A simple quantum test for smooth horizons

Kyriakos Papadodimas,a Suvrat Rajub and Pushkal Shrivastavab

a
International Centre for Theoretical Physics, Strada Costiera 11, 34151 Trieste, Italy
b
International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Shivakote,
Bengaluru 560089, India.
arXiv:1910.02992v1 [hep-th] 7 Oct 2019

E-mail: [email protected], [email protected],

[email protected]

Abstract: We develop a new test that provides a necessary condition for a quantum state to
be smooth in the vicinity of a null surface: “near-horizon modes” that can be defined locally
near any patch of the null surface must be correctly entangled with each other and with
their counterparts across the surface. This test is considerably simpler to implement than a
full computation of the renormalized stress-energy tensor. We apply this test to Reissner-
Nordström black holes in asymptotically anti-de Sitter space and provide numerical evidence
that the inner horizon of such black holes is singular in the Hartle-Hawking state. We then
consider BTZ black holes, where we show that our criterion for smoothness is satisfied as
one approaches the inner horizon from outside. This results from a remarkable conspiracy
between the properties of mode-functions outside the outer horizon and between the inner
and outer horizon. Moreover, we consider the extension of spacetime across the inner horizon
of BTZ black holes and show that it is possible to define modes behind the inner horizon
that are correctly entangled with modes in front of the inner horizon. Although this provides
additional suggestions for the failure of strong cosmic censorship, we lay out several puzzles
that must be resolved before concluding that the inner horizon will be traversable.
Contents

1 Introduction 1

2 Entangled modes across a null surface 5

3 The Reissner-Nordström black hole in AdS 13

4 The BTZ black hole 23

5 Conclusions and discussion 33

A Scattering in the Reissner-Nordström geometry at large angular momenta 36

1 Introduction

In addition to an outer event horizon—which characterizes a black hole—charged and rotat-

ing black holes may have additional horizons in their interior. These inner horizons are of
theoretical interest, because they can also be Cauchy horizons. Cauchy horizons present a
problem for Laplacian determinism, because given smooth initial data in the past there is no
unique way to evolve it past such a horizon even though the metric is not singular there.
This difficulty is sometimes bypassed by appealing to “strong cosmic censorship”. This
is the idea that although the Cauchy horizon exists for the static solution, the moment one
considers generic initial data leading to the formation of a realistic black hole, the inner
horizon will collapse to form a singularity and mark the end of spacetime. The simplest
physical intuition for this conjecture is as follows. In the static solution, the Cauchy horizon
is to the future of all events outside the black hole. On the other hand, an infalling observer
can reach the Cauchy horizon in a finite amount of proper time after crossing the outer
horizon. Therefore, if one perturbs the exterior of the black-hole solution, then in the finite
amount of time that it takes the observer to reach the inner horizon, the observer will receive
radiation from all the events outside the black hole, which stretch out for an infinite amount
of time [1]. The strong cosmic censorship conjecture is based on the intuition that this infinite
“blue shift” effect will destabilize the inner horizon [2] .
However, it is not straightforward to prove such a result [3]. Somewhat surprisingly,
in several cases, it is possible to extend the metric continuously across the horizon [4] and
only the derivatives of the metric diverge. The classical analysis has therefore focused on the
severity of this divergence and the question of whether solutions to the equations of motion
can be continued weakly past the horizon. The answer to this question depends on the precise

–1–
smoothness conditions that are placed on the initial data [5]. One key physical issue in this
analysis is a competition of the decay of perturbations outside the horizon, controlled by the
quasinormal frequencies of the black hole, and the blue shift effect near the inner horizon
controlled by its surface gravity [6]. This has been the subject of extensive recent discussions
[7, 8] and we refer the reader to the introduction of [9] for a clear and concise discussion of
the issues involved in the classical problem.
However, the world is not classical. So what happens when quantum effects are taken
into account? This issue has received far less attention than the classical question. Moreover,
the literature so far has focused on the behaviour of the renormalized stress-energy tensor
[10, 11, 12, 13, 14] near the inner horizon. But this calculation is not only tedious, it also
suffers from inherent ambiguities [15].
In this paper, we will propose a simple quantum-mechanical criterion that must be sat-
isfied for the horizon to be smooth. Our test is much simpler to implement than a full com-
putation of the stress-energy tensor, although we emphasize that it provides only a necessary
and not a sufficient condition for the horizon to be smooth.
The test is as follows. Consider a scalar field, φ propagating in a background metric in
d + 1 dimensions. To investigate the smoothness of a null surface we consider two spacetime
points 1 and 2 in its vicinity; the points may be on the same side of the surface or on opposite
sides. Then we demand that it should be possible to define a state of the quantum fields so
that the short distance behaviour of the two-point function is given by
1
hφ(1)φ(2)i = N d−1 + ..., (1.1)
s 2

where s is the square of the geodesic distance between the two points and . . . denotes terms
that are less-singular than the displayed term. (The normalization N is chosen for convenience
and described in section 2.)
At first sight, the condition (1.1) may appear trivial. After all, this is the term that
is thrown away in computations of the stress-tensor. However, on further examination, the
simple condition (1.1) turns out to yield a surprising number of riches. We first describe the
physical intuition that suggests that condition (1.1) should be important, and then describe
its mathematical consequences.
The physical intuition behind our test is that a smooth state in any spacetime is char-
acterized by the existence of entangled degrees of freedom across any null surface. This is
what allows spacetime to form a “universal quantum channel”; the large degree of entangle-
ment makes it possible to send arbitrary kinds of communications between one region and
a nearby region. In fact, in any quantum field theory, a finite-energy state has so much en-
tanglement between the various local degrees of freedom that by manipulating the degrees
of freedom in any region of spacetime, it is possible to create arbitrary states in the Hilbert
space [16]. The leading short-distance behaviour of the two-point function displayed in (1.1)
is one quantification of the large amount of entanglement between neighbouring regions.

–2–
 To distil this entanglement, we define appropriate near-horizon modes by integrating
the field in local Rindler coordinates on both sides of the surface. Since the Rindler modes
oscillate an infinite number of times near the surface, we can extract a sharp mode even by
restricting ourselves to a very small region near the null surface. These near-horizon modes
depend on a frequency, ω0 and we show that if a state of the quantum fields, |Ψi satisfies
(1.1) then the action of these modes must be related through

(a − e−πω0 e
a† )|Ψi = 0, (1.2)

and their two-point functions must be given by

1
hΨ|aa† |Ψi = hΨ|ea† |Ψi =
ae . (1.3)
1 − e−2πω0
The reader will recognize that the Minkowski vacuum is annihilated by a similar com-
bination of left and right Rindler modes [17]. The result above is slightly different since it
uses only local physics; relatedly, the modes that appear are not defined by integrating the
field over the half-space, but just in the neighbourhood of a null surface. In particular, the
relations (1.2) and (1.3) may fail to hold for globally defined Rindler modes in an excited
state even if it has finite Minkowski energy. On the other hand, for the near-horizon modes,
(1.2) and (1.3) must hold in any smooth state.
The significance of (1.2) and (1.3) can be brought out by considering a charged spherically
symmetric black hole. In such a situation, the near-horizon modes reduce to globally defined
Schwarzschild modes smeared with a function that is sharply peaked in frequency space. So
the two-point function of the near-horizon modes picks up only the part of the correlators of
the Schwarzschild modes that is proportional to a delta function in the frequency-difference.
In this simple setting, our result therefore implies the following: for the horizon to be smooth
in a given quantum state, the two-point function of Schwarzschild modes must contain a piece
proportional to a delta function in the frequency-difference with a coefficient that is precisely
the Boltzmann factor corresponding to the temperature of the horizon.
This criterion is powerful because it is not easy for a state to be smooth at both the outer
and the inner horizons. For a state to be smooth at the outer horizon, the fields there must
be entangled at the temperature corresponding to the outer horizon; then the propagation
of fields between the inner and outer horizon must precisely “heat” (or “cool”) them to
the temperature of the inner horizon. This leads to an infinite sequence of constraints on the
reflection and transmission coefficients that control the propagation of fields from the outer to
the inner horizon. Indeed, as we show in section 3, a simple computation of these coefficients
can be used to easily demonstrate that charged black holes in anti-de Sitter space have singular
horizons without going through a full computation of the renormalized stress-tensor.
We can also apply our test to the rotating BTZ black hole. Here, as one might expect from
the analysis of [12], we find that our test is satisfied. This follows from a conspiracy between
the properties of the reflection and transmission coefficients in the black-hole interior and the
behaviour of normalizable modes outside the outer horizon. Since our test only provides a

–3–
necessary condition for smoothness, it does not guarantee that the renormalized stress-energy
tensor will be finite. Indeed, our test is satisfied even for those parameters of the BTZ black
hole where the analysis of [12] suggests that the stress-tensor will diverge.
On the other hand, our test, in some respects, is also stronger than a check of the stress-
tensor since it involves modes on both sides of the horizon. In the case of the BTZ black hole,
we are therefore forced to explore whether field operators can be extended smoothly past the
inner horizon.
This extension turns out to be nontrivial since the standard construction of mirror oper-
ators [18, 19] that applies to the outer horizon cannot be directly applied to the inner horizon
of the BTZ black hole. This can be understood in terms of the monogamy of entanglement.
In the free-field limit, the modes near the inner horizon can be written as linear-combinations
of modes near the outer horizon. Since the modes near the outer horizon are entangled with
modes outside the outer horizon, they cannot also be entangled with fresh-modes behind the
inner horizon. We demonstrate a precise version of this result in the text.1
Surprisingly, it turns out to be possible to reuse the modes between the inner and outer
horizon in such a manner that not only are the constraints of locality respected, the modes
also have the correct two-point function dictated by the temperature of the inner horizon.
This also provides the unique extension of quantum fields in the near-horizon region just
beyond the inner horizon if the inner horizon is traversable.
At first sight, our results provide additional evidence for the claim that strong cosmic
censorship fails for the BTZ black hole. If so, this would lead to a very puzzling situation from
the point of view of the AdS/CFT correspondence [21]. If the inner horizon is traversable,
an infalling observer in an eternal BTZ black hole can simply exit the system as described by
the two CFTs. This would imply that the eternal BTZ black hole is not completely described
by two entangled CFTs [22] and this system must be supplemented by additional degrees of
freedom.
However, as we explain in section 5, the question of traversability of the inner horizon is
more delicate. For instance, if one perturbs the exterior of the black hole, although the per-
turbation initially dies off at the quasinormal frequency, unitarity requires a nonperturbative
S
tail (of size e− 2 ) to survive at late times [22]. What is the effect of this tail on the inner
horizon? Relatedly, the boundary CFT undergoes spontaneous fluctuations over exponential
long time-scales. These fluctuations, which appear at arbitrarily late times, can also destroy
the inner horizon. In this paper, we do not present a solution to these questions, but leave
them open for further work.
The plan of the paper is as follows. In section 2 we describe our test in greater detail,
define the near-horizon modes carefully and derive (1.2) and (1.3) starting with (1.1). We
then apply this test to charged asymptotically AdS black holes in various dimension in section
3 and to the BTZ black hole in section 4. We conclude in section 5. The Appendix presents
1
We emphasize that this argument is very different, in flavour, from the monogamy arguments that were
used in [20] since it can be implemented entirely at the level of effective field theory.

–4–
a WKB analysis of the propagation of large angular-momentum waves in AdS black holes,
which is of interest since in this regime (1.2) and (1.3) are automatically satisfied at the inner
horizon if they are satisfied at the outer horizon.

2 Entangled modes across a null surface

In quantum field theory, to test whether the spacetime in the vicinity of a surface is smooth,
we can test whether it is possible to transmit “messages” in that region by turning on a
source for a field at one point and measuring the response of the field at another. While the
long-distance propagation of the source will depend on the nature of the spacetime, in the
short-distance limit, we expect that fields in the neighbourhood of the source will respond in
a universal manner. The response of the one-point function of the field to a source depends
only on the commutator, but the response of higher-point functions also requires a specific
short-distance behaviour for the two-point function. This, in turn, requires degrees of freedom
that constitute the quantum field to be entangled with each other in a specific manner, as we
now explain.
In the neighbourhood of any null surface, the quantum fields can be expanded in a set of
modes corresponding to left and right movers. We will choose the convention that the “left
movers” are those that smoothly cross the surface, so that the phase factor multiplying these
modes varies smoothly as we move along the surface. On the other hand, the modes that
move parallel to the null surface (i.e. whose surfaces of constant phase are parallel to the null
surface) will be called “right movers”. See Figure 1.

Figure 1: Near horizon right-moving modes of a free scalar field near a null surface (black
line). The red modes correspond to the operator a while the blue ones to e
a (see eq. (2.12)).
The arrows indicate that the modes propagate parallel to the surface.

The specific technical result that we intend to show is that it is possible to define appro-
priate “right moving” modes on the two sides of the null surface, so that their action on the
state of the system is related in a specific manner.2
2
We use the term “entanglement” to denote this relationship although our result is primarily about cor-

–5–
 We consider a small portion of a null surface, such that the metric is continuous across the
surface. In (d + 1)-dimensions, we introduce coordinates U, V and transverse coordinates, ξ a
with 2 ≤ a ≤ d. We use a calligraphic font for these coordinates in this section to distinguish
them from the coordinates that appear in the black-hole geometry. The Greek indices below
can take values in a larger range, 0 ≤ µ, ν ≤ d, and we set ξ 0 = U and ξ 1 = V. The null
surface under consideration is that of U = 0, and on this surface, we will consider the small
patch near ξ µ = 0, where we write the metric as

ds2 = −dUdV + δαβ dξ α dξ β + gµν

(1) µ ν
dξ dξ . (2.1)
(1)
We will assume that gµν → 0 as ξ µ → 0 i.e. as we approach the patch under consideration
the metric is well approximated by the first two terms in (2.1).

Near-horizon 2-point function

Now consider a scalar field propagating in this background. The precise assumption that
we will make is that if the spacetime is “reasonable” then, in the short-distance limit the
two-point function of the scalar field will be given simply by an inverse power of the geodesic
distance. This is true at least if the ultraviolet physics of the scalar operator is controlled by
a free-fixed point (at scales much lower than the Planck scale). A similar assumption can be
made for fermions, or fields with spin.
Consider two insertions of the field at two nearby, but distinct points, which may be on
the same side or on opposite sides of the null surface. We denote the state of the full system
by |Ψi. In AdS/CFT, this state may be considered to be a state in the Hilbert space, as
defined by the boundary conformal field theory. The assumption above implies that if this
state is “reasonable” then
1
hΨ|φ(U1 , V1 , ξ1a )φ(U2 , V2 , ξ2a )|Ψi = N d−1 + R. (2.2)
s 2
In the equation above, the leading-order square of the geodesic distance s is given by

s = (−δU δV + δαβ δξ a δξ b ), (2.3)

where δU = U1 − U2 − i; δV = V1 − V2 − i and we use the (d − 1)-vector, δ ξ, ~ with components

~a a a
δ ξ = ξ1 − ξ2 to denote the displacement in the transverse directions between the two points.
Here, R can be any function of the variables (Ui , Vi , ξia ) that is less-singular than the displayed
leading term as the points come close to each other. We have also introduced −i regulators
with δU and δV, which reflect that fact that we are interested in a Wightman function. This
will be important when we consider commutators below. The normalization of the short-
distance singularity
Γ(d − 1)
N = d , (2.4)
2d π 2 Γ( d2 )
relation functions, and not about any quantum-information measure of entanglement. The reason for our
terminology is that — as shown in [23] — it is possible to use these correlators between near-horizon modes
to extract Bell pairs from opposite sides of the horizon.

–6–
can be fixed by considering a canonically normalized field in flat-space and explicitly checking
its short-distance behaviour.
The important equation (2.2) makes the assumptions that the short-distance singularities
of the two-point function arise only when the geodesic distance vanishes. Note that the re-
(1)
maining part of the metric gµν does not appear in the leading singular part of this expression,
although it is important for the subleading terms captured in the function R.
We will show that (2.2) forces a specific form of entanglement between right movers across
the surface U = 0. To see this, we take a further limit of (2.2) and repeat the manipulations
that led to equation (4.3) of [24]. If we first differentiate the two-point function, we will find
that
d2 − 1 (δV)2
hΨ|∂U1 φ(U1 , V1 , ξ1a )∂U2 φ(U2 , V2 , ξ2a )|Ψi = − N d+3 + ∂U1 ∂U2 R. (2.5)
4 s 2
We now consider the limit δV → 0. In this limit the derivative of the two-point function
above goes to zero unless δ ξ~ = 0. But this point, where the transverse displacement vanishes,
gives a delta function contribution. We can check this, and also determine the normalization
of the delta function by performing an integral over the transverse coordinates
− d+3
(δV)2
Z Z 
1 1 1 2
lim d d−1
δ ξ~ d+3 = lim 2 d−1
d−1
d ~
δξ 1 + δ ξ~2
δV→0 s 2 δV→0 (δU) (−δUδV) 2 (−δUδV)
(2.6)
1
= κN ,
(δU)2
where Z d−1
1 4 π 2
κN = dd−1
δ ξ~ d+3 = . (2.7)
(1 + δ ξ~2 ) 2
2
d − 1 Γ( d−1
2 )
It is important that, in the integral above, we take the range of the integral over the transverse
δ ξ~
coordinates to be (−∞, ∞) in order to be able to change variables from δ ξ~ → √−δU δV
. This
is only a trick to determine the normalization. It does not require that the expression for the
two-point function remain valid for an infinite separation in transverse coordinates or even
that the transverse coordinates have an infinite extent.
The various normalization constants in the equations above simplify when taken together
using
d2 − 1 1
κN N = . (2.8)
4 4π
Note that if we were to perform the operations above on a term in the two-point function
that diverges with a smaller power of geodesic distance or on a regular function, which may
appear inside R, we would obtain a result that is non-singular in the limit U1 → U2 .
So we finally find
1 1 ~
lim hΨ|∂U1 φ(U1 , V1 , ξ1a )∂U2 φ(U2 , V2 , ξ2a )|Ψi = − δ d−1 (δ ξ)
V1 −V2 →0 4π (U1 − U2 − i)2 (2.9)
+ lim ∂U1 ∂U2 R̃.
V1 −V2 →0

–7–
If the reader is worried about the ultralocal delta function that appears above, she should
note that in the applications below we will always use (2.9) after integrating both sides with
test functions. Typically, we will consider fields that are separated by V in the V coordinate,
and smeared over a range U in the U coordinate and ξ in the transverse coordinates. Then
the formula above holds provided we take V  U  ξ .

Near-horizon modes
We now define some modes by integrating the field over a very short distance on both sides
of the U = 0 surface and for a very short distance over the transverse coordinates. We will
show that the short distance structure of the two-point function above forces these modes to
have a specific two-point correlation function. We follow the procedure given in [23] to define
the modes.
The intuition is the following: as we approach the null surface from below U → 0− we
can find solutions of the wave equation that behave like (−U)iω0 . These solutions undergo an
infinite number of oscillations until U = 0. If we think in terms of the tortoise-like coordinate
u = log(−U) these solutions behave like plane waves. This approximation becomes better as
we approach the null surface at U → 0 or u = −∞. We define modes a corresponding to
wave packets with highly peaked frequency ω0 and centered very far towards u = −∞, or
equivalently around a point U0 very near the null surface U = 0. We make sure that these
modes are unit-normalized (as opposed to having delta-function normalization). We define
these modes as Fourier modes with approximate frequency ω0 characterized by a window of
support in position space centered around U0 and with a very large width in u-coordinates
(though a small region in U coordinates).
This is achieved by introducing a “tuning function” T (U) which has the property that
T (U) is real and has support only for U ∈ [Ul , Uh ], where 0 < Ul  U0  Uh and so that Uh
is much smaller than the characteristic curvature scale of the geometry. This tuning function
is assumed to vanish smoothly at the end-points of its support and normalized carefully as
follows. First, we define its Fourier transform via
∞  iν ∞  −iν
U U
Z Z
1 dU
T (U) = s(ν) dν; s(ν) = T (U) . (2.10)
−∞ U0 2π 0 U U0

Then we normalize the “tuning function” by demanding

Z
|s(ν)|2 dν = 1. (2.11)

We choose s(ν) to be sharply peaked around ν = 0, which corresponds to T (U) being almost
constant in the domain [Ul , Uh ].
We now also integrate over a small volume Vol in the transverse coordinates and define

–8–
the modes

dd−1 ξ a −∞
−U −iω0
Z Z  
1 a
a= √ √ dU∂U φ(U, V = 0, ξ ) T (−U),
πω0 0 Vol U0
Z d−1 a Z ∞  iω0 (2.12)
1 d ξ a U
a= √
e √ ∂U φ(U, V = −, ξ ) T (U).
πω0 Vol 0 U0

The domain of integration, which is controlled by T (U) is a very small region on both sides
of the null surface and also a very small patch in the transverse coordinates. The modes have
a nontrivial dependence on the choice of ω0 and, moreover, such modes can be defined in the
vicinity of any point of the null surface. But we have suppressed this dependence on the left
hand sides of (2.12) to lighten the notation.
In the rest of this section we will show how the short-distance behaviour of the two-point
function (2.9) determines the two-point correlators of these modes.
Before moving on, we emphasize that the modes a, e a depend on the choice of U0 , Ul , Uh
and also the precise choice of the tuning function. We are interested in the behavior of these
modes in the limit U0 → 0 and UUh0 , UU0l  1 and where s(ν) is sharply peaked. The statements
we are going to make about the modes a, e a correspond to the behavior of these modes in
this particular limit. However, in the equations below, we will not display the dependence on
these various cutoffs explicitly, since the equations can be made arbitrarily precise by taking
these cutoffs to be as small as necessary.

Commutators
If we assume that the field operators satisfy canonical commutation relations then we find
that at equal values of V we have
i ~
[φ(U1 , V, ξ1a ), ∂U2 φ(U2 , V, ξ2a )] = δ d−1 (δ ξ)δ(U 1 − U2 ). (2.13)
2
This commutators has, inbuilt into it, the fact that the fields commute at spacelike separation.
We note that the commutator is also consistent with the two-point correlator that we found
above because
 
1 1 1
Im (h∂U1 φ(U1 , V, ξ2a ), ∂U2 φ(U2 , V, ξ1a )i) = Im − = δ 0 (U1 − U2 )δ d−1 (δ ξ),
~
4π (U1 − U2 − i)2 4
(2.14)
1 ∂ 1 ∂
which can be seen from the identity − (x−i)2 = ∂x x−i = ∂x P x + iπδ (x). 1 0

We can substitute these commutation relations into the definition of the modes to com-
pute their commutators.
Z
† 1
[a, a ] = [∂U1 φ(U1 , V = 0, ξ1a ), ∂U2 φ(U2 , V = 0, ξ2a )]
πω0 Vol
(2.15)
U1 −iω0 U2 iω0
   
d−1 a d−1 a
× − − T (−U1 )T (−U2 )dU1 dU2 d ξ1 d ξ2 .
U0 U0

–9–
Using the canonical commutators (2.13), and doing the trivial integral in the transverse
directions, we find that

U1 −iω0 U2 iω0
Z    
† i
[a, a ] = ∂U1 δ(U1 − U2 ) − − T (−U1 )T (−U2 )dU1 dU2
2πω0 U0 U0
Z −∞ Z −∞ (2.16)
i 1 2 dU1
= ∂U1 T (−U1 )T (−U1 )dU1 + T (−U1 ) .
2πω0 0 2π 0 U1
In doing the integral over the delta function in the U coordinates, we were careful that the
U2 integral proceeds with dU2 < 0, and so the delta function sets U1 = U2 and gives an
additional minus sign. The first term above vanishes if the tuning function vanishes at both
its end points. For the second term, using the normalization of the tuning function above we
find
[a, a† ] = 1. (2.17)
A similar calculation for e
a leads to
[e a† ] = 1,
a, e (2.18)
and also
[a, e a† ] = 0.
a] = [a, e (2.19)

Cross-correlators
For the cross-correlators we have
Z
1
hΨ|aea|Ψi = dU1 dU2 ∂U1 ∂U2 hΨ|φ(U1 , V = 0, ξ1a )φ(U2 , V = 0, ξ2a )|Ψi
πVol ω0 (2.20)
× (−U1 )−iω0 (U2 )iω0 T (−U1 )T (U2 )dd−1 ξ1a dd−1 ξ2a .
In the small domain where the integrand above has support, note that the regular parts of
the two-point function just drop out and the only contribution comes from the singular terms
identified above.
Now, note that we may write

U2 −iω
Z ∞
e−πω
 
1 1
= ω − dω, (2.21)
(U1 − U2 )2 (−U1 )U2 −∞ 1 − e−2πω U1
where U1 < 0 and U2 > 0. If |U1 | > |U2 | this identity follows from completing the ω integral
in the upper half plane and picking up the poles at ω = in; else we close the contour in the
lower half-plane and pick up the poles at ω = −in.
Substituting the short distance limit (2.9) into the expression above and doing the trans-
verse integral we find that

e−πω U2 −iω
Z  
1 dU1 dU2
hΨ|aea|Ψi = 2 ω − dωT (−U1 )T (U2 )(−U1 )−iω0 (U2 )iω0
4π ω0 U1 U2 1 − e−2πω U1
e−πω
Z
1
= ω |s(ω − ω0 )|2 dω.
ω0 1 − e−2πω
(2.22)

– 10 –
In the limit where s(ω) is very sharply peaked around ω = 0, and using the normalization
condition (2.11), the two-point function just reduces to
e−πω0
hΨ|ae
a|Ψi = . (2.23)
1 − e−2πω0
This gives us the universal form of entanglement for short distance modes on opposite sides of
the horizon. As in [23], this two-point function can be used to extract Bell pairs from either
side of the null surface.
Taking the Hermitian conjugate of this relation leads to
e−πω0
hΨ|a†e
a† |Ψi = . (2.24)
1 − e−2πω0
Finally, by a similar calculation we find
hΨ|a†e a† |Ψi = 0.
a|Ψi = hΨ|ae (2.25)
Self-correlators
We can also determine correlators of the a and e
a modes with their own conjugates. In this
case, we need the identity
U2 −i −iω
Z ∞  
−1 1 1
= ω e dω, (2.26)
(U1 − U2 − i)2 U1 U2 −∞ 1 − e−2πω U1
when U1 < 0 and U2 < 0. The integral on the right can again be done by closing the ω
contour in the upper or the lower half plane and picking up the poles at ω = in. Note that
the i term ensures that the integral converges both at large positive and large negative ω.
Using this identity, we see that
 −iω
U1 −iω0 U2 iω0
   
U2
Z
† dU1 dU2 1
hΨ|aa |Ψi = dωT (−U1 )T (−U2 ) − −
U1 U2 1 − e−2πω U1 U0 U0
Z
1
= |s(ω − ω0 )|2 dω.
1 − e−2πω
(2.27)
Again, in the limit where s(ω) is sharply peaked we find that
1
hΨ|aa† |Ψi = 1 + hΨ|a† a|Ψi = . (2.28)
1 − e−2πω0
A small subtlety that we note is that it is the i prescription that is important for ensuring
that haa† i and ha† ai have distinct values. Indeed, the reader might have wondered whether
1 e−2πω
it is possible to replace 1−e−2πω → 1−e −2πω in the identity (2.26) since both these terms have

the same residues at the pole ω = ±in. However, with an additional factor of e−2πω in the
numerator, the integral will not converge for large negative ω since the numerator and the
denominator now grow at the same rate but the factor of e−ω grows in that regime. It is
the i prescription that forces us to use the measure displayed in the equation (2.26) rather
than the one with an additional exponential factor. This, in turn, ensures that there is no
exponential factor in the numerator on the right hand side of (2.28).

– 11 –
 Relating the action of modes on the state
Above, we derived the two-point functions between the various modes. However, by putting
these results together, we can derive a stronger relationship that relates the action of the left-
movers on the quantum state to the action of the right movers.
Let us write
a|Ψi = c1 a|Ψi + c2 a† |Ψi + |χi,
e (2.29)
where c1 , c2 are constants to be determined and |χi is orthogonal to the vectors produced by
the action of a and a† on |Ψi so that

hχ|a|Ψi = hχ|a† |Ψi = 0. (2.30)

Now, from (2.25) we have

hΨ|a†e
a|Ψi = 0 ⇒ c1 = 0. (2.31)
We can now set c2 using

e−πω0
hΨ|ae
a|Ψi = ⇒ c2 = e−πω0 . (2.32)
1 − e−2πω0
Finally, note that
e−2πω0
a†e
hΨ|e a|Ψi = ⇒ hχ|χi = 0. (2.33)
1 − e−2πω0
where we have also used the two-point correlator of a with its conjugate. As similar procedure
a† . Therefore, we reach the simple relations
can be followed for the action of e

a|Ψi = e−πω0 a† |Ψi;

e a† |Ψi = eπω0 a|Ψi.
e (2.34)

Null surfaces with spherical or translational symmetry

The analysis above can be simplified slightly if the null surface under consideration has
spherical or translational symmetry. Since the case of spherical symmetry is the case that
will be important for us later, we indicate the generalization explicitly.
Consider a spherically symmetric null surface so that the metric is given by

ds2 = −dUdV + r02 dΩ2d−1 + gµν

(1) µ ν
dξ dξ . (2.35)
(1)
We now impose that gµν → 0 provided that U, V → 0, and, as above, Greek indices, µ, ν run
over all coordinates.
We can now define modes by integrating all over the (d − 1)-sphere and extracting a
particular spherical harmonic.
d−1
−U −iω0

r 2
Z
a = √0 ∂U φ(U, V = 0, Ω) T (−U)dUY`∗ (Ω)dd−1 Ω,
πω0 U0
d−1 Z
(2.36)
 iω0
r0 2 U
a= √
e ∂U φ(U, V = −, Ω) T (U)dUY` (Ω)dd−1 Ω,
πω0 U0

– 12 –
Now, rather than being defined in the neighbourhood of a point, the modes carry an angular
momentum, denoted by `. These modes depend both on ` and on ω0 but we have suppressed
that dependence to lighten the notation.
A precise repetition of the analysis above, leads to the same result for the action of these
modes on the state
a|Ψi = e−πω0 a† |Ψi; e
e a† |Ψi = eπω0 a|Ψi. (2.37)
Note that the orthogonality of the spherical harmonics implies that the modes are entangled
only if the same value of ` is used in both lines of (2.36).

A note of caution
The relation (2.37) holds for the special modes that we have defined, which use information
from the field just next to the null surface. In particular, even in the simple background of a
Schwarzschild black hole, these modes must be distinguished from the global Schwarzschild
modes that are defined by integrating the field all over the exterior or all over the interior.
The difference between these modes is manifest in their response to perturbations. If we
consider perturbations of the horizon generated by just throwing in some matter on top of
an equilibrium black hole then this will change the form of the two-point function between
Schwarzschild modes in the interior and the exterior. However, this perturbation does not
affect the nature of the entanglement very close to the horizon. In particular, the entanglement
between the modes a and e a, which are defined by integrals very close to the null surface U = 0,
is unaffected by a smooth deformation.
One way is that if we think of Schwarzschild modes with frequencies ω and ω 0 , then the
near-horizon modes defined above pick up the coefficient of the δ(ω − ω 0 ) term in their two-
point function. The analysis above tells us that this coefficient must be universal and cannot
be changed by a smooth deformation. We will elucidate this point further in our analysis of
the Reissner-Nordström black hole below, where we discuss global modes and explicitly relate
them to the near-horizon modes defined above.

3 The Reissner-Nordström black hole in AdS

We now apply the considerations above to the Reissner-Nordström (RN) geometry in anti-de
Sitter space. The techniques that we have developed here can be just as easily applied to
charged black holes in flat-space or in de Sitter space. The reason for considering AdS is
not only that this allows us to make a link with the AdS/CFT conjecture but also because
there is a canonical choice of a quantum state with asymptotically anti-de Sitter boundary
conditions—the Hartle-Hawking state. In flat space, it was shown through an elaborate
computation of the renormalized stress-tensor that the Hartle-Hawking state led to a singular
stress-tensor on the horizon of this black hole [11, 13]. Here, we will show how a similar
conclusion can be reached for charged black holes in flat space much more easily using our
test.

– 13 –
 This section is divided into four parts. First, we review the features of the classical RN
geometry, which also serves to introduce necessary notation. Next, we describe the global
expansion of fields propagating on this background. Then we relate these global modes to
the near-horizon modes described in the previous section. The criterion that the near-horizon
modes should be correctly entangled leads to specific constraints on the two-point functions of
the global modes. Although we were not able to check these constraints analytically, it is easy
to check them numerically in any dimension. In the last subsection, we present evidence that
the Hartle-Hawking/Kruskal state is singular at the inner horizon both for asymptotically
AdS RN black holes and for asymptotically flat RN black holes in various dimensions.

3.1 The classical geometry

The classical geometry of the RN black hole in AdS, in general spacetime dimension d + 1, is
described by the metric

ds2 = −f (r)dt2 + f (r)−1 dr2 + r2 dΩ2d−1 , (3.1)

where
A B
f (r) = r2 + 1 − + . (3.2)
rd−2 r2(d−2)
Here we have set the radius of AdS to 1 and and the constants A and B are related to the
mass, M and charge Q by

16πM (8πQ)2
A= ; B= , (3.3)
(d − 1)Vd−1 2(d − 1)(d − 2)Vd−1

where Vd−1 is the volume of the unit (d − 1)-sphere [25]. We will consider non-extremal black
holes where f (r) has first order zeros at r− and r+ , which are the positions of the inner and
outer horizons respectively.
As usual, it is convenient to introduce the tortoise coordinate
dr
dr∗ = . (3.4)
f (r)
Near the outer horizon we have r∗ → −∞ and we choose the origin of r∗ so that as one
approaches the horizon, we have
1 2κ+ r∗
(r − r+ ) → e , (3.5)
2κ+

where the surface gravity at the horizons is, as usual, given by κ+ = f 0 (r+ )/2. This fixes r∗
to a constant asymptotic value near the boundary of AdS. Moreover, near the future outer
horizon, we also have t → ∞ and, as usual, spacetime can be continued past this horizon by
introducing coordinates
1 κ+ (r∗ −t) 1 κ+ (r∗ +t)
U =− e ; V = e . (3.6)
κ+ κ+

– 14 –
In these coordinates, the right future outer horizon is at U = 0, V > 0. After crossing this
horizon, in the forward wedge, we define the t, r∗ coordinates by
1 κ+ (r∗ −t) 1 κ+ (r∗ +t)
U= e ; V = e . (3.7)
κ+ κ+
Within the forward wedge, we can move to the left outer horizon where, again, r∗ → −∞
but t → −∞. This horizon has V = 0. It is possible to cross this left horizon to reach a left
asymptotic region.
However, in the forward wedge, it is also possible to reach the region r∗ → ∞, which
marks the inner horizon. Within the forward wedge, the right inner horizon has t → +∞,
and the left inner horizon has t → −∞. Near the inner horizon, the relationship between the
tortoise and the ordinary radial coordinate becomes
ζ 2 −2κ− r∗
r − r− → e . (3.8)
κ−
Here, the surface gravity of the inner horizon is κ− = −f 0 (r− )/2. Note that one unavoidably
obtains an additional constant, ζ, in the relationship between r − r− and r∗ having fixed a
similar constant to 1 near the outer horizon. Naively, it appears that a simple change of
coordinates will allow the continuation of the geometry beyond the inner horizon as well, and
this naive extension leads to an extended spacetime diagram. In particular, define
ζ −κ− (r∗ +t) ζ κ− (r∗ +t)
U0 = − e V0 =− e . (3.9)
κ− κ−
This places the right inner horizon at U 0 = 0 and the left inner horizon at V 0 = 0. When we
cross the right inner horizon, we may use the coordinates
ζ −κ− (r∗ +t) ζ κ− (r∗ +t)
U0 = e V0 =− e . (3.10)
κ− κ−
A similar change of coordinates allows an extension across the left inner horizon.
The naive extended Penrose diagram is shown in Figure 2.

3.2 Quantum fields on the RN background

We now consider a scalar quantum field propagating on the background above. We will
assume that the scalar field is governed by an effective action
1 √  µν
Z
−g g ∂µ φ∂ν φ − m2 φ2 .

Seff = (3.11)
2
Since the field equations are linear, we expect to be able to expand the field as
X Z dω
φ= √ Fω,` (r∗ )e−iωt Y` (Ω) + h.c, (3.12)
`
2π

where Fω,` are operators. We now describe the operators Fω,` in more detail in various limits.
This also serves to define our notation.

– 15 –
Figure 2: Maximal extension of the Reissner Nordström spacetime. The dots indicate repe-
titions of the displayed pattern.

Field expansion near the boundary of AdS

Near the boundary of AdS, we demand that the field be normalizable. This corresponds to
a situation where no sources have been turned on for the field. Near the boundary we have
lim [r∆ Fω,` (r)] = Oω,` , (3.13)
r→∞
where Oω,` are the modes of the boundary operator of dimension ∆ that is dual to the bulk
field φ.

Field expansion just outside the outer horizon

The normalizable boundary conditions link the left and right moving modes outside the
horizon. As we approach the outer horizon, the tortoise coordinate r∗ → −∞. A simple
consideration of the wave-equation resulting from the action (3.11) shows that as we approach
the horizon from outside, so that r → r+ but r > r+ , the field has the expansion
1 
iωr∗ −iδω,` −iωr∗

Fω,` −→ √ d−1 aω,` e + e e . (3.14)
r→r+
2ωr+2
With the normalization above, the operators aω,` are canonically normalized
[aω,` , aω0 ,`0 ] = δ(ω − ω 0 )δ`,`0 . (3.15)

– 16 –
Field expansion just inside the outer horizon
Just inside the outer horizon, the left-moving modes must be the same as the left-movers
outside the horizon by continuity of the field. However, it is possible to have new right-moving
modes. Therefore, as r → r+ but r < r+ , the field has the expansion
1  
Fω,` −→ √ d−1 ea†ω,` eiωr∗ + aω,` e−iδω,` e−iωr∗ . (3.16)
r→r+ 2
2ωr+

Field expansion just outside the inner horizon

As we approach the inner horizon, we find that r∗ → ∞. As we approach it from outside,
i.e. r → r− but r > r− , we find that the radial mode functions simplify again and we may
write
1  
eb† eiωr∗ + bω,` e−iωr∗ .
Fω,` −→ √ d−1 ω,` (3.17)
r→r− 2
2ωr−

The modes b and eb are not independent and must be related to the modes a, e a just inside the
outer horizon. The relationship can be obtained by evolving the expansion given in (3.16)
forward in time until one reaches the inner horizon. We will consider this relationship in more
detail in subsection 3.5.

Field expansion just inside the inner horizon

For the sake of completeness, we also describe the field expansion just inside the inner
horizon. This is an academic exercise since we have not yet found any state where the
conditions for smoothness are met even outside the inner horizon. However, if such a state
were to be constructed, then the expansion behind the inner horizon would be relevant.
Just inside the inner horizon, with r → r− and also r < r− , we may write
1  
eb† eiωr∗ + c† e−iωr∗ .
Fω,` −→ √ d−1 ω,` ω,` (3.18)
r→r−
2ωr−2

Here we have imposed the fact that the “right movers”, eb cross over smoothly whereas the
c are some new modes which may or may not be related to the earlier modes. Notice that
both eb and c have the wrong energy with respect to the Schwarzschild Hamiltonian. This is
because, in our coordinates as we move up on the diagram 2, the value of t decreases.
Note that this expansion relies only on the continuity of the metric, and does not make
any assumptions about what happens deep inside the inner horizon.

3.3 Relationship between near-horizon modes and global modes

We can define near-horizon modes near all the horizons in this geometry. We will define
modes on both sides of the right outer horizon, and on both sides of the right inner horizon.
It is, of course, possible to define modes near the left-horizons but in our description (which
treats the horizons symmetrically) this is redundant and so we will avoid it for now.

– 17 –
 So we define

U −iω0
Z  
1
a= √ ∂U φ(U, V = 0, Ω) − T (−U )Y`∗ (Ω) dU dd−1 Ω,
πω0 U0
Z  iω0
1 U
a= √
e ∂U φ(U, V = −, Ω) T (U )Y` (Ω) dU dd−1 Ω,
πω0 U0
(3.19)
U 0 −iω1
Z  
1 0 0
b= √ ∂U 0 φ(U , V = 0, Ω) − 0 T (−U 0 )Y`∗ dU 0 dd−1 Ω,
πω1 U0
Z  0 iω1
1 0 0 0 U
c= √ ∂U φ(U , V = −, Ω) T (U 0 )Y` (Ω) dU 0 dd−1 Ω.
πω1 U00

Here ω1 = κκ+−ω0 . All these modes depend on a choice of ω0 and `0 and while we suppress
these quantities in the notation we will make some choices for them later.
The integrals in (3.19) can be performed as follows. For the first line of (3.19) we find
that

U −iω0
Z  
1  −iωt ∗ iωt

a= √ ∂U Fω,` (r∗ )e + Fω,` (r∗ )e − T (−U )dU dω. (3.20)
π 2ω0 U0

Now, since the tuning function has support for only very small values of U , we may expand the
mode function using the approximation (3.14). We then find that up to the O () dependence
on the cutoffs, described in section 2,

1
Z iω  i κω
  −U −iω0 dω
a= √ aω,` (κ+ ) κ+
∂U (−U ) + T (−U )dU √
2π ω0 U0 ω
 ω
i √ −U i( κ+ −ω0 )
Z 
1 iω dU (3.21)
= √ κ+
ω(κ+ U0 ) aω,` T (−U ) dω
2π ω0 κ+ U0 U
Z r
ω iω ω dω
= i s(ω0 − )(κ+ U0 ) κ+
aω,` .
κ+ ω0 κ+

Note that the normalization of the modes is correct since, using the expression above, and
also the identity [aω,` , aω0 ,` ] = δ(ω − ω 0 ), we can check that
Z
† ω 2 ω dω
[a, a ] = |s(ω0 − )|
κ+ ω0 κ2+
Z 0
(3.22)
0 2ω 0
= |s(ω0 − ω )| dω = 1.
ω0

where we used that s(ω) is very sharply peaked around ω = 0.

Similarly, we can express the near-horizon modes in terms of the global modes near the
other horizons as well. We find that that in the limit under consideration precisely the same

– 18 –
function s appears in these other expressions.
Z r
−iω
κ+ ∗
ω ω dω
a = −i (κ+ U0 ) s (ω0 −
e ) aω,` .
e
κ+ ω0 κ+
Z r
κ− −iω ω ω dω
b = i ( U0 ) s(ω1 −
κ−
) bω,` . (3.23)
ζ κ− ω1 κ−
Z r
κ− iω ω ω dω
c = −i ( U0 ) κ− s∗ (ω1 − ) cω,` .
ζ κ− ω1 κ−

As advertised, in each case the near-horizon modes can be written as global modes smeared
with a function that is sharply peaked in frequency space.

3.4 Constraints on two-point functions

From the results of section 2, we now find the following constraints on the near-horizon modes
defined above. As explained there, smoothness of the outer horizon implies that
1
hΨ|ea† |Ψi = hΨ|aa† |Ψi =
ae ;
1 − e−2πω0
(3.24)
e−πω0
hΨ|e a† a† |Ψi =
aa|Ψi = hΨ|e .
1 − e−2πω0
On the other hand, upon applying these constraints near the inner horizon we find that
1
hΨ|bb† |Ψi = hΨ|cc† |Ψi = ;
1 − e−2πω1
(3.25)
e−πω1
hΨ|bc|Ψi = hΨ|b† c† |Ψi = ,
1 − e−2πω1
where ω1 = κκ− +
ω0 as above.
Now, as discussed earlier, the modes b, b† can in principle be expressed as linear com-
binations of a, ea, a† , e
a† by solving the wave equation in the region between the two horizons.
Hence it is not obvious whether equations (3.24) and (3.25) can hold simultaneously. In fact
as we will show numerically in the next subsections, these equations are incompatible in the
case of the Hartle-Hawking state for the AdS-RN black hole. This statement implies that
this state is singular on the inner horizon of the AdS-RN black hole.
Before we continue with this analysis, let us see what the constraints (3.24), (3.25) imply
for the two-point function of the global modes a, b, e a. Let us assume that the two-point
function of the global modes is given by

hΨ|aω,` a†ω0 ,` |Ψi = A1 (ω)δ(ω − ω 0 ) + A2 (ω, ω 0 );

hΨ|e a†ω0 ,` |Ψi = Ae1 (ω)δ(ω − ω 0 ) + Ae2 (ω, ω 0 );
aω,` e
(3.26)
aω0 ,` |Ψi = C1 (ω)δ(ω − ω 0 ) + C2 (ω, ω 0 );
hΨ|aω,` e
hΨ|bω,` b†ω0 ,` |Ψi = B1 (ω)δ(ω − ω 0 ) + B2 (ω, ω 0 ),

– 19 –
where, in each case, we have separated the correlator into one part that is proportional to
a delta function and another part that is assumed to be smooth at ω = ω 0 . (We have
also used the spherical symmetry to set the same value for ` in all the operators above.)
Then, substituting the formulas above, we find that the coefficients of the delta function are
completely fixed by demanding smoothness at the outer and the inner horizon. In particular,
we find that
1
A1 (ω) = Ae1 (ω) = ;
− κ2π ω
1−e +

− κπ ω
e +
C1 (ω) = ; (3.27)
− 2π ω
1 − e κ+
1
B1 (ω) = ,
− κ2π ω
1−e −

where we have used the relationship between ω1 = κ+ ω0 /κ− . Note that the factors that
appear in the exponentials above are just the standard inverse-temperatures of the inner and
outer horizons given by β± = κ2π± .
Therefore we see that the two-point function of the near-horizon modes picks only the
delta-function piece in the two-point function of the global modes and completely fixes that
piece. This explains why a smooth deformation of the geometry cannot change the two-point
function of the near-horizon modes: a smooth deformation of the geometry may change the
smooth part of the two-point function of the global modes, but it cannot alter the coefficient
of the delta function in this two-point function. This is the only term that the near-horizon
modes are sensitive to and it must take a given value near a smooth horizon.
Now, as we explained above, in any given state of the system, the b modes are obtained
by evolving the a and e a modes in the region between the inner and outer horizon of the
black hole. Therefore, given the correlators of those modes, it is possible to check whether
the constraints (3.27) are satisfied. We will show that for the RN geometry, in various
dimensions, these constraints cannot all be satisfied simultaneously.

3.5 Numerical results for the Reissner-Nordström black hole

We now check whether our constraints are satisfied in the AdS Reissner-Nordström geometry.
In this situation, where the propagation is linear, and using the time-translation isometry
of the geometry we can write
b = κ1 a + κ a† .
e2 e (3.28)
It is the time-translation invariance that allows us to relate modes with the same value of ω,
and we note that this equation may be modified if interactions are strong, or if the geometry
is strongly time-dependent. Therefore,

hbb† i = haa† i|κ1 |2 + |e a†e

κ2 |2 he ai + 2Re (κ1 κ e∗2 hae
ai)
1 2 −2πω0 2 −πω0 ∗

(3.29)
= −2πω
|κ 1 | + e |e
κ 2 | + 2e Re (κ1 κ 2 ) .
1−e 0
e

– 20 –
So the question of checking whether the near-horizon constraints (3.24),(3.25) are satisfied as
we approach the near-horizon reduces to evaluating the Bogoliubov coefficients, κ1 , κ e2 and
determining if the combination above is in agreement with (3.25).By computing the two-point
function of the modes near the inner horizon we define

δ = hb† bi(1 − e−2πω1 ) − 1. (3.30)

which is the fractional difference from the expected Boltzmann factor. If the constraints are
satisfied we will have δ = 0.

Description of the numerical algorithm

These Bogoliubov coefficients can be computed by solving the radial (ordinary) differential
equation that fixes the evolution of the field. We briefly describe our algorithm. A solution
of the wave-equation with frequency ω and angular momentum, ` can be written in the form
φω,` (r∗ )e−iωt Y` (Ω). The radial part of the solution φω,` obeys the equation
 
d d−1 dφω,` φω,`
r + ω 2 φω,` − `(` + d − 2)f (r) 2 − m2 φω,` f (r) = 0, (3.31)
dr∗ dr∗ r
where r is determined by solving the auxiliary equation r0 (r∗ ) = f (r). This auxiliary equation
must be solved first; demanding the behaviour (3.5) near the outer horizon allows us to fix
the value of r∗ near the boundary of AdS as r → ∞. Solving the auxiliary equation between
the inner and outer horizon allows us to determine ζ, which appears in (3.8).
The equation (3.31) is now solved, as a set of two first-order differential equations, starting
from the boundary with initial conditions set by φω,` (r) → r1∆ at large r. As r → r+ , we can
dφ
then read off the phase δω,` by matching the behaviour of φω,` and φ0ω,` = drω,` ∗
to (3.14).
Now, starting at a point in the middle of the inner and outer horizons, we solve the
equation towards r∗ → −∞ (the outer horizon) and also r∗ → ∞ (the inner horizon) with
two separate initial conditions: once with φω,` = 0, φ0ω,` = 1 and a second time with φω,` =
1, φ0ω,` = 0. Near both the inner and the outer horizons, the asymptotic behaviour of the
mode is given by (3.16) and (3.17). By determining the coefficients of the terms that oscillate
as e±iωr∗ for both sets of initial conditions, we can solve for the Bogoliubov coefficients κ1
and κ e2 . This immediately leads to a result for δ.
The procedure above is quite simple, and can be implemented using any standard nu-
merical library. In our own case, to solve the differential equations, we used an eighth-order
explicit Runge-Kutta solver with the Dormand and Prince coefficients [26] as implemented in
C using code from CALCODE [27]. We also used the root-finding routines in the GNU scien-
tific library [28] to locate the horizons and GNU parallel [29] to parallelize the calculations.

Results
In Figure 3 we present a plot for δ vs the frequency for the angular momentum ` = 0 for
various values of the radii of the inner and outer horizon and in dimensions d = 3, 4, 5, 6.
The black holes range from those that are near-extremal to those where there is a significant
separation of scales between the two horizons. The AdS radius is fixed to unity.

– 21 –
 δ δ
1.0

1.5
0.8
rp=1.08 rm=0.92
rp=1.07 rm=0.78
rp=1.02 rm=0.63
rp=1.03 rm=0.37 0.6
1.0 rp=1.19 rm=0.52
rp=1.27 rm=0.26
0.4 rp=1.3 rm=0.45
rp=1.44 rm=0.2
rp=1.63 rm=0.32
0.5
0.2

ω ω
1 2 3 4 5 0 1 2 3 4 5

(a) d = 3 (b) d = 4
δ δ
0.6

0.5
rp=1.08 rm=0.96 0.3 rp=1.07 rm=0.97
0.4 rp=1.02 rm=0.74 rp=1.02 rm=0.8
rp=1.14 rm=0.65 0.2 rp=1.11 rm=0.72
0.3
rp=1.22 rm=0.59 rp=1.18 rm=0.68
0.2 rp=1.46 rm=0.46 rp=1.36 rm=0.56
0.1
0.1

ω ω
1 2 3 4 5 1 2 3 4 5

(c) d=5 (d) d = 6

Figure 3: A plot of the fractional difference, δ, vs the frequency for ` = 0 for various values
of the inner and outer temperature. Non-zero values of δ indicate that the constraints of
section 2 are not satisfied.

These graphs are already sufficient to show that the Hartle-Hawking state is singular
at the inner horizon of these geometries. This is because the criterion of section 2 must be
satisfied for each value of frequency and angular momentum, as a necessary condition, for
the state to be smooth. Instead the graphs above all display non-zero values of δ for generic
choices of frequency and angular momentum.
Our numerical results are quite robust. In fact, the largest source of numerical error
arises from the relative phase between κ1 and κ e2 in (3.29). This phase requires a careful
determination of ζ and a treatment of the wave-equation near the horizon. However, even
this is not a significant source of error since, unlike the case of [30], we are considering a
specific mode and so the wave-equation simplifies greatly near the horizon. Moreover, it is
not difficult to check that if one keeps the angular momentum fixed and increases ω then
beyond a point, no choice of relative phase between κ1 and κ e2 will yield δ = 0. This allows
us to reach the physical conclusions above — that the inner horizon is not smooth in the
Hartle-Hawking state — with confidence.
However, it is also interesting to understand how the fractional difference varies with
angular momentum. In Figure 4, we present a plot for δ vs the angular momentum for
frequency fixed at the AdS scale for d = 4. The reader will note the remarkable fact that
at large `, the fractional difference tends to zero. This can be explained via a semiclassical
WKB analysis in this limit, as we demonstrate in Appendix 5. This WKB analysis serves as
an additional check on our numerical algorithm.

– 22 –
 We note that in [13], it was pointed out that the renormalized expectation value of φ2 is
less singular than expected as one approaches the inner horizon. This is directly a consequence
of the large ` behaviour of Figure 4, since the fastest divergence of φ2 as one approaches the
inner horizon is controlled entirely by the large-` modes.

δ
0.20

0.15 rp=1.08 rm=0.92

rp=1.02 rm=0.63
rp=1.19 rm=0.52
0.10
rp=1.3 rm=0.45
rp=1.63 rm=0.32
0.05

0.00 ω
10 20 30 40

Figure 4: A plot of the fractional difference, δ, vs the angular momentum with frequency
fixed to the AdS scale for various sizes of the inner and outer horizons.

4 The BTZ black hole

We now turn to a discussion of the rotating BTZ black hole. It was argued in [12] that a
rotating BTZ black hole that is close enough to extremality violates strong cosmic censorship,
and so this presents an excellent test-case for our criterion.
We start by reviewing the geometry of the BTZ black hole, and the propagation of
quantum fields in this geometry. We then show that it is indeed the case that the near-horizon
modes, as one approaches the inner horizon from the outside, have the correct occupation
number. This corresponds to the fact that if we set the local temperature of the field correctly
near the outer horizon, it is automatically red-shifted by the geometry so that the local
temperature of the modes near the inner horizon coincides exactly with the temperature of
the inner horizon! This is in contrast to what we found numerically in the previous section
for the AdS-RN black hole in higher dimensions.
However, as explained in section 2, if we want to extend the spacetime behind the inner
horizon, we also require the existence of modes behind the inner horizon. The reader may
worry that the extension of quantum fields behind the inner horizon is not unique. Neverthe-
less, as we have emphasized the near-horizon modes just behind the inner horizon are fixed by
requirements of smoothness and do not require knowledge of the dynamics deep behind the
inner horizon.
When one crosses the outer horizon, interior modes can be understood using the standard
construction of the mirror operators [19]. However, we show that this construction fails at
the inner horizon due to the monogamy of entanglement. In particular, since the modes

– 23 –
between the inner and the outer horizon are already entangled with the modes outside the
outer horizon, they cannot also be entangled with new modes behind the inner horizon.
Nevertheless we show, remarkably, that it is possible to reuse the modes between the
inner horizon and the outer horizon behind the inner horizon as modes for the field behind
the inner horizon. This is dependent on the fact that the modes outside the inner horizon
are correctly populated. The modes that we write down uniquely fix the behaviour of the
field just behind the inner horizon although the usual ambiguities associated with a Cauchy
horizon arise if we attempt to probe deeper into the geometry.
We will follow the notation of [12] in large part.

4.1 Classical geometry and propagation of fields

The BTZ solution can be written in the form
dr2
ds2 = −f (r)dt2 + + r2 (dφ − Ω(r)dt)2 , (4.1)
f (r)

where
(r2 − r+
2 )(r 2 − r 2 )
− r+ r−
f (r) = ; Ω(r) = . (4.2)
r2 r2
Here we have set the AdS radius to 1.
The positions of the inner and outer horizon are at r+ and r− . The angular velocities
Ω± and surface gravities κ± of the two horizons are given by
2 − r2
r+
r∓ −
Ω± = ; κ± = . (4.3)
r± r±

The tortoise coordinate is defined as in (3.4) and we once again adopt the convention
1 2κ+ r∗
that as one approaches the outer horizon, (r − r+ ) = κ+ e . We can check that near the
inner horizon this leads to a relationship of the form (3.8) with
r− +r+ r+
+1
ζ=2 2r−
(r+ − r− ) r− . (4.4)

We note that our conventions for r∗ differ from those of [12] by an overall additive constant.
We now consider a minimally coupled scalar field φ propagating in this geometry. Just
as above, we may expand the field in various regions. However, the expansion in terms of
creation and annihilation operators is slightly more subtle than in the non-rotating case, as
we explain below.
Near the boundary of AdS we may, as usual, expand the field as
Oω,m
Z
1 X
φ −→ dω ∆ e−iωt eimφ + h.c, (4.5)
r→∞ 2π r
m

where ∆(∆ − 2) = m2 and Oω,m are the Fourier modes of the primary operator of dimension
∆ that is dual to the field φ.

– 24 –
 Near the outer horizon, as we approach it from outside on the right, the expansion of the
field φ is3
XZ dω+  
−iδω+ ,m −iω+ κ+ r∗
φ −→ √ √ a ω+ ,m eiω+ κ+ r∗
+ e e e−iκ+ ω+ t eim(φ−Ω+ t) + h.c,
r→r+
m
2π r+ 2ω +
(4.6)
where
ω − mΩ+ ω − mΩ−
ω+ = ; ω− = . (4.7)
κ+ κ−
A subtlety here is that in the expression above, the question of whether the coefficient of the
mode function is a creation or annihilation operator is determined by the positivity of ω+
and not of ω. With this convention the operators above are canonically normalized

[aω+ ,m , a†ω0 0
0
] = δ(ω+ − ω+ )δmm0 . (4.8)
+ ,m

The expansions (4.5) and (4.6) fix the relationship between the operators aω+ ,m and
Oω,m and also the phase δ. These can be determined by solving the wave-equation between
the boundary and the outer horizon, which can be done analytically [31]. We will use the
solutions as written in the conventions of [12].
1
aω+ ,m = Oω,m
C
κ+ iω+ 2 Γ 21 (∆ − iω+ − iω− ) Γ 12 (∆ − iω+ + iω− )
 


2 ∆ 1
C= √ (r+ − r− ) 2 √ √ (4.9)
2 r+ 2ω+ Γ(∆)Γ(−iω+ )
2iω+ 1

1


Γ(iω+ )Γ 2 (∆ − iω+ − iω− ) Γ 2 (∆ − iω+ + iω− )

−iδ κ+
e = √
.
Γ(−iω+ )Γ 21 (∆ + iω+ − iω− ) Γ 12 (∆ + iω+ + iω− )


2
We now proceed with the expansion of the field just inside the outer horizon
XZ dω+  
−iκ+ ω+ t im(φ−Ω+ t) iκ+ ω+ t −im(φ−Ω+ t)
φ −→ √ √ a ω+ ,m e e + a
e ω+ ,m e e e−iω+ κ+ r∗
r→r+
m
2π r+ 2ω +

+ h.c.
(4.10)
Once we are near the inner horizon, we may write
XZ dω−  
−iκ− ω− t im(φ−Ω− t) iκ− ω− t −im(φ−Ω− t)
φ −→ √ √ bω− ,m e e + b
eω− ,m e e e−iω− κ− r∗
r→r−
m
2π r− 2ω−

+ h.c,
(4.11)
where
ω − mΩ−
ω− = . (4.12)
κ−
3
We find it more convenient notationally to define the modes aω+ ,m in a somewhat asymmetric conventions,
so that there is no phase factor eiδω+ ,m in front of eiω+ κ+ r∗ .

– 25 –
Note that in the expansion near the inner horizon, the classification of operators into creation
and annihilation operators is determined by the sign of ω− . Finally, if the field extends across
the inner horizon, we may write
XZ dω−  
φ −→ √ √ cω− ,m eiω− κ− r∗ + ebω− ,m e−iω− κ− r∗ eiκ− ω− t e−im(φ−Ω− t) + h.c.
r→r−
m
2π r− 2ω−
(4.13)

4.2 Near-horizon modes and constraints from entanglement

We now describe the relationship between the near-horizon modes and the global modes
described above, and explain the constraints that the analysis of section 2 places on the
two-point functions of the global modes.
Here, it is more convenient to use Eddington-Finkelstein coordinates, rather than Kruskal
coordinates. These coordinates can be defined in the vicinity of both the inner and the outer
horizon. Near the outer horizon, we set v+ = r∗ + t so that the metric is given by

ds2 = −f (r)dv+
2
+ 2dv+ dr + r2 (dφ − Ω(r)dt)2 . (4.14)

Now define x+ = r+ − r and θ+ = φ − Ω+ t. Then, as we approach the horizon which is at

x+ = 0, the metric becomes

ds2 = −2dv+ dx+ + r+

2 2
dθ+ + O (x+ ) , (4.15)

which is the form that was required in section 2.

We now define near-horizon modes using the general prescription outlined in 2. In par-
ticular, define
x+ −iω0
Z   r
1 −imθ+ r+
a= √ ∂x+ φ(x+ , v+ = 0, θ+ ) − T (−x+ )dx+ e dθ+ ,
πω0 U0 2π
(4.16)
x+ iω0
Z   r
1 imθ+ r+
a= √
e ∂x+ φ(x+ , v+ = −, θ+ ) − T (x+ )dx+ e dθ+ .
πω0 U0 2π
Such modes can be defined for any value of v+ but, as the reader can see from the relationship
with the global modes below, shifting the value of v+ only rescales the mode by a phase.
Similarly, we can define near-horizon modes near the inner horizon. There, with v− =
r∗ − t, and x− = r− − r, and θ− = φ − Ω− t, as we approach the inner horizon at r− = 0, the
metric becomes
ds2 = −2dv− dx− + r+ 2 2
dθ− + O (x− ) . (4.17)
Therefore, the prescription of section 2 tells us that near-horizon modes can be defined near
the inner horizon using
x− −iω1
Z   r
1 imθ− r−
b= √ ∂x− φ(x− , v− = 0, θ− ) − T (−x− )dx− e dθ− ,
πω1 U0 2π
(4.18)
x− iω1
Z   r
1 r −
c= √ ∂x− φ(x− , v− = −, θ− ) − T (x− )dx− e−imθ− dθ− .
πω1 U0 2π

– 26 –
As in the section above, ω1 = κκ+−ω0 .
Using the expansion of the field near the various horizons, we can relate these modes to
the global modes. In doing the relevant integrals, the reader should keep in mind that the
expressions above, which are given in terms of r∗ and t need to be transformed to x± , v± . So,
for instance,
eiκ+ ω+ (r∗ −t) = e2iκ+ ω+ r∗ e−iκ+ ω+ v+ = (κ+ x+ )iω+ e−iκ+ ω+ v+ . (4.19)
Making similar substitutions in the other near-horizon expansions and using the definition
(2.10), we find
Z r
iω+ ω+
a = i s(ω0 − ω+ )aω+ ,m (κ+ U0 ) dω+ ;
ω0
Z r
∗ −iω+ ω+
a = −i s (ω0 − ω+ )e
e aω+ ,m (κ+ U0 ) dω+ ;
ω0
Z r (4.20)
κ− U0 iω− ω−
b = i s(ω1 − ω− )bω− ,m ( ) dω− ;
ζ ω0
Z r
κ− U0 −iω− ω−
c = −i s∗ (ω1 − ω− )cω− ,m ( ) dω− .
ζ ω0

The constraints of section 2 now lead to the following constraints on these modes. At
the outer horizon we have
   
a − e−πω0 e
a† |Ψi = 0; a† − eπω0 e
a |Ψi = 0; [a, e
a] = 0;
1 (4.21)
hΨ|aa† |Ψi = hΨ|e a† |Ψi =
ae .
1 − e−2πω0
These constraints are automatically met in the Hartle-Hawking state. For the occupation of
a this follows since, in that state, the global modes are populated as4
1
hΨ|aω+ ,m a†ω0 0 |Ψi = 0
δ(ω+ − ω+ )δmm0 . (4.22)
+ ,m 1 − e−2πω+
Moreover, the global e
a modes can be constructed using the mirror-operator construction [19],
which yields modes whose two-point function is again
1
hΨ|e a†ω0
aω+ ,m e 0 |Ψi = 0
δ(ω+ − ω+ )δmm0 , (4.23)
+ ,m 1 − e−2πω+
and which are moreover entangled with the modes outside the horizon through

aω+ ,m |Ψi = e−πω+ e

a†ω+ ,m |Ψi; a†ω+ ,m |Ψi = eπω+ e
aω+ ,m |Ψi;
(4.24)
aω+ ,m |Ψi = e−πω+ αω† + ,m |Ψi;
e a†ω+ ,m |Ψi = eπω+ aω+ ,m |Ψi.
e

Using (4.20), we can then check that all the relations in (4.21) are satisfied.
4
Notice that the occupation levels in (4.22) know about the temperature of the black hole via the factor
κ+ which enters in the relation (4.7) between ω+ and ω.

– 27 –
 Of more interest to us are the constraints that the analysis of section 2 places on the
modes near the inner horizon. Here the constraints of section 2 can be divided into two parts.
The first part is that as we approach the inner horizon from the exterior, the modes should
be correctly populated
1
hΨ|bb† |Ψi = . (4.25)
1 − e−2πω1
The second condition is that the modes behind the inner horizon should be correctly entangled
with the modes in front
   
b − e−πω1 c† |Ψi = 0; b† − eπω1 c |Ψi = 0; [b, c] = 0. (4.26)

We check (4.25) in section 4.3 and analyze (4.26) in section 4.4.

4.3 Checking the constraints on approaching the inner horizon

The constraint (4.25) can be checked by using the propagation of fields between the inner and
outer horizon and the propagation of fields from the boundary to the outer horizon. First we
note that by virtue of the Killing isometry of the geometry we have

b = κ1 a + κ a† ,
e2 e (4.27)

Here the Bogoliubov coefficients, κ1 and κ e2 can be obtained from the reflection and trans-
mission coefficients given in [12], after accounting for our different normalizations and also
accounting for the relationship between local and global modes described. They are given by
q
ω−
ω+ Γ(iω− )Γ (1 + iω+ )
 iω+
iδ1 κ+
κ1 = e √ 1

1

eiδ
2 Γ 2 (−∆ + iω− + iω+ ) + 1 Γ 2 (∆ + iω− + iω+ )
q (4.28)
ω−

κ+
−iω+
ω +
Γ(iω− )Γ (1 − iω+ )
e2 = eiδ1 √
κ
,
Γ 2 (−∆ + iω− − iω+ ) + 1 Γ 12 (∆ + iω− − iω+
1


2

where the important phase eiδ is specified in (4.9) and the irrelevant common phase-factor is
given by
√ !iω− iω+
κ
2ζ (U κ )
0 + +
eiδ1 = iω− . (4.29)
κ− κ−
(U0 κ− )
Note that, as we will check below, these Bogoliubov coefficients correctly satisfy

|κ1 |2 − |e
κ2 |2 = 1. (4.30)

Therefore we find that the condition that must be met for (4.25) to be satisfied is that

1 −2πω+ e−πω+
2 e ∗ ∗ 1
|κ1 |2 −2πω
+ |e
κ 2 | −2πω
+ (κ κ
1 2 + κ κ
2 1 ) −2πω
= . (4.31)
1−e 1−e 1−e 1 − e−2πω−
+ +
e e +

– 28 –
 At first sight this might seem a little surprising. But, in fact, the identity above is true as
π
can be checked by just repeatedly using the Gamma-function identity Γ(iz)Γ(−iz) = z sinh(πz) .
In particular, we find that
1 1
e∗2 = −
κ1 κ (cosh(πω− ) − cosh(π(ω+ + i∆)));
2 sinh(πω− ) sinh(πω+ )
1 1
κ∗1 κ
e2 = − (cosh(πω− ) − cosh(π(ω+ − i∆)));
2 sinh(πω− ) sinh(πω+ )
(4.32)
1 1
|κ1 |2 = (cosh(π(ω− + ω+ )) − cos(π∆));
2 sinh(πω− ) sinh(πω+ )
1 1
κ2 |2 =
|e (cosh(π(ω− − ω+ )) − cos(π∆)).
2 sinh(πω− ) sinh(πω+ )

Putting these results together, with a little algebra, we find that (4.31) follows!
We should emphasize that this result arises as a result of a nontrivial conspiracy between
the reflection and transmission coefficients that control the propagation between the outer
and the inner horizon, and the phase factor e−iδ that arises from propagation outside the
outer horizon. This is reminiscent of the conspiracy between properties of the mode functions
in these regions that the authors of [12] noticed when they were considering the classical
problem in the same background.
Our test does not appear to be sensitive to the constraint, r+∆r −
−r− > 1, that was found
to be necessary in [12] for the stress-tensor to be regular at the inner horizon. Indeed, from
our point of view, this constraint is somewhat surprising since (4.31) tells us that the state
is as smooth as possible at the inner horizon: all the modes are occupied at just the right
temperature. It would be interesting to understand this additional constraint through a
mode-sum calculation of the stress-tensor near the inner horizon [32].

4.4 Extending the field behind the inner horizon

The second part of our test in section 2 was that operators in front of the horizon must also
be correctly entangled with operators behind it. In this case, the key-point is to focus on the
operators c. One might have thought that since the c operators could, in principle, be “new”
operators, one could just write them down using the standard mirror operator construction
[19]. However, this is not possible due to the monogamy of entanglement. Since the modes
just outside the inner horizon are linear combinations of modes near the outer horizon (4.27),
and since those modes are already entangled with modes outside the outer horizon as in
(4.20), (4.21), the modes near the inner horizon cannot also be entangled with fresh modes
inside the inner horizon. Note that this application of the monogamy of entanglement is
very different from the one used in the fuzzball or firewall arguments, since it can be phrased
entirely at the level of effective field theory and involves only simple operators. Nevertheless,
it turns out to be possible to cleverly reuse the modes behind the inner and outer horizon to
generate modes behind the inner horizon.

– 29 –
 No new modes
The fact that any new modes behind the inner horizon decouple can be seen from the
following argument. Let us write

c=κ a + κ4 a† + κ5 d + κ6 e† .
e3 e (4.33)

where d and e denote candidate new modes which commute with e a and a† . We will first show
†
that the new modes d and e can consistently be set to zero above.
This is done as follows. First, we use the maximal entanglement of a and e
a to show that
the new modes cannot have any correlators with them. We start with
 
a − e−πω0 e
a† |Ψi = 0, (4.34)

and therefore,  
hΨ|d a − e−πω0 e
a† |Ψi = 0. (4.35)

a† ,
But since d computes both with a and with e
 
hΨ| a − e−πω0 e a† d|Ψi = 0. (4.36)

But note that we also have

a† .
hΨ|a = eπω0 hΨ|e (4.37)
Substituting this into the previous equation, we find that

a† d|Ψi = 0.
2 sinh (πω0 ) hΨ|e (4.38)

After employing similar reasoning for the two-point function with a, we conclude that

a† d|Ψi = 0.
hΨ|ad|Ψi = hΨ|e (4.39)

Similarly, we have
hΨ|ae† |Ψi = hΨ|e
a† e† |Ψi = 0. (4.40)
The constraints (4.26) require

c|Ψi = e−πω1 b† |Ψi. (4.41)

Expanding both the left and the right hand sides using (4.33) and (4.27), we see that
   
κ a + κ4 a† + κ5 d + κ6 e† |Ψi = e−πω1 κ∗1 a† + κ
e3 e e∗2e
a |Ψi. (4.42)

But since d|Ψi and e† |Ψi are orthogonal to all the other vectors that appear above, we see
immediately that
(κ5 d + κ6 e† )|Ψi = 0. (4.43)
But now using that
c† |Ψi = eπω1 b|Ψi, (4.44)

– 30 –
we can also conclude that
(κ∗5 d† + κ∗6 e)|Ψi = 0. (4.45)
But from (4.45) and (4.43), we find that

|κ5 |2 − |κ6 |2 = 0. (4.46)

So not only are the new modes not entangled with the old modes, they cannot even contribute
to the “norm” of the oscillators behind the horizon!
The results (4.39), (4.40) and (4.46) together constitute an important result. They show
us that we cannot define new modes behind the inner horizon that automatically have the
right entanglement with modes in front of the inner horizon. The reader might worry why
the mirror operator construction [19] fails at the inner horizon. The reason is given by the
equation (4.34). This means that if we look at the set of simple operators outside the inner
horizon, the state is not separating with respect to these operators.
Therefore, for the remainder of this analysis, we will set

κ5 = κ6 = 0. (4.47)

A pictorial representation of our result that one cannot arbitrarily define new modes
behind the inner horizon is shown in Figure 5

Figure 5: A figure of the outer and inner horizons and the modes near them. The purple
modes near the right inner horizon are linear combinations of the green and violet modes near
the outer horizons. But the green and violet modes are already entangled together. So, it is
nontrivial to find the brown modes behind the inner horizon with the correct entanglement.

Reusing old modes

At first the reader might conclude that the analysis above implies that it is impossible to
extend the field behind the inner horizon. However, it turns out that is possible to cleverly
reuse the old modes by choosing appropriate values for κe3 and κ4 in (4.33) so as to satisfy
the entanglement constraints.

– 31 –
 The concern is that the constraints might overdetermine κ e3 and κ4 . However, the idea is
to use the entanglement constraints between c and b to solve for κ e3 and κ4 and then simply
check whether [c, b] = 0 and [c, c† ] = 1. This is done as follows. We first note using (4.33)
(and setting κ5 = κ6 = 0)

c|Ψi = (e a + κ4 a† )|Ψi = (e
κ3e κ3 + κ4 eπω0 )e
a|Ψi, (4.48)

where we have used the entanglement between a and e a. On the other hand, using the entan-
glement that is required between c and b and using (4.27) we find that

c|Ψi = e−πω1 b† |Ψi = e−πω1 (κ∗1 a† + κ

e∗2e
a)|Ψi
(4.49)
= e−πω1 (κ∗1 eπω0 + κ
e∗2 )e
a|Ψi,

where, in the last line, we again used the entanglement between the a and the e a modes.
Equating the final expressions on the two sides of the equations above, and using the fact
a|Ψi =
that e 6 0 leads to the condition that

e3 + κ4 eπω0 = e−πω1 (κ∗1 eπω0 + κ

κ e∗2 ). (4.50)

We can perform a similar analysis using the action of c† . Here we find that

c† |Ψi = (e
κ∗3e
a† + κ∗4 a)|Ψi = (e
κ∗3 + κ∗4 e−πω0 )e
a† |Ψi. (4.51)

On the other hand, the smoothness of the inner horizon requires

c† |Ψi = eπω1 b|Ψi = eπω1 (e a† + κ1 a)|Ψi

κ2e
(4.52)
= eπω1 κ1 e−πω0 + κ

†
e2 ea |Ψi.

Taking the coefficients of the right hand sides of the two results above, and then taking the
complex conjugate leads to the relation

e3 + κ4 e−πω0 = eπω1 κ∗1 e−πω0 + κ

e∗2 .


κ (4.53)

The relations (4.50) and (4.53) give linear equations for κ

e3 and κ4 that can be solved in
terms of κ1 , κ
e2 . In particular the solution is
1
κ
e3 = [κ∗ sinh(πω1 ) + κ
e∗2 sinh[π(ω0 + ω1 )]] ;
sinh(πω0 ) 1
(4.54)
1
κ4 = [κ∗ sinh[π(ω0 − ω1 )] − κ
e∗2 sinh(πω1 )] .
sinh(πω0 ) 1

These solutions are subject to further consistency checks. The first consistency check is that
we need [c, b] = 0. This leads to the requirement

κ1 κ4 − κ
e2 κ
e3 = 0. (4.55)

– 32 –
Using (4.50) and (4.53) we find, after some algebra, that

1 e−πω1
|κ1 eπω0 + κ
e2 |2 − |κ1 + κ
e2 eπω0 |2 e2πω1 .


κ1 κ4 − κ
e2 κ
e3 = (4.56)
2 coth(πω0 ) − 1

On the other hand, from the relations

1
hΨ|bb† |Ψi = 1 + hΨ|b† b|Ψi = , (4.57)
1 − e−2πω1
we find that
1 − e−2πω0
e2 eπω0 |2 = e2π(ω0 −ω1 )
|κ1 + κ , (4.58)
1 − e−2πω1
and also that
1 − e−2πω0
e2 e−πω0 |2 =
|κ1 + κ . (4.59)
1 − e−2πω1
Substituting these relations in (4.56) we find that indeed (4.55) is satisfied.
The second consistency check is that the solution must satisfy [c, c† ] = 1. This is just the
requirement that
κ3 |2 − |κ4 |2 = 1.
|e (4.60)
But this is also satisfied since
e−2πω1 
κ3 |2 − |κ4 |2 = πω0 2 πω0 2 4πω0

|e −|κ1 e + κ 2 | + |κ1 + κ 2 e | e = 1. (4.61)
(e2πω0 − 1)
e e

One might have hoped that by placing reflecting boundary conditions at the singularity,
one would correctly reproduce the coefficients κ
e3 and κ4 i.e. we quantize the field behind the
inner horizon assuming that the geometry is given the naive analytic continuation and then
also impose φ = 0 at the timelike singularity at r = 0. However, a simple calculation shows
that this does not give the right values of κ
e3 and κ4 . It is possible that more sophisticated
reflecting boundary conditions could be used to reproduce the values of κ e3 and κ4 but we
have not yet discovered them.

5 Conclusions and discussion

Summary of results. In this paper, we have developed a new test that provides a neces-
sary condition for a quantum state to be smooth in the vicinity of a null surface. The condition
is that the near-horizon modes defined by (2.12), which have canonical commutators, must
have the self-correlators given by (2.27) and must be entangled with each other through the
correlators (2.23). This is a universal result, which relies just on the short-distance properties
of the correlator of the scalar field.
When one considers static black holes, where the field can also be expanded in global
modes that are obtained as Fourier transforms of the field, then this test can be translated
into the statement that the two-point correlator of such global modes has a delta-function

– 33 –
piece in the frequency-difference, and the coefficient of this delta function is the same in any
smooth state.
We applied this test to charged black holes in anti-de Sitter space for boundary dimension,
3, 4, 5, 6, where we were able to easily verify, for a range of parameters, that the naturally
defined Hartle-Hawking state was not smooth on the inner horizon. So, even in the absence
of external perturbations, quantum fluctuations destabilize the inner horizon for such black
holes. These results can be understood to be the AdS analogue of the results of [13] for
charged flat-space black holes in four dimensions but they also demonstrate the utility of our
test.
We then considered the BTZ black hole, which was recently argued to violate strong
cosmic censorship in [12]. Here, we found, that part of our test — which pertained to the
occupation number of modes as we approach the inner horizon — was automatically satisfied
in the Hartle-Hawking state as a result of some non-trivial identities involving the various
reflection and transmission coefficients in the black-hole geometry.
However, our test has a second element: it requires the existence of modes behind the
inner horizon that are correctly entangled with modes in front. This presents an additional
complication: although, one might naively assume that one could expand the field in terms
of new degrees of freedom behind the inner horizon, this turns out to be prohibited by the
monogamy of entanglement. Instead, one is forced to reuse the degrees of freedom between
the inner and outer horizon to construct near-horizon modes behind the inner horizon. We
identified the correct combination that could be used to expand the field just behind the inner
horizon.
We clarify that this does not completely fix the dynamics behind the inner horizon. If
one considers points that are finitely separated from the inner horizon, then the correlators
of field insertions at such points require information that is not contained in the near-horizon
modes. The arbitrariness in these correlators corresponds to the freedom that one has in
extending the metric and the fields behind the Cauchy horizon.
Our results seem to support the claim that strong cosmic censorship is violated for the
BTZ black hole. If so, the arbitrariness described above would present interesting challenges
for the AdS/CFT interpretation of such black holes. The AdS/CFT conjecture suggests that a
boundary CFT describes all phenomenon in the bulk geometry but a traversable inner horizon
would imply that there are events (behind the inner horizon) that cannot be described by the
boundary CFT in an obvious way. While this would be an interesting scenario, we believe
that it is premature to conclude that strong cosmic censorship is violated for the BTZ black
hole.

Unresolved questions about strong cosmic censorship. Our reasons for believing
that strong cosmic censorship continues to be a subtle issue, at the quantum level, are as
follows. First, we emphasize that from the point of view of the exterior, strong cosmic
censorship is a statement about the infinite future, since any events at the inner horizon are

– 34 –
to the future of all events outside the outer horizon. However, at very late times, we expect
a host of ill-understood and nonperturbative phenomena in the black-hole exterior.
The first such phenomena is that, even for a fixed angular momentum, perturbations
outside the black hole do not die off exponentially for arbitrarily long times. Instead at a
S
time of O (S) where S is the black hole entropy, the perturbation reaches a size e− 2 and
develops a fat tail, where it does not decay any further [22]. What does this fat tail imply for
strong cosmic censorship? Even the classical arguments in favour of the violation of strong
cosmic censorship rely on a careful balance between the blue-shift near the inner horizon and
the decay outside. If even an exponentially small amount of energy falls into the black hole
at very late times, this could
 Sdestabilize
 the inner horizon.
At even later times O ee there are even more exotic phenomena. Since the BTZ black
hole is believed to be dual to a theory with a discrete spectrum of states, at very long times,
we expect the entire system to under Poincare recurrence. Thus, if the black hole was formed
from collapse by matter thrown in from the boundary, then Poincare recurrence implies that
that the black hole should eventually spontaneously turn into a white hole that emits matter,
which then proceeds to bounce off the boundary, and re-collapse into the black hole. What
does this imply for the inner horizon?
These nonperturbative effects are ill understood and not addressed by our effective field
theory analysis. Nevertheless, they are crucial to address the issue of strong cosmic censorship
precisely because this analysis poses peculiar questions about events in the infinite future.

Discussion. Our work points to several directions for future work. One technical issue,
which we believe would be interesting to understand better, is as follows. In the Hartle-
Hawking state, the occupancy of the modes near the inner horizon of the BTZ black hole,
which we presented in section 4, provides enough information to compute the stress-tensor
at the inner horizon using mode sums. It would be nice to understand the constraint found
in [12]— which suggests that the stress-tensor is finite only β = r∆+ > 1 — from the
r− −1
perspective of direct mode sums.
It is also natural to use our test to understand both flat-space and de Sitter black holes.
The de Sitter case is particularly interesting because, classically, not only do charged de-
Sitter black holes violate even the weak formulations of strong cosmic censorship [8], even
non rotating and neutral black holes in de Sitter space have two horizons: the black-hole and
the cosmological horizon. So an interesting exercise is to consider various possible quantum
states and examine their smoothness on all these horizon by the methods discussed in this
paper. We hope to return to this analysis in forthcoming work.
Acknowledgments: We would like to thank Jose Barbon, Jan de Boer, Chandramouli
Chowdhury, Alessandra Gnecchi, Monica Guica, Chandan Jana, R. Loganayagam, Shiraz
Minwalla, Ruchira Mishra, K. Narayan, Harvey Reall, Erik Verlinde and Amitabh Virmani
for helpful discussions. PS is grateful to TIFR (Mumbai) for hospitality while this work was
being completed. The work of SR is supported, in part, by a Swarnajayanti fellowship of the
Department of Science and Technology (India).

– 35 –
Appendix

A Scattering in the Reissner-Nordström geometry at large angular mo-

menta

In Section 3.5, we noticed that the fractional difference between the expected Boltzmann
factor and the numerical Boltzmann factor at the inner horizon for large ` modes in Reissner-
Nordström black holes tends to zero. Here, we will explain this feature by using WKB
approximation to solve wave the equation at large `. The exact form of the metric is given
in Section 3.
ds2 = −f (r)dt2 + f (r)−1 dr2 + r2 dΩ2d−1 . (A.1)
We reproduce the near horizon form of f (r).

f (r) = 2κ± (r − r± ); r → r± . (A.2)

We consider a massless scalar Klein-Gordon field φ. The wave equation can be solved with
an ansatz of the form
1
φω,` (r, t, Ω) = ψ`,ω (r∗ )e−iωt Y`m (Ω), (A.3)
r(d−1)/2
which leads to
00
ψ`,ω (r∗ ) − V (r∗ )ψ`,ω (r∗ ) = 0,
(A.4)
 
f (r) (d − 3)(d − 1) (d − 1) 0
V (r∗ ) = −ω 2 + 2 `(` + d − 2) + f (r) + rf (r) ,
r 4 2
dr
where r∗ is the tortoise coordinate, defined by dr∗ = f (r)

Large-` behaviour
At large `, the potential near outer horizon takes the following form.

2κ+ (r − r+ ) 2
V (r∗ ) ≈ −ω 2 + 2 ` . (A.5)
r+

Solving the ψ-equation, Eq. (A.4), with the above potential yields,
   
` p ` p
ψ`,ω (r∗ ) ≈ AB Kiω/κ+ f (r) + BB Iiω/κ+ f (r) . (A.6)
r+ κ+ r+ κ+

In the regime, `(r − r+ )  1 and r − r+  r+ , this solution can be approximated as

!1 √ r !1 " √ √ #
2 2
r ` f (r) ` f (r) ` f (r)
π r κ − 1 r κ −πω −
AB p+ + e r+ κ+
+ BB p+ + ie e r+ κ+
+e r+ κ+
. (A.7)
2 ` f (r) 2π ` f (r)

– 36 –
The near horizon Bessel solution can be matched with a WKB solution in the large-` limit.
For large-`, the WKB approximation is valid at all radial points where f (r) is finite (and
nonzero). R r∗ √ R r∗ √
ψwkb (r∗ ) ≈ Awkb e− V dr∗
+ Bwkb e V dr∗
. (A.8)
Near the boundary,

ψwkb (r∗ ) ≈ Awkb e−`r∗ + Bwkb e`r∗ , 1  r  `. (A.9)

Near the outer horizon,

√ √
`
X − r+ κ+ f (r) −X
`
f (r) 1
ψ(r∗ ) ≈ Awkb e e + Bwkb e e r+ κ+
,  r − r+  r+ , (A.10)
`
where,
Z ζ∗∞ p
X= V (r)dr∗ (A.11)
ζ∗+

with 1`  ζ+ − r+  r+ and 1  ζ∞  ` close to boundary. To a very good approximation,

we can write Z 0 p
f (r)
X≈l dr∗ ≡ lΛ, Λ > 0. (A.12)
−∞ r
Near the boundary, the normalizable mode behaves as
√  p 
ψ(r∗ ) ≈ N r∗ J d −ir∗ `2 − ω 2 , r  `, (A.13)
2

where N is a normalization constant. For large-`, the normalizable mode behaves as

i 1  −`r∗ iπ d 3iπ

ψ(r∗ ) ≈ N √ √ e e 8 + ielr∗ e 8 d , 1  r  `. (A.14)
` 2π
Matching this with the WKB solution Eq. (A.9), we get

Awkb iπ
= −ie− 4 d . (A.15)
Bwkb
Near horizon behaviour would now become
 √ √ 
i 1 − iπ
`
d lΛ − r+ κ+ f (r)
`
−lΛ r+ κ+ f (r)
ψ(r∗ ) ≈ N √ √ −ie 4 e e +e e . (A.16)
` 2π
`
√
f (r)
We see that the coefficient of e r+ κ+ is exponentially suppressed. Hence, in the large `
limit,
BB = 0 (A.17)
Close to horizon,  
` p
ψ`,ω (r∗ ) = AB Kiω/κ+ f (r) (A.18)
r+ κ+

– 37 –
Now we consider the near horizon (`(r − r+ )  1) limit.
"  iω #
AB ` κ+ iω iωr∗
ψ`,ω (r∗ ) = Γ(− )e + (ω → −ω)
2 2r+ κ+ κ+
!1/2   (A.19)
AB πκ+ iδω,`
iωr∗
−iδω,`
−iωr∗
= e 2 e +e 2 e ,
2 ω sinh( κπω
+
)

where
  iω !
δω,` ` κ+ iω
= arg Γ(− ) ; large-`. (A.20)
2 2r+ κ+ κ+
The mode expansion of the scalar field just outside the horizon as r → r+ is
Z
dω 1 1 X −iωt h i
φ(r, t, Ω) = √ d−1 e Y`,m (Ω)aω,` e−iδω,` e−iωr∗ + eiδω,` eiωr∗ + h.c. (A.21)
2π 2ω 2
r `,m+

Scattering between horizons for large `

The mode expansion just inside the outer horizon, as r → r+ + 0− is as follows.
Z
dω 1 1 X −iωr∗ h −iδω,` −iωt iωt
i
φ(r, t, Ω) = √ e e aω,` e Y`,m (Ω) + ã ω,` e Ȳ`,m (Ω) + h.c..
2π 2ω d−1
r 2 `,m +
(A.22)
While, just outside the inner horizon, as r → r− + 0+ ,
Z
dω 1 1 X −iωr∗ h −iωt iωt
i
φ(r, t, Ω) = √ e bω,` e Y`,m (Ω) + b̃ω,` e Ȳ`,m (Ω) + h.c. (A.23)
2π 2ω d−1 2
r `,m
−

To compute the occupation numbers of the modes at the inner horizon, we need to relate the
operators a and b. Consider a solution, Z(r), to the wave equation such that
 −i ω  
ω `ζ κ− ` p
Z(r) ≈ Γ(1 + i ) J
i κω |f (r)| ; r − r−  r− , (A.24)
κ− 2r− κ− − r− κ−

where ζ is the factor introduced in the definition of the tortoise coordinate (3.8). Close to
the inner horizon,
Z(r) = e−iωr∗ ; r → r− + 0+ (A.25)
When we move slightly away from the inner horizon such that `(r − r− )  1 but r − r−  r− ,

−i ω !1
 2
ω `ζ κ− r− κ−
Z(r) ≈Γ(1 + i ) p
κ− 2r− κ− 2π` |f (r)| (A.26)
 πω

`
√   √ 
i r κ |f (r)| iπ − πω −i r `κ |f (r)|
− iπ
× e 4e 2κ − e − − +e4 e 2κ − e − − .

– 38 –
Away from the horizon, we can also use the large-` WKB approximation.
1 h R r∗ √ R r∗ √ i
Z(r) ≈ A+ ei −∞ V dr∗ + B+ e−i −∞ V dr∗
V 1/4 (A.27)
1 h R r∗ √ R r∗ √ i
Z(r) ≈ 1/4 A− ei ∞ V dr∗ + B− e−i ∞ V dr∗ ,
V
with,
R∞ √ R∞ √
A− = A+ ei −∞ V dr∗
= A+ eiθ B− = B+ e−i −∞ V dr∗
= B+ e−iθ . (A.28)

Matching the WKB solution with the Bessel expansion near the inner horizon, we get
 −i ω r
ω `ζ κ− κ− iπ − 2κ
πω
A− = Γ(1 + i ) e 2 e −;
κ− 2r− κ− 2π
 −i ω r (A.29)
ω `ζ κ− κ− 2κ
πω
B− = Γ(1 + i ) e −.
κ− 2r− κ− 2π

Close to the outer horizon, `(r+ − r)  1 but r+ − r  r+ , this solution can also be written
as
s πω   
2π e 2κ+ i π4 −i π4 κ+
πω ` p
Z(r) ≈ (A+ e − B+ e e )J−i κω |f (r)|
κ+ e 2πω
κ+
−1
+ r+ κ+
  (A.30)
π π πω `
+(B+ e−i 4 − A+ ei 4 e κ+ )Ji κω
p
|f (r)| .
+ r+ κ+

At the outer horizon, we can expand the Bessel function and substitute for A+ and B+
to get,
Z(r) = Aω,` e−iωr∗ + Bω,` eiωr∗ ; r → r+ − 0+ , (A.31)
where, for large-`,
h − πω π πω π πω i δω,l
Aω,` = Nω,` e 2κ− ei( 2 −θ) − e 2κ− e−i( 2 −θ) e κ+ e−i 2 ;
h − πω π πω πω π
i δω,l
Bω,` = Nω,` e 2κ− ei( 2 −θ) e κ+ − e 2κ− e−i( 2 −θ) ei 2 ;
v πω
ω
0 u sinh(π κ+ )
u
iδω,` e 2κ+
Nω,` = e t ; (A.32)
sinh(π κω− ) e 2πω
κ+
−1
v  
−i ω u ω
0 i3π

`ζ κ− u
u Γ 1 + i κ−
eiδω,` = e 4 t  .
2r− κ− Γ 1−i ω κ−

We can check that the Bogoliubov coefficients satisfy,

|Aω,` |2 − |Bω,` |2 = 1. (A.33)

– 39 –
Using the mode expansion near the horizons,

bω,` = e−iδω,` A∗ω,` aω,` − Bω,`

∗
ã†ω,` . (A.34)

Using the constraints on operators a and ã, Eq. (3.27), and large-` Bogoliubov coefficients,
Eq. (A.32), we get the Boltzmann factor at inner horizon.
1
hbω,` b†ω0 ,`0 i = δ(ω − ω 0 )δ`,`0 `  1. (A.35)
− 2πω
1−e κ−

References

[1] R. Penrose, Structure of space-time, in Battelle Rencontres: 1967 lectures in mathematics and
physics (C. de Witt and J. Wheeler, eds.). New York: Benjamin, 1968.

[2] M. Simpson and R. Penrose, Internal instability in a Reissner-Nordstrom black hole, Int. J.
Theor. Phys. 7 (1973) 183–197. J. M. McNamara, Instability of black hole inner horizons,
Proceedings of the Royal Society of London. A. Mathematical and Physical Sciences 358 (1978)
499–517. S. Chandrasekhar and J. B. Hartle, On crossing the cauchy horizon of a
reissner–nordström black-hole, Proceedings of the Royal Society of London. A. Mathematical
and Physical Sciences 384 (1982) 301–315. E. Poisson and W. Israel, Inner-horizon instability
and mass inflation in black holes, Phys. Rev. Lett. 63 (1989) 1663–1666. E. Poisson and
W. Israel, Internal structure of black holes, Phys. Rev. D41 (1990) 1796–1809.

[3] A. Ori, Inner structure of a charged black hole: An exact mass-inflation solution, Phys. Rev.
Lett. 67 (1991) 789–792. A. Ori, Structure of the singularity inside a realistic rotating black
hole, Phys. Rev. Lett. 68 (1992) 2117–2120. F. Mellor and I. Moss, Stability of Black Holes in
De Sitter Space, Phys. Rev. D41 (1990) 403. C. M. Chambers and I. G. Moss, Stability of the
Cauchy horizon in Kerr-de Sitter space-times, Class. Quant. Grav. 11 (1994) 1035–1054,
[gr-qc/9404015]. P. R. Brady, C. M. Chambers, W. Krivan and P. Laguna, Telling tails in the
presence of a cosmological constant, Phys. Rev. D55 (1997) 7538–7545, [gr-qc/9611056].
M. Dafermos, The Interior of charged black holes and the problem of uniqueness in general
relativity, Commun. Pure Appl. Math. 58 (2005) 0445–0504, [gr-qc/0307013]. M. Dafermos,
Stability and instability of the Reissner-Nordstrom Cauchy horizon and the problem of
uniqueness in general relativity, Contemp. Math. 350 (2004) 99–113, [gr-qc/0209052].
K. Murata, H. S. Reall and N. Tanahashi, What happens at the horizon(s) of an extreme black
hole?, Class. Quant. Grav. 30 (2013) 235007, [1307.6800]. D. Christodoulou, The formation of
black holes in general relativity, in The Twelfth Marcel Grossmann Meeting: On Recent
Developments in Theoretical and Experimental General Relativity, Astrophysics and Relativistic
Field Theories (In 3 Volumes), pp. 24–34, World Scientific, 2012. S. Bhattacharjee, S. Sarkar
and A. Virmani, Internal Structure of Charged AdS Black Holes, Phys. Rev. D93 (2016)
124029, [1604.03730].

[4] M. Dafermos, Black holes without spacelike singularities, Commun. Math. Phys. 332 (2014)
729–757, [1201.1797]. M. Dafermos and J. Luk, The interior of dynamical vacuum black holes
I: The C 0 -stability of the Kerr Cauchy horizon, 1710.01722.

– 40 –
 [5] M. Dafermos and Y. Shlapentokh-Rothman, Rough initial data and the strength of the blue-shift
instability on cosmological black holes with λ¿ 0, Classical and Quantum Gravity 35 (2018)
195010.

[6] V. Cardoso, J. L. Costa, K. Destounis, P. Hintz and A. Jansen, Quasinormal modes and Strong
Cosmic Censorship, Phys. Rev. Lett. 120 (2018) 031103, [1711.10502].

[7] S. Hod, Strong cosmic censorship in charged black-hole spacetimes: As strong as ever, Nucl.
Phys. B941 (2019) 636–645, [1801.07261]. O. J. C. Dias, H. S. Reall and J. E. Santos, Strong
cosmic censorship for charged de Sitter black holes with a charged scalar field, Class. Quant.
Grav. 36 (2019) 045005, [1808.04832]. Y. Mo, Y. Tian, B. Wang, H. Zhang and Z. Zhong,
Strong cosmic censorship for the massless charged scalar field in the Reissner-Nordstromde
Sitter spacetime, Phys. Rev. D98 (2018) 124025, [1808.03635]. S. Hod, Quasinormal modes
and strong cosmic censorship in near-extremal KerrNewmande Sitter black-hole spacetimes,
Phys. Lett. B780 (2018) 221–226, [1803.05443]. J. Luk and S.-J. Oh, Strong cosmic censorship
in spherical symmetry for two-ended asymptotically flat initial data I. The interior of the black
hole region, 1702.05715. R. Luna, M. Zilho, V. Cardoso, J. L. Costa and J. Natrio, Strong
Cosmic Censorship: the nonlinear story, Phys. Rev. D99 (2019) 064014, [1810.00886].

[8] O. J. C. Dias, F. C. Eperon, H. S. Reall and J. E. Santos, Strong cosmic censorship in de Sitter
space, Phys. Rev. D97 (2018) 104060, [1801.09694].

[9] O. J. C. Dias, H. S. Reall and J. E. Santos, Strong cosmic censorship: taking the rough with the
smooth, JHEP 10 (2018) 001, [1808.02895].

[10] W. A. Hiscock, Stress-energy tensor near a charged, rotating, evaporating black hole, Physical
Review D 15 (1977) 3054.

[11] N. Birrell and P. Davies, On falling through a black hole into another universe, Nature 272
(1978) 35.

[12] O. J. C. Dias, H. S. Reall and J. E. Santos, The BTZ black hole violates strong cosmic
censorship, 1906.08265.

[13] O. Sela, Quantum effects near the Cauchy horizon of a Reissner-Nordstrm black hole, Phys.
Rev. D98 (2018) 024025, [1803.06747].

[14] A. R. Steif, The Quantum stress tensor in the three-dimensional black hole, Phys. Rev. D49
(1994) 585–589, [gr-qc/9308032].

[15] R. M. Wald, Quantum field theory in curved spacetime and black hole thermodynamics.
University of Chicago Press, 1994.

[16] R. Haag, Local quantum physics: Fields, particles, algebras, 2nd ed. Springer, 1992.

[17] N. Birrell and P. Davies, Quantum fields in curved space. Cambridge Univ Press, 1986.

[18] K. Papadodimas and S. Raju, Black Hole Interior in the Holographic Correspondence and the
Information Paradox, Phys. Rev. Lett. 112 (2014) 051301, [1310.6334].

– 41 –
[19] K. Papadodimas and S. Raju, State-Dependent Bulk-Boundary Maps and Black Hole
Complementarity, Phys. Rev. D89 (2014) 086010, [1310.6335].

[20] S. D. Mathur, What the information paradox is not, 1108.0302. A. Almheiri, D. Marolf,
J. Polchinski and J. Sully, Black Holes: Complementarity or Firewalls?, JHEP 02 (2013) 062,
[1207.3123].

[21] J. M. Maldacena, The Large N limit of superconformal field theories and supergravity, Int. J.
Theor. Phys. 38 (1999) 1113–1133, [hep-th/9711200].

[22] J. M. Maldacena, Eternal black holes in anti-de Sitter, JHEP 0304 (2003) 021,
[hep-th/0106112].

[23] S. Raju, A Toy Model of the Information Paradox in Empty Space, 1809.10154.

[24] K. Papadodimas and S. Raju, Remarks on the necessity and implications of state-dependence in
the black hole interior, Phys. Rev. D93 (2016) 084049, [1503.08825].

[25] A. Chamblin, R. Emparan, C. V. Johnson and R. C. Myers, Charged AdS black holes and
catastrophic holography, Phys. Rev. D60 (1999) 064018, [hep-th/9902170].

[26] E. Hairer, S. Nørsett and G. Wanner, Solving ordinary differential equations I: nonstiff
problems. Springer-Verlag, New York, 1993.

[27] C. K. Raju, CALCODE: An ODE Solver, 1999.

[28] M. Galassi, J. Davies, J. Theiler, B. Gough, G. Jungman, M. Booth et al., GNU Scientific
Library Reference Manual (Network Theory Ltd.), 2009.

[29] O. Tange, Gnu parallel - the command-line power tool, ;login: The USENIX Magazine 36 (Feb,
2011) 42–47.

[30] A. Lanir, A. Ori, N. Zilberman, O. Sela, A. Maline and A. Levi, Analysis of quantum effects
inside spherical charged black holes, Phys. Rev. D99 (2019) 061502, [1811.03672].

[31] V. Balasubramanian and T. S. Levi, Beyond the veil: Inner horizon instability and holography,
Phys.Rev. D70 (2004) 106005, [hep-th/0405048]. E. Keski-Vakkuri, Bulk and boundary
dynamics in BTZ black holes, Phys. Rev. D59 (1999) 104001, [hep-th/9808037].

[32] P. Candelas, Vacuum Polarization in Schwarzschild Space-Time, Phys. Rev. D21 (1980)
2185–2202.

– 42 –