Published for SISSA by Springer

Received: November 17, 2023

Accepted: January 8, 2024
Published: January 26, 2024

Null Raychaudhuri: canonical structure and the

dressing time

Luca Ciambelli ,a Laurent Freidela and Robert G. Leighb,a

JHEP01(2024)166
a
Perimeter Institute for Theoretical Physics,
31 Caroline St. N., Waterloo ON, N2L 2Y5, Canada
b
Illinois Center for Advanced Studies of the Universe & Department of Physics,
University of Illinois,
1110 West Green St., Urbana IL 61801, U.S.A.
E-mail: [email protected], [email protected],
[email protected]

Abstract: We initiate a study of gravity focusing on generic null hypersurfaces, non-

perturbatively in the Newton coupling. We present an off-shell account of the extended
phase space of the theory, which includes the expected spin-2 data as well as spin-0, spin-1
and arbitrary matter degrees of freedom. We construct the charges and the corresponding
kinematic Poisson brackets, employing a Beltrami parameterization of the spin-2 modes.
We explicitly show that the constraint algebra closes, the details of which depend on the
non-perturbative mixing between spin-0 and spin-2 modes. Finally we show that the spin zero
sector encodes a notion of a clock, called dressing time, which is dynamical and conjugate
to the constraint.
It is well-known that the null Raychaudhuri equation describes how the geometric data
of a null hypersurface evolve in null time in response to gravitational radiation and external
matter. Our analysis leads to three complementary viewpoints on this equation. First, it
can be understood as a Carrollian stress tensor conservation equation. Second, we construct
spin-0, spin-2 and matter stress tensors that act as generators of null time reparametrizations
for each sector. This leads to the perspective that the null Raychaudhuri equation can be
understood as imposing that the sum of CFT-like stress tensors vanishes. Third, we solve
the Raychaudhuri constraint non-perturbatively. The solution relates the dressing time to
the spin-2 and matter boost charge operators.
Finally we establish that the corner charge corresponding to the boost operator in the
dressing time frame is monotonic. These results show that the notion of an observer can
be thought of as emerging from the gravitational degrees of freedom themselves. We briefly
mention that the construction offers new insights into focusing conjectures.

Keywords: Classical Theories of Gravity, Gauge Symmetry, Space-Time Symmetries,

Black Holes

ArXiv ePrint: 2309.03932

Open Access, © The Authors.

https://doi.org/10.1007/JHEP01(2024)166
Article funded by SCOAP3 .
Contents
1 Introduction 1
2 Null hypersurfaces 6
2.1 Geometry 6
2.2 Dynamics 8
3 Presymplectic phase space 9
3.1 Diffeomorphisms and canonical generators 9

JHEP01(2024)166
3.2 Boost symmetry 11
3.3 Simplifying assumptions 11
3.4 Bulk Lagrangian 14
4 Kinematic Poisson bracket 16
4.1 Unimodular decomposition and complex structure 16
4.2 Beltrami parameterization 18
4.3 Propagators 19
4.4 Constraint algebra 20
5 Time and boost operator 22
5.1 Time operator 23
5.2 Presymplectic structure in dressing time 25
5.3 Boost operator 27
6 Conclusions and discussion 31
A Inverse of the symplectic 2 form 33
B Composite brackets 34
C Shear bracket 36
D Dressing frame symplectic structure 37

1 Introduction

It is a familiar property of classical gravitational theories such as general relativity that

diffeomorphisms in spacetime play the role of a gauge symmetry. Working on a fixed
background such as Minkowski spacetime, when suitable gauge fixing and linearization is
performed, the field equations have wave solutions. The simplest notion of quantization
then gives rise to corresponding helicity-2 graviton particle states, which can be regarded
as the most basic prediction of a quantum theory. Critically, it is in this sense in which
quantum gravity may have a correspondence limit, reducing to classical gravity at least in
the weak-field regime. This is however quite far from a full quantum gravitational theory,
in which one should expect background independence along with a formulation that does
not rely on a corresponding perturbative expansion in the Newton coupling. In fact, it is

–1–
well-known that if one approaches quantum gravity from the latter point of view in terms of
an effective Lagrangian field theory, then that theory is non-renormalizable, signaling that
new physics is required at high energy and/or short distance scales. It is also not obvious that
there is not new physics in the infrared as well, which may be associated with complications
that may occur in the correspondence limit.
There are several top-down approaches to quantum gravity postulating new physics at
high energy scales. The new physics either comes under the form of an infinite tower of new
degrees of freedom as in string theory or under some fundamental postulate of discreteness of
geometry as in loop quantum gravity or causal dynamical triangulations. The perspective

JHEP01(2024)166
that we plan to pursue here is more modest and aims at a bottom-up approach towards
quantum geometry relying on a precise and non-perturbative semi-classical understanding of
the coupling between the matter and gravitational degree of freedom and the backreaction it
implies on null geometry. The goal is to provide some basic unifying features that ought to
be present in the semi-classical limit of any top-down approach to quantum gravity.
Top-down approaches to quantum gravity have in common one feature, that performing
a semi-classical limit is not necessarily straightforward. This is mainly due to the fact that a
theory of quantum gravity needs to be fundamentally non-local and this in turn implies that
it is challenging to match at the quantum level the classical symmetries. Indeed, classical
gravity possesses many symmetries: some are trivial redundancies, while some are physical
symmetries that organize observables. This is an old discussion, going back to the original
works of Noether [1]. The point here is that while diffeomorphisms are gauge symmetries
and thus the theory must be discussed, either classically or quantum mechanically, with the
corresponding constraints, the charge operators do not vanish even on-shell, but localize
on codimension-2 surfaces (which herein will be referred to as ‘corners’ for brevity). It has
been proposed in recent years that corners play a central role in any gravitational theory,
an idea that has been referred to as the corner proposal [2–11] (see also the reviews [12–14]
and references therein). Universal algebras of charges at corners have been found, while their
quantum counterparts remain to be fully understood.
Where are these symmetries and how are the observables organized in classical and then
quantum gravity? A bottom-up approach to quantum gravity, while potentially incomplete,
has the virtue of automatically addressing these questions. Indeed, in this approach classical
gravity is not the endpoint of a limiting procedure, it is the starting point toward the quantum
regime, in which symmetries can be incorporated from the start. Whether one intends to
address classical evolution or to quantize the theory, a choice of Cauchy surface is made, which
given a classical spacetime would be a hypersurface that is either timelike or null. Once chosen,
there are several notions of phase space that we might address. First, there is the phase
space arising from some choice of fields in spacetime (leading to fields on the hypersurface).
In this case generally, because of the presence of gauge symmetries, the symplectic structure
derived from a Lagrangian would be degenerate (i.e., it is presymplectic). Second there is
a reduced phase space where one eliminates some subset of the fields as ‘pure gauge,’ thus
obtaining perhaps a simpler theory with a non-degenerate symplectic structure. In each case,
there is a further complication in that the (gauge) symmetries may or may not be canonically
represented on the phase space. In this regard, it has recently been realized [8, 15–17] that

–2–
one may first pass to an extended phase space, upon which the gauge currents are represented
by Hamiltonian vector fields on the space of fields. Each of these approaches is a possible
starting point for a quantum theory, some better than others presumably. Nonetheless, it is
important to prepare the classical ground as suitably as possible for quantization. The key
point is to insure that all aspects of the geometry are included in the gravitational phase
space, not only the spin-2 degrees of freedom but also the background geometry degrees of
freedom that enters as we will see in the form of spin-0 and spin-1 canonical pairs. The spin-0
pair which plays a prominent role in our analysis consists of the area form conjugate to the
surface tension, a boost connection encoding a combination of inaffinity and expansion.

JHEP01(2024)166
In this paper, we will focus on the classical gravity problem on generic finite distance
null hypersurfaces. There are certainly important examples of null hypersurfaces that
have been extensively studied in the literature. This includes the null conformal boundary
in asymptotically flat spacetimes; in this context the reduced phase space was long ago
understood beginning with Ashtekar and Streubel [18]. Another important example of a
null hypersurface that has been much studied is a Killing horizon [19]. By a generic null
hypersurface, we mean one that might be located anywhere in a spacetime without simplifying
assumptions. In fact, one may consider ‘segments’ of hypersurfaces that are taken to end
on specific corners, including on a (classical) singularity. Such a situation makes contact
with the familiar discussions of the classical focusing of null congruences and the classical
singularity theorems.
Gravity on null hypersurfaces, and in general null physics, possesses remarkable properties
leading to dramatic simplifications (compared to space-like or time-like hypersurfaces) that we
will demonstrate in this paper. A lot of efforts in quantum gravity have been unsuccessfully
focused on the Wheeler-DeWitt equation which represents canonical quantization of con-
straints associated with spacelike surfaces [20, 21]. One of the challenges in this case is that
the constraints are elliptic equations, whose solutions are non-local and no non-perturbative
treatment is known. On the other hand, constraints projected on null surfaces present
themselves as equations of evolution along null time and are therefore much simpler to solve
non-perturbatively. The canonical structure localizes along one-dimensional null rays and they,
therefore, provide a more direct and simpler insight into the quantum nature of geometry.
A conceptual complication is that the geometry on a null surface is non-Riemannian.
In particular, there is the degeneracy of the metric [22], and the corresponding intricacies
in projecting to the transverse and tangent subspaces [23–25]. In recent years, a superior
understanding of null geometry has been achieved, with the realization that Carrollian physics
plays a central role [26, 27]. The null geometry is correspondingly understood in terms
of data on a line bundle over a Euclidean space (which from a spacetime perspective is a
corner) called a Carroll structure.
In fact, it is now well-established how to encapsulate the symmetries and physical laws on
null hypersurfaces [28–44]. One approach is to consider the induction of Carrollian geometry,
including a metric, bundle structures and a connection, along with a symplectic structure,
from a spacetime theory. Another equivalent approach is to formulate a Carrollian stress
tensor directly on the null hypersurface. While this might be viewed as an effective ‘fluid’
perspective, we will in fact demonstrate below that these two approaches are completely

–3–
equivalent. From the first point of view, there are important constraints that describe the
properties of the null congruence (and from the Carroll perspective, the geometry of the
Carroll structure) while from the latter point of view, the constraints correspond to the
(broken) Ward identities of the Carrollian stress tensor. From each perspective, there are
modes that we will refer to as spin-0, spin-1, and spin-2 fields, this nomenclature making
reference to representation theory relevant to the codimension-2 base of the Carroll structure.
As just mentioned, in the latter perspective the Einstein equations correspond to Carrollian
conservation equations. In particular, they correspond to the Raychaudhuri [45–47] (see
also [48])1 and Damour [49, 50] constraint equations, which involve all of the spin-0, spin-
1, and spin-2 gravitational data as well as matter fields. The canonical charges can be

JHEP01(2024)166
constructed, and they reduce to codimension-2 charges on-shell of the constraints.
The relationship between the intrinsic Carrollian presymplectic data and the induced
gravitational presymplectic structure, off-shell of the constraints, has been studied in [51, 52];
see also [53] for an earlier proposal for the stress tensor on null boundaries. We here prove,
using the framework of [54–56], that the bulk action, improved by a suitably chosen boundary
Lagrangian, induces exactly the Carrollian canonical presymplectic potential, plus corner
terms. Indeed, the final action is related to the one found in [57] by corner Lagrangians.
Building the gravitational presymplectic data on a generic null hypersurface is not an
easy task, and the problem has attracted tremendous attention in recent years [51, 52, 57–67].
The study of the characteristic initial value problem on null hypersurfaces was initiated much
earlier by Sachs [68] and pursued in [69–78] using also the constrained Hamiltonian formalism.
The ensemble of works of Reisenberger stands out in this avenue of research [79–83].
The main body of this manuscript is devoted to a full analysis of the presymplectic
structure and Poisson bracket on a generic null surface in four spacetime dimensions. As
we mentioned above, this involves an intricate interplay between spin-0,1,2 data. In this
paper, we put much of our attention on the spin-0 and spin-2 contributions, as our intent
is to focus on the Raychaudhuri constraint. The spin 0 canonical data include the surface
tension, which is often treated as a background structure. Here we see that it has to be
understood dynamically as responding, through the constraint, to the presence of spin 2 and
matter degrees of freedom. It is profitable to formulate the spin-2 data (corresponding to the
degenerate metric and shear) in terms of the Beltrami parameterization [84], the use of which
in this context was first implemented by Reisenberger; see also [85, 86] for another recent use
of the Beltrami parameterization in gravity. We can then invert the off-shell presymplectic
structure and read the resulting kinematic Poisson brackets.
More generally though, there is a non-trivial mixing between the spin-0 and spin-2
modes which means that we cannot simply decouple the spin 0 modes as is assumed in a
perturbative treatment. The Poisson brackets are local on the base, and their non-locality
is only along the null fibre direction. We find that the constraint algebra closes only when
both the spin-2 and spin-0 contributions are included. This means that one cannot treat the
spin-2 gravitational radiation independently of the geometry of a generic null hypersurface.
Our formulation is however non-perturbative in Newton’s constant, and thus can be regarded
as including gravitational backreaction.
1
The Raychaudhuri equation on null hypersurfaces was first derived in [46].

–4–
 The constraint algebra that we referred to above is of course off-shell as the constraints
themselves have not been imposed. The Poisson brackets may thus be regarded as kinematic.
In a classical theory we would then look for a way to pass to the physical subspace upon
which the constraints are imposed. We will show that the Raychaudhuri constraint generating
the null time reparametrizations can be understood as the vanishing of a total CFT-like stress
tensor, which is the sum of spin-0, spin-2, and matter contributions. In a quantum theory,
the details are of course different, the constraint becoming an operator. Given the nature of
the Raychaudhuri constraint, we expect anomalies to play a key role at the quantum level.
In the conclusion, we will comment upon these options but leave details to future work.
There is a particular linear combination of a diffeomorphism along ℓ and internal boost,

JHEP01(2024)166
which we denote Mf′ with f a function on the null geometry, that has the effect of preserving
the reduced space of the spin-0 and spin-2 modes. Interestingly, we find that in the classical
theory there is a canonical transformation the result of which has the interpretation of
measuring null time via a field variable that emerges from the spin-0 sector. There is a
choice of such a clock, called dressing time, that completely relegates to the corner the spin-0
symplectic pair on-shell. In the special case of Killing horizons this clock coincides with affine
time, but is generally different. Remarkably, the dressing time clock variable is canonically
conjugate to the Raychaudhuri constraint. All other fields, including the spin-2 gravitational
mode and arbitrary matter fields, are dressed in the sense of becoming invariant under
diffeomorphisms in the dressing time frame. We thus arrive at an interpretation in which
an observer (in this case, simply a clock) emerges directly from the gravitational degrees of
freedom. In a quantum context, this is closely related to the crossed product construction of
refs. [87–89]. Since the clock is associated with the spin-0 modes, one might have naively
thought that these were pure gauge and had no physical impact. However, we will see very
explicitly that this is not the case — they instead give rise to a clock which is dynamical
throughout the bulk of the hypersurface but is conjugate to a constraint. Promoting the
clock to be dynamical has another virtue, which is that we can solve the Raychaudhuri
constraint non-perturbatively.
The boost charge we construct is given by the corner area, which is automatically positive.
In the dressing time only, it is further monotonic on-shell, if the classical null energy condition
is imposed. When quantum matter is added to the system, this charge represents therefore a
good candidate to describe the generalized entropy [90], and extend the quantum focusing
conjecture [91–93] to arbitrary null hypersurfaces.
The raison d’être of this manuscript is thus to appreciate that null hypersurfaces have
surprising properties, leading to simplifications in dealing with gravity: the gravitational
data can be fully recast as intrinsic Carrollian data, the symplectic structure can be inverted,
the constraint can be solved non-perturbatively and is conjugate to the dressing time, it
can be further understood as a balance equation for CFT-like stress tensors associated to
the spin-0, spin-2 and matter sectors, the boost charge is positive and monotonic. All
these features are essential in preparing the ground for future explorations. The long-term
idea is that null physics offers a non-perturbative window to study quantum gravity from
a bottom-up approach.
The paper is organized as follows. We begin in section 2 by reviewing the salient
ingredients of the geometry 2.1 and dynamics 2.2 of null hypersurfaces. Then, section 3
is devoted to the canonical presymplectic phase space. From the general analysis 3.1 we

–5–
find a set of assumptions 3.3 isolating the spin-0 and spin-2 contributions. We then show
how to relate this intrinsic analysis to the bulk gravitational Lagrangian 3.4. In section 4
we construct the Poisson brackets of the dynamical fields. The key ingredients are the
complex structure 4.1 and the Beltrami parameterization 4.2. They lead us to the notion
of propagators 4.3, which are instrumental in our analysis. We use them in 4.4 to compute
the constraint algebra, and in particular the highly non-trivial mixing between the spin-0
and spin-2 Hamiltonians. Moving from an arbitrary coordinate null time to the dressing
time, we show in section 5 that the spin-0 data are pushed to the corner, and are replaced
by the full constraint operator in the bulk. We conclude showing in 5.3 that the boost
charge is positive and monotonic in the dressing time frame. We then offer some outlook in

JHEP01(2024)166
section 6. Technical details supporting the results of section 4 appear in appendices A, B,
and C. Similarly, appendix D contains useful materials for section 5.

2 Null hypersurfaces

We review here the salient ingredients of the geometry and dynamics of null hypersurfaces.
The main references are [37, 51, 52], see also [13].

2.1 Geometry
An (n + 1)-dimensional null hypersurface N is intrinsically described as a manifold endowed
with a Carrollian structure (qab , ℓb ). It consists of a nowhere-vanishing vector field ℓ = ℓa ∂a
and a rank-n metric qab such that ℓa qab = 0. Here, for simplicity, we will take n = 2
(corresponding to a spacetime dimension d = n + 2 = 4). Any such Carrollian structure is
unique up to a local internal boost sending ℓ to eλ ℓ. Mathematically this means that N is
a fibred manifold with fibration π : N → S, such that the null direction ℓ is in the kernel
of dπ and S is a Riemann surface that we will usually think of as a 2-sphere.
A generic null hypersurface in a manifold has creases and caustics, see e.g. [94, 95]. The
Carrollian structure is applicable only to portions of null hypersurfaces in which creases and
caustics are not present. In this regard, we are performing a quasi-local analysis on N , and
thus we assume that N is an open set with topology S × I.
We denote C an arbitrary transverse cut, i.e., a closed codimension-1 surface on N with
non-degenerate induced metric. The degenerate metric qab on N determines a spatial area
form εC such that ιℓ εC = 0, where ιℓ is the interior product. Moreover, the choice of ℓ
determines a volume form εN on N , such that ιℓ εN = εC . The expansion θ, defined below,
relates the two forms through dεC = θεN . In the following, we will assume that the manifold
N is such that its completion N has boundaries (corners) ∂N = C∂ with C∂ = Cpast Cfuture .
S

To construct a connection on N and describe dynamics on a null hypersurface, it is

necessary to introduce an Ehresmann connection: an Ehresmann connection is a 1-form
k = ka dxa dual to the null vector ℓ, ιℓ k = 1. In the context of Carrollian physics, this was
described in [37, 96]. The Ehresmann connection defines a notion of horizontality, where Y
is a horizontal vector field on N if ιY k = 0. A general vector ξ ∈ T N can be decomposed
as ξ = f ℓ + Y , with Y horizontal. With the Ehresmann connection one can decompose
the volume form on N as εN = k ∧ εC , and define the horizontal projector qa b = δa b − ka ℓb

–6–
satisfying ℓa qa b = 0 = qa b kb and qa b qbc = qac . Such a construction has been introduced
in [23]; see also [24]. The Carrollian literature is vast. Its link with BMS symmetries of null
hypersurfaces, and their gravitational data has been appreciated in [28–43].
The set of data (qab , ℓb , ka ) defines a ruled Carrollian structure on N . The Carrollian
acceleration φ and vorticity w, which are a transverse 1-form and 2-form, are defined via
dk = φ ∧ k + w. Setting the vorticity to zero corresponds to a choice of foliation for N , that
is, a global section of the bundle. With the ruled Carrollian structure we can uniquely raise
indices of horizontal tensors. As an illustrative example, given the expansion tensor, defined
as the Lie derivative along ℓ of the degenerate metric,

JHEP01(2024)166
1
θab := Lℓ qab , (2.1)
2
the tensor θa b is defined as the unique tensor satisfying θa c qcb = θab and θa b kb = 0. We
can then decompose the expansion tensor in terms of the expansion θ and shear σa b of
the null surface:
1
θa b = θqa b + σa b , with σa a = 0. (2.2)
2
A ruled Carrollian structure also allows us to introduce the notion of Carrollian con-
nections Da discussed in the literature in [51, 52, 97].2 We choose Da to be a torsionless
connection preserving the identity, that is, Da (qb c + kb ℓc ) = 0. This Carrollian connection
defines a 1-form ω = ωa dxa on N given by

Da εN = −ωa εN . (2.3)

This 1-form can be decomposed into a transverse and horizontal part as ωa = κka + πa ,
where πa is the Hájiček connection and κ the inaffinity of ℓ.3 The 1-form ω is a boost
connection since, under the boost ℓ → eλ ℓ and k → e−λ k, it transforms as ω → ω + dλ. The
boost-covariantized derivative of ℓ is horizontal and given by the expansion tensor

(Da − ωa )ℓb = θa b . (2.4)

The Carrollian connection chosen only preserves the metric when derivatives and tensor
indices are taken to be horizontal. In general, when the expansion is non-vanishing the
induced metric is not preserved by the connection, and one has

Da qbc = −kb θac − kc θab . (2.5)

To recapitulate, this Carrollian connection has been found requiring absence of torsion, and
metricity in the horizontal directions. In particular, these properties guarantee that we can
write the expansion tensor à la Brown-York, that is,
1
θab = Lℓ qab = D(a ℓc qb)c . (2.6)
2
2
As discussed in [52], the Carrollian connection can be chosen to be induced by the bulk Levi-Civita
′ ′ ′
connection ∇a through Da Tb c := Πa a Πb b ∇a′ Tb′ c Πc′ c following the framework of [23]. Here, Πa b = qa b +ka ℓb
is the bulk Rigging projector, which becomes δab on the null hypersurface.
3
The symbol κ is often used to indicate the surface gravity. However, inaffinity and surface gravity are
in general different, coinciding only for Killing horizons. For more details, see for instance [98], where the
inaffinity is denoted κ2 while the surface gravity κ1 .

–7–
2.2 Dynamics
We now turn our attention to the dynamics induced from Einstein’s equations. It is well-
known that the projection of Einstein’s equations along a null hypersurface leads to the
Raychaudhuri [45] and Damour [49] equations, see also [99] and [38, 52]. These are given
by (Rℓℓ = ℓa Rab ℓb )

(Lℓ + θ)[θ] = µθ − σa b σb a − Rℓℓ , (2.7)

b b b b
qa (Lℓ + θ) [πb ] + θφa = (Db + φb )(µqa − σa ) + qa Rbℓ , (2.8)

where we have introduced the surface tension 4 of N

JHEP01(2024)166
θ
µ := κ + (2.9)
2
and Da = qa b Db denotes the horizontal derivative. We denote Lξ = ιξ d+dιξ the spacetime Lie
derivative along ξ. One can check that these equations are covariant under the boost symmetry
ℓ → eλ ℓ due to the fact that µ transforms as a boost connection µ → eλ (µ + ℓ[λ]). We also
note that the Carrollian vorticity w does not contribute, while the Carrollian acceleration
φ contributes only to the Damour equation. The surface tension µ is an important datum
in our construction, and will play a central role in the rest of the manuscript.
Equations (2.7) and (2.8) can alternatively be understood as conservation equations of
a Carrollian fluid. One defines the Carrollian fluid energy-momentum tensor in terms of
the rigged projector and the Carrollian connection as

8πG Ta b := Da ℓb − δa b Dc ℓc . (2.10)

The tensor Da ℓb = ωa ℓb + θa b is the Weingarten map [24]. The Carrollian fluid energy-
momentum tensor can be decomposed in terms of the quantities defined above as Ta b =
τa ℓb + τa b , where

8πG τa = πa − θka , (2.11)

b b b
8πG τa = σa − µqa . (2.12)

As shown in [51] (see also [43, 52]), Einstein’s equations pulled back on N can then
simply be written as fluid conservation equations

Db Ta b = Taℓ
mat
, (2.13)
mat is the bulk matter energy-momentum tensor. This generalizes to null surfaces the
where Tab
Brown-York result [100] and is in agreement with the membrane paradigm [99]. The Carrollian
energy-momentum tensor can be seen as the c → 0 limit of a relativistic energy-momentum
tensor. In this fluid analogy, θ is the fluid energy, πa the heat flux, −µ the pressure, σab the
viscous stress tensor, and Taℓmat the external energy flux [44, 97, 101].

The Raychaudhuri equation (2.7) can be derived from (2.13) by projecting the latter on ℓ

8πG ℓb Da Tb a = −(Lℓ + θ)[θ] + µθ − σa b σb a . (2.14)

Similarly, the Damour equation is obtained by projecting (2.13) on qc b , see [52].

4 d−3
These equations can be extended to any spacetime dimension d ≥ 3. For example, one has µ = κ + d−2
θ.

–8–
3 Presymplectic phase space

As shown in [62], the Raychaudhuri and Damour equations (2.7), (2.8) can be understood as
(non)-conservation equations of corner symmetry charges associated with diffeomorphisms
along ℓ and time-dependent diffeomorphisms of the base S. We are going to review that the
corresponding charge aspects are −θεC and πa εC , respectively. To see this, we first introduce
the canonical presymplectic potential5 intrinsic to N
1 ab
Z Z  
can can
Θ = θ = εN τ δqab − τa δℓa . (3.1)
N N 2

JHEP01(2024)166
This presymplectic potential is the null analogue of the Brown-York potential [100], and it is
uniquely selected by demanding that the configuration variables are the Carrollian geometric
data qab and ℓa . This expression first appeared in [51], and shows that −τa and 12 τ ab are
the momenta conjugate to the Carrollian vector ℓ = ℓa ∂a and the degenerate Carrollian
metric qab , respectively. The tensor τ ab is defined as the unique symmetric tensor such that
qac τ cb = τa b and τ ab kb = 0. Note that while qab is not invertible, by q ab we mean the unique
symmetric tensor for which q ab kb = 0. This satisfies qa b = qac q cb . The presymplectic two-form
associated to this potential is derived from Einstein-Hilbert gravity in section 3.4.

3.1 Diffeomorphisms and canonical generators

To evaluate the canonical charges associated with diffeomorphisms one uses that the integrand
of the presymplectic potential is manifestly invariant under diffeomorphisms. We introduce
field space calculus Lξ̂ = Iξ̂ δ + δIξ̂ , with ξˆ the Hamiltonian vector field associated with
the diffeomorphism ξ ∈ T N ; see [12, 14] for details. To proceed, we study how the fields
transform. Since at this stage no assumptions have been made on field space, we have that
the fields transform covariantly via the Lie derivative

Lξ̂ qab = ξ c Dc qab + qac Db ξ c + qcb Da ξ c (3.2)

Lξ̂ ℓa = ξ b Db ℓa − ℓb Db ξ a (3.3)
Lξ̂ τ ab = ξ c Dc τ ab − τ ac Db ξ c − τ cb Da ξ c (3.4)
Lξ̂ τa = ξ b Db τa + τb Da ξ b . (3.5)

We then find that Lξ̂ Θcan = Fξcan , where the canonical flux Fξcan is a corner term given by

1 ab
Z  
Fξcan = ι ξ εN τ δqab − τa δℓa . (3.6)
C 2
Here, and in the following, we denote C the boundary of N , previously called C∂ . The
integrand of the flux is a spacetime 2-form and field space one-form simply given by the
contraction ιξ θcan . The canonical flux vanishes when ξ|C ∈ T C, or when the phase space
is properly extended [15–17].
The charge is then extracted from

Iξ̂ Ωcan = −δQξ + Fξcan Qξ = Iξ̂ Θcan , (3.7)

5
In this section, we deal with the off-shell presymplectic phase space.

–9–
which shows that Iξ̂ Θcan is the canonical charge for diffeomorphism symmetry. This result
holds assuming δξ = 0.
We therefore find that the canonical charge is the sum of the constraint (the equations
of motion) plus the corner charge
Z  
can
Iξ̂ Θ = ε N T a b D b ξ a − τa ξ b D b ℓ a
ZN  
= εN Ta b (Db − ωb )ξ a
NZ Z
=− εN ξ a D b Ta b + εb ξ a Ta b , (3.8)
N C

JHEP01(2024)166
where we used τb Da ℓb = Ta b ωb in the second equality, eq. (2.3) in the third, and we denoted
εb = ι∂b εN . The first term in the last equality of (3.8) vanishes on-shell,6 showing that the
charges associated with a vector field ξ tangent to N are corner charges given by
Z Z
Qξ = εN Ta b (Db − ωb )ξ a =
ˆ εb ξ a Ta b , (3.9)
N C

where = ˆ denotes the on-shell evaluation.

From now on, we choose the transverse cut C to be such that εb pulls back to kb εC .
The diffeomorphism ξ is given by an arbitrary7 diffeomorphism f along ℓ and a transverse
diffeomorphism, Y a , such that8

ξ a = f ℓ a + Y b qb a . (3.10)

Let us compute the charge Qf ℓ = Mf associated to ξ a = f ℓa . We have

1 1
Z   Z
Mf = εN f σa b σb a − θ(ℓ + µ)[f ] =
ˆ− εC f θ. (3.11)
8πG N 8πG C

The last step could have been derived directly from (3.9), since ℓa Ta b εb = −θεC . On the
other hand, for a diffeomorphism along C, ξ a = qb a Y b , we have (QY = PY )
1 θ
Z    
PY = εN σc b (Db − πb )Y c + πc Dℓ − Y c − µDc Y c − θY c φc (3.12)
8πG N 2
1
Z
=
ˆ ε C Y a πa , (3.13)
8πG C
where Dℓ = ℓa Da . As expected, the charges become corner charges on-shell.
If the canonical flux vanishes at the corner Fξcan = 0, which is equivalent to the condition
f |C = 0, one can simply derive the charge algebra from the Noether analysis. Indeed, under
this assumption δQξ = Iξ̂ Ωcan and we have

ˆ ζ̂) = −I I Ωcan = I δI Θcan = L I Θcan = I Θcan = −Q[ξ,ζ] ,

{Qξ , Qζ } = Ωcan (ξ, (3.14)
ξ̂ ζ̂ ξ̂ ζ̂ ξ̂ ζ̂ [ξ̂,ζ̂]
R
6
The derivation is given for pure gravity. If matter is present, there is another contribution N εN ξ a Taℓ
mat
=
mat
Iξ̂ Θ to the first term, and so the first term generally vanishes on-shell.
7
The time-independent and linear-in-time modes of f are the supertranslation and superboost parame-
ters, respectively.
8
We continue to require the field independence of the vector field, i.e., δξ = 0, and thus one has to select f
and Y a such that this holds.

– 10 –
 ˆ ζ̂] = −[ξ,
where the last equality is a consequence of [ξ, dζ]. This means that

{Mf , Mg } = −M(f ℓ[g]−gℓ[f ]) (3.15)

{PY , Mf } = −MY [f ] (3.16)
{PY , PY ′ } = −P[Y,Y ′ ]Lie . (3.17)

We note that one could have kept the dissipative terms and instead computed the charge
algebra using the extended phase space introduced in [15–17].

3.2 Boost symmetry

JHEP01(2024)166
As already noticed, Carrollian fluids possess another symmetry, which is the internal local
boost symmetry generated by the infinitesimal parameter λ. Given the canonical presymplectic
potential, we can compute the associated charges. This symmetry acts on the fields as

Lλ̂ ℓ = λℓ, Lλ̂ ϵN = −λϵN Lλ̂ qab = 0, Lλ̂ θ = λθ, Lλ̂ µ = ℓ(λ) + λµ. (3.18)

From this, we derive

1 1
Lλ̂ τ ab = λτ ab − ℓ(λ)q ab , Lλ̂ τa = Da λ. (3.19)
8πG 8πG
To compute the charge, assuming from now on δλ = 0, we start by evaluating
1 1 1
Z   Z 
Lλ̂ Θcan = − εN ℓ(λ)q ab δqab + Da λδℓa = − δ εN ℓ(λ) , (3.20)
8πG N 2 8πG N

where we used that δεN = ( 12 q ab δqab − ka δℓa )εN . We also have

1
Z
can
Iλ̂ Θ = εN λθ. (3.21)
8πG N

Then, we find that the boost charge is a corner charge even off-shell
1
Z 
Iλ̂ Ωcan = Lλ̂ Θcan − δIλ̂ Θcan = − δ εC λ . (3.22)
8πG C

This means that the boost charge is given by the corner area element
1
Z
Gλ = εC λ. (3.23)
8πG C

The charge algebra derived earlier can therefore be extended by the boost generator to give

{PY , Gλ } = −GY [λ] (3.24)

{Mf , Gλ } = −Gf ℓ[λ] . (3.25)

3.3 Simplifying assumptions

In this paper, we will make a set of simplifying assumptions that will bring us to a framework
p
suitable to study the Raychaudhuri equation. We start by denoting Ω := det(q) > 0 the
spatial measure factor. We also choose adapted coordinates xa = (v, z, z), where (z, z) are
complex coordinates for the base of the Carrollian bundle and such that

ℓ = e−α ∂v , (3.26)

– 11 –
where α labels the time scale. We will restrict our analysis to variations that do not change
the Carrollian structure:

δℓ = −δα ℓ. (3.27)

In these coordinates, the measure and expansion on N are given by

(0) (0)
εN = eα ΩεN , εC = ΩεC , θ = e−α ∂v log Ω, (3.28)

(0) (0)
where we recall that εC = ιℓ εN and we introduced the bare measure εN = dv ∧ εC with
(0)

JHEP01(2024)166
εC = d2 z. The variation of the measure is then

δεN = δα εN + δ log Ω εN δεC = δ log Ω εC . (3.29)

While the spin-1 contribution drops with these assumptions, the spin-0 and spin-2
contributions to the presymplectic potential read

1
Z  
Θcan = εN 1 ab
2 σ δqab − θδα − µδ log Ω . (3.30)
8πG N

The presymplectic 2-form Ωcan can be written more simply if one rewrites everything with
respect to the rescaled null generator ℓ̃ := ∂v = eα ℓ. The associated surface tension
is µ̃ = eα µ + ∂v α, while the associated shear is σ̃ ab = eα σ ab . Introducing the measure
ε̃N = dv ∧ εC = e−α εN , we have9

1 1
Z   Z
can
Ω = ε̃N 1
−δ µ̃∧δ log Ω+ 2Ω δ(Ωσ̃ ab )∧δqab + εC (δα∧δ log Ω) . (3.31)
8πG N 8πG C

The variation δα appears only in the last term, which is a corner contribution. The corner
contribution and physics is ultimately crucial in order to understand gravity and edge modes.
However, we first need to solve the constraints on the bulk of the hypersurface.
We thus want to impose δα = 0, and further α = 0. This allows us to identify the
Carrollian null vector ℓ simply with ∂v , and require the symmetries to preserve this notion
of time. To systematically achieve this, we must require the symmetry transformations to
be such that δα = 0. We note that the assumption (3.27) is preserved only by a transverse
diffeomorphism Y that does not depend on the coordinate v, such that [ℓ, Y ] = 0. This selects
among all diffeomorphisms on N those that are automorphisms of the Carrollian bundle.
Such symmetries have been called Carrollian diffeomorphisms in [35, 37, 97].
With this restriction, we recall that the symmetries act on α as

Lfˆα = e−α ∂v f = ∂v f Lλ̂ α = −λ LŶ α = 0, (3.32)

where we used that α = 0. We have also introduced the pragmatic notation Lfˆ = Lξ̂ with
ξ = f ∂v and LŶ = Lξ̂ with ξ = Y . We remark that, imposing α = 0, the diffeomorphism
along ℓ is explicitly field-independent, ξ = f ℓ = f ∂v .
9 can
Once
R rewritten in

terms of the rescaled data, there is an extra exact variational term in Θ given by
−δ N εN α ∂v log Ω , which will vanish with our assumption α = 0.

– 12 –
 We therefore see that in order to preserve the condition δα = 0 we need to combine the
action of the diffeomorphism along ℓ with a specific boost generated by λf = ∂v f :

L′fˆ := Lf + Lλf . (3.33)

This transformation is such that L′fˆα = 0 and thus L′fˆℓ = 0.

The combined transformation acts on the field space as

L′fˆqab = Lf qab , L′fˆσa b = ∂v (f σa b ) (3.34)

L′fˆµ = ∂v ((∂v + µ)f ), L′fˆΩ = f ∂v Ω, L′fˆθ = ∂v (f θ). (3.35)

JHEP01(2024)166
To summarize, we have shown that there is a consistent way to set δα = 0 by combining
diffeomorphisms along ℓ and boosts. These two symmetries cease to exist as independent
transformations on this reduced phase space. They combine together and act via L′ , while the
transverse diffeomorphism is unaffected. To stress the reduction, we will also use the notation
L′Ŷ , which is simply equal to LŶ . It is important to stress that under this transformation
the surface tension µ transforms anomalously, i.e. as a connection, not a scalar. We remark
that, given (3.35), if the surface tension µ is zero, it would remain zero on the phase space if
f is at best linear in v. This selects among the Carrollian diffeomorphisms the Weyl BMS
subgroup of symmetries studied in [102].
Dropping the tilde notation, with these restrictions the canonical presymplectic 2-form
reduces to
1
Z  
can 1 ab
Ω = εN − δµ ∧ δ log Ω + 2Ω δ(Ωσ ) ∧ δqab , (3.36)
8πG N

where all quantities are now computed using the null generator ℓ = ∂v . In particular
1 ′ ′ Ω ′ ′ q a′ b ′
 
ab
σ = q aa q bb (∂v qa′ b′ − θqa′ b′ ) = q aa q bb ∂v . (3.37)
2 2 Ω
This is our starting point for the kinematic Poisson bracket, in the next section.
We compute the canonical charges directly from the presymplectic 2-form
1
Z  
(0)
If′ˆΩcan = εN −L′fˆµδΩ+ 12 L′fˆ(Ωσ ab )δqab +δµL′fˆΩ− 12 δ(Ωσ ab )L′fˆqab (3.38)
8πG N
1
Z  
(0)
= εN δµf ∂v Ω−(∂v (∂v +µ)(f ))δΩ−δ(Ωf σa b σb a )+ 12 ∂v (f Ωσ ab δqab ) (3.39)
8πG N
1
Z  
(0)
= εN δ(−∂v2 f Ω+µf ∂v Ω−Ωf σa b σb a )+∂v ( 12 f Ωσ ab δqab −f µδΩ) , (3.40)
8πG N

where in the second step we used (3.37). We can rewrite this as

If′ˆΩcan = −δMf′ + Ff , (3.41)

f ι∂v θcan , while the first term gives the

R
where the last term is the expected flux Ff := C
new charge
1 1
Z Z
(0) (0)
Mf′ = εN f C + εC (Ω∂v f − f ∂v Ω) , (3.42)
8πG N 8πG C

– 13 –
where C := ∂v2 Ω − µ∂v Ω + Ωσa b σb a is the Raychaudhuri constraint. As a consistency check, we
note that the new charge is simply the sum of the old charge associated to a diffeomorphism
along ℓ and the internal boost charge: Mf′ = Mf + G∂v f . The transverse diffeomorphism
charges are unchanged, PY′ = PY when Y is Carrollian.
In the absence of fluxes, we compute that the charge algebra

{Mf′ , Mg′ } = −M(f

′
ℓ[g]−gℓ[f ]) (3.43)
{PY′ , Mf′ } = −MY′ [f ] (3.44)
{PY′ , PY′ ′ } = −P[Y,Y
′
′]
Lie
(3.45)

JHEP01(2024)166
has the same structure as before, as expected.

3.4 Bulk Lagrangian

So far, we have confined our attention to the intrinsic Carrollian structure on null hypersurfaces.
We have shown that the Carrollian conservation laws — the Carrollian equations of motion —
encode the Einstein constraints, once projected onto the null hypersurface. But to proceed,
we also need to show that the presymplectic structure is the same, as we are interested in
the kinematic Poisson brackets of the dynamical data.
To do so, consider the 4-dimensional Einstein-Hilbert action coupled to matter

1
 
LEH = R + Lmat ε, (3.46)
16πG

|g|d4 x is the volume form, and we put c = 1. Its local presymplectic potential
p
where ε =

1
θEH = ∇µ (g νσ g µρ δgσρ − g νµ g ρσ δgρσ )εν (3.47)
16πG
 
1 mat − G µν
is found through the variation δLEH = −E + dθEH , where E = 16πG 8πGTµν µν δg ε
√ mat
−2 δ( |g|L )
mat = √
are the equations of motion and Tµν δg µν .
|g|
Consider a null hypersurface N , and introduce coordinates such that the Carrollian
vector on N and the 1-form normal to N are given by10

ℓ = e−α ∂v , n = eᾱ dr. (3.48)

In [52], it has been shown that the Einstein-Hilbert presymplectic potential is related to
the Carrollian canonical presymplectic potential (3.1) as11

θEH = θcan + δℓN − dθC , (3.49)

where we defined
1 1
ℓN = εN (κ + θ), θC = εC δ(ᾱ − α). (3.50)
8πG 16πG
10
The quantity ᾱ ensures the normalization condition ιk n = 1.
11
While we assume here that the spin-1 field (V A in [52]) vanishes, this equation holds in full generality.

– 14 –
We recall that κ is the inaffinity, which is related to α and ᾱ via κ = ℓ(ᾱ) + · · · , where the
dots represent terms linear in the fields that do not contribute to the presymplectic potential
associated to ℓN , and thus are not relevant in the following.
Our goal is to show that the bulk Einstein-Hilbert action can be improved by a boundary
Lagrangian, such that the full bulk potential is equivalent to the Carrollian potential plus
a corner contribution.
Let us review the procedure outlined in [54–56]. Essentially, two bulk Lagrangians
differing by a boundary Lagrangian

L1 = L2 − dℓbdy , (3.51)

JHEP01(2024)166
have symplectic potentials related via

θ1 = θ2 − δℓbdy + dθbdy , (3.52)

where θbdy is the symplectic potential of ℓbdy .

Given the result (3.49), it is natural to define an improved bulk Langrangian given by

Ltot = LEH − dℓN . (3.53)

To compute θtot , we thus need to evaluate the corner potential θN . Using δℓN = −E + dθN ,
and since κ + θ = ℓ(ᾱ + ln Ω) + · · · where the dots are terms with no time derivatives,
we explicitly find
1  
θN = δ ᾱ + Ω−1 δΩ εC . (3.54)
8πG
Therefore the total presymplectic potential becomes

θtot = θEH − δℓN + dθN (3.55)

= θcan + d(θN − θC ) (3.56)

1  
= θcan + d εC (δα + δ ᾱ + 2Ω−1 δΩ) (3.57)
16πG
1
= θcan + dδℓcor + d(εC (δα + δ ᾱ)), (3.58)
16πG
1
where we used (3.49) in the second line and introduced the corner Lagrangian ℓcor = 8πG εC
in the last step.
Adding this corner Lagrangian to the total action, one exactly retrieves the action found
in [57], when α = 0, studied further in [65]. The boundary Lagrangian is the null analogue of
the Gibbons-Hawking term, [58, 59], while the corner term is the null analog of the Hayward
term [103]. The symplectic 2-form Ωtot is
1
Z
tot tot can
Ω = δΘ =Ω + δεC ∧ δ(α + ᾱ). (3.59)
16πG C

This exactly matches (3.36), if we further require δ(α + ᾱ)|C = 0. This shows that the
bulk Einstein-Hilbert action can be improved by a boundary Lagrangian, such that the
final gravitational presymplectic structure on a null hypersurface is exactly the intrinsic
Carrollian presymplectic structure.

– 15 –
4 Kinematic Poisson bracket

In this section we compute the kinematic12 Poisson bracket of the Carrollian presymplectic
phase space variables. Doing so requires performing integration by part, so in this section
we assume that all variations vanishes on ∂N .

4.1 Unimodular decomposition and complex structure

Our starting point is the canonical presymplectic 2-form (3.36), that we rearranged into

1
Z    
(0)
Ωcan = 1 ab

JHEP01(2024)166
εN δ 2 Ωσ ∧ δqab − δµ ∧ δΩ . (4.1)
8πG N

We can decompose the metric into its unimodular and determinant parts, 13 such that

qab = Ω q ab det(q) = 1. (4.2)

Then, the shear can be written as σa b = 12 ∂v q ac q cb , where q ab satisfies q ab kb = 0 and

q ac q cb = qb a , with qb a the horizontal projector. The Raychaudhuri equation can then be
written, using (2.7), as a constraint equation
 
mat
C = ∂v2 Ω − µ∂v Ω + Ω σ 2 + 8πGTvv = 0, (4.3)

where σ 2 := σa b σb a . We observe that, compared to (2.7), this equation seems more regular in
the presence of creases and caustics, where Ω → 0, provided that µ has sufficiently regular
behaviour. While we have excluded these situations in our analysis, it would be interesting
to study the regularity of our constraints further.
The canonical presymplectic 2-form becomes

1 1 
Z  
′ ′

can (0)
Ω = εN δ ∂v q a′ b′ q aa q bb ∧ δ (Ωq ab ) − δµ ∧ δΩ . (4.4)
8πG N 4

To invert it and thus derive the kinematic Poisson bracket, it is convenient to introduce
null coframes satisfying ma ma = 0 = mb mb and ma ma = 1 = ma ma , together with
ℓa ma = 0 = ℓa ma and ma ka = 0 = ma ka . For the rest of this section indices are raised
and lowered using q ab , so that ma = q ab mb . These frames can be used to decompose the
metric and the tensor ϵa b as follows

q ab = ma mb + ma mb , (4.5)
b b b
ϵa = ma m − ma m . (4.6)

The tensor ϵa b satisfies the important properties

ϵa b ϵb a = 1 ϵa b mb = ma ϵa b mb = −ma . (4.7)
12
This means the bracket before imposing the constraints.
13
This tensor is degenerate so its determinant vanishes. By det(q), we mean the determinant of the
non-degenerate part of this tensor, which is co-rank 1.

– 16 –
The sphere complex structure will have non-trivial Poisson brackets, and it is given by
Ja b = iϵa b .14 The physical results we will derive are real combinations involving this complex
structure, which is a useful tensor in the following.
The variation of the coframe is decomposed as

δma = ma ∆ + ma ω (4.8)
δma = ma ∆ − ma ω (4.9)
a a a
δm = −m ∆ + m ω (4.10)
a a a
δm = −m ∆ − m ω. (4.11)

JHEP01(2024)166
Here we introduced two complex one-forms in field space, ∆ and ω, whose explicit expressions
depend on the parameterization chosen. We note that ω corresponds to a connection for
frame rotations, and that it is pure imaginary, in order to preserve the condition ma ma = 1.
Therefore, we have that

δq ab = 2ma mb ∆ + 2ma mb ∆, (4.12)

b b b
δϵa = 2ma m ∆ − 2ma m ∆. (4.13)

We see that both variations are independent of ω. Further, we note that ma mb and its
complex conjugate form a basis of the space of 2 × 2 traceless symmetric matrices. This means
that the complex structure variation contains the same information as the metric variation

δϵa b ϵb c = −ϵa b δϵb c = δq ab q bc . (4.14)

The v-derivative of m and m can also be decomposed in a similar fashion

∂v ma = ma ∆v + ma ωv ∂v ma = ma ∆v − ma ωv , (4.15)

where ∆v and ωv are two complex scalars given by I∂ˆv ∆ = ∆v , etc. Then

σa b = ma mb ∆v + ma mb ∆v . (4.16)

A straightforward computation which uses that

δσ ab = [(δ − 2ω)∆v ]ma mb + [(δ + 2ω)∆v ]ma mb − (∆v ∆ + ∆v ∆)q ab , (4.17)

leads to the final expression

1
Z  
(0)
Ωcan =


εN δΩ ∧ δµ + ∆v ∆ + ∆v ∆ + Ω [(δ − 2ω)∆v ] ∧ ∆ + [(δ + 2ω)∆v ] ∧ ∆ .
8πG N
(4.18)
This is the expression we need to invert, in order to find the kinematic Poisson bracket.
14
In the following, in an abuse of notation, we will refer to ϵa b itself as the complex structure.

– 17 –
4.2 Beltrami parameterization
To proceed and construct the inverse of the canonical 2-form, it is useful to introduce the
Beltrami parameterization [84], see also [79, 80]. This parameterization is a convenient
way of describing the degrees of freedom of a 2-dimensional metric. First of all, given that
ℓ = ∂v , the condition ℓa qab = 0 reduces to qva = 0 = qvv and mv = 0 = mv . Then, the
null coframe can be parameterized as
1 ζ ζ
mz = √ = mz , mz = √ , mz = √ , β := 1 − ζζ, (4.19)
β β β

JHEP01(2024)166
where we introduced the Beltrami differential ζ(x) and its complex conjugate ζ(x), such that
the unimodular metric can be written as (recall xa = (v, z, z))

|dz + ζdz|2
q ab (x)dxa dxb = 2(ma dxa )(mb dxb ) = 2 . (4.20)
β
Examining the variation of the null frame given above leads to the explicit formulae
δζ ¯ = δζ , ζδζ − ζδζ
∆= , ∆ ω= = −ω. (4.21)
β β 2β
Similarly, the v-derivative of m and m gives
∂v ζ ∂v ζ ζ∂v ζ − ζ∂v ζ
∆v = , ∆v = , ωv = = −ω v , (4.22)
β β 2β
which again are related to the variations by I∂ˆv ∆ = ∆v , etc. With the Beltrami differential,
we can process (4.18) further. We begin by noting the identities

δ∂v ζ δζζ∂v ζ
(δ − 2ω)∆v = +2 = (∂v − 2ωv )∆, (4.23)
β β2
δ∂v ζ δζζ∂v ζ
(δ + 2ω)∆v = +2 = (∂v + 2ωv )∆. (4.24)
β β2
Introducing the ω-covariant derivatives

Dv = ∂v − 2ωv , Dv = ∂v + 2ωv , (4.25)

we arrive at
1
Z   
can (0)

Ω = εN δΩ ∧ δµ + ∆v ∆ + ∆v ∆ + Ω Dv ∆ ∧ ∆ + Dv ∆ ∧ ∆ . (4.26)
8πG N

In this section, we will be interested in the constraints in the bulk of the null hypersurface.
Therefore, we assume that the variations of the Beltrami differentials vanish at the corner,
such that we can freely integrate by parts in v this expression to rewrite it as

can 1
Z
(0)

√ √ 
Ω = εN δΩ ∧ δµ + ∆v ∆ + ∆v ∆ + 2Dv ( Ω∆) ∧ Ω∆ . (4.27)
8πG N

In the next subsection, we will derive the kinematic Poisson bracket from this canonical
presymplectic structure. Note that we may already expect some complications, as the spin-0
and spin-2 degrees of freedom are coupled together.

– 18 –
4.3 Propagators
To properly construct the inverse of (4.27), a crucial ingredient is a certain propagator.
Denoting x1 = (v1 , z1 , z 1 ) and x2 = (v2 , z2 , z 2 ), this is defined as15
R v2
1 −2 ωv dv
P 12 =√ e v1 H(v1 − v2 )δ (2) (z1 − z2 ), (4.28)
Ω1 Ω2

where Ω1 = Ω(x1 ) etc., and H(v) is the odd Heaviside function discussed in appendix A.
Given that ω v = −ωv , it satisfies the property
R v2

JHEP01(2024)166
1 2 ωv dv
P12 = P 12 =√ e v1 H(v1 − v2 )δ (2) (z1 − z2 ) = −P 21 . (4.29)
Ω1 Ω 2

This propagator is crucial because it is the covariant antiderivative of the delta function.
Indeed, using θ1 = ∂v1 log Ω1 , we have

δ (3) (x1 − x2 )
R v2
1 1
 
−2 ωv dv
D v1 + θ1 P 12 = √ e v1 ∂v1 H(v1 − v2 )δ (2) (z1 − z2 ) = , (4.30)
2 Ω 1 Ω2 Ω1

where we denote δ (3) (x) := δ(v)δ (2) (z). Note that this implies

δ (3) (x1 − x2 )
R v2
1 1
 
−2 ωv dv
D v2 + θ2 P 12 = √ e v1 ∂v2 H(v1 − v2 )δ (2) (z1 − z2 ) = − . (4.31)
2 Ω 1 Ω2 Ω2

Using this, we can obtain the kinematic Poisson bracket. We relegate details to ap-
pendix A, and show here the final result, which is

{Ω1 , µ2 } = 8πG δ (3) (x1 − x2 ) (4.32)

 
{µ1 , µ2 } = 4πG ∆v1 ∆v2 P 12 + ∆v1 ∆v2 P12 (4.33)
{µ1 , ζ2 } = −4πG ∆v1 β2 P 12 (4.34)
{µ1 , ζ 2 } = −4πG ∆v1 β2 P12 (4.35)
{ζ1 , ζ 2 } = 4πG β1 β2 P12 (4.36)
{ζ 1 , ζ2 } = 4πG β1 β2 P 12 . (4.37)

These are the fundamental building blocks from which we will derive brackets among composite
functionals. The spin-2 brackets are compatible with those found in [80]. We observe that
the field µ does not commute with the metric determinant or itself. We expect that this
feature will have important repercussions on the quantum theory.
As shown in appendix B, from here one can compute the bracket of the complex structure

{ϵ1a b , ϵ2c d } = −16πG P12 m1a mb1 m2c md2 − 16πG P 12 m1a mb1 m2c md2 , (4.38)

which shows that there is a non-trivial non-commutative structure for the geometric data.
15
We call these quantities “propagators” as they can be intuitively seen as non-trivial holonomies connecting
(propagating) the geometric data between the null times v1 and v2 .

– 19 –
 Furthermore, we also derive in appendix C the shear tensor bracket

{σa1b , σc2d } = 4πG m1a mb1 ϵ2d 1b 2 d 1b 2d
c ∆v2 Dv1 P12 + ϵa mc m2 ∆v1 Dv2 P 12 − ∆v1 ∆v2 P12 ϵa ϵc

+ m1a mb1 m2c md2 Dv1 Dv2 P 12 + c.c. , (4.39)

where c.c. denotes complex conjugation. This result is important in the derivation of the
constraint algebra, together with
 
{σa1b σb1a , µ2 } = 2{∆v1 ∆v1 , µ2 } = −8πG ∆v1 ∆v1 Dv1 P 12 + Dv1 P12 . (4.40)

JHEP01(2024)166
4.4 Constraint algebra
The Hamiltonian associated to diffeomorphisms along ℓ was given above in (3.11). We can
split it into its spin-0 and spin-2 contributions16

1
Z  
(0) (2) (0)
Mf′ = εN Ω f σa b σb a − θ(ℓ + µ)[f ] = Mf + Mf , (4.41)
8πG N

such that one has the usual spin-2 Hamiltonian

1
Z
(2) (0)
Mf := εN Ωf σa b σb a , (4.42)
8πG N

and
1 1
Z Z
(0) (0) (0)
Mf := − εN Ωθ(ℓ + µ)[f ] = εN (∂v2 Ω − ∂v Ωµ)f. (4.43)
8πG N 8πG N

We wish to compute the brackets of these spin-0 and spin-2 contributions. Instead of
simply reporting the final result, it is instructive to present here a detailed derivation. This
serves as an illustrative example of the kind of manipulations needed in this section. We
start with the spin-2 piece. Using that {Ω1 , ζ2 } = 0 = {Ω1 , ζ 2 }, we get

1
Z Z
(2) (0) (0)
{Mf , Mg(2) } = ε ε f1 Ω1 g2 Ω2 {σb1a σa1b , σd2c σc2d } (4.44)
(8πG)2 N1 N1 N2 N2
1
Z Z
(0) (0)
= ε ε f1 Ω1 g2 Ω2 σb1a {σa1b , σc2d }σd2c . (4.45)
(4πG)2 N1 N1 N2 N2

Inserting (4.39), and using that σa b ϵb a = 0, we process this expression into the form

1
Z Z 
(2) (0) (0)
{Mf , Mg(2) } = ε N1 εN2 f1 Ω1 g2 Ω2 σb1a m1a mb1 m2c md2 Dv1 Dv2 P 12
4πG N1 N2

+ m1a mb1 m2c md2 Dv1 Dv2 P12 σd2c (4.46)
1
Z Z  
(0) (0)
= ε N1 εN2 f1 Ω1 g2 Ω2 ∆v1 ∆v2 Dv1 Dv2 P 12 + ∆v1 ∆v2 Dv1 Dv2 P12 .
4πG N1 N2
(4.47)
16
In this section, since we are interested in the constraints, we are not considering corner contributions.
Therefore, M ′ = M , but we are working in the framework of subsection 3.3, and thus on the field space
spanned by L′ . We also assume that the flux vanish, hence f |C .

– 20 –
Then, using (4.30), (4.31), and their conjugates, we obtain
1
Z  
(2) (0)
{Mf , Mg(2) } = εN Ω Dv (f ∆v )g∆v −f ∆v Dv (g∆v )+Dv (f ∆v )g∆v −f ∆v Dv (g∆v )
8πG N
1
Z Z  
(0) (0)
+ ε N1 εN2 f1 ∂v1 Ω1 g2 ∂v2 Ω2 ∆v1 ∆v2 P 12 +∆v1 ∆v2 P12 (4.48)
16πG N1 N2
1
Z
(0)
=− ε Ω(f ∂v g−g∂v f )∆v ∆v
4πG N N
1
Z Z  
(0) (0)
+ ε N1 εN2 f1 ∂v1 Ω1 g2 ∂v2 Ω2 ∆v1 ∆v2 P 12 +∆v1 ∆v2 P12 (4.49)
16πG N1 N2
(0) f1 ∂v1 Ω1 g2 ∂v2 Ω2
Z Z  
(2) (0)

JHEP01(2024)166
= −Mf ∂v g−g∂v f + ε N1 ε N2 ∆v1 ∆v2 P 12 +∆v1 ∆v2 P12 ,
N1 N2 16πG
(4.50)

where we also used σa b σb a = 2∆v ∆v .

We conclude that the spin-2 algebra does not close. Comparing the last term with (4.33),
we rewrite the result as
1
Z Z
(2) (2) (0) (0)
{Mf , Mg(2) } = −Mf ∂v g−g∂v f + ε ε f1 ∂v1 Ω1 g2 ∂v2 Ω2 {µ1 , µ2 }. (4.51)
(8πG)2 N1 N1 N2 N2
This proves that the non-closure of the spin-2 sector is entirely due to the non-commutativity
of the field µ, that one can trace back to the coupling of the spin-0 and spin-2 fields in
the presymplectic form.
Similarly, for the spin-0 sector, we have
1
Z Z
(0) (0) (0)
{Mf , Mg(0) } = ε ε f1 g2 {∂v21 Ω1 −∂v1 Ω1 µ1 , ∂v22 Ω2 −∂v2 Ω2 µ2 } (4.52)
(8πG)2 N1 N1 N2 N2
(0)
εN ∂ v Ω (0) f1 g2
Z Z Z
(0)
f ∂v2 g−g∂v2 f +


= εN 1 εN 2 {∂v1 Ω1 µ1 , ∂v2 Ω2 µ2 }
N 8πG N1 N2 (8πG)2
(4.53)
1 1
Z Z Z
(0) (0) (0)
=− ε (f ∂v g−g∂v f ) ∂v2 Ω+ ε ε f1 g2
8πG N N (8πG)2 N1 N1 N2 N2
 
× µ1 ∂v2 Ω2 {∂v1 Ω1 , µ2 }+∂v1 Ω1 {µ1 , ∂v2 Ω2 }µ2 +∂v1 Ω1 ∂v2 Ω2 {µ1 , µ2 } (4.54)
1
Z
(0)
εN (f ∂v g−g∂v f ) ∂v2 Ω−∂v Ωµ


=−
8πG N
1
Z Z
(0) (0)
+ ε ε f1 g2 ∂v1 Ω1 ∂v2 Ω2 {µ1 , µ2 } (4.55)
(8πG)2 N1 N1 N2 N2
1
Z Z
(0) (0) (0)
= −Mf ∂v g−g∂v f + ε ε f1 g2 ∂v1 Ω1 ∂v2 Ω2 {µ1 , µ2 }. (4.56)
(8πG)2 N1 N1 N2 N2
The same bi-local contribution seen in the spin-2 sector appears here as well, and thus also
the spin-0 sector does not close.
As expected, but nevertheless quite remarkably, the mixed bracket contains exactly this
extra term. Utilizing similar manipulations as those used for the spin-0 algebra, one arrives at17
1
Z Z
(2) (0) (0)
{Mf , Mg(0) } =− ε N1 εN2 f1 g2 ∂v2 Ω2 ∂v1 Ω1 {µ1 , µ2 }, (4.57)
(8πG)2 N1 N2
17 (0) (2) (2) (0)
By symmetry, one can also show {Mf , Mg } = {Mf , Mg }.

– 21 –
such that the algebra of the full constraints properly closes
(2) (0) (2) (0)
{Mf′ , Mg′ } = {Mf + Mf , Mg(2) + Mg(0) } = −M(f ∂v g−g∂v f ) − M(f ∂v g−g∂v f ) = −M(f
′
∂v g−g∂v f ) .
(4.58)
So the split into the spin-0 and spin-2 part is a delicate procedure. We expect that this split
can be consistently done perturbatively in G. We are thus unveiling here a non-perturbative
effect mixing graviton propagation and the underlying geometry of null hypersurfaces. A full
quantum theory would therefore be complete only when keeping the spin-0 geometric data.
Indeed, one should think of the spin-0 sector (as well as the disregarded spin-1 sector) as
completing the extension of the phase space of the theory such that a proper representation

JHEP01(2024)166
of the constraint algebra is obtained, and the constraints act canonically on the fields, as
we now show.
Indeed, before concluding this section we display the action of the spin-2 and spin-0
Hamiltonians on our fields. A long yet straightforward computation gives
1
Z
(2) (0)
{Mf , µ(x2 )} = εN1 f (x1 )∂v1 Ω(x1 ){µ(x1 ), µ(x2 )}, (4.59)
8πG N1

while
1
Z
(0) (0)
{Mf , µ(x2 )} = ∂v22 f (x2 ) + ∂v2 (µ(x2 )f (x2 )) − εN1 f (x1 )∂v1 Ω(x1 ){µ(x1 ), µ(x2 )},
8πG N1
(4.60)
such that
(0) (2)
{Mf′ , µ(x2 )} = {Mf + Mf , µ(x2 )} = ∂v22 f (x2 ) + ∂v2 (µ(x2 )f (x2 )). (4.61)

Not only this is an important consistency check (see (3.35)), it also shows how both the
spin-0 and spin-2 Hamiltonians are needed to recover the correct µ-transformation.
This is not the case for Ω, on which only the spin-0 Hamiltonian acts non-trivially
(2)
{Mf , Ω(x2 )} = 0, (4.62)
(0)
{Mf , Ω(x2 )} = f (x2 )∂v2 Ω(x2 ), (4.63)

which again shows that the constraint acts canonically (3.35).

5 Time and boost operator

In the preceding discussion, we have used a kinematical coordinate v to parameterize null

curves along the Carroll structure, and we have seen that the kinematic Poisson brackets
involve the spin-0 fields in a non-trivial way. Their inclusion has the important effect of giving
a proper representation of the constraint algebra. In this section, we show that time can be
promoted to a dynamical variable conjugate to the Raychaudhuri constraint. We focus on a
particular choice of time, called the dressing time, where the time variable is constructed
as the holonomy of the surface tension. In this frame the spin-0 data are relegated to the
corner. Furthermore, the constraint can be solved non-perturbatively on sections of the null
surface where the expansion doesn’t vanish.

– 22 –
5.1 Time operator
We begin by considering a diffeomorphism V : v → V (v, σ), combined with a local boost
such that ℓ remains equal to ∂v . The phase space variables (Ω, µ, qab , σab ) can be written as
functions of V in terms of their pullbacks, which we formalize as follows,
∂v2 V
Ω = Ω̃ ◦ V, qab = q̃ab ◦ V, σab = ∂v V (σ̃ab ◦ V ), µ = ∂v V (µ̃ ◦ V ) + . (5.1)
∂v V
By the composition of maps we mean that Ω(v) = Ω̃ ◦ V (v) = Ω̃(V (v)), etc. We note that µ
transforms non-linearly: this is due to the fact that it is a boost connection, and as noted
above, we are considering a combined diffeomorphism and internal boost.18 Its transformation

JHEP01(2024)166
ensures that if O is a tensor of boost weight s such that O = (∂v V )s (Õ ◦ V ) then

(∂v − sµ)O = (∂v V )s+1 [(∂V − sµ̃)Õ] ◦ V, (5.2)

is a tensor of boost weight s + 1. In this equation we use that everything is happening locally
in σ. For instance, for a boost weight-0 tensor O, we have that O(v, σ) = Õ(V (v, σ), σ) is
simply denoted Õ ◦ V and by ∂V we mean the derivative keeping σ fixed.19 So in the following
we regard V = V (v) as a shortcut notation. We note that the Raychaudhuri constraint
C = ∂v2 Ω − µ∂v Ω + Ω(σa b σb a + 8πG Tvvmat ) is of weight two and thus satisfies

C = (∂v V )2 C̃ ◦ V (5.3)

where C̃ = ∂V2 Ω̃ − µ̃∂V Ω̃ + Ω̃(σ̃a b σ̃b a + 8πG T̃Vmat

V ).
The non-linear transformation of µ means that it is possible to choose the time variable
V such that µ̃ = 0, or in other words
∂v2 V
µ= . (5.4)
∂v V
This should be read as trading the dynamical variable µ for V . The quantity V can be
thought of as a “rest frame” time for time itself where the boost connection vanishes. In the
following we refer to V satisfying (5.4) as the dressing time. It is important to appreciate
that the choice (5.4) corresponds to employing a dynamical field variable as a clock.
This relation defines V only up to an affine transformation V → AV + B, which means
that there are residual symmetries preserving µ̃ = 0. These are the BMSW transformations
discussed in section 3.4. Eq. (5.4) is easily integrated in terms of two pieces of data, the values
of V (v, σ) and ∂v V (v, σ) at particular values of v, which we write as a, b. The general solution is
Z v Z v′ !
′ ′′ ′′
V(a,b) (v, σ) = V (a, σ) + ∂v V (b, σ) dv exp dv µ(v , σ) . (5.5)
a b

We note that whereas µ transforms as a connection under the internal boost symmetry, V
transforms as a Goldstone field. The original field space variables are dressed version of
18
In (5.1), we are giving a finite transformation. Thus the given non-linear shift in the transformation of
µ corresponds to a finite transformation of the form λ−1 V ∂v λV , with λV = ∂v V . This is an instance of the
combined ℓ-diffeomorphism/boost whose infinitesimal form was given in (3.33).
19
Of course when we change frame we also have to change the tangential derivative ∂A O = (∂˜A Õ) ◦ V , where
˜
∂A = ∂A − ∂A V ∂v is the tangential derivative along V =cst.

– 23 –
the tilded variables σ̃ab , etc., which are therefore invariant under diffeomorphism. From
this expression we see that ∂v V = exp ( bv µ(v ′ , σ)dv ′ ) is simply a Wilson line for the boost
R

connection. It is invertible unless µ = ±∞. Suppose we extend our construction through

a caustic, where θ → ±∞. If κ remains finite, one has that µ → ±∞, which is indeed
this singular point in the Jacobian.
An analogous transformation is routinely applied in the case where the expansion vanishes
such as on a Killing horizon. In this case the dressing time V is equal to the affine time λ,
because µ = κ, and thus κ = ∂v2 λ/∂v λ, which using notation similar to the above corresponds
to κ̃ = κaff = 0. Conversely, for an expanding surface, since µ = κ + 2θ , the surface tension
differs from the inaffinity and the two times are different. We can express that V = Vaff ◦ λ

JHEP01(2024)166
where Vaff (λ, σ) is the relative time given by20
Z λ q
Vaff (λ, σ) = A dλ′ Ωaff (λ′ , σ) (5.6)
b

where A, b are constants and we denote Ω = Ωaff ◦ λ. This follows simply from the chain
rule,21 which gives
!
∂v2 V ∂2λ ∂λ2 Vaff
= v + ∂v λ ◦λ (5.7)
∂v V ∂v λ ∂λ Vaff
∂2 V
together with the fact that µaff = ∂λλ Vaff
aff
= 12 θaff .
We conclude this discussion by noticing that the Raychaudhuri equation can be solved
non-perturbatively. Solving Raychaudhuri by choosing the area element Ω : v → Ω(v) as a
time-coordinate. This is possible if one assumes that the expansion ∂v Ω does not vanish.22
We can express this areal clock relative to the dressing clock as

V = V̄ ◦ Ω, (5.8)

such that
!
∂2V ∂2Ω 2 V̄
∂Ω
µ = v = v + ∂v Ω ◦ Ω. (5.9)
∂v V ∂v Ω ∂Ω V̄
We can then add and subtract the Raychaudhuri constraint to write
! !
∂2Ω ∂ 2 V̄ C mat )
Ω(σ 2 +8πGTvv ∂ 2 V̄
µ = v +∂v Ω Ω ◦Ω = µ+ − +∂v Ω Ω ◦Ω. (5.10)
∂v Ω ∂Ω V̄ ∂v Ω ∂v Ω ∂Ω V̄

Denoting the shear and the matter stress tensor in the areal frame respectively by (σ̄, T̄ mat )
mat = (∂ Ω)2 T̄ mat ◦ Ω, we see that on-shell this reduces to
where σ = ∂v Ω(σ̄ ◦ Ω) and Tvv v ΩΩ
!
2 mat ∂ 2 V̄
−Ω(σ̄ + 8πGT̄ΩΩ ) + Ω ◦ Ω = 0, (5.11)
∂Ω V̄
20
The square root appears in 4 bulk dimensions. In spacetime dimension d we have Vaff (λ) =
Rλ d−3
′ ′
A b
dλ [Ωaff (λ )] d−2 instead.
21
Explicitly, ∂v (Vaff ◦ λ) = ∂v λ(∂λ Vaff ) ◦ λ.
22
When the expansion vanishes, solving Raychaudhuri imposes that gravitational and matter states are in
their vacuum, if we assume positivity of energy.

– 24 –
which integrates to
Z Ω
∂Ω V̄ = exp dΩ′ Ω′ (σ̄ 2 + 8πGTΩΩ
mat
)(Ω′ ). (5.12)
b

This is a non-perturbative solution of the Raychaudhuri equation. We note that in the

absence of shear, the exponential’s argument is exactly the boost Hamiltonian, where the
right-hand side is the matter ANEC operator in the areal time.

5.2 Presymplectic structure in dressing time

The significance of the dressing time can be seen more clearly by considering the presymplectic

JHEP01(2024)166
structure. We are interested here in the general case, where both gravity and matter are
present. To evaluate the presymplectic structure we compute the variation of the phase
space variables in the dressing frame, where µ̃ = 0,

∂ 2 δV ∂v δV ∂v δV
 
δµ = v −µ = ∂v , (5.13)
∂v V ∂v V ∂v V
δΩ = (δ Ω̃) ◦ V + (∂V Ω̃ ◦ V )δV, (5.14)

δqab = (δ q̃ab ) ◦ V + (∂V q̃ab ◦ V )δV, (5.15)

δσab = (δσ̃ab ◦ V )∂v V + ∂v ((σ̃ab ◦ V )δV ). (5.16)

As a simple model of matter, we consider a real scalar field with presymplectic struc-
ture on N
Z
mat (0)
Ω = εN δ(Ω∂v φ) ∧ δφ, (5.17)
N

and energy-momentum Tvv mat = (∂ φ)2 . The discussion could be easily extended to an
v
arbitrary matter sector. The matter fields can be expressed with respect to the dressing
clock in a similar fashion to the above,

φ = φ̃ ◦ V, ∂v φ = ∂v V (∂V φ̃) ◦ V. (5.18)

We define the total presymplectic structure

Ωtot = Ωcan + Ωmat , (5.19)

where Ωcan is the canonical 2-form (3.36). From the transformations listed above we can
establish that it can be rearranged as a sum of a bulk and a corner term, in the dressing
frame. We relegate details of this computation to appendix D, and display here the final
result. The bulk term is given by
1
Z    
(0)
Ωtot
N = εN ∂ v V δ(Ω̃σ̃ ab ) ∧ δ q̃ab + δ Ω̃∂V φ̃ ∧ δ φ̃ ◦ V
N 16πG
1
Z  
(0)
+ εN δ ∂v V [C̃ ◦ V ] ∧ δV, (5.20)
8πG N

where  
C̃ = ∂V2 Ω̃ + Ω̃ σ̃ 2 + 8πG (∂V φ̃)2 (5.21)

– 25 –
is the total Raychaudhuri constraint in the dressing frame. We see as expected that the
spin-0 term has disappeared from the bulk presymplectic form and has been replaced by
the pairing between the dressing time and the constraint.
On the other hand, the corner term is given by
1 ∂v δV ∂v Ω
Z   
(0)
Ωtot
C =− ∧ δΩ − δV ∧ δ
εC
8πG C ∂v V ∂v V
!
Ω̃
Z
(0)
+ εC δV ∧ σ̃ ab δ q̃ab + Ω̃∂V φ̃δ φ̃ ◦ V. (5.22)
C 16πG

The terms in the second line come from spin-2 and matter fields. They are the flux terms

JHEP01(2024)166
tot ◦ V , with θ̃ tot = θ̃ (2) + θ̃ mat = ε(0) ( Ω̃ σ̃ ab δ q̃ + Ω̃∂ φ̃δ φ̃). These are the
R
C δV ∧ ι∂v θ̃ N 16πG ab V
flux terms introduced in (3.6), Ff = C f ι∂v θtot .
R

It is instructive for the computation of the charge to decompose this off-shell symplectic
structure in a sum of symplectic terms attached to the spin-0, spin-2 and matter contributions.
There is a bulk split
(0) (2)
Ωtot mat
N = ΩN + Ω N + Ω N , (5.23)

and a corner split

(0) (2)
Ωtot mat
C = ΩC + Ω C + Ω C . (5.24)

Splitting likewise the Raychaudhuri constraint C̃ = C̃ (0) + C̃ (2) + C̃ mat , with

C̃ (0) = ∂V2 Ω̃, C̃ (2) = Ω̃ σ̃ 2 , C̃ mat = 8πG Ω̃(∂V φ̃)2 , (5.25)

we gather
1
Z  
(0) (0)
ΩN = εN δ ∂v V C̃ (0) ◦ V ∧ δV (5.26)
8πG N
1
Z   Z
(2) (0) (2)
ΩN = ε δ ∂v V C̃ ◦ V ∧ δV + ∂v V δ θ̃(2) ◦ V (5.27)
8πG N N N
1
Z   Z
(0)
Ωmat
N = ε δ ∂v V C̃ mat
◦ V ∧ δV + ∂v V δ θ̃mat ◦ V. (5.28)
8πG N N N

We thus see that the bulk part of the presymplectic structure is the same for the spin-2
and matter sectors, while the spin-0 sector is different. This is also the case for the corner
part of the presymplectic structure
1 ∂v δV ∂v Ω
Z   
(0) (0)
ΩC =− εC ∧ δΩ − δV ∧ δ (5.29)
8πG C ∂v V ∂v V
Z
(2)
ΩC = δV ∧ ι∂v θ̃(2) ◦ V (5.30)
ZC
Ωmat
C = δV ∧ ι∂v θ̃mat ◦ V. (5.31)
C

This is the main result of this section. In the dressing time, the presymplectic structure
in the spin-2 and matter sectors displays a pattern where the corner contribution is given
entirely by the canonical flux. As already commented upon, the spin-0 sector is peculiar: it

– 26 –
has no bulk potential, θ̃(0) = 0, and on the corner it gives rise to the charge. The dressing
time is therefore guaranteeing a split between the spin-0 sector and the rest.
In an arbitrary frame, the spin-0 data µ̃ contributes in the bulk. Indeed, one can just
keep all terms containing µ̃ in the previous computation. One finds (see appendix D), apart
from the additional term involving µ̃ appearing in the Raychaudhuri constraint C̃ and the
1 (0)
contribution to the canonical flux C δV ∧ ι∂v θ̃(0) ◦ V with θ̃(0) = − 8πG
R
εN µ̃δ Ω̃, simply an
additional term of the form
1
Z
(0)
− εN ∂v V (δ µ̃ ∧ δ Ω̃) ◦ V. (5.32)
8πG N

JHEP01(2024)166
As announced, this means that the symplectic pair (µ̃, Ω̃) remains dynamical in the bulk in an
arbitrary frame. Thus, to reiterate, the dressing time is special in that the spin-0 contribution
to the symplectic structure is entirely given by the dressing time and the constraint C̃. In
other words, in dressing time there are no spin-0 conjugate pairs in the bulk, on shell. This
result has far reaching consequences: it implies, as we are about to see, that the canonical
boost charge in the dressing time is monotonic.

5.3 Boost operator

Now that we have introduced the time V and the corresponding dressed fields (q̃ab , σ̃ab ), it is
interesting to redo the canonical analysis in their terms. First, the dressing time transforms
under infinitesimal diffeomorphisms simply as

L′fˆV = f ∂v V. (5.33)

This transformation implies (3.35). Then, the dressed variables are by construction gauge
invariant23
L′fˆΩ̃ = 0, L′fˆq̃ab = 0, L′fˆσ̃ab = 0, L′fˆφ̃ = 0. (5.34)

The proof of this follows from the fact that, on one hand, we know (see (3.34)) that
L′fˆqab = f ∂v qab , while on the other hand we have

L′fˆqab = L′fˆ(q̃ab ◦ V ) = (L′fˆq̃ab ) ◦ V + (∂V q̃ab ) ◦ V L′fˆV. (5.35)

Under the transformations (5.33), (5.34), this expression gives f ∂v (q̃ab ◦ V ). A similar analysis
can be done for the other fields.
One can now evaluate the charge. To do so, we use that L′fˆ[∂v V C̃ ◦ V ] = ∂v [f ∂v V C̃ ◦ V ].
Let us show the result sector by sector, starting from the bulk contributions

1 1
Z  Z
(0) (0) (0)
If′ˆΩN = − δ εN f (∂v V )2 C̃ (0) ◦V
+ εC f ∂v V δV C̃ (0) ◦V (5.36)
8πG 8πG C
ZN
1 1
 Z
′ (2) (0) 2 (2) (0)
IfˆΩN = − δ εN f (∂v V ) C̃ ◦V + εC f ∂v V δV C̃ (2) ◦V (5.37)
8πG 8πG C
ZN
1 1
 Z
′ mat (0) 2 mat (0)
IfˆΩN = − δ εN f (∂v V ) C̃ ◦V + εC f ∂v V δV C̃ mat ◦V, (5.38)
8πG N 8πG C
23
In an arbitrary frame, one also has L′fˆµ̃ = 0.

– 27 –
and the corner contributions (we use (5.4), and process the spin-0 part)
Z 
1 1
Z  
(0) (0) (0)
If′ˆΩC =− 2
ε f µδΩ+∂v V [∂V Ω̃◦V ]δV − δ εC (Ω∂v f −f ∂v Ω) (5.39)
8πG C C 8πG C
Z
(2)
If′ˆΩC = f ∂v V [ι∂v θ̃(2) ◦V ] (5.40)
ZC
′ mat
IfˆΩC = f ∂v V [ι∂v θ̃mat ◦V ]. (5.41)
C

For the spin-2 and matter parts, an explicit computation shows that the corner contributions
coming from the bulk constraint combine with the dressing-time flux to produce the full

JHEP01(2024)166
canonical flux
1 (0) (2)
 
∂v V ε [C̃ ◦ V ]δV + ι∂v θ̃(2) ◦ V = ι∂v θ(2) , (5.42)
8πG C
1 (0) mat
 
∂v V εC [C̃ ◦ V ]δV + ι∂v θ̃mat ◦ V = ι∂v θmat . (5.43)
8πG
On the other hand, the same procedure for spin-0 gives rise to the canonical charge:
1 1
Z Z Z 
(0) (0) (0)
εC f ∂v V [C̃ (0) ◦V ]δV +If′ˆΩC = f ι∂v θ(0) − δ εC (Ω∂v f − f ∂v Ω) . (5.44)
8πG C C 8πG C

1 (0)
In these expressions, we have introduced the symplectic potentials θ(0) = − 8πG εN µδΩ,
(0)
Ω (0)
θ(2) = εN 16πG σ ab δqab , and θmat = εN Ω∂v φδφ.
To recap, we have found that each sector contributes as
1 1
Z  Z Z 
(0) (0)
If′ˆΩ(0) = − δ εN f C (0) + f ι∂v θ(0) − δ εC (Ω∂v f −f ∂v Ω) (5.45)
8πG N C 8πG C
1
Z  Z
(0)
If′ˆΩ(2) = − δ εN f C (2) + f ι∂v θ(2) (5.46)
8πG
ZN  ZC
′ mat 1 (0) mat
IfˆΩ = − δ εN f C + f ι∂v θmat , (5.47)
8πG N C

where we denote C (i) = (∂v V )2 C̃ (i) ◦ V , see (5.3). While the constraint is typically understood
as a classical conservation equation for intrinsic data on the null geometry, we see here that
it can alternatively be thought of as the vanishing of the sum of stress tensors from each
sector. Indeed, the three CFT-like stress tensors

C (0) = ∂v2 Ω − µ∂v Ω C (2) = Ωσa b σb a C mat = 8πG ΩTvv

mat
, (5.48)

satisfy the Raychaudhuri constraint

C (0) + C (2) + C mat = 0. (5.49)

This interpretation is clear for the matter sector (5.47), and for the spin 2 sector (5.46), but
it can also be applied to the gravitational spin-0 sector as shown in (5.45). We know that
the constraint vanishing corresponds to the gauging of diffeomorphism invariance. What
is interesting is that we now understand this gauging as a balancing between the spin-0
stress tensor and the matter like degrees of freedom (matter + spin-2). Notice that both

– 28 –
the spin-2 and matter sector are manifestly positive stress tensors. On the other hand, the
spin-0 stress tensor is not.
So we see here that the main feature of the dressing frame is to split completely the charge
contributions of the different sectors. The spin-2 and the matter sector do not contribute
to the charge, but they generate the bulk constraints and the corner flux. On the other
hand, the spin-0 sector has a multipurpose function. It contributes to the flux at the corner,
plus it generates the canonical charge. In an arbitrary time frame, the spin-0 sector would
appear also in the bulk, and the sectors would be mixed.
By construction, once regrouping all terms together, the final result is

JHEP01(2024)166
If′ˆΩtot = −δMf′ + Ff , (5.50)

f ι∂v θtot the flux, θtot = θ(0) + θ(2) + θmat and

R
with Ff = C
1 1
Z Z
(0) (0)
Mf′ = εN f C + εC (Ω∂v f − f ∂v Ω). (5.51)
8πG N 8πG C

This is simply the result of section 3.3. The boost charge depends both on a choice of cut
C and a vector field f (v, σ)∂v as given by (5.51). The function f appearing in the vector
field transforms as a scalar in our phase space. Indeed, consider two vector fields ξ = f ∂v
and ζ = h∂v , we have (Lλ̂h f = 0)

L′ζ̂ ξ = (Lĥ + Lλ̂h )ξ = h∂v f ∂v . (5.52)

This proves that f is a scalar on our phase space, and thus f = f˜ ◦ V . Using this, we can
write the charge in the dressing time as
1
Z  
(0)
Mf′˜(C) =
ˆ ε̃C Ω̃∂V f˜ − f˜∂V Ω̃ , (5.53)
8πG C
(0) (0)
where we denote ε̃C ◦ V = εC ∂v V the measure in dressing time, and we have now made
explicit that this charge depends on the chosen cut C.
We now specialize this general charge for the boost generator f˜T (V ) = V − T , where the
parameter T (σ) is a function on S. Calling Mf′˜ (C) = MT′ (C), we get
T

1
Z Z
(0) (0)
MT′ (C) =
ˆ ε̃C (Ω̃ − (V − T )∂V Ω̃) = ε̃C S̃T , (5.54)
8πG C C

where we defined the boost charge aspect,

1
S̃T (V ) :=(Ω̃(V ) − (V − T )∂V Ω̃(V )), (5.55)
8πG
and we made explicit that it depends on the symmetry parameter T , and the time V at which
it is evaluated. We recall that the choice of a time function V0 is equivalent to specifying the
cut CV0 = {V = V0 }. So evaluating S̃T (V ) at a specific time V = V0 is the same as evaluating
it at the cut CV0 . We then see from the previous expression that the boost charge evaluated
at the specific cut C = CT is the area in the dressing time, on-shell
1
Z
(0)
MT′ (CT ) =
ˆ ε̃C Ω̃. (5.56)
8πG C

The positivity of this charge is therefore guaranteed by construction.

– 29 –
 Since the following discussion is true per point on C, we focus on the boost charge aspect.
Consider then its time derivative
1
ˆ−
∂V S̃T (V ) = (V − T )∂V2 Ω̃(V ). (5.57)
8πG
In the dressin g time where µ̃ = 0, we can use the Raychaudhuri constraint (5.21) to get
the flux formula for the boost
1
∂V S̃T (V ) =
ˆ (V − T )Ω̃(V )(σ̃ 2 + 8πGT̃Vmat
V )(V ). (5.58)
8πG

JHEP01(2024)166
The key point is that in the dressing time, there is no spin zero contribution to the right-
hand side.
Assuming that the boost is evaluated in the future V ≥ T of the cut, this leads to two
important results: first, one sees that if the radiation is absent and the matter is in its vacuum
state, then the flux vanishes and the charge is conserved, even if the null surface is expanding,
solving therefore a longstanding puzzle; see e.g., the recent works [104, 105]. This means
that the dressing time charge is a proper notion of energy that is conserved not only on
non-expanding horizons but also for expanding but non-radiative null hypersurfaces. Second,
we see that if we impose the classical null energy condition on the matter fields T̃Vmat V ≥ 0,
we obtain monotonicity of the boost charge aspect on the entire phase space

∂V S̃T (V ) ≥ 0. (5.59)

The bound is saturated by the shearless configuration, and when matter is in its vacuum
state. Notice that this result is intimately tied to the choice of dressing time, where µ̃ = 0.
Indeed, it is only in this frame that the Raychaudhuri equation ensures the inequality written
above, since in any other frame there would be an additional spin zero contribution to the
stress tensor given by −µ∂v Ω, which is not manifestly positive. The boost charge aspect thus
provides a notion of entropy that satisfies positivity and monotonicity even in the presence of
expansion. Consider then the difference between the boost aspects S̃T (V ) and S̃T (T ). We find

1  
∆S̃ = ∆Ω̃ − (V − T )θ̃(V ) , (5.60)
8πG
where we employed the notation ∆F = F (V ) − F (T ) for V ≥ T , to denote the difference
of fields evaluated at different cuts CV and CT .
When the second cut CV is expansion free, the difference in the boost charges is the
1
area difference, ∆S̃ = 8πG ∆Ω̃. This is the key mechanism at play in the construction of a
local entropy formula for Black Hole horizons as given in [106] (see also [107]), where the
second cut is taken to be at V = ∞ and the definition of an event horizon implies that
the asymptotic expansion vanishes. In general, the expansion at the second cut contributes
to (5.60) and is subtracted from the area contribution.
As already remarked, the dressing time coincides with the affine time for marginally
trapped surfaces, where the expansion vanishes. An analysis of positivity of the boost charge
has been recently performed in [108]. For arbitrary null hypersurfaces, our result provides a
generalization of the boost charge that ensures its positivity and monotonicity. The latter is

– 30 –
important for strong subadditivity of entropy, if one can interpret this charge as the modular
Hamiltonian of the system. Testing the properties of this charge when perturbative quantum
matter is added, one could also investigate whether this charge is a good candidate to describe
the generalized entropy [90]. We plan to focus on this in the future, and study whether this can
be used to generalize the quantum focusing conjecture [91–93] to arbitrary null hypersurfaces.

6 Conclusions and discussion

In this paper, we have constructed a very general account of the classical dynamics of the
geometric data on an arbitrary null geometry, which can be interpreted as a null hypersurface

JHEP01(2024)166
in some classical spacetime. As such, we focus on the presymplectic canonical potential
defined on the null geometry, which takes the general form (3.1). This presymplectic form
contains the expected spin-2 degrees of freedom consisting of a metric q on the base of the
Carroll structure and the shear σ. In addition, there are spin-0 modes Ω and µ, as well as
spin-1 modes that we have not elaborated upon in this paper.
The spin-2 modes are usefully rewritten in terms of Beltrami differentials. In those terms
it is possible to invert the symplectic form to obtain canonical kinematic Poisson brackets on
phase space. As detailed in section 4.3, the result is an intricate pattern that is local on the
base and generally extended non-locally in the null fibre direction. The brackets as given
demonstrate that there is non-trivial mixing between the spin-0 and spin-2 sectors. This
mixing is crucial — in fact we have shown explicitly that were it not for this mixing, the
constraint algebra would not be properly represented on phase space.
In addition to local diffeomorphisms, there is an internal boost symmetry which gives
rise to a corner charge, its constraint vanishing identically. We focused in this paper on the
diffeomorphisms along the null fibre, and one finds that there exists a combination of them
and internal boosts leaving the fibre ℓ invariant. The corresponding charge is

1 1
Z Z
(0) (0)
Mf′ = εN f C + εC (Ω∂v f − f ∂v Ω) , (6.1)
8πG N 8πG C

where
C = ∂v2 Ω − µ∂v Ω + Ω(σa b σb a + 8πG Tvv
mat
)

is the Raychaudhuri constraint. In the above discussion, we showed how to solve the equation
C = 0 non-perturbatively. It is interesting to note that the constraint has contributions
from three terms coming from the spin-0, spin-2 and matter sectors, each of which has been
established to be a stress tensor density generating null-time reparametrizations. It seems
natural to reinterpret the constraint then, not as a conservation law, but as the vanishing
of a total stress tensor, expressing the gauging of diffeomorphism invariance along the null
geometry. A natural possibility then arises, which is that at the quantum level, this constraint
is anomalous in some sectors. This is a future direction worth exploring.
While this is a satisfying result, the Poisson brackets are kinematic and correspondingly
the fields involved transform non-trivially under diffeomorphisms. The key to constructing
invariant observables is to introduce a dynamical clock variable, which we have shown in
section 5 to be implicit in the spin-0 sector. A useful way of thinking of the Carroll geometry

– 31 –
is that it defines a null congruence through each point of the base, and the dynamical variables
simply encode the properties of that congruence. In such terms, it would be a standard
assumption that one should use affine time to describe the null congruence. As we showed in
section 5, this however is only appropriate in the special case where the expansion vanishes
(such as on Killing horizons). In general a better notion of a dynamical clock is given by the
dressing time, in which the canonical variable µ is set to zero, by a canonical transformation.
This is possible precisely because of the presence of the internal boost symmetry which renders
µ as a connection — locally then, µ can be rewritten in terms of a dynamical time variable V .
This addresses precisely the oft-lamented fact that time is pure gauge in gravitational theories.
Remarkably, rewriting the theory in terms of this particular clock variable has several key

JHEP01(2024)166
properties. First, it is in fact canonically conjugate to the Raychaudhuri constraint on the
extended phase space. Second, it renders the other variables (spin-2 gravity modes as well as
arbitrary matter fields) invariant under local diffeomorphisms.
The reader may well be confused by this state of affairs. On the one hand we gave a
description in which the spin-0 fields were fully dynamical (in the sense of appearing off-shell
in the presymplectic form) throughout the bulk of the null geometry. On the other hand,
we have an equivalent description in which the spin-0 degrees of freedom are replaced by
a clock and constraint, while other dynamical fields are dressed. In fact, this should be
regarded as an expected structure: we see explicitly that solving the constraint, or more
precisely reducing to the constraint surface in phase space, corresponds to requiring the
clock to be dynamical variables determined by the spin 2 and matter degrees of freedom
while simultaneously dressing all other fields. We have also seen that the spin-0 fields make
important contributions to the codimension-2 corner charges. In the language of [8, 15, 16, 89],
we can interpret the original spin-0 formulation to mean that we are working on an extended
phase space, the significance of which the symmetries are represented by Hamiltonian vector
fields whose algebra of Hamiltonians closes. From the point of view of the extended phase
space, the clock and constraint are dynamical variables, satisfying canonical Poisson brackets.
By passing to the second formulation, we have effectively imposed the constraints in the
sense that all other fields are dressed. The precise sense in which this has happened though is
interesting. It was not done by setting the constraint to zero, but by performing a canonical
transformation in which the constraint becomes a dynamical variable conjugate to the clock.
In a quantum theory, the constraint commutes with the dressed operators, but does not
commute with the clock. This realizes on a general null geometry the idea that the constraints
evolve the system in time, as well known in the timelike ADM formulation of gravity. Thus,
interpreting a constraint operator as ‘zero’ is subtle and irrecovably tied up with what we
might want to think of as a notion of time in a diffeomorphism-invariant quantum theory.
We remark that to achieve this we had to include the surface tension µ from the beginning,
which is a fundamental ingredient of the theory, that we traded for the dressing time.
When specified to a boost, our new charge Mf′ has the interesting property of being posi-
tive and monotonic, with the inequality saturated by the classical non-radiating configuration.
Furthermore, given that the dressing time coincides with the affine time for non-expanding
horizons, it reduces to the usual boost charge used to study the laws of thermodynamics on
marginally trapped surfaces. In the future, we plan to investigate whether this charge operator
is a good candidate for a generalization of the quantum focusing conjecture to arbitrary null
hypersurfaces, and in particular its relationship to the modular Hamiltonian [19, 93, 109–111],

– 32 –
given the suggestive form of eq. (5.12). Finally let us remind the reader that a limitation of
our analysis is that it is valid away from creases and caustics. These should be thought of as
boundaries of the null regions we are studying here. It would be interesting to understand
better how the geometrical matching of null portions through these boundaries is achieved
in terms of the corner charges.
As advertised at various points of the manuscript, one of our principle motivations is to
study the quantization of the system. In particular, our results suggest that comparing the
various available pathways to quantization may yield important insights and perhaps surprises.
Another followup project is to extend the work to the Damour constraint, accounting for
the transverse diffeomorphisms. We expect that a similar structure will be present, with

JHEP01(2024)166
the Damour constraint conjugate to additional dressing variables playing the proper role of
observer devices. We expect to report on these topics soon.

Acknowledgments

This paper is the result of a focused research group held at BIRS, Banff, in November
2022, titled “Symmetries of Gravity at the Black Hole Horizon”. We are indebted to BIRS
for the warm hospitality and for creating such a stimulating environment. We thank Ted
Jacobson, Marc Klinger, Jerzy Kowalski-Glikman, and Aron Wall for discussions and feedback.
Research at Perimeter Institute is supported in part by the Government of Canada through the
Department of Innovation, Science and Economic Development Canada and by the Province
of Ontario through the Ministry of Colleges and Universities. LC is grateful to University
of Milano Statale (and in particular Antonio Amariti) and University of Trento (and in
particular Valter Moretti) for hospitality during the completion of this work. The work of
RGL is partially supported by the U.S. Department of Energy under contract DE-SC0015655,
and RGL thanks the Perimeter Institute for supporting collaborative visits.

A Inverse of the symplectic 2 form

To invert the canonical 2 form

1
Z
(0)
 √ √ 
Ωcan =


εN δΩ ∧ δµ + ∆v ∆ + ∆v ∆ + 2Dv ( Ω∆) ∧ Ω∆ , (A.1)
8πG N

we rewrite it as a bi-local expression

1
Z Z
(0) (0)
Ωcan = ε N1 εN2 δZ α (x1 )Ωcan β
αβ (x1 , x2 ) ∧ δZ (x2 ) (A.2)
16πG N1 N2

where Z α is the basis of our fields and x1 = (v1 , z1 , z 1 ).

Choosing Z α (x1 ) = (Ω(x1 ), µ(x1 ), ζ(x1 ), ζ(x1 )), we get

∆v1 ∆ v1
 
0 1 β1 β1
 
1  −1 0
 0 0 
Ωcan
 (3)
αβ (x1 , x2 ) =  ∆v √  δ (x1 − x2 ) (A.3)
8πG  Ω1 Ω 2
− 0 0 −2 β1 β2 Dv1 
2
β2 √ 
∆ v2 Ω1 Ω2
− β2 0 −2 β 1 β2 D v 1 0

– 33 –
where we have distributed the x1 and x2 dependency in the matrix, and the delta function
is defined via
Z
(0)
εN1 δ (3) (x1 − x2 )f (x1 ) = f (x2 ). (A.4)
N1

We want to find the distributional inverse. The delicate term is

√
can 1 Ω1 Ω2
Ωζζ (x1 , x2 ) = − Dv1 δ (3) (x1 − x2 ). (A.5)
4πG β1 β2

Its inverse is (δ (3) (x1 − x2 ) = δ(v1 − v2 )δ (2) (z1 − z2 ))

JHEP01(2024)166
R v2
β1 β2 −2 ωv dv
(Ωcan
ζζ
(x1 , x2 ))−1 = −4πG √ e v1 H(v1 − v2 )δ (2) (z1 − z2 ) = −4πGβ1 β2 P 12 ,
Ω1 Ω2
(A.6)
where H(−v) = −H(v) is the odd Heaviside step function satisfying

∂v1 H(v1 − v2 ) = δ(v1 − v2 ), (A.7)

and we see why the propagator, introduced in (4.28), is such a key object in this theory.
Using this, we can eventually inverse the matrix (A.3),

−2δ (3) (x1 − x2 )

 
0   0 0
 (3)
2δ (x1 − x2 ) − ∆v1 ∆v2 P 12 + ∆v1 ∆v2 P12 β2 ∆v1 P 12 β2 ∆v1 P12 

Ωαβ
can (x1 , x2 ) = 4πG 
 
.
 0 β1 ∆v2 P12 0 −β1 β2 P12 
0 β1 ∆v2 P 12 −β1 β2 P 12 0
(A.8)

With our conventions established in (3.14), one has Ωαβ α β

can (x1 , x2 ) = −{Z (x1 ), Z (x2 )},
and thus the non-vanishing Poisson brackets are

{Ω1 , µ2 } = 8πGδ (3) (x1 − x2 ) (A.9)

 
{µ1 , µ2 } = 4πG ∆v1 ∆v2 P 12 + ∆v1 ∆v2 P12 (A.10)
{µ1 , ζ2 } = −4πGβ2 ∆v1 P 12 (A.11)
{µ1 , ζ 2 } = −4πGβ2 ∆v1 P12 (A.12)
{ζ1 , ζ 2 } = 4πGβ1 β2 P12 (A.13)
{ζ 1 , ζ2 } = 4πGβ1 β2 P 12 . (A.14)

B Composite brackets

To explicitly construct brackets among the coframe elements, we exploit the fact that the
Poisson bracket is a derivation, and we introduce the pragmatic notation

{ζ(x1 ), ζ(x2 )} = {ζ1 , ζ 2 } = Lζ̂1 ζ 2 = Iζ̂1 δζ 2 . (B.1)

– 34 –
We then have, for instance,

{ζ1 , m2a } = Lζ̂1 m2a = Iζ̂1 δm2a = Iζ̂1 (∆2 m2a + ω2 m2a ) = Iζ̂1 ∆2 m2a + Iζ̂1 ω2 m2a . (B.2)

Using then

Iζ̂1 δζ2 = {ζ1 , ζ2 } = 0 Iˆ δζ 2 = {ζ 1 , ζ 2 } = 0 (B.3)

ζ1

Iˆ δζ2 = 4πGβ1 β2 P 12 Iζ̂1 δζ 2 = 4πGβ1 β2 P12 , (B.4)

ζ1

we compute

JHEP01(2024)166
Iζ̂1 δζ2 Iˆ δζ 2
ζ1
Iζ̂1 ∆2 = =0 I ˆ ∆2 = =0 (B.5)
β2 ζ1 β2
Iˆ δζ2 Iζ̂1 δζ 2
ζ1
I ˆ ∆2 = = 4πGβ1 P 12 Iζ̂1 ∆2 = = 4πGβ1 P12 (B.6)
ζ1 β2 β2
ζ2 Iζ̂1 δζ 2 −ζ 2 Iζ̂1 δζ2 ζ2 Iˆ δζ 2 −ζ 2 Iˆ δζ2
ζ1 ζ1
Iζ̂1 ω2 = = 2πGζ2 β1 P12 I ˆ ω2 = = −2πGζ 2 β1 P 12 .
2β2 ζ1 2β2
(B.7)

Thus we get

{ζ1 , m2a } = 2πGζ2 β1 P12 m2a {ζ1 , ma2 } = 2πGζ2 β1 P12 ma2 , (B.8)

and also
! !
ζ ζ
{ζ 1 , m2a } = 4πGβ1 P 12 m2a − 2 m2a {ζ 1 , ma2 } = −4πGβ1 P 12 ma2 + 2 ma2 . (B.9)
2 2

From these commutators one can derive their conjugates. For instance

{ζ 1 , m2a } = {ζ1 , m2a } = 2πGζ 2 β1 P 12 m2a . (B.10)

Then we have

Lm1a ζ2 = {m1a , ζ2 } = −{ζ2 , m1a } = −2πGζ1 β2 P21 m1a = 2πGζ1 β2 P 12 m1a , (B.11)

and so on.
Repeating the same steps as before for Im̂1a etc., we arrive to the coframe commutators.
We display the two fundamental blocks, while the others can be derived by conjugation
 
{m1a , mb2 } = πG −2ζ1 P 12 m1a mb2 +2ζ2 P12 m1a mb2 −ζ 1 ζ2 P12 m1a mb2 −ζ1 ζ 2 P 12 m1a mb2 (B.12)
 
{m1a , mb2 } = πG −4P 12 m1a mb2 +2ζ1 P 12 m1a mb2 +ζ 1 ζ2 P12 m1a mb2 −2ζ 2 P 12 m1a mb2 +ζ1 ζ 2 P 12 m1a mb2 .

From here, a lengthy yet straightforward computation leads to the commutators

{m1a mb1 , m2c md2 } = −4πGP12 m1a mb1 m2c md2 − 4πGP 12 m1a mb1 m2c md2 , (B.13)

{m1a mb1 , m2c md2 } = 4πGP 12 m1a mb1 m2c md2 + 4πGP12 m1a mb1 m2c md2 . (B.14)

– 35 –
 Given our parameterization, the projector q cb q ba = qc a is given by Diag(0, 1, 1), and
thus a good consistency check is that one obtains
{qa1 b , qc2 d } = 0 {ϵ1a b , qc2 d } = 0. (B.15)
On the other hand, one has
{q 1ab , q 2cd } = 16πGP 12 m1a m1b m2c m2d + 16πGP12 m1a m1b m2c m2d , (B.16)
proving that our degenerate metric satisfies a non-commutative geometric structure. This
is further confirmed by
{ϵ1a b , ϵ2c d } = −16πGP12 m1a mb1 m2c md2 − 16πGP 12 m1a mb1 m2c md2 ,

JHEP01(2024)166
(B.17)
which is the result reported in the main text.

C Shear bracket

We here show the bracket of the shear tensor, as given in (4.16).

A brute force computation yields
 
{m1a , ∆v2 } = 2πG ζ1 ∂v2 P 12 m1a −ζ1 ζ2 ∆v2 P 12 m1a +2ζ2 ∆v2 P12 m1a −ζ 1 ζ2 ∆v2 P12 m1a (C.1)
 
{m1a , ∆v2 } = 2πG 2∂v2 P12 m1a −ζ 1 ∂v2 P12 m1a −2ζ 2 ∆v2 P12 m1a +ζ 1 ζ 2 ∆v2 P12 m1a +ζ1 ζ 2 ∆v2 P 12 m1a ,

and
 
{∆v1 , ∆v2 } = 4πG −ζ1 ζ2 ∆v1 ∆v2 P 12 −ζ1 ζ2 ∆v1 ∆v2 P12 +ζ2 ∆v2 ∂v1 P12 +ζ1 ∆v1 ∂v2 P 12 (C.2)
 
{∆v1 , ∆v2 } = 4πG ∂v1 ∂v2 P 12 −ζ2 ∆v2 ∂v1 P 12 −ζ 1 ∆v1 ∂v2 P 12 +ζ 1 ζ2 ∆v1 ∆v2 P12 +ζ 1 ζ2 ∆v1 ∆v2 P 12 .

We have to evaluate
¯ v m1 mb , ∆v m2 md + ∆
{σa1b , σc2d } = {∆v1 m1a mb1 + ∆ ¯ v m2 md }
1 a 1 2 c 2 2 c 2

¯ v m1 mb , ∆v m2 md }
= {∆v1 m1a mb1 , ∆v2 m2c md2 } + {∆ (C.3)
1 a 1 2 c 2
1 b ¯ ¯ 1 b ¯
+ {∆v m m , ∆v m m } + {∆v m m , ∆v m2 md }
1 a 1
2 d
2 c 2 1 a 1 2 c 2

¯ v m1 mb , m2 md }
= ∆v2 {∆v1 m1a mb1 , m2c md2 } + {∆v1 m1a mb1 , ∆v2 }m2c md2 + ∆v2 {∆ 1 a 1 c 2
¯ 1 b 2 d ¯
+ {∆v m m , ∆v }m m + ∆v {∆v m m , m m } + {∆v m m , ∆v }m2 md
1 b 2 d 1 b ¯
1 a 1 2 c 2 2 1 a 1 c 2 1 a 1 2 c 2
¯ v {∆
+∆ ¯ v m1 mb , m2 md } + {∆
¯ v m1 mb , ∆
¯ v }m2 md . (C.4)
2 1 a 1 c 2 1 a 1 2 c 2

Putting together all these brackets, there are many cancellations, and the remaining non-
vanishing contributions are
  
{σa1b , σc2d } = 4πG m1a mb1 ϵ2d
c ∆v1 ∆v2 ζ 1 P12 + ∆v2 ∂v1 P12 − ∆v2 ζ1 ∆v1 P12
 
2 d
+ ϵ1b 1b 2d
a mc m2 ∆v1 ∆v2 ζ 2 P 12 + ∆v1 ∂v2 P 12 − ∆v1 ζ2 ∆v2 P 12 − ∆v1 ∆v2 P12 ϵa ϵc

+ m1a mb1 m2c md2 ∆v2 ζ 2 ∂v1 P 12 − ∆v2 ζ 2 ζ 1 ∆v1 P 12 + ∂v1 ∂v2 P 12
− ζ2 ∆v2 ∂v1 P 12 − ζ 1 ∆v1 ∂v2 P 12 + ζ 1 ζ2 ∆v1 ∆v2 P 12
 
+ ∆v1 ζ1 ∂v2 P 12 − ∆v1 ζ1 ζ2 ∆v2 P 12 + ∆v1 ∆v2 ζ1 ζ 2 P 12 + c.c. , (C.5)

– 36 –
where c.c. denotes complex conjugation.24 Using the relation 2ωv = ζ∆v − ζ∆v , and recalling
Dv = ∂v − 2ωv and Dv = ∂v + 2ωv , we obtain

{σa1b , σc2d } = 4πG m1a mb1 ϵ2d 1b 2 d 1b 2d
c ∆v2 Dv1 P12 + ϵa mc m2 ∆v1 Dv2 P 12 − ∆v1 ∆v2 P12 ϵa ϵc

+ m1a mb1 m2c md2 Dv1 Dv2 P 12 + C.C. . (C.6)

D Dressing frame symplectic structure

In this appendix we work out the symplectic form in the dressing time, to show the detailed

JHEP01(2024)166
computation used in section 5.2. An important property used in the derivation below is
the fact that δ commutes with the coordinate time derivative, δ∂v = ∂v δ, and from this
one can establish that

([∂V , δ]Õ) ◦ V = 0. (D.1)

Let us begin with the matter field, whose presymplectic potential has been introduced
in (5.17). Using (5.18), we get
Z
(0)
Ωmat = εN δ(Ω∂v φ)∧δφ
ZN
(0)
= εN (δΩ∂v φ+Ω∂v δφ)∧(δ φ̃+∂V φ̃δV )◦V
ZN
(0)
= εN ((δ Ω̃+∂V Ω̃δV )◦V ∂v φ+ Ω̃∂v ((δ φ̃+∂V φ̃δV )◦V ))∧(δ φ̃+∂V φ̃δV )◦V
ZN  
(0)
= εN ∂v V δ(Ω̃∂V φ̃)∧δ φ̃◦V +δ(∂v V Ω̃(∂V φ̃)2 )◦V ∧δV +∂v (δV ∧(Ω̃∂V φ̃δ φ̃)◦V ) .
N

This is the matter contribution: the first term appears in the first line of (5.20), and gives
rise to N ∂v V δ θ̃mat ◦ V , the second term is the matter contribution to the constraint on
R

1 R (0) mat ◦ V ) ∧ δV , and finally the third term is the

the second line of (5.20), 8πG N εN δ(∂v V C̃
corner contribution appearing in (5.22), which becomes (5.31).
Next, we consider the spin-2 contribution to (3.36), which is processed in a very similar
way to the matter sector (q is like φ while σ is like ∂v φ) to give25

1
Z
can(2) (0)
Ω = εN δ(Ωσ ab ) ∧ δqab (D.2)
16πG N
1
Z
(0)
= ε ∂v V δ(Ω̃σ̃ ab ) ∧ δ q̃ab ◦ V + 2δ(∂v V Ω̃σ̃ 2 ) ◦ V ∧ δV
16πG N N
+ ∂v (δV ∧ (Ω̃σ̃ ab δ q̃ab ) ◦ V ) .


(D.3)

This displays exactly the same pattern as the matter contribution: the first term appears
in the first line of (5.20), and gives rise to N ∂v V δ θ̃(2) ◦ V , the second term is the spin-2
R

1 R (0) (2) ◦ V ) ∧ δV ,
contribution to the constraint on the second line of (5.20), 8πG N εN δ(∂v V C̃
and finally the third term is the corner contribution appearing in (5.22), which becomes (5.30).
24
The tensor ϵa b is purely imaginary ϵa b = −ϵa b .
25
We have used the identity σ̃ ab ∂V q̃ab = 2σ̃ 2 .

– 37 –
 The spin-0 part is different from the rest. Starting again from (3.36), and using (5.13),
we get
1
Z
(0)
Ωcan(0) = − ε δµ∧δΩ (D.4)
8πG N N
1 ∂v δV 1 ∂v δV
Z   Z  
(0) (0)
=− εN ∂ v ∧δ Ω̃◦V − εN ∂ v ∧∂V Ω̃◦V δV
8πG N ∂v V 8πG N ∂v V
1
Z
(0)
+ εN ∂V2 Ω̃◦V δ∂v V ∧δV
8πG N
1 1
Z   Z  
(0) 2 (0)
= εN δ ∂V Ω̃◦V ∂v V ∧δV − εN ∂v ∂V δ Ω̃◦V ∧δV
8πG N 8πG N

JHEP01(2024)166
1 ∂v δV
Z  
(0)
− ε ∂v ∧(δ Ω̃+∂V Ω̃δV )◦V
8πG N N ∂v V
1 1 ∂v δV ∂v Ω
Z   Z   
(0) 2 (0)
= ε δ ∂V Ω̃◦V ∂v V ∧δV − ε ∂v ∧δΩ−δV ∧δ .
8πG N N 8πG N N ∂v V ∂v V
We thus see that the first term is the spin-0 contribution to the constraint on the second line
1 R (0) (0) ◦ V ) ∧ δV , while the last term is the corner contribution (5.29).
of (5.20), 8πG N εN δ(∂v V C̃
There are extra contributions to the spin-0 part when considering a generic clock,
since one has
∂v δV
 
δµ = ∂v δV µ̃ ◦ V + ∂V µ̃ ◦ V ∂v V δV + ∂v V δ µ̃ ◦ V + ∂v . (D.5)
∂v V
There are three additional terms, so that the final result is
1
Z  
(0)
Ωcan(0) = εN δ (∂V2 Ω̃ − µ̃∂V Ω̃) ◦ V ∂v V ∧ δV (D.6)
8πG N
1
Z
(0)
− εN ∂v V δ µ̃ ∧ δ Ω̃ ◦ V (D.7)
8πG N
1
Z  
(0)
− εN ∂v δV ∧ µ̃δ Ω̃ ◦ V (D.8)
8πG N
1 ∂v δV ∂v Ω
Z   
(0)
− ε ∂v ∧ δΩ − δV ∧ δ . (D.9)
8πG N N ∂v V ∂v V
We see that in a generic time frame we restore the bulk µ̃-contributions to the constraint (first
line), to the bulk presymplectic structure (second line), and to the corner term (third line).

Open Access. This article is distributed under the terms of the Creative Commons
Attribution License (CC-BY4.0), which permits any use, distribution and reproduction in
any medium, provided the original author(s) and source are credited.

References
[1] E. Noether, Invariant Variation Problems, Gott. Nachr. 1918 (1918) 235 [physics/0503066]
[INSPIRE].
[2] L. Freidel and A. Perez, Quantum gravity at the corner, Universe 4 (2018) 107
[arXiv:1507.02573] [INSPIRE].
[3] W. Donnelly and L. Freidel, Local subsystems in gauge theory and gravity, JHEP 09 (2016) 102
[arXiv:1601.04744] [INSPIRE].

– 38 –
 [4] M. Geiller, Edge modes and corner ambiguities in 3d Chern-Simons theory and gravity, Nucl.
Phys. B 924 (2017) 312 [arXiv:1703.04748] [INSPIRE].
[5] A.J. Speranza, Local phase space and edge modes for diffeomorphism-invariant theories, JHEP
02 (2018) 021 [arXiv:1706.05061] [INSPIRE].
[6] M. Geiller, Lorentz-diffeomorphism edge modes in 3d gravity, JHEP 02 (2018) 029
[arXiv:1712.05269] [INSPIRE].
[7] W. Donnelly, L. Freidel, S.F. Moosavian and A.J. Speranza, Gravitational edge modes,
coadjoint orbits, and hydrodynamics, JHEP 09 (2021) 008 [arXiv:2012.10367] [INSPIRE].
[8] L. Ciambelli and R.G. Leigh, Isolated surfaces and symmetries of gravity, Phys. Rev. D 104

JHEP01(2024)166
(2021) 046005 [arXiv:2104.07643] [INSPIRE].
[9] L. Freidel, R. Oliveri, D. Pranzetti and S. Speziale, Extended corner symmetry, charge bracket
and Einstein’s equations, JHEP 09 (2021) 083 [arXiv:2104.12881] [INSPIRE].
[10] L. Ciambelli and R.G. Leigh, Universal corner symmetry and the orbit method for gravity, Nucl.
Phys. B 986 (2023) 116053 [arXiv:2207.06441] [INSPIRE].
[11] W. Donnelly, L. Freidel, S.F. Moosavian and A.J. Speranza, Matrix Quantization of
Gravitational Edge Modes, arXiv:2212.09120 [DOI:10.1007/JHEP05(2023)163] [INSPIRE].
[12] L. Ciambelli, From Asymptotic Symmetries to the Corner Proposal, PoS Modave2022 (2023)
002 [arXiv:2212.13644] [INSPIRE].
[13] L. Freidel, M. Geiller and W. Wieland, Corner symmetry and quantum geometry,
arXiv:2302.12799 [INSPIRE].
[14] L. Ciambelli et al., Cornering Quantum Gravity, arXiv:2307.08460 [INSPIRE].
[15] L. Ciambelli, R.G. Leigh and P.-C. Pai, Embeddings and Integrable Charges for Extended
Corner Symmetry, Phys. Rev. Lett. 128 (2022) 171302 [arXiv:2111.13181] [INSPIRE].
[16] L. Freidel, A canonical bracket for open gravitational system, arXiv:2111.14747 [INSPIRE].
[17] M.S. Klinger, R.G. Leigh and P.-C. Pai, Extended phase space in general gauge theories, Nucl.
Phys. B 998 (2024) 116404 [arXiv:2303.06786] [INSPIRE].
[18] A. Ashtekar and M. Streubel, Symplectic Geometry of Radiative Modes and Conserved
Quantities at Null Infinity, Proc. Roy. Soc. Lond. A 376 (1981) 585 [INSPIRE].
[19] A.C. Wall, A proof of the generalized second law for rapidly changing fields and arbitrary
horizon slices, Phys. Rev. D 85 (2012) 104049 [Erratum ibid. 87 (2013) 069904]
[arXiv:1105.3445] [INSPIRE].
[20] R.L. Arnowitt, S. Deser and C.W. Misner, The dynamics of general relativity, Gen. Rel. Grav.
40 (2008) 1997 [gr-qc/0405109] [INSPIRE].
[21] B.S. DeWitt, Quantum Theory of Gravity. I. The Canonical Theory, Phys. Rev. 160 (1967)
1113 [INSPIRE].
[22] M. Henneaux, Geometry of zero signature spacetime, Bull. Soc. Math. Belg. 31 (1979) 47
[INSPIRE].
[23] M. Mars and J.M.M. Senovilla, Geometry of general hypersurfaces in space-time: Junction
conditions, Class. Quant. Grav. 10 (1993) 1865 [gr-qc/0201054] [INSPIRE].
[24] E. Gourgoulhon and J.L. Jaramillo, A 3+1 perspective on null hypersurfaces and isolated
horizons, Phys. Rept. 423 (2006) 159 [gr-qc/0503113] [INSPIRE].

– 39 –
[25] E. Gourgoulhon, 3+1 formalism and bases of numerical relativity, gr-qc/0703035 [INSPIRE].
[26] J.-M. Lévy-Leblond, Une nouvelle limite non-relativiste du groupe de Poincaré, A. Inst. H.
Poincare Phys. Theor. 3 (1965) 1.
[27] N.D. Sen Gupta, On an analogue of the Galilei group, Nuovo Cim. A 44 (1966) 512 [INSPIRE].
[28] C. Duval, G.W. Gibbons and P.A. Horvathy, Conformal Carroll groups and BMS symmetry,
Class. Quant. Grav. 31 (2014) 092001 [arXiv:1402.5894] [INSPIRE].
[29] C. Duval, G.W. Gibbons, P.A. Horvathy and P.M. Zhang, Carroll versus Newton and Galilei:
two dual non-Einsteinian concepts of time, Class. Quant. Grav. 31 (2014) 085016
[arXiv:1402.0657] [INSPIRE].

JHEP01(2024)166
[30] C. Duval, G.W. Gibbons and P.A. Horvathy, Conformal Carroll groups, J. Phys. A 47 (2014)
335204 [arXiv:1403.4213] [INSPIRE].
[31] L. Donnay, G. Giribet, H.A. González and M. Pino, Supertranslations and Superrotations at the
Black Hole Horizon, Phys. Rev. Lett. 116 (2016) 091101 [arXiv:1511.08687] [INSPIRE].
[32] L. Donnay, G. Giribet, H.A. González and M. Pino, Extended Symmetries at the Black Hole
Horizon, JHEP 09 (2016) 100 [arXiv:1607.05703] [INSPIRE].
[33] F. Hopfmüller and L. Freidel, Gravity Degrees of Freedom on a Null Surface, Phys. Rev. D 95
(2017) 104006 [arXiv:1611.03096] [INSPIRE].
[34] R.F. Penna, Near-horizon Carroll symmetry and black hole Love numbers, arXiv:1812.05643
[INSPIRE].
[35] L. Ciambelli et al., Flat holography and Carrollian fluids, JHEP 07 (2018) 165
[arXiv:1802.06809] [INSPIRE].
[36] V. Chandrasekaran, É.É. Flanagan and K. Prabhu, Symmetries and charges of general
relativity at null boundaries, JHEP 11 (2018) 125 [Erratum ibid. 07 (2023) 224]
[arXiv:1807.11499] [INSPIRE].
[37] L. Ciambelli, R.G. Leigh, C. Marteau and P.M. Petropoulos, Carroll Structures, Null Geometry
and Conformal Isometries, Phys. Rev. D 100 (2019) 046010 [arXiv:1905.02221] [INSPIRE].
[38] L. Donnay and C. Marteau, Carrollian Physics at the Black Hole Horizon, Class. Quant. Grav.
36 (2019) 165002 [arXiv:1903.09654] [INSPIRE].
[39] L. Ciambelli and C. Marteau, Carrollian conservation laws and Ricci-flat gravity, Class. Quant.
Grav. 36 (2019) 085004 [arXiv:1810.11037] [INSPIRE].
[40] H. Adami et al., T-Witts from the horizon, JHEP 04 (2020) 128 [arXiv:2002.08346] [INSPIRE].
[41] V. Chandrasekaran and A.J. Speranza, Anomalies in gravitational charge algebras of null
boundaries and black hole entropy, JHEP 01 (2021) 137 [arXiv:2009.10739] [INSPIRE].
[42] H. Adami, M.M. Sheikh-Jabbari, V. Taghiloo and H. Yavartanoo, Null surface thermodynamics,
Phys. Rev. D 105 (2022) 066004 [arXiv:2110.04224] [INSPIRE].
[43] L. Freidel and P. Jai-akson, Carrollian hydrodynamics from symmetries, Class. Quant. Grav. 40
(2023) 055009 [arXiv:2209.03328] [INSPIRE].
[44] A.C. Petkou, P.M. Petropoulos, D.R. Betancour and K. Siampos, Relativistic fluids,
hydrodynamic frames and their Galilean versus Carrollian avatars, JHEP 09 (2022) 162
[arXiv:2205.09142] [INSPIRE].
[45] A. Raychaudhuri, Relativistic cosmology. I, Phys. Rev. 98 (1955) 1123 [INSPIRE].

– 40 –
[46] R.K. Sachs, Gravitational waves in general relativity. VI. The outgoing radiation condition,
Proc. Roy. Soc. Lond. A 264 (1961) 309 [INSPIRE].
[47] L.D. Landau and E.M. Lifschits, The Classical Theory of Fields, Pergamon Press, Oxford, U.K.
(1975) [INSPIRE].
[48] S. Kar and S. SenGupta, The Raychaudhuri equations: A Brief review, Pramana 69 (2007) 49
[gr-qc/0611123] [INSPIRE].
[49] T. Damour, Black Hole Eddy Currents, Phys. Rev. D 18 (1978) 3598 [INSPIRE].
[50] T. Damour, Quelques propriétés mécaniques, électromagnétiques, thermodynamiques et
quantiques des trous noirs, Ph.D. thesis, Université Pierre et Marie Curie, Paris 6, France (1979)

JHEP01(2024)166
[INSPIRE].
[51] V. Chandrasekaran, E.E. Flanagan, I. Shehzad and A.J. Speranza, Brown-York charges at null
boundaries, JHEP 01 (2022) 029 [arXiv:2109.11567] [INSPIRE].
[52] L. Freidel and P. Jai-akson, Carrollian hydrodynamics and symplectic structure on stretched
horizons, arXiv:2211.06415 [INSPIRE].
[53] G. Jafari, Stress Tensor on Null Boundaries, Phys. Rev. D 99 (2019) 104035
[arXiv:1901.04054] [INSPIRE].
[54] L. Freidel, M. Geiller and D. Pranzetti, Edge modes of gravity. Part I. Corner potentials and
charges, JHEP 11 (2020) 026 [arXiv:2006.12527] [INSPIRE].
[55] L. Freidel, M. Geiller and D. Pranzetti, Edge modes of gravity. Part II. Corner metric and
Lorentz charges, JHEP 11 (2020) 027 [arXiv:2007.03563] [INSPIRE].
[56] L. Freidel, M. Geiller and D. Pranzetti, Edge modes of gravity. Part III. Corner simplicity
constraints, JHEP 01 (2021) 100 [arXiv:2007.12635] [INSPIRE].
[57] L. Lehner, R.C. Myers, E. Poisson and R.D. Sorkin, Gravitational action with null boundaries,
Phys. Rev. D 94 (2016) 084046 [arXiv:1609.00207] [INSPIRE].
[58] K. Parattu, S. Chakraborty, B.R. Majhi and T. Padmanabhan, A Boundary Term for the
Gravitational Action with Null Boundaries, Gen. Rel. Grav. 48 (2016) 94 [arXiv:1501.01053]
[INSPIRE].
[59] K. Parattu, S. Chakraborty and T. Padmanabhan, Variational Principle for Gravity with Null
and Non-null boundaries: A Unified Boundary Counter-term, Eur. Phys. J. C 76 (2016) 129
[arXiv:1602.07546] [INSPIRE].
[60] W. Wieland, Fock representation of gravitational boundary modes and the discreteness of the
area spectrum, Annales Henri Poincare 18 (2017) 3695 [arXiv:1706.00479] [INSPIRE].
[61] W. Wieland, New boundary variables for classical and quantum gravity on a null surface, Class.
Quant. Grav. 34 (2017) 215008 [arXiv:1704.07391] [INSPIRE].
[62] F. Hopfmüller and L. Freidel, Null Conservation Laws for Gravity, Phys. Rev. D 97 (2018)
124029 [arXiv:1802.06135] [INSPIRE].
[63] R. Oliveri and S. Speziale, Boundary effects in General Relativity with tetrad variables, Gen.
Rel. Grav. 52 (2020) 83 [arXiv:1912.01016] [INSPIRE].
[64] H. Adami et al., Symmetries at null boundaries: two and three dimensional gravity cases, JHEP
10 (2020) 107 [arXiv:2007.12759] [INSPIRE].
[65] H. Adami et al., Null boundary phase space: slicings, news & memory, JHEP 11 (2021) 155
[arXiv:2110.04218] [INSPIRE].

– 41 –
[66] M.M. Sheikh-Jabbari, On symplectic form for null boundary phase space, Gen. Rel. Grav. 54
(2022) 140 [arXiv:2209.05043] [INSPIRE].
[67] H. Adami et al., Hydro & thermo dynamics at causal boundaries, examples in 3d gravity, JHEP
07 (2023) 038 [arXiv:2305.01009] [INSPIRE].
[68] R.K. Sachs, On the Characteristic Initial Value Problem in Gravitational Theory, J. Math.
Phys. 3 (1962) 908 [INSPIRE].
[69] R. Gambini and A. Restuccia, The Initial Value Problem and the Dirac Bracket Relations in
Null Gravidynamics, Phys. Rev. D 17 (1978) 3150 [INSPIRE].
[70] R. Penrose, Null Hypersurface Initial Data for Classical Fields of Arbitrary Spin and General

JHEP01(2024)166
Relativity, Gen. Rel. Grav. 12 (1980) 225 [INSPIRE].
[71] C.G. Torre, Null Surface Geometrodynamics, Class. Quant. Grav. 3 (1986) 773 [INSPIRE].
[72] J.N. Goldberg, D.C. Robinson and C. Soteriou, Null hypersurfaces and new variables, Class.
Quant. Grav. 9 (1992) 1309 [INSPIRE].
[73] J.N. Goldberg and C. Soteriou, Canonical general relativity on a null surface with coordinate
and gauge fixing, Class. Quant. Grav. 12 (1995) 2779 [gr-qc/9504043] [INSPIRE].
[74] R.A. d’Inverno, P. Lambert and J.A. Vickers, Hamiltonian analysis of the double null 2 + 2
decomposition of general relativity expressed in terms of self-dual bivectors, Class. Quant. Grav.
23 (2006) 4511 [gr-qc/0604084] [INSPIRE].
[75] W. Wieland, Generating functional for gravitational null initial data, Class. Quant. Grav. 36
(2019) 235007 [arXiv:1905.06357] [INSPIRE].
[76] W. Wieland, Gravitational SL(2, R) algebra on the light cone, JHEP 07 (2021) 057
[arXiv:2104.05803] [INSPIRE].
[77] M. Mars and G. Sánchez-Pérez, Double null data and the characteristic problem in general
relativity, J. Phys. A 56 (2023) 035203 [arXiv:2205.15267] [INSPIRE].
[78] M. Mars and G. Sánchez-Pérez, Covariant definition of double null data and geometric
uniqueness of the characteristic initial value problem, J. Phys. A 56 (2023) 255203
[arXiv:2301.02722] [INSPIRE].
[79] M.P. Reisenberger, The Symplectic 2-form and Poisson bracket of null canonical gravity,
gr-qc/0703134 [INSPIRE].
[80] M.P. Reisenberger, The Poisson bracket on free null initial data for gravity, Phys. Rev. Lett.
101 (2008) 211101 [arXiv:0712.2541] [INSPIRE].
[81] M.P. Reisenberger, The symplectic 2-form for gravity in terms of free null initial data, Class.
Quant. Grav. 30 (2013) 155022 [arXiv:1211.3880] [INSPIRE].
[82] A. Fuchs and M.P. Reisenberger, Integrable structures and the quantization of free null initial
data for gravity, Class. Quant. Grav. 34 (2017) 185003 [arXiv:1704.06992] [INSPIRE].
[83] M.P. Reisenberger, The Poisson brackets of free null initial data for vacuum general relativity,
Class. Quant. Grav. 35 (2018) 185012 [arXiv:1804.10284] [INSPIRE].
[84] E. Beltrami, Saggio di interpetrazione della geometria non-euclidea, (in Italian) Giornale di
Mathematica 6 (1868) 284.
[85] L. Baulieu, Leaf of Leaf Foliation and Beltrami Parametrization in d > 2 dimensional Gravity,
arXiv:2109.06681 [INSPIRE].

– 42 –
 [86] L. Baulieu and T. Wetzstein, BRST BMS4 symmetry and its cocycles from horizontality
conditions, JHEP 07 (2023) 130 [arXiv:2304.12369] [INSPIRE].
[87] E. Witten, Gravity and the crossed product, JHEP 10 (2022) 008 [arXiv:2112.12828]
[INSPIRE].
[88] K. Jensen, J. Sorce and A.J. Speranza, Generalized entropy for general subregions in quantum
gravity, JHEP 12 (2023) 020 [arXiv:2306.01837] [INSPIRE].
[89] M.S. Klinger and R.G. Leigh, Crossed Products, Extended Phase Spaces and the Resolution of
Entanglement Singularities, arXiv:2306.09314 [INSPIRE].
[90] A.C. Wall, Maximin Surfaces, and the Strong Subadditivity of the Covariant Holographic

JHEP01(2024)166
Entanglement Entropy, Class. Quant. Grav. 31 (2014) 225007 [arXiv:1211.3494] [INSPIRE].
[91] N. Engelhardt and A.C. Wall, Quantum Extremal Surfaces: Holographic Entanglement Entropy
beyond the Classical Regime, JHEP 01 (2015) 073 [arXiv:1408.3203] [INSPIRE].
[92] R. Bousso, Z. Fisher, S. Leichenauer and A.C. Wall, Quantum focusing conjecture, Phys. Rev.
D 93 (2016) 064044 [arXiv:1506.02669] [INSPIRE].
[93] R. Bousso et al., Proof of the Quantum Null Energy Condition, Phys. Rev. D 93 (2016) 024017
[arXiv:1509.02542] [INSPIRE].
[94] M. Siino and T. Koike, Topological classification of black hole: Generic Maxwell set and crease
set of horizon, Int. J. Mod. Phys. D 20 (2011) 1095 [gr-qc/0405056] [INSPIRE].
[95] M. Gadioux and H.S. Reall, Creases, corners, and caustics: Properties of nonsmooth structures
on black hole horizons, Phys. Rev. D 108 (2023) 084021 [arXiv:2303.15512] [INSPIRE].
[96] X. Bekaert and K. Morand, Connections and dynamical trajectories in generalised
Newton-Cartan gravity II. An ambient perspective, J. Math. Phys. 59 (2018) 072503
[arXiv:1505.03739] [INSPIRE].
[97] L. Ciambelli et al., Covariant Galilean versus Carrollian hydrodynamics from relativistic fluids,
Class. Quant. Grav. 35 (2018) 165001 [arXiv:1802.05286] [INSPIRE].
[98] T. Jacobson and G. Kang, Conformal invariance of black hole temperature, Class. Quant. Grav.
10 (1993) L201 [gr-qc/9307002] [INSPIRE].
[99] R.H. Price and K.S. Thorne, Membrane paradigm, in Black holes: the membrane paradigm, K.S.
Thorne, R.H. Price, D.A. MacDonald eds., Yale University Press (1986).
[100] J.D. Brown and J.W. York Jr., Quasilocal energy and conserved charges derived from the
gravitational action, Phys. Rev. D 47 (1993) 1407 [gr-qc/9209012] [INSPIRE].
[101] L. Rezzolla and O. Zanotti, Relativistic hydrodynamics, Oxford University Press (2013).
[102] L. Freidel, R. Oliveri, D. Pranzetti and S. Speziale, The Weyl BMS group and Einstein’s
equations, JHEP 07 (2021) 170 [arXiv:2104.05793] [INSPIRE].
[103] G. Hayward, Gravitational action for space-times with nonsmooth boundaries, Phys. Rev. D 47
(1993) 3275 [INSPIRE].
[104] V. Chandrasekaran and E.E. Flanagan, The gravitational phase space of horizons in general
relativity, arXiv:2309.03871 [INSPIRE].
[105] G. Odak, A. Rignon-Bret and S. Speziale, General gravitational charges on null hypersurfaces,
JHEP 12 (2023) 038 [arXiv:2309.03854] [INSPIRE].
[106] T. Jacobson and R. Parentani, Horizon entropy, Found. Phys. 33 (2003) 323 [gr-qc/0302099]
[INSPIRE].

– 43 –
[107] E. Bianchi, Horizon entanglement entropy and universality of the graviton coupling,
arXiv:1211.0522 [INSPIRE].
[108] A. Rignon-Bret, Second law from the Noether current on null hypersurfaces, Phys. Rev. D 108
(2023) 044069 [arXiv:2303.07262] [INSPIRE].
[109] T. Faulkner, R.G. Leigh, O. Parrikar and H. Wang, Modular Hamiltonians for Deformed
Half-Spaces and the Averaged Null Energy Condition, JHEP 09 (2016) 038
[arXiv:1605.08072] [INSPIRE].
[110] H. Casini, E. Teste and G. Torroba, Modular Hamiltonians on the null plane and the Markov
property of the vacuum state, J. Phys. A 50 (2017) 364001 [arXiv:1703.10656] [INSPIRE].

JHEP01(2024)166
[111] S. Hollands and A. Ishibashi, News versus information, Class. Quant. Grav. 36 (2019) 195001
[arXiv:1904.00007] [INSPIRE].

– 44 –