Dynamical Horizons: Energy, Angular Momentum, Fluxes and Balance Laws

Dynamical Horizons: Energy, Angular Momentum, Fluxes and Balance Laws

Abhay Ashtekar* and Badri Krishnant
Center for Gravitational Physics and Geometry and Center for Gravitational Wave Physics,
Department of Physics, Penn State University, University Park, PA 16802, USA

(Dated: February 6, 2008)

Dynamical horizons are considered in full, non-linear general relativity. Expressions of fluxes of
energy and angular momentum carried by gravitational waves across these horizons are obtained.
Fluxes are local, the energy flux is positive and change in the horizon area is related to these fluxes.
The flux formulae also give rise to balance laws analogous to the ones obtained by Bondi and Sachs
at null infinity and provide generalizations of the first and second laws of black hole mechanics.

PACS numbers: 04.25.Dm, 04.70.Bw

O
O

(N

>
O

m ■
> ■

o :

00 .

o ■
■

o :

(N .
O ■

"o !

i

5h .

bo

Black holes which are themselves in equilibrium but in
possibly time-dependent space-times can be modelled by
isolated horizons [1]. Over the past three years, proper-
ties of isolated horizons were studied in detail. In par-
ticular, the framework enabled one to extend the laws of
black hole mechanics [2, 3] and has been used to extract
physics from initial data of widely separated black holes
[4] and from numerical simulations of the final phases of
black hole collisions [1, 5]. The purpose of this letter is
to outline a generalization of these ideas to fully dynami-
cal situations in which matter and gravitational radiation
can fall into black holes.

Our definition of a dynamical horizon is geared to prac-
tical applications in astrophysical situations, particularly
those considered in numerical relativity.

I. DEFINITION AND NOTATION

Definition: A smooth, three-dimensional, space-like
sub-manifold H in a space-time is said to be a dynamical
horizon if it is foliated by a preferred family of 2-spheres
such that, on each leaf S, the expansion 9m of a null
normal £ a vanishes and the expansion 9i n ) of the other
null normal n a is strictly negative.

Thus, a dynamical horizon H is a 3- manifold which
is foliated by marginally trapped 2-spheres. As shown
below, the area of these 2-spheres necessarily increases.
An example is provided by continuous segments of world
tubes of apparent horizons that naturally arise in numer-
ical evolutions of black holes. In contrast to event hori-
zons, dynamical horizons can be located quasi-locally;
knowledge of the full space-time is not required. The
condition that H be space-like is implied by a stronger
but physically reasonable restriction that the derivative
of 9(t) along n a be negative [6]. Finally, the requirement
that the leaves be topologically S 2 can be replaced by

'Electronic address: ashtekaragravity.psu.edu
t Present address: Max-Planck-Institut fur Gravitationsphysik,
Albert-Einstein-Institut, Am Muhlenberg 1, D-14476 Golm, Ger-
many; Electronic address: badkri@aei-potsdam.mpg.de

the weaker condition that they be compact. One can
show that the topology of S is necessarily S 2 if the flux
of matter or gravitational energy across H is non-zero. If
these fluxes were to vanish identically, H would become
isolated and replaced by a null, non-expanding horizon
[2]-

Dynamical horizons are closely related to Hayward's
trapping horizons [6]. However, while the definition of
trapping horizons imposes a condition on the derivative
of off H, our conditions refer only to geometric quan-
tities which are intrinsically defined on H. But in cases of
physical interest, the additional condition would be satis-
fied and dynamical horizons will be future, outer trapping
horizons. Nonetheless, our analysis and results differ con-
siderably from those of Hayward's. While his framework
is based on a 2+2 decomposition, ours will be based on
the ADM 3+1 decomposition. Our discussion includes
angular momentum, our flux formulae are new and our
generalization of black hole mechanics is different. Our
analysis is geared to providing tools to extract physics
and perform checks on numerical simulations of dynam-
ical black holes. Therefore we will restrict ourselves to
dynamical horizons with zero charge.

Let us begin by fixing notation. Let r a be the unit
time-like normal to H and denote by V a the space-time
derivative operator. The metric and extrinsic curvature
of H are denoted by q a t and K a f, :— q a c qb d ^cTd respec-
tively; D a is the derivative operator on H compatible
with q a b and 1Z a b its Ricci tensor. Leaves of the preferred
foliation of H will be called cross- sections of H. The unit
space-like vector orthogonal to S and tangent to H is de-
noted by r a . Quantities intrinsic to S will be generally
written with a tilde. Thus, the two-metric on S is q a b, the
extrinsic curvature of S C H is K a b '■= Q a °% DcTd, the
derivative operator on (S, q a b) is D a and its Ricci tensor
is IZab- Finally, we will fix the rescaling freedom in the
choice of null normals via l a := r a +r a andn a := f a — r a .

We first note an immediate consequence of the defini-
tion. Since 6m = and 9i n ) < 0, it follows that K > 0.
Hence the area as of S increases monotonically along r a .
Thus the second law of black hole mechanics holds on H .
We will obtain an explicit expression for the change of
area in part III.

2

Our main analysis is based on the fact that, since H
is a space-like surface, the Cauchy data {q a b,K ab ) on H
must satisfy the usual scalar and vector constraints

H s := K + K 2 - K ab K ab = \^GT ab r a r b (1)
Hy := D b (K ab - Kq ab ) = 8irGT bc ? c q a b . (2)

We will often focus our attention on a portion AH C H
which is bounded by two cross-sections Si and S 2 -

II. ANGULAR MOMENTUM

The angular momentum analysis is rather straight for-
ward and is, in fact, applicable to an arbitrary space-like
hypersurface. Fix any vector field ip a on H which is tan-
gential to the cross-sections of H. Contract ip a with both
sides of eqn. (2). Integrate the resulting equation over
the region AH C H , perform an integration by parts and
use the identity L v q ab = 2D^ a ip^ to obtain

^i K ^ a?bd2V -^i K ^ bd2V

^ Aff (^^li rt ^) rf3y (3)

where P ab := K ab - Kq ab . It is natural to identify the
surface integrals with the generalized angular momentum
associated with those surfaces and set:

J s ] --^cj s K ^ a r b d 2 V (4)

where we have chosen the overall sign to ensure compat-
ibility with conventions normally used in the asymptoti-
cally flat context. The term 'generalized' emphasizes the
fact that the vector field ip a need not be an axial Killing
field even on S; it only has to be tangential to our cross-
sections.

The flux of this angular momentum due to matter
fields and gravitational waves are respectively

Ji v) = - f T ab T a v »d 3 V, (5)

JAH

J™ = -^af^^V, (6)
and we get the balance equation

4 v) -j[ v) =jM+jM. (7)

As expected, if (p a is a Killing vector of the three-metric
q a b, then the gravitational flux vanishes: = 0.

It is convenient to introduce the angular momentum
current j v := —K ab ip a r b so that (4) becomes —
(8TrG)-i$ s j*d*V.

III. ENERGY FLUXES AND AREA BALANCE

As is usual in general relativity, the notion of energy is
tied to a choice of a vector field. Here, we will consider
vector fields £ a = N£ a where the lapse N is constructed
as follows. Let r be a radial coordinate on H defined such
that the cross-sections of H are level surfaces of r. Then
r a oc D a r. It turns out that in order to get the balance
law for energy, we must tie our lapse functions N to radial
coordinates such that D a r = N r r a . (Since £ a = N£ a ,
as usual the term 'lapse' refers to space-time evolution;
not to 'evolution' along f a .) Thus each r determines a
permissible lapse function N r . If we use a different radial
coordinate r' , then the lapse is rescaled according to the
relation

fir'

N r ,=N r —. (8)
ar

Thus, although the lapse itself will in general be a func-
tion of all three coordinates on H, the relative factor be-
tween any two permissible lapses can be a function only
of r. We denote the resulting permissible vector fields by
£" r ) := N r £ a . Recall that, on an isolated horizon, physi-
cal fields are time independent and null normals can be
rescaled by a positive constant [2] . Now the horizon fields
are 'dynamical', i.e., r-dependent, and the rescaling free-
dom is by a positive function of r.

We are interested in calculating the flux of energy
associated with £^ for any radial coordinate r. De-
note the flux of matter energy across AH by J 7 ^ '■=
J AH T ab T a ^ b r - ) d 3 V. By taking the appropriate combina-
tion of (1) and (2) we obtain

T ™ =ii^J AH Nr {Hs + 2?aH ^ } d3y ■ (9)

Since H is foliated by two-spheres, we can perform a 2 + 1
split of the various quantities on H. Using the Gauss
Codazzi relation we rewrite 1Z in terms of quantities on
S:

K = K + K 2 - K ab K ab + 2D a a a (10)

where a a = r b D b r a - r a D b r b . Next, the fact that the
expansion 8^ of £ a vanishes leads to the relation

K + K = K ab r a r b . (11)

Using (10) and (11) in eqn. (9) and simplifying, we obtain
the result

/ N r Ud 3 V = 16nG [ T ab ? a tf r) d 3 V

J AH JAH

+ [ AV{|a| 2 + 2|C| 2 } d 3 V (12)

JAH

where \a\ 2 = <7 ab a ab with <j ab := q a c q b d W c £ d -
\qabq cd ^c£d, the shear of f, and |C| 2 = CCa with

3

£<j ._ q ab r c \7 c £ b . Both u ab and £ a are tensors intrin-
sic to S. To simplify the left side of this equation, note
that the volume element d 3 V on H can be written as
d 3 V = N~ 1 dr d 2 V where d 2 V is the area element on S.
Using the Gauss-Bonnet theorem, the integral of N r lZ
can then be written as

/ N r Kd 3 V = ( dr((fnd 2 v)=8ir(r2-r 1 ).
Jam J ri \Js J

(13)

Substituting this result in cqn. (12) we finally obtain

This is the key result we were looking for. Let us now
interpret the various terms appearing in this equation.
The first integral on the right side of this equation is

(r)

the flux Tm of matter energy associated with the vector
field Since £?s is null and f time-like, if T a b sat-

isfies, say, the dominant energy condition, this quantity
is guaranteed to be non-negative. It is natural to inter-
pret the second term as the flux Tg r ^ of £" r ^ -energy in the
gravitational radiation:

9 ■ 16nGj AH

iV r {| ( r| 2 + 2|C| 2 } d 3 V. (15)

This expression shares four desirable features with the
Bondi-Sachs energy flux at null infinity. First, it does
not refer to any coordinates or tetrads; it refers only to
the given dynamical horizon H and the evolution vec-
tor field £( r y Second, the energy flux is manifestly non-
negative. Third, all fields used in it are local; we did not
have to perform, e.g., a radial integration to define any
of them. Finally, the expression vanishes in the spheri-
cally symmetric case: if the Cauchy data (q a b, Kab) and
the foliation on H is spherically symmetric, a a b = and
C a =0.

To conclude this section, let us choose for our radial
coordinate the area radius R := a/ Air. Then,

R2 _R]_ = wh) , wh)
2G 2G m 9

(16)

Thus, as promised in part I, we have obtained an explicit
formula relating the change in the area of the horizon to
fluxes of matter and gravitational £(fl)-energy.

IV. MASS AND THE FIRST LAW

Let us now combine the results of parts II and III to
obtain the physical process version of the first law for H
and a mass formula for an arbitrary cross-section of H.

Denote by E*( RS > the £(#)-energy of cross-sections S of
H. While we do not yet have the explicit expression for

it, we can assume that, because of the influx of matter
and gravitational energy, will change by an amount

A E^ R ) = Tm^ +Fg R ^ as we move from one cross section
to another. Therefore, if we define effective surface grav-
ity ha associated with £^ as Ur := 1/2R, the infinites-
imal form of (16) implies (k R /8wG)da = dE^ R K For a
general choice of the radial coordinate r, (14) yields a
generalized first law:

8ttG

da = dE^

(17)

where the effective surface gravity k r of is given by

■ (18)

dR

where

Mr

(r)

This rescaling freedom in surface gravity is analogous to
the rescaling freedom which exists for Killing horizons,
or more generally, isolated horizons. The new feature in
the present case is that we have the freedom to rescale
the surface gravity (and £ a ) by a positive function of the
radius instead of just by a constant. This is just what
one would expect in a dynamical situation. Finally, note
that the differentials appearing in (17) are actual varia-
tions along the dynamical horizon due to an infinitesimal
change in r and are not variations in phase space as in
some of the formulations [2, 3, 7] of the first law.

To include rotation, pick a vector field <p a on H such
that ip a is tangent to the cross-sections of H, has closed
orbits and has affinc length 2tt.( At this point, (p a need
not be a Killing vector of q a b-) Consider time evolution
vector fields t a which are of the form t a = N r i a — tttp a
where N r is a permissible lapse associated with a radial
coordinate r and il an arbitrary function of r. Evaluate
the quantity J AH T ab ? a t b d 3 V using (3) and (14):

T2 - n
2G

Mi nrd ' v -i ar " 2v

- [ 2 dQ(f f d 2 v\ = [ T ab ? a t b d 3 V

Jfl! JS J J AH

-tL L sip ° bc « q " >d3v ■ (19)

Again, if we denote by E 1 the t-energy associated with
cross-sections S of H, the right side of (19) can be inter-
preted as AE l . If we now restrict ourselves to infinites-
imal AH, the three terms in the curly brackets combine
to give d(ttJ) — Jdfl and we obtain

dr — k r

h FldJ = — —

2G 8irG

da + ndJ = dE l .

(20)

This equation is our generalization of the first law for dy-
namical horizons. Since the differentials in this equation
are variations along H, this can be viewed as a 'physi-
cal process version of the first law'. Note that for each

4

allowed choice of lapse N r , angular velocity O(r) and vec-
tor field tp a on H, we obtain a permissible time evolution
vector field t a = N r £ a — fl(p a and a corresponding first
law. This situation is very similar to what happens in the
isolated horizon framework where we obtain a first law for
each permissible time translation on the horizon. Again,
the generalization from that time independent situation
consists of allowing the lapse and the angular velocity to
become r-dependent, i.e., 'dynamical'.

For every allowed choice of (N r , 0(r), ip a ), we can in-
tegrate eqn. (20) on H to obtain a formula for E* on any
cross section but, in general, the result may not be ex-
pressible just in terms of geometric quantities defined lo-
cally on that cross-section. However, in some physically
interesting cases, the expression is local. For example,
In the case of spherical symmetry, it is natural to choose
$1 = and R as the radial coordinate in which case we
obtain E* = R/2G. This is just the irreducible (or Hawk-
ing) mass of the cross-section. Even in this simple case,
(19) provides a useful balance law, with clear-cut inter-
pretation. Physically, perhaps the most interesting case
is the one in which q a b is only axi-symmetric with ip a
as its axial Killing vector. In this case we can naturally
apply, at each cross-section S of H, the strategy used
in the isolated horizon framework to select a preferred
t a : Calculate the angular momentum J defined by the
axial Killing field ip, choose the radial coordinate r (or
equivalently, the lapse N r ) such that

and choose £1 such that

R 4 - AG 2 J 2
2R 3 VR i + 4G2J2

n = n o (R) :=

2GJ

rTWTW 2 ^'

(21)

(22)

This functional dependence of k r on R and J is exactly
that of the Kerr family. (The condition on surface gravity
can always be implemented provided the right side of (21)
is positive, which in the kerr family corresponds to non-
extremal horizons. The resulting r and N r are unique.)
With this choice of N r and fl, the energy Eg is given by
the well known Smarr formula

E

2(

k a a
8^G

VR 4 + AG 2 J 2
2GR

(23)

as a function of its angular momentum and area, each
cross-section is assigned simply that mass which it would
have in the Kerr family. However, there is still a balance
equation in which the flux of gravitational energy T^°^ is
local and positive definite (see (19)). (The gravitational
angular momentum flux which, in general, has indeter-
minate sign vanishes due to axi-symmetry.) Motivated
by the isolated horizon framework, we will refer to this

canonical E ° as the

associated with cross-sections

S of H and denote it simply by M. Thus, among the
infinitely many first laws (20), there is a canonical one:

dM

k Q

da + fl dJ .

, 8ttG

We conclude with three remarks

(24)

i) Note that the mass and angular momentum depend
only on local fields on each cross section S and changes
in these quantities over finite regions AH of H have been
related to matter and gravitational radiation fluxes, de-
termined by the local geometry of H.

ii) Unlike the vector fields — N r £ a , general permis-
sible vector fields t a is not necessarily causal. Therefore
the matter flux J AH T a i,t a T h d 3 V need not be positive.
Similarly, if ip a is not a Killing field of q a b, the gravita-
tional flux need not be positive. Therefore, although the
area a always increase along f a , E* can decrease. This
is the analog of the Penrose process in which 'rotational
energy' is extracted from the dynamical horizon.

iii) While the infinitesimal version eq (20) of the first
law is conceptually more interesting, the finite balance
equation (19) is likely to be more directly useful in the
analysis of astrophysical situations. In particular, the
presence of an infinite number of these balance equations
can provide useful checks on numerical simulations in the
strong field regime.

Acknowledgements We would like to thank Chris
Beetle, Steve Fair hurst and Jerzy Lewandowski for stim-
ulating discussions. This work was supported in part
by the NSF grant PHY-0090091, the NSF Cooperative
Agreement PHY-0114375 and the Eberly research funds
of Penn State. BK was also supported through Duncan
and Roberts fellowships.

[1] A. Ashtekar, C. Beetle, O. Dreyer, S. Fairhurst, B. Krish-

nan, J. Lewandowski and J. Wisniewski, Phys. Rev. Lett.

85, 3564-3567 (2000).
[2] A. Ashtekar, S. Fairhurst, and B. Krishnan, Phys. Rev.

D62 104025 (2000).
[3] A. Ashtekar, C. Beetle, and J. Lewandowski, Phys. Rev.

D64, 044016 (2001).
[4] B. Krishnan, Thesis dissertation, Penn State (2002); A.

Ashtekar and B. Krishnan (in preparation).
[5] O. Dreyer, B. Krishnan, E. Schnetter and D. Shoemaker,

gr-qc/0206008.
[6] S. Hayward, Phys. Rev. D49, 6467-6474 (1994).
[7] R. M. Wald, Quantum Field Theory in Curved Spacetime

and Black Hole Thermodynamics (University of Chicago

Press) (1994).