Professional Documents
Culture Documents
In this chapter we shall introduce the causal boundary of a suitable Lorentz manifolds or space-
times. A short introduction can be found in [H-E] or a detailed one in [G-K-P].
We shall consider only time oriented Lorentz manifolds, which are future and past distin-
guishing, thus:
De…nition 1 A time oriented Lorentz manifold (M; g) is future and past distinguishing if
for any p; q 2 M
I + (p) = I + (q) () p = q , I (p) = I (q) .
De…nition 2 A space-time (or Lorentz manifold) (M; g) is asymptotically simple if the fol-
lowing condition holds:
2
Remark 3 The condition g = ge is the de…nition of ge and g to be conformal met-
rics. In case of conformal metrics the causality of the vectors do not change. The geodesics will
change, however one can prove that the lightlike-geodesics remain the same, see equation (8) in
the appendix or e.g. Appendix D in [W].
Remark 5 For the sake of simplicity we will identify (M ) and M thus we will assume hereon
f.
that M M
Proposition 6 In case of an asymptotically simple and empty Lorentz manifold the connected
components of the boundary @M are lightlike hypersurfaces if dimM 6= 3.
Proof. In Appendix A we calculate the Ricci curvature r in terms of the conformal metric
ge = e2f g, see equation (10) : Substituting f = ln we have our case.
n o
r (X; Y ) = re (X; Y ) + (n 2) Hess ^ln + d (ln ) d (ln ) (X; Y )
2
(n ^
2) grad (ln ) ge (X; Y ) + e (ln ) ge (X; Y )
e
g
1
To eliminate the ln we have
g (ln ) ; V = V (ln ) =
ge grad 1
V( ) = 1 g ;V
ge grad for every vector …eld V , thus
g (ln ) = 1 grad
grad g ;
g
^
Hess ln (X; Y ) = g
g (ln ) ; Y
e rX grad = ge rX grad
;Y = ge X 1 g +
grad 1 g ;Y
rX grad =
X( )Y ( ) ] (X; Y ) = ] (X; Y )
2 + 1 Hess 1
2 (d d ) (X; Y ) + 1 Hess
(d (ln ) d (ln )) (X; Y ) = X (ln ) Y (ln ) = X( )Y2 ( ) = 12 (d d ) (X; Y )
g (ln ) ; grad
ge grad g (ln ) = 12 ge grad g ; grad
g
e (ln ) = P Hessln E fi "i = P
fi ; E Efi ( )E
fi ( )
2
]
+ 1 Hess fi ; E
E fi "i = 1
2
g ; grad
ge grad g +
i i
2
1 e = 1
2
g
grad + 1 e
e
g
Thus
X ( )Y ( ) 1 ] 1
r (X; Y ) = re (X; Y ) + (n 2) 2
+ Hess (X; Y ) + 2 (d d ) (X; Y )
1 2 1 2 1e
(n 2) g
grad ge (X; Y ) + g
grad + ge (X; Y )
2 e
g 2 e
g
1 ] 1 2 1e
= re (X; Y ) + (n 2) Hess (X; Y ) + (3 n) g
grad ge (X; Y ) + ge (X; Y ) :
2 e
g
1
If %; e
%X are the Ricci endomorphisms, i.e. g (% (X) ; Y ) = r (X; Y ) and then ge 2 % (X) ; Y =
r (X; Y ). Thus in the above equation
1 1e g 1 2 1e
ge % (X) ; Y = ge e
% (X) + (n 2) rX grad + (3 n) g
grad + X; Y
2 2 e
g
f; z 2 @M . Thus if n 6= 3 then
for all X 2 Tz M
2
g
grad g ; grad
= ge grad g = 0 holds at every z 2 @M: (1)
e
g
2
Our goal is to describe the topology of @M in case of an asymptotically simple and empty
manifold.
De…nition 7 By property (4) in de…nition 2, every inextendable lightlike geodesic has a future
and past endpoint in @M . Let J + be those points of @M which are future endpoints and J
those points of @M which are past endpoints of such lightlike geodesics.
Remark 8 A) J + [ J = @M ;
B) J + \ J = ;;
C) J + and J are connected if dim M 3;
D) Through every point p 2 @M there is a unique lightlike geodesic @M , which will be called
a generator.
Now a simple continuity argument yields C). If L+ T M is the submanifold of all the future
directed lightlike vectors, then every v 2 L+ has an open neighbourhood Uv L+ such that
every inextendable lightlike geodesic exp (t w) ; w 2 Uv ; t 2 Iw R, has its future endpoint
on the same connected component of J + . By property (4) in de…nition 2 this component is
unique for every inextendable lightlike geodesic. Thus every vector in L+ corresponds to a
unique connected component of J + and this mapping is locally stable. As L+ is connected if
dim M 3 we have that every inextendable lightlike geodesic must end on the same connected
component, thus there is only one connected component of J + . A similar argument proves the
connectedness of J .
By A), B), C), if p 2 J + , then there is a neighbourhood Up M f for which J + (q; Up ) \ M =
+ +
;; 8q 2 J \ Up . As J is lightlike by proposition 6, there is a unique lightlike integral curve
J + through p. If \ Up is not a lightlike geodesic, then there would be points q1 ; q2 2 such
that q2 2 I + (q1 ) : As J + (q1 ; Up ) \ M = ; and q2 2 @M thus would contradict to q2 2 I + (q1 ).
But this would contradict to q2 2 I + (q1 ), thus must be a lightlike geodesic.
We want describe the topology of J + ; J and M . We will need some lemmas …rst. We
recall the following proposition.
Lemma 10 Let : [0; 1] ! M be a future directed lightlike geodesic with (0) ; (1) 2 @F
where F is a future set. Then (t) 2 @F , 8t 2 [0; 1] :
3
For a future set, we can divide @F into four disjoint sets. For every q 2 @F there may or
may not exist points p 2 @F \ J (q) and f 2 @F \ J + (q). The di¤erent possibilities are the
following:
@FN : the set of those points for which 9p; 9f . In this case there is a causal curve from p to
q and from q to f which must be an unbroken lightlike geodesic as p and f are achronal.
@F : the set of those points for which @p; 9f . By the achronality of @F , the causal curve
from q to f is a lightlike geodesic. Moreover, by lemma 10 this geodesic lies in @F and its past
extension leaves @F and never returns. Thus q is a past endpoint of a lightlike geodesic in @F
which is past inextendable in @F .
@F+ : the set of those points for which 9p; @f . Then as above Thus q is a future endpoint of
a lightlike geodesic in @F which is future inextendable in @F .
@F0 : the set of those points for which @p; @f . Then q has a neighbourhood Uq such that for
every point r 2 Uq \ @F the points q and r are acausally related, i.e. @F0 is an open and acausal
hypersurface.
Note that by the achronality of @F we have that @F \J (q) = @F \(J (q) I (q)) = @F \
def
E (q) where E (q) = J (q) I (q) is the past horismos. Similarly @F \J + (q) = @F \E + (q).
The following characterization of the four disjoint set is useful.
Lemma 11 Let Uq be a neighbourhood of q 2 @F , where F is a future set. Then
(i) if I + (q) I + (F Uq ) then q 2 @FN [ @F+ ;
(ii) if I (q) I (M F Uq ) then q 2 @FN [ @F :
Proof. As we can change future and past and (M F ) is a past set, it is enough to prove (i).
If (i) holds for Uq then we can take smaller geodetically convex neighbourhood K Uq of q for
which K is compact and (i) will hold for K. Let xn 2 I + (q) \ K; xn ! q a sequence. As (i)
holds for K we have a point yn 2 F K and a past directed timelike curve from xn to yn : This
curve lies completely in F and intersects @K the …rst time at some point pn . Thus pn 2 F \ @K
and there is a past directed timelike geodesic from xn to pn which must lie also completely in F .
A suitable subsequent of pn will converge to a p 2 @K \ F as K is compact. By the convergence
we will have a past directed causal geodesic from x to p 2 F . Since I (q) \ F = ; as q 2 @F ,
we have that must be lightlike, p must be in @F and by lemma 10 @F as well. Thus
q 2 @FN [ @F+ .
As an accumulation and application of the above result we have:
Corollary 12 If K is a closed set, then the boundary of the future set F = I + (K) is an achronal
imbedded closed hypersurface and for every point q 2 @F , we have q 2 @FN [ @F+ . Thus @F K
is a union of lightlike geodesics, called generators, and every point of @F K is an interior
point of a generator or a future endpoint of a generator. Note that the past endpoints of the
generators can be only on J or on K.
De…nition 13 We call M causally simple if for every compact set K we have
@J + (K) = J + (K) I + (K) = E + (K) and @J (K) = J (K) I (K) = E (K) :
Thus J + (K) and J (K) are closed in M .
Lemma 14 An asymptotically simple and empty space is causally simple.
Proof. Let K be a compact set. As @J + (K) = @I + (K), by corollary 12 we have that @J + (K) =
(@I + (K) K) [ K. As @I + (K) K is a union of generators, we have to show, that every
generator has its past endpoint in K because then @I + (K) K J + (K), thus @J + (K)
J + (K) which implies @J + (K) = J + (K) I + (K). By corollary 12 the past endpoints of the
generators are in K or in J . Since it can not be in J we have (i).
4
Proposition 15 An asymptotically simple and empty space is globally hyperbolic.
Proof. Let be a volume element on M such that (M ) = 1. (This means every open set has
positive volume and it is a continuous measure.) Let
def def
f + (p) = J + (p) f (p) = J (p) :
lim f + (xn ) = \1 +
i=1 J (xi ) :
n
\p2 J + (p) = ;:
Trivially \p2 I + (p) \p2 J + (p). If p ; p 2 ; p << p << p+ then J + (p+ ) J + (p)
I + (p ) follows \p2 J + (p) \p2 I + (p). Thus
Since \p2 I + (p) is a future set, it must be open in M , but \p2 J + (p) is closed in M . Thus
\p2 J + (p) equals M or ;: By strong causality \p2 J + (p) 6= M .
Therefore, f + (p) ! 0 if p tends to the future of . Similarly, if is past inextendable
and p 2 tends to the past of , then f (p) ! 0. Thus there will be a unique point on an
inextendable timelike curve, for which f + (p) = f (p) ; p 2 . Thus every causal curve intersects
the set
def
H = p 2 M j f + (p) = f (p)
exactly once. By the above facts it is easy to see, that H is a topological hypersurface, which
is spacelike (acausal). But then H is a Cauchy hypersurface which implies that M is globally
hyperbolic, see [O’N], pp. 422-423, corollary 39.
5
Proof. Let be a lightlike generator of J + , let p 2 : Consider the set J p; M \ M thus
def
the causal past of p relative to M and its intersection with M . Since P = I p; M \ M is a
past set in M , by the analogue of lemma 11 if q 2 @ (P ) then q 2 PN [ P : Since all the future
endpoints of the generators of @P are in J + ; the future endpoint of the generator through q
must be p. But then q 2 J p; M \ M thus I p; M \ M J p; M \ M . The converse is
trivial, thus I p; M \ M = J p; M \ M hold, so J p; M \ M is closed.
Let us assume that J + C; M . If p1 ; p2 2 ; p1 p2 , then J p2 ; M \ M
I p1 ; M \ M , since a causal curve to p2 with the segment of between p1 and p2 is a
broken causal curve, except we take , but this is not in M . This gives that
\p2 J p; M \ M: = \p2 I p; M \ M
\p2 J p; M \ M = ; (2)
\p2 J p; M \ M \ C 6= ;
[p2 J p; M \ M = [p2 I p; M \ M
Theorem 17 Let (M; g) be an asymptotically simple and empty space which is orientable and
dim M = 4. Then J + and J is topologically R S2 and M is topologically R4 .
Proof. The …rst step is to examine the set of the lightlike geodesics in M . We will give two
topological model of this space which will be the same.
In the proof of proposition 15 we saw, that there is a Cauchy hypersurface H M . Thus
every inextendable lightlike geodesic intersects H exactly once. We can identify an inextendable
lightlike geodesic with the point where it intersects H and the future lightlike direction of the
geodesic at that point. The future lightlike directions at a point p 2 H yield the future light cone.
6
Let X be a future directed timelike unit length vector …eld on M . If we consider Sp H Tp H the
unit sphere, then the elements of the translated sphere X (p) + Sp H give all the future directed
lightlike directions. Thus Sp H can be identi…ed with the future directed lightlike directions in a
natural way. Therefore
def
SH = [p2H Sp H
can be identi…ed with the set of inextendable lightlike geodesics. An inextendable lightlike
geodesic is identi…ed with v 2 Sp H; p 2 H if \ H = p and 0p is parallel to v + X (p).
Moreover for every 2 Sp H; p 2 H there is a unique inextendable lightlike geodesic through p
with direction v. This identi…cation gives a topology on the set of inextendable lightlike geodesics.
Roughly speaking the inextendable lightlike geoesics 1 and 2 are close if their intersection with
H is close to each other and their directions are close there. The manifold SH is an S2 bundle
over H since Sp H u S2 . Now since M is orientable and time oriented, these yield an orientation
of H. But every oriented 3-manifold is parallelizable, see ***, which gives that its unit bundle
SH is a trivial S2 bundle over H. Thus the …rst topological model of the set of inextendable
lightlike geodesics in M is
S2 H.
The second model is similar, but instead of H we consider J + . Every inextendable lightlike
geodesic has a future endpoint in M f which is on J + . If the geodesic has endpoint p 2 J + then
its past direction is in the past lightlike cone which will denoted by Lp M f Tp M f. The past
2
directed lightlike directions can be identi…ed with S but there is one direction, let it denoted
by lp ; which is the past directed lightlike direction in Tp J + ; i.e. the unique tangent lightlike
direction of J + at p since Tp J + is lightlike by proposition 6. The direction lp does not correspond
to a lightlike geodesic in M; because the lightlike geodesic corresponding to this tangent lightlike
direction remains in J + by property (P 2). We have Lp M f lp = S2 fpointg t R2 . Therefore
the endpoints of inextendable lightlike geodesics in M and their past directions can be identi…ed
with the elements of the bundle
n o
[p2J + f
Lp M lp = J + e R2
which can be a twisted bundle, see ***. As J + is connected, by remark 8, M is on one side
of J + and M is orientable we have that J + is orientable as well. As above J + must be then
parallelizable since it is an orientable 3-manifold.
It is clear that for every p 2 J + and lightlike direction v 2 Lp M f lp there is a unique
lightlike geodesic in M with future endpoint p and direction v. Thus we have a second model of
the set of inextendable lightlike geodesics in M . This gives an other topology on the set where
two lightlike geodesics 1 and 2 are close if their future endpoint on J + and their past directions
there are close.
We have to show that the two topology on the set of inextendable lightlike geodesics is the
same. If we consider an inextendable lightlike geodesic in M which future endpoint is p then we
can extend it beyond p in M f. Let e the extended lightlike geodesic. It intersects the hypersurfaces
+
H and J transversally, thus it has a compact segment between this two hypersurfaces. By the
continuity of the variations of e it is easy to see, that if take a small variation of the intersection
point e \ H and the lightlike direction of e there, then intersection of the corresponding lightlike
geodesic with J + and its lightlike direction there will be close to e \ J + and to the direction of
e at e \ J + . This will yield that if U is a neighbourhood of e in one of the models, then we can
construct a neighbourhood V of e in the other model such that V U . This gives that the open
sets de…ned by the two models will be the same on the set of inextendable lightlike geodesics in
M.
7
Thus we have that
S2 H t J + e R2 :
We can simplify the right side because if we take a point x 2 M and its causal future
@J + C; M in the closure of M then by lemma 16 we have that every lightlike generator of J +
intersects @J + C; M exactly once. Since every point of J + is on a generator which intersects
def
K = @J + C; M \ J + we have that J + is a lightlike line bundle over K. Since we have past
and future directions, this lightlike line bundle is trivial, thus
J+ = K R.
Therefore we have
S2 H t (K R) e R2 = K e R3
Now we will introduce ideal points of space-times, which extends the above idea for more general
cases. However, this de…nition won’t always yield a nice result, i.e. the boundary won’t be a
smooth manifold in the general case.
First we introduce a set with the help of future and past sets, then we show how the elements
of this set correspond to the ordinary and ideal points of the manifold.
In this chapter we also assume that our Lorentz manifold M is future and past distinguishing.
Note that a past (or future) set is open, as I (P ) is open. We will prove the results hereon
only for past sets, for future sets we will have dual theorems.
The notion refers to the set of all points of the curve, thus for a curve : D ! M where
D R is its domain: I ( ) = [t2D I ( (t)) :
Proof. Assume that P = I ( ) for some timelike curve2 : D ! M . The proof goes by
contradiction. Suppose that P = Q [ R, where Q and R are past sets and proper subsets of P .
As they are proper subsets Q " R and R " Q. Thus there are points
q2Q R; r 2 R Q: (3)
As q; r 2 P = I ( ) we have points on the curve (tq ) ; (tr ) with tq; tr 2 D for which q 2
I ( (tq )) ; r 2 I ( (tr )). But then either (tq ) (tr ) or (tr ) (tq ). We can assume
that (tq ) (tr ) ; thus q; r 2 I ( (tr )). As (tr ) is in Q or in R which are past sets3 , we
have that both q and r are in Q or in R which contradicts (3).
2 The …rst part of the proof is also valid for causal curves whith a slight modi…cation.
3 If is a causal curve then as (tr ) is in the open set I + (q) \ I + (r), there is a point c 2 I + (q) \ I + (r) \
I ( (tr )), which is in one of the past sets Q or R, thus q; r 2 I (c) follows that both q and r are in Q or R
which is a contradiction.
8
Let P be an IP. For the other direction we …rst prove that for any two points p; q 2 P there
is an other point r 2 P such that p and q are in the past of r.
For any p 2 P consider the decomposition of P into two past sets
P =I P \ I + (p) [ I P I + (p) :
P =I P \ I + (p)
must hold. Thus for any two points p; q 2 P from q 2 P = I (P \ I + (p)) follows that there is
a point r 2 P \ I + (p) for which q 2 I (r). Moreover by r 2 I + (p) we have that p 2 I (r) as
well.
Now let p0 ; p1; : : : a countable dense set in P . Then by the above facts there is a point r1 2 P
such that p0 ; p1 2 I (r0 ). There is a point r2 2 P such that r1 ; p2 2 I (r2 ) : Thus in general
there is a point ri 2 P such that ri 1 ; pi 2 I (ri ). By the construction r1 < r2 < : : : which gives
a timelike curve joining the points of this sequence. As pk 2 I (rk ) we have that pk 2 I ( )
for every index k 2 N.
Now let p 2 P , as P \ I + (p) is an open set, there is a point pk 2 P \ I + (p). Since pk 2 I ( )
we have that p 2 I (pk ) I ( ) ; which follows
P = I ( ):
Remark 20 The above proposition shows, that in de…nition 18 we can use the equivalent form,
that the set P = I ( ) are called IP where is a timelike curve.
If is a timelike curve which has a future endpoint p, then it is trivial that I ( ) = I (p).
This suggests the following de…nition.
Remark 22 As we stated in the footnote in the proof of proposition 19, if is a causal curve,
then I ( ) is an IP. By proposition 19 there is a timelike curve such that I ( ) = I ( ) :
The following technical lemma will be useful and in some sense it will characterize those point
of a past set, which are "negligible" in some sense.
Lemma 23 Let S be a set and p be a point for which I (p) I (S). Assume that p has
a neighbourhood N such that I (S N ) = I (S), then there exist a point q 6= p and a past
directed causal geodesic from q to p for which I (p) $ I (q) I (S).
Proof. By shrinking N if necessary, we can assume that the closure N is a compact geodetically
convex neighbourhood. Let pn 2 I (p) be a sequence for which pn ! p. Then we can choose an
other sequence rn 2 S such that pn 2 I (rn ) because pn 2 I (p) I (S). Let qn 2 @N be the
…rst intersection of an arbitrary past directed timelike curve from rn to pn with the boundary
@N , thus pn qn rn 2 S; qn 2 @N . As @N is compact we can assume that qn ! q 2 @N .
9
Clearly q 6= p and p q: Since N was geodetically convex, there is a unique past directed causal
geodesic from q to p. Moreover p 2 J (q) ) I (p) I (J (q)) = I (q). As p 6= q the past
distinguishing property of M yields I (p) $ I (q).
Let x 2 I (q) than q 2 I + (x) thus there will be a point qn 2 I + (x). But then x << qn
rn 2 S follows that x 2 I (S) which concludes the proof.
Proof. For part a) assume on the contrary that I ( ) is a PIP thus I ( ) = I (p? ) but
p? is not the future endpoint of : Then p? has a neighbourhood N which is disjoint from ,
therefore I ( N ) = I ( ). Thus by lemma 23 there is a point q 6= p? for which I (p? )
I (q) I ( ). As q 6= p? and the Lorentz manifold is future distinguishing we must have
I (p ) $ I (q) I ( ) contradicting I (p? ) = I ( ) :
?
In corollary 12 we proved that the most of the boundary of a future set (corresponding to a
compact set) is ruled by lightlike geodesics. We will show a similar results in the next proposition.
Proof. This trivially follows from the fact that if N is a geodetically convex, suitably small neigh-
bourhood of q, and qn ! q a converging sequence, then for the relative past sets I (qn ; N ) !
I (q; N ) holds. Thus if x 2 I (q) then there is a timelike curve from q to x and a point
y 2 I (q; N ) on the curve for which x << y << q. By the convergence, there will be a qn for
which y 2 I (qn ; N ) holds. As q 2 @P we can choose the converging sequence qn 2 P . Therefore
the above argument shows that x 2 I (qn ; N ) P if qn is suitably close to q.
Proposition 26 Let P be a past set. Its boundary is ruled by future endless lightlike geodesics
if and only if P is a union of TIPs.
Proof. First let us suppose that @P is ruled by future endless lightlike geodesics. Let p 2 P
and a future endless timelike curve through p. If intersects @P at some point q and is the
future endless lightlike geodesics through q then by the above lemma 25 we have p 2 I (q)
I ( ) P . By theorem 24 I ( ) is a TIP. Thus
I ( i N ) = I ( i ) ; 8 i ; i 2 I:
10
Therefore
Thus by lemma 23 there exists point x+ and a past directed causal geodesic segment from x+
to x such that I (x+ ) P . As x is on the boundary of the past set P the geodesic segment
cannot be timelike, since I (x+ ) P . Therefore must be lightlike and x+ 2 @P . Assume
that we extend to a maximal geodesic on @P . It cannot have a future endpoint x? since the
above argument shows there is a future directed lightlike geodesic segment e @P from x? to
an y 2 @P . The segments and e can be joined at x? and there is no break, because then [ e
would be a broken causal curve from x to y which would yield x 2 I (y) P which would be a
contradiction. Thus @P must be future endless.
As we saw in the case of asymptotically simple Lorentz manifolds the boundary points
J + ; J are the endpoints of inextendable lightlike geodesics, see de…nition 7. First we will
show in this easier case the correspondence between the causal boundary points (J + [ J ) and
the TIPs and TIFs of the manifold. However, we will need an extra condition.
If M is an asymptotically simple manifold then
def
M = M [J
f:
is its closure in M
Thus from hereon,
De…nition 27 If p 2 J + then let I (p) be the set of all points of the timelike curves, which
future endpoint in M is p. Similarly we can de…ne I + (p) if p 2 J .
well. Therefore there is a timelike geodesic (with respect to Mf; ge which lies in M , ends at
p and intersects @M transversally. But by remark 3 this is also a timelike curve in M . Since
I (p), this shows that I (p) is non-empty and by its de…nition it is a past set.
As I ( ) is a TIP by theorem 24, therefore I (p) is the union of TIPs. Thus, proposition 26
yields that the boundary of I (p) is ruled by lightlike geodesics, which are future endless in M .
It will be true, that all the lightlike geodesics on the boundary of I (p) will have the same
endpoint p, which will be proved later.
11
Proof. First, we prove the inclusion
I ( ) I (p) :
If x 2 I ( ) then there is a point y 2 for which x 2 I (y) : Let be a timelike curve from
x to y. Let yp be the segment of from y to the future where its future endpoint p is also
included. The curve [ yp is a causal curve with respect to both metrics g and ge which is not
a lightlike geodesic and joins x and p. Since [ yp M except at the endpoint, we have a
variation of this curve which is timelike, joins x with p and it is in M , see ***.4 Thus x 2 I (p)
by de…nition.
For the inclusion I (p) I ( ) let x 2 I (p). Then there is a timelike curve ' M which
starts at x and its future endpoint is p. First we show that there is a timelike smooth curve from
x in M which future endpoint is p and meets @M transversally. To prove this we use the proof
of remark 28. That proof showed, we have a timelike geodesic with respect to ge which lies in
M , ends at p and meets @M transversally. Let V f be a geodesically convex neighbourhood
M
of p in Mf, thus there are no conjugate points in V with respect to the metric ge. Let z 2 \ V .
The timelike geodesics from z with respect to the metric ge cover a neighbourhood of p and if
they are close enough to they intersect @M transversally. Thus if W V f is a suitably
M
small neighbourhood of p, then for every q 2 W there is a timelike geodesic q : [0; 1] ! M with
respect to ge such that q (0) = z; q (1) = q. Let : [0; 1] ! M f be the timelike geodesic joining
z and p, thus (t) 2 M if t 2 [0; 1). Now let I ^ (r; V ) denote the chronological future/past of r
in the manifold (V; ge) similarly J ^(r; V ) denotes the relative causal future/past. If q 2 W \ '
then for every t 2 (0; 1] since z 2 I ^ (t) ; V and p 2 I + ^
q (t) ; V the timelike geodesic
q
intersects @ I + ^
q (t) ; V for some parameter st 2 (0; 1). Thus for every parameter t 2 (0; 1]
f; such that
there is a parameter st 2 (0; 1) and a future directed lightlike geodesic t : [0; 1] ! M
Note, that t (0) ; t (1) 2 M for every t 2 (0; 1], but t can leave M , thus intersect @M . Since
z 2 M , the geodesic t M if t is close enough to 0. Moreover t is continuous in t. Therefore
if 1 is not in M , then 1 must intersects @M . Thus there is a …rst parameter t? 2 (0; 1] for
which t? \ @M 6= ;. But then t? cannot intersect @M transversally by the continuity of t in
t. Therefore t? is a lightlike geodesic, which starts in M and is tangent to @M contradicting
to property (P 2). Thus 1 must lie completely in M . Since 1 is a lightlike geodesic connecting
q 2 ' and (s1 ). This 1 [ j[s1 ;1] is a broken causal curve which joins q and p, moreover it is
in M except at the endpoint p where the curve meets @M transversally. Since x 2 I (q) it is
easy to …nd a smooth timelike curve in M starting at x which future endpoint is p and the last
segment is which ensures the transverality of this timelike curve.
This yields that there is a suitably small neighbourhood U f of p such that for all the
M
f which starts at x, intersects @M transversally,
points y 2 U , there is a the timelike curve in M
if it leaves M it never returns, moreover it ends at y. But since the endpoint of is p, we can
take a point y 2 \ U . Therefore the above timelike curve joining x and y is entirely in M and
is also timelike with respect to g. Thus x 2 I ( ) which concludes the proof.
Proposition 31 For every p 2 J + the set I (p) is a TIP and conversely if P is a TIP and
P 6= M then P = I (p) for one and only one p 2 J + .
4 We have to be cautious at this point since the variation must remain in M . The timelike parallel vector…eld
along [ yp which gives the direction of the variation must be transversaal to @M to p and inward pointing.
12
Proof. If p 2 J + then there is a lightlike geodesic in M which future endless in M but its
future endpoint in M is p. By theorem 24 I ( ) is a TIP. By the above lemma 30 we have
I (p) = I ( ) : Thus I (p) is a TIP.
For the converse direction if P is a TIP, then
P =I ( )
I + (r; N ) \ P = ;:
Since N was geodetically convex in M f and the causal relations are conformal invariants, the
def
boundary C ? = @I + (r; N ) = @J + (r; N ) is a smooth lightlike hypersurface except at the vertex
r of the lightcone. The manifold J + is also a smooth hypersurface. By our assumption (P 2)
on J , the lightlike geodesic is transversal to J + . Thus the smooth hypersurfaces J + and C ?
are transversal and therefore the intersection J + \ C ? is locally a smooth (n 2)-dimensional
submanifold at p, which is necessarily spacelike by (P 2). As M is on one side of J + and C ? is
transversally intersecting it at p, it is easy to see the following.
Let x 2 I (p; N ) \ M suitably close to p. Then any timelike future inextendable curve from
x in N must intersect J + [ C ? and therefore leave P:
ÁBRA***
Now as p is the future endpoint of and I ( ) P there must be a point x 2 I ( ) which
is in I (p; N ) \ M arbitrary close to p. As x 2 I ( ) P , we have by (4) that x can be joined
to a point of by a future directed timelike curve in P . But x can be also connected to a point
of N by a future directed timelike curve in P by (5). This curve leaves N thus it is must
intersect J + [ C ? , therefore leave P which is a contradiction.
Since the future endpoint of is p lemma 30 yields that
I (p) = I ( ) = P:
Now assume, that there is an other p0 2 J + ; p0 6= p for which I (p0 ) = P . Then we could
repeat the above argument replacing with p0 to get a contradiction.
Note that in the above proof we proved also that for any future endless lightlike geodesic
2 @P its future endpoint is p: Because can have only one future endpoint, which is the future
endpoint of all of the future endless lightlike geodesic 2 @P . The above proposition also shows
that if b is an other timelike future endless curve for which P = I b , then it has must have
a future endpoint in M and this must be p.
***ELLENPÉLDA ÁBRÁKKAL P2 nélkülire
Now as we see in an asymptotically simple space-time if the property (P2) holds, then the
boundary points, the ideal points, correspond to TIPs and TIFs. Our goal is to consider the
13
abstract set of PIPs, TIPs, PIFs,TIFs and give a topology of this set to get a topological space
c
M which would be equivalent to M . By that we mean, that in most of the cases, e.g. if M is
c
asymptotically simple, the constructed M would be the manifold itself with the causal boundary
we would expect. Moreover, we will require that an embedding : M ,! M ] exists as well.
We give some basic ideas here as this topic has many open questions.
Let M be a Lorentz manifold (or a space-time) which is future and past distinguishing.
Without a property, which implies this distinguishing, we cannot even reconstruct M itself, as
it may happen in the most extreme case that we have only one IP and IF which is M itself.
De…nition 32 Let M # denote the set of all the IPs of M " and denote the set of all the IFs.
As we saw earlier for each p 2 M we have a PIP and PIF (I (p) and I + (p)). It is clear that we
have to identify these two points of M " and M # . To do this consider the set M " [ M # . For every
point p 2 M let us identify the points I (p) 2 M # and I + (p) 2 M " . After these identi…cation
on M " [ M # we get the set M ] ; where for a point P 2 M # (or P 2 M " ) we write P 2 M ] for
the corresponding point in M ] . By the identi…cation, if p 2 M then (I (p)) = (I + (p)) . Thus
the set M ] naturally includes M and the TIPs and TIFs:
As the IPs and IFs are open sets of M we have to describe the interior of these sets in M ]
and call them open sets of M ] .
De…nition 33 Let A 2 M # , then
def
Aint = P 2 M ] j P 2 M " ; P \ A 6= ;
, if B 2 M " then
def
B int = P 2 M ] j P 2 M # ; P \ B 6= ;
. In both cases Aint and B int will be considered as open sets of M ] .
If we consider asymptotically simple space-times the above de…nition is the most natural that
we can think of, see ÁBRA***.
Using only that the above sets are open, we can reconstruct the open sets of M in M ] ,
because the following holds. Let p 2 M consider the sequences p+ +
i 2 I (p) and pi 2 I (p)
where pi ! p the sets I + pi \ I p+ i are open sets of M and yield a base of the topology of
M at p. Thus, if we consider for every p 2 M the above family I + pi \ I p+ i of open sets.
Then the topology they generate (see below) coincides with the topology of M .
By generation we mean that if F = fA j 2 g is a family of sets of M where is an index
set, then the -algebra they generate is the family of sets
def
F = fA 1
\ \A n
jA i
2 F; i 2 Ng [ f[ 2 A j g:
I.e. we take all the …nite intersections and …nite or in…nite unions. If
F in = Aint j A 2 M # or A 2 M "
are open then F in should be also open as well and it gives back all the open sets of M . It’s trivial
that they all the elements of F in are open and for every U M open set for every point p 2 U
there is a base element for which + pi \ I p+ i U thus the union of such base elements for
in
all the points of U yields U and is in F by de…nition.
The problem is that at the boundary points in case of an asymptotically simple space-time
F doesn’t contains natural open sets if the boundary is lightlike, see ÁBRA***. To solve this
problem we de…ne also the exteriors of I (p) ; p 2 M and hope that it will be enough.
14
De…nition 34 Let A 2 M # , then
def
Aext = P 2 M ] j P 2 M " ; and for all S M for which P = I + (S) we have I (S) * A :
The formulation is more di¢ cult here, because we must exclude the boundary points.
Let F ex = Aext j A 2 M # or A 2 M " then we de…ne the open sets of M ] as the sets of
F ex [ F in . From now on we consider M ] with this topology as a topological space.
There is still a problem as some boundary points in the asymptotically simple case are both
in M # and in M " , typically in case of a timelike boundary. Such doubly represented points can
not be separated by open sets of M ] yielding that M ] as a topological space is not a Hausdor¤
space.
Therefore we take the "smallest possible" indenti…cations among the points of M ] which
make M ] a Hausdor¤ space with its topology. One can prove that if R M ] M ], 2
]
are relations such that M =R is a Hausdor¤ space for each 2 , then R := \ 2 R is also a
relation for which M ] =R is a Hausdor¤ space.
c def
In most of the cases M = M ] =R with the topology we de…ned above is M with the causal
c
boundary and one can prove that we have a natural : M ,! M dense embedding, see [G-K-P].
However it turns out, that in some cases this de…nition is not working properly, see Rácz [R],
where an alternative de…nition is given, which yields better results in case of the Taub space-time.
The above constructions only yield the topology but not the di¤erential topology of the causal
boundary. In [C] Chrusciel shows that if there is a "nice" extension a Lorentz manifold which
yields a causal boundary, then there is a maximal extensions of its boundary which is unique.
However we still do not know when and how this extension exists. We cannot construct from M
the this extension and its di¤erential properties.
An possible way to get some results is to assume global hyperbolicity as it is done in some
papers and achieve results in special cases.
3 Applications
Let us see some examples, how the causal boundary looks like in well known space-times.
First let us Consider the Schwarzschild space-time and how it is achieved. It models a lonely
star or black hole without any further objects. Since there are no other objects we require that
this space-time is static:
15
In contrast with the Robertson-Walker space-time, space is not warped in time but static.
As a star is usually spherically symmetric if no rotations are present, we want our to be
spherically symmetric. Thus if our star is centered at the origin of R = R3 then (R; gR ) is
spherically symmetric. Let us consider the spherical coordinates on R3 f0g = S2 R+ , where
gS2 and gR+ are the canonical metrics of the unit sphere and R+ , moreover (s; rb) denotes the
spherical coordinates. Then by the spherical symmetry it easily follows that
c2 (b
gR = b r) S2 (gS2 ) + bb2 (b
r) R+ (gR+ ) ;
where bb; b
c : R+ ! R+ are smooth functions. It is a natural assumption that the surface of the
spheres around the star are strictly increasing with the radius from 0 to 1. Thus changing the
parametrization of the radius rb to a suitable other parametrization r, we can assume that in the
new parameter b r) changes to r2 . This is some kind of normalization. Of course bb (b
c2 (b r) also
changes to some b (r) function.
An other consequence of the spherical symmetry is that the function a depends only on the
radius r, thus instead of a S2 R+ we can simply write a R+
Thus our metric on L = I S2 R+ is of the form
2
g= (a R+ ) I (gI ) + r2 S2 (gS2 ) + b2 (r) R+ (gR+ )
2) b Vb
Ric X; = 0;
3) Ric Vb ; W
c = Ric\
F (V; W ) c fc
g Vb ; W #;
where f # = ff + (d 1) gB (gradff;grad
2
f)
:
Thus the Ricci ‡atness at in…nity yields the far away from the star we have
A) RicB (X; Y ) 1r Hessr (X; Y ), if X; Y 2 X (B)
B) Ric\F (V; W ) c fc
g Vb ; W # , if U; V 2 X (F )
For the left hand side of A), since dim B = 2 the sectional curvature of the 2-dimensional
plane at every point b = (t; r) 2 B can be calculated and it is independent of the base in which
it is taken. So for the base …elds @t ; @r we have K(t;r) = gB (R(@ t ;@r )@r ;@t )
a2 (r)b2 (r) : Therefore in the base
…elds @t ; @r :
A1) RicB (@t ; @t ) = a2 (r) K(t;r)
16
A2) RicB (@r ; @r ) = b2 (r) K(t;r)
A3) RicB (@r ; @t ) = 0
For the right hand side of A) …rst we have to calculate the Levi-Civita connections with the
Koszul formula:
a0 (r) b0 (r)
= : (7)
a (r) b (r)
0
A simple rearrangement is a0 (r) b (r) + a (r) b0 (r) = 0 which is (a (r) b (r)) = 0, thus
a (r) b (r) const: Using the limit equation (6) we have that a (r) b (r) 1 thus
1
a (r) = .
b (r)
For the equation B) consider the parametrization ( ; ') 7! (cos sin '; sin sin '; cos '), where
( ; ') 2 [0; 2 ] [0; ]. Calculating the Ricci tensor we have
2
RicF = RicS = gS2 ;
which gives the eft hand side. For the right hand side: r = div (gradB r) = gB r@t b2@(r)
r
; @t 1
a2 (r) +
0
1 a (r) b0 (r) 0
a0 (r) b0 (r) Eq. (7) 0
gB r@r b2@(r)
r
; @r 1
b2 (r) = b2 (r) a(r) + b3 (r) 2 bb3(r) 1
(r) = b2 (r) a(r) b(r) = 2 bb3(r)
(r) :
0
Therefore, r# = 2 rbb 3(r) 1
(r) + r 2 b2 (r) : Now B) implies
b0 (r) 1
gS2 = r2 gS2 2 + 2 2 ;
rb (r) r b (r)
0 0
thus 1 = 2 rb (r)
b3 (r) +
1
b2 (r) = r
b2 (r) hence r
b2 (r) = r + C. Rearranging this we get
r ! 1
C
b (r) = 1+ :
r
17
De…nition 36 The manifold L = R (R+ f2M g) S2 with coordinates (t; r; s) equipped with
the metric
1
g = h (r) R (gR ) + h (r) R+ (gR+ ) + r2 S2 (gS2 )
is called Schwarzschild space-time. Consider PI = R (2M; 1) and PII = R (0; 2M ) with
the metric h (r) R (gR ) + h 1 (r) R+ (gR+ ). The warped products PI r S2 and PII r S2 are
parts of the Schwarzschild space-time which are called Schwarzschild exterior space-time
and Schwarzschild blackhole of mass M .
From now on P refers to PI or PII . The problem is that at radius r = 2M the metric is not
de…ned properly. To …nd the TIPs, TIFs for the causal boundary we have to deal with the parts
separately. By the geodesic equations for warped products, see [O’N] or [Sz] we have that:
It can be seen, that must be a pregeodesic of S2 , thus it is constant, or moves along a great
circle. But if const and is a lightlike geodesic, then is a lightlike curve in P: Since for
every TIP, TIF its boundary is a union of lightlike geodesics ending at the same boundary point
(for nice space-time at least). Therefore, it is natural to assume that for every TIP or TIF, there
are lightlike generators of the boundary, for which the part is constant. This yields that:
Remark 38 If C is the causal boundary of P then C S2 is the causal boundary of the Schwarz-
schild space-time.
Now we have only deal with PI and PII . To have a better understanding, we will introduce
a suitable coordinate transformation. For further details on the orbit, observers, etc. of the
Schwarzschild space-time see [O’N] pp.362-386.
4 Kruskal-Szekeres space-time
We want to have a nice space-time diagram of PI and PII . i.e. we seek a conformal transformation
of P such that lightlike geodesic will be represented by straight lines. In general, a space-time
diagram suppresses 2 dimension, by us the S2 part is suppressed if we draw only P . Moreover
the lightlike geodesics are lines of 45 .
First consider the plane R2 with coordinates (u; v). The function
def r
1 r
f (r) = (r 2M ) e 2M = rh (r) e 2M h(r)
r
1 r
is a strictly monotone increasing function on R+ since f 0 (r) = e 2M 2M > 0: Thus it has a
2M
smooth inverse f 1 : +
e ; 1 ! R . Let
def 2M def
Q = (u; v) j uv > R2 and H = f(u; v) 2 Q j uv = 0g :
e
The region Q is a connected component between two hyperbolas. The set Q H has 4 connected
components, which are QI ; QII ; QIII ; QIV see ÁBRA***.
18
1
0 F f (uv)
De…nition 39 Consider the metric gQ = 1 given in the base
F f (uv) 0
def 8M 2 1 r
@u ; @v on Q, where F (r) = r e
2M . The Lorentz manifold (Q; gQ ) is called the Kruskal
plane.
Proof. Let (t; r) = 2M ln uv ; f 1 (uv) denote the coordinates on PI [ PII and @t ; @r the
corresponding base …elds. Since @u f 1 (uv) = f 0 (f 11 (uv)) v replacing f 1 (uv) = r we have
r
@u f 1 (uv) = e1 2M 2M r v. It can be disturbing that r is a coordinate function on PI [ PII but
it is implicitly de…ned by f 1 (uv). This mixed notion will help to simplify the calculation.
r 2M r 2M
As T (@u ) = 2M u @t + e
1 2M r v@
2M
r and T (@t ) = v @t + e
1 2M
r u@r we can check that
T is isometric as the metric on PI [ PII is de…ned by gP = h (r) R (gR ) + h 1 (r) R+ (gR+ ).
We calculate
4M 2 r 2 4M 2 2 1
gP (T (@u ) ; T (@u )) = 2
h (r) + e1 2M v h (r) =
( ) u r2
2 2
4M r 2 4M 4M 2 r 2 4M 2
v2 2 h (r) + e 1 2M
h 1
(r) = v 2
h (r) + e 2M h(r) h 1 (r) = 0;
(uv) r2 f 2 (r) r2
4M 2 1 2Mr 2 4M
2
gP (T (@t ) ; T (@t )) = 2
h (r) + e 2
u2 h 1 (r) =
( ) v r
2 2
4M 1 2M 2
r 4M 4M 2 r 2 4M 2
u2 2 h (r) + e 2
h 1
(r) = u 2
2
h (r) + e 2M h(r)
2
h 1 (r) = 0;
(uv) r f (r) r
4M 2 r 2 4M
2
gP (T (@u ) ; T (@t )) = h (r) + e1 2M vuh 1
(r) =
uv r2
4M 2 r
h(r)
2 4M 2 1 8M r
h (r) + e 2M f (r) h (r) = e 2M h(r) = F (r) ;
f (r) r2 e
which shows that T is isometric. To show that is a di¤eomorphism, we can explicitly give
2 t 2 2
its inverse function. As Mt = ln uv we have e M = uv . Using that f 2 (r) = (uv) we get
t p t p
u = e 4M f (r) and v = e 4M f (r).
def
De…nition 41 The space-time K = Q f 1 (uv) S2 is called Kruskal space-time and we
denote the regions Ki = Qi f 1 (uv) S2 ; i 2 fI; II; III; IV g. The region v > 0 in K is called
def
the truncated Kruskal space-time which we denote by K + The set H = K Q is usually
referred as horizon, however this is misleading. If we remove H (0; 0) 2M S2 then it a
union of four lightlike hypersurfaces which are horizons.
19
Proposition 42 The regions (QI [ QII ) f 1 (uv) S2 and (PI [ PII ) r S2 are isomorphic by
id.
1
The mapping embeds the Schwarzschild space-time isomorphically into the truncated
Kruskal space-time and the horizon H \ K + joins the Schwarzschild black hole and the exterior
space-time. Note that H \ K + corresponds to t = 1: Thus they are joined at in…nity in the
Schwarzschild model, which we couldn’t see. Moreover, the singularity at r = 2M , t 2 R of the
Schwarzschild space-time is "mapped" to the sphere (0; 0) 2M S2 .
Now with the help of TIPs and TIFs, we can draw the Penrose space-time diagram of Q,
since it is 2-dimensional Lorentz manifold, thus all the lightlike curves are lightlike pre-geodesics,
see ÁBRA***.
By our remark 38 and proposition 42 we now understand the causal structure of the Schwarz-
schild space-time.
5 Appendix
5.1 A (conformal metric)
Let ge = g be conformal metrics. Then the angles and orthogonality, causality does not change.
From the Koszul formula
2e e X Y; Z
g r = X (e
g (Y; Z)) + Y (e
g (X; Z)) g (X; Y )) + ge ([X; Y ] ; Z)
Z (e ge ([X; Z] ; Y ) ge ([Y; Z] ; X) =
2e
g (rX Y; Z) + X ( ) g (Y; Z) + Y ( ) g (X; Z) Z ( ) g (X; Y ) :
By taking orthonormal base …elds E1 ; : : : ; En with respect to g and substituting these in the
above equality in Z we get that
e X Y = rX Y + 1 (X ( ) Y + Y ( ) X
r grad g (X; Y )) ;
2
1
where grad is taken with respect to g. Let S (X; Y ) = 2 (X ( ) Y + Y ( ) X grad g (X; Y ))
denote the symmetric (1; 2) tensor. Thus
e X Y = rX Y + S (X; Y )
r (8)
20
and
Thus
e (Z; X) Y
R = R (Z; X) Y + Hessf (Z; X) Y + Hessf (Z; Y ) X g (X; Y ) rZ grad f
Hessf (X; Z) Y Hessf (X; Y ) Z + g (Z; Y ) rX grad f
+S (Z; S (X; Y )) S (X; S (Y; Z))
= R (Z; X) Y + Hessf (Z; Y ) X g (X; Y ) rZ grad f
Hessf (X; Y ) Z + g (Z; Y ) rX grad f
+X (f ) S (Z; Y ) g (X; Y ) S (Z; grad f ) (Z (f ) S (X; Y ) g (Z; Y ) S (X; grad f ))
As
X (f ) S (Z; Y ) Z (f ) S (X; Y )
= X (f ) (Z (f ) Y + Y (f ) Z grad f g (Z; Y )) Z (f ) (X (f ) Y + Y (f ) X grad f g (X; Y ))
= Y (f ) (X (f ) Z Z (f ) X) grad f g (X (f ) Z Z (f ) X; Y ) :
And
Thus
e (Z; X) Y
R = R (Z; X) Y + Hessf (Z; Y ) X Hessf (X; Y ) Z
+g (Z; Y ) rX grad f g (X; Y ) rZ grad f
+Y (f ) (X (f ) Z Z (f ) X) grad f g (X (f ) Z Z (f ) X; Y )
+g (grad f; grad f ) (g (Z; Y ) X g (X; Y ) Z)
21
Now we must be careful with the contraction. The left hand side must be contracted in an
orthonormal base of ge but the right hand side must be contracted with respect to an orthonormal
def
base of g. If E1 ; : : : ; En are orthonormal base …elds with respect to g and "i = g (Ei ; Ei ),
fi = p1 Ei ; i = 1 : : : ; n are
then j "i = g (Ei ; Ej ) = 1 ge (Ei ; Ej ) = ge p1 Ei ; p1 Ej . Thus E
i
orthonormal base …elds of ge.
P e (Ei ; X) Y; Ei "i =
If we contract both sides with respect of g and E1 ; : : : ; En , then i g R
P e p1 Ei ; X Y; p1 Ei "i = P ge R e p1 Ei ; X Y; p1 Ei "i = re (X; Y ). This is sur-
i g R i
prisingly the contraction with respect to ge. P
P By the contraction If " i = g (E i ; Ei ), then at the contraction i Hessf (Ei ; Y ) g (X; Ei ) "i =
i g r g(X;Ei )"i Ei grad f; Y P= g (r X grad f; Y ) = Hess f (X; Y ). The contraction
P of g (Z; Y ) rX grad f
is Hessf (X; Y ). We have i g (X; Y ) g (rEi grad f; Ei ) "i = g (X; Y ) i Hessf (Ei ; Ei ) "i =
g (X; Y ) f . The contraction of Y (f ) X (f ) Z is nX (f ) Y (f ). For Y (f ) Z (f ) X we will get
Y (f ) g (grad f; X) = Y (f ) X (f ), for grad f g (X (f ) Z Z (f ) X; Y ) we will have X (f ) Y (f )
g (X; Y ) g (grad f; grad f ). The contraction of the last term g (grad f; grad f ) (g (Z; Y ) X g (X; Y ) Z)
is g (grad f; grad f ) g (X; Y ) (1 n)
Or in a shorter form
n o
2
re = r + (n 2) kgrad f kg f g (n 2) fHessf df df g :
This is the "new" Ricci tensor, in terms of the "old" metric ge: Since we can reverse the roles
e 2f ge = g we can express the "old" Ricci tensor in terms of the "new" metric ge if we change
f ! f: Thus
2 n o
r = re + (n ^
2) grad ( f) ^
( f ) ge (n ^
2) Hess ( f) d ( f) d ( f ) (10)
e
g
2 n o
= re + ff (n ^
2) grad (f ) ge + (n ^f + df
2) Hess df
e
g
References
[C] Chrusciel, P. Conformal boundary extensions of Lorentzian manifolds, Journal of Dif-
ferential Geometry 84, (2010), pp. 19-44
[G-K-P] Geroch R., Kronheimer E. H., Penrose R., Ideal points in space-time, Proc. R. Soc.
Lond. A. 327, (1972), pp. 545-567
[H-E] Hawking, S.W., Ellis G. F. R., The large scale structure of space-time, CUP, 1973
[O’N] O’Neill, B. Semi-Riemannian geometry,
22
[R] Rácz, I., Causal Boundary of Spacetimes. PhysicalL Review D Particles Fields Gravi-
tation and Cosmology, 36. (1987), pp. 1673-1675.
[Sz] Szeghy, D, Fizika és geometria I, short lecture notes.
[W] Wald R., General Relativity
23