ANhiALS OF PHYSICS 108, 1-36 (1977)

Singularities and Causality Violation*.+

FRANK J. TIPLER

Department of Mathematics, Unirersity of California. Berkcle-v, California 94720

ReceivedNovember30, I976

A numberof importanttheorems in GeneralRelativity haverequireda causalityassump-

tion; for example,the Gerochtopologychangetheorem,andmostof the Hawking-Penrose--
Gerochsingularitytheorems.It is shownin this paperthat the causalityconditioncan be
replacedby weakercausalityconditions,andin somecases removedaltogether.In particular.
(a) it is shownthat if the Einsteinequations(andthe weakenergycondition)hold on the
“topology-changing”space-time considered by Geroch,thentopologychangecannot occur.
No causalityassumption is neededin the proof. Furthermore,it is shownthat if topology
changeoccurswithin a finite region,thenthis changeof topologymustbeaccompanied by
singularities.
(b) It is shownthat causalityviolation cannotpreventthe Hawking-Penrose-
Gerochsingularities unlessthe causalityviolation begins“at infinity”-a region whichis
free of matter and gravitational radiation-and this seems very unlikely.

I. TNTR~DUCTI~N

It has long been known that many solutions to the Einstein Equations possess
causal anomalies in the form of closedtimelike lines (CTL). In fact. all known asymp-
totically flat vacuum solutions with nonzero angular momentum contain such
anomalies [I, 21, and there are indications that a very rapidly rotating star would have
them also [3]. These considerations do not guarantee that our universe actually
contains such a causality violating region, however, and the question of whether or
not such a region exists in the real universe is clearly an important question to answer.
Furthermore, we would like to know if it would be possible to rmnufactzcrr such a
region, say by speeding up the rotation of a star.
I shall provide a partial answer to the second question in this paper; 1 shall show
that, in general, any attempt to evolve CTL from regular initial data will cause
singularities to form in space-time. Thus if by the word “manufacture” we mean
“construct using only ordinary materials everywhere,” then the theorems of this paper
will conclusively demonstrate that a CTL-containing region cannot be manufactured.
For a singularity is a region where the matter density becomesinfinite [4], and matter
with arbitrarily large density clearly cannot be considered “ordinary material.”
Tt does not, of course, follow from this result that causality violating regions cannot
exist, for the entire universe could be a causality violating set. In this case the notion
* Part of a Ph.D. thesissubmittedto thePhysicsFaculty of the Universityof Maryland.
1 Work supported in part by the National Science Foundation under grant No. MCS-76-21525.
1
Copyright ,O 1977 by Academic Press, Inc.
All rights of reproduction in any form reserved.
2 FRANK J. TIPLER

of “regular initial data” could not even be defined. Nor does it follow that it
is impossible to create CTL. After all, singularities are expected to occur in black
hole explosions in any case [5]. Thus the first question, the question concerning the
existence of causality violating regions, is still open.
The proofs contained in this paper will be based on the “Global Techniques”
developed by Hawking, Penrose, and Geroch. In order to make this paper largely
self-contained, the basic lemmas from Global Techniques will be given in Section 2.
The proofs of these lemmas can be found in such works as Geroch [6], Penrose [7, 81,
and Hawking and Ellis [9]. I shall also include in this section some necessary lemmas
from the theory of ordinary differential equations.
The next five sections will discuss the relationship between CTL and singularities
from several points of view. Section 3 will begin with a summary of a theorem
(published elesewhere [lo]) which shows that a time machine cannot be manufactured
in an asymptotically flat space-time without the formation of singularities. This
section will conclude with a theorem showing that any causality violation which
begins in a small region and expands must be accompanied by the formation of
singularities.
The attempt to define which is meant by the phrase “small region” leads to a
consideration of the relationship between CTL and the topological notion of com-
pactness, and this relationship is discussed in Sections 4, 5, 6. Section 4 is devoted to
the statement and proof of a very general theorem which states that causality violation
which begins in a finite region with compact boundary must result in the formation
of singularities. Section 5 and 6 will apply this theorem (and certain modifications
of it) to various problems in General Relativity. In particular, Section 5 will apply
this theorem to the problem of topology change. It will be shown that topology change
cannot occur in a compact region if the weak energy condition holds. Also included
in this section is a proof that if topology change occurs in a finite (but not compact)
region, the change must be accompanied with singularities. Previous topology change
theorems [ 111 have assumed the nonexistence of CTL; no such assumption is necessary
in the above-mentioned theorems. Section 6 will be devoted to compact space-times.
It will be shown that causality is violated at every point in a generic compact space-
time in which the weak energy condition holds.
These results suggest that it might be possible to prove a conjecture made by
Geroch [12] and Hawking and Ellis [9, p. 2721: Causality violation cannot prevent
the formation of the singularities predicted by the Hawking-Penrose-Geroch singu-
larity theorems. (It is important to prove that causality violation cannot prevent these
singularities; one should not merely dismiss the possibility with word play as in “their
[CTL] existence in itself would be singularity” [13], or as in “...collapse presumably
produces singularities-or a violation of causality, which is also a rather singular
occurrence !” [I 4, p. 9351. This conjecture is dealt with in Section 7, and to some extent
it is proven. It is shown that causality violation can prevent singularities only if the
causality violation begins at infinity.
Finally, Section 8 concludes the paper with an attempt to answer the question:
Are CTL possible ?
 SINGULARITIES AND CAUSALITY VIOLATION 3

The notation, terminology, and conventions used in this paper are the same as
those of Hawking and Ellis [HE] [9], unless otherwise noted. (However, it will be
assumedthat /1 = 0; the Einstein equations are R,,,, - ig,,,R = 8~7’,,,,.)

2. PRELIMINARIES

The object of study in this paper is a space-time, which is a pair (M, g) where M
is a real, four-dimensional, connected C” Hausdorff manifold and g is a C” Lorentz
metric. (M, g) is C” inextendibie, orientable, and space and time orientable.
The Global Techniques used to analyze the structure of space-time are naturally
divided into two parts, the first being the study of a space-time’scausal structure. and
the second being the analysis of conjugate points along a geodesic. The two basic
relationships of the first are given in the

DEFINITION. A point p will be said to chronologically precede q (written p .:. ~1)

if and only if there is a smooth (Ca) future-directed nondegenerate timelike curve
from p to q. (A degenerate curve is a curve which consists of only one point. Thus
p < p will mean that there is a CTL passing through p.) Similarly, we will say that
p causally precedes q (written p < q) if and only if there is a future-directed causal
(timelike or null) curve fromp to q. We will allow the possibility that the causalcurve
is degenerate, so we have p ( p for all p in M. A closed causal curve is signified
by the existence of a pair of distinct points p, q such that p 4’ y and c/ 4 p. The
relations -< and < have the following properties.

LEMMA I.

a << b implies a < 6:

a < 6, b < c implies a .:< (3;
a < b < c implies a < c:
a < b < c implies a +< c;
a < b < c implies a < c.

We can use the relations < and < to construct sets.

DEFINITION. The set I’-(p) = {q E M p <q) is called the chronological ,firrure

of p; I-(p) z {q E M 1q < p> is the chronological past of q. We also have J ’ ( p)
{q E M / p < q), the causaljirture ofp and J-(p) = {q E M ~q < p) the causalpast of p.
The chronological or causal future of a set S is defined by I’(S) ~-: UI,ESI (p),
J+(s) G u BESJ+(p), respectively, and similarly for the pasts of S. I+(S) is open but
J+(S) is not necessarily closed. The boundary of I+(s) will be denoted i’(S).

DEFINITION. A set S will be called achronal if there are no points p. (1E S such
4 FRANK J. TIPLER

that p < q. A set S is said to be acausal if there are no points p, q E S for which p < q.
The edge of an achronal closed set S (denoted edge (S)) is the set of points p in S such
that if r < p < q, with y a timelike curve from r to q, p E y, then every neighborhood
of y contains a timelike curve from r to 4 not meeting S.

LEMMA 2. Every point p in the achronal set f+(q) - q is the future endpoint of a
null geodesic on f+(q) which can be extended into the past on Z’-(q) either indejnitely,
or until it meets q.

Another way to state this lemma is as follows. f+(q) is generated by null geodesics
which either have no past endpoints or have past endpoints at q. Generators which
have no past endpoints never intersect as they are extended into the past.
As noted earlier, the existence of a CTL through a point p is denoted by p <p.
We shall say that a space-time (M, g) satisfies the chronology condition if there are
no points p E M such that p < p. Similarly, a space-time will be said to satisfy the
causality condition if there are no closed causal curves. The set of points at which the
chronology condition does not hold is the disjoint union of sets of the form
Z+(q) n Z-(q) # m, q E M. The set of points at which the causality condition does
not hold is the disjoint union of sets of the form J+(q) n J-(q) # (q), q E M. Another
causality requirement that is often imposed on a space-time is given by the following.

DEFINITION. The strong causality condition is said to hold at p if every neighbor-

hood of p contains a neighborhood of p which no non-spacelike curve intersects
more than once. Thus if a space-time satisfies the strong causality condition at every
point, then not only are there no closed causal curves, but there are also no “almost”
closed curves; that is, no causal curve from p can ever return to the vicinity of p.
It is not possible for a causal curve to be “imprisoned” inside a compact set if strong
causality holds. The notion of “imprisonment” is made precise by the following

DEFINITION. A causal curve y that is future-inextendible is said to be total/y future

imprisoned in S if it enters and remains within a compact set S. A future-inextendible
causal curve which does not remain within any compact set but continually reenters
a compact set S is said to be partially future imprisoned in S.
We want to prove that causality violation which evolves from regular initial data
must be accompanied by singularities; to do this we must put some restrictions on what
is meant by “regular initial data.” The set upon which it is most natural to give initial
data is a partial Cauchy surface which is defined to be a closed acausal set S with no
edge (i.e., edge (S) = a). A partial Cauchy surface is thus a spacelike hypersurface
which no causal curve intersects more than once. We shall take the existence of such
a set as a necessary condition for the existence of “regular initial data.” For any
partial Cauchy surface S, there are sets within which events are completely determined
by data on S. These sets are given by
 SINGULARITIES AND CAUSALITY \‘IOL.ATION 1

DEFINITION. The ,future domain of dependence or CUUC~IJ; development of S is

defined by

Z?t(S) G {p 1p t A4 and every past inextendible timelike

curve containing p intersects Si.

Similarly for the past history of S, we define

ii-(S) = {p 1p E M and every future inextendible timelike

curve containing p intersects Sj

3s the past domain ?f dependenceof S, and we call D(S) ::I D+(S) LJ Z>-(S) the totul
domain of’ dependenceof S. The future boundary of D+(S), that is, the limit of the
region that can be predicted from knowledge of data on S, is defined by

H b(S) is called the-ftirure Catchy horizon; H-(S), the past Cauchy horizon. and H(S),
the total Cauchy horizon are correspondingly defined. H+(S) and Dt(S) have the
properties (for S closed and achronal)

(a) D.(S) is closed,

(b) H,(S) is achronal and closed,
(c) Pa(s) = H’(‘S)LJ s.
(d) edgeS : edge H+(S).

(The above statements are still true if (+) is replaced by (-).)

If D(S) M, then the entire future and past can be predicted from data on S;
in this case we call S a Cuuchy surface. If (M, g) contains a Cauchy surface S, then
H(S) =m::‘ . Some other important properties of Cauchy surfaces and partial Cauchy
surfaces are given in the following five lemmas.

LEMMA 3. JfS is a partial Cauchy suTface, then H+(S), if nonempt?: is generated

bj, ml1 geodesicsegmentswhich have no past endpoints: these segmentsnever intersect
as the). are e.utendedinto the past.

LEMMA 4. The strong causalit~~condition holds OHint D+(S), ,for S achronal and
closed.

LEMMA 5. If S is closed and achronal, then Jtir each set qf points q < p with
p E int D-‘(S) and q ES, there is a timelike geodesicfrom q to p which attains the
maximum Iength ,for timelike curves connecting q and p. Furthermore, to each
6 FRANK J. TIPLER

p E int s+(S) there is a timelike geodesicorthogonal to S which attains the maximum

length for timelike curves connecting p and S. (The term “length” of a causal curve
denotes its proper time integral.)

LEMMA 6. Let S be a Cauchy surface for the space-time (M, g). Then M is homeo-
morphic to R1 x S, and for each a ERI, {a} x S is a Cauchy surface for (M, g).
Thus topology change is impossibleif the space-timecontains a Cauchy surface.

LEMMA 7. Zf p E int D+(S), then Z+(S) n J-(p) is compact, provided S is a partial

Cauchy surface.

The second part of Global Techniques is the study of conjugate points and the
relationship of these points to the causal structure. Roughly speaking, two points
p and q on a causal geodesic y(t) are said to be conjugate along r(t) if a geodesic which
is infinitesimally close to y(t) intersects r(t)(at p and q. More precisely, p and q will
be conjugate along y(t) if the expansion 0 of a geodesic congruence containing y(t)
becomes infinite at p and q [HE, pp. 96-97; lOO-1011. The expansion 8 satisfies the
equation
dl9ldt = -R,,KaKb - 20~ - (l/n) P (2.1)

where Ku is the tangent vector to the geodesic, t is an affine parameter along y(t),
and n = 3 for timelike geodesics, and n = 2 for null geodesics. The function o2
is called the shear of the congruence and is positive definite. For null geodesics it
satisfies the equation

da,,ldt = -CmanbKaKb - 6u,, G.2)

where 2a2 = gmncrnn and m, n = I,2 label the two spacelike directions of a pseudo-
orthonormal frame parallel propagated along y(t) [HE, p. 861. It can be shown [HE,
pp. 97, 1001 that p and q are conjugate along y(t) if and only if a function y, defined
by 8 = (l/y) dy/dt, satisfies y = 0 at q and p. If we define a new function z by the
relation zn = y, then B = (n/z) dz/dt and (2.1) becomes

(d2z/dt2) + H(t) z = 0 (2.3)

where
H(t) z (l/n)(R,bK”Kb + 203. (2.4)

Since z” = y, y will be zero at p and q if and only if z = 0 and p at q. Thus we.have

reduced the problem of finding conjugate points to the problem of discovering the
location of zeros in solutions to Eq. (2.3). This, fortunately, is a well-known problem
in the theory of ordinary differential equations. This problem has an extensive
literature from which we will take the following two lemmas.
 SINGULARITIES AND CAUSALITY VIOLATION 7

LEMMA 8. Let H(t) be continuous and nonnegative in the interval (a, $ ~1. !f
li~+i~f (2.5)
I 2 t2H(t) > 6.
Then every solution to (2.3) has infinitely many zeros on (a, $- co).
The proof can be found in Ref. [ 151.

Let H(t) be continuous on (a, + co). If

LEMMA 9.

s% H(t) dt = -1-m
0
then all solutions to (2.3) have injinitely many zeros on (a, i- 03‘).
The proof can be found in Ref. [16].
We shall also need the following proposition, a result which is /jot in the literature.

PROPOSITION 1. A sufficient condition for the existence of a solution to (2.3) with

at least t++‘ozeros in the interval [t, , m) is that there exist mrmbers t, . t, M’ith
t,, 4 tl -< t, such that

(2.6)

assumingthat H(t) is continuous and H(t) 3 0 in [t, , CO).

ProqjI We can always find a solution z(t) which has a zero at to We will show that
this solution has another zero in (to, co); for assume it does not. Then without loss
of generality we can assume z(t) > 0 and dzjdt > 0 in (t, , a), since if dz/dt c 0
at any point in (to , a). we would have a zero in (t,, , co) by the condition H(t) 3.- 0.
From (2.3) we obtain

dz dz f.,
z(t) H(t) dr.
27 +t2 =-I dt t=f, -- .fI
I’

Since dzjdt :‘; 0, we have z(t) 3 z(t,) for any t :> t, . Since H(r) ‘-:, 0. we have

Thus
dz 12
Jiit=!lstl z(t) H(t) dt < ;

, dz
z(tJ H(t) dt

- (h - to) $ 1,:1 .r,‘? H(t) dz

G dr f-f1 I I

I -~ (t, -- t,) if’ H(r) dt

- ‘I
8 FRANK J. TIPLER

By hypothesis, the factor in brackets is negative. If dzjdt It+ > 0, then dzjdt jtZtl < 0,
implying a zero of z(t) in (tz , co). If dzjdt It+ = 0, then dzjdt It+ < 0 since
J;: z(t) H(t) dt > 0 b y assumption. In either case, z(t) must have a zero in (to, co).
This contradicts the assumption that z(t) has no zeros in (to , co). Thus z(t) must have
at least one zero in (to, co), and hence two zeros in [t, , co). i
Proposition 1 concludes our digression into the theory of ordinary differential
equations. Now we will connect this mathematical interlude with physics. Recall that
most of the theorems required the assumption H(t) 3 0. From Eq. (2.4) we see that
for causal geodesics this corresponds to R,,,KaKb + 2g2 >, 0, or since a2 is intrinsically
nonnegative, to RGbKaKb3 0. The Einstein equations tell us that when Ka is null,
Rab = 8nT,,KaKb, so the above inequality will hold whenever T,,KaKb b 0. This is
assured by the Weak Energy Condition, which says that the energy-momentum tensor
at each point p in M obeys the inequality T,, W” Wb > 0 for any timelike vector Wa
in T, , where TP is the space of all tangent vectors to M at p. By continuity we will
have TabKaKb > 0 for any null vector KO in T, . Physically, this condition says that
the energy density as measured by any observer is nonnegative, and this holds for all
known forms of matter (see [HE, pages 89-911 for a more detailed discussion of this
point; see, however, Epstein et al. [17]). But the weak energy condition insures
RabKaKb> 0 only for null geodesics. For timelike geodesics, we need a different
condition to insure this.

DEFINITION. We shall say that the energy-momentum tensor satisfies the Strong
Energy Condition at all p E M if for every timelike Wa E T, , we have

T,,WaWb > $WaW,T.

Combining the Einstein equations and the strong energy condition, we get
R,, WaWb > 0 for all causal vectors W”. Physically, the strong energy condition says
that gravitation is always attractive, and this is true for all known forms of matter
(see [HE, p. 951).
The final locally defined energy condition we will need is the following

DEFINITION. We will say that the Ubiquitous Energy Condition holds on a set S
if the energy-momentum tensor at each point p ES satisfies TabKaKb> 0 for all
non-spacelike vectors Ka E T, , (If S is not specified, S = M will be implied.)
Physically, the ubiquitous energy condition says that the energy density is nonzero
for every observer at every point of S. Furthermore, for all observed forms of matter,
the condition TabPUb > 0 for all timelike vectors V implies TabKaKb> 0 for all
null vectors Ka also; the ubiquitous energy condition assumes this to be a property
of all physically reasonable Tab . Thus matter consisting entirely of radiation moving
in one special direction-Type II matter in the notation of Hawking and Ellis
[9, p. 89]-is ruled out by the ubiquitous energy condition even though T,, # 0.
However, such an energy-momentum tensor is extremely unlikely.
The ubiquitous energy condition was apparently originally proposed by Aristotle
 SINGULARITIES AND CAUSALITY VlOLATlOh 9

(Nature abhors a vacuum), and later defended by numerous authors, among them
G. W. Leibniz, who supported it with an argument which is cogent even in the world
view of General Relativity: At any point in space-time we expect there will be a
little randomly oriented radiation present, even in what would otherwise be a perfect
vacuum. The microwave background radiation, for example, is expected to be present
everywhere in space-time, except perhaps where there is matter to shield it out. This
random background radiation would be sufficient to satisfy the ubiquitous energy
condition. Even in radiation shielded regions there would be quantum mechanical
zero-point radiation (Sakurai [18, p. 331; seehowever, Epstein et al. [17]) which would
in itself be sufficient to satisfy the condition. Thus the ubiquitous energy condition
seemsto be an eminently reasonable condition to impose on the whole of space-time.
though for our purposes we will need to impose it only on certain compact sets.
The ubiquitous energy condition has been imposed (in effect) on a compact set by
Hawking [19] in one of his early singularity theorems.

DEFIXIUON. A space-time will be said to satisfy the generic condition if ecer!

timelike or null geodesiccontains at least one point p at which

(2.71

where K” is the tangent vector to the geodesicat 17.

If (3.7) is satisfied at some point p, then the tidal force ROaiil (or R,n4n4) is nonzero
at p along a timelike (or null) geodesic [HE, p. IOl]. Thus physically this condition
says that every geodesic feels tidal force at one point in its history at the very least.
To seethat this condition is reasonable, see[HE, p. 1011.In the theorems of this paper
the generic condition and the ubiquitous energy condition will be used for the same
purpose: to focus geodesiccongruences.
The relationship between the causal structure and conjugate points lies in the
following two lemmas.

LEMMA 10. A timelike geodesic curve y j>om q to p is qf’ maxima/ proper tinw
IengtII if’nd on/J, if there is no point conjugate to q along y in the iflterral (y. p).

LFMM~ 1 1. If there is a point r in (q, p) conjugate to q along (I causal geodesic, 1’.

then there is a timelike curve,from q to p.

3. FINITELY VICIOUS SPACE-TIMES

One of the major purposes of this paper is to show that it is not possible to manu-
facture a time machine-which we will define to be a region of high curvature gener-
ating a chronology violating set Vsuch that Vintersects the earth’s world line-without
the formation of singularities. One way of showing this is to note that a manufactured
time machine would have to embody the following features. The time machine would
10 FRANK J. TIPLER

have to be constructed in some localized region, for it is beyond our power to mani-
pulate all the matter in the universe. Thus we would expect the gravitational field
generated by the time machine to decrease in strength as we move away from it,
eventually becoming negligible at large distances. That is, the time machine is formed
in asymptotically flat space. (In the first approximation, complications introduced
by cosmology are ignored.) The time machine would be built from normal matter
(matter satisfying the weak energy condition) in a universe which is initially free of
CTL; that is, the time machine evolves from regular initial data in the asymptotically
flat space-time. Thus we will require the existence of a partial Cauchy surface S in
the space-time so that we can define “initial data.” We will further require (M, g)
to be “partially asymptotically predictable” from S, which essentially means that
D+(S) comes to an end because of the formation of CTL or singularities and not
because of the choice of S [20]. Otherwise, we could not say that the CTL evolved
from S. Finally, in order to use the time machine, it must be possible for an observer
initially far away from the time machine, on earth, say, to travel to the time machine,
go backwards in time, and return to earth before he left. In order for this to happen,
the time machine must not be shielded from outside observers by an event horizon;
in symbols, J-(9+) n V is nonempty, where V is the chronology-violating set. (Note
that this condition does not preclude the existence of an event horizon. It merely
says that some CTL are to be found outside of an event horizon.)
The following theorem shows that such a time machine cannot be constructed
without the formation of singularities; CTL cannot arise from regular initial data
in any asymptotically flat, geodesically complete space-time.

THEOREM 1. An asymptotically flat space-time (M, g) cannot be null geodesically

complete if

(1) RabKaKb> Ofur ail null vectors Ka;

(2) the generic condition holds on (M, g);
(3) (AI, g) is partially asymptotically predictable from a partial Cauchy surface S;
(4) the chronology condition is violated in

J+(s) n J-(3+).

(Note that condition (1) follows from the Einstein equations and the weak energy
condition.)
The idea behind the proof is quite simple; we first show that under the above con-
ditions there exists a null geodesic which never leaves H+(S). This geodesic cannot be
complete, for it can be shown that (1) and (2) imply that all complete null geodesics
have a pair of conjugate points. This would be impossible, by Lemma 11 and the
achronality of H+(S). The rigorous proof has been published elsewhere [lo].
The theorem above has two major weaknesses from the physical point of view.
First of all, the proof depends on the existence and structure of asymptotic infinity;
the theorems ignore complications due to cosmology. This objection is perhaps not
 SINGULAI~TIE~ AND CAUSALITY VIOLATION II

roe serious, for the condition of asymptotic flatness is used only to show that at least
one generator of H+(S) could be continued into the future for infinite affine parameter
length while remaining in H+(S). This state of affairs would be expected even in a
nonasymptotically flat universe, unless there were many causality violating regions
whose Cauchy horizons intersected, as in Fig. 1. The other weakness is the fact that

,,/u,,
\ -
partial Cauchy surface S /’

FIG. 1. Multiregion causality violation.

although we know singularities must occur with the formation of CTL in asympto-
tically flat space-times,we do not know where they occur. In this paper it will be shown
that if causality violation arisesin ajnite region from regular initial data, then singu-
larities must occur. and further they must occur in the finite region. The word “finite”
can be made more precise in several ways; however, note that this word is usually
associatedin somemanner with the mathematical notation of compactness(in standard
cosmology, for example, ajnite universe is one whose spacelike sections are topo-
logically compact). Thus one way of making “finite” more precise is given in the
following

DEFINITION. A space-time (M, g) will be called jnitelv vicious if it contains a

partial Cauchy surface S; if it violates the causality condition in a subset of J’-(S);
and if it has a hypersurface slicing S(T) with the properties

(i) S is one of the sliceswith S(0) = S:

(ii) there is a closed interval [T1, TJ of the slice parameter such that if
-7E [T, 9~~1, then S(7) n D+(S) is spacelike, and S(T) n H+(S) is compact. Also, if
TV, TV are any two numbers in [TI , TJ with 73 > T:~, then s(‘T4) n a+(S) lies to the
future of S(7.J n D+(S);
(iii) let B be the region of space-time in D+(S) between s(T1) and s(T2) inclusive,
and let y be any segment of a generator of H+(‘S) with y n s(T2) i; B and y C B.
Then y can be extended in H+(S) n B such that the extension intersects each s(T)
for 7 E [T 1 , TJ exactly once. (The adverb “finitely” comes from the compactness of
f+(S) n s(T); “vicious” is the adjective Carter [21] applies to any set containing
CTL.)

The precise definition of a finitely vicious space-time is somewhat complicated.

but the physical situation it is meant to model is actually very simple, ar.d is illustrated
in Fig. 2. If causality violation occurs somewhere in the future of a partial Cauchy
surface S, a Cauchy horizon, H+(S), must develop to separate the acausal region
12 FRANK J. TIPLER

from the interior of D+(S), since by Lemma 4, the strong causality condition holds
on int B+(S). In the region where causality violation begins, H+(S) may have a very
strange structure-indeed, the structure of the entire region may be very strange-but
far away from this region we would expect space-time to take on familiar features.
For example, we would expect that it would be possible to slice the space-time with
a sequence of hypersurfaces, such that these hypersurfaces become spacelike far away
from the region where causality violation begins. Furthermore, if we continue suffi-

FIG. 2. A finitely vicious space-time. The shaded region is region B, and S is a partial Cauchy
surface.

ciently far into the future along the hypersurface slicing (increasing T), we should,
if the strong gravitational fields which give rise to the causality violation are
“localized” in space, come at last to a region of space-time where the light ray trajec-
tories have the familiar property of intersecting a spacelike hypersurface once and
only once. This region I have denoted with the letter "B" in the definition of “finitely
vicious.” Condition (iii) in this definition has been stated so as to allow the possibility
that a generator of H+(S) is closed or almost closed; it is possible that a generator of
H+(S) could intersect the region B several times. However, such a generator must
leave B, enter the strong field region and reenter B; it cannot close or almost close
entirely in B. The portions of the hypersurfaces in B have many of the properties
of Cauchy surfaces. (Condition (iii) focuses attention on those generators of H+(S)
which end on S(Q-,), the future boundary of the region B, because it is possible for
null geodesics to enter H+(S) as generators as the geodesics move into the past.
However, Lemma 3 assures us that a generator of H+(S) will never leave H+(S)
in the past direction; if a null geodesic is a generator of N+(S) when it hits S(T~), it
will be a generator of H+(S) when it hits S(7r), and when it hits any hypersurface
S(T) in between.)
We want to consider at present only those space-times in which the causality
violating region is “localized in space,” so we must find some way of making “locali-
zation in space” precise. One way of accomplishing this is to require H+(S) 0 S(7)
to be compact since H+(S) is the boundary between the acausal region and the causal
 SINGULARITIES AND CAUSALITY VIOLATION I?

region. The remainder of condition (ii) is devoted to making certain that the word
“space” has a meaning.
If a time machine were manufactured in our universe, our space-time would
probably be finitely vicious. For it should be possible to find a sequenceof hyper-
surfaces through the world tube of our galaxy which would be spacelike except in th:
immediate vicinity of the time machine. The boundary between the causal and acausal
region would begin in the strong gravitational field near the machine, and expand
until it finally intersected the world tube of the earth, where the properties of null
geodesicsare that required by condition (iii). We would expect the boundary to
expand in all directions into regions of the galaxy where the “local” causal behavior
is similar to that of the earth. For a certain period, the period between 71 and TV.
say, the boundary H+(S) would be confined within our galaxy and so S(T) n H’(S)
would be contained within a compact set. (If H+(S) expands in all directions away
from the strong field region, then there should be a neighborhood of H+(s) n 3-r)
with compact closure.) Since both H+(S) and S(T) are closed. it follows that
H*(S) n S(T) is itself compact.
The expansion of the Cauchy horizon is an essential feature of a time machine.
for the region of causality violation must expand to encompassthe earth. This expan-
sion also guarantees the occurrence of singularities, as shown by

THEOREM 2. Suppose that a space-time (M, g) is jinitely riciow, and supposethat

the area of H+(S) n S(TJ is strictly greater than the area of H+(S) n S(TJ. Then if
the weak energy, condition and Einstein equations hold, the space-time (M, g) is ml/l
geodesically incomplete.

(The area of H-(S) n S(T) is defined in the same manner as black hole area: see
[HE, p. 3181.)

Prooj: The proof is essentially the time reverse of the proof of Hawking’s well-
known Black Hole Area Law [HE, pp. 318-3191. Since any genrator of H+(S) which
intersects s(T2) intersects siTI) exactly once, the area of H+(S) f7 s(T) can increase
from s(T1) to S(TJ only if the expansion 8 of someof the generators of H*(S) is positive
somewherein B. This meansthat 0 < 0 along at least one generator y as we move into
the past. If y were geodesically complete, then by the weak energy condition. the
Einstein equations, and Eq. (2.1), some of the generators of H+(S) would intersect to
the past of H+(S) n s(T2). This is impossible by Lemma 3. Thus y cannot be past
complete: the space-time (M, g) is null geodesically incomplete. 1
Notice that Theorem 2 provides us with information about the location of the
singularity in a finitely vicious space-time with an expanding Cauchy horizon. A null
generator of H+-(S) ends in a singularity somewhere between S(T~) and S = S(O).
The concept of finite viciousness was introduced as one method of making the
notion of “finite region” precise. One could prove that the construction of a time
machine would generate singularities without using this concept. First note that 6’ Y- 0
somewhere along at least one null geodesic generator of H+(S), for the region of
14 FRANK J. TIPLER

causality violation must expand to include the earth. We then infer from the proof of
Theorem 2 that the condition 6 > 0 is a sufficient condition for the occurrence of
singularities.

4. COMPACT BOUNDED VICIOUS SPACE-TIMES

Another way we can give meaning to the expression “causality violation begins
in a finite region” is to say “causality violation begins in a region of space-time with
compact boundary.”

DEFINITION. A space-time (M, g) will be called compact bounded vicious (cb.

vicious) if it contains a partial Cauchy surface S and a compact hypersurface A with
the properties

(i) A is the boundary of a closed set B which has nonempty intersection with
a causality violating region. A is not required to be connected;
(ii) A n D+(S) is nonempty, but B n Z-(S) = 0;
(iii) if H+(S) n A is nonempty, then U n A is spacelike and connected, where
U is some neighborhood of H+(S) n A.

Figure 3 gives an example of a c.b. vicious space-time. The definition is straight-

forward; only a few clarifying remarks need be made. First of all, the first part of

H+(S)
A is spocelike in
o nbd of H+(S) l-l A

A--

FIG. 3. A c.b. vicious space-time. Causality violation occurs somewhere inside the cone H+(S).
A is the boundary of the cylinder; B is the entire cylindrical solid, including A.

condition (ii) is necessary in order to avoid situations like the one depicted in Fig. 4,
where the compactness of C gives us absolutely no restriction on the size of the
causality violating region. The second part of condition (ii) tells us that B is located
entirely to the future of S. We state this requirement as B n Z-(S) = m rather than
B C J+(S) in order to include space-times like the one in Fig. 5.
Since causality is violated in B, but int D+(S) n B # o, we know that H+(S) n B
 SINGULARITIES AND CAUSALITY VIOLATION 15

is nonempty. However, in some c.b. vicious space-times it is possible to choose A

such that N l(S) n A is empty; it is possible to do this, for example, in the Taub-NUT .
universe, as is shown in Fig. 6. Also, the set A pictured in this figure is an example of
a set A which is not connected. The fact that A is not required to be cpnnected con-
stitutes a slight modification in the definition of hypersurface given in HE. where
a hypersurface is said to be an orientable imbedded paracompact C” connected
Hausdorff three-dimensional submanifold of M without boundary [HE, pp. 14, 441.
But the hypersurface A possesses all of the other properties listed in the preceding
sentence.
If the region where causality violation begins is “localized’‘-say it is confined to
a small region in our galaxy-it should be possible to find an A sufficiently large so that
A is spacelike in a neighborhood U of H+(S) n A. The justification of this possibility

3
FIG. 4. A set C which has all the properties of the set B in a c.b. vicious space-time except (ii).
Causality is violated everywhere in J+(S) - d+(S); everywhere inside the cone forming H%S).

FIG. 5. A c.b. vicious space-time in which B n 1-G) = 4, but B (T J+(S). Except for the cuts and
identifications, the space-time is Minkowskian. The ellipse is the set A. B is the region enclosed by
A. By Theorem 3, the set B containssingularities. We cannoteliminatea priori the possibilitythat a
singularity will “shield” some portion U from a causal curve with past endpoints on S; that is,
lInJ+(S) = 4, but CJCB.

jgj/IO8/1-2
16 FRANK J. TIPLER

has already been given in the discussion of finitely vicious space-times. U n A is

required to be connected in order to avoid “A’s” like that pictured in Fig. 7.
The reader may wonder why the boundary of B was the set required to be compact
rather than B itself. In other words, why not define a “compact vicious” space-time
rather than a “compact bounded vicious” space-time? The answer to this question
is simple. It will be shown in the next theorem that B cannot be compact in a physically
realistic space-time. This fact makes it difficult to prove the occurrence of incomplete
causal geodesics in B, for it is possible that the noncompactness of B results from
causal curves in B “going off to infinity” rather than terminating in singularities.

FIG. 6. The Tat&NUT universe, a c.b. vicious space-time. CTL occur to the future of H+(S).

\ cut out

‘S

FIG. 7. A set “A” which has all the properties of the set A in a c.b. vicious space-time except
U n “A” is not spacelike and connected in a neighborhood U of H+(S) n “A”.
 SINGULARITIES AND CAUSALITY VIOLATION 17

In the following theorem, we shall eliminate this possibility by making use of the
property of int I’%(S) given by Lemma 5. To each point p E int fil(S), there is a time-
like geodesicfrom Stop of maximal length. We shall say that the part of B in int fi (S)
is$nite if there is an upper bound to the lengths of causal curves from S to points
in [int fi-t(S)] n B. We can express this in the notation of HE by definining 4 p, cl)
for points p, q E M to be zero if 4 $ P(p) and otherwise to be the least upper bound
of the lengths of future-directed piecewise non-spacelike curves from p to r/. For
sets S and U, we define d(‘S, U) to be the least upper bound of u’cp, q), p ES. q t 1,’
[HE-p. 2151. Thus [int b’(S)] n B is finite if and only if cf(S, [int B.(S)] n B) is
finite; it is finite if no causal curve from S can remain in [int Zji(S)] n B forever. WC
can now prove that any physically realistic finite c.b. vicious space-time is singular.

THEOREM 3. If’ the ubiquitous energy condition and the Einsteiw equations hold OII
the set B of’ a c.b. vicious space-time, then B is noncompact. Furthertllorr. if’
rt(S. [int fit(S)] n B) is finite, then the space-time is timelike geodesical!l, incomplete:
there must be ati incomplete timelike geodesic in B.
Scholium. The above theorem is still true if we replace the assumption that the
ubiquitous energy condition holds on B with two much weaker assumptions (I)
the weak energy condition, and (2) the assumption that there is at least 01~’ point
(1E Ha+-(S)n B such that KaKbK~,R,~,,,c,K,.~$7’0 at q (K” is the tangent vector to a
generator of H’(S) through q). Thus B will be noncompact if at least one generator
of H-(S) in B feels tidal force at least one point in its history in B. (Condition (2)
can be replaced by the generic condition.)
We will need the following two propositions in the proof of Theorem 3.

PROPOSITION 2. Let S be a partial Cauchlvsurjbce. and let K be a compact WI II IlicIt

is contained in int 13-(S) u S. Then J-(K) n S is compact.
Prooj: Since the open sets I-(p). p E int Zj+(‘S), cover [int I?‘(S)] u S. and since
K is compact, there are a finite number of points pit int Do+ such that
KC ui IP(‘p;). Since I-( pi) C Jm(pi), we have K C ui J-( pi). Thus J-(K) C (J, J ( IJ]).
which gives

Jo n SC J (I\) n J-(S) C [v Jp(pi)] n J!(S) =- u [J(p,) n J’(S)].

l
By Lemma 7. ./ ( p,) n J+(S) is compact, so vi [J-( p,) n J’(S)] is compact. Nob.
since by definition S is closed, {l,Ji [J-( p,‘) n J’(S)]] n S is compact. But

J (K) n S = J-(K) n S n SC \u [J-(pi) n J*(S)]; n s

which implies J-(K) n S is contained in a compact set. Thus J-(K) n S is compact. a

PROPOSITION 3. Let S be a partial Cauch~~swface. !f H-is) is nonemptj., therr

18 FRANK J. TIPLER

any generator of H+(S) which is totally past imprisoned inside a compact set B is
geodesically complete in the past direction.
The proof, like the proposition, is an obvious generalization of Lemma 8.5.5
of [HE, pages 295-2971. However, it should be noted that the HE proof of Lemma
8.5.5 contains a few algebraic errors; see my Ph.D. thesis for details [IO].

Proof of Theorem 3. Suppose that B is compact. Since the causality condition is

violated in B and B n D+(S) # 0, H+(S) must be nonempty, with some generators
of H+(S) intersecting A in the past direction unless H+(S) is entirely contained in B.
If H+(S) C B, then H+(S) is compact since H+(S) is closed and B is compact. If
H+(S) n A # 0, then any generator y of H+(S) which once enters B can never leave
B. For if y did leave B it would have to intersect A at least once more. But this is
impossible since A n U is spacelike and connected in some neighborhood U
of H+(S) n A, and the initial intersection of the generator y with A defines the future
side of the spacelike region of A around H+(S) n A as facing outward from int B.
Thus if the generator y were to leave B, it would have to be future directed upon
leaving B; this cannot occur since the space-time is time orientable.
By Proposition 3 the generators of H+(S) which are totally past imprisoned in B
are geodesically complete in the past direction. Consider the expansion 0 of the
tangent vectors a/ax to the null geodesic generators of H+(S) n B. Suppose that 6’ > 0
at some point p on a generator y of H+(S) n B. Then by the ubiquitous (or weak)
energy condition, the Einstein equations, and Eq. (2.1) some of the generators of
H+(S) would intersect to the past of p. This is impossible by Lemma 3. Therefore,
0 < 0 on H+(S) n B.
Now consider the family of differentiable maps

pz : H+(S) n B + H+(S) n B

defined by taking a point q E H+(S) n B a distance z to the past along the null geodesic
generator through q, where z is the proper distance in the metric g& . The positive
definite metric gib is defined by introducing a future-directed vector field V which is
geodesic in a neighborhood U of H+(S) n B with compact closure. Then g’(X, Y) =
g(X, Y) + 2g(X, V) g(Y, V). Let dA be the area measured in the metric gAb of a small
element of H+(S) n B. Under the map pZ ,

(d/dz) dA = -0 dA.

Now since H+(S) n B is compact, the integral

dA
sH+kS)nB
must be finite. Since d/dz(dA) > 0, and since z is not bounded above, this is possible
only if every geodesic generator y of H+(S) n B is a closed null geodesic which is
entirely contained in B, for in this case z would be cyclic. Furthermore, it is possible
 SINGULARITIES AND CAUSALITY VIOLATION 19

for L to be cyclic only if % -= 0 on H,(S) A B. But by Eq. (2.1), % L 0 on H l.(S) n B

only if -R,,&aKti - 20~ x 0 on H+(S) n B. Using the Einstein equations, together
with either the ubiquitous energy condition or the weak energy condition and the
generic condition, we see that this is impossible. (The Einstein equations and the
ubiquitous energy condition imply RnbKUKh :. 0 everywhere on H+(S) n B. The
Einstein equations, the weak energy condition, and the generic condition (or con-
dition (2) of the Scholium to Theorem 3) imply that R,,KaKh + 20~ > 0 at least one
point on some y.) We have a contradiction and thus H+(S) n B must be noncompact.
This implies that B is also noncompact, since H-‘(S) is closed.
Furthermore, since B+(S) is closed, D+(S) n B is noncompact. For suppose it were
compact. Then H’(S) n B would be a closed subset of n;(S) n B, hence compact,
in contradiction to the above result. Now assume H-!(S) n A I (r:. Since A n U
is spacelike and connected in some neighborhood U of H’(S) n A, there is a neigh-
borhood W C U of H’(S) n A such that I-(H’(S) n A) n WC int B. (That is. any
timelike curve with future endpoint on H+(S) n A immediately enters int B as it
moves into the past, and remains in int B for some proper time at least.)
Now let V be a neighborhood of H k(S) n A contained in U. Then A C.n
is compact, since it is a closed subset of the compact set A. We must have
f-(H-(S) n B) n ,1 C A - V n A, since if this were I?O/ true, there would be a past-
directed timelike curve from int B which intersects A -- c’ on Ieauing int Z?. But
V n A is spacelike and past-directed timelike curves can intersect V n A only upon
entering int B. since the space-time is time orientable, and A is orientable. Further-
more. we have
I (‘H-t(S) n B) C A n int D-l(S) u S (4.1)

since I-( Hi+) ~ /Z(S) C int D+(s) u S and in addition A n Z-(S) == .~j. Thus

/-(H+(S) n B) n A C A - V n A n [int ii+(S) U S].

Define A - V n A n [int r?-+-(S)u S] =- K. Note that K is compact, sinceA - V n A

consists of disjoint sets, some in int a+(s) u S, and the others not. (Recall that
A -- V n A n H-(S) == ? .) Thus by Proposition 2, J-(K) n S is compact.
Since B n I-(S) :m: LI!, any timelike curve drom B which intersects S must intersect
A. Thus
I--( H’(S) n B) n SC I-([I-(H t(S) n B)] n A) n S. (4.3)

By (4.2), the set on the RHS of (4.3) is contained in Z-(K) n S. Furthermore,

I-(K) n S CJ-onS L

so we have, finally

I-(w(S) n B) n S C L.
20 FRANK J. TIPLER

By a similar argument I-(int D+(S) n B) n SC L. Combining these results, we can

conclude that 1-(6+(S) n B) n S is contained within a compact set L. (If
H+(S) n A = a, we set K = A n [int I?+(S) u S] and L = J-(K) n S.)
Now write d(S, [int D+(S)] n B) = a. Suppose the space-time (M, g) is timelike
geodesically complete. Let /-!I: L x [0, 01] -+ M be the differentiable map which takes
a point p E L a distance s E [0, a] up the future-directed geodesic through p orthogonal
to L. Then /3(L x [0, a]) is compact, and it contains the set int D+(S) n B since by
Lemma 5, there is to each point q E int B+(S) n B a future-directed geodesic y
orthogonal to S of maximal length from S to q, and by assumption, y has length
less than or equal to (Y. Furthermore, since I-(D+(S) n B) n S C L, y intersects S
in L. Thus
D+(S) n B = D+(S) n B C /3(L x [0, a])
and hence D+(S) n B is compact. But we have already shown that D+(S) n B
is noncompact. We have a contradiction, so (il4, g) cannot be timelike geodesically
complete; there must be an incomplete timelike geodesic in B. 1

5. SINGULARITIES, CTL, AND TOPOLOGY CHANGE

Besides its intrinsic interest, Theorem 3 has a number of important applications,

most notably to the question of topology change. By Lemma 6, topology change can
occur only if a Cauchy horizon forms. However, this Cauchy horizon could be due
either to the formation of singularities or to causality violation (or both). Many
workers have discussed the causal case [II, 22, 231, but whether or not causality
violation could give rise to nonsingular topology change has up to now been an open
question. We can use Theorem 3 to show singularities must accompany topology
change in a physically realistic space-time, whether or not causality violation occurs.
To show this for closed universes we will require a lemma due to Geroch [II, 221.

LEMMA 12. Let B be a compact subset of a space-time such that the boundary of B
is the disjoint union of two compact spacelike 3-manifolds, S and S’. Suppose that the
causality condition holds on B. Then S and S’ are dfleomorphic, andfurther B is topo-
logically S x [0, I].
We first use Theorem 3 to show that if B is compact, then, in a physically realistic
space-time, topology change cannot occur at a/I.

THEOREM 4. Let B be a compact subset of a space-time such that the boundary

of B is the disjoint union of two compact spacelike 3-manifolds, S and S’. If

(1) the weak energy condition and the Einstein equations hold on B;
(2) KaKbK~,R,~,,~,Kf~ # 0 at at least one point p E B on every null geodesic which
is totally past-imprisoned in B.
Then S and s’ are diffeomorphic, and further B is topologically S x [0, 11.
 SINGULARlTlES AND CAUSALITY VIOLATIOK 21

Comment. Since we are using the notation of HE, both S and S’ are connected.
The theorem is still true even if one (or both) is not connected, but the proof is a bit
more complex than the one given here. (Condition (2) could be replaced by: (3) the
generic condition holds on B.)

Proqf: Since S and S’ are spacelike and orientable, and the space-time is time
orientable, any causal curve from S into B can intersect S again only after first leaving
B through S’. Similarly for S’. Thus we can attach B to a space-time (M, g’) in which
S and S’ are acausal and B n I-(S) = D. If S and S’ are not diffeomorphic, then by
Lemma 12, causality violation occurs in B. The space-time (M’, g’) is c.b. vicious.
with B in the definition of c.b. vicious space-time chosen to equal the B of the present
theorem, and A =mS u S’. Since S C d+(S), condition (ii) holds, S is a partial Cauchy
surface since it is a 3-D acausal spacelike manifold without boundary. Since S’ is
connected, if H-+(S) n A = H+(S) n S’ is nonempty, then S’ n U is obviously space-
like and connected in some neighborhood U of H+(S) n A. Conditions (I) and (2)
are the same as those in the Scholium to Theorem 3, so the conclusions to Theorem 3
hold here. (Note that any generator of H+(S) which intersects B is totally past-
imprisoned in B; in fact, such a generator is totally,firture imprisoned in B.) Thus B
must be noncompact, contrary to assumption. Hence. S and S’ are diffeomorphic.
and further B is topologically S x [0, 11. 1
We can state Theorem 4 in a somewhat different form. If topology change c/oes
occur. then B is not compact and contains a singularity if B is finite.

THEOREM 5. Let B be a four-dimensional region qf a space-time such that the

boundary qf B is the disjoint union qf two compact partial Cauchy surfaces, S and S’.
Suppose that
( I) the weak energy condition and the Einstein equations hold on the space-time:
(2) the generic condition holds on the space-time:
(3) d(S. [int D+(S)] n B) isjnite;
(4) S and S’ are not dl@eomorphic.
Then the space-time is timelike geodesicall,v incomplete and B is not compact.
The proof is similar to that of Theorem 4, and so is omitted. 1
We can prove a theorem analogous to Theorem 4 for open universes, provided we
assume the possible topology change to be localized in a compact region. Following
Geroch [22, pp. 51-531 we define a 3-manifold S to be externally Euclidean if there
exists a connected compact set C of S such that S - C is diffeomorphically S2 . R.
That is, outside of the region C, S is identical with ordinary Euclidean space with a
3-ball removed. Let A4 be a four-dimensional subset of space-time whose boundary
is the disjoint union of two externally Euclidean spacelike 3-manifolds S and S’.
Suppose that there exists a connected compact set K of M such that M - K is diffeo-
morphic to S* ‘< R x [0, I], where for each fixed number 01E [0, 11 the submanifold
22 FRANK .I. TIPLER

S2 x R of M is spacelike and for each fixed point p of S2 x R the line [0, I] of M

is timelike. Then M will be called externally Lorentzian. K may be described as a
timelike world tube between S and S’ in which any “topology change” must take place.
Analogous to Lemma 12, we have

LEMMA 13 (Geroch [22 p. 551). Let M be an externally Lorentzian portion of

space-time, the boundary of M being the disjoint union of two spacelike externally
Euclidean 3-manifolds, S and S’. Suppose M has no closed timelike curves. Then S
and S’ are dl@eomorphic,and M is dl@eomorphicallyS x [0, 11.
We can strengthen this result.

THEOREM 6. Let M be an externally Lorentzian portion of space-time, the boundary

of M being the disjoint union of two spacelike externally Euclidean 3-manifolds, S
and S’. If S is a partial Cauchy surfacefor the entire space-time, and a Cauchy surface
for M - K, and in addition we assume
(1) the weak energy condition and the Einstein equationshold on M;
(2) the generic condition holds on the space-time.
Then S and S’ are d@eomorphic, and S is a Cauchy surfacefor M.
(The Cauchy surface condition makes certain that the breakdown in prediction is
restricted entirely to K; M - KC int D+(S).)

Proof. If the conclusion does not hold, then causality is violated in K. We now
proceed as in Theorem 4, showing that M can be regarded as part of a c.b. vicious
space-time. Choose B = K and A = 8K. (A will have to be smoothed out at 8K n S
and 8K n S’, to make sure that A is a C” manifold, but this is a minor technicality.)
The remainder of the proof is very similar to that of Theorem 4, and will
omitted. 1

6. CTL AND COMPACT SPACE-TIMES

Another connection between the concepts of compactness and causality violation

is provided by compact space-times.

LEMMA 14. (Geroch [ll]; HE [p. 1891). If(M, g) is compact, then the chronology
condition is violated in the space-time. 1

The proof is easy. We simply note that M can be covered by open sets of the form
I+(q) with q E M and recall that the chronology condition holds at q only if q is not
in I+(q). Thus M can be covered by afinite number of sets I+(q) only if the chronology
condition is violated in ‘M.
The important thing to notice about this proof is that it is entirely topological in
 SINGULARITIES AND CAUSALITY VIOLATION 23

nature: there is no reference to the dynamics of space-time-there is no reference to

Einstein’s equations. If one usesEinstein’s equations, we can prove a stronger theorem
than Lemma 14; we can prove that in a generic, compact space-time, the chronology
condition is violated at euery point.

DEFINITION. A space-time (M, g) will be called tot&y LGYOUS if I-(q) n i-(q) -- M

for some q E M.
Notice that if P(q) n I-(q) = M is true for one q E M, it will be true for all q c M
since the choronoly violating set, denoted by V, is the disjoint union of open sets
of the form P(q) n I-(q). Thus if (M, g) is totally vicious, then every point in ,iM
can be connected to every other point by both a future-directed and a past-directed
timelike curve. The causality violation is total.

THEOREM 7. /f iM, g) is compact, then CM, g) is totally vicious provided theJoNowing

conditions hold

( I) the Einstein equations and the weak energy condition.

(2) the generic condition.

Proof. By Lemma 14, (M, g) violates the chronology condition, so there is a

point cl e M for which I+(q) n I-(q) # O. If (M, g) were not totally vicious, the set
t’(q) u f-(q) would be nonempty. Suppose f’(q) f G Now j’(q) is generated by
null geodesic segmentswhich have no past endpoints. For a generator of f~+(q) can
have an endpoint only at q, and q cannot be an endpoint without violating the
achronolity of I-Yq) (since q < q). We can now argue as in Proposition 3 to show that
all generators of f+(q) are geodesically complete in the past direction. But this contra-
dicts conditions (1) and (2) as an argument similar to the one found in the proof of
Theorem 3 will show. If f+(q) # @, we can obtain a contradiction in the same way.
Thus i’(q) u j-(q) must be empty, which means that (M, g) is totally vicious. m

7. CAUSALITY VIOLATION AND THE HAWKINGPENROSE-GEROCH

SINGULARITY THEOREMS

Besidesthe Geroch Topology Change Theorems, there is another important class

of theorems in General Relativity which use the causality condition: the Hawking-
Penrose-Geroch Singularity Theorems. In an earlier section of this paper, I showed
the causality condition to the nonessentialin the first classof theorems; in this section,
I shall argue that the causality condition is an unnecessary assumption in the singu-
larity theorems. More precisely, 1 shall generalize the two most important singularity
theorems, showing that singularities can be prevented by causality violation on/~
if the causality violation begins at “infinity,” an unlikely possibility.
As the first step in the generalization, recall the two most important singularity
theorems.
24 FRANK J. TIPLER

HAWKING-PENROSE THEOREM [HE, p. 2661. Space-time (M, g) is not timelike

and null geodesically complete if
(1) RabKaKb 3 0 for every non-spacelike vector Ka (this can be inferred from
the strong energy condition and the Einstein equations);
(2) the generic condition is satisjied;
(3) the chronology condition holds on M;
(4) there exists at least one of the following:
(i) a compact achronal set, without edge,
(ii) a closed trapped surface,
(iii) a point p such that on every past (or every future) null geodesic from
p the divergence 8 of the null geodesics from p becomes negative (i.e., the null geodesics
from p are focussed by the matter or curvature and start to reconverge).

PENROSE’S THEOREM [HE, p. 2631. Space-time cannot be nuNgeodesically complete

if
(1) R,,KaKb 3 0 for all null vectors Ka (this can be inferredfrom the weak energy
condition and the Einstein equations).
(2) there is a noncompact Cauchy surface in M;
(3) there is a closed trapped surface in M.

(For cosmological applications, condition (3) in Penrose’s theorem can be replaced

by condition 4(iii) of the Hawking-Penrose theorem; in open universes, we need only
the weak energy condition to infer the existence of singularities.)
The generalization of the above theorems will be based on the following two defi-
nitions.

DEFINITION. A space-time (M, g) is said to be asymptotically deterministic if

(i) (M, g) contains a partial Cauchy surface S such that
(ii) either N(S) = H+(S) u H-(S) is empty, or, if not, then
liz [inf T,,KQK6] > 0

on at least one of the null geodesic generators y(s) of H(S), where a is the past limit of
the affine parameter along y if y E H+(S), and the future limit if y E H-(S). (K” is
the tangent vector to y.)

DEFINITION. The matter tensor will be said to be past stochastic along a causal
geodesic segment y if there exist numbers a > 0, b > 0, and an integral number c of
disjoint affine parameter intervals (sl , s,), (sg , s&..., (si , s$+&... along y, each interval
satisfying 1sj - sj+r 1 > b (and sj > ++I for all j) with TabKaKb > a at every point
 SINGULARITIES AND CAUSALITY VIOLATION 25

in every interval. (K” is the tangent vector to y.) Furthermore, c is finite if y has a
past endpoint or is past incomplete, and infinite if y is past complete. Future stochastic
matter tensors are defined similarly.
These detinitions are based on the following physical reasoning. Ifcausality violation
occurs in a space-time with a partial Cauchy surface S, then H(S) must be nonempty.
For the sake of argument, suppose the causality violation occurs to the future of
S, giving H’(S) 7’: i‘ . If the formation of a Cauchy horizon is due to causality
violation, then we would expect that the region where H-(S), and hence the causality
violation, begins would contain matter. It requires nonzero gravitational fields to
“tip over the light cones” sufficiently far to give causality violation, and it seems
unlikely that these fields would occur in empty space,for nonzero gravitational fields
will give rise to matter via pair creation. (There are exceptions to this; for example.
the Taub-NUT universe and the modified Minkowski spacepictured in Fig. 5, but
in these casesthere are singularities on the Cauchy horizon, and hence H+-(S) is not
due entirely to causality violation. Furthermore, in the actual universe, particle
creation would be expected to occur around such singularities.)
The region where H+(S) “begins” is an open set of the space-time which contains
the past inextendible portions of the generators of H.(S).. In asymptotically deter-
ministic space-times, this region is nonempty in the sensethat a certain component
of the matter tensor as measuredin a pseudo-orthonormal frame parallel propagated
into the past along at least one of the generators of W(S) does not have a vanishing
lower bound. It is possible, of course, that the particular component we have selected
could vanish as the afhne parameter approached its past limit without all components
vanishing. However, physical considerations show this to be unlikely. Suppose, for
example. that the matter can be represented as a perfect fluid. with matter tensor
[HE, p. 701
To, = (p L P) K, v,, -~'~p&l, .
Thus
T,,bKaK” = (p -- p)( V,,IY~)~
which can vanish s -+ a if
case I : p -* 0 and p --, 0 (thus 7,,, --, 0).
case2: p+ -p, with p H 0, p H 0.
case 3: ( VJP) 4 0,
(or both 2 and 3). Case 2 cannot occur if we impose the strong and weak energy
condition on the space-time. As for case 3, pick a point b on y(s) with V,,K” -#. 0.
Then the frequency shift which a photon traveling along y(s) undergoes is given by
v41
,, l’,.K”
-b'h --; __~
11.. .~,K”

where the frequency is measuredby observers moving with the fluid. Thus if V’,.K’ + 0
as s + a a photon would suffer an infinite blue shift as it moved from a to b: a space-
26 FRANK J. TIPLER

time which allowed this type of behavior would be unstable, because a photon from
a would arrive at b with infinite energy. {See Ellis and King [25, p. 1541 for a detailed
discussion of this type of instability.) Note that in asymptotically deterministic space-
times, the limit
lim
s-a inf TabKaKb (7.1)

is not required to exist; it is merely required that (7.1) be greater than zero.
However, it is possible to have TabKaKb + 0 as s --f a while (7.1) equals zero, for
it is conceivable that TabKaKb could “fluctuate” as s -+ a, with TabKaKb = 0 at an
infinite number of points if a = - co. Even if this occurred, we would still expect
that the matter tensor would be at least stochastic along at least one generator y(s)
of H+(S) unless T,bKaKb approaches zero in some average sense as s --f a; for example,
if y(s) intersected a proton every so often, then the matter tensor would be past
stochastic along y(s). Note that were y(s) past complete, it would be possible to have

lim Ci”=l I ‘i - ‘i~i = 0

N-00 1 s1 - sN+l /

so that the nonzero regions are of zero measure in the entire history of y(s), and still
have the matter tensor past stochastic along y(s).
In summary, then, it seems very reasonable to assume that a space-time in which
the Cauchy horizon is due to causality violation is asymptotically deterministic,
or at least it has a generator of H(S) along which the matter tensor is past (future)
stochastic as the generator of H+(S) (H-(S)) approaches its past (future) affine para-
meter limit. With this assumption, we can eliminate the causality condition from
Penrose’s theorem and the Hawking-Penrose theorem.

THEOREM 8. (Generalized Penrose’s Theorem). Space-time (M, g) cannot be

null geodesically complete if

(1) R,,KaKb > 0 for all null vectors Ka;

(2) there is a closed trapped surface in M;
(3) the space-time is asymptotically deterministic, and the Einstein equations hold;
(4) the partial Cauchy surface defined by (3) is noncompact.
Conditions (3) and (4) can be replaced by: There exists a noncompact partial
Cauchy surface S with the property
(3’) if H(S) is nonempty, then there is at least one generator y of H+(S) (or
H-(S)) along which the matter tensor is past (future) stochastic as y approaches its
past (future) affine parameter limit. Further, the Einstein equations hold.

Proof. If H(S) is empty, then S is a noncompact Cauchy surface for (M, g); by
Penrose’s theorem, the space-time is null geodesically incomplete. If H(S) is nonempty,
 SINGULARITIES AND CAUSALITY VIOLATION 27

for all partial Cauchy surfaces S, there must exist a partial Cauchy surface S’ and a
generator y(s) of H--(S) (say) along which

If y(.r) were future complete, then by the Einstein equations, we have along y(s)

(7.2)

where H(s) ~(RahKQKh+ 2~~). By Lemma 9, condition (l), and Eq. (2.3). y(s)
must have an infinite number of conjugate points, since zeros of (2.3) correspond to
conjugate points of y. But this is impossible becauseof Lemma 11 and the achonality
of H-(S-j. Thus y(s) must be future incomplete. (Were y(s) in H+(S) rather than in
H-(S) we could proceed in the same way to deduce the past incompleteness of y.)
If condition (3’) rather than (3) holds, then Eq. (7.2) still holds, and the argument
proceedsas before. In all cases,the space-time is null incomplete, since y(s) is a portion
of a null geodesic. 1

THEOREM 9 (Generalized Hawking-Penrose Theorem). Space-time (M. g) is

not timelike and null geodesically complete [f
( 1) R,,JPKb 13 0,for every non-spacelikevector K”;
(2) the generic condition is satisjed;
(3) there exists at least ohe of the,foIlowing:
(i) a compact achronal set without edge.
(ii) a closed trapped surface,
(iii) a point p such that on every past (or every future) null geodesic,from
p the dil’ergence 6 ?f the null geodesics,fromp becomesnegative,
(4) the space-timeis asymptotically deterministic, and the Einstein equationshold.
Condition (4) can be replaced by condition (3’) of Theorem 8. The proof of
Theorem 9 is similar to the proof of Theorem 8, and so will be omitted. m
There is additional evidence that H(S) which arisesfrom causality violation can be
nonsingular only if it begins in empty space.

DEFINITION. The matter tensor will be said to be future vanishing along a future
complete causal geodesicsegment y(s) if there are no numbers sI , s2(s? ID s,) in any
affine parameter interval of the form [sO, + co) such that s, :- s,, and

(7.3)

where K” is the tangent to y(s). Past vanishing matter tensors are defined analogously.
28 FRANK J. TIPLER

PROPOSITION 4. Let S be a partial Cauchy surface. Then the matter tensor must
befuture vanishing along every future complete generator of H-(S), provided the weak
energy condition and the Einstein equationshold.
Proof. Suppose that y(s) is a future complete generator of H-(S) along which
Tab is not future vanishing. Then there are affine parameter values s, < s1 < sZ
such that (7.3) holds. By the Einstein equations, the weak energy condition, Eq. (2.3)
and Proposition 1, y(s) must have a pair of conjugate points, since zeros of (2.3)
correspond to conjugate points of y(s). But this is impossible because of Lemma 11
and the achronality of H-(S). Thus Tab must be future vanishing along y(s). (If
“future” is replaced by “past” and “H-(S)” is replaced by “H+(S)” in Proposition 4,
the resulting statement is also true.) 1
Thus there can be no “patches” of matter near the upper bound of the affine para-
meter of a future complete generator y(s) of H-(S), for (7.3) would be expected to
hold in this case. Note that requiring the matter tensor to be not future vanishing along
a future complete null geodesic segment y(s) is a much weaker restriction on the tensor
than requiring it to be future stochastic along the segment; if we have T,,,KaKb 3 a
on one affine parameter interval ( 5%- s1 [ > b, the matter tensor will not be future
vanishing, provided the interval is sufficiently near s = + co. If y(s) intersects one
proton near infinity, Tab will not be future vanishing. However, either requirement
imposed on y(s) would make it impossible for y(s) to be a generator on H-(S), since
in either case y(s) would have a pair of conjugate points.
Not only must the matter vanish as s --j + co along a future complete generator
of H-(S); we can also show that the tidal force components of the Weyl tensor cannot
approach a nonzero limit as s -+ + co.

PROPOSITION 5. Let y be a future complete null geodesic segment with tangent

vector Ka and a$ine parameter s. Suppose that there exists a parallel propagated
pseudo-orthonormal frame with E, = K in which some components of the tensor
KcKd&Cblcdre& satisfy
lim inf [ KcKdK&b~ed[rK~~ j # 0. (7.4)
3++m
Then if the weak energy condition and the Einstein equations, y(s) cannot be a
generator of H-(S), where S is a partial Cauchy surface. (There is a similar proposition
for H+(S).)
Proof. Suppose on the contrary that y(s) is a generator of H-(S). If the limit (7.4)
is to be nonzero, then there must be a value s,, of the affine parameter for which some
component of the tensor KcKaK&bjcd[eKfl is nonzero for all s E (so, + co). Now in
a pseudo-orthonormal frame, this implies that some component of CmcdnKCKd
is nonzero in the same interval [HE, p. 1011.
Let h,, be the component of Cmdhnwhich is nonzero along y(s) (and satisfies
lim inf,,,, [ h,, ( + 0; that is, 1h,, [ > c > 0 for all s E (so, +co)). Now Eq. (2.2)
(dlds) o,,, = --Cmdna- k,,, = h,, - eo,, (2.2)
 SlNGULARITlES AND CAUSALITY VIOLATION 29

tells us h,, gives rise to a shear unln of a geodesic congruence about y(s). This equation
is a linear, first-order ordinary differential equation which is easily integrated, giving

where A,,L, is a constant (the solution can be found in any elementary book on ordinary
differential equations; for example, Hildebrand [26, p. 71).
We want to prove that under the above assumptions y(s) has a pair of conjugate
points between s0 and s = + co; we will assume that there are no conjugate points
in (s,, , + co) and derive a contradiction. If there are no conjugate points in (s(, , t- x ).
then 0 > 0 in some interval (sl , $ co), s1 3 s,, , by the weak energy condition the
Einstein equations, and Eq. (2.1). Thus the function

is monotone increasing. Furthermore, f(s) must diverge as s ---+ I- w. since if it

converged (a monotone function must either diverge or converge: it cannot merely
not have a limit). we would have

(7.5)

But

which implies that there exists an affine parameter se > s1 for which / (T,, / .A 0
for all s E (sZ, T cc) and further lim,s,+, inf / omn i > 0 (in fact, lim,,,, inf 1 u’nrn’
+ co). Using this together with the weak energy condition, we find that lim,+,,, .I
inf H(S) > 0, where H(s) = i(RabKaKb + 20~) = fr(RahKaKb -C omnumn). Thus
lim s->i-u s2H(s) = + co, which by Lemma 8 implies that all solutions to (2.3) have
an infinite number of zeros, hence conjugate points, in (s, . i- co), contrary to
assumntion.
30 FRANK J. TIPLER

Therefore, we must have lims++mf(.s) = + 03. Write

From Eq. (7.5) we can obtain

= c 1K / - ) A;‘, 1 cffs).

Using 1’Hopital’s rule, we get

.fzlf(s’) ds’ _ lim (d/ds) sIl efcs’) ds

hi K = lim ef’s’ - ,$-cc (d/ds) ef(@
s-m
ef’s’
= lim
s-m efts’f’(s)

By assumption B > 0. By Eq. (2.1), and the weak energy condition, 8 is monotone
decreasing {i.e., dO/ds < 0). Thus lim,,, I/@ must either exist or diverge to + co.
Therefore,
lim 1 umlz 1 2 c hi K - / AL, / li+z e-f(s)
s+a

since f(s) diverges, the second term vanishes. The first term is bounded below by some
positive number since 0 < co. Hence limsem / a,, 1 > 0, and this implies
J” H(s) ds > J” oz ds = + co. By Lemma 9 and Eq. (2.3), this means an infinite
number of conjugate points in (sO + co)- Thus whether or not f(s) diverges, y(s)
must have an infinite number of conjugate points; this contradicts the assumption
that it has no conjugate points. Thus there must be a pair of conjugate points on y(s),
which would be impossible if y(s) were a generator of H-(S), by Lemma 11, and the
achronality of H-(s). We conclude (finally!) that the y(s) cannot be a generator of
=w I
Proposition 5 does not claim that the entire Weyl tensor must vanish as s + + co;
first of all, it refers only to the tidal force components, and second, it does not eliminate
the possibility that CnaJn4 could fluctuate about zero, Cman4 could be alternately
positive and negative as s -j + co; for there is no Weyl tensor analog of the weak
energy condition to prevent this behavior. However, if Weyl tensor did fluctuate in
this manner, we would expect to see some matter present, because a varying gravi-
tational field should give rise to particle creation, not much, to be sure, but by Propo-
sition 4, we do not need much to prevent a nonsingular I?(S). Furthermore, it seems
unlikely that the tidal force components of the Weyl tensor would vanish in all
pseudo-orthonormal frames with E, = K without the entire tensor vanishing also.
Since the curvature tensor is determined by the Ricci tensor (or matter tensor) and
 SINGULARITIES AND CAUSALITY VIOLATION 31

the Weyl tensor, the theorems and propositions proved in this section collectively
strongly suggestthat H(S) can be generated by portions of complete geodesicsonly
if the curvature vanishes in the region where H(S) begins. As argued earlier, this
seemsvery unlikely if H(S) arises from a violation of the causality condition, since
in general CTL would occur only if curvature were present to “tip over the light
cones.”
However, violation of the causality condition is not the only pathology which
could give rise to a Cauchy horizon; for example, H(s) could be due to singularities
which arise at “infinity,” as in the Reissner-Nordstrom solution (Fig. 8) and the

I\r=O singularity

FIG. 8. The Reissner-Nordstriim solution, a space-time in which H(S) begins at intiity

Kerr solution. But the Cauchy horizon in the Reissner-Nordstrom solution is thought
to be unstable [HE, p. 161; Simpson and Penrose [27]). In fact, the generic condition
is not satisfied on any null geodesicy(s) such that a segment of y(s) is a generator of
H(S). Thus what the theorems and propositions of this section probably show is
that H(S) is generated at least in part by incomplete null geodesics,whether or not
causality violation occurs. (Unless S is a “bad” partial Cauchy surface in empty space.
as discussedin Ref. [lo].)
There is another singularity theorem, due to Hawking. which does not need a
causality assumption.

HAWKING’S THEOREM [HE, p. 2721. Sparetime is nor finrefike geodesica& rom-

plete (f
(1) RabKaKb> 0,for every non-spacelikevector K”;
(2,) there exists a compact spacelike three-surface S (without edge);
(3) the unit normals to S are everywhere converging (‘or everywhere diverging)
Oil s.

j9j/IO8/1-3
32 FRANK J. TIPLER

The problem with this theorem is that condition (3) is much too strong; the space-
time must be contracting (or expanding) everywhere. It is possible to generalize this
theorem to arbitrary initial data in a closed universe in the following sense. Suppose
a closed universe contracts (not necessarily everywhere), causality violation occurs
in the regions of high density, and then the universe reexpands. We can show that,
provided the causality violation begins in a finite region, the “bounce” must be
accompanied by singularities.

THEOREM 10. Let B be a four-dimensional region of a space-time such that the

boundary of B is the disjoint union of two compact partial Cat&y surfaces, S and S’.
Supposethat

(1) R,JPKb b 0 for every null vector Ka;

(2) the generic condition holds;
(3) d(S, [int D+(S)] n B) isJinite;
(4) the causality condition is violated in B.
Then the space-time is timelike geodesically incomplete.

(The proof is similar to that of Theorem 5, and so will be omitted.) 1

It was tacitly assumed above that if causality violation arises from regular initial
data, then it is possible to find a partial Cauchy surface S whose Cauchy horizon lies
on the boundary of a chronology-violating set V, at least in the region where the
causality violation begins. That is, it was assumed that l+(V) n H+(S) was not empty
and contained a geodesic segment of nonzero affine parameter length. However,
finding such a partial Cauchy surface might be difficult. Therefore, I shall restate
here some of the theorems, propositions, and definitions given above in a form which
does not depend on the existence of H(S), but only on the structure of l+(V). The
physical justifications for the theorems propositions, etc., are the same as those
for their analogs given above. For example, we would expect matter to be present in
the region where causality violation begins. This suggests the following analog to
“asymptotically deterministic”

DEFINITION. A space-time (M, g) is said to be asymptotically causal if there is no

point p for which I+(p) = M and each point q such that I+(q) n I-(q) f IZI, we have

1~2 [inf T,,K”K”] > 0 (7.6)

on at least one of the null geodesic generators y(s) of f+(q), where a is the past limit
of the afhne parameter along y. We then have

THEOREM 11. Suppose that a space-time (M, g) has a point q such that I+(q) n
I-(q) # o, but I+(q) # M. Then (M, g) is not null geodesically complete, provided
 SINGULARITIES AND CAUSALITY VJOLATION 33

(1) the Einstein equationshold;

(2) the weak energy condition holds;
(3) at least one of the following holds:
(a) (M, g) is asymptotically causal,
(b) the mutter tensor is past stochastic along at least one of the generators
of 1 k(q).

Proof. Since Z!-(q) + M, i+(q) is nonempty. Further, f+(q) is achronal and is

generated by null geodesic segments which have no past endpoints since q E Z+(q) n
Z-(q) = int[Z+(q) n Z-(q)]. If condition (3a) holds, then along at least one of these
segments, y(s). the matter tensor must satisfy (7.6). If y(s) were past complete, then by
(1) and (2) we would have along y(s)

(7.7)

where H(s) = $(R,,KUKb + 20~). By Lemma 9, conditions (l), (2), and Eq. (2.3),
y(s) must have an infinite number of conjugate points since zeros of (2.3) correspond
to conjugate points of y(s). But this is impossible because of Lemma 11 and the
achronality of f+(q). Thus y(s) must be past incomplete.
If condition (3b) holds, then Eq. (7.7) still holds, and the argument proceeds as
before. 1
Theorem 9, the Generalized Hawking-Penrose Theorem, has an obvious analog.
Conditions (4) and (4’) of this theorem are respectively replaced by conditions (5)
and (5’).
(5) If (M, g) does not satisfy the chronology condition, then (M, g) is asymp-
totically causal and the Einstein equations hold.
(5’) If (M, g) has a point y for which Z+(q) n Z-(q) is nonempty, then Z-(q) .c-M,
and there is at least one generator y of Z+(q) along which the matter tensor is past
stochastic as y approaches its past affine parameter limit. Further, the Einstein
equations hold.
The proof of this analog to Theorem 9 is omitted since it is essentially the same
as the proof of Theorem 9. m
PROPOSITION 6. (analogous to Proposition 4). Suppose there is a q E M such
that Z+(q) n Z-(q) is nonempty and Z-(q) # M. Then the matter tensor must be ,fiture
vanishing along every jurture complete generator of Z-(q), provided the weak energjs
condition and the Einstein equations hold. (Note that no generator of f-~(q) has a ,future
endpoint.) 1
The analog to Proposition 5 is obtained by replacing “a partial Cauchy surface S”
with “a point 4 E M such that Z+(q) n I-(r/) # 3 and Z-(q) # M,” and L‘H-(.~)”
with “Z-(q).”
34 FRANK J. TIPLER

8. CONCLUSION

The main purpose of this paper is to answer the question: “Is it possible to construct
a time machine?” If by this we mean: “Is it possible to evolve CTL from regular
initial data everywhere using known materials?,” then the answer is almost certainly
No!
The import of the singularity theorems proved herein is this: CTL cannot in general
arise in finite regions from regular initial data without some matter first passing
through such extreme conditions that we cannot trust our knowledge of material
behavior. The singularity theorems can be believed unless the weak energy condition
is violated or the manifold picture of space-time breaks down. Most proposals [28]
for the former would first require the matter to reach a density of 1O54g/cm3, and the
latter is not expected to occur until the matter density becomes log4 g/cm3. These
numbers are respectively 40 and 80 orders of magnitude above the most extreme
matter densities with which we are familiar (nuclear densities, and it’s debatable how
familiar we are with nuclear matter). Clearly, matter at these densities must be
considered “unknown material.”
Nevertheless, this result does not preclude the existence of CTL which arise from
regular initial data; it is quite possible that the singularity on H+(S) is restricted to a
very small region of space-time, with most of the matter forming the time machine
avoiding the singularity. Indeed, a singularity is the future end of the event horizon
when a black hole evaporates [5,29], and most of the matter in the universe avoids
this singularity.
Furthermore, Hawking has argued [29] that what comes out of the singularity is
completely random; he contends that, assuming global causality, the singularity emits
with equal probability every configuration of particles compatible with such external
constrants as energy and angular momentum conservation. However, it seems to me
that if what comes out of a singularity is to be truly completely random, we should
allow for the possibility that CTL could “come out of” a singularity. That is, we
should extend Hawking’s “Randomicity Principle” beyond the limited domain of
particle emission and say that vitually any metric compatible with various external
constraints such as those listed above could arise from a singularity. Among these
metrics will be metrics containing CTL. Thus it is possible that CTL occur in the
regions of space-time containing black hole explosions. But conversely, CTL, if they
do occur, almost certainly must be associated with singularities, those “points” at
which our knowledge of physics breaks down.
“The demonstration that no possible combination of known substances, known
forms of machinery, and known forms of force can be united in a practicable machine
by which men shall [travel back in time], seems to the writer as complete as it is
possible for the demonstration of any physical fact to be.“’

1 The sentence in quotes is a slight modification of the conclusion of Newcomb’s classic paper
[30] proving the impossibility of heavier-than-air flying machines.
 SINGULARITIES AND CAUSALITY VlOLATlON 35

ACKNOWLEDGMENTS

I would like to express my deepest gratitude to my thesis advisor, Professor Dieter R. Brill, for his
extremely helpful comments and criticisms. I would also like to thank Professor Robert H. Gowdy,
and J. Isenberg, L. Lindblom, J. Nester, and P. Yasskin for helpful discussions.

REFERENCES
I. B. CARTER, Pi7y.r. Rw. 174 (1968), 1559.
2. G. W. GIBBONS AND R. A. RUSSELL-CLARK, Pl7ys. Ret-. Left. 30 (1973), 398.
3. F. J. TIPLER. Phys. Rev. D 9 (1974), 2203.
4. C. J. S. CLARKE, Conz~n. Math. P/zys. 41 (1975), 65. A cautionary remark is in order. Clark
shows only that some component of the curvature tensor diverges in a frame parallel-propagated
along any incomplete timelike curve contained in int b(S). His result is not strictly applicable
to most of the theorems of the present paper, which generally show that some of the lrlrllgenerators
of H l(S) are incomplete. However, I would expect p.p. curvature singularities to develop in this
case also.
5. B. S. DEWITT, Pl7ys. Rep. C 19 (1975), 295.
6. R. GEKOCH, Spacetime structure from a global viewpoint, i/l “General Relativity and Cosmology,
Proc. Int. School in Phys. ‘Enrico Fermi”’ (R. K. Sachs, Ed.), pp. 71 -IO?, Academic Press.
New York, 1971.
7. R. PENROSE, “Techniques of Differential Topology in Relativity,” Vol. 7 of the Regional Con-
ference Series in Applied Math.. Society of International Applied Mathematics. Philadelphia,
1972.
8. R. PENROSL, The structure of spacetime, ifr “Battelle Recontres” (C. M. Dewitt and J. A. Wheeler,
Eds.), Benjamin, New York, 1968.
9. S. W. HAWKING AND G. F. R. ELLIS, “The Large Scale Structure of Space-time.” Cambridge
Univ. Press, Cambridge, 1973.
10. F. J. TIPLER. Pl~>>.s.
Rev. Lett. 37 (1976). 879: Ph.D. thesis, University of Maryland. 1976 (available
from University Microfilms).
11. R. GEROCH, J. Mat/r. Pl7yx 8 (1967), 782.
12. R. GEROCH, Singularities, i/z “Relativity” (S. Fickler, M. Carmeli. and L. Witten. Eds.). p. 765,
Plenum. New York, 1970.
13. I. D. NOVIKOV AND YA. B. ZEL’DOVICH, Am7. Rer. Aswon. Astropl7ys. II ( 1973), 3X7.
14. C. W. MISNER. K. S. THORL’E, AL’D J. A. WHEELER. “Gravitation,” Freeman. San Franciszo,
1973.
15. P. HARTF*IAX. “Ordinary Differential Equations,” Wiley, New York. 1964.
16. A. WIXTER, Quwt. J. Appt. Math. 7 (1949), 115.
17. H. EPSTEIN, V. GLASER. AND J. JAFFE, Nuorto Ci777ento 36 (1965). 1016.
18. J. J. SAKUKAI, ‘*Advanced Quantum Mechanics,” Addison-Wesley, Reading, Mass., 1967
19. S. W. HAWKIUG, Phys. Rer. Lett. 17 (1966), 444.
20. An asymptotically flat space-time is said to be partir7ll.v as.v777ptotica/!~~ predictaide from a partial
Cauchy surface SifE)nh & I~;andD= ny + : for all generators h of .fi and all generators
:’ of .I -. A more detailed discussion of the meaning of this definition can be found in Ref. [IO).
21. B. CARTER, Phvs. Rer. 174 (1968). 1559; J. Gen. Rev. Guar. l(l971). 349.
22. R. GEROCH, Ph.D. Thesis, Princeton University, Princeton, 1967.
23. W. KUNDT. Corrrm. Mat/r. P/7ys. 4 (1967), 143; D. R. BRILL. Thoughts on topology change, in
“Magic Without Magic: John Archibald Wheeler” (J. R. Klauder. Ed.), Freeman, San Francisco,
1972; R. H. GOWDY, A7217. Physics 83 (1974). 203.
24. D. R. BRILL. A simple derivation of the general redshift formula. i/z “Methods of Local and
Global Differential Geometry in General Relativity” (D. Farnsworth, J. Fink, J. Porter, and A.
Thompson, Eds.), Springer-Verlag, Berlin, 1972.
36 FRANK J. TIPLER

25. G. F. R. ELLIS AND A. R. KING, Comm. Math. Z’ZzJx 38 (1974), 119.

26. F. B. HILDEBRAND, ‘Advanced Calculus for Applications,” Prentice-Hall, Englewood Clifts,
N. J., 1965.
27. M. SIMPXIN AND R. PENROSE,Znt. J. TZreoret. P&s. 7 (1973), 183.
28. F. W. HEHL, P. VON DER HEYDE, AND G. D. KERLICK, Phys. Rev. D 10 (1974), 1066.
29. S. W. HAWKING, Phys. Rev. D 14 (1976), 2460.
30. S. NEWCOMB, “Side-lights on Astronomy: Essays and Addresses by Simon Newcomb,” p. 345,
Harper, New York, 1906.