PHYSICAL REVIEW D VOLUME 13, NUMB ER 8 15 APRIL 1976

Path-integral derivation of black-hole radiance*

J. B. Hartle
Department of Physics, University of California, Santa Barbara, California 93106
and California Institute of Technology, Pasadena, California 91125

S. O'. Hawkingt
Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, England
and California Institute of Technology, Pasadena, California 91125
(Received 17 November 1975)
The Feynman path-integral method is applied to the quantum mechanics of a scalar particle moving in the
background geometry of a Schwarzschild black hole. The amplitude for the black hole to emit a scalar particle
in a particular mode is expressed as a sum over paths connecting the future singularity and infinity. By
analytic continuation in the complexified Schwarzschild space this amplitude is related to that for a particle to
propagate from the past singularity to infinity and hence by time reversal to the amplitude for the black hole
to absorb a particle in the same mode. The form of the connection between the emission and absorption
probabilities shows that a Schwarzschild black hole will emit scalar particles with a thermal spectrum
characterized by a temperature which is related to its mass, M, by T = k c'/8m GMk. Thereby a conceptually
simple derivation of black-hole radiance is obtained. The extension of this result to other spin fields and other
black-hole geometries is discussed.

I. INTRODUCTION AND SYNOPSIS Figure 1 shows the Penrose diagram for the
Schwarzschild geometry. The unshaded part of this
The Feynman path-integral method' is a natural diagram represents the geometry outside a spheri-
way to formulate the quantum mechanics of matter cally symmetric collapsing body. The shaded part
fields moving in curved background spacetimes. '' should be replaced by the geometry inside. Let
In this method the amplitude K(x, x') for a particle us now consider the probability that a particle is
to propagate from one space time point x' to anoth- emitted by the black hole and detected by an ob-
er x is expressed as an integral over all the paths server a constant distance away in a positive-fre-
connecting the two points. The integral has the quency mode peaked about some point A when
form there are no incoming particles in the distant past.
(xpx' )/0
This probability can be related to the amplitude to
~(x xI) Q e fs
propagate from some point B on the future singu-
paths
larity to the observation point A. This in turn can
where S(x, x') is the classical action for a particu- be represented as a sum over paths of the form in
lar path connecting x' and x. The amplitude K is Eq. (1.1), where the paths summed over are those
called the propagator. which begin at the point B on the future singularity
This formulation of quantum mechanics has sev- and end at the observation point A. A typical such
eral advantages. Because the sum is over paths path (BCA) is shown in Fig. 1. These are exactly
in the four-dimensional space time and because S the paths which correspond to a pair of particles
is a four-dimensional scalar the expression for K being created (near C), one of which falls into the
is manifestly covariant. Expressing the propaga- black hole and the other of which propagates out
tor as a sum over paths gives it a direct physical to the observer. We do not sum over paths which
interpretation. Since the propagator is expressed start on 8 since they would represent the propaga-
directly as a functional integral a problem of find- tion of incoming particles in the distant past. We
ing an approximate form for it reduces immediate- do not sum over paths which pass through the
ly to a problem of approximating the functional in- shaded region since that should properly be re-
tegral. For these reasons the Feynman path-inte- placed by the interior of the collapsing star and
gral method is an attractive way to do quantum will not contribute to the particle production at late
mechanics in curved space times. In this paper times (as we shall show in more detail subsequent-
we shall use the path-integral method to derive the ly). We consider only propagation from the future
thermal radiation emitted by black holes. ' In the singular ity.
following we shall give a qualitative outline of our If one attempts to evaluate the production proba-
methods and results. The details and proofs will bility by applying the method of stationary phase
be presented in the subsequent sections. to the integrals involved, then it is readily seen

13 2188
 PATH-INTEGRAL DERIVATION OF BLACK- HOLE RADIANCE 2189

that there are no real stationary paths which con- plitude to propagate from a corresponding point D
nect the future singularity to a positive-frequency on the past singularity. In turn, by symmetry un-
mode for a stationary exterior observer. Such der time reversal, the modulus of the latter am-
paths would be permissible classical paths with the plitude is the same as that of the amplitude to pro-
positive energy connecting the two surfaces and pagate from the time-reversed point of A into the
there are none (although these are classical paths future singularity at the time-reversed point of D.
with negative energy). However, if the coordinates This shows that by appropriately distorting the
of the point B on the future singularity are dis- contours of integration into the complex coordinate
placed to complex values then stationary-phase plane the amplitude for a black hole to emit parti-
paths can be found. These paths are, in fact, in cles can be related to the amplitude for it to ab-
the real manifold and connect the past singularity sorb. If we consider the amplitude for the black
to the external observer. Thus, the amplitude to hole to emit particles in a definite mode with ener-
propagate to the external observer from a point B gy E as measured by a distant observer then this
on the future singularity can be related to the am- connection is

(probability to emit a particle with energy E) =e "s~" && (probability to absorb a particle with energy E),
(1.2)

where ~ is the surface gravity of the black hole; in

the case of a Schwarzschild black hole of mass M,
v =1/4M. (Here as in the following we are using
units where h =c = G =1.) This connection between
emission and absorption is exactly that necessary
to establish the result of Ref. 5 that a Schwarzs-
child black hole emits particles with a thermal
spectrum corresponding to a temperature T
= g/2'. To see this, imagine surrounding the
black hole by a thermal cavity and adjusting the
temperature, T, until the whole system is in equi-
librium. The temperature of the radiation is then
the temperature of the black hole. In equilibrium
the rate of emission particles by the black hole
FIG. 1. The unshaded part of the diagram represents must exactly equal the rate of absorption. Since
the Schwarzschild geometry outside a spherically sym- the ratio of the probability of having N photons in
metric collapse. The world line 0 is that of an observer, a particular mode in the cavity with energy F. to
who remains outside the black hole at a fixed radius. The the probability of having N- 1 photons in the same
nonstationary path BGA corresponds to particle produc- mode is exp(-E/kT) this equilibrium condition will
tion by the black hole. A pair of particles is created near
C. One falls into the future singularity at B while the be true when T=~/2mk.
other propagates out to the observer and is detected In the following we shall give the details of this
at A. The amplitude that a particle is produced by the simple argument. In Sec. II the quantum mechan-
black hole and detected by the observer in a given mode ics of scalar particles moving in a Schwarzschild
at late times can be expressed as an integral over the background geometry is formulated in terms of
amplitude to propagate between a point B on the future path integrals. Section III contains a derivation of
singularity and a point A on the observer's world line. the necessary analytic properties of the propaga-
In turn the propagator can be expressed as a sum over
the paths which connect these points. By analytically
tor on the complexified manifold, and in Sec. IV
continuing the point B into the complexified Schwarzs- the thermal radiation from a Schwarzschild black
child space, the amplitude to propagate to A from a real hole is deduced. Section V discusses the generali-
point B on the future singularity can be related to the zation to Reissner-Nordstrom and Kerr black
amplitude to propagate to A from a reflected point D on holes and higher-spin particles.
the past singularity. This latter process is just the
time-reversed proces of absorption of a particle by the II. PATH-INTEGRAL QUANTUM MECHANICS OF A SCALAR
black hole. In this way the probability for a black hole PARTICLE IN A SCHWARZSCHILD BACKGROUND
to emit a scalar particle can be related to the probability
for it to absorb one. The relation implies the thermal In this section we shall formulate the quantum
radiance. mechanics of a scalar particle moving in a curved
2190 J. B. HARTLE AND S. W. HAWKING

background geometry in terms of Feynman path separated. Amplitudes for more restricted sets
integrals. Our work here is a generalization and of observations may be constructed by summing
interpretation of that of Feynman for a scalar field this amplitude over the unobserved positions. For
'
in a flat-space background, and as far as the gen- example, the amplitude that an observation of
eral formulation of path integrals in curved back- spacetime position at one parameter time yields
grounds goes closely parallels the previous work the value x' and a second observation a parameter
'
of the DeWitts. ' The considerations of this sec- time 8' later yields the value x is
tion are intended to motivate the definition of the
propagator in Sec. III as a solution of the inhomo-
geneous scalar wave equation with certain bound-
ary conditions. As a consequence in the present
X(W, x, x') =
f )pxXf—
ex[w e ', d{x, )dxw
Q

section we shall not be uniformly mathematically (2.2)

precise, but this is a familiar situation when work-
where the integral is over all the unobserved posi-
ing with path integrals.
tions at parameter times between 0 and W. In
The path of a scalar particle through spacetime
other words, the integral is a functional integral
may be specified by giving the four coordinates
over all paths which have x(0) = x' and x(W) = x.
x as a function of a parameter time w. Letting
The parameter w has been introduced as an ob-
x stand for all four coordinates we write this as
servable which plays a role analogous to ordinary
x =x(2{)). Suppose we consider the motion of a par-
time in nonrelativistic quantum mechanics. How- '
ticle which starts at a spacetime point x' at w =0
and arrives at x at w= W. An action functional
ever, there is no experiment in which it is direct-
which describes the classical motion of such a ly observed (since particles do not carry clocks).
All physical observations can be obtained from the
particle is
amplitude K(x, x') for a particle to be localized at
two spacetime points x' and x. K(x, x') is called
(2.1) the propagator. This amplitude can be constructed
in two steps: first by summing over all paths
where g is the metric on the curved spacetime and
which connect x' to x in a given parameter time
x represents the tangent vector whose components
W and then by summing over the unobserved val-
are dx /du). The path which extremizes S satis-
fies the geodesic equation with w as an affine pa- ue, W. The first sum is just F in Eq. (2.2). In the
second sum an appropriate weight must be as-
rameter. Thus, for timelike paths w may be taken
signed to each elapsed parameter time W. In flat
to be a constant multiple of the proper time while
for spacelike paths it could be taken to be the same space if the scalar particle has a rest mass m this
weight is' i exp(-im'W). It is natural to adopt this
multiple of the proper distance.
also for the curved-space case. The expression
The action functional of Eq. (2.1) is not the usual ,

for the propagator then takes the form'

one in which the integrand is [-g(x, x)]'~'. How-
ever, (2.1) is a perfectly valid classical action
which has obvious advantages for a path-integral
formulation in that it is quadratic in the four-ve-
X( , x)x=i
f
0
dWex(2( (I*W)F(W, x, x'), (2-.3)

locities (see the Appendix). In contrast to the where E is given by Eq. (2.2). The restriction of
usual form, the action in Eq. (2.1) continues ana- the integral to positive 8' is the requirement that
lytically from timelike to spacelike paths. In add- the particles always propagate forward in parame-
ition it gives correctly the relativistic quantum ter time.
mechanics of a scalar particle in flat spacetime From this definition it easily follows that K(x, x')
and is therefore a natural generalization to curved is symmetric in x and x'. Letting w = 8'- w' leaves
backgrounds. We shall not discuss other choices the action in Eq. (2.1)=unchanged but interchanges
of the action further here. x and x' in the sense that as functions of ge', x(0)
The basic assumption of the Feynman path-in- =x and x(W) =x'. Thus E(W, x, x') is symmetric in
tegral method is that the amplitude for a particle x and x' and it follows immediately from Eq. (2.3)
to travel a particular path in spacetime is propor- that K(x, x') is also.
tional to exp(iS[x(2())]). To have a clearer idea of Equations (2.2) and (2.3) are the basic relations
what this means imagine dividing the parameter which are needed to define the quantum mechanics
time w into many small intervals at values w, . of a free scalar particle moving in a curved space-
The amplitude for observations of spacetime posi- time. Before this definition is complete the inte-
tion at each parameter time w, to yield the set of grals in Eqs. (2.2) and (2.3) need to be given mean-
values (x, =x(co, )) is proportional to exp(iS[x(N))]) ing. We now turn to this question but for simplicity
in the limit as the intervals become infinitesimally and definiteness restrict our attention to scalar
13 PATH-INTEGRAL DERIVATION OF BLACK- HOLE RADIANCE 2191

particles propagating in the Schwarzschild geome- the class of paths defined above look like when the
try. analytic continuation reaches real values of 8' and
The first problem is the definition of the path in- the contours of the path integration are deformed
tegral in E(l. (2.2). To solve this we analytically to real coordinate values? It seems clear that the
continue the variables in this formal expression to resulting class of paths will not be confined by any
values where the integral is well defined. In par- finite boundaries in the Kruskal coordinates. In
ticular zg and 5' are continued to negative imagi- particular, they will cross and recross the singu-
nary values -iw and -iA, respectively, and the larities at x=0. These singularities are poles in
coordinates are continued to a domain where the the metric considered as functions of the complex
metric has signature +4. In the case of the coordinates. The path integral across the singu-
Schwarzschild geometry which in Kruskal coordi- larity is defined by giving a prescription for which
nates z, y, 8, y has the form way the contour of integration goes around the
ds2 = (32M'e "/'"/x)( dz'-+ dy2) + x'dQ', pole. In turn this is determined by the analytic
(2.4)
continuation of the path from the positive-definite
with dQ' =d8'+ sin'8d(/]' and r(y, z) defined by section and the deformation of the contours to real
z2+y2 —(y/2M values everywhere except near x=0. If the paths
l} re/2M (2. 5)
cross r =0 then they extend into the Schwarzschild
this can be accomplished by letting z =i/ and keep- geometries with negative mass. This is illus-
ing f real. Then the analytically continued metric trated schematically in Fig. 2(b).
y is given by the line element The actual computation of F(Q, x, x') is greatly
da' = (32M'e ' '"/r)(dr'+dy')+r'dQ', (2 6)
facilitated by noticing that it satisfies a parabolic
partial differential equation
with x now defined by 8F
P+y& —(~/2M l )er/2// (2 9)
(2.7)
The analytically continued expression for F be- where 0'=y ~V Vz and V indicates covariant dif-
comes ferentiation with respect to the metric y. The de-
0 rivation of this result is reviewed in the Appendix.
P(II, x, x')= f '
Ilx[te]exp ——,
0
X(x, x]dte
(2.8)
The boundary conditions on E(l. (2. 9) which yield

where x is to be understood as x =(f, y, 8, (/]) and

x as dx /d(d The sp.ace covered by the coordinate
ranges -«f«, -«y «, 0& e~ g, 0& y&2g
is complete, has the topology A'&&S', and, since
r ranges only over values greater than 2M, the
metric y is regular with the exception of the trivial
polar singularities at 8 =0 and n. The integration
now extends over all paths in this space which
start with (0= 0 at x' and end with &o = Q at x (see
Fig. 2). This path integral can be precisely de-
"
fined. Our basic assumption" is that the function
F when defined in this way and analytically con- (b)
tinued back to real values of the coordinates and
parameter time variable gives the correct propa- FIG. 2. (a) A compactified representation of a con-
stant 8, constant fIt} slice of the positive-definite space-
gator defined heuristically by E(l. (2.2). time whose metric is given in Eq. (2.6). The heavy
This procedure not only gives definition to the circle represents infinity. There are no singularities.
integral in E(l. (2. 2) but also identifies the class A typical path connecting two points x' and x is shown.
of paths over which integration is done. Imagine (b) A Penrose diagram for the Schwarzschild geometry
taking a particular path in the space with positive- showing in addition the regions of negative mass (or
definite metric and continuing both coordinates r ~0) above and below the singularities. A typical mem-
and parameter to complex values. The complexi- ber of the class of paths continued analytically to this
real section from the positive-definite spacetime repre-
fied path is now a two-dimensional sheet in the sented in (a) is shown. Such paths may cross and re-
space of complex coordinates given by the four cross the singularities at r =0. Integrations over paths
complex analytic functions x(((]) of the complex pa- which cross the singularities are specified by choosing
rameter ~. The analytic functions are completely contours of integration which are the analytic continu-
fixed by their real values for real ~. What does ations of those in the positive-definite section.
2192 J. B. HARTLE AND S. W. HAWKING

F as defined by the path integral are first that In general there will be several geodesics con-
necting x and x'. In that case the small-8' be-
F(O, x, x') = 5(x, x'), (2.1o) havior of F will be
where 5(x, x') is the four-dimensional 5 function in
the space with positive-definite metric [equal to F(W i) —((, (s, (x, x )/(~w)
6'~'(x —x')y ' ']. Second, F must vanish as x ap-
proaches the infinity of the space with positive- && W '[D, (x, x')+O(W ')],
definite metric. The first boundary condition is
simply the requirement that at 0 =0 the particle
(2.15)
be localized at x'. The second follows from the where the sum is over each class of geodesics
exponential damping of the integral as x tends to which connect x to x'. This small-W behavior will
infinity. not be uniformly valid over the whole range of
The solution to Eq. (2.8) and associated boundary values of x. In particular where neighboring geo-
desics which start at x' intersect (caustics) we ex-
and analytic in 0 except at the origin. "
conditions will be analytic in x where the metric is
The solu- pect the approximation to break down. For exam-
tion can thus be continued back to physical values ple, s(x, x') will have a branch point at such an
of the coordinates and parameter time. There it intersection, but we know from general considera-
will satisfy the analytically continued equation tions that F has none.
If Eq. (2.15) is integrated over a smooth function
QF of x then as TV tends to zero there will be a signi-
i = — F, (2.11)
ficant contribution to the integral only for values
where Cl'=g ~V V~ and V is covariant differentia- of x for which s, (x, x') is nearly stationary keeping
tion with respect to the metric Z. Equation (2.11) x' fixed. This will happen only for x close to x'.
can be thought of as the Schrodinger equation for In other words, F(o, x, x') is proportional to a 6
propaga, tion in the parameter time W. function. The proportionality factor is just unity
The second problem involved in giving meaning because the normalization factors in Eq. (2.15)
to Eq. (2.3) for the propagator is the integral over must also give rise to Eq. (2.10). Thus,
To resolve this we need the asymptotic form IimF(W, x, x') = 5(x, x') . (2.16)
of F(W, x, x') for both small and large W. The be- lV 0

havior of F for small W is essentially given by the For large values of 0 standard estimates" for
definition of the path integral itself. For small W the solution of the parabolic equation (2.9) defined
and fixed x and x' the action [Eq. (2.1)] for a typi- by the boundary condition in Eq. (2.10) show that
cal path will become large. The only paths which F(Q, x, x') decreases at least as fast as 0 ' for x
contribute significantly to the path integral are the and x'. Physically this is nothing more than the
stationary paths, i.e. , the geodesics connecting x spreading of an initially localized wave packet with
and x'. For these paths the action is increasing Q. Thus, at large 0, F can be ex-
pressed as
S(W, x, x') =-,'s(x, x')/W, (2.12)
where s(x, x') is the square of the geodesic distance
F(A, x, x') =0 '[F,(x, x')+F, (x, x')0 '+ ].
between x and x'. If there were a single geodesic (2.17)
connecting x and x' then for small W we could We will assume that expansion can be continued
write back to real values of 8' and x by term. Thus, in
F(W, x, x') =exp[is(x, x')/(4W)]N (W, x, x'), particular we have for large W

where N(W, x, x') is a real normalization factor. F(W, x, x') =O(W-') . (2.18)
Equation (2.11) can then be solved for small-W be- The large-S' behavior of F shows that the inte-
havior of N. When note is taken of the identity gral in Eq. (2.3) always converges at the upper
(V s)(V s) =4s, (2.13) limit. This is nqt the case at the lower limit where
F diverges as W ' [Eq. (2.15)]. To make the inte-
one finds that gral finite we shall insert a convergence factor
N(W, x, x') =D(x, x') W-'+ ~
(2.14) exp(-e/W), where e is a small positive constant.
Physical quantities are to be computed with & finite
where D is independent of R', its exact form will and then the limit & -0 is to be taken. Thus,
not concern us. The reader will recognize this as
essentially the WEB approximation to the solution
of Eq. (2.11). Considerably more detail on the de-
sc(x, e) =I
f awaxp(-(I'w-
0
~iw)s'(w, x, x') .

rivation can be found in Refs. 2 and 8(d). (2.19)

 PATH-INTEGRAL DERIVATION OF BLACK- HOLE RADIANCE 2193

This method of regularization corresponds physi- the inhomogeneous wave equation which is regular
cally to the requirement that the particle propagate at infinity and for which the singularities corre-
forward in the parameter time 8'. It correctly sponding to propagation along future-directed null
gives the usual Feynman propagator for a scalar geodesics lie below the real t axis, while those
particle in flat spacetime as we shall show in the corresponding to propagation along the past-di-
next section. With Eq. (2.19) our definition of the rected null geodesics lie above the real t axis.
propagator is now essentially precise. We shall Elsewhere in the complex f plane K(x, x') is ana-
now examine the consequences. lytic. We shall now consider the analogous bound-
ary conditions for Eq. (3.1) in the Schwarzschild
III. ANALYTICITY PROPERTIES OF THE PROPAGATOR geometry.
To begin with let us consider the case in which
In the preceding section an integral representa- x' is exterior to the black hole and x lies on the
tion for the Feynman propagator K(x, x') was de- horizon. It is convenient to use null Kruskal co-
rived in terms of the propagator for a definite pa- ordinates U and V in which the Schwarzschild
rameter time W, F(w, x, x'). It follows immedi- metric takes the form
ately from Eq. (2. 9), the parameter time Schro- '
ds' = —(32~'e " /r)d Ud V+ r'dQ', (3.4)
dinger equation, Eq. (2.11), and the boundary can-
ditions, Eqs. (2.15) and (2.17), that K(x, x') is a with z defined by
solution of the inhomogeneous wave equation in the ~'"
UV = (1 —r/2M)e" (3.5)
Schwarz schild background
'-m')K(x, x') =- 5(x, x'). On the horizon x=2M and either U=O or V=O. We
( (3.1)
analytically continue the nonzero member of the
As an alternative to the path integral, K(x, x') could pair (U, V) to complex values and refer to the sur-
be defined as a solution to Eq. (3.1) with suitable face thus obtained as the complexified horizon.
boundary conditions. This approach is a useful one Since the metric is analytic in the Kruskal coordi-
because some properties of K(x, x') can be deduced nates on the complexified horizon, the function
directly from the differential equation and because E(w, x, x') will also be analytic there. Any singu-
it is more easily generalizable to the propagation larities in K(x, x') will therefore come from the
of particles with higher spin. In this section we end points of the integration over O'. From the
shall derive the boundary conditions for K(x, x') asymptotic expression [Eq. (2.1'1)] for E(w, x, x')
from its path-integral definition. at large values of 8' one easily sees that the inte-
First, we illustrate the procedure with the ex- gral converges for large 8' for all complex values
ample of a massless scalar particle in flat space. of x. Any singularities of K(x, x') must therefore
The solution of the parameter time Schrodinger come from the W=O end point. To analyze these,
equation for F which satisfies the boundary condi- divide the interval [0, ~] in W into two pieces
tion of Eq. (2.10) is [0, Wp] and [Wp, ~], where Wp is small. The inte-
F(W, x, x') =i(4~W) 'e"'" "'"" ' (3.2) gral from 8', to infinity gives a contribution
K, (x, x') to K(x, x') which is analytic in x. In the
where s(x, x') denotes the square of the Minkowski part from 0 to 8', our expression for the small-W
interval between x and x'. From Eq. (2.19) the behavior of E(W, x, x') may be used wherever it is
propagator is then valid. The result for K(x, x') when m =0 is

K x, x')
i
=- 4m' 1
(3.3) K(x, x') =K, (x, x')
s(x, x')+is '
e iss(x, s )/4lVp'
which is the correct Feynman propagator. In any s,(x, x')+i&
coordinates in which the Minkowski metric is ana-
lytic, s(x, x') will be an analytic function of the co- This expression can be used to continue K(x, x') to
ordinates and K(x, x') will also be analytic except values of x off the complexified horizon. With
at the poles where s(x, x') =- ic. These poles cor- m &0, K(x, x') also has singularities whenever
respond to the null geodesics connecting x and x'. s, (x, x') = —ie, i.e., slightly displaced from wher-
It is regularity at infinity plus the location of these ever there is a null geodesic connecting x' to the
poles in the complex coordinate plane that uniquely complexified horizon. We shall now locate these
fixes K(x, x') as a, solution of the inhomogeneous points.
wave equation. More concretely, in the usual rec- To start with we shall show that all null geodes-
tangular Minkowski coordinates with x = (f, x) and ics which start from real values of x' intersect the
x'=(f', x'), K(x, x') has poles at t —f' complexified horizon on the real section, i.e. , at
=+(~x- x' —ie). Thus K(x, x') is that solution of
~
real values of U and V. For definiteness let us
2194 J. B. HARTLE AND S. W. HA WEIN G

consider first the geodesics which connect a real f(r) = [& —b'x '(l —2M/x)]'/' to avoid the positive
x' exterior to the hole to the future horizon. In- real r axis and to have r =r' and r =2M on the
stead of the affine parameter V on the horizon it same sheet of the Riemann surface of f.The com-
is convenient to use the Killing time v related to f
plex analytic structure of for other values of b
it by V= exp(xv). This is just the familiar advanced is then fixed by analytic continuation in that vari-
time of the Eddington-Finkelstein coordinates able.
which cover the region V~0. Complex null geo- A given complex value of b and a given contour
desics may be represented by giving the four co- in the r plane connecting r = r' with =2M should
r
ordinates as functions of a complex affine parame- determine a unique complex null geodesic. For
ter which we shall call X. The geodesic is thus example, for every real b between 0 and SWSM
really a two-dimensional sheet in the complex co- and a purely real contour between r = r' and = 2M
r
ordinate space. there is a unique real null geodesic. For the same
From the relation V= exp(vv), a given value of value of b a second contour which could not be ob-
n and one displaced from it by Imv = 2v/z corre- tained from the first by a smooth distortion would
spond to the same value of V. A consequence of determine a different complex null geodesic.
this is that by studying the null geodesics whose However, it is easily verified from Eqs. (3.8)
v coordinates are confined to a strip of width 2m/v and (3.9) that there are no poles in the integrands
in the complex plane one learns about the null geo- of these equations to prevent one contour which
desics for all other values of v. It is convenient connects =r' and =2M on the given sheet from
r r
to choose a strip which includes the real v axis. being distorted into any other. The most conven-
We may then suppose that at X=O the coordinates ient choice for the contour, therefore, is simply
assume the real starting values v', r', 8', y'. Our to take it to lie along the real axis provided that it
question is what complex values of v in this strip does not intersect a cut of f(r).
with real values of 8 and p and r=2M lie on the For a given contour the integral in Eq. (3.9) de-
two-dimensional sheet which represents a complex fines a function y/b which is a multivalued func-
null geodesic '? tion of b2. In order to obtain a unique connection
Null geodesics in this stationary spherically between y and b it will be necessary to restrict b'
symmetric spacetime may without loss in generali- to a given sheet of the Riemann surface of p/b. We
ty be taken to be in the equatorial plane, 8 = v/2. shall call this the physical sheet. This sheet must
They are characterized by two constants of the include the real axis between b'=0 and b'=27M
motion e and l which may take complex values. which corresponds to the physical real null geodes-
The definitions of these constants are ics. There is a branch point of the function rp/5 at
b' =27M'. It is therefore convenient to define the
2M dv ~dq
1— l= (3 'l) physical sheet as the plane cut along the positive
real axis from 2VM' to infinity and containing the
Then from the null condition g(x, x) = 0 we derive physical real values from b'=0 to 2VM'. It is then
the familiar expressions easily verified that the contour in Eq. (3.9) can
always be chosen to lie along the real axis. With
2M
1 b' on the physical sheet the integral in Eq. (3.9)
J,. 1 —2M/r [1 —~ v '(I —2M/r)]' 'I ' then defines a unique connection between p and b.
Of interest in the present, instance are null geo-
(3.8)
desics which have real values of p at the end point.
and An elementary analysis of the integral in Eq. (3.9)
shows that since the contour can be chosen real
(3.9) for b' on the physical sheet, the imaginary part of
the integrand is always of one sign and the imagi-
Here we have written b for the impact parameter nary part of p does not vanish unless b' is real
l/e. The multiplicative arbitrariness in the affine and between 0 and 27M'. However, these values
parameter X implies that the invariantratio b is suf- of b mean that v at the end point will also be real.
ficient to completely characterize a particular null Thus, complex null geodesics starting from x' in-
geodesic. tersect the complexified future horizon only for
The purely real geodesics connecting x' to the real values of V. A similar conclusion clearly
future horizon correspond to real values of b be- holds for the past horizon. There are then singu-
tween 0 and S~M. For these values, x=r' and larities of K on the complexified horizon slightly
r =2M can be connected by a purely real contour. displaced from the real values of U and V at which
For this reason for real b between 0 and 3~3M it null geodesics from x' intersect the horizon ac-
is convenient to take the cuts of cording to the relation s(x, x') = —ie We shall now.
 PATH-INTEGRAL DERIVATION OF BLACK- HOLE RADIANCE 2195

determine the direction of these displacements. propagator for regions other than the complexified
Suppose that &p is the end point on the future horizon. To complete the program outlined in the
horizon of a real null geodesic which starts at x'. Introduction we shall be concerned in particular
If Vp repre sents the value of V as soc iated with xp, with the analytic properties in the Schwarzschild
then for values of V near to Vp s will behave as coordinate t. This is connected to the null coordi-
~S
s(x x')= ~V —(V- V)+ ~ ~ ~
. (3.10)
nates U and V through the relations

U I y 2M 2/2 (r-t)/4N) ~
Xp
U&0, V&0 (region II)
V (I 2r/2~)1/2e(r+ t)/4N I
A positive value of (Bs/()V)„means that the solu-
tion of s(x, x ) =- ie is in the lower half plane while (3. 13a)
a negative value means that it is in the upper half
plane. To determine the correct sign let k5V be U (tr/2M I ) t/2 (r t) /4-
the displacement vector from the null geodesic to NU&, V& 0 (region I)
I)'/2 (r+ ')/4N
a neighboring geodesic which starts at x' but ends V = (y/2)if
on the future horizon a small affine parameter dis- (3. 13b)
tance 5V to the future of V, . Thus on the horizon
k = 8/SV. If I is the tangent vector to the null geo- and similar relations with the signs of U and V
desic then on the horizon l k &0. The equation of changed in the quadrants reflected in the origin of
geodesic deviation implies that the U- V plane. These relations are indicated sche-
matically on a Penrose diagram in Fig. 3.
d '(I' k)/d)). ' = 0, (3.11) For definiteness let us first consider the case
when x' is exterior to the black hole and x is in
and this relation can be used to propagate l'k back
region II. The portion of the future horizon with
along the null geodesic to X=O, where both geodes-
ics originate. One finds
l'k =cX, (3.12)
where c is a negative constant. Since the tangent
vector along the neighboring geodesic is the sum
of I and k()V, Eq. (3.12) implies that the neighbor-
ing geodesic is timelike. Thus s(x, x ) is negative
as x runs along the neighboring curve and
(ss/s V)„&0. The singularities of the propagator
corresponding to s(x, x') = —ie therefore lie in the
upper half plane and the propagator will be analytic
in the lower half V plane on the complexified
horizon.
In a similar manner the analytic properties of
FIG. 3. Penrose diagram for the Schwarzschild
A
the propagator on the past complexified horizon
geometry. The amplitude for a black hole to emit
can be deduced. For x' located at a real point out- particles which are detected in a mode of energy E by
side the black hole, and x on the complexified past an observer in region I may be related [Eq. f4.4)] to the
horizon, K(x, x') will be analytic in the upper half integral of exp(-iEt) times the propagator to go from a
U plane. point x on a surface C+ of constant r in region II to a
The analytic properties which we have deduced point x' on the detector's world line in region I. The
from the path integral for the propagator on the integral is over the coordinate t on C+. The propagator
is analytic in the coordinate t except for singularities at
complexified horizon may now be considered as those values where a null geodesic from x intersects the
boundary conditions which define the propagator as complexified surface C+. One of these values is the
a particular solution of the inhomogeneous wave real value of t corresponding to the intersection with
equation, Eq. (3.1). For all of our subsequent re- C+ of the radial real future-directed null geodesic from
sults we could have started from this definition of x. As shown in the text, if x has a time coordinate with
the propagator in terms of its analytic properties an imaginary part -4Clf, it actually corresponds to the
on the complexified horizon, but such a definition point x in region III, which is x reflected in the origin.
There is thus another singularity in t with imaginary
would lack the physical motivation which our defi-
part -4mM corresponding to the radial real past-directed
nition in terms of the path integral gives. null geodesic from x intersecting the surface C which
The inhomogeneous scalar wave equation togeth- is the reflection of C+. The two singularities just dis-
er with the boundary conditions just deduced may cussed are repeated at intervals of S~M in Imt. The lo-
be used to derive the analytic properties of the cation of the singularity is illustrated in Fig. 4(a).
2196 J. B. HARTLE AND S. W. HAWKING

V~ 0 together with the part of the past horizon region II when continued in t to t —4mMi then equals
with U ~ 0 are a complete characteristic Cauchy the propagator from a point x" in region III to x'
surface for region II. The propagator in the in- in the exterior region. The point x" is just x re-
terior region is uniquely determined through the flected in the origin of the U- V plane. This identi-
differential equation by the initial data on this ty will be the basis of our derivation of black-hole
Cauchy surface. These are just the values of the radiance in the next section, but for the present
propagator on the relevant parts of the horizon. we simply note that it implies that immediately
Complex values of t with r, 8, and y fixed cor- below the line o =-4@M there are singularities
respond to certain complex values of U and V ac- corresponding to the real null geodesic which con-
cording to Eq. (3. 13a). If we let t =7 +iv then in nect a point on the surface of constant r in region
particular III to x'. The singularities in this case lie below
U-(U)e ' "" V=(V(e"'" (3. 14)
the real axis because x" lies in region III and from
the relations analogous to Eq. (3.14) the small pos-
The problem of determining the propagator at a itive imaginary value of U and a negative value of
certain complex value of t may be considered as V which locate the pole correspond to a negative
a problem of solving the wave equation, Eq. (3.1), imaginary value of t.
for a fixed value of v in the real coordinates U~~
In this way the analytic properties of the propa-
and V~. Since the metric is independent of t the
~
gator K(x, x') in the variable f become apparent.
equation is hyperbolic for any value of 0 and the For fixed 0, y, r in region II with x' located in re-
characteristic initial-value problem is well posed. gion I they are illustrated in Fig. 4(a). The propa-
The analyticity of the propagator on the complexi- gator is periodic in v=1m(t) with period 8mM. The
fied horizon in the upper half U plane and in the regions of analyticity corresponding to the upper
lower half V plane implies that the Cauchy data half U plane and lower half Vplane on the com-
for this real problem are regular provided
& 0 & 0. The standard existence and uniqueness
-4' plexified horizon are the shaded strips of width
There are periodic singularities corres-
4mM.
theorem for the hyperbolic characteristic initial ponding to the real null geodesics which connect
value problem guarantees that there will be a so- x' to a point on the curve of given 8, y, r either in
lution for the propagator for this range of 0. To the past or in the future.
determine whether the resulting solution is analy- If x and x' are both located in region I the propa-
tic in t we need only verify that the Cauchy-Rie-
mann condition is satisfied. This is (a bar denotes
gator is still periodic in 0 with period 8'
consequence of Eq. (3.13b). Now, however, there
as a

complex conjugation) are real values of t both in the past and in the fu-
ture of x' for which there are real null geodesics
(3.15) connecting it to the fixed r, 9, y curve. For the
values to the future of x' the corresponding singu-
where the derivative with respect to t is being larities are displaced slightly above the real t
taken at constant r. Evidently this condition is axis. For the values of t to the past, the corres-
satisfied by the data for -4'& o & 0 since K is an ponding singularities are displaced slightly below
analytic function of U and V in the appropriate half the real t axis. These singularities are shown in
planes on the complexified horizon. Furthermore Fig. 4(b). There are no singularities correspond-
(8/st)„commutes with ' so that determining ing to real null geodesics near o =+4' since the
(8 K/Bt )„may be regarded as a problem of solving corresponding real points lie in region IV, every
the wave equation with zero data on the character- point of which is separated by a spacelike interval
istic Cauchy surfaces. The unique answer to this from x'.
problem is (aK/st)„=0. One concludes, therefore, The propagator K(x, x ) is periodic in imaginary
that for x' in the exterior region and fixed r, 8, y t because it follows from the path-integral defini-
in the region U&0, V&0, K(x, x') is analytic in t tion that the propagator is an analytic function of
in a strip of width 4' below the real axis.
The strip of analyticity cannot be extended above
the Kruskal coordinates U and V except at the
singularities we have described. However, the
the real axis because immediately above it there Schwarzschild coordinate, t, has a logarithmic
are singularities corresponding to the real null singularity as a function of U and V and is multi-
geodesics which connect a value of t on the sur- valued; it is defined only modulo 8wiM. Thus if
face of given r to x' (see Fig. 3). The strip of the propagator has a singularity at some value of
analyticity cannot be pushed below 0 =-4aM' either. U and V, it will have periodic singularities at in-
From Eq. (3.14), this value of v corresponds to a tervals of 8giM when expressed as a function of
U and V which are again in the real section but re- t. The propagator is similar to that suggested by
versed in sign. The propagator K(x, x') with x in Unruh. " By contrast, the propagator proposed by
13 PATH-INTEGRAL DERIVATION OF BLA CK- HOLE RADIANCE 2197

Boulware'4 is not periodic in t because it is not horizons will not see any particles. This is an il-
analytic on the two horizons. lustration of the fact that the concept of particles
The existence of periodic singularities in t im- is observer-dependent. "
In the next section we
plies that observers moving on lines of constant shal1. show that the propagator constructed here
r, 8, y in the extended Schwarzschild solution will will give for observers at a constant distance from
detect particles. This is very similar to the fact the black hole the same rate of particle production
that, as Unruh has pointed out, observers moving as was obtained in Ref. 5 through a study of the
on world lines of uniform acceleration in Minkow- mixing of positive and negative frequencies in a
ski space will also detect particles. On the other gravitational collapse.
hand, an observer moving along either of the two
IV. BLACK-HOLE RADIANCE
In the preceding section we demonstrated the
Im t
analyticity of the propagator K(x, x') in a strip in
M 12aM the t plane when x is in region II and x' is in
region I of the Schwarzschild geometry. We shall
SvrM SvrM X X
now use this anaiyticity to derive the thermal
M 4~M radiation from a Schwarzschild black hole.
Suppose we surround a Schwarzschild black hole
X X
Re t Re t by particle detectors at some large constant ra-
4' M -4mM dius R. These detectors measure particles com-
X
—SwM ing out from inside the surface in modes
X X fi(t', r', e', y') which are purely positive-frequen-
—12wM cy (with respect to t') solutions of the scalar
wave equation. The amplitude that a particle is
detected in a mode f, (x') having started in a mode
(a) (b)
lt&(x) on some surface which bounds a region in-
FIG. 4. The analytic structure of the propagator in terior to 8 is
the complex t plane. (a) shows the analytic structure of
K (x, x') for x' fixed outside the black hole (region I) and da~ x' ~" x ~
x' ~~K x', x ~„~~ x,
fixed values of r, 8, y inside the future horizon (region
IQ. The propagator is periodic in Im(a) with period (4. 1)
8' as a consequence of the relation of t [Eq. (3.14)]
to the coordinates U and V in which the metric is analy- where the integral over x' is taken over the sur-
tic everywhere except at the physical singularity. The face r'=A and the integral over x is over the
shaded regions are the regions of analyticity in t which interior bounding surface. The notation aB&b
are deduced from the analyticity of the propagator in the means ab „— a „b.
upper half U plane and lower half V plane on the com- Suppose for the moment that the particle detec-
plexified horizon. The crosses locate the singularities tors are confined to a time interval t'E(- t,', t,'),
which correspond to the real null geodesics which con- where t,' is very large. Eventually the limit
nect x' to the curve of constant r, 8, y. A typical sit- '-~
t, will be taken. The interior bounding surface
uation is illustrated in Fig. 3. There are singularities
immediately above the real axis corresponding to the mentioned above can then be taken to be a space-
null geodesics which connect x' to the fixed r, 8, y like surface through the precollapse star at -t,',
curve in region II. There are singularities below the a complete spacelike surface inside the future
Imt =-4vrM line corresponding to the real geodesics horizon and a timelike surface connecting this
which connect x' to the fixed r, 0, y curve in region III. to the ~' =R surface outside at t' = t,'. The space-
In each case there is an infinite sequence of singularities like surface inside the future horizon will be
(only a few of which are shown) which arises because taken to be part of a constant-x surface C, out-
there are nu11 geodesics which spiral an arbitrary num-
ber of times near r =3M and thus can connect the fixed side the matter and a spacelike extension inside
r, 9, y curve to x' at increasingly large values of it. The complete spacelike surface inside the
1t t'1. The singularit—ies at other values of Imt are future horizon could have been chosen to be the
duplicates of these as a consequence of the periodicity future singularity were it not convenient to avoid
of the propagator in lmt with period 8aM. (b) shows mathematical complications associated with the
the similar analytic structure when both x and x' are in singularity in the metric at r =0 by keeping it
Region I. The propagator remains periodic in Imt with
away from those points.
period 8mlVi. There are now infinite sequences of null
We now calculate the total probability that a
geodesics which connect x' to a curve of fixed r, 0, y
in the future and in the past. Correspondingly there are particle is measured by a detector in a mode
singularities above and below the real t axis. These are f, (x') which is peaked in time about some late
periodically repeated in Im t with period BxM. time to. This probability will be the sum of the
2198 J. B. HARTLE AND S. W. HAWKING

square of the amplitude in Eq. (4. 1) over all modes abilities are computed the formal square of the 5
h, (x) which are consistent with our knowledge of function will be replaced by a density-of-states
the bounding surface. There will be no contri- factor in the usual way. Factoring out this ener-
bution from modes which are localized on the gy-conservation 6 function there remains of Eq.
spacelike part at t' = —t,' because we are assum- (4.1).
ing that there are no scalar particles in the initial
state. The contribution of the other two surfaces
f do(B )
f dd(R) f, (R') — (R', B) —dt(R),
d
is significantly restricted by the fact that the prop-
agator K(x, x') in the exterior geometry of a col- (4.2)
lapsing star at late times will be a function only of
where R and R' denote the coordinates r, 8, y and
the difference t'- t as a consequence of the time r', 8', y' respectively, f, (R) and hz(R) denote the
transition invariance of the Schwarzschild geom-
angular parts of the respective modes, and de(R)
etry. Thus if f, (x') is peaked about a late time to
and d()'(R') are appropriately weighted angular inte-
only times t comparable to t,' will contribute in
grals. The crucial information about the emis-
Eq. (4. 1). In particular there will be no contribu-
sion is contained in the amplitude 4z defined by
tion from the timelike surface which starts at t,'
since by choosing t,' large enough this only inter-
Ss(R', R) = J( dte ' 'K(0, 8'; t, R). (4.3)
sects values of t much larger than t,'. Further- ~00

more, the only part of the spacelike surface in-

side the horizon which contributes is the part with Making use of the symmetry of K(x, x') under in-
tp and for sufficiently late times this is well terchange of x and x' this can also be written in
outside the matter. The conclusion of this is that what will be the more convenient form
if one is interested in the probability of production
of particles at late times the details of the col-
lapse may be ignored. The only part of the ampli-
d (B', B) =
I dt d ' ' K(t(; 0,tR'). , (4.4}

tude in Eq. (4. 1}which contributes to the probabil-

The amplitude 8~ is the component with energy E
ity of particle production is that in which the inte-
of the amplitude to propagate from the surface C,
gral over x is taken over the complete spacelike
to a point (0, 8') outside the black hole.
surface inside the horizon, and this can be ideal-
Following the program outlined in the Introduc-
ized as a surface C, of constant r between 0 and
tion we now wish to relate the amplitude for emis-
2M in the exact Schwarzschild geometry. We are
sion as contained in Eqs. (4.2) and (4.4) to an
explicitly assuming here that the propagator be-
amplitude for the black hole to absorb a particle
tween a point interior to the future horizon and a
in the same mode by distorting the contours of
point outside the black hole in the geometry of a
integration into the complex plane. To do this it
collapsing star is well approximated at late times
is enough to concentrate on the amplitude 8a(R', R}
by the propagator we have obtained in the analyti-
and distort the contour of the t integration in Eq.
cally extended Schwarzschild metric.
(4.4) downward by an amount —4vMi. This distor-
To compute the total probability that a particle
tion is permissible since the main result of the
is detected we next note that there is no informa-
preceding section is that K(t, R;0, R) is analytic
tion on the state of the particle on the future sin-
in a strip of width 4@M below the real axis. Equa-
gularity. The total probability is obtained by sum-
tion (4.4) becomes
ming the modulus squared of Eq. (4. 1) over a
complete set of states on C+. It is not necessary
to carry this sum out in detail to derive the black- &a(R', R)=e " 'JI dte 'a'K(t —iv/x, R;O, R'),
hole radiance as we shall now show.
Of chief interest is the amplitude for a black
hole to emit a mode with a definite positive ener-
(4.5)
gy E. The time dependence of such a mode is where we have written the surface gravity of the
f-exp(- iEt'). Because of the time translation black hole g instead of 1/4M. Equation (3.14)
invariance the modes h, (x) in the complete set on shows that translating t by an amount —iv/a is
C, may also be classified into modes with the time equivalent to reflecting the Kruskal coordinates
dependence exp(- iEt), although since t is a space- U and V in the origin, The integral in Eq. (4.5)
like coordinate inside the horizon, E is not to be can thus be interpreted as the component with en-
interpreted as a local energy. The fact that K is ergy E of the amplitude to propagate to a point
a function only of the difference t'- t means that (0, R') outside the black hole from the surface C
the integral over t and t' in Eq. (4.1) will lead to in region III, which is C, reflected in the origin
an energy-conservation 5 function. When prob- of the U-V plane. If this integral is inserted in
13 PATH-INTEGRAL DERIVATION OF BLACK-HOLE RADIANCE 2199

Eq. (4.2) in place of Ss we obtain minus the ampli- is exactly equal to the modulus squared of the
tude for a particle to be detected in a mode f, of amplitude for the black hole to absorb a particle
definite energy E having started on C in region which starts at (0, R') in a mode f, with energy E
III in a mode h& with the same energy. The minus and arrives at C+ in a mode. h& with the same en-
sign occurs because the appropriate normal to ergy. When the sum over the complete set of
the surface C is reversed. By time-reversal states h& is performed we have the general rela-
invariance the modulus squared of this amplitude tion

black hole to emit a particle in a mode with energy E)

=e "
(probability for a Schwarzschild
~" x (probability for a Schwarzschiid black hole to absorb a particle in the same mode). (4 5)

This is the fundamental connection between emis- which x' is fixed outside the black hole and x is
sion and absorption stated in the Introduction. integrated over a spacelike surface interior to the
This connection shows that a black hole wiQ- emit horizon which divides the past and future. It is
particles with a thermal spectrum characterized convenient to call this surface C, and take for it a
by a, temperature T = z/2wk. Thus we recover" surface of constant s in the usual Boyer-Lindquist
the result of Ref. 5. coordinates such that r &r &r, . In order to gen-
eralize our result to the Kerr geometry we shall
V. ROTATION, CHARGE, AND SPIN thus need a set of coordinates in which the metric
is regular over at least regions I, II, and III shown
In this section we shall comment on the general-
in Fig. 5(a). Fortunately Carter" has given such
ization of our results to particles with higher spin
a set of coordinates. To avoid lengthy redefinition
and to black holes with rotation and charge.
we shall use his notation wherever it does not dif-
For fields of spin greater than zero it is diffi- fer from that used elsewhere in this paper. The
cult to express the propagator in terms of an in-
reader is referred to Carter's paper for symbo1s
tegral over paths. However, we shall assume that not defined here.
the analytic properties of the higher-spin propaga-
Region I in the Kerr geometry can be covered by
tors are the same of those we have derived for Boyer-Lindquist coordinates (t, r, 8, q) with r &r,
the scalar field. Namely, we shall assume that if
(Carter uses f, q for our't, y). Regions II and III
U and V are affine parameters along the past and
can be covered by a similar patch withe &r&r .
future horizons, respectively, both increasing
Regions I and II can be covered by a coordinate
toward the future, the propagator from a point
patch of the Kerr-Newman form involving an ad-
outside the black hole to a point on the horizon is
vanced time v. Similarly regions I and 1II can be
analytic in the upper half U plane and in the lower
covered by a patch involving a retarded time u.
half V plane. If this assumption is taken then the
(Carter uses u for our v and -m for our M. ) In
generalization of our results to higher-spin par-
region I these coordinates are connected by rela-
ticle is immediate since the derivation of black-
hole radiance given in the preceding section for
Schwarzschild black holes and that to be given be-
low for more general black holes depend only on
this analyticity property and the correct construc-
tion of the emission and absorption amplitudes.
We now proceed to the generalization of our argu-
ment to the Kerr and Reissner-Nordstrom black
holes. For simplicity we treat these two cases
separately, leaving it to the reader to join the
arguments together for the general rotating
charged black hole.

A. The Kerr black hole

Figure 5(a) shows the familiar Penrose diagram (a) (b)
for the axis of a Kerr black hole. Following our
argument for the Schwarzschild case, the ampli- FIG. 5. (a) The Penrose diagram for the axis of a
tude for the black hole to emit a particle can be Kerr black hole. (b) The Penrose diagram for a B,eiss-
related to an integral of the propagator K(x, x') in ner-Nordstrom black hole.
2200 J. B. HARTLE AND . S. W. HAWKING

tions of the form surface of constant & inside the future horizon
of exp(- jEt) times the propagator to go from a
(5. 1)
point x on that surface to a point x' outside the
(5.2) hole. In the Kerr case we will be interested in
the amplitude for the emission of a particle of
where df/dr =(r'+a')/b, . Carter then introduces
energy E and an angular momentum along the
a new azimuthal Killing angle cp' defined by axis of rotation m (m is not to be confused with
+ = 0' —+ty
'P (5 8) the rest mass). In a similar fashion this ampli-
tude can be related to an integral of the form
where cu, = a/(r, ~+ a') is the angular frequency of
the horizon. He also introduces two new null co-
ordinates x and y which correspond to our U and
V. These are defined as
U= —8 "", V=e', region I, (5.4a) ~00 0
dye '' "'K(t, (p, R;0, 0, R').
U=e "", V=e"", region II, (5.4b)
(5.5)
U=-e '", V = —e"", region III, (5.4c)
Here K is the propagator in the Kerr geometry
where z=-,'(r, -r )/(r, '+a') is the surface gravity which, because of the time-translation invariance
of the black hole. Thus V is Carter's x while U is and axial symmetry, depends only on the differ-
the negative of Carter's y. In this new (U, V, r, y') ence in the t and y coordinates of x and x'. The
patch the metric is analytic in regions I, II, and quantities R and R' stand for the x, 6 coordinates
III [see Carter's Eq. (26)). The future horizon is of x and x'. The integral is over the surface C, .
located at U=O and V is an affine coordinate along Arguments similar to those given in the Schwarzs-
it. The past horizon is at V=0 and U is an affine child case will show that, for fixed r, L9, y', K is
coordinate along it. The coordinate x may be analytic with U in the upper half plane and V in
considered to be defined in terms of U and V by the lower half plane. We can therefore distort
Eqs. (5. 1), (5.2), and (5.4). Figure 5(a) gives a the contour of the t integration downward by an
schematic representation of the various definitions. amount —iw/x keeping r, 8, y' fixed since that
In the Schwarzschild case the amplitude for a amounts to rotating U by an angle m and V by an
black hole to emit a particle with energy E was angle -m. Keeping y' fixed means from Eq. (5.3)
ultimately related to an integral he(R', R) over a that y —y —imur, /~ in the process. Thus,

+oo 2 'If
ge„(R', R)= e " + " dt dkpK(t —iv/g, y- iv&o /i&, R;0, 0, R'). (5.6)
~oo 0

Since this displacement of t is equivalent to [see Eq. (5.4)] U- —U and V- —V the integration is now over
the reflected surface of constant r which is in region 111 and which we have shown as C in Fig. 5(a). The
remaining integral can be related to the amplitude for the black hole to absorb a particle of energy E and
angular momentum m. Thus we have
(probability for a Kerr black hole to emit a mode with energy Z and angular momentum m)
=e "ie " +it" x (probability for a Kerr black hole to absorb a mode with Z and m). (5.7)

This is exactly the relation necessary to establish geometry. The propagator for a particle of
that a rotating black hole will emit particles with charge q is a solution of the wave equation
an expected number per mode proportional to
g"8(V
'' in-'
—iq A ) (V s —iq A8)K(x, x') = —5(x, x'),
(exp[(E —m&u, )2w/g] —1) so that z/2nkmay be
terpreted as the black hole's temperature. (5.8)
in the Reissner-Nordstr5m background geome'try.
B. Reissner-Nordstrom black hole In the usual gauge the only nonvanishing compo-
The situation with the Reissner-Nordstr5m nent of A„ is
black hole is similar. Here, however, we shall
investigate the amplitude for the black hole to A, (x) =e/r, (5.9)
emit particles of charge q. Figure 5(b) shows where e is the charge on the black hole. How-
the Penrose diagram for the Reissner-Nordstr5m ever, such a gauge K cannot be expected to be
 P AT H-IN T E GRAL DE BIVA TION OF BLACK- HOLE RADIANCE 2201

analytic in the upper half U plane or lower half where R stands for the coordinates r, 8, y.
V plane as is required by our boundary conditions In the new gauge K will be analytic in the upper
because the components of A„(x} will not be ana- half U plane and the lower half V plane. The co-
lytic on the horizon in coordinates which are ana- ordinates U and V here are related to those of
lytic there. For example, if we use the (u, r, 8, q)) Carter as described in Eq. (5.4) and above. As
coordinates which are analytic on the future hori- in the Kerr case this analyticity implies that the
zon, A„(x) = (e/r)(1 —2M/r+e'/r'} ', which di- contour of the t integration in Eq. (5.12} may be
verges at r = ~, . A gauge in which A„ is stationary distorted downward in the complex t plane by an
and regular on both horizons can be found by amount —im/x keeping r fixed since this amounts
making the transformation to rotating U by n' and V by w. Thus,
+p +/+A py (5.10) d g ()(' R)=x'" """I d(x-" -'e)'
where
A =4't, (5.11)
xK(t -Ar/a, R; 0, R').
and C is the potential
on the horizon equal to
e/r, . In the new gauge A„will be analytic in the (5.ls)
same domain as the metric but it will not vanish
at infinity. This means that the time dependence Since displacing t by —im/x is equivalent to the
of a mode of energy E at large distance from the reflection U- —U and V- —V this integral may be
hole will not be exp(-iEt) but rather written
exp[- (i(E —q4) t].
As in the Schwarzschild and Kerr cases the Ss(R', R) =e" ' "
Jt dte ' 'K(t, R;O, R').
amplitude for the black hole to emit a particle
with energy E and charge q can be expressed in
terms of an integral of the propagator over a (5.14)
constant-r surface C, which lies between the
future horizon at r, and the inner horizon at r . where the integral is now over the reflected sur-
By the time invariance of the propagator the face C illustrated in Fig. 5(b). The integral in
amplitude for emission depends on the integral
Eq. (5.14) can be related, as before, to the ampli-
' ' K(t, R;O, R'), tude for the black hole to absorb a particle of
Ss(R', 8) = dt 8 '
(

charge q. Following through the arguments which

(5.12) led to Eq. (4.6) we have

(probability for a Reissner-Nordstr6m black hole to emit a particle of charge q and energy E
=e '"(s '@)"x (probability for a Reissner-Nordstrom black hole
to absorb a particle of charge q and energy E) 15).
(5.
This is exactly the relation necessary to establish that a rotating black hole will emit scalar charged
particles with an expected number per mode proportional to [exp[(E —q4)/kT] —I] ', where kT = z/2w.
(See Refs. 5 and 19.)

APPENDIX: DERIVATION OF THE DIFFUSION EQUATION FOR F(o„x,x')

The amplitude F(Q, x, x') is defined by the path integral

' 0
P((), x, x') = f ()x[x]exp —— y(x, d)dtx

where y is the positive-definite metric of Eq. (2.6), x=dx/d(d, and the sum ranges over all paths with
x(0) =x' and x(Q) =x. We shall now derive the diffusion equation [Eq. (2.8)] for F and in the process dis-
cuss the interpretation of the differential 6x[(d].
Divide the interval [0, Q] up into %+1 intervals each e long. With a natural assignment of the weight to
the integrals over spacetime the path integral in Eq. (2.1) may be interpreted as
d4 4 d'
Ii(Q, x, x') = lim I
& [y(xz)] &
~ ~

& [y(x, )] exp g S(e,

&=0
x&+i, x&) (A2)
2202 J. B. HARTLE AND S. W. HAWKING

where x, =x xN y=x, and S is given by integral d~zexp —45 Szzs e

S(e, x;„,x, )=4 d(oy(x, x), (A3) = (4wz)'.

0 (A8)

evaluated along the geodesic path which connects Because the odd integrations in z vanish, the
x& at + =0 with x„,
at ~ =e. The constant A is a first-order term in e is
normalization fixed by the requirement that the
amplitude for a particle to propagate from one z —=B
BQ
ns
(~E —-R 8E}. (A9)
point to any other point in the spacetime is unity.
This is equivalent to where the curvature term comes from the expan-
sion
A = Jl d'x[y(x)]' 'exp[- S(e, x, x')]. (A4) y'~' =1- 6'R z~zz+ 0(z'), (Al0)
The reason that Eq. (A2) is correct is that as z and B 8 is the integral
becomes smaller and smaller the action for the
paths which connect fixed x& to x;„will become
larger and larger and hence their contribution to B"8= t d'z(z "z8/4) exp(- M„„z"z"/z)
the integral will be exponentially damped. The
dominant contribution will come from the station- =2ea '. (Al 1)
ary path for which S is a minimum. This is a
geodesic between the points. Using this expression we finally arrive at the
The diffusion equation can be derived by con- diffusion equation for E by replacing the partial
sidering the relation between E(Q+e, x, x') and derivatives in the normal coordinates by covariant
E(Q, x, x'). From Eq. (A2) this is derivatives 0 with respect to the metric y. The
equation is
E(Q+ e, x, x') = d4y[y(y)]'~' exp[- S(c, x, y)]

x E(Q, y, x')/A. (A5)

—=(y
eQ
ng-
v v, — a}s'.
—, (A12)

Write y=x+z and let the integration be over z. On This is exactly Eq. (2.8) taking into account the
the right expand S and F in powers of z. On the fact that R vanishes for Schwarzschild geometry.
left expand I' in powers of e. Analysis of the in- The factor -R/3 in Eq. (A12) is a consequence
tegral shows that only the first few terms of the of our particular choice of weight in the coordinate
expansion on the right contribute to the part of E integrals in Eq. (A2}. If we had replaced
linear in e giving an expression for BE/BQ. This exp[- S(e, x, , „x,.)] in the integrals with
analysis has been ca.rried out by Cheng in a gen-
eral coordinate system. 4 However, the calcula- [y(x„,)/y(x, )]'~' exp[- S(e, x, „,x;) ],
tions are considerably simplified if a Riemann then the resulting equation would have been
normal coordinate system is introduced at x. In
such a coordinate system () p y"
ag
V V8+ —,(s —1)R E. (A13)
y„a(y) = 5„z ——,'R„yz sz~z'+ O(z'). (A8)
Thus any amount of scalar curvature can be had
where 5 & denotes the Kronecker 5. From the both here and in the equation for K by the appro-
definition of normal coordinates it follows that the priate choice of the action. For the vacuum
geodesics from the origin are straight lines so black-hole solutions we are considering 8 vanishes
that x"=z "/z and and these equations are all identical and we will
not consider this issue further. However, See the
S(e, x, x+z) =-,'5„zz "zz/e. (A7) remarks in Ref. 3.
From the form of S we see that the only significant
part of the integral as e-0 comes from the region ACKNOWLEDGMENTS
where z"-c' '.
Thus only terms O(z') in the ex-
pansion of (y)'~' and E need to kept under the in- We are grateful for the hospitality of Kip
tegral and the limits may be extended to infinity. Thorne and the relativity group at Caltech. We
Then, the zeroth-order term in an expansion of would also like to thank Werner Israel and Richard
Eq. (A5) gives the normalization condition Feynman for helpful conversations.
 PATH-INTEGRAL DERIVATION OF BLACK- HOLE RADIAN CE 2203

*Work supported in part by the National Science Founda- Relativistic Quantum Fields (McGraw-Hill, New York,
tion under Grants Nos. MPS-75-01398 and GP-43905. 1965).
)Sherman Fairchild Distinguished Scholar at the Cali- See, for example, E. Nelson, J. Math. Phys. 5, 332
fornia Institute of Technology. (1964), and M. Kac, Probability and Belated ToPics
~R. P. Feynman, Rev. Mod. Phys. 20, 327 {1948). in the Physical Sciences (Interscience, New York,
2C. Morette, Phys. Rev. 81, 848 (1951). 1959) .
B. S. DeWitt, Rev. Mod. Phys. 29, 377 (1957). This is analogous to the "Euclidicity postulate" in
4K. S. Cheng, J. Math. Phys. 13, 1723 (1972). Euclidean quantum field theory. See, for example,
S. W. Hawking, Nature 248, 30 (1974); Commun. Math. K. Symanzik, in Local Quantum Theory, Proceedings
Phys. 43, 199 (1975). of the International School of Physics "Enrico Fermi"
6S. W. Hawking (unpublished). Course 45, edited by R. Jost (Academic, New York,
7R. P. Feynman, Phys. Rev. 76, 769 (1949). 1969), and R. F. Streater, Rep. Prog. phys. 38, 771
For other discussions of proper time in connection with (1975).
relativistic quantum mechanics see (a) V. Fock, Phys. ~3See, for example, S. D. Eidel'man, Parabolic Systems
Zeit. Sowet. Un. 12, 404 (1937); (b) Y. Nambu, Prog. (North-Holland, Amsterdam, 1969), p. 178.
J.
Theor. Phys. 5, 82 (1950); (c) Schwinger, Phys. D. Boulware, Phys. Rev. D 11, 1404 (1975).
Rev. 82, 64 (1951); {d) B. S. DeWitt, Phys. Rep. 19C, ~5W. G. Unruh (unpublished).
295 (1975). ~~A. Ashtekar and A. Magnon (unpublished).
90ne might have though that the parameter time W, ~'A remark is probably in order here comparing the
which in the case of timelike paths is a multiple of the conventions used here and in Ref. 5. We have chosen
proper time, is determined completely by x, x' and the the action in Eq. (2.1) so that, taking account of our
path and not an independently specifiable quantity. signature convention (+2), the variable part of it for
However, that is the case only for differentiable paths, nonrelativistic paths reduces to + f d t (dx/d t)t. This
and nondifferentiable paths are being summed over is the usual choice for the nonrelativistic action and
here. The situation is analogous to that in nonrelativ- leads to a Schrodinger equation of the usual form and
istic quantum mechanics where one can ask for the a time dependence for solutions with energy E of the
amplitude to propagate from a space point x to a space form exp(-iEt). In Ref. 5 wave functions with definite
point x' along a given path in a definite time t. If the frequency had time dependence exp(i~t). Exactly the
path were differentiable t mould be determined, but same physical results would be obtained with this con-
the paths summed over in the Feynman integral for the vention provided one started with an action of the oppo-
propagator are not differentiable. site sign.
~
The definition of E (x, x') corresponds to the I+ of B. Carter, Phys. Rev. 174, 1559 {1968).
Ref. 7 and the Az of J. D. Bjorken and S. D. Drell, G. W. Gibbons, Commun. Math. Phys. 44, 245 (1975).