The class of continuous timelike curves determines the

topology of spacetime *
David B. Malament
Department of Philosophy, University of Chicago, Chicago. Illinois 60637
(Received 23 August 1976)
The title assertion is proven, and two corollaries are established. First, the topology of every past and
future distinguishing spacetime is determined by its causal structure. Second, in every spacetime the path
topology of Hawking, King, and McCarthy codes topological, differential, and conformal structure.

1. SUMMARY sumed to be temporally orientable and endowed with a

particular temporal orientation.
Suppose one has two spacetimes (M,g) and (M',g')
together with a bij ection f: M - M', where both f and Given subsets A and 0 of M with 0 open, r (A, 0) is
1"1 preserve continuous timelike curves; i. e., if 1': the set of points q in 0 such that there exists a future
1- M is a continuous time like curve in (M, g), then directed smooth timelike curve 1': 1- 0 (Where I~R is
foy: I-M' is a continuous timelike curve in (M',g'); connected) and points tv t2 E I such that t1 < t2 , y(tl) E A,
and symmetrically for f- 1 • We show that f must be a and y(t 2 )=q. r(A,O) is called the chronological future
homeomorphism. In this sense the class of continuous of A relative to O. The causal future of A relative to 0,
timelike curves in spacetime determines its topology. J+(A,O), is the union of An 0 with the set of points q
in 0 such that there exists a future directed smooth
The result is of interest because, at least in some causal curve (i. e" a smooth curve whose tangent vec-
sense, we directly experience whether events on our tors are everywhere nonvanishing, nonspacelike, and
worldlines are "close" or not. That experience alone, future directed) 1': 1-0 and points t1 , t2E I such that
it appears, allows a complete determination of topo- tl <t 2 , y(UEA, and y(t 2 )=q. Finally, the horismos
logical structure. The result also has two consequences future of A relative to 0, E+(A,O), is the set ~(A,O)
which are of independent interest. -r(A,O). These sets have duals r(A,O), J"(A,O), and
It is well known that in all strongly causal spacetimes E - (A, 0) which are defined analogously (substitute past
the Alexandroff topology is equal to the manifold topo- directed curves for future directed curves), I(A,O) is
logy.1 Hence, at least in strongly causal spacetimes, if the union r(A, 0) U r(A, 0), The sets J(A, 0) and
one knows of all points p and q whether it is possible E(A,O) are defined similarly.
that a particle travel from p to q, then one can recover If A={P}, we write r(p,O) instead of r(A,O) and
the topology of spacetime. The question naturally r(p) instead of r(p,M). Similarly for the other I, J,
arises whether the condition of strong causality is E sets. The relations qEI+(p,O), qEJ+(p,O), and
necessary for this recovery. We show that it is not. q E E+(p, 0) will sometimes be written as p« q(O),
The weaker condition of past and future distinguish- p<q(O), p-q(O). Furthermore, p«q(M), p<q(M), and
ability suffices. One has the follOwing result: If (M,g) p - q(M) will sometimes be written as p« q, p < q, and
and (M', g') are past and future distinguishing space- p-q.
times and if f: M - M' is a causal isomorphism (i. e. ,
a bijection where both f and f- 1 preserve the causal The I, J, E sets have the following basic properties. 3
connectibility relation «), then f must be a homeo- If qEr(p,o), then pEI-(q,O) and conversely (Similarly
morphism. But we also show that the assertion be- for the J and E sets). Both r (p, 0) and r(p, 0) are
comes false if the hypothesis of past and future distin- open. If p« q(O) and q < r(O), then p« r(O), Similarly,
guishability is relaxed to that of future distinguishability if p < q(O) and q« r(O), then p« r(O). If p - q(O), then,
(or past distinguishability) alone. if 1': [0,1]- 0 1S a future directed smooth causal curve
with ,(0) = p and y(1) = q, , must be a null geodesic.
A second consequence of our theorem is an improve-
ment of a result of Hawking, King, and McCarthy. 2 An open set 0 is com'ex iff given any two points p and
They define a path topology on spacetimes and prove q in 0 there is a geodesic ,: [0,1]- 0 with ,(0) = p,
that, in the presence of strong causality, the path to- ,(1) = q and, is unique (up to reparametrization). If 0
pology "codes" (standard) topological, differential, and is an open convex set, then, for all points p in 0,
conformal structure. We show that their hypothesis of J+(p,O)=CI[r(p,O)]=the closure in 0 of r(p,O); and
strong causality is unnecessary. Indeed their result is E+(p,O)=Bnd[r(p,O)]=the boundary of r(p,o) in O.
true of all spacetimes. {These assertions are false in general if 0 is not con-
vex. But J·(p,O)~cI[r(p,o)l and E+(p,O)CBnd[r(p,O)]
are always true.} Dual assertions hold for J- and E-.
2. STANDARD DEFINITIONS AND RESULTS The open convex sets form a basis for the manifold
In what follows a spacetime (M, g) is taken to be a topology; i. e., given any point p and any open set U con-
connected, four-dimensional smooth manifold without taining p, there is an open convex set 0 with p E 0 c:: U.
boundary M, together with a smooth pseudo-Riemannian A set A is achronal in 0 iff for all points p and q in
metric of Lorentz signature g. Spacetimes are as- An 0, it is not the case thatp«q(O).

1399 Journal of Mathematical Physics, Vol. 18, No.7, July 1977 Copyright © 1977 American Institute of Physics 1399

Downloaded 02 Apr 2013 to 128.233.210.97. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jmp.aip.org/about/rights_and_permissions
 A spacetime (M,g) is chronological iff it admits no condition obtains but with« replaced by < in (*). Final-
closed, future-directed smooth timelike curves. (M, g) ly, we say that y is a future directed null geodesic iff
is causal iff it admits no closed, future-directed smooth the above condition obtains but with (*) replaced by
causal curves.
tl't2EI and tl <t 2=">y(t 1 )-y(t2 ) (0).
A spacetime (M ,g) is future (resp. past) distinguish-
ing iff for all p and q: I+(p)=r(q) ~ p=q (resp. rep) Note that every future directed continuous null geodesic
=I-(q) ~ p=q). Equivalently, (M,g) is future (resp. can be reparametrized so as to become a (smooth)
past) distinguishing iff for all p in M and all open sets future directed null geodesic. (The corresponding as-
a containing p, there exists an open set 0 1 with sertions for continuous timelike and causal curves are
pEal ~ a such that no future (resp. past) directed false.) Dual definitions can be given for past directed
smooth timelike curve through p which leaves 0 1 ever continuous timelike (causal, null geodesic) curves.
returns to it. The sets r(A, 0), J+(A, 0), peA, 0) could be redefined
Finally, a spacetime is strongly causal iff, for all in terms of continuous curves, but doing so would not
points p and all open sets a containing p, there exists affect the resultant point sets. For example, p« q(O)
an open set 0 1 with p E 0 1 r;: a such that no future direct- (according to our definition involving smooth timelike
ed smooth timelike curve which leaves 0 1 (whether or curves) iff there is a future directed continuous time-
not if passes through p) ever returns to 0 1 , like curve y: 1- C and points t1 , t2 E / with 11 < t2 ,
y(t 1 ) = p, and y(t 2 ) = q.
If (M, g) is a spacetime and Or;: M is a connected
When there is no chance of confusion we shall not
open set, then we may think of (O,gla) as a spacetime
distinguish between curves y: 1- M and their point set
in its own right. If a is convex, (O,glo) is necessarily
images y(I]. Also, we shall sometimes refer, Simply,
strongly causal.
to continuous (causal, null geodesic) curves and it
These "causality conditions" can be ordered in terms should be understood that the curves are either future
of (strictly) increasing strength: or past directed.
strong causality
U 3. FROM TOPOLOGICAL STRUCTURE TO
future and past distinguishability
U DIFFERENTIAL AND CONFORMAL STRUCTURE
future (or past) distinguishability
U We shall prove that the class of future directed con-
causality tinuous timelike curves determines the topology of
tl
chronology spacetime. Having done so, it will follow automatically
that this class of curves also determines the differen-
The respective converSe implications are all false. tial and conformal structure of spacetime. This is all
If (M ,g) is a spacetime, the Alexandroff topology on that one can hope for since all conformally equivalent
M, TA' is the coarsest topology on M in which all sets Lorentz metrics on a manifold induce the same con-
rep) and I-(q) are open. The collection of all sets of tinuous timelike curves.
form rep) n I-(q) form a basis for TA' If T is the That differential and conformal structure will follow
(standard) manifold topology on M, then it is always on the heels of topological structure is a consequence
true that TAr;: T. But the condition TA= T is equivalent of:
to strong causality. Suppose (M, g) is strongly causal.
Then the condition that a set A C M be open (in is n Hawking's theorem 4 : Suppose (M,g) and (M' ,g') are
spacetimes and f: M - M' is a homeomorphism where
explicitly definable in terms of the relation «: A is
open iff, for all points p in A, there exist points rand r
both f and 1 preserve future directed continuous null
s in A such that p E r(r) n /-(s) c: A. geodesics. Then f is a smooth conformal isometry.

Given two spacetimes (M,g) and (M' ,g'), a bijection To avail ourselves of this result, we need a simple
f: lVl- M' is a smooth isometry iff f and r 1 are smooth, lemma.
and f* (g) =g'. f is a smooth conformal isometry iff f Lemma 1: Suppose (M ,g) and (M' ,g') are spacetimes
and r 1 are smooth, and there is a smooth nonvanishing and f: M - M' is a homeomorphism where both f and
map 0: M'-R such that f*(g) =02g ,. r 1 preserve future directed continuous timelike curves.
So far "causal structure" has been developed entirely Then both f and f- 1 preserve future directed continuous
in terms of smooth curves. For our purposes it is null geodesics.
essential to work with the larger class of continuous Proof; It suffices to observe that the future directed
curves. Suppose y: / - M is a continuous curve. We continuous null geodesics of a spacetime (M,g) can be
say that y is future directed and timelike iff, for all characterized in terms of its future directed continuous
to E I and all open convex sets a containing f(t o), there timelike curves and its topology.
exists an open (i. e., open in the relative topology on
I) subinterval Y r;: I containing to such that First, given any open set U and points p,q in U, we
have that q E Bnd[r (p, U)] iff for all future di rected
tE Y and t < to=">y(t)« y(to ) (0), continuous time like curves a: (0,1) - U, if aCto) = q for
(*) some to where 0 < to < 1, then there exist t 1 , 12 where
t E 7 and to < t=">y(to)« y (t) (0).
0< t1 < to < t2 < 1 such that a (t 1 ) <i r(p, U), but a (t 2 )
We say that y is future directed and causal iff the above Er(p,U).

1400 J. Math. Phys., Vol. 18, No.7, July 1977 David B. Malament 1400

Downloaded 02 Apr 2013 to 128.233.210.97. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jmp.aip.org/about/rights_and_permissions
 Next, note that if y: 1 - M is a continuous curve, then differential, and conformal structure of spacetime as
y is a future directed null geodesic iff for all to Eland given and abstracting a causal connectibility relation
all open sets 0 containing y(t o), there exists an open «, we ask if the construction can be turned "on its
set U ~ 0 containing y(t o) such that for all t1 , t2 E 1 with head" with the relation « construed as primitiveo It
t1 < '2' if y(t l ), y(t 2 ) E U then y(t 2 ) E Bnd[I+(y(t l ) , U)]./ turns out that it can be if the spacetime in question is
sufficiently well behaved in its causal structure. "Suf-
ficiently well behaved" means "at least past and future
4. THE PRINCIPAL RESULT AND ITS CONSEQUENCES distinguishing. "
Theorem 1: Suppose (M,g) and W',g') are space- If (M,g) and (M',g) are spacetimes, a mapj:M-M'
times and j: M - M' is a bij ection where both j and l r is a causal isomorphism iff j is a bijection and for all
preserve future directed continuous timelike curves. points p and q in M: p«q<:=:>j(P)«f(q). Our result fol-
Then j is a homeomorphism. (By Hawking's theorem j lows from the following lemma o
must also be a smooth conformal isometry.)
Lemma 3: Suppose (M,g) and (M',g') are past and
A proof of the theorem is given in the next section o future distinguishing spacetimes and that j: M - M' is a
causal isomorphism. Then j and j_l preserve future
As it is stated, the hypothesis of the theorem is
directed continuous time like curves 0

slightly stronger than necessary. It suffices that j and

r 1
take (past or future directed) continuous timelike Proof: Suppose y: 1- M is an arbitrary future directed
curves to (past or future directed) continuous timelike continuous timelike curve in (M, g). Suppose p = y(t o )
curves. 5 This follows immediately from the follOwing with to E I, and suppose 0' is an arbitrary open convex
lemma. set containing j(p). We must show that there exists
an open subinterval 7 ~ 1 with to E 7 such that
Lemma 2: Suppose (M,g) and (M' ,g') are spacetimes
and j: M - M' is a bij ection. Suppose further that both t E 7 and t < to=,>{j° y)(t)« f (p) (0'),
j and r 1 preserve continuous time like curves. Then
either: (a) Both j and r 1 preserve future directed con- t E 7 and to < t='>j(p)« (joy)(t) (0'). (* )
tinuous timelike curves, or (b) both j and 1 take future r
directed continuous timelike curves to past directed Since (M', g') is future distinguishing, there is an
continuous time like curves. open set U' with j(p) E U' ~ 0' such that no future direct-
ed timelike curve from j(p) which leaves U' ever re-
Prooj: Let p be any point in M. Suppose there are
enters. Letj(q) be any point in r(j(p), u'). Since
future directed continuous timelike curves y and a
j(p)«j(q), we must have p« q. So there must exist
through p such that jo y, but not joa, is future directed
an open convex set 0 with P E 0 ~ rl(qL Since,), is a
in (M' ,g'). Let y- be the "lower segment" of y with
future directed continuous timelike curve, there must
future end point po Let 0+ be the "upper segment" of a
exist an open subinterval II ~ 1 with to E II such that
with past end point p. Then the continuous time like
curve which results from "linking" y- with 0+ is one t ElI and to < t='>p« y(t) (oL
whose image under j is not a continuous timelike curve
We claim now that
at aU. This is impossibleo So at least as restricted to
continuous timelike curves through some particular p«y(t) (O)=>j(p)« (joy)(t) (0').
point in M, f either systematically preserves or system-
For, ifp«y(t)(O), wehavep«y(t)«q. Hencej(p)
atically reverses orientation.
«(joy)(t)«j(q). So there exists a future directed
Let A (resp. B) be the set of points in M at which j smooth timelike curve through j(p), if 0y)(t), and j(q)
preserves (respo reverses) orientation. We show A is in sequence. We know that this curve cannot leave U'
open. Suppose p is in A and p« q for some point q. between j(p) and j(q)o So we must have (fc y)(t)
Then there is an open set 0 with PE O~I-(q). Let y be E r(j (p), U') ~ r(j(p), 0').
a future directed continuous timelike curve with initial
point p and terminal point q. Suppose now there is a A parallel argument using past distinguishability
point rEO n B. Let a be any future directed continuous of (M', g') establishes that there is an open subinterval
time like curve with initial point r and terminal pOint 72 ~ 1 with to E I2 such that:
q. Then the result of linking y with a is not a continuous t E 72 and t < to ='>(jo Y)(0« j(p) (0'),
time like curve, but its image under j is a continuous
timelike curve. This is impossible since 1 preserves r Hence the set 7 ={t E IJt;;, to} U {t E l/t '" to} is an open
continuous time like curves. Therefore, 0 ~ A and so subinterval of 1 with to E T which satisfies (*)./
A is open as claimed. A symmetric argument establishes Thus we have
that B is open.
Theorem 2: Suppose (M ,g) and (lVi' ,g') are past and
It thus follows that j either systematically preserves future distinguishing spacetimes and f: M - M' is a
or systematically reverses the orientation of continuous causal isomorphism. Then j is a homeomorphism. (By
timelike curves. The same argument applies to 1 and, r Hawking's theorem j must also be a smooth conformal
of course, f preserves orientation iff 1 does too. / r isometry. )
We consider now the question whether the topological As was the case with Theorem 1, Theorem 2 can be
structure of spacetime can be recovered from its recast so as to be completely "time symmetric" in
causal structure. Rather than thinking of the topological, formulation. 5 Let T be the symmetric causal connect-

1401 J. Math. Phys., Vol. 18, No.7, July 1977 David B. Malament 1401

Downloaded 02 Apr 2013 to 128.233.210.97. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jmp.aip.org/about/rights_and_permissions
 ibility relation on spacetime points defined by pTq ~ P But j is a causal isomorphism. The important thing to
«q or q«p. Given two spacetimes (M,g) and (M',g')
a map j: M - M' is a symmetric causal isomorphism
notice here is that every point below the t == equator
has all points in both upper slabS in its chronological
°
iff j is a bij ection and for all points p and q in M: future.
pTq ~ j(p)Tj(q)o To recast Theorem 2 in symmetric
form, it suffices to prove the following lemma and in- This establishes that the condition in the hypothesis
voke Lemma 2. of Theorem 2 cannot be relaxed to future distinguish-
ability. A symmetric example (with excisions below the
Lemma 4: Suppose (M ,g) and (M' ,g') are past and
future distinguishing spacetimes and that j: M - M' is
°
t == equator) shows that it cannot be relaxed to past
distinguishability either.
a symmetric causal isomorphism. Then j and 1"1 pre-
serve continuous timelike curves. Finally, we use Theorem 1 to generalize a result of
Hawking, King, and McCarthy. 2 They define the path
One proves the lemma by compounding the construc- topolOKY on a spacetime to be the finest topology which
tions of Lemmas 2 and 3. We skip the argument as it is induces on all continuous timelike curves the same
somewhat tedious and involves no new ideas. topology induced on them by the standard manifold topol-
The following example shows that the hypothesis of ogy. Equivalently, if (M ,g) is a spacetime with A:: M,
past and future distinguishability in Theorem 2 (and A is open in the path topology on M iff given any con-
hence Lemma 3) cannot be relaxed to either future dis- tinuous timelike curve y: 1- M there exists a (stan-
tinguishability or past distinguishability alone. We give dard) open set 0 such that y[Il n A == y[I] no. Their
the example in a two-dimensional version to simplify interest in the new topology is motivated in part by the
matters. belief that, in some sense, we "experience" continuity
along future directed continuous timelike curves. The
Start with the two-dimensional plane carrying a standard topology, they claim, has no comparable
metric: physical significance.
Hawking, King, and McCarthy prove that given any
strongly causal spacetime (M,g), ifj: M-M is a
with respect to global Cartesian coordinates t, x. Next homeomorphism with respect to the path topology, then
form a vertical cylinder by identifying the point (t, 0) j must be a smooth conformal isometry But along the 0

with all points (t, 2n) for all n. Finally excise two way they prove the following:
closed half-lines: {(t ,x): x = 0 and t ~ o} and {(t ,x): x = 1 Lemma 52; If (i'v1, g) is a spacetime and f: M - M is a
and t?oO} (see Fig. 1.) Along the "equator" t=O the homeomorphism with respect to the path topology, then
metric reduces to the form ds 2 = dtdx and its associated both f and 1"1 preserve continuous timelike curves.
null cones are horizontal, pointing in the direction of
increasing x. But as I t I - <Xl, the cones "tip to the left" Thus it follows immediately that we have
and asymptotically approach the upright position they Theorem 3; If (M ,g) is an arbitrary spacetime and f.
have in Minkowski spacetime. Because of the excisions M - M is a homeomorphism with respect to the path
the spacetime is future distinguishing. But it is not past topology, then f is a smooth conformal isometry.
distinguishing. Every point on the t == 0 equator has for
its chronological past the entire region of the space- One can easily reformulate the theorem so as to be
time falling below the equator. parallel in form to Theorems 1 and 2. One simply takes
f; M - M' to be a path topology homeomorphism be-
Now let j be a bijection of the spacetime onto itself tween arbitrary spacetimes (M,g) and (M',K'). The
defined by conclusion is affected not at all.
( ) (t, x) if t < 0,
j: I.x - (t.x+l) ift""'O.

j leaves the "lower open half" of the spacetime fixed, 5. PROOF OF THEOREM 1
but reverses the position of the two upper slabs. j is
°
surely discontinuous along the t = equator; it "cuts"
continuous timelike curves which cross the equator.
If it were assumed that j preserves all continuous
curves, it would follow immediately that j is continuous.
Given any sequence {PJ converging to p, one could
find a continuous curve "threading" all the Pi in se-
quence and then p. Its image would have to be a con-
tinuous curve threading all the j(Pj) in sequence and
thenj(p). Hence {j(p)} would have to converge toj(p).
X~O Under our hypotheses, however, this construction can
v
only cope with sequences {Pi} which converge chrono-
excise excise logically to p. The problem is with those sequences
FIG. 1. {Pj} which converge to P but are locally spacelike re-
, toO lated to p •
. ~----
Our proof is rather long and so is divided into a se-
quence of lemmas. The crucial idea is this: To show
v
------ that j is continuous at p, one proves that one may as
well assume that j is continuous over a nice-looking

1402 J. Math. Phys., Vol. 18, No.7, July 1977 David B. Malament 1402

Downloaded 02 Apr 2013 to 128.233.210.97. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jmp.aip.org/about/rights_and_permissions
 region near p (Lemma E). Then one uses continuous
null geodesic segments in that "safe region" to char-
acterize the convergence of points to p. This does the
trick because (by Lemma 1 above) continuous null geo-
desics in the safe region are necessarily preserved by
f· FIG. 2.
In what follows LJ (resp. LJ') is taken to be the set of
points at which j (resp. j-1) is discontinuous.
°
Lemma A: If is an open set in M, 0' is an open
convex set in M', and j[O] C 0', then CM - f). °
Pro oj: Let p be any point in 0. To show j is continuous
at p, it suffices to show that given any open set lj' con-
taining j (p), j"1 [0' n V'] is open in M. Since 0' is con- tE Y and to < t:=;;.j(p) < (j y)(t) (0').0

vex, the spacetime (0' ,gIOl) is strongly causal. So the

Alexandroff topology on 0' is equal to the relative mani- Just as in the proof of Lemma B we can show that there
°
must exist an open set in M containing p such that
fold topology induced in 0' Thus V' n 0' is open in the
A lexandroff topology on 0'. But j I 0 : ~ 0' is certainly ° j[I(p, 0)] c:: 0'. By moving to a subset we may take to
be convex. We choose I c:: I containing to so that:
°
continuous with respect to the Alexandroff topologies on
° and 0'. So j"l[V' nO'] must be open in the Alexandroff
topology on 0. A fortiori j-1[ V' no'] is open in (the
tEl and t<to:=;;'oov(t)<p (0),

manifold topology on) M. / tE Y and to < t:=;;.p <yet) (0).

Lemma B: Given pin M, there is an open set in

M containing p such that I(p, 0) c:: M -f). (So f is at
° Now if yU) < p, then every continuous timelike curve
segment through yet) intersects I-(p, 0). Hence every
least continuous over "local futures and pasts. ") continuous timelike curve segment through (j 'Y let) 0

intersects I-{j(p), 0'). Thus (j c y)(t) E cliI-(j(p), 0')]

Prooj: Let 0' be an open convex set containingj(p). and therefore, since 0' is convex, (j0y)(tkJ-(j(p),O').
We show first that there is an open set containing p ° Thus the first half of (*) is established. The second
such that j[I+(p, 0)] c:: 0'. half is symmetric. Hence joy is a future directed con-
Suppose there is no such 0, Then given any open 0 1 tinuous causal curve, (The argument for j-1 is, of
containing p there must be a point PI in 01 such that course, symmetric.)/
Pl E rep, 01) but j(p) do' Since I-(p1' 01) is open, we
<
Lemma D: (il LJ is closed in M; fJ' is closed in M'.
can find an open set 02 c:: 01 containing p such that
02 c 1- (PI' OJ. There must exist a P2 in 02 such that (ii) For all p EM, P E f) iff f (p) E j)'.
P2Er(p,oz)c::r(p,Ol) butj(Pz)do'. Clearly Pz«P1(0). (iii) If P E fJ, then there is an inextendible future
Continuing in this way, we can generate a nested se- directed continuous causal curve through P fully con-
quence of open sets 01 ~ O2 2° 3 ' •• all containing p, and tained inj) .
a sequence of points {Pi} where, for all i, Pi r- 0i' PI • 1
Prooj: Suppose j is continuous at p. Let 0' be any
«Pi(Oi)' P«Pi(Oi)' but j(Pi) <;1 0' (see Fig. 2). Further-
more, we may choose the {oj so that they converge to open convex set containing j(p). Let be an open set
with p EO C::j"1[0']. Then, applying Lemma A, we have
°
P (Le., so that their intersection is {P}). Now we can
certainlv join p. 1 to p. with a continuous future directed that 0 c:: M -fJ. Thus M - fJ is open. Similarly M' - f)'
timelik~ curve ~+egme~t Yi contained in 0i' Linking is open. So CO 0

these segments together and adjoining the point p, we Suppose Pis inDo Then there exists a sequence {p!}
obtain a future directed continuous timelike curve y which converges to P and an open convex set 0' in M'
through P which "threads" all the Pi' By our construc- which contains j(p) but none of thej(Pf)' We can find
tion no initial segment of je y can intersect 0'. But this sequences {r l } and {Sf} converging chronologically to p
is impossible since j ' y is a continuous timelike curve from below and above respectively such that for each
through j (p). i there is a local future directed continuous time like
Therefore, as claimed, there is an open set con- curve Yi through PI with initial point r l and terminal
taining p-call it 0l-such that j[J+(p, 01)] CO'. Simi- paint Si' The only accumulation point of the ')If is p.
larly, there is an open set 0z such that j[I-(p, 02)] c:: 0'. Now {j(r!)} and {j(s/)} must converge to f(Pl, So
Let o=oln Oz. Then clearly, f[I(p,O)]C::O'. It now (passing to a subsequence if necessary) we may assume
follows by Lemma A that 1(p, 0) c:: M -f) .J that all JOYI begin and end in 0'. But since j(Pf) <i 0',
Lemma C: j and j-1 preserve continuous causal each of these curves JOYI must leave 0' as well. There
curves. will be a future directed inextendible continuous causal
curve A through j(p) every point of which is an ac-
Prooj: Let y: 1- M be a future directed continuous
cumulation point of the joy f .6 Since the only accumula-
causal curve in M with y(to)= p for some to E I. Let 0'
tion point of the Y i is P, it must be the case that
be any open convex set containing j(P). We must show
A - {f (p)} c:: fJ'. Since fJ' is closed, it follows that A
that there exists an open subinterval J c:: I containing to
C::f)'. Thus p Ef) :=;;. f (p) ED'. The converse is sym-
such that:
metric. So we have (ii). For (iii) we need only repeat
tEl and t < to:=;;.{j oy)(O <j(p) (0'), this past argument with respect to f{ p) and rl. /
1403 J. Math. Phys., Vol. 18, No.7, July 1977 David B. Malament 1403

Downloaded 02 Apr 2013 to 128.233.210.97. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jmp.aip.org/about/rights_and_permissions
 Lemma E: IfD * ¢, then there exists an open convex
set °withD n °*¢ such that:
(i) D is achronal in 0.
(ii) Through each point P inD n there passes a uni- ° FIG. 3.
que continuous null geodesic r,o such that r,o nOr:;;, D '
(iii) Given any continuous null geodesic r which in-
tersectsL) n 0, either rn Or:;;,L) or rn onD is a
singleton.
Proof: First note that (ii) and (iii) follows from (i) in it follows that these segments converge to fo Q-.
view of Lemma D. For (i) supposeL) *jZ) but no exists
satisfying the required conditions. Let 01 be any open
° Now recall that no point f (p /) lies within 0'. So,
though the f Qj converge to fO 0-, they must all leave
0

convex set meetingD with compact closure. By our 0' before reaching their respectivef(p/), Letf(q) be
assumption we can find points r 1 and S2 in 01 n D such any point of the null geodesic extension of fo 0-. We
that r l «Sl (° 1 ), Now let 02 be any open convex set where claimf{q)ED'. For suppose to the contrary that
r l E 02 r:; , l-(Sl> 01)' Repeating the argument with respect f (q) E M - L)', Then, since f (q) is a convergence point
to 02' we can find points r 2 and S2 in 02 n L) such that of {joQj}, q must be a convergence point of {OJ}. This
r 2 « S2(02)' Certainly S2« Sl (OJ. Continuing in this is impossible since q ri 0-.
fashion, we generate a sequence {SI} in 01 n D with
S i+l « S i (01) for all i, This sequence must have an ac-
In our construction we assumed that 0 satisfied the
cumulation point s. But now if we apply Lemma B to "not before P;" clause for an infinite subset of Q j , Drop-
s, we find that there must exist an open set contain-
ing S such that J+(s, 0) r:; , M - D. This leads to a con-
° ping that assumption, we have the following conclusion.
If 0- and 0+' are the respective lower and upper segments
tradiction since eventually all the s I must enter of Q, then either the future null geodesic extension of
]+(s, ° 1 ), /
f 00- or the past null geodesic extension of f Q+ is a
future directed continuous causal curve segment through
0

Proof of the Theorem: Suppose L) *¢ and

Lemma E. Let P be any point in D n with correspond-
is as in

ing r p' Clearly l(r p no, 0) r:; , M - L). There must exist
° ° f(P) lying withinj)'. But this is true of all future
directed continuous null geodesic segments; Q was
chosen arbitrarily. Thus, since f is a bij ection, it fol-
a sequence {PI} converging to P and an open convex set lows that there exist distinct future directed continuous
0' containingf(p) but none of thef(Pi). causal curves through f(P) lying withinj)'. Their pre-
Let Q be any future directed continuous null geodesic images under /"1 must be distinct future directed con-
segment through P distinct from r,o which is sufficiently tinuous causal curves through plying withinL), But this
"short" that f 0 Q is fully contained in 0'. There exist contradicts our assumption thatLJ is achronal in 0.
continuous null geodesic segments Q i within 0, passing Thus, j) is empty, and, hence, ,0' is empty as well, /
through PI respectively, which converge to Q in the
sense that every open set which intersects Q intersects
ACKNOWLEDGMENTS
eventually all Qj • We may choose {Q;} so that it has no
convergence points off Q, Eventually all Q; enter I wish to thank Professor Robert Geroch for several
r,
l(r p 0, 0) and hence M -j). It follows from Lemma suggestive discussions, and Professor Howard Stein for
catching a number of errors in an earlier version of
E (iii) that, for eventually all i, Q/1 j) is either empty or
a singleton. The intersection point of Q i withD (if there this paper.
is one) comes either "before Pi'" at PI itself, or "after
PI'" Without loss of generality we may assume that
there is an infinite subset of {QI} in each member of IE. H. Kronheimer and R. Penrose, Proc. Camb. Phil. Soc.
which the intersection point withL) (if there is one) does 63, 481 (1967).
not come before Pi' Now let Qj be the "lower-half" of 'IS. W. Hawking, A. R. King, and P.J. McCarthy, J. Math.
Q i with future end point Pi included. By moving to a sub- Phys. 17, 174 (1976).
3Proofs of these and subsequent claims can be found in S. W.
sequence we can thus find a sequence of continuous null

erties (see Fig. 3):

°
geodesic segments {Qi} in with the following prop-
Hawking and G. F. R. Ellis. The Large Scale Structure of
SPacetime (CambridgeU.P., Cambridge, 1973); R. Penrose,
Techniques of Differential Topology in Relativity (SIAM,
Philadelphia, 1972).
(i) {Qi} converges to the lower half Q- of Q, but has <I A proof is given in Hawking, King, and McCarthy (Ref. 2).
no convergence points off Q- • The theorem is not formulated in exactly this from, but the
argument carries over intact.
(ii) For each i, Qj n L) r:; , {PI}' 5This version of the theorem is applicable to all temporally
orientable spacetimes whether or not a particular temporal
From (ii), Lemma C, and Lemma 1, it follows that orientation is distinguished.
each image curve f Qj is a continuous null geodesic
0 GHawking and Ellis (Ref. 3) prove this in detail in their
segment in M'. From (0 and the fact that Q- -{P}r:;;, M -L), Lemma 6.2.1.

1404 J. Math. Phys., Vol. 18, No.7, July 1977 David B. Malament 1404

Downloaded 02 Apr 2013 to 128.233.210.97. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jmp.aip.org/about/rights_and_permissions