c 2002)

General Relativity and Gravitation, Vol. 34, No. 7, July 2002 (

Gravitational Collapse: The Role of General

Relativity1

R. Penrose
Department of Mathematics, Birkbeck College - London
[Current address: Mathematical Institute, University of Oxford, 24–29 St. Giles,
Oxford, OX 13 LB, UK]
(Rivista del Nuovo Cimento, Numero Speziale I, 257 (1969))

Stars whose masses are of the same order as that of the sun (M ) can find a
final equilibrium state either as a white dwarf or, apparently, (after collapse and
ejection of material) as a neutron star. These matters have been nicely discussed
in the lectures of Hewish and Salpeter. But, as they have pointed out, for larger
masses no such equilibrium state appears to be possible. Indeed, many stars are
observed to have masses which are much larger than M —so large that it seems
exceedingly unlikely that they can ever shed sufficient material so as to be able to
fall below the limit required for a stable white dwarf (∼1.3M : Chandrasekhar
[1]) or neutron star (∼ 0.7M : Oppenheimer-Volkoff [2]) to develop. We are thus
driven to consider the consequences of a situation in which a star collapses right
down to a state in which the effects of general relativity become so important that
they eventually dominate over all other forces.
I shall begin with what I think we may now call the “classical” collapse picture
as presented by general relativity. Objections and modifications to this picture will
be considered afterwards. The main discussion is based on Schwarzschild’s solu-
tion of the Einstein vacuum equations. This solution represents the gravitational
field exterior to a spherically symmetrical body. In the original Schwarzschild
co-ordinates, the metric takes the familiar form

1 Reprinted with the kind permissions of Societa Italiana di Fisica and of the author.

1141

0001-7701/02/0700-1141/0
c 2002 Plenum Publishing Corporation
1142 R. Penrose

ds 2 = (1 − 2m/r)dt 2 − (1 − 2m/r)−1 dr 2 − r 2 (dθ 2 + sin2 θ dϕ 2 ). (1)

Here θ and ϕ are the usual spherical polar angular co-ordinates. The radial co-
ordinate r has been chosen so that each sphere r = const, t = const has intrinsic
surface area 4πr 2 . The choice of time co-ordinate t is such that the metric form
is invariant under t → t + const and also under t → −t. The static nature of
the space-time is thus made manifest in the formal expression for the metric. The
quantity m is the mass of the body, where “general-relativistic units” are chosen,
so that
c=G=1
that is to say, we translate our units according to
1 s = 3 · 1010 cm = 4 · 1038 g.
When r = 2m, the metric form (1) breaks down. The radius r = 2m is referred
to as the Schwarzschild radius of the body.
Let us imagine a situation in which the collapse of a spherically symmetrical
(nonrotating) star takes place and continues until the surface of the star approaches
the Schwarzschild radius. So long as the star remains spherically symmetrical, its
external field remains that given by the Schwarzschild metric (1). The situation
is depicted in Fig. 1. Now the particles at the surface of the star must describe
timelilke lines. Thus, from the way that the “angle” of the light cones appears
to be narrowing down near r = 2m, it would seem that the surface of the star
can never cross to within the r = 2m region. However, this is misleading. For
suppose an observer were to follow the surface of the star in a rocket ship, down
to r = 2m. He would find (assuming that the collapse does not differ significantly
from free fall) that the total proper time that he would experience as elapsing,
as he finds his way down to r = 2m, is in fact finite. This is despite the fact
that the world line he follows has the appearance of an “infinite” line in Fig. 1.
But what does the observer experience after this finite proper time has elapsed?
Two possibilities which suggest themselves are: i) the observer encounters some
form of space-time singularity—such as infinite tidal forces—which inevitably
destroys him as he approaches r = 2m; ii) the observer enters some region of
space-time not covered by the (t, r, θ, ϕ) co-ordinate system used in (1). (It would
be unreasonable to suppose that the observer’s experiences could simply cease
after some finite time, without his encountering some form of violent agency.)
In the present situation, in fact, it is possibility ii) which occurs. The easiest
way to see this is to replace the co-ordinate t by an advanced time parameter v
given by
v = t + r + 2m log(r − 2m),
whereby the metric (1) is transformed to the form (Eddington [3], Finkelstein [4])
Gravitational Collapse 1143

Figure 1. Spherically symmetrical collapse in the usual Schwarzschild co-ordinates.

1144 R. Penrose


ds 2 = (1 − 2m/r)dv 2 − 2 dr dv − r 2 dθ 2 + sin2 θ dϕ 2 . (2)
This form of metric has the advantage that it does not become inapplicable at
r = 2m. The whole range 0 < r < ∞ is encompassed in a nonsingular fashion
by (2). The part r > 2m agrees with the part r > 2m of the original expression (1).
But now the region has been extended inwards in a perfectly regular way across
r = 2m and right down towards r = 0.
The situation is as depicted in Fig. 2. The light cones tip over more and more
as we approach the centre. In a sense we can say that the gravitational field has
become so strong, within r = 2m, that even light cannot escape and is dragged
inwards towards the centre. The observer on the rocket ship, whom we considered
above, crosses freely from the r > 2m region into the 0 < r < 2m region. He
encounters r = 2m at a perfectly finite time, according to his own local clock,
and he experiences nothing special at that point. The space-time there is locally
Minkowskian, just as it is everywhere else (r > 0).
Let us consider another observer, however, who is situated far from the star.
As we trace the light rays from his eye, back into the past towards the star, we find
that they cannot cross into the r < 2m region after the star has collapsed through.
They can only intersect the star at a time before the star’s surface crosses r = 2m.
No matter how long the external observer waits, he can always (in principle) still
see the surface of the star as it was just before it plunged through the Schwarzschild
radius. In practice, however, he would soon see nothing of the star’s surface—only
a “black hole”—since the observed intensity would die off exponentially, owing
to an infinite red shift.
But what will be the fate of our original observer on the rocket ship? After
crossing the Schwarzschild radius, he finds that he is compelled to enter regions
of smaller and smaller r. This is clear from the way the light cones tip over
towards r = 0 in Fig. 2, since the observer’s world line must always remain a
timelike line. As r decreases, the space-time curvature mounts (in proportion to
r −3 ), becoming theoretically infinite at r = 0. The physical effect of space-time
curvature is experienced as a tidal force: objects become squashed in one direction
and stretched in another. As this tidal effect mounts to infinity, our observer must
eventually2 be torn to pieces—indeed, the very atoms of which he is composed
must ultimately individually share this same fate!
Thus, the true space-time singularity, resulting from a spherically symmet-
rical collapse, is located not at r = 2m, but at r = 0. Although the hypersurface
r = 2m has, in the past, itself been frequently referred to as the “Schwarzschild
singularity”, this is really a misleading terminology since r = 2m is a singularity
merely of the t co-ordinate used in (1) and not of the space-time geometry. More

2 In fact, if m is of the order of a few solar masses, the tidal forces would already be easily large
enough to kill a man in free fall, even at r = 2m. But for m > 108 M the tidal effect at r = 2m
would be no greater than the tidal effect on a freely falling body near the Earth’s surface.
Gravitational Collapse 1145

Figure 2. Spherically symmetrical collapse in Eddington-Finkelstein co-ordinates.

1146 R. Penrose

appropriate is the term “event horizon”, since r = 2m represents the absolute

boundary of the set of all events which can be observed in principle by an
external inertial observer. The term “event horizon” is used also in cosmology
for essentially the same concept (cf. Rindler [5]). In the present case the horizon
is less observer-dependent than in the cosmological situations, so I shall tend to
refer to the hypersurface r = 2m as the absolute event horizon3 of the space-time
(2).
This, then, is the standard spherically symmetrical collapse picture presented
by general relativity. But do we have good reason to trust this picture? Need we
believe that it necessarily accords, even in its essentials, with physical reality? Let
me consider a number of possible objections:
a) densities in excess of nuclear densities inside,
b) exact vacuum assumed outside,
c) zero net charge and zero magnetic field assumed,
d) rotation excluded,
e) asymmetries excluded,
f) possible λ-term not allowed for,
g) quantum effects not considered,
h) general relativity a largely untested theory,
i) no apparent tie-up with observations.
As regards a), it is true that for a body whose mass is of the order of M ,
its surface would cross r = 2m only after nuclear densities had been somewhat
exceeded. It may be argued, then, that too little is understood about the nature of
matter at such densities for us to be at all sure how the star would behave while still
outside r = 2m. But this is not really a significant consideration for our general
discussion. It could be of relevance only for the least massive collapsing bodies,
if at all. For, the larger the mass involved, the smaller would be the density at
which it would be expected to cross r = 2m. It could be that very large masses
indeed may become involved in gravitational collapse. For m > 1011 M (e.g. a
good-sized galaxy), the averaged density at which r = 2m is crossed would be
less than that of air!
The objections b), c), d), e) and, to some extent, f) can all be partially handled
if we extract, from Fig. 2, only that essential qualitative piece of information which
characterizes the solution (2) as describing a collapse which has passed a “point of
no return”. I shall consider this in more detail shortly. The upshot will be that if a
collapse situation develops in which deviations from (2) near r = 2m at one time
are not too great, then two consequences are to be inferred as to the subsequent

3 In a general space-time with a well-defined external future infinity, the absolute event horizon would
be defined as the boundary of the union of all timelike curves which escape to this external future
infinity. In the terminology of Penrose [6], if M is a weakly asymptotically simple space-time, for
example, then the absolute event horizon in M is I˙− [I+ ].
Gravitational Collapse 1147

behaviour. In the first instance an absolute event horizon will arise. Anything
which finds itself inside this event horizon will not be able to send signals to the
outside worlds. Thus, in this respect at least, the qualitative nature of the “r = 2m”
hypersurface in (2) will remain. Similarly, an analogue of the physical singularity
at r = 0 in (2) will still develop in these more general situations. That is to say,
we know from rigorous theorems in general-relativity theory that there must be
some space-time singularity resulting inside the collapse region. However, we do
not know anything about the detailed nature of this singularity. There is no reason
to believe that it resembles the r = 0 singularity of the Schwarzschild solution
very closely.
In regard to c), d) and f) we can actually go further in that exact solutions are
known which generalize the metric (2) to include angular momentum (Kerr [7])
and, in addition, charge and magnetic moment (Newnan et al. [8]), where a cos-
mological constant may also be incorporated (Carter [9]). These solutions appear
to be somewhat special in that, for example, the gravitational quadrupole moment
is fixed in terms of the angular momentum and the mass, while the magnetic-
dipole moment is fixed in terms of the angular momentum, charge and mass.
However, there are some reasons for believing that these solutions may actually
represent the general exterior asymptotic limit resulting from the type of collapse
we are considering. Any extra gravitational multipole moments of quadrupole
type, or higher, can be radiated away by gravitational radiation; similarly, extra
electromagnetic multipole moments of dipole type, or higher, can be radiated
away by electromagnetic radiation. (I shall discuss this a little more later.) If this
supposition is correct, then e) will to some extent also be covered by an analysis
of these exact solutions. Furthermore, b) would, in effect, be covered as well,
provided we assume that all matter (with the exception of electromagnetic field—
if we count that as “matter”) in the neighbourhood of the “black hole”, eventually
falls into the hole. These exact solutions (for small enough angular momentum,
charge and cosmological constant) have absolute event horizons similar to the
r = 2m horizon in (2). They also possess space-time curvature singularities,
although of a rather different structure from r = 0 in (2). However, we would
not expect the detailed structure of these singularities to have relevance for a
generically perturbed solution in any case.
It should be emphasized that the above discussion is concerned only with
collapse situations which do not differ too much initially from the spherically
symmetrical case we originally considered. It is not known whether a gravitational
collapse of a qualitatively different character might not be possible according to
general relativity. Also, even if an absolute event horizon does arise, there is
the question of the “stability” of the horizon. An “unstable” horizon might be
envisaged which itself might develop into a curvature singularity. These, again,
are questions I shall have to return to later.
1148 R. Penrose

As for the possible relevance of gravitational quantum effects, as suggested

in g), this depends, as far as I can see, on the existence of regions of space-time
where there are extraordinary local conditions. If we assume the existence of an
absolute event horizon along which curvatures and densities remain small, then
it is very hard to believe that a classical discussion of the situation is not amply
adequate. It may well be that quantum phenomena have a dominating influence
on the physics of the deep interior regions. But whatever effects this might have,
they would surely not be observable from the outside. We see from Fig. 2 that such
effects would have to propagate outwards in spacelike directions over “classical”
regions of space-time. However, we must again bear in mind that these remarks
might not apply in some qualitatively different type of collapse situation.
We now come to h), namely the question of the validity of general relativity
in general, and its application to this type of problem in particular. The inadequacy
of the observational data has long been a frustration to theorists, but it may be that
the situation will change somewhat in the future. There are several very relevant
experiments now being performed, or about to be performed. In addition, since it
has become increasingly apparent that “strong” gravitational fields probably play
an important role in some astrophysical phenomena, there appears to be a whole
new potential testing-ground for the theory.
Among the recently performed experiments, designed to test general rela-
tivity, one of the most noteworthy has been that of Dicke and Goldenberg [10],
concerning the solar oblateness. Although the results have seemed to tell against
the pure Einstein theory, the interpretations are not really clear-cut and the matter
is still somewhat controversial. I do not wish to take sides on this issue. Probably
one must wait for further observations before the matter can be settled. However,
whatever the final outcome, the oblateness experiment had, for me, the importance
of forcing me to examine, once more, the foundations of Einstein’s theory, and
to ask what parts of the theory are likely to be “here to stay” and what parts
are most susceptible to possible modification. Since I feel that the “here to stay”
parts include those which were most revolutionary when the theory was first put
forward, I feel that it may be worth-while, in a moment, just to run over the
reasoning as I see it. The parts of the theory I am referring to are, in fact, the
geometrical interpretation of gravity, the curvature of space-time geometry and
general-relativistic causality. These, rather than any particular field equations, are
the aspects of the theory which give rise to what perhaps appears most immediately
strange in the collapse phenomenon. They also provide the physical basis for the
major part of the subsequent mathematical discussion.
To begin with, let us agree that it is legitimate to regard space-time as consti-
tuting a four-dimensional smooth manifold (or “continuum”). I do not propose to
give a justification of this, because on an ordinary macroscopic level it is normally
taken as “obvious”. (On the other hand, I think that at a deeper submicroscopic level
it is almost certainly “false”, but this is not likely to affect the normal discussion
Gravitational Collapse 1149

of space-time structure—except perhaps at a space-time singularity!) Next, we

must establish the existence of a physically well-defined metric ds which defines
for our manifold a (pseudo-) Riemannian structure, with signature (+ − −−).
The meaning of ds is to be such that when integrated along the world line of any
particle, it gives the lapse of proper times as experienced by that particle. Thus,
the existence of ds depends on the existence of accurate clocks in nature. These
clocks must behave locally according to the laws of special relativity. Also, for any
two such clocks following the same world line, the time rates they register must
agree with one another along the line and should not depend on, say, differing
histories for the two clocks. That such clocks do seem to exist in nature, in effect,
is a consequence of the fact that any mass m has associated with it a natural
frequency mh−1 . Thus, the existence of accurate clocks comes down ultimately,
via quantum mechanics, to the existence of well-defined masses in nature, whose
relative values are in strict proportion throughout space-time. Of course, it might
ultimately turn out that the mass ratios of particles are not constant throughout
space-time. Then different particles might define slightly different (conformally
related) metrics for space-time. But the evidence at present is strongly against any
appreciable difference existing.
If two neighbouring events in space-time have a separation such that ds 2 ≥ 0,
then according to special relativity, it is possible for one to have a causal influence
on the other; if ds 2 < 0, then it is not. We expect this to persist also on a global
scale. Thus, it is possible, of two events, for one to influence the other causally if
and only if there is a timelike or null curve connecting them.
The existence of a physically well-defined metric and causal structure for
space-time, then, seems to be fairly clearly established. It is not so clear, however,
that this metric, as so defined, is going to be nonflat. However, we can take the
experiment of Pound and Rebka [11] as almost a direct measurement establishing
the nonflat nature of space-time. (For this, strictly speaking, the experiment would
have to be repeated at various points on the Earth’s surface.) The measured ds
near the Earth’s surface and the ds further from the Earth’s surface cannot both be
incorporated into the same Minkowskian framework because of the “clock slow-
ing” effect (cf. Schild [12]). Furthermore, owing to energy balance considerations
it is clear that it is with gravitational fields that this “clock slowing” effect occurs
(owing to the fact that it is energy, i.e. mass which responds to a gravitational
field). Thus gravitation must be directly related to space-time curvature.
Since we have a (pseudo-) Riemannian manifold, we can use the standard
techniques of differential geometry to investigate it. In particular, we can construct
a physically meaningful Riemann tensor Rabcd and thence its Einstein tensor
1
Gab = Rab − Rgab .
2
1150 R. Penrose

Because of the contracted Bianchi identities we know that this satisfies the
usual vanishing divergence law. But we also have a symmetric tensor Tab , namely
the local energy-momentum tensor (composed of all fields but gravitation), which
must satisfy a similar vanishing divergence law. It does not then necessarily follow
that
Gab + λg ab = −8πTab (3)

for some constant λ, but it is worth remarking that if we do not postulate this equa-
tion, then we have not just one, but two (linearly unrelated) conserved “energylike”
quantities, namely Gab and Tab . In fact, this is just what happens in the theory
of Brans and Dicke [13]. (Such a motivation for the choice of Einstein’s field
equations (3), does not to my mind have quite the force of the earlier argument,
so alternatives to (3) are certainly well worth considering.) Finally, the geodesic
motion of monopole test particles may be taken as a consequence of the vanishing
divergence condition on Tab (Einstein and Grommer [14]).
So I want to admit the possibility that Einstein’s field equations may be wrong,
but not (that is, in the macroscopic realm, and where curvatures or densities are
not fantastically large) that the general pseudo-Riemannian geometric framework
may be wrong. Then the mathematical discussion of the collapse phenomenon
can at least be applied. It is interesting that the general mathematical discussion of
collapse actually uses very little of the details of Einstein’s equations. All that is
needed is a certain inequality related to positive-definiteness of energy. In fact, the
adoption of the Brans-Dicke theory in place of Einstein’s would make virtually
no qualitative difference to the collapse discussion.
The final listed objection to the collapse picture is h), namely the apparent
lack of any tie-up with observed astronomical phenomena. Of course it could be
argued that the prediction of the “black hole” picture is simply that we will not
see anything—and this is precisely consistent with observations since no “black
holes” have been observed! But the real argument is really the other way around.
Quasars are observed. And they apparently have such large masses and such
small sizes that it would seem that gravitational collapse ought to have taken over.
But quasars are also long-lived objects. The light they emit does not remotely
resemble the exponential cut-off in intensity, with approach to infinite red shift,
that might be inferred from the spherically symmetrical discussion. This has led
a number of astrophysicists to question the validity of Einstein’s theory, at least
in its applicability to these situations.
My personal view is that while it is certainly possible (as I have mentioned
earlier) that Einstein’s equations may be wrong, I feel it would be very premature
indeed to dismiss these equations just on the basis of the quasar observations.
For, the theoretical analysis of collapse, according to Einstein’s theory, is still
more or less in its infancy. We just do not know, with much certainty, what the
consequences of the theory really are. It would be a mistake to fasten attention just
Gravitational Collapse 1151

on those aspects of general-relativistic collapse which are known and to assume

that this gives us essentially the complete picture. (It is perhaps noteworthy that
many general-relativity theorists have a tendency, themselves, to be a bit on the
sceptical side as regards the “classical” collapse picture!) Since it seems to me
that there are a number of intriguing largely unexplored possibilities, I feel it may
be worth-while to present the “generic” general-relativistic collapse picture as I
see it, not only as regards the known theorems, but also in relation to some of the
more speculative and conjectural aspects of the situation.
To begin with, let us consider what the general theorems do tell us. In order
to characterize the situation of collapse “past a point of no return”, I shall first
need the concept of a trapped surface. Let us return to Fig. 2. We ask what
qualitative peculiarity of the region r < 2m (after the star has collapsed through)
is present. Can such peculiarities be related to the fact that everything appears
to be forced inwards in the direction of the centre? It should be stressed again
that apart from r = 0, the space-time at any individual point inside r = 2m
is perfectly regular, being as “locally Minkowskian” as any other point (outside
r = 0). So the peculiarities of the 0 < r < 2m region must be of a partially
“global” nature. Now consider any point T in the (v, r)-plane of Fig. 2 (r < 2m).
Such a point actually represents a spherical 2-surfacc in space-time, this being
traced out as the θ, ϕ co-ordinates vary. The surface area of this sphere is 4π r 2 .
We imagine a flash of light emitted simultaneously over this spherical surface
T. For an ordinary spacelike 2-sphere in flat space-time, this would result in an
ingoing flash imploding towards the centre (surface area decreasing) together
with an outgoing flash exploding outwards (surface area increasing). However,
with the surface T , while we still have an ingoing flash with decreasing surface
area as before, the “outgoing” flash, on the other hand, is in effect also falling
inwards (though not as rapidly) and its surface area also decreases. The surface
T (v = const, r = const < 2m) of metric (2) serves as the prototype of a trapped
surface. If we perturb the metric (2) slightly, in the neighbourhood of an initial
hypersurface, then we would still expect to get a surface T with the following
property:
T is a spacelike closed4 2-surface such that the null geodesics which meet it
orthogonally all converge initially at T.
This convergence is taken in the sense that the local surface area of cross-section
decreases, in the neighbourhood of each point of T , as we proceed into the future.
(These null geodesics generate, near T , the boundary of the set of points lying
causally to the future of the set T .) Such a T is called a trapped surface.
We may ask whether any connection is to be expected between the existence
of a trapped surface and the presence of a physical space-time singularity such

4 By a “closed” surface, hypersurface, or curve, I mean one that is “compact without boundary”.
1152 R. Penrose

as that occurring at r = 0 in (2). The answer supplied by some general theorems

(Penrose [6, 15], Hawking and Penrose [16]) is, in effect, that the presence of
a trapped surface always does imply the presence of some form of space-time
singularity.
There are similar theorems that can also be applied in cosmological situations.
For example (Hawking and Penrose [16], Hawking [17–19]) if the universe is
spatially closed, then (excluding exceptional limiting cases, and assuming λ ≤ 0)
the conclusion is that there must be a space-time singularity. This time we expect
the singularity to reside in the past (the “big bang”). Other theorems (Hawking and
Penrose [16], Hawking [17, 19], can be applied also to spatially open universes.
For example, if there is any point (e.g. the Earth at the present epoch) whose past
light cone starts “converging again” somewhere in the past (i.e. objects of given
size start to have larger apparent angular diameters again when their distance
from us exceeds some critical value), then, as before, the presence of space-time
singularities is implied (λ ≤ 0). According to Hawking and Ellis ([20], cf. also
Hawking and Penrose [16]) the presence and isotropy of the 3◦ K radiation strongly
indicates that the above condition on our past light cone is actually satisfied. So the
problem of space-time singularities does seem to be very relevant to our universe,
also on a large scale.
The main significance of theorems such as the above, is that they show that
the presence of space-time singularities in exact models is not just a feature of
their high symmetry, but can be expected also in generically perturbed models.
This is not to say that all general-relativistic curved space-times are singular—
far from it. There are many exact models known which are complete and free
from singularity. But those which resemble the standard Friedmann models or the
Schwarzschild collapse model sufficiently closely must be expected to be singular
(λ ≤ 0). The hope had often been expressed (cf. Lindquist and Wheeler [21],
Lifshitz and Khalatnikov [22]) that the actual space-time singularity occurring in
a collapsing space-time model might have been a consequence more of the fact that
the matter was all hurtling simultaneously towards one central point, than of some
intrinsic feature of general-relativistic space-time models. When perturbations are
introduced into the collapse, so the argument could go, the particles coming from
different directions might “miss” each other, so that an effective “bounce” might
ensue. Thus, for example, one might envisage an “oscillating” universe which
on a large scale resembles the cycloiclal singular behaviour of an “ oscillating”
spatially closed Friedmann model; but the detailed behaviour, although perhaps
involving enormous densities while at maximal contraction, might, by virtue
of complicated asymmetries, contrive to avoid actual space-time singularities.
However, the theorems seem to have ruled out a singularity-free “bounce” of this
kind. But the theorems do not say that the singularities need resemble those of the
Friedmann or Schwarzschild solutions at all closely. There is some evidence (cf.
Misner [23], for example) that the “generic” singularities may be very elaborate
Gravitational Collapse 1153

and possess a qualitative structure very different from that of their smoothed-out
counterparts. Very little is known about this, however.
It is worth mentioning the essential basic assumptions that enter into the
theorems. In the first place we require an “energy condition” which, by virtue of
Einstein’s equations (3), may be stated as a negative-definiteness condition on the
Ricci tensor:
t a ta = 1 implies Rab t a t b ≤ 0, (4)
that is to say, the time–time component R00 of Rab is nonpositive in any or-
thonormal frame. If we assume λ = 0 in Einstein’s equations (3), then (4)
becomes
t a ta = 1 implies Tab t a t b ≥ 21 Tcc .
This, when referred to an eigenframe of Tab , can be stated as
X
E + pα ≥ 0 and E + pα ≥ 0, (5)

where α = 1, 2, 3. Here E is the energy density (referred to this frame) and

p1 , p2 , p3 are the three principal pressures. If (3) holds with λ < 0, then it is still
true that (4) is a consequence of (5). The significance of the energy condition (4)
lies in the effect of Raychaudhuri [24] which states that whenever a system of
timelike geodesics normal to a spacelike hypersurface starts converging, then this
convergence inevitably increases along the geodesics until finally the geodesics
cross over one another (assuming the geodesics are complete).
There is a corresponding focussing effect in the case of null geodesics. This
depends on the “weak energy condition”:
l a la = 0 implies Rab l a l b ≤ 0. (6)
This condition (6) is a consequence of (4) (as follows by a limiting argument) but
not conversely. If we assume Einstein’s equations (3) with, now, any value of λ,
then (6) is equivalent to
E + pa ≥ 0 (7)
for α = 1, 2, 3. The conditions (7) are, in fact, a consequence of the nonnegative
definiteness of the energy density:
t a ta = 1 implies Tab t a t b ≥ 0
(that is T00 ≥ 0 in each orthonormal frame). Thus, there is a strong physical basis
for (6). The physical basis for (4) is not quite so strong, but provided λ ≤ 0,
we would certainly expect (4) to hold for all normal matter. (Note that if E >
0 ≥ λ, only large negative pressures could cause trouble with (4). Usually people
only worry about large positive pressures!) It is the “strong” condition (4) that is
1154 R. Penrose

required for the proofs of most of the theorems, but much can be said, concerning
the qualitative nature of a collapse situation, even on the basis of the “weak”
condition (6) alone (cf. Penrose [15]).
A remark concerning the condition on the cosmological constant λ seems
appropriate here. It is a weakness of the theorems that most of them do require
λ ≤ 0 for their strict applicability. However, it would appear that the condition
λ ≤ 0 is only really relevant to the initial setting of the global conditions on the
space-time which are required for applicability of the theorem. If curvatures are
to become large near a singularity, then (from dimensional considerations alone)
the λ-term will become more and more insignificant. So it seems unlikely that a
λ-term will really make much difference to the singularity structure in a collapse.
The relevance of λ is really only at the cosmological scale.
Most of the theorems (but not all, cf. Hawking [19]) require, as an additional
assumption, the nonexistence of closed timelike curves. This is a very reasonable
requirement, since a space-time which possesses closed timelike curves would
allow an observer to travel into his own past. This would lead to very serious
interpretative difficulties! Even if it could be argued, say, that the accelerations
involved might be such as to make the trip impossible in “practice” (cf. Gödel
[25]), equally serious difficulties would arise for the observer if he merely reflected
some light signals into his own past! In addition closed timelike curves can lead
to unreasonable consistency conditions on the solutions of hyperbolic differential
equations. In any case, it seems unlikely that closed timelike curves can substitute
for a space-time singularity, except in special unstable models.
Some of the theorems require an additional “generality” condition, to the
effect that every timelike or null geodesic enters some region in which the curvature
is not everywhere lined up in a particular way with the geodesic. (More precisely,
t[a Rb]cd[e tf ] t c t d 6= 0 somewhere along the geodesic, t a being its tangent vector.)
This condition plays a role in the mathematics, but from the physical point of view
it is really no condition at all. We would always expect a little bit of matter or
randomly oriented curvature along any geodesic in a physically realistic solution.
It is only in very special limiting cases that we would expect the condition to be
violated. (Curiously enough, however, practically every explicitly known solution
does violate the condition!)
Finally, it should be remarked that none of the theorems directly establishes
the existence of regions of approaching infinite curvature. Instead, all one obtains
is that the space-time is not geodesically complete (in timelike or null directions)
and, furthermore, cannot be extended to a geodesically complete space-time.
(“Geodesically complete” means that geodesics can be extended indefinitely to
arbitrarily large values of their length or affine parameter—so that inertially mov-
ing particles or photons do not just “fall off the edge” of the space-time.) The most
“reasonable” explanation for why the space-time is not inextendible to a complete
space-time seems to be (and I would myself believe this to be the most likely, in
Gravitational Collapse 1155

general) that the space-time is confronted with, in some sense, infinite curvature
at its boundary. But the theorems do not quite say this. Other types of space-time
singularity are possible, and theorems of a somewhat different nature would be
required to decide which is the most likely type of singularity to occur.
We must now ask the question whether the theorems are actually likely to
be relevant in the case of a collapsing star or superstar. Do we, in fact, have any
reason to believe that trapped surfaces can ever arise in gravitational collapse? I
think a very strong case can be made that at least sometimes a trapped surface
must arise. I would not expect trapped surfaces necessarily always to arise in a
collapse. It might depend on the details of the situation. But if we can establish
that there can be nothing in principle against a trapped surface arising—even if
in some very contrived and outlandish situation—then we must surely accept that
trapped surfaces must at least occasionally arise in real collapse.
Rather than use the trapped-surface condition, however, it will actually be
somewhat easier to use the alternative condition of the existence of a point whose
light cone starts “converging again”. From the point of view of the general the-
orems, it really makes no essential difference which of the two conditions is
used. Space-time singularities are to be expected in either case. Since we are
here interested in a collapse situation rather than in the “big bang”, we shall be
concerned with the future light cone C of some point p. What we have to show
is that it is possible in principle for enough matter to cross to within C, so that
the divergence of the null geodesics which generate C changes sign somewhere
to the future of p. Once these null geodesics start to converge, then “ weak energy
condition” (6) will take over, with the implication that an absolute event horizon
must develop (outside C). As a consequence of the stronger “energy condition”
(4) it will also follow that space-time singularities will occur.
Since we ask only that it be possible in principle to reconverge the null rays
generating C, we can resort to an (admittedly far-fetched) “gedanken experiment”.
Consider an elliptical galaxy containing, say, 1011 stars. Suppose, then, that we
contrive to alter the motion of the stars slightly by eliminating the transverse
component of their velocities. The stars will then fall inwards towards the centre.
We may arrange to steer them, if we like, so as to ensure that they all reach the
vicinity of the centre at about the same time without colliding with other stars. We
only need to get them into a volume of diameter about fifty times that of the solar
system, which gives us plenty of room for all the stars. The point p is now taken
near the centre at about the time the stars enter this volume (Fig. 3). It is easily
seen from the orders of magnitude involved, that the relativistic light deflection
(an observed effect of general relativity) will be sufficient to cause the null rays
in C to reconverge, thus achieving our purpose.
Let us take it, then, that absolute event horizons can sometimes occur in a
gravitational collapse. Can we say anything more detailed about the nature of the
resulting situation? Hopeless as this problem may appear at first sight, I think there
1156 R. Penrose

Figure 3. The future light cone of p is caused to reconverge by the falling stars.

is actually a reasonable chance that it may find a large measure of solution in the
not-too-distant future. This would depend on the validity of a certain result which
has been independently conjectured by a number of people. I shall refer to this
as the generalized5 Israel conjecture (abbreviated GIC). Essentially GIC would
state: if an absolute event horizon develops in an asymptotically flat space-time,
then the solution exterior to this horizon approaches a Kerr-Newman solution
asymptotically with time.

5 Israel conjectured this result only in the stationary case, hence the qualification “generalized”. In
fact, Israel has expressed sentiments opposed to GIC. However, Israel’s theorem [26, 27] represents
an important step towards establishing of GIC, if the conjecture turns out to be true.
Gravitational Collapse 1157

The Kerr-Newman solutions (Kerr [7], Newman et al. [8]) are explicit asymp-
totically flat stationary solutions of the Einstein-Maxwell equation (λ = 0) in-
volving just three free parameters m, a and e. As with the metric (1), the mass, as
measured asymptotically, is the parameter m (in gravitational units). The solution
also possesses angular momentum, of magnitude am. Finally, the total charge
is given by e. When a = e = 0 we get the Schwarzschild solution. Provided
that
m2 ≥ a 2 + e2
the solution has an absolute event horizon. Carter [9] has shown how to obtain
all the geodesics and charged orbits for this solution, reducing the problem to a
single quadrature. Thus, if GIC is true, then we shall have remarkably complete
information as to the asymptotic state of affairs resulting from a gravitational
collapse.
But what reason is there for believing that GIC has any chance of being
true? One indication comes from a perturbation analysis of the Schwarzschild
solution (Regge and Wheeler [28], Doroshkevich et al. [29] which seems to
indicate that all perturbations except rotation have a tendency to be damped out.
Another indication is the theorem of Israel [26] which states, in effect, that the
Schwarzschild solution is the only static asymptotically flat vacuum solution with
an absolute event horizon (although there is a nontrivial side-condition to the
theorem; cf. also Thorne [30] for the axially symmetric case). Israel [27] has
also generalized his result to the Einstein-Maxwell theory, finding the spherically
symmetric Reissner-Nordstrom solution to be the only asymptotically flat static
solution with an absolute event horizon. Carter [31] has made some progress, in
the vacuum rotating case, towards the objective of establishing the Kerr solution
(e = 0) as the general asymptotically flat stationary solution with an absolute event
horizon. In addition, there are solutions of the vacuum equations known (Robinson
and Trautman [32]), which are suitably asymptotically flat and nonrotating, which
apparently possess absolute event horizons, but are nonstatic. As time progresses
they become more and more symmetrical, approaching the Schwarzschild solution
asymptotically with time [33]. In the process, the higher multipole moments are
radiated away as gravitational radiation.
The following picture then suggests itself. A body, or collection of bodies,
collapses down to a size comparable to its Schwarzschild radius, after which
a trapped surface can be found in the region surrounding the matter. Some way
outside the trapped surface region is a surface which will ultimately be the absolute
event horizon. But at present, this surface is still expanding somewhat. Its exact
location is a complicated affair and it depends on how much more matter (or
radiation) ultimately falls in. We assume only a finite amount falls in and that GIC
is true. Then the expansion of the absolute event horizon gradually slows down
to stationarity. Ultimately the field settles down to becoming a Kerr solution (in
1158 R. Penrose

the vacuum case) or a Kerr-Newman solution (if a nonzero net charge is trapped
in the “black hole”).
Doubts have frequently been expressed concerning GIC, since it is felt that
a body would be unlikely to throw off all its excess multipole moments just as it
crosses the Schwarzschild radius. But with the picture presented above this is not
necessary. I would certainly not expect the body itself to throw off its multipole
moments. On the other hand, the gravitational field itself has a lot of settling-down
to do after the body has fallen into the “hole”. The asymptotic measurement of
the multipole moments need have very little to do with the detailed structure of
the body itself; the field can contribute very significantly. In the process of settling
down, the field radiates gravitationally—and electromagnetically too, if electro-
magnetic field is present. Only the mass, angular momentum and charge need
survive as ultimate independent parameters. (Presumably the charge parameter e
would be likely to be very small by comparison with a and m.)
But suppose GIC is not true, what then? Of course, it may be that there
are just a lot more possible limiting solutions than that of Kerr-Newman. This
would mean that much more work would have to be done to obtain the detailed
picture, but it would not imply any qualitative change in the set-up. On the other
hand there is the more alarming possibility that the absolute event horizon may
be unstable! By this I mean that instead of settling down to become a nice smooth
solution, the space-time might gradually develop larger and larger curvatures in
the neighbourhood of the absolute event horizon, ultimately to become effectively
singular there. My personal opinion is that GIC is more likely than this, but various
authors have expressed the contrary view.6
If such instabilities are present then this would certainly have astrophysical
implications. But even if GIC is true, the resulting “black hole” may by no means
be so “dead” as has often been suggested. Let us examine the Kerr-Newmann
solutions, in the case m2 > a 2 + e2 in a little more detail. But before doing
so let us refer back to the Schwarzschild solution (2). In Fig. 4, I have drawn
what is, in effect, a cross-section of the space-time, given by v − r = const.
The circles represent the location of a flash of light which had been emitted at
the nearby point a moment earlier. Thus, they indicate the orientation of the light
cones in the space-time. We note that for large r the point lies inside the circle,
which is consistent with the static nature of the space-time (i.e. one can “stay
in the same place” while retaining a timelike world line). On the other hand for
r < 2m the point lies outside the circle, indicating that all matter must be dragged
inwards if it is to remain moving in a timelike direction (so, to “stay in the same
place” one would have to exceed the local speed of light). Let us now consider the
corresponding picture for the Kerr-Newman solutions with m2 > a 2 + e2 (Fig.

6 Some recent work of Newman [34] on the charged Robinson-Trautman solutions suggests that new
features indicating instabilities may arise when an electromagnetic field is present.
Gravitational Collapse 1159

Figure 4. Spatial view of spherical “black hole” (Schwarzschild solution).

5). I shall not be concerned, here, with the curious nature of the solution inside
the absolute event horizon H , since this may not be relevant to GIC. The horizon
H itself is represented as a surface which is tangential to the light cones at each of
its points. Some distance outside H is the “stationary limit” L, at which one must
travel with the local light velocity in order to “stay in the same place”.
I want to consider the question of whether it is possible to extract energy
out of a “black hole”. One might imagine that, since the matter which has fallen
through has been lost for ever, so also is its energy content irrevocably trapped.
However, it is not totally clear to me that this need be the case. There are at least
two methods (neither of which is very practical) which might be construed as
mechanisms for extracting energy from a “black hole”. The first is due to Misner
[35]. This requires, in fact, a whole galaxy of 2N “black holes”, each of mass m.
We first bring them together in pairs and allow them to spiral around one another,
ultimately to swallow each other up. During the spiraling, a certain fraction K of
their mass-energy content is radiated away as gravitational energy, so the mass of
the resulting “black hole” is 2m(1 − K). The energy of the gravitational waves
is collected and the process is repeated. Owing to the scale invariance of the
1160 R. Penrose

gravitational vacuum equations, the same fraction of the mass-energy is collected

in the form of gravitational waves at each stage. Finally we end up with a single
“black hole” of mass 2N m(1 − K)N . Now, the point is that however small K may
in fact be, we can always choose N large enough so that (1 − K)N is as small as
we please. Thus, in principle, we can extract an arbitrarily large fraction of the
mass-energy content of Misner’s galaxy.
But anyone at all familiar with the problems of detecting gravitational radi-
ation will be aware of certain difficulties! Let me suggest another method which
actually tries to do something a little different, namely extract the “rotational en-
ergy” of a “rotating black hole” (Kerr solution). Consider Fig. 5 again. We imagine
a civilization which has built some form of stabilized structure S surrounding the
“black hole”. If they lower a mass slowly on a (light, inextensible, unbreakable)
rope until it reaches L, they will be able to recover, at S, the entire energy content of
the mass. If the mass is released as it reaches L then they will simply have bartered
the mass for its energy content. (This is the highest-grade energy, however, namely
wound-up springs!) But they can do better than this! They also build another
structure S ∗ , which rotates, to some extent, with the “black hole”. The lowering
process is continued, using S ∗ , to beyond L. Finally the mass is dropped through
H, but in such a way that its energy content, as measured from S, is negative! Thus,
the inhabitants of S are able, in effect, to lower masses into the “black hole” in
such a way that they obtain more than the energy content of the mass. Thus they
extract some of the energy content of the “black hole” itself in the process. If we
examine this in detail, however, we find that the angular momentum of the “black
hole” is also reduced.
Thus, in a sense, we have found a way of extracting rotational energy from the
“black hole”. Of course, this is hardly a practical method! Certain improvements
may be possible, e.g., using a ballistic method.7 But the real significance is to
find out what can and what cannot be done in principle since this may have some
indirect relevance to astrophysical situations.
Let me conclude by making a few highly speculative remarks. In the first
place, suppose we take what might be referred to, now, as the most “conservative”
point of view available to us, namely that GIC is not only true, but it also represents
the only type of situation that can result from a gravitational collapse. Does it
follow, then, that nothing of very great astrophysical interest is likely to arise
out of collapse? Do we merely deduce the existence of a few additional dark
“objects” which do little else but contribute, slightly, to the overall mass density
of the universe? Or might it be that such “objects”, while themselves hidden from
direct observation, could play some sort of catalytic role in producing observable

7 Calculations show that this can indeed be done. A particle p0 is thrown from S into the region
between L and H, at which point the particle splits into two particles p1 and p2 . The particle p2
crosses H , but p1 escapes back to S possessing more mass-energy content than p0 !
Gravitational Collapse 1161

Figure 5. Rotating “black hole” (Kerr-Newman solution with m2 > a 2 + e2 ). The inhabitants of the
structures S and S ∗ are extracting rotational energy from the “black hole”.

effects on a much larger scale. The “seeding” of galaxies is one possibility which
springs to mind. And if “black holes” are born of violent events, might they not
occasionally be ejected with high velocities when such events occur! (The one
1162 R. Penrose

thing we can be sure about is that they would hold together!) I do not really want
to make any very specific suggestions here. I only wish to make a plea for “black
holes” to be taken seriously and their consequences to be explored in full detail.
For who is to say, without careful study, that they cannot play some important part
in the shaping of observed phenomena?
But need we be so cautious as this? Even if GIC, or something like it, is
true, have we any right to suggest that the only type of collapse which can occur
is one in which the space-time singularities lie hidden, deep inside the protective
shielding of an absolute event horizon? In this connection it is worth examining
the Kerr-Kewman solutions for which m2 < a 2 + e2 . The situation is depicted in
Fig. 6. The absolute event horizon has now completely disappeared! A region of
space-time singularity still exists in the vicinity of the centre, but now it is possible
for information to escape from the singularity to the outside world, provided it
spirals around sufficiently. In short, the singularity is visible, in all its nakedness,
to the outside world!
However, there is an essential difference between the logical status of the
singularity marked at the centre of Fig. 6 and that marked at the centres of
Figs. 4 and 5. In the cases of Figs. 4 and 5 there are trapped surfaces present, so
we have a theorem which tells us that even with generic perturbation a singularity
will still exist. In the situation of Fig. 6, however, we have no trapped surfaces,
no known theorem guaranteeing singularities and certainly no analogue of GIC.
So it is really an open question whether a situation remotely resembling Fig. 6 is
ever likely to arise.
We are thus presented with what is perhaps the most fundamental unanswered
question of general-relativistic collapse theory, namely: does there exist a “cosmic
censor” who forbids the appearance of naked singularities, clothing each one in
an absolute event horizon? In one sense, a “cosmic censor” can be shown not to
exist. For it follows from a theorem of Hawking [19] that the “big bang” singularity
is, in principle, observable. But it is not known whether singularities observable
from outside will ever arise in a generic collapse which starts off from a perfectly
reasonable nonsingular initial state.
If in fact naked singularities do arise, then there is a whole new realm opened
up for wild speculations! Let me just make a few remarks. If we envisage an
isolated naked singularity as a source of new matter in the universe, then we do
not quite have unlimited freedom in this! For although in the neighbourhood of
the singularity we have no equations, we still have normal physics holding in
the space-time surrounding the singularity. From the mass-energy flux theorem
of Bondi et al. [36] and Sachs [37], it follows that it is not possible for more
mass to be ejected from a singularity than the original total mass of the sys-
tem, unless we are allowed to be left with a singularity of negative total mass.
(Such a singularity would repel all other bodies, but would still be attracted by
them!)
Gravitational Collapse 1163

While in the realm of speculation concerning matter production at singular-

ities, perhaps one further speculative remark would not be entirely out of place.
This is with respect to the manifest large-scale time asymmetry in the behaviour
of matter in the universe (and also the apparent large-scale asymmetry between
matter and antimatter). It is often argued that small observed violations of T (and
C) invariance in fundamental interactions can have no bearing on the cosmological
asymmetry problem. But it is not at all clear to me that this is necessarily so. It is
a space-time singularity (i.e. presumably the “big bang”) which appears to govern

Figure 6. A “naked singularity” (Kerr-Newman solution with m2 < a 2 + e2 ).

1164 R. Penrose

the production of matter in the universe. When curvatures are fantastically large—
as they surely are at a singularity—the local physics will be drastically altered.
Can one be sure that the asymmetries of local interactions will not have the effect
of being as drastically magnified?
When so little is known about the geometrical nature of space-time singu-
larities and even less about the nature of the physics which takes place there, it is
perhaps futile to speculate in this way about them. However, ultimately a theory
will have to be found to cope with the situation. The question of the quantization of
general relativity is often brought up in this connection. My own feeling is that the
purpose of correctly combining quantum theory with general relativity is really
somewhat different. It is simply a step in the direction of discovering how nature
fits together as a whole. When eventually we have a better theory of nature, then
perhaps we can try our hands, again, at understanding the extraordinary physics
which must take place at a space-time singularity.

REFERENCES

[1] S. Chandrasekhar: Roy. Astr. Soc. Month. Not., 95, 207 (1935).
[2] J. R. Oppenheimer, G. Volkoff: Phys. Rev., 55, 374 (1939).
[3] A. S. Eddington: Nature, 113, 192 (1924).
[4] D. Finkelstein: Phys. Rev., 110, 965 (1958).
[5] W. Rindler: Roy. Astr. Soc. Month. Not., 116, 6 (1956). (8)
[6] R. Penrose: in Battelle Rencontres (ed. C. M. De Witt and J. A. Wheeler) (New York, 1968).
[7] R. P. Kerr: Phys. Rev. Lett., 11, 237 (1963).
[8] E. T. Newman, E. Couch, K. Chinnapared, A. Exton, A. Prakash and R. Torrence: Journ. Math.
Phys., 6, 918 (1965).
[9] B. Carter: Phys. Rev., 174, 1559 (1968).
[10] R. H. Dicke and H. M. Goldenberg: Phys. Rev. Lett., 18, 313 (1967).
[11] R. V. Pound and G. A. Rebka: Phys. Rev. Lett., 4, 337 (1960).
[12] A. Schild: in Relativity Theory and Astrophysics, Vol. 1: Relativity and Cosmology (ed. J.
Ehlers)(Providence, R.I. 1967).
[13] C. Brans and R. H. Dicke: Phys. Rev., 124, 925 (1961).
[14] A. Einstein and J. Grommer: S. B. Preuss. Akad. Wiss., 1, 2 (1927).
[15] R. Penrose: Phys. Rev. Lett., 14, 57 (1965).
[16] S. W. Hawking and R. Penrose: Proc. Roy. Soc., A (in press) (1969).(9)
[17] S. W. Hawking: Proc. Roy. Soc., A294, 511 (1966).
[18] S. W. Hawking: Proc. Roy. Soc., A295, 490 (1966).
[19] S. W. Hawking: Proc. Roy. Soc., A300, 187 (1967).
[20] S. W. Hawking and G. F. R. Ellis: Astrophys. Journ., 152, 25 (1968).
[21] R. W. Lindquist and J. A. Wheeler: Rev. Mod. Phys., 29, 432 (1957).
[22] E. M. Lifshitz and I. M. Khalatnikov: Advances in Phys., 12, 185 (1963).
[23] C. W. Misner: Phys. Rev. Lett., 22, 1071 (1969).

(8) The correct page number is 662. This paper was reprinted in Gen. Rel. Grav., 34, 133 (2002) [Editor].
(9) The reference is: Proc. Roy. Soc. London A314, 529 (1970) [Editor].
 Gravitational Collapse 1165

[24] A. K. Raychaudhuri: Phys. Rev., 98, 1123 (1955).(10)

[25] K. Gödel in: Albert Einstein Philosopher Scientist, edited by P. A. Schilpp (New York, 1959),
p. 557.
[26] W. Israel: Phys. Rev., 164, 1776 (1967).
[27] W. Israel: Commun. Math. Phys., 8, 245 (1968).
[28] T. Regge and J. A. Wheeler: Phys. Rev., 108, 1063 (1957).
[29] A. G. Doroshkevich, Ya. B. Zel’dovich and I. D. Novikov: Žurn. Ėksp. Teor, Fiz. 49, 170 (1965).
English trans., Sov. Phys. JETP, 22, 122 (1966).
[30] K. S. Thorne: Ph. D. thesis, Princeton University Princeton, N. J. (1965).
[31] B. Carter: personal communication (1969).
[32] I. Robinson and A. Trautman: Proc. Roy. Soc. A265, 463 (1962).
[33] J. Foster and E. T. Newman: Journ. Math. Phys. 8, 189 (1967).
[34] E. T. Newman: personal communication (1969).
[35] C. W. Misner: personal communication (1968).
[36] H. Bondi, M. G. J. van der Burg A. W. K. Metzner: Proc. Roy. Soc. A269, 21 (1962).
[37] R. K. Sachs: Proc. Roy. Soc. A270, 103 (1962).

(10) This paper was reprinted in Gen. Rel. Grav. 32, 749 (2000) [Editor].