Maximal Analytic Extension of The Kerr Metric: Additional Information On J. Math. Phys

Maximal Analytic Extension of the Kerr Metric
Robert H. Boyer and Richard W. Lindquist
Citation: J. Math. Phys. 8, 265 (1967); doi: 10.1063/1.1705193

View online: http://dx.doi.org/10.1063/1.1705193
View Table of Contents: http://jmp.aip.org/resource/1/JMAPAQ/v8/i2
Published by the American Institute of Physics.
Additional information on J. Math. Phys.

Journal Homepage: http://jmp.aip.org/
Journal Information: http://jmp.aip.org/about/about_the_journal
Top downloads: http://jmp.aip.org/features/most_downloaded
Information for Authors: http://jmp.aip.org/authors
Downloaded 28 May 2012 to 171.67.34.69. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
JOURNAL OF MATHEMATICAL PHYSICS VOLUME 8. NUMBER 2 FEBRUARY 1967
Maximal Analytic Extension of the Kerr Metric*

ROBERT H. BOYERt
Department of Applied Mathematics, University of Liverpool, Liverpool, England
AND
W. LINDQUIST
RICHARD
Scott Laboratory of Physics, Wesleyan University, Middletown, Connecticut
(Received 19 July 1966)
Kruskal's transformation of the Schwarzschild metric is generalized to apply to the stationary, axi-
ally symmetric vacuum solution of Kerr, and is used to construct a maxiInal analytic extension of
the latter. In the low angular momentum case, at < mt, this extension consists of an infinite sequence
Einstein-Rosen bridges joined in time by successive pairs of horizons. The number of distinct asymp-
totically flat sheets in the extended space can be reduced to four by making suitable identifications.
Several properties of the Kerr metric, including the behavior of geodesics lying in the equatorial plane,
are examined in some detail. Completeness is demonstrated explicitly for a special class of geodesics,
and inferred for all those that do not strike the ring singularity.
I. INTRODUCTION expected in more realistic models. We here lay

the foundations for such a study of stationary
A FAMILIAR feature of exact solutions to the
field equations of general relativity is the pres- axially symmetric models of rotating bodies, by
analyzing in some detail the empty space metric
ence of singularities. Although long recognized in
such special metrics as those of Schwarzschild and of Kerr. 5
Reissner-Nordstrom, and in the spherically sym- Of course one must be careful to distinguish
metric Friedmann models, these singularities have true singularities, to which the Penrose-Hawking
been dismissed as nonphysical, due perhaps to the theorems refer, from "pseudosingularities" that re-
high degree of symmetry assumed for the solutions. 1 flect merely a poor choice of coordinates. The latter
Recent work by Penrose and Hawking2 suggests can always be removed by the familiar device of
that this traditional view may need to be revised; covering the manifold with a faInily of coordinate
it is quite possible that gravitational collapse leads patches. It has long been recognized that the singu-
inevitably to a singular state if trapped surfaces are larity at r = 2m in the standard form of the
once formed 3-provided that one stubbornly per- Schwarzschild metric is of this type; transformations
sists in applying the classical field equations to to coordinate frames which remove this apparent
regions of arbitrarily high curvature. 4 Thus there singularity have been given by several authors. s
is some point in studying the complete geometry Closely related to the problem of singularities
of exact solutions, even though they are idealizations is that of completeness. 7 Solutions of the field equa-
of physically reasonable gravitating systems, in order 6R. P. Kerr, Phys. Rev. Letters 11, 237 (1963).
to learn something about the type of behavior to be 6A. S. Eddington, Nature 113, 192 (1924); this trans-
formation was rediscovered by D. Finkelstein, Phys. Rev. 110,
* Work supported in part by the Aerospace Research 965 (1958). Other forms of the metric which are nonsingular
Laboratories, Office of Aerospace Research, and by the at the Schwarzschild radius were exhibited by G. Lemaitre,
Office of Scientific Research, U. S. Air Force. Ann. Soc. Sci. Bruxelles 53A, 51 (1933), and by J. L. Synge,
t Deceased. Proc. Roy. Irish Acad. A53, 83 (1950). C. Fronsdal, Phys. Rev.
1 See, e.g., E. M. Lifshitz and 1. M. Khalatnikov, Zh. 116,778 (1959); and J. Plebanski, Bull. Acad. Polon. Sci. 10,
Eksperim. i Toor. Fiz. 39, 149, 800 (1960) [English transl.: 373 (1962), independently constructed the maximal analytic
SOVlet Phys.-JETP 12, 108,558 (1961)]. extension of the Schwarzschild manifold by imbedding it in a
t R. Penrose, Phys. Rev. Letters 14, 57 (1965); S. W. six-dimensional flat space. A transformation to a coordinate
Hawking, ibid. 15,689 (1965). See also S. W. Hawking and G. frame which displays this maximal extension was first given
F. R. Ellis, Phys. Letters 17, 246 (1965). by M. D. Kruskal, Phys. Rev. 119, 1742 (1960).
a The collapse of a spherically symmetric mass has been 7 By "completeness" we mean affine completeness: the
studied in detail by M. M. May and R. H. White, Phys. Rev. property that all geodesics can be continued to arbitrarily
141, 1232 (1966). They find that unless the collapse is halted large values of their affine path parameters. A standard theo-
before any fraction of the total mass has fallen within its own rem of Riemannian geometry states that geodesic complete-
Schwarzschild radius, a singularity invariably ensues. ness is equivalent to completeness as usually defined-namely,
• Wheeler has repeatedly emphasized that, from a deeper that all Cauchy sequences converge-if the metric is positive
point of view, these singularities must be nonphysical, since definite; however, this theorem breaks down on manifolds with
quantum effects will necessarily alter the complexion of the an indefinite metric. For a careful discussion of different possi-
problem completely in regions of high curvature. See B. K. ble and inequivalent definitions of completeness for manifolds
Harrison, K. S. Thorne, M. Wakano, and J. A. Wheeler, with an indefinite metric, see W. Kundt, Z. Physik 172, 488
Gravitation Theory and Gravitational Collapse (University of (1963); C. W. Misner, J. Math. Phys. 4, 924 (1963); and M.
Chicago Press, Chicago, Ill., 1965). Fierz and R. Jost, Helv. Phys. Acta 38, 137 (1965).
265
266 R. H. BOYER AND R. W. LINDQUIST
tions are usually presented locally, in terms of a r = f, 8 = 8, <P = ;p, (1.1)

coordinate system adapted to the symmetries of
a given problem. Thus the manifold on which the dt = dl + 2m df/(f - 2m)
metric is to be defined is left unspecified at the start, (where, to agree with our later notation, we use
and determined only afterwards by imposing various bars to denote the original coordinates of Schwarz-
global and topological conditions. Ideally, one would schild) leads to a form of the Schwarzschild metric
like the resulting manifold to be geodesically com- free of singularities at f = 2m:
plete and free of singularities, but the Penrose-
Hawking theorems show that in many cases these ds 2 = dr2 + r2(d~ + sin2 8 d<P2 )
two aims are incompatible. Of course, one can always
eliminate singularities from a given metric by rede- - dt 2 + (2m/r)(dr + dt)2. (1.2)
fining the manifold to exclude the singular points; This metric, interpreted on the manifold r > 0,
this is not a very satisfactory solution, however, if o ~ () ~ 7f', 0 ~ <p < 27f', - 00 < t < 00, is in fact an
physical test particles run off the edge of the space- analytic extension of the original; it is complete
time so obtained in a finite proper time. It seems for t ~ + 00 -excepting those geodesics which
preferable, we believe, to require geodesic complete- strike the true singularity at r = O-but not for
ness, insofar as possible, even though this leads t ~ - 00 • One notes that this form has the structure
in general to singularities. Thus if mr and mr' are of a flat space metric plus the square of a null
analytic manifolds, with mr' :> mr, we call mr' a vector; this feature serves as the point of departure
maximal analytic extension of mr if every geodesic in the Kerr-Schildo theory, to which we refer briefly
of mr' is either complete or terminates at a singular in Sec. II. The alternative transformation
point. Such an extension need not exist,S and even
if it does it will not be unique, because of the freedom r' = f, ()' = 8,
(1.3)
available to identify points in mr' in a variety of
dt' = dl - 2mdf/(f - 2m)
ways without disturbing analyticity.
Analytic Extensions of the Schwarzschild and leads to an expression for di similar to (1.2), but
Reissner-NordstriSm Metrics with dt replaced by -dt', which is complete for
t' ~ - 00. One thus obtains two coordinate patches,
Eddington6 long ago pointed out that the trans- shown in Fig. 1, whose domain of overlap is the
formation
region external to the Schwarzschild pseudosingu-
v larity. This is still not the maximal analytic exten-
sion, however. The latter has been given in a
particularly simple form by Kruskal, 6 and is sum-
marized by the transformation equations
u ± v = l(r/2m) - 11' exp [(r ± l)/4m]. (1.4)
The corresponding manifold, illustrated by the now

familiar Kruskal diagram (Fig. 1), consists of two
asymptotically flat universes whose spacelike sec-
tions are joined together on a 2-sphere of minimum
~ (E) patch ~ (E/) patch area-the "throat" of an Einstein-Rosen bridge. 10
FIG. 1. A Kruskal diagram for the Schwarzschild metric,
The area of this throat changes with time, reaching
showing a portion of the (u, v) coordinate plane. In this plane a maximum value of 47f'(2m)2 at u = 0, v = 0, and
null curves appear as straight lines of slope ± 1; the null lines collapsing to zero both in the future and the past,
u ± v = 0 define the horizons. The curves r = const are
hyperbolas having these horizons as asymptotes, with the at those points (u = 0, v = ±1, for instance) for
r -= 0 hyperbola a singular surface. The shaded regions show which r = O.
the portion of the (u, v) plane in which the Eddington c0-
ordinates (E) and (E') are regular; their overlap is the region A similar analysis for the Reissner-Nordstrom
of reltUlarity u > 0, US > Vi (r > 2m, - < I < Q» for
Q)
solution (the metric outside a single mass m with
stancfard Schwarzschild coordinates. The fourth unshaded
region, u < 0 and US > v2, is also asymptotically flat, and charge q) has been carried through by Graves and
represents the second sheet of an Einstein-Rosen bridge.
II R. P. Kerr and A. Schild, Am. Math. Soc. Symposium,
SA simple counter-example has been constructed by C. W. New York, 1964; also Galileo Quatercentenary, Florence,
Misner, University of Maryland Tech. Rept. No. 529 (1965), 1964.
p.14. 10 A. Einstein and N. Rosen, Phys. Rev. 48, 73 (1935).
MAXIMAL ANALYTIC EXTENSION OF THE KERR METRIC 267
Brill.l1 This metric has two pseudosingularities at simply a barrier that prevents continuation of the
r = r ± == m ± (m 2 - q2)t (1.5) metric to negative values of r. As we see presently,
on the Kerr manifold r = 0 defines a similar barrier,
and a true singularity at r = O. Graves and Brill but one of lower dimension, and the extension to
displayed transformations of the Kruskal type which negative values of r is meaningful and necessary.
would eliminate either pseudosingularity, but not Trapped surfaces exist here, too; however, the Pen-
both. Nevertheless, their results are sufficient to rose-Hawking theorems do not apply to the
portray the full manifold, for by piecing together Reissner-Nordstrom and Kerr manifolds, for these
successive coordinate patches one can extend the do not admit the required Cauchy initial hyper-
metric first across one singularity and then across surface.
the other. It turns out that the resulting manifold is When q2 > m 2, Eq. (1.5) breaks down; there
still not complete, however, and one must continue are then no horizons and much of the previous
to add similar patches indefinitely. The picture which discussion ceases to apply. In fact, the original
emerges is thus altogether different from the coordinate patch covers the maximal extension.
Schwarzschild case. One can view the manifold as
an Einstein-Rosen bridge, whose throat oscillates Kerr Metric
between a minimum and a maximum area, given by We have reviewed the above results at length
411"r~ and 411"r!. After each cycle the bridge is found
since they bear directly on the problem we wish to
to be attached to a new pair of (asymptotically consider-that of obtaining a corresponding maximal
flat) sheets, isometric to the original pair but never- analytic continuation of the Kerr metric. This solu-
theless topologically distinct from them (but see tion, it may be recalled, describes a possible exterior
the end of Sec. III). field outside a rotating body; it is the only known
Horizons example of a stationary vacuum metric with gravita,-
tional mass and rotation that is asymptotically flat.
The pseudosingularities r = r "" like r = 2m in
Like the Schwarzschild metric it is algebraically
the Schwarzschild metric, have an important geo-
special (i.e., of type D in the Petrov-Pirani classifica,-
metric and physical significance: They determine
tion13); thus it contains two geodesic shear-free null
the location of horizons. In stationary space-times,
congruences. a It admits two Killing vectors, as-
these may provisionally12 be defined as stationary
sociated with time translations and rotations about
null hypersurfaces: Whenever, for some function
an axis of symmetry, and it contains two parameters
fer, 6, tp), the equation (grad f)2 = 0 is satisfied m and a which can be identified with the total mass
on f = const, then this hypersurface is called a
and angular momentum per unit mass of the source.
horizon. It acts locally as a one-way membrane of
Although no one has yet succeeded in displaying
infinite red shift (see note added in proof).
explicitly an interior. metric which fits smoothly
Within the horizons of the Schwarzschild and
onto Kerr's exterior solution, there do not appear
Reissner-Nordstrom metrics, when such occur, there
to be any difficulties in principle in integrating
are trapped surfaces in the sense of Penrose. 2 More-
the combined equations of hydrodynamics and gravi-
over, both metrics contain real singularities, located tation for the interior case, provided that the shape
at r = 0 in the coordinates normally used. Time- of the body is chosen appropriately.15 For purposes
like paths which cross the horizons of the latter,
of this paper, however, we neglect the presence
however, need not strike the singularity, but can of any such source, and ask rather for the maximal
be continued onto another sheet and thence out extension of the empty space metric.
to (another) spatial infinity. The surface r = 0 is Kerr's solution contains two event horizons, which
11 J. C. Graves and D. R. Brill, Phys. Rev. 120,1507 (1960).
The case m2 = q' has been studied by B. Carter, Phys. Letters 13 A. Z. Petrov, Sci. Notices, Kazan State Univ. 114, 55
21, 423 (1966). (1954); F. A. E. Pirani, Phys, Rev. lOS, 1089 (1957).
12 B. Carter (private communication) has stat~d a thEl?rem 14 R. Debever, Compt. Rend. 249, 1324 (1959); R. Penrose,
which suggests a more stringent defini~ion of ~ hor~zon. Bl:'lefly: Ann. Phys. (N. Y.) 10, 171 (1960); R. K. Sachs, Proc. Roy.
"Let a space-time admit a group of lsometries wlth p-dlmen- Soc. (London) A264, 309 (1961).
sional integral surfaces, and let the tangen~ p vector b~ (4-p)- 16 R. H. Boyer, Proc. Cambridge. Phil. Soc. 61, 52? (1965).
surface orthogonal. Then the loc~ of !lullity of the ~ill~n~ p A. G. Doroshkevich, Ya. B. Zel'dovlCh and I: D. Novlkov, ~h.
vector is itself a null hypersurface. This sl!rface-the ~Illing Eksperim. i Teor. Fiz. 49, 170 (1965) [English transl.: SOVIet
horizon"-appears in the Schwarzs~hild and Rels~ner Phys.-JETP 22, 122 (1966)] claim that the inter!or solu~ion
Nordstrom metrics with ~ = 1 or 3 and m the Kerr metriC (as must also display some type of vortex or convectIve motlOn,
we see later) with p. = 2. For a different ~efinition of an "event because the Kerr metric has other off-digaonal components
horizon," see W. Rmdler, Monthly Notlces Roy. Astron. Soc. besides g" •• However, their argument is based upon a false pre-
116, 662 (1956). mise; see for example Eq. (2.13).
in an appropriate coordinate system are located at Kerr and Schild showed that the vacuum field
equations require the null congruence to be geodesic
r = r", == m ± (m
2
-
2
a )! (1.6)
(with respect to either gafl or 1/"p). By choosing H
plus a true singularity again formally defined by suitably, ka can be so normalized that
r = O. The correspondence with the Reissner-
Nordstrom result (1.5) is rather striking. Thus one (2.2a)
might expect the maximal extension of the Kerr V fI being the covariant derivative based upon ga{J'
manifold to be topologically very similar to the We suppose this to be the case, and define an affine
Graves-Brill construction. A main object of our parameter p. by k" = dx a/dp.. Then, as Kerr and
work is to justify this expectation, by displaying a Schild noted, one finds
transformation analogous to that of Kruskal. We
give this in detail in Sec. III, after first exhibiting Ra{J"'/6kfJk6 = -WH/dp.2)k"k"'/. (2.2b)
several other useful coordinate frames in Sec. II. It follows that k a is a multiple Debever-Penrosel4
Kruskal's method, as generalized by Graves and vector, and consequently, by the Goldberg-Sachsls
Brill, cannot be applied directly-except to the theorem, that the null congruence defined by k" is
two-dimensional subspace containing the symmetry shear-free as well.
axis and the time coordinatel6-because the metric In their analysis, Kerr and Schild give rules for
depends on the polar angle 0 in a complicated way. constructing the general empty-space metric of the
But when we combine a transformation analogous form (2.1). We do not quote these here, but merely
to (1.4) with an appropriate change in the azimuthal remark that with the exception of the Kerr metric
angle 'P, whose effect is to straighten out the null the representation (2.1) is unique, so that the metric
congruences in the neighborhood of the event horizon is of type II in the Petrov-Pirani classification.
being considered, we find, happily, that the resulting
metric is regular across the horizon at all values of O. Kerr Metric in Explicit Form
The proof that our extension is maximal requires
Let us consider the exceptional case, in which the
a demonstration that geodesics which do not strike a
line element can be represented in the form (2.1)
true singularity can be continued to infinite length.
in two distinct ways:
We show this in Sec. IV by studying the geodesic
equations. Some particular features of the geodesics (2.3a)
themselves, which seemed to us to be interesting
(2.3b)
and curious, are described briefly in Sec. V.
a
Both k and l" are then principal null vectors (i.e.,
II. PROPERTIES OF THE KERR METRIC
double Debever-Penrose vectors), which implies that
Kerr-Schild Theory the metric is of type D (type I degenerate). This
Kerr and Schild9 have studied solutions of the case is of particular interest, for it describes the
vacuum field equations for which the metric has stationary vacuum solution with rotation first ob-
the form 17 tained in a different way by Kerr. 5 He gives the
explicit presentation
gafl = 1/afl + 2Hk akfl· (2.1a)
3
d
Here 1/afl is the metric of Minkowski space, k" a ds
2
= x + dy + dz - dt + (r4 2mr
2 2 2
+ a2z2)
2
null vector field, H a scalar field. It does not matter

whether k" is defined to be null with respect to the X [rex dx + y dy) + a(x dy - y dx) + z dz + dtJ2
flat background metric 1/ afl or the full metric gafl' r2 + a2 r'
since (2.4)
gaflk fl = 1/ aflk fl == ka r being defined by
and therefore g aflk"kfl = 0 implies 1/ aflk"kfl = 0 and [(x
2
+ y2)/(r2 + a2)] + Z2/r 2 = 1, (2.5)
conversely. In fact,
which corresponds to one of the two forms (2.3). We
(2.1b) denote the coordinate system (x, y, z, t) as the (M)
18 B. Carter, Phys. Rev. 141, 1242 (1966), has independently
frame, and identify it with the form (2.3a) in which
worked out the analytic extension of this subspace.
17 Notational conventions: Greek letters range and sum from 18 J. N. Goldberg and R. K. Sachs, Acta Phys. Polan. 22,13
1 to 4; gafl has signature (+++-). (1962).
MAXIMAL ANALYTIC EXTENSION OF THE KERR METRIC 269
the k congruence is displayed. The (altogether dif- x 2 + y2 Z2

ferent) coordinate system (x', y', z', t') adapted to a2 sin2 0 - a2 cos2 0 = 1 (2.9)
the 1 congruence we call the (M') frame. The trans-
it follows that the 0 = const surfaces are hyper-
formation equations relating these two systems can
boloids of one sheet, confocal to the ellipsoids.
be worked out from formulas given later in this
Actually, since z has the same sign as cos 8, the
section.
surface 0 = const is only a half-hyperboloid, trun-
The form (2.4) of the Kerr metric is inconvenient
cated at its waist, lying in the half-space z ;e: 0
for many applications, since r is a complicated func-
according as 0 ~ !7r. Note that at r = 0 the ellipsoid
tion of x, y, z. Kerr has given alternative and more
degenerates to a disk, x 2 + y2 = a2 sin2 0, z = o.
suitable form, in which r is one of the coordinates;
The boundary of this disk [where r = 0, 0 = !7r
he applies to Eq. (2.4) the transformation
and therefore l:(r, 0) = 0] is of particular importance,
x = (r2 +
a 2)! sin 0 cos [cp - tan- 1 (air)], since it is precisely this set of points at which the
metric (2.7) becomes singular.
y = (r2 + a 2)! sin 0 sin [cp - tan- 1 (air)], (2.6)
The surfaces cp = const have the appearance of
z = r cos 8, bent planes, which are approximately vertical for
large r but flatten out and become horizontal at
and obtains
the edge of the disk (Fig. 2). Letting cp be fixed but
d,s2 = dr + 2a sin 0 dr dcp + (r2 + a ) sin 0 dcp2
2 2 2 2
arbitrary, set
+ l: d02 - dt 2 + (2mr/l:)(dr + a sin2 0 dcp + dt)2. ~ = x cos cp + y sin cp, 1/ = -xsincp + y coscp.
(2.7) Equation (2.6) then yields
Here
~ = r sin 0, 1/ = -a sin 0, z = r cos 8,
(2.8)
or equivalently
6
This is a generalization of the Eddington form of
(2.10)
the Schwarzschild metric [cf. Eq. (1.2)]; accordingly,
we refer to the coordinate system (r, 0, cp, t) as the This defines a ruled quartic surface, whose genera-
(E) frame. Note that in the asymptotic region, r ~ tors are given by 1/ = -a sin 0, z = ~ cot 8, with 0
co, Eq. (2.7) reduces to Eq. (1.2), which justifies the held constant on a given generator. (Hence each
interpretation of the parameter m as the total mass line is also a generator of the corresponding hyper-
of the source. (We therefore assume m > 0.) By boloid 0 = const.) Since ~ = x, 1/ = Y for cp = 0, we
comparing higher-order terms with the weak-field have in fact constructed the surface cp = O. And
metric for a rotating body as given by Landau- because of the way (~, 1/) are related to (x, y), it is
Lifshitz,19 or by calculating the corrections to the obvious that all the other cp = const surfaces are
perihelion precession formula and comparing with
the Lense-Thirring result,20 one can show that the
parameter a is just the angular momentum per unit
mass of the source. We see later that the Kerr
solution radically changes its character for lal > m.
(r, 6, <g) Coordinate Surfaces
To explore the properties of the principal null

congruences in greater detail, we find it convenient
to study the relation between the (E) and (M)
coordinates, by viewing the surfaces r, 0, cp = const
in a Euclidean 3-space whose Cartesian coordinates
are (x, y, z). From Eq. (2.5) it is clear that the sur-
faces r = const are confocal ellipsoids, while from
19 L. D. Landau and E. M. Lifshitz, Classical Theory of
Fields (Addison-Wesley Publishing Company, Inc., Reading,
Mass., 1962), 2nd ed., p. 359. FIG. 2. The ruled quartic surface '" = const, shown im-
20 R. H. Boyer and T. G. Price, Proc. Cambridge Phil. Soc. bedded in a Euclidean 3-space with Cartesian coordinates
61, 531 (1965). (~, 71, z).
merely replicas of this one, and can be obtained by coordinates, since they reduce to the standard
rotating it about the z axis. Schwarzschild coordinates when a = O.
Throughout the foregoing we have tacitly assumed The metric (2.13) has, in addition to the true
that r is positive. However, the disk r = 0, which singularity at ~ = 0, a pair of pseudosingularities
has been viewed as a membrane of discontinuity, at the real zeros of d(r). The latter are located at
may just as well be considered a two-sided aperture
to a second sheet on which r is negative. Such a (1.6)
2 2
continuation to negative r values is permissible and thus exist only when a ::::; m • The close simi-
because the metric (2.7) remains regular at r = 0 larity of this equation with the one arising in the
(provided 0 ¥- t1l"). We hasten to remark that this Reissner-Nordstrom problem has already been re-
second sheet is not to be confused with the other marked, as has the identification of the surfaces
side of an Einstein-Rosen bridge,IO which one en- r = r", with stationary null surfaces, or horizons.
counters in the familiar Kruskal procedure. We Let I = 0 be any null hypersurface containing the
construct an analog of the latter in Sec. III; it has Killing vectors aI alP and aI at; then
propert;ies quite different from the r < 0 extension
contemplated here. For one thing, the behavior of
I = I(r, 8), (grad 1)2 = 0,
the metric as r _ - ex> , with m > 0 by definition and from the contravariant form of the metric
and therefore mlr negative, describes the sort of [given in Eq. (2.15)] we deduce
geometry one would expect far from a particle of
negative mass. On the other hand, the two sides of
d(r)(aljar)2 + (aljaO)2 = O. (2.14)
an Einstein-Rosen bridge are isometric, so both de- The only solutions of this equation periodic in 0
scribe the field outside a positive-mass body. are the ellipsoids r = r",. The cases a = m and a = 0
In the Kerr-8child theory the most general metric (Schwarzschild) are seen to be exceptional: In the
is determined by an analytic function of one com- former case the two horizons coalesce, while in the
plex variable. For the special case of the Kerr solu- latter case r_ = 0, which is not a null surface at all
tion, this function has a branch point at the ring but an essential singularity.
singularity r = 0, 0 = t1l". Thus one can properly Because of these additional spurious singularities
view the continuation of (2.7) to negative r values the (8) coordinates are an inappropriate tool to
as an analytic continuation onto the second Riemann use in studying the analytic properties of the vacuum
surface of this function. By passing twice in the metric; evidently in any such investigation the (E)
same direction through the ring one returns to the coordinates are far superior. 21 but the (8) coordi-
original Riemann sheet, and thus to a manifold nates have one important advantage which we ex-
isometric to the original one, which may for sim- ploit in later sections, namely, they treat both of
plicity be identified with it. the principal null congruences on an equal footing.
"Schwarzschild-Like" Coordinates Principal Null Congruences
Making the transformation The inverse (contravariant) form of Eq. (2.7) IS
f = r, jj = 0,
(grad)2 = ~-{ (r2 + a2)(:J2 - 2a(:r)(:)
dip = dIP + a drl d(r) , (2.11)
dl = dt - 2mr drl d(r) ,
+ S~2 8 (:IPY + (:oYJ
where
(2.12) - (:J - (2;r)(:r - :J, (2.15)
we find that which is again of the form: fiat-space metric plus

the square of a null vector [cf. Eq. (2.1b)]. Thus
ds 2 = I + d( 2) + (r2 + a2) sin2 8 dip2 - dl2
~(dr2 d
the contravariant components of the null congruence
+ (2mr/~)(a sin 2
8 dip + dZY. (2.13) k can be read off immediately:
B
This form has only one off-diagonal component, k == (k', k , k", k') = (-1,0,0,1). (2.16a)
g ~1I and is thus invariant under the transformation
21 This feature of the original Eddington coordinates was
ip - -ip, l - -l. We refer to (r, 0, ip, l) as (8) first clearly recognized by D. Finkelstein. s
MAXIMAL ANALYTIC EXTENSION 0:' THE KERR K~T1itIC 171
With this normalization Eq. (2.280) is satisfied, and

consequently t (or r) is an affine parameter along the
congruence. Having chosen k' > 0, we can say that
k is future-pointing, and since k' < 0, we conclude
tha.t the k congruence is 'tngoing.
Next reca:ll that k must appear as a linear con-
gruence when expressed in (M) coordinates. Since
k' = k" = 0, such lines must lie in the surfaces
8, rp = const, and are therefore the common gen-
erators of these surfaces. As t increases, the rays
proceed inwards toward the disk r = 0, cross it,
°
then emerge onto the r < sheet keeping the same
, and rp values. The rays lying in the equatorial FIG. 3. The horizon r = r +, viewed from the (M) frame with
pla.ne 8 = !1r are exceptions. These are tangent to (x, 1/, z) as Cartesian coordinates. The principal null con-
the ring singularity at }; = 0, meeting it after a gruence 1 lying in the horizon is shown, as well as portions of
the k congruence lying in the coordinate surface fI = ,,/4
finite lapse of affine parameter r.22 and", = const.
It is not easy to detect the remaining principal
null vector 1 which is hidden in (2.7) or (2.15). the corresponding horizons (see Fig. 3). The affine
Equation (2.13) offers a clue; it is invariant under parameter for this special case is not obvious-it
the transformation df -+ -df which interchanges evidently cannot be r, and it turns out that t does
ingoing and outgoing rays. In system (8), therefore, not work either-but we postpone this question
k and 1 differ only in the sign of their radial com- until Sec. IV [see in particular Eq. (4.11)].
ponents. Applying the transformation (2.11) to k This result appears at first sight to violate the
gives us the following: symmetry between the two principal null con-
gruences. However, from Eq. (2.3b), there must
Jr: = (k', k i , k', kl)
exist another coordinate frame-(E') say-adapted
= [-1,0, -a/A, (r2 + a 2)/A] (2.16b) to the 1 congruence, in which the roles of k and 1 are
interchanged. This frame is not hard to find. Using
a.nd consequently
Eq. (2.11) as a guide, we make the obvious trans-
1 "" (l' , z, , l' , ll) formation
= [+1,0, -a/A, (r2 + a 2)/A]. (2.17b) f = r', ii = 8',
It is ihen a simple matter to determine the com- dip = drp' - a dr' /A, (2.19)
ponents of 1 in the (E) frame:
dl = dt' + 2mr' dr' / A
1 = (l', Zl, Z", l')
and obtain a line element just like (2.7), except that
2
= [+1,0, -2a/A, (r2 + a + 2mr)/A]. (2.1780) the sign of dr (or equivalently, of drp a.nd dt) is
reversed:
Note that f = r is an affine parameter along both
ray systems, but that t is an affine parameter only ds2 = dra - 2a sin2 6 dr drp'
for the ingoing congruence.
Equation (2.1780) fails at the horizons r = r,.,
+ (r2 + a2) sin2 e drp,2 + l: dif - dt,2
but if we rewrite it as + (2mr/};)(dr - 2
a sin 8 drp' - dt')'. (2.20)
1 = N(A, 0, -2a, r2 + a + 2mr)
2
A further transformation, from (E') to the associated
Minkowski (M') frame (x', y', z', t'), proceeds es-
with N an unspecified normalization factor, we can
sentially via Eqs. (2.6), except that rp - tan-l(a/r)
take the limit r -+ r., without difficulty. The result,
is replaced by rp' + tan-l(a/r), One finds
1., = N .,(0,0, -2a, 4mr ,), (2.18) k = (k r ', k~', k'i", k'')
displays the key role of l± as the null generators of = [-1, 0, -2a/ A, (r2 + a2 + 2mr)/ A], (2.16c)
12 This shows, incidentally, that the ring singularity is real,
for it is easily seen that d'lH /dp.2, which is an algebraic in- 1 = (r', lB', l'P', I'') = (+1,0,0, +1), (2.17e)
variant of the Riemann tensor [see Eq. (2. 2a»), becomes
infinite there. which is just the reverse of (2.16a), (2.1780), as
lent analytic extensions of the original (E) frame,

according ag we choose r to lie in the different in-
tervals - 00 < r < r _, r _ < r < r +, r + < r < 00.
Putting these several pieces together to make a
smooth manifold is the job of the next section.
r= ...... m. ANALYTIC EXTENSIONS
"Kruskal-Like" Coordinates
Let us mark the three domains r > r +, r _ <
~ (E')patch
r < r+, r < r_ within the (E) patch ag regions
to, 1, 2}, respectively. Transforming from (E) to
(E') via Eq. (2.21) we obtain three similar domains,
FIG. 4. The pair of (E) and (E') coordinate patches for the
to', I', 2'1. We assume that the transformation is
Kerr metric, drawn in a fashion roughly analogous to Fig. 1. one-to-one (and therefore analytic) between {Ol
Even-numbered regions, which are asymptotically fiat, are and {O'l; it then follows that {II and {I'} are
separated by an odd-numbered interior region whose bound-
aries are the two horizons r±. Although the drawing is only inequivalent-one has a boundary crossed by the
me~nt t;o be sy~bolic, the compression of the asymptotic k congruence, the other by the 1 congruence--and
regIOns IS suggestive of a Penrose conformal transformation
[R. Penr,?se, Relativity, Groups and Topology, B. DeWitt and likewise for {21 and {2'1. Putting the two patches
C. DeWItt, Eds. (Gordon and Breach Science Publishers, together, we obtain the enlarged domain shown
New York, 1964)].
schematically in Fig. 4.
is to be expected. Note that in this frame it is the Now take anotper (E') patch, whose sections are
k congruence which provides the generators for the labeled {O*, 1*, 2*}, and tie it into the middle region
horizons. of the original (E) patch, by requiring that the
This result is not paradoxical, for the two co- domains {II and {i *} coincide. The manifold has
ordinate frames are not equivalent. Combining Eqs. now grown to the size of Fig. 5.
(2.11) and (2.19), one gets the direct transformation We have drawn the two regions {O*} and {I'} as
equations (E) - t (E'). For a2 > m 2 this trans- contiguous, but this is unfair, for there is no reagon
formation is one-to-one over the entire range of in the foregoing to suppose that they are in any way
2 2
r values, but for a :::; m , which is the interesting related. We now wish to demonstrate that the picture
cage, the (E) and (E') coordinate patches only is in fact a reasonable one, by displaying a single
partially overlap. It is a straightforward matter coordinate system in which the four regions
to show that to, 1, 0*, I'} are linked together as shown. Such
a linkage looks very similar to a Kruskal diagram
t' = (Fig. 1). Our object, therefore, is to find a trans-
formation analogous to (1.4), such that both seg-
ments of the horizon r = r+ appear regular when
(2.21) viewed from these new coordinates.
and The tagk is not a trivial one. For one thing, the
principal null congruences of the Schwarzschild
<p
,
= <p + (2
a
m - a2)t
jr - r+j
In - - ,
r - r_
metric are straightened out in Kruskal coordinates
to the lines u ± v = const, 8, ip = const. One might
from which it is clear that either (t, <p) or (t', <p') hope, by combining a transformation of the azi-
must diverge at r±. We thus obtain three inequiva- muthal angle with a suitable generalization of Eq.
(1.4), to achieve a similar simplification here. But
the corresponding vectors k and 1 of the Kerr metric
are not in general 2-surface-forming, so that such
FIG. 5. The diagram of Fig. 4, a search is pointless. We settle for a frame in which
enlarged by the addition of the principal null rays merely lie in the surfaces
another (E') patch labeled
10*,1*, 2*}. The shaded portion u ± v = const. This solution, while less elegant
defines the Kruskal or (K) than Kruskal's, is adequate for our purposes, and
patch.
is perhaps as close ag one can come to a generaliza-
tion of it.
Since we want to treat both congruences impar-
M A X I MAL A N A L Y TIC EXT ENS ION 0 F THE K ERR MET RIC 273
tially, we start from the (8) frame. From the expres- Although this is only defined for r > r+. l finite
sions for k and 1 in these coordinates [Eqs. (2.16b), (i.e., for u > \vl), the inverse transformation,
(2.17b»), it may be seen that as r ---t r+ both become given implicitly by
asymptotic to a family of helices winding around
the horizon, given by .T,( )
't!r == (r---
2m
r+)(r-- -
2m
r-)-'er/a+ =u-v
2 2
(3.5)
dr: dO: diP: dl = 0 : 0: -a: 2mr+. and
We seek a transformation of the azimuthal co- l = 0"+ tanh- 1 [2uv/(u2 +v 2
)]
ordillate which will straighten out the helices. Many

is well defined over the cut plane u2 ;6 v2 • In partic-
obvious possibilities suggest themselves, but one
ular, 'It(r) increases monotonically from - 00 to 0
encounters later difficulties unless one picks the
to + 00 as r runs from r _ to r + to + 00, and is an
new coordinate to be a linear combination of Cp and
analytic function of r over this interval, so that
l with constant coefficients, plus any function of r
r( u, v) is an analytic function over the u, v plane,
and 0, and this restricts the choice to
while leu, v) diverges at u = ±v and is analytic
w = Cp + (a/2mr+)l + "fer, 0). (3.1) elsewhere. As the coordinates (u, v, w, O)-which
For simplicity we take "fer, 0) = O. It is quite pos- we henceforth call the (K) frame-vary over their
sible that other choices might lead to a more trac- respective ranges
table form for the metric; we have not investigated -00 <u < 00, -00 <v < 00,
this in any detail.
Through Eq. (3.1) we have untwisted the null o ::; w < 211",
curves in the limit r ---t r+ (at the expense, inciden- they cover the four regions to, 1, l' and O*} of
tally, of twisting them in the neighborhood of Fig. 5; the boundaries between adjacent regions
spatial infinity, which may be disagreeable but is are the pair of lines u = ±v on which r = r+. Note
not serious-Kruskal-type coordinates are not at that region {O*}, in which u < \v\ (and there-
all well adapted to the asymptotically flat portions fore r > r+), forms the second sheet of an Einstein-
of the manifold anyway.) The metric computed Rosen bridge.
from (2.13) after this single change of variables is In the following it is convenient to know the
still singular at r+, of course, but now we know how direct transformation from (E) to (K) coordinates.
to deal with it. We make a further transformation This is given by
(r, t) ---t (u, v), such that the integral curves of the
two principal congruences lie in the hypersurfaces
u ± v = const, although not in w = const. Ac-
cording to Eqs. (2.16b), (2.17b), these integral curves (3.6)
satisfy a) a (r - r_)
(2mr+
w=cp+ - - t--ln - - .
r+ 2m
or The transformation from (E') to (K) is obtained
similarly, with t replaced by -t'. Equation (3.6)
(r ;mr+)(r ~r-r' exp (r ~ l) evidently defines an analytic homeomorphism of
the domain r > r_, - 00 < t < 00 onto the half-
= const = F(u ± v), (3.2) plane u + v > 0, i.e., onto the region to, I}.
with On the other hand, we could have mapped the
(3.3) domain - 00 < r < r +, - 00 < t < 00 homeomor-
phically into a (different) Kruskal patch by regu-
and
larizing about the other event horizon, r = r _. 28
p == r_/r+o The latter transformation, (E) ---t (K') say, takes
From the work of Kruskal and Graves-Brill we are the form
led to take23 F(x) = x 2 , and consequently
u ± v = (r - r+)!(r - r_)-,/2 exp (r ± l). (3.4)
2m 2m 20"+
sa Had we wanted the transformation to yield a metric
regular across r_, the appropriate choice would have been
F(x) = X-b.
We postpone a discussion of the full import of this would have been desirable to have a coordinate
result until later in the section, in order to settle a system in which, like Kruskal's, the null curves took
fundamental point. an especially simple form (see Fig. 6). Cohen and
Brill/' in their study of metrics for slowly rotating
Analyticity of the Metric in (K) Coordinates
bodies, have applied the Kruskal procedure to a
There remains the very important but rather line element differing from the Schwarz schild form
tedious job of working out the components of the by terms of first order in the angular velocity. We
Kerr metric in (K) coordinates, and demonstrating recover their results if we drop from Eq. (3.8) all
that they are indeed analytic functions of u and v terms of order a 2 j this leaves
throughout the (u, v) plane. We spare you the de-
di ~ r2(d{l2 + sin 2
(I dw 2)
tails, and merely assert that after much labor one
can cast the metric in the following form: + (32m3/r)e-r/2m(du2 - dv2)
2
ds = 1: dfi + 41T;f1:(r2 + a2)-2(du 2
dv2)
- + (4a/r) sin {Ie- r/2m
2
+ 1:- 1 sin 2 (I[(r2 + a2) dw + a(m2 - a2)-if X (r2 + 2mr + 4m2) dw (v du - u dv). (3.10)
X (r + r +)(r - r _)-I(V du - U dV)]2 To this order, the curves u ± v = const, w, {I = const
are null, just as in the Kruskal form of the Sch-
+ (1:.::\)-11[21T+f1:(r2 + a 2)-I(v du - u dv)Y
warzschild metric.
- [21T+f1:+(r! + a2)-\v du - u dv)
Construction of a Maximal Analytic Manifold
2
- a sin 0.::\ dwY} . (3.8)
What we have done to extend the doma.in {O, 1)
Here we have introduced we can also do to the domain {I, 2). We use the
alternative transformation (3.7) to produce a new
(3.9a) (K') patch, consisting of the three regions {I, 2, 2* I
and already introduced plus a new domain {3), which,
like {I}, is bounded by two pairs of horizons (see
fer) = .::\(r)/if(r) Fig. 7). Here also the asymptotically flat portions,
= 4m2 [(r - r _)/2m]2m l r+e-rlo+. {2) and {2*), form two sheets of an Einstein-Rosen
(3.9b)
bridge. While isometric to each other, these are not
Since 1: and f are both analytic and nonzero through- isometric to {OJ and {O*}-they have, it may be
out the whole (u, v) plane, it is clear that only the recalled, the sort of geometry one would associate
term in (3.8) with .::\(r) in the denominator can lead with a negative mass source.
to difficulty. Closer inspection shows, however, that A similar extension can be applied to the domain
the numerator of this term has a simple zero at {I', 2'). In view of the picture which is rapidly
r = r + also, so the quotient remains analytic. emerging, we find it preferable to change our nota-
Had we chosen to regularize across r _ instead, tion slightly, and to write {-I, -2} instead of
using Eq. (3.7) or its equivalent, we would have {1', 2'). Thus the new Kruskal patch encompasses
obtained much the same result, except for the in- the regions {-I, -2, -2*, -3) of Fig. 7.
evitable branch point at r(u, v) = 0 and {I = !11". This procedure can clearly be continued indefi-
The line element (3.8) is much too cumbersome nitely in both directions, and generates an infinite
to be of any direct use. Fortunately, once we have chain of overlapping (E) and (E') patches. The
established its analyticity we need make no further result can be viewed even more simply as a chain
use of it. This is to some extent regrettable, for it of overlapping "Kruskal" patches, {K 2ft } , isometric
to (K) or (K') according as n is even or odd. We
FIG. 6. The null cone at a point can, in fact, eliminate any reference whatever to
v in the horizon r = r+. The cone is
lopsided and spills out of the angle the auxiliary (E) and (E') frames, by presenting
defined by the principal null vectors the transformation equations that connect (K) and
k and 1; in addition, the shaded (K'); these take the surprisingly simple form
portion of the cone behind the
(u, v) plane is smaller than the por-
tion jutting out in front of it. (u + vy+( -u' - v'Y- = 1,
(3.11a)
Although the shape of the cone
7'------ u
varies from point to point, the pro-
jections of k and 1 onto the plane
(-u + vy+(u' - v'Y- = 1,
always have slope ±1. 24 D. R. Brill and J. M. Cohen, Phys. Rev. 143, 1011 (196t1).
and ~
w' - w = (2a/r _) In [(u + v)/( -u + v)]
= - (2a/r +) In [Cu' - v' )/( -u ' - v')]. (3.lIb)
These transformations are one-to-one and analytic
in the domains '" {11, {31, ... of overlap, and FIG. 7. A chain of
Kruskal patches, with
have the following simple properties: They preserve those of (K) and (K')
straight lines of slope ±I as well as the straight type alternating and
partially overlapping.
lines v/u = const and the hyperbolas v2 - u 2 = The complete figure
const, while mapping the north quadrant of (K) extends indefinitely in
both directions, to form
onto the south quadrant of (K' ). a maximal analytic ex-
The manifold we have constructed has many tension of the Kerr
manifold.
curious properties. It has infinitely many disjoint
and asymptotically fiat sheets, for one thing, and
therefore does not admit a Cauchy surface. In fact,
(K) patch
one can get from any such even-numbered sheet to
§ (K1 patches
any other-with the sole exception of the companion
sheet on the other side of the "bridge"-by following
a suitable timelike or lightlike curve. The timelike
curve u = O(u' = 0), w, 0 = const is particularly
noteworthy, for it defines the location of the "throat" zons, in the past light cone of the source which
of an Einstein-Rosen bridge. The throat itself (i.e., emitted it-and this may be felt to be a bit
the 2-surface u = 0, v = const) has an area which unrealistic. 26
pulsates with time, in close analogy with the The foregoing construction presupposes, of course,
Reissner-Nordstrom case. 12 Since the throat is de- that the metric admits two distinct horizons. For
formed, due to the effects of rotation, its area is a2 > m2 the horizons disappear and the transforma-
no longer given by the simple formula 471"r2 obtained tion to Kruskal coordinates loses all meaning; all
for the Schwarzschild and Reissner-Nordstrom met- that remains are two asymptotically fiat spaces
rics, but by the more complicated expression joined at the disk r = O. The exceptional case lal =
m deserves special comment. Carter has studied
area = 271" {(r2 + a2) this in detaill6 ; although his work was confined to
+ (g/a)(-~)-tsin-I [a(-~/g)!]) (3.12) the symmetry axis (0 = 0, 71"), it is clear that his
conclusions apply with equal force to the full metric:
with g(r) = (r2 + a2)2 - a2~. At the horizons r = r ±, One builds up a ladder of alternating (E) and (E')
where the extrema occur, this simplifies to patches (with or without identifications) in the
area± = 871"mr ±. (3.12) simple sequence { ... - 2, 0, 2, ... I. The odd-
numbered sheets disappear; there is nothing to
The manifold sketched in Fig. 7 is not the only correspond to the second Kruskal sheets { ... - 2*,
possible maximal extension, because of the freedom 0*, 2*, ... I; and in fact there is no need at all to
still remaining to make topological identifications. introduce (K)-type coordinates to cover the mani-
The simplest manifold of this more general type fold in this case.
arises if we identify {K2n I with {K 2"H I. It is covered
IV. GEODESIC COMPLETENESS
by only two Kruskal patches, {Ko I and {K2 1 say,
with regions {-II and {31 identified, so it consists Geodesic Equations, First Integrals
of four asymptotically fiat sheets (two with positive For the study of geodesics the original (E) co-
mass, two negative) glued together with two interior ordinates are particularly convenient. Let p. be an
regions. 26 This manifold, like the others obtained affine path parameter, normalized to give proper
in the same way, is violently acausal-a properly time along timelike geodesics, and use a dot to
aimed signal will emerge, after crossing four hori-
2S B. Carter has pointed out that for small negative values
16 This particular choice was made by Graves and Brill.11 of r and values of 8 near the equator g'l''P goes negative. Hence
It is clear from their work, however, that the identification is as long as one adopts a top'ology in which rp is treated as an
unnecessary; one can also describe the Reissner-Nordstrom angular coordinate there will necessarily exist closed timelike
metric by an infinite chain of Kruskal patches. curves in this region.
denote differentiation with respect to p.. The geodesic '1:. 2t 2 - (al + 2mry)2
equations can be extracted in the usual way from
= d[ -e'1:. + ('1:. + 2mrh2 p~ - (l2/sin2 fJ)].
the variational principle 8f L dp. = 0, with a
Lagrange function given by (4.5)
2L = t 2 + 2a sin 2 Of.p Problem of Completeness
+ (r2 + a sin fJ.p2 + '1:.(i _ j'
2
)
2
The Kerr metric. evidently cannot be imbedded
isometrically in a complete analytic manifold, since
+ (2mr/'1:.)(t + a sin fJ.p + i)2
2
(4.1) there exist geodesics-the principal null rays lying
as in Eq. (2.7). in the equatorial plane, for example-which strike
We are chiefly interested not in the equations of the ring singularity at finite values of their affine
motion themselves, but rather in their first integrals. parameters. If one excludes all such geodesics, it
Since tp and t are cyclic, we obtain two integrals is reasonable to ask whether the remainder can be
immediately: continued to arbitrarily large values of p.. We argue
below that this is so, provided that the manifold
P~ == aL/a.p = I,
(4.2a) is chosen to be the analytic extension described
P, == aL/al = -/" previously. In this sense, therefore, the extension
can be regarded as maximal.
with land 'Y representing the angular momentum/ Since it is impossible to solve the geodesic equa-
mass and energy/mass of a test particle moving tions exactly, except in a few very special cases, we
along the given geodesic. (On null geodesics p. is base the argument on the first integrals (4.4) and
defined only up to a linear transformation, and con- (4.5). Starting with any set of initial values, one
sequently only the ratio III' is meaningful. We extends the solution either until a singular point
remove the arbitrariness by taking 'Y = 1 in this is reached, or until one or more coordinates diverge.
case.) A third integral is given by L itself: There is clearly no problem if r diverges-since the
2L = -e, (4.2b) metric is asymptotically flat and therefore complete
for r ~ ± <Xl-but only if.p, I, or 0 diverge at finite
the indicator e being +1, 0, -1 for timelike, null, values of r. As one sees from the above equations,
and spacelike geodesics, respectively. this happens at '1:. = 0, sin (J = 0 or Ll = O. If the
The remaining two conjugate momenta are readily first possibility occurs nothing can be done about
found to be it; the second is obviously a consequence of the
Pr = ('1:.t + al + 2mry)/ d,
spheroidal-type coordinates here employed, and can
(4.3) be eliminated by transforming to those of the (M)
pe = '1:.0, type. Thus it is only the apparent divergence of
.p and t at the horizons that needs to be examined
which shows, incidentally, that we can expect dif-
carefully. [It follows from Eq. (4.5) that t remains
ficulties with the equations of motion at the ring
finite at r ,o; the same is true of 0.] In fact, .p and i
singularity ('1:. = 0) and at the horizons (d = 0).
diverge if and only if Pr diverges, and it is easy
It is clear from the form of L that Po = 0 is consistent
to see that this happens if and only if r", has the
with the equations of motion if (J = 0 or 'II" (axial
same sign as al + 2mr ",/" for Eqs. (4.3) and (4.5)
case) or if (J = !'II" (equatorial case); in these special
imply
cases the first integrals (4.2) yield a complete solu-
tion by quadratures, which we consider in greater -e'1:. + ('1:. + 2mrh2 - p; - (l2/sin 2 fJ)
detail in Sec. V. P. = -
'1:.t - (al + 2mry)
From Eq. (4.2a) we can express .p and i in terms (4.6)
of r and the constants of integration; the resulting
which yields a finite limit whenever t", and
formulas are
(al + 2mr,o/') have opposite sign-that is, whenever
.p = '1:.- 1 [( l/sin' fJ) - apr], the geodesic is ingoing. [Some confusion over the
(4.4)
meaning of the term" ingoing" can arise here, unless
i = '1:.- 1 h('1:. + 2mr) + 2mrpr], one is careful. On the positive r sheet the hypersur-
with pr given by (4.3). faces t = const are everywhere spacelike: (grad t)2 =
We note also the following form of Eq. (4.2b), -1 - 2mr/'1:., which is certainly negative for r > O.
which is important in later arguments: Hence an ingoing path is properly defined as one
MAXIMAL AN ALYTIC EXTENSION OF THE KERR METRIC 277
for which dr/dt < O. For some values of l and "I Eqs. The first integrals (4.5), (4.6) then yield the com-
(4.3), (4.4) suggest that dr/dJL is positive on an mon solution
ingoing path, but this simply means that the affine
(4.9)
parameter has been chosen to increase into the past.]
We therefore conclude that all ingoing paths can in agreement with Eq. (2.18). To find t(JL) it is
be continued across the horizons. Some subsequently necessary to solve one of the geodesic equations,
strike the ring singularity, others continue on to and the simplest to use is Pr = aL/ar. A straight-
r = - (Xl (and are thus complete), a third class forward computation gives
reaches a turning point and start back. On the
return trip, however, both t and cP diverge as the
't' = =r(21T±)-lt2 ,
horizon is approached. This is not surprising; the with IT ± defined by Eq. (3.3). Hence
particle has merely run off the (E) coordinate patch.
To continue to follow its motion transform to an t(JL) = ±21T ± In (JL - JLo) (4.lOa)
(E') system, or equivalently to an appropriate set or
of (K) coordinates. It is clear that a result similar
± t/2cr±
to Eq. (4.6) must also apply to the outgoing tra- JL-JLo=e . (4. lOb)
jectory in this case; hence the particle re-emerges We emphasize that this result is exact.
into an asymptotically flat" universe," but of course On the event horizon r = r +, JL - t + (Xl as
on a sheet different from the first. If the effective t - t + (Xl, while JL - t JLo as t - t - (Xl. This is just
total energy r == "12 - E is positive, the particle what one would have expected, since the (E) co-
escapes to infinity; if negative, it reaches another ordinates are known to be incomplete at r + for
turning point and starts back toward r = r + again, t - t - (Xl. A full picture of the horizon r = r+ is
in which case we transform to another set of (E) provided by viewing it in (K) coordinates; then the
coordinates and repeat the cycle. Clearly JL can be path equations become
made as large as we please by piecing together
sufficiently many such cycles, which proves com- W(JL) = const
pleteness for this case. (4.11)
Of course there are many geodesics which do
not fall into any of the above categories, such as and are evidently complete. A similar result applies
the ones which oscillate to and fro between a maxi- at r = r _, but with the time directions reversed;
mum and minimum radius outside r +, and those this too is consistent with the complete picture of
which spiral in towards (or out from) an unstable this horizon when viewed from (K') coordinates
circular orbit. However, these lead in general to no (as in Fig. 7).
difficulties with completeness, since t remains finite Qualitatively similar results emerge when one
as t increases to infinity. analyzes the geodesics that approach r ± asymp-
There remains one further class to be considered, totically (Le., as t - t ± (Xl). All timelike curves of
namely, the geodesics for which r = 0 at r ±, or this type reach a vertex such as u = v = 0 in a
equivalently, for which finite proper time and can be extended without dif-
ficulty; similar remarks apply to the null curves.
(4.7) There are others, necessarily spacelike, which ap-
proach r + from region {I} as t - t + (Xl (or r _ as
This class includes spacelike geodesics tangent to
the horizons (these present no problems), geodesics
t - t - (Xl), and these are in fact complete.
of all three types which approach the horizon asymp- V. Equatorial Geodesics
totically as t - t ± (Xl, and finally, and most impor-
tantly, the principal null geodesics that are the The first integrals obtained in the previous section
generators of the horizons. We study the latter in yield a complete description when 0 = 0 or 11' or
detail below. when 0 = !11'. The former case, which corresponds
to motion along the symmetry axis, has been in-
Completeness within the Horizons vestigated by Carter16 ; we confine our attention,
20
To obtain the solution for the principal null ray therefore, to the latter. Boyer and Price have
1 lying in r± set ~ = 0, E = 0 and also shown that the orbit equation for equatorial geodes-
ics leads to a precession of the pericenter in agree-
(j = 0, "I = l = O. (4.8) ment, through third order, with the approximate
calculations of Lensa-Thirring. In the present sec- tudes of a and A. If A2 < a 2 there are clearly no
tion we concentrate on the qualitative features of real positive roots, so collapse to the singularity
geodesics in the strong field regions. Our analysis inevitably occurs. Conversely, if A2 is sufficiently
is patterned after the study of orbits in the Schwarz- large, incoming rays reach a pericenter &nd return
schild metric first carried out by Darwin27 and later to infinity, while outgoing rays from r = 0 reach an
extended by Mielnik and Pleba:6.ski. 28 Of course, all apocenter and fall back in. Thus one expects that
these studies are largely academic exercises, since there should exist two critical impact p8.l"ameters,
in realistic situations (except possibly the late phases Al > 0 and A2 < 0 say, such that light signals spiral
of gravitational collapseS) the geometry in these in to r = 0 if A2 < A < AI, and "bounce" back out to
regions, and hence the geodesics themselves, differ infinity otherwise. At these critical values of A, if;(p)
considerably from the empty-space results, due to has a double zero, which defines the corresponding
the nonzero stress-energy tensor there. Nevertheless, critical radii PI, P2. A light ray at such a radius, and
the study of the Kerr metric, even as an ideal case, with the correct value for A, will travel around in a
has, we believe, some real value, for it helps to circular orbit indefinitely, but such an orbit is, of
clarify the role which angular momentum plays in course, unstable.
general relativistic models. In particular, because These predictions are borne out in the Schwarz-
the character of the orbits in the interior regions schild case by Darwin's analysis. There, it may be
changes markedly as soon as the central body is recalled, the critical radii are both located at p = i
given some angular momentum, it seems worth- (or r = 3m), and the critical impact parameters are
while to point out those features of Darwin's analysis A = ±3(!)t. Most importantly, there do not exist
which are unique to the Schwarzschild problem, and any light rays whose pericenters lie inside r = 3m.
those which persist in the a ¢ 0 case as well. If one imagines the parameter a (or a) being in-
creased gradually from zero, one expects PI (a) and
Null Geodesics p2(a) to depart smoothly from the Darwin value,
It is simplest, and most instructive, to begin with and this is precisely what happens. In fact, one can
the null rays. We normalize the affine parameter give fairly simple closed expressions for the critical
along the rays by taking 'Y = 1. Then the energy radii. These must be double zeros of if;(p) &nd thus
integral (4.5) reduces to must satisfy the condition
(5.1) pep - !)2 _ 2a = O.
For convenience set (3 = 2m and introduce the Solving this by standard methods, one finds%t
dimensionless variables
PI = ! + asec (t COS-I 2a), (S.4a.)
P = r/{3, A = l/{3, a = a/{3. (5.2) pz = ! + a sec (t COS-I 2a + iT),
From (5.1) it is clear that turning points occur
at the zeros of the cubic polynomial
provided that a :s: !; there is only one critical r&dius,
at
(5.3)
PI = ! + a sech (t cosh- 1
2a), (S.4b)
The location of these zeros is thus fundamental
to a qualitative understanding of the null tra- if a > t. (Note that PI is always larger than P2'
jectories. This is to be expected, since the centrifugal barrier
Applying the rule of signs, we see that if;(p) = 0 is stronger if A is positive.)
has always one real negative root: a ray sent in from However, when a ¢ 0 a completely novel feature
r = - is thus repelled and ultimately deflected
<X)
emerges: for a small range of impact parameters,
back to - <x). (The outgoing principal null ray, with As :s: A < - a, pericenters exist inside the inner
A = - a, is an exceptional case; it alone strikes the horizons, and, in fact, for all values of P between 0
r = 0 singularity from this direction.) On the positive and p_. This remarkable property permits one to
sheet there are consequently either zero or two transmit information from one positive sheet to
turning points, depending on the relative magni- another by bouncing a light signal off a centrifugal
barrier inside p_.
17 C. Darwin, Proc. Roy. Soc. (London) A249, 180 (1959);
A263 39 (1961). J~ A third zero, PI = t + a sec(! C08-1 2a - f,..), has a
18 B. MIelnik and J. Plebatiski, Acta Phys. Polon. 21, 239 somewhat different interpretation: it describes the maximum
(1962). possible apocenter within the inner horizon.
This reeult is illustrated by Fig. 8, which shows r

I Ir
2
the portion of the X-p plane in which light tra- ,
I
I
I
t
!1 I I'
jectories are possible. The boundary curve separating I
I
!I : :1
allowed and forbidden regions is obtained from Eq. I
,
I
1\ ! !i
(5.3) with if;(p) = 0; it is given in explicit form as ,
I
I
:
' \
1 I
I
i il
I,
I I \ I II
X = (1 - p)-I{ -a ± prep - p+)(p - p_)]i}. (5.5) I
I
I ' \ \\
I
I
f \. ~ \ 1
Not surprisingly, the shape of this curve is qualita- I

O+----7.0;;-1--'--"""+'................'+1"'r-.....................-'-L""'t-...:.:-~'~::!+,P.,.
\ \
tively different in the two cases a < ! (where " . \ 1 ':::;:-:- '
horizons exist) and a > ! (where they do not).

\ \ ,/'
\ I \ / a.o~
In the latter case pericenters occur in the region '. J \., ,/ .3---
v '---, .6--
p > PI for X positive [with PI given by Eq. (5.4b)],
and over the entire region p > 0 for X negative. FIG. 9. Minimum pericenter for timelike geodesics as a
function of effective total energy r, shown for a == 0, .3, .6.
Timelike Geodesics The graph for a = 0.3 has three components, corresponding
to }.. < 0, p < p_,}.. < 0, and p > p+, }.. > 0; that for a==0.6
A very similar analysis can be carried through has two components, corresponding to}.. < 0 and}.. > O. Note
that the p < p_ portion (when it exists) covers the entire
for the timelike geodesics; it is complicated, however, available energy range r ~ -1. Stable bound orbits in the
by the presence of an additional energy parameter, exterior region occur for p > Per it, where Per It is the value
of P at which the graph has zero slope-see also Fig. 11.
PI = -"'{ or equivalently r = "'{2 - 1, which governs
the type of motion that results. In the Schwarzschild positive energy. At p = 2 the required total energy
case, it may be recalled, the minimum possible is zero, and it decreases to a minimum value of
pericenter for a particle trajectory changes with the r = -t at p = 3 (or r = 6 m). These features are
effective total energy r in the fashion shown by reflected in the stability of circular orbits: they
Fig. 9. There are no pericenters for p < !. To reach are stable if and only if p exceeds 3. We therefore
a point between ! < p < 2 and be deflected back recognize in the Schwarzschild problem three char-
out again the particle must come from infinity with acteristic radii-in addition to the famous "singu-
larity" at p = I-given by p = !, 2, and 3. Let us
try to determine the corresponding characteristic
radii in the Kerr metric.
First of all, the minimum pericenters analogous
to p = ! are given once more by Eqs. (5.4a) or
(5.4b). But the striking feature of the null case per-
sists here as well: at all energies (i.e., for all r ~ -1)
pericenters are found within the region 0 < p ~ p_,
so that these minima are not absolute ones if a
differs from zero.
To see how the character of the motion changes
A as r is varied, a diagram of the X-p plane, showing
4
allowed and forbidden regions for various values of
r, may again be helpful (see Fig. 10). The turning
points, which separate these two regions, are still
given by the zeros of a cubic polynomial:
1/i(PJ == r/ + / + (ra 2
- X2)p + (A + "'{a)2. (5.6)
Solved explicitly for A, this gives
X = (1 - p)-I{-",{a
FIG. 8. A diagram of the }..-p plane, showing ~he regions in
which light trajectories can exist. Shaded regIOns are for-
± [pcp - p+)(p - p_)(rp + I)]i), (5.7)
bidden; their boundaries define the turning points for null analogous to Eq. (5.5). By studying the double
rays. Two cases, a == 0.48 and a = 0.52, are showniTthe
former contains horizons, at p_ = 0.36 and p+ = 0.64. .Note zeros of 1/i(p) one constructs the plot of r vs minimum
the small forbidden region extending from p = 0 to p == p- pericenter shown in Fig. 9. The critical energy cor-
when }.. is negative in the former case. When a ~ i the two
forbidden regions for}.. < 0 coalesce. responding to Darwin's value r erlt = -t splits into
do not yield so easily; however, the orbit equation

is again relatively simple. One finds 20
2
dcp a A(u)
(5.10)
du = D(u) ± D(u) [B(u) )t
.5 10 with B(u) defined as above, and

P A(u) = l - fJ(l + 'Ya)u,
-t D(u) = 1 - fJu + aV == 2
a (u+ - u)(u_ - u).
The sign of the second term in Eq. (5.10) is to be
-2 r chosen to agree with that of u. With a = 0 this
a =-1
b • -.5 reduces to dcpldu = ±l[B(u)rl, whose solution in
C '-;2
-3
d • -.05
terms of Jacobi functions is immediate. The general
case requires elliptic integrals of the third kind,
which are clumsier to deal with and not very il-
FIG. 10. A portion of the A-p plane (cf. Fig. 8), showing
regions in which timelike trajectories can exist. Graphs were luminating; for this reason we avoided detailed dis-
plotted for a = 0.48 and for several values of r < o. In each cussions of exact solutions in the previous sections.
case the allowed region lies to the left of the curve. An addi-
tional forbidden region, not shown on the graph, extends from For purposes of illustration, however, we think it is
p = 0 to p = p_, much like the one shown in Fig. 8. Allowed instructive and not too painful to work out one
regions also exist on the negative sheet for r > o. Graphs
(a) and (b) illustrate the case of no bound orbits; (c) has bound example in detail, and we accordingly derive here
orbits for A < 0, (d) for A > 0 as well. an exact formula for the deflection of a beam of
light confined to the equatorial plane in the Kerr
two values, r:rit say. As Fig. 10 makes clear, stable field.
bound orbits in the exterior region are possible if The total deflection Acp is obtained from Eq.
r;rit < r < 0 and A is negative; if r:rit < r < 0, (5.10) on integrating the right-hand side over a
they are possible for positive A as well. contour extending from u = 0 to the branch point
It is easy to see that r attains these critical values u = lid at pericenter and back to u = 0 again.
whenever the cubic polynomial 1/i(p) has a triple The first term, being analytic, does not contribute
zero. Consequently,
Peril
(5.8a) 6
with Pori t in turn a real root of the following quartic:
p4 _ 6/ + (9 - 60/)/ - 14elp + 9a4 = o. (5.8b)
4
There are in general two such roots, with correspond-
ing critical energies r:rit, shown in Fig. 11.
Exact Solutions; The Deflection of Light 2
The above work is based almost exclusively on

the energy integral, Eq. (4.5). This does not differ
substantially from the corresponding formula in the
Schwarzschild problem-at least, not as long as one
restricts his attention to the equatorial plane-so
that the solution for r(p.) should be basically the 0.5
same. Indeed, on making the standard transforma-
tion u = llr one gets
ii = 4
u B(u) (5.9a)
1.0
with B(u) a cubic polynomial:
B(u) = r + €fJu FIG. 11. Values of critical radii and corresponding critical
energies, calculated from Eq. (5.8) for 0 :$ a :$ 1. Curves
+
(ra - nu + fJ(l + 'Ya)V.
2 2
(5.9b) labeled ( +) have A > 0; those labeled ( - ) have A < O. Porit
gives the minimum radius-in units of 2m-for which stable
Hence u(p.) can be expressed in terms of elliptic bound orbits are possible (excluding those orbits which pene-
trate the inner horizon and bounce back onto another sheet of
functions. The other first integrals, for IjJ and t, the manifold).
MAXIMAL AN ALYTIC EXTENSION OF THE KERR METRIC 281
to the total path integral, so one is left with

(5.15)
1/d
A(u) du
flIP = 2 l 0 D(u) [B(u)]i'
(5.11)
This result has also been obtained by Skrotskii30
using the weak-field metric of Landau-Lifshitz, and
To bring this to standard form, split the rational
from a more general viewpoint by Plebaflski. 31 The
portion of the integrand into partial fractions, as
value for the correction term a/dis difficult to esti-
A(u) A+ + A_ , mate in the case of the sun, since its angular momen-
D(u) = u+ - u u_ - u tum is not well known. Assuming essentially uniform
and write the cubic B(u) as rotation throughout its interior, at the rate 14.3 0 / day
observed for sun spots near the equator, one com-
(3(l + al(81 - U)(8 2 - u)(u - 8 a) putes a ~ 1.9 km and consequently aid ~ 3 X 10- 6
with 81 ~ 8 2 ~ 0 ~ 8 a• By definition, 82 = l/dj the for a light ray grazing the sun's disk, which is unde-
remaining two zeros can be expressed in terms of tectable by several orders of magnitude. Measure-
l, a and d as follows: ments of the solar oblateness are not inconsistent
with a value of aid several times as large as this32 j
81 = (Q +P - (3)/2{3d, 83 = -(Q - P + (3)/2(3d, however, this would require a mass quadrupole
(5.12) moment Q0 of order 10- 4 M 0R0', which in tum
where would contribute a correction term to 0 of about the
same magnitude. 33
P = del - a)/(l + a), Q2 = (P _ 2(3) (P + 3(3)
Note added in proof: It has been pointed out by
have been chosen to agree as closely as possible
C. V. Vishveshwara (University of Maryland Tech.
with Darwin's notation.
Rept. No. 589) that if one considers "stationary"
Introducing the further parameters
sources and observers (Le., those whose world lines
k2 = 82 - 83 , • 2
sIn .1,
'Yo = --83
-- are the Killing trajectories a/at), then the surface
81 - 83 82 - 83 of infinite red shift occurs at gil = 0, which does
one obtains after some manipulation not coincide with the null horizon in the Kerr case
(unless a = 0).
flfJ = 4(l + a)-I(d/Q)l {A+(u+ - 83)-1
ACKNOWLEDGMENTS
X [Il(a!, k) - Il(1/;o, a!, k)] + A_(u_ - 83)-1
X [Il(a:, k) - Il(1/;o, a:, k)]}, (5.13) Weare much indebted to Professor A. Schild
and members of the Center for Relativity Theory of
where Il(,l, k), II (1/;0, 0/, k) are, respectively, the the University of Texas for their hospitality and
complete and incomplete elliptic integrals of the many stimulating conversations. We are also grate-
third kind. It can be shown that the above result ful to Brandon Carter for many helpful discussions,
reduces to flIP = 7r when {3 = O. This is of course and for placing several unpublished results at our
to be expected, since the Kerr metric becomes flat disposal.
for (3 = 0, and serves as a check on the intermediate
calculations. In the limit a ~ 0 one finds A+ Added in proof: I acknowledge above all my debt
0, A_/(u_ - 83) = l, and a_ = 0, so to a dear friend, collaborator and co-author, whose
penetrating insight and ingenuity were responsible
flIP ~ 4(P/Q)i[K(k) - F(1/;o, kJ] for bringing this work to a successful conclusion .
• _0
= 4(P/Q)iF(1/;I' k), (5.14) To the extent that this paper conveys these traits,
it bears the imprint of his thought, and stands
with cot 1/;1 = (1 - k2)! tan 1/;0, in agreement with here as a memorial to his accomplishment.-R.W.L.
Darwin's result.
30 G. V. Skrotskii, Dokl. Akad. Nauk SSR 114, 73 (1957)
From Eq. (5.13) one readily deduces the cor- [English transl.: Soviet Phys.-Doldady 2,226 (1957)].
rections to the familiar deflection formula due to 31 J. Plebanski, Phys. Rev. 118, 1396 (1960).
at R. H. Dicke, Nature 202, 432 (1964), has suggested that
rotation of the central body. We assume {3/d and the interior of the sun might have a rotational period as small
a/ d small, and keep terms to order {3a/ d2 • Setting as 25 h, without leading to an unreasonably large visual oblate-
ness or violating stellar structure theory. This would increase
flIP = 7r + 0, aid to 6 X 10-5, still too small to be significant.
33 The quadrupole field adds to Eq. (5.15) a term! (4mld)
where 0 is the deflection angle as usually defined, (kid)', where Q = mk'is the quadrupole moment. Taking a
graVitational oblateness of 5 X 10-5 for the sun, as suggested
we find by Dicke," we calculate (kld)'0 = 7 X 10-5•

Maximal Analytic Extension of The Kerr Metric: Additional Information On J. Math. Phys

Uploaded by

Copyright:

Available Formats

Maximal Analytic Extension of The Kerr Metric: Additional Information On J. Math. Phys

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Maximal Analytic Extension of The Kerr Metric: Additional Information On J. Math. Phys

Uploaded by

Copyright:

Available Formats

Maximal Analytic Extension of the Kerr Metric

Robert H. Boyer and Richard W. Lindquist

Citation: J. Math. Phys. 8, 265 (1967); doi: 10.1063/1.1705193

Additional information on J. Math. Phys.

Maximal Analytic Extension of the Kerr Metric*

I. INTRODUCTION expected in more realistic models. We here lay

tions are usually presented locally, in terms of a r = f, 8 = 8, <P = ;p, (1.1)

The corresponding manifold, illustrated by the now

null vector field, H a scalar field. It does not matter

the k congruence is displayed. The (altogether dif- x 2 + y2 Z2

(r, 6, <g) Coordinate Surfaces

To explore the properties of the principal null

we find that which is again of the form: fiat-space metric plus

With this normalization Eq. (2.280) is satisfied, and

lent analytic extensions of the original (E) frame,

ordillate which will straighten out the helices. Many

This reeult is illustrated by Fig. 8, which shows r

Not surprisingly, the shape of this curve is qualita- I

horizons exist) and a > ! (where they do not).

do not yield so easily; however, the orbit equation

.5 10 with B(u) defined as above, and

The above work is based almost exclusively on

to the total path integral, so one is left with

You might also like