A Classical Firewall Transformation as a Canonical

Transformation

Nathaniel A. Strauss, Bernard F. Whiting

Department of Physics, University of Florida, Gainesville, Florida, USA
arXiv:2309.09891v1 [gr-qc] 18 Sep 2023

E-mail: straussn@ufl.edu, bernard@phys.ufl.edu

Abstract. The firewall transformation put forward by ’t Hooft in recent years

has made ambitious claims of solving the firewall problem and the black hole
information paradox while maintaining unitary evolution. However, the theory
has received limited attention from the community, especially in regards to its
foundations in purely classical gravitational physics. This paper investigates the
underlying assumptions of ’t Hooft’s firewall transformation before quantization.
We find that the limiting procedure used by ’t Hooft in order to obtain an
identification of the quantum operators for ingoing and outgoing particles near
a black hole is not consistent. We propose a correction, which involves a more
relaxed approximation regime. In the new approximation regime, we find a new
classical analog for the firewall transformation for spherical shells, which allows
evolving the spherical shells’ dynamics past their point of collision. In the classical
theory, no firewall is removed, as both ingoing and outgoing matter is present on
every spacelike hypersurface, and it does not appear that any firewalls will be
removed after a canonical quantization.

Keywords: general relativity, black hole, quantum mechanics, firewall, information

A Classical Firewall Transformation as a Canonical Transformation 2

1. Introduction

The behavior of matter near black holes is nontrivial in both quantum and classical
physics. In this paper, we investigate two non-trivialities and how they interact: (1)
particle creation near the black hole event horizon as predicted by quantum field
theory and (2) the gravitational interaction of particles in a black hole background
as predicted by general relativity. We start by discussing particle creation near
black holes, known as Hawking radiation, which stems from a fusion of general
relativity and quantum field theory. In quantum field theory, it can be shown that
differently accelerating observers will disagree on the nature, or even existence of
observed particles. [1] In other words, there are many possible vacuum states, which
are tailored to different observational frames. [2, 3] Applying the idea to a black hole
background, a freely falling observer and an observer hovering just outside the horizon
are accelerating with respect to one another, so if the freely falling observer does
not detect particles, the observer hovering above the horizon will detect a thermal
distribution of particles, which is known as Hawking radiation. [4] There already
existed a problem of information loss in classical mechanics as matter that falls beyond
the event horizon looses its identity. From a quantum mechanical perspective, the
existence of Hawking radiation severely exacerbates the issue. This happens because
the entanglement between the black hole and the Hawking radiation results in the
initial pure state evolving into a mixed state. Furthermore, because Hawking radiation
eventually causes the black hole to completely evaporate, the matter entangled to the
Hawking radiation will eventually cease to exist. [5] One resolution to the problem is
to use unitary time evolution to include the Hawking radiation in the far past of the
spacetime. However, in order for the Hawking radiation to escape the black hole, it
must have had very high energy near the event horizon in the far past, leading to a
divergent stress energy on the horizon. [5, 6] This divergent stress energy is known as
the firewall problem, and is not generally considered to be a real physical phenomenon.
Rather, most physicists believe the firewall will disappear in a complete theory. [5, 6]
A proposal to resolve the firewall problem that preserves unitarity time evolution
has been put forward by ’t Hooft [7–12], and has continued to be developed in recent
years with some engagement from the community in the context of searching for the
quantum microstates of black holes (e.g. [13–17]). ’t Hooft’s proposal is intended to
resolve the firewall problem and information problem by including, into the quantum
theory, the gravitational interaction of particles entering and leaving the black hole.
The basic idea of ’t Hooft’s framework is that as the outgoing Hawking particles
interact with ingoing matter, they leave informational “footprints” on each other,
from which one can recover information that is lost in the standard theory. One
allows firewalls to appear in the quantum theory, but one can also transform away
any firewall that appears using a so-called “firewall transformation”. [8, 9] ’t Hooft’s
model only considers radially moving particles for simplicity and splits the particles
into two categories: ingoing and outgoing. As high energy particles pass by each
other near the event horizon, they interact gravitationally, causing a Shapiro time
delay and an exchange of energy. Normally the Shapiro time delay is thought of as
an effect on particles scattering off a large massive body, and the time delay is the
difference between the times it takes for the particle to reach some distant point with
and without passing by the massive object. In this context, however, the time delay
is caused by ingoing and outgoing particles passing by each other, not by the black
hole. For example, in the context of this paper, the Shapiro time delay for an outgoing
A Classical Firewall Transformation as a Canonical Transformation 3

particle is the difference between the times it takes for the particle to reach a point
far from the black hole with and without passing by one or more ingoing particles.
’t Hooft suggests that including both the time delay caused by the high energy
ingoing and outgoing particles and also keeping the high-energy particles in the
theory would be double-counting the effect of the high-energy particles, so we can
formally ignore the high-energy particles after including their gravitational effect
on the surrounding particles. [8] As a consequence of this approach, the firewall
transformation can be written in the following form for radially moving particles: [10]
X
ûin (Ωj ) = f (Ωj , Ωi )p̂out (Ωi ), (1)
i
X
ûout (Ωj ) = f (Ωj , Ωi )p̂in (Ωi ), (2)
i

where ûin/out and p̂in/out represent an ingoing or outgoing particle’s quantum position
and momentum operators (in Kruskal coordinates), respectively, Ωj and Ωi represent
the angular position of the particles relative to the black hole, and f (Ωj , Ωi ) gives the
shift in particle j’s Kruskal position operator due to particle i. For example, a firewall
on the future event horizon caused by the divergent momentum of ingoing particles can
be removed by including the Shapiro time delay on the Kruskal-coordinate position
of outgoing Hawking radiation.
In the authors’ view, it is unclear whether the quantum firewall transformation
given by (1) or (2) transforms away firewalls physically or merely formally. In other
words, it is possible that the firewall physically exists in the theory even after it
formally disappears from the notation. In order to elucidate the issue, and to orient
us toward quantization, we take a close look at the classical analogs of (1) and (2) in a
Hamiltonian framework. In order to follow ’t Hooft’s procedure as closely as possible,
we specialize the Hamiltonian framework to Kruskal coordinates. For simplicity, our
toy model will consist of two spherical shells of null matter, one ingoing and one
outgoing.
The structure of the paper is as follows. In Section 2, we summarize the Shapiro
time delay effect for two spherical shells of null matter passing by each other in a
Schwarzschild background. In Section 3, we precisely describe the approximations
necessary to obtain the classical analogs of (1) and (2), even though the outcome
is difficult to reconcile with ’t Hooft’s procedure. In Section 4, we introduce the
canonical theory for the two shells of null matter. In Section 5, we show that we
can account for the time delay via a canonical transformation when we measure the
time delay between the past of both shells and the future of both shells. Finally, in
Section 6, we summarize our results and suggest further work is required in order
to better understand to what degree ’t Hooft’s quantum firewall transformation may
solve the firewall problem.

2. The Shapiro Time Delay and the Firewall Transformation

The firewall transformation ’t Hooft has proposed is motivated by including the

gravitational shockwave of ingoing and outgoing particles passing by each other, which
has been considered a Shapiro time delay effect. [11, 12] In this section we will give an
overview of the general relativity needed to understand the gravitational shockwave in
a Schwarzschild background in the simplified case of radially moving spherical shells.
A Classical Firewall Transformation as a Canonical Transformation 4

We make use of results derived via the ADM Hamiltonian formalism, which is oriented
towards canonical quantization. [18] The simplification to spherical shells removes all
angular dependence to the resulting shockwave that point particles would create, but
this angular dependence does not affect the larger context of the degree to which ’t
Hooft’s firewall transformation solves the firewall problem.
We begin with the Schwarzschild spacetime and insert two radially moving,
spherical shells, as shown in Figure 1. In each of the four regions separated by the
shells, the spacetime is given by the metric equation
 2Mi  2  2Mi −1 2
ds2 = − 1 − dTi + 1 − dR + R2 dΩ2 (3)
R R
in ordinary Schwarzschild coordinates, where

dΩ2 = dθ2 + sin2 θdϕ2 , (4)

and Mi is the Schwarzschild mass in region i of Figure 1. The metric can also be
written in the Kruskal coordinates

ds2 = 2gUi Vi dUi dVi + R2 dΩ2 , (5)

where
1 − 2Mi /R 16Mi3 −R/2Mi
gUi Vi = 8Mi2 = e . (6)
Ui Vi R
From this point forward, we suppress the indices i when the expression applies to each
region of Figure 1 separately. For example, the Kruskal coordinates are related to the
Schwarzschild coordinates implicitly by
 R 
UV = − 1 eR/2M , (7)
2M
 R 
V /U = sign − 1 eT /2M (8)
2M
in each region separately.
Null particles traveling radially travel along lines of constant U if they are
outgoing and along lines of constant V if they are ingoing, as will be shown in Section 4.
It has long been known that the gravitational effect of a radially moving particle is an
angular-dependent shift in one of the Kruskal coordinates U or V when matching the
Kruskal coordinates across the trajectory of the particle. [19] In the case of a spherical
shell of particles, the shift in the Kruskal coordinate of the shell depends only on the
radius at the point of intersection, being angularly independent, and can be found
via the definitions (7) and (8). Let the total conserved energies of the ingoing and
outgoing shells be

Ein = M1 − M4 , (9)
Eout = M4 − M3 (10)

as measured in region 4 and

Ẽin = M2 − M3 , (11)
Ẽout = M1 − M2 (12)
A Classical Firewall Transformation as a Canonical Transformation 5

0
=
V3

2
U
=
0
3 1

0
=
3
U

V4
=
4

0
Figure 1: A Penrose diagram for a Schwarzschild black hole spacetime including two
spherical shells of null matter, one ingoing and one outgoing. The ingoing shell is the
dotted line and the outgoing shell is the dashed line. Lines of constant U run from
bottom left to top right and lines of constant V run from bottom right to top left,
both at an angle of 45 degrees. The diagram is partitioned into four labeled regions,
separated by the two shells, which collide exterior to the event horizon. Note that
region 1 lies entirely within the exterior of the black hole, while regions 2, 3, and 4
each contain part of the exterior and part of the interior. Each of the four regions is
installed with its own set of Kruskal coordinates (Ui , Vi ) and Schwarzschild mass Mi ,
and the visible coordinate axes are labeled in each region. Because the event horizon
of the black hole moves when matter crosses it, the U coordinate shifts when crossing
the ingoing shell and the V coordinate shifts when crossing the outgoing shell.

as measured in region 2. Defined this way, all E’s are positive.

By direct calculation using (7), we can find the matching conditions for the
Kruskal coordinates across the shells to first order in the energies:

eR/2M4 R2 Ein
U1,in − U4,in ≈ − , (13)
4M43 V4,in
eR/2M3 R2 Ẽin
U2,in − U3,in ≈ − , (14)
4M33 V3,in
eR/2M2 R2 Ẽout
V1,out − V2,out ≈ − , (15)
4M23 U2,out
eR/2M3 R2 Eout
V4,out − V3,out ≈ − ,, (16)
4M33 U3,out

where Ui,in/out and Vi,in/out are the Kruskal coordinates evaluated along the ingoing
or outgoing shell, respectively, as measured in region i. Note that the shift in the
Kruskal coordinate depends on the radius at which one matches the coordinates on
each side of the shell. For example, a line of constant U in region 4 will match to a
different constant value of U in region 1, depending on the radius at which that line
impacts the ingoing shell, which travels along a line of constant V . Also, we make a
A Classical Firewall Transformation as a Canonical Transformation 6

gauge choice in our Schwarzschild time coordinate such that the Kruskal coordinate
along the trajectory of each shell does not change when crossing that shell:

U1,out = U2,out , (17)

U4,out = U3,out , (18)
V1,in = V4,in , (19)
V2,in = V3,in . (20)

This above four gauge choices are not simultaneously consistent in the general theory,
but are consistent to first order in the energies, the approximation regime in which
we work. The result is a change in the Schwarzschild time coordinate (and not the
Schwarzschild radial coordinate) when crossing the shell, which we call a Shapiro time
delay.
When matching the coordinates across all four regions in Figure 1, a consistency
condition emerges called the Dray-’t Hooft-Redmount formula [19, 20], which can be
written

(R0 − 2M1 )(R0 − 2M3 ) = (R0 − 2M2 )(R0 − 2M4 ), (21)

where R0 is defined by
 R 
0
Ui,out Vi,in = − 1 eR0 /2Mi (22)
2Mi
and represents the Schwarzschild radial coordinate where the two shells meet, and is
assumed to lie in the exterior of all four Schwarzschild regions. By rearranging (21),
we can show separately
Ein Eout
Ẽin = Ein + , (23)
R0 − 2M4
Ein Eout
Ẽout = Eout − . (24)
R0 − 2M4
We can thus think of the Dray-’t Hooft-Redmount formula as describing an exchange
of energy between the two shells at the collision. Also, the conserved energy of the
shells on the interior of the shells is larger because the Schwarzshchild mass is smaller
there, which, roughly speaking, increases the shells’ gravitational potential energy. We
see that the outgoing shell loses energy after the shells meet, and the ingoing shell
gains the same amount of energy after the shells meet. Nevertheless, this exchange
of energy is second order in Ein and Eout , so we neglect it for the remainder of the
paper (see Section 3). Note however that, unlike (13) through (16), the results (21)
through (24) are all actually exact.
We have made a gauge choice in our Schwarzschild time coordinate such that the
Kruskal coordinate along the trajectory of each shell does not change when crossing
that shell. However, that Kruskal coordinate of each shell will change when the two
shells meet. Also, the shift in the Kruskal coordinate approaches a finite value at the
horizon of the black hole, corresponding to the shift in location of the event horizon
due to the energy of the shell. In this paper we focus on the situation of the two
shells colliding very close to the Schwarzschild horizon, i.e. Ui,out Vi,out ≪ 1 and
A Classical Firewall Transformation as a Canonical Transformation 7

r0 /2M − 1 ≪ 1. In this regime we can approximate the shifts in the constant Kruskal
coordinates of the shells from (13) through (16) and (17) through (20):

eEin
U1,out − U3,out ≈ − , (25)
M3 V3,in
eEin
U2,out − U4,out ≈− , (26)
M4 V4,in
eẼout
V1,in − V3,in ≈ − , (27)
M3 U3,out
eẼout
V2,in − V4,in ≈ . (28)
M4 U4,out

These equations give the shifts in the constant Kruskal coordinate for each shell when
the shells meet near the event horizon. After the shells meet, they continue to move
along their new constant Kruskal coordinates. Because we are working to first order
in Ein , Eout , Ui,in , and Vi,out , the right hand side of (25) through (28) will be the same
on both sides of the shell, up to higher order terms (see Section 3). At this point,
we’ve found that the change in Kruskal coordinate of the outgoing shell is proportional
to the energy of the ingoing shell and vice versa, which forms the basis of a classical
understanding for the firewall transformation.

3. The approximation scheme

In obtaining the classical expressions (25) through (28), which form the basis for the
firewall transformation, we have made several approximations, which we will make
more precise here. Firstly, we assume that the two shells meet near the event horizon
of the black hole:

Uout ∼ εout , (29)

Vin ∼ εin , (30)
R0
|Uout Vin | = − 1 eR0 /2M ∼ εout εin , (31)
2M
where the tilde signifies “is of the order” and εout , εin ≪ 1. We also neglect the change
in energies of the shells across the shells given by (23) and (24), imposing

2Eout
∼ δout , (32)
R0 − 2M
2Ein
∼ δin , (33)
R0 − 2M
where δin/out ≪ 1. The above expressions imply

Ein ∼ δin εin εout , (34)

Eout ∼ δout εin εout , (35)
2Ein Eout
Ẽin − Ein = −(Ẽout − Eout ) = ∼ δin δout εin εout . (36)
R0 − 2M
A Classical Firewall Transformation as a Canonical Transformation 8

Therefore, we neglect terms of order δin δout εin εout . Now, using the notation of this
section, we can show precisely to what order (25) through (28) are accurate. For
example,

eR0 /2M4 R02 Ein 2

U1,out − U3,out = − + O(δin εin ε2out ) (37)
4M 3 V3,in
eEin
=− + O(δin εin ε2out ), (38)
M V3,in
eR0 /2M2 R02 Eout 2
V1,in − V3,in = − + O(δout ε2in εout ) (39)
4M 3 U3,out
eEout
=− + O(δout ε2in εout ). (40)
M U3,out

Meanwhile, we can calculate directly

eEin
∼ δin εout , (41)
M V3,in
eEout
∼ δout εin . (42)
M U3,out

Therefore, in order to obtain (25) through (28), we work to third order in the regime
εin , δin , εout , δout ≪ 1.
We now compare (25) through (28) to ’t Hooft’s firewall transformation in (1)
and (2). In ’t Hooft’s expressions, the initial Kruskal coordinate of each particle is
taken to zero, i.e. the simultaneous limits εin → 0 and εout → 0. We see here that
taking both limits also results in the change in both Kruskal coordinates going to zero,
i.e. δin εout → 0 and δout εin → 0. Thus, it is not consistent to take both initial Kruskal
coordinates to zero while also keeping a non-zero shift in both Kruskal coordinates
after crossing the shell. However, there is no problem taking exactly one of the limits
εin → 0 or εout → 0, while keeping a non-zero shift in the appropriate coordinate. For
example, taking the limit εin → 0 means taking the simultaneous limits V3,in → 0 (seen
from (30)) and Ein → 0 (seen from (34)). We can take this limit such that eEin /M V3,in
remains of order δin εout , as given by (41). The result is V1,in = V3,in = 0, whereas
U1,out − U3,out ∼ δin εout . To prevent this issue from obscuring any further results, we
refrain from taking either of the limits εin → 0 or εout → 0 for the remainder of the
paper.

4. The Canonical Formalism for Spherical Shells in a Schwarzschild

Background

We begin with the Hamiltonian formulation for a single shell. In [18], Louko et al.
investigated the effect of a single spherical shell of null matter in a Schwarzschild
background in great detail in the ADM formalism. In Appendix A we apply their
results to the Kruskal coordinates, and find two expressions for the Hamiltonian
and canonical momentum, with Schwarzschild time as the foliation time coodinate
and one of the two Kruskal coordinates as the foliation radial coordinate (see (A.15)
A Classical Firewall Transformation as a Canonical Transformation 9

and (A.16)):
1 ηU εU − 1
HU = pU U, (43)
2M 2
1 ηV εV + 1
HV = pV V, (44)
2M 2
where

εU = sign U, (45)
εV = sign V, (46)
ηU = sign pU , (47)
ηV = sign pV . (48)

Here, pV is the momentum conjugate to the Kruskal coordinate V and pU is the

momentum conjugate to the Kruskal coordinate U .
The two Hamiltonians are related by a canonical coordinate transformation
generated by a type two generating function. For example, when transforming from
V to U , the generating function

G2 = sign(R − 2M )e−T /2M pU V (49)

results in the transformation equations

 R 
U = sign − 1 e−T /2M V, (50)
2M
 R 
pU = sign − 1 eT /2M pV , (51)
2M
which is consistent with (8). Furthermore, in transforming from V to U using G2 as
the generating function, the Hamiltonian changes from HV to HU .
Via canonical coordinate transformations between the canonical variables (U, pU )
and (V, pV ), we can obtain a meaningful Hamiltonian and meaningful equations of
motion using either Kruskal coordinate, regardless of whether the shell is ingoing or
outgoing. In the remainder of this paper, we are only concerned with the dynamics
in the exterior of the Schwarzschild horizon, so we take εU = εV = 1. We also
wish to clearly distinguish the ingoing shell from the outgoing shell, regardless of
canonical variables, either (U, pU ) or (V, pV ), we use to describe that shell. We use
the subscript “in” to signify the quantity for the ingoing shell, where ηU = ηV = −1.
Likewise, we use the subscript “out” to signify the quantity for the outgoing shell,
where ηU = ηV = +1. For example, the Hamiltonians are
1
HU in = − pU in Uin , (52)
2M
HV in = 0, (53)
HU out = 0, (54)
1
HV out = pV out Vout , (55)
2M
where for each shell, we use either HU or HV , and we can switch between them via
a canonical coordinate transformation. Notice in one set of variables for each shell
A Classical Firewall Transformation as a Canonical Transformation 10

the Hamiltonian is zero, which means in those variables both the canonical coordinate
and the momentum are constants of motion. This justifies our earlier assertion that
ingoing shells travel along constant V and outgoing shells travel along constant U .
By coupling the action for the shell to the gravitational action, Louko et al.
found a relationship between shell’s momenta and the radial derivative of the foliation
coordinates. [18] Applying those results to the Kruskal coordinates, we find the
following equations of motion for the momenta:
4M Ein T /2M
pU in = − e , (56)
Vin
4M Ein
pV in =− , (57)
Vin
4M Eout
pU out = , (58)
Uout
4M Eout −T /2M
pV out = e . (59)
Uout
Expressions (56) through (59) are defined in each region of Figure 1 separately,
though the change in each symbol on the right hand side is at most second order
in δin or δout , so in our approximation regime, we can assume the momenta of the
shells are unambiguous all along each shell, and at the point of intersection. We
have a physically meaningful canonical momentum in both sets of Kruskal variables,
regardless of whether the particle is ingoing or outgoing, and regardless of in which
region of Figure 1 we observe the shell.
Returning to the case when two intersecting shells are present throughout the
spacetime, we can now rewrite (25) through (28) in terms of the canonical variables:
e
U1,out − U3,out = p3,V in , (60)
4M32
e
V1,in − V3,in =− p3,U out , (61)
4M32
or equivalently
e
U2,out − U4,out = p4,V in , (62)
4M42
e
V2,in − V4,in = p4,U out . (63)
4M42
In one set the two compared regions are spacelike separated and in the other they are
timelike separated. In order to streamline the notation and concisely summarize the
matching of the Kruskal coordinates of the shells in the different regions, we write
′ e
Uout − Uout = σout pV in , (64)
4M 2
e
Vin′ − Vin = σin pU out , (65)
4M 2
where the values of σout and σin are given in Table 1, depending on the regions in
′
which Uout , Uout , Vin , and Vin′ are defined. We consider these two equations to be our
precise classical analogs of ’t Hooft’s firewall transformation equations, (1) and (2).
The goal for the remainder of the paper is to include (64) and (65) in the canonical
theory, oriented towards canonical quantization.
A Classical Firewall Transformation as a Canonical Transformation 11

σout σin
′
Uout region Vin′ region
Uout region 1 2 3 4 1 2 3 4
1 0 0 −1 −1 1 0 1 1 0

Vin region
2 0 0 −1 −1 2 −1 0 0 −1
3 1 1 0 0 3 −1 0 0 −1
4 1 1 0 0 4 0 1 1 0
(a) (b)

Table 1: These two tables give the values of σout and σin in (64) and (65) depending
′
on the regions in which Uout , Uout , Vin , and Vin′ are defined. Table 1a gives the value
of σout and Table 1b gives the value of σin .

5. The Firewall Transformation as a Canonical transformation

The firewall transformation as written in (1) and (2) is a simple change of basis in
the Hilbert space for the wavefunction describing the quantum particles. Perhaps the
most straightforward way to obtain ’t Hooft’s firewall transformation after canonical
quantization is to demand the relations (64) and (65) hold as canonical coordinate
transformations. Consider a canonical Hamiltonian containing one ingoing shell and
one outgoing shell:
pV out Vout pU in Uin
Hc = − , (66)
2M 2M
where the Hamiltonian is defined in each region of Figure 1 separately. In order to use
our expressions for the firewall transformation (64) and (65), we need to first perform
a canonical coordinate transformation to the canonical variables from (Uin , pU in ) and
(Vout , pV out ) to (Uout , pU out ) and (Vin , pV in ). This renders the Hamiltonian equal to
zero, since the canonical coordinates and momenta are constants of the motion, but
they are still physically meaningful via (56) through (59). Suppose now we transform
to new canonical variables
′ e
Uout (Uout , pV in ) = Uout + σout pV in , (67)
4M 2
e
Vin′ (Vin , pU out ) = Vin + σin pU out , (68)
4M 2
p′U out = pU out , (69)
p′V in = pV in , (70)
which are consistent with (64) and (65). We can check if the transformation is
canonical by calculating the Poisson brackets:
′
{Uout , p′U out } = 1, (71)
{Vin′ , p′V in } = 1, (72)
′
{Uout , p′V in } = 0, (73)
{Vin′ , p′U out } = 0, (74)
{p′U out , p′V in } = 0, (75)
′ e
{Uout , Vin′ } = (σin − σout ), (76)
4M 2
A Classical Firewall Transformation as a Canonical Transformation 12

where {, } signifies the Poisson bracket in the old variables (Uout , pU out ) and (Vin , pV in ).
Thus, the transformation is only canonical when σout = σin .
By examining Table 1, we find two significant cases when σout = σin : (1) the
primed variables are defined in region 2 and the unprimed variables are defined in
region 4 and (2) the unprimed variables for the ingoing shell and the primed variables
for the outgoing shell are defined in region 1, and the primed variables for the ingoing
shell and the unprimed variables for the outgoing shell are defined in region 3. In both
cases, the unprimed variables describe the shells before the collision, and the primed
variables describe the shells after the collision. The other cases for which σout = σin
involve either switching the primed and unprimed regions in the above two cases or
the trivial cases where the primed and unprimed regions are identical. Thus, we have
found that we can interpret (64) and (65) as a canonical transformation between the
canonical variables in the past of the point of collision and the future of the point of
collision.
Thus, we see that the canonical transformations identified in cases (1) and
(2) above define how to evolve the two null shells past the collision point without
introducing the gravitational degrees of freedom into the canonical theory. The fact
that the transformation is canonical only when evolving between the past and future
of the shells suggests that canonical transformations between other regions are missing
evolutionary data. The canonical transformation approach is also advantageous
because it is straightforward to quantize, the transformation equations surviving
as equalities between operators. However, before or after quantization, it does not
appear that the canonical “firewall transformation” given by (64) and (65) removes
any firewalls from the geometry, since both shells remain present in the theory on all
spacelike hypersurfaces.

6. Discussion

The firewall transformation, given by (1) and (2) [10], is inspired by the purely classical
Shapiro time delay that results from energetic bodies passing by each other and aims
to resolve the firewall problem while preserving unitarity of the time-evolution in
the quantum theory. However, it remains unclear to the present authors to what
degree the firewall transformation actually solves the firewall problem. In order to
elucidate the issue, we have proposed methods for including the Shapiro time delay
into the classical theory, specialized to the case of radially moving spherical shells of
null matter, and we have oriented our approach towards quantization by using the
canonical ADM formalism. Our first result is from Section 3, in which we showed
that taking the initial Kruskal coordinates of both the ingoing and outgoing shells
to zero is inconsistent with keeping a nonzero shift in those coordinates. This result
is not consistent with ’t Hooft’s procedure [10] of taking both limits to zero in the
quantum theory, which obscures the physical meaning of ’t Hooft’s quantum firewall
transformation. We have corrected the issue by taking only one or neither of the
Kruskal coordinates of the ingoing or outgoing shells to zero. In Section 5, we showed
that we can impose the Shapiro time delay between the canonical variables before and
after the collision by demanding the canonical variables for the shells in each region
are related by a canonical coordinate transformation, which we interpret as a way to
evolve the shells beyond the point of collision.
In the quantum case discussed by ’t Hooft, the firewall transformation removes
all ingoing particles or all outgoing particles at the horizon, which in turn removes
A Classical Firewall Transformation as a Canonical Transformation 13

the firewall that those particles create. However, in our classical analog, we have
a Hamiltonian including two spherical shells at all times, even after performing a
canonical transformation to include the Shapiro time delay. Exactly how one proceeds
from the classical theory to the quantum firewall transformation has so far not been
made explicit in the literature. Thus, either the quantum firewall transformation’s
removal of firewalls is a purely quantum effect, or the quantum firewall transformation
merely obscures any firewalls by removing them formally without removing them
physically. We leave to future work any investigation of the corresponding quantum
theory for the classical approach we have outlined in this paper. Regardless of whether
or not the quantum firewall transformation removes firewalls, it appears to preserve at
least some information normally lost beyond the event horizon of the black holes in the
observables for the outgoing particles, since the firewall transformation relates ingoing
and outgoing observables to each other. Furthermore, if the firewall transformation is
able to provide a unitary scattering matrix for particles interacting near a black hole,
that would be a significant step forward.
We make one final remark concerning case (2) of Section 5. When referring to the
firewall transformation, ’t Hooft has often spoken in terms of suppressing half of the
physical variables. Exactly this happens if we refer to physical degrees of freedom only
in region 1 in the context of case (2), while suppressing altogether the physical degrees
of freedom in region 3 (which could, nevertheless, be reconstructed by reference to the
firewall transformation). A feature that is not quite captured by our shell model is
that every ingoing particle will interact with every outgoing particle. It might have
been the case that every ingoing particle lost energy during each interaction, and
every outgoing particle gained energy as a result of each interaction. Then we could
imagine that, finally, all ingoing particles would end up with zero energy, and could
be ignored anyway, while all outgoing particles would start out with zero energy, and
hence could be ignored to begin with – so that some of what ’t Hooft seeks could
perhaps be recaptured – but none of that is what our calculations indicate. Instead,
the ingoing particles all gain energy as they interact with each outgoing particle, and
the outgoing particles all lose energy as they interact with each ingoing particle, so
both sets of particles are maximally energetic as they cross their respective horizons,
exactly where the firewalls would seem to exist. It is precisely here that ’t Hooft
wishes all particles to be ignored. However, to us, the choice to ignore all particles in
region 3, where indeed they are most energetic, seems to be somewhat bewildering,
and is a choice yet to be fully fathomed.

Acknowledgements

NAS acknowledges support from the Graduate Student Fellowship, the CLAS
Dissertation Fellowship, and a teaching assistantship from the University of Florida.
BFW acknowledges support from NSF through grant PHY-1607323, and from the
University of Florida. Hospitality at the Observatoire de Meudon, the Institut
d’Astrophysique de Paris, and at the Albert Einstein Institute in Potsdam throughout
the course of this work is also gratefully acknowledged.
REFERENCES 14

References

[1] Unruh W G 1976 Phys. Rev. D 14(4) 870–892 URL https://doi.org/10.1103/

PhysRevD.14.870
[2] Fulling S A 1973 Phys. Rev. D 7(10) 2850–2862 URL https://doi.org/10.
1103/PhysRevD.7.2850
[3] Davies P C W 1975 J. Phys. A Math. Theor. 8 609–616 URL https://doi.org/
10.1088/0305-4470/8/4/022
[4] Hawking S 1974 Nature 248 30–31 URL https://doi.org/10.1038/248030a0
[5] Mathur S D 2009 Class. Quant. Grav. 26 224001 URL https://doi.org/10.
1088/0264-9381/26/22/224001
[6] Almheiri A, Marolf D, Polchinski J and Sully J 2013 J. High Energy Phys. 02
062 URL https://doi.org/10.1007/JHEP02(2013)062
[7] ’t Hooft G 2016 Found. Phys. 46 1185–1198 URL https://doi.org/10.1007/
s10701-016-0014-y
[8] ’t Hooft G 2016 Found. Phys. 47 1503–1542 URL https://doi.org/10.1007/
s10701-017-0122-3
[9] ’t Hooft G 2016 The quantum black hole as a hydrogen atom: Microstates without
strings attached URL https://doi.org/10.48550/arxiv.1605.05119
[10] ’t Hooft G 2018 Found. Phys. 48 1134 – 1149 URL https://doi.org/10.1007/
s10701-017-0133-0
[11] ’t Hooft G 2021 Universe 7 298 URL https://doi.org/10.3390/
universe7080298
[12] ’t Hooft G 2022 Quantum clones inside black holes URL https://doi.org/10.
48550/ARXIV.2206.04608
[13] Hogan C 2020 Class. Quant. Grav. 37 095005 URL https://doi.org/10.1088/
1361-6382/ab7964
[14] Zeng D f 2022 Nucl. Phys. B 977 115722 URL https://doi.org/10.1016/j.
nuclphysb.2022.115722
[15] Kwon O 2022 Phenomenology of Holography via Quantum Coherence on Causal
Horizons URL https://doi.org/10.48550/arXiv.2204.12080
[16] Slagter R J 2022 Int. J. Mod. Phys. A 37 2250176 URL https://doi.org/10.
1142/S0217751X22501767
[17] Egorov V, Smolyakov M and Volobuev I 2023 Phys. Rev. D 107(2) 025001 URL
https://doi.org/10.1103/PhysRevD.107.025001
[18] Louko J, Whiting B F and Friedman J L 1998 Phys. Rev. D 57(4) 2279–2298
URL https://doi.org/10.1103/PhysRevD.57.2279
[19] Dray T and ’t Hooft G 1985 Nucl. Phys. B 253 173 – 188 URL https:
//doi.org/10.1016/0550-3213(85)90525-5
[20] Redmount I H 1985 Prog. Theor. Phys. 73 1401–1426 URL https://doi.org/
10.1143/PTP.73.1401
REFERENCES 15

Appendix A.

In this appendix, we derive the Hamiltonians for spherical shells of null matter via
the ADM canonical formalism for general relativity with spherical symmetry. The
ADM formalism with spherical symmetry admits the following decomposition of the
metric: [18]

ds2 = −N 2 dt2 + Λ2 (dr + N r dt)2 + R2 dΩ2

= −(N 2 − Λ2 N r 2 )dt2 + 2Λ2 N r dtdr + Λ2 dr2 + R2 dΩ2 , (A.1)

where t and r are the foliation coordinates, N (t, r) is the lapse, N r (t, r) is the shift,
and Λ(t, r) and R(t, r) are the canonical variables of the metric. The action for a
spherical shell of matter is
Z Z q
S = dtL = −m dt N̂ 2 − Λ̂2 (ṙ + N̂ r )2 , (A.2)

where m is the rest mass of the shell, r(t) is the canonical coordinate r for the shell, a
dot is a partial derivative with respect to t, and a hat signifies the quantity is evaluated
on the shell. The momentum conjugate to r is

δL mΛ̂2 (ṙ + N̂ r )
p= =q . (A.3)
δṙ
N̂ 2 − Λ̂2 (ṙ + N̂ r )2

Solving directly for the velocity,

N̂ /Λ̂
ṙ = η q − N̂ r , (A.4)
2 2
1 + Λ̂ m /p 2

where η ≡ sign p. Thus, in the null limit m → 0, we have

N
ṙ = η − N r. (A.5)
Λ
We can then construct the Hamiltonian:

H = pṙ − L
 N̂ 
= p η − N̂ r . (A.6)
Λ̂
In this paper, we specialize the approach to Kruskal coordinates, given by the
metric

ds2 = 2gU V dU dV + R2 dΩ2 , (A.7)

where
1 − 2M/R 16M 3 −R/2M
gU V = 8M 2 = e . (A.8)
UV R
REFERENCES 16

The Kruskal coordinates are related to the Schwarzschild coordinates implicitly by

 R 
UV = − 1 eR/2M , (A.9)
2M
 R 
V /U = sign − 1 eT /2M . (A.10)
2M
In the construction of the shell Hamiltonians we utilize hybrid coordinate systems
using the Schwarzschild time coordinate T as the foliation time coordinate t and
either Kruskal coordinate U or V as the foliation radial coordinate r. The metrics are

ds2 = 2e−T /2M gU V (dV 2 − V 2

2M dT dV ) + R dΩ
2
(A.11)
= 2eT /2M gU V (dU 2 + U 2
2M dT dU ) + R dΩ ,
2
(A.12)

and comparing with (A.1), we find the corresponding ADM variables:

2M V
N2 = 1 − , Λ2 = 2gU V e−T /2M , Nr = − , r = V, t=T (A.13)
R 4M
with V as the radial coordinate and
2M U
N2 = 1 − , Λ2 = 2gU V eT /2M , Nr = , r = U, t=T (A.14)
R 4M
with U as the radial coordinate. Inserting the above expressions into (A.6), we find
the Hamiltonians specialized to the Kruskal coordinates with the Schwarzschild time
coordinate:
 N̂ 
HV = p η − N̂ r
Λ̂
1 ηε + 1
= pV̂ , (A.15)
2M 2
 N̂ 
HU = p η − N̂ r
Λ̂
1 ηε − 1
= pÛ (A.16)
2M 2

where ε is the sign of V̂ or Û , respectively, and r is equal to V̂ or Û , respectively.

Equations (A.15) and (A.16) represent our starting point in Section 4.