PHYSICAL REVIEW D VOLUME 41, NUMBER 6 15 MARCH 1990

Internal structure of black holes

Eric Poisson and Werner Israel

Canadian Institute for Advanced Research Cosmology Program, Theoretical Physics Institute,
University of Alberta, Edmonton, Alberta, Canada Ti5G 2JI
(Received 16 October 1989)
The gravitational effects associated with the radiative tail produced by a gravitational collapse
with rotation are investigated. It is shown that the infinite blueshift of the tail's energy density
occurring at the Cauchy horizon of the resulting black hole causes a classically unbounded inflation
of the effective internal gravitational-mass parameter of the hole. Since this effect is causally discon-
nected from any external observer, the black-hole external mass remains bounded. The mass
inflation phenomenon causes the spacetime curvature to grow to Planckian scales on a spacelike hy-
persurface in the vicinity of the Cauchy horizon, beyond which the classical laws of general relativi-
ty break down. A consequence is that an observer's trip to this hypersurface embraces all but the
last Planck time of the entire black-hole classical history.

I. INTRODUCTION lapse are not sufficient to predict unambiguously what

happens to the future of the Cauchy horizon.
Black-hole theory is without any doubt one of the ma- In the analytically extended Kerr manifold, there is a
jor triumphs of classical general relativity. By a series of relatively spacious region beyond the Cauchy horizon: it
theorems' it has been established (subject to the plausible takes roughly the same proper time for a free-falling ob-
assumption of cosmic censorship ) that the external grav- server to travel from the event horizon to the Cauchy
itational field of a black hole relaxes to a Kerr-Newman horizon as from the Cauchy horizon to the ring singulari-
field described solely by three parameters: the hole's ty. The presence of a region beyond the Cauchy horizon
mass, charge, and angular momentum. This remarkable is an embarrassment: there is no way of predicting the
result is usually referred to as the no-hair theorem. The course of events in this region and signals coming from
mechanism responsible for such a relaxation of the exter- the singularity could alter the physics in an unforeseeable
nal field has been elucidated by Price who showed that manner. There is a further problem associated with the
perturbations developing on the surface of a spherically Cauchy horizon: it is a surface of infinite blueshift. If we
collapsing star produce the emission of gravitational radi- recall that infalling gravitational radiation is expected to
ation which carries away all the initial characteristics of propagate inside the black hole with paths approaching
the star's gravitational field except the mass, charge, and that of the null generators of the Cauchy horizon, we
angular momentum parameters. This radiation then in- realize that the energy density of this radiation will suffer
teracts with the spacetime curvature: while some of it es- an infinite blueshift as it approaches the Cauchy horizon.
capes to infinity, some is backscattered and absorbed by A free-falling observer would see the entire future history
the resulting black hole. Most of this backscattering of the Universe flash before his eyes before encountering
occurs soon after the emission: the amplitude of the in- a wall of infinite density at the Cauchy horizon.
falling flux typically decays according to an inverse The first-order analysis of this phenomenon was under-
power law with advanced time. taken by many people and that perturbations diverge to
Attempts have been made to extend these results to linear order has been verified many times over. This re-
the black-hole internal gravitational field. For a sult suggests that in a generic gravitational collapse, a
Schwarzschild black hole, it has been demonstrated ' singularity developing at the Cauchy horizon would seal
that the asymptotic portion of spacetime near the singu- off the "Kerr tunnel" that leads to other asymptotically
larity (corresponding to large values of advanced time v) flat universes, but so far, nobody has attempted to ana-
is virtually free of aspherical perturbations propagated lyze the situation beyond linear order or to evaluate the
from the surface of the star since the gravitational radia- effect on the black-hole geometry of such an infinite
tion becomes infinitely diluted as it reaches the singulari- growth of the perturbations at the Cauchy horizon. The
ty. But this result breaks down as soon as the charge or question of interest is to evaluate whether the perturba-
angular momentum of the black hole is nonzero. The tions, allowed to act as a source in Einstein's equations,
internal structures of the Reissner-Nordstrom and Kerr can trigger the formation of a singularity of sufficient
solutions differ drastically from that of the Schwarzschild strength (e.g. , stronger than that of a surface layer) to
solution: the singularity is now timelike and both of effectively stop the evolution of spacetime at the Cauchy
these spacetimes possess a Cauchy horizon, a null hyper- horizon. This property would of course take care of the
surface beyond which predictability breaks down. problems associated with the breakdown of predictability
Indeed, initial data specified (say) at the onset of the col- occurring there.

1796 1990 The American Physical Society

41 INTERNAL STRUCTURE OF BLACK HOLES 1797

This question has been addressed by the present au- hole is charged and initially described by the Reissner-
thors in two papers. We have shown that the answer is Nordstrom solution. This might appear to be a highly
yes: the combination of the outflux emitted by the col- restrictive idealization, but we believe that this model can
lapsing star and its backscattered, blueshifted radiative capture the essential physics, since the global structures
tail propagating near the Cauchy horizon provoke a of the Kerr and Reissner-Nordstrom black holes are very
tremendous inflation of the black-hole internal mass pa- similar. Both manifolds possess a Cauchy horizon and
rameter, which in fact becomes classically unbounded at according to our physical picture, the mass inflation
the Cauchy horizon. Mass inflation then produces the phenomenon only depends on the two general features
inflation of curvature which too grows to infinity at the described above: the presence of a highly blueshifted
Cauchy horizon. The Cauchy horizon is the ultimate influx and a separation between the Cauchy and inner ap-
brick wall at which the evolution of spacetime is forced parent horizons. The spherical model should thus allow
to stop. a qualitative understanding of the phenomenon without
The mechanism responsible for this catastrophic be- introducing too many diSculties in the mathematical
havior is precisely the combination of the effects of the analysis. The second assumption. is that we will model
two radiation fluxes. At first sight, it might seem that the gravitational radiation emitted from the surface of
only the blueshifted influx is the critical ingredient and the collapsing star and backscattered by the background
that one could artificially turn off the outflux of radiation curvature as two intersecting radial streams of infalling
without altering the general conclusions. Indeed, one and outgoing lightlike particles following null geodesics.
would expect the gravitational effects of this blueshifted We furthermore assume that the streams do not interact
energy to increase the black-hole gravitational mass pa- with each other so that they are separately conserved.
rameter to arbitrarily large values. [It might seem that This too might appear as an important idealization, but
inside the black hole, the mass parameter has no direct since we are only interested in the highly blueshifted
operational meaning since one cannot go to infinity in or- modes propagating near the Cauchy horizon, it is ap-
der to measure it. But the mass parameter always parent that any further scattering should not be impor-
possesses an important local meaning: it determines the tant and that Issacson's effective stress-energy descrip-
"Coulomb" (Petrov type-D) component of the local cur- tion for these high-frequency modes should be an ade-
vature. An increase in the mass parameter would be quate description. Indeed, since the radius of curvature
manifested locally, for example, by an increase in the ti- of the background spacetime is always larger than the
dal forces felt by an (extended) observer falling radially blueshifted wavelength of the infalling modes, the gravi-
inward near the Cauchy horizon, even though this ob- ton geometric optics approximation should be an accu-
server would register the energy density due to the infal- rate approximation. As for the outgoing modes, we will
ling radiative tail as almost unblueshifted and vanishingly see that their detailed description is not at all important
small. ] for our conclusions: their only role is to irradiate the
The expectation of a mass growth near the Cauchy Cauchy and inner apparent horizons in order to produce
horizon due to the blueshifted radiative tail is not real- a separation.
ized: the mass parameter remains almost unchanged as The mass inflation phenomenon completely changes
long as the outflux is not turned on. This apparent mira- our understanding of the black-hole internal structure.
cle can be explained as follows: while the Cauchy hor- Instead of having the Cauchy horizon acting as a curtain
izon is a surface of infinite blueshift for our own asymp- beyond which a macroscopically large region of space-
totically flat universe, the inner apparent horizon of the time does not allow predictability from initial data
hole is a surface of infinite redshift for future asymptoti- specified earlier, we have near the Cauchy horizon a mi-
cally flat universes. In the situation where only the influx croscopic region of spacetime where the curvature is ex-
is considered, the two horizons coincide and the blueshift tremely high. If we imagine, as was speculated recent-
and redshift cancel exactly thus preventing any spectacu- ly, ' that curvature is naturally bounded at Planckian
lar increase in the mass parameter. When one turns on magnitude, we find that the effective mass parameter of
the outflux, however, the situation is radically different. the black-hole interior can reach the incredibly high
The outgoing radiation escaping the surface of the col- value of mo/mp, —10 universe masses for a black hole
lapsing star crosses the Cauchy horizon and focuses its of ten solar masses (mo is the external mass of the hole
null generators. But the apparent horizon now contracts and mp~ the Planck mass). In view of this fantastic in-
faster (in fact, it debates very rapidly) and the surfaces of crease of the mass, the charge and angular momentum of
infinite blueshift and redshift become distinct. The can- the black hole become totally irrelevant to the descrip-
cellation can therefore no longer occur and the mass pa- tion of the geometry near the Cauchy horizon, which is
rameter increases dramatically. One can think of the then accurately described by the Schwarzschild metric
infinitely blueshifted influx as being the essential feature with an enormous mass parameter. It is then easy to
explaining mass inflation, but the phenomenon needs to show that an observer's trip to the Cauchy horizon cov-
be triggered by an arbitrarily small amount of outgoing ers all but the last Planck time of the black hole's classi-
radiation which produces the necessary separation be- cal history. The spacetime region near the Cauchy hor-
tween the Cauchy and apparent horizons. izon can then be imagined as a "fat cigar" (topology
In order to formulate the problem mathematically, it is S XR+ ) inside which the curvature is Planckian.
necessary to make a few simplifying assumptions. The The paper is organized as follows: motivated by the
first is that we will use a spherical model where the black discussion given above, we will consider as our model a
1798 ERIC POISSON AND WERNER ISRAEL

background Reissner-Nordstrom spacetime irradiated by G, b = —[2rr , +g,

b & (1 'r,
r— r
2— r) ]/r
simultaneous infalling and outgoing radial fluxes. We (2.2)
will then derive in Sec. II the basic field equations for 6&0=sin 06&&=rOr —'r —, R .
spherical spacetimes, expressed in a form covariant under
The semicolon denotes covariant differentiation with
arbitrary transformations of the coordinates of the radial
two-spaces (8, $)=const. In Sec. III we will explore
some exact solutions to the field equations: namely, the
static Reissner-Nordstrom solution and the dynamic in-
going and outgoing generalized Vaidya solutions which
respect to the two-dimensional

dimensional metric g &], we write

metric g, b [we use the
stroke ( ) to denote the same with respect to the four-
~

1(—g' ll. b for any

scalar field P and R denotes the two-dimensional Ricci
scalar associated with the radial two-spaces (we use the
: ,
describe a charged, spherical black hole irradiated by a
superscript 4 to indicate four-dimensional quantities,
pure in- or outflux. We will also derive the Dray-
't Hooft —Redmount (DTR) relation which describes the when ambiguities might arise}.
We write Einstein's field equations as
gravitational field before and after the collision of two
spherical thin shells propagating at the speed of light, one G,p=8m(E p+T p), (2.3)
expanding, the other contracting. The DTR relation is
an exact solution to Einstein's equations for infalling and where E & is the Maxwellian contribution to the stress-
outgoing fluxes which can be described by 5-function energy tensor representing the static electric field of a
pulses. We then proceed in Sec. IV with the formulation point charge of strength e located at the origin r =0:
of the mass inflation phenomenon by formally integrating E &=P,~diag( —1, —1, 1, 1), P, —e /8nr (2.4)
Einstein s equations with continuous infalling and outgo- ~

ing radial fluxes. We show than it is possible to reach our Expression (2.4) is valid in any system of coordinates
conclusions making no additional assumption other than (x', 8, $) and follows from the hypothesis that the elec-
that given in our model. To evaluate the growth rate of tromagnetic field (as seen by static observers) should be
the gravitational mass, however, we must make a formal purely electric and radial. The non-Maxwellian contribu-
perturbation expansion in terms of the product of the flux tion to the stress-energy tensor T & will later describe our
luminosities. Section V is finally devoted to quantum cross flow of lightlike radiation. For the moment, we
considerations: we speculate on the internal structure of shall decompose it according to
the black hole should curvatures be bounded at Plancki-
an magnitude. We summarize our conclusions in Sec. VI. (2.5)
Various technical details which might obscure the main
where P represents a tangential pressure; we will also
line of thought are relegated to the Appendixes.
denote the two-dimensional trace T', by the symbol T.
The field equations then read
II. FIELD EQUATIONS
FOR SPHERICAL SPACETIMES 2rr , + (1
&
e /r . 'r,
r— r Hr)g, b
2— =—8nrT, b, — .
(2.6)
We have given some justification in the Introduction r r —'r g —e /r =8~r P.
for using a spherical model in which the black hole is ini-

:
tially described by the Reissner-Nordstrom solution. To At this point, it is useful to introduce the scalar fields
describe the effects of the gravitational radiation propa- m f
(x'), (x'), and x(x') defined as
gating inside the hole, we will be interested in the field g' r, r b=f— —2m/r+e 1 /r
equations describing a charged, spherical black hole per-
turbed by cross-flowing radial fluxes of infalling and out-
going radiation. This idealized model should illustrate
a = —'B„f = —(m e
—, /r)/r— (2.7)

the basic physics behind the mass inflation phenomenon Substituting Eqs. (2.7) into (2.6), we obtain
and is sufficiently simple to allow a detailed mathematical r ,b (a+ Clr)g, b—
=. 4vrrT, b, —
analysis. We now proceed to derive the field equations (2.8)
describing the above situation. In doing so, it is most r —e /r —'rR—, =8m. rP .
convenient to use a coordinate system x =(x', 8, $)
(a =1,2), where the coordinates x' of the "radial" two- Taking the trace of the first of Eqs. (2.8) yields
spaces (0, $) =const are left unspecified. The field equa- Or = — 2x+4m. r T which can be substituted back to yield
tions as thus written will be covariant under arbitrary r ,b+zg, b = [Link](T, b —g, b T) —
. . (2.9)
transformations of the two-dimensional coordinates. The
spacetime metric will then be written as Substituting the same result into the second of Eqs. (2.8)
—g b then gives
dS x dx +r dA (2. 1)
R —2B„&=m(8T —2P) . (2. 10)
where g, b is the metric on the radial two-spaces, dQ the
line element on the unit two-sphere, and r a function of Equations (2.9} and (2. 10) constitute our basic field equa-
x' measuring the area of the two-spheres x'=const. The tions. It is however useful to derive an additional equa-
various geometric quantities for such a metric are given
in Appendix A, where it is shown that the Einstein tensor
tion by using Eqs. (2.7}: we obtain = —(2/r)m,
—2ar „which can be inverted to yield m, in terms of f,
is given by f, and r,
. We can then use the field equation (2.9) us-
41

ing f, =2g 'r

m, =4nr(. T,
br „. . little algebra then gives
A
5— , T)r b .
INTERNAL STRUCTURE OF BLACK HOLES

(2. 1 1)
Note that Um depends only on the product p;~, and
not on the individual linear contributions.
„,
The third
wave equation requires a little more work. We want to
1799

We note that Eq. (2. 11) follows directly from the conven- find an expression for Olnf which will be used in Sec. III
tional form of the field equations (obtained when one uses to derive the DTR relation. First note that, for any func-
r and t as coordinates) and the requirement of two- tion it,
dimensional covariance.
Olney=(QOQ f'—
g, )If (2. 18)
The conservation equations (E +T )~&=0 can be
obtained by using the Christoffel symbols given in Ap- Now, we can calculate f by taking the second deriva-
pendix A. Since E ~ and T ~ are separately conserved, tive of the first of Eqs. (2.7) and get

Of = —2(Om/r —m' r, /r +aOr+s, r'),

we find that
(2. 19)
2Tab} ( 2);ap (2. 12)
which can be simplified further if we use Eqs. (2. 16}, and
The angular components of T ~~13=0 are satisfied trivial- (2. 14) in a,= ,'Rr, — —m,Ir . A little algebra then
ly, while the conservation equations for the electromag- yields
netic stress-energy tensor confirm that P, must be pro-
portional to r as was stated earlier.
~

Of = —2 m/r +4m'r, /r +4afR. —

. (2.20)
In summary, we find that the field equations for spheri-
cal spacetimes for which the stress-energy tensor can be
The product f'f,involves the product m'm, which
can be recast in terms of m in the following way:
decomposed into two separately conserved Maxwellian m'm, = ——,'rfOm. Substituting this result into the ex-
and non-Maxwellian contributions are given by Eqs. (2.9)
and (2. 10) and that energy conservation is expressed by
pression for f 'f,yields

Eq. (2. 12). Then Eq. (2. 11) follows directly from the
definition of m and Eq. (2.9). We note that similar equa-
f 'f, = —2f Om Ir + 8ar'm, /r +4m f. (2.21)

tions were previously written down in Ref. 10. Collecting the results finally gives
We now specialize to the problem of interest. We want Olnf =16rrf (f ,r
2sr)T' r— b
—R, (2.22)
to construct a stress-energy tensor T which would de-
scribe cross-flowing streams of infalling and outgoing which is the result wanted. Note that Olnf is linear in
lightlike particles following radial null geodesics. It is T' and that f 2sr =1 — e /r — The .two-dimensional
easy to see that it must take the form Ricci scalar is given explicitly in Eq. (2.14).
4
ab ab Pinra Ib +Poui a b (2. 13)
where I, is a radial null vector pointing inwards and n, is III. EXACT SOLUTIONS:
a radial null vector pointing outwards. The scalars p;„
„,
and p, represent the energy density of the fluxes but do
not have direct operational meaning since the null vec-
REISSNER-NORDSTROM, VAIDYA,
AND THE DTR RELATIONS

tors can be arbitrarily normalized. Since we now have

P = T =0, the field equations simplify to We will explore in this section a few exact solutions of
the field equations derived in the previous section. The
6+Kg b
= 4&rT b solution corresponding to a vanishing non-Maxwellian
(2. 14)
R =28„a=2(2m —3e Ir)/r3, stress-energy tensor is the Reissner-Nordstrom solution
and describes the geometry of a static, charged, spherical
=4mr T, "r b, (r T'") b=0 . (2. 15) black hole. We shall explore this solution in Sec. III A.
m, .
If we switch on either a pure influx, or pure outflux, of
radiation (described T,b =p;„I,lb or
From this system of equations, it is possible to derive
three one-dimensional wave equations of the form Og=p
which are very useful since, as we will show in Sec. III,
„,
T,b =p, n, nb), we obtain
either by
the charged Vaidya solution"
(a generalization of the original Vaidya solution) which
describes a charged black hole whose mass varies with ei-
they can be formally integrated. First, taking the trace of
the first of Eqs. (2. 14} while noting that the trace of T,b is ther advanced time (in the case of pure inflow) or retard-
ed time (in the case of pure outflow). We shall consider
zero yields
this solution in Sec. III B. An exact solution to the full
Gr= —2z. (2. 16) field equations (inflow and outflow} exists in the special-
ized case where the fluxes can be modeled as thin shells
Second, taking the derivative of the first of Eqs. moving at the speed of light. The energy densities p;„and
(2. 15} and using the conservation equation yields
Om =4mr T' r b. Equation (2. 14) . can then be used to
, „,
p, can then be expressed as 5 functions and the field
equations can be integrated in the generalized form of the
give Dray —'t Hooft —Redmount (DTR) relation. ' This rela-
Om = (4') r T' T,b . — (2. 17} tion connects the masses in the different regions of space-
time separated by the shells. We shall explore this solu-
This equation wiH be very important in what follows. tion in Sec. III C.
 ERIC POISSON AND WERNER ISRAEL

„.
1800 41

A. The Reissner-Nordstrom solution relates m;„(v)and p;„as

The Reissner-Nordstrom solution dm;„(v)/dv =4mr p, (3.7)
ds = dt
fo— +fo 'dr +r dQ The ingoing Vaidya solution represents a spherical,
(3. 1) charged black hole which is irradiated by an influx of ra-
fo=1 2—
mo/r +e /r dial lightlike radiation corning from Jz (Fig. 1) and fol-
is the solution to Einstein s equations with vanishing T &. lowing ingoing null geodesics. Similarly, the outgoing
The solution can be easily obtained by choosing x'=(t, r) charged Vaidya solution
and by writing g, &dx'dx"= e~— fdt
:
g=P(r) and =1 2m—(r)/r+e /r . Field equation
f
+f
'dr, where
(2. 15) yields that m (r) =const—mo and the tt component
of Eq. (2. 14) yields that P'=0. The factor e ~ is therefore
a constant that can be absorbed by a rescaling of coordi-
ds

f,„,
=du(2dr
=1 —2m, „,
(u)/r
f,„,
d—
u)+r
+e /r
dA

represents the black-hole interior being irradiated by an

(3.8)

nate t to give Eq. (3. 1). outfiux of lightlike radiation coming from 2L. Alterna-
For later use, we will now define other convenient tively, we can imagine that the radiation really comes
coordinate systems to express the Reissner-Nordstrom from the surface of the collapsing star (this is obviously
metric: namely, the Eddington-Finkelstein null coordi- the correct interpretation since the "parallel" universe to
nates and the Kruskal null coordinates. So we define ra- the left does not appear when a star is present) but since
dial null coordinates u and v (the Eddington-Finkelstein
coordinates) which we choose such that they both run
dm, „, =
( u ) /du 4m „,
r p, , (3.9)
forward inside the black hole's event horizon (see Fig. 1): for T,& =p, „,nt„
n, n, =—
B, u, the mass parameter actu-
= —t+r*, ally increases even though the star loses mass. This prop-
u v =t+r*, (3.2) erty corresponds to the fact that an observer near ZL
where r'=Jdr/fo(r) Note .that far from the black would measure an increase of mass as radiation pours
hole, u ~— t+r, v ~t+r
and that u is in fact an ad- into the hole from his domain.
One could have chosen other null coordinates to ex-
vanced time for observers in universe IL. The metric
(3.1) then takes the form press the Vaidya solutions (3.6) and (3.8). But the
Eddington-Finkelstein coordinates are a good choice
ds =fodu dv +r dQ (3.3) since they reduce to ordinary advanced and retarded
times far from the black hole. Also, since the mass func-
Coordinates u, v are singular on the black-hole horizons, tions are objects one can measure at infinity (they have a
but if one focuses attention on one horizon at a time, it is
direct operational meaning), the derivatives dm;„/dv and
possible to rescale the coordinates such that the metric
becomes manifestly regular on that horizon. Let us then
focus attention on the inner horizon r =ro —mo —(mo
—e )'~ and define the rescaled coordinates U, V (the
Kruskal null coordinates) as
U= —e Kpg
V= —e K
pU
(3 4)
where = —fo(ro)/2=(mo —e
ao— )' /ro is the surface
gravity of the inner apparent horizon. That the metric is
now regular at r =ro follows easily. We find that near
the horizon fo= — 2UV and the metric element gUr
hence reads

gUv= ' fo/a'oUV = —I/—ao . (3.5)

We will make use of the results derived here in Sec. IV.

B. The charged Vaidya solutions

The ingoing charged Vaidya solution"

ds =dv(2dr f;„dv)+r dQ—
(3.6)
f;„=1—2m;„(v)/r+e /r
FIG. 1. A con, formal diagram representing the Reissner-
is the solution to Einstein's equations with T,I, =p;„I,II, Nordstrom spacetime: we illustrate here the coordinates
where the normalization of I, is such that I, = —B, v. defined in the text. Coordinate v is defined to be an advanced
Coordinate v has here the same meaning as in Eq. (3.2): time for observers in region I& and u is defined to be an ad-
it reduces to the ordinary advanced time t +r far from vanced time for observers in region II (note that coordinate t
the black hole. It is easy to see that field equation (2. 15) runs backward in this region).
 INTERNAL STRUCTURE OF BLACK HOLES 1S01

„,
dm, /du also have direct operational meaning: they are
the rates of variation of the black-hole gravitational mass
yields, when substituted into Eq. (2. 14):
o = —B„a=—(2m 3e—Ir)lr (3. 12)
as measured far from the black hole. It is through Eqs.
(3.7) and (3.9) that the energy densities p;„and p, ac-
quire operational meaning.
„, The stress-energy
we now choose
tensor takes the usual form (2. 13) and

C. The DTR relation I. = —a. v, n. = —a. U. (3. 13)

The conservation equations (2. 15) express the fact that
We are now ready to derive an exact solution to the
Einstein equations when we have a crossbow of infalling
and outgoing lightlike radiation which can be modeled as
„,
the product r p;„(rp, ) is independent
U( V) so that we can write
of coordinate

spherical thin shells moving with the speed of light, one

expanding, the other contracting (see Fig. 2). The shells
p;„=L;„( „,=L,„,
V)/4nr, p, ( U) I4~r (3. 14)
divide the radial two-space into four different sectors, L;„and L,„,
The luminosities have no direct operational
each possessing a different mass and a different function meaning since they depend on the definition of the coor-
f. The relationship between the masses is precisely the dinates U and V. Substituting these results into Eq. (2. 17)
(generalized) DTR relation' which we shall now derive yields
(note that the original DTR relation is only valid for a
vanishing charge). In order to do so, and for later use, it Clm = —2(re ) 'L;„(V)L, „,( U) (3.15)

„, „,
is necessary to express the field equations (2. 14)—(2. 17) in
radial double-null coordinates in which the two-
and we note that lnf is linear in L;„and L, without
dimensional metric assumes the form any contribution from the bilinear L;„L,
. It is precisely
this property which allows the integration of the field
g b
x'dx = —2e dUdV, (3. 10) equations to obtain the DTR relation, as we will see
presently.
where sr =0(U, V); for the time being we leave the null Before we can do so, however, we need to derive a very
coordinates arbitrary: we do not necessarily assume that useful result. We will show, as is well known, that any
they are the Kruskalized coordinates of Eq. (3.4). We equation of the form 01(|=p can be formally integrated.
find that, for any scalar function P, Suppose we are interested in the solution for g subject to
(3. 1 1) boundary conditions on a characteristic sector X de-
scribed by the equations U = U, , V = V, . It follows from
The Ricci scalar is calculated to be R = —200. and this a straightforward application of Green's identity that the
solution at event ( U, V) (assumed to be located to the fu-
ture of sector X) is given by

Q(U, V)= ——,' f f e p'dU'dV'+f(U„V)

l 1

+f(U, V)) —f(U„V,), (3. 16)

where we denote o'=0(U', V'), etc. The simplicity of
this result comes from the fact that the Green's function
for the operator 0, when expressed in double-null coordi-
nates, is given by the product of two Heaviside step func-
tions. One can see that the solution at event (U, V) is
given by a contribution from the boundary conditions
specified on sector X, and by a surface integral over the
past radial light cone of event (U, V). It is straightfor-
ward to differentiate Eq. (3. 16) twice and to use Eq. (3.11)
to verify that it is truly a solution of g=p. This in-
tegral formula will be very useful in the following section.
The DTR relation now follows directly from Eq. (3. 16)
applied to g = lnf, by recalling — that p is linear in L;„and
„,
L, which are here given by 5 functions. If we take our
integration over an arbitrarily small lightlike rhombus
FIG. 2. Crossing null shells: two concentric, spherical thin around the point of collision, event q (Fig. 2), the absence
shells propagating with the speed of light collide without in-
teraction at event q, hence separating spacetime into four radial
of any bilinear term L;„L, „,
in p ensures that the integral
contribution will be arbitrarily small. We therefore get
sectors A, . . . , D. The energy content of the infalling she11 is the generalized DTR relation
given by the difFerence mc — m& and that of the outgoing shell
(first expanding from the surface of the collapsing star but then
forced to collapse to the singularity because of the gravitational
f„(q)fs(q) =fc(q)fp(q), (3.17)
pull) by mD — m&. The solution of the Einstein equations, the
DTR relation, relates the mass m& to the masses m&, . . . , mD. m „,
which expresses the relationship between the four masses
. . . , mD. It is remarkable that an exact solution to
1802 ERIC POISSON AND WERNER ISRAEL 41

the complicated cross-flow field equations does exist and energy tensor is given by
that it is so simple.
T,i, = [L;„(
V) /4m r ]I, lb + [L, „, ( U) /4m r ]n, nb, (4.2)
IV. MASS INFLATION where 1, = — 8 V, n, = —8, U.
The relation between the
We now have everything in hand to formulate the mass mass function and the metric element e is
inflation phenomenon. We consider a situation where an
initially static Reissner-Nordstrom black hole with mass
1 2—m (U, V)/r+e /r =f = —2e (BUr)(B„r) (4. 3)
m, is perturbed by a crossflow of outgoing and infalling and the field equations for m are
lightlike fluxes (see Fig. 3). More precisely, at advanced
time V, we switch on the infalling flux, while at retarded
time U& we switch on the outgoing flux. Therefore for
BUm
= L, „(—
„,
= —L, ( U)e d~r,
(4.4)
Bvm V)e dUr .
U & U&, V& V, spacetime geometry is described by the
static Reissner-Nordstrom solution with mass m, . For Its wave equation is given by Eq. (3.15).
V & V& but U & U„ the ingoing Vaidya solution takes The relation between L;„(V) and the measured quanti-
over with a mass function m;„(V) which reduces to m, at ty dm;„(v)/dv (v is, as always, the Eddington-Finkelstein
V = Vi and to the asymptotic limit mo on J"+„.Similarly, advanced time) can be evaluated in the pure inflow region
for U & U, but V & V, , spacetime is described by the out- where m = m;„, = f;„and where the solution is known
f
crossflow region V & V„U
„,
going Vaidya solution with m, ( Ui ) =m, . In the
& U„spacetime will be de-
to be the ingoing charged Vaidya solution. By continui-
ty, L;„(V) will assume the same form in the cross-flow re-
scribed by the solution to the field equations (2. 14) and gion. By rewriting the second of Eqs. (4.4) in terms of v,
(2. 15) with the appropriate boundary conditions de- and by using Eq. (3.6), it is easy to show that
scribed above. We shall now attempt to find this solu- dm;„(v)
tion. (4.5)
dU
Let us first recall the basic field equations. We shall
continue to use our system of arbitrary radial null coordi- Similarly, we can show from comparing the first of Eqs.
nates since they allow us to use the integral equation (4.4} to the outgoing Vaidya solution in the pure outflow
(3.16). The metric is then given by region that
d$2= —2e ~dUdV+7 (4. 1}
L, „, =
(U)
u dm, (u)„, (4.6)
while the non-Maxwellian contribution to the stress- dQ

To get the actual expression for rn;„(v) we recall the

analysis of Price which showed that the amplitude of the
backscattered gravitational radiation (modeled as a per-
turbing test field) varies as v "at late advanced times. If
I is the multipole order of the perturbing field, n is given
by 21+2. Now the energy density of the ingoing radia-
tion, which is proportional to dm;„/dv, will vary as v
so that the ingoing mass function will reach the asymp-
totic limit mo as mo — m,
' '+ '.
„(v)-v
The dominant
contribution to the influx will come from the quadrupole
moment 1=2 so that typically mo — m;„-U So we ".
can take
dm, „(
v) /dv —v (4.7)
with p =—4(I +1) ~ 12. As for L, (U), we will see later „,
that it is not necessary to write down a precise expression
for it. We will only assume that it is a positive quantity:
the radiation escaping the surface of the collapsing star
should have a positive energy density.
To integrate Einstein s equations in the cross-flow re-
gion, we can make use of the integral formula (3. 16} to
find a formal expression for the mass function. Substitut-
FIG. 3. Background Reissner-Nordstrom spacetime per- ing Eq. (3. 15) into Eq. (3. 16) yields
turbed by cross-flowing streams of radial radiation: we turn on
the influx at time V = V, and the outflux at U = U, . If we sup-
pose that the fluxes are later turned off, spacetime is character-
m ( U, V) = I l
J l
(r'e ') 'L,
„( „,
V')L ( U')dU'd V'

ized by four static regions with masses mo, . . . , m 3 and by + m;„(V}+m, „, — ( U) m, , (4.8)
inflow, outflow, and cross-flow regions. This figure represents
the continuous counterpart of Fig. 2. with L, „(V) given by Eqs. (4.5) and (4.7), while L, „,
( U) is
41 INTERNAL STRUCTURE OF BLACK HOLES 1803

left arbitrary. So far, we have left the coordinates U and role in the contraction of the Cauchy horizon which then
V unspecified; we shall now remedy this situation. Ingo- contracts only moderately. Of course, this is not so for
ing radiation is switched on at V = Vi, while the outflux the inner apparent horizon [described by the function
is switched on at U = U&. To make things conceptually rAH(U, V) defined by f
(rAH )=0] which deflates catas-
clear, we can suppose that the fluxes are switched off at trophically as the mass inflates to infinity. We therefore
later times U2 and V2, we are of course interested in the conclude, from our intuitive analysis, that re will rather
situation where V2 tends to the value of Von the Cauchy tend to go to zero in approaching the Cauchy horizon,
horizon. In this view, spacetime is characterized by four hence feeding further mass inflation. The mass parame-
static regions with masses ma . . m3 and by inflow, ter is therefore bound to inflate to classically arbitrarily
outflow, and cross-flow regions (Fig. 3). Each of the stat- large values.
ic regions can be described by a Reissner-Nordstrom A1though we cannot prove that re actually goes to
solution with appropriate mass, and each static region zero, we will now proceed with a formal proof that re
possesses its own set of Eddington-Finkelstein null coor- cannot go to infinity as we approach the Cauchy horizon.
dinates, defined as in Sec. III. We choose our coordinate We construct the object /=re and derive an expression
u to be the Eddington-Finkelstein advanced time defined for Cling which can be formally integrated using Eq.
with respect to static region m0. With this choice, the (3.16). We will show that the solution for ln(re ) thus
Cauchy horizon lies on the u = ~ surface and u reduces obtained, if not bounded, must nevertheless be bounded
to ordinary advanced time near 2z. Similarly, we take u above. This shows that re cannot go to infinity and
to be the Eddington-Finkelstein retarded time defined verifies our conclusion that the mass parameter becomes
with respect to static region m, . With this choice, u unbounded at the Cauchy horizon. Using Eq. (2. 18) we
reduces to ordinary advanced time for observers near J'I can calculate that Cling=2 o +Or lr fir,
hich can
w—

and the outer horizon of region m& lies on the surface be expressed in terms of r and e by using Eqs. (3. 12),
u = —00. We then choose our coordinate V to be the (2. 16), and (2.7):
Kruskalized advanced time associated with the inner ap- Ging= (3e —r ) /r (4. 13)
parent horizon r =rv=m0 —(m02 — e 2 ) 1/2.
Substituting this in Eq. (3. 16) yields
V= —e Kpu
(4.9)
where v0=(ma e)'~ Ira,— such that V=O on the Cau-
in/( U, V) = in[/,„(Ui, V)g, „, ( U, V, )/P( U„Vi) ]
chy horizon. Similarly, we choose U to be the Kruskal- ——
f f g'r' (3e
' r' )dU'dV— ', (4. 14)
I
ized retarded time associated with the inner horizon 1

r=r, =m, —(m,2 —e 2 ) 1/2. „,

where P;„(g, ) denotes the function re which can be
obtained from the known ingoing (outgoing) charged
U= —e K u
1
(4. 10) Vaidya solution. Now, the first term is finite since it can
where a.
, =(m, e)'~ Ir „s—
uch that U = —~
on the be shown that g;„(Ui, V) is actually finite on the Cauchy
outer horizon of region m &. Note that both coordinates horizon. This nontrivial statement is proved in Appendix
take negative values inside the black hole. B: essentially, 1(, is the radius of the Cauchy horizon r0
„

Now that we have chosen our system of coordinates, times the metric element gzz of the ingoing Vaidya solu-
we can substitute Eqs. (4.7) and (4.9) into Eq. (4.5) to get tion evaluated near the Cauchy horizon; the Appendix
shows that this metric element is indeed regular near the
—ln(
L,„(V)-[ —V)] IV -v i'e (4. 1 1) Cauchy horizon. This is a consequence of the fact that,
when only influx is present, the surfaces of infinite blue-
If we note that, for V~O, shifts and redshifts coincide, thus preventing an un-
f L;„(V)dV-[—ln( —V)] —V)
V
~/( bounded increase of the mass parameter, as discussed in
1 the Introduction. The second term is more serious, it
-u ~ep
KpV
(4. 12) might well be unbounded and what is crucial here is the
sign of the contribution from the integral. If the integral
we conclude from Eq. (4.8) that m ( U, V) will diverge as is actually unbounded, the dominant contribution will
V~O unless the product re goes to infinity quickly come from values of the integrand near the Cauchy hor-
enough. Now, if m(U, V) tends to become singular on izon. But it follows directly from ~e~ &m0 (a necessary
the Cauchy horizon, we expect from Eq. (4.3) that e condition to the existence of the black hole) that ra & ~e~
will tend to go to zero, provided that the product of the so that in the vicinity of the Cauchy horizon, the contri-
derivatives (Bvr)(Bi,r) remains well behaved. We expect bution from the integral will be negative definite. This
this to be the case since our coordinates U and V are well ensures that, if unbounded, in'( is necessarily not bound-
behaved in the vicinity of the Cauchy horizon and be- ed below, but bounded above. This means that the prod-
cause we do not expect the Cauchy horizon to collapse uct re cannot go to infinity on the Cauchy horizon and
too quickly due to the influence of the outflux of radia- hence cannot forbid the infinite increase of the mass pa-
tion. Indeed, the behavior of the Cauchy horizon [de- rameter.
scribed by the function rcH(U)=r (U, V=O)] is solely Some remarks are now in order: mathematically, the
ruled by the amount of outgoing radiation crossing it and mass inflation phenomenon is expressed as the divergence
focusing its generators. The infinite blueshift plays no of integral (4.8) as we approach the Cauchy horizon. We
1804 ERIC POISSON AND %ERNER ISRAEL 41

have shown that this integral must blow up, without ac- In conclusion, some comments on how the aspherici-
tually having to evaluate it explicitly, which we cannot ties inevitably present in a realistic collapse might be ex-
do exactly. We therefore know that mass actually pected to affect this idealized spherical picture. The
inflates, but we do not know how fast. We have given a broad aspects of mass inflation should be generic, since
physical interpretation of this result in the Introduction: they depend only on the qualitative features of infinite
mass inflation can be explained by the combined effect of blueshift at the Cauchy horizon and the separation of
the infinitely blueshifted influx which piles up on the Cauchy and apparent horizons under transverse irradia-
Cauchy horizon, and of the arbitrary outflux which pro- tion. Because angular momentum J is nearly conserved
duces the crucial separation of the Cauchy and inner hor- in the collapse (it would be conserved exactly for axial
izons. While the Cauchy horizon contracts moderately, symmetry), the Kerr parameter a = J/m should become
the inner horizon deflates much more rapidly. The mass negligibly small in comparison with m in the mass-
inflation phenomenon can be understood further from an inflated geometry near the Cauchy horizon. Thus the
application of the DTR relation (3.17) (Fig. 2). If the two asymptotic field is expected to be very nearly
concentric null shells cross through each other near the Schwarzschild.
inner horizon of sector 8, f~ will be very small, but since What about the effects of nonrotational (quadrupole)
the product fcfD is finite and nonvanishing, the small-
ness of f~ must be balanced by the largeness of
closer to the inner horizon the shell propagates, the
The f„. asymmetries? For a slightly aspherical, uncharged and
nonrotating collapse it has been shown that the "tail"
(corresponding to large advanced time u) of the spacelike
f
larger „will be. This can be interpreted as a violent in-
crease of the mass parameter at the intersecting point.
singularity formed at r =0 relaxes asymptotically
stationary, exactly Schwarzschild-like form. This argu-
to a

The DTR relation has been previously used by Blau' to ment is easily adapted to the situation where a Cauchy
elaborate Eardley's analysis' on the death of white holes horizon is present: a spacelike curve just outside the
by accretion and blueshift on the past horizon. However, Cauchy horizon, where cur vatures reach Planckian
the DTR relation does not offer any description of what values, now assumes the role of the curve r =0. Asym-
happens at the interaction point, in contrast with our metries of the collapsing star can infiuence the (infinitely
continuum analysis. In particular, the characteristic time long) tail of this curve only through a very narrow win-
scale over which mass inflation occurs is not given by the dow of the star's history, and they are exponentially red-
schematic DTR analysis. shifted. The geometry of the Cauchy horizon's tail is
A crude estimate of the mass parameter growth rate thus determined solely by the asymptotics of the external
can be obtained if we formally expand the integral of Eq. field as U~ ~. But this is known to tend to a Kerr-
(4.8) in powers of bilinear L;„L,
first-order term, we can then take for re
„,
If we kee. p only the
their back-
Newman form, characterized by an inner-horizon surface
gravity Kp which is a constant (independent of geographi-
ground values obtained from the static Reissner- cal location on the surface). Thus the exponential mass-
inflation factor e ' is uniform, and nonuniformities
UK
Nordstrom solution with mass m p (to this order
mp — m, ). So, from Eq. (3.5), take re =rp/Kp and b, m/m-b, 8 of the mass aspect m (H, qr, v) on angular
write, using Eq. (4. 11), scale 60 will not grow exponentially, but should remain
roughly constant during inflation. Near the Cauchy hor-
L;„(V) =ez (rp/~p) [ —ln( —V)]
(4. 15) izon these nonuniformities are negligible on length scales
p'2
comparable with the local (near-Planckian) radius of cur-
where e is an arbitrary dimensionless constant. Using
vature. On such length scales, the geometry should ap-
Eq. (4. 12) then yields pear locally indistinguishable from a Schwarzschild solu-
tion of very large mass.
m(U, V)=mpe 1(U)(apv) ~e ' (4. 16) Thus, there appears to be no reason to expect our
analysis and results to be significantly affected by the
'
where I ( U) is the fraction cpm p
'

star's mass radiated away. The dominant contribution,

f U1 „,
L, ( U')d U' of the effects of rotation and asymmetries. ' Of course, a de-
tailed analysis will be needed to confirm these tentative
conclusions.
as noted above, comes from the quadrupole moment and
so

m ( U, V) =mpe I ( U)([Link]) ' e ' (4. 17) V. QUANTUM CONSIDERATIONS

with e representing a dimensionless quadrupole moment. How can quantum mechanics modify the classical pic-
Our crude estimate indicates that the mass parameter ture described above? We shall attempt to find some
inflates exponentially with Eddington-Finkelstein ad- clues to an answer in this section. It is now widely be-
vanced time with a characteristic time scale of 1/Kp. lieved that singular behavior in general relativity signals a
This is most certainly an underestimate since, as we men- breakdown of the theory, and that it should not appear
tioned earlier, the product re does not actually remain once a good theory of quantum gravity is formulated.
constant but probably decreases to zero near the Cauchy Although there is hope from superstring theory, a quan-
horizon, hence precipitating further mass inflation. A tum theory has not yet been discovered and much of the
numerical integration of the field equations would be quantum effects associated with gravity are as yet uncer-
necessary in order to say more. ' tain. Nevertheless, it was speculated lately ' that be-
41 INTERNAL STRUCTURE OF BLACK HOLES 1805

cause of these unknown quantum effects, spacetime cur-

(5.3)
vature should always be subject to an upper bound of
Planckian magnitude. This corresponds to the most (the calculation is done in Planck units). This shows that
naive perception that one can have of a "nonsingular an observer's trip to the vicinity of the Cauchy horizon
singularity": a region of spacetirne where curvature is as covers all but the last Planck time of the black hole's en-
extreme as we can imagine. As discussed in Ref. 5, one tire classical history. This estimate is however modified if
possible mechanism for slowing down the infinite rise of vacuum polarization allows curvature to remain bounded
curvature is vacuum polarization associated with beyond the transition hypersurface X&. Let us assume
creation of virtual particles near the black-hole singulari- specifically that metric (5. 1) is valid at and near X& but
ty. If the stress thus induced is a tension along the axes that according to the analysis of Ref. 5, it would take the
of the three-cylinders of constant time, spacetirne tends form
to remain regular near the origin.
Let us imagine what could be the consequences of such ds = adr—
2+e '(df), +e «'~dt ) (5.4)
an upper bound in curvature for the black-hole inner
structure. We will come back later to the hard problem beyond X& (a is of order unity and is a measure of the
of justifying this assumption. Classically, the interaction curvature beyond X&, ~ is the proper time for an observer
of infalling and outgoing radiation inside the black hole at fixed t, 8, and P, and g is an undetermined function de-
produces an infinite rise in the hole's mass parameter. pending on the equation of state of the quantum materi-
Quantum mechanically, we might expect an enormous in- al). According to this new expression for the metric, the
crease, but not quite an infinite one. We recall that be- trip from X& to the origin would take an amount of time
cause of mass inflation and because the fluxes of radiation of the order of ro. The possibility that vacuum polariza-
conserve (at least approximately) charge and angular tion would increase this estimate even further is not ex-
momentum, the only relevant parameter describing the cluded.
black-hole geometry near the Cauchy horizon is precisely The picture emerging from these quantum considera-
the mass. We therefore infer that in this region, space- tions is that the region within X& is represented by a "fat
time can be adequately described by the Schwarzschild cigar" (topology S X R + ) of uncertain thickness en-
metric veloping the Cauchy horizon and inside which the evolu-
tion of spacetime is determined by unknown quantum
ds = —(1 2m/r—) 'dr +(1 —2m/r)dt +r dQ. , (5. 1) effects which would, presumably, forbid curvature to
where m is the inflated final mass parameter. How large grow beyond Planckian scales.
can it become? Spacetime curvature m lr has become, Let us now review some of the possibilities concerning
according to our hypothesis, of order unity in Planck the quantum evolution of spacetime geometry. Here, of
units. Since r is then of the order of the Cauchy horizon course, we enter the realm of very uncertain speculation.
One possibility would be to apply quantum cosmology in
mo (if we assume that e ~
I
-
radius, itself of the order of the black-hole external mass
m o ), we can conclude that its current minisuperspace formulation where a wave
function depending on the gravitational and external
m -mo/mp~, (5.2) fields degrees of freedom is constructed. As mentioned
earlier, our problem does not share quantum cosmology's
where mp& is the Plank mass. For a black hole of ten so- difficulty concerning the formulation of boundary condi-
lar masses, the internal mass parameter reaches the order tions. In view of metric (5. 1), the simplest minisuper-
of 10 times the mass of the observable Universe. space model for the black-hole interior would be the
If we suppose that we can accept the classical descrip- two-dimensional model
tion up to the transition hypersurface X& on which cur-
vature becomes Planckian as the mass parameter reaches
ds2= X(r)dr +—
a (r)dt +b (r)dQ (5.5)
its limiting value (5.2), we find that the transition region where a and b represent the gravitational degrees of free-
is spacelike since, classically, mass increases further to dom (X is not a dynamical variable). A quantum model
infinity on a null hypersurface occurring later. The sepa- of this kind, for pure gravity, has been considered by
ration between X& and the Cauchy horizon is of the or- Narnbu and Sasaki, ' using the Einstein-Hilbert action.
der of the Planck time, as we will verify below. So, for all Classically, the trajectories representing the evolution of
practical purposes, the Planckian phase, or end of the Schwarzschild spacetime must go toward the singularity.
classical phase, appears to arise at a spacelike surface, as Should the wave function be peaked around some classi-
would be the case for the Schwarzschild singularity. It is cally forbidden trajectory which escapes the singularity,
interesting to note that the quantum evolution of the one could interpret this as a quantum gravitational
black-hole internal structure starts from a known and avoidance of singular behavior. Unfortunately, Narnbu
well-defined initial state on X&. This is to be contrasted and Sasaki find that the wave function is exponentially
with the situation in quantum cosmology where initial damped for classically forbidden trajectories. This model
conditions have to be postulated. therefore offers little hope for quantum gravitational
Using metric (5. 1) but assuming that 2m/r ))1,it is justification of singularity avoidance. Further studies in-
easy to see that for an observer at fixed t, 8, and P, the cluding matter fields' do not modify this pessimistic con-
trip from the quantum radius ro= —m' -rcH to the ori- clusion.
gin takes about one Planck time: This negative result is an illustration of Wheeler s prin-
 ERIC POISSON AND %'ERNER ISRAEL

ciple of unanimity. ' Roughly, it states that if the solu- tie reason to doubt that our conclusions should be valid
tions to the classical equations of motion all exhibit for a generic, rotating black hole. ''
singular behavior (except perhaps for a set of measure According to this classical picture (if we follow it to its
zero) then the quantum solutions should also be singular. most extreme consequences) the singularity arising in a
According to this view, any model based on the Einstein- generic black hole would be null rather than being space-
Hilbert action would not be freed of singularities by a like as favored by strong cosmic censorship. However,
quantum analysis. To avoid singular behavior, one would in a more pragmatic viewpoint where we consider curva-
hence need to seek for alternative Lagrangians leading to tures of Planckian magnitude to be (classically) "singu-
nonsingular classical solutions. This is a rather natural "
lar, it makes little difference since the transition hyper-
step since it was realized long ago that the Einstein- surface on which curvature reaches unity in Planck units
Hilbert action should not be valid at extreme regimes. ' is a spacelike hypersurface. It is nevertheless an interest-
Quantization of matter fields on a classical spacetime ing question of principle to evaluate whether more realis-
geometry typically induces quadratic terms in curvature tic perturbing fields could succeed in producing an actual
in the effective gravitational Lagrangian and it is likely spacelike singularity. We will leave this question open
that higher-order terms would need to be included as for the time being.
well. A particular Lagrangian in ten dimensions is sin- The complicated field equations derived here do not al-
gled out by superstring theory. The Einstein- low us to find an explicit expression for the mass function
Hilbert+ Gauss-Bonnet action m(U, V). Equation (4. 17) represents probably a very
crude underestimate of its growth rate and to be able to
Jd' x& —g [R+a(R &r&R
~~ 4R &R—~+R )]
say more, a numerical integration of the field equations
should be performed. In the process, we would be able to
(5.6}
gather more information about the relevant quantities of
is the unique action principle leading to second-order the problem, specifically, the behavior of the metric ele-
differential equations for the ten-dimensional metric g & ment e and radius function r. Numerical integration of
(a-lp&, the square of the Planck length). It would be in- the equations is now in progress' and the results will be
teresting to attempt to find regular classical solutions to published elsewhere; for completeness, we derive in Ap-
the field equations associated with this choice of Lagrang- pendix C the basic first-order system of differential equa-
ian for a generalized Kantowski-Sachs metric of the form tions necessary to write the numerical code.

ACKNOWLEDGMENTS
ds = —A(r)dr +B(r)dt +r dQ +ap (r)g„zdx "dx
The mass inAation phenomenon, one of the most spec-
(5.7} tacular manifestations of the nonlinear aspects of general
where the nature of the six-dimensional line element relativity, first dawned on us when we derived the gen-
"dx is rather uncertain. The behavior of the eralized form of the Dray —'t Hooft —Redmount (DTR) re-
g~~dx
scale factor p(r) should reflect the compactification of ex- lation presented in Sec. III. Shortly afterward, Steven
tra dimensions at the classical regime. Such an investiga- Blau informed us of his very interesting and independent
work' in which the original DTR relation is applied to
tion is now underway. '
show that white holes are typically engulfed by black
holes soon after their formation. We are grateful to him
VI. CONCLUSION for those enlightening discussions. Don Page's interest in
our work was very stimulating and numerous discussions
It is likely that our views about the quantum evolution- with him helped the elaboration of this paper. Our
ary phase of spacetime inside black holes will change and warmest thanks to him. This research was supported by
that the many uncertainties surrounding it will not be the Canadian Institute for Advanced Research, Natural
resolved soon. But a definite classical picture appears to Sciences and Engineering Research Council of Canada,
be emerging up to the formation of the "fat cigar singu- and Fonds pour la Formation de Chercheurs et 1'Aide a
larity" near the Cauchy horizon and before the ambient la Recherche, Quebec, Canada.
curvature m/r actually reaches Planck values, there
does not seem to exist any mechanism capable of stop-
ping the inAation phenomenon. For instance, scattering APPENDIX A: GEOMETRIC QUANTITIES
of the infiow by the now inflated curvature never becomes FOR SPHERICAL SPACETIMES
effective. Indeed, the radius of curvature, behaving like
m ' —V~' always remains larger than the blueshifted We list in this appendix the relevant geometric quanti-
ties of a spherical spacetirne geometry whose metric is ex-
~

wavelength of the ingoing modes, behaving like V~. Our

~

detailed mathematical analysis is based on a very ideal- pressed in the form

ized spherical model, but the basic physical mechanism dS2 g d& ad& b+ &2d Q2 (A1)
(the combination of a highly focused and blueshifted
shower of radiation propagating along the Cauchy hor- where x'=(x', x ) are arbitrary coordinates spanning
izon and the arbitrary irradiation of the latter by outgo- the "radial" two-spaces (8, $)=const. Function r(x') is
ing radiation emitted from the surface of the collapsing defined in the usual way from the proper area of a two-
star) is independent of the model. We have therefore lit- sphere x'=const: A =4m-r . The symbol d0 denotes
41 INTERNAL STRUCTURE OF BLACK HOLES 1807

the two-dimensional line element of the unit two-sphere: we will show that, when expressed in regular coordinates,
d = d 8 + sin 8 d P . It is most convenient to express
0 the metric is manifestly regular there. Interestingly
the diverse four-dimensional quantities derived from the enough, even though certain elements of the curvature
four-dimensional metric g & in terms of the two- tensor diverge, the invariant R &z&R ~~ remains regular
dimensional quantities derived from g, b and derivatives at the Cauchy horizon. This is a consequence of the fact
of r. We will denote by a semicolon the covariant deriva- that the singularity in the energy density is null.
tive with respect to the two-metric, whereas a stroke (~) We look at the ingoing charged Vaidya solution
will denote the same with respect to the full four-
ds =dv(2dr fd—
u)+r dQ
dimensional metric. The d'Alembertian g of any scalar (81)
field P will be the two-dimensional quantity f =1 —2m(v)/r+e /r
where the mass function is given by
We will express the four-dimensional quantities with the m (v) = mo —p(u), (82)
superscript 4, whereas we will leave the two-dimensional
quantities free of additional indices. For example, the ab where p(v)-u 'r ".
The slow increase of the mass pa-
component of the four-dimensional Ricci tensor will be rameter is produced by an influx of lightlike particles de-
noted as R,b whereas the two-dimensional Ricci tensor scribed by the stress-energy tensor
will be written as R,&. We use throughout the conven- T p=pl lp, (83)
tions of Misner, Thorne, and Wheeler.
With the notation described above, the Christoffel sym- where 1 = —8
u and 4mr p=dm /dv. We recall that be-
bols associated with the four-dimensional metric are cause of the large blueshifts occurring near the Cauchy

bc bc
'88 =sin gI' = —rr' horizon (which we shall verify below) the effective stress-
energy tensor description (83) should be an accurate
(A3)
I'ee=I &&=r, /r, I'&&= —sin& cos8, 1 ~e&=cot8 . description for the backscattered gravitational radiation
falling into the black hole at late advanced times. We
The Riemann curvature tensor is will first show that the energy density measured by a
4
R abed —
—2
Ra @pe sin ORapgp
free-falling observer with four-velocity u,
Rabcd ~ r;a$
(A4) p, b, —=T u =p(1 u (84)
R e&e& = r sin 8( 1 r'r, ) . — pu )

becomes infinite at the Cauchy horizon v = ~.

It is only
Contracting over the first and third indices yields the necessary to compute the v component of the four-
Ricci tensor velocity; if we choose a radial observer, the normalization
u u =—
R,b —R,i, —2r , /r,
b . condition 1 yields (the overdot denotes
(A5) differentiation with respect to the observer's proper time)
Ree =sin =1 (rCIr +r'r—
, ),
OR&&
u =[r —(r +f)' ']If =2r'/f,
and the Ricci scalar is
where the approximation holds near the Cauchy horizon.
R =R +2(1 —2r r —r'r, )/r (A6) Therefore
We finally arrive at the Einstein tensor (u&1 )i=o 2 1/f2 (86)

G, i, = —[2rr ,b+g, b(1 r'r. , —2rHr—)]/r which is infinite at the Cauchy horizon. It is possible to

Gzz=sin OG&&=rQr —'r R . —,

(A7) integrate (85) explicitly if we approximate
asymptotic expression fo= 1 2mv/r—
by its
+e /r . This is in
f
These equations can be compared with the usual re- fact a good approximation since m (v) is slowly varying
sults obtained when one chooses a particular system of near the Cauchy horizon. We find that on the path of the
coordinates for x '.
For example, if one sets free-falling observer
g, bdx'dx = —2h du dv, one recovers the results derived f = —2e (87)
in Waugh and Lake. Similarly, by choosing
g, bdx 'dx = — e dt + e dr one recovers the . results where i~o=(mo e)' /rcH s— uch that combining Eqs.
given in Misner, Thorne, and Wheeler, page 844 of Ref. (82), (84), and (87), we arrive at
23.
p -v ~e
2KO U

APPENDIX B: VAIDYA SOLUTION which manifestly blows up at the Cauchy horizon. Note
IN DOUBLE-NULL COORDINATES that not only p, b, blows up, but also its integrated value
over the path of the observer.
A free-falling observer crossing the Cauchy horizon It is not at first sight obvious that metric (81) is regular
measures an infinite amount of energy density for the at the Cauchy horizon since the coordinates themselves
shower of infalling radiation propagating along it. We become singular there: both r and v are constant on the
will verify this statement explicitly in this appendix, and Cauchy horizon. Even a rescaling of v of the form
1808 ERIC POISSON AND WERNER ISRAEL 41

V= —e '
does not remove this difficulty. But because
m (v) is a slowly varying function, it is possible to ap-
proximately transform metric (Bl} into a double-null
form which is manifestly regular at the Cauchy horizon. (816)
(Such a transformation is not possible in general. ) We
now proceed with this transformation. We suppose that where we recognize in front of the large parentheses the
the influx of radiation is switched on at advanced time v, static expression associated with the initial mass mo. It is
and that rn (u, ) =mo, not only do we suppose that mo is manifest that the correction terms are bounded at v ~~
the asymptotic limit of the mass function, we also sup- provided that f
„"p(u')dv'&
1
~.
This property is clearly
pose that it is its initial value. This choice has the impor- satisfied by our choice for p(u).
tant advantage of allowing us to define the various quan-

:
It is now easy to transform to coordinates U and V in
tities of interest (e.g. , coordinate u, function ~, below) which the metric is well behaved. This has the effect of
with respect to ma rather than another value of the mass changing the coefficient , 'f,
in —Eq. (816) to the expres-
parameter. The difference mo — m(u)—p(v) will there- sion (3.5). This finally shows that metric (Bl), when ex-
fore be assumed to be zero for v v, , and will be assumed( pressed in well-behaved coordinates, is perfectly regular
to be non-negative thereafter until it decreases asymptoti- at and near the Cauchy horizon. Note however that
cally to zero as v goes to infinity. Only this asymptotic second derivatives in V will produce a divergent curva-
behavior will be of interest to us and we will not be con- ture of the order of p, b, . The curvature invariant
cerned with the initial decrease of the mass function. Al-
though we are interested in the situation where the g &
sg~~rs=4gm (u)/r (817)
asymptotic behavior of p(u) is that of an inverse power
however, remains regular. The reason lies in the fact that
law, the following calculation only assumes that }u(u)/mo
the singularity is null: any contraction of the kind l 1
is small enough so that we can ignore second-order
will give a contribution of zero.
corrections.
We start by expanding f
[m(u), r] around the initial
value of the mass mo and around function r, (u, v), the
solution of APPENDIX C: FIRST-ORDER EQUATIONS
FOR NUMERICAL INTEGRATION
dr, = ,' f, (du + du—), (89)
For completeness, we include here the system of first-
where f, =f [mo, r, ] (the subscript stands for "static"). order differential equations which would enable a numeri-
We thus obtain cal analyst to integrate Einstein's equations numerically.
— All we will do is recast the field equations (4. 1)—(4.4) in a
f =f, +2@/r, 2~, (r r, ), — (810)
more convenient form. We shall choose for our null
where coordinates the usual Schwarzschild retarded and ad-
vanced times.
2 2)1/2/r2 (811)
( Define first additional variables x and y according to
is a function of r,
dinates by
Funct. ion r(u, u) is related to the coor- e "= d„r/ , f,'
e ——
r=r}„r/,'f . - (C 1)

Substituting this in Eq. (4.3) yields

dr = '
f (e~du +du)—, (812)
I p
2J
—(x+y)
(C2)
where e~ plays the role of an integrating factor. We
therefore find that the metric element which we want to and then into Eqs. (4.4) gives
evaluate is

g„,=h = ,'fest=Br/B—u . (813)

B„m= e "dm, „, /du,
(u )
(C3)
B, m =elm;„(u)/du .
To find r (u, u), we use Eqs. (89), (810), and (812}to ob- It is then straightforward to derive additional differential
tain a differential equation for the difference r — r, : equations for x and y starting from Eqs. (Cl). We find
B(r —r, )/Bu +tc, (r —
r, ) =p lr, , (814)
B„y= —(2e "lrf)dm, (u)ldu, „, (C4)
which can be integrated to yield B,x = —(2e /rf}dm;„(v)ldu .

r r, = f, —du'p(u') lr,'f,',
f (815) Since x is determined up to an arbitrary rescaling of coor-
1
dinate u, it is not necessary to write down an expression
)
where we suppose that u v, . For u &u„p(v)=0, and for B„[Link] same is true for B,y.
r(u, v) reduces to r, (u, u). [We use the notation In the initial static Reissner-Nordstrom region (u & u, ,
'=
r, r, (u, v'), etc. ] I—
t is now straightforward to show that u &u&), we have x =y =0, m(u, u)=m&, and r(u, u) is
differentiation of Eq. (815) with respect to u and use of given implicitly by u + v =2r*. In the pure inffow region
Eq. (813) yields )
(u & u, , v u, ), we have B,x given by Eq. (C4), y =0,
41 INTERNAL STRUCTURE OF BLACK HOLES 1809

m (u, v) = m;„(v),and r},r = l2. An expression for is

f B„r t)„r= l2. As before, an equation for t)„ris not needed.
f
not needed since by continuity, r(u „U)is known. Final- All the above quantities can then be obtained in the
the outflow region (u )u, , u & v, ) we have cross-Aow region by numerical integration of Eqs.
ly, in pure
x =0, B„y given by Eq. (C4), m (u, u) =m (u) and „, (C 1)—(C4) with the boundary conditions given above.

'R. M. Wald, General Relativity {University of Chicago Press, ' I. H. Redmount, Prog. Theor. Phys. 73, 1401 (1985); T. Dray
Chicago, 1984), pp. 322—324. and G. 't Hooft, Commun. Math. Phys. 99, 613 (1985).
2R. Penrose, in Theoretical Principles in Astrophysics and Gen- S. K. Blau, Phys. Rev. D 39, 2901 (1989).
eral Relativity, edited by N. R. Lebovitz, W. H. Reid, and P. ' D. M. Eardley, Phys. Rev. Lett. 33, 442 (1974).
O. Vandervoort (University of Chicago Press, Chicago, 1978), '5D. Goldwirth and T. Piran (in preparation).
p. 217; W. Israel, Can. J. Phys. 64, 120 (1986). ' 'Added in proof: This has recently been shown explicitly for
3R. H. Price, Phys. Rev. D 5, 2419 (1972); 5, 2439 (1972). arbitrary rotating black holes, with the infalling and outgoing
4A. G. Doroshkevich and I. D. Novikov, Zh. Eksp. Teor. Fiz. radiation modeled schematically by thin shells: C. Barrabes,
—
74, 3 (1978) [Sov. Phys. JETP 47, 1 (1978)]. W. Israel, and E. Poisson, University of Alberta Report No.
sE. Poisson and W. Israel, Class. Quantum Grav. 5, L201 (1988). CIAR UA90-1, 1990 (unpublished).
J. M. McNamara, Proc. R. Soc. London A358, 449 (1978); ' Y. Nambu and M. Sasaki, Prog. Theor. Phys. 79, 96 (1988).
A364, 121 (1978); Y. Gursel, V. D. Sandberg, I. D. Novikov, '7V. A. Belinskii and I. M. Khalatnikov, Zh. Eksp. Teor. Fiz.
and A. A. Starobinsky, Phys. Rev. D 19, 413 (1979); R. A. 63, 1121 (1972) [Sov. Phys. JETP 36, 591 (1973)]; R.
Matzner, N. Zamorano, and V. D. Sandberg, ibid. 19, 2821 Laflamme and E. P. Shellard, Phys. Rev. D 35, 2315 (1987).
{1979);S. Chandrasekhar and J. B. Hartle, Proc. R. Soc. Lon- ' J. A. Wheeler, Gen. Relativ. Gravit. 8, 713 (1977).
don A284, 301 (1982); N. Zamorano, Phys. Rev. D 26, 2564 '9See, e.g. , N. D. Birrell and P. C. W. Davies, Quantum Fields
(1982). in Curved Space (Cambridge University Press, Cambridge,
~E. Poisson and W. Israel, Phys. Rev. Lett. 63, 1663 (1989); England, 1982), Chap. 6.
Phys. Lett. B 233, 74 (1989). J. Madore, Phys. Lett. 111A, 283 (1985); B. Zumino, Phys.
R. A. Isaacson, Phys. Rev. 160, 1263 (1968); 160, 1272 (1968). Rep. 109, 109 (1986); H. Ishihara, Phys. Lett. B 179, 217
M. A. Markov, Pis'ma Zh. Eksp. Teor. Fiz. 36, 214 (1982) (1986); K. Maeda, ibid. 186, 151 (1987); N. Deruelle and L.
[JETP Lett. 36, 265 (1982)]; M. R. Bernstein, Bull. Am. Phys. Farina-Busto, CERN Report No. TH. 5303/89, 1989 (unpub-
Soc. 16, 1016 (1984); V. Frolov, M. A. Markov, and V. F. lished).
Mukhanov, Phys. Lett. B 216, 272 (1989); Phys. Rev. D 41, 2'E. Poisson and W. Israel (in preparation).
383 (1990). R. Penrose, in General Relativity: An Einstein Centenary Sur-
' V. A. Berezin, V. A. Kuzmin, and I. I. Tkachev, Phys. Rev. D
vey, edited by S. W. Hawking and W. Israel (Cambridge Uni-
36, 2919 (1987), Appendix A. versity Press, Cambridge, England, 1979).
"W. B. Bonnor and P. C. Vaidya, Gen. Relativ. Gravit. 1, 127 2 C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation

(1970); B. T. Sullivan and W. Israel, Phys. Lett. 79A, 371 (Freeman, San Francisco, 1973).
(1980). B. Waugh and K. Lake, Phys. Rev. D 34, 2978 (1986).