Isolated Horizons: Hamiltonian Evolution

and the First Law

1,2 ∗ 1 † 1 ‡
Abhay Ashtekar , Stephen Fairhurst and Badri Krishnan

1. Center for Gravitational Physics and Geometry

Department of Physics, The Pennsylvania State University
University Park, PA 16802, USA
arXiv:gr-qc/0005083v3 25 Aug 2000

2. Institute for Theoretical Physics,

University of California, Santa Barbara, CA 93106, USA

Abstract

A framework was recently introduced to generalize black hole mechanics by

replacing stationary event horizons with isolated horizons. That framework
is significantly extended. The extension is non-trivial in that not only do the
boundary conditions now allow the horizon to be distorted and rotating, but
also the subsequent analysis is based on several new ingredients. Specifically,
although the overall strategy is closely related to that in the previous work,
the dynamical variables, the action principle and the Hamiltonian framework
are all quite different. More importantly, in the non-rotating case, the first law
is shown to arise as a necessary and sufficient condition for the existence of
a consistent Hamiltonian evolution. Somewhat surprisingly, this consistency
condition in turn leads to new predictions even for static black holes. To
complement the previous work, the entire discussion is presented in terms of
tetrads and associated (real) Lorentz connections.

I. INTRODUCTION

The zeroth and first laws of black hole mechanics refer to equilibrium situations and
small departures therefrom. The standard treatments [1–5] restrict themselves to stationary

∗ E-mail: ashtekar@gravity.phys.psu.edu

† E-mail: fairhurs@phys.psu.edu

‡ E-mail: krishnan@phys.psu.edu

1
space-times admitting event horizons and small perturbations from stationarity. While this
simple idealization is a natural starting point, from physical considerations it seems overly
restrictive. (See [6,7] and especially [8] for a detailed, critical discussion.) A framework,
which is tailored to more realistic physical situations was introduced in [6] and the zeroth
and first laws were extended to it in [7–9]. This analysis generalizes black hole mechanics in
two directions. First, the notion of event horizons is replaced by that of ‘isolated horizons’.
While the former can only be defined retroactively, requiring access to the entire space-time
history, the latter can be defined quasi-locally. Second, the underlying space-time need not
admit any Killing field; isolated horizons need not be Killing horizons. The static event
horizons normally used in black hole mechanics [1–3,10] and the cosmological horizons in de
Sitter space-times [11] are all special cases of isolated horizons. Furthermore, since space-
times can now admit gravitational and matter radiation, there is a large class of other
examples.
The framework developed in [8] for generalizing black hole mechanics was based on two
restrictive assumptions. First, only undistorted, non-rotating horizons were considered. That
is, the boundary conditions used in [8] implied that the intrinsic 2-metric of the horizon is
spherically symmetric1 and that the imaginary part of the Weyl tensor component Ψ2 —
which encodes the angular momentum — vanishes. Second, while a rather general class of
matter fields was allowed, it was assumed that the only relevant charges — i.e., hair — are
the standard electromagnetic ones. The second assumption was weakened in [9,12] which
allowed dilaton couplings and Yang-Mills fields. In this paper, we allow for distortion and
more general matter sources. Distortion plays an important role in several astrophysical
situations, e.g., in problems involving black holes immersed in external fields, or surrounded
by matter rings, and especially in the problem of black hole collisions. Post-Newtonian
considerations suggest that, during black hole coalescence, individual horizons are distorted
due to the Coulomb attraction even in the regime in which the black holes are sufficiently
far from each other for the gravitational radiation falling into their horizons to be negligible.
This phenomenon is also seen in numerical simulations.
The extensions [9,12] which incorporated dilatonic and Yang-Mills charges did not involve
a significant generalization of the basic framework developed in [8]. The present paper, on
the other hand, does. We begin with substantially weaker boundary conditions (formulated
in terms of (real) tensor fields rather than the spinors used in [8]), and show that they imply
constancy of surface gravity (and electro-static potential) on the horizon. This property turns
out to be necessary and sufficient for the usual action principle of tetrad gravity to continue
to be valid in presence of isolated horizons. The action leads to a covariant phase space,
constructed from solutions to the field equations. Ref [8], by contrast, used the canonical
phase space based on spinorial variables which is tailored for quantization but which contains

1 However,it allowed space-times, such as the Robinson-Trautman solutions, in which there is no

space-time Killing field whatsoever in any neighborhood of the isolated horizon [15].

2
technical complications that are unnecessary to the classical mechanics of isolated horizons.
Up to this point, distortion and rotation are both incorporated. However, in the last step, i.e.,
in the discussion of the first law, we restrict ourselves to non-rotating horizons. (Rotation is
incorporated in [13].)
To formulate the first law one must first define the energy E∆ associated with any iso-
lated horizon ∆. Since there can be radiation in the spacetime outside isolated horizons,
the ADM energy EADM is not a good measure of E∆ [7,8]. Instead, as in the work of Brown
and York [14], the strategy is to define the energy of the horizon using a Hamiltonian frame-
work. Experience with the phase space formulation of general relativity suggests that, in
the presence of boundaries, the Hamiltonian Ht generating time-translation along a suitable
vector field ta acquires surface terms. The idea is to define E∆ as the surface term at ∆ in
the Hamiltonian. The key issue then is that of selecting the ‘appropriate’ time-translation
ta . Since one expects the volume term in the expression of Ht to be a linear combination
of constraints and thus vanish when evaluated on solutions, the problem reduces to that of
specifying the boundary values of ta (or, equivalently, of the lapse and shift fields). The
conditions at infinity are obvious and, in any case, will not affect the surface term at ∆.
Thus, we need to focus only on the boundary value of ta on ∆.
In the non-rotating case it is clear that, at the horizon, ta should be proportional to the
null normal to ∆. However, our boundary conditions do not select the null normal uniquely;
there is a freedom to rescale2 the normal by a constant (on ∆) which can vary from one
space-time to another. This freedom is physically important because, amongst other things,
the surface gravity is sensitive to it. Suppose we fix this freedom by tying the boundary
value of ta to fields on the horizon, e.g., by demanding that surface gravity be a specific
function of the area and charges. We can then ask whether the time-evolution generated by
this ta preserves the symplectic structure. It turns out that the answer is not always in the
affirmative. On the horizon, the evolution vector field ta and the electromagnetic potential
Aa have to be tied to the horizon parameters appropriately. These conditions impose a
constraint on the surface gravity κ(t) and the electric potential Φ(t) . Somewhat surprisingly,
t
the constraint is precisely the first law δE∆ = (κ(t) /8πG) δa∆ +Φ(t) δQ∆ . Thus the evolution
defined by t is Hamiltonian if and only if the first law holds.3 In this sense, the first law is
a

even more fundamental than it is generally taken to be. Conceptually, this is perhaps the

2 This is not surprising since this freedom exists already on Killing horizons. If the space-time
is asymptotically flat and admits a static Killing field globally, one can eliminate this freedom
by restricting oneself to that Killing field which is unit at infinity. However, this strategy is not
available if there is radiation in the exterior region, or, as in the static solutions representing
distorted black holes, the metric fails to be asymptotically flat.

3 In the undistorted context, while this role of the first law was known to the authors of [8], its
importance was not fully appreciated. The importance was noticed independently in [16] and used
effectively in [12] to extract physical information on spherical black holes with Yang-Mills hair.

3
most striking feature of the present framework.
The requirement that the first law hold is not sufficient to fix ta uniquely. Although
every ta must be a null normal to ∆, the rescaling of ta from one spacetime to another can
depend on the horizon parameters and the first law does not fully determine this parameter
dependence. There is an infinite family of parameter-dependent vector fields ta each defining
t
a consistent Hamiltonian evolution and a horizon energy E∆ . By contrast, at infinity all
these vector fields must tend to a time-translation of an universal flat metric, used in the
construction of the phase space. Hence, there is only the familiar 3-parameter freedom in
t
the definition of EADM , associated with the choice of a rest-frame. Furthermore, in any one
space-time, we can eliminate it by simply going to the rest frame and thus extract the total
mass MADM of the system. Although not necessary for mechanics, it is natural to ask if one
can define a similar notion of mass of isolated horizons. The answer is in the affirmative
in the Einstein-Maxwell theory. Let us require that ta should not only lead to a consistent
Hamiltonian evolution but agree, on static solutions, with the static Killing field which is
unit at infinity. Then the horizon value of ta is uniquely determined for all space-times
in the phase-space. There is a preferred notion of time-translation, say tao . We can set
to
M∆ = E∆ and regard M∆ as the mass of an isolated horizon ∆. In the earlier work on
non-distorted horizons [7–9,12], the discussion of the first law was carried out only in the
context of these preferred evolution vector fields tao . That derivation is more closely tied
to the traditional discussion of the laws in the static context but is not necessary from the
more general perspective of isolated horizons. Nonetheless, the availability of a canonical
definition of the mass M∆ is useful for other applications of this framework, e.g. to numerical
relativity.
The paper is organized as follows. In Section II, we specify the boundary conditions defin-
ing general isolated horizons, allowing both distortion and rotation. We explain the role of
these conditions, compare them with those used in [6–9,12] and work out their consequences,
including the zeroth law. In Section III we introduce the Lagrangian framework based on
tetrads and (real) Lorentz connections and in Section IV, the covariant phase space. The
first law is discussed in Section V. Upto this point, the focus is on the Einstein-Maxwell the-
ory (although incorporation of the dilaton is straightforward). In Section VI we extend the
framework to incorporate Yang-Mills fields. The horizon mass is introduced in Section VII
and subtleties associated with the dilaton and Yang-Mills fields are discussed. For the conve-
nience of readers who may not be familiar with distorted black holes, Appendix A presents
a variety of examples and, for convenience of readers who work in the Newman-Penrose
formalism, Appendix B summarizes the structure of isolated horizons in that framework.
We have attempted to make this paper self-contained in terms of methodology and tech-
nical details. However, the motivation behind isolated horizons and certain properties of our
Hamiltonian are the same as those discussed in detail in [8]. Since the inclusion of distortion
does not add anything substantial to these issues, we have refrained from repeating that
discussion in this paper.

4
 II. STRUCTURE OF ISOLATED HORIZONS AND THE ZEROTH LAW

In this section, we will introduce the basic definitions of isolated horizons and analyze
their immediate consequences. The definitions will become progressively stronger. However,
even the strongest boundary conditions are significantly weaker than requiring the horizon
to be a Killing horizon for a local Killing vector field. By proceeding in steps, we will be
able to keep track of the precise assumptions that are needed to obtain various results. Also,
the availability of a hierarchy of definitions will be useful in other applications —such as
numerical relativity and quantum gravity— which lie beyond the scope of the present paper.
Let us begin by introducing some notation. Throughout this paper we assume that all
manifolds and fields under consideration are smooth. Let M be a 4-manifold equipped with
a metric gab of signature (−, +, +, +). Let ∆ be a null hypersurface of (M, gab ). A future
directed null normal to ∆ will be denoted by ℓ. (In this paper, the term ‘null normal’ will
always refer to a future directed null normal.) Let qab = b gab be the degenerate intrinsic
←−
metric on ∆.4 A tensor q ab on ∆ will be called an ‘inverse’ of qab if it satisfies q ab qac qbd =
b qcd .
Thus q ab is unique only up to addition of terms of the form ℓ(a V b) for some vector field V
tangential to ∆. The expansion θ(ℓ) of a specific null normal ℓ is defined by θ(ℓ) = q ab ∇a ℓb ,
where ∇a is the derivative operator compatible with gab . It is straightforward to check that
θ(ℓ) is independent of the choice of q ab . With this structure at hand, we can now introduce
our first definition.

A. Non-expanding Horizons

Definition 1: A 3-dimensional sub-manifold ∆ of a space-time (M, gab) is said to be a

non-expanding horizon if it satisfies the following conditions:

• (i) ∆ is topologically S 2 × IR and null;

• (ii) The expansion θ(ℓ) of ℓ vanishes on ∆ for any null normal ℓ;

• (iii) All equations of motion hold at ∆ and the stress-energy tensor Tab of matter fields
at ∆ is such that −Tba ℓb is future directed and causal for any future directed null normal
ℓ.

Note that if conditions (ii) and (iii) hold for one null normal ℓ they hold for all.
The role of these conditions is as follows. The restriction on topology is geared to the
structure of horizons that result from gravitational collapse. However, it can be weakened.

4 Equalitieswhich hold only at ∆ will be denoted by ‘= b ’ and the pullback of a covariant index
will be denoted by an arrow under that index, e.g. ω a will denote the pullback of the 1-form ωa to
←−
∆.

5
One can retain the requirement that the horizon have compact cross-sections but replace S 2
by a manifold with higher genus. Our main analysis will extend to this case in a straight-
forward manner. More generally, we can allow ∆ to have non-compact cross-sections, as for
example in the case of certain acceleration horizons. The results presented in this section,
including our derivation of the zeroth law will go through. However, since such horizons
extend to infinity, our Hamiltonian framework will have to be modified appropriately. Fi-
nally, one could envisage incorporation of NUT charge. This extension would be even more
subtle because, if all fields are smooth, ∆ would be topologically S 3 and ℓ would provide
a Hopf fibration. In this case, ∆ would not admit any cross-sections which are everywhere
transverse to ℓ. This extension will be discussed elsewhere.
Requirement (iii) is analogous to the dynamical conditions one imposes at infinity. While
at infinity one requires that the metric (and other fields) approach a specific solution to the
field equations (the ‘classical vacuum’) , at the horizon we only ask that the field equations
be satisfied. The energy condition involved is very weak; it is implied by the (much stronger)
dominant energy condition that is typically used. Thus, the first and the last conditions are
quite tame.
The key condition is (ii). It implies, in particular, that the horizon area is constant
‘in time’ and thus incorporates the idea that the horizon is isolated without having to
assume the existence of a Killing field. We will denote the area by a∆ and refer to R∆
2
defined by a∆ = 4πR∆ as the horizon radius. All these conditions are satisfied on any
Killing horizon (with 2-sphere sections) if gravity is coupled to physically reasonable matter
(including perfect fluids, Klein-Gordon fields, Maxwell fields possibly with dilatonic coupling,
Yang-Mills fields).
Although the conditions in the definition are quite weak, they have surprisingly rich
consequences. We will now discuss them in detail. In some of this analysis it is convenient
to use a null-tetrad and the associated Newman-Penrose quantities (see Appendix B and
references therein). Given a specific null normal field ℓa to ∆, we can introduce a complex
null vector field ma tangential to ∆ and a real, future directed null field na transverse to
∆ so that the following relations hold: n · ℓ = −1, m · m = 1 and all other scalar products
vanish. The quadruplet (ℓ, n, m, m) constitutes a null-tetrad. There is of course an infinite
number of null tetrads compatible with a given ℓ, related to one another by restricted Lorentz
rotations. Our conclusions will not be sensitive to this gauge-freedom.
(a) Properties of ℓ: Since ℓa is a null normal to ∆, it is automatically twist free and geodesic.
We will denote the acceleration of ℓa by κ(ℓ)

ℓa ∇a ℓb =
b κ(ℓ) ℓb . (II.1)

Note that the acceleration is a property not of the horizon ∆ itself, but of a specific null
normal to it: if we replace ℓ by ℓ′ = f ℓ, then the acceleration changes via κ(ℓ′ ) = f κ(ℓ) + Lℓ f .
Since the twist of ℓ vanishes, the Raychaudhuri equation implies:
1 2
b κ(ℓ) θ(ℓ) − θ(ℓ)
L ℓ θ(ℓ) = − σσ − Rab ℓa ℓb
2

6
where σ = ma mb ∇a ℓb is the shear of ℓ in the given null tetrad. Since θ(ℓ) vanishes on ∆,
we conclude: σ σ + Rab ℓa ℓb =
b 0. The condition on the stress-energy tensor ensures that
a b a b
Rab ℓ ℓ = 8πGTab ℓ ℓ is non-negative on ∆. Hence, we conclude:

σ=
b 0, and Rab ℓa ℓb =
b 0. (II.2)

Thus, in particular, every null normal ℓ is free of expansion, twist and shear.
(b) Conditions on the Ricci tensor : The second equation in (II.2) implies that the vector
−Ra b ℓb is tangential to ∆. The energy condition and the field equations imply this vector
must also be future causal. This means that Ra b ℓb must be proportional to ℓa and hence,
R a b ℓb = 0. In the Newman-Penrose formalism this condition translates to:
←
−

1 1
Φ00 = Rab ℓa ℓb =
b 0 and Φ01 = Φ10 = Rab ℓa mb =
b 0. (II.3)
2 2
Since this statement is equivalent to R a b ℓb = 0, it is gauge invariant, i.e. it does not depend
←−
upon the specific choice of null normal ℓ and m.
(c) A natural connection 1-form on ∆: Since ℓ is expansion, shear and twist free, there exists
a one-form ωa intrinsic to ∆ such that

∇a ℓb =
b ω a ℓb (II.4)
←
−

which in turn implies

Lℓ qab =
b 2∇
←−a
ℓb =
−−−
b 0.

Thus, every null normal ℓ is a ‘Killing field’ of the degenerate metric on ∆. Furthermore,
we will now show that
2
ǫ := im ∧ m (II.5)

is also invariantly defined. Since Lℓ qab = 0, the space S of integral curves of ℓ is naturally
equipped with a non-degenerate metric q ab (so that qab on ∆ is the pull-back of q ab ). Denote
by ǫab the unique (up to orientation) unit alternating tensor on (S, q ab ). 2ǫ is the pull-back
to ∆ of ǫ. Although ℓ is a ‘Killing field’ of the intrinsic horizon geometry, the space-time
metric gab need not admit a Killing field in any neighborhood of ∆. Robinson-Trautman
metrics [15] and the Kastor-Traschen solutions [17] provide explicit examples of this type.
The 1-form ω will play an important role throughout this paper. It has an interesting
geometrical interpretation. We can regard ω as a connection on the line bundle T ∆⊥ over
∆ whose fibers are the 1-dimensional null normals to ∆. Under the rescalings ℓ 7→ ℓ̃ = f ℓ,
of the null normal ℓ, it transforms via:

ωa 7→ ω̃a = ωa + ∇a ln f. (II.6)
←
−

7
(d) Induced Connection on ∆: Each metric submanifold M of M admits a natural connection
— one which is torsion-free and compatible with the induced metric on M. This connection
is also canonically induced by the space-time connection ∇. However, since the induced
metric qab on ∆ is degenerate, there exist infinitely many connections compatible with it. A
general null sub-manifold inherits a unique (torsion-free) derivative operator D from ∇ if and
only if its null normal ℓ satisfies ∇
←−a
ℓb = 0. Therefore, the conditions imposed in Definition
−−−
1 guarantee that every non-expanding horizon has a unique intrinsic derivative operator D.
The action of D on a vector field X a tangent to ∆ and on a 1-form ηa intrinsic to ∆ is given
by

Da X b =
b ∇
←−
−a X̃
b
and Da ηb =
b ∇a−η̃
←− −−b .

where X̃ b and η̃b are arbitrary extensions of X b and ηb to the full space-time manifold M.
It is easy to show that D is independent of the extensions.
The 1-form ω captures only part of the information in D. The full connection D on ∆
plays an important role in extracting physics in the strong field regime near ∆ [18]. However,
it is not essential to the discussion of isolated horizon mechanics.
(e) Conditions on the Weyl tensor : Let us begin with the definition of the Riemann tensor,
[∇a ∇b − ∇b ∇a ]X c = −2Rabd c X d . If we set X c = ℓc and pull back the indices a and b, then
using (II.4), we obtain:

b −2Cabd c ℓd
b −2Rabd c ℓd =
[Da ωb − Db ωa ]ℓc = (II.7)
←−
− ←−
−

where Cabc d is the Weyl tensor. The last equality follows from Rab ℓb =
b 0. Thus, if v is any
←
−
1-form on ∆ satisfying v · ℓ =
b 0, contracting the previous equation with vc we get

Cabd c vc ℓd =
b 0.
←−
−

Let us choose a null tetrad and set v to be m or m. Then

Ψ0 := Cabcd ℓa mb ℓc md =
b 0 and Ψ1 := Cabcd ℓa mb ℓc nd = Cabcd ℓa mb mc md =
b 0 , (II.8)

where we have used the trace-free property of the Weyl tensor in the second equation. It is
also clear that equations (II.8) are independent of which null normal ℓ, and vector fields m
and m we choose to construct the null tetrad; equation (II.8) is gauge invariant.
(f) Curvature of ω: Let us contract (II.7) with nc and use ℓa na = −1. Then we have:
c d c d
b Cabd ℓ nc =
2D[a ωb] = b Cabcd ℓ n . (II.9)
←−
− ←−
−

Expanding the Weyl tensor in terms of the Ψ’s, one obtains

Cabcd ℓc nd =
b 4(Re [Ψ2 ])n[a lb] + 2Ψ3 ℓ[a mb] + 2Ψ3 ℓ[a mb]
−2Ψ1 n[a mb] − 2Ψ1 n[a mb] + 4i(Im [Ψ2 ])m[a mb] . (II.10)

8
where

b Cabcd ℓa mb mc nd
Ψ2 := and b Cabcd ℓa nb mc nd .
Ψ3 := (II.11)

Substituting this expression into (II.9), pulling back on the two free indices and taking into
account (II.8) and (II.5), we obtain

b 2(Im [Ψ2 ]) 2ǫ .
dω = (II.12)

This relation will play an important role in what follows. Note that, because Ψ0 and Ψ1
vanish on ∆, Ψ2 is gauge invariant.

Remark: It is interesting to compare the structure of ∆ with that of null-infinity I (in

the usual conformal gauge, in which the conformal factor is chosen such that the null-
normal to I is divergence-free). Both are null surfaces and can be regarded as ‘line bundles’
over a base space S of the integral curves of null normals. (For brevity, we will ignore a
caveat concerning completeness of fibers.) The null normals are Killing fields of the intrinsic
degenerate metric so that this metric is the pull-back to the 3-surface of a positive-definite
metric on S. In both cases, the space-time connection induces an intrinsic connection on the
3-surface [19]. These connections capture physically important information. However, there
are a number of differences as well. Since I is constructed by a conformal completion, the
conformal freedom permeates all geometric structures at I. In particular, given a physical
space-time, the intrinsic metric and the derivative operator are known only up to conformal
transformations. On the other hand, since I is at infinity, in some ways its structure is both
more rigid and simpler. First, without loss of generality, we can assume that the metric
on S is a 2-sphere metric; the issue of distortion is physically irrelevant at I. Second, the
Weyl tensor vanishes identically at I and the curvature of the intrinsic connection captures
non-trivial information about the next order space-time curvature. By contrast, at ∆ only
four components of the Weyl curvature vanish and four other components are coded in the
curvature of the intrinsic connection on ∆. In spite of these differences, one can carry over
some techniques from null infinity to extract physical information about isolated horizons. In
particular, using the analogs of techniques which have been successful at I, one can introduce
preferred cross sections of and Bondi-type expansions near ∆ [18].
This concludes our analysis of the consequences of the boundary conditions defining non-
expanding horizons. Note that, even though ℓ is a Killing field for the intrinsic, degenerate
metric qab on ∆, it is not an infinitesimal symmetry for other geometrical fields such as the
intrinsic connection D or components of the curvature tensor. In the next sub-section, we
will make the structure more rigid by suitably restricting the choice of ℓ.

B. Weakly Isolated horizons

The time-independence of the intrinsic metric qab captures the idea that ∆ is isolated in
a suitable sense. While this condition has rich consequences, the resulting structure is still

9
not sufficient for physical applications. In particular, since ℓ can be rescaled by an arbitrary
positive definite function, the acceleration κ(ℓ) is not necessarily constant on ∆. Therefore,
we need to impose additional restrictions on the physical fields at ∆ to establish the zeroth
law. Since ℓ is already a symmetry of the intrinsic metric, it is natural to require it also be
a symmetry of the ‘extrinsic curvature’. However, the standard definition of the extrinsic
curvature is not applicable to null surfaces. Nonetheless, given a null normal ℓ, we can
construct a tensor field Ka b := Da ℓb , defined intrinsically on ∆, which can be thought of as
the analogue of the extrinsic curvature.5 Indeed, on a metric sub-manifold, if we replace ℓ
by the unit normal, Ka b is precisely the extrinsic curvature. It is then natural to demand
that, on an isolated horizon, Ka b also be time-independent: Lℓ Ka b = b 0. As a consequence
of (II.4), this is equivalent to imposing Lℓ ω = b 0.
Let us examine the above condition. As we will show at the end of this section, given
a non-expanding horizon we can always find a null normal ℓa which satisfies Lℓ ω = b 0. The
behavior of this condition under rescalings of ℓ is complicated by the fact that ω itself depends
upon the choice of null normal (see equation (II.6)). However, under a constant rescaling
ℓ 7→ ℓ̃ = cℓ, the connection 1-form ω is unchanged. Therefore, if ℓ satisfies the condition
Lℓ ω =b 0, so does any ℓ̃ related to ℓ by constant rescaling. This suggests we introduce an
equivalence relation: Two future-directed null normals ℓ and ℓ̃ belong to the same equivalence
class [ℓ] if and only if ℓ̃ = cℓ for some positive constant c.
The above considerations lead us to the following definition:
Definition 2: A weakly isolated horizon (∆, [ℓ]) consists of a non-expanding horizon ∆,
equipped with an equivalence class [ℓ] of null normals to it satisfying

Lℓ ω =
b 0 for all ℓ ∈ [ℓ]. (II.13)

As pointed out above, if this last equation holds for one ℓ, it holds for all ℓ in [ℓ].
A Killing horizon (with 2-sphere cross-sections) is automatically a weakly isolated hori-
zon, (provided the matter fields satisfy the energy condition of Definition 1). Given a non-
expanding horizon ∆, one can always find an equivalence class [ℓ] of null-normals such that
(∆, [ℓ]) is a weakly isolated horizon. However, condition (II.13) does not by itself single out
the appropriate equivalence class [ℓ]. As indicated in Section II D, one can further strengthen
the boundary conditions and provide a specific prescription to select the equivalence class
[ℓ] uniquely. However, for mechanics of isolated horizons, these extra steps are unnecessary.

5 We are grateful to Thibault Damour for pointing out that Ka b is called the Weingarten map and
is analogous to extrinsic curvature. This comment suggested the above motivation for our condition
on the connection 1-form ω. For an alternate, and in a sense weaker, condition see the remark at
the end of Section II D. From the viewpoint of intrinsic structures on ∆ discussed in Section II A, it
is perhaps more natural to ask that ℓ be a symmetry of the full intrinsic connection D (see section
II D and [18].) However, this stronger condition is not necessary for the laws of mechanics discussed
here.

10
In particular, our analysis will not depend on how the equivalence class [ℓ] is chosen. The
adverb ‘weakly’ in Definition 2 emphasizes this point.
The condition (II.13) has several consequences which are relevant for this paper.
(a) Surface gravity: In the case of Killing horizons ∆K , surface gravity is defined as the
acceleration of the Killing field ξ normal to ∆K . However, if ∆K is a Killing horizon for ξ,
it is also a Killing horizon for c ξ for any positive constant c. Hence, surface gravity is not
an intrinsic property of ∆K , but depends also on the choice of a specific Killing field ξ. (Of
course the result that the surface gravity is constant on ∆K is insensitive to this rescaling
freedom.) In asymptotically flat space-times admitting global Killing fields, this ambiguity is
generally resolved by selecting a preferred normalization in terms of the structure at infinity.
For example, in the static case, one requires the Killing field ξ to be unit at infinity. However,
in absence of a global Killing field or asymptotic flatness, this strategy does not work and
we simply have to accept the constant rescaling freedom in the definition of surface gravity.
In the context of isolated horizons, then, it is natural to keep this freedom.
A weakly isolated horizon is similarly equipped with a preferred family [ℓ] of null normals,
unique up to constant rescalings. Therefore, it is natural to interpret κ(ℓ) as the surface
gravity associated with ℓ. Under the permissible rescalings ℓ 7→ ℓ̃ = cℓ, the surface gravity
transforms via: κ(ℓ̃) = cκ(ℓ) . Thus, while ω is insensitive to the rescaling freedom in [ℓ], κ(ℓ)
captures this freedom fully. One can, if necessary, select a specific ℓ in [ℓ] by demanding that
κ(ℓ) be a specific function of the horizon parameters which are insensitive to this freedom,
e.g., by setting κ(ℓ) = 1/2R∆ , where R∆ is the horizon radius (related to the horizon area
2
a∆ via a∆ = 4πR∆ ).
(b) Zeroth law : The boundary conditions of Definition 2 allow us to define surface gravity κ(ℓ)
of a weakly isolated horizon (∆, [ℓ]). We will now show that the surface gravity is constant
on ∆. In other words, the zeroth law holds for weakly isolated horizons.
Recall from (II.12) that on a non-expanding horizon, dω = b 2 Im [Ψ2 ] 2ǫ for any choice of
null normal ℓ. Since 2ǫ is the pull-back to ∆ of the alternating tensor ǫ on the space S (of
orbits of ℓ), clearly ℓ · 2ǫ =
b 0. Therefore,

ℓ · dω =
b 0

for every null normal ℓ. In particular, on a weakly isolated horizon this equation holds for
any ℓ ∈ [ℓ]. Moreover, each of these restricted null normals also satisfies

0=
b Lℓ ω =
b d(ℓ · ω) + ℓ · dω

Hence, we conclude:

d(ℓ · ω) =
b d(κ(ℓ) ) =
b 0,

where we have used the definition (II.1) of κ(ℓ) . Thus, surface gravity is constant on ∆.
Although this proof of the zeroth law appears extremely simple, the argument is not as
trivial as it might first appear since we have used a number of consequences of the boundary

11
conditions derived in section II A. In contrast to earlier derivations [2,10] we do not require
the presence of a Killing field even in a neighborhood of ∆. Therefore the proof applies also
to space-times such as the Robinson-Trautman solutions [15] which do not admit a Killing
field. Also, ∆ need not be ‘complete’ — it may be of finite affine length with respect to
any ℓ — and may not admit the analog of a ‘bifurcate surface’ on which the Killing field
vanishes. Finally, the field equations are used rather weakly; we only need to assume that
−(Ra b − 21 Rδ a b )ℓb is a future directed causal vector.
Surface gravity does not have a definite value on a weakly isolated horizon. The value of
κ(ℓ) depends upon the choice of null normal ℓ ∈ [ℓ]. Since all the normals ℓ to ∆ are future
directed, the rescaling constant c is necessarily positive. Therefore, if the surface gravity is
non-zero (respectively, zero) with respect to one ℓ, it is non-zero (respectively, zero) with
respect to any other ℓ ∈ [ℓ]. This rescaling freedom is the same as the one discussed above
in the context of Killing horizons.
We will conclude this sub-section with three remarks.
i) Freedom in the choice of [ℓ]: Given a non-expanding horizon ∆, it is natural to ask if
one can always select an equivalence class [ℓ] of null normals such that (∆, [ℓ]) is a weakly
isolated horizon. As indicated earlier in this section, the answer is in the affirmative and,
furthermore, there is a considerable freedom in the choice of [ℓ]. Let us examine this issue
in some detail.
Since ℓ · dω = b 0 for any null normal ℓ to a non-expanding horizon, it follows that a null
normal ℓ satisfies Lℓ ω = b 0 if and only if dκ(ℓ) =
b 0. Thus, to find a family [ℓ] required in the
definition of weakly isolated horizons, it is necessary and sufficient to find a null normal ℓ
for which the surface gravity is constant. On a non-expanding horizon, the surface gravity
transforms as follows.
If ℓ 7→ ℓ̃ =
b f ℓ, then κ(ℓ̃) =
b f κ(ℓ) + Lℓ f.

Hence, starting with any ℓ, we can simply solve for f by requiring that κ(l̃) be constant on
∆. The solution is not unique. If κℓ is constant, given any non-zero function g satisfying
Lℓ g =
b 0 and a constant κ̃, let us set
κ̃
b g e−κ(ℓ) v +
f =
κ(ℓ)

where v satisfies Lℓ v =b 1. Then, we obtain an ℓ̃ 6∈ [ℓ] for which κ(ℓ̃) = b κ̃. This is the
only freedom if both κℓ and κ(ℓ̃) are to be constant. Thus, each non-expanding horizon
gives rise to an infinite family of weakly isolated horizons. Put differently, although one
can easily obtain weakly isolated horizons from non-expanding ones by choosing appropriate
null normals [ℓ], a specific weakly isolated horizon carries much more information than the
non-expanding horizon it comes from. At the end of Section II D, we will indicate how one
can further strengthen the boundary conditions to give a prescription for selecting a specific
[ℓ]. However, the analysis of this paper does not depend on how this selection is made.
ii) How does Definition 2 compare with that used in the undistorted, non-rotating case?
As one would expect, the definition given in [6–8] is significantly stronger. Furthermore, it

12
was tied to a foliation from the beginning. More precisely, it assumed that there exists a
foliation to which ω is normal, with ω = −κn, and it required that the expansion Re [µ] of the
null normal n to the leaves of the foliation be constant on ∆ (see appendix B for definitions
of the NP spin coefficients). Although it was shown that the foliation is unique if it exists,
the heavy reliance on the foliation right from the beginning made that definition less elegant
and its invariant content less transparent. Also, since we now allow dω to be non-zero and
impose no conditions on Re [µ], we can now incorporate rotation and distortion.
iii) Alternate boundary conditions: In the definition of isolated horizons, we required
Lℓ ω = b 0, which in particular implies Lℓ κ(ℓ) ≡ Lℓ (ℓ · ω) =
b 0. Thus, the definition itself
assured us that κ(ℓ) is ‘time independent’ and to prove the zeroth law we had to show that
it is also independent of ‘angles’. Could we have used another definition in which the ‘time
dependence’ of κ(ℓ) was not explicitly required but followed from other conditions? The
answer is in the affirmative: In place of Lℓ ω = b 0, we could have required that ∆ admit
a foliation on which the expansion Re [µ] of n and the Newman-Penrose spin coefficient π
which carries the angular momentum information are ‘time independent’ [13]. Then the
field equations would have implied that κ(ℓ) is time-independent. Furthermore, all results
of this paper go through (and were in fact first obtained) with these modified boundary
conditions. Note however that the new condition is neither weaker nor stronger than the
one we used. Both require that π be time independent. In addition, the present definition
of isolated horizons requires that κ(ℓ) be time independent while the alternative definition
would have required, instead, that Re [µ] be time independent. In this paper, we chose the
present definition because it can be stated without reference to a foliation.

C. Electromagnetic Field

We shall now describe the form of the electromagnetic field at an isolated horizon and
introduce a partial gauge fixing at the horizon which will allow us to introduce the notion
of an electric potential. In the next three sections — where we discuss the action, phase
space and first law — we will assume that the only matter field present at the isolated
horizon are Maxwell fields. However, as our discussion will make it clear, this restriction is
made primarily for simplicity. The overall framework is rather general and can accommodate
matter for which there exists a well-defined action principle and a (covariant) Hamiltonian
framework. In particular, in Section VI, we describe how to extend the formalism to include
Yang Mills Fields.
The isolated horizon boundary conditions restrict matter primarily through conditions
on the stress-energy tensor Tab . Let us begin with Tab ℓa ℓb =b 0, a direct consequence of
the boundary conditions and Raychaudhuri equation. (This restriction arises due to the
fact that ∆ is a non-expanding horizon; the subsequent stronger boundary conditions do
not further constrain F). Although this condition is weak, it turns out to have interesting
consequences on the form of the electromagnetic field, F, at ∆. The stress-energy tensor for
electromagnetism is given in terms of the field strength F as

13
 1 1
Tab = [Fac Fb c − gab Fcd Fcd ]. (II.14)
4π 4
Let us contract this expression with ℓa ℓb and examine the consequences for F. This gives
b Tab ℓa ℓb =|
0= b ℓa mb Fab |2 , (II.15)
where, to obtain the last expression, we have used the anti-symmetry of F and the fact that
the metric at the horizon can be expressed in terms of a null-tetrad as gab = −2ℓ(a nb) +
2m(a mb) . An immediate consequence of (II.15) is that
ℓa F =
b 0.
ab
←−−−−−
(II.16)
In order to obtain a similar expression for ⋆ F recall that the stress energy tensor can be
rewritten as Tab = − 4π [ Fac ⋆ Fb c − 14 gab ⋆ Fcd ⋆ Fcd ]. Applying the same argument which led to
1 ⋆

(II.16), we obtain a similar restriction on ⋆ F, namely

ℓ←−a⋆ F −− =
−−−−ab
b 0. (II.17)
These two restrictions tell us there is no flux of electromagnetic radiation across the horizon.
It is straightforward to show that (II.16), (II.17) and the form of the metric at ∆ place
further restrictions on Tab :
Tab ℓa mb =
b 0 Tab ℓa mb =
b 0
Tab ma mb =
b 0 Tab ma mb =
b 0.

The first two equations contain no new information since we already knew from general
arguments (see equation (II.3)), that Φ10 = b 0 and Φ01 =
b 0. However, the last two equations
do place further restrictions on the stress energy tensor. Since the equations of motion are
enforced at the boundary, we see immediately that they are equivalent to further restricting
the Ricci tensor at the horizon by requiring:
1 1
Φ02 := Rab ma mb =
b 0 and Φ20 := Rab ma mb =
b 0. (II.18)
2 2
This result need not hold for general matter fields; it relies on the properties of the electro-
magnetic stress-energy tensor (II.14).
The next task is to define the electric and magnetic charges of the horizon. Since the
horizon is an inner boundary of spacetime, the normal to a 2-sphere cross section of the
horizon will naturally be inward pointing. Bearing this in mind, we define the electric and
magnetic charges of the horizon as
I I
1 ⋆ 1
Q∆ :=
b − F and P∆ :=
b − F. (II.19)
4π S∆ 4π S∆

For these definitions to be meaningful, the values of Q∆ and P∆ should be independent of

the cross section of the horizon S∆ . We will now show that ∆ being a non-expanding horizon
guarantees this is the case. Let us first evaluate

14
 L ℓF
←
−
=
b ℓ · dF
←−−
+ d(ℓ · F)−.
←−−−−−−−

The first term on the right hand side vanishes due to Maxwell’s equations on ∆, while the
second term is zero due to the previous restriction on F, (II.16). Therefore we conclude
that F
←
−
is Lie dragged by ℓ. An identical argument for ⋆ F leads to the analogous conclusion.
Therefore we obtain
L ℓF
←
−
=
b 0 and L ℓ ⋆←−F− =
b 0. (II.20)
This result, along with (II.16) and (II.17), guarantees that Q∆ and P∆ are independent of
the choice of cross section S∆ of the horizon. Note that this result was obtained using only
the boundary conditions; equations of motion in the bulk are not needed.
Finally, let us examine the remaining freedom in the electromagnetic field. The boundary
conditions do not restrict Fab na mb and ⋆ Fab na mb at all. These components describe the
electromagnetic radiation flowing along the horizon. Therefore, isolated horizon boundary
conditions allow electromagnetic radiation arbitrarily close to —and even at— the horizon,
provided none crosses it.
So far we have confined ourselves to the field strengths F and ⋆ F. However, in the action
principle and the Hamiltonian framework we have to consider also the Maxwell potential A.
Now, if the magnetic charge is non-zero, either one has to allow ‘wire singularities’ in the
vector potentials or regard A as a connection on a non-trivial U(1)-bundle. (If we regard it
as a connection on a IR+ -bundle, the magnetic charge is necessarily zero.) Since we wish to
deal only with smooth fields, we will not allow ‘wire singularities’ in the potentials. If we
work with bundles, the magnetic charge is quantized whence the space of histories has several
disconnected components. Thus, in the first law, we will not be able to consider variations
δ of fields with δP 6= 0. As far as mechanics of isolated horizons is concerned, there is
essentially no loss of generality if we restrict ourselves to the case P∆ = 0. Therefore, in the
next three sections, while working with Maxwell fields, we will do so. As usual, our final
results can be formally extended to the case of non-vanishing magnetic charge by performing
a duality rotation on F. (As discussed in Section VI, the situation is rather different in the
case of Yang-Mills fields.)
Recall that the first law in the Einstein-Maxwell case involves the electro-static potential
Φ. In static space-times, one typically sets Φ = −ξ a Aa where ξ is the static Killing field and
the gauge is chosen such that the vector potential A tends to zero at infinity and satisfies
Lξ A = 0 everywhere in space-time. We now need a strategy to define the electric potential
Φ without reference to a Killing field or infinity. To this end, we introduce the following
definition:
Definition 3: The electromagnetic potential A will be said to be in a gauge adapted to the
weakly isolated horizon (∆, [ℓ]) if it satisfies
L ℓA
←−
=
b 0. (II.21)

Mathematically, this restriction is analogous to this condition L ℓ ω = b 0 imposed on the

gravitational field in Definition 2. However, while the condition on ω is a physical restriction

15
on the form of the gravitational field at ∆, the condition on A is a gauge choice; it can
always be imposed without physically constraining the electromagnetic field strength.
Given an electromagnetic potential A in a gauge adapted to (∆, [ℓ]), we can now define
the scalar potential Φ(ℓ) at the horizon in an obvious fashion:

Φ(ℓ) :=
b −ℓ · A.

The key question now is whether our boundary conditions are strong enough to ensure that
Φ(ℓ) is constant on ∆. Only then can we hope to use this notion of the scalar potential in the
first law. Note that this question is rather similar to the one we asked of surface gravity κ(ℓ)
in Section II B. By using arguments completely analogous to those that led us to the zeroth
law, we will now show that the answer to the present question is also in the affirmative. In
a gauge adapted to the horizon,

0=
b L ℓA
←−
= ℓ−·−−F− − dΦ
b ← ←−−(ℓ)
.

As we saw above, the boundary conditions imply ℓ←−·−−F− = b 0 (equation (II.16)). Hence, it
follows immediately that Φ(ℓ) is constant on the horizon. We can regard this result as the
‘electromagnetic part’ of the zeroth law of isolated horizon mechanics.
We will see in the next section that these zeroth laws play a key role in making the
gravitational and the electromagnetic action principles well-defined in presence of isolated
horizons. As with surface gravity κ(ℓ) , the functional dependence of Φ(ℓ) on the horizon
parameters varies with the choice of ℓ ∈ [ℓ]. We will see in Section V that the Hamiltonian
framework constrains these dependencies in an interesting fashion.

D. Other definitions and remarks

In this sub-section, we introduce two new definitions which are important to the general
framework of isolated horizons.
The first is concerned with rotation. From one’s experience with the Newman-Penrose
framework, one expects the gravitational contribution to angular momentum to be coded in
the imaginary part of Ψ2 . This expectation will be shown to be correct in [13]. Therefore,
in the Einstein-Maxwell theory, we introduce the following definition:
Definition 4: A weakly isolated horizon (∆, [ℓ]) will be said to be non-rotating if Im [Ψ2 ]
vanishes on ∆.

If (∆K , [ξ]) is a Killing horizon and ξ is a hypersurface orthogonal, time-like vector field near
∆K , on physical grounds one would expect the horizon to be non-rotating. Is this expectation
compatible with our definition? The answer is in the affirmative. For, in this case, one can
show that Bab := Cacbd ξ c ξ d vanishes in the region where ξ a is time-like. Hence, by continuity,
it also vanishes on ∆ forcing Im [Ψ2 ] to vanish there. Similarly, if the space-time admits a
hypersurface orthogonal, rotational Killing field ϕa in a neighborhood of ∆ Im [Ψ2 ] again

16
vanishes on ∆. The definition is again compatible with one’s intuition that the horizon
should be non-rotating in this case. In this paper, while we allow for presence of rotation in
the first four sections, we will restrict ourselves to non-rotating horizons in the proof of the
first law in Section V.
Finally, for completeness, let us introduce a stronger notion of ‘isolation’ by strengthening
the boundary conditions of Definition 2.
Definition 5: A weakly isolated horizon (∆, [ℓ]) is said to be isolated if

[Lℓ , D]V =
b 0 (II.22)

for all vector fields V tangential to ∆ and all ℓ ∈ [ℓ]

As before, if any one ℓ satisfies this condition, so do all ℓ ∈ [ℓ]. However, unlike (II.13),
condition (II.22) is a genuine restriction in the sense that it can not always be met by a
judicious choice of null normals. Generically it does suffice to single out the equivalence class
[ℓ] uniquely [18]. In particular, in the Kerr family, the only [ℓ] which satisfies (II.22) is the
one containing constant multiples of the globally defined Killing field which is orthogonal to
the horizon. Every Killing horizon is of course an isolated horizon. Thus, even though (II.22)
is a stronger condition than (II.13), it is still very weak compared to conditions normally
imposed. For most physical applications, e.g., to numerical relativity, it is appropriate to
work with isolated horizons. For mechanics of isolated horizons, however, we can —and
will— work with the larger class of weakly isolated horizons.
The following consequences of the boundary conditions defining weakly isolated horizons,
derived in this section, will play an important role in the subsequent discussion:

b ω a ℓb ;
a) ∇ a ℓb =
←
−

b ℓa ωa , the surface gravity defined by the null normal ℓa , is constant on ∆;

b) κ(ℓ) =

c) There is a natural (area) 2-form 2ǫ on ∆ satisfying L ℓ 2ǫ =

b 0 and 2ǫab ℓb =
b 0.

d) The electromagnetic potential A is chosen to satisfy L ℓ A =

b 0 and in this gauge the
a
scalar potential Φ(ℓ) := −ℓ Aa is constant on ∆; and

e) The electromagnetic field satisfies ℓ←a−−F−−ab− =

b 0; and ℓ←−a⋆ F −− =
−−−−ab
b 0.

III. ACTION

In this paper we use the first order formulation of general relativity in terms of tetrads
and connections. Since tetrads are essential to incorporate spinorial matter, it is natural to
base the framework on tetrads from the beginning. The use of a first order formalism, on the
other hand, is motivated primarily by mathematical simplicity. In the first order framework,
the action and the Hamiltonians can be expressed entirely in terms of differential forms
which significantly simplify the variational calculations. The previous paper [8] which dealt

17
with undistorted horizons used spinors and self-dual connections, while here we choose to
use orthonormal tetrads and real, Lorentz connections. For analyzing mechanics of isolated
horizons, there are two advantages to this. First, the Hamiltonian and symplectic structure
are now manifestly real which simplifies evaluation of the boundary terms at the horizon.
Second, the analysis can now be extended to other space-time dimensions in a straightforward
manner. However, these simplifications, come with a price. Since, at present, the self dual
variables appear to be indispensable for non-perturbative quantization, the results obtained
here will have to be re-expressed in terms of self-dual variables in order to extend the analysis
[20] of the quantum horizon geometry and black hole entropy to include rotation.

A. Preliminaries

Let us begin with the first order action for Einstein-Maxwell theory on a 4-dimensional
manifold M which is topologically M × IR, where M is an oriented Riemannian 3-manifold
without boundary (the complement of a compact set of) which is diffeomorphic to (the
complement of a compact set of) IR3 . Thus, topological complications of M, if any, are
confined to a compact set. In this subsection we shall only give the relevant formulae. For
details, see e.g. [21]. Our basic fields will consist of a triplet (eIa , AaI J , Aa ) defined on M
where eIa denotes a co-tetrad, AaI J the gravitational (Lorentz) connection and Aa the electro-
magnetic connection. Here, lower case latin letters refer to the tangent space of M while
the upper case letters I, J etc. refer to an internal four dimensional vector space V with a
fixed metric ηIJ of signature (− + ++). The co-tetrad eIa is an isomorphism between the
tangent space Tp (M) at any point p and the internal space V . Using it, we define a metric
on M by gab := eIa eJb ηIJ which also has signature (− + ++). The Lorentz connection AaI J
acts only on internal indices and defines a derivative operator

Da kI := ∂a kI + AaI J kJ ,

where ∂ is a fiducial derivative operator which, as usual, will be chosen to be flat and torsion
free. Finally, Aa is the U(1) electromagnetic connection 1-form on M. (As noted in Section
II C, we will assume that the magnetic charge is zero.) All fields will be assumed to be
smooth and satisfy the standard asymptotic conditions at infinity.
The 2-forms ΣIJ
1
ΣIJ := ǫIJKL eK ∧ eL
2
constructed from the co-tetrads will play an important role throughout our calculations. In
particular, the action for an asymptotically flat space-time (with no internal boundary) is
given by (see e.g. [21])
Z Z Z
−1 1 1
S(e, A, A) = ΣIJ ∧ FIJ + ΣIJ ∧ AIJ − F ∧ ⋆F . (III.1)
16πG M 16πG τ∞ 8π M

18
Here F and F are the curvatures of the gravitational and electromagnetic connections A and
A respectively:

FI J = dAI J + AI K ∧ AK J and F = dA ,
⋆
Fab = 12 ǫab cd Fcd is the dual of F defined using ea I , and τ∞ is the time-like cylinder at
infinity. The boundary term at τ∞ ensures the differentiability of the action.
Let us briefly examine the equations of motion arising from the action. Varying the
action with respect to the connection, one obtains

DΣ = 0.

This condition implies that the connection D defined by A has the same action on internal
indices as the unique connection ∇ compatible with the co-tetrad, i.e., satisfying ∇a eIb = 0.
When this equation of motion is satisfied, the curvature F is related to the Riemann curvature
R of ∇ by

Fab IJ = Rab cd eIc eJd .

Varying the action with respect to eIa and taking into account the above relation between
curvatures, one obtains Einstein’s equations
 
1
Gab = 8πGTab cd
≡ 2G Fac Fbd g − gab Fcd Fcd
4
where Gab is the Einstein tensor and Tab the electromagnetic stress energy tensor. Finally,
variation with respect to the electromagnetic connection, A, yields Maxwell’s equation

d⋆ F = 0 .

B. Internal boundary ∆

Let us now consider the variational principle for asymptotically flat histories which admit
a weakly isolated horizon ∆ as their internal boundary. The manifold M under consideration
has an internal boundary ∆, topologically S 2 × IR. As before, M is topologically M × IR,
where M is now an oriented manifold with an internal, 2-sphere boundary, whose topological
complications are again confined to a compact region. Space-time is bounded to the future
and past by two (partial Cauchy) surfaces M ± , extending to spatial infinity (see Figure 1).
Following Definition 2 of weakly isolated horizons, we will equip ∆ with a fixed equivalence
class of vector fields [ℓ] which are transversal to its 2-sphere cross-sections (where, as before,
ℓ ∼ ℓ′ if and only if ℓ = b cℓ′ for a constant c). It is also convenient to fix an internal null
I I I I
tetrad (ℓ , n , m , m ) on ∆, each element of which is annihilated by the fiducial, flat internal
connection ∂.

19
 +
S

∆ +
M
M

-
S
-
M
io
FIG. 1. The region of space-time M under consideration has an internal boundary ∆ and is
bounded by two partial Cauchy surfaces M ± which intersect ∆ in the 2-spheres S ± and extend to
spatial infinity io .

The permissible histories consist of smooth triplets (e, A, A) on M satisfying boundary

conditions at infinity and on ∆. The boundary conditions at infinity are, as before, the
standard ones which ensure asymptotic flatness. Since the asymptotic behavior and bound-
ary integrals at infinity play only a secondary role in our analysis, we shall not spell out
the precise fall-off requirements. At ∆, the histories are subject to three conditions: i) the
tetrads e should be such that the vector field ℓa := ℓI eaI defined by each history belongs to
the equivalence class [ℓ] fixed on ∆; ii) the tetrad e and the gravitational connection A should
be such that (∆, [ℓ]) is a weakly isolated horizon for the history; and, iii) the electromagnetic
potential A is in a gauge adapted to the horizon, i.e., Lℓ A = b 0.
Remark: In space-time, we have the freedom to perform a local, internal Lorentz rotation
on the tetrad eaI (and the gravitational connection AaI J ). All these tetrads define the same
Lorentzian metric gab . Since ℓa is required to be a null normal to ∆, the permissible gauge
rotations are reduced on ∆ to the sub-group (IR+ × E 2 )loc of local null rotations preserving
the null direction field ℓ. (Here IR+ is the group of rescalings of ℓa , na which leaves ma fixed
and E 2 is the 3-dimensional Euclidean group consisting of rotations in the ℓ-m, ℓ-m and
m-m planes.) Condition i) above — dictated by the existence of a preferred equivalence
class [ℓ] in Definition 2 — further reduces the internal gauge freedom to IR+ × (Eloc 2
), i.e.,
+ +
reduces the group IRloc of local ℓ-n rescalings to the group IR of global rescalings. Thus,
while any one space-time (M, gab ), still defines infinitely many histories due to the freedom
of tetrad-rotations, this freedom is somewhat reduced at ∆ because of the structure fixed by
the boundary conditions.6

6 Nonetheless, from a space-time perspective, the multiplicity of histories can still be rather sur-

20
 Given any tetrad eaI , the internal null vectors (ℓI , nI , mI , mI ) fixed on ∆ trivially provide
a null tetrad (ℓa , na , ma , ma ). In terms of these vectors, we can express ← Σ−IJ as:
IJ
Σ
←
−
b 2ℓ[I nJ] 2ǫ + 2n ∧ (imℓ[I mJ] − imℓ[I mJ] ),
= (III.2)

where, as before, 2ǫ = im ∧ m is the pull-back to ∆ of the natural alternating tensor on

the 2-sphere S of integral curves of ℓa associated with the given history. The weak isolation
of (∆, [ℓ]) restricts the form of the connection A at ∆. To see this, recall that one of the
equations of motion requires the connection D defined by A to have the same action on
internal indices as ∇. Hence, ∇a ℓI = b ∂a ℓI + A a I J ℓJ =
b A a I J ℓJ where, in the second step
←
− ←
− ←−
we have used the fact that the flat derivative operator, ∂, has been chosen to annihilate the
internal tetrad on ∆. Since ∇a eIb = 0 by definition of ∇, and ∇a ℓb = b ωa ℓb (see II.4) it follows
←
−
that A
←−IJ
ℓJ =
b ωℓI . Hence, on ∆, A←−
has the form:

A
−IJ
←
=
b −2l[I nJ] ω + CIJ , (III.3)

where the 1-form CIJ satisfies CIJ ℓJ =b 0.

With this background material at hand, we are now ready to consider variations of the
action (III.1) in the presence of an inner boundary representing a weakly isolated horizon.
A key question is whether a new surface term at the horizon is necessary to make the
variational principle well-defined. We will show that, thanks to the zeroth law, such a term
is not needed.
A simple calculation yields:
Z Z Z
1 1
δS(e, A, A) = Equations of Motion · δφ − ΣIJ ∧ δAIJ − δA ∧ ⋆ F. (III.4)
M 16πG ∆ 4π ∆

where, in the first term, φ stands for the basic fields (e, A, A) in the action. Note that, as
in the case without an internal boundary, the variation of the boundary term at infinity
precisely cancels the contribution arising from the variation of the bulk terms.
In order to show that the action principle is viable, it is necessary to show that the terms
at the horizon vanish due to the boundary conditions imposed there. Let us begin with the
gravitational term. Using (III.2) and (III.3) it can be re-expressed as
Z
1
− δω ∧ 2ǫ. (III.5)
8πG ∆

Since 2ǫ is the pull-back to ∆ of the alternating tensor on the 2-sphere S of integral curves of
ℓ, it follows that Lℓ 2ǫ =
b 0. Furthermore, the weak isolation of the horizon ensures Lℓ ω = b 0

prising. For example, if gab is the Schwarzschild metric with mass M > 0, there is a history in
which the surface gravity κ(ℓ) is positive and another in which it is zero. This redundancy can be
eliminated by working with isolated, rather than weakly isolated horizons.

21
and, since the null normal ℓa defined by any tetrad belongs to the fixed equivalence class [ℓ]
at the horizon, we have δℓ = b cδ ℓ for some constant cδ . These two facts imply Lℓ δω =
b 0. Thus
the entire integrand is Lie dragged by ℓ. In the variational principle, however, all fields are
fixed on the initial and final hypersurfaces, say M ± . In particular, δω necessarily vanishes on
the initial and final cross sections of the horizon. Therefore, the integrand in (III.5) vanishes
on the initial and final cross sections and is Lie dragged by ℓ. This immediately implies
(III.5) is zero.
Let us now consider the electromagnetic term. Since every A is in a gauge adapted to the
isolated horizon, Lℓ A = b 0. Furthermore, δℓa = b cδ ℓa , so we conclude Lℓ δA =
b 0. Next, (II.20)
⋆
ensures Lℓ F = b 0. Thus, the integrand of the electromagnetic surface term is Lie dragged
a
by ℓ . An identical argument to the one presented above implies that the electromagnetic
surface term in (III.4) also vanishes. Therefore, the variation of the action (III.1) continues to
yield Einstein-Maxwell equations in spite of the presence of an inner boundary representing
a weakly isolated horizon.
It is instructive to re-examine the key step in the above argument. Suppose we only had a
non-expanding horizon. Then, the gravitational surface term could still be reduced to (III.5),
and 2ǫ and ⋆←−F− would still be Lie-dragged by ℓ. However, in this case, we could not argue
that ω and A are also Lie-dragged. As we saw in Sections II B and II C, these conditions are
equivalent, respectively, to the constancy of the surface gravity κ(ℓ) and the electromagnetic
potential Φ(ℓ) on ∆. In this sense, given a non-expanding horizon as the inner boundary,
the gravitational and electromagnetic zeroth laws are the necessary and sufficient conditions
one must impose for the viability of the standard, first order, tetrad action principle.
Remark: Note that (III.1) is not the unique viable action for the problem: as usual, there
is freedom to add suitable boundary terms without affecting the viability. Specifically, we
are free to add any horizon boundary term which is composed entirely of fields which are
Lie dragged by ℓ, for example the intrinsic horizon metric qab and fields ω, 2ǫ, and A ←−
. Then,
due to the argument given above, the new action would also be viable. However, as usual,
this freedom will not affect the definition of the symplectic structure which underlies the
Hamiltonian treatment of the next section.

IV. COVARIANT PHASE SPACE

Let us now construct the phase space of space-times containing weakly isolated hori-
zons. In the next section, we will use this framework to construct Hamiltonians generating
suitable time translations and define the energy of an isolated horizon. In [8], the phase
space was constructed by performing a Legendre transform of the action. This procedure
leads to a ‘canonical’ framework in which the phase space consists of configuration and mo-
mentum variables defined on a spatial hypersurface. With the self-dual connections used in
[8], the gravitational configuration variable turns out to be a connection and its conjugate
momentum, a 2-form so that the Hamiltonian description can again be given in terms of
forms. With the full Lorentz connections now under consideration, the situation turns out

22
to be more complicated. Specifically, one encounters certain second class constraints and,
when these are solved, one ends up with the same canonical phase space that one would
have obtained through a second order formalism. In the Hamiltonian framework, then, the
simplicity we encountered in Section III is lost. More specifically, constraint functions and
Hamiltonians now contain terms involving second derivatives of the basic canonical vari-
ables which make variations rather complicated. (For details, see chapters 3 and 4 in [21].)
Therefore, in this section we will not use a Legendre transform. Instead, we will construct
the ‘covariant phase space’ from the space of solutions to field equations (see, e.g., [22]). As
in Section III, all expressions will now involve only the basic form-fields and their exterior
derivatives and variational calculations will continue to be simple.
To specify the phase space, let us begin as in Section III by fixing a manifold M with
an internal boundary ∆ (see figure 1). As before, we will equip ∆ with an equivalence
class [ℓ] of vector fields transverse to its 2-sphere cross-sections. To evaluate the symplectic
structure and Hamiltonians, we will often use a partial Cauchy surface M in the interior of
M which intersects ∆ in a 2-sphere S. Points of the covariant phase space Γ will consist of
histories considered in Section III which satisfy field equations. More explicitly, Γ consists
of asymptotically flat solutions (e, A, A) to the field equations on M such that i) the vector
field ℓa := ℓI eaI belongs to the equivalence class [ℓ] fixed on ∆, ii) in each solution, (∆, [ℓ]) is
a weakly isolated horizon; and, iii) the electromagnetic potential A is in a gauge adapted to
the horizon, i.e., Lℓ A = b 0.
Our next task is to use the action (III.1) to define the symplectic structure Ω on Γ. It is
convenient to make a brief detour and first introduce two new fields which can be regarded
as ‘potentials’ for the surface gravity κ(ℓ) and the electric potential Φ(ℓ) . Given any point
(e, A, A) in the phase space Γ, let us define scalar fields ψ and χ on ∆ as follows:
i) Lℓ ψ =b (ℓ · ω) =b κ(ℓ) and Lℓ χ =
b (ℓ · A) =
b −Φ(ℓ) ; and
ii) ψ and χ vanish on S , the intersection of M − with ∆.7
−

Note that these conditions associate with each point of Γ a unique pair (ψ, χ) on ∆ and in
the ‘extremal’ case κ(ℓ) = 0, ψ vanishes identically.
We wish to use the standard procedure [22] involving second variations of the action to
define the symplectic structure.8 Let us recall the main steps of this procedure. One first
constructs the symplectic current J: Given a point γ in the phase space Γ and two tangent

7 Condition ii) serves only to fix the freedom to add constants to ψ and χ. One could envisage
replacing it by a different condition. Our results will be insensitive to this choice.

8 Actually this procedure provides a pre-symplectic structure, i.e., a closed 2-form on the phase-
space which, however, is generally degenerate. The vectors in its kernel represent infinitesimal
‘gauge transformations’. The physical phase space is obtained by quotienting the space of solutions
by gauge transformations and inherits a true symplectic structure from the pre-symplectic structure
on the space of solutions. The 2-form Ω introduced below is indeed degenerate. However, for
simplicity, we will abuse the notation somewhat and refer to Ω as the symplectic structure.

23
vectors δ1 and δ2 at that point, J provides a closed 3-form J(γ; δ1 , δ2 ) on M. Integrating dJ
over the part M̃ of space-time under consideration, one obtains
Z I
0= d J(γ; δ1, δ2 ) = J.
M̃ ∂ M̃

Now, if there is no internal boundary, one can choose M̃ to be a region bounded by any two
Cauchy surfaces M1 and M2 so that the boundary is given by ∂ M̃ = M1 ∪ M2 ∪ τ∞ , where
τ∞ is the time-like
R
‘cylinder at infinity’. In simple cases, the asymptotic conditions ensure
that the
R
integral τ∞ J(γ; δ1 , δ2 ) vanishes. Then, taking orientations into account, it follows
that M J(γ; δ1 , δ2 ) is independent of the choice of Cauchy surface M. One then sets the
symplectic structure to be
Z
Ω|γ (δ1 , δ2 ) = J(γ; δ1 , δ2 ) .
M

In our case, the second variation of the action (III.1) yields the following symplectic
current:
−1 1 ⋆
J(γ; δ1 , δ2 ) = [δ1 ΣIJ ∧ δ2 AIJ − δ2 ΣIJ ∧ δ1 AIJ ] − [δ1 F ∧ δ2 A − δ2 ⋆ F ∧ δ1 A]. (IV.1)
16πG 4π
Using the fact that the fields γ ≡ (e, A, A) satisfy the field equations and the tangent vectors
δ1 , δ2 satisfy the linearized equations off γ, one can directly verify that J(γ; δ1 , δ2 ) is in fact
closed as guaranteed by the general procedure involving second variations. It is now natural
to choose M̃ to be a part of our space-time M bounded by two partial Cauchy surfaces
M1 , M2 , the time-like cylinder τ∞ and a part ∆ ˜ of the isolated horizon bounded by M1 and
M2 . Again, the asymptotic conditions ensure that the integral of J over τ∞ vanishes. Hence,
Z Z Z
( + + )J(γ; δ1 , δ2 ) = 0.
M1 M2 ˜
∆

However, this does not immediately provide us the conserved symplectic structure because
˜ does not vanish in general. Since the isolation of the horizon implies
the integral of J over ∆
that there are no fluxes of physical quantities across ∆, one might expect that, although
non-zero, the integral over ∆ ˜ would be ‘controllable’. This is indeed the case. Using the
forms (III.2) and (III.3) of ←
Σ− and A
←
−
on ∆ and the definitions of the potentials ψ and χ, it is
easy to verify that the pull-back of the symplectic current to ∆ is itself exact:

J
←
−
(γ; δ1 , δ2 ) =
b dj(γ, δ1 , δ2 )

where the 2-form j on ∆ is given by:

1 1
j(γ, δ1 , δ2 ) = [δ1 ψ δ2 (2ǫ) − δ2 ψδ1 (2ǫ) ] + [δ1 χδ2 ⋆ F − δ2 χδ1 ⋆ F].
8πG 4π
˜ in 2-spheres S1 and S2 respectively, we have:
Hence, if M1 and M2 intersect ∆

24
 Z Z Z
J(γ; δ1 , δ2 ) = −( + )j(γ; δ1 , δ2 ).
˜
∆ S1 S2

The negative sign appearing in the above expression is due to the choice of orientation of
S∆ , which is induced from M rather than from ∆. Using these results we can define the
symplectic structure as:
Ω|γ (δ1 , δ2 ) =
Z I
−1 1
[δ1 ΣIJ ∧ δ2 AIJ − δ2 ΣIJ ∧ δ1 AIJ ] + [δ1 (2ǫ) δ2 ψ − δ2 (2ǫ) δ1 ψ]
16πG M 8πG S
Z I
1 1
− [δ1 ⋆ F ∧ δ2 A − δ2 ⋆ F ∧ δ1 A] + [δ1 ⋆ F δ2 χ − δ2 ⋆ F δ1 χ] (IV.2)
4π M 4π S
Again, using field equations one can directly verify that the right side of (IV.2) is independent
of the choice of the partial Cauchy surface; the symplectic structure is ‘conserved’. We will
use (Γ, Ω) as our covariant phase space.
Note that, even though the action did not contain a surface term at the horizon, the sym-
plectic structure does. So, the overall situation is the same as in the undistorted, non-rotating
case considered in [8]. Finally, our discussion of the action principle and our construction
of the covariant phase space is applicable to all weakly isolated horizons ∆; nowhere did we
have to restrict ourselves to the non-rotating case.

V. HAMILTONIAN EVOLUTION AND THE FIRST LAW

To discuss the first law, we must first define horizon energy, which in turn requires a time
evolution field ta on M. Given a vector field ta satisfying appropriate boundary conditions,
δt := (L t e, L t A, L t A) satisfies the linearized equations for any γ := (e, A, A) in Γ and thus
defines a vector field on Γ.9 This δt can be interpreted as the infinitesimal generator of time
evolution on the covariant phase space. It is then natural to ask if this vector field is a
phase space symmetry, i.e., if L δt Ω vanishes everywhere on Γ. The necessary and sufficient
condition for this to happen is that there exist a function Ht —the Hamiltonian generating
the t-evolution— such that
δHt = Ω(δ, δt ) (V.1)
for all vector fields δ to Γ. On general grounds, one expects Ht to contain a surface term
t t
EADM at infinity representing the total (i.e. ADM) energy, and a surface term E∆ at the
horizon which can be interpreted as the horizon energy, both tied to the evolution field ta .

9 Inthe Lie-derivatives, the internal indices are treated as scalars; thus L t eIa = tb ∂b eIa + eIb ∂a tb . To
make δt a well-defined vector field on Γ, we now exclude M ± from M and let M and ∆ be without
future and past boundaries. Whenever needed, these boundaries, M ± and S ± , can be added by
taking the obvious closure of M.

25
 A key question then is to specify the appropriate boundary conditions on ta . It is clear
that, at infinity, ta should be an asymptotic time-translation, i.e., should approach a time-
translation Killing field of the flat metric used to specify the boundary conditions. At the
horizon, on the other hand, the metric is not universal and the space-time defined by a generic
point γ of the covariant phase space does not admit any Killing field near ∆. Therefore,
specification of the boundary conditions at ∆ is not as straightforward as that at infinity. It
is for this reason that we now assume that (∆, [ℓ]) is a non-rotating, weakly isolated horizon
for all points γ ≡ (e, A, A) of the phase space. The problem of specifying the appropriate
boundary conditions on ta in the rotating case is more complicated . However, it has been
addressed successfully and will be discussed in [13].
Recall that the internal boundary ∆ of M is equipped with a specific equivalence class
[ℓ] of vector fields. As discussed in Section II, these ℓ are the isolated horizon analogs of
constant multiples of Killing fields on the Killing horizons in static space-times. Therefore,
in the non-rotating case, it is natural to demand that, on ∆, the evolution vector field ta
should belong to the equivalence class [ℓ]. This automatically ensures that (L t e, L t A, L tA)
satisfy the appropriate boundary conditions to define a tangent vector at each point of the
phase space Γ. However, unlike at infinity, the geometry at the horizon is not fixed once
and for all. Therefore, it is natural to allow the precise value of the evolution vector field
ta on ∆ to vary from one point of the phase space to another. In more familiar terms,
this corresponds to allowing the (boundary values of) lapse and shift fields to depend on
dynamical fields (e, A, A) themselves, a procedure routinely used in numerical relativity and
gauge-fixed calculations in canonical gravity. Following the current terminology in numerical
relativity, we will refer to such ta as live evolution vector fields. The use of live fields turns
out to be necessary to ensure that δt is a phase space symmetry, i.e., yields a Hamiltonian
evolution on (Γ, Ω).
Let us fix a live ta whose restriction to the horizon belongs to the equivalence class [ℓ] at
all points of the phase space. To analyze if δt is a Hamiltonian vector field, it is simplest to
compute the 1-form Xt on Γ defined by
Xt (δ) = Ω(δ, δt ). (V.2)
Now δt is Hamiltonian — i.e., L δt Ω = 0 on Γ — if and only if Xt is closed, i.e.,
ddXt = 0
where dd denotes the exterior derivative on (the infinite dimensional) phase space Γ. If this
is the case then, up to an additive constant, the Hamiltonian is given by
ddHt = Xt .
To calculate the right side of (V.2), it is useful to note the following identities from
differential geometry:
L t A = t · F + D(t · A) L t Σ = t · DΣ + D(t · Σ) − [(t · A), Σ]
(V.3)
L t A = t · F + d(t · A) L t ⋆ F = t · (d ⋆ F) + d(t · ⋆ F)

26
Using these, the field equations satisfied by (e, A, A) and the linearized field equations for
δ, we obtain the required expression of Xt :

Xt (δ) := Ω(δ, L t ) (V.4)

Z Z
−1 1
= Tr [(t · A)δΣ − (t · Σ) ∧ δA] − (t · A)δ(⋆ F) − (t · ⋆ F) ∧ δA . (V.5)
16πG ∂M 4π ∂M
Note that the expression involves integrals only over the 2-sphere boundaries S∞ and S∆ of
M, the partial Cauchy surface used in the evaluation of the symplectic structure; there is no
volume term.
The integrals at infinity can be evaluated easily by making use of the fall-off conditions.
As one would expect, the electromagnetic term vanishes (because A falls off at least as 1/r
while F falls off as 1/r 2 ) while the gravitational term yields precisely the ADM energy EADM
t
a
associated with the asymptotic time-translation defined by t . At the horizon, we can use
equations (III.2) and (III.3) to show that t·Σ contracted on internal indices with δA vanishes
and (II.17) implies t · ⋆ F = 0 leaving
Z Z
1 1
Xt (δ) = − (t · ω)δ(2ǫ) − (t · A)δ(⋆F) + δEADM
t
8πG S∆ 4π S∆
1 t
=− κ(t) δa∆ − Φ(t) δQ∆ + δEADM (V.6)
8πG
where, in the last step, we have used the fact that both t · ω = κ(t) and t · A = −Φ(t) are
constant on the horizon and the definition (II.19) of electric charge.
The necessary and sufficient condition for the existence of a Hamiltonian is that Xt be
closed. Clearly, this is equivalent to
1
ddκ(t)∧
∧ dda∆ + ddΦ(t)∧
∧ ddQ∆ = 0 , (V.7)
8πG
where∧∧denotes the antisymmetric tensor product on Γ. Now, (V.7) trivially implies that the
surface gravity κ(t) and the electric potential Φ(t) at the horizon defined by ta can depend
only upon the area and charge of the horizon. Other factors, such as the ‘shape’ of the
distorted horizon, can not affect the values of κ(t) or Φ(t) . Finally, (V.7) is the necessary and
t
sufficient condition that there exists a function E∆ , also only of a∆ and Q∆ such that

t 1
δE∆ = κ(t) δa∆ + Φ(t) δQ∆ . (V.8)
8πG
t
Since E∆ is a function only of a∆ and Q∆ , it is a function of fields defined locally at the
t
horizon. As noted before, it is natural to interpret E∆ as the horizon energy defined by the
a
time translation t . The total Hamiltonian is given by:
t t
Ht = EADM − E∆ . (V.9)

Let us summarize. Eq (V.8) is a necessary and sufficient condition for the 1-form Xt to
be closed. Therefore, the vector field δt on Γ defined by the space-time evolution field ta is

27
Hamiltonian if and only if the first law (V.8) holds. Thus (V.8) is a restriction on the choice
of the live vector field ta : While any ta (which preserves the boundary conditions) defines
an evolution flow on the phase space, it is only when
1
κ(t) dda∆ + Φ(t) ddQ∆
8πG
is an exact 1-form on Γ that this flow is Hamiltonian (i.e., preserves the symplectic struc-
ture). At first, this restriction seems somewhat surprising because, in absence of internal
boundaries, every vector field ta (which tends to a fixed Killing field of the flat metric at
infinity) defines a Hamiltonian evolution. However, even in this context, there is no a priori
reason to expect this tight correspondence to hold if one allows general, live vector fields
ta whose boundary values at infinity can change from one space-time to another. Finally,
we will see in Section VII that every space-time belonging to the phase space Γ admits an
infinite family of vector fields ta for which Xt is closed. Therefore, in particular, the first
law does not restrict the ‘background’ space-times (or the variations δ) in any way. Indeed,
for any space-time in our phase space, there is an infinite family of first laws, one associated
with each permissible ta .
We will conclude this section with a few remarks.
i) Form of Ht : The Hamiltonian (V.9) contains only surface terms. This may seem
surprising because, in the canonical framework, the familiar Hamiltonian contains a volume
integral consisting of a linear combination of constraints. While the volume term vanishes ‘on
shell’ and does not contribute to numerical value of the canonical Hamiltonian on physical
states, it is nonetheless crucial for obtaining the correct evolution equations since derivatives
of the Hamiltonian transverse to the constraint surface are needed to construct the Hamil-
tonian vector field. The covariant phase space, by contrast, consists only of solutions to the
field equations whence the issue of taking ‘off shell’ derivatives never arises. In diffeomor-
phism invariant theories, the Hamiltonian on the covariant phase space is always made of
surface terms.10 If space-time has several asymptotic regions, the boundary term in each
region defines the standard energy corresponding to that region. Therefore, in the present
t
case, it was natural to interpret E∆ as the horizon-energy defined by the ta evolution. Fi-
nally, we should emphasize that we used a covariant phase space only for simplicity. The
final results go through (and, in fact, were first obtained) in a canonical framework as well.
ii)Comparisons: As noted in the Introduction, all treatments of the first law for non-
rotating but possibly distorted horizons available in the literature refer to static space-times.
The isolated horizon framework, by contrast, does not refer to a Killing field at all and thus
allows a significantly larger class of physically interesting situations. On the other hand,

10 In particular, therefore, Hamiltonians generating diffeomorphism which have support away from
the boundaries vanish identically. Unlike their counterparts on the canonical phase space, the
infinitesimal phase-space motions induced by such space-time vector fields are in the kernel of the
covariant symplectic structure.

28
since it relies on a Hamiltonian framework, we cannot incorporate phenomenological matter
if it does not admit a phase space description. Other treatments based on Hamiltonian
methods generally restrict themselves to static space-times with a non-zero surface gravity.
This assumption is essential there because those treatments use ‘bifurcate’ surfaces in an
important way and these do not exist in the extremal static solutions where the surface
gravity vanishes. In contrast, the results of this section, do not refer to a bifurcate surface
and go through irrespective of whether κ(t) is non-zero or zero. In a realistic collapse, the
physical space-time is not expected to have the bifurcate surface. The present analysis uses
only the portion of the physical space-time in which the horizon has settled down with
no further in-going radiation, rather than an analytical continuation of the near horizon
geometry used in certain approaches. Finally, in contrast to other treatments, we have an
infinite family of first laws, one for each evolution field ta for which δt is a phase space
symmetry.
iii)Non-uniqueness of energy: Each permissible, live ta defines a horizon energy E∆ t
. At
first it seems surprising that there is so much freedom in the notion of energy. Let us compare
the situation at I + , which, like ∆, is null. There, we only have a 3-parameter freedom which,
furthermore, can be eliminated simply by fixing a rest frame. How does this difference arise?
Recall that energy is (the numerical value of) the generator of an unit time translation.
At infinity, all 4-metrics in the phase space approach the same flat metric. Hence, we can
simply fix a unit time-translation Killing field tao of that flat metric near infinity and use its
restriction to I + as the unit time-translation for all metrics in the phase space. By contrast,
there is no fixed 4-metric near ∆ to which all the metrics in our phase space approach.
Hence, we do not have the analog of tao ; only the equivalence class [ℓ] is now common to
all the metrics. If, for a given metric g̃ab in our collection, we select the time translation
represented by a specific ℓ̃a in [ℓ], a priori we do not know which vector field la in [ℓ] would
represent the ‘same’ time-translation for another geometry gab . One might imagine using the
seemingly simplest strategy: just fix a ℓao in [ℓ] and demand that ta approach that ℓa0 for all
points γ in the phase space. Unfortunately, the strategy is not viable because such a ta fails
to define a Hamiltonian evolution in Γ.11 Finally, if we restrict ourselves to globally static
space-times, we can overcome this difficulty by always working with the Killing field which
is unit at infinity. However, in absence of global Killing fields, the behavior of the evolution
vector field ta near the horizon is unrelated to its behavior near infinity. Nonetheless, as we
shall show in Section VII, if one has sufficient control on the space of static solutions of the
theory under consideration, it is possible to select a preferred energy function on the phase
space and use it as the mass of the isolated horizon. In the Einstein-Maxwell case, all static
solutions with horizons are explicitly known whence the strategy is viable.

11 As we saw above, a necessary condition for δt to be a Hamiltonian vector field on Γ is that κ(t)
is a function only of a∆ and Q∆ . Therefore, if we can find a tangent vector δ in the phase space
with δa∆ = δQ∆ = 0 but δκ(t) 6= 0, ta can not define a Hamiltonian evolution. It is easy to find
such a tangent vector δ for this ta .

29
 VI. YANG-MILLS FIELD

In the previous three sections we restricted our attention to Einstein-Maxwell theory.

We will now indicate how Yang-Mills fields can be included. This section is divided into
three parts. In the first, we discuss restrictions on the Yang-Mills fields due to the horizon
boundary conditions and introduce the notion of a ‘Yang-Mills gauge adapted to the horizon’.
In the second part, we discuss the action principle and construct the covariant phase space
for Einstein-Yang-Mills theory. Using this formalism, in the third subsection, we introduce
a Hamiltonian generating time evolution and extend the first law to the Yang-Mills case.

A. Preliminaries

We will restrict ourselves to compact gauge groups G and Yang-Mills connections defined
on trivial bundles. Since the bundle is trivial, the connection gives rise to a smooth, globally
defined Lie algebra valued 1-form, A. As usual, the Yang-Mills derivative operator D will be
defined as Dλ = ∂λ + [A, λ], where ∂ is a flat Yang-Mills connection, and the field strength,
F, via

F := dA + A ∧ A. (VI.1)

The stress energy tensor, T, is given in terms of the field strength as

1 1
Tab = [Fac i Fb c i − gab Fcd i Fcd i ], (VI.2)
4π 4
where the label i runs over the internal indices in the Lie algebra of the group G.
Let us begin by examining how the isolated horizon boundary conditions restrict the
form of the field strength, F, on ∆. Since the Yang-Mills stress energy tensor has the same
form as the Maxwell one, (II.14), the analysis is completely analogous to that of Section
II C. Therefore, we shall not include derivations of the results, but instead highlight the
differences.
Recall that, on a non-expanding horizon, Rab ℓa ℓb = b 0 whence Tab ℓa ℓb =b 0. This has
several consequences for the Yang-Mills field. In particular, one concludes

ℓa F =
b 0
ab
←−−−−−
and ℓ←−a⋆ F −− =
−−−−ab
b 0. (VI.3)

These two restrictions guarantee there is no flux of Yang-Mills field across the horizon.
Making use of the specific form of the stress energy tensor, we also conclude
1 1
Φ02 = Rab ma mb =
b 0 and Φ20 = Rab ma mb =
b 0. (VI.4)
2 2
Our next task is to define the Yang-Mills equivalents of the electric and magnetic charges
of the horizon. Naively, one might consider integrating F and ⋆ F over a 2-sphere cross section

30
of the horizon as was done in the Maxwell theory. However, these 2-forms now have a free
internal index and are only gauge covariant rather than gauge invariant. Since there is no
preferred internal basis at the horizon, the integrals would fail to be well-defined. Therefore,
we must look for 2-forms which are gauge invariant. A natural quantity to consider is
the norm of F, defined by the Killing-Cartan form Kij on the Lie-algebra of G and the
(contravariant) 2ǫ on the horizon:
h i1
|F| := (2ǫ · F)i (2ǫ · F)j Kij 2
(VI.5)

(Although the contravariant 2ǫ is ambiguous up to terms of the type ℓ[a V b] where V a is any
vector field tangential to ∆, this ambiguity does not affect |F| because of (VI.3). The norm
of ⋆ F is defined analogously. These two quantities are gauge invariant and allow us to define
the electric and magnetic Yang-Mills charges of the horizon:
1 I 1 I
QY∆M :=
b − | ⋆ F | 2ǫ and P∆Y M :=
b − |F| 2ǫ . (VI.6)
4π S∆ 4π S∆
Recall that the unusual signs in the definitions of the charges arise due to the orientation of
the S∆ — the normal to the two sphere is inward pointing. In Maxwell theory, the magnetic
charge is zero unless we consider either connections on non-trivial bundles or allow ‘wire
singularities’. As is well known, this is not true for Yang-Mills theory: the magnetic charge
can be non-zero even if we restrict attention to smooth fields on a trivial bundle.
We would now like to verify that the charges defined in (VI.6) are independent of the
cross section of the horizon S∆ on which the integration is performed. The isolated horizon
boundary conditions guarantee this is the case. First, recall the geometric identity

L ℓF
←
−
=
b ℓ · DF
←−−−
− [(ℓ · A), F
←
−
]+←
D(ℓ · F)−.
−−−−−−−−
(VI.7)

A similar expression for ⋆ F is also true. The first term on the right hand side vanishes due
to the field equations and the third term is zero due to the previous restriction on F, (VI.3).
Therefore at the isolated horizon,

L ℓF
←
−
=
b −[(ℓ · A), F] and b −[(ℓ · A), ⋆ F] .
L ℓ ⋆←−F− = (VI.8)

In the Maxwell case, F and ⋆ F are Lie dragged by ℓ. However, for non-Abelian fields, this is
not a gauge invariant statement; the terms on the right hand sides of (VI.8) are necessary
for gauge invariance. Although the field strength and its dual are not Lie dragged along ℓa ,
recalling that L ℓ 2ǫ =
b 0 and using the cyclic property of the trace, it is straightforward to
demonstrate that their norms are:

L ℓ |F|=
b 0 and L ℓ | ⋆F | =
b 0. (VI.9)

This result, along with (VI.3), guarantees that the charges QY∆M and P∆Y M are independent
of the choice of cross section S∆ of the horizon.

31
 Let us consider the remaining components of the Yang-Mills field. The boundary con-
ditions place no restrictions on Fab na mb and ⋆ Fab na mb at all. As in the electromagnetic
case, these components describe the radiation flowing along the horizon. The isolated hori-
zon boundary conditions allow radiation arbitrarily close to — and even at — the horizon,
provided none crosses it.
We have so far restricted our attention to the field strength and its dual. However, in
the action principle and phase space, the basic variable will be the Yang-Mills connection
A. Let us begin with the definition of the Yang-Mills equivalent of the electric potential.
Recall that, given an ℓ, the electric potential was defined in Section II C as Φ(ℓ) =
b −(ℓ · A).
This definition is not appropriate in the Yang-Mills case since the resulting potential has a
free internal index and is therefore not gauge invariant. Instead, we define the Yang-Mills
potential, ΦYM
(ℓ) to be negative the norm of (ℓ · A):

ΦYM
(ℓ) :=
b − | (ℓ · A) | . (VI.10)

This gives us a gauge invariant potential at the horizon.

As in Maxwell theory, we need to constrain the form of A at the horizon. Several
considerations motivate our choice of these boundary conditions. First they must be chosen
so that the action principle is well defined. Second, if the gauge group is U(1) the boundary
conditions should reduce to those given in Section II C for the electromagnetic field. Finally,
we should be able to show that the Yang-Mills electric potential is constant on the horizon.
These considerations suggest the following definition:
Definition 6: The connection A will be said to be in a gauge adapted to the isolated horizon
(∆, [ℓ]) if it satisfies the following two conditions

• (i) The Yang-Mills potential is constant on the horizon

dΦYM
(ℓ) =
b 0 (VI.11)

• (ii) The dual of the field strength (⋆ F) and (ℓ · A) point in the same Lie algebra
direction,

(ℓ · A)i ∝ (2ǫ · ⋆ F)i (VI.12)

on the horizon.

These boundary conditions satisfy the requirements discussed above. First, it is straight-
forward to show that in the U(1) case, condition (i) is equivalent to requiring L ℓ A =b 0 and
(ii) is redundant. Second, as we shall see in Section VI B, these boundary conditions are also
sufficient to make the variational principle well defined.
It is not difficult to show that these conditions can always be satisfied. The remaining
gauge freedom is simply A → g −1 Ag + g −1 ∂g, where g satisfies L ℓ g =
b 0.

32
 B. Action and Phase Space

In this section we will consider the first order action for Einstein-Yang-Mills theory
on the manifold M described in Section III B. The basic fields will consist of the triplet
(eIa , AaI J , Aia ), where eIa and AaI J are the tetrad and Lorentz connection and Aia is the Yang-
Mills connection. The gravitational fields, eIa and AaI J satisfy the same boundary conditions
(at ∆ and infinity) as in Section III B. Furthermore, we require the Yang-Mills fields to be
in a gauge adapted to the horizon and assume they fall off sufficiently fast at infinity.12 The
Einstein-Yang-Mills action is
−1 Z 1 Z 1 Z
S(e, A, A) = ΣIJ ∧ FIJ + ΣIJ ∧ AIJ − Tr[F ∧ ⋆ F] . (VI.13)
16πG M 16πG τ∞ 8π M
The gravitational part of the action has previously been discussed in Section III, therefore
we shall only consider in detail the Yang-Mills terms and verify the variational principle is
well defined. Taking into account the results of Section III, a variation of the action can be
expressed as
Z
1 Z
δS(e, A, A) = Equations of Motion · δφ − Tr[δA ∧ ⋆ F]. (VI.14)
M 4π ∆
We must demonstrate the boundary term at ∆ vanishes due to the conditions imposed on
the Yang-Mills fields. Using (VI.3) and (VI.12), one can show that the trace in (VI.14) can
be replaced by a product of norms:

Tr[δA ∧ ⋆ F] = (δΦYM YM ⋆
(ℓ) − cδ Φ(ℓ) ) | F |
∆
ǫ,

where ∆ ǫ = n ∧ 2ǫ is the volume form on ∆ and δℓ = cδ ℓ.

In the action principle, variations are performed keeping data fixed at the initial and final
slices. In particular δΦYM
(ℓ) and cδ vanish there. However, the boundary conditions guarantee
YM
that Φ(ℓ) , and hence its variation, is constant on ∆. Since, δΦYM
(ℓ) and cδ vanish on the initial
cross section of the horizon and are constant, it follows that δΦYM
(ℓ) =
b 0 and cδ =b 0. Therefore,
the Yang-Mills horizon boundary term vanishes; the action principle is well defined in the
presence of Yang-Mills fields. As in the Einstein-Maxwell case, boundary conditions played
a crucial role in demonstrating the viability of the action.

12 More specifically we require the fall off conditions on the Yang-Mills connection to be such that
all integrals, in particular the symplectic structure, are finite and yet the asymptotic electric and
magnetic charges are not forced to vanish. While it is not trivial to meet these conditions (for
examples the conditions used in [23,12] appear not to lead to a well-defined symplectic structure)
they can be met. However, as in the rest of the paper, for brevity, we will not spell out the boundary
conditions at infinity in detail.

33
 We now wish to construct the covariant phase space and symplectic structure. As be-
fore, points in covariant phase space Γ will consist of histories which satisfy the Einstein-
Yang-Mills field equations, appropriate falloff conditions at infinity and the isolated horizon
boundary conditions at ∆. Before proceeding further, we shall once again need to introduce
an additional field at the horizon. This can be regarded as ‘potential’ for the Yang-Mills
potential ΦYM . Given any point (e, A, A) in the phase space Γ, let us define the scalar field
χ on ∆ as follows: i) Lℓ χ = −ΦYM −
(ℓ) ; and ii) χ vanishes on S , the intersection of M
−
with
∆. These conditions are identical to those imposed in the Maxwell case.
Once again, we take second variations of the action in order to obtain a symplectic
structure. Since the gravitational terms are exactly the same in Section III, we shall only
describe in detail the Yang-Mills part of the symplectic structure. The second variation of
the action (VI.13) yields the following symplectic current:
1
J(γ; δ1 , δ2 ) = Jgrav − Tr[δ1 ⋆ F ∧ δ2 A − δ2 ⋆ F ∧ δ1 A]. (VI.15)
4π
Using the fact that the field equations and linearized field equations are satisfied, one can
directly verify that J(γ; δ1 , δ2 ) is indeed a closed 3-form. We again choose the spacetime
region of interest, M̃, to be that part of the spacetime M bounded by M1 , M2 , infinity and
a portion ∆ ˜ of the isolated horizon. Integrating dJ over M̃ and using asymptotic falloff
conditions, we obtain
Z Z Z
( + + )J(γ; δq1 , δ2 ) = 0.
M1 M2 ˜
∆

The integral of J over ∆˜ does not vanish but, as in Section III, the pull back of J to ∆ is
exact. Therefore, we can express the integral over ∆ ˜ of J as integrals over the initial and
final 2-spheres S1 and S2 . Using these results, and keeping track of orientations, we obtain
the symplectic structure:

Ω|γ (δ1 , δ2 ) =
1 Z IJ IJ 1 I
− [δ1 Σ ∧ δ2 AIJ − δ2 Σ ∧ δ1 AIJ ] + [δ1 2ǫδ2 ψ − δ2 2ǫδ1 ψ]
16πG M 8πG S
Z I
1 ⋆ ⋆ 1
− [δ1 F ∧ δ2 A − δ2 F ∧ δ1 A] + [δ1 (| ⋆ F | 2ǫ)δ2 χ − δ2 (| ⋆ F | 2ǫ)δ1 χ] . (VI.16)
4π M 4π S
Full use of the isolated horizon boundary conditions has been made in obtaining this sym-
plectic structure. In particular, to obtain the given form of the Yang-Mills surface term, we
have used the fact that (ℓ · A) and (2ǫ · ⋆ F) point in the same direction in the Lie algebra.
Using field equations one can directly verify that the right side of (VI.16) is independent of
the choice of the partial Cauchy surface; the symplectic structure is ‘conserved’. We will use
(Γ, Ω) as our covariant phase space.

34
 C. Hamiltonian and First Law

In this subsection we will generalize the arguments of Section V to obtain an expression

for the energy of the horizon in Einstein-Yang-Mills theory. To do so, we must specify a
time evolution vector field ta . As before, we require ta to be a member of the preferred
equivalence class [ℓa ] at the horizon (this again requires restriction to non-rotating isolated
horizons) and approach unit time translation asymptotically. Given ta we can calculate the
infinitesimal generator of time evolution, δt = (L t e, L tA, L t A), and determine whether it is
Hamiltonian. Recall that δt is Hamiltonian if and only if the 1-form Xt on the phase space
defined by

Xt (δ) := Ω(δ, δt ) (VI.17)

is closed. Let us calculate Xt . The gravitational part will be identical to the expression
obtained in Section V, therefore we shall concentrate on the Yang-Mills terms. As with
the Maxwell field, the Lie derivatives of the Yang-Mills fields can be re-expressed using the
following identities:

L t A = t · F + D(t · A) L t ⋆ F = t · (D⋆ F) − [ℓ · A, ⋆ F] + D(t · ⋆ F) . (VI.18)

Making use of these expressions, the field equations satisfied by (e, A, A) and the linearized
field equations for δ, we obtain the required expression for Xt :
−1 Z 1 Z
Xt (δ) = (t · A)δΣ − (t · Σ) ∧ δA − Tr[(t · A)δ(⋆F) − (t · ⋆ F) ∧ δA].
16πG ∂M 4π ∂M
(VI.19)

As before the expression involves integrals only over the 2-sphere boundaries S∞ and S∆
t
of M; there is no volume term. The gravitational terms yield δEADM at infinity and
−(1/8πG)κ(t) δa∆ at the horizon. The Yang Mills term at infinity vanishes due to fall-
off conditions, therefore we need only calculate the Yang-Mills contribution at the horizon.
This is composed of two terms, the second of which vanishes due to the restriction (VI.3)
which guarantees ℓ · ⋆←−F− =
b 0. Since we are in a gauge adapted to the horizon, (ℓ · A) and ⋆ F
point in the same internal direction. This allows us to replace the trace in the first term by
norms:

Tr[(t · A)δ(⋆F)] = −ΦYM ⋆ 2

(t) δ(| F | ǫ)

and guarantees that ΦYM (t) is constant on the horizon. Making use of the definition of Yang-
Mills electric charge, (VI.6), we obtain
1
Xt (δ) = − κ(t) δa∆ − ΦYM YM t
(t) δQ∆ + δEADM .
8πG
Recall that the necessary and sufficient condition for the existence of a Hamiltonian is that
Xt be closed. Clearly, this is equivalent to

35
 1
ddκ(t) ∧ dda∆ + ddΦYM dQYM
(t) ∧ d ∆ = 0, (VI.20)
8πG
Once again, we conclude that the surface gravity κ(t) and the Yang-Mills potential ΦYM (t) at
a YM
the horizon defined by t can depend only upon the area a∆ and charge Q∆ of the horizon.
t
Finally, (VI.20) is also the necessary and sufficient condition that there exist a function E∆ ,
YM
also only of a∆ and Q∆ such that

t 1
δE∆ = κ(t) δa∆ + ΦYM YM
(t) δQ∆ . (VI.21)
8πG
t
As before, E∆ is interpreted as the horizon energy defined by the time translation ta . We
conclude that the vector field δt is Hamiltonian if and only if the first law, (VI.21), holds.
We will conclude this section with a few remarks.
i) The derivation of the first law and its final form are completely analogous to those in
the Einstein-Maxwell theory. By contrast, in the discussion of the first law for undistorted
isolated horizons of [12], certain restrictions were imposed on the permissible variations δ in
the Einstein-Yang-Mills case. In our treatment, subtleties arise only in the definition of a
canonical horizon mass (see Section VII C) rather than the discussion of the first law itself.
ii) Although the Yang-Mills magnetic charge P∆YM will generically not be zero, no term
involving δP∆YM arises in the first law.
iii) How does our result compare with those previously available [23,24]? In [24] the first
law for Yang-Mills fields is proved for globally stationary spacetimes and small perturbations
from one such space-time to another. Assuming the Yang-Mills fields fall off sufficiently fast
at infinity, (in the non-rotating case) the first law of [24] then reads
Z
1
δM = κδa + Tr[φ δ ⋆ F] .
8πG Hor

Here, M is the ADM mass evaluated at infinity, while all terms on the right hand side are
evaluated at the horizon. Due to a different gauge choice at the horizon, the authors define
a Lie algebra valued potential φ and leave the ‘ΦδQ’ term inside an integral. However, the
general form of this first law is the same as ours. In this sense, our framework generalizes
the results of [24] to non-static contexts.
In [23], the first law is proved for globally stationary spacetimes and arbitrary small
departures therefrom. However, there are a number of important differences between these
results and the ones obtained in this paper. In the non-rotating case, the first law of [23]
reads
1
δMADM + V δQ = κδa
8πG
where V and Q are the Yang-Mills potential and charge evaluated at infinity while κ and a
are of course evaluated at the horizon. Because of the non-Abelian nature of the Yang-Mills
field, unlike in the Maxwell case, the charge Q evaluated at infinity is now quite different

36
from the charge evaluated at the horizon and, as in the Maxwell theory, the potential V
evaluated at infinity has no direct bearing on the potential at the horizon. Furthermore,
that calculation makes an essential use of the bifurcation 2-sphere and all fields are required
to be smooth there. This restriction implies that the Yang-Mills potential at the horizon
vanishes. (The same is true if one restricts the analysis of [23] to the Maxwell case.)
The first law derived in the isolated horizon framework is valid also in presence of radi-
ation in the exterior space-time region and makes no reference to the bifurcation 2-sphere.
(Although we restricted ourselves to the non-rotating case, rotation has been incorporated in
this framework in [13].) Furthermore, it has the aesthetically pleasing feature that all quan-
t
tities that appear in (VI.21) — including the energy E∆ , the potential ΦYM
(t) and the charge
YM
Q∆ — are evaluated at the horizon. In particular, one can now meaningfully consider
the physical process version in which one does an experiment at the horizon by dropping a
test particle/field and changing the horizon charge infinitesimally. More generally (VI.21) is
genuinely a law governing the mechanics of the horizon.

VII. HORIZON MASS

For notational simplicity, we will say that a (live) vector field ta is permissible if it
gives rise to a Hamiltonian evolution. We saw in Sections V and VI that each permissible
vector field ta defines a horizon energy E∆
t
. In the phase space framework, E∆ t
has a direct
interpretation: it is the surface term at the horizon in the expression of the Hamiltonian
generating the ta -evolution. However, in many physical applications — such as the study of
black hole mergers — one is interested in properties of a specific space-time, rather than the
full phase space. Then, it is useful to have at one’s disposal a canonical notion of energy,
the analog of the ADM energy in the rest frame at infinity. This quantity could then be
interpreted as the horizon mass. In this section, we will introduce this notion in detail.
The discussion is divided into three parts. In the first, we consider the Einstein-Maxwell
theory; in the second, we discuss dilatonic couplings [9]; and, in the third, we analyze the
Einstein-Yang-Mills system.

A. Einstein-Maxwell theory

In Section V we showed that ta is permissible if and only if (V.7) holds on the phase
space. We will now construct a large family of permissible evolution fields ta . Fix any
regular function κo of two variables a∆ and Q∆ . Then, given any point γ ≡ (e, A, A) of Γ,
we define (the boundary value of) the vector field ta as follows. Consider the vector field ℓa
on ∆ defined by the tetrad, ℓa = eaI ℓI , and denote by κ(ℓ) the surface gravity associated with
it. Then, κo = cκ(ℓ) for some constant c. Let us set ta = cℓa . Repeating this procedure at
each phase-space point γ, we obtain a live vector field ta with κ(t) = κo . (The resulting c will
be constant on ∆ but a function on the phase space.) Next, consider the electro-magnetic
potential, which is guaranteed to be constant on ∆ by our boundary conditions but whose

37
value at any phase space point is so far completely free. We will now use (V.7) to fix it.
Equation (V.7) implies:
∂κ(t) ∂Φ(t)
= .
∂Q∆ ∂a∆
Since κ(t) = κo is known, we can simply integrate the equation for Φ(t) as a function of a∆
and Q∆ . Furthermore, the solution is unique if we impose the physical condition that Φ(t)
should vanish whenever Q∆ = 0. Thus, starting from any regular function κo of a∆ and Q∆ ,
we have obtained a permissible evolution field ta . Conversely, it is easy to verify that every
permissible vector field arises via this construction. There is clearly a very large family of
such live vector fields.
An obvious question is if there is a ‘canonical’ or ‘natural’ choice of ta ? We will now
show that the answer is in the affirmative. Recall that, in the Einstein-Maxwell theory,
there is precisely a 2-parameter family of globally static solutions admitting horizons: the
Reissner-Nordström family. (Since A is required to be a globally defined connection on a
trivial U(1) bundle, the magnetic charge is zero on the entire phase space.) Let us focus on
this family. Denote by ξ a the static Killing field which is unit at infinity. Its surface gravity
is a specific function of a∆ and Q∆ :
!
1 GQ2
κ(ξ) = 1 − 2∆ .
2R∆ R∆
2
As before, R∆ is the horizon radius, defined by a∆ = 4πR∆ . We can therefore use κ(ξ) in
place of κo in the above construction. The resulting permissible, live vector field tao agrees
with ξ a on the horizon of every static solution. This property is satisfied only if we set
κo = κ(ξ) .
to
Next, we can ‘integrate’ (V.8) to obtain the horizon energy E∆ . Although a priori there is
the freedom to add a constant, we can fix it by requiring that the energy vanish as a∆ and Q∆
tend to zero. Indeed, we have no choice in this since one cannot construct a quantity with
dimensions of mass from the fundamental constants that appear in the Einstein-Maxwell
theory. (Einstein-Yang-Mills theory does admit such a constant and we will see in Section
VII C that it leads to an interesting modification of the situation discussed here.) Let us
define the horizon mass via
to
M∆ = E∆ .

To justify this definition, let us begin by restricting to static solutions. In each static solution,
we are free to extend tao such that it coincides with the Killing field ξ a . General symplectic
arguments imply that, on any connected component of the space of static solutions, the
numerical value of the total Hamiltonian, generating evolution along ξ a , must be constant
(see, e.g., [8].) In the Einstein-Maxwell case, there is a single connected component and, by
the dimensional argument given above, the numerical value of the Hamiltonian must vanish
on it. Hence, from (V.9) it follows that, on any static solution,

38
 Hto = MADM − M∆ = 0.

On a general solution, of course, MADM would be greater than M∆ , the difference being equal
to the energy in radiation. If the horizon is complete in the future and time-like infinity i+
satisfies certain regularity conditions, as in [8] one can argue that the difference is precisely
the total energy radiated across I + and hence M∆ equals the future limit of the Bondi mass.
These considerations support our interpretation of M∆ as the horizon mass.
Finally, since we now have a canonical evolution field tao , we can drop the suffix t on surface
gravity and electromagnetic potential and write the first law (V.8) in the more familiar form:
1
δM∆ = κδa∆ + ΦδQ∆ .
8πG
In contrast to treatments based on static space-times, the quantities that enter this law are
all defined at the horizon. Therefore, as pointed out in [8], it is now possible to interpret this
law also in the ‘active’ sense where one considers physical processes which increase the area
and the charge of a given horizon. To our knowledge, the standard proofs of this physical
version [5,10] are not applicable to processes in which the background has non-zero electric
charge and the process changes it infinitesimally.
We will conclude with a few remarks.
i) In the above discussion, the permissible evolution field tao was constructed by setting
a
to = cℓa where c is given by κo ≡ (1/2R∆ )(1 − G(Q∆ /R∆ )2 ) = cκ(ℓ) . For c to be well-
defined, it is necessary that κo vanishes whenever κ(ℓ) does. Therefore, for the mass to be
well-defined, we must excise those points from the phase space at which κ(ℓ) vanishes but κo
does not. However, this is not a serious limitation. In particular, we still retain all static
solutions including the extremal ones at which κo vanishes.
ii) Since we have a specific κo , we can use (V.7) to obtain the corresponding electro-
static potential: Φo = Q∆ /R∆ . Furthermore, by integrating (V.8) it is easy to express M∆
explicitly in terms of the horizon parameters:
!
1 R∆ GQ2
M∆ = κa∆ + ΦQ∆ = 1 + 2∆ . (VII.1)
4πG 2G R∆

Thus, the functional dependence of M∆ on the horizon parameters at any point of the phase
space is the same as in static space-times. Note that this is a result of the framework, not
an assumption. Its derivation involved two distinct steps. First, and most importantly, the
first law (V.8) arose as a necessary and sufficient condition for the existence of a consistent
Hamiltonian framework. Second, the freedom in ta was exploited in order to construct the
preferred, permissible evolution field tao . It is quite significant that M∆ can be expressed so
simply using just the parameters defined locally at the horizon even when there is radiation
arbitrarily close to it. This fact is likely to play an important role in the problem of extracting
physics in the strong field regimes from numerical simulations of black hole collisions [18].
It is important to notice that although we made use of our knowledge of static solutions to
arrive at a canonical tao and the mass function M∆ , the final result (VII.1) makes no reference

39
to these solutions. M∆ is a simple function of the parameters which can be directly computed
from the geometry of any one isolated horizon.
iii) In the earlier work [8] on undistorted horizons, one restricted oneself to the preferred
evolution field tao from the very beginning (although this vector field was selected using a
different but equivalent procedure).13 The a priori freedom in the choice of a permissible ta
was not discussed and the first law appeared only in the more familiar form, given above.

B. Dilatonic coupling

The Einstein-Maxwell-dilaton system was studied in some detail in the undistorted case
in [9]. We will revisit it here in the more general context considered in this paper because it
brings out a subtlety in the definition of the horizon mass M∆ and the associated first law.
The dilaton is a scalar field φ which can couple to the Maxwell field in a non-standard
fashion. The coupling is governed by a constant α. If α=0, one obtains the standard
Einstein-Maxwell-Klein-Gordon theory and the situation then is completely analogous to
the Einstein-Maxwell theory considered above. If α=1, the theory represents the low energy
limit of string theory. In this case, there are some interesting differences from the Einstein-
Maxwell theory considered in this paper. To bring out these differences, in this subsection
we will set α=1. (The situation for a general value of α is discussed in [9] where one can
also find details on the material summarized below.)
In the standard formulation, the theory has three charges, all defined at infinity; the
ADM mass MADM , the usual electric charge Q∞ and another charge Q̃∞ :
I I
1 ⋆ 1
Q∞ = F and Q̃∞ = e−2φ ⋆ F .
4π S∞ 4π S∞

Q̃ is conserved in space-time (i.e. its value does not change if the 2-sphere of integration is
deformed) while Q is not. From the perspective of the isolated horizons, it is more useful to
use a∆ , Q∆ , Q̃∆ as the basic charges14 :

13 This strategy seems to have generated a misunderstanding (see, e.g., [29]) that the first law was
obtained in [8,9,12] merely by identifying the parameters labeling a general isolated horizon with
those of static horizons and then using the Smarr formulas available in the static context. This
was not the case. Rather, static solutions were used only to select the appropriate normalization
of the evolution vector field ta at the horizon. The Hamiltonian framework was then used to define
the horizon mass without any reference to Smarr formulas. As in this section, the mass was then
shown to reproduce the Smarr-type formulas on general horizons.

14 In the undistorted case, the dilaton is constant on ∆ and hence we can replace Q∆ by φ∆ as in
[9].

40
 I I
1 ⋆ 1
Q∆ = F, and Q̃∆ = e−2φ⋆ F .
4π S∆ 4π S∆

Although the standard electric charge is not conserved in space-time, it is conserved along
∆ whence Q∆ is well-defined.
It is straightforward to extend the construction of the phase space to include the dilaton.
The only difference is that the charge Q in equations (V.6) - (V.8) is replaced by Q̃. With
this minor change, the discussion of the first part of Section VII A is also unaffected. Thus,
given any function κo of a∆ and Q̃∆ , we can construct a permissible, (live) evolution field ta .
The difference arises in the next step where we constructed a preferred tao . With the
dilatonic coupling, the theory has a unique [25] three parameter family of static solutions
which can be labeled by (a∆ , Q∆ , Q̃∆ ). As in the Reissner Nordström family, these solutions
are spherically symmetric. In terms of these parameters, the surface gravity κ(ξ) of the static
Killing field which is unit at infinity is given by:
" #" #− 12
1 Q∆ Q̃∆ Q∆ Q̃∆
κ(ξ) = 1 + 2G 2
1 − 2G 2
.
2R∆ R∆ R∆
The problem in the construction of the preferred tao is that we need a function κo which
depends only on a∆ and Q̃∆ . Therefore, we can no longer set κo = κ(ξ) on the entire phase
space because κ(ξ) depends on all three horizon parameters.
To extract the mass function M∆ on the phase space, we can proceed as follows. Let
us foliate Γ by Q∆ = const surfaces. On each leaf, κ(ξ) trivially depends only on a∆ and
Q̃∆ and so we can set κo = κ(ξ) . Therefore, by the procedure outlined in Section VII A,
to
we obtain a (live) vector field tao and can define the mass M∆ (γ) = E∆ (γ) for all points γ
on this leaf. Repeating this procedure for each leaf, we obtain a live vector field tao and a
mass function M∆ everywhere on Γ. However, the surface gravity κ(to ) now depends on all
three parameters, rather than just a∆ and Q̃∆ . Therefore, the first law (V.8) cannot hold
for arbitrary variations δ and consequently δto fails to be a Hamiltonian vector field. Put
differently, although there is a multitude of permissible, live vector fields, each leading to a
first law, none of them can coincide with the Killing field ξ a (which is unit at infinity) on all
static solutions. This is a significant departure from the Einstein-Maxwell case considered
above.
Nonetheless, (modulo the caveat discussed in the first remark at the end of Section
VII A) the above procedure does provide us with a well-defined mass function M∆ on the
entire phase space which can be expressed in terms of the horizon parameters as
" #− 12
R∆ Q∆ Q̃∆
M∆ = 1 − 2G 2
.
2G R∆
It equals MADM in static space-times and has other properties which motivated our inter-
pretation of M∆ as the horizon mass in the Einstein-Maxwell case. Since this function is
well-defined on the entire phase space, we can simply vary it and express the result in terms
of the horizon parameters. The result is:

41
 1
δM∆ = κδa∆ + Φ̂δ Q̂∆
8πG

where κ = κ(to ) , Φ̂2 = (Q∆ Q̃∆ /R∆ 2

) and Q̂2∆ = Q∆ Q̃∆ . Thus, although there is still a
first law in terms of tao and M∆ , it does not have the canonical form (V.8) because tao is
not a permissible vector field. More generally, in theories with multiple scalar fields [26], if
one focuses only on static sectors, one obtains similar ‘non-standard’ forms of the first law
with work terms involving scalar fields. This reflects the fact that there is no permissible
vector field ta , defined for all points of the phase space, which coincides with the properly
normalized Killing field on all static solutions. In the undistorted case, the analysis was
carried out only in terms of the vector field tao and the horizon mass M∆ [9]. The resulting
first law had the above form.
Alternatively, one can restrict oneself to variations δ which are tangential to the leaves
of the phase space foliation constructed above. Since tao is a permissible vector field for any
one leaf, we obtain the standard first law

δM∆ = (1/8πG)κ(to ) δa∆ + Φ(to ) δ Q̃∆

for the restricted variations. The idea of using such restricted variations was suggested in
[12] in the context of Yang-Mills fields (although the foliations and other details were not
spelled out there).
To summarize, because there is now a three parameter family of static solutions rather
than two —or, more precisely, because the standard surface gravity κ(ξ) in static space-times
depends on a∆ , Q̃∆ and Q∆ — a canonical, permissible evolution field is no longer available.
However, there is still a multitude of permissible evolution fields and corresponding first
laws. Furthermore, one can still define a canonical mass function M∆ on the entire phase
space.

C. Yang-Mills fields

In Einstein-Maxwell theory, with and without the dilaton, there is no way to construct
a quantity with the dimensions of mass from the fundamental constants in the theory. The
situation is different for Einstein-Yang-Mills theory because the coupling constant g has
dimensions (LM)−1/2 . The existence of a quantity with units of mass has interesting conse-
quences which we will now discuss.
Let us begin with a summary of the known static solutions in Yang-Mills theory. First,
the Reissner-Nordström family constitutes a continuous two parameter set of static solutions
of the Einstein-Yang-Mills theory, labelled by (a∆ , QYM
∆ ). In addition, there is a 1-parameter
family of ‘embedded Abelian solutions’ with (a fixed) magnetic charge P∆o , labelled by
(a∆ , P∆o ). Finally, there are families of ‘genuinely non-Abelian solutions’. For these, the
analog of the Israel theorem for Einstein-Maxwell theory fails to hold [27]; the theory admits
static solutions which need not be spherically symmetric. In particular, an infinite family of

42
solutions labelled by two integers (n1 , n2 ) is known to exist. All static, spherically symmetric
solutions are known and they correspond to the infinite sub-family (n1 , n2 = 0), labelled by
a single integer. However, the two parameter family is obtained using a specific ansätz, so
there may well exist other static solutions. Although the available information on static
solutions is quite rich, in contrast to the Einstein-Maxwell-Dilaton system, one is still rather
far from having complete control of the static sector of the Einstein-Yang-Mills theory.
The zeroth and first laws do hold in the Einstein-Yang-Mills case. At present, however,
we can only hope to repeat the strategy used in the last two sub-sections to define a canonical
mass function M∆ on portions of the phase space. In order to define it on the full phase
space, of the ‘uniqueness’ and ‘completeness’ conjectures of [12] will have to hold (possibly
with a suitable modification.15 Nevertheless, new insight into the static solutions can be
obtained by restricting attention to certain leaves of the phase space. The basic idea is
taken from [12] but applied in a slightly different manner to the more general context of
distorted horizons.
Consider a connected component of the known static solutions, labelled by ~n ≡ (n1 , n2 ).
This is a 1-dimensional sub-space of the phase space which we denote S~n . Each point in S~n
can be labelled by the value of the horizon area a∆ . Calculate the surface gravity κ(ξ) for
this family, where ξ a is the static Killing field which it unit at infinity and set κo = κ(ξ) in
the construction sketched in Section VII A. We then obtain a live vector field tao , and the
corresponding first law

to 1
δE∆ = κ(t ) δa∆
8πG o
on the full phase space.
to (~
n)
When restricted to S~n , we can interpret E∆ as the horizon mass M∆ and replace κ(to )
by the function β~n (a∆ ) used in the literature: β~n = 2κ(to ) R∆ . Then, by integrating the first
law along S~n , one obtains:
Z
(~
n) 1 R∆
M∆ = β~n (x)dx
2G 0

(t )
where, we have used the fact that, since E∆o is a surface integral at ∆, it vanishes as the
horizon area goes to zero. Thus, the horizon mass is completely determined by β~n (a∆ ).
to to
Next, we use the fact that the Hamiltonian given by Hto = EADM − E∆ (see (V.9)) is
constant on each connected, static sector if tao coincides with the static Killing field on the
entire sector. By construction, our tao has this property. In the Einstein-Maxwell case, since

15 For example it may be appropriate to restrict oneself to the class of space-times admitting
isolated horizons which are complete in the future. Physically, this is the most interesting case since
such horizons would represent the asymptotic geometry resulting from a gravitational collapse or
black hole mergers.

43
 to
there is no constant with the dimension of energy, it follows that the restriction of HADM to
the static sector must vanish. The situation is quite different in Einstein-Yang-Mills
√ −1 theory
where the Yang-Mills coupling constant g provides a scale. In c = 1 units, (g G) ∼ Mass.
Therefore, we can only conclude
(~
n) (~
n)
√
MADM = M∆ + (g G)−1 C (~n)

for some ~n-dependent constant C (~n) . As the horizon radius shrinks to zero, the static solution
[27,28] under consideration tends to the solitonic solution with the same ‘quantum numbers’
√ soliton,(~n)
~n. Hence, by taking this limit, we conclude (g G)−1 C (~n) = MADM . Therefore, we have
the following interesting relation between the black hole and solitonic solutions:
Z
BH,(~
n) 1 R∆
soliton,(~
n)
MADM = β~n (x)dx + MADM
2G 0

where the integral of β~n is evaluated on the 1-dimensional ‘parameter space’ of S~n (given by
the horizon radius). Furthermore,
√ −1as is clear from the above discussion, the ADM mass of
the soliton is a multiple of (g G) . Thus, somewhat surprisingly, the derivation of the first
law in the isolated horizon framework has led to an interesting relation between the ADM
masses of black holes and their solitonic analogs in the static sector.

VIII. DISCUSSION

In the first part of this paper, we introduced the notions of weakly isolated and isolated
horizons. In contrast with earlier work [8,9,12], the definitions allow for the possible presence
of distortion and rotation at the horizon. In addition, the present definitions are more
geometric and intrinsic; in particular, they never refer to a foliation.
The notion of an isolated horizon, unlike that of an event horizon, is completely quasi-
local. One can test if a given 3-surface in space-time is (weakly) isolated or not simply by
examining space-time geometry at the surface. Furthermore, space-times admitting an iso-
lated horizon ∆ need not admit any Killing field even in a neighborhood of ∆. In particular,
they can admit radiation in the exterior region. Therefore, such space-times can serve as
more realistic models of late stages of a gravitational collapse or black hole merger. In [18]
the near ∆ geometry of vacuum solutions is examined in detail using similar techniques to
those used at null infinity. The resulting structure — presence of a preferred ‘rest frame’,
constrains on possible isometries, Bondi-type expansions of the metric — should be useful
to extract physics in the strong field regime of general relativity, especially in the problem
of binary black hole collisions.
This paper, however, focused on another aspect of isolated horizons: extensions of the
zeroth and first laws of black hole mechanics. All previous discussions of these laws were
restricted to perturbations of stationary black holes. Using Lagrangian and Hamiltonian
frameworks, we extended these laws to arbitrary space-times admitting non-rotating iso-
lated horizons in Einstein-Maxwell-dilaton and Einstein-Yang-Mills theory. Furthermore,

44
the analysis suggests that it should be rather easy to incorporate other forms of matter,
provided they admit Lagrangian and Hamiltonian descriptions.
The generalization of black hole mechanics presented in this paper has several interesting
features. First, all quantities that enter the first laws are defined locally at the horizon ∆.
In standard treatments, some quantities such as area and surface gravity are defined at the
horizon. Others, like energy and sometimes [23] even the Yang-Mills/Maxwell charge and
potential, are evaluated at infinity. In part because of this ‘mismatch’, to our knowledge the
‘physical process version’ of the first law [10] had not previously been established for processes
which change the charge of the black hole. Since all quantities in the present treatment
are defined locally at the horizon, it is now straightforward to establish the law for such
processes [8]. Secondly, other treatments based on a Hamiltonian framework [10,32,23] often
critically use the bifurcate 2-surface which does not exist in the extremal case. Therefore,
extremal black holes are often excluded from the first law. The present analysis never makes
reference to bifurcate surfaces (which do not exist in physical space-times resulting from
gravitational collapse). Therefore, our discussion of the first law holds also in the extremal
case. Thirdly, with obvious modifications of boundary conditions at infinity, our analysis
includes cosmological horizons where thermodynamic considerations are also applicable [11].
Finally, and perhaps most importantly, our analysis sheds new light on the ‘origin’ of the
first law: it arose as a necessary and sufficient condition for the existence of a Hamiltonian
generating time evolution. A new feature of our framework is the existence of an infinite
family of first laws corresponding to the infinite family of ‘permissible’ vector fields ta . (A
vector field ta is permissible if it is Hamiltonian, that is, induces canonical transformations on
the phase space.) In theories where we have sufficient control on the space of static solutions,
such as Einstein-Maxwell, one can select a natural evolution field tao . Corresponding to
evolution along this tao , there is a canonical notion of energy which can be interpreted as
the mass of the isolated horizon. There exist also preferred values of surface gravity and
electric potential and a canonical first law. This additional structure is extremely useful
in other applications of the framework, such as extraction of physical information from
numerical simulations of black-hole collisions. However, it is not essential to the discussion
of mechanics: our derivation of the first law in Sections V and VI does not require any
knowledge of the static sector of the theory.
The Hamiltonian approach to black hole mechanics has appeared in the literature before,
most notably in the work of Brown and York [14]. The spirit of the Brown-York approach
is similar to ours. In particular, they do not restrict themselves to stationary situations.
However, in that work, the focus is on an outer, time-like boundary whereas our focus
is on the inner, null boundary representing the isolated horizon. Conserved quantities in
presence of internal boundaries were recently discussed also by Julia and Silva [30] in a more
general context of theories with gauge symmetries. As in our framework, their treatment
exploits the simplifications that occur in a first order formalism and the final surface-integral
expressions of conserved charges are dictated by the precise boundary conditions imposed at
the internal boundaries. Their treatment is based on superpotentials and is thus complements
the Hamiltonian methods used here and in [14].

45
 In this paper, the Lagrangian and Hamiltonian frameworks are based on real tetrads
and Lorentz connections. It is therefore quite straightforward to extend our analysis to
any space-time dimension. Indeed, it has already been extended to 2+1 dimensions in [31].
However, our phase space — and especially the explicit symplectic structure used here — is
tailored to the Einstein-matter system. While it should be possible to extend it to higher
derivative theories of gravity as in [32], that task would not be as simple.

ACKNOWLEDGMENTS

We are grateful to Chris Beetle and Jerzy Lewandowski for countless discussions. We
thank Alejandro Corichi and Daniel Sudarsky for stimulating correspondence, Jiri Bicak
and Werner Israel for pointing out references on distorted black holes and Thibault Damour,
Sean Hayward and Bob Wald for helpful comments. This work was supported in part by the
NSF grants PHY94-07194, PHY95-14240, INT97-22514 and by the Eberly research funds of
Penn State. SF was supported in part by a Braddock Fellowship.

APPENDIX A: EXAMPLES OF DISTORTED HORIZONS

Because of the no hair theorems in the Einstein-Maxwell theory, distorted horizons have
received a rather limited attention in the literature. Therefore, in this appendix we will
discuss a few explicit examples in Einstein-Maxwell theory. For a general construction and
an existence result, see [18,33].
To obtain explicit solutions, one has to impose symmetries. All solutions considered in
this section will be static and axi-symmetric. As one would expect from the no-hair theorems,
they fail to be asymptotically flat, whence they fail to represent isolated black holes in the
standard sense. Nonetheless, they all satisfy the isolated horizon boundary conditions. That
framework also serves to ‘explain’ the otherwise surprising feature that the surface gravity of
these solutions depends only on the area and the charge and is insensitive to the distortion
parameters.
In the literature on static, distorted black holes, it is generally assumed that the solution is
valid only in a finite region around the horizon and its distant behavior is suitably modified by
the far-away matter which causes the distortion. For undistorted isolated horizons, Robinson-
Trautman space-times [15] offer interesting examples of vacuum, asymptotically flat solutions
which admit isolated horizons but no Killing fields whatsoever. Distorted analogs of these
solutions are not known but presumably exist. It would be interesting to find them.

1. Black Hole in a magnetic universe

Let us begin with a simple example: an uncharged black hole in an ‘external magnetic
field’ which distorts the horizon. The specific solution we wish to consider is static and

46
axisymmetric and was first obtained in the Ernst-potential framework [34]. The magnetic
field is non-zero on the horizon. Thus, one has to consider the full set of Einstein-Maxwell
equations on the horizon.
The space-time has topology S 2 × IR2 and the metric is given by
"   #
2 2 2M dr 2 r 2 sin2 θ 2
ds = F − 1− dt2 + + r 2 dθ2 + dφ (A.1)
r 1 − 2M/r F2
where
1
F = 1 + B02 r 2 sin2 θ.
4
B0 is a constant and on the axis the magnetic field is given by B = B0 dr. Because F diverges
at infinity, the metric fails to be asymptotically flat. For M = 0, the metric reduces to that
of the Melvin universe and for M 6= 0 it admits a Killing horizon at r = 2M. To examine the
behavior at the horizon, let us first cast the metric in the Eddington-Finkelstein coordinates
(v, r, θ, φ) where dv = dt + (1 − 2Mr
)−1 dr:
 
2 2 2M 2 2 2 2 2 r 2 sin2 θ 2
ds = −F 1− dv + 2F dvdr + F r dθ + dφ . (A.2)
r F2
Since the metric is not asymptotically flat, the standard procedure of normalizing the Killing
field to be unit at infinity is not applicable. Thus, we only have an equivalence class [ℓ] of
∂
(preferred) null normals to the horizon, ℓ ∝ ∂v . Let ∆ be the Killing horizon and assume
∂
the Killing field ∂v is a member of the equivalence class [ℓ]. It follows trivially that (∆, [ℓ])
is a non-rotating isolated horizon. If B0 6= 0, the scalar curvature 2R of the horizon 2-metric
has θ-dependence; the horizon is distorted. However, an explicit calculation shows that, as
in the Schwarzschild space-time, the surface gravity κ is given by 1/2r ≡ 1/2R∆ and the
electrostatic potential Φ vanishes on ∆. At first, it is quite surprising that while the presence
of distortion affects µ, Φ11 , Ψ2 and 2R, it does not affect κ or Φ. However, as we saw in
Section V, this result is to be expected from the general Hamiltonian considerations.

2. Distorted black holes as special cases of Weyl solutions

In this sub-section, we will review the construction of a large family of distorted black
holes starting from Weyl solutions [35] and a recent generalization of these results to include
electric charge [36].
A general class of static, axisymmetric spacetimes was found by Weyl in 1917 [37]. The
metric for such a spacetime can be cast in the following form:

ds2 = −e2ψ dt2 + e2(γ−ψ) (dρ2 + dz 2 ) + e−2ψ ρ2 dφ2 (A.3)

where ψ and γ are smooth functions of ρ and z. Einstein’s vacuum equations expressed in
terms of ψ and γ take a particularly simple form. The equation for ψ,

47
 ψ,ρ
ψ,ρρ + + ψ,zz = 0, (A.4)
ρ

is simply the Laplace equation in flat space with cylindrical coordinates (ρ, z, φ). (In addition,
ψ has to be independent of the angular coordinate φ.) Given a solution for ψ, the function
γ can be determined by simple integration:
2
γ,ρ = ρ[ψ,ρ − ψ,z2 ]
γ,z = 2ρ[ψ,ρ ψ,z ] (A.5)

The Schwarzschild metric is of course a particular solution to these equations and corre-
sponds to choosing for ψ and γ:
  !
1 L−M 1 L2 − M 2
ψ = ψS := ln and γ = γS := ln , (A.6)
2 L+M 2 L2 − η 2
q q
where L = 12 (l+ + l− ), η = 12 (l+ − l− ) with l+ = ρ2 + (z + M)2 and l− = ρ2 + (z − M)2
and M is the mass of the Schwarzschild solution. Note that ψ is just the Newtonian potential
due to a rod of length 2M placed symmetrically about the origin on the z-axis. Both ψS and
γS diverge logarithmically in the limit ρ → 0 (for | z |≤ M). In order to recast this solution
in the standard Schwarzschild form, one must transform from (z, ρ) to the Schwarzschild
coordinates (r, θ) by

z = (r − M) cos θ and ρ2 = r 2 (1 − 2M/r) sin2 θ (A.7)

This coordinate transformation shows that the horizon, r = 2M, corresponds to the line
segment ρ = 0, | z |≤ M in Weyl coordinates. Therefore, the Weyl coordinates cover only
the exterior of the horizon.
Now, the key point is that (A.4), the only field equation one has to solve, is linear. Hence
we can ‘distort’ the Schwarzschild solution simply by adding to ψS any solution ψD of the
flat space Laplace equation which is regular along the z−axis [35]. Thus, we can set

ψ = ψS + ψD and γ = γS + γD . (A.8)

Substituting these expressions into (A.4) and (A.5) and using the forms of the Schwarzschild
functions ψS and γS , one can show that at ρ = 0,

γD |ρ=0 =
b 2ψD |ρ=0 . (A.9)

This fact plays an important role in analyzing the horizon structure.

In Schwarzschild coordinates, the distorted metric takes the form

e2(γD −ψD )
ds2 = −e2ψD (1 − 2M/r)dt2 + dr 2 + e2(γD −ψD ) r 2 dθ2 + r 2 sin2 θe−2ψD dφ2. (A.10)
(1 − 2M/r)

48
As usual, the metric has a coordinate singularity at r = 2M. Let us therefore introduce the
Eddington-Finkelstein coordinate v as before. The metric can be re-expressed in (v, r, θ, φ)
coordinates as

ds2 = −e2ψD (1 − 2M/r)dv 2 + (1 − 2M/r)−1 e2ψD (e2(γD −2ψD ) − 1)dr 2

+ 2e2ψD dvdr + e2(γD −ψD ) r 2 dθ2 + e−2ψD r 2 sin2 θdφ2 . (A.11)

Using condition (A.9) it is not difficult to show that the coefficient of dr 2 in the metric is
regular at r = 2M [35].
∂
It is immediately obvious from (A.11) that the r = 2M surface is a Killing horizon of ∂v .
However, we can not select a preferred normalization for this vector field since the metric is
not asymptotically flat. As in the last sub-section, let ∆ be the Killing horizon and choose
∂
ℓ ∝ ∂v . Then, it is straightforward to verify that (∆, [ℓ]) is a complete, non-rotating isolated
∂
horizon. Let us calculate the value of surface gravity for ℓ = b ∂v . We obtain

1 1
b (e2ψD −γD )
κ= =
b (A.12)
2r 2r
where we arrived at the last expression by using (A.9). Again, while spin coefficient Re [µ],
the Weyl component Ψ2 and the scalar curvature 2R of the horizon metric all depend on the
distortion function ψD , somewhat surprisingly the surface gravity κ(ℓ) does not.
The natural question is whether the above framework can be extended to obtain distorted
Reissner-Nordström solutions. This turns out to be non-trivial because the key equation
(A.4) now acquires a source term from the electromagnetic field and this field itself depends
non-trivially on ψ through the Maxwell equations. At first, the coupled system appears to
be hopelessly difficult. However, there exists a prescription [38] for defining a new potential
ψ̃ in terms of ψ and the electromagnetic field such that ψ̃ satisfies the flat space Laplacian
(A.4). Using this method, the known distorted black hole solutions were recently generalized
to the charged case [36]. The distorted Reissner-Nordström solution is given by the metric

ds2 = −(1 − 2M/r + Q2 /r 2 )e2ψD dt2 + (1 − 2M/r + Q2 /r 2 )−1 e2(γD −ψD ) dr 2

+e2(γD −ψD ) r 2 dθ2 + e−2ψD r 2 sin2 θdφ2 . (A.13)

The forms of ψD and γD are now substantially more complicated than in the uncharged case.
Nonetheless, it is still possible to show that (A.9) continues to hold. As before this equality
2
implies that the apparent singularity at rH −2MrH + Q2 = 0 is only a coordinate singularity.
∂
The surface defined by r = rH is a Killing horizon of ∂t . There is once again, no natural
way to normalize the Killing field, so we only have an equivalence class [ℓa ] of null normals
to the Killing horizon. (∆, [ℓ]) is a non-rotating isolated horizon.
∂
The surface gravity of ∂t is given by
!
1 Q2
κ= 1− 2 . (A.14)
2rH rH

49
Again, the surface gravity is independent of the distortion of the horizon and has the same
dependence on the horizon radius R∆ (which turns out to be equal to rH ) and charge Q as in
Reissner-Nordström spacetime. Considerations of Section V suggest this peculiar behavior
of κ in all these examples is not accidental but can be ‘explained’ from general Hamiltonian
considerations which led us to the first law.

APPENDIX B: THE NEWMAN-PENROSE FORMALISM

1. Notation and Conventions

Let us begin with a summary of the Newman-Penrose formalism (see [39] or [40–42] for
a complete account). Apart from the spacetime signature which we take to be (−, +, +, +),
we will follow the conventions used in [42]. Consider a tetrad of null vectors n, ℓ, m and m
(n and ℓ are real while m is complex) which satisfy

n.ℓ = −1 n.m = 0 n.m = 0

(B.1)
ℓ.m = 0 ℓ.m = 0 m.m = 1 .

The directional derivatives along the basis vectors are denoted by

D = ℓ a ∇a ∆ = na ∇a δ = ma ∇a δ = ma ∇a . (B.2)

The full the information contained in the connection is expressed in terms of twelve complex
scalars called the Newman-Penrose spin coefficients defined as follows:
1
κ = −ma ℓb ∇b ℓa ǫ = 2
(ma ℓb ∇b ma − na ℓb ∇b ℓa ) π = ma ℓb ∇b na
1
σ = −ma mb ∇b ℓa β = 2
(ma mb ∇b ma − na mb ∇b ℓa ) µ = ma mb ∇b na
1
(B.3)
ρ = −ma mb ∇b ℓa α = 2
(ma mb ∇b ma − na mb ∇b ℓa ) λ = ma mb ∇b na
1
τ = −ma nb ∇b ℓa γ = 2
(ma nb ∇b ma − na nb ∇b ℓa ) ν = ma nb ∇b na .

It is sometimes more useful to express these definitions in terms of covariant derivatives of

the basis vectors:
Dℓ = (ǫ + ǫ)ℓ − κm − κm Dn = −(ǫ + ǫ)n + πm + πm
∆ℓ = (γ + γ)ℓ − τ m − τ m ∆n = −(γ + γ)n + νm + νm
δℓ = (α + β)ℓ − ρm − σm δn = −(α + β)n + µm + λm (B.4)
Dm = πℓ − κn + (ǫ − ǫ)m ∆m = νℓ − τ n + (γ − γ)m
δm = λℓ − σn + (β − α)m δm = µℓ − ρn + (α − β)m .

The ten independent components of the Weyl tensor are expressed in terms of five complex
scalars Ψ0 , Ψ1 , Ψ2 , Ψ3 and Ψ4 . The ten components of the Ricci tensor are defined in terms

50
of four real and three complex scalars Φ00 , Φ11 , Φ22 , Λ, Φ10 , Φ20 and Φ21 . These scalars are
defined as follows:
1 1
Ψ0 = Cabcd ℓa mb ℓc md Φ01 = R ℓa mb
2 ab
Φ10 = R ℓa m b
2 ab
1 1
Ψ1 = Cabcd ℓa mb ℓc nd Φ02 = R ma mb
2 ab
Φ20 = R ma mb
2 ab
a b c d 1 1
Ψ2 = Cabcd ℓ m m n Φ21 = R ma nb
2 ab
Φ12 = R ma nb
2 ab
(B.5)
1 1
Ψ3 = Cabcd ℓa nb mc nd Φ00 = R ℓa ℓb
2 ab
Φ11 = R (ℓa nb
4 ab
+ ma mb )
1 R
Ψ4 = Cabcd ma nb mc nd Φ22 = R na nb
2 ab
Λ = 24
.

The six components of the Electromagnetic-Field 2−form Fab can be defined in terms of
three complex scalars:
1
φ0 = −ℓa mb Fab φ1 = − (ℓa nb − ma mb )Fab φ2 = na mb Fab . (B.6)
2
The eight real Maxwell equations dF = 0 and d⋆ F = 0 can be written as a set of four complex
equations:

Dφ1 − δφ0 = (π − 2α)φ0 + 2ρφ1 − κφ2 (B.7)

Dφ2 − δφ1 = −λφ0 + 2πφ1 + (ρ − 2ǫ)φ2 (B.8)
∆φ0 − δφ1 = (2γ − µ)φ0 − 2τ φ1 + σφ2 (B.9)
∆φ1 − δφ2 = νφ0 − 2µφ1 + (2β − τ )φ2 . (B.10)

2. Boundary Conditions

In this section we describe the isolated horizon boundary conditions in the Newman-
Penrose formalism. We will restrict ourselves to Einstein-Maxwell theory with zero cosmo-
logical constant.
In a null-tetrad adapted to the null hypersurface ∆, take ℓ to be a null normal, m and
m tangent to ∆ and n transverse to ∆. Since ℓ is hypersurface orthogonal and null, it is
geodesic. This implies that16 κN P =
b 0 and Im [ρ] =
b 0. Thus

Dℓb := ℓa ∇a ℓb =
b (ǫ + ǫ)ℓb . (B.11)

The surface gravity is therefore given by κ(ℓ) = ǫ + ǫ and the expansion of ℓ is θ(ℓ) =
b Re [ρ].
For a non-expanding horizon ∆, the conditions on ℓ imply ρ = b 0 and the Raychaudhuri
equation then implies σ = b 0 and Φ00 = 12 Rab ℓa ℓb =
b 0. Furthermore, from (II.16) (which is

16 We will denote the NP spin coefficient κ by κN P to distinguish it from the surface gravity κ(ℓ) .

51
a consequence of the energy condition), it follows that φ0 =
b 0. This leads to the following
conditions on the Ricci tensor at the horizon
Φ00 =
b 0 Φ01 =
b 0 Φ10 =
b 0
(B.12)
Φ02 =
b 0 Φ20 =
b 0 Φ11 =
b −2Gφ1 φ1 .

The first Maxwell equation (B.7) gives

Dφ1 =
b 0 which implies DΦ11 =
b 0. (B.13)

Also, as shown in (II.8)

Ψ0 =
b 0 and Ψ1 =
b 0. (B.14)

The intrinsically defined one-form ωa defined in (II.4) is given by

ωa = −κ(ℓ) na + (α + β)ma + (α + β)ma . (B.15)

It is often convenient to choose the null tetrad such that dn

←−−
= 0 which implies

b µ
µ= and π=
b α+β. (B.16)

In this case we get a foliation of ∆ spanned by m and m. Furthermore, by an appropriate

b ǫ and thus the foliation is Lie
spin transformation, we can choose ǫ to be real so that ǫ =
dragged along ℓ:

L ℓ ma = (ǫ − ǫ)ma =
b 0. (B.17)

The one-form ω now becomes

ωa = −κ(ℓ) na + πma + π ma . (B.18)

Let us consider a weakly isolated horizon (∆, [ℓ]). The condition L ℓ ω = 0 is equivalent to
requiring

Lℓπ =
b 0 and L ℓ κ(ℓ) =
b 0 (B.19)

and as we proved in Section II B, these conditions imply that the surface gravity κ(ℓ) is
constant on ∆.
As mentioned in Section II B, a weakly isolated horizon with non-zero surface gravity
admits a natural foliation. In the Newman-Penrose framework this foliation can be charac-
terized as follows: It is the unique foliation on each leaf of which the pull-back of the 1-form
πma + π ma is divergence-free. This condition was first introduced by Háj́iček [43] in the
context of stationary spacetimes.
Finally, since our boundary conditions require that L ℓ κ(ℓ) = b 0, in a sense, a part of the
zeroth law is simply assumed. As mentioned in Section II B, we could have used a slightly

52
different set of boundary conditions which make no direct requirement on κ(ℓ) and yet lead
to the zeroth law (as well as the results of Sections III - VII).
Let (∆, [ℓ]) be a non-expanding horizon, equipped with an equivalence class [ℓ] of null
normals to related to each other by constant positive rescalings. As above introduce a null
tetrad where ℓ is an element of [ℓ], m and m are tangent to the foliation, n is curl free and
ǫ is real. In place of Definition 2, let us assume that ∆ admits a foliation by a family S∆
of 2−spheres transverse to [ℓ] such that the NP spin coefficients in an associated null-tetrad
satisfy:

Lℓµ =
b 0 and Lℓπ =
b 0. (B.20)

These conditions now replace the requirement L ℓ ω = b 0 used in the definition of a weakly
isolated horizon. We can prove the zeroth law from these conditions as follows. First,
consider the definition 2∇[a ∇b] ξc = Rabc d ξd of the Riemann tensor. In the NP formalism,
these are written as a set of 18 complex equations known as the ‘field equations’. For our
purposes, we need only three of these equations [42]

Dα − δǫ = (ρ + ǫ − 2ǫ)α + βσ − βǫ − κλ − κγ + (ǫ + ρ)π + Φ10

Dβ − δǫ = (α + π)σ + (ρ − ǫ)β − (µ + γ)κ − (α − π)ǫ + Ψ1
Dµ − δπ = (ρ − ǫ − ǫ)µ + σλ + (π − α + β)π − νκ + Ψ2 + 2Λ .

Adding the first equation to the complex conjugate of the second equation and imposing our
boundary conditions gives

δ(ǫ + ǫ) =
b δκ(ℓ) =
b 0 (B.21)

while the third equation reduces to

Ψ2 =
b (ǫ + ǫ)µ . (B.22)

Equation (B.21) tells us that surface gravity is constant on each leaf of the foliation. It now
only remains to show that it is also constant along ℓ. To show this we turn to the Bianchi
identity: ∇[a Rbc]de = 0. In the NP formalism, this is written as a set of nine complex and
two real equations. We shall need only two of these equations [42]

DΨ2 − δΨ1 + ∆Φ00 − δΦ01 + 2DΛ

= −λΨ0 + 2(π − α)Ψ1 + 3ρΨ2 − 2κΨ3 + σΦ02
+(2γ + 2γ − µ)Φ00 − 2(α + τ )Φ01 − 2τ Φ10 + 2ρΦ11
(B.23)
DΦ11 − δΦ10 + ∆Φ00 − δΦ01 + 3DΛ
= (2γ + 2γ − µ − µ)Φ00 + (π − 2α − 2τ )Φ01 + (π − 2α − 2τ )Φ10
+2(ρ + ρ)Φ11 + σΦ02 + σΦ20 − κΦ12 − κΦ21 .

53
 Subtracting these equations, imposing our boundary conditions and using Λ = 0, we get
DΨ2 = b 0. Combining this result with (B.22) gives D(ǫ + ǫ) = b 0. This completes the proof
of the zeroth law within the alternate definition of weak isolation. Most of the results of this
paper were first obtained using that definition. However, since that notion is tied so heavily
to the presence of a foliation, its intrinsic meaning is somewhat obscure. Therefore, it was
then replaced by Definition 2 used in the main body of the paper.

54
 REFERENCES
[1] J.D. Bekenstein, Black holes and entropy, Phys. Rev. D7 2333 (1973);
J.D. Bekenstein, Generalized second law of thermodynamics in black hole physics Phys.
Rev. D9 3292 (1974);
J.D. Bekenstein and A. Meisels, Einstein A and B coefficients for a black hole Phys.
Rev. D15 2775 (1977).
[2] J.W. Bardeen, B. Carter and S.W. Hawking, The four laws of black hole mechanics
Commun. Math. Phys. 31 161 (1973).
[3] B. Carter, Black hole equilibrium states Black Holes, ed B DeWitt and C DeWitt (Gor-
don and Breach: New York) (1973).
[4] S.W. Hawking, Particle creation by black holes Commun. Math. Phys. 43 199 (1975).
[5] M. Heusler, Black Hole Uniqueness Theorems (Cambridge University Press: Cambridge)
(1996).
[6] A. Ashtekar, A. Corichi and K. Krasnov, Isolated horizons: the classical phase space,
Adv. Theor. Math. Phys. 3 418-471 (1999) (Preprint gr-qc/9905089).
[7] A. Ashtekar, C. Beetle and S. Fairhurst, Isolated horizons: a generalization of black hole
mechanics Class. Quantum Grav. 16 L1 (1999).
[8] A. Ashtekar, C. Beetle and S. Fairhurst, Mechanics of isolated horizons, Class. Quantum
Grav. 17 253-298 (2000).
[9] A. Ashtekar and A. Corichi, Laws governing isolated horizons: inclusion of dilaton
coupling, Class. Quantum Grav. 17 1317-1332 (2000).
[10] R.M. Wald, Quantum Field Theory in Curved Spacetime and Black Hole Thermodynam-
ics (University of Chicago Press) (1994).
[11] G. Gibbons and S.W. Hawking, Cosmological event horizons, thermodynamics, and
particle creation Phys. Rev. D15 2738 (1977).
[12] A. Corichi and D. Sudarsky, Mass of colored black holes Phys. Rev. D61 101501 (2000);
A. Corichi, U. Nucamendi and D. Sudarsky, Einstein-Yang-Mills Isolated Horizons:
Phase Space, Mechanics, Hair and Conjectures (preprint gr-qc 0002078) (2000).
[13] A. Ashtekar, C. Beetle and J. Lewandowski, Mechanics of rotating isolated horizons (in
preparation) (2000); C. Beetle, Isolated horizons and black hole mechanics Penn State
Ph.D. dissertation (2000).
[14] D. Brown and J.W. York, Quasilocal energy and conserved charges derived from the
gravitational action Phys. Rev. D 47 1407–1419 (1993)(gr-qc/9209012);
[15] P.T. Chruściel, On the global structure of Robinson-Trautman space-time, Proc. Roy.
Soc. London A 436 299-316 (1992).
[16] C. Beetle and A. Corichi (in preparation).
[17] D. Kastor and J. Traschen, Phys. Rev. D 47 5370 (1993);
K. Nakao, T. Shiromizu and S. Hayward, Horizons of the Kastor-Traschen multi-black-
hole cosmos, Phys. Rev. D 52 796-808 (1995).
[18] A. Ashtekar, C. Beetle and J. Lewandowski, Geometry of general isolated horizons (in
preparation) (2000).

55
 A. Ashtekar, C. Beetle, O. Dreyer, S. Fairhurst, B. Krishnan, J. Lewandowski and J.
Wiśniewski, Generic Isolated Horizons and their Applications, Phys. Rev. Lett. In Press.
[19] A. Ashtekar, Radiative degrees of freedom of the gravitational field in exact general
relativity J. Math. Phys. 22, 2885-2895 (1981);
A. Ashtekar and M. Streubel, Symplectic geometry of radiative modes and conserved
quantities at null infinity Proc. R. Soc. (London) A376, 585-607 (1981).
[20] A. Ashtekar, J. Baez, A. Corichi and K. Krasnov, Quantum geometry and black hole
entropy Phys. Rev. Lett. 80 904-907 (1998);
A. Ashtekar, J. Baez, and K. Krasnov, Quantum geometry of isolated horizons and black
hole entropy, pre-print NSF-ITP-99-153 (1999).
[21] A. Ashtekar, Lectures on Non-Perturbative Canonical Gravity, Notes prepared in col-
laboration with R. S. Tate, (World Scientific) (1991).
[22] A. Ashtekar, L. Bombelli and O. Reula, The covariant phase space of asymptotically flat
gravitational fields Mechanics, Analysis and Geometry: 200 years after Lagrange ed. M
Francaviglia (North-Holland) (1991);
R.M. Wald and A. Zoupas, General Definition of “Conserved Quantities” in General
Relativity and Other Theories of Gravity Phys. Rev. D61 084027 (2000).
[23] D. Sudarsky and R.M. Wald, Extrema of mass, stationarity and staticity, and solutions
to the Einstein-Yang-Mills equations, Phys. Rev. 46 1453-1474 (1992).
[24] M. Heusler and N. Straumann, The first law of black hole physics for a class of non-linear
matter models, Class. Quant. Grav. 10 1299-1321 (1993).
[25] A.K.M. Masood-ul-Alam, Uniqueness of a static charged dilaton black hole, Class.
Quant. Grav. 10 2649-2656 (1993).
[26] G. Gibbons, R. Kallosh and B. Kolb, Moduli, scalar charges and the first law of black
hole thermodynamics Phys. Rev. Lett. 77 4992-4994 (1996).
[27] B. Kleihaus and J. Kunz, Static black holes with axial symmetry Phys. Rev. Lett. 79
1595-1598 (1998);
B. Kleihaus and J. Kunz, Static axially symmetric Einstein-Yang-Mills-Dilaton solu-
tions: 1. Regular solutions Phys. Rev. D57 834-856 (1998);
B. Kleihaus and J. Kunz, Static axially symmetric Einstein-Yang-Mills-Dilaton solu-
tions: 2. Black hole solutions Phys. Rev. D57 6138- 6157 (1998).
[28] R. Bartnik and J. McKinnon, Particlelike solutions of the Einstein-Yang-Mills Equations
Phys. Rev. Lett. 61 141-144 (1988);
P. Bizon, Colored Black Holes Phys. Rev. Lett. 64 2844-2847 (1990).
D. Brown and J.W. York, Microcanonical functional integral for the gravitational field
Phys. Rev. D 47 1420–1431 (1993) (gr-qc/9209014).
[29] R.M. Wald, The thermodynamics of black holes Living Reviews in Relativity (1999).
[30] B. Julia and S. Silva, Currents and superpotentials in classical gauge theories: II. Global
aspects and the example of affine gravity (preprint gr-qc/0005127) (2000).
[31] A. Ashtekar, O. Dreyer and J. Wiśniewski, Isolated horizons in 2+1 dimensions (in
preparation).
[32] R.M. Wald, Black hole entropy is Noether charge. Phys. Rev. D48 3427-31 (1993);

56
 V. Iyer and R.M. Wald, Some Properties of Noether charge and a proposal for dynamical
black hole entropy. Phys. Rev. D50 846-64 (1994).
[33] J. Lewandowski, Space-times admitting isolated horizons Class. Quantum Grav. (in
press) Preprint gr-qc/9907058.
[34] F.J. Ernst, Black holes in a magnetic universe J. Math. Phys. 17 No.1, 54-56 (1976).
[35] L. Mysak and G. Szekeres, Behavior of the Schwarzschild Singularity in Superimposed
Gravitational Fields Can.J.Phys. 44 617 (1966);
R. Geroch and J.B. Hartle, Distorted Black Holes J. Math. Phys. 23(4) 680 (1982).
[36] S. Fairhurst and B. Krishnan, Distorted black holes with charge (in preparation).
[37] H. Weyl, Zur Gravitationstheorie Ann. Physik 54 117-145 (1917).
[38] R. Gatreau, R.B. Hoffman and A. Armenti jr, Static Multi-Particle Systems in General
Relativity Il Nuovo Cimento 7B(1) 71 (1972).
[39] E.T. Newman and R. Penrose, An approach to gravitational radiation by a method of
spin coefficients J. Math. Phys 3 566-578 (1962).
[40] R. Penrose and W. Rindler, Spinors and Space-Time (Cambridge University Press:
Cambridge) (1984).
[41] S. Chandrasekhar, The Mathematical Theory of Black Holes (Oxford Classic Texts in
the Physical Sciences) (1998).
[42] J. Stewart, Advanced General Relativity (Cambridge Monographs on Mathematical
Physics) (1991).
[43] P. Háj́iček, Stationary electrovacuum spacetimes with bifurcate horizons J. Math. Phys.
16 518-522 (1975).

57