Home Search Collections Journals About Contact us My IOPscience

Notes on isolated horizons

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2000 Class. Quantum Grav. 17 5025

(http://iopscience.iop.org/0264-9381/17/24/307)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 155.69.4.4
This content was downloaded on 24/08/2015 at 23:56

Please note that terms and conditions apply.

Class. Quantum Grav. 17 (2000) 5025–5045. Printed in the UK PII: S0264-9381(00)17326-5

Notes on isolated horizons

G Date
The Institute of Mathematical Sciences, CIT Campus, Chennai 600 113, India
E-mail: shyam@imsc.ernet.in

Received 21 September 2000

Abstract. A general analysis for characterizing and classifying ‘isolated horizons’ is presented in
terms of null tetrads and spin coefficients. The freely specifiable spin coefficients corresponding to
isolated horizons are identified and specific symmetry classes are enumerated. For isolated horizons
admitting at least one spatial isometry, a standard set of spherical coordinates are introduced and
an associated metric is obtained. An angular momentum is also defined.

PACS numbers: 0420, 0470

1. Introduction

Stationary black holes in asymptotically flat spacetime obey the famous laws of black hole
mechanics [1] which admit a thermodynamical interpretation [2] supported by the Hawking
effect [3]†. These provide an arena to study the interplay of classical gravity, quantum
mechanics and thermodynamics [4]. These involve the global notions of event horizon,
asymptotic flatness and stationarity. There have been two generalizations by relaxing some
of the global conditions, namely the ‘trapping horizons’ of Hayward [5] and the ‘isolated
horizons’ of Ashtekar et al [6].
Hayward [5] introduced the quasi-local idea of ‘trapping’ horizons by abstracting the
property that these are foliations of (suitably) marginally trapped surfaces. He gave a
classification of ‘trapping horizons’ and also defined quantities satisfying a generalized version
of all the laws of black hole mechanics. The entire analysis is strictly quasi-local with no
reference to any asymptotics.
Among these classes of horizons, is a special case wherein a horizon is a null hypersurface.
These horizons have the property that the area of its two-dimensional spacelike surfaces remains
constant along its null generators and the second law reduces trivially to the statement that the
area is constant. Since one expects these areas to change only when there is energy flow across
the horizon, these horizons suggests an intuitive idea of ‘isolation’.
With a somewhat different notion of ‘isolated horizons’, recent work by Ashtekar et al
[6] also seeks to replace event horizons and yet obtain an analogue of the zeroth and the
first law of black hole mechanics. In this generalization, though reference to asymptotics is
retained, stationarity of usual black holes is relaxed. In particular, Ashtekar et al formulate
these notions and their associated quantities such as mass, in terms of variables and actions
suitable for passage to a non-perturbative quantization. This generalization permits one to deal
† See also references in the reviews in [4].

0264-9381/00/245025+21$30.00 © 2000 IOP Publishing Ltd 5025

5026 G Date

with a situation wherein a collapse proceeds in stages punctuated by a series of ‘non-activity’

stages across a collapsed ‘core’. Each such stage of ‘non-activity’ is modelled by an ‘isolated
horizon’. Loss of stationarity comes with a price which makes this generalization non-trivial
and leads to infinitely many forms of the first law [7]. This framework allows Ashtekar et al
to compute the entropy of the so-called non-rotating isolated horizons [8].
The latter notion of isolated horizons is more restrictive than that of trapping horizons
in the sense that isolated horizons are null hypersurfaces, while trapping horizons have no
such restriction. The restriction is not overly strong, the space of solutions admitting isolated
horizons is infinite dimensional [9]. The restriction, however, gives some control over the
space of solutions and recently Ashtekar et al [7] have given a very interesting interpretation
of the first law(s).
In both the generalizations mentioned above, there are two parts. The first is the geometric
characterization of an appropriate horizon. This involves both the distinguishing features of
an appropriate three-dimensional sub-manifold of a spacetime and the definitions of quantities
such as surface gravity, mass, area, angular velocity, angular momentum, charge, etc associated
with the horizon. The second part involves showing that the quantities thus defined do satisfy
the laws of black hole mechanics (or a generalized version thereof).
In this work we consider a uniform treatment of general isolated horizons only. Our aim
is to arrive at a characterization such that the zeroth and the first law of black hole mechanics
hold. One should observe at the outset that the zeroth law is a property of a single solution
of the matter–Einstein field equation admitting an isolated horizon. The first law, however,
involves the comparison of certain quantities associated with several such solutions and thus
is a statement concerning properties of the full space of such solutions. Establishing the first
law is therefore expected to be more involved.
A remark concerning the slight difference in the approach of Ashtekar and co-workers
and ours is in order. In the approach of Ashtekar et al, an action principle plays an important
role, particularly for the first law and of course in the subsequent computation of the entropy.
It is thus natural to view an isolated horizon as an ‘inner boundary’ of a suitable class of
spacetime manifolds and transcribe characterization of isolated horizons in terms of boundary
conditions on the variables entering the action formulation. On the other hand, as a first
step, we wish to translate the physical idea behind the isolated horizon into a geometrical
characterization and try to get a handle on the space of solutions admitting such horizons.
For this an action formulation is unnecessary and it is natural to view an isolated horizon as
a suitable null hypersurface of a solution spacetime. Since a tetrad formulation is closer to
(metric) geometry, it is a natural choice for us. This has the further advantage that it can
be generalized to other dimensions. If one could obtain the laws of black hole mechanics
without the use of an action principle, then one could also treat phenomenological matter.
One should also note that since both approaches capture the same physical idea, one does
not expect different results at the level of geometrical characterization. What we do seek
though is a direct and systematic arrival at a geometrical characterization of the physical
notions.
The basic defining property of these horizons is that they are all null hypersurfaces.
This naturally suggests the use of a null tetrad formulation. In terms of suitably adapted
null tetrads and their corresponding spin coefficients, this alone immediately implies that the
coefficients κ = 0 and ρ is real. To this one adds the requirement that the expansion of the
null geodesic generators should be zero (ρ = 0), which incorporates the properties of being
isolated (constancy of area) and potential marginal trapping. Since these are supposed to be
hypersurfaces of physical spacetimes, it is only natural to require that Einstein equations hold
and that the stress tensor satisfies a suitable energy condition. The Raychoudhuri equation
 Notes on isolated horizons 5027

and the energy conditions then imply a number of consequences, one of which is that the null
geodesic congruence is also shear free (σ = 0).
There is, however, a good deal of freedom in choosing the null tetrads, the freedom to make
local Lorentz transformations. These have been conveniently classified into three types [10].
Having obtained a null direction field, the relevant freedom is reduced to only the so-called
type I and III transformations. Due to this freedom, one cannot characterize isolated horizons
in terms of the spin coefficients by themselves unless these are invariant under the local Lorentz
transformation. (The particular values of the spin coefficients in the previous paragraph are
indeed invariant.) One can either impose conditions on quantities that are invariant under
these local Lorentz transformations or one could fix a convenient ‘gauge’ and then use the
corresponding spin coefficients for characterization. In this work we use a combination of
both.
The paper is naturally divided into two parts. The first part uses a gauge-fixing procedure
to identify freely specifiable spin coefficients. The second part uses invariant quantities to
analyse symmetry classes of the horizons and identifies suitable associated quantities.
The paper is organized as follows. In section 2 we discuss the basic conditions
characterizing isolated horizons and arrive at the zeroth law. We also obtain the ‘freely
specifiable’ spin coefficients by a gauge-fixing procedure. In section 3 we discuss a symmetry
classification (isometries). A characterization of ‘rotating horizons’ is discussed and an angular
momentum is identified. This section uses invariant quantities and constructs a natural set of
coordinates on the null hypersurface. We assume the existence of at least one spatial symmetry
for this purpose. A candidate parametrization of a ‘mass’ is also given. Section 4 contains
a summary and remarks on the first law. The notation and conventions used are those of
Chandrasekhar [10] and some of these are collected in appendix A for the reader’s convenience.
The metric signature is (+−−−). Appendix B is included to illustrate our procedures for the
Kerr–Newman family.

2. Characterization of isolated horizons

2.1. Basic conditions and the zeroth law

Our definition of an isolated horizon will turn out to be the same as the most recent one given
by Ashtekar et al [7]. We will, however, not try to define it completely intrinsically nor view
it as a boundary to be attached to an exterior asymptotically flat spacetime. We will explicitly
view an isolated horizon as a null hypersurface in a four-dimensional spacetime which is a
solution of a set of Einstein–matter field equations. The matter–stress tensor is required to
satisfy suitable energy conditions and the solution is required to be causally well behaved. To
keep the logic and the role of each of the conditions transparent, we will begin with minimal
conditions, see their implications and add further conditions as needed.
(I) An isolated horizon is a null hypersurface in a solution of four-dimensional Einstein–
matter field equations with matter satisfying the dominant energy condition (which also
implies the weak energy and the null energy conditions). Thus, the Einstein equation
holds on  and the matter–stress tensor satisfies:
• the weak energy condition and, in particular, Tµν µ ν
0;
• Tµν ν is causal. It is future (past) directed according to µ .
One can always write these equations with reference to a chosen tetrad. The null
hypersurface character of , singles out one null direction and makes the use of a null tetrad
natural. We will thus assume that we have made a (arbitrary) choice of null tetrad, , n, m, m̄,
5028 G Date

such that , at points of , is along the direction of the null normal. There is of course a large
class of null tetrads to choose from and this is parametrized by the group of type I and III
local Lorentz transformations (see appendix A). We will refer to the type I transformations as
(complex) boosts, type III scaling as scaling and type III rotations of m, m̄ as rotations.
The hypersurface orthogonality of  implies that the null congruence defined by µ ∂µ is
a geodetic congruence (κ = 0) and is twist-free (ρ is real). The geodesics, however, are not
affinely parametrized in general. The shear of this congruence is given in terms of σ . Vanishing
of κ already makes the twist and shear invariant under boost transformations. Furthermore,
the null geodesics generate  and hence  has the topology of R × 2 (we assume there
are no closed causal curves). The R part can be an interval, while 2 could be compact or
non-compact. At present, one need not stipulate these global aspects. However, later we will
restrict to 2 being spherical. The topology, guarantees the existence of at least one foliation,
not necessarily unique, as follows.
Choose any two-dimensional sub-manifold 2 of  such that it is transversal to the
direction field of the null normals. Fix an arbitrary null vector field along the null direction.
Under diffeomorphisms generated by the null vector field, we will generate images of 2 and
hence a foliation of . The leaves of such a foliation will be diffeomorphic to 2 . One will
therefore have a natural (and arbitrary) choice of null directions tangential to the leaves. Once
we make a choice of m, m̄, the null tetrad is uniquely completed. It then follows immediately
that the null congruence defined by nµ ∂µ is also twist-free (µ is real). Since the leaves are
obtained by the diffeomorphism generated by  and the m, m̄ are tangential to the leaves, it
follows that,
(L m)µ nµ = (L m̄)µ nµ = 0. (1)
This, in turn, implies that (α + β̄ − π) = 0 must hold. Note that this procedure can always
be followed and hence we can always have a choice of null tetrad such that µ is real and
α + β̄ = π.
Remark. In the above we made two arbitrary choices, the vector field  and the initial
transversal 2 . We have not defined m, m̄ by any transport from those chosen on initial
2 . There is no loss of generality though. By the available freedom of making local boosts,
scaling and rotations, we can change the initial null normal by local scaling and also the
initial leaf (and hence the foliation) by local boosts. These transformations are not completely
general though since we want to preserve the foliation property. We will refer to these restricted
transformations as residual transformation. These will become progressively further restricted.
At this stage, the rotation parameter is completely free and so is the scaling parameter, while
the boosts parameter satisfies
Da − ( − ¯ )a = 0
D ā + ( − ¯ )ā = 0 (2)
δ ā − (ᾱ − β)ā = δ̄a − (α − β̄)a.

Thus from the null hypersurface property and a procedure of choosing null tetrads we have
deduced that for every given initial choice of null normal vector field and initial 2 , there exists a
choice of null tetrads such that the spin coefficients satisfy, κ = ρ− ρ̄ = µ− µ̄ = α+ β̄ −π = 0.
Furthermore we are free to make the residual transformations.
For the next condition one can give two different arguments. The  and the n congruences
are orthogonal to the leaves. If the n congruence were also geodesic (ν = 0) then depending on
the expansion of the n-congruence, we could have the leaves as (marginally) trapped surfaces.
 Notes on isolated horizons 5029

Indeed, we would like to have this property, at least in the black hole context, so at the least
we require  to be expansion-free (ρ + ρ̄ = 0). We will, however, leave ν unspecified for the
moment.
Alternatively, this condition also shows that the induced metric on the leaves does not vary
from leaf to leaf and hence the area of a leaf is a constant or equivalently, the Lie derivative of
m ∧ m̄ projected to 2 is zero. Since the area is expected to change only when there is energy
flow across the horizon, this captures the notion of ‘isolation’. In the terminology of Ashtekar
et al [7], this is the specification of a non-expanding horizon.
(II)  congruence is expansion-free (ρ + ρ̄ = 0). A number of consequences follow from
these conditions. The first of the energy conditions in conjunction with the Raychoudhuri
equation for the expansion of  congruence and the zero-expansion condition above implies
that
• shear () = 0 (σ = 0),
• Rµν µ ν = 0 on  and by Einstein equations
• Tµν µ ν = 0 on .
Note that since the  geodesics are not affinely parametrized, the Raychoudhuri equation
has an extra term ( + ¯ )θ (). This, however, does not affect the conclusion.
Tνµ ν being causal and Tµν µ ν = 0 then implies that Tνµ ν = eµ , with e being non-
negative. Again using the Einstein equation this implies that


ν R
Rµν  = 8πe + µ . (3)
2
Hence we obtain the Ricci scalars 00 = 01 = 10 = 0. Furthermore, using definitions of
the Ricci scalars we deduce,
11 + 3 = −4πe ≡ −E ( 0). (4)
For future use we also note that the conservation of the stress tensor and the above form for it
implies that,
 · ∇ E ≡ D E = 0. (5)
The Bianchi identities from appendix A imply,
D(2 − 11 − ) = 0. (6)
Equation (h) from item 5 of appendix A gives,
2 + 2 = −δπ − π(π̄ − ᾱ + β) + Dµ + µ( + ¯ ). (7)
Combining the above equations we obtain
−K ≡ 2 − 11 −  = −π π̄ − {δπ − (ᾱ − β)π} + Dµ + µ( + ¯ ) + E (8)
and D of the left-hand side is zero by the Bianchi identity. K introduced above is the (complex)
curvature of 2 as defined by [11]. Restricting now to compact leaves without boundaries one
has

(K + K̄) = 4π(1-genus) (Gauss–Bonnet) (9)
2

(K − K̄) = 0 (Penrose). (10)
2
5030 G Date

We note in passing that the following can be checked explicitly:

ω+ ≡ πm + π̄ m̄ ⇒ dω+ ∼ (δπ − δ̄ π̄ )m ∧ m̄
(11)
ω− ≡ i(π m − π̄ m̄) ⇒ dω− ∼ (δπ + δ̄ π̄ )m ∧ m̄.
Here the underlined derivatives are the compacted (rotation-covariant) derivatives defined in
appendix A. From this equation (10) follows and the integral of the right-hand side of the
second of the above equations over any leaf also vanishes.
Substituting for K in the above equations and specializing to a spherical topology
(genus = 0) gives,

 
δπ + δ̄ π̄
2π = π π̄ + − E − Dµ − ( + ¯ )µ, (12)
2 2 2

0= (δπ ± δ̄ π̄). (13)
2
We also deduce that,
¯ π̄ }
−(K − K̄) = 2 − ¯2 = −{δπ − (ᾱ − β)π } + {δ̄ π̄ − (α − β)
= δπ − δ̄ π̄. (14)
Note that all of these equations are manifestly invariant under rotations. Also observe that if
2 is not real, then π cannot be set to zero.
Since 2 is invariant under boost, scaling and rotation transformation when κ = 0, its value
has a physical meaning, incidentally, , 11 are also invariant under these transformations.
Thus, the complex curvature K of the leaves is an invariant.
Above we obtained an imaginary part of the complex curvature in terms of derivatives of
π, π̄ in the gauge chosen. Since K is invariant under boost transformations, the right-hand
side of equation (14) above, must also be invariant under residual boost transformations. This
implies that we must have,
Dπ ≡ Dπ + ( − ¯ )π = 0. (15)
Now, equations (b) + (d̄) of item 5 of appendix A imply that,
δ̄( + ¯ ) = Dπ + ( − ¯ )π = 0. (16)
Thus,  + ¯ is constant on each leaf but it may vary from leaf to leaf. This now limits the
scaling freedom to scaling by a factor which is constant on each leaf. We may now exhaust
this freedom by setting  + ¯ to a constant on . Alternatively, we may note that invariance of
the Dπ = 0 condition under residual boost transformations implies that,
D( + ¯ ) = 0. (17)
We have therefore already obtained that the real part of  is a constant on . This of course
reduces the scaling to scaling by a constant factor.
Thus at this stage we have κ, ρ, σ, (α + β̄ − π ), 00 , 01 are zero and µ is real. As a
consistency check of the equations with the gauge choice, we also deduced that  + ¯ is constant
over the horizon. We have neither required ν to be zero nor that µ is strictly negative (positive).
Physically we have already captured a geometrical property of  that it is potentially foliated
by marginally trapped surfaces in a physical spacetime.
We already have a definition of area, namely the area of a leaf with respect to the induced
metric on 2 and that this area is ‘constant’. What could be a candidate for a ‘surface gravity’?
In the usual case of stationary black holes, it is the acceleration at the horizon of the Killing
 Notes on isolated horizons 5031

vector normal to the horizon, with suitable normalization of the stationary Killing vector at
infinity. Presently, we do not have any stationary Killing vector. The topology of , however,
suggests that  serves to define an evolution along . Thus, its acceleration, κ̃ ≡ ( + ¯ ) is a
natural candidate. Indeed, as we have seen above, κ̃ is constant over ! Identifying κ̃ with
surface gravity (modulo a constant scale factor to be fixed later) we already have the zeroth
law. The rest of the logic is similar to [6]. The Bianchi identity implies,
(D + κ̃)(Dµ) = 0. (18)

Remarks.
1. We have deduced the zeroth law using just the two conditions (non-expanding horizon),
use of foliation and use of the residual local Lorentz transformation. The notion of weak
isolation [7] is not necessary. This is an alternative mentioned by Ashtekar et al in [7].
Although we have fixed the residual scaling freedom to a constant scaling only, we are
still left with full rotation freedom and residual boost freedom.
2. All of the above works with any initial choice of . In the process though we obtained the
scaling to be restricted to a constant scaling only. Thus if we regard two s as equivalent
if they differ by a constant non-zero factor, then all of the above holds for any given
equivalence class of . We could, however, begin with an equivalence class such that
Dµ = 0. As long as an initial µ is non-zero we can always perform a local scaling
transformation to the new µ to satisfy D  µ = 0. This will fix the initial  and hence its
equivalence class. Thus it is possible to choose a unique equivalence class of  such that
Dµ = 0. Note that this condition is preserved by the residual transformations. At this
stage we do not need to make such a choice, though we will use this towards the end of
this section.
3. We have obtained two constant quantities, surface gravity and area, associated with . We
could obtain κ̃ in terms of integrals of certain expressions over a leaf via equation (12).
Note that in the absence of µ being constant on a leaf, Gauss–Bonnet does not give κ̃
directly in terms of the area.

2.2. Freely specifiable spin coefficients

As noted earlier, the zeroth law refers to a single solution, while the first law refers to a class
of solutions. To gain an understanding of such a class, we now proceed to identify freely
specifiable spin coefficients corresponding to (non-expanding) horizons. For this of course we
have to choose a suitable gauge.
Since we view  as a sub-manifold of a solution, we now consider excursion off-. We
still continue with an arbitrary initial choice of the null normal vector field and initial 2 (now
restricting to spherical topology). On  we have constructed null tetrads and therefore have
nµ ∂µ defined. Consider geodesics specified by points on  and the nµ ∂µ . We need only
infinitesimal geodesics to go infinitesimally off-. In this neighbourhood we construct null
tetrads by parallel transporting the null tetrads from , n · ∇(tetrad) = 0. The equations
of item 2 from appendix A then immediately give γ , ν, τ to be zero off- and therefore, by
continuity also on .
Thus, on , we have six of the 12 spin coefficients as zero, namely, γ , κ, ν, ρ, τ, σ .
Furthermore, we have the real part of  as a constant, µ real and π = α + β̄.
Now one can always choose a gauge such that the imaginary part of  is zero on  (see the
transformation equations of appendix A). This has two consequences. Firstly, it reduces the
5032 G Date

rotation freedom, so far unrestricted, to rotations by parameter which is constant along the null
generators. It is still local along the leaves. Secondly, the eliminant equation of the appendix
implies that α − β̄ is constant along the null generators of . Thus both the combinations,
α ± β̄ are now constant along the null generators and  is a real constant. Now only α ± β̄, µ
and λ are non-trivial functions on . Are all of these freely specifiable?
Not yet! The spin coefficients have still to satisfy the Einstein equations. Only the Ricci
scalars enter in these equations and Weyl scalars can be thought of as derived quantities via
the 18 complex equations of [10]. The eliminant equation gives directly relations among the
spin coefficients.
In appendix A, we have collected equations from [10]. These equations use the conditions
derived from the non-expanding horizon conditions discussed above. Since we are interested
in quantities defined on , the equations involving D  derivatives are omitted. These can be
understood as specifying the derivatives off  in terms of quantities specified on .
We see immediately that the eliminant equations give no conditions on the non-trivial
spin coefficients mentioned above. From the full set of 18 equations of [10] one can see the
following. All quantities are evaluated on .
• 00 appears in equation (a) which is identically true.
• 01 appears in equations (c)–(e) and (k) together with 1 . These serve to give 1 = 0
and D  κ = 0. This, in turn, implies that κ is zero in an infinitesimal neighbourhood of .
• 02 appears in equations (g) and (p). These serve to give D  σ = −02 and also
Dλ − δ̄π − π 2 + κ̃λ = 02 . (19)
This is a non-trivial condition among the non-vanishing coefficients and can be thought
of as a differential equation for λ given the other quantities.
• 12 appears in equations (i), (m), (o), (r) together with ¯ 3 . Use (r) to eliminate 3 . The

remaining equations show that D (π − α − β̄) = 0 and 12 determines only the off-
derivatives. One consequence of these is that π = α + β̄ in a neighbourhood of .
• 22 appears only in equation (n) and specifies D  µ. With the equation being real, it
follows that µ is real in a neighbourhood.
• 11 ,  appear in combinations with 2 in equations (f), (h), (l), (q). One can use (l) to
determine 2 and to eliminate the complex curvature of leaves, K, from the remaining
equations. Two of the equations then determine D   and D  ρ, while the remaining one
gives
Dµ − δπ − π π̄ + κ̃µ = α ᾱ + β β̄ − 2αβ − δα + δ̄β + 11 + 3. (20)
Again this can be thought of as a differential equation for µ in terms of the remaining
quantities. Note that the equation (q) implies that if 2 is real then ρ is real in the
neighbourhood.
• Lastly, equation (j) just gives 4 in terms of D  λ.
In effect we have obtained two (differential) conditions on two of the non-trivial
coefficients, µ and λ and are left with just α ± β̄ as freely specifiable. Since these are constant
along the null generators, these need to be given only on a leaf.
The differential equations for λ and µ can be ‘reduced’ further. Recall the remark
concerning the initial choice of the null normal vector field. Generically we can choose it to be
such that Dµ = 0. This is consistent with the differential equation due to the Bianchi identity
and the zeroth law. This fixes a unique equivalence class of null normals. We could consider
also imposing the Dλ = 0 condition. If 02 is zero, as is the case for vacuum, Einstein–
Maxwell and Einstein–Yang–Mills (Einstein–YM) systems, the condition is consistent with
 Notes on isolated horizons 5033

the differential equation. If it is non-zero, then consistency with the differential equation
requires D02 = 0 which, in turn, via the Bianchi identity (e) requires D  0 = 0. Constancy
of µ and λ along the generators is precisely the extra condition for isolated (as opposed to
weakly isolated) horizons that is imposed by Ashtekar et al [7].
Without the constancy of µ, λ along the generators, we see that π = (α + β̄), (α − β̄), µ
and λ may be freely specified on a leaf. κ̃ may then be determined via the Gauss–Bonnet
integral. Not all of these specifications give different (non-expanding) horizons since we still
have a good deal of residual transformations apart from the initial choices of  and foliation.
With the constancy of µ, λ along generators is imposed, we fix the initial equivalence
class of , determine λ and κ̃µ in terms of α ± β̄ and 02 specified on a leaf. The residual
rotation can be fixed completely by making α − β̄ real. We are then left with only the constant
scaling and the dependence on the initial foliation (residual boosts). Note that when  is real,
α − β̄ is invariant under boost transformations. Thus our identification of freely specifiable
spin coefficients is ‘gauge invariant’. The residual transformations now just serve to demarcate
equivalent isolated horizons.

Remarks.

1. Apart from seeing the role of Einstein equations via the 18 equations of [10], we also
determined the Weyl scalars. Of the five Weyl scalars, only 2 and 3 are determined by
freely specifiable coefficients on a leaf and Ricci scalars. Thus the (real) curvature of the
leaves, and hence the metric on the leaves is also determined. The freely specifiable data
consists of just one complex function, π , and one real function, α − β̄, on a leaf.
2. It is apparent from our analysis and also proved by Lewandowski [9] that the space of
solutions admitting isolated horizons is infinite dimensional. In the above analysis we
have addressed the freedom of local Lorentz transformations naturally present in a tetrad
formulation. We also have the diffeomorphism invariance though. Thus two sets of data
related by a diffeomorphism on a leaf must be regarded as giving the ‘same’ isolated
horizon. This is of course implicit in an analysis based on an ‘initial-value problem’
formulation [9]. The corresponding diffeomorphism classes must be characterized by
diffeomorphism invariants. Integrals of scalars (e.g. invariant combinations of spin
coefficients) over the leaf are just one such set of invariants. We will obtain a few of
these in the next section.

In summary, we have shown that every solution of Einstein–matter equations with

matter satisfying the dominant energy condition and admitting non-expanding horizon admits
foliations and a corresponding choice of null tetrads modulo constant scalings such that the
free data consists of one complex function π and one real function α − β̄. Furthermore, with κ̃
being identified as (unnormalized) surface gravity, the zeroth law holds for all such solutions.
In the next section we will consider the symmetry properties of isolated horizons and
see that if leaves admit at least one isometry, there is a unique choice of foliation and the
equivalence is reduced to that implied by constant scaling only.

3. Constants associated with isolated horizons

In this section we discuss symmetries of isolated horizons and corresponding ‘conserved’

quantities. This discussion is carried out in terms of invariant quantities so that the conclusions
are not tied to any particular gauge choice. Under certain conditions we also see how this helps
to fix a unique foliation.
5034 G Date

3.1. Symmetries of 
In the previous section we naturally obtained two constant quantities associated with , the
surface gravity and the area. Noting that quantities constructed out of the spin coefficients
and their derivatives and which are invariant under the residual transformations are physical
characteristics of a given isolated horizon, . An example is the complex curvature K defined
above. To look for further characteristic quantities associated with a given  we analyse the
symmetries of these horizons.
Consider a fixed solution containing . This is characterized by a set of physical quantities.
By definition, a symmetry of  is a diffeomorphism of  which leaves these quantities
invariant. Since the induced metric on leaves is one such quantity, a symmetry must be
an isometry of the leaves. However, there are further invariant quantities built from ingredients
other than the induced metric on leaves, e.g. (K − K̄). We now look for further such invariant
quantities.
Since residual transformations permit a change of foliation, it is desirable to make
use of quantities which are invariant under these. Thus in order to classify the symmetry
classes and distinguish a rotating case, we look for invariant forms and vector fields on .
Any vector field (or form) can be expressed as an expansion in terms of the appropriate
tetrad basis. Depending on the transformation properties of the expansion coefficients under
residual Lorentz transformations, these expansions will (or will not) preserve their form. By
invariant vectors (forms) we mean form invariance under residual transformations. The general
coordinate transformations of course play no role in the discussion. The use of such ‘invariant’
quantities frees us from having to keep track of the particular choices of tetrad bases. The
demand of form invariance puts restrictions on the expansion coefficients which can then be
taken in a convenient manner as seen below.
Noting that the tangent space of  is spanned by the tangent vectors , m, m̄, while the
cotangent space is spanned by the cotangent vectors n, m, m̄, it is easy to see that a generic
invariant vector field X on  is parametrized as
1
X≡ (ζ + π C̄ + π̄C) + Cm + C̄ m̄ (21)
−κ̃
where C transforms as π under rotation and is invariant under a boost, while ζ is real
and invariant under both sets of transformations. These are our candidates for generating
symmetries of .
The condition that X should be a Killing vector (of the metric on  and not the four-
dimensional spacetime metric) requires, in terms of the rotation covariant derivatives (see
appendix A):
δ C̄ = 0, δ̄C = 0, δC + δ̄ C̄ = 0. (22)
The third of the above equations can be solved identically by setting C ≡ iδ̄f where f is an
2
invariant function, while the first two require f to satisfy δ 2 f = 0 = δ̄ f .
While looking for an invariant 1-form, one should note that when restricted to forms the
boost transformations have their -dependent terms dropped. Then there are two types of such
1-forms (real and spacelike):
ω ≡ −κ̃n + πm + π̄ m̄, ω(X) = ζ ;
(23)
ω̃ ≡ C̃m + C̃¯ m̄, ω̃(X) = −(C C̃¯ + C̄ C̃).
Since the 1-form ω is invariant and built out of spin coefficients, it is a physical
characteristic of  and thus a symmetry-generating invariant vector field must leave this
 Notes on isolated horizons 5035

invariant, LX ω = 0. This immediately implies that D(ζ ) = 0. It also gives two differential
equations for ζ , namely,
δζ = −C̄(K − K̄), δ̄ζ = C(K − K̄). (24)
This has several implications. Firstly, X(ζ ) = 0, i.e. ζ is constant along the integral curves of
X. Secondly, taking D of the equations, using the commutators of the derivatives and Bianchi
identity implies that DC = 0. Finally, the integrability conditions for these equations, which
require that X(K − K̄) = 0 and δC + δ̄ C̄ = 0, are automatically true by our definition of
symmetry. It then follows that the solution for ζ is unique up to an additive constant. Note
that when K is real, ζ must be a constant. These integrability conditions will also allow us to
set up an adapted (θ, φ) coordinate system on leaves.
One can construct an invariant vector field Y (not necessarily a symmetry generator)
which is orthogonal to a symmetry generator X by taking C going to iC where  is a real
and invariant function. We would also like to use a symmetry generator along . Thus we
define two further invariant vector fields,
1
Y ≡ (ζ − i(π C̄ − π̄ C)) + i(Cm − C̄ m̄) (25)
−κ̃
1
Z≡ . (26)
−κ̃
Evidently, X, Y are real, spacelike and mutually orthogonal (for all ). Z is of course a null
symmetry generator and is orthogonal to X and Y . Observe that ω(X) = ζ = ω(Y ), ω(Z) = 1.
We would now like to have all these vector fields commute so that parameters of their
integral curves can be taken as coordinates for . While commutativity of X, Z is a statement
concerning the nature of the isometry algebra, commutators involving Y are just stipulations
on Y which in no way affects properties of X or Z.
The commutativity of the vector fields X, Y with Z requires:
Dζ = 0, DC = 0, D = 0. (27)
Note that commutativity of Z, X is already implied by equation (24), while that of Z, Y is a
condition on . These enable us to construct the invariant vector fields as follows. For any leaf,
assume we could find a real function f , satisfying δ 2 f = 0. Extend it to  by Lie dragging
by . This implies that f so constructed, satisfies the double-derivative condition on all the
leaves. Furthermore, these conditions themselves are invariant under residual transformations.
We could similarly define , ζ on .
The commutativity of X, Y with X a Killing vector (of leaves) requires X() = 0 and
X(ζ ) − Y (ζ ) = 2iC C̄(K − K̄). (28)
The above equation automatically holds due to equation (24). The only additional information
we have obtained is that the as yet arbitrary  function is constant along integral curves of
X. There are no conditions implied on X due to the demands of commutativity. Several
consequences can now be derived.
The integral curves of X (the Killing orbits) in general leave a leaf. We could, however,
choose a foliation such that these orbits are confined to leaves. This means that the coefficient
of  must be made zero. This can be effected by a boost transformation. This still leaves a
one-parameter freedom of boost residual transformation. This can be fixed by demanding that
integral curves of Y be similarly confined to leaves. The boost parameter effecting this is given
by,
π̄ iζ
a+ = (1 + i). (29)
κ̃ 2κ̃C
5036 G Date

That the parameter a satisfies the conditions of being a residual transformation requires
X(ζ ) = 0, which is already seen to be true. For the transformation parameter to be well
defined, it is necessary that ζ vanishes wherever C does. Since C must vanish for at least
one point on a leaf, ζ must also vanish for at least one point. Thus, when K is real, which
implies that ζ is a constant, ζ in fact must be zero.
Note that this fixes the boost freedom completely. The left-hand side is nothing but the
transformed value of π̄ divided by the surface gravity. Thus in effect, this boost transformation
has fixed for us a particular foliation. In this foliation we have the X, Y vector fields purely
tangential to the leaves. Consequently, we call the vector field X a ‘rotational’ symmetry
generator.
In this foliation ζ = −(π C̄ + π̄C) and we see that ζ is zero iff there exists a gauge in
which π is zero. This fact will be used in defining the angular momentum in terms of ζ .
Since the leaves are compact, the Killing vector field is complete. Suppose for the moment
that its orbits are closed curves. Provided that the range of the Killing parameter is the same
for all the orbits, we can adjust this range to be 2π by a constant scaling of X and identify
the Killing parameter to an angular coordinate φ. This will genuinely make the spatial Killing
vector correspond to axisymmetry. When can this be done?
First, we need to argue that orbits of the Killing vector, X, are closed. Every vector field
on S 2 will have at least one zero (or fixed points). These fixed points can be classified as
elliptic, hyperbolic, etc by standard linearizations†. The third of equation (22) implies that X
is ‘area preserving’ and thus has zero divergence. This implies that its zeros are all elliptic
and hence orbits in the vicinity are closed. Since the orbits cannot intersect, all orbits in fact
must be closed. The vector field Y in contrast is not divergence-free and thus its orbits are not
closed.
For an integral curve γ of X, the line integral of the invariant 1-form ω is equal to the
integral of ω(X) = ζ along γ . Since ζ is constant along such curves, its integral will be
ζ (γ ) × I (γ ), where I (γ ) is the range of the Killing parameter along γ . Now one can consider
a family of such γ s labelled by an infinitesimal integral curve of Y . Differentiate the integrals
with respect to the parameter along Y . On the left-hand side use Stokes theorem together with
dω = (K − K̄)m ∧ m̄, while on the right-hand side use a Taylor expansion. This gives the
left-hand side as,
   
ω− ω= dω = (K − K̄)m ∧ m̄ ∼ −2iC C̄(K − K̄)(β)I (β)δβ.
γ (β+δβ) γ (β) cyl cyl

(30)
Here the surface integral is over an infinitesimal cylinder formed by the family. In the last step
of course we have used the mean-value theorem. The right-hand side gives,

{Y (ζ )I (β) + ζ (β)Y (I (β))}δβ. (31)

Equation (28) satisfied by ζ then implies that the parameter range does not vary along Y and
hence is a constant. We can now introduce the azimuthal angle φ. Clearly, ζ and  are now
independent of φ.
We have used the vector field Y to show that an azimuthal coordinate can be introduced.
This is independent of the choice of . Can we also introduce the polar angle θ ? The answer
is yes as can be seen below.
Every vector field on S 2 must vanish for at least one point. X vanishes precisely when
C = 0. If  is non-singular then Y also vanishes precisely where ever X vanishes. One can
† See any modern book on dynamical systems, e.g. [12].
 Notes on isolated horizons 5037

‘decompactify’ the leaf by removing such a point and see that C must vanish at one and only
one more point. Furthermore, the integral curves of Y must ‘begin’ and ‘end’ at these two
points. This may also be seen via the Poincaré–Bendixon theorem [12]. Furthermore, since
X, Y commute, the diffeomorphisms generated by X take orbits of Y to orbits of Y . Choosing
a ‘meridian’ we could suitably adjust  and hence select a Y so that the parameter along this
curve ranges from 0 to π . Lie dragging by X then defines these for other longitudes. In this
way we can introduce the standard spherical polar angles on the leaf (boost gauge fixed). The
metric can also be written down as

ds 2 = −2C C̄(2 dθ 2 + dφ 2 ). (32)

That such a choice of  is possible can be seen by construction. For the above form of the
metric, one can compute the Ricci scalar in terms of . We also have (K + K̄) as the curvature
of the leaves. Equating these two gives a first-order differential equation for . Its smooth
behaviour on the leaf, except for the poles, fixes the constant integration and determines 
in terms of C and (K + K̄). For the Kerr–Newman family it reproduces the precise metric
components. Thus existence of an invariant Killing vector (spatial), permits us to determine
both ζ and  and also allows us to introduce an adapted set of coordinates θ, φ on the spherical
leaves. Note that while on each of the leaves we can introduce these coordinates, the choice of
prime meridians on different leaves is arbitrary. These could be related by diffeomorphisms
generated by Z.

Remark. Commutativity of the invariant vector fields, X, Y, Z, not only allows us to introduce
spherical polar coordinates on leaves, it also allows us to introduce coordinates on  itself.
With the excursion off- defined via parallel transport along geodesics defined by nµ ∂µ , one
can naturally also introduce coordinates in the neighbourhood.

On a spherical leaf, we can have either none, one or three isometries i.e. either (a) no such
f , (b) precisely one such f , or (c) precisely three such f s. If we found two such functions
(two Killing vectors), then their commutator is either zero or is also a Killing vector. On S 2 we
cannot have two commuting Killing vectors, therefore the commutator  must be non-zero and
a Killing vector. One can consider the commutator X ≡ X, X  of two Killing vectors and
see that it is also an invariant Killing vector provided the ζ  satisfies the same conditions. This
is possible only if K − K̄ = 0. Thus, as expected, we see that maximal symmetry is possible
only if 2 is real. In this case of course one does not expect any rotation. The converse need
not be true as shown by the ‘distorted’ horizons [7].

Remark. We have fixed the boost freedom by demanding X, Y be tangential to a leaf. When
we have three such pairs of vector fields, are all of these automatically tangential to the same
leaf? The answer is yes. Since three isometries implied K is real, all the ζ, ζ  , ζ  are constants
and hence zero. Furthermore, tangentiality of any one implies π = 0. Therefore, all three
Killing vectors are tangential to the same leaf.

In the case of single isometry we have both possibilities, namely K − K̄ is zero or non-
zero. These two can be seen intuitively as hinting that while a ‘rotating’ body is expected to
be distorted (bulged) a distorted body need not be rotating. Since maximal symmetry implies
‘no rotation’ and also that K is real, we label the two cases as:
 is non-rotating iff K is real (i.e. 2 is real) and is rotating iff K has a non-zero imaginary
part.
We have already noted that in the non-rotating case ζ must be zero. This means that
π can be transformed to zero. One could also see directly from the equations of residual
5038 G Date

transformations that it is possible to choose a gauge such that π = 0. For the rotating case, of
course, ζ is non-zero and π is also non-zero in every gauge.

3.2. ‘Conserved’ quantities

While ζ introduced above, could be non-zero, it is some function of the adapted coordinate θ.
Its integral over the leaf is clearly a constant which is zero iff K is real. This integral is therefore
a candidate for defining angular momentum. Indeed, the explicit example of Kerr spacetime,
discussed in appendix B, shows that the candidate agrees with the angular momentum of the
Kerr spacetime. It also provides a proportionality factor. Unlike the Kerr black hole, however,
this angular momentum is defined in terms of quantities intrinsic to . Also, for the Kerr–
Newman solution, one can find ζ and its integral. Interestingly, one does not obtain the usual,
total angular momentum (J = Ma) of the Kerr–Newman spacetime, but one obtains the total
angular momentum minus the contribution of the electromagnetic field in the exterior. Indeed,
one can explicitly check that the integral of ζ is precisely equal to the Komar integral [13]
evaluated at the event horizon. This justifies the case for taking the following as the angular
momentum of an isolated horizon. Thus we define,

1
J ≡− ζ (im ∧ m̄). (33)
8π S 2
Apart from the ‘rotational’ symmetry generator defined above, we see that gZ is also a
symmetry generator provided g is a constant. Following an identical logic as for the rotational
symmetry generator, we see that the analogue of J is proportional to the area. Thus area
is the ‘conserved’ charge associated with Z. Unlike X whose normalization is fixed due to
compactness of its orbits, S 1 , normalization of Z is not fixed. Clearly, arbitrary constant linear
combinations of Z and the well defined ‘rotational’ X is a symmetry generator. Let us denote
this as
ξa,b ≡ (−aZ − bX); ζ̃a,b ≡ ω(ξa,b ) = −(a + bζ ). (34)
We have introduced the suffixes a, b to remind ourselves that the suffixed quantities depend on
the arbitrary constants a, b. We define, by analogy with J , a corresponding conserved quantity
as,

1 a
M̃a,b ≡ − ζ̃a,b = Area + 2bJ. (35)
4π S 2 4π
Note that unlike ζ which is determined via a differential equation involving (K − K̄), ζa,b
is not completely determined by (K − K̄). A Smarr-like relation still follows because of the
symmetries we have and our parametrization of the general isometry, ξa,b in terms of arbitrarily
introduced constants a, b.
Interestingly, the usual Smarr relation for the Kerr–Newman family, with M, J replaced
by the Komar integrals evaluated at the event horizon has exactly the same form as above with
a = κ̃KN and b = 6KN ! The Maxwell contribution is subsumed in the Komar integrals at the
horizons and there is no explicit ‘Q’ term.
It is thus suggested that Ma,b may be identifiable as the mass of the isolated horizon.
It is defined on the same footing as the angular momentum. For b = 0, ξa,b is a spacelike
isometry (an analogue of the stationary Killing vector) as in the case of the Kerr–Newman
family. However, a, b are arbitrary parameters. Furthermore, while b is the analogue of the
angular velocity, it is not clear why it must be non-zero when and only when J is non-zero.
The notion of angular velocity is tricky though because, on the one hand, ‘rotation’ must
be defined with respect to some observers (in the usual case asymptotic observers), on the
 Notes on isolated horizons 5039

other hand, it is signalled by the imaginary part of 2 which is a local property. At present we
have not been able to resolve this issue.

Remarks.

1. The above candidate definitions are not the same as those given by Ashtekar et al [7].
There need not be any inconsistency since our definitions need not enter a first law in
the usual manner. Some functions of our quantities may become the mass and angular
momentum which will enter a first law.
2. Our candidate identifications are based on analogy and consistency with the Kerr–Newman
family. The symmetry generators, Z, X, are invariant vector fields. In particular, they
are insensitive to the constant scaling freedom which is unfixed as yet. Provided, the
constants a, b, are also invariants, the general symmetry generator ξa,b will also be an
invariant vector field. If a, b are proposed to be the surface gravity and the angular velocity
based on analogy and dimensional grounds, these will be subject to the constant scaling
freedom and one will have to face the normalization issue as in [6, 7]. If, on the other
hand, a, b (equivalently ξa,b ) could be directly tied to the spacetime under consideration,
and are calculable then a first law variation may be obtained directly. In the absence of a
stationary Killing vector there does not appear to be any natural choice of ξa,b .

Without relating the constants a, b to any other intrinsic properties of  and/or normalizing
them suitably, Ma,b cannot be fully identified as the mass of . Since we do not address the
first law in this work, we do not pursue the identification further.
In summary, in this section we assumed the existence of at least one spatial symmetry of
 and deduced a candidate definition of the angular momentum and a possible identification of
the mass. We also saw how this enables us to fix a unique foliation together with an adapted set
of spherical polar angular coordinates. At present, we have not been able to obtain a candidate
definition of the angular velocity. Whether or not at least one spatial isometry must exist is
also not addressed. These will be addressed elsewhere.

3.3. The special case of spherical symmetry

If a solution admitting an isolated horizon has maximal symmetry for the leaves, we have
already seen that complex curvature must, in fact, be real. Furthermore, it must be constant due
to maximal symmetry. Our gauge choice fixing the foliation then has π = 0. Equations (19)
and (8) which determine λ and µ then become,

κ̃λ = 20 , κ̃µ = −K − E ⇒ κ̃δµ = −δ E . (36)

If λ = 0 can be shown, then equation (m) of item 5 of appendix A will give 3 = 21 .
Likewise δ E = 0 will give µ to be a constant. It appears that maximal symmetry alone does
not imply either λ or δµ to be equal to zero. For the Maxwell and YM matter, the form
of the stress tensor together with the energy condition implies 02 = 0 and then λ = 0
does follow. Since the spacetime itself is not guaranteed to be spherically symmetric in the
neighbourhood, while δ E = 0 is a reasonable condition, it does not seem to be forced. If
these two conditions, however, are imposed, then the previous results on non-rotating and
spherically symmetric isolated horizons [6] are recovered. Incidentally, in this case positivity
of K (spherical topology) and of E implies that κ̃µ is negative. Thus for positive surface
gravity, µ must be negative implying marginal trapping.
5040 G Date

4. Summary and conclusions

In this work we have presented a general analysis of isolated horizons by manipulating the
basic equations of the null tetrad formalism. By using the existence of foliations (causally
well behaved solutions) and keeping track of local Lorentz transformations we have shown
that non-expanding conditions (as opposed to weak isolation) are adequate to obtain the zeroth
law. We also used a gauge-fixing procedure to identify freely specifiable spin coefficients. By
using invariant vector fields and 1-forms, we analysed the symmetries of isolated horizons and
showed that if at least one spatial isometry exists, then a unique foliation can be fixed. Further
more a natural choice of spherical polar coordinates exists for the leaves. We defined associated
‘conserved’ quantities which are consistent with the Kerr–Newman family. In appendix B we
have illustrated our procedures for the Kerr–Newman family.
We have not taken explicit examples of various matter sectors (except the Kerr–Newman
family). For the Einstein–Maxwell, Einstein–Yang–Mills with or without dilaton has been
discussed in [6]. We have also not invoked any action principle. Our hope was and still is to
deduce a first law without the use of an action principle. Action formulation(s) and its use has
already been discussed in detail in [6, 7].
The first law for isolated horizons is much more subtle. A very interesting formulation
and perspective is given by Ashtekar et al [7]. Here we will be content with some remarks.
The usual stationary black holes are parametrized by finitely many parameters (three for
the Kerr–Newman family) and the first law also involves only a few parameters. In contrast, the
space of solutions with isolated horizons is infinite dimensional and yet a first law is expected
to involve only a few parameters. Even granting the infinitely many forms of the first law [7],
each of these still involves a few parameters only with area, angular momentum and charges
playing the role of independently variable quantities. Why only a few quantities are expected
a priori to enter the first law of mechanics for isolated horizons is unclear. Ashtekar et al have
incorporated asymptotic flatness in an action principle and used the covariant phase space
formulation to deal with the space of solutions directly. They traced the existence of the first
law as a necessary and sufficient condition for a Hamiltonian evolution on the covariant phase
space of isolated horizons. It would be nicer to have a direct and quasi-local ‘explanation’ of
the first law.

Acknowledgments

It is a pleasure to acknowledge the illuminating and fruitful discussions Abhay Ashtekar

provided during his visit to the Institute of Mathematical Sciences in February 2000.
Stimulating discussions with Madhavan Varadarajan, Richard Epp and Samuel Joseph of the
Raman Research Institute during a visit there are gratefully acknowledged. I wish to thank
Abhay Ashtekar for his critical comments on an earlier draft of this paper.

Appendix A. Summary of relevant equations for spin coefficients

In this appendix we collect together the relevant and useful formulae from [10]. The
conventions are those of Chandrasekhar’s book.

1. Null tetrad basis in the tangent and cotangent spaces:

E1 = µ ∂µ ≡ D, E 1 = nµ dx µ ,
E2 = nµ ∂µ ≡ D  , E 2 = µ dx µ ,
 Notes on isolated horizons 5041

E3 = mµ ∂µ ≡ δ, E 3 = −m̄µ dx µ ,
E4 = m̄µ ∂µ ≡ δ̄; E 4 = −mµ dx µ .
2. Covariant derivatives in terms of spin coefficients:
µ;ν = (γ + γ̄ )µ ν + ( + ¯ )µ nν − (α + β̄)µ mν − (ᾱ + β)µ m̄ν − τ̄ mµ ν − κ̄mµ nν
+σ̄ mµ mν + ρ̄mµ m̄ν − τ m̄µ ν − κ m̄µ nν + ρ m̄µ mν + σ m̄µ m̄ν
nµ;ν = −( + ¯ )nµ nν − (γ + γ̄ )nµ ν + (α + β̄)nµ mν + (ᾱ + β)nµ m̄ν + νmµ ν + π mµ nν
−λmµ mν − µmµ m̄ν + ν̄ m̄µ ν + π̄ m̄µ nν − µ̄m̄µ mν − λ̄m̄µ m̄ν
mµ;ν = ν̄µ ν + π̄µ nν − µ̄µ mν − λ̄µ m̄ν − τ nµ ν − κnµ nν + ρnµ mν − σ nµ m̄ν
+(γ − γ̄ )mµ ν + ( − ¯ )mµ nν − (α − β̄)mµ mν + (ᾱ − β)mµ m̄ν .
3. Local Lorentz transformations:
These are given using κ, ρ, σ, 0 , 1 , 00 set equal to zero. These values are invariant
under both the sets of local Lorentz transformations.

Type I (boosts)
 = , n = n + ām + a m̄ + a ā, m = m + a, m̄ = m̄ + ā;
2 = 2 , 3 = 3 + 3ā2 , 4 = 4 + 4ā3 + 6ā 2 2 ;
  = , τ  = τ,
π  = π + 2ā + D ā, α  = α + ā, β  = β + a,
γ  = γ + aα + ā(β + τ ) + a ā,
λ = λ + ā(2α + π) + 2ā 2  + δ̄ ā + āD ā,
µ = µ + aπ + 2āβ + 2a ā + δ ā + aD ā,
ν  = ν + aλ + ā(µ + 2γ ) + ā 2 (τ + 2β) + a ā(π + 2α) + 2a ā 2 (D  + āδ + a δ̄ + a āD)ā.

Type III (scaling and rotations)

 = A−1 , n = An, m = eiθ m, m̄ = e−iθ m̄,
2 = 2 , 3 = A3 e−iθ , 4 = A2 e−2iθ 4 ,
µ = Aµ, τ  = eiθ τ, π  = e−iθ π, λ = Ae−2iθ λ, ν  = A2 e−iθ ν,
( + ¯ ) = A−1 {( + ¯ ) − A−1 DA}, ( − ¯ ) = A−1 {( − ¯ ) + iDθ },
(α + β̄) = e−iθ {(α + β̄) − A−1 δ̄A}, (α − β̄) = e−iθ {(α − β̄) + iδ̄θ }.
4. Type III rotation-covariant derivatives (‘compacted covariant derivatives’):
These are defined for a quantity that transforms homogeneously under rotations.
X  = einθ X, n an integer
DX ≡ DX − n( − ¯ )X; D  X  = einθ DX
δX ≡ δX + n(ᾱ − β)X; δ  X  = ei(n+1)θ δX

δ̄X ≡ δ̄X − n(α − β̄)X; δ̄ X  = ei(n−1)θ δ̄X.
5042 G Date

The corresponding commutators are given by

 
δ, D X = (ᾱ + β − π̄ )DX
 
δ̄, D X = (α + β̄ − π )DX
 
δ, δ̄ X = −n(K + K̄)X.
5. Riemann tensor component equations:
There are in total 18 such equations together with their complex conjugates. However,
since we are interested in values only on the horizon, we regard the equations involving
D  derivatives as specifying the same in terms of values obtained on . These therefore
give no conditions on  itself. Three of these equations, equations (a), (b) and (k) of
[10] have already been implied by the energy conditions, Raychoudhuri’s equation and
the zero-expansion condition on the  congruence. These give, 0 , 1 , 00 , 01 and 10
to be zero. This leaves us with six equations, two of the eliminant equations and two from
the Bianchi identities. These are listed below for convenience in terms of the ‘compacted
covariant derivatives’ where relevant. The equation labels refer to the equations from
Chandrasekhar’s book.
Dα − δ̄ = −α( − ¯ ) + (π − α − β̄) (d)

Dβ − δ = β( − ¯ ) + (π̄ − ᾱ − β) (e)

Dλ − δ̄π = π 2 − λ( + ¯ ) + 20 (g)

Dµ − δπ = π π̄ − µ( + ¯ ) + 2 + 2 (h)

δα − δ̄β = α ᾱ + β β̄ − 2αβ − (µ − µ̄) − 2 + 11 +  (l)

δλ − δ̄µ = π(µ − µ̄) + µ(α + β̄) − λ(ᾱ + β) − 3 + 21 . (m)

6. Eliminant equations:
δ(α + β̄ − π) − δ̄(ᾱ + β − π̄) = −D(µ − µ̄) − 2(αβ − ᾱ β̄) − (ᾱ − β)π + (α − β̄)π̄
(b )

D(ᾱ − β) + δ( − ¯ ) = (2ᾱ − π̄ )( − ¯ ). (f )

7. Bianchi identities (terms which vanish are dropped):

−D2 − D  00 − 2D = 0 (b )

−D(11 + 3) − D  00 = 0. (i )

8. Change of the metric on 2 under its diffeomorphism:

The metric on 2 is −(m ⊗ m̄ + m̄ ⊗ m)µν . We are interested in the Lie derivative of this
metric with respect to a vector field on 2 projected back onto 2 . For a vector field of
the form, X = Cm + C̄ m̄, one obtains,
LX (m ⊗ m̄ + m̄ ⊗ m) = 2δ̄C(m ⊗ m) + 2δ C̄(m̄ ⊗ m̄) + (δC + δ̄ C̄)(m ⊗ m̄ + m̄ ⊗ m)
LX (m ⊗ m̄ − m̄ ⊗ m) = (δC + δ̄ C̄)(m ⊗ m̄ − m̄ ⊗ m).
 Notes on isolated horizons 5043

Appendix B. Example of a Kerr–Newman family

In this appendix we describe the Kerr–Newman family in terms of the spin coefficients. While
these are given by [10], a slight modification is needed. Furthermore, these are also used to
illustrate our gauge-fixing procedure.
Metric:

2 2 
η2   sin θ η2
ds 2 = 2 dt 2 − (dφ − ω dt)2 − dr 2 − η2 dθ 2 , (B1)
 η 2 
where
η2 ≡ r 2 + a 2 cos2 θ,  2 ≡ (r 2 + a 2 )2 − a 2 sin2 θ
a(2Mr − Q2> )
ω≡ ,  ≡ r 2 + a 2 − 2Mr + Q2>
2
ξ ≡ r + ia cos θ, ξ̄ ≡ r − ia cos θ.
The area, surface gravity and angular velocity of the event horizon are given by (r+ is the
radius of the event horizon),
area = 4π(r+2 + a 2 )
r+ − M
surface gravity =
2Mr+ − Q2∗
a
angular velocity = 2 .
r+ + a 2
The null tetrad (principle congruences) definitions (the components refer to t, r, θ, φ,
respectively):
1 2 1 2
µ ≡ (r + a 2 , , 0, a), nµ ≡ (r + a 2 , −, 0, a),
 2η2



1 i 1 −i
mµ ≡ √ ia sin θ, 0, 1, , m̄µ ≡  − ia sin θ, 0, 1, .
2ξ sin θ 2ξ̄ sin θ
These are not well defined on the horizons ( = 0) since the coordinates are singular
at the horizon. , n are both future directed and ‘out-going’ and ‘in-coming’, respectively.
Outside the event horizon one can make a scaling by A = 1/, and obtain the  to be well
defined on the horizons. Though n is ill-defined on the horizon, it does not matter since we do
not need to use it. This scaling changes some of the spin coefficients given by Chandrasekhar.
Following are these changed values, on horizons, except γ which is not needed:
1
κ = ρ = σ = λ = ν = 0,  = r − M, µ=− ,
2ξ̄ η2
ia sin θ ia sin θ ia sin θ cot θ cot θ
π= √ , τ =− √ , α= √ − √ , β= √ .
2ξ̄ 2 2η2 2ξ̄ 2 2 2ξ̄ 2 2ξ
Note that we already have α + β̄ = π, and that the gauge-covariant derivative of π is zero.
However, µ is complex. We can make a boost transformation to make this real without
disturbing the ‘gauge condition’. Such a transformation parameter, b, satisfies

µ − µ̄ + bπ − b̄π̄ + 2b̄β − 2bβ̄ + δ b̄ − δ̄b = 0.

5044 G Date

In general this gives a one-parameter family of solutions. However, regularity on the spherical
leaves fixes this uniquely to
a sin θ
b = −i √ .
2 2r(r 2 + a 2 )
It can be checked explicitly that this is consistent with the commutation relations.
Since  is real, α − β̄ is unchanged. π and µ change to,

 ia sin θ 1 r −M
π = √ + (B2)
2 ξ̄ 2 r(r 2 + a2)

1 a 2 sin2 θ M 2
µ = − 2(r 2 − a 2 cos2 θ) + 2 η + a 2 sin2 θ . (B3)
4r(r 2 + a )η
2 2 r + a2 r
This is manifestly negative definite, consistent with marginal trapping.
For subsequent computations we note that all the quantities we will need will be functions
only of θ and hence effectively the derivatives δ, δ̄ are invariant and can be taken to be
1 1
δ = √ ∂θ , δ̄ = √ ∂θ , effectively. (B4)
2ξ 2ξ̄
The function f , that gives the rotational Killing vector, can be determined by noting that
ᾱ − β can be expressed as −δ(ln(sin θ/ξ )). Solving δ 2 f = 0 gives

f = −Ĉ cos θ + constant, Ĉ = −(r 2 + a 2 )

iĈ (B5)
C = iδ̄f = √ sin θ.
2ξ̄
The constant has been fixed by demanding that the range of the Killing parameter is 2π. This
is explained below.
One can solve the equations determining ζ and obtain

 2

ĈM η − 2r 2 r 2 − a2 ĈQ2∗ r 1 1
ζ = + + − 2 . (B6)
a η 4 (r + a )
2 2 2 a η 4 (r + a 2 )2
The condition that ζ must vanish where C vanishes, is used to fix the constant of integration.
It can be checked directly that the invariant vector field X is already tangential to leaves
(has the coefficient of  to be zero). So no further boost transformation is needed. In terms
of the coordinate basis provided by the t, r, θ, φ, X has non-vanishing components along the
t and φ directions, implying that the adapted azimuthal angle does not coincide with the φ.
This is not surprising since on the event horizon φ is singular. One can make a coordinate
transformation to the usual non-singular coordinates r∗ , φ̃ + [10] and see explicitly that in terms
of this coordinate basis, X has a non-vanishing component only along φ̃ + . Demanding that
this be equal to 1, fixes the constant Ĉ and identifies φ̃ + as the adapted azimuthal coordinate.
This coordinate transformation of course does not affect any of the spin coefficients. It does
not affect the area 2-form either.
It is straightforward to check that the integral of ζ over the sphere precisely equals
the angular momentum Komar integral [13] for the Kerr–Newman solution evaluated at the
horizon,
 
1 1
J ≡− ζ ds = − ζ (r 2 + a 2 ) sin θ dθ dφ. (B7)
8π S 2 8π S 2
 Notes on isolated horizons 5045

For comparison we include the Komar integrals corresponding to the stationary and the
axial Killing vectors on t = constant and r = constant surfaces,
1 a2 r 2 + a2  a  a 
M(r) = M − Q2 + 3+ arctan −
r 2r 2ar 2 r r
2 2   (B8)
2 3a 3a 3
(r + a )
2
a a
J (r) = Ma − Q + + arctan − .
4r 4r 3 4a 2 r 2 r r
Next, the Ricci scalar of a diagonal metric on S 2 is given by


1 ∂θ gφφ 1
R = − ∂θ + ∂ 2 gφφ . (B9)
2 det(g) det(g) θ
From our procedure of introducing θ, φ coordinates, we have gθ θ = −2C C̄2 , gφφ =
−2C C̄. We have found above C for the Kerr–Newman family and we also have R = K + K̄.
Thus we obtain a differential equation for . This can be solved easily to obtain (for both Kerr
and Kerr–Newman curvatures)
1 (r 2 + a 2 )2 sin2 (θ )
= . (B10)
2 η4
This reduces our standard form of the metric to the usual one for the Kerr–Newman solution.

References

[1] Bardeen J M, Carter B and Hawking S W 1973 Commun. Math. Phys. 31 161
[2] Bekenstein J D 1973 Phys. Rev. D 7 2333
Bekenstein J D 1974 Phys. Rev. D 9 3292
[3] Hawking S 1975 Commun. Math. Phys. 43 199
[4] Wald R M 1999 Preprint gr-qc/9912119
Ashtekar A 2000 Ann. Phys. 9 178
(Ashtekar A 1999 Preprint gr-qc/9910101)
[5] Hayward S 1994 Phys. Rev. D 49 6467
Hayward S 1994 Class. Quantum Grav. 11 3025
For a recent review see Hayward S 2000 Preprint gr-qc/0008071
[6] Ashtekar A, Beetle C and Fairhurst S 1999 Class. Quantum Grav. 16 L1
Ashtekar A, Corichi A and Krasnov K 1999 Preprint gr-qc/9905089
Ashtekar A, Beetle C and Fairhurst S 2000 Class. Quantum Grav. 17 253
Ashtekar A and Corichi A 2000 Class. Quantum Grav. 17 1317
Corichi A, Nucanmendi U and Sudarsky D 2000 Phys. Rev. D 62 044046
(Corichi A, Nucanmendi U and Sudarsky D 2000 Preprint gr-qc/0002078)
Ashtekar A, Beetle C, Dreyer O, Fairhurst S, Krishnan B, Lewandowski J and Wisniewski J 2000 Phys. Rev.
Lett. 85 3564
(Ashtekar A, Beetle C, Dreyer O, Fairhurst S, Krishnan B, Lewandowski J and Wisniewski J 2000 Preprint
gr-qc/0006006)
[7] Ashtekar A, Fairhurst S and Krishnan B 2000 Phys. Rev. D 62 104025
(Ashtekar A, Fairhurst S and Krishnan B 2000 Preprint gr-qc/0005083)
[8] Ashtekar A, Baez J and Krassnov K 2000 Preprint gr-qc/0005126
[9] Lewandowski J 2000 Class. Quantum Grav. 17 L53
[10] Chandrasekhar S 1998 Mathematical Theory of Black Holes (Oxford Classics Series) (Oxford: Oxford University
Press)
[11] Penrose R and Rindler W 1995 Spinors and Space-Time vol 1 (Cambridge Monographs on Mathematical
Physics) (Cambridge: Cambridge University Press)
[12] Tabor M 1989 Chaos and Integrability in Nonlinear Dynamics (New York: Wiley-Interscience)
Robinson C 1995 Dynamical Systems (Boca Raton, FL: Chemical Rubber Company)
[13] Wald R M 1984 General Relativity (Chicago, IL: University of Chicago Press)