UCD-02-02

February 2002
gr-qc/0203001
arXiv:gr-qc/0203001v1 1 Mar 2002

Near-Horizon Conformal Symmetry

and Black Hole Entropy

S. Carlip∗
Department of Physics
University of California
Davis, CA 95616
USA

Abstract
Near an event horizon, the action of general relativity acquires a
new asymptotic conformal symmetry. Using two-dimensional dilaton
gravity as a test case, I show that this symmetry results in a chiral
Virasoro algebra with a calculable classical central charge, and that
Cardy’s formula for the density of states reproduces the Bekenstein-
Hawking entropy. This result lends support to the notion that the
universal nature of black hole entropy is controlled by conformal
symmetry near the horizon.

∗
email: [email protected]
1. Introduction
Since the seminal work of Bekenstein [1] and Hawking [2] in the early 1970s, we have
understood that black holes are thermodynamic objects, with characteristic temperatures
and entropies. The Bekenstein-Hawking entropy depends on both Planck’s constant h̄ and
Newton’s gravitational constant G, and offers one of the few known “windows” into quantum
gravity. In particular, an understanding of the microscopic statistical mechanics of black
hole thermodynamics may give us valuable information about the fundamental degrees of
freedom of quantized general relativity. Until quite recently, though, standard derivations of
the Bekenstein-Hawking entropy involved only macroscopic thermodynamics, and a statistical
mechanical description was more a hope than a reality.
In the past few years, this situation has changed dramatically. Today, indeed, we face the
opposite problem: we have many candidate descriptions of black hole statistical mechanics,
all of which yield the same entropy despite counting very different states. In particular, there
are two string theoretical descriptions, one based on counting D-brane states [3] and another
involving a dual conformal field theory [4]; an approach in loop quantum gravity that counts
spin network states [5]; and a slightly more obscure method [6] based on Sakharov’s old idea of
induced gravity [7]. The problem of “universality” is to explain why these approaches agree,
and why they agree with the original semiclassical computations [2, 8] that know nothing of
the details of quantum gravity.
One possible answer is that black hole thermodynamics may be controlled by a symmetry
inherited from the classical theory. This idea has its roots in an observation by Strominger [9]
and Birmingham et al. [10] that black hole entropy in three spacetime dimensions can be
obtained from Cardy’s formula [11] for the density of states of a two-dimensional conformal
field theory at the “boundary” of spacetime. A number of authors have tried to extend such
arguments to black holes in arbitrary dimensions [12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23],
but while these calculations seem to have the right “flavor,” none is yet fully satisfactory
[24,25,26,27,28]. In particular, all such proposals so far require awkward boundary conditions
at black hole horizons, and most have serious difficulties in two spacetime dimensions, where
there does not seem to be enough room at the one-dimensional horizon for the required degrees
of freedom.
In this paper, I point out three new ingredients that lead to an improved description of the
near-horizon symmetries of a black hole, and show how they may overcome these difficulties.
The new ingredients are the following:

1. Conformal symmetry: In the presence of a stationary black hole—or, more generally,

a black hole with a momentarily stationary region near its horizon—the Einstein-Hilbert
action of general relativity acquires a new conformal symmetry. Indeed, let ∆ be a
segment of such a horizon (see figure 1), and let N be a “momentarily stationary”
neighborhood, that is, a neighborhood admitting a Killing vector χa for which ∆ is a
Killing horizon. If f is an arbitrary smooth function that vanishes outside N , then
under the transformation
gab → ∇c (f χc )gab (1.1)

1
 ✁S
r
✁❤❤
✁ ❤❤❤❤
❤❤❤
✁ ❤❤C❤2
❤❤❤
✁ ❤❤❤❤
❤❤❤
✁ ❤❤
❤ i0
✥
∆✁ ✥✥ ✥✥
✥✥
✁N ✥✥✥
✁ ✥✥ ✥✥✥
✥✥✥
✁
✥✥✥✥✥ C1
✁ ✥
✥ ✥ ✥✥✥
✥
✁
✁

Figure 1: A black hole spacetime: horizon ∆, two partial Cauchy surfaces C1 and C2 , a
“reference” cross-section S of ∆, and a neighborhood N of the horizon.

the action in n dimensions transforms, up to possible boundary terms, as

1 Z c ab 1 n−2Z
δI = ∇c (f χ )g Gab ǫ = f χc ∇c Rǫ = 0, (1.2)
16πG N 16πG 2 N

where ǫ is the volume element and the last equality follows from the fact that χa is a
Killing vector. The addition of matter to the Lagrangian will not change this result, as
long as the matter fields have the same symmetries as the metric in N .
For (1.1) to be a genuine symmetry, of course, it must preserve the relevant space of
fields; that is, the new metric must also admit a Killing vector in N . It is straightforward
to check that this will be the case if
(χa ∇a )2 f = 0. (1.3)
Below, we shall generalize this argument to the case of an asymptotic symmetry, for
which (χa ∇a )2 f (x) approaches zero as x approaches ∆.
The transformation (1.1) is not a symmetry of the full Einstein-Hilbert action, of
course, since a generic metric admits no local Killing vector. If one is interested in
quantum gravitational questions about black holes, though, one should restrict the action
to field configurations in which a black hole is present [29]. For such configurations, a
symmetry of the form (1.1) is present at least as an asymptotic symmetry.
2. Horizon symplectic form: In the presence of a horizon, the canonical symplectic form
of general relativity—that is, roughly, the Poisson brackets—picks up a new contribution
from the horizon. This is most easily seen in the covariant canonical formalism [31], in
which the symplectic form for a collection of fields φ is given by an integral
Z
Ω[φ; δ1 φ, δ2 φ] = ω[φ; δ1 φ, δ2 φ] (1.4)
C

of a closed form ω over a (partial) Cauchy surface C. Consider the two surfaces C1 and
C2 of figure 1. The fact that ω is a closed form ensures that
Z ∆∩C2
ΩC1 [φ; δ1 φ, δ2 φ] = ΩC2 [φ; δ1 φ, δ2 φ] + ω[φ; δ1φ, δ2 φ], (1.5)
∆∩C1

2
 where the integral on the right-hand side is over the portion of the horizon joining C1
and C2 . For the isolated horizon boundary conditions of Ref. [30], the restriction of ω
to ∆ is exact, and the horizon integral can be absorbed in Ω. In general, though, there
is no reason to expect such a simple outcome. Instead, to define a symplectic structure
that is independent of the choice of Cauchy surface C, one must choose a “reference”
cross-section S of the horizon and define
Z Z ∆∩C
Ω̂C [φ; δ1 φ, δ2 φ] = ω[φ; δ1φ, δ2 φ] + ω[φ; δ1φ, δ2 φ] (1.6)
C S

where the second integral is over the portion of the horizon connecting S and C1 . This
term is already implicit in [30], where the boundary contribution to Ω is fixed in terms
of a reference cross-section that is used to determine the relevant “integration constant.”
The Poisson brackets thus include a contribution from the horizon itself. As we shall
see below, for an asymptotic symmetry of the sort we are interested in here, this horizon
contribution will dominate.

3. Asymptotic symmetry: The horizon of a generic black hole need not have a stationary
neighborhood N . The nonexpanding horizon boundary conditions of Ashtekar et al. [30],
for example, require a Killing vector only on the horizon itself. What we really need is
the notion of an asymptotic symmetry, in which the spacetime is “almost” stationary
as one approaches the horizon.
Traditionally, an “asymptotic symmetry” in general relativity has meant an exact
symmetry, i.e., a diffeomorphism, that preserves some extra asymptotic structure. Here
we have a slightly different situation: a symmetry of the action that may be exact only
at the horizon, but that can be made arbitrarily good by shrinking the neighborhood N
in which the parameter f has its support. This circumstance is probably best viewed
as an instance of a weakly broken symmetry. In particular, we can find an approximate
Killing vector χa near the horizon (e.g., in the manner of [32]) and a metric ḡ for which
χa is an exact Killing vector, and write g = ḡ + h, where h = 0 at the horizon. The
Lagrangian L[ḡ + h] is then invariant up to terms of order h, and it may be shown
that the would-be Noether current for the transformation (1.1) is conserved up to terms
of order h. While more work is required to fully understand this sort of symmetry, it
is reasonably clear that if h is sufficiently smooth, an asymptotic symmetry near the
horizon should become an exact symmetry for fields located on the horizon itself

2. The two-dimensional black hole

We can now ask whether the new symmetry (1.1) places any restrictions on black hole
thermodynamics. In general, one ought not expect a symmetry to determine anything as
“microscopic” as a density of states. There is one important exception, though: for a one- or
two-dimensional conformal symmetry described by a Virasoro algebra
c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n,0 (2.1)
12

3
with central charge c, the Cardy formula [11, 33] tells us that the number of states having
eigenvalue ∆ of L0 goes asymptotically as
 s 
 ceff
∆
ρ(∆) ∼ exp 2π with ceff = c − 24∆0 , (2.2)
 6 

where ∆0 is the lowest eigenvalue of L0 . The question is thus whether the symmetry (1.1)
can be described by such an algebra.
To answer this question, it is useful to focus on a particular example, two-dimensional
dilaton gravity. This is not as restrictive as it may seem, since general relativity in any
dimension can be dimensionally reduced via a Kaluza-Klein mechanism to two-dimensional
gravity coupled to “matter” fields, and I shall argue below that the extra fields do not affect
the conclusions. Still, this work should be considered a first step, which can presumably be
considerably generalized.
The action for dilaton gravity can be written in the form [34]
1 1
Z Z  
I= L= φR + 2 V [φ] ǫ, (2.3)
2G ℓ
where ǫ is the two-dimensional volume form and V is an arbitrary function of the dilaton field
φ. (The kinetic term for φ has been absorbed into φR by field redefinition.) Strictly speaking,
one cannot define the expansion of a null congruence in two dimensions, but the analog in
dilaton gravity is
1
ϑ = ℓ a ∇a φ (2.4)
φ
where ℓa is the null normal. All known exact black hole solutions, including dimensionally
reduced descriptions of higher-dimensional black holes, have null horizons with vanishing ϑ.
As in previous work [12, 13], we will start with a “stretched horizon,” in this case a null
surface ∆˜ with null normal ℓa for which ϑ is small but nonzero. Near a genuine horizon, we
can take ϑ to be a measure of how far we have “stretched” away; in the end, we will take the
limit ϑ → 0.
In two dimensions, the vector ℓa determines a unique “orthogonal” null vector na , such
that ℓa na = −1. We extend na from ∆ ˜ by requiring that

na ∇a nb = 0, (2.5)

from which it follows that

∇a ℓb = −κna ℓb , ∇a nb = κna nb (2.6)

where κ is the “surface gravity.” Note, though, that unlike a timelike or spacelike unit vector,
a null normal does not have a fixed normalization: by rescaling ℓa → f ℓa , one can change κ
˜ [30],
almost arbitrarily on a fixed null surface ∆

κ → ℓa ∇a f + κf, na ∇a f = 0, (2.7)

4
where the last condition ensures that (2.5) is preserved. We shall use this freedom below to
choose a convenient form for κ.
Observe from (2.6) that
∇a ℓb + ∇b ℓa = κgab , (2.8)
so ℓa is a conformal Killing vector. We will see later that the natural scaling of ℓa leads to a
surface gravity κ proportional to ϑ, so ℓa is actually an approximate Killing vector near the
horizon.
The application of the transformation (1.1) to two dimensions is a bit tricky, both because
a new field φ is present and because the field equations of dilaton gravity differ from those of
ordinary general relativity. In general, we should expect φ as well as gab to transform, and it
is easy to check that under a transformation

δgab = ∇c (f ℓc )gab = (ℓc ∇c f + κf )gab

δφ = (ℓc ∇c h + κh), (2.9)

the Lagrangian (2.3) satisfies δL ∼ ϑ. We thus have an asymptotic symmetry in the sense
described earlier. In particular, by restricting f and h to have their support in a small region
near a horizon, we can make the variation δI arbitrarily small. For now, the relationship of f
and h will remain unspecified; we shall see later that the choice that makes the transformation
(2.9) canonical actually implies that δL ∼ ϑ2 .
Equation (2.9) is not enough to determine the separate variations of ℓa and na . This is to
be expected, since the normalization of ℓa is not fixed; the only restriction, from (2.6), is that
na ∇a (nb δℓb ) = 0. We are thus free to choose δℓa = 0, which then implies that

ℓa δna = ℓc ∇c f + κf
δκ = ℓb ∇b (ℓc ∇c f + κf ) (2.10)
δs = ℓa ∇a (ℓb ∇b h + κh)

where s = ℓa ∇a φ = ϑφ. It follows that

[δ1 , δ2 ]gab = (ℓc ∇c {f1 , f2 } + κ{f1 , f2 })gab with {f1 , f2 } = (ℓa ∇a f1 )f2 − (ℓa ∇a f2 )f1 , (2.11)

giving the standard conformal algebra.

To express the transformations (2.9) in Hamiltonian form, we need the symplectic form
Ω of (1.6). This can be computed by Wald’s methods [31, 34]. For variations that have their
support only in a small neighborhood N of ∆, ˜ the main contribution will come from the
˜ ˜ one finds that
integral along ∆. Restricting the symplectic form of Ref. [34] to ∆,
1
Z  
Ω̂ = ℓa ∇a (δ1 φ)ℓb δ2 nb − ℓa ∇a (δ2 φ)ℓb δ1 nb ǫ̂, (2.12)
2G ˜
∆

˜ Since ∆
where ǫ̂ = n is the induced volume element on ∆. ˜ is null, one can integrate by parts,
and use (2.10) to obtain
1
Z
Ω̂[δ1 , δ2 ] = − (δ1 φδ2 κ − δ2 φδ1 κ) ǫ̂. (2.13)
2G ˜
∆

5
3. Hamiltonian and Virasoro algebra
The next question is whether the transformation (2.9) is canonical, that is, whether it is
generated by a “Hamiltonian” L. Such a Hamiltonian must satisfy [31]
1
Z h i
δL[f, h] = Ω̂[δ, δf,h ] = − δφ ℓb ∇b (ℓc ∇c f + κf ) − δκ (ℓc ∇c h + κh) ǫ̂ (3.1)
2G ˜
∆

where again s = ℓa ∇a φ and I have integrated by parts to obtain the last equality. The variation
δ can be thought of as an exterior derivative on the space of fields, and the integrability
condition for (3.1) is that δ 2 L[f, h] = 0. If we assume that the parameters f and h are
field-independent, so δf = δh = 0, it is easy to see that this condition requires that δs be
proportional to δκ. In particular, this proportionality must hold for variations of the form
(2.9), and this, together with the requirement that f and h be field-independent, implies that
κ ˜
= constant on ∆, (3.2)
s
sf = κh. (3.3)
Despite appearances, (3.2) is not a real restriction on the geometry, since it can always be
satisfied by rescaling ℓa as in (2.7). As noted earlier, this relation makes the transformation
(2.9) an even better approximate symmetry. Indeed, it may be checked that now
s
Z Z    
δ φRǫ = 2 na ∇a f (ℓa ∇a s − κs) + na ∇a f ℓa ∇a κ ǫ, (3.4)
κ
and the integrand goes as ϑ2 near ∆. While no corresponding suppression appears automati-
cally in the potential term in (2.3), the variation of that term can easily be arranged to be of
order ϑ2 by an appropriate choice of κ/s on ∆. ˜
With the relation (3.3) between h and f , (3.1) can be easily integrated, yielding
1 1
Z Z
L[f ] = s (2ℓa ∇a f + κf ) ǫ̂ = − (2ℓa ∇a s − κs) f ǫ̂. (3.5)
2G ˜
∆ 2G ˜
∆

We must next choose a basis for the functions f on ∆. ˜ Since the normalization of ℓa is not
fixed—even (3.2) determines it only up to a constant—the corresponding light cone coordi-
nate has no intrinsic physical meaning. There is, however, a natural coordinate on ∆,˜ the
dilaton φ itself, which by the two-dimensional version of the Raychaudhuri equation should
be monotonic on ∆. ˜ Let
z = e2πiφ/φ+ , (3.6)
where φ+ is the value of φ on the horizon, so z → 1 at ∆.∗ We can then choose a basis of
functions to be proportional to z n , with the proportionality constants determined from (2.11)
and the requirement that the fn satisfy the standard Diff(S 1 ) commutation relations:
φ+ n
fn = z , {fm , fn } = i(m − n)fm+n . (3.7)
2πs
∗
This choice is almost unique, in that φ+ is the only natural quantity in the theory having the right dimension.
In principle, though, we could have chosen w = z α to define our modes. This would leave the central charge
(3.12) unchanged, but would shift the Hamiltonian (3.10).

6
Note that with this choice, the consistency condition (1.3) is satisfied asymptotically:

2πn2 n ˜ → ∆.
(ℓa ∇a )2 fn = − sz → 0 as ∆ (3.8)
φ+
In terms of these modes, the Hamiltonian (3.5) becomes

ℓ a ℓ b ∇a ∇b φ
!
1 κ φ+ dz
Z
L[fn ] = − 1+2 sfn . (3.9)
2G s ˜
∆ κ 2πi z

On shell, though, ∇a ∇b φ ∝ gab [34], so the last term in (3.9) vanishes, giving

1 κ φ+ 2
L[fn ] = − δn0 . (3.10)
2G s 2π
It remains for us to compute the Poisson brackets {L[fm ], L[fn ]}. This is most easily done
directly from equation (3.1), and a straightforward computation yields
2πi s 3
{L[fm ], L[fn ]} = δfm L[fn ] = − n δm+n,0 . (3.11)
G κ
This may be recognized as the expression for a central term in the Virasoro algebra, with
central charge
24π s
c=− . (3.12)
G κ
Inserting (3.10) and (3.11) into the Cardy formula (2.2), we obtain a density of states
2πφ+
log ρ(L0 ) = , (3.13)
G
giving exactly the standard Bekenstein-Hawking entropy for the two-dimensional dilaton black
hole [34].
In contrast to previous work on Virasoro algebras at the horizon, this derivation has the
nice feature that the central charge (3.12) does not depend on the particular black hole being
considered. The algebra may therefore be viewed as a universal one, with different black holes
represented by different values (3.10) of L0 .
For simplicity, I have dealt only with two-dimensional black holes. An extension to higher
dimensions would clearly be of interest. As noted above, though, higher-dimensional general
relativity may be dimensionally reduced in the manner of Kaluza and Klein to two-dimensional
dilaton gravity coupled with extra “matter” fields (see, for example, [35]). It is fairly easy to
see that these added terms cannot contribute to the classical central charge (3.12), although
they might give quantum corrections. The algebra derived here is thus more universal than
it might seem.
As also noted above, we should probably worry further about the making the notion of an
“asymptotic symmetry” used here more rigorous. It may be useful to exploit a generalization
of the symmetry (1.1) that exists in the presence of a conformal Killing vector η a ,

∇a ηb + ∇b ηa = κgab . (3.14)

7
It is not hard to check that the transformation
n−2
 
c
gab → η ∇c f + κf gab (3.15)
2
leaves the Einstein-Hilbert action invariant provided that f is chosen to satisfy
Z
g ab ∇a κ∇b f ǫ = 0. (3.16)

Moreover, if the original metric gab admits a conformal Killing vector, it is easily checked that
the transformed metric does as well. Maintaining the condition (3.16) is more complicated, but
at least one solutions exists: if both κ and f are functions of a single null coordinate v, (3.16)
holds automatically, and is preserved by (3.15). Work on understanding the implications of
this extended symmetry is in progress.

Acknowledgments
I would like to thank Peter Beach Carlip, age 2 34 , for all the “black hole information” and
“quantum gravity” he gave me during this research. This work was also supported in part by
Department of Energy grant DE-FG03-91ER40674.

References
[1] J. D. Bekenstein, Phys. Rev. D7 (1973) 2333.

[2] S. W. Hawking, Nature 248 (1974) 30.

[3] A. Strominger and C. Vafa, Phys. Lett. B379 (1996) 99

[4] O. Aharony et al., Phys. Rept. 323 (2000) 183.

[5] A. Ashtekar et al., Phys. Rev. Lett. 80 (1998) 904.

[6] V. P. Frolov and D. V. Fursaev, Phys. Rev. D56 (1997) 2212.

[7] A. D. Sakharov, Sov. Phys. Dokl. 12 (1968) 1040; reprinted in Gen. Rel. Grav. 32 (2000)
365.

[8] G. W. Gibbons S. W. Hawking, Phys. Rev. D 15 (1977), 2752.

[9] A. Strominger, J. High Energy Phys. 9802 (1998) 009.

[10] D. Birmingham et al., Phys. Lett. B424 (1998) 275.

[11] J. A. Cardy, Nucl. Phys. B270 (1986) 186; H. W. J. Blöte, J. A. Cardy, and M. P.
Nightingale, Phys. Rev. Lett. 56 (1986) 742.

[12] S. Carlip, Phys. Rev. Lett. 82 (1999) 2828.

8
[13] S. Carlip, Class. Quant. Grav. 16 (1999) 2828.

[14] S. N. Solodukhin, Phys. Lett. B454 (1999) 213.

[15] R. Brustein, Phys. Rev. Lett. 86 (2001) 576.

[16] S. Das et al., Phys. Rev. D63 (2001) 024023.

[17] J. Jing and M.-L. Yan, Phys. Rev. D63 (2001) 024003.

[18] M. Cadoni and S. Mignemi, Phys. Rev. D59 (1999) 081501.

[19] M. Cadoni et al., Phys. Rev. D63 (2001) 125021.

[20] D. J. Navarro et al., Nucl. Phys. B580 (2000) 311.

[21] F.-L. Lin and Y.-S. Wu, Phys. Lett. B453 (1999) 222.

[22] S. Carlip, Phys. Lett. B508 (2001) 168.

[23] M.-I. Park, preprint hep-th/0111224.

[24] M.-I. Park and J. Ho, Phys. Rev. Lett.83 (1999) 5595.

[25] M.-I. Park and J. H. Yee, Phys. Rev. D61 (2000) 088501.

[26] S. Carlip, Phys. Rev. Lett. 83 (1999) 5596.

[27] O. Dreyer et al., Class. Quant. Grav. 18 (2001) 1929.

[28] J. Koga, Phys. Rev. D64 (2001) 124012.

[29] S. Carlip, Constrained dynamics and quantum gravity 1999, Nucl. Phys. Proc. Suppl. 88
(2000) 10.

[30] A. Ashtekar et al., Phys. Rev. D62 (2000) 104025.

[31] R. M. Wald, Phys. Rev. D48 (1993) 3427.

[32] R. A. Matzner, J. Math. Phys. 9 (1968) 1657.

[33] S. Carlip, Class. Quant. Grav. 15 (1998) 3609.

[34] J. Gegenberg et al., Phys. Rev. D51 (1995) 1781.

[35] J. H. Yoon, Phys. Lett. B451 (1999) 296.