arXiv:0710.

4334v3 [hep-th] 28 Sep 2008

Local bulk operators in AdS/CFT and the fate of the BTZ

singularity

Alex Hamilton, Daniel Kabat, Gilad Lifschytz, and David A. Lowe

Presented by DK at the 2007 Sowers workshop and GL at the pre-strings 2007 workshop

Abstract. This paper has two parts. First we review the description of lo-
cal bulk operators in Lorentzian AdS in terms of non-local operators in the
boundary CFT. We discuss how bulk locality arises in pure AdS backgrounds
and how it is modified at finite N . Next we present some new results on BTZ
black holes: local operators can be defined inside the horizon of a finite N BTZ
black hole, in a way that suggests the BTZ geometry describes an average over
black hole microstates, but with finite N effects resolving the singularity.

1. Introduction
Quantum gravity in asymptotically anti-de Sitter space is dual to a conformal
field theory on the boundary of AdS [Mal98]. One of the most interesting questions
raised by this duality is: how does approximately local bulk gravitational physics
emerge from the CFT?
To address this one needs some way of probing local physics in the bulk. We are
mostly interested in the semiclassical limit of small Planck length. In this limit we
should be able to recover the traditional results of quantum field theory in curved
space [BD82]. So it’s natural to ask: how can a local quantum field in the bulk of
AdS be represented in the boundary CFT?
This question was addressed in [BDHM98, BGL99, Ben00] and was further
developed by the present authors in [HKLL06a, HKLL06b, HKLL07]. In the
latter works it was shown that local operators in the bulk could be represented as
non-local operators in the CFT. The CFT operators turn out to have support on
a compact region of the complexified boundary. This representation makes several
properties manifest. It makes it clear why bulk locality is exact at large N , but
breaks down at finite N , in exactly the manner required by holography. It also

1991 Mathematics Subject Classification. Primary 81T30, 81T20; Secondary 83C57.

Key words and phrases. AdS/CFT correspondence, BTZ black hole.
AH was supported by a Columbia University Initiatives in Science and Engineering grant
and by DOE grant DE-FG02-92ER40699.
DK was supported by DOE grant DE-FG02-92ER40699.
GL is supported in part by Israeli science foundation grant number 568/05.
DL was supported by DOE grant DE-FG02-91ER40688-Task A.
1
2 ALEX HAMILTON, DANIEL KABAT, GILAD LIFSCHYTZ, AND DAVID A. LOWE

provides a simple CFT description of the horizon and singularity of a BTZ black
hole in the large N limit.
An outline of this paper is as follows. In section 2 we review the representation
of local bulk operators in terms of operators on the complexified boundary. In
section 3 we use these boundary operators to discuss bulk locality and holography
from the point of view of the CFT. In section 4 we extend the construction to
the BTZ black hole and discuss the horizon and singularity in the large N limit.
We conclude in section 5 with some speculation on the fate of the horizon and
singularity at finite N . Sections 2 – 4 are a review; the results in section 5 are new.

2. Local operators in the semiclassical limit

In Poincaré coordinates the metric on Lorentzian AdSD is
R2

(2.1) ds2 = 2 −dT 2 + |dX|2 + dZ 2 .
Z
Here R is the AdS radius. The Poincaré horizon is at Z = ∞, while the CFTd=D−1
lives on the boundary at Z = 0. Consider a scalar field of mass m in AdS, with
normalizable fall-off near the boundary.
φ(T, X, Z) ∼ Z ∆ φ0 (T, X) as Z → 0
The parameter ∆ is related to the mass of the field by
r
d d2
∆= + + m2 R 2 .
2 4
We will refer to φ0 as the boundary field. It’s dual to an operator of dimension ∆
in the CFT.
(2.2) φ0 (T, X)SUGRA ↔ O(T, X)CFT
The question is, can we express φ in terms of φ0 ? If so, then we can use (2.2) to
find the CFT operator dual to a local operator in the bulk.
For now we’ll study this in the semiclassical limit
ℓS , ℓP → 0 in the bulk
N, λ → ∞ on the boundary
Here ℓS and ℓP are the bulk string and Planck lengths, while N and λ are parameters
for some kind of ’t Hooft large-N expansion in the CFT whose details won’t matter
for us. The basic idea is to represent
Z
φ(T, X, Z) = dT ′ dd−1 X ′ K(T ′ , X′ |T, X, Z) φ0 (T ′ , X′ )

using a kernel or smearing function K. Since AdS has a timelike boundary, this
is not a standard Cauchy problem, and neither existence nor uniqueness of K is
guaranteed. Indeed in [HKLL06a] we discuss examples where both existence and
uniqueness are violated.
A cure for these problems, at least in a pure semiclassical AdS background, is
to make a Wick rotation to de Sitter space. Define a new set of boundary spatial
coordinates by setting X = iY. This turns the AdS metric (2.1) into
R2

(2.3) ds2 = 2
−dT 2 − |dY|2 + dZ 2 .
Z
 LOCAL BULK OPERATORS IN ADS/CFT AND THE FATE OF THE BTZ SINGULARITY 3

bulk point

dS boundary Y

Figure 1. The field at a bulk point in de Sitter space can be

expressed in terms of data on the past de Sitter boundary. The slice
Y = 0 also describes a region in AdS. So we can also regard this
as expressing the field in AdS in terms of data on the complexified
AdS boundary.

This is de Sitter space in flat FRW coordinates, with Z playing the role of conformal
time. The past boundary of de Sitter space is at Z = 0.
In de Sitter space we have a standard Cauchy problem. As shown in Fig. 1 we
can use a retarded de Sitter Green’s function to solve for the bulk field in terms
of data on the past boundary. The explicit analytic expressions are pretty simple:
the field at a point in AdS can be expressed as [HKLL07]

Z  2 ∆−d
Γ ∆ − d2 + 1 ′ d−1 ′ Z − T ′2 − |Y′ |2
φ(T, X, Z) = dT d Y
π d/2 Γ(∆ − d + 1) T ′ 2 +|Y′ |2 <Z 2 Z

(2.4) φ0 (T + T ′ , X + iY′ ) .

Note that we have to integrate over a compact region of the de Sitter boundary (the
region inside the past light-cone of the bulk point). Equivalently we integrate over
a compact region of the complexified AdS boundary (the region spacelike separated
from the bulk point).
By construction this lets us reproduce bulk correlation functions in the semi-
classical limit.
(2.5) Z
hφ(x1 , Z1 )φ(x2 , Z2 )iSUGRA = dx′1 dx′2 K(x′1 |x1 , Z1 )K(x′2 |x2 , Z2 )hO(x′1 )O(x′2 )iCF T

This is guaranteed to work, just because φ0 and O have identical correlators. Al-
though somewhat trivial, this result has an interesting corollary. In the semiclassical
limit of vanishing Planck length, bulk operators will commute at spacelike separa-
tion. Therefore the corresponding smeared boundary operators will commute with
each other. This is true even though, as shown in Fig. 2, the smearing functions
might overlap on the boundary.
4 ALEX HAMILTON, DANIEL KABAT, GILAD LIFSCHYTZ, AND DAVID A. LOWE

Figure 2. Smearing functions for two bulk points separated only

in the Z direction. The smearing functions overlap on the bound-
ary, nonetheless the smeared operators commute at infinite N .

3. Bulk locality and holography at finite N

What we’ve done so far is exact in the semiclassical limit; it can be regarded as
a set of statements about free wave equations in a pure AdS background. In this
section we’ll remain in a pure AdS background, but ask what happens at finite N .
First we need to decide what smearing functions to use. One possibility is to
use the same smearing functions at finite N . For example, in N = 4 Yang-Mills
the operator
Z
(3.1) Φ(T, X, Z) = dT ′ d3 X ′ K(T ′ , X′ |T, X, Z) TrF 2

can be defined at any N , where K is the kernel appearing in (2.4). In a pure AdS
background we believe these must be the right operators to use at finite N , just
because the construction is singled out by the symmetries. To see this, introduce a
distance function on AdS.
(3.2)  
geodesic distance Z 2 + Z ′2 + |X − X′ |2 − (T − T ′ )2
σ(T, X, Z|T ′ , X′ , Z ′ ) = cosh =
R 2ZZ ′

σ is invariant under AdS isometries. In Poincaré coordinates the isometry

(T, X, Z) → λ(T, X, Z)

acts as a scale transformation on the boundary. Z has conformal weight −1, and
since K ∼ limZ ′ →0 (σZ ′ )∆−d , we see that K transforms covariantly under AdS
isometries with conformal weightR d − ∆. But given an operator of dimensions ∆,
this is exactly what we need for dd x KO to behave like a scalar field in the bulk!
 LOCAL BULK OPERATORS IN ADS/CFT AND THE FATE OF THE BTZ SINGULARITY 5

T ∆X

Figure 3. The smearing functions have support on the two jagged

lines. For |∆X| > Z they are spacelike separated.

That is, the smearing functions we have defined provide the unique covariant way
to map a primary field in the CFT to a scalar field in the bulk.12
This might seem very strange from the point of view of holography. In (3.1)
we’ve defined a continuous infinity of bulk operators. How can this be compatible
with the holographic bound [SW98], which should only allow a finite number of
degrees of freedom in any given region in the bulk?
We believe the resolution is that at finite N not all the operators defined in (3.1)
commute at spacelike separation. To see this, consider a fixed-T hypersurface in the
bulk, with operators placed at some radial position Z. The smearing functions we
have defined have an extent in the time direction ∆T = Z.3 As shown in Fig. 3 the
operators will be spacelike separated on the boundary provided the bulk operators
have a spatial separation |∆X| > Z. In this case the operators are guaranteed to
commute, just by locality of the boundary theory. At finite N we do not expect
operators separated by |∆X| < Z to commute. (This is unlike the semiclassical
situation discussed in section 2, where operators could commute even though they
overlapped.) So we expect 1/Z d−1 commuting operators per coordinate area on the
boundary. Using the AdS metric (2.1), this means we expect 1/Rd−1 commuting
operators per proper area in the bulk. Equivalently we expect one commuting
operator per unit cell, where the cell volume ∼ Rd−1 . At finite N , bulk locality
breaks down on distances set by the AdS radius of curvature!
This is a bit disturbing, in that it seems the commuting operators we can build
from TrF 2 aren’t sufficient to describe a local bulk dilaton on distances less than
an AdS radius. We don’t have a complete resolution of this puzzle. It’s probably
too strong a condition to require that all operators describing the dilaton commute
exactly. Take the set of commuting operators built from TrF 2 . Perhaps there are
1Thus the smearing functions should have a purely group-theoretic interpretation in terms
of representations of the (complexified) isometry group SO(d, 2), along the lines of [VK91]. We
are grateful to Djordje Minic for discussions on this point.
2
To avoid a possible confusion: by construction, the operators defined in (3.1) satisfy a free
wave equation. So they do not have the right interactions to be identified with the bulk dilaton
field at finite N . Nevertheless, they should serve as good probes of local physics in the bulk.
3They also have an extent in the imaginary spatial directions, but as shown in [HKLL07]
it’s only the extent in time that matters here.
6 ALEX HAMILTON, DANIEL KABAT, GILAD LIFSCHYTZ, AND DAVID A. LOWE

additional operators which can be used to describe the dilaton, which do not all
commute, but whose commutators are so small at low energies that they can be
ignored.4 The operators built from TrF 2 using our smearing functions are good
first candidates for the job, since their commutators do vanish in the large-N limit.
But it could also be that operator mixing is important, so that operators besides
TrF 2 can contribute.
One might also wonder about the holographic bound. Here things work out
very nicely. The holographic bound in AdS states that the entropy per longitudinal
coordinate area is bounded by s < N 2 /Z d−1 . Each operator in the CFT gives
an entropy density ∼ 1/Z d−1 , so bound is saturated if we have N 2 commuting
operators in the CFT. This seems quite reasonable, as N = 4 Yang-Mills involves
N × N matrices and has a central charge ∼ N 2 .

4. Semiclassical BTZ black hole

We now turn to excitations of AdS3 , in particular we will study non-extremal
BTZ black holes. But first let’s consider AdS3 in accelerating or Rindler-like coor-
dinates.
r2 − r02 2 R2
(4.1) ds2 = − dt + dr2 + r2 dφ2
R2 r2 − r02
−∞ < t, φ < ∞ 0<r<∞
Here r0 is an arbitrary parameter with units of length.5 We’ll frequently work in
terms of the rescaled coordinates
t̂ = r0 t/R2 φ̂ = r0 φ/R .
It’s straightforward to construct smearing functions in these coordinates. The Wick
rotation φ = iy turns the AdS3 metric (4.1) into de Sitter space, now expressed in
static coordinates.
r2 − r02 2 R2
(4.2) ds2 = − dt + dr2 − r2 dy 2
R2 r2 − r02
−∞ < t < ∞ y ≈ y + 2πR/r0 0<r<∞
(The periodicity in y is necessary to avoid a singularity at r = 0.) One can use a
retarded de Sitter Green’s function to construct smearing functions in AdS3 . Alter-
natively, one can just translate our previous result (2.4) into Rindler coordinates,
to find6
Z
(4.3) φ(t, r, φ) = dt′ dy ′ KRindler(t + t′ , φ + iy ′ |t, r, φ) φ0 (t + t′ , φ + iy ′ )

(∆ − 1)2∆−2  σ ∆−2
KRindler = lim
πR3 r ′ →∞ r′

where σ is the invariant distance between (t, r, φ) and (t + t′ , r′ , φ + iy ′ ), and the

integral is over spacelike separated points on the complexified AdS boundary. A
BTZ black hole can be obtained from AdS3 by identifying φ ≈ φ+2π. This produces
4More precisely, we expect matrix elements of the commutators between low-energy states
to be small.
5We denoted this parameter r in our previous work.
+
6The boundary field φ is defined slightly differently in Rindler coordinates: φ(t, r, φ) ∼
0
φ0 (t, φ)/r ∆ as r → ∞.
 LOCAL BULK OPERATORS IN ADS/CFT AND THE FATE OF THE BTZ SINGULARITY 7

an orbifold singularity at r = 0. But making this identification doesn’t change the

smearing functions at all: if the boundary field has the necessary periodicity, so
will the bulk field. So we can use the same smearing functions (4.3) in a BTZ
background.
In the semiclassical limit this gives a clear picture of the BTZ horizon and
singularity. First, the horizon. The integral in (4.3) is over points on the complex-
ified boundary that are spacelike separated from the bulk point. As the bulk point
approaches the (future, past) horizon the integration region extends to (t = +∞,
t = −∞). Thus to probe the horizon requires an infinite time from the boundary
point of view. This fits nicely with the bulk definition of a horizon, as bounding
the region where light rays cannot escape to infinity.7
What about the BTZ singularity? With scalar fields as probes we cannot
directly study the bulk geometry. However it turns out that the orbifold singularity
generates a divergence in scalar correlators. To see this, we use the fact that in
the semiclassical limit we can make the bulk correlator periodic with an image sum
[LO94].
∞
X
(4.4) hφ(t, r, φ)φ(t′ , r′ , φ′ )iBTZ = hφ(t, r, φ)φ(t′ , r′ , φ′ + 2πn)iAdS
n=−∞
But r = 0 is a fixed point of the isometry φ → φ + const., so correlators in AdS are
φ-independent at r = 0. If we compute a bulk correlator in a BTZ background, the
image sum diverges when one of the bulk points is located at the singularity.
The same divergence arises from the boundary point of view. The CFT dual
to AdS3 in Rindler coordinates lives on R1,1 , while the CFT dual to BTZ lives on
R × S 1 . In the semiclassical limit the BTZ boundary correlator can be given the
necessary periodicity with an image sum.
∞
X
hφ0 (t, φ)φ0 (t′ , φ′ )iBTZ = hφ0 (t, φ)φ0 (t′ , φ′ + 2πn)iAdS
n=−∞
To recover a bulk correlator we convolve this with our smearing functions as in
(2.5). Again the image sum diverges when one of the bulk points is located at the
singularity. So we also get a divergent correlator at r = 0 from the boundary point
of view.
For future reference it’s useful to study the divergence in a little more detail.
Consider a point in AdS located near r = 0, at
(t, r, φ) with t = 0, r → 0, φ = 0
and a second point in AdS located near the boundary, at
(t′ , r′ , φ′ ) with r′ → ∞.
The invariant distance (3.2) between these points is approximately
 
r′ r ′ ′
σ≈ cosh φ̂ + sinh t̂ .
r0 r0
7One can also describe bulk points that are located inside the horizon [HKLL07]. However
this requires the use of smearing functions with support on both the left and right boundaries of
the extended Kruskal diagram. For a bulk point inside the future horizon the smearing function
extends to t = +∞ on the right boundary and t = −∞ on the left boundary, where time has the
same orientation on the left and right. From the boundary point of view this means we are using
operators that act on both copies of the thermofield-doubled CFT.
8 ALEX HAMILTON, DANIEL KABAT, GILAD LIFSCHYTZ, AND DAVID A. LOWE

The AdS bulk correlation function [IS95] decays exponentially at large φ′ .

1 1
hφ(t, r, φ)φ(t′ , r′ , φ′ )iAdS = √ √

4πR σ 2 − 1 σ + σ 2 − 1 ∆−1
(r0 /2r′ )∆
≈  ∆
r
2πR r0 cosh φ̂′ + sinh t̂′
′
(4.5) ∼ e−∆|φ̂ | for |φ̂′ | > φ̂max ∼ log(r0 /r)
This behavior means the BTZ image sum (4.4) is cut off at |n| ∼ 2πr R
0
log rr0 , which
in turn means the BTZ correlator diverges logarithmically as r → 0.8
 ∆
′ ′ ′ log(r0 /r) r0
(4.6) hφ(t, r, φ)φ(t , r , φ )iBTZ ∼ as r → 0
2π 2 r0 2r′ sinh t̂′
5. BTZ at finite N
What happens to the BTZ black hole at finite N ? Although definitive state-
ments are hard to come by, there are a few interesting observations to make. First,
there’s the issue of what smearing functions to use. In a pure AdS background
we gave a symmetry argument that the same smearing functions should be used
at any value of N . For excited states such as BTZ, where the symmetries are
broken, this argument is not valid.9 Nonetheless one might be tempted to use the
same semiclassical smearing functions for BTZ even at finite N . This is indeed a
reasonable prescription for bulk points outside the horizon. However as the bulk
point approaches the (future) horizon the semiclassical smearing functions extend to
t = +∞ on the boundary, and for points inside the horizon they grow exponentially
with time.10
K(t̂, φ̂|·) ∼ e(∆−d)t̂ as t → ∞
In the semiclassical limit this causes no problems. CFT correlators at infinite N
decay exponentially [BSS02], making the convolution of a boundary correlator
with a smearing function well-defined.
hφ0 (t̂, φ̂)φ0 (0, 0)iBT Z ∼ e−∆t̂
But when N is finite such behavior cannot persist indefinitely. At finite N , the CFT
on R × S 1 has a finite thermal entropy. Eventually the discrete spectrum of the
CFT becomes important and causes correlators to oscillate quasi-periodically rather
than decay exponentially [DLS02]. So we could not hope to use our semiclassical
smearing functions to reproduce sensible bulk correlators. It seems there are a
couple options.
(1) Perhaps this supports the fuzzball picture of Mathur and collaborators
[Mat05], in which the geometry inside the horizon differs radically from
what one would expect based on the traditional black hole metric. In this
8This corrects a normalization error in [HKLL07].
9From the boundary point of view BTZ is related to a CFT on a Euclidean 2-torus, and the
bulk-to-boundary map could depend on the modular parameter of the torus in a way that cannot
be determined from modular invariance.
10This is the behavior on the right boundary. For points inside the future horizon the
smearing function also extends to t = −∞ on the left boundary, and grows as e−(∆−d)t̂ in the far
past.
 LOCAL BULK OPERATORS IN ADS/CFT AND THE FATE OF THE BTZ SINGULARITY 9

case one would need to know the exact microstate of the black hole to
make sense of the interior of the horizon.
(2) Maybe there is some modification to the smearing functions which gives
sensible bulk correlators even for points inside the horizon. “Sensible”
means with small 1/N corrections to the semiclassical result, except near
r = 0, where the divergence (4.6) should be smoothed out. This would
support the picture that the semiclassical geometry is a good description,
even inside the horizon, but with quantum gravity effects resolving the
singularity.
(3) It could be that options 1 and 2 are compatible, if one is able to recover the
semiclassical BTZ metric from the fuzzball picture by a suitable averaging
procedure [AdBM06, BCL+ 07].
We conclude by presenting a prototype construction to show how option 3 might
be realized. The difficulty is that for points inside the horizon the semiclassical
smearing functions grow exponentially with time. A simple cure is to define a
modified smearing function K e which vanishes if t̂ is larger than some cutoff time
t̂max : 
e t̂, φ̂|·) = K(t̂, φ̂|·) if |t̂| < t̂max
K(
0 otherwise
(We impose the cutoff on both the left and right boundaries of the black hole.) We
can use this modified smearing function to define – purely within the CFT – a set
of “bulk operators.”
Z
e def e ′ , φ′ |t, r, φ) O(t′ , φ′ )
(5.1) φ(t, r, φ) = dt′ dφ′ K(t

With the cutoff in place, the operators φe have finite correlation functions.
Imposing a cutoff in this way might seem very arbitrary. But in fact there is a
good physical motivation for the cutoff which sets an upper bound on tmax . With
n(E) = (# CFT states with energy < E) = eS(E)
the density of states
dn
= βeS
dE
implies a spacing between energy levels
1
(5.2) ∆E = S .
βe
This spacing corresponds to a time (the Heisenberg time [Sre99])
tH = βeS .
By this time CFT correlators begin to oscillate quasi-periodically rather than decay
exponentially [BR03, BR04]. So imposing a cutoff on the smearing functions at
tmax . tH is the minimal change necessary to obtain well-defined bulk correlators.11
Moreover, such a cutoff has a nice physical interpretation. From the CFT point of
view measurements with a duration exceeding tH can resolve individual microstates
11In general t
max should be set by the time at which CFT correlators begin to behave quasi-
periodically. This could occur before tH , so really tH is an upper bound on tmax . The main point
is that at finite N the upper bound is finite. We are grateful to Hong Liu and Massimo Porrati
for discussions on this topic.
10 ALEX HAMILTON, DANIEL KABAT, GILAD LIFSCHYTZ, AND DAVID A. LOWE

of the CFT. So putting a cutoff at tmax . tH implies an average over microstates.

This means the smearing functions we constructed based on the classical BTZ
geometry break down at the horizon when N is finite, unless one averages over
microstates. This suggests that the region inside the horizon of the classical BTZ
geometry isn’t a good description of any individual microstate of the black hole.
Rather the region inside the horizon only provides a good description of ensemble
averages over black hole microstates.
With a cutoff at tmax . tH , K e represents the minimal modification to the
smearing functions necessary to plausibly represent BTZ correlators at finite N . In
the semiclassical limit we expect tmax → ∞, so the bulk operators we have defined
have the correct semiclassical limit.
Ideally at this point we would compute correlation functions of the operators
φe in the CFT at finite N . Such a calculation might be within reach [Wit07]. But
for the time being we will regard tmax as a fixed ad hoc cutoff and study correlators
of the operators φe in the large-N limit of the CFT. Up to small 1/N corrections,
this should be a good guide to behavior at finite N .
We first work in Rindler coordinates on AdS3 and consider the correlation
function
(5.3) e r, φ) φ(t
h φ(t, e ′ , r′ , φ′ ) iAdS
between a bulk operator located at (t = 0, r, φ = 0) and an operator near the
boundary at (t′ , r′ , φ′ ) with r′ → ∞. In the semiclassical limit the boundary corre-
lator
(r02 /2)∆
(5.4) hφ0 (t̂, φ̂)φ0 (0, 0)iAdS =  ∆
2πR cosh φ̂ − cosh t̂

decays exponentially at spacelike separation. So the bulk-boundary correlator (5.3)

will be exponentially small provided the boundary point is spacelike separated
from the support of the smearing function. As shown in Fig. 4, with the modified
smearing functions the cutoff at tmax means that – even for a bulk point inside the
Rindler horizon – the correlator will decay exponentially at large φ′ .
To extend this discussion to BTZ we use the image charge construction (4.4).
For points inside the horizon t̂max serves to cut off the AdS correlator at |φ̂′ | ≈ t̂max .
However as shown in (4.5) the AdS correlator is already exponentially small when
|φ̂′ | > φ̂max . Thus when we perform the image sum there are two possible regimes.
(1) For t̂max > φ̂max , or equivalently for r > r0 e−t̂max , the additional cutoff
at t̂max isn’t important. So using K e rather than K makes a negligible
change to the BTZ correlator away from r = 0. We can therefore probe a
large region inside the BTZ horizon, roughly the region
r0 e−t̂max < r < r0 ,
using only a finite time interval on the boundary, and without seeing
significant deviations from the semiclassical result. Strictly speaking this
means there is no horizon, at least not in the sense of section 4 where the
horizon corresponded to integration over infinite time.
(2) However for t̂max < φ̂max , or equivalently for
0 < r < r0 e−t̂max ,
 LOCAL BULK OPERATORS IN ADS/CFT AND THE FATE OF THE BTZ SINGULARITY 11

Figure 4. Penrose diagram of the (t, φ) plane. The support of

Ke is indicated by the jagged line. Points in the shaded region are
spacelike separated from the support of K. e When the smearing
function extends from −t̂max to +t̂max the shaded region is char-
acterized by |φ̂| > t̂max + |t̂|.

the additional cutoff at t̂max is crucial. It serves to regulate the image

R
sum, cutting it off at |n| ≈ 2πr 0
t̂max so that
 ∆
e ′ , r′ , φ′ ) iBTZ ≈ t̂max
e r, φ) φ(t
h φ(t,
r0
.
2π 2 r0 2r′ sinh t̂′
Note that the correlator is independent of r, and the divergence at r = 0
has been eliminated!
We find it appealing that the same effect that eliminates the horizon also gets rid
of the divergence. Note that the effects we have discussed are very robust: they
are independent of any details of the CFT and only rely on the generic thermal
behavior (5.2). Our results are compatible with the detailed study of extremal BTZ
black holes in [B+ 07].
To summarize: we’ve defined a set of operators in the CFT (5.1) which should
have well-defined correlation functions even at finite N . As a guide to the behavior
of these operators we studied their correlation functions in the large N limit. For
bulk points well outside the horizon the cutoff at tmax has no effect on the smearing
functions. For bulk points inside the horizon but well away from the singularity the
cutoff at tmax makes an exponentially small change in correlators. But for points
very near r = 0 the cutoff at tmax becomes important and renders correlation func-
tions finite. It seems reasonable that working with the true finite-N correlation
functions of the CFT, rather than their semiclassical large-N limit, should only
make a small change in these results. If so, this would support option 3: after a
12 ALEX HAMILTON, DANIEL KABAT, GILAD LIFSCHYTZ, AND DAVID A. LOWE

suitable average over microstates, enforced by a cutoff at tmax . tH , the semiclas-

sical BTZ geometry becomes a good description, even inside the horizon, but with
quantum gravity effects resolving the singularity.

Acknowledgements
DK is grateful to Mark Sowers and the organizers of the 2007 Sowers workshop
for a delightful and stimulating conference. GL would like to thank Vijay Bala-
subramanian and especially Masaki Shigemori for discussions and the organizers of
pre-strings 2007 for a stimulating conference.

References
[AdBM06] Luis F. Alday, Jan de Boer, and Ilies Messamah, The gravitational description of
coarse grained microstates, JHEP 12 (2006), 063.
[B+ 07] Vijay Balasubramanian et al., Typicality versus thermality: An analytic distinction,
eprint (2007), hep–th/0701122.
[BCL+ 07] Vijay Balasubramanian, Bartlomiej Czech, Klaus Larjo, Donald Marolf, and Joan
Simon, Quantum geometry and gravitational entropy, eprint (2007), arXiv:0705.4431
[hep–th].
[BD82] N. D. Birrell and P. C. W. Davies, Quantum fields in curved space, Cambridge Univ.
Press, 1982.
[BDHM98] Tom Banks, Michael R. Douglas, Gary T. Horowitz, and Emil J. Martinec, Ads
dynamics from conformal field theory, eprint (1998), arXiv:hep–th/9808016.
[Ben00] Iosif Bena, On the construction of local fields in the bulk of ads(5) and other spaces,
Phys. Rev. D62 (2000), 066007.
[BGL99] Vijay Balasubramanian, Steven B. Giddings, and Albion E. Lawrence, What do cfts
tell us about anti-de sitter spacetimes?, JHEP 03 (1999), 001.
[BR03] J. L. F. Barbon and E. Rabinovici, Very long time scales and black hole thermal
equilibrium, JHEP 11 (2003), 047.
[BR04] , Long time scales and eternal black holes, Fortsch. Phys. 52 (2004), 642–649.
[BSS02] Danny Birmingham, Ivo Sachs, and Sergey N. Solodukhin, Conformal field theory
interpretation of black hole quasi- normal modes, Phys. Rev. Lett. 88 (2002), 151301.
[DLS02] Lisa Dyson, James Lindesay, and Leonard Susskind, Is there really a de sitter/cft
duality, JHEP 08 (2002), 045.
[HKLL06a] Alex Hamilton, Daniel Kabat, Gilad Lifschytz, and David A. Lowe, Holographic rep-
resentation of local bulk operators, Phys. Rev. D74 (2006), 066009.
[HKLL06b] , Local bulk operators in ads/cft: A boundary view of horizons and locality,
Phys. Rev. D73 (2006), 086003.
[HKLL07] , Local bulk operators in ads/cft: A holographic description of the black hole
interior, Phys. Rev. D75 (2007), 106001.
[IS95] Ikuo Ichinose and Yuji Satoh, Entropies of scalar fields on three-dimensional black
holes, Nucl. Phys. B447 (1995), 340–372.
[LO94] Gilad Lifschytz and Miguel Ortiz, Scalar field quantization on the (2+1)-dimensional
black hole background, Phys. Rev. D49 (1994), 1929–1943.
[Mal98] Juan M. Maldacena, The large n limit of superconformal field theories and supergrav-
ity, Adv. Theor. Math. Phys. 2 (1998), 231–252.
[Mat05] Samir D. Mathur, The fuzzball proposal for black holes: An elementary review,
Fortsch. Phys. 53 (2005), 793–827.
[Sre99] Mark Srednicki, The approach to thermal equilibrium in quantized chaotic systems,
J. Phys. A32 (1999), 1163.
[SW98] Leonard Susskind and Edward Witten, The holographic bound in anti-de sitter space,
eprint (1998), arXiv:hep–th/9805114.
[VK91] N.Ja. Vilenkin and A.U. Klimyk, Representation of lie groups and special functions,
Kluwer Academic Publishers, 1991.
[Wit07] Edward Witten, Three-dimensional gravity revisited, eprint (2007), arXiv:0706.3359
[hep–th].
 LOCAL BULK OPERATORS IN ADS/CFT AND THE FATE OF THE BTZ SINGULARITY 13

Department of Physics, Columbia University, New York NY 10027 USA

Current address: Department of Mathematics and Applied Mathematics, University of Cape
Town, Rondebosch 7700, South Africa
E-mail address: [email protected]

Department of Physics, Columbia University, New York NY 10027 USA

E-mail address: [email protected]

Department of Mathematics and Physics and CCMSC, University of Haifa at Oranim,

Tivon 36006 ISRAEL
E-mail address: [email protected]

Department of Physics, Brown University, Providence RI 02912 USA

E-mail address: [email protected]