
Horwich (1989); Lewis (1986); MacBcath (1982); Mellor (1981); Smith (1986); Thorn
(1975); and Weir (1988).
3. Although it is a truism, it needs repeating that philosophy of science quickly
becomes sterile when it loses contact with what is going on in science.
4. This appellation is suggested by some passages in H. G. Wells' Time Machine
(1968), but I do not claim to have captured what Wells meant by time travel. For a
review of and references to some of the science fiction literature on time travel, see
Gardner (1988).
5. One way to make (CO) precise is to require that there exists on M a continuous,
non-vanishing, tirnelike vector field.
 Time Travel 201

6. Good treatments of Godel's model are to be found in Malament (1985), Pfarr

(1981), and Stein (1970). Godel's original (1949a) cosmological model is a dust-filled
universe, i.e., T°* = nUaUb where n is the density of the dust and U" is the normed
four-velocity of the dust. This model is a solution to Einstein's field equations only for
a non-vanishing cosmological constant A = -co2, where co is the magnitude of the
rotation of matter. Alternatively, the model can be taken to be a solution to
EFE with A = 0 and Tab = fiUaUb + ((o2/8n)gab; however, this energy-momentum
tensor does not seem to have any plausible physical interpretation. Godel's original
model gives no cosmological redshift. Later Godel (1952) generalized the model to
allow for a redshift.
7. Recall from chapter 3 that a time slice is a spacelike hypersurface without edge.
8. Recall from chapter 3 that a partial Cauchy surface is a time slice that is
achronal, i.e., is not intersected more than once by any timelike curve.
9. Some philosophers apparently think time travel is logically or conceptually
impossible; see Hospers (1967, p. 177) and Swinburne (1968, p. 169).
10. Sometimes the local-to-global property may fail in causally nice spacetimes
because singularities develop in solutions to some field equation. But one may regard
such a failure as indicating that the field is not a fundamental one; see chapter 3.
11. The need for such a distinction has been previously noted by Bryson Brown
(1992).
12. There would be an easy victory here if it were the case that in a world which
is nomologically accessible from the actual world and which has a spacetime
structure M, gab containing CTC, it is a law that the spacetime is M, gab. But on no
account of laws with which I am familiar will it be the case that in, say, a Godelian
universe it is a law that the spacetime is Godelian. However, it could be complained
that in effect we are treating the spacetime structure as lawlike in taking it to be a
fixed background on which to solve for consistency constraints; but this complaint
seems to me to be another version of that in the first therapy (Tl).
13. For a review of the horizon problem and attempted solutions, see chapter 5.
14. The slogan here is that there is no difference in laws without some difference
in non-nomic facts. Some authors take this supervenience thesis to be a necessary truth
(i.e., to hold for all possible worlds). David Lewis (1986, pp. ix-xii) takes it to be a
contingent truth which holds only for possible worlds near the actual world.
15. However, the matter is complictaed by the fact that Huygens' principle (see
chapter 5) does not hold in curved spacetimes.
16. Godel took the possibility of time travel to support the conclusion that time
is "ideal." GodePs argument is discussed in an appendix to this chapter.
17. For more on this distinction, see Earman (1994).
18. The weak energy condition (Tab V Vb > 0 for any timelike V] entails the
null energy condition (TallKaKb > 0 for any null K"). Taken together, the weak energy
condition and EFE entail the null convergence condition (RabK"Kb > 0 for any null
K"), which does the real work in Hawking's theorem.
19. Ori (1993) and Ori and Soen (1994) argue that a time machine solution can
satisfy the weak energy condition right up to the time when the chronology violation
starts.
20. For more details, see Earman (1994).
21. Sec Yurtscver (1991) for a definition of this notion and for a more precise
specification of the stability property.
22. See Yurtsevcr (1991) for a formulation of the relevant smoothness conditions.
202 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

23. The space of initial conditions has a natural topology so that a generic set of
solutions can be taken to be one that corresponds to an open set of initial conditions.
24. This condition requires that Jr TabK"Kb rf/l > 0 where K" = dx"/dl is the
tangent to the null geodesic y and /I is an affine parameter of y.
25. The energy density depends on the behavior of the two-dimensional cross-
sectional area of a pencil of geodesies.
26. Results of this character have been announced by Klinkhamriier and Thorne
(1990) in a never published preprint; see Echeverria et al. (1991).
27. Valuable information about the development of Godel's ideas on time are to
be found in Malament (1995) and Stein (1990, 1995).
28. The reference here is to Jeans (1936). Jeans claimed that in GTR "time
regained a real objective existence, although only on an astronomical scale, and with
reference to astronomical phenomena" (p. 22).
29. I owe this illustration to David Malament (1995). By examining the early
versions of Godel's manuscripts for (1949b), Malament was led to hypothesize that
Godel started his investigation by searching for solutions to EFE with rotating matter
and subsequently discovered that his particular rotating solution contained CTCs.
30. At least if the CTCs cannot be unwound by passing to a covering spacetime,
as is the case with the Godel universe. For more on observational indistinguishability,
see chapter 7.
31. My assessment of Godel's argument for idealism was developed in corre-
spondence with Steven Savitt. Our ideas are entangled but somewhat different. The
reader is urged to consult Savitt (1994).
 7
Eternal Recurrence, Cyclic Time,
and All That

7.1 Introduction

The idea of a cyclic or repeating time finds an astonishingly broad acceptance

in the history of thought, being found among such diverse and widely
separated peoples as the Stoics of ancient Greece, the Hindus of India, the
Taoists of China, and the Mayans of Central America.1 Christianity was
generally hostile to the idea since it clashed with the doctrine of the uniqueness
of Christ.2 The scientific revolution produced a decidedly ambivalent stance.
On the one hand, Newtonian mechanics seemed to hold out the possibility
of a "clockwork" universe. But on the other hand Newton himself worried
that because of the instability of the solar system God would from time to
time have to wind up the clock. Nineteenth and early twentieth century
physics produced another tension. Poincare's recurrence theorem showed that
a closed mechanical system will almost surely return to approximately the
same state; but the second law of thermodynamics seemed to indicate that
the universe is destined to wind down to a heat death.
Doing full justice to this complex subject would require a separate
book-length treatment. My aim is not so high; indeed, my remarks here will
be narrowly confined to the implications of GTR and, more particularly, to
the obstacles that spacetime singularities and the censorship of naked
singularities pose to eternal recurrence. It is necessary at the outset to
distinguish between two related but different ideas which are sometimes
confused: first, the idea that numerically distinct but otherwise similar states
occur over and over again in an open time; and second, the idea that the
universe progresses through a series of changes only to return to the
numerically identical state. Although terminology differs in these matters, I
will use eternal recurrence to refer to the former, and circular, cyclic, or closed
time to refer to the latter.
203
204 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

7.2 Tolman on eternal recurrence

The classical form of the second law of thermodynamics implies an increase

in entropy with increasing time and thus appears to be in conflict with the
idea that the universe can undergo regular periodic changes. However,
Richard C. Tolman's investigation of the extension of thermodynamics to
general relativity convinced him that the relativistic version of the second law
permitted thermodynamic change to take place "at a finite rate entirely
reversibly without any increase in entropy at all" (Tolman 1931b, p. 1759;
see also 193 la). Tolman was thus led to consider the possibility of the periodic
behavior of a non-stationary universe. He found, however, that for the simple
case of a universe filled with a uniform distribution of matter, his requirements
for thermodynamic reversibility clashed with the conditions for periodic
behavior in the sense of continual expansions and contractions between fixed
finite limits. From the modern perspective it is easy to see the problem even
if thermodynamical considerations are set aside. As mentioned in chapter 5,
the symmetry properties of a homogeneous and isotropic universe, which
Tolman was assuming, force the stress—energy tensor to have the form of a
perfect fluid: Tab = (fi + p)U"Ub + pgab, where fi is the matter density and
p is the pressure. If ^ > 0, p > 0, and the cosmological constant A is set to 0
(as Tolman assumed), then EFE entail that for the FRW universe with
k = + 1 (closed space sections with constant positive curvature), whose line
element is

the scale factor a(t] ("radius of the universe") must go to 0 at some time in
the past or future. The same conclusion holds if /: + 3p > 0 and A is not too
large.
Having excluded truly periodic behavior, Tolman turned to what he
termed "quasi-periodic behavior" where the volume for a spatially finite
universe increases from 0 to an upper bound and subsequently contracts to
0, and so on ad infinitum. The solutions he found "could not be regarded
analytically as strictly periodic" (Tolman 1931b, p. 1765). Again the reason
is easy to see from the standard treatment in terms of the FRW models. In
these models the early stages of the universe are assumed to be radiation
dominated. Then as a consequence of EFE (for A = 0), the scale factor a(t)
behaves approximately as i1'2. So as the big bang (t = 0) is approached,
a(f) —> 0 + , a(f) -> +00, and d(t] -» — co, which are hardly the analytical
conditions for a minimum for a(t). A similar remark applies equally to the
big crunch.
Nevertheless, Tolman continued to speak of quasi-periodic behavior.

It is evident physically that contraction to zero volume could only be followed

by another expansion, and in addition, as noted by Einstein in a similar
 Eternal Recurrence, Cyclic Time, and All That 205

connection, the idealization on which our considerations have been based

can be regarded as failing in the neighborhood of zero volume. Hence from
a physical point of view it seems reasonable to consider that solutions of the
kind which we are considering could correspond to a series of successive
expansions and contractions. (Tolman 1931b, p. 1765)

Similar sentiments are expressed in a joint article with Morgan Ward

published in the following year (Tolman and Ward 1932, pp. 841-842). In
both instances Tolman was wavering between two ideas. The first is that
when the idealizations of perfect homogeneity and isotropy are removed, the
singularities can be avoided and strictly periodic behavior can be achieved.
For a time Einstein also harbored the hope that spacetime singularities would
disappear when the idealizations were removed (see chapter 1). This hope
was expressed, for example, in Einstein (1931), the paper to which Tolman
referred in the above quotation. I will have more to say on this matter in
section 7.4 below. The second idea is that even if the idealizations are accepted
and the universe contracts to zero volume "it is evident physically that
contraction to zero volume could only be followed by another expansion."
But on reflection there is nothing physically evident in the notion that the
contraction to zero volume would result in the sudden reversal of the sign of
a(t), followed by renewed expansion. Indeed, as I will now argue, the notion
of the universe recycling itself though a succession of big bangs and big
crunches is physically meaningless for the models under consideration.

7.3 Extending through the big bang and the big crunch

The most straightforward, if not the only sensible way, to cash in the notion
of the universe recycling itself is to exploit the concept of the extension of a
spacetime (as explained in chapter 2). Here we seem to quickly run into a
dead end rather than a recycling. It was noted that for the radiation-
dominated regime, which holds near t = 0, a(t) ~ / 1/2 . As a result the FRW
metric is not extendible as a C2 (or even C2~) metric. This follows from the
fact that the Kretschmann curvature scalar K(t) —'•Rabcd(t}Rabcd((] blows up
rapidly as t -> 0 + .
But why should the extended metric be C2? If continuity/ differentiability
(c/d) conditions on the metric are relaxed, is it possible that the FRW metrics
can be extended through the big bang? The answer to the second question
is trivially yes if no c/d conditions are put on the extended metric. In answer
to the first question, there may be no sufficient reason to require that the
extended metric be C2; but there should be enough c/d to assure that EFE
make sense at least distributionally (see chapter 2). If not, then although a
mathematically well-defined extension through the big bang or the big crunch
may exist, it can justifiably be deemed physically meaningless. In chapter 2
it was argued that the requirement in question is implemented by the
206 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

condition that the metric be regular in the sense of Geroch and Traschen
(1987). This condition is enough to prove some negative results.
The metric (7.1) lives on the manifold M = S3 x (0, b), (0, b) <= R,
0 < b < + co. Try to imagine that M is imbedded by the natural inclusion
map into M = S3 x (a, b), — oo < a < 0. (This seems to have been Tolman's
implicit assumption.) This means that the coordinates r, 6, (j), t are Cx for
a < I < b. In these coordinates the contravariant components g", gee, g^ of
the metric are not locally bounded at t = 0 since they behave as l/t.3 This
does not settle the matter since we have to deal with the possibility that
M = S3 x (0, b) can be imbedded as a proper subset in a larger M in such
a fashion that the r, 6, <j), t coordinates break down at the boundary t = 0.
This would open up the possibility that in some new coordinate system
f, 6, (f>, t which belongs to the C100 atlas of M but which is related to the
r, 9, (f>, t coordinates by a transformation which is singular at t = 0, the
components of the FRW metric remain locally bounded at t = 0. It is hard
to envision how this sort of behavior could occur. But the example of the
Schwarzschild metric discussed in chapter 2 should give one pause; for it took
decades before general relativists realized that an imbedding exists in which
the Droste coordinates break down at the Schwarzschild radius in such a way
that a smooth continuation of the Schwarzschild metric is possible. I
conjecture that the FRW case is not analogous to the Schwarzschild case in
this respect, but since I have no proof, I turn to a different and weaker
demonstration of the inextendibility of the FRW metric.
Not only does one want EFE to make sense distributionally, one also
wants the local conservation law Va Tab = 0 to make sense in the same way.
To obtain the conservation law as a consequence of EFE, the Bianchi identity
ViaRbc]de = 0 is needed. And for this identity to be meaningful in the sense of
distributions it is necessary to require not only that the metric be regular but
also that Rabcd be locally square integrable. 4 It follows that the absolute value
of the Kretschmann curvature scalar K must be locally integrable. But if
integrability is judged with respect to the FRW metric this is not the case if
one tries to extend to t = 0. For example, in the k = 0 case, \K(t)\ ~ l/aa(t).
In the radiation-dominated era shortly after the big bang, a(t) ~ t1/2 so that
|A"(<)| ~ l/t4. Using the volume measure from the FRW metric, the volume
integral of |^T(i)| is proportional to t~5/z, which diverges as t —> 0+ . Of course,
it is possible to define other volume measures with respect to which Rabcd is
locally square integrable at the big bang; but unless some physical significance
can be assigned to such measures, the physical significance of the extension
remains moot. 5
To the extent that this argument is successful, a similar argument can be
used to show that it is not physically meaningful to extend through other
scalar polynomial curvature singularities with strong blowup behavior. But
as we know from chapter 2, there are many other types of spacetime
singularities, and it is not at all evident whether and how the non-existence
of physically meaningful extensions can be proved in the general case.
 Eternal Recurrence, Cyclic Time, and All That 207

If the assumption of radiation dominance in the early universe is dropped

in favor of matter dominance, then another route is opened to the conclusion
that there is no physically meaningful extension through the big bang. For
then the big bang singularity will be of the strong curvature type discussed
in section 3.5; namely, for every causal geodesic approaching the big bang,
all volume forms tend to 0. So in any extension that is C° every physical
object would be crushed to zero volume. Arguably such an extension is not
physically meaningful since the identity of physical objects is lost (see Ori
1991b).

7.4 Finding God in the big bang

It has been charged that attempts to find a prior cause of the big bang,
whether physical or metaphysical, are incoherent. The gist of the objection
is that the search for a prior cause assumes the existence of instants of time
prior to the big bang, an assumption incompatible with the cosmological
models that are the basis for our belief in the big bang. Thus, Adolf Griinbaum
has written:

To suggest or assume tacitly that instants existed after all before the big bang
is simply incompatible with the physical correctness of the putative big bang
model at issue, and thus implicitly denies its soundness. . . . It is now clear
that the physical correctness of this model is also implicitly denied by
someone who addresses any of the following questions to it: "What happened
before the big bang?", "What prior events caused matter to come into
existence at t = 0?", "What prior events caused the big bang to occur at
t = 0?" . . . it is altogether wrongheaded . . . to complain that—even when
taken to be physically adequate—the putative big bang model fails to answer
questions based on assumptions which it denies as false. (Griinbaum 1991,
pp. 238-239)

I must demur slightly from these opinions. By itself, a model of the big
bang—say, a standard FRW model—is neither compatible nor incompatible
with the notion that there are instants of time before the initial singularity.
The fate of that notion depends on our choice of extendibility conditions. I
have argued above that under plausible constraints on what is to be counted
as a physically meaningful extension, there are no physically meaningful
extensions through the big bang of the standard models. Perhaps the reader
will find my argument convincing, perhaps not. But my first point is that
some such argument is needed. My second point is that even if my argument,
or some other, succeeds, it remains open that there is some mathematically
meaningfully extension—involving lower continuity/differentiability condi-
tions than those required for a physically meaningful extension—and that
God or some other metaphysical cause operates in this mathematical time. I
will return to this point below.
208 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

William Lane Craig (1991, 1994) has contended that even if it is conceded
there is no meaningful sense in which there are moments of time before the
big bang, the theist who wants to see God at work in the big bang still has
options available (see also Craig and Smith 1993). Thus, it is claimed that
the Creator's act of causing the universe to begin to exist could be conceived
to be simultaneous with the universe's beginning to exist. Or He could be
conceived to "exist timelessly and to cause tenselessly the origin of the universe
at the Big Bang singularity" (Craig 1994, p. 329). But if the argument of
section 7.3 is effective in showing that there is no physically meaningful
extension to times before the big bang, it is equally effective in showing that
there are no meaningful extensions to t = 0.
Why is it that theists want to find God in the big bang? Of course, if one
is determined to find God, He can be found everywhere. But what in
particular is it about the big bang, as opposed, say, to a flower, that makes
His presence evident? Here is an excerpt of a recent letter to The Wall Street
Journal which encapsulates one answer to our question that I have found to
be not uncommon in popular thought:

Reductionist science has made vast strides in the past 400 years, but it has
now hit the brick wall of the Big Bang. Most physicists admit their 'standard
model' cannot explain it. One example: If all the mass of the universe were
packed into that one dimensionless point just before the Big Bang, gravita-
tional attraction would be intense beyond comprehension. How then to
explain a momentary suspension of that attractive force, suddenly reversed
into an incredibly huge explosive force, followed at once by the re-emergence
of gravity? An act of God perhaps?6

Rather than locating an- instant of time at which God does his work, what
the author of this letter actually provides is a sketch of a reductio of the
assumption that there is a spacetime event corresponding to a state where all
the mass of the universe is "packed into that one dimensionless point." The
technical argument of the preceding section confirms the reductio.
A seemingly more sophisticated but not essentially different response is
that something cannot begin to exist without a cause; so if there is no physical
cause of the beginning to exist, there must be a metaphysical one. Here I am
in complete agreement with Professor Griinbaum in that the standard big
bang models are not compatible in any obvious way with the idea that the
universe has a physically uncaused beginning. Indeed, these models imply
that for every time t there is a prior time t' and that the state at l' is a cause
(in the sense of causal determinism) of the state at t? Craig (1994) has
responded that a beginning for time does not require that time have a first
instant but only that time be finite in the past.8 So for Craig the FRW models
of the big bang exhibit a temporal structure in which time began to exist
even though there is no first instant of time. Consequently, for Craig the
principle Whatever begins to exist has a cause applies to these models. However,
on Craig's reading this principle is not an obvious "metaphysical truth";
 Eternal Recurrence, Cyclic Time, and All That 209

in particular, it is not a consequence of the widely held principle Every event

has a cause, which is satisfied in the FRW big bang models without any help
from the theists.
As explained in chapter 2, there are ways of doctoring standard relativistic
cosmological models to represent singularities as boundary points attached to
the spacetime manifold. In this way one could hope to reclaim a "first instant"
for time, opening the way for an application of the principle Every event has
a cause and the invocation of a deistic cause in the absence of a physical
cause. But in the most widely cited method of doctoring—the ^-boundary
approach—the big bang is represented as a single point that is not Hausdorff-
separated from interior points.9 This counterintuitive consequence serves to
emphasize the fact that the boundary points are ideal elements, a warning
that remains in effect even if some other means of adding the boundary points
that escapes the counterintuitive features of the ^-boundary construction were
to be found. Nothing prevents the theist from seeing God as operating at these
ideal points. But since ideal points are not points of spacetime, the sense in
which God can be said to cause or bring about the universe by operating at
these points is very remote from the usual causal notions of science and
everyday life that are concerned with connections between events in space
and time. This is not to say that theistic talk about God creating the universe
is illegitimate. But it is to say that such talk finds no special purchase in the
big bang. Even in models with no big bang and with time extending infinitely
far into the past, ideal points corresponding to t = — oo could be attached to
the spacetime manifold and God'j helping hand could be seen at work there.
I have a parallel reaction to the complaint that the scientific models of
the big bang leave much unexplained—why, for example, the universe began
with the matter content it did and not some other, and why the expansion
from the big bang obeys EFE and not some other empirical equations (see
Quinn 1993). The point behind the complaint is perfectly correct: science
leaves unexplained the most fundamental laws it has been able to uncover,
and it cannot say why one rather than another of the myriad histories
compatible with these laws has been actualized. But again, this observation
applies to all scientific modeling of natural phenomena, not just to the big
bang models. As far as I can see, the big bang models offer no special
advantages to the deists.
Speaking purely personally now, it strikes me as bordering on the
sacrilegious to see God's creative force as able to operate only at a singularity
or ideal point. It is more to His glory if He operates everywhere and
everywhen, and if He operates independently of such contingencies as whether
there is an initial singularity and, if so, what type it is. Those who want to
find God in the big bang should beware of falling into the trap of relegating
God to the diminishing interstices left by modern science. Once the trap is
recogni/ed it is easy to escape using God's supernatural attributes. If there is
no first instant for the physical universe or no prior physical time to the big
bang at which God can operate, no matter. The Creator "may be conceived
210 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

to exist in a metaphysical time" and thus "to exist temporally prior to the
inception of physical time" (Craig 1994, p. 328). The constraints of physics
cannot bind the Creator. But precisely to the extent that a supernatural cause
of the beginning of the universe does not have to answer to the constraints
of nature, scientists qua scientists are entitled to ignore it.

7.5 No recurrence theorems

After this digression into theology, let us return to the question of eternal
recurrence. In section 7.3 it was argued, contra Tolman, that eternal
recurrence cannot be achieved in the context of standard big bang-big crunch
models. But as Tolman noted, these models involve very unrealistic idealiza-
tions. It remains to be seen whether no recurrence results can be demonstrated
under more realistic conditions.
We must first investigate the possibility of cheap counterexamples to no
recurrence. EFE admit a wide class of stationary solutions. Technically,
stationarity of the metric gab means that there is a timelike vector field V such
that £y&6 = 0; or equivalently, V(a Vb) = O.10 (Such a V is called a Killing
field.) This is the invariant way of saying that the spacetime metric does not
change with time since it guarantees that for any spacetime point there
is a neighborhood covered by a coordinate system {x*,x2,x3, i] such that
V = (d/dl)a and 8gab/dl — 0. But by itself stationarity is not enough for a
counterexample to no recurrence. Godel spacetime (see chapter 6) is station-
ary, but there is no reasonable sense in which this spacetime exhibits eternal
recurrence; indeed, Godel spacetime does not contain any global time slices
so that it does not make sense either to affirm or deny that in this setting
things are the same at two different "instants."
To remedy this defect we can require the spacetime to be static, that is
the spacetime is not only stationary but also the Killing field V is hypersurface
orthogonal, i.e., V[aVbVc^ = O.11 But staticity is still not sufficient to produce
a counterexample since it is compatible with closed and cyclical time
structure, which will be studied in section 7.6. So let us also posit that the
spacetime possesses an open time structure by requiring it to admit a time
function t whose level surfaces coincide with the orthogonal hypersurfaces of
the timelike Killing field and whose value increases along every future-
directed timelike curve. Such a spacetime is trivially periodic, periodic for
every period, at least with respect to the spacetime geometry since the
geometry of any orthogonal hypersurface of V is the same as any other.
(Following along the trajectories of V gives the isometry of one hypersurface
onto another.)
Even after all this fiddling, two more things are needed to produce a
counterexample to no recurrence. First, it is necessary not only that the spatial
geometry but also that all the physical fields on spacetime be the same at the
two instants in question. Of course, in any cosmological model M, gab, Tab
 Eternal Recurrence, Cyclic Time, and All That 211

that satisfies EFE, the energy-momentum tensor Tab will inherit the sym-
metries of the metric so that in a stationary spacetime where £ygai, = 0, it
will also be the case that £v Tab = 0. But it is not automatic that the fields
that generate Tab inherit the symmetries of the metric. For example, when
Tab is generated by an electromagnetic field, there are stationary solutions to
the Einstein—Maxwell equations for which the Maxwell tensor Fab charac-
terizing the electromagnetic field is not stationary, i.e., £vFab ^ 0. It has also
been conjectured that there are static solutions to the Einstein-Maxwell
equations where the electromagnetic field does not inherit the symmetry, but
no specific example is known (see Michalski and Wainwright 1975). Thus,
in searching among the static solutions of EFE for a counterexample to no
recurrence one needs either to confine attention to vacuum solutions (Tab = O) 12
or else one must check that the source fields are in fact static.
Second, to assure that the static solution is a genuine counterexample to
no recurrence, the solution should be maximal. This rules out the possibility
that the solution is merely a piece of a larger spacetime that is non-static.
That the worry here is a real one is illustrated by the exterior Schwarzschild
solution, which is a static vacuum solution. But this solution is extendible,
and its maximal analytic extension—Kruskal spacetime—is non-stationary.
Even with all of these caveats in place, counterexamples to no recurrence
can be found among the static solutions to EFE. The static Einstein universe
(see Hawking and Ellis 1973, p. 139) is a case in point. However, this
cosmological model relies on a non-zero value for the cosmological constant
A. And it is also unstable; for a fixed value of A, the smallest change in the
mass density will cause the universe to evolve. From here on I propose to
concentrate on solutions to EFE with vanishing A. Then in searching for
static counterexamples to no recurrence there are various candidates among
spatially open models. Negative mass Schwarzschild spacetime is a spatially
open, static, inextendible, vacuum solution to EFE with A = 0 (see chapter
3). But as the name indicates, it has the unphysical property of a negative
ADM mass. It also violates cosmic censorship since the central singularity is
naked. The truncated Schwarzschild spacetime of Janis, Newman, and
Winicour (1968) is a static solution corresponding to a massless scalar field.
It coincides with the exterior Schwarzschild solution for r > a but the
Schwarzschild sphere r = a becomes a point singularity. This singularity is
also naked. There are also static plane wave solutions, but these lack Cauchy
surfaces (see Penrose 1965) and therefore violate strong cosmic censorship.
Weyl (1917) produced a class of axisymmetric static solutions. But in those
cases where maximal extensions have been obtained—e.g., the Curzon
monopole solution (see Scott and Szekeres 1986a, b)—the resulting spacetime
is either non-static or contains naked singularities. These examples raise the
following question: Are there inextendible static solutions to EFE with A = 0
that do not violate cosmic censorship and are stable? Apparently the answer
is not known. Recently it has been found that for a stress—energy source
corresponding to a Yang-Mills field, there are solutions to EFE with A = 0
212 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

which are static, singularity free, asymptotically Minkowskian, and topologic-

ally IR4 (see Smoller, Wasserman, Yau, and McLeod 1991; Smoller and
Wasserman 1993). In these particle-like solutions the Yang-Mills repulsive
force balances the gravitational attraction to achieve a static situation.
However, these solutions are unstable.
Even if there were a physically satisfactory static counterexample to no
recurrence, the counterexample would be of a very special sort and would
not refute the conjecture that in a universe which harbors some change
(non-stationarity), the change never brings about a return to an exactly
similar state. Since I cannot prove or refute this conjecture, I turn to cases
where there is hope of proving a general no recurrence theorem.
All of the potential counterexamples to no recurrence seem to be spatially
open. This suggests that a tight no recurrence result may be found for spatially
closed universes. More concretely, we can begin by abstracting some of the
features of the spatially closed FRW models, which we know from section 7.2
are incompatible with eternal recurrence and, indeed, with any recurrence. 13
The FRW models contain Cauchy surfaces. That is a feature we want to
preserve since for an interesting form of recurrence the recurrence should
result from deterministic evolution rather than by chance. For such cases
recurrence and eternal recurrence coincide, since Laplacian determinism
entails that if an exactly similar state recurs once, it will recur over and over
again ad infinitum. We know that the existence of a Cauchy surface implies
that the spacetime manifold M is diffeomorphically £ x IR. And to realize
our hope of proving no recurrence we want to restrict attention to the case
where £ is compact, which expresses in a rigorous way the notion that the
universe is spatially closed. Next, the standard energy conditions (see chapter
3) and EFE (with A = 0) allow us to infer the convergence condition that
RabV"Vb > 0 for any non-spacelike V. And finally we may assume that all
timelike and null geodesies feel a tidal force at some moment so that the
timelike and null genericity conditions hold. All of this (plus some differen-
tiability conditions on the metric) allows us to apply the Hawking—Penrose
singularity theorem (see chapter 2) to conclude that the spacetime is timelike
or null geodesically incomplete. The question then becomes whether the
singularities indicated by the geodesic incompleteness prevent recurrence, as
they do in the FRW case.
For purposes of reductio assume that recurrence happens. Then as already
noted, determinism implies that recurrence happens infinitum. Tipler (1980)
showed that this eternal recurrence plus the compactness of the space sections
allows the construction of a certain complete timelike geodesic. Then by the
same conjugate point constructions used in the classic Hawking—Penrose
singularity theorems (see chapter 2), he showed that the existence of such a
geodesic leads to a contradiction. This is an important result but it does not
give an unassailable answer to our question. The constructions of the
Hawking-Penrose theorems require that the metric be C2.14 But as argued
in chapter 2, the c/d conditions lor a physically meaningful spacetime may
 Eternal Recurrence, Cyclic Time, and All That 213

be lower than C2, perhaps low enough so as not to preclude that in some
non-C2 but physically meaningful extension eternal recurrence takes place.
Of course, it may be that the singularities demonstrated in Tipler's result are
strong enough that they cannot be removed by a physically meaningful
extension, as I argued was the case for the FRW big bang models. But until
this is shown, we do not have a really solid argument against eternal
recurrence for generic, spatially closed, deterministic models.15
Under some further conditions, which imply the uniqueness and stability
of the Cauchy development of the initial data for EFE, Tipler (1980) was
also able to establish that for the class of spatially finite models in question,
not only is return to an exactly similar state impossible but so is return to a
state arbitrarily close to the initial one.16 This result shocks intuitions trained
on classical mechanics, where the famous Poincare recurrence theorem shows
that a spatially bounded system with a finite number of degrees of freedom
and finite energy almost always returns arbitrarily closely to the initial state.17
It is clear that the GTR is much less hospitable than is classical physics to
the notion of eternal return and that the reasons have to do with the formation
of spacetime singularities. But because of uncertainties about how to charac-
terize essential singularities and also because of the difficulties in proving the
existence of such singularities in a general setting, it is unlikely that we will
have in hand in the near future a precise knowledge of just how inhospitable
GTR is to eternal recurrence.

7.6 Cyclic time

As noted in section 7.1, eternal recurrence should not be confused with a

cyclic time structure. 18 As used here the former refers to the recurrence of
similar states in an open time, while the latter refers to return to numerically
the same state in a circular or closed time. It will become apparent, however,
that there is an intimate connection between eternal recurrence and cyclic
time. But first it is important to be more precise about what it means for time
to be open or, alternatively, to be closed.
One approach would be to relativize the open versus closed distinction
to observers. Thus, if we idealize an observer as an inextendible future-
directed timelike curve y, we could say that for observer 0(y) time is open
(respectively, closed) just in case y is open (respectively, closed). For garden
variety spacetime this definition entails that for every observer time is open.
In Godel spacetime the definition entails that time for any non-accelerated
observer is open, whereas time for some sufficiently strongly accelerated
observers is closed. Although it is hard to fault this definition on its own terms,
it does seem overly narcissistic. For instance, in the cylindrical spacetime of
Fig. 7.1, time for 0(y) is open since y never returns to the same spacetime
location. But there is also a clear intuitive sense in which time itself in this
universe should be counted as circular.
214 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 7.1 An observer with an open world line in a temporally closed spacetime

To get away from the narcissism of particular observers we could look

at a congruence ^ of inextendible timelike curves (i.e., through every point
of the spacetime there passes exactly one member of *<?). Since we are dealing
with time orientable spacetimes, such a congruence always exists. We could
then say that relative to <<?, time in M, gab is open (respectively, closed) just
in case every y e <$ is open (respectively, closed). The Godel universe is dust
filled. The world lines of the dust particles are timelike geodesies and,
therefore, are open. Thus, the world lines of the dust form a congruence #
relative to which time in the Godel universe is (according to the present
definition) open. There is no alternative (€' relative to which time in the Godel
universe is (according to the present definition) closed. In the cylindrical
spacetime of Fig. 7.1 there is obviously a congruence of closed timelike curves.
But only slightly less obviously, there is also a congruence of open timelike
curves. The latter is open to the charge of group narcissism. It is true that
no world line of this group returns to the same event in spacetime, but however
one partitions the cylinder by connected time slices, each world line of the
group in question returns to the same slice—which argues that time can only
plausibly be considered as closed in this example. A related objection to the
present approach is that, as will be illustrated below, there are spacetimes
which can plausibly be considered to have a closed temporal structure even
if there are no CTCs.
The defender of the observer-oriented approach can shrug off these
objections. But for purposes of capturing the idea of a cyclic time structure
in the sense of return to the same state, a definition couched in terms of time
slices seems inescapable. The following idea then suggests itself.

DEFINITION 7.1
A time-oriented spacetime M, gab has an open time structure just in case there
is a linear time function, i.e., a map t: M ~+ K where (i) t is C°, (ii) for every
r 6 IR in the range of t, t^1 (r) is a time slice, and (iii) for any p, q 6 M such
that p « q, t(p) < t(q}.19

One way to justify this definition is to verify that the quotient topology is in
fact R. To explain what this means, suppose that = is ari equivalence relation
on the topological space X. The quotient topology assigned to the equivalence
classes Xj = is the finest topology in which the projection map n: X -> X/ =
 Eternal Recurrence, Cyclic Time, and All That 215

is continuous. In the present case, the equivalence relation for the linear time
function t is defined by the condition that for p, q € M, p = q just in case
t(p) — t(q). To show that the associated quotient topology is indeed R it
suffices to show that t is an open map. Towards this end, note that if Def.
7.1 holds, sets of the form /" (q) n /"*" (p] form a basis for the manifold topology
of M (see Hawking and Ellis 1973, pp. 196—197). It is easy to verify that
t(I~ (q) n I+ (p)) = (t(p), t(q)} and, consequently, that t is an open map.
Instead of Def. 7.1 we could have started with

DEFINITION 7.2
A time-oriented spacetime M, gab has an open time structure just in case there
is a collection ff of time slices that partition M and the quotient topology of
is R.

We have already seen that Def. 7.1 entails Def. 7.2. The converse is not hard
to establish.
In parallel with Defs. 7.1 and 7.2 one can try to characterize a closed or
circular time structure in the following two ways.

DEFINITION 7.3
A time-oriented spacetime M, gab has a closed time structure just in case there
is a circular time function, i.e., a map u: M —> Sl where (i) u is C°, (ii) for
any s e S1 , u ~ l ( s ) is a time slice, and (iii) for any distinct p, q,j, z 6 M, if
there is a future-directed timelike curve which goes from p to q to y to z and
which does not reintersect the slice u~1(u(p)), then u(p),u(j>) separate
u(q), u(z) on the circle S1 .

DEFINITION 7.4
A time-oriented spacetime M, gab has a closed time structure just in case there
is a collection y of time slices that partition M and the quotient topology
M/Sf is S1.

Just as Defs. 7.1 and 7.2 are equivalent, so are Defs. 7.3 and 7.4.
One potential drawback to the present approach is that it is not
applicable to many acausal spacetimes— Godel spacetime, for example, which
does not admit any time slices. However, the conclusion that Godel spacetime
cannot be taken to have either an open or a closed time structure strikes me
as correct. Similarly, it seems to me correct to conclude in line with Defs. 7.1
through 7.4 that the spacetime of Fig. 7.2, which can be partitioned by time
slices, is temporally neither open nor closed.
Somewhat more surprising and harder to swallow is the fact that
according to Def. 7.1 through 7.4 there are spacetimes that can be considered
to be both temporally open and closed. A relevant example is illustrated in
Fig. 7.3 where two half-lines have been removed from the two-dimensional
cylindrical spacetime of Fig. 7.1. If one likes, by introducing an appropriate
216 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

Fig. 7.2 A spacctime that is partitioned by time slices but is temporally neither open
nor closed
. 1 . . . 1

Fig. 7.3 A spacetime that may be considered to be both temporally open and closed

conformal factor in the metric, the lines can be taken to represent irremovable
curvature singularities. The removed lines overlap sufficiently that the
spacetime is stably causal (i.e., a slight widening of the light cones does not
result in CTCs; see chapter 6). By a theorem of Hawking and Ellis (1973,
Prop. 6.4.9) the spacetime admits of a global time function so that Def. 7.1
is fulfilled and time can be considered open. 20 On the other hand there is a
more obvious partition SP by time slices such that the quotient topology Mfff
isS 1 .
An even nastier surprise comes from Robert Geroch's 21 observation that
Defs. 7.3 and 7.4 allow us to count Minkowski spacetime as having a circular
time structure. To see this, choose the partition ,9" such that the level surfaces
of some inertial time coordinate that lie at constant multiples of some chosen
unit are counted as components of the same slice. One peculiar feature of this
slicing is that the slices are not achronal. But achronality is not in general a
reasonable condition to impose in cases of circular time structures; for
instance, achronality is violated in the spacetime of Fig. 7.1, which certainly
counts as a paradigm case of a circular time structure. Another notable feature
of the Geroch slicing of Minkowski spacetime is that the slices have discon-
 Eternal Recurrence, Cyclic Time, and All That 217

nected components; in fact, each slice has a countably infinite number of

them. Similarly, some of the level surfaces of any linear time function for the
spacetime of Fig. 7.3 will be disconnected. Thus, requiring connectedness of
the time slices would with a single stroke get rid of both of the counterintuitive
examples. But the price to be paid for soothing our intuitions is too high. For
example, in the case of two-dimensional Minkowski spacetime with a closed
ball removed, it seems natural to count the level surfaces of an inertial time
coordinate as good time slices, even though some of them will be disconnected.
If alternatively the disconnected components are regarded as different time
slices, then in this spacetime the topology of time, as characterized by the
quotient topology, will have a branching and then recoalescing structure
which is non-Hausdorff. The former alternative, which counts the topology
of time as R, seems preferable.
My proposal is that we keep Defs. 7.1 and 7.2 as they are and that we
simply swallow the consequence that time in the spacetime of Fig. 7.3
can be considered open. By contrast, however, I find it intolerable for time
in Minkowski spacetime to be counted as closed. Thus, while Defs. 7.3 and
7.4 supply necessary conditions for a closed time, they cannot be considered
to give sufficient conditions. The clue to an appropriate strengthening lies in
the fact that those examples where we have no compunction about saying
that a spacetime exhibits a closed time structure result from making identifica-
tions in a covering spacetime with an open and periodic time structure. If we
make this feature the definition of a closed time structure, then Defs. 7.3 and
7.4 will, of course, be satisfied, but Minkowski spacetime will no longer be
counted as exhibiting a closed time structure. I thus propose to replace Defs.
7.3 and 7.4 with

DEFINITION 7.5
A time-oriented spacetime M, gab has a closed time structure just in case M, gab
results from identifying the time slices modulo A e IR in a temporally open
and A-periodic covering spacetime.

If this proposal is accepted, we have the intimate connection, hinted at

above, between eternal recurrence and cyclic time. But the connection serves
to draw much of the interest of the latter; for a cyclic time structure is never
intrinsic but only arises because the Great Topologist has made identifications
in a larger spacetime. Furthermore, the various no-go results for eternal
recurrence discussed in previous sections also tell against the physical
possibility of cyclic time.
When he wrote The Philosophy of Space and Time, Hans Reichenbach
thought that we always have the option of interpreting away a CTC in favor
of an open timelike curve on which similar events are repeated over and over
ad infinitum (see Reichenbach 1958, pp. 140-143, 272-273). This belief is
mistaken if the interpreting away takes the form of unwinding the CTG in a
covering spacetime. As we know from chapter 6, Godel spacetime is a relevant
218 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

example; it contains CTCs, but because it is simply connected it is its own

universal covering spacetime. But given the present proposal for defining a
closed time structure, Reichenbach's position has more merit for cases where
GTCs inhabit a spacetime with a closed time structure. But it still remains
to ask whether the two descriptions—a cyclic time structure in the base
spacetime versus an open but periodic time structure in a covering spacetime—
are equivalent in a sense so strong that the choice between them can be
regarded as conventional.
At a minimum we would want it to be the case that the two spacetimes
are observationally indistinguishable. Clark Glymour (1977) has proposed a
criterion for observational indistinguishability which runs as follows:

DEFINITION 7.6
The spacetimes M, gab and M', g'ab are observationally indistinguishable just
in case for every inextendible timelike curve y of M, gab there is an inextendible
timelike curve y' of M', g'ab such that I~ (y) and /~ (y') are isometric, and
similarly with M, gab and M', g'ab interchanged.

This definition uses the idealization, already employed above, of observers as

inextendible timelike curves, 22 and it assumes that 'observation' is to be given
a causal reading so the events that an observer can in principle observe are
precisely those within her past light cone.23 Then the idea is that no observer
will be able to tell which of the two spacetimes she inhabits just in case for
any observer of the one spacetime there is an observer of the other such that
the past light cones of the two are the same in the sense of being isometric.24
Applying Def. 7.6 to examples of spacetimes with closed time structures yields
mixed results for Reichenbach. The spacetime of Fig. 7.3 possesses a closed
time structure according to my proposed analysis in Def. 7.5; and according
to Def. 7.6, it is observationally indistinguishable from its universal covering
spacetime. But this is an untypical case since the spacetime of Fig. 7.3 also
possesses an open time structure. More typical are cases like that of Fig. 7.1
where the spacetime and its universal covering are not observationally
distinguishable according to Def. 7.6. On behalf of Reichenbach it can be
argued that although the spacetimes in question are in principle observation-
ally distinguishable, no observer can be in a position to know which of the
two she inhabits, at least not if mind—body identity is true; for by construction
the corresponding states of the two spacetimes are identical as regards the
value of any physical quantity 25 and, thus, the corresponding mental states
that supervene on the body states of the observers must also be the same.

7.7 Conclusion

For those interested in exotic time structures, GTR proves to be something

of a tease. To illustrate how time can "branch," one can cite examples of
 Eternal Recurrence, Cyclic Time, and All That 219

general relativistic spacetimes that can be partitioned by a collection of time

slices where initially the slices are connected but later become disconnected.
By regarding the disconnected components as different time slices, the
quotient topology becomes a branching "Y". But typically the disconnected
pieces can be uniformly regarded as different components of the same slices
so that the quotient topology is the standard IR; indeed, if Penrose's form of
the cosmic censorship hypothesis (see chapter 3) holds, then there is a partition
by time slices in which all of the slices are connected.
Mathematical examples of general relativistic spacetimes illustrating
eternal recurrence and cyclic time structures can also be given. But because
of the occurrence of singularities the physics of GTR is distinctly unfriendly
towards the former; and if the proposed characterization of cyclic time
structure as a rolling up of open periodic spacetime is accepted, GTR is
equally unfriendly towards the latter. Just how unfriendly GTR can be is a
matter that cannot be resolved until we settle some delicate questions about
the nature of essential singularities.
One interesting by-product of the analysis is that the open time versus
closed time dichotomy is not a dichotomy, for there are spacetimes that exhibit
neither structure and others that exhibit both. In this sense the global
structures of general relativistic spacetimes are more exotic than popular and
philosophical imaginations have been able to grasp. According to the analysis
given here, a spacetime can be considered to have a closed time structure
even though it does not contain CTCs. But in those cases where CTCs are
involved, all the problems about causality discussed in chapter 6 are
implicated.

Notes
1. For a nice overview and references to the literature, see Tipler (1980).
2. In "The City of God" (Bk XIII, Ch. 13) St. Augustine responded to the idea
of "cycles of time," in which there should be a "constant renewal and repetition of
the order of nature" as follows: ". . . far be it, I say, from us to believe this. For since
Christ died for our sins; and, rising from the dead, He dieth no more" (Augustine
1948, pp. 191-192).
3. Nor is there any other admissible coordinate system in which all the
contravariant components are locally bounded. For if the components g'j are locally
bounded in the coordinate system {x'} in the atlas of C™ charts for M, then likewise
the components g"j in any other coordinate system {x"} belonging to the atlas are
locally bounded. (This is just to say that the requirement that gab and gab be locally
bounded is an invariant one.) The transformation rule for the contravariant com-
ponents of the metric isg"J = (8x"/dxm) (dx'J/dx")gm". The dx'k/8xp are certainly locally
bounded if the x'k are C™ functions of the xp. So if the gm" are locally bounded, g"J is
the sum of products of locally bounded functions and, thus, is itself locally bounded.
4. At least if the distribution must arise from a locally integrable tensor field; see
chapter 2.
5. But nothing prevents God from using such a measure to peer through the big
bang. The following section indulges in a bit of theology.
220 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

6. Letters to the Editor section, Wall Street Journal, June 16, 1993.
7. This point was made forcefully by Torretti (1979).
8. What are the necessary and sufficient conditions for a spacetime M, gab to have
a temporal structure that is finite in the past? The reader may want to think about
why the following is not a sufficient condition: every past-directed timelike half-curve
has finite proper length.
9. Perhaps theists will find this feature attractive since it means that although
God operates at the beginning of time, He is nevertheless near every event.
10. £v denotes the Lie derivative with respect to V. £Vl?ai> = VcVcgab + gci,VaVc +
gacVb Vc. This is equal to V, Vb + Vb Va since Vcgab = 0.
11. Equivalently, the rotation or twist tensor (Oab(V) of V vanishes; see chap-
ter 6.
12. It is typically assumed that the vanishing of Tab implies the vanishing of the
source fields. I will follow this assumption here.
13. In an FRW universe that expands from a zero volume and then recontracts
to a zero volume, there will be a time t = t* of maximal expansion. The time slices
t = t* + d, d > 0, situated symmetrically about t = t* carry the same intrinsic spatial
geometry. But this is not a counterexample to no recurrence since the extrinsic
curvatures of the slices are different. If £" is the normed timelike vector field orthogonal
to the t = constant slices, the extrinsic curvature of the slices is defined by Kab —:Va£,b.
At t = t* — d the universe is expanding (K"a > 0) while at t — t* + d the universe is
contracting (K"a < 0). In a static spacetime the extrinsic curvature of the orthogonal
hypersurfaces of the timelike Killing field vanishes.
14. Actually, as noted in chapter 2, C2~ will suffice.
15. I differ here with Torretti (1983, p. 337, note 26) on the substance and
significance of Tiplcr's no recurrence theorem.
16. The same caveats expressed above about extendibility apply here as well.
The relevant notion of closeness of states is made precise by using Sobolev spaces; see
Tipler (1980).
17. For a system of point masses, the closeness of states is measured in the natural
Euclidean metric on the classical phase space.
18. This section relies on and at the same time corrects some of my ideas in
Barman (1977).
19. Recall that p « q means that there is a future-directed timelike curve from
p to q. The present definition of a linear time function differs only slightly from the
definition of a global time function. For a global time function t it is required that t
be differentiable and that Vt be a timelike vector field. It follows that t is strictly
increasing along timelike curves (clause (iii) of Def. 7.1) and that the level surfaces
of t are spacelike hypersurfaces without edges (clause (ii) of Def. 7.1). Def. 7.1 is
adopted here in order to have a symmetry with the definition of a circular time function
given below.
20. The reader may wish to try to draw the level surfaces of a global time function
for the spacetime of Fig. 7.3.
21. Private communication.
22. The reader may want to consider the implications of modifying Def. 7.6 so
as to reflect the fact that actual observers only live for a finite amount of time; see
Malairicrit (1977).
23. For some caveats about this dogma, sec chapter 5.
24. One might want to strengthen Def. 7.6 by requiring not only that / (y) and
 Eternal Recurrence, Cyclic Time, and All That 221

/" (/) are isometric (which guarantees that the two are geometrically the same)
but also that all physical fields are the same at the corresponding points of the
isometry. As noted above, EFE do not guarantee that the latter follows from the
former.
25. At least if Def. 7.6 is strengthened in the manner indicated in note 24.
 8
Afterword

In a paper addressed to the measurement problem in QM, J. S. Bell and

M. Nauenberg wrote:
It seems that the quantum mechanical description will be superseded. In
this it is like all theories made by man. But to an unusual extent its
ultimate fate is apparent in its internal structure. It carries in itself the
seeds of its own destruction. (Bell and Nauenberg 1966, p. 285)
Do spacetime singularities signal that classical GTR contains the seeds of
its own destruction? Recall from chapter 1 Peter Bergmann's report of his
and Einstein's opinions:
It seems that Einstein always was of the opinion that singularities in
classical field theory are intolerable. They are intolerable because a singular
region represents a breakdown of the postulated laws of nature. I think one
can turn this argument around and say that a theory that involves
singularities and involves them unavoidably, moreover, carries within itself
the seeds of its own destruction. (Bergmann 1980, p. 156)
Couple these opinions with the Hawking-Penrose singularity theorems, and
the seeds of self-destruction in GTR bloom.
It seems to me, however, that as regards self-destruction there is an
important difference between QM and GTR. The measurement problem in
QM shows, prima facie, that the theory is empirically inadequate in the
worst way: it cannot account for the fact that measurement procedures
yield definite outcomes. As hope fades that some clever interpretational
ploy will resolve this problem, it becomes more likely that some overhaul of
quantum dynamics will be needed and, thus, that Bell and Nauenberg's
moral is correct. By contrast, there is no correspondingly blatant failure of
classical GTR to save the phenomena in connection with its prediction of
singularities in the gravitational collapse of stars and in the big bang origin
of the universe. In the former case, GTR's predictions about black holes
appear to be gaining empirical successes. In the latter case, there are
puzzles connected with the particle horizons of the standard big bang
model. But as argued in chapter 5, these puzzles are to a large extent
concerned not with empirical adequacy but with the nature of scientific
222
 Afterword 223

explanation; and in any case, the most popular fixes for the horizon
problem do not banish the initial singularity.
Of course, the acceptability of a scientific theory concerns much more
than the ability of the theory to save the phenomena. And it is evident that
for many physicists singularities present a formidable obstacle to the acceptance
of classical GTR. The first issue of this year's General Relativity and Gravitation
contains an article entitled "No More Spacetime Singularities?"; it pro-
nounces that "even as a classical theory, general relativity is deficient
as a theory of spacetime because it predicts the existence of singularities"
(Kostelecky and Perry 1994, p. 7). 1 But what exactly is it about singularities
that makes GTR "deficient"? The authors do not say. In the Misner-
Thorne-Wheeler bible, Gravitation, Charles Misner speaks of the "abhor-
rence" of the theoretical prediction of infinite curvature and infinite density
which is "particularly heightened by the correlative prediction that these
infinities occurred at a finite proper time in the past, and would—if they
recur—occur again at some finite proper time in the future" (p. 813).2
Relativity theory, in both its special and general forms, implies all sorts of
things that seem abhorrent to intuitions trained on Newtonian physics. But
generally the conclusion to be drawn is not that something is wrong with
relativity theory but rather that intuitions need retraining. What then is it
about spacetime singularities that calls for a modification of the theory rather
than intuitions?
As if responding to this question, John Wheeler opined in Gravitation
that the singularities of gravitational collapse confront physics with "its
greatest crisis ever" (p. 1198). What is this crisis? For Wheeler it is encapsulated
in the "paradox" of gravitational collapse: GTR says "'This is the end' and
physics says 'there is no end'" (p. 1197), or "'collapse ends physics'; 'collapse
cannot end physics'" (p. 1198). Granted, there is a superficial paradox in
saying "The laws of GTR entail the existence of singularities," and then
adding in the same breath "And in doing so they entail their own demise."
But the air of paradox here is due to a loose way of speaking. For the
singularities entailed by the theory do not exist at any spacetime location,
and there is no event in spacetime where the laws fail to hold (chapter 2).
Perhaps this dissolution of Wheeler's paradox is too facile, for it is
achieved by refusing to count as part of spacetime any location where the
metric and, hence, the laws of GTR are not well defined. Perhaps then the
residuum of a paradox lies in the fact that in entailing singularities GTR
demonstrates its own incompleteness. This is the opinion of Brandenberger
et al.
Singularities are undesirable for a theory which claims to be complete, since
their existence implies that spacetime cannot be continued past them. The
spacetime structure becomes unpredictable already at the classical level . . .
The presence of singularities is an indication that G[T]R is an incomplete
theory. Wheeler even talks about a "crisis in physics." (Brandenberger et
al. 1993, p. 1629)
224 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

In response to this "crisis" Brandenberger et al. propose to modify Einstein's

field equations in such a way that singularities are avoided without sacrificing
well-tested predictions of standard GTR.3 But by its own lights GTR is not
guilty of incompleteness because it entails black hole and big bang singularities.
For either the singularities are essential or not. If they are inessential (i.e.,
removable), what's the beef? If they are essential and cannot be removed by
any extension in which the laws of GTR make sense, then by the lights of
the theory there is nothing further to be said. The theory is hardly convicted
out of its own mouth of incompleteness for failing to answer questions about,
for example, what happens 'before' the big bang and 'after' the big crunch,
at least not if the theory implies, as argued in chapter 7, that such questions
are not physically meaningful.
Perhaps the seeds of self-destruction are rooted not in incompleteness but
in falsity. Perhaps, that is, in entailing singular behavior GTR is committed
to empirically false predictions. But to sustain an interesting form of the seeds
of its own destruction, the reason for thinking that GTR is committed to false
predictions would have to be stated in the rubric of classical GTR. I know
of no such reasons. Indeed, leaving aside possible quantum effects, GTR
continues to pass every empirical test with flying colors (see Will 1993). A
much different charge is that considerations from quantum physics suggest
GTR breaks down in the vicinity of curvature singularities. Thus Robert
Wald writes "The prediction of singularities undoubtedly represents a
breakdown of general relativity in that its classical description of gravitation
and matter cannot be expected to remain valid at the extreme conditions
expected near a spacetime singularity" (Wald 1984a, p. 212). Behind Wald's
remark is the idea that when curvature becomes sufficiently strong, quantum
effects "which invalidate classical general relativity will play a dominant
role" (ibid., p. 212). Of course, we know the singularities of GTR need not
involve unbounded curvature (chapter 2). And even when they do, it remains
a pious hope that some quantum theory of gravity, yet to be formulated, will
contain mechanisms for the avoidance of singularities. Leaving aside such
demurrers, the main point I wish to emphasize is that we have strayed far
from the original contention that even taken on its own terms, classical GTR
contains the seeds of its own destruction because it entails singularities. At
least we have come far enough to break the analogy with the measurement
problem in QM.
Having found no obvious merit in the charge that, even taken on its own
terms, classical GTR stands convicted out of its own mouth of some heinous
crime for pronouncing the existence of singularities, it is well to consider the
opposite attitude that singularities are seeds of confirmation rather than seeds
of self-destruction. After all, spacetime singularities are a feature that separates
GTR from all of its predecessor Newtonian and special relativistic theories
and from some of its competitor theories of gravitation. 4 So by confirming
the existence of singularities, the theory would receive a big boost in empirical
support. In addition, it is not hard to develop a fondness for certain types of
 Afterword 225

singularities. For example, a black hole singularity can be appreciated both

as the ultimate garbage dump, able to take care of any waste disposal problem
without the need to recycle, and as a source of extractable energy (see Wald
1984a, pp. 324-330).
There are two obstacles to this embrace of singularities. The first concerns
how knowledge of the existence of singularities is to be achieved. To say that
spacetime singularities exist is not to say that there are such and such events
in spacetime whose presence can be detected, if only indirectly. (And the fact
that we lack an attractive procedure for attaching singular points to the
spacetime manifold speaks against talking about spacetime singularities as if
they were objects, even ideal objects.) Rather to say that spacetime singularities
exist—or better, that spacetime is singular—is to say that the large-scale
structure of spacetime has such and such features, where the such and such
features may be complicated and abstruse (chapter 2). But though difficult,
the matter is not altogether desperate. By way of analogy, it is not easy to
establish reasonably secure knowledge claims about other large-scale features
of spacetime, for example, as to whether space is open or closed; for such
claims must rely on a number of auxiliary theoretical assumptions, each of
which stands in need of its own justification. And aside from a pervasive
skepticism and antirealism, which would deny all knowledge claims outside
of the realm of the directly observable, I do not see any reason in principle
why claims about large-scale features of spacetime—including its singularity
structure—cannot be established, at least by the standards that scientists use
for evaluating other theoretical claims.
The second and more serious obstacle to the embrace of singularities is
that no one wants to hug a naked singularity. (Exceptions: those with a taste
for weirdness and those who want to perform supertasks; see chapter 4.)
Indeed, if cosmic censorship fails for GTR, then it would seem that classical
GTR is convicted out of its own mouth of the sin of incompleteness. At least
the conviction is sustained if determinism is required for completeness for
non-quantized theories, for violations of cosmic censorship are inextricably
bound up with a breakdown in determinism (chapter 3). Perhaps one can
also argue that violations of cosmic censorship would show that classical GTR
is incomplete in a stronger sense. The premise required is not that determinism
holds but the weaker premise that all physical processes be law governed. The
argument would be completed by showing that classical GTR places no
constraints, not even statistical ones, on what can emerge from a naked
singularity. To date the evidence pro and con on whether GTR contains
built-in mechanisms for enforcing cosmic censorship has been both scanty and
mixed. Progress has been slow and probably will continue to be so owing to
the difficulty in formulating and proving censorship theorems and in con-
structing counterexamples. It is to be hoped that the growing activity in
numerical relativity will give us more insight into this crucial issue.
The presence of CTCs in solutions to Einstein's field equations can serve
to buttress the charge that classical GTR is an incomplete theory insofar as
226 BANGS, CRUNCHES, WHIMPERS, AND SHRIEKS

CTCs and other acausal features involve violations of cosmic censorship.

Independently of cosmic censorship, CTCs would also support a separate
charge of incompleteness if they were deemed to be conceptually or physically
impossible, for then some selection principle over and above the laws of
classical GTR would be needed to exclude them. However, it was argued in
chapter 6 that the grandfather paradox and its ilk do not speak in favor of
such an impossibility but are simply crude devices for bringing out the
existence of consistency constraints entailed by the presence of CTCs. In some
instances these constraints may aspire to law status. When the aspirations are
fulfilled, GTR is not thereby shown to be incomplete for the actual world;
rather, what is shown is that in some possible worlds which are nomologically
accessible from the actual world and which contain CTCs, there are laws
over and above those of classical GTR. But on an empiricist conception of
laws this is hardly surprising since in traveling to other possible worlds—
whether or not those worlds contain CTCs—one should be prepared to find
that the laws are not the same as those of the actual world.5
Suppose for sake of discussion the reader is willing to seriously entertain
my position that the fact GTR entails the existence of spacetime singularities
need not mean it contains the seeds of its own destruction and that a
generalized horror singulariti is not justified. How does it affect the search
for a quantum theory of gravity? Obviously, the banishment of all spacetime
singularities is no longer to be taken as a desideratum for quantum gravity.
But beyond that obvious consequence the way forward is not clear. Although
there may be no sound basis for a general horror of singularities, some types
of singularities are accompanied by features that are justly cause for concern—
the acausality of CTCs and the gross failure of determinism associated with
naked singularities being the principle ones. For those for whom these
concerns amount to alarm, quantum gravity looms as a savior. In chapter 6
it was seen that at present the prospects for proving chronology protection
theorems in classical GTR seem dim. 6 Quantum gravity promises help since
semiclassical calculations indicate that in some situations quantum fields
diverge strongly on chronology horizons. Thus, one may hope that a full
quantum theory of gravity will contain the mechanisms to prevent the
manufacture of CTCs.7 And one can also hope that if classical GTR lacks
the resources to prevent the development of other types of naked singularities
not associated with CTCs, then quantum gravity can supply the resources
for censorship; at present, however, there is no clear basis for this latter hope.
Philosophers of science can be justly proud of their contributions to the
foundations of QM and, in particular, to clarifying the measurement problem
and to elucidating the meaning of the Bell inequalities. But thus far their
meager efforts towards understanding the foundations of the other great
theory of modern physics, GTR, and, in particular, towards understanding
the problem of spacetime singularities docs not merit any corresponding pride.
This book is an initial effort to set out some of the many facets of the problem,
to explain its intrinsic interest, and to indicate some of its implications for the
 Afterword 111

foundations of physics and the philosophy of science. It was written in the

faith that, if adequately revealed, the problem of spacetime singularities will
not remain the orphan of the philosophy of science and that if adopted as a
rightful child it will enrich not only the philosophy of space and time but
other members of the family as well. My faith can only be vindicated by the
work of other, more able hands.

Notes
1. The authors go on to consider string theory as a way of avoiding singularities.
2. Misner goes on to consider ameliorating the problem by pushing the initial
singularity into the infinite past; see Misner, Thome, and Wheeler (1973, pp.
813-814).
3. A different proposal for achieving the same aim is given in Cornish and Moffat
(1994).
4. Singularities are not very effective in separating GTR from other classical
relativistic theories of gravity since in proving the existence of singularities Einstein's
field equations are used in a weak way—essentially to derive the consequence that
RabVVb > 0 for any non-spacelike F"; see chapter 2.
5. By an empiricist conception of laws I mean one which makes the laws of a
world supervene on the occurrent features of that world; see chapter 6.
6. On this matter, see also Earman (1994).
7. Even if quantum gravity presents the manufacture of CTCs, it does not follow
that quantum gravity is incompatible with CTCs in general. Quantum field theory
on curved spacetimes provides some hints on this matter. Interacting quantum fields
lose the property of unitarity in the presence of CTCs. However, a reasonable
probability interpretation may still be possible (see Friedman et al. 1992).
 References

Abbott, P. (ed) (1986). Inflationary Cosmology. Philadelphia: World Scientific.

Akdeniz, K. G., Arik, M., Mutus, A., and RizaSglu, E. (1991). "Coasting Kaluza-
Klein Cosmology," Modern Physics Letters A, 6, 1543-1546.
Albert, D. (1992). Quantum Mechanics and Experience. Cambridge, MA: Harvard
University Press.
Albrecht, A., and Steinhardt, P. J. (1982). "Cosmology for Grand Unified Theories
with Radiatively Induced Symmetry Breaking," Physical Review Letters, 48,
1220-1223.
Allis, V., and Koetsier, T. (1991). "On Some Paradoxes of the Infinite," British
Journal for the Philosophy of Science, 42, 187-194.
Anderson, P. (1983). "Effects of Quantum Fields on Singularities and Particle
Horizons in the Early Universe," Physical Review D, 28, 271-285.
. (1984). "Effects of Quantum Fields on Singularities and Particle Horizons
in the Early Universe. II," Physical Review D, 29, 615-627.
Arntzenius, F. (1992). "The Common Cause Principle," in D. Hull, M. Forbes, and
K. Okruhlik (cds), PSA 1992, Volume 2. East Lansing, MI: Philosophy of
Science Association, 227-237.
Augustine, St. (1948). "City of God," in W. J. Gates (cd), The Basic Writings of
Saint Augustine, Volume II. New York: Random House, 3—663.
Barrow, J. D. (1988). "The Inflationary Universe: Modern Developments," Quarterly
Journal of the Royal Astronomical Society, 29, 101-117.
, and Silk, J. (1980). "The Structure of the Early Universe," Scientific
American, 242 (April), 118-128.
, and Tiplcr, F. J. (1986). The Anthropic Cosmological Principle. Oxford: Clarendon
Press.
Bell, J. S., and Nauenberg, M. (1966). "The Moral Aspect of Quantum Mechanics,"
in A. De Shalit (ed), Preludes in Theoretical Physics. Amsterdam: North-
Holland, 279-286.
Benaccrraf, P. (1962). "Tasks, Super-Tasks, and the Modern Eleatics," Journal of
Philosophy, LIX, 765-784.
, and Putnam, H. (eds) (1983). Philosophy of Mathematics. 2nd ed. Cambridge:
Cambridge University Press.
Bergmann, P. G. (1942). Introduction to the Theory of Relativity. New York: Prentice-Hall.
. (1980). Remarks made included in "Open Discussion, Following Papers by
S. W. Hawking and W. G. Unruh," in H. Woolf (cd), Some Strangeness in the
Proportion. Reading, MA: Addison-Wesley, 156.
228
 References 229

Birkhoff, G. D. (1923). Relativity and Modern Physics. Cambridge, MA: Harvard

University Press.
Blau, S. K., and Guth, A. H. (1987). "Inflationary Cosmology," in S. W. Hawking
and W. Israel (eds), Three Hundred Years of Gravitation. Cambridge: Cambridge
University Press, 524-603.
Borner, G. (1988). The Early Universe. Berlin: Springer-Verlag.
Boulware, D. G. (1992). "Quantum Field Theory in Spaces with Closed Timelike
Curves," Physical Review D, 46, 4421-4441.
Brandenberger, R., Mukhanov, V., and Sornborger, A. (1993). "Cosmological Theory
without singularities," Physical Review D, 48 1629-1642.
Brans, C. H., and Randall, D. (1993). "Exotic Differentiable Structures and General
Relativity," General Relativity and Gravitation, 25, 205-221.
Brown, B. (1992). "Defending Backwards Causation," Canadian Journal of Philosophy,
22, 429-444.
Budic, R., Isenberg, J., Lindblom, L., and Yasskin, P. B. (1978). "On the Determina-
tion of Cauchy Surfaces from Intrinsic Properties," Communications in Mathe-
matical Physics, 61, 87-95.
Carroll, J. W. (1994). Laws of Nature. New York: Cambridge University Press.
Carroll, S. M., Farhi, E., and Guth, A. H. (1992). "An Obstacle to Building a Time
Machine," Physical Review Letters, 68, 263-266.
Carter, B. (1971). "Causal Structure in Spacetime," General Relativity and Gravitation,
1, 349-391.
Chandrasekhar, S., and Hartle, J. B. (1982). "On Crossing the Cauchy Horizon of a
Reissner-Nordstrom Black-Hole," Proceedings of the Royal Society (London) A, 348,
301-315.
Chapman, T. (1982). "Time Travel," and "Two Times," in Time: A Philosophical
Analysis. Dordrecht: D. Reidel, 134-145.
Charlton, N., and Clarke, C. J. S. (1990). "On the Outcome of Kerr-Like Collapse,"
Classical and Quantum Gravity, 7, 743-749.
Chihara, C. S. (1965). "On the Possibility of Completing an Infinite Process,"
Philosophical Review, 74, 74-87.
Choquet-Bruhat, Y., and Geroch, R. (1969). "Global Aspects of the Cauchy Problem
in General Relativity," Communications in Mathematical Physics, 14, 329-335.
, De Witt-Morette, C., and Dillard-Bleich, M. (1982). Analysis, Manifolds and
Physics. Revised edition. Amsterdam: North-Holland.
Chrusciel, P. T. (1992). "On Uniqueness in the Large of Solutions of Einstein's
Equations ('Strong Cosmic Censorship')," Contemporary Mathematics, 132. Provi-
dence: American Mathematical Society, 235-273.
, and Isenberg, J. (1993). "Non-Isometric Vacuum Extensions of Vacuum
Maximal Globally Hyperbolic Spacetimes," Physical Review D, 48, 1616-1628.
, Isenberg, J., and Moncrief, V. (1990). "Strong Cosmic Censorship in
Polarized Gowdy Spacetimes," Classical and Quantum Gravity, 1, 1671-1680.
Clarke, C. J. S. (1973). "Local Extensions in Singular Space-Times," Communica-
tions in Mathematical Physics, 32, 205-214.
. (1975a). "The Classification of Singularities," General Relativity and Gravitation,
6, 35-40.
. (1975b). "Singularities in Globally Hyperbolic Space-Time," Communica-
ions in Mathematical Physics, 41, 65-78.
. (1977). "Time in General Relativity," in J. Earman, C. Glymour, and J.
230 References

Stachel (cds), Foundations of Space- Time Theories, Minnesota Studies in the Philosophy
of Science, Volume VIII. Minneapolis: University of Minnesota Press, 94—
108.
. (1988). "Singularities, Problems and Prospects," in B. R. Iyer, A. Kembhavi,
J. V. Narlikar, and C. V. Vishveshwara (eds), Highlights in Gravitation and
Cosmology. Cambridge: Cambridge University Press, 15—29.
. (1993). Analysis of Space-Time Singularities. Cambridge: Cambridge University
Press.
, and Krolak, A. (1985). "Conditions for the Occurrence of Strong Curvature
Singularities," Journal of Geometry and Physics, 2, 127—143.
, and Schmidt, B. G. (1977). "Singularities: The State of the Art," General
Relativity and Gravitation, 8, 129-137.
Clauser, J. F., and Home, M. A. (1974). "Experimental Consequences of Objective
Local Theories," Physical Review D, 10, 526-535.
Collins, C. B., and Hawking, S. W. (1973). "Why is the Universe Isotropic?"
Astrophysical Journal, 180, 317-334.
, and Stewart, J. M. (1971). "Qualitative Cosmology," Monthly Notices of the
Royal Astronomical Society, 153, 419-434.
Cornish, N. J., and Mofat, J. W. (1994). "Nonsingular Gravity Without Black Holes,"
Journal of Mathematical Physics, 35, 6628-6643.
Craig, W. L. (1991). "Pseudo-dilemma?," Nature, 254, 347.
. (1994). "Professor Griinbaum on Creation," Erkennlnis, 40, 325-341.
, and Smith, Q. (1993). Theism, Atheism, and Big Bang Cosmology. Oxford:
Clarendon Press.
Grosswcll, K. (1993). "The Quest for the Cosmological Constant," New Scientist, 137
(20 February), 23-27.
Curzon, H. E. J. (1924a). "Bipolar Solutions of Einstein's Gravitation Equations,"
Proceedings of the London Mathematical Society, Series 2, 23, xxix.
. (1924b). "Cylindrical Solutions of Einstein's Gravitation Equations," Proceed-
ings of the London Mathematical Society, Series 2, 23, 477—480.
Cutler, C. (1992). "Global Structure of Gott's Two-String Spacetime," Physical Review
D, 45, 487-494.
De, U. K. (1969). "Paths in Universes Having Closed Time-Like Lines," Journal of
Physics A (Series 2), 2, 427-432.
dc Sitter, W. (1917a). "On the Curvature of Space," Koninklijke Akademie van
Wetenschappen te Amsterdam. Proceedings of the Section of Sciences, 20, 229-242.
. (1917b). "On Einstein's Theory of Gravitation and its Astronomical Conse-
quences. Third Paper," Monthly Notices of the Royal Astronomical Society, 78, 3-28.
. (1918). "Further Remarks on the Solutions of the Field-Equations of
Einstein's Theory of Gravitation," Koninlijke Akademie van Wetenschappen te
Amsterdam. Proceedings of the Section of Sciences, 20, 1309—1312.
Deser, S. (1993). "Physical Obstacles to Time-Travel," Classical and Quantum Gravity,
10, S67-S73.
,Jakiw, R., and 't Hooft, G. (1992). " Physical Cosmic Strings Do Not Generate
Closed Timelikc Curves," Physical Review Letters, 68, 267—269.
Deutch, D. (1991). "Quantum Mechanics near Closed Timelike Lines," Physical Review
D, 44, 3197-3217.
Dieckmann, J. (1988). "Cauchy Surfaces in Globally Hyperbolic Space-Times,"
Journal of Mathematical Physics, 29, 578-579.
 References 231

Dominici, D., Holman, R., and Kim, C. W. (1983). "Horizon Problem in Brans-Dicke
Cosmology," Physical Review D, 28, 2983-2986.
Droste, J. (1917). "The Field of a Single Center in Einstein's Theory of Gravita-
tion, and the Motion of a Particle in That Field," Koninklijke Akademie
van Wetenschappen te Amsterdam. Proceedings of the Section of Sciences, 19, 197-
215.
Dummett, M. (1986). "Causal Loops," in R. Flood and M. Lockwood (eds), The
Nature of Time. New York: Basil Blackwell, 135-169.
Dwyer, L. (1975). "Time Travel and Changing the Past," Philosophical Studies, 27,
341-350.
. (1977). "How to Affect, but Not Change the Past," Southern Journal of
Philosophy, 15, 383-385.
. (1978). "Time Travel and Some Alleged Logical Asymmetries between Past
and Future," Canadian Journal of Philosophy, 8, 15—38.
Eardley, D. M. (1987). "Naked Singularities in Spherical Gravitational Collapse,"
in B. Carter and J. B. Hartle (eds), Gravitation in Astrophysics, Cargese 1987. New
York: Plenum Press, 229-235.
Earman, J. (1977). "How to Talk About the Topology of Time," Nous, 11, 211-
226.
. (1986). A Primer on Determinism. Dordrecht: D. Reidel.
. (1989). World Enough and Spacetime: Absolute vs. Relational Theories of Space and
Time. Cambridge, MA: MIT Press.
. (1992). Bayes or Bust? A Critical Examination of Bayesian Confirmation Theory.
Cambridge, MA: MIT Press.
. (1993). "In Defense of Laws: Reflections on Bas van Fraassen's Laws and
Symmetries," Philosophy and Phenomenological Research, LIII, 413-419.
. (1994). "Outlawing Time Machines: Chronology Protection Theorems,"
Erkenntnis. In press.
, and Glymour, C. (1980). "The Gravitational Red Shift as a Test of General
Relativity: History and Analysis," Studies in History and Philosophy of Science, 11,
175-214.
, and Janssen, M. (1993). "Einstein's Explanation of the Motion of Mercury's
Perihelion," in J. Earman, M. Janssen, and J. D. Norton (eds), The Attraction
of Gravitation: New Studies in the History of General Relativity, Einstein Studies,
Volume 5. Boston: Birkhauser, 129-172.
, and Norton, J. D. (1994). "Infinite Pains: The Trouble with Supertasks," to
appear in S. Stich (ed), Paul Benacerraf: The Philosopher and His Critics. New
York: Blackwell.
Echeverria, F., Klinkhammer, G., and Thorne, K. S. (1991). "Billiard Balls in
Wormhole Spacetimes with Closed Timelike Curves; Classical Theory," Physical
Review D, 44, 1077-1099.
Eddington, A. S. (1923). The Mathematical Theory of Relativity. Cambridge: Cambridge
University Press.
. (1924). "A Comparison of Whitehead's and Einstein's Formulae," Nature,
113, 192.
Ehring, D. (1987). "Personal Identity and Time Travel," Philosophical Studies, 52,
427-433.
Einstein, A. (1914). "Die Formale Grundlagc der Allegemeiner Relativitatstheorie,"
Kb'niglich Preussische Akademie (Berlin). Sitzungsberichte, 1030-1085.
232 References

. (1915). "Erklarung der Perihelbewegung des Merkur aus der Allgemeinen

Relativitatstheorie," Koniglich Preussische Akademie (Berlin). Sitzungsberichte, 844-
847.
. (1916). "Die Grundlagen der Allgemeinen Relativitatstheorie," Annalen der
Physik, 49, 769-822. Reprinted in English translation in W. Perrett and G. B.
Jeffrey (eds), The Principle of Relativity. New York: Dover, 1957.
. (1918). "Kritisches zu Einer von Hrn. De Sitter Gegebenen Losung der
Gravitationsgleichungen," Koniglich Preussische Akademie der Wissenschaften (Berlin).
Sitzungsberichte, 270-272.
. (1927). "Allgemeine Relativitatstheorie und Bewegungsgesetz," Preussische
Akademie der Wissenschaften (Berlin). Plysikalisch-Mathematische Klasse. Sitzungs-
berichte, 235-245.
. (1931). "Zum Kosmologischen Problem der Allgemeinen Relativitatstheorie,"
Preussische Akademie der Wissenschaften (Berlin). Physikalisch-Mathematische Klasse.
Sitzungsberichte, 235-237.
. (1939). "On a Stationary System with Spherical Symmetry Consisting of
Many Gravitating Masses," Annals of Mathematics, 40, 922-936.
. (1941). "Demonstration of the Non-Existence of Gravitational Fields with a
Non-Vanishing Total Mass Free of Singularities," Tucumdn Universidad National
Revisita, Series A, 2, 11-15.
. (1949a). "Autobiographical Notes," in P. A. Schilpp (ed), Albert Einstein:
Philosopher-Scientist. New York: Harper and Row, Volume 1, 1-95.
. (1949b). "Remarks Concerning the Essays Brought Together in This
Co-Operative Volume," in P. A. Schilpp (ed), Albert Einstein: Philosopher-Scientist.
New York: Harper and Row, Volume 2, 665-688.
. (1954). Ideas and Opinions. New York: Crown.
. (1955). The Meaning of Relativity. 5th cd. Princeton: Princeton University
Press.
. (1961). Relativity, The Special and General Theory. New York: Crown Books.
First published in 1922 as Uber die Spezielle und die Allgemeine Relativitats-
theorie, Gemeinverstandich. Braunschweig: Vieweg.
, and Grommer, J. (1927). "Allgemeine Relativitatstheorie und Bewegungs-
gesetz," Preussische Akademie der Wissenschaften (Berlin). Physikalisch-Mathematische
Klasse. Sitzungsberichte, 2-13.
, and Infeld, L. (1949). "On the Motion of Particles in General Relativity
Theory," Canadian Journal of Mathematics, 1, 209-241.
, and Pauli, W. (1943). "On the Non-Existence of Regular Stationary Solutions
of Relativistic Field Equations," Annals of Mathematics, 44, 131-137.
, and Rosen, N. (1935). "The Particle Problem in the General Theory of
Relativity," Physical Review, 48, 73-77.
, and Rosen, N. (1936). "Two-Body Problem in General Relativity Theory,"
Physical Review, 49, 404-405.
, Infeld, L., and Hoffmann, B. (1938). "The Gravitational Equations and the
Problem of Motion," Annals of Mathematics, 39, 65-100.
Eisenstaedt, J. (1989). "The Early Interpretation of the Schwarzschild Solution," in
D. Howard and J. Stachel (eds), Einstein and the History of General Relativity,
Einstein Studies, Volume 1. Boston: Birkhauser, 213-233.
. (1993). "Lemaitrc and the Schwarzschild Solution," in J. Earman, M.
Jarissen, and J. D. Norton (eds), The Attraction of Gravitation: New Studies in the
 References 233

History of General Relativity, Einstein Studies, Volume 5. Boston: Birkhauser,

353-389.
Ellis, G., and Tavakol, R. (1994). "Mixing Properties of Compact K = -1 FLRW
Models," in D. Hobhill, A. Burd, and A. Coley (eds), Deterministic Chaos In
General Relativity. New York: Plenum Press, 237-250.
Ellis, G. F. R. (1971). "Topology and Cosmology," General Relativity and Gravitation,
2, 7-21.
. (1988). "Does Inflation Necessarily Imply Q = 1?" Classical and Quantum
Gravity, 5, 891-901.
, and Schmidt, B. G. (1977). "Singular Space-Times," General Relativity and
Gravitation, 8, 915-953.
, and Schreiber, G. (1986). "Observational and Dynamical Properties of Small
Universes," Physics Letters A, 115, 97-107.
, and Sciama, D. W. (1972). "Global and Non-Global Problems in Cosmology,"
in L. O'Raifeartaigh (ed), General Relativity: Papers in Honour of J. L. Synge.
Oxford: Clarendon Press, 35-59.
, and Stoeger, W. (1988). "Horizons in Inflationary Universes," Classical and
Quantum Gravity, 5, 207-220.
, Maartens, R., and Nel, S. D. (1978). "The Expansion of the Universe,"
Monthly Notices of the Royal Astronomical Society, 184, 439-465.
Finkelstein, D. (1958). "Past-Future Asymmetry of the Gravitational Field of a Point
Particle," Physical Review, 110, 965-967.
Finlay-Freundlich, E. (1951). "Cosmology," in International Encyclopedia Of Unified
Science, Volume I, No. 8. Chicago: University of Chicago Press.
Fischer, A. E., and Marsden, J. E. (1979). "The Initial Value Problem and the
Dynamical Formulation of General Relativity," in S. W. Hawking and W.
Israel (eds), General Relativity: An Einstein Centenary Survey. Cambridge: Cambridge
University Press, 138-211.
Fleming, G. N. (1989). "Lorentz Invariant State Reduction and Localization," in A.
Fine and J. Leplin (eds), PSA 1988, Vol. 2. East Lansing, MI: Philosophy of
Science Association, 112-126.
Fock, V. (1939). "Sur la Mouvement des Masses Finies d'Apres la Theorie de
Gravitation Einsteinienne," Academic of Sciences URSS, Journal of Physics, 1,
81-116.
Forster, M. (1986). "Unification and Scientific Realism Revisited," in A. Fine and
P. Machamer (eds), PSA 1986, Vol. 1. East Lansing, MI: Philosophy of
Science Association, 394-405.
Friedman, J. L., and Morris, M. S. (1991a). "The Cauchy Problem for the Scalar
Wave Equation Is Well Defined on a Class of Spacetimes with Closed Timelike
Curves," Physical Review Letters, 66, 401-404.
, and Morris, M. S. (1991b). "The Cauchy Problem on Spacetimes with Closed
Timelike Curves," Annals of the New York Academy of Sciences, 631, 173-
181.
, Morris, M. S., Novikov, I. D., Echeverria, F., Klinkhammer, G., Thorne,
K. S., and Yurtsever, U. (1990). "Cauchy Problem in Spacetimes with Closed
Timelike Curves," Physical Review D, 42, 1915-1930.
, Papastamatiou, N. J., and Simon, J. Z. (1992). "Failure of Unitarity for
Interacting Fields on Spacetimes with Closed Timelike Curves," Physical Review
D, 46, 4456-4469.
234 References

Frolov, V. P. (1991). "Vacuum Polarization in Locally Static Multiply Connected

Spacetime and Time-Machine Problem," Physical Review D, 43, 3878-3894.
, and Novikov, I. D. (1990). "Physical Effects in Wormholes and Time
Machines," Physical Review D, 42, 1057-1065.
Gamov, G. (1970). My World Line, New York: Viking Press.
Gardner, M. (1988). "Time Travel," in Time Travel and Other Mathematical Bewilder-
ments. New York: W. H. Freeman, 1-14.
Gautreau, R., and Anderson, J. L. (1967). "Directional Singularities in Weyl
Gravitational Fields," Physics Letters A, 25, 291-292.
Geroch, R. P. (1967). "Topology in General Relativity," Journal of Mathematical
Physics, 8, 782-786.
. (1968a). "What is a Singularity in General Relativity?" Annals of Physics, 48,
526-540.
. (1968b). "Local Characterization of Singularities in General Relativity,"
Journal of Mathematical Physics, 9, 450-465.
. (1969). "Limits of Spacetimes," Communications in Mathematical Physics, 13,
180-193.
. (1970a). "Singularities," in M. Carmeli, S. I. Fickler, and L. Witten (eds),
Relativity. New York: Plenum Press, 259-291.
. (1970b). "Domain of Dependence," Journal of Mathematical Physics, 11,
437-449.
. (1977). "Prediction in General Relativity," in J. Earman, C. Glymour, and
J. Stachel (eds), Foundations of Space-Time Theories, Minnesota Studies in the
Philosophy of Science, Volume VIII. Minneapolis: University of Minnesota Press,
81-93.
, and Horowitz, G. T. (1979). "Global Structure of Spacetimes," in S. W.
Hawking and W. Israel (eds), General Relativity: An Einstein Centenary Survey.
Cambridge: Cambridge University Press, 212-293.
, and Jang, P. S. (1975). "Motion of a Body in General Relativity," Journal
of Mathematical Physics, 16, 65—67.
, Liang, C., and Wald, R. M. (1982). "Singular Boundaries of Space-Times,"
Journal of Mathematical Physics, 23, 432-435.
, and Traschen, J. (1987). "Strings and Other Distributional Sources in
General Relativity," Physical Review D, 36, 1017-1031.
Gibbons, G. W., and Hawking, S. W. (1992). "Kinks and Topology Change," Physical
Review Letters, 69, 1719-1721.
, Hawking, S. W., and Siklos, S. T. C. (eds) (1983). The Very Early Universe.
Cambridge: Cambridge University Press.
Glymour, C. (1977). "Indistinguishable Space-Times and the Fundamental Group,"
in J. Earman, C. Glymour, and J. Stachel (eds), Foundations of Space-Time
Theories, Minnesota Studies in the Philosophy of Science, Volume VIII. Minneapolis:
University of Minnesota Press, 50-60.
Godel, K. (1949a). "An Example of a New Type of Cosmological Solutions of
Einstein's Field Equations of Gravitation," Reviews of Modern Physics, 21,
447-450.
. (1949b). "A Remark about the Relationship between Relativity Theory and
Idealistic Philosophy," in P. A. Schilpp (ed), Albert Einstein: Philosopher-Scientist.
New York: Harper and Row, Vol. 2, 557-562.
. (1952). "Rotating Universes in General Relativity Theory," Proceedings of the
 References 235

International Congress of Mathematicians (1950), 1. Providence: American Mathe-

matical Society, 175-181.
Goldwirth, D. S., Perry, M. J., and Piran, T. (1993). "The Breakdown of Quantum
Mechanics in the Presence of Time Machines," General Relativity and Gravitation,
25, 7-13.
Gott, J. R. (1991). "Closed Timelike Curves Produced by Pairs of Moving Cosmic
Strings: Exact Solutions," Physical Review Letters, 66, 1126—1129.
Gowdy, R. H. (1977). "Instantaneous Cauchy Surfaces, Topology Change, and
Exploding Black Holes," Journal of Mathematical Physics, 18, 1798-1801.
Grant, J. D. E. (1993). "Cosmic Strings and Chronology Protection," Physical Review
D, 47, 2388-2394.
Griinbaum, A. (1968). Modern Science and Reno's Paradoxes. London: George Allen and
Unwin.
. (1969). "Can an Infinitude of Operations be Performed in a Finite Time?"
British Journal for the Philosophy of Science, 20, 203-218.
. (1991). "Creation as a Pseudo-Explanation in Current Physical Cosmology,"
Erkennlnis, 35, 233-254.
Guth, A. H. (1981). "Inflationary Universe: A Possible Solution to the Horizon and
Flatness Problems," Physical Review D, 23, 347-356.
, and Steinhardt, P. (1989). "The Inflationary Universe," in P. Davies (ed),
The New Physics. Cambridge: Cambridge University Press, 34-60.
Hakim, R. (1984). "The Inflationary Universe: A Primer," in Laboratoire "Gravi-
tation et Cosmologie Relativistes," Universite Pierre et Marie Curie et
C.N.R.S., Institut Henri Poincare, Paris (ed), Gravitation, Geometry, and Relativis-
tic Physics, Lecture Notes in Physics, Vol. 212. Berlin: Springer-Verlag, 302-332.
Harrison, J. (1971). "Dr. Who and the Philosophers: Or Time-Travel for Beginners,"
Proceedings of the Aristotelian Society, Supplementary Volume 45, 1-24.
Hartle, J. B. (1983). "Quantum Cosmology and the Early Universe," in G. W.
Gibbons, S. W. Hawking, and S. T. C. Siklos (eds). The Very Nearly Universe.
Cambridge: Cambridge University Press, 59-89.
Havas, P. (1989). "The Early History of the 'Problem of Motion' in General Rela-
tivity," in D. Howard and J. Stachel (eds), Einstein and the History of General
Relativity, Einstein Studies, Volume 1. Boston: Birkhauser, 234-276.
. (1993). "The General Relativistic Two-Body Problem and the Einstein-
Silberstein Controversy," inj. Earman, M. Janssen, and J. D. Norton (eds),
The Attraction of Gravitation: New Studies in the History of General Relativity, Einstein
Studies, Volume 5. Boston: Birkhauser, 88-125.
Hawking, S. W. (1965). "On the Hoyle-Narlikar Theory of Gravitation," ,
of the Royal Society (London) A, 286, 313-319.
. (1967). "The Occurrence of Singularities in Cosmology. III. Causality and
Singularities," Proceedings of the Royal Society (London) A, 300, 187-201.
. (1979). "Comments on Cosmic Censorship," General Relativity and Gravitation,
10, 1047-1049.
. (1980). "Theoretical Advances in General Relativity," in H. Woolf (ed),
Some Strangeness in the Proportion. Reading, MA: Addison-Wesley, 145-152.
. (1988). A Brief History of Time. New York: Bantum Books.
. (1992). "Chronology Protection Conjecture," Physical Review D, 46, 603—611.
, and Ellis, G. F. R. (1973). The Large Scale Structure of Space-Time. Cambridge:
Cambridge University Press.
236 References

, and Penrose, R. (1970). "The Singularities of Gravitational Collapse and

Cosmology," Proceedings of the Royal Society (London) A, 314, 529-548.
, and Stewart, J. M. (1993). "Naked and Thunderbolt Singularities in Black
Hole Evaporation," Nuclear Physics B, 400, 393-415.
Hilbert, D. (1917). "Die Grundlagen der Physik: Zweiter Mitteilung," Konig-
liche Gesellschaft der Wissenschqften zu Gottingen. Mathematische-Physikalische Klasse.
Nachrichten, 55-76.
Hochberg, D., and Klephart, T. W. (1994). "Can Semi-Classical Wormholes Solve
the Cosmological Horizon Problem?" General Relativity and Gravitation, 26,
219-223.
Hogarth, M. L. (1992). "Does General Relativity Allow an Observer to View an
Eternity in a Finite Time?" Foundations of Physics Letters, 5, 173-181.
. (1993). "Predicting the Future in Relativistic Spacetimes," Studies In History
and Philosophy of Science, 24, 721-739.
. (1994). "Non-Turing Computers and Non-Turing Computability," in D.
Hull, M. Forbes, and R. M. Burian (eds), PSA 1994, Vol. 1. East Lansing:
Philosophy of Science Association, 126-138.
Horowitz, G. T. (1979). "Finding a Statement of Cosmic Censorship," General Relativity
and Gravitation, 10, 1057-1061.
Horwich, P. (1989). "Time Travel," in Asymmetries in Time. Cambridge, MA: MIT
Press, 111-128.
Hospers, J. (1967). An Introduction to Philosophical Analysis. 2nd ed. Englewood-Cliffs:
Prentice-Hall.
Hiibner, P., and Ehlers, J. (1991). "Inflation in Curved Model Universes with
Non-Critical Density," Classical and Quantum Gravity, 8, 333-346.
Israel, W. (1984). "Does a Cosmic Censor Exist?" Foundations of Physics, 14, 1049-1059.
. (1986). "The Formation of Black Holes in Nonspherical Collapse and Cosmic
Censorship," Canadian Journal of Physics, 64, 120—127.
Janis, A. I., Newman, E. T., and Winicour, J. (1968). "Reality of the Schwarzschild
Singularity," Physical Review Letters, 20, 878-880.
Jeans, J. (1936). "Man and the Universe," in J.Jeans, W. Bragg, E. V. Appleton, E.
Mellanby, J. B. S. Haldane, and J. Huxley (eds), Scientific Progress. New York:
Macmillan, 13-38.
Johnson, R. A. (1977). "The Bundle Boundary in Some Special Cases," Journal of
Mathematical Physics, 18, 898-902.
Joshi, P. S. (1985). "Topological Properties of Certain Physically Significant Subsets
of Spacetimes," in N. Dadhich, J. Krishna Rao, J. V. Narliker, and C. V.
Vishveshwara (eds), A Random Walk in Relativity and Cosmology. New York: John
Wiley, 128-136.
. (1993). Global Aspects in Gravitation and Cosmology. Oxford: Clarendon Press.
, and Dwivedi, I. H. (1992). "Naked Singularities in Non-Self-Similar
Gravitational Collapse of Radiation Shells," Physical Review D, 45, 2147-2150.
, and Saraykar, R. V. (1987). "Cosmic Censorship and Topology Change in
General Relativity," Physics Letters A, 120, 111-114.
Kahn, C., and Kahn, F. (1975). "Letters from Einstein to de Sitter on the Nature of
the Universe," Nature, 257, 451-454.
Kerszberg, P. (1989). The Invented Universe. Oxford: Clarendon Press.
Kim, S.-W., and Thorne, K. S. (1991). "Do Vacuum Fluctuations Prevent the
Creation of Closed Timelike Curves?" Physical Review D, 43, 3929-3947.
 References 237

King, A. R. (1974). "New Types of Singularity in General Relativity: The General

Cylindrically Symmetric Stationary Dust Solution," Communications in Mathe-
matical Physics, 38, 157-171.
Klein, F. (1918a). "Bermerkungen iiber die Beziehungen des De Sitter'schen Koordi-
natensystem B zu der Allgemeinen Welt Konstanter Positiver Krummung,"
Koninklijke Akademie van Wetenschappen te Amsterdam. Proceedings, 21 (1918-1919),
614-615.
. (1918b). "Uber die Integralform der Erhaltungssatze und die Theorie der
Raumlich-Geschlossen Welt," Kb'nigliche Gesellschaft der Wissenschaften zu Got-
tingen. Nachrichten, 394-423.
Klinkhammer, G. (1992). "Vacuum Polarization of Scalar and Spinor Fields near
Closed Null Geodesies," Physical Review D, 46, 3388-3394.
, and Thorne, K. S. (1990). "Billiard Balls in Wormhole Spacetimes with
Closed Timelike Curves. II. Quantum Theory," preprint.
Kodama, H. (1979). "Inevitability of a Naked Singularity Associated with the Black
Hole Evaporation," Progress in Theoretical Physics, 62, 1434-1435.
Kostelecky, V. A., and Perry, M. (1994). "No More Spacetime Singularities?" General
Relativity and Gravitation, 26, 7—12.
Kramer, D., Stephani, H., Herlt, E., and MacCallum, M. (1980). Exact Solutions of
Einstein's Field Equations. Cambridge: Cambridge University Press.
Kriele, M. (1989). "The Structure of Chronology Violating Sets with Compact
Closure," Classical and Quantum Gravity, 6, 1607-1611.
. (1990a). "A Generalization of the Singularity Theorem of Hawking and
Penrose to Space-Times with Causality Violations," Proceedings of the Royal
Society (London) A, 431, 451-464.
. (1990b). "Causality Violations and Singularities," General Relativity and
Gravitation, 22, 619-623.
Kristian, J., and Sachs, R. K. (1966). "Observations in Cosmology," Astrophysical
Journal, 143, 379-399.
Krolak, A. (1986). "Towards a Proof of the Cosmic Censorship Hypothesis," Classical
and Quantum Gravity, 3, 267-280.
. (1987a). "Towards a Proof of the Cosmic Censorship Hypothesis in Cosmo-
logical Space-Times," Journal of Mathematical Physics, 28, 138-141.
. (1987b). "Strong Cosmic Censorship and the Strong Curvature Singularities,"
Journal of Mathematical Physics, 28, 2685-2687.
Kruskal, M. D. (1960). "Maximal Extension of Schwarzschild Metric," Physical
Review, 119, 1743-174-5.
Kuroda, Y. (1984). "Naked Singularities in the Vaidya Space-Time," Progress of
Theoretical Physics, 72, 63-72.
Lake, K. (1988). "Comment on 'Naked Singularities in Self-Similar Spherical
Gravitational Collapse'," Physical Review Letters, 60, 241.
, and Hellaby, C. (1981). "Collapse of Radiating Fluid Spheres," Physical
Review D, 24, 3019-3022.
Lanczos, C. (1922a). "Ein Vereinfaschendes Koordinatensystem fur die Einsteinschen
Gravitationsgleichungen," Physikalische ^eitschrift, 23, 537-539.
. (1922b). "Bemerkung zur de Sitterschen Welt," Physikalische ^eitschrift, 23,
539-543.
Landau, L. (1987). "On the Violation of Bell's Inequalities in O_uantum Theory,"
Physics Letters A, 120, 54-56.
238 References

, and Lifshitz, E. M. (1962). The Classical Theory of Fields. 2nd English ed.
Reading, MA: Addision-Wesley.
Lemaitre, G. (1932). "L'Univers en Expansion," Publication du Laboratoire d'Astronomie
et de Geodesie de I'Universite de Louvain, 9, 171—205. Also in Societe Scientifique de
Bruxelles. Annales A, 53, 51-85.
Lewis, D. (1973). Counter/"actuals. Cambridge, MA: Harvard University Press.
. (1975). "Causation," in E. Sosa (ed), Causation and Conditionals. Oxford:
Oxford University Press, 180-191.
. (1986). "The Paradoxes of Time Travel," in Philosophical Papers, Volume 2.
Oxford: Oxford University Press, 67-80.
Lichnerowicz, A. (1939). Sur Certains Problemes Globaux Relatifs au Systeme des Equations
d''Einstein. Paris: Hermann.
Lifshitz, E. M., and Khalatnikov, I. M. (1963). "Investigations in Relativistic
Cosmology," Advances in Physics, 12, 185-249.
Lightman, A., and Gingerich, O. (1991). "When Do Anomalies Begin?" Science, 255,
690-695.
Linde, A. D. (1980). "Gauge Theories, Time-Dependence of the Gravitational
Constant and Antigravity in the Early Universe," Physics Letters B, 93, 394—396.
. (1982). "A New Inflationary Scenario: A Possible Solution of the Horizon,
Flatness, Homogeneity, Isotropy, and Primordial Monopole Problems," Physics
Letters B, 108, 389-393.
. (1984). "The Inflationary Universe," Reports on Progress in Physics, 47,
925-986.
Lossev, A., and Novikov, I. D. (1992). "The Jinn of the Time Machine: Non-Trivial
Self-Consistent Solutions," Classical and Quantum Gravity, 9, 2309—2321.
Lowe, D. L. (1993). "Semiclassical Approach to Black Hole Evaporation," Physical
Review D, 47, 2446-2453.
MacBeath, M. (1982). "Who Was Dr. Who's Father?" Synthese, 51, 397-430.
MacCallum, M. A. H. (1971). "Mixmaster Universe Problem," Nature, 230, 112-113.
. (1979). "Anisotropic and Inhomogeneous Relativistic Cosmologies," in S.
W. Hawking and W. Israel (eds), General Relativity: An Einstein Centenary Survey.
Cambridge: Cambridge University Press, 533—580.
MacMillan, D. R. (1961). "Cartesian Products on Contractible Open Manifolds,"
Bulletin of the American Mathematical Society, 67, 51-514.
Malament, D. (1977). "Observationally Indistinguishable Space-Times," in J. Ear-
man, C. Glymour, and J. Stachel (eds), Foundations of Space-Time Theories,
Minnesota Studies in the Philosophy of Science, Volume VIII. Minneapolis: Univer-
sity of Minnesota Press, 61—80.
. (1985). "Minimal Acceleration Requirements for 'Time Travel' in Godel
Space-Time," Journal of Mathematical Physics, 26, 774—777.
. (1987). "A Note about Closed Timelike Curves in Godel Space-Time,"
Journal of Mathematical Physics, 28, 2427-2430.
— . (1988). Private communications.
. (1995). "Introductory Note for 1949b," in S. Feferman et al. (eds), Kurt Godel,
Collected Works, Volume III. New York: Oxford University Press, 261-269.
Maudlin, T. (1990). "Time-Travel and Topology," in A. Fine, M. Forbes, and L.
Wessels (eds), PSA 1990, Vol. 1. East Lansing, MI: Philosophy of Science
Association, 303-315.
Max.enko, G. F., Unruh, W. G., and Wald, R. M. (1985). "Does a Phase Transition
 References 239

in the Early Universe Produce the Conditions Needed for Inflation?" Physical
Review D, 31,273-282.
Mellor, D. H. (1981). "Prediction, Time Travel and Backward Causation," in Real
Time. Cambridge: Cambridge University Press, 160—187.
Menotti, P., and Seminara, D. (1993). "Closed Timelike Curves and the Energy
Condition in 2 + 1 Dimensional Gravity," Physics Letters B, 301, 25-28.
Michalski, H., and Wainwright, J. (1975). "Killing Vector Fields and the Einstein-
Maxwell Field Equations in General Relativity," General Relativity and Gravita-
tion, 6, 289-318.
Mikheeva, E. V., and Novikov, I. D. (1992). "Inelastic Billiard Ball in a Spacetime
with a Time Machine," preprint.
Mill, J. S. (1843). A System of Logic, Ratiocinative and Inductive. London: John W. Parker.
Misner, C. (1963). "The Flatter Regions of Newman, Unti, and Tamburino's
Generalized Schwarzschild Space," Journal of Mathematical Physics, 4, 924—937.
. (1967). "Taub-NUT Space as a Counterexample to Almost Anything," in
J. Ehlers (ed), Relativity Theory and Astrophysics: I. Relativity and Cosmology, Lectures
in Applied Mathematics, Volume VIII. Providence: American Mathematical
Society, 160-169.
. (1968). "The Isotropy of the Universe," Astrophysical Journal, 151, 431-457.
. (1969). "Mixmaster Universe," Physical Review Letters, 22, 1071-1074.
, Thorne, K. S., and Wheeler, J. A. (1973). Gravitation. San Francisco: W. H.
Freeman.
Moncrief, V. (1981). "Infinite-Dimensional Family of Vacuum Cosmological Models
with Taub-NUT (Newman-Unti-Tamburino)-type Extensions," Physical
Review D, 23, 312-315.
, and Isenberg, J. (1983). "Symmetries of Cosmological Horizons," Communica-
tions in Mathematical Physics, 89, 387—413.
Morris, M. S., Thorne, K. S., and Yurtsever, U. (1988). "Wormholes, Time Machines,
and the Weak Energy Condition," Physical Review Letters, 61, 1446-1449.
Newman, R. P. A. C. (1984a). "A Theorem of Cosmic Censorship: A Necessary and
Sufficient Condition for Future Asymptotic Predictability," General Relativity
and Gravitation, 16, 175-192.
. (1984b). "Cosmic Censorship and Conformal Transformations," General
Relativity and Gravitation, 16, 943-953.
. (1986). "Cosmic Censorship and the Strengths of Singularities," in P. G.
Bergmann and V. de Sabbata (eds), Topological Properties and Global Structure of
Space-Time. New York: Plenum Press, 153-168.
(1989). "Black Holes Without Singularities," General Relativity and Gravitation,
21, 981-995.
, and Clarke, C. J. S. (1987). "An R 4 Spacetime with a Cauchy Surface Which
is Not K3," Classical and Quantum Gravity, 4, 53-60.
Norton, J. D. (1987). "Einstein, the Hole Argument and the Reality of Space," in J.
Forge (ed), Measurement, Realism, and Objectivity. Dordrecht: D. Reidel, 153—188.
Novikov, I. D. (1989). "An Analysis of the Operation of a Time Machine," Soviet
Journal of Experimental and Theoretical Physics, 68, 439—443.
. (1992). "Time Machine and Self-Consistent Evolution in Problems with
Self-Interaction," Physical Review D, 45, 1989-1994.
Oppenhcimer, J. R., and Snyder, H. (1939). "On Continued Gravitational Contrac-
tion," Physical Review, 56, 455-459.
240 References

, and Volkoff, G. M. (1939). "On Massive Neutron Cores," Physical Review,

55, 374-381.
Ori, A. (1991a). "Rapidly Moving Cosmic Strings and Chronology Protection,"
Physical Review D, 44, 2214-2215.
. (1991b). "Inner Structure of a Charged Black Hole: An Exact Mass-Inflation
Solution," Physical Review Letters, 67, 789-792.
. (1993) "Must Time-Machine Construction Violate the Weak Energy Con-
dition?" Physical Review Letters, 71, 2517-2520.
, and Piran, T. (1987). "Naked Singularities in Self-Similar Spherical Gravita-
tional Collapse," Physical Review Letters, 59, 2137-214-0.
, and Piran, T. (1988). "Self-Similar Spherical Gravitational Collapse and the
Cosmic Censorship Hypothesis," General Relativity and Gravitation, 20, 7-13.
, and Piran, T. (1990). "Naked Singularities and Other Features of Self-Similar
General-Relativistic Gravitational Collapse," Physical Review D, 42, 1068-
1090.
, and Soen, Y. (1994). "Causality Violation and the Weak Energy Condition,"
Physical Review D, 49, 3990-3997.
Ozsvath, I. (1967). "Homogeneous Lichnerowicz Universes," Journal of Mathematical
Physics, 8, 326-344.
Padmanabhan, T., and Seshadri, T. R. (1987). "Horizon Problem and Inflation,"
Journal of Astrosphysics and Astronomy, 8, 275-280.
, and Seshadri, T. R. (1988). "Does Inflation Solve the Horizon Problem?"
Classical and Quantum Gravity, 5, 221-224.
Pais, A. (1982). 'Subtle is the Lord. . .' The Science and Life of Albert Einstein. Oxford:
Clarendon Press.
Papapetrou, A. (1985). "Formation of a Singularity and Causality," in N. Dadhich,
J. Krishna Rao, J. V. Narlikar, and C. V. Vishveshwara (eds), A Random Walk
in Relativity and Cosmology. New York: John Wiley, 184-191.
, and Hamoui, A. (1967). "Surfaces Caustiques Degenerees dans la Solution
de Tolman. La Singularite Physique en Relativite Generale," Annales de Institut
Henri Poincare A, 6, 343-364.
Parker, L., and Fulling, S. A. (1973). "Quantized Matter Fields and the Avoidance
of Singularities in General Relativity," Physical Review D, 7, 2357-2374.
Patzelt, H. (1990). "On Horizons in Homogeneous Isotropic Universes," Classical and
Quantum Gravity, 7, 2081-2087.
Penrose, R. (1964). "Conformal Treatment of Infinity," in C. De Witt and B. De
Witt (eds), Relativity, Groups and Topology. New York: Gordon and Breach,
565-584.
. (1965). "A Remarkable Property of Plane Waves in General Relativity,"
Reviews of Modern Physics, 37, 215-220.
. (1969). "Gravitational Collapse: The Role of General Relativity," Revisita
del Nuovo Cimento, Serie I, 1, Numero Speciale, 252-276.
. (1973). "Naked Singularities," Annals of the New York Academy of Sciences, 224,
125-134.
. (1974). "Gravitational Collapse," in C. DeWitt (ed), Gravitational Radiation
and Gravitational Collapse. Dordrecht: D. Reidel, 82-91.
. (1977). "Space-Time Singularities," in R. RufHni (ed), Proceedings of the First
Marcel Grossmann Meeting on General Relativity. Amsterdam: North-Holland,
173-181.
 References 241

. (1978). "Singularities of Space-Time," in N. R. Lebovitz, W. H. Reid., and

P. O. Vandervoort (eds), Theoretical Principles in Astrophysics and Relativity.
Chicago: University of Chicago Press, 217-243.
. (1979). "Singularities and Time-Asymmetry," in S. W. Hawking and W.
Israel (eds), General Relativity: An Einstein Centenary Survey. Cambridge: Cam-
bridge University Press, 581-638.
. (1986). Book review of G. W. Gibbons, S. W. Hawking, and S. T. C. Siklos
(eds), The Very Nearly Universe, in Observatory, 106, 20-21.
. (1988). "Fundamental Asymmetry in Physical Laws," Proceedings of Symposia
in Pure Mathematics, 48, 317-328.
. (1989a). The Emperor's New Mind. Oxford: Oxford University Press.
. (1989b). "Difficulties with Inflationary Cosmology," Annals of the New York
Academy of Sciences, 271, 249-264.
Pfarr, J. (1981). "Time Travel and GodePs Space," General Relativity and Gravitation,
13, 1073-1091.
Pitowsky, I. (1990). "The Physical Church Thesis and Physical Computational
Complexity," lyyun, 39, 81-99.
Politzer, H. D. (1992). "Simple Quantum Systems in Spacetimes with Closed Timelike
Curves," Physical Review D, 46, 4470-4476.
. (1994). "Path Integrals, Density Matrices, and Information Flow with Closed
Timelike Curves," Physical Review D, 49, 3981-3989.
Pollock, M. D. (1981). "On the Solution to the Cosmological Horizon Problem
Proposed by Zee," Physical Review D, 24, 1045-1048.
Quinn, P. (1993). "Creation, Conservation, and the Big Bang," in J. Earman, A. I.
Janis, G. Massey, and N. Rescher (eds), Philosophical Problems of the Internal and
External Worlds: Essays Concerning the Philosophy of Adolf Grunbaum. Pittsburgh:
University of Pittsburgh Press, 589-612.
Raine, D. J. (1981). "Mach's Principle and Space-Time Structure," Reports on Progress
in Physics, 44, 1151-1195.
Ramsey, F. P. (1928-29). "Law and Causality," in D. H. Mellor (ed), Foundations:
Essays in Philosophy, Logic, Mathematics, and Economics (1978). Atlantic Highlands:
Humanities Press, 128-151.
Raychaudhuri, A. K., and Modak, B. (1988). "Cosmological Inflation with Arbitrary
Initial Conditions," Classical and Quantum Gravity, 5, 225-232.
Rees, M. J. (1972). "Origin of the Cosmic Microwave Background Radiation in a
Chaotic Universe," Physical Review Letters, 28, 1669-1671.
Reichenbach, H. (1958). The Philosophy of Space and Time. New York: Dover.
. (1971). The Direction of Time. Berkeley: University of California Press.
Rendall, A. D. (1992a). "The Initial Value Problem for a Class of General Relativistic
Fluid Bodies," Journal of Mathematical Physics, 33, 1047-1053.
. (1992b). "Cosmic Censorship and the Vlasov Equation," Classical and
Quantum Gravity, 9, L99-L104.
Rindler, W. (1956). "Visual Horizons in World-Models," Monthly Notices of the Royal
Astronomical Society, 116, 662-677.
Ryan, M. P., and Shepley, L. C. (1975). Homogeneous Relativistic Cosmologies. Princeton:
Princeton University Press.
Salmon, W. (1984). Scientific Explanation and the Causal Structure of the World. Princeton:
Princeton University Press.
242 References

Sato, H. (1980). "Horizon of the Universe and the Broken-Symmetric Theory of

Gravity," Progress of Theoretical Physics, 64, 1498-1499.
Savitt, S. (1994). "The Replacement of Time," Australasian Journal of Philosophy, 72,
463-474.
Schilpp, P. A. k (ed) (1949). Albert Einstein: Philosopher-Scientist. Two volumes. New
York: Harper and Row.
Schmidt, B. (1971). "A New Definition of Singular Points in General Relativity,"
General Relativity and Gravitation, 1, 269-280.
Schoen, R., and Yau, S.-T. (1983). "The Existence of a Black Hole due to
Condensation of Matter," Communications in Mathematical Physics, 90, 575—
579.
Schwarzschild, K. (1916). "Uber das Gravitationfeld eines Massenpunktes nach der
Einsteinschen Theorie," Kb'niglich Preussische Akademie der Wissenschaften (Berlin).
Sitzungsbenchte, 189-196.
Sciama. D. W. (1963). "Retarded Potentials and the Expansion of the Universe,"
Proceedings of the Royal Society (London), A, 273, 484-495.
Scott, S. M. (1992). "When Is a Pseudo-Ricmannian Manifold Non-Singular?" in
Z. Perjes (ed), Relativity Today. Commach, NY: Nova Science Publishers,
165-194.
, and Szekeres, P. (1986a). "The Curzon Singularity. I: Spatial Sections,"
General Relativity and Gravitation, 18, 557—570.
, and Szekeres, P. (1986b). "The Curzon Singularity. II. Global Picture,"
General Relativity and Gravitation, 18, 571-583.
Seifert, H. J. (1979). "Naked Singularities and Cosmic Censorship: Comment on the
Current Situation," General Relativity and Gravitation, 10, 1065-1067.
. (1983). "Black-Holes, Singularities, and Topology," in E. Schmutzer (ed),
Proceedings of the 9th International Conference on General Relativity and Gravitation.
Cambridge: Cambridge University Press, 133—147.
Shapiro, S. L., and Teukolsky, S. A. (1991a). "Formation of Naked Singularities: The
Violation of Cosmic Censorship," Physical Review Letters, 66, 994-997.
, and Teukolsky, S. A. (1991b). "Black Holes, Naked Singularities and Cosmic
Censorship," American Scientist, 79, 330-343.
Siklos, S. T. C. (1979). "Singularities and Invariants," General Relativity and Gravitation,
10, 1003-1004.
. (1981). "Nonscalar Singularities in Spatially Homogeneous Cosmologies,"
General Relativity and Gravitation, 13, 433—441.
Silberstein, L. (1936). "Two-Centers Solution of the Gravitational Equations, and
the Need for a Reformed Theory of Matter," Physical Review, 49, 268-
270.
Smith, J. W. (1986). "Time Travel and Backward Causation," in Reason, Science and
Paradox. London: Groom Helm, 49—58.
Smoller, J. A. and Wasserman, A. G. (1993). "Existence of Infintely Many Smooth,
Static, Global Solutions of the Einstein/Yang-Mills Equations," Communications
in Mathematical Physics, 151, 303-325.
, Wasserman, A. G., Yau, S.-T., and McLeod, J.B. (1991). "Smooth Static
Solutions of the Einstein/Yang—Mills Equations," Communications in Mathematical
Physics, 143, 115-147.
Sober, E. (1988). "The Principle of Common Cause," inj. H. Fetzer (ed), Probability
and Causality. Dordrecht: D. Reidcl, 211-228.
 References 243

, and Barrett, M. (1992). "Conjunctive Forks and Temporally Asymmetric

Inference," Australasian Journal of Philosophy, 70, 1~23.
Stachel, J. (1968). "Structure of the Curzon Metric," Physics Letters A, 27, 60-61.
. (1979). "The Genesis of General Relativity," in H. Nelkowski, A. Hermann,
H. Posner, R. Schrader, and R. Seiler (eds), Einstein Symposium Berlin, Lecture
Notes in Physics, Volume 100. Berlin: Springer-Verlag, 428-442.
. (1986). "Einstein and the Quantum: Fifty Years of Struggle," in R. G.
Colodny (ed), From Quarks to Quasars. Pittsburgh: University of Pittsburgh Press,
349-385.
. (1993). "Lanczos' Early Contributions to General Relativity," preprint.
Stein, H. (1970). "On the Paradoxical Time-Structures of Godel," Philosophy of Science,
37, 589-601.
. (1990). "Introductory Note to 1949a," in S. Feferman, J. W. Dawson, Jr.,
S. C. Kleene, G. H. Moore, R. M. Solovay, and J. van Heijenoort (eds),
Kurt Godel, Collected Works, Volume II. New York: Oxford University Press,
199-201.
. (1995). "Introductory Note to 1946/9," in S. Feferman et al. (eds), Kurt Godel,
Collected Works, Volume III. New York: Oxford University Press, 202-229.
Steinmuller, B., King, A. R., and Lasota, J. P. (1975). "Radiating Bodies and Naked
Singularities," Physics Letters A, 51, 191-192.
Stewart, J. M. (1968). "Neutrino Viscosity in Cosmological Models," Astrophysical
Letters, 2, 133-135.
Summers, S. J., and Werner, R. (1987a). "Bell's Inequalities and Quantum Field
Theory. I. General Setting," Journal of Mathematical Physics, 28, 2440-2447.
, and Werner, R. (1987b). "Bell's Inequalities and Quantum Field Theory.
II. Bell's Inequalities are Maximally Violated in the Vacuum," Journal of
Mathematical Physics, 28, 2448-2456.
Susmann, R. A. (1988). "On Spherically Symmetric Shear-Free Perfect Fluid Con-
figurations (Neutral and Charged). II. Equation of State and Singularities,"
Journal of Mathematical Physics, 29, 945-970.
Swinburne, R. (1968). Space and Time. London: Macmillan.
Synge, J. L. (1950). "The Gravitational Field of a Particle," Proceedings of the Royal
Irish Academy A, 53, 83-114.
. (1960). Relativity: The General Theory. Amsterdam: North-Holland.
Szekeres, G. (1960). "On the Singularities of a Riemannian Manifold," Publicationes
Mathematicae, Debrecen, 7, 285-301.
Thorn, P. (1975). "Time-Travel and Non-Fatal Suicide," Philosophical Studies, 27,
211-216.
Thomson, J. F. (1954-55). "Tasks and Super-Tasks," Analysis, XV, 1-13.
't Hooft, G. (1992). "Causality in (2 + 1)-dimensional Gravity," Classical and Quantum
Gravity, 9, 1335-1348.
Thorne, K. S. (1991). "Do The Laws of Physics Permit Closed Timelike Curves?"
Annals of the New York Academy of Sciences, 631, 182—193.
. (1994). Black Holes and Time Warps: Einstein's Outrageous Legacy. New York:
W. W. Norton.
Tipler, F. J. (1974). "Rotating Cylinders and the Possibility of Global Causality
Violation," Physical Review D, 9, 2203-2206.
. (1976). "Causality Violation in Asymptotically Flat Space-Times," Physical
Review Letters, 37, 879-882.
244 References

. (1977a). "Singularities and Causality Violation," Annals of Physics, 108,

1-36.
. (1977b). "Singularities in Conformally Flat Space-Times," Physics Letters A,
64, 8-10.
. (1978). "Energy Conditions and Spacetime Singularities," Physical Review D,
17, 2521-2528.
. (1979). "What Is a Black Hole?" General Relativity and Gravitation, 10,
1063-1067.
. (1980). "General Relativity and the Eternal Return," in F. J. Tipler (ed),
Essays in General Relativity. New York: Academic Press, 21-37.
. (1985). "Note on Cosmic Censorship," General Relativity and Gravitation, 17,
499-507.
, Clarke, C. J. S., and Ellis, G. F. R. (1980). "Singularities and Horizons—A
Review Article," in A. Held (ed), General Relativity and Gravitation, Volume 2.
New York: Plenum Press, 97-206.
Tolman, R. C. (1931a). " On the Problem of the Entropy of the Universe as a Whole,"
Physical Review, 37, 1639-1660.
. (1931b). "On the Theoretical Requirements for a Periodic Behaviour of the
Universe," Physical Review, 38, 1758-1771.
, and Ward, M. (1932). "On the Behavior of Non-Static Models of the
Universe when the Cosmological Term Is Omitted," Physical Review, 39,
835-843.
Torretti, R. (1979). "Mathematical Theories and Philosophical Insights in Cosmol-
ogy," in H. Nelkowski, A. Hermann, H. Posner, R. Schrader, and R. Seiler
(eds), Einstein Symposium Berlin, Lecture Notes in Physics, Volume 100. Berlin:
Springer-Verlag, 320-335.
. (1983). Relativity and Geometry. New York: Pergamon.
Turner, M. S. (1987). "Inflation in the Universe, Circa 1986," in M. A. H.
MacCallum (ed), General Relativity and Gravitation: Proceedings of the llth Inter-
national Conference on General Relativity and Gravitation. Cambridge: Cambridge
University Press, 223-246.
van Bendegam, J. P. (1994). "Ross's Paradox is an Impossible Super-Task," British
Journal for the Philosophy of Science. Forthcoming,
van Fraassen, B. C. (1980). The Scientific Image. Oxford: Clarendon Press.
. (1989). Laws and Symmetry. Oxford: Clarendon Press.
van Stockum, W. J. (1937). "The Gravitational Field of a Distribution of Particles
Rotating about an Axis of Symmetry," Proceedings of the Royal Society of Edinburgh,
57, 135-154.
Vickers, J. A. G. (1987). "Generalized Cosmic Strings," Classical and Quantum Gravity,
4, 1-9.
Visser, M. (1993). "From Wormhole to Time Machine: Remarks on Hawking's
Chronology Protection Conjecture," Physical Review D, 47, 554-565.
. (1994). "Van Vleck Determinants: Traversable Wormhole Spacetimes,"
Physical Review D, 49, 3963-3980.
Wald, R. M. (1973). "On Perturbations of a Kerr Black Hole," Journal of Mathematical
Physics, 14, 1453-1461.
. (1980). "Dynamics in Nonglobally Hyperbolic, Static Space-Times," Physical
Review, 21, 2802-2805.
. (1984a). General Relativity. Chicago: University of Chicago Press.
 References 245

. (1984b). "Black Holes, Singularities and Predictability," in S. M. Christenson

(ed), Quantum Theory of Gravity. Bristol: Adam Hilger, 160-168.
. (1986). "Inflation and Phase Transitions," in E. W. Kolb, M. S. Turner, D.
Lindley, K. Olive, and D. Seckel (eds), Inner Space/Outer Space. Chicago:
University of Chicago Press, 341-344.
. (1992). "Correlations Beyond the Cosmological Horizon," to appear in the
proceedings of the workshop "Origin of Structure in the Universe," Point
d'Oyre, Belgium, 1992.
, and Iyer, V. (1991). "Trapped Surfaces in the Schwarzschild Geometry and
Cosmic Censorship," Physical Review D, 44, 3719-3722.
, and Yurtsever, U. (1991). "General Proof of the Averaged Null Energy
Condition for a Massless Scalar Field in Two-Dimensional Curved Spacetime,"
Physical Review D, 44, 403-416.
Weir, S. (1988). "Closed Time and Causal Loops: A Defence Against Mellor,"
Analysis, 48, 203-209.
Wells, H. G. (1968). Time Machine. New York: Bantam Books.
Weyl, H. (1917). "Zur Gravitationstheorie," Annalen der Physik, 54, 117-145.
. (1919). Raum-^eit-Materie. 3rd ed. Berlin: Julius Springer.
Wheeler, J. A., and Feynman, R. P. (1949). "Classical Electrodynamics in Terms of
Direct Interparticle Action," Reviews of Modern Physics, 21, 425-433.
Will, C. M. (1993). Theory and Experiment in Gravitational Physics. Revised edition.
Cambridge: Cambridge University Press.
Yodzis, P. Seifert, H.-J., and Miiller zum Hagen, H. (1973). "On the Occurrence of
Naked Singularities in General Relativity," Communications in Mathematical
Physics, 34, 135-148.
Yourgrau, P. (1991). The Disappearance of Time. Cambridge: Cambridge University
Press.
Yurtsever, U. (1990). "Test Fields on Compact Space-Times," Journal of Mathematical
Physics, 31, 3064-3078.
. (1991). "Classical and Quantum Instability of Compact Cauchy Horizons
in Two Dimensions," Classical and Quantum Gravity, 8, 1127-1139.
Zee, A. (1980). "Horizon Problem and the Broken-Symmetric Theory of Gravity,"
Physical Review Letters, 44, 703-706.
Zemach, E. M. (1968). "Many Times," Analysis, 28, 145-151.
This page intentionally left blank
 Index

(Italicized page numbers denote definitions]

Abbott, P. 158 n.25 Friedmann-Robinson- Walker

affine parameterization of a geodesic, 60 spacetime
n.9; 120 asymptotically delayed, 30, 36
generalized affine length (g.a.l.), 35, 36 big crunch, 20, 46, 89, 204, 205, 210,
generalized affine parameter (g.a.p.), 224. See also
35; 37 Friedmann—Robinson—Walker
Akdeniz, K. G., 148 spacetime
Albrecht, A, 152 black holes, 20, 55, 64, 67, 70, 73, 75,
Allis, V., 122 82, 85-8, 101 n.25, 131,
Anderson, J. L., 18 168-69, 222
Anderson, P., 57, 148 evaporation of, 58, 90-9
anti-de Sitter spacetime, 69, 75, 76-7, interior of, 100 n.6
81, 113-16 mini-, 92
Arnowitt-Deser-Misner (ADM) mass, "no hair" theorems for, 67
26n.22, 67, 102 n.34, 211 Blau, S. K., 158 n.25
Arntzenius, F., 136, 137, 158 n.13 Borner, G., 143, 146, 158 n.17
Augustine, Saint 219 n.2 Brandenburger, R., 223, 227 n.3
Brans, C. H., 50
Brans-Dicke theory (BDT), 149
6-boundary. See singularities, Brown, B., 200 n.2, 201 n . l l
i-boundary approach Budic, R., 79
Barrett, M., 158 n.13
Barrow, J. D., 123 n.8, 135, 158 n.25 Carroll, J. W., 177
Bayesianism, 145, 158 n.21 Carroll, S. M., 169, 200 n.l
Bell, J. S., 222 carry along (of geometric objects), 61 n.13
Bell's inequalities, 137, 157 n.12, 226 Carter, B., 164, 166
Benacerraf, P., 103-4 Cauchy horizon, 82, 85, 96, 171
Bergmann, P., 11-12, 26 n.21, 222 future and past, 66, 99 n.3, 114
Bianchi identity, 47-8, 206 Cauchy sequence, 33
big bang, 10, 14, 15, 16, 19, 20, 21, 57, Cauchy surface, 44-5, 53, 68-71,
58, 65, 67, 68-9, 93, 124, 133, 73-80, 94, 95, 97, 102 n.32,
134, 143, 144-50, 152, 154, 107-8, 110, 117, 125, 128,
155-156, 157 n. 11, 204-209, 219 138-9, 166, 212. See also
n.5, 222, 224. See also globally hyperbolic spacetime

247
248 Index

Cauchy surface (cont.) completely stable, 191

future and past, 68 generically stable, 191
maximal, 74 quantum instability, 192
partial, 68, 73, 74, 82, 85, 89-90, 92, chronology protection, 190-2, 226
96, lOOn.10, 100 n . l l , 168, 189, Chrusciel, P. T., 49-50
190, 191, 201 n.8 Church's theorem, 23-4
causal past and future of a point p, 44 Church's thesis (or proposal), 120, 123
causal simplicity condition, 767 n.12
causality conditions circular (or cyclic or closed) time, 24,
hierarchy of, 160, 164-6, 169. See also 203, 213-19
chronology condition; future Clarke, C. J. S., 25 n.3, 32, 37-8, 46,
distinguishing and past 50, 56-7, 60 n.2, 62 n.18, 84,
distinguishing condition; simple 169, 187, 200 n.l
causality; strong causality; Clauser, J. F., 136
temporal orientability closed or almost closed causal curves,
causally benign spacetimes, 180. See also 42, 55, 82, 100 n.8, 166
wave equation closed time structure, 215, 216, 217, 218,
causally non-benign spacetimes, 181 219
causally precedes, 165 closed timelike curves (CTCs), 21, 23,
causally regular (neighborhoods in 26 n.31, 42, 51, 54-5, 58, 82, 96,
spacetime), 180, 181. See also 161, 166, 167-94, 197, 198, 201
wave equation n.12, 202 n.30, 214, 216-19,
causally stable spacetimes, 100 n.8, 113, 225-26, 227 n.7. See also time
166-1, 216 travel
causation. See also principle of common Collins, C. B., 144, 145, 147, 148
cause Collins-Hawking theorem, 144, 145-6
backward, 160, 162-4 compatibilism. See free will
counterfactual analysis of, 127 completeness conditions (for spacetimes)
principle that every event has a i-completeness, 36-7, 38, 40-1, 42,
cause, 209 44, 46, 49, 54, 57, 62 n. 17, 62 n.22
principle that whatever begins to exist bounded acceleration completeness,
has a cause, 208 35, 36
Ghandrasekar, S., 78, 114 geodesic completeness, 30—1, 33, 36,
chaotic cosmology, 146, 148 40, 41, 48, 50-1, 53, 77, 92, 101
Chapman, T., 164, 200 n.2 n.20, 212
Charlton, N., 169, 200 n.l conformally flat spacetimes, 88, 134
Chihara, C. S., 108-9 conservation law, 13, 47, 132, 178, 180,
Choquet-Bruhat, Y., 45, 62 n.27, 63 184, 185 206
n.29, 79, 97 consistency constraints, 161, 173, 174—6,
Christoffel symbol, 60 n.7 179, 181-3, 186-8, 193-4, 201
chronological past and future of a point n.12, 226. See also time travel
P,42 constraint equations, 48, 79, 100 n.16,
chronologically precedes, 165 125-6
chronology condition, 54-5, 165 Cornish, N. J., 227 n.3
violations of, 167-80, 188, 190, 193, 201 cosmic censorship hypothesis (CCH),
n.l9, 226 11, 23, 28, 44-6, 56, 57, 58, 59,
chronology horizon, 189-93 64-99, 101 n.25, 101 n.30, 102
compactly causally generated, 190 n.34, 115, 119, 190, 194, 203,
compactly generated, 189, 192 211, 219, 225, 226
 Index 249

strong cosmic censorship (SCC), 46, domains of dependency (of an achronal

68-9, 73, 105, 110, 115, 211 spacelike surface)
weak cosmic censorship (WCC), 69, future and past, 66
70-1, 73 Dominici, D., 149
cosmic microwave background radiation Droste coordinates, 5-6, 31, 42, 100
(CMBR), 135, 142-9, 153-4, 156 n.l4, 206. See also Schwarzschild
cosmic strings, 38, 169 solution
cosmological constant, 5, 7, 14, 15, 16, Droste, J., 5
22, 25n.4, 51, 81, 88, 114, 134, Dwivedi, I. H., 87
150-1, 211. See also Einstein field dynamical (or evolution) equations, 48,
equations. 125-6
Coulomb field, 127
covariant derivative operator, 25 n.19, Eardley, D. M., 64, 83
60n.7 Earman, J., 25 n.6, 57, 99 n.l, 111, 122
Craig, W. L., 208, 210 n.l, 122 n.2, 123 n.10, 123 n.ll,
curvature conjecture, 56—7 123 n.12, 158 n.21, 178, 201
curvature scalar, 25 n.4, 29, 30, 100 n.17, 201 n.20, 227 n.6
n.16. See also Kretschmann Echeverria, F., 184, 185, 187, 200 n.l,
curvature scalar 202 n.26
Curzon solution Eddington, A. S., 7-9
monopolar, 211 Eddington-Finkelstein coordinates, 8-9,
(-Silberstein) bipolar, 17-19, 38 25 n.14
Cutler, C., 169, 200 n.l Ehlers, J., 151
Einstein, A., 4-22, 24, 24-5 n.3, 25 n.4,
de Sitter spacetime, 7, 8, 10, 20-1, 25 25 n.6, 25 n.12, 25 n.15, 26 n.21,
n.12, 131-3, 151, 153 26 n.22, 26 n.25-9, 26 n.31, 27,
de Sitter, W., 7, 10, 20 29-30, 59, 65, 158 n.16, 200,
De, U. K., 168 204, 205, 222
decoupling time, 134-5, 146, 149, 150, Einstein field equations (EFE), 4, 5, 7,
151, 153, 155 8, 12-18, 20, 21, 23, 25 n.4, 25
Deser, S., 169, 200 n.l n.19, 26 n.22, 27, 29, 38,40,46,
determinism, 58, 59, 62 n.35, 69, 77, 82, 47-8, 49-51, 55-6, 57, 59, 73,
83-4, 93-4, 97-8, 99, 99 n.l, 99 79, 94, 97, 99, 125, 132, 134,
n.2, 101 n.24, 136, 166, 167, 171, 138, 144, 150, 152, 189, 194-7,
189, 193, 225-6. See also free will 200, 201 n.6, 201 n.18, 202 n.29,
by fiat (in GTR), 98 204-6, 209, 210, 211, 212, 221
global Laplacian, 44, 171 n.24, 225. See also Bianchi
Laplacian, 19-20, 212 identity; conservation law;
local, 93, 171 constraint equations, dynamical
determinism maximal (or hole-free) equations
spacetimes, 98, 102 n.36 cosmologic member, 15. See also
development (of initial data), 78—80, cosmological constant
101 n.17, 213 vacuum field equations, 5, 7, 14, 17,
maximal, 76, 79, 80, 98, 99 25 n.4, 29, 40, 47, 48, 55, 79, 89, 95
Dieckman, N. J., 46, 125 initial value problem, 47-8, 79, 94.
diffeomorphism, 31, 61 n.13 See also development (of initial
distributions, 13, 46-7, 60 n.5, 62 n.27, data)
62 n.29, 62 n.30, 82, 99, 205-6, Einstein tensor, 25 n.4, 47, 81
219 n.4 Eisenstaedt, J., 25 n.3, 25 n.8
250 Index

Ellis, G., 25 n.3, 30, 32, 34, 36, 37, 38, extrinsic curvature. See second
40, 41, 42, 46, 49, 51, 53, 61 fundamental form
n.15, 62 n.18, 62 n.20, 77, 92,
98, 99, lOOn.6, 101 n.29, 113, false vacuum, 151. See also inflationary
114, 122 n.4, 125, 127, 132, 141, cosmology
148, 149, 150, 154, 158 n.27, 159 Fermat, P., 106
n.31, 165, 167, 191,211,215,216 Fermat's last theorem, 105-7, 116-18,
end point (of a curve), 30, 43, 44, 61 120
n.10, 62 n.24 Feynman, R. P., 187
energy conditions, 6, 57-8, 82, 84, 98, finite compactness. See metric space,
112-13, 114, 119, 145, 167, 169, finite compactness
191 Finkelstein, D., 25 n.14
dominant, 81, 144, 158 n.20 Finlay-Freundlich, E., 16-17
logical relations between types of, 101 first fundamental form, 79
n.21, 201 n.18 Fischer, A. E., 48
null, 63n.34, 191, 201 n.18 Fleming, G., 157 n.12
strong, 51, 81, 144 Fock, V., 13
weak, 57, 63 n.34, 81, 167, 201 n.18, Forster, M., 137
201 n.19 frame, 29
energy-momentum tensor, 25 n.4, 47, parallel propagation (p.p.) of, 29, 40,
51, 57, 80, 81, 89, 132, 134. See 60 n.8
also conservation law; energy framed spacetime, 32, 62 n.15
conditions freewill, 23, 171-2, 178-9
eternal recurrence, 13, 24 203, 210, 212, Friedman, J. L., 96, 169, 174, 180, 181,
213, 217, 219, 220n.l3, 220 n.15 191, 193, 200 n.l, 227 n.7
no recurrence theorems, 210-13 Friedmann, A., 15
eternal time machine spacetime, 184 Friedmann-Robinson-Walker (FRW)
event horizon conjecture (EHC), 89. See spacetime, 15, 16, 19, 36, 39, 44,
also horizon problem 46, 48, 57, 71, 131, 132, 133-5,
exotic differentiable structures, 45-6, 50 138, 148-50, 157 n.10, 158 n.26,
expansion of geodesic congruence, 51, 178, 204, 205, 206, 208, 209,
52, 63 n.32 212, 213, 220 n.13
explanation, 223. See also principle of big bang singularities in, 14, 15, 16, 71
common cause big crunch singularities in, 20, 46
anthropic, 147 Frobenius' theorem, 196
deductive-nomological, 138-9 Frolov, V. P., 169, 200 n.l
purely local, 176 Fulling, S. A., 57
robust, 146, 157 future distinguishing and past
exponential map, 62 n.19 distinguishing condition, 165
extensions of spacetimes, 6, 31—Z, 43, 46,
47, 50, 79-80, 94-7, 99, 205, 208 Gamov, G., 15
containing CTCs, 189, 192-3 Gardner, M., 200 n.4
continuity/differentiability (c/d) Gautreau, R., 18
conditions for, 46, 48, 50, 59 general theory of relativity (GTR), 3, 4,
global, 38, 50 11, 12, 14, 15, 16, 18, 21, 22-4,
local, 37, 50, 61 n.14 26 n.31, 27, 28, 44, 46, 51, 56-7,
maximal, 32, 45, 62 n.15 58, 59, 64-5, 66-7, 73, 80-1,
proper, 31 84-7, 90, 92, 94, 96, 97, 98, 125,
through singularities, 205—7, 208 134, 137-8, 141-2, 144, 147,
 Index 251

149, 157 n.3, 157 n.5, 158 n.16, 192, 193-4, 197, 226. See also
163, 200, 202 n.28, 203, 213, time travel
218-29, 222-6, 227 n.3, 227 n.4 Grommer, J., 12
GTR chauvinism, 97 Griinbaum, A., 104, 207, 208
incompleteness of, 223-6 Guth, A. H., 152, 158 n.25
initial value formulation of, 47—8
initial value problem in, 78-9 Hakim, R., 135
geodesic postulate, 12, 13, 25 n.20; half-curve, 30, 62 n.24, 101 n.31, 105
geodesies, 29, 30, 31, 51, 52, 53, 55, 60 completeness (or incompleteness) of,
n.9, 63 n.33, 100 n.7, 100-1 n.16, with respect to some parameter,
128, 128-30, 132-3. See also 30, 43
generic conditions Hamoui, A., 82
conjugate points, 53, 212 Hartle,J. B., 78, 114, 143, 148
generic conditions, 51, 55, 167 Havas, P., 25 n.3, 25 n.15, 25 n.28, 25
incomplete, 31, 34, 41, 43, 55, 62 n.29
n.24, 69, 84, 88 Hawking radiation. See black holes,
marginally outgoing, 88, 101 n.28 evaporation of
Geroch, R. P., 13, 25 n.20, 27, 32, 33, Hawking, S. W., 14, 23, 34, 36, 41, 42,
34-5, 36, 41, 45, 46-7, 56, 60 46,49, 51, 53, 54, 58, 61 n.15,
n.2, 62 n.15, 62 n.19, 70, 74, 75, 62 n.18, 62 n.20, 64, 65, 67, 77,
79, 90, 91, 97, 98, 101 ri.19, 112, 82, 91, 99, 100 n.6, 101 n.26, 110
157 n.5, 206, 216 n.29, 113, 114, 122 n.4, 132, 142,
Gingerich, O., 155 144, 147, 158 n.25, 165, 167,
global-to-local property, 173, 174 169, 189, 190, 191, 192, 200 n.l,
globally hyperbolic spacetime, 44—5, 46, 201 n.18, 215, 216
52, 57, 62 n.26, 68, 76-80, 90, Hawking—Penrose theorem, 23, 48—9,
95, 96, 101 n.32, 102 n.36, 51-6, 59, 212, 222
104-6, 110, 117, 122-3 n.4, 166. Hellaby, C., 86
See also Cauchy surface . Higgs fields, 151, 158 n.28
Glymour, C., I l l , 218 Hilbert, D., 5-6, 10, 25 n.8, 27, 32
God, 32-3, 101 n.32, 203, 207-10, 219 Hochberg, D., 147
n.5, 220 n.9 Hoffman, B., 12
Godel, K 21, 22, 160, 161, 168, 169, Hogarth, M. L., 104, 107, 113, 120, 157
194, 200, 201 n.6, 201 n.16, 202 n.5
n.27, 202 n.29, 202 n.31 hole argument, 13
Godcl spacetime, 21-2, 54, 63 n.32, 102 hoop conjecture, 101 n.25
n.36, 110, 164, 167, 168, 169, Hopf—Rinow theorem, 34
177, 181, 184, 188, 194-9, 200-1 horizon coordinate distance, 134, 151, 152
n.6, 201 n.12, 202 n.30, 210, horizon problem, 10, 21, 24, 124, 133,
213-15, 217-18. See also time 134-5, 137, 139, 142-57, 158
travel, Godelian n.17, 159 n.32, 223
Goldwirth, D. S., 193, 200 n.l horizons, 124, 130, 158 n.26. See also
Gompf, R., 62 n.26 Cauchy horizon; chronology
Gott, J. R., 169, 200 n.l horizon
Gott spacetime, 169, 192 apparent horizon, 87, 101 n.26
Gowdy spacetime, 88-9, 95 event horizon, 20-21, 87-88, 89, 101
grand unified theories (GUI's), 150 n.26. See also event horizon
grandfather paradox, 24, 161, 170 3, conjecture
174, 175, 179, 181, 183, 184-6, absolute, 70, 100 n.6
252 Index

horizons (cant.) King, A. R., 40, 86

event horizon (cant.) Klein, F., 7
future (FEH), 130 Klinkhammer, G., 200 n.l, 202 n.26
particle horizons, 20-1, 57, 124, Kodama, H., 90
128-31, 132, 133-4, 137-56, Koetsier, T., 122 n.2
177. See also horizon problem Kostelecky, A. V., 63 n.36, 223
Horne.J. F., 136 Kretschmann curvature scalar (or
Horowitz, G. T., 74, 75, 101 n.19 invariant), 9, 18-19, 32, 205, 206
Horwich, P., 200 n.2 Kriele, M, 55
Hubble constant, 15, 150 Kristian,J., 125
Huygens' principle, 140, 201 n.15 Krolak, A., 84, 88
Hiibner, P., 151 Kruskal coordinates, 43
Kruskal extension (of the Schwarzschild
ideality of time, 194-200 solution), 32, 39, 48, 70, 87, 211.
development of Godel's ideas about See also Schwarzschild solution
the, 202 n.27, 202 n.29 Kruskal, M. D., 6, 32
incompatibilism. See free will
Infeld, L., 12, 13-14 Lake, K., 85-6
inflationary cosmology (or model), 21, Lanczos, C., 7, 8, 25 n.l2
24, 124, 149-57, 158 n.25, 158 Landau, L., 51, 137
n.26, 158-9 n.29, 159 n.30, 159, large fraction condition, 139. See also
n.32 principle of common cause
internal infinity, 92, 101 n.31 Lasorta, J. P., 86
intrinsic curvature. See first fundamental laws of nature (or physical laws), 11,
form 19-20, 23, 49, 65, 97, 156-7
Isenberg, J., 88, 90, 95, 102 n.31 n.12, 161, 177, 188, 194, 198,
isometric imbedding (of spacetimes), 31, 201 n.12, 222. Ste also physical
32, 37, 61 n.15. See also possibility
extensions of spacetimes empiricist conceptions of, 178, 226,
Israel, W. H., 64, 89, 101 n.18 227 n.5
Iyer, V., 87 Mill-Ramsey-Lewis (MRL) account
of, 178-9, 182-3, 188, 194
Jacobi field, 63 n.33 supervenience of, on occurrent facts,
Janis, A. I., 101 n.31, 211 177-8, 201 n.14
Jeans, J., 196, 202 n.28 see also physical possibility
Johnson, R. A., 36 Leibniz's principles of sufficient reason
Joshi, P. S., 63 n.35, 64, 73, 74, 87, 94, and plentitude, 32-3, 49
157 n.8, 176 Lemaitre, G., 9, 10
Lewis, D., 127, 178, 200 n.2, 201 n.14
Kahn, C., 25 n.9 Liang, C., 36
Kahn, F., 25 n.9 Lie derivative, 100 n.16, 220 n.10
Kant, I, 65 Lifshitz, E. M., 51
Kephart, T. W., 147 light cone, 130, 157 n.7. See also null cone
Kerr spacetime, 67, 168 Lightman, A., 155
Kerr—Newman spacetime, 88, 169 Linard-Wiechart potential, 141
Kerszberg, P., 25 n. 10 Linde, A. D., 148, 152
Khalatnikov, I. M., 51 local-to-global property, 173, 177, 201
Killing field, 89, 90, 210, 220 n.13 n.10
Kim, S.-VV., 192, 200 n.l Loreritz gauge equation, 158 n.14
 Index 253

Lorentz, H. A., 12 Norton, J., 13, 25 n.7, 25 n.18, 122 n.l,

Lorentz metric, 60 n.3, 164 122 n.2, 123n.lO, 123 n.l 1
Lowe, D. L., 91 Novikov, I. D., 169, 174, 184, 186, 200
n.l
Novikov's piston device, 186-7
Maartens, R., 125 null cone, 75, 757 n.7. See also light cone
MacCallum, M. A. H., 135, 148 null convergence condition, 201 n.18
Mach's principle, 6, 25 n.12, 141-2, null energy condition. See energy
156, 158 n.16 conditions,
McLeod.J. B., 26 n.22, 212 null infinity, future and past, 70, 100
MacMillan, D. R., 46 n.6, 184
Malament, D., 107, 112, 121, 122 n.4,
168, 200 n.6, 202 n.27, 202 n.29,
observationally indistinguishable
220 n.22
spacetimes, 199, 200, 218
Malament-Hogarth (M-H) spacetime,
open time structure, 210, 214, 215, 218,
lOOn.13, 104, 107-19
219
Marsden, J. E., 48
Oppenheimer, J. R., 14, 86
Maudlin, T., 157 n.12, 187
Oppenheimer-Snyder-Volkoff model of
maximal spacetime, 32, 79-80
gravitational collapse, 14, 86, 89
Maxwell's equations, 125, 140-1, 172,
Ori, A., 83, 84, 169, 200 n.l, 201 n.19,
183, 211
207
Mazenko, G. F., 159 n.30
outer product (of tensors), 47, 60 n.4
metric space, 33—4:, 36, 40
Ozsvath, L, 168
Cauchy complete, 33, 34, 36, 40
finite compactness, 40
Michalski, H., 211 Padmanabhan, T., 152, 153, 155
Mikheeva, E. V., 184, 200 n.l Pais, A., 26 n.25
Mill,J. S., 178 Papapetrou, A., 82
Minkowski spacetime, 13, 28, 30, 31, paradox of gravitational collapse, 223
34, 36, 44, 45, 62 n.l5, 84, 105, parallel transport (of a vector field), 35,
128-32, 140, 161, 163, 166, 173, 40, 60 n.7, 60 n.9
180, 183, 195, 216-17 Parker, L., 57
Misner, C., 9, 38, 41, 62 n.22, 146, 148, partial order, 61 n.l5
223, 227 n.2 Patzelt, H., 153
Misner spacetime, 38, 41, 43, 76, 191. Pauli, W., 14
See also Taub-NUT spacetime Penrose conjecture. See Weyl curvature
missing points, 27, 29, 31, 33, 40-2, 44, hypothesis.
59, 62 n.23. See also singularities Penrose, R., 11, 14, 23, 28, 33, 44, 46,
mixmaster universe, 148 51, 56, 64, 65, 67, 71, 73, 75, 85,
Modak, B., 152 88, 92, 93, 95, 100 n.9, 100 n.12,
Moffat, J. W., 227 n.3 101 n.20, 114, 141, 147, 152, 158
Moncrief, V., 88, 90 n.29, 159 n.30, 211, 219
Morris, M. S., 96, 169, 191, 200 n.l perihelion motion of Mercury, 5, 25 n.6
Perry, M., 63 n.36, 223
Pfarr, J., 200 n.6
Nauenberg, M., 222 physical possibility, 161, 163, 174-6,
Nel, S. D., 125 178, 217. See also consistency
Newman, E. T., 211 constraints; laws of nature
Newman, R. P. A. C., 46, 54, 75, 84, 88 physical possibility,, 174-5, 176, 182
254 Index

physical possibility (cont.) Raychaudhuri, A. K., 152

physical possibility2, 174-5, 176, 179, Raychaudhuri equation, 52
182 recursive sets, 107, 120, 122 n.3
Piran, T., 83, 84 recursively enumerable (r.e.) sets, 107,
Pitowsky, I., 105, 106, 107, 108, 110, 122 n.3
116, 119, 123 n.12 redshift/blueshift effect, 111-13, 169,
Pitowsky spacetime, 105 197, 199-200
Planck time, 147, 149 Reeh-Schleider theorem, 137, 144
Plato machine, 103, 109-10, 117, 119. Rees, M. J., 135
See also supertasks regular points, 12, 37. See also
Poincare conjecture, 46, 62 n.26 quasi-regular points
Poincare recurrence theorem, 203, 213 Reichenbach, H., 135-6, 217-18
Pollock, M. D., 148 Reissner-Nordstrdm spacetime, 34,
positive pressure criterion, 144, 158 n.20 77-8, 85, 114, 116-17, 118, 119
prediction, deterministic, 66, 73, 128, Kendall, A. D., 88, 101 n.24
156, 157 n.5, 157 n.6 Ricci tensor, 25 n.4, 40
principle of common cause (PCC), 23, Riemann tensor, 13, 29, 47, 116
135, 136-4$, 146, 156, 158 n.13, physical components of, 29, 38
159 n.32. See also screening off Riemannian space, 33, 34, 40
condition Rindler, W., 130, 131, 132, 133
probabilistic common cause, 136 Rosen, N., 17-19
principle of self-consistency (PSC), 174. rotation (or twist) of geodesic
See also consistency constraints; congruence, 52, 63 n.32, 196, 220
time travel n.ll
Putnam, H., 103, 104 Ryan, M. P., 62 n.22

quantum field theory (QFT), 58,

136-7, 143-4, 169, 191, 192, 227 Sachs, R. K., 95, 125
n.7 Salmon, W., 136
quantum mechanics (QM), 3-4, 16, 26 Saraykar, R. V., 73, 74
n.27, 57, 63 n.36, 93, 136, 192, Sato, H., 148
193, 194, 226 Savitt, S., 202 n.31
measurement problem of, 3-4, 22, Schmidt, B., 30, 32, 36, 37, 38, 40, 57,
93-4, 157 n.12, 222, 224, 226 62 n.17, 98
quantum theory of gravity, 3, 24, 56-8, Schoen, R., 52
90, 224, 226, 227 n.7 Schreiber, G., 148, 149
semiclassical approach to, 57, 58, Schwarzschild, M., 5
90-1, 169, 192 Schwarzschild radius, 5, 6, 9, 10, 31, 32,
quasi-regular points, 37, 41. See also 86, 205, 206. See also
quasi-regular singularity; regular Schwarzschild singularities
points Schwarzschild solution, 5-6, 8, 9, 14,
Quinn, P., 209 31-2, 43, 75, 102 n.34, 206, 211.
quotient topology, 214-15, 216, 217, 219 See also Kruskal extension (of the
Schwarzschild solution)
radiation dominance condition, 150, negative mass, 75, 100 n.14, 211
151, 155, 204-7 truncated, 211
Rainc, D. J., 158 n.16 Sciama, D. W., 127, 141, 142
Ramsey, F. P., 178 Scott, S. M., 19, 42, 43, 44, 46, 62 n.23,
Randall, D., 50 211
 Index 255

screening off condition, 136. See also n.34, 203, 215, 225-6. See also
principle of common cause cosmic censorship hypothesis
second fundamental form (or extrinsic future, 78, 89, 90, 93
curvature), 48, 79, 100 n.16, 220 in the first sense (FNS,), 75, 77, 89
n.13 in the second sense (FNS 2 ), 76,
second law of thermodynamics, 14-7, 77, 88
203, 204 locally, 74, 89
Seifert, H.J., 82-3 non-scalar polynomial (or whimper),
self-similar spacetimes, 101 n.23 40.
Seshadri, T. R., 152, 153, 155 quasi-regular, 19, 38, 50. See also
Shapiro, S. L., 64, 87, 88, 94 quasi-regular points
shear of geodesic congruence, 52, 63 scalar polynomial (s.p.), 38, 84
n.32 blowup, 38-9
signature (of a metric), 60 n.3, 62 n.16, oscillatory, 59-40
164 Schwarzschild, 6, 8-10, 25 n.3, 26 n.21
Siklos, S. T. C., 39, 40, 158 n.25 "pragmatic attitude" towards, 10
Silberstein, L., 12, 17-19, 26 n.29 shell-crossing, 82
Silk,J., 135 shell-focussing, 83
simple causality condition, 165 thunderbolt, 91, 92, 101 n.30
singularities, 3-4, 5, 22-30, 33-44, small universes, 148, 149
46-59, 101-2 n.32, 105, 119, Smith, J. W., 200 n.2
203-11, 222-7. See also Smith, Q,., 208
completeness conditions; Smoller,J. A., 26 n.22, 212
Hawking-Penrose theorem; Snyder, H., 14, 86
missing points Sober, E., 136, 158 n. 13
as artifacts of idealizations, 14, 205 Sobolev space, 48, 220 n.16
^-boundary approach, 36-7, 209 Soen, Y., 200 n.l, 201 n.19
cone, 30, 32, 38, 61 n.14 Somerfeld radiation condition, 141, 177
coordinate, 8-9 spacetime, 28. See also spacetime
curvature, 37-9, 55-6, 59, 190, 224 manifold; Lorentz metric;
irremovable, 215 spacetime metric
strong, 84, 86, 87, 206 spacetime manifold, 6, 12, 41, 50, 73, 164
definitions of, 28-31, 35, 36, 41, 46, attaching singular points to, 28, 36,
6 0 n . l , 62 n.19 59, 209
Einstein's attitude towards, 5-20, 24, boundary of, 43
24 n.3, 25 n.21, 26 n.27 envelopment of, 42-3
Ellis-Schmidt classification of, 37 spacetime metric, 6, 11, 12, 13, 25 n.4,
essential (or genuine), 6, 8-9, 33, 42, 25n.5, 28, 58, 164, 210
43, 46, 48-9, 50, 57, 59, 213, continuity/differentiability (c/d) of,
219, 224 32, 46, 48-50, 60 n.3, 98-9, 205
extending spacetime metric through. regular, 6, 13, 46-7, 206
See extensions of spacetimes, spatial metric of geodesic congruence,
through singularities 63 n.33
harmless, 80 special theory of relativity (STR), 12,
infinite blueshift, 77, 85 28, 126, 142, 195-6
living with, 56, 65-7, 224-5 Stachel, J., 6, 19, 25 n.12, 26 n.24, 26
naked, 20, 28, 37, 44, 58, 59, 63 n.35, n.25, 26 n.31
65-70, 73-8, 80-7, 88, 90, 91, static spacetime, 96, 119, 210-12, 220
93, 94-6, 100 n.14, 101 n.20, 102 n.13
256 Index

stationary spacetime, 204, 210-12 Tipler, F. J., 21, 25 n.3, 26 n.21, 55, 58,
Stein, H., 200 n.6, 202 n.27 60 n.2, 64, 70, 84, 123 n.8, 167,
Steinhardt, P., 152 n.25, 158 n.25 169, 191, 212, 213, 219 n.l, 220
Steinmuller, B., 86 n.15, 220n.l6
Stewart, J. M., 91, 148 Tolman, R. C., 204-5, 206, 210
Stoeger, W., 150, 154, 158 n.27, 159 n.31 Tolman-Bondi spacetime, 83, 87
strong causality condition, 42, 44, Torretti, R., 25 n.l7, 220 n.7
89-90, 92, 100 n.15, 165-ff totally vicious spacetimes, 167, 168, 169
strong energy condition. See energy trapped surface, 51, 52, 53, 54-5, 88-9,
conditions 101 n.27
Summers, S. J., 137 Traschen,J., 13, 46, 206
super 7t machine, 103, 108-9,. 119. See Turing computability, 120
also supertasks Turner, M. S., 143, 144, 152, 158 n.25
supertasks, 23, 44, 103-20, 223 twin paradox spacetime, 184, 185, 186,
Sussman, R. A., 29, 35, 36 187, 188
Swinburne, R., 201 n.9
Synge,J. L., 60 n.l unified field theory, 16, 26 n.25
Szekeres, D., 6, 19, 27, 60 n.l, 211 Unruh, W. G., 159 n.30

Vaidya spacetime, 87
Taub-NUT spacetime, 38-40, 41, 43, van Bendegem, J. B., 122 n.2
77, 82, 89, 90, 95, 168-9, 191 van Fraassen, B. C., 136, 178
temporal orientability, 99 n.4, 165, 214 van Stockum spacetime, 21, 169
Teukolsky, S., 64, 87-8, 94, 101 n.25 van Stockum, W. J., 21, 169
Thomson, J. F., 103 Vickers, J. A. G., 38
Thomson lamp, 103, 104, 110, 119. See Volkoff, G. M., 86
also supertasks
't Hooft, G., 169, 200 n.l Wainwright, J., 211
Thorne, K. S., 9, 10, 101 n.25, 200 n.l, Wald, R. M., 29, 30, 31, 36, 53, 60 n.3,
202 n.26, 223, 227 n.2 60 n.9, 63 n.33, 63 n.34, 64, 67,
time directionality, 22. See also temporal 77, 84, 87, 88, 90, 99 n.6, 99
orientability n.l6, 102 n.34, 123 n.6, 132, 144,
time functions, 195-6, 199, 210 157 n.2, 157 n.3, 157 n.4, 159
circular, 215, 220 n.l9 n.30, 166, 191, 224, 225
global, 70, 111, 166, 197-9, 216, 220 Ward, M., 205
n.19 Wasserman, A. G., 26 n.22, 212
linear, 214, 217, 220 n.l9 wave equation, 96, 140, 173, 180-4, 190
time machines, 23, 24, 160, 161, 163, advanced representation, 142
188, 190, 192, 193, 200 n.4, 201 Kirchhoff retarded representation, 140,
n.19. See also time travel 142
time order, 22 weak derivatives (of a function), 62 n.28
time slice, 68, 168, 172, 180, 189, 201 weak energy condition. See energy
n.8, 210, 214-17, 219-20 conditions.
time travel, 22, 24, 54, 160-94, 201 n.9, Werner, R., 137
201 n.l6. See also closed timelike Weyl axially symmetric solutions, 17,
curves; grandfather paradox; 211. See also Curzon solution,
time machines (-Silberstein) bipolar
Godelian, 160-76 Wcyl curvature hypothesis, 100 n.9, 147
Wellsian, 160-2, 200 n.4 Weyl, H., 17, 211
 Index 257

Weyl tensor, 40, 62 n.21 Yodzis, P., 82

Wheeler, J., 9, 187, 223, 227 n.2 Yourgrau, P., 199
white holes, 70, 100 n.9 Yurtsever, U., 63 n.34, 169, 179, 180,
Winicour, J., 211 183, 191, 200 n.l, 201 n.21, 201
wormholes, 96, 147-8, 169, 180, 181, n.22
184-8, 191, 192. See also eternal
time machine spacetime; twin
paradox spacetime Zee, A., 148
Zemach, E. M., 164
Yau, S.-T., 26 n.22, 52, 212 Zorn's lemma, 32, 61 n.15