Lecture notes prepared for the Solvay Doctoral School on

Quantum Field Theory, Strings and Gravity.

Lectures given in Brussels, October 2017.
arXiv:1801.07064v4 [hep-th] 8 Feb 2019

Advanced Lectures on
General Relativity

Lecturing & Proofreading:

Geoffrey Compère

Typesetting, layout & figures:

Adrien Fiorucci

Fonds National de la Recherche Scientifique (Belgium)

Physique Théorique et Mathématique
Université Libre de Bruxelles and International Solvay Institutes
Campus Plaine C.P. 231, B-1050 Bruxelles, Belgium

Please email any question or correction to: [email protected]

Abstract — These lecture notes are intended for starting PhD students in theoretical physics who
have a working knowledge of General Relativity. The 4 topics covered are (1) Surface charges as con-
served quantities in theories of gravity; (2) Classical and holographic features of three-dimensional
Einstein gravity; (3) Asymptotically flat spacetimes in 4 dimensions: BMS group and memory effects;
(4) The Kerr black hole: properties at extremality and quasi-normal mode ringing. Each topic starts
with historical foundations and points to a few modern research directions.
 Table of contents

1 Surface charges in Gravitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.1 Introduction : general covariance and conserved stress tensor . . . . . . . . . . . . . . 7
1.2 Generalized Noether theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.1 Gauge transformations and trivial currents . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Lower degree conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Surface charges in generally covariant theories . . . . . . . . . . . . . . . . . . . 13
1.3 Covariant phase space formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1 Field fibration and symplectic structure . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.2 Noether’s second theorem : an important lemma . . . . . . . . . . . . . . . . . 17
Einstein’s gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Einstein-Maxwell electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.3 Fundamental theorem of the covariant phase space formalism . . . . . . . . . . 20
Cartan’s magic formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Noether-Wald surface charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Some residual ambiguities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.4 Conserved surface charges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.1 Definition of the charges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.2 Integrability condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4.3 Conservation criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.4 Charge algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Representation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.5 Conserved charge formula for General Relativity . . . . . . . . . . . . . . . . . 28
1.5 Conserved charges from the equations of motion . . . . . . . . . . . . . . . . . . . . . . 29
1.5.1 Anderson’s homotopy operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5.2 Invariant presymplectic current . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5.3 Expression of Barnich-Brandt’s charge for Einstein’s gravity . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Three dimensional Einstein’s gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2.1 Overview of typical properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.1.1 A theory without bulk degrees of freedom . . . . . . . . . . . . . . . . . . . . . 34
2.1.2 Einstein-Hilbert action and homogeneous spacetimes . . . . . . . . . . . . . . . 35
2.2 Asymptotically anti-de Sitter phase space . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.1 Global properties of AdS3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.2 Asymptotically AdS3 black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
BTZ black holes and more. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Main properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Table of contents 3
 Penrose diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Identifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Symmetries of the quotient space . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
BTZ phase space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.2.3 Asymptotically AdS3 spacetimes . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Asymptotic symmetry algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
Phase space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.3 Asymptotically flat phase space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.1 Flat limit and the BMS3 group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.2 Constant representatives : spinning particles and flat cosmologies . . . . . . . . 57
2.4 Chern-Simons formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4.1 Local Lorentz triad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.4.2 Chern-Simons action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.4.3 General covariance and charges . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4.4 AdS3 phase space in the Chern-Simons formalism . . . . . . . . . . . . . . . . . 64
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3 Asymptotically flat spacetimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.1 A definition of asymptotic flatness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1.1 Asymptotic structure of Minkowski spacetime . . . . . . . . . . . . . . . . . . . 69
Conformal compactification of Minkowski spacetime . . . . . . . . . . . . . . . 69
Singular points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.1.2 Gravity in Bondi gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.1.3 Initial and late data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.2 Asymptotic symmetries : the BMS4 group . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.1 Supertranslations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.2.2 Lorentz algebra and its extensions . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.2.3 Gravitational memory effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Displacement memory effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Spin memory effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3 Scattering problem and junction conditions . . . . . . . . . . . . . . . . . . . . . . . . . 83
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4 Rotating black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.1 The Kerr solution — Review of the main features . . . . . . . . . . . . . . . . . . . . . . 86
4.1.1 Metric in Boyer-Lindquist coordinates . . . . . . . . . . . . . . . . . . . . . . . . 86
4.1.2 Killing horizon and black hole thermodynamics . . . . . . . . . . . . . . . . . . 86
4.1.3 Ergoregion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.1.4 Event horizon and Cauchy horizon . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.1.5 Linear stability of black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.2 Extremal rotating black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.1 Near horizon geometries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

Table of contents 4
 4.2.2 Extremal BTZ black holes and their dual CFT description . . . . . . . . . . . . . 94
Extremal BTZ geometry and near-horizon limit . . . . . . . . . . . . . . . . . . 94
Chiral zero temperature states and chiral sector of a CFT . . . . . . . . . . . . . 95
Chiral Cardy formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2.3 The Kerr/CFT correspondence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Virasoro symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Conjugated chemical potential and Cardy matching . . . . . . . . . . . . . . . . 99
Frolov-Thorne vacuum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3 Black hole spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3.1 Fundamentals of the Newman-Penrose formalism . . . . . . . . . . . . . . . . . 101
4.3.2 Fundamentals of Petrov’s classification . . . . . . . . . . . . . . . . . . . . . . . 104
4.3.3 Quasi-normal mode ringing of Kerr . . . . . . . . . . . . . . . . . . . . . . . . . 107
Separation of variables and Teukolsky master equation . . . . . . . . . . . . . . 108
Solving the angular equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Solving the radial equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Geometric interpretation in the eikonal limit . . . . . . . . . . . . . . . . . . . . 112
Schwarzschild spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Kerr spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

Table of contents 5
Conventions and notations — We employ units such that the speed of light c “ 1 but we will
keep G explicit. The spacetime manifold is denoted by the couple pM, gµν q. The signature of the
Lorentzian metric gµν obeys the mostly plus convention p´, `, `, ...q. The dimension of spacetime
is generally n. If necessary, we will write explicitely n “ d ` 1 where d represents the number
of spatial dimensions. We use the unit normalized convention for symmetrization and antisym-
1 1 µν
metrization, T pµνq “ 2 pT
µν ` T νµ q and krµνs “ 2 pk ´ kνµ q. We employ Einstein’s summed index
convention: double indices in an expression are implicitly summed over. Finally, we follow the
conventions adopted in the textbook Gravitation [1] by Wheeler, Thorne and Misner, concerning the
definition of various objects in Relativity. In particular, the Riemann-Christoffel tensor is determined
“ Bα Γ ´ Bβ Γ `Γ κα Γ νβ ´Γ κβ Γ να .
µ µ µ µ µ
as R ναβ νβ να
κ κ In this convention, the n-sphere has a positive Ricci
curvature scalar R “ Rα α where the Ricci tensor is Rαµαν .

The notation of spacetime coordinates is as follows. Greek indexes µ, ν, . . . span the full dimension
of spacetime n “ d ` 1, so µ P t0, . . . , du. Often the index 0 represents a timelike coordinate. Latin
indexes will designate the other coordinates x a with a P t1, . . . , du. Capital latin letters will be used to
denote angular coordinates x A among the spacelike coordinates.

Some conventions concerning objects of exterior calculus have also to be detailed. The volume form
?
is denoted by eµ1 ¨¨¨µn . It is a tensor so it includes ´g. We will keep the notation ε µ1 ¨¨¨µn for the
?
numerically invariant pseudo-tensor with entries ´1, 0 or 1. We have eµ1 ¨¨¨µn “ ´gε µ1 ¨¨¨µn , see the
appendix B of Wald’s book [2] for details. A general pn ´ pq-form (p P N, p ď n) is written as boldface
?
X “ X µ1 ¨¨¨µ p ´gpdn´p xqµ1 ¨¨¨µ p developed in the base :

1
pdn´p xqµ1 ¨¨¨µ p “ ε µ ¨¨¨µ ν ¨¨¨ν dx νp`1 ^ ¨ ¨ ¨ ^ dx νn .
p!pn ´ pq! 1 p p`1 n

We will always invoke Hodge’s duality between p-forms and pn ´ pq-forms to define objects in the
?
more convenient way. For example, the Lagrangian density L (equal to ´g times the Lagrangian
scalar) will be identified to the n-form L “ L dn x. A vector field J µ will be regarded as a pn ´ 1q-form
?
J “ J µ ´gpdn´1 xqµ , since J µ is the Hodge dual of a 1-form. An antisymmetric 2 tensor kµν “ krµνs
?
will be identified with its Hodge dual: a pn ´ 2q-form k “ krµνs ´gpdn´2 xqµν where the antisymetri-
sation arises naturally from the definition of the natural basis of pn ´ 2q-forms. And so on ! (We keep
?
the factors of ´g explicit to easily vary them!)

Under an infinitesimal diffeomorphism generated by the vector χµ , a general field Φi with arbitrary
index structure summarized by the abstract index i will be modified by the Lie derivative δχ Φi “
`Lχ Φi . As a final remark, we note that waved equalities («) represent any equation that holds if and
only if the Euler-Lagrange equations of motion formulated in the theory of interest are satisfied.

Conventions and notations 6

 Lecture 1

Surface charges in Gravitation

The main purpose of this first lecture is to introduce the concept of canonical surface charges in a
generally covariant theory of gravity, whose General Relativity is the most famous representative.

As a starter, we will show that a conserved stress tensor can be generated for any classical field the-
ory, simply by coupling it to gravity and using general covariance of the so-enhanced theory. Then
we will enter into the main point we have to discuss, and motivate why we cannot define conserved
currents and charges in Noether’s fashion for generally covariant theories, and more globally, for
theories that include gauge transformations. This quite tricky fact will lead us to extend Noether’s
first theorem to formulate lower degree conservation laws, which will be exploitable for theories such
as Einstein’s gravity. On the way, we will discuss about the symplectic structure of abstract spaces
of fields, and use the covariant phase space formalism to derive a magnificent and powerful result
linking this structure and the lower degree conserved forms that we are looking for. We will then be
able to compute surface charges associated to these quantities and study their properties and their
algebra. Along the text, some pedagogical examples will be provided, namely for pure Einstein’s
gravity, and Maxwell’s electrodynamics, enhanced in a curved background. Finally, we will present
another possible definition of these surface charges and use the latter definition as an efficient tool to
derive the conserved charges of Chern-Simons theory. We will finally discuss the residual ambigui-
ties of the conserved quantities...

1.1 Introduction : general covariance and conserved stress tensor

Before considering a theory of gravity, let us first consider a relativistic field theory of matter. It will
allow us to analyse a remarkable relation between the very fundamental concept of general covariance
of a theory, and the conservation of a stress tensor associated to this theory. It is a very nice way to
obtain a divergence-free tensor without directly invoking Noether’s theorem.

Let us start with the theory of special relativity. There is a background structure in the theory: the flat
metric. In Cartesian coordinates pt, x, y, zq, the metric takes the most refined form that a Lorentzian
metric could take : ηµν “ diagp´1, `1, `1, `1q, the so-called metric of Minkowski. The symmetries
of spacetime are defined as vector fields ξ µ that preserve this background structure, i.e. Lξ ηµν “ 0.
These symmetries preserve all distances and are by right recognized as isometries. In components,
we have ξ ρ Bρ ηµν ` 2ηρpµ Bνq ξ ρ “ 0. The first term is trivially zero, so it remains Bpµ ξ νq “ 0. The most
general solution is given by ξ µ “ aµ ` brµνs xν . The isometries of Minkowski spactime thus depend
upon 10 parameters : the 4 components of aµ that encode global translations and the 6 matrix ele-
ments of brµνs associated with Lorentz transformations (rotations and boosts). Under the Lie bracket,
these vectors give rise to the Poincaré algebra.

Lecture 1. Surface charges in Gravitation 7

 1.1. Introduction : general covariance and conserved stress tensor

What happens in an arbitrary coordinate system? Thanks to general covariance, we can express all
tensorial equations, such as Lξ ηµν “ 0, in an arbitrary frame by remplacing partial derivatives Bµ
by covariant derivatives ∇µ . The covariant derivative is defined with the Levi-Civita connection
compatible with the metric. In a general spacetime, although curvature appears, it is still possible to
consider free-falling observers, around which a local Lorentzian frame can be constructed. So locally,
we find again Bpµ ξ νq “ 0 and general covariance can again be invoked to arrive to

Result 1 (Killing equation)

A necessary and sufficient condition for a vector field to be an isometry of spacetime pM, gq
is that it verifies the so-called Killing equation : Lξ gµν “ 0 ðñ ∇pµ ξ νq “ 0.

Since the Lie bracket is a tensorial quantity, the Killing vectors of Minkowski spacetime form the
Poincaré algebra independently of the coordinates chosen to express the isometries.

After obtaining this very crucial formula, we can now show the following theorem :

Result 2 (Existence of a conserved stress tensor)

Any relativistic field theory in Minkowski spacetime admits a symmetric stress tensor T µν
that is divergence-free when the equations of motion hold, ∇µ T µν « 0.

Let’s consider an arbitrary theory of matter fields collectively denoted by Φ M “ pΦiM qiPI (with a
totally general index structure i belonging to a set of such structures I). The theory is described by
a Lagrangian density LrΦ M s. One can always couple this theory to gravity by introducing a non-flat
metric gµν into the Lagrangian: LrΦ M s Ñ LrΦ M , gµν s. The coupling is said minimal when it consists
in merely replacing the Minkowski metric by the general metric, standard derivatives by covariant
?
derivatives and with the necessary mutation of the volume form : dn x Ñ ´gdn x. After that, we can
define a natural symmetric tensor :
2 δL
T µν fi ? (1.1)
´g δgµν
thanks to the Euler-Lagrange derivative, rigorously defined as
ˆ ˙ ˆ ˙
δL BL BL BL
@Φ P Φ :
i
fi ´ Bµ ` Bµ Bν `¨¨¨ (1.2)
δΦ i BΦi B Bµ Φ i B Bµ Bν Φ i

for theories of any order in derivatives. The compact notation Φ “ tpΦiM qiPI , gµν u now encompasses
at the same time the original matter fields and the metric gµν of spacetime. We therefore have a natu-
ral symmetric candidate stress-tensor in the original field theory, namely T µν rΦ M , ηµν s where we sub-
stituted back the metric gµν to ηµν . What remains to be done is to show that T µν rΦ M , ηµν s expressed
δLrΦ M s
by (1.1) is covariantly conserved when the original matter equations hold, i.e. when δΦiM
“ 0.

Lecture 1. Surface charges in Gravitation 8

 1.1. Introduction : general covariance and conserved stress tensor

Let’s begin by performing an arbitrary variation of the Lagrangian density:

BL δL δL
δL “ δΦi ` Bµ δΦi ` ¨ ¨ ¨ “ δΦi i ` Bα Θα rδΦ; Φs. (1.3)
BΦ i δBµ Φ i δΦ

The first term simply contains the Euler-Lagrange equations of motion. The second one collects the
remnants of the inverse Leibniz rule, which was applied in order to factorize the variation of fields
δΦi without any derivative acting on it. In other words, this second term is nothing else than a
boundary term, expressed as a total derivative, or more precisely, the divergence of a vector field
density Θµ rδΦi , Φi s named the bare presymplectic potential. In the more convenient language of forms,
we can rewrite this equation as
δL
δL “ δΦi ` dΘrδΦ; Φs . (1.4)
δΦi

Here, Θ “ Θµ pdn´1 xqµ is a pn ´ 1q-form, and L “ Ldn x is the n-form naturally associated to the
Lagrangian density L. The total derivative dΘ is thus also a n-form, by virtue of the definition of the
exterior derivative d :
” ı
dΘ “ dx ν Bν Θµ pdn´1 xqµ “ Bν Θµ pdn xqδµν “ Bµ Θµ dn x. (1.5)

Let us now analyse the variation of L under an infinitesimal diffeomorphism generated by ξ µ . We

get

δL δL
δξ L “ δξ gµν ` δξ ΦiM i ` dp¨ ¨ ¨ q (1.6)
δgµν δΦ M
δL δL
“ Lξ gµν ` δξ ΦiM i ` dp¨ ¨ ¨ q (1.7)
δgµν δΦ M
δL δL
“ 2∇ µ ξ ν ` δξ ΦiM i ` dp¨ ¨ ¨ q (1.8)
δgµν δΦ M
δL
“ dn x ´g T µν ∇µ ξ ν ` δξ ΦiM i ` dp¨ ¨ ¨ q.
a
(1.9)
δΦ M

Let us now substitute gµν by the original Minkowski metric ηµν and let’s impose the matter field
equations. We are then “on-shell” in the original theory. We still have covariant derivatives since we
work in arbitrary coordinates. We get

δξ L « dn x ´g ∇µ pT µν ξ ν q ´ dn x ´g ∇µ T µν ξ ν ` dp¨ ¨ ¨ q
a a
(1.10)
« d n x Bµ ´g T µν ξ ν ´ dn x ´g ∇µ T µν ξ ν ` dp¨ ¨ ¨ q
`a ˘ a
(1.11)
« ´dn x ´g ∇µ T µν ξ ν ` dp¨ ¨ ¨ q.
a
(1.12)

Since general covariance requires that the total variation of the Lagrangian density on any diffeomor-
phism must be a total derivative, including when the equations of motion hold, it implies immedi-
ately the conservation of the stress tensor of the original matter theory!

δξ L “ dp¨ ¨ ¨ q ùñ ∇µ T µν |gµν “ηµν « 0 . (1.13)

Lecture 1. Surface charges in Gravitation 9

 1.2. Generalized Noether theorem

In this relativistic matter theory we can now build a 4-vector from the stress tensor : J µ “ T µν ξ ν
?
which is conserved (or such that its Hodge dual form J “ J µ ´gpdn´1 xqµ is closed), provided that
the diffeomorphism ξ µ is an isometry of spacetime.

∇µ J µ “ ∇µ T µν ξ ν ` T µν ∇pµ ξ νq “ 0 ` 0 ùñ dJ “ 0. (1.14)

The integral of J on an arbitrary Cauchy surface1 Σ produces a scalar quantity Q “

ş
Σ J which is
conserved when the system evolves, provided that fields decay sufficiently rapidly at the boundary
BΣ. Let us choose a coordinate x0 “ t such that the Cauchy surface is described as the surface t “ 0.
Then pdn´1 xq0 “ dΣ is the volume form on the surface. We have
ż ż ż ż
Bt Q “ Bt J 0 pdn´1 xq0 “ dΣ B0 J 0 “ ´ ~ ¨ ~J “ ´
dΣ ∇ ~J ¨ d~S “ 0.
Σ Σ Σ BΣ

To each isometry corresponds such a conserved charge :

Isometry Origin Conserved charge

ş
Translations Minkowski is homogeneous Pµ “ Σ dΣ T µ0
ş
Lorentz transformations Minkowski is isotropic and relativistic Mµν “ Σ dΣ px µ T ν0 ´ x ν T µ0 q

All these features are not surprising, since there is a fundamental result that permits to deduce imme-
diatly the existence of dynamical invariants when the theory of interest possesses some continuous
symmetries. This is the next topic to which we now turn!

1.2 Generalized Noether theorem

1.2.1 Gauge transformations and trivial currents

Let us begin by reviewing one of the most famous statements ever established in modern physics :
Noether’s first theorem. Proven in 1916, and considered as a "monument of mathematical thought"
by Einstein himself, it is not abusive to say that most of modern physical works rely on this result.
We will present it without proof, but in a quite modernized form. Let’s first clarify the terminology:
global symmetries preserve the Lagrangian up to a boundary term. Gauge transformations are global
symmetries whose generator arbitrarily depends upon the coordinates.

Result 3 (Noether’s first theorem)

Take any physical theory described by a Lagrangian density L defined on a spacetime man-
ifold pM, gq that admits global symmetries, some of which might be gauge transformations.
It exists a bijection between :

B The equivalence classes of global continuous symmetries of L, and

B The equivalence classes of conserved vector fields J µ , the so-called Noether currents.

1 A Cauchy surface is a subset of M which is intersected by every maximal causal curve exactly once. Once the initial

data is fixed on such a codimension 1 surface, the field equations lead to the evolution of the system in the entire spacetime.

Lecture 1. Surface charges in Gravitation 10

 1.2. Generalized Noether theorem

On the one hand, we say that two global symmetries of L are equivalent if and only if they differ only
by a gauge transformation and another symmetry whose generator is trivially zero on shell. On the
µ µ
other hand, we declare that two currents J1 and J2 are equivalent if and only if they differ by a trivial
current of the form
µ µ
J2 “ J1 ` Bν krµνs ` tµ (1.15)
µ µ
where krµνs is an skew tensor p2, 0q and tµ « 0. So we have Bµ J2 « Bµ J1 . Using this formulation of
Noether’s theorem, a thorny problem arises immediately. Imagine that you have a pure gauge the-
ory, i.e. a gauge theory with no non-trivial global symmetry at your disposal. From Noether’s first
theorem, it exists only one equivalence class of conserved currents: the trivial ones. In particular,
for generally covariant theories, any transformation like x µ Ñ x µ ` ξ µ is a gauge transformation,
thus the natural symmetries, also called isometries, are associated to trivial currents in a similar way.
We can define a charge by integrating on a Cauchy slice Σ as we saw before, but it reads simply
ş ş
Q “ Σ J µ pdn´1 xqµ « BΣ krµνs pdn´2 xqµν when the equations of motion hold. Q is manifestly com-
pletely arbitrary, because krµνs is totally unconstrained ! Let us make the issue explicit by computing
the Noether current of General Relativity.
´ ? ¯
R ´g
We consider the Hilbert-Einstein Lagrangian density coupled to matter L “ 16πG ` L M dn x where
R is the scalar Ricci curvature and L M is the Lagrangian density of matter fields. The conserved
stress-tensor built from varying the Lagrangian is
?
µν 2 δL 1 1 δpR ´gq 2 δL M 1 ` µν µν ˘
T fi ? “? `? “´ G ´ 8πGTM « 0 (1.16)
´g δgµν ´g 8πG δgµν ´g δgµν 8πG

as we exactly retrieve Einstein’s field equations. The Noether current associated to a diffeomorphism
ξ µ is therefore trivial, J µ “ T µν ξ ν « 0.

1.2.2 Lower degree conservation laws

We can sketch a solution to this puzzle simply by considering more carefully the expression of the ar-
ş ş
bitrary Noether charge Q “ BΣ krµνs pdn´2 xqµν “ BΣ k. We see that Q reduces to the flux of k through
the boundary BΣ2 , and depends only on the properties of this pn ´ 2q-form in the vicinity of BΣ. This
suggests to invoke lower degree conservation laws. Indeed, let us imagine that we are able to define
uniquely a pn ´ 2q-form k “ krµνs pdn´2 xqµν such that dk “ Bν krµνs pdn´1 xqµ “ 0. Thanks to such an
ş
object, we can define an integral charge Q “ S k on any surface S which will be conserved when
we change surfaces without crossing any singularity (such as the source of the charge!). Seeking for
conserved pn ´ 2q-forms is the right path to obtain a canonical notion of charges in gauge theories.

While the first Noether theorem maps each symmetry to a class a conserved currents (or equivalently
closed pn ´ 1q-forms J “ J µ pdn´1 xqµ ), it exists a generalized version of it which precisely focuses on
lower degree conserved forms. This result was established by Barnich, Brandt and Henneaux in 1995
[3] using cohomological methods, and we present it here without proof.
2 Remember that Σ being a Cauchy slice (and so by definition a pn ´ 1q-dimensional volumic object), BΣ is nothing but

a pn ´ 2q-surface ! Take n “ 4 to clarify the role of any geometric structure...

Lecture 1. Surface charges in Gravitation 11

 1.2. Generalized Noether theorem

Result 4 (Generalized Noether theorem)

Take any physical theory described by a Lagrangian density L defined on a spacetime man-
ifold pM, gq which admits global symmetries, some of which might be gauge transforma-
tions. It exists a bijection between :

B The equivalence class of gauge parameters λpx µ q that are field symmetries, i.e. such
that the variations of all fields Φi defined on M vanish on shell (δλ Φi « 0).
Two gauge parameters are equivalent if they are equal on-shell.

B The equivalence class of pn ´ 2q-forms k that are closed on shell (dk « 0).
Two pn ´ 2q-forms are equivalent if they differ on-shell by dl where l is a pn ´ 3q-form.

Note that the expression of conserved pn ´ 2q-forms remains ambiguous but the conserved charge
ş
Q “ BΣ krµνs pdn´2 xqµν is not ambiguous! We can always add to k the divergence of a pn ´ 3q-form
and a pn ´ 2q-form that is trivial on shell. But since the integral of an exact form is zero by Stokes’
theorem, it does not modify the conserved charge. We must now understand how we can use this
theorem in a general theory with gauge invariance, how we can extract these conserved forms out of
any theory and discuss the properties of such conserved charges.

As an example, let us show how we can understand the electric charge in classical electrodynamics
under the light of this powerful theorem. We denote the 4-potential by Aµ and the matter current
µ
by JM . Gauge transformations transform Aµ Ñ Aµ ` Bµ λ. We are interested in the non-trivial field
symmetries: the gauge parameters λ ff 0 such as δλ Aµ « 0. There is only one set of symmetries, the
constant gauge transformations, λpx µ q “ c P R0 . As we will derive below, the conserved pn ´ 2q-form
is given by kc rAs “ cF µν pdn´2 xqµν . It is conserved on-shell dkc « 0 outside of matter sources as a
consequence of Maxwell’s equations, Dµ F µν « JM ν . We can integrate k
c“1 on any surface S outside of
ű
matter sources, e.g. t, r both constant and r large in order to get the electric charge Q E “ S kc“1 “
~E ¨ ~er dS. As an exercice, we can check that it conserved in time,
ű
S
¿ ¿
d tr n´2
QE “ Bt k 2pd xqtr “ ´ B A k Ar dS “ 0. (1.17)
dt
S S

In the first equation, we evaluated the form on a contant t, r slice (the two contributions add up
since krt “ ´ktr ). In the second equation, we developed the radial component of dk “ 0, namely
Bt ktr ` B A k Ar “ 0 where pA “ θ, φq are the angular coordinates. The last equation follows from the
fact that the integration of a closed form on a sphere vanishes (assuming of course that the field
strength obeys the free Maxwell equations so without crossing the trajectories of electrons!). Another
point to notice is that ¿ ¿
d tr n´2
QE “ Br k 2pd xqtr “ ´ B A ktA dS “ 0. (1.18)
dr
S S

after using the time component of dk “ 0, namely Br ktr ` B A ktA “ 0 and after assuming again that the
field strength is free at S. More generally, we obtain the standard Gauss law that only the homology

Lecture 1. Surface charges in Gravitation 12

 1.2. Generalized Noether theorem

class of the integration surface matter (i.e. which sources are included in the surface).

1.2.3 Surface charges in generally covariant theories

In electrodynamics, we have just seen that it exists exactly one equivalence class of field symme-
tries, i.e. gauge transformations that vanish but such that the gauge parameter itself is non-zero.
A representative of this non-trivial symmetry is the global gauge transformation by 1 everywhere
in spacetime. The generalized Noether theorem asserts that it is uniquely associated with the con-
served electric charge (we still need to check that explicitly). Now, these symmetries are easily de-
rived because Maxwell theory is linear. General relativity is a non-linear theory and life is more
complicated. Gauge transformations in such a generally covariant theory are diffeomorphisms, and
the ones that do not transform the metric are the isometries whose generators are the Killing vec-
tors δξ gµν “ Lξ gµν « 0. However, for a general spacetime, there is no Killing vector since gµν has no
isometries. So the generalized Noether theorem (Result 4) cannot be applied to any generally defined
diffeomorphism ξ µ . Correspondingly, it seems hopeless to write a formula describing a conserved
pn ´ 2q-form for some suitably defined symmetries ξ µ in a generally covariant theory.

There are however two particular cases where the theorem is just enough: for a family of solutions
with shared exact isometries, and for a set of solutions with a shared asymptotic isometry. Both cases
make good employ of the linearized theory around a suitably chosen solution. Let us consider a
solution ḡµν of general relativity – denoted as the background field – which we perturb by adding
an infinitesimal contribution gµν “ ḡµν ` hµν . It is not difficult to show that the Lagrangian density
linearized around ḡµν and expanded in powers of hµν is gauge-invariant under the transformation
δξ hµν “ Lξ ḡµν , where ξ µ is an arbitrary diffeomorphism. Thus, if the background admits some
Killing symmetries (Lξ ḡµν “ 0), their generators also define a set of symmetries of the linearized
theory δξ hµν “ 0. We can then apply the Result 4 to claim the existence of a set of conserved pn ´ 2q-
forms kξ rh; ḡs if hµν satisfies the linearized equations of motion around ḡµν . This is the key to define
canonical conserved charges in generally covariant theories!

A general method to define symmetries and associated conserved quantities in generic spacetimes
consists in introducing boundary conditions in an asymptotic region where the linearized theory
can be applied around a reference ḡµν that admits several Killing vectors. These isometries are only
asymptotically defined and relevant, and so are named asymptotic symmetries. In effect, they form the
closest analogue in gravity of the group of global symmetries in field theories without gravity. The
most obvious illustration of it can be found by looking at the class of asymptotically flat spacetimes.
A rough definition of such a spacetimes is provided with metrics that approach the Minkowski met-
ric when some suitably defined radial coordinate r is running to infinity with gµν ´ ηµν “ O p1{rq.
Far from the sources of gravitation, the spacetimes becomes approximatively flat : we thus take
ḡµν “ ηµν as background to linearize the theory. Relevant symmetries ξ̄ µ include the 10 symmetries
of Minkowski spacetime that generate the Poincaré algebra, and to each one, a conserved pn ´ 2q-form
is associated by virtue of the generalized Noether theorem. Then we can integrate these pn ´ 2q-forms
on a 2-sphere at infinity to get Poincaré charges of spacetime! As it turns out, there are even more
symmetries leading to additional conserved charges that only exist in gravity, the BMS charges, as

Lecture 1. Surface charges in Gravitation 13

 1.3. Covariant phase space formalism

we will discuss further in these lectures.

So far we considered boundary conditions where the linearized theory can be directly applied asymp-
totically. But there are more ways where the linearized theory is useful. Let’s formalize the concepts
a bit more. The set of metrics that obey the boundary conditions form a set G. A particular met-
ric is singled out in this set: the reference or background solution ḡµν P G. So far we considered
the simple case where all asymptotic charges only depend upon ḡµν and the linearized perturba-
ű
tion hµν “ gµν ´ ḡµν : the charge is then S k ξ rg ´ ḡ; ḡs. But in general, the charge might depend
non-linearly on gµν . The way to define the charge is to linearize the theory around each gµν , by
considering an abstract field variation δgµν . The charge is then defined as
żg¿
Qξ rg; ḡs ” k ξ rdg1 ; g1 s (1.19)
ḡ
S

where we integrate the pn ´ 2q-form both on a 2-sphere S and on a path in G joining the reference
solution ḡµν (e.g. Minkowski) to the solution of interest gµν . The charge is conserved as long as ξ µ
is an asymptotic symmetry in the sense that dkξ rδg; gs « 0 for all gµν P G and all variations that are
“tangent” to G. It is not clear at this point if the charge is independent of the path chosen in G to re-
late ḡµν to gµν . In order to define with more care this construction (in particular in which sense δ can
be viewed as an exterior derivative on the field space G, how “tangent to G” can be defined, discuss
the independence of the path in G, . . . ), and in order to compute from first principles the conserved
pn ´ 2q-forms promised by the result 4, we have to develop more formalism...

But before, let us mention a last but important conceptual point. The fact that the energy, in particu-
lar, is a surface charge in General Relativity can be interpreted as gravity being holographic! Indeed,
in quantum gravity the energy levels of all states of the theory can be found by quantizing the Hamil-
tonian. In the classical limit, the Hamiltonian is a surface charge. If this remains true in the quantum
theory (as it does for example in the AdS/CFT correspondence) knowing the field on the surface
bounding the bulk of spacetime will allow to know all possible states in the bulk of spacetime.

1.3 Covariant phase space formalism

1.3.1 Field fibration and symplectic structure

We work again on a target spacetime M which is a Lorentzian manifold provided with a set of coor-
dinates tx µ u. Let us set aside the metric tensor gµν for the moment. On each point P P M, a tangent
space TP M of vectors vµ can be constructed, which admits a natural coordinate basis tBµ u. The dual
space of TP M is the so-called cotangent space TP‹ M which contains 1-forms wµ spanned by a related
natural basis tdx µ u. Conversely, vectors are also associated with functions on 1-forms through the
interior product.
i : TP M Ñ Linear functions on TP‹ M;
(1.20)
ξ ÞÑ r iξ : TP‹ M Ñ R : w ÞÑ iξ w fi ξ µ Bµ w s.

Lecture 1. Surface charges in Gravitation 14

 1.3. Covariant phase space formalism

We can extend this definition to promote the interior product to an operator iξ : Ωk pMq Ñ Ωk´1 pMq
where Ωk pMq is the set of k-forms defined on M, simply by requiring that iξ w ” ξ µ Bdx
B
µ w, @w P

Ωk pMq. We have also at our disposal a differential operator d “ dx µ Bµ , the exterior derivative that
induces the De Rham complex. Starting from scalars (or pedantically 0-forms), successive applications
of d lead to higher order forms :

Ω0 pMq Ñ Ω1 pMq Ñ Ω2 pMq Ñ ¨ ¨ ¨ Ñ Ωn´1 pMq Ñ Ωn pMq Ñ 0. (1.21)

To be short, we have a first space which is the manifold M of coordinates tx µ u equipped with a natu-
ral differential operator d. Using d, we get forms of higher degree, since d : Ωk Ñ Ωk`1 . One can also
use iξ to ascend the chain of Ω’s.

Now we consider the fields only. We designate them by the compact notation Φ “ pΦi qiPI where
the fields Φi also include the metric field gµν . Fields are abstract entities without dependence in the
coordinates. In order to be complete, the field space or jet space consists in the fields Φi and a set of
“symmetrized derivatives of fields” tΦi , Φiµ , Φiµν , . . . u. In field space, we can again select a “point”
pΦi , Φiµ , Φiµν , . . . q and the cotangent space at that point is then defined as pδΦi , δΦiµ , δΦiµν , . . . q. The
symmetrized derivative is defined such that

B j pµ νq j BΦixy 1
i
Φ “ δ δ
α β i δ , so in particular i
“ . (1.22)
BΦµν αβ BΦ xy 2

The variational operator δ is defined as

B B B
δ “ δΦi i
` δΦiµ i
` δΦiµν `¨¨¨ (1.23)
BΦ BΦµ BΦiµν

It is convenient to use the convention that all δΦi , δΦiµ , . . . are Grassmann odd. It implies that δ2 “ 0.
δ is then an exterior derivative on the field space and each δΦi , δΦiµ , δΦiµν , . . . is a 1-form in field space.

We can now put the manifold and the field space together and we get the jet bundle or variational bicom-
plex. The jet bundle is a manifold with local coordinates px µ , Φpµq
i q where pµq stands for any set of sym-

metrized multi-indices. The fields are all fibers above the target manifold. Taking a section of the fiber,
we obtain the coordinate-dependent fields and their derivatives tΦi px µ q, Bµ Φi px µ q, Bµ Bν Φi px µ q, . . . u.
The standard differential operator d is still defined on the jet space as d “ dx µ Bµ but now

B B B
Bµ ” ` Φiµ i
` Φiµν `¨¨¨ (1.24)
Bx µ BΦ BΦiν

Thenceforth we have two Grassmann-odd differential operators at our disposal: d and δ and the for-
malism ensures that they anti-commute tδ, du “ 0 as you can check. We have a “variational bicomplex”.
i ’s is a pp, qq-form.
A form with p dx µ ’s and q δΦpµq

Lecture 1. Surface charges in Gravitation 15

 1.3. Covariant phase space formalism

Φi , Φiµ , . . .
Φi , Φiµ , . . .
Φi , Φiµ , . . . Field space J TP˚ J “ SpantδΦiI u
Field fibration

Section “ tΦi p xµ q , B α Φi p xµ q , ...u

yµ ‚
xµ ‚ Spacetime M TP‹ M “ Spantdx µ u
zµ ‚

Horizontal derivative = exterior derivative d

Jet bundle “ tpp xµ , Φpµq
i qu
Vertical derivative = variational operator δ

Figure 1.1: Elements from the variational bicomplex structure.

The classical physics of the fields is encoded into a Lagrangian density n-form L and a set of boundary
conditions. Since the Lagrangian is a n-form and depends on the fields, it is a natural object in this
variational bicomplex structure. Let us now revise the formula giving an arbitrary variation of the
Lagrangian density (1.4). Remember that the boundary terms arise after iterative applications of
inverse Leibniz rule. Now, δ has been defined as a 1-form that anticommutes with dx µ so we should
now write
δL
δL “ δΦi ´ dΘrδΦ; Φs . (1.25)
δΦi
We can get back to (1.4) by contracting each side of (1.25) with the inner product iδa where

B
iδa fi δa ΦiI (1.26)
BδΦiI

and we take δa Φ I Grassmann even by definition. The minus sign is compensated by the fact that δ
needs to anticommute with d to reach Θ.

Lecture 1. Surface charges in Gravitation 16

 1.3. Covariant phase space formalism

We name ΘrδΦ; Φs the presymplectic potential. It depends by definition on the fields and their varia-
tions, but not explicitly on the coordinates. It is a pn ´ 1, 1q-form! As a consequence, δΘ is a pn ´ 1, 2q-
form! It is the so-called presymplectic form :

ωrδΦ, δΦ; Φs “ δΘrδΦ; Φs. (1.27)

In order to go back to a notation where variations are more familiar Grassmann even quantities, one
can contract both sides of the equation with the inner product iδ2 iδ1 where iδa is defined as above. The
operator iδ1 hits either the first or second δ so there are two terms; in each case the remaining δ is
replaced by δ2 . Taking into account the sign obtained by anticommuting δ with d we obtain

iδ2 iδ1 ω fi ωrδ1 Φ, δ2 Φ; Φs “ δ1 Θrδ2 Φ; Φs ´ δ2 Θrδ1 Φ; Φs. (1.28)

Our main goal consists in linking the symplectic form that we have just defined on the jet space to
conserved pn ´ 2q-forms that we announced before. But before that, let us make a necessary inter-
mezzo about the second Noether theorem on continuous symmetries, which we will use afterwards
as a lemma !

1.3.2 Noether’s second theorem : an important lemma

Each gauge symmetry of a Lagrangian gives rise to an identity among its equations of motion. This
fundamental property of gauge theories leads to Noether’s second theorem:

Result 5 (Noether’s second theorem)

Given a generally covariant Lagrangian n-form L “ Ldn x and an arbitrary infinitesimal

diffeomorphism ξ µ , one has „ 
δL δL
δξ Φ “ dSξ
i
;Φ
δΦi δΦ
where Sξ is a n ´ 1 form proportional to the equations of motion and its derivatives. The
equality also holds for other types of gauge transformations where ξ µ is then replaced by
an arbitrary gauge parameter of the other type.

Instead of giving a formal proof of this relation, we prefer verify it for two famous gauge theories !

Lecture 1. Surface charges in Gravitation 17

 1.3. Covariant phase space formalism

Einstein’s gravity
1 ?
Let us focus first on the Einstein-Hilbert Lagrangian density L “ 16πG R ´g, where R is the Ricci
curvature associated to the metric field gµν .
ˆ ? ˙
δL 1 1 δ p ´g Rq
dn x ´g ?
a
δξ gµν “ δξ gµν (1.29)
δgµν 16πG ´g δgµν
1
dn x ´g p´G µν qLξ gµν
a
“ (1.30)
16πG
1
dn x ´g G µν ∇µ ξ ν
a
“´ (1.31)
8πG
1 1
dn x ´g ∇µ pG µν ξ ν q ` dn x ´g ∇µ G µν ξ ν
a a
“´ (1.32)
8πGˆ ˙ 8πG
1
“ d n x Bµ ´
a
´g G µν ξ ν (1.33)
8πG
1
pdn´1 xqµ ´g G µν ξ ν
a
ùñ Sξ “ ´ (1.34)
8πG

and the second Noether theorem is proven for this case. In the crucial fourth step, we used Bianchi’s
identities ∇µ G µν “ 0, which is the identity among the equations of motion directly related to general
covariance.

Einstein-Maxwell electrodynamics

As an exercice, we can also show a similar result for classical electrodynamics that is minimally cou-
pled to Einstein’s gravity. The field is the 4-vector potential Aµ . We thus consider the Maxwell field
into a curved spacetime manifold described by its metric tensor gµν . They are two gauge symme-
tries in the game : the classical invariance of electrodynamics Aµ Ñ Aµ ` Bµ λ, and also the invari-
ance under diffeomorphisms, guaranteed by the generally covariant property of a theory coupled
to gravity. The minimal coupling assumption leads us to the Lagrangian n form L “ LG ` LEM “
1 ? n 1? n
16πG R ´gd x ´ 4 ´gF Fαβ d x, where Fαβ “ Bα A β ´ B β Aα is the antisymmetrical Faraday tensor,
αβ

which is gauge invariant and contains the physical electric and magnetic fields.

Let us show as a little lemma that the electromagnetic stress tensor is conserved on-shell. We have :

µν 2 δL EM 1
TEM “ ? “ F µα F να ´ F αβ Fαβ gµν . (1.35)
´g δgµν 4

To do this calculation, the useful formulas are :

a 1a
δ ´g “ ´ggµν δgµν ; δgµν “ ´gµα gνβ δgαβ . (1.36)
2

Recalling that Brα Fβγs “ 0 or equivalently ∇rα Fβγs “ 0 since all Christoffel symbols cancel out by

Lecture 1. Surface charges in Gravitation 18

 1.3. Covariant phase space formalism

antisymmetry, it can be checked that ∇rµ F ναs “ ´ 12 ∇ν Fαµ . We can now check:

µν 1
∇µ TEM “ ∇µ F µα F να ` Frµαs ∇rµ F ναs ´ F αβ ∇ν Fαβ (1.37)
2
1 µα ν 1 αβ ν
“ ∇µ F F α ´ F ∇ Fαµ ´ F ∇ Fαβ
µα ν
(1.38)
2 2
1 1
“ ∇µ F µα F να ` F µα ∇ν Fµα ´ F αβ ∇ν Fαβ (1.39)
2 2
“ ∇µ F F α ,
µα ν
(1.40)

which is indeed zero on-shell.

Let us now compute the left-hand-side of the second Noether theorem. Pay attention to the fact
that in the case of Einstein-Maxwell theory, the gauge parameter is a couple pξ µ , λq (where ξ µ is a
diffeomorphism, and λ a gauge transformation for Aµ ).

δL δL EM BL EM `a ˘ a
p1q “ “ ´Bν “ Bν ´gF νµ “ ´g∇ν F νµ ; (1.41)
δAµ δAµ BBν Aµ
?
δL δL EM δLG ´g µν 1 a
p2q “ ` “ TEM ´ ´gG µν . (1.42)
δgµν δgµν δgµν 2 16πG

And thus, since Φi “ tAµ , gµν u :

δL δL δL
δ Φi “ δ Aµ ` δ gµν . (1.43)
δΦi pξ,λq δAµ pξ,λq δgµν pξ,λq

The potential field Aµ varies under the two gauge transformations : δpξ,λq Aµ “ Lξ Aµ ` δλ Aµ “
Lξ Aµ ` Bµ λ, while gµν is only affected by diffeomorphisms δpξ,λq gµν “ δξ gµν “ Lξ gµν . It remains to
compute the Lie derivative of Aµ on the flow of ξ µ :

Lξ Aµ fi ξ ρ Bρ Aµ ` Aρ Bµ ξ ρ “ ξ ρ Fρµ ` ξ ρ Bµ Aρ ` Aρ Bµ ξ ρ “ ξ ρ Fρµ ` Bµ pξ ρ Aρ q. (1.44)

Inserting all these expressions into (1.43) leads us to

ˆ ˙
δL 1 µν 1
Φ “ ´g∇ν F ξ Fρµ ` Bµ pξ Aρ ` λq ` 2 ´g
i
a “ ρ ‰ a
δ νµ ρ
T ´ G µν
∇µ ξ ν (1.45)
δΦi pξ,λq 2 EM 16πG
ˆ ˙
a µν a a µν 1
“ ´g∇µ TEM ξ ν ` ´g∇ν F ∇µ pξ Aρ ` λq ` ´g TEM ´
νµ ρ
G µν
∇µ ξ ν (1.46)
8πG

where we used (1.40) in the first term. We now apply the inverse Leibniz rule on the third term, and
we remember that G µν is divergence-free, to obtain :
ˆ ˙ ˆ ˙
a µν 1 a µν 1 a µν
´g TEM ´ G µν ∇µ ξ ν “ ´g∇µ TEM ´ G µν ´ ´g∇µ TEM ξ ν (1.47)
8πG 8πG
ˆ ˙
a µν 1 a a µν
“ Bµ ´gTEM ´ ´gG µν
´ ´g∇µ TEM ξ ν (1.48)
8πG
ˆ ˙
δL a µν
“ Bµ 2 ξ ν ´ ´g∇µ TEM ξ ν . (1.49)
δgµν

Lecture 1. Surface charges in Gravitation 19

 1.3. Covariant phase space formalism

The last term is compensated as expected by the first term of (1.46). We then have
ˆ ˙
δL δL
Φ “ ´g∇ν F ∇µ pξ Aρ ` λq ` Bµ 2
i
a νµ ρ
δ ξν (1.50)
δΦi pξ,λq δgµν
ˆ ˙
a “ ‰ δL
“ ´g∇µ ∇ν F pξ Aρ ` λq ` Bµ 2
µν ρ
ξν (1.51)
δgµν
ˆ ˙
“a ‰ δL
“ Bµ ´g∇ν F pξ Aρ ` λq ` Bµ 2
µν ρ
ξν (1.52)
δgµν
„ 
δL ρ δL µ
“ Bµ pξ Aρ ` λq ` 2 ξ ν fi Bµ Spξ,λq . (1.53)
δAµ δgµν

We get the second line thanks to the Bianchi identity ∇µ ∇ν F µν “ 0, and the third one by virtue of
? ?
the property ´g∇µ p¨ ¨ ¨ q ” Bµ p ´g ¨ ¨ ¨ q. We are thus left with a pn ´ 1q-form Spξ,λq that satisfies the
second Noether theorem :
„ 
δL ρ δL
Spξ,λq “ pξ Aρ ` λq ` 2 ξ ν pdn´1 xqµ . (1.54)
δAµ δgµν

We have just proven that the second Noether theorem was valid for both diffeomorphisms and elec-
tromagnetic gauge transformations!

1.3.3 Fundamental theorem of the covariant phase space formalism

Cartan’s magic formula

Since we are considering generally covariant theories, we can always identify the variation along a
diffeomorphism and the Lie derivative along its flow : δξ ” Lξ when acting on tensors. In turn, the
Lie derivative of a tensor can subdivided into several operations using Cartan’s magic formula

Lξ p¨ ¨ ¨ q “ d iξ p¨ ¨ ¨ q ` iξ dp¨ ¨ ¨ q, (1.55)

which makes it useful for deriving algebraic relations. Here, recall that the involution along a vector
ξ µ is defined as iξ “ ξ µ Bµ and d “ dx µ Bµ . We can easily prove it when the Lie derivative acts
on a scalar field φ, since it reduces to the directional derivative on the integral curves of ξ µ . So
Lξ φ “ ξ µ Bµ φ “ iξ dφ which is correct because iξ φ “ 0 (the space of p´1q-forms is empty !). The proof
that Cartan magic’s formula holds for all forms follows by induction from this observation. Indeed,
one can show that, for any integer k, Ωk is generated by scalars, their exterior derivative, and some
exterior products. So we can accept that it is true for all tensors without exhaustively do the proof
here !

Noether-Wald surface charge

Let us now take the variation of L along any infinitesimal diffeomorphism ξ µ :

p1.55q
δξ L “ Lξ L “ dpiξ Lq ` iξ dL “ dpiξ Lq ` 0 (1.56)
p1.4q δL
“ Lξ Φ ` dΘrLξ Φ; Φs. (1.57)
δΦ

Lecture 1. Surface charges in Gravitation 20

 1.3. Covariant phase space formalism

By virtue of Noether’s second theorem (Result 5), we get :

„  ˆ „ ˙
δL µ δL
dpiξ Lq “ dSξ ; Φ ` dΘrLξ Φ; Φs ùñ Bµ ξ L ´ Θ rLξ Φ; Φs ´ Sξ
µ µ
;Φ “ 0. (1.58)
δΦ δΦ

The standard Noether current of field theories is the Hodge dual of the conserved n ´ 1 form

Jξ fi iξ L ´ ΘrLξ Φ; Φs with dJξ “ dSξ ñ dJξ « 0. (1.59)

Now, a fundamental property of the covariant phase space is that a closed form that depends linearly
on a vector ξ µ and its derivatives is locally exact. Therefore, this Noether current can be written as
Jξ “ Sξ ` dQξ . The proof is simple. It relies on the existence of an operator Iξ such that

d Iξ ` Iξ d “ 1. (1.60)

Acting with Iξ on dpJξ ´ Sξ q we get

0 “ Iξ dpJξ ´ Sξ q “ Jξ ´ Sξ ´ dIξ pJξ ´ Sξ q (1.61)

so we deduce that Qξ “ Iξ pJξ ´ Sξ q. The operator Iξ is in fact given by

1 B B
@ωξ P Ωk pMq, Iξ ωξ “ ξα ωξ ` pHigher derivative termsq. (1.62)
n ´ k BBµ ξ α Bdx µ

Since only terms proportional to at least one derivative of ξ α matter and neither Sξ nor iξ L do contain
derivatives of ξ µ we have Iξ Sξ “ Iξ iξ L “ 0 and we have more simply

Qξ rΦs “ ´Iξ Θrδξ Φ; Φs. (1.63)

We will call this pn ´ 2q-form the Noether-Wald surface charge.

We are now ready to state and prove the fundamental theorem:

Result 6 (Fundamental theorem of the covariant phase space formalism)

In the Grassmann odd convention for δ, contracting the presymplectic form with a gauge
transformation δξ Φi , it exists a pn ´ 2, 1q-form kξ rδΦ; Φs that satisfies the identity

ωrδξ Φ, δΦ; Φs « dkξ rδΦ; Φs

where Φi solves the equations of motion, and δΦi solves the linearized equations of motion
around the solution Φi . The infinitesimal surface charge kξ rδΦ; Φs is unique, up to a total
derivative that does not affect the equality above, and it is given in terms of the Noether-
Wald surface charge and the presymplectic potential by the following relation :

kξ rδΦ; Φs “ ´δQξ rδΦ; Φs ` iξ ΘrδΦ; Φs ` total derivative

Lecture 1. Surface charges in Gravitation 21

 1.3. Covariant phase space formalism

Proof

We are considering the jet space where δ is Grassmann odd and anticommutes with the exterior
derivative d. Let us compute
„ 
δL
δSξ ; Φ “ δJξ rδΦ; Φs ´ δdQξ rδΦ; Φs (1.64)
δΦ
“ δiξ LrΦs ´ δΘrLξ Φ; Φs ` dδQξ rδΦ; Φs (1.65)
“ ´iξ δLrΦs ´ δΘrLξ Φ; Φs ` dδQξ rδΦ; Φs (1.66)
ˆ ˙
δLrΦs i
“ ´iξ δΦ ´ dΘrδΦ; Φs ´ δΘrLξ Φ; Φs ` dδQξ rδΦ; Φs (1.67)
δΦi
« iξ dΘrδΦ; Φs ´ δΘrLξ Φ; Φs ` dδQξ rδΦ; Φs. (1.68)

Cartan’s magic formula implies that Lξ ΘrδΦ; Φs “ diξ ΘrδΦ; Φs ` iξ dΘrδΦ; Φs, so we get
„ 
δL
δSξ ; Φ « Lξ ΘrδΦ; Φs ´ diξ ΘrδΦ; Φs ´ δΘrLξ Φ; Φs ` dδQξ rδΦ; Φs (1.69)
δΦ
“ δξ ΘrδΦ; Φs ´ δΘrδξ Φ; Φs ` d δQξ rδΦ; Φs ´ iξ ΘrδΦ; Φs
` ˘
(1.70)
fi ωrδξ Φ, δΦ; Φs ´ dkξ rδΦ; Φs (1.71)

where kξ rδΦ; Φs “ ´δQξ rδΦ; Φs ` iξ ΘrδΦ; Φs ` dp¨ ¨ ¨ q. Now we are about to conclude : the form Sξ
vanishes identically by definition on shell, and if δΦi solves the linearized equations of motion, its
variation vanishes too. So we have proven the fundamental theorem of the covariant phase space
formalism.

Some residual ambiguities

The fundamental theorem allows to uniquely define (up to an exact form) the infinitesimal surface
charge kξ from the presymplectic form. Now, is the definition of the presymplectic form unambigu-
ous?

First notice that the presymplectic potential Θ is ambiguous. If we add a boundary term dM to
the Lagrangian density L Ñ L ` dM, we get exactly Θ Ñ Θ ` δM. However, since ω “ δΘ, this
transformation has no effect on the presymplectic form because δ2 “ 0. Second, we defined Θ from
an integration by part prescription, which gives the canonical definition of Θ but our derivation
goes through by modifying Θ Ñ Θ ´ dB and therefore ω Ñ ω ´ δdB “ ω ` dδB fi ω ` dωB .
This ambiguity reflects our ignorance on how to select the boundary terms in the presymplectic
form. In principle, most of these ambiguities should be related to the so-called “corner terms” in
the action principle, but a generic derivation has not been proven (see one specific example in [4]).
Fortunately, this ambiguity is irrelevant for exact symmetries of the fields (Killing symmetries in the
case of Einstein’s theory) as can be shown quickly:

ωrδξ Φ, δΦ; Φs Ñ ωrδξ Φ, δΦ; Φs ` dωB rδξ Φ, δΦ; Φs ùñ kξ Ñ kξ ` ωB rLξ Φ, δΦ; Φs “ kξ ` 0. (1.72)

Lecture 1. Surface charges in Gravitation 22

 1.4. Conserved surface charges

1.4 Conserved surface charges

Now we will show how the Result 6 can help us defining surface charges for generally covariant and
other gauge theories.

1.4.1 Definition of the charges

We defined so far a pn ´ 2q-form kξ rδΦ; Φs with special properties. We now integrate kξ on a closed
surface S of codimension 2 (e.g. a sphere at time and radius fixed). Doing it, we are left with the
local variation of charge between the two solutions Φi and Φi ` δΦi , where Φi satisfies the equations of
motion, and δΦ their linearized counterpart around Φ. We denote this by
¿
δ{ Hξ rδΦ; Φs “ kξ rδΦ; Φs. (1.73)
S

We denote δ{ instead of δ in order to emphazise that the right-hand-side is not necessarily an exact
differential on the space of fields. If it is the case, the charge will be said integrable, otherwise it is not.

1.4.2 Integrability condition

Let us comment a bit on this very important concept of integrability. δ{ Hξ is a functional depending
on the fields Φi and their variations δΦi . It is obviously a 1-form from the point of view of the fields,
and a scalar on the manifold. But nothing tells us that this 1-form is exact for the exterior derivative δ,
i.e. we are not sure that it exist some Hξ rΦs such as δ{ Hξ rδΦ; Φs “ δpHξ rΦsq. A necessary condition for
allowing the existence of a Hamiltonian generator Hξ associated with ξ is the so-called integrability
condition : ¿ ¿
δ1 kξ rδ2 Φ; Φs ´ δ2 kξ rδ1 Φ; Φs “ 0, @δ1 Φ, δ2 Φ P T rΦs. (1.74)
S S

It is also a sufficient condition if the space of fields does not have any topological obstruction, which
is most often the case.

If the charge is integrable, Hξ exists. In order to define it, we denote by Φ̄i some reference field
configuration, and we continue to denote by Φi our target configuration. Then we select a path γ
linking Φ̄i and Φi in field space, and we perform a path integration along γ
ż ¿
Hξ rΦ; Φ̄s “ kξ rδΦ; Φs ` Nξ rΦ̄s. (1.75)
γ
S

Here Nξ rΦ̄s is a charge associated with the reference Φ̄i that is not fixed by this formalism (it can be
fixed in other formalisms, e.g. the counterterm method in AdS{CFT). The definition of Hξ does not
depend on the path γ chosen precisely because the integrability condition is obeyed.

Lecture 1. Surface charges in Gravitation 23

 1.4. Conserved surface charges

1.4.3 Conservation criterion

Let us suppose from now on that the integrability condition (1.74) is obeyed. The surface charge
Hξ rΦ, Φ̄s is clearly conserved on shell under continuous deformations of S if and only if dkξ rδΦ; Φs «
0 or, equivalently,
Hξ rΦ; Φ̄s is conserved ðñ ωrδξ Φ, δΦ; Φs « 0. (1.76)

We repeat again that “on shell” means here : “Φi solves the equations of motion and δΦi solves
the linearized equations of motion around Φi ”. In many cases, asking for conservation in the entire
spacetime is too stringent, but one at least requires conservation at spatial infinity, far from sources
and radiation. The conservation condition implies that the difference of charge between two surfaces
S1 and S2 vanishes,
ż ¿ ż ¿ ż ż ż ż
ˇ ˇ pStokesq p6q
Hξ ˇS ´ Hξ ˇS “ kξ ´ kξ “ dkξ « ω«0 (1.77)
1 2
γ γ γ C γ C
S1 S2

where C is the codimension one surface whose boundary is S1 Y S2 .

In gravity, we get conserved charges in two famous cases that we have already discussed :

B If ξ µ is an exact (Killing) symmetry, we know that δξ gµν “ Lξ gµν “ 0 so ωrδξ g, δg; gs “ 0.

Therefore, any Killing symmetry is associated with a conserved surface charge in the bulk of
spacetime.

B For asymptotic symmetries, the Killing equation Lξ gµν “ 0 is only verified in an asymptotic
sense when r Ñ 8, so ωrδξ g, δg; gs Ñ 0 only in an asymptotic region. As a consequence, the
charges associated to ξ µ will be conserved only in the asymptotic region.

Morever, we mention a third particular case: the so-called symplectic symmetries, which are vectors ξ µ
that are no longer isometries of gµν but still lead to a vanishing presymplectic form. They also lead
to conserved charges in the bulk of spacetime (see examples in [5, 6]).

1.4.4 Charge algebra

In special relativity, we have 10 Killing vectors and a bracket between these Killing vectors: the Lie
bracket. Under the Lie bracket, the 10 Killing vector form the Poincaré algebra. Moreover, the charges
associated with these vectors represent the algebra of symmetries and also form the Poincaré algebra
under a suitably defined Poisson bracket between the charges. What we want to do now is to derive
this representation theorem for gravity, and for more general gauge theories.

We only consider the most important case of asymptotic symmetries. Let us consider a set G of field
configurations that obeys some boundary conditions. A vector ξ µ is said to be an allowed diffeomor-
phism if its action is tangential to G. In other words, the infinitesimal Lie variation of the fields is a
µ
tangent vector to G, which therefore preserves the boundary conditions that define G. The set tξ a u
µ
c ξ . One can integrate
of such vectors fields form an algebra for the classical Lie bracket rξ a , ξ b sµ “ Cab c
these infinitesimal transformations to obtain their global counterparts, which form a group of allowed

Lecture 1. Surface charges in Gravitation 24

 1.4. Conserved surface charges

transformations, always preserving G.

We assume that the boundary conditions are chosen such that any allowed vector ξ µ asymptotically
solves the Killing equation. We can then define a conserved charge Hξ , which we assume is integrable
and finite3 . Two cases can be distinguished :

B If Hξ is non-zero for a generic field configuration, the action of ξ µ on fields is considered to

have physical content. For example, boosting or rotating a configuration changes the state of
the system.

B If Hξ is zero, the diffeomorphism ξ µ is considered to be a gauge transformation, and does

nothing more than a change of coordinates. These diffeomorphisms are also called trivial gauge
transformations.

We define the asymptotic symmetry group as the quotient

Allowed diffeomorphisms
Asymptotic symmetry group “
Trivial gauge transformations

which extracts the group of state-changing transformations. This is the closest concept in gravity to
the group of global symmetries of field configurations obeying a given set of boundary conditions.
Let us now derive the representation of the asymptotic algebra obeyed by the charges themselves !

Representation theorem

First, we need a Lie bracket for the charges. The definition is the following : for any infinitesimal
diffeomorphisms χµ , ξ µ , we define
¿
tHχ , Hξ u fi δξ Hχ “ kχ rδξ Φ; Φs. (1.78)
S

d t
ş
The last equality directly follows from the definition of the charge (1.75), in the same way as dt 0 dt1 f pt1 q “
f ptq. To derive the charge algebra, we have to express the right-hand-side as a conserved charge for
some yet unknown diffeomorphism. The trick is to use again a reference field Φ̄i to re-introduce a
path integration :
¨ ˛
¿ ¿ ¿
tHχ , Hξ u “ ˝ kχ rδξ Φ; Φs ´ kχ rδξ Φ̄; Φ̄s‚` kχ rδξ Φ̄; Φ̄s (1.79)
S S S
¨ ˛
ż ¿ ¿
“ ˝ δkχ rδξ Φ; Φs‚` kχ rδξ Φ̄; Φ̄s (1.80)
γ
S S

3 If the charge is not finite, it means that the boundary conditions constraining the fields on which we are defining the

charges are too large and not physical. If the charge is not integrable, one could attempt to redefine ξ µ as a function of the
fields to solve the integrability condition.

Lecture 1. Surface charges in Gravitation 25

 1.4. Conserved surface charges

by virtue to the fundamental theorem of integral calculus. The first term needs some massaging.
After using the integrability condition, one can show that
ż ¿ ż ¿
δkχ rδξ Φ; Φs “ krχ,ξs rδΦ; Φs (1.81)
γ γ
S S

and with all fields Φi and their variations δΦi on shell. The proof will be given below for the inter-
ested reader. So we are left with :
ż ¿ ¿
tHχ , Hξ u “ krχ,ξs rδΦ; Φs ` kχ rδξ Φ̄; Φ̄s (1.82)
γ
“ Hrχ,ξs ` Kχ,ξ rΦ̄s (1.83)

where we defined ¿
Kχ,ξ rΦ̄s fi kχ rδξ Φ̄; Φ̄s ´ Nrχ,ξs rΦ̄s. (1.84)

The charge algebra is now determined. It reproduces the diffeomorphism algebra up to an extra
functional Kχ,ξ rΦ̄s that depends only on the reference Φ̄i . For this reason, it commutes with any
surface charge Hξ under the Poisson bracket, and so it belongs to the center of this algebra. Thus
we obtain a central extension when we consider the charges instead of the associated vectors. We can
show that the central extension is antisymmetric under the exchange χµ Ø ξ µ , and

Krχ1 ,χ2 s,ξ rΦ̄s ` Krξ,χ1 s,χ2 rΦ̄s ` Krχ2 ,ξs,χ1 rΦ̄s “ 0,
µ µ
@ξ 1 , ξ 2 , χµ . (1.85)

In other words, Kχ,ξ rΦ̄s forms a 2-cocycle on the Lie algebra of diffeomorphisms, and furthermore
confers to t¨, ¨u a rightful structure of Lie bracket, since the presence of the central extension affects
neither the properties of antisymmetry nor Jacobi’s identity. A central extension Kχ,ξ which cannot
be absorbed into a normalization of the charges Nrχ,ξs is said to be non-trivial. So we have proved
the representation theorem :

Result 7 (Charge representation theorem)

Assuming integrability, the conserved charges associated to a Lie algebra of diffeomor-

phisms also form an algebra under the Poisson braket tHχ , Hξ u fi δξ Hχ , which is isomor-
phic to the Lie algebra of diffeomorphisms up to a central extension

tHχ , Hξ u “ Hrχ,ξs ` Kχ,ξ rΦ̄s.

It remains to prove the remaining equality (1.81), which we provide here for the interested reader.
For that purpose, we need some algebra on the variational bicomplex. We define the operator δQ “
QiI BΦ
B i B i B
i ` δQ I BδΦi and we recall that i Q “ Q I BδΦi . One can check that
I I I

rδQ , ds “ 0, rδQ , δs “ 0, tiQ , Iξ u “ 0, (1.86)

tiQ , δu “ δQ , riQ1 , δQ2 s “ irQ1 ,Q2 s , (1.87)
rδQ1 , δQ2 s “ ´δrQ1 ,Q2 s , rQ1 , Q2 si ” δQ1 Q2i ´ δQ2 Q1i . (1.88)

Lecture 1. Surface charges in Gravitation 26

 1.4. Conserved surface charges

In particular for gravity, we are interested in the operator δLξ Φ that acts on tensor fields as a Lie
derivative, δLξ Φ “ `Lξ in our conventions. Note that the commutator in (1.88) is consistent with the
standard commutator of Lie derivatives:

Φ Lξ 2 Φ
i i
rδLξ
1
Φ , δLξ 2 Φ sΦ “ δLξ
1
´ p1 Ø 2q (1.89)
ΦΦ
i
“ Lξ 2 δLξ ´ p1 Ø 2q (1.90)
1

“ Lξ 2 Lξ 1 Φi ´ p1 Ø 2q (1.91)
“ ´Lrξ 1 ,ξ 2 s Φi ´ p1 Ø 2q (1.92)
ΦΦ
i
“ ´δLrξ ´ p1 Ø 2q. (1.93)
1 ,ξ 2 s

With these tools in mind, let us start the proof. Applying the operator Iξ on the fundamental relation
of Result 6, we obtain the definition of the surface charge form from the presymplectic form,

Iξ ωrδξ Φ, δΦ; Φs « Iξ dkξ rδΦ; Φs (1.94)

« kξ rδΦ; Φs ` dp. . . q. (1.95)

Contracting with iδχ Φ we further obtain

kξ rδχ Φ; Φs « Iξ ωrδχ Φ, δξ Φ; Φs ` dp. . . q. (1.96)

We would like to compute

δkξ rδχ Φ; Φs « δIξ ωrδχ Φ, δξ Φ; Φs ` dp. . . q (1.97)

« ´Iξ δωrδχ Φ, δξ Φ; Φs ` dp. . . q, (1.98)
« ´Iξ δiδξ Φ iδχ Φ ωrδΦ, δΦ; Φs ` dp. . . q. (1.99)

We would like to use the fact that the presymplectic structure is δ-exact, δωrδΦ, δΦ; Φs “ 0, so we
will (anti-)commute the various operators as

δiδξ Φ iδχ Φ “ ´iδξ Φ δiδχ Φ ´ δδξ Φ iδχ Φ (1.100)

“ iδξ Φ iδχ Φ δ ` iδξ Φ δδχ Φ ´ δδξ Φ iδχ Φ (1.101)
“ iδξ Φ iδχ Φ δ ` δδχ Φ iδξ Φ ` irδξ Φ,δχ Φs ´ δδξ Φ iδχ Φ (1.102)
“ iδξ Φ iδχ Φ δ ` δδχ Φ iδξ Φ ´ iδrξ,χs Φ ´ δδξ Φ iδχ Φ . (1.103)

The first term does not contribute as announced. The second and fourth term in fact combine to a
contraction of the integrability condition (1.74) after using the definition (1.95) of the surface charge
in terms of ω. We refer to [7] for this piece of the proof. We are then left with

δkξ rδχ Φ; Φs « Iξ ωrδrξ,χs Φ, δΦ; Φs ` dp. . . q (1.104)

« krξ,χs rδΦ; Φs ` dp. . . q (1.105)

which proves (1.81).

Lecture 1. Surface charges in Gravitation 27

 1.4. Conserved surface charges

This closes our presentation of the covariant phase space formalism. What we have discussed is in
fact one general method to derive canonical conserved charges in gauge theories. Let us now make
some explicit calculations in General Relativity, to illustrate a bit all these formulae that we have just
written...

1.4.5 Conserved charge formula for General Relativity

?
Let us consider the Hilbert-Einstein Lagrangian density L “ 1
16πG ´g R. The only field Φi to take
into account is the metric tensor gµν , whose local variation will be denoted as hµν “ δgµν (convention:
δ is Grassmann even). First, we need the expression of a general perturbation of L : the calculation is
straightforward and you should already performed it during your gravitation classes, in particular
when you extracted the Einstein’s equations from the variational principle, so it is left as an exercise :
?
´g µν
δL “ ´ G hµν ` Bµ Θµ rh; gs; (1.106)
? 16πG
´g
Θ rh; gs “
µ
p∇ν hµν ´ ∇µ hνν q (1.107)
16πG

where ∇α is the Levi-Civita connection compatible with gµν . If the variation is contracted with the
action of a diffeomorphism ξ µ , we are able to explicit the presymplectic superpotential :
?
´g ´ ¯
Θ rLξ g; gs “
µ
2∇ν ∇pµ ξ νq ´ 2∇µ ∇ν ξ ν . (1.108)
16πG

Recalling the definition of Riemann’s curvature tensor, one gets easily that ∇µ ∇ν ξ ν “ ∇ν ∇µ ξ ν `
ν µ α
R α νξ « ∇ν ∇µ ξ ν because the last term is proportional to the Ricci tensor which vanishes on shell
for pure gravity without matter. So :
?
´g
Θ rLξ g; gs «
µ
∇ ν p∇ ν ξ µ ´ ∇ µ ξ ν q . (1.109)
16πG

Knowing the symplectic prepotential gives us access to the Noether-Wald charge (1.63) after some
derivations :
? ?
´g ´g µ ν n´2
Qξ “ ´Iξ Θrδξ g; gs “ µ ν ν µ n´2
p∇ ξ ´ ∇ ξ q pd xqµν “ ∇ ξ pd xqµν (1.110)
16πG 8πG

which is often called Komar’s term, in reference to the Komar’s integrals that give the mass and angu-
lar momentum of simple spacetimes when (1.110) is evaluated on the asymptotic 2-sphere. The last
ingredient we need is

B
´ iξ Θ “ ´ξ ν Θµ pdn´1 xqµ “ pξ µ Θν ´ ξ ν Θµ q pdn´2 xqµν . (1.111)
Bdx ν

The total surface charge is kξ rh; gs “ ´δQξ rgs ´ iξ Θrh; gs where the last minus sign is valid in the
Grassmann even convention for hµν since iξ is Grassmann odd! Finally, after some tensorial algebra,
we are left with :
? ˆ ˙
´g n´2 1
kξ rh; gs “ pd xqµν ξ ∇σ h
µ νσ
´ ξ ∇ h ` ξσ∇ h
µ ν ν µσ
` h∇ ξ ´ h ∇ρ ξ .
ν µ ρν µ
(1.112)
8πG 2

Lecture 1. Surface charges in Gravitation 28

 1.5. Conserved charges from the equations of motion

One can explicitly prove that this object is conserved when gµν and hµν are on shell and ξ µ is a Killing
vector of gµν :
rµνs
dkξ rh; gs « 0 ðñ Bν k ξ pdn´2 xqµν « 0. (1.113)

Don’t forget that it remains an ambiguity on this surface charge, which appears when we attempt
to add a boundary term to the presymplectic form. If we impose that this term is only made up of
covariant objects, the form of this term is highly constrained. Indeed, one can be convinced that the
only boundary symplectic form constituted from gµν is :

1
pδgq σ ^ pδgqσν pdn´2 xqµν .
µ
Erδg, δg; gs “ (1.114)
16πG

When the variations are generated by an infinitesimal diffeomorphism ξ µ , (1.114) results in

1
Erδξ g, δg; gs “ p∇µ ξ σ ` ∇σ ξ µ q pδgqσν pdn´2 xqµν . (1.115)
16πG

It is not surprising to obtain a contribution proportional to the Killing equation, since we have al-
ready shown that charges associated to exact symmetries do not suffer from any ambiguity ! The
charge kξ rh; gs ` α Erδξ g, h; gs is the Iyer-Wald charge [8] when α “ 0 and the Abbott-Deser charge [9]
when α “ 1.

Let us conclude this section by performing a concrete calculation on the most simple black hole met-
ric: the Schwarzchild metric. In spherical coordinates pt, r, θ, φq, we can describe the region outside
the horizon by

2m ´1 2
ˆ ˙ ˆ ˙
2m
gµν rms “ ´ 1 ´ dt2 ` 1 ´ dr ` r2 dΩ2 with dΩ2 “ dθ 2 ` sin2 θdφ. (1.116)
r r

Bg
Only the mass parameter m labels the family of metrics. Therefore, hµν fi δgµν rδm, ms “ Bmµν δm.
We find δgµν dx µ dx ν “ 2δm 2 ` 2δm 1 ´ 2m ´2 dr 2 . Choosing a 2-surface S on which both t and r are
` ˘
r dt r r
constant and fixing ξ “ Bt , a direct evaluation of (1.112) with the natural orientation etrθφ “ `1 shows
that : ¿ ż 2π żπ
δm δm δm
δ{ Hξ “ dΩ “ dφ dθ sin θ “ “ δM. (1.117)
4πG 0 0 4πG G
S

where M “ m{G is the total mass of spacetime. So the charge is trivially integrable and, after a simple
path integration between the Minkowski metric (m “ 0) and a target metric with given m ą 0, we
get the right result according to which M is the total energy of the Schwarzschild black hole !

1.5 Conserved charges from the equations of motion

In this section, we quickly discuss another way to define conserved charges through pn ´ 2q-forms.
This will lead to a particular prescription to fix the boundary ambiguity in the presymplectic form.
This method is sensitively the same as the Iyer-Wald’s one, and it also relies on the link between the
symplectic structure of the space of fields and lower degree conserved currents.

Lecture 1. Surface charges in Gravitation 29

 1.5. Conserved charges from the equations of motion

1.5.1 Anderson’s homotopy operator

We first introduce a more formal procedure for performing integration by parts on expressions that
do not necessarily involve ξ µ but must involve the fields Φi . It involves the fundamental operator,
p
called Anderson’s homotopy operator IδΦ , which bears some ressemblance with the operator Iξ con-
structed and used above. Using the Grassmann odd convention for δ its constitutive relations are

n δ
´dIδΦ ` δΦi “δ when acting on n-forms ; (1.118)
δΦi
p p`1
´dIδΦ ` IδΦ d “ δ when acting on p-forms pp ă nq. (1.119)

As an exercise, the reader can convince him/her-self that the correct definition is
„ 
n i B i B i B B
IδΦ “ δΦ ´ δΦ Bν ` Bν δΦ `¨¨¨ , (1.120)
BBµ Φ i BBµ Bν Φ i BBµ Bν Φ i Bdx µ
„ 
n´1 1 i B 1 i B 2 i B B
IδΦ “ δΦ ´ δΦ Bν ` Bν δΦ `¨¨¨ (1.121)
2 BBµ Φ i 3 BBµ Bν Φ i 3 BBµ Bν Φ i Bdx µ

where higher derivative terms are omitted.

1.5.2 Invariant presymplectic current

Recalling that the Lagrangian density can be promoted to a n-form L, we can use (1.118) on it :

δL δL
δL “ δΦi n
´ dIδΦ L fi δΦi i ´ dΘrδΦ; Φs. (1.122)
δΦi δΦ

So the definition Θ “ IδΦ

n L fixes the boundary term ambiguity in Θ. Note the global sign in front

of Θ, because d and δ are both Grassmann-odd and anticommute. We always define the Iyer-Wald
presymplectic current as ωrδΦ, δΦ; Φs “ δΘrδΦ; Φs, and using (1.119), we can apply IδΦ
n on both

sides of (1.122) to get :

ˆ˙
n n δL i n n
IδΦ
“ δL IδΦ
δΦ ´ IδΦ dIδΦ L (1.123)
δΦi
ˆ ˙
n δL n´1 n
“ IδΦ δΦi i ´ δIδΦ n
L ´ dIδΦ IδΦ L (1.124)
δΦ
ˆ ˙
n n n i δL n´1 n
ùñ IδΦ δL ` δIδΦ L “ IδΦ δΦ ´ dIδΦ IδΦ L. (1.125)
δΦi

n s “ 0 because δ2 “ 0, the left-hand-side is nothing but 2 δI n L “ 2 ωrδΦ, δΦ; Φs, and so :

Since rδ, IδΦ δΦ

ωrδΦ, δΦ; Φs “ WrδΦ, δΦ; Φs ` dErδΦ, δΦ; Φs (1.126)

where we have isolated and defined the invariant presymplectic current :

ˆ ˙
1 n i δL
WrδΦ, δΦ; Φs fi IδΦ δΦ . (1.127)
2 δΦi

Lecture 1. Surface charges in Gravitation 30

 1.5. Conserved charges from the equations of motion

It differs from the Iyer-Wald one by a boundary term that reads as :

1 n´1 n
ErδΦ, δΦ; Φs fi ´ IδΦ IδΦ L. (1.128)
2

This formulation allows us to choose W instead of ω as symplectic form to build conserved surface
charges. It is called invariant because it is defined in terms of the equations of motion and does not
depend upon the boundary terms added to the action.

Let us consider again an infinitesimal diffeomorphism ξ µ . In order to compute Wrδξ Φ, δΦ; Φs we

need to contract either of the two δΦ on the right-hand side of (1.127). There are therefore two terms.
Now, it is a mathematical fact of the variational bicomplex that these two terms are equal,
ˆ ˙ ˆ ˙
δL δL
ILn ξ Φ δΦi n
“ ´IδΦ Lξ Φi (1.129)
δΦi δΦi

The proof is given in the Appendix of [7] (denoted as Proposition 13). Therefore,
ˆ ˙
δL
WrLξ Φ, δΦ; Φs fi iLξ Φ W « n
´IδΦ Lξ Φ .
i
(1.130)
δΦi

The trick to progress is to consider Noether’s second theorem

δL
dSξ “ Lξ Φi . (1.131)
δΦi
n to both sides :
and apply Anderson’s operator IδΦ
ˆ ˙
δL
n
IδΦ dSξ “ n
IδΦ Lξ Φi (1.132)
δΦi
n´1
“ δSξ ` dIδΦ Sξ . (1.133)

If Φi is on shell and δΦi is also on shell in the linearized theory, the variation δSξ vanishes. Using
(1.130) we are left with a familiar formula :

WrLξ Φ, δΦ; Φs « dkξBB rδΦ; Φs (1.134)

where now the invariant surface charge form or Barnich-Brandt charge form is
„ 
δL
kξBB rδΦ; Φs “ n´1
IδΦ Sξ ;Φ . (1.135)
δΦ

The surface charges are obtained by integration on a 2-surface and on a path in field space, as before.

1.5.3 Expression of Barnich-Brandt’s charge for Einstein’s gravity

The computation of the Barnich-Brandt’s charge for General Relativity can be performed thanks to
the formula (1.135), and with the mere knowledge of Sξ already derived, see (1.34). But it is not
necessary, since using (1.126) we have kξBB rδΦ; Φs “ kξIW rδΦ; Φs ` Erδξ Φ, δΦ; Φs. Therefore, the two

Lecture 1. Surface charges in Gravitation 31

 1.5. Conserved charges from the equations of motion

formulations differ by this ambiguous term, which can be computed explicitly with (1.128). In doing
so we get exactly (1.114), so the Barnich-Brandt’s local charge for Einstein’s theory reads as follows
? ˆ ˙
µν ´g 1 1 ρν 1 ν µ σ
kξ rh; gs “ ξ ∇σ h
µ νσ
´ ξ ∇ h ` ξσ∇ h
µ ν ν µσ
` h∇ ξ ´ h ∇ρ ξ ` h σ ∇ ξ .
ν µ µ
(1.136)
8πG 2 2 2

This formula was also obtained by Abbott and Deser by a similar procedure involving integrations
by parts, without using formal operators [9, 10, 11].

When we will show that 3-dimensional Einstein’s gravity can be reduced to a couple of Chern-Simons
theories, we will use this formulation of conserved charges (instead of the Iyer-Wald one) to compute
the charges in that alternative formalism, simply because it is faster.

References
Many textbooks on QFTs explain Noether’s theorem in detail, as e.g. the book of di Franscesco et al.
[12].

The covariant phase space formalism was developed in [13, 8]. Several introductions to the formal-
ism can be found in research articles including e.g. Section 3 of [14] and Appendix A of [15]. For a
list of references on the definition of the symplectic structure for Einstein gravity, see e.g. Section 4.4
of [16].

The cohomological formalism for defining the surface charges (and in particular the proof of unique-
ness or “Generalized Noether theorem") was developed in [3, 17, 7]. For additional details on the
variational bicomplex, see e.g. Appendix C of [18] or Section I.1. of [19].

The Hamiltonian formalism, which also leads to a complete theory of conserved charges and which
is equivalent to the covariant phase space formalism, was developed in [20, 21, 22]. Due to a lack of
time, it was not covered in these lectures.

The proof of the representation theorem for the algebra of (integrable) charges can be found in the
case of the Hamiltonian formalism in [22] and for the covariant formalism in [7].

The definition of surface charges using the equations of motion was developed for Einstein gravity
and higher curvature theories in [9, 10, 11]. Only the Abbott-Deser formula for Einstein gravity was
covered here. These definitions are equivalent to the cohomological formalism.

If you need to use surface charges for various theories of second order in derivatives (Einstein,
Maxwell, Chern-Simons, scalars), the explicit formulae for the surface charges with all signs and
factors right up to my knowledge can be found in Section 4.4. of [14]. For those interested, a Math-
ematica package is also available to compute surface charges in several theories with tutorial on my
homepage.

Lecture 1. Surface charges in Gravitation 32

 Lecture 2

Three dimensional Einstein’s gravity

General Relativity is the commonly accepted modern paradigm of gravitation, and is supported
nowadays by compelling experimental evidence, which consolidates its position as the unavoidable
model of gravitation. Yet, it is a very complex theory presenting a lot of puzzles to physicists, both
at the classical and quantum level. Classically, one has to deal with laborious analytic or numerical
calculations and some conceptual issues that still remain unresolved. Quantum mechanically, Ein-
stein’s theory is not well-behaved and after several years of intense searches, quantum gravity is still
an elusive theory without direct experimental prospect.

Now, it exists a reduced version of Einstein’s general relativity where the physical objects are more
under control and where its quantization, if it is consistent, is within our reach : 3-dimensional Ein-
stein’s gravity to which this lecture is dedicated. As a toy-model, it is a very useful framework thanks
to which we can experiment some techniques and derive features, some of which extend to the phys-
ical 4d case. As one removes a spatial dimension, the solution space is reduced as well, in fact to
constant curvature solutions in the absence of matter. But by considering topological and asymptotic
properties, Einstein’s solutions still contain black holes and infinite-dimensional asymptotic symme-
tries and so save the theory from its apparent triviality.

Anti-de Sitter spacetime, one of the three constant curvature spacetimes, was studied in the early
80’s. The seminal paper of Brown and Henneaux in 1986 [23] gave a definition of asymptotically
AdS3 spacetimes together with an analysis of its infinite dimensional symmetries. The paper was
ignored for long, until the discovery of the AdS{CFT correspondence by Maldacena in 1997 [24, 25].
Notably in 1988, Witten attempted a quantization of 3d gravity [26] by using its equivalent Chern-
Simons representation, constructed by Achúcarro and Townsend in 1986 [27]. Most of the community
considered the theory as trivial, until a black hole solution, the BTZ black hole, was derived in 1992
by Bañados, Teitelboim and Zanelli [28, 29], which shares some features with the 4d Kerr solution. In
string theory, many higher dimensional supersymmetric black holes contain the BTZ black hole in
their near-horizon geometry, and 3d Einstein gravity becomes a universal tool to understand black
hole entropy, as emphasized by Strominger in 1997 [30]. The black hole entropy of any BTZ black
hole can be “holographically” computed using Cardy’s formula of a dual putative CFT2 thanks to
an extension of Maldacena’s AdS{CFT correspondence. Today, it is not yet clear whether pure 3d
Einstein gravity makes sense quantum mechanically without a string theory embedding. Several at-
tempts have been made but are not conclusive [31, 32].

In the last years, the holography community has paid a particular attention to 4d asymptotically flat
spacetimes where another infinite dimensional symmetry group, the BMS group, plays a particular
role. There is also a 3d analogue of this BMS group, and again it is useful to understand the toy
model in parallel to the physical 4d case.

Lecture 2. Three dimensional Einstein’s gravity 33

 2.1. Overview of typical properties

Through this course, we will review the typical properties of 3d gravity, which are mostly due to the
vanishing of the Weyl curvature. Next we will turn to the AdS3 phase space: we will describe global
features of AdS3 itself, and then give several elements on the Brown-Henneaux boundary conditions
and the resulting asymptotic group, without forgetting a long discussion on BTZ black holes. We will
show that the asymptotically flat phase space can be obtained from the flat limit of the AdS3 phase
space. Finally, we will shortly present the Chern-Simons formulation of 3d gravity, which reduces
the theory to the one of two non-abelian gauge vector fields!

2.1 Overview of typical properties

2.1.1 A theory without bulk degrees of freedom

Beyond the fact that the analytic complexity of General Relativity is reduced in 2 ` 1 dimensions, it
occurs a particular phenomenon that heavily constraints the gravitational field. To see how 3d gravity
is special, let us consider a n-dimensional spacetime manifold pM, gq. On M, the intrinsic curvature is
encoded in the Riemann tensor Rµναβ , which natively possesses n4 components. This number can be
n2 2
reduced to 12 pn ´ 1q independent components, after taking into account several well-known indicial
symmetries :

B Antisymmetry on the pair pµ, νq Ñ N “ npn ´ 1q{2 independent choices of pµ, νq ;

B Antisymmetry on the pair pα, βq Ñ N “ npn ´ 1q{2 independent choices of pα, βq ;

B Symmetry under permutation of the pairs Ñ reduces the number of independent choices of
pairs from N 2 to NpN ` 1q{2 ;

B Identity Rµpναβq “ 0. Any repeated index would lead to zero after using the above three proper-
ties. Indeed, for a repeated index in the last three symmetrized indices, R1p233q “ R1332 ` R1323 “
0, and for repeated mixed indices, R1p123q “ R1231 ` R1312 “ R1231 ` R1213 “ 0. Each identity
with distinct indices (up to reschuffling the indices) brings new constraints. The number of
new constraints brought by this last identity is therefore equal to C “ npn ´ 1qpn ´ 2qpn ´ 3q{4!.
We see that if n ă 4, it brings no additional restriction.

Finally we are left with the correct number of independent components : Fpnq “ NpN ` 1q{2 ´ C “
n2 2
12 pn ´ 1q. Now we recall that the Riemann tensor can be written with an explicit decomposition
between the trace-part (the Ricci tensor Rµν “ Rαµαν ) and the traceless conformally invariant part (the
Weyl tensor Wµναβ , sometimes also denoted as Cµναβ ) :

2 ` ˘ 2
Rµναβ “ Wµναβ ` gαrµ Rνsβ ` Rαrµ gνsβ ´ Rg g . (2.1)
n´2 pn ´ 1qpn ´ 2q αrµ νsβ

According to Einstein’s equations Gµν “ 8πGTµν the source of curvature that governs the Ricci ten-
sor is the local distribution of matter energy-momentum and stresses. Outside the sources, the Ricci
curvature vanishes (Rµν “ 0). Yet, the Weyl curvature might not vanish. It carries the gravitational
information of the sources on the local observer (gravitational waves, Newtonian potential and ad-
ditional Einsteinian potentials).

Lecture 2. Three dimensional Einstein’s gravity 34

 2.1. Overview of typical properties

But now in 3d, a simplification occurs, because the number of independent components in the Rie-
n2 2 9
mann tensor 12 pn ´ 1q “ 12 ˆ p9 ´ 1q “ 6 perfectly matches with the number of independent com-
ponents of the Ricci tensor, which is symmetrical, and thus owns 21 npn ` 1q “ 3
2 ˆ 4 “ 6 independent
components ! So the Weyl tensor is identically zero. That implies that there is no gravitational degree
of freedom in 3d Einstein’s gravity because gravitational information cannot propagate. Also, there
is no Newtonian potential, and so the masses do not attract ! This astounding fact allows to extend
the mass spectrum of the theory to negative masses (as long as the spectrum is bounded from below
for stability reasons): indeed negative masses cannot give rise to repulsion since two masses do not
“feel” each other in 3d gravity !

2.1.2 Einstein-Hilbert action and homogeneous spacetimes

Until now we didn’t use the equations of motion. Now let us write the Einstein-Hilbert action, and
obtain another characterization of the local triviality of 3d gravity :
ż
1
d3 x
a
Srgs “ ´gpR ` 2Λq (2.2)
16πG M

where R is the Ricci scalar curvature, and Λ the cosmological constant. In the case of pure gravity, the
equations of motion are Gµν ` Λgµν “ Rµν ` pΛ ´ R2 qgµν “ 0. Taking the trace we find that R “ 6Λ
(since gµν gµν “ n “ 3). On shell, the Ricci tensor is completely determined by the metric tensor
Rµν « 2Λgµν . Inserting this expression into (2.1) after deleting the Weyl tensor, one gets :

Rαβµν « Λ gαµ gβν ´ gαν gβν .

` ˘
(2.3)

This is the curvature tensor of an homogeneous spacetime or constant curvature spacetime. The solu-
tions are thus distinguished by the sign of the cosmological constant Λ :

B If Λ ą 0, the solutions of Einstein’s field equations are locally the de Sitter space ;

B If Λ “ 0, the solutions of Einstein’s field equations are locally Minkowski (flat) ;

B If Λ ă 0, the solutions of Einstein’s field equations are locally the anti-de Sitter space.

Remark that the spacetime being locally homogeneous everywhere is consistent with the non-interaction
between masses. Indeed, if we consider a set of test masses, the region that separates them is locally
an homogeneous spacetime where no preferential direction can be chosen, and in particular there is
no direction of attraction. The solution space of 3d gravity is however not trivial! There are particles
(local defects), black holes (due to topological properties) and asymptotic symmetries, as discussed
later on!

Lecture 2. Three dimensional Einstein’s gravity 35

 2.2. Asymptotically anti-de Sitter phase space

2.2 Asymptotically anti-de Sitter phase space

2.2.1 Global properties of AdS 3

Let us focus on negative curvature solutions. We introduce a
length scale ` P R`
0 such as Λ “ ´1{` . Since we are first in-
2 Future timelike infinity

terested in the global homogeneous solution, we provide our-

selves with global coordinates pt, r, φq, in the sense that any com-
plete geodesic can be maximally extended in this coordinate sys-
τ
tem. t is a timelike coordinate, while r, φ are spacelike coordi-
nates, r P R` is the luminosity distance and φ P r0, 2πs is the

Spacelike infinity
angular coordinate. The maximally symmetric solution of Ein-
stein’s equation with negative curvature reads as :
˙´1
r2 r2
ˆ ˙ ˆ
2 2
ds “ ´ 1 ` 2 dt ` 1 ` 2 dr2 ` r2 dφ2 . (2.4)
` `

This spacetime is called the anti-de Sitter spacetime (AdS3 ). The

length scale ` determines the distance at which the curvature
sets in. At shorter distances (r ! `), the metric (2.4) is simply
`2 2 2
Minkowski. Near infinity (r " `), we get ds2 „ r2
dr ` `r 2 p´dt2 `
`2 dφ2 q so the asymptotic boundary is a cylinder, with the φ circle φ

as a base and extending along the timelike coordinate t. Given Past timelike infinity
that there is a potential barrier to reach infinity, one may think
Figure 2.1: Penrose diagram of
about AdS3 as a “spacetime in a box”, and its Penrose diagram
global AdS3 spacetime.
can be easily obtained by compactifying the luminosity distance
to bring the asymptotic cylinder to a finite distance. Before to do
that, let us introduce a new coordinate system in which the luminosity distance r is remplaced by
another (dimensionless) radial coordinate ρ as r fi ` sinh ρ. The line element takes the form
´ ¯
ds2 “ `2 dρ2 ´ cosh2 ρ dτ 2 ` sinh2 ρ dφ2 (2.5)

where we have rescaled τ “ t{` for later convenience. We wish to compactify the radial direction
given by ρ. So let us define a "conformal radial coordinate" q such as dq “ dρ{ cosh ρ, or cosh ρ “
1{ cos q. Since ρ ě 0, cosh ρ P r1, 8r and it is mapped to q P r0, π{2r. In the patch pq, τ, φq the metric is
brought into the form

`2 `2 `
ds2 “ 2 2 2 2
´dτ 2 ` dΩ2D .
˘
2
pdq ´ dτ ` sin q dφ q “ 2
(2.6)
cos q cos q

Surfaces of constant τ are half-spheres or disks (since the colatitude q must be lower than π{2) with
boundary at q “ π{2 (ρ “ 8). At this boundary, the conformal metric is simply p´dτ 2 ` dφ2 q as
expected : one can thus actually visualize AdS3 as an infinite cylinder as represented by the figure
2.1. But note that since the range of the timelike coordinate τ remains infinite while that of q is fi-
nite, there is no way to compress AdS3 into a finite range of coordinates for both τ and q if we want

Lecture 2. Three dimensional Einstein’s gravity 36

 2.2. Asymptotically anti-de Sitter phase space

to preserve the condition that lightrays are diagonal. If we attempt to perform another conformal
transformation to reduce τ in a finite range, the circles generating the conformal boundary will be
mapped to points, since the interval of the conformal radius q will be squeezed to a single point. So
we admit that the conformal diagram is that we have drawn : it is an infinite cyclinder, whose upper
and lower boundary, rejected to infinity, are respectively future and past timelike infinities.

We have still more to learn from the metric written in the new global coordinates pρ, τ, φq. First we
clearly remark that AdS3 has the topology of R3 . It is also clear that AdS3 is static, because Bτ is a
trivial Killing vector of gµν (since no component depends on τ) that is also orthogonal to the constant
time slices (gτ a “ 0). Since the staticity is manifest in this coordinate system, it bears the name of
global static coordinates. We can calculate the trajectories of particles in AdS3 , which is also more easy
in this system. Let us focus only on radial null geodesics x µ pλq which have to verify
ˆ ˙2 ˆ ˙2
2 2 2 2 dρ 2 dτ
ds “ 0 ñ dρ “ cosh ρ dτ ñ “ cosh ρ . (2.7)
dλ dλ

To integrate this equation on λ, we can use a dynamical invariant which is the energy of the null ray:

dx ν dτ dτ
E fi ´Bτ ¨ u fi ´gµν pBτ qµ “ ´gττ “ cosh2 ρ . (2.8)
dλ dλ dλ

So we have to integrate now

˙2
E2
ˆ
dρ d
“ 2
ùñ sinh ρ “ `E ðñ sinh ρpλq “ Epλ ´ λ0 q (2.9)
dλ cosh ρ dλ

for outgoing lightrays. We find thus that spatial infinity ρ Ñ 8 is reached when the affine parameter
λ Ñ 8. But for the coordinate time τ, we directly integrate

dτ E E
“ 2
“ 2 pλ ´ λ q2
ùñ τ “ arctan rEpλ ´ λ0 qs (2.10)
dλ cosh ρpλq 1 ` E 0

and thus when λ Ñ 8, τ “ π{2, so null geodesics reach infinity after a finite coordinate time interval
(see the blue curve on figure 2.1) ! AdS3 spacetime is said not to be globally hyperbolic: one needs
boundary conditions at spatial infinity to arbitrarily extend the dynamics in coordinate time in the
bulk of spacetime. Equivalently, Cauchy surfaces do not exist.

Concerning timelike geodesics, it can be shown thanks to a cautious examination of the AdS3 geodesic
equation that they cannot reach the boundary, and are in fact reflected at large distances. The Christof-
fel symbols contain a potential proportional to r2 that blows up at infinity and confines timelike ob-
jects to the center of spacetime. Some timelike geodesics are drawn in red on the conformal diagram
in Figure 2.1.

To study the symmetries of AdS3 , the most convenient way is to first realize that one can immerse
it into the 4-dimensional manifold Rp2,2q , the space R4 provided with the pseudo-riemannian metric
ζ ab “ diagp´1, ´1, `1, `1q. On this manifold, we denote by X0 and X̃0 the timelike coordinates,

Lecture 2. Three dimensional Einstein’s gravity 37

 2.2. Asymptotically anti-de Sitter phase space

and (X1 , X2 ) the spacelike ones. AdS3 can be viewed as an hyperboloid isometrically immersed into
Rp2,2q :
AdS3 ” H fi tX µ P Rp2,2q | ´ X02 ´ X̃02 ` X12 ` X22 “ ζ ab X a X b “ ´`2 u. (2.11)

We immediately see that a natural parametrization of this hypersurface is the following1 :

$
’
’
’ X0 “ ` cosh ρ cos τ
’
& X̃
0 “ ` cosh ρ sin τ
X µ P H ðñ pρ P R` , τ, φ P r0, 2πrq. (2.12)
’
’
’ X1 “ ` sinh ρ cos φ
’
% X
2 “ ` sinh ρ sin φ

2
The application is isometric because the pull-pack of the ambient metric dSp2,2q “ ζ ab dX a dX b on H
exactly reproduces the metric of AdS3 in global coordinates :
ˇ ´ ¯
2
dSp2,2q “ `2 dρ2 ´ cosh2 ρ dτ 2 ` sinh2 ρ dφ2 . (2.13)
ˇ
ˇ
H

We can now easily analyze the isometries of AdS3 . The host space Rp2,2q is 4-dimensional flat space
with 2 timelike directions. It possesses the maximal number of symmetries allowed in 4d, namely 10.
The isometries consist of the 4 translations along each direction (because the metric does not depend
on the coordinates X a ) and the 6 matricial transformations that preserve the metric ζ : M´1 ζ M “
ζ ñ M P SOp2, 2q by definition. All these symmetries cannot survive on H : in particular, the
hypersurface H is clearly not invariant under translations X a Ñ X a ` A a , but acting with SOp2, 2q
still preserves the condition ζ ab X a X b “ ´`2 , so we have obtained 6 exact symmetries of AdS3 . Since
it cannot have more than 6 global isometries, we have proven that AdS3 admits SOp2, 2q as Killing
isometry group. The generators of sop2, 2q are given by :

B B
Jab “ Xb ´ Xa b (2.14)
BX a BX

where Xa “ ζ ab X b . After expanding in global coordinates, one can see that J01 “ Bτ generates time
evolution on H, whereas J12 “ Bφ generates rotations. The most general Killing vector is naturally
given by 12 ω ab Jab , and is thus determined by an antisymmetric ω ab “ ω rabs tensor in 4d.

2.2.2 Asymptotically AdS 3 black holes

We now set aside the global homogeneous case to consider more evolved geometries, but which
asymptote nevertheless to AdS3 when approaching spatial infinity in a sense that we will make pre-
cise below. Such spacetimes are called asymptotically AdS3 . The space of such spacetimes will be quite
rich, in particular because of the presence of black hole solutions. There is no contradiction with our
earlier derivation that all solutions to Einstein’s equations are locally AdS3 . A 3d black hole does not
possess any curvature singularity. It is a black hole because it admits an event horizon, which turns
out to protect a “causal singularity”. We will first present the solution and discuss its properties, and
afterwards we will see how to obtain such a solution starting from AdS3 itself !
1 Note that the application defined by the system above, and which sends AdS3 into Rp2,2q , is not injective, because it
contains 2π-periodic functions of τ. So it wraps AdS3 around H an infinity of times. But locally the injectivity is however
ensured, so we talk about immersion rather than embedding.

Lecture 2. Three dimensional Einstein’s gravity 38

 2.2. Asymptotically anti-de Sitter phase space

BTZ black holes and more. . .

Let us consider a set of Boyer-Lindquist-like coordinates pt, r, φq, where t is an asymptotically timelike
coordinate, r is an asymptotically radial coordinate, and φ is a polar angle identified as φ „ φ ` 2π.
Imposing that the spacetime behaves like AdS3 when r Ñ 8, and that the spacetime is stationary
and axisymmetric, a natural ansatz to consider is

dr2 ˘2
ds2 “ ´N 2 prqdt2 ` 2
` φ
` r dφ ` N prqdt . (2.15)
N 2 prq

This solution for Nprq and N φ prq was found for first time in 1992 by Bañados, Teitelboim, and Zanelli,
and describes in some range of the parameters, as we are about to show, the so-called BTZ black hole.
The boundary conditions are fixed such as

r2 16G2 J 2
N 2 prq “ ´8MG ` ` pLapse functionq; (2.16)
`2 r2
4GJ
N φ prq “ ´ pAngular draggingq. (2.17)
r2

We can explicitly verify that the BTZ solution has 2 Killing vectors Bt and Bφ since the metric coeffi-
cients depend only on r. But the solution is not static because the term dtdφ is not invariant under
time reversal. The quantities M and J which naturally appears in the metric are respectively the
surface charges associated to Bt and Bφ evaluated on the circle at infinity S “ tx µ P M | t “ Cst, r “
Cst Ñ 8u (or actually any circle homotopic to it) :
¿
kBt rδg; gs “ δM Ñ M “ total mass of the black hole; (2.18)

¿S
k´Bφ rδg; gs “ δJ Ñ J “ total angular momentum of the black hole. (2.19)
S

The minus sign in the definition of J is conventional. We have a 2-parameter family of solutions, with
the scale length ` being fixed. In fact, it is a black hole for special values of the parameters, but it is
also more than that as we will now show.

Main properties

Global AdS3 spacetime is recovered if N φ “ 0 ô J “ 0 and N “ 0 ô M “ ´1{8G. In that case, r “ 0

is the origin of polar coordinates and there is no singularity there.

More generally, we can check that the solution is asymptotically AdS3 simply by taking the limit
r Ñ 8 in (2.15). One can compute the curvature scalar R and show that it is equal to ´6{`2 ev-
erywhere. So again, the solution does not contain any curvature singularity. Starting from infin-
ity and going inwards, the first particular surface we encounter is the limit of staticity gtt prerg q “
´N 2 prerg q ` r2 rN φ prerg qs2 “ 0 below which Bt has positive norm, and so a static observer with a 4-
velocity colinear to Bt cannot exist since t is no more a timelike coordinate. The critical "radius" is
?
given by rerg “ ` 8GM, and this surface, called ergocircle in analogy to the Kerr metric, exists if and
only if M ą 0. Another set of critical values of r are the roots of the lapse function N 2 prq “ 0. The

Lecture 2. Three dimensional Einstein’s gravity 39

 2.2. Asymptotically anti-de Sitter phase space

latter equation is quadratic equation in r2 , so we will find two roots in terms of r2 . We are interested
in positive r so we choose the positive roots, which are
g d
f
J 2
ˆ ˙
? f
r˘ “ ` 4GM 1 ˘ 1 ´ . (2.20)
e
M`

Let us denote by H˘ the surfaces tx µ | r “ r˘ u. They exist if and only if

|J| ď M` ; M ą 0. (2.21)

This limits the spectrum of allowed black holes. They are said extremal if |J| “ M ` or equivalently
r` “ r´ . Another special place is r “ 0 beyond which Bφ becomes timelike: this is the causal singu-
larity. Let us now demonstrate that H` is a rightful (outer) event horizon. Taking advantage of our
knowledge about the Kerr black hole, we claim that the surface r “ r` is in fact a Killing horizon
ruled by integral curves of the helicoidal vector ξ “ Bt ` Ω H Bφ for a certain Ω H . To find Ω H , we use
the trick to solve gµν ξ ν “ 0 at r “ r` which gives Ω H “ ´gtt {gtφ |r“r` “ ´gtφ {gφφ |r“r` “ ´N φ pr` q “
2 “ r {r `. This is the angular velocity of the horizon. By construction, ξ µ ξ “ 0 on H . As an
4GJ{r` ´ ` µ `
exercise, you can show that ξ µ actually generates the horizon ξ µ Dµ ξ ν “ κξ ν where the surface gravity
2 ´ r 2 q{`2 r . These two expressions make sense in the range (2.21). So we have
is given by κ “ pr` ´ `
proven that H` was the rightful outer event horizon, and from now we can use the name "black hole"
without abuse. The event horizon is found to be a Killing horizon, so the rigidity theorem is obeyed,
since BTZ spacetime is stationary and axisymmetric. Moreover, note also that r´ ď r` ď rerg . So
there is a non-trivial region beyond the ergosphere and still outside the horizon where the observers
experience some frame dragging due to the rotation of the black hole. This ergoregion is a supplemen-
tary feature that BTZ black hole shares with the Kerr solution in 4d.

Now let us look at the thermodynamical properties. Hawking’s temperature reads simply as

κ r2 ´ r2
TH “ “ ` 2 ´. (2.22)
2π 2π ` r`

Note that TH “ 0 for extremal cases where also Ω H “ 1{`. Now we can define the entropy of the BTZ
black hole :
ż 2π
1 1 ?
SH “ pPerimeter of the horizonq “ dφ gind (2.23)
4G 4G 0

where gind is the determinant of the induced metric on the horizon at constant value of time, so
ds2ind “ ds2 rr “ r` , t “ Csts “ r`
2 dφ2 , so

ż 2π
1 ? πr`
SH “ dφ gind “ . (2.24)
4G 0 2G

We can check the first law TH δS H “ δM ´ Ω H δJ, which is quite obvious here :

2 ´ r2
r` 2 ´ r2
r`
r` r´ ´ π
8GM “ 2
´
, 4GJ “ ùñ δM ´ Ω H δJ “ 2
δr` “ TH δS H . (2.25)
` ` 2π ` r` 2G

Lecture 2. Three dimensional Einstein’s gravity 40

 2.2. Asymptotically anti-de Sitter phase space

Penrose diagrams

To analyze the causal features of the BTZ spacetime, there is nothing like a Penrose diagram ! The
procedure is sensitively the same as for the Kerr black hole. The idea is to introduce a set of Kruskal
coordinates pU, V, r, φq in the vicinity of each root of the lapse function N 2 prq “ 0. On each Kruskal
patch, we wish to write the line element as
˘2
ds2 “ Ω2 pU, VqpdU 2 ´ dV 2 q ` r2 N φ pU, Vqdt ` dφ .
`
(2.26)

where t “ tpU, Vq. If J “ 0, r´ “ 0 and there is only one non-trivial root r` : in this case the Kruskal
patch around r` actually covers all the spacetime. Let us start with r` : we leave as an exercise (see
the original [29]) to show that Kruskal coordinates around r` are defined by the patch K` :
$ c´ ¯´ ¯r {r
’ ´r`r` r`r´ ´ `
& U` “
’ sinhrκtpU` , V` qs;
If r P sr´ , r` s : c´ r`r` ¯ ´ r´r´ ¯ (2.27)
r {r
’
% V` “ ´r`r` r`r´ ´ `
’
r`r` r´r´ coshrκtpU` , V` qs;

$ c´ ¯´ ¯r´ {r`
’ r´r` r`r´
& U` “
’
r`r` r´r´ coshrκtpU` , V` qs;
If r P rr` , 8r: c´ ¯´ ¯r´ {r` (2.28)
’
% V` ´r`r` r`r´
’ “ r`r` r´r´ sinhrκtpU` , V` qs;

2 ´ r 2 q{`2 r . Within the patch K , the angular coordinate φ is chosen such

and we recall that κ “ pr` ´ ` ` `
as N φ pr` q “ 0 to ensure that the metric element N φ dt remains regular at r` . Up to this change of
coordinates, the BTZ metric takes the right form in K` with
˙r´ {r`
pr2 ´ r´
2 qpr ` r q2 ˆ
` r ´ r´
Ω2` prq “ for r P sr´ , 8r . (2.29)
κ 2 r 2 `2 r ` r´

We can define a similar Kruskal patch K´ around r´ . The expressions are quite similar, up to some
permutations p` Ñ ´q. The two patches have a non-trivial overlap K “ K´ X K` . Just as in the
3 ` 1 Kerr metric, one may maximally extend the geometry by gluing together an infinite number of
copies of patches K` , K´ through their overlap K. Since we have at our disposal a set of Kruskal-like
coordinates, the Penrose compactification is quite straightforward. We introduce another change of
coordinates pU, Vq Ñ pp, qq such as
ˆ ˙ ˆ ˙
p`q p´q
U ` V “ tan ; U ´ V “ tan . (2.30)
2 2

This transformation is a bijection for the usual determination of the arctangent function, which lies
between ´π{2 and π{2. Let us start with the non-rotating case J “ 0. Only the outer horizon at r`
exists, since r´ “ 0 coincides with the causal singularity. In this case, spacelike infinity r “ 8 is
mapped to the (vertical) lines p “ ˘π{2, the singularity r “ 0 is mapped to the (horizontal) lines
q “ ˘π{2, and finally the horizon lies at p “ ˘q. This gives the following Penrose diagram:

Lecture 2. Three dimensional Einstein’s gravity 41

 2.2. Asymptotically anti-de Sitter phase space

r“0

r“

r`
r“
r`
r“8

r“8
q

r“0
p

Figure 2.2: Penrose diagram for a static BTZ black hole (J “ 0).

In the rotating case, we have to perform the change of coordinates pU, Vq Ñ pp, qq in each Kruskal
patch, so we find the following Penrose diagram:

r“

r´
r“
r“

r“

r´
r`

r´
r“

r“

II II
r`

r´

r“0

r“0
III III

K
II
r“

r“
r´

r`
r“

r“

r“

r`
r´

r`

r“
r`
r“8

r“8

r“8

r“8
r“0

r“0

III III I I I I

II
r“

r´
r“

K
r´
r“0

r“0

III III
r“

r“

II II
r`

r´
r“

r“
r`

r´

(a) Penrose diagram for K´ (b) Penrose diagram for K` (c) Complete diagram

Figure 2.3: Penrose diagrams for non-extremal BTZ black holes

As we have already seen, the K parts are identified because they represent the overlap between the
patches K´ and K` . In the K´ patch, the original black hole coordinates covered K and one region
labeled by III, while in the K` patch, they covered K and one region labeled by I. To obtain a“maximal
causal extension" (i.e. where all causal curves extend maximally), we must include the others regions
in each diagram, and then glue together an infinite sequence of them. The resulting Penrose diagram
bears strong ressemblence with the Kerr one but the asymptotics and singularities differ.

Lecture 2. Three dimensional Einstein’s gravity 42

 2.2. Asymptotically anti-de Sitter phase space

Let us now comment on the extremal case. For the non-massive case M “ 0 (which is also non-
rotating), the metric reduces to ds2 “ ´pr{`q2 dt2 ` pr{`q´2 dr2 ` r2 dφ2 , and we can directly define null
dimensionless coordinates U “ pt{`q ´ p`{rq, V “ ´pt{`q ´ p`{rq such that ds2 “ r2 dUdV ` r2 dφ2 .
This allows to directly write the compact Penrose coordinates (2.30), in which the metric reads as

`2 cos p ` cos q
ds2 “ pdp2 ´ dq2 q ` r2 dφ2 ; r “ ´` . (2.31)
sin2 p sin p

So the origin r “ 0 is mapped to the segment of the line p “ π ˘ q running from p “ 0 to p “ π.

On the other hand, spacelike infinity is mapped to the segment of the line p “ π, and the Penrose
diagram is simply a closed triangle. Now we finish with the extremal rotating case |J| “ M`. In this
case only the outer horizon survives, and the metric can be written as a function of r` instead of M.
pr2 ´r` 2 q2
The lapse function becomes simply N 2 prq “ r 2 `2
. The appropriate null coordinates are U “ t ` r‹
and V “ ´t ` r‹ where r‹ is the so-called tortoise coordinate

r `2 `2
ż ˇ ˇ
‹ dr ˇ r ´ r` ˇ
r “ “´ 2 2q
` ln ˇ ˇ. (2.32)
N 2 prq 2pr ´ r` 4r` ˇ r ` r` ˇ

Defining the Penrose coordinate again as before, we get

4N 2 prq`2 pdp2 ´ dq2 q sin p r‹

ds2 “ ` r2 pN φ dt ` dφq2 where r is solution of “ . (2.33)
pcos p ` cos qq2 cos p ` cos q `

We see that the horizon r “ r` is represented by lines at ˘45, whereas r “ 0 lies at p Ñ kπ (the limit
is taken from above values) and r “ 8 at p Ñ kπ (the limit is this time taken from below values). The
region beyond the horizon (0 ă r ă r` ) is mapped onto a triangle bounded by the lines p “ 0 (which
is r “ 0) and p “ q “ π, p ´ q “ π which is quite similar to the non-rotating case. The outer region
r ą r` is obtained thanks to another determination of the arctangent function, and is a symmetric
triangle too. To obtain the full spacetime, we have to glue the two triangles along the common edge
r “ r` at 45˝ . Once this is done, we can go safely accross the gluing edge because the root of Nprq is
compensated by the cancellation of the denominator in the pp, qq term in the metric. As before, the
maximal extension is build by including an infinite sequence of triangles (see Figure 2.4).

We conclude here the description of the BTZ geometry. What remains to be done is see how to reach
the black hole solution from the global AdS3 spacetime. Since there is no topological defect for M ą 0,
the process at work is in fact a fold of AdS3 on itself after that some points have been identified. We
discuss this in more details in the next paragraph.

Lecture 2. Three dimensional Einstein’s gravity 43

 2.2. Asymptotically anti-de Sitter phase space

r`
r“

r“
BH r`
0 ge
om
r“ etr
y
r`

r“8

r“8
r“0
r“

r“ r“
0 r`

r`
r“

(a) Penrose diagram for M “ J “ 0 (b) Penrose diagram for |J| “ M `

Figure 2.4: Penrose diagrams for extremal BTZ black holes.

Identifications

Any Killing vector ξ µ generates a 1-parameter sub-

group of isometries of AdS3 . It acts on points as P Ñ Asymptotic box
esξ P where s is a continuous parameter, and locally as
P “ px µ q Ñ P1 “ px µ ` ξ µ q. If we restrict s to discrete
values k ∆s where k P Z and ∆s is a basic step con- Sp0q Spsq

ventionally fixed as 2π, we are left with the so-called

identification subgroup. The identified quotient space is P esξ P
obtained by identifying points that belong to a given ξ
orbit of the identification subgroup. Geometrically, we
can view this operation as follows. Let us consider a
surface S0 in AdS3 . Each point of this surface belongs
Identification
to an orbit of ξ µ , and we can apply an element of the
identification subgroup to get another surface Sp2πkq Figure 2.5: Identifications in AdS3 .
whose points have coordinates e2πkξ P, P P S0 , k P Z.
The quotient process consists in gluing the surface S0 “ Sp0q with Sp2πq, Sp´2πq, Sp4πq, and so on !

Since ξ µ is a Killing vector, the gluing leads to a continuous spacetime and the quotient space inherits
from anti-de Sitter space a well defined metric which has constant negative curvature. Indeed, let us
choose a coordinate system pt, r, ϕq in which ξ “ B ϕ and ϕ P R. So Lξ gµν “ ξ ρ Bρ gµν ` 0 “ B ϕ gµν “
0 ñ gµν pt, r, ϕq “ gµν pt, rq. When we perform identifications along ϕ, gµν is not a multivalued func-

Lecture 2. Three dimensional Einstein’s gravity 44

 2.2. Asymptotically anti-de Sitter phase space

tion of the ϕ coordinate. So the metric remains locally smooth and is still a solution of Einstein’s
equations. The identification process makes the curves joining two points that are on the same orbit
to be closed in the quotient space. In order to preserve causality in the quotient space or, equivalently,
prevent the appearance of closed causal curves, a necessary condition (but not sufficient in general)
is that the Killing vector ξ µ must be spacelike : ξ µ ξ µ ą 0. Let us see what happens particularly for
the BTZ black hole.

One can show that the non-extremal BTZ black hole solutions are obtained by making identifications
in AdS3 by the discrete group generated by the Killing vector

r` r´
ξ“ J12 ´ J03 (2.34)
` `
where the Jab are the Killing vectors of AdS3 , belonging to the sop2, 2q algebra [29]. To see geomet-
rically in what consist these identifications, the most “simple” thing to do is to find a coordinate
system pt, r, φq in which ξ “ Bφ . The answer is simply that AdS3 takes the form (2.15) but with φ P R.
Recognizing that Bφ is given by (2.34) takes more effort, which is narrated in [29]. It follows that the
non-extremal BTZ black hole is obtained by realizing periodic identifications along (2.34) in AdS3 !

But all is not resolved, since ξ µ is not spacelike everywhere in AdS3 . In fact, before performing
identifications, we better remove the regions where ξ 2 ď 0 in global AdS3 in order to avoid closed
timelike curves. We are left with a spacetime denoted by AdS3 which is geodesically incomplete,
since before the removal, some geodesics traveled from regions ξ 2 ą 0 to regions ξ 2 ď 0. The critical
surface ξ 2 “ 0 appears as a singularity in the causal structure of spacetime, since continuing beyond
it would produce closed timelike curves. We retrieve a familiar feature of 3 ` 1 black holes : the only
incomplete geodesics are those that hit the singularity. In 4d it is a curvature singularity but here it
is a causal singularity. Remember that in 2 ` 1 dimensions the Weyl tensor is identically zero while
the Ricci tensor is determined by the cosmological constant by the vacuum Einstein’s equations, so
curvature singularities cannot appear. It turns out however that a horizon prevents the asymptotic
observer to detect the causal singularities. This is a form of “cosmic censorship”.

Going back to AdS3 without identifications, we can recognize its boundary ξ 2 “ 0 as a surface.
Explicitly, we have
2
r` 2
r´
ξ 2 “ ξ a ξ a “ ζ ab ξ a ξ b “ pX 2
´ X 2
q ` pX̃ 2 ´ X22 q. (2.35)
`2 0 1
`2 0
Therefore, AdS3 is bounded by the planes X0 “ ˘X1 and X̃0 “ ˘X2 intersecting the conformal cylin-
der. The locus ξ 2 “ 0 is therefore a connected, diamond-shaped region depicted below.

Symmetries of the quotient space

The BTZ solution is stationary and axisymmetric: it has 2 commuting Killing vectors Bt and Bφ . One
may ask whether there are any other independent Killing vectors, since it comes from an identifi-
cation of a maximally symmetric spacetime that has 4 more isometries. We will show that the two
aforementioned Killing vectors are the only two isometries of the BTZ solution.

Lecture 2. Three dimensional Einstein’s gravity 45

 2.2. Asymptotically anti-de Sitter phase space

Figure 2.6: A finite-time section of the conformal representation of AdS3 .

The red sections are the null surfaces on which ξ 2 “ 0.

Before any identification, the spacetime has 6 independent Killing vectors which are the generators of
sop2, 2q. After identification, some of these vectors become multivalued, and so are no longer proper
symmetries of the quotient space. A necessary and sufficient condition for a vector η µ P sop2, 2q to in-
duce a well-defined vector field on the quotient space is that η µ be invariant under the identification
subgroup pe2πξ q‹ η “ η. Here we consider η µ and ξ µ as sop2, 2q matrices. Since η is a Killing vector,
this condition is equivalent to pe2πξ qηpe2πξ q´1 “ η or re2πξ , ηs “ 0. A theorem due to Chevalley and
Jordan states that any matrix η can be fragmented in two commuting parts s and n : ξ “ s ` n. s is
semi-simple with real eigenvalues, and n is nilpotent. Any matrix commuting with e2πξ must also
commute separately with e2πs and e2πn , so we get rs, ηs “ rn, ηs “ 0. This implies rξ, ηs “ 0.

So the problem of finding all Killing vectors of the BTZ solution is equivalent to finding all the sop2, 2q
matrices that commute with ξ. Similarly to sop4q “ sop3q ‘ sop3q, we know that sop2, 2q “ sop2, 1q ‘
sop2, 1q (sop2, 1q is the Lorentz algebra in 2 ` 1 dimensions) and we can decompose ξ “ ξ ` ` ξ ´ , η “
η ` ` η ´ , where the "`" parts are the self-dual parts, and the "´" the anti-self-dual ones. Since the
sum between the algebras is direct, we are left with rξ ` , η ` s “ rξ ´ , η ´ s “ 0. Now, recall that the only
elements of sop2, 1q that commute with a given non-zero element of sop2, 1q are only the multiples of
that element. After decomposing (2.34) as ξ ` ` ξ ´ , one finds that both ξ ` and ξ ´ are non-zero. We
conclude that ξ ˘ 9 η ˘ and there is no additional Killing vector. The identification kills 4 out of the 6
Killing isometries of AdS3 and only Bt and Bφ survive.

Lecture 2. Three dimensional Einstein’s gravity 46

 2.2. Asymptotically anti-de Sitter phase space

BTZ phase space

We conclude this quite long section by a summary of the different spacetimes that are described by
the line element (2.15). At fixed `, it is only described by two parameters, which are the total mass M
and the total angular momentum J, so we can represent the phase space in a pJ, Mq-plane.

B For M ą 0 and |J| ď M`, we find all BTZ black holes whose main features were explained.
They are identifications of AdS3 that possess a Killing horizon. The latter is a rightful event
horizon that shields a causal (but not a curvature) singularity. These solutions are delimited
by the extremal ones, which obey |J| “ M `. The special BTZ black hole with zero mass has
M “ J “ 0.

B For M ą 0 and overspin |J| ą M`, the spacetime leaves exposed the chronological singularity
(closed timelike curve or C.T.C.) at r “ 0. This is a naked singularity, which is usually consid-
ered unphysical.

B Solutions with M ă 0 and J “ 0 represent particles in AdS3 sitting at r “ 0. If M ą ´1{8G, they

produce a conical defect around it2 . When M “ ´1{8G, the conical singularity disappears and
the spacetime, smooth everywhere, is nothing but global AdS3 . When M ă ´1{8G, the angular
defect becomes an excess. The energy spectrum of these solutions is not bounded from below
and has a mass lower than the natural ground state, global AdS3 . These solutions are therefore
also usually considered unphysical and discarded. If J ‰ 0, the spacetime is identified with a
twist in time around the conical defect/excess: we thus find spinning particles with mass M and
angular momentum J. Again when |J| ą ´M` the spacetime contains naked closed timelike
curves...

M` “ |J| M ` “ |J|

BTZ black holes

Naked sing. Naked sing.

β
J
1
´ 8G Defects C.T.C.
AdS3

Excesses

Figure 2.7: Solutions described by the BTZ metric.

The slope of extremal lines is tan β “ `´1 .
Figure adapted with permission from [33].Copyrighted by the American Physical Society.

2 We save more detailed explanations for the lecture on 3d asymptotically flat spacetimes where conical defects also

appear. Doing so we do not develop the same explanations twice !

Lecture 2. Three dimensional Einstein’s gravity 47

 2.2. Asymptotically anti-de Sitter phase space

2.2.3 Asymptotically AdS 3 spacetimes

In the previous section, we have studied the phase space of stationary and axisymmetric solutions of
Einstein’s 3d gravity that also asymptote to AdS3 spacetime. We have found many kinds of solutions,
including BTZ black holes and spinning particles... Is it possible to formalize a phase space that
includes all these solutions, and contains even more solutions?

Boundary conditions

Let us denote by P this phase space. Any point of P is a solution of 3d Einstein’s equations that
is asymptotically AdS3 at spatial infinity. To define precisely the assumptions under which a given
metric gµν belongs or not to P , we have to construct a set of boundary conditions that tell us how gµν
behaves when a suitable spacelike coordinate runs to infinity. The problem of fixing the asymptotic
behaviour of fields in gravity is not straightforward because the choice of appropriate fall-offs at
infinity is not unique at all ! We will consider the boundary conditions that obey the following
criteria:

B P has to contain the solutions we have already analyzed : BTZ black holes, spinning particles,...
and of course AdS3 itself. So the boundary conditions must not to be too restrictive ;

B P needs to lead to finite and integrable surfaces charges. It restricts the possibilities for P . For
example, a geometry gµν that has infinite energy is not allowed, and must be avoided by the
boundary conditions;

B The asymptotic symmetry group must at least contain SOp2, 2q, the group of exact symmetries
of global AdS3 .

These three requirements lead to the boundary conditions of Brown and Henneaux [23]. However,
keep in mind that several other choices of boundary conditions exist if one for example changes the
third requirement (see [34] and further developments for alternatives).

Remember that the asymptotic symmetry group is the quotient group between “large" diffeomor-
phisms, associated to non-vanishing canonical charges, and the “gauge" diffeomorphisms, which act
trivially on the phase space and are not associated with any charge. In their original work, Brown and
Henneaux started with fall-off conditions for all components of the metric. We find more pedagogical
and illuminating to first reduce the coordinate system by removing all gauge diffeomorphisms. It
indeed allows to isolate in detail the structure of the asymptotic symmetry group which then coin-
cides with all remaining (non-trivial) diffeomorphisms that obey the boundary conditions. Gauge
diffeomorphisms are typically removed by fixing a coordinate system such that no diffeomorphism
depending arbitrarily on all coordinates is still allowed. Such a coordinate system exists for asymp-
totically AdS3 spacetimes, as stated in the Fefferman-Graham theorem:

Lecture 2. Three dimensional Einstein’s gravity 48

 2.2. Asymptotically anti-de Sitter phase space

Result 8 (Fefferman-Graham coordinates)

Any asymptotically AdS3 spacetime can be written in the neighborhood of the boundary
as

ds2 “ `2 dρ2 ` e2ρ gp0qab dx a dx b ` O peρ q (2.36)

dr2 r2
“ `2 ` 2 gp0qab dx a dx b ` O prq pr “ `eρ q. (2.37)
r2 `
The coordinates pρ, x a q are called the Feffermann-Graham coordinates : ρ is a spacelike co-
ordinate such as ρ Ñ `8 represents the boundary, and px a q are the coordinates defined
on this boundary, x1 “ t being a timelike coordinate and x2 “ φ is a spacelike coordinate
with closed periodic orbits (an angle). The geometry at the boundary is dictated by the 2d
metric gp0qab , whose associated covariant derivative will be denoted by D a .

In this framework, the Brown-Henneaux boundary conditions are simply Dirichlet boundary condi-
tions for the boundary metric:
gp0qab “ ηab is fixed (2.38)

where ηab dx a dx b “ ´dt2 ` `2 dφ2 . Imposing Fefferman-Graham gauge, the 3 diffeomorphisms de-
pending on arbitrary functions of the 3 coordinates are reduced to the residual diffeomorphisms
which are 3 diffeomorphisms depending on arbitrary functions of 2 coordinates. In that sense, the
trivial bulk diffeomorphisms are removed, and only the more interesting residual diffeomorphisms
remain. The boundary conditions then reduce this set of residual diffeomorphisms to the asymptotic
symmetries. In summary, the asymptotic symmetry group can be computed simply by inspecting
the set of non-trivial diffeomorphisms that preserve both the Fefferman-Graham asymptotic expan-
sion and the Brown-Henneaux boundary conditions. Let us consider an arbitrary diffeomorphism
ξ “ ξpρ, x a q. The preservation of (2.36) leads us to

Lξ gρρ “ 0 ñ 2gρµ Bρ ξ µ “ 2gρρ Bρ ξ ρ “ 2Bρ ξ ρ “ 0 ô ξ ρ pρ, x a q “ Rpx a q; (2.39)

Lξ gρa “ 0 ñ gµa Bρ ξ µ ` gµρ Ba ξ µ “ gab Bρ ξ b ` Ba R “ 0. (2.40)

If we denote by g ab the inverse of gab , we get

ż
b ab b a b a
Bρ ξ “ ´g Ba R ñ ξ pρ, x q “ V px q ´ dρ g ab Ba R. (2.41)

So we are left with the most general diffeomorphism that preserve the Fefferman-Graham gauge,
which depends upon 3 functions of 2 coordinates as announced. Now we impose also the boundary
conditions :

Lξ gab “ O peρ q ñ ξ ρ Bρ gab ` ξ c Bc gab ` gac Bb ξ c ` gbc Ba ξ c “ O peρ q (2.42)

ô 2e2ρ gp0qab R ` e2ρ V c Bc gp0qab ` gp0qac Bb V c ` gp0qbc Ba V c “ O peρ q
` ˘
(2.43)
ô 2gp0qab R ` LV gp0qab “ 0. (2.44)

Lecture 2. Three dimensional Einstein’s gravity 49

 2.2. Asymptotically anti-de Sitter phase space

Taking the trace of this latter equation, and recalling that the boundary metric is flat, we get 4Rpx a q “
ab pD V ` D V q so
gp0q a b b a
1
Rpx a q “ ´ Dc V c ùñ D a Vb ` Db Va “ Dc V c ηab . (2.45)
2
In conclusion, the asymptotic vectors are determined at leading order by a boundary vector field V a
which is in fact a conformal Killing vector on the boundary. To solve the conformal Killing equation,
let us introduce lightcone coordinates on the boundary x˘ “ pt{`q ˘ φ, such as ds2p0q “ gp0qab dx a dx b “
´dt2 ` `2 dφ2 “ ´`2 dx` dx´ . From (2.45), we get

´` 2
2B` V` “ 0 ñ V` “ V` px´ q ô V ´ “ gp0q V` “ ´ V` ñ V ´ “ V ´ px´ q (2.46)
`2
and symmetrically V ` “ V ` px` q. The conformal Killing vectors on the boundary are naturally di-
vided into left-moving and right-moving fields on the cylinder, and we obtain thus 2 infinite families
of generators:
ż
p`q 1
ξ ` `
“ V px qB` ´ B` V Bρ ` dρ g`´ B` B` V ` B´ ;
`
(2.47)
2
ż
1
ξ p´q “ V px qB´ ´ B´ V Bρ ` dρ g`´ B` B´ V ´ B´ .
´ ´ ´
(2.48)
2

The first term contains the independent function that defines the generator. The second term is com-
pletely fixed in terms of the boundary conformal vectors in order to preserve the Feffermann-Graham
p0q
gauge and it is also leading. The third term is subleading, since g ab “ e´2ρ gab ` . . . . Therefore, both
the first and second term will bring the leading contribution to the conserved charges.

Asymptotic symmetry algebra

Let us look at this asymptotic symmetry algebra of vector fields ! One word of caution however: the
asymptotic symmetry algebra requires finite conserved charges, which we will have to check next!
p`q ` p´q
First, we develop each vector field into Fourier modes ξ m “ ξ p`q pV ` “ eimx q, ξ p´q “ ξ n pV ´ “
e inx´ q and compute their usual Lie bracket. First, ξ p`q and ξ p´q always commute since they depend at
leading order on opposite boundary lightcone coordinates, and the subleading orders trivially follow
by the Fefferman-Graham gauge,

p´q µ p`q α p´q µ p´q α p`q µ

” ı ´ ¯ ´ ¯ ´ ¯ ´ ¯
p`q
ξm , ξn “ ξm Bα ξ n ´ ξn Bα ξ m “ 0. (2.49)

The asymptotic algebra is thus a direct sum of two chiral subalgebras. Moreover

p`q α p`q α
” ı ´ ¯ ´ ¯ ´ ¯ ´ ¯
p`q p`q p`q p`q
ξm , ξn “ ξm Bα ξ n ´ ξn Bα ξ m (2.50)
`
´ `¯ `
´ `
¯
“ eimx B` einx ´ einx B` eimx ` op1q (2.51)
” ı
p`q p`q p`q p`q
“ ipn ´ mqξ m`n ñ i ξ m , ξ n “ pm ´ nqξ m`n . (2.52)

The same calculation can be performed for the vectors ξ p´q . We have just proven that each chiral
subalgebra is isomorphic to the Witt algebra, which is the centerless algebra of circle diffeomorphisms.
We can now verify that the asymptotic algebra contains the generators of AdS3 exact symmetries. Let

Lecture 2. Three dimensional Einstein’s gravity 50

 2.2. Asymptotically anti-de Sitter phase space

p`q p`q p`q

us consider simply the subset tξ ´1 , ξ 0 , ξ 1 u. They form a closed subalgebra under the Lie bracket
” ı ” ı ” ı
p`q p`q p`q p`q p`q p`q p`q p`q p`q
i ξ1 , ξ0 “ ξ1 ; i ξ 1 , ξ ´1 “ 2ξ 0 ; i ξ 0 , ξ ´1 “ ξ ´1 (2.53)

p´q p´q p´q

which we recognize as the slp2, Rq algebra. But since it also holds for tξ ´1 , ξ 0 , ξ 1 u and we know
that slp2, Rq » sop2, 1q, the asymptotic algebra contains a set of 6 generators which form a subalgebra
isomorphic to slp2, Rq ‘ slp2, Rq » sop2, 1q ‘ sop2, 1q » sop2, 2q, so the asymptotic symmetry group is
the natural extension of the exact symmetry group of the asymptotic AdS3 space !

Now, we would like to compute the charges. While the boundary conditions are sufficient to compute
the charges, it is useful to first make a detour to the solution space, which will allow us to make the
integral on fields in (1.19) totally explicit!

Phase space

Now that the asymptotic symmetries have been characterized, let us go back to the description of P
itself. Recall that P is the set of Einstein’s solutions which can be written as (2.36) in the neighborhood
of the boundary and obey Brown-Henneaux boundary conditions. In Feffermann-Graham gauge, the
remaining metric coefficients are given as an asymptotic expansion in terms of r “ `eρ with r Ñ 8.
Solving order by order Einstein’s equations, one finds that the expansion miraculously stops at second
order:
dr2 r2 `2 `4
ˆ ˙
2
ds “ `2 2 ` 2 gp0qab ` 2 gp2qab ` 4 gp4qab dx a dx b . (2.54)
r ` r r
Moreover, the second order gp4qab is completely fixed by the leading orders

1
gp4qab “ g gcd g (2.55)
4 p2qac p0q p2qdb

and finally the trace of gp2qab and its covariant divergence D a gp2qab are also fixed by the equations of
motion and the boundary conditions to be zero ! So the form of the metric is nearly totally fixed! In
fact the only metric components that are left over to vary independently are the traceless part and
divergence-free part of gp2qab , which is naturally interpreted as a boundary stress tensor. To make
these remaining boundary degrees of freedom explicit, we use again lightcone coordinates at the
boundary. The vanishing trace condition reads as gp2q`´ “ 0 and the conservation condition gives

D a gp2qab “ 0 ñ B ` gp2q`` “ 0 ô B´ gp2q`` “ 0 ô gp2q`` px a q ” `2 L` px` q (2.56)

and immediately gp2q´´ px a q ” `2 L´ px´ q (the `2 factor is conventional). Therefore, each metric in P
can be written into the form

dr2
ˆ ´q
˙ˆ `q
˙
2 2 L´ px 2 L` px
ds “ `2 2 ´ rdx ´ ``
dx ´ ´
rdx ´ ` dx `
. (2.57)
r r r

This is the most general (analytic) Einstein solution which obeys the Brown-Henneaux boundary
conditions. One may check that all the solutions we have discussed in the previons section can be
brought to this form. For example, the BTZ black hole (2.15) can be written as (2.57) with constant

Lecture 2. Three dimensional Einstein’s gravity 51

 2.2. Asymptotically anti-de Sitter phase space

L` and L´ such as M “ pL` ` L´ q{p4Gq and J “ `pL` ´ L´ q{p4Gq. In this patch the outer horizon
lies at r “ 0 !

Let us finally complete our discussion by computing the charges associated to this phase space, and
check the representation theorem. The set of metrics is now defined by (2.57) and an arbitrary vari-
Bgµν Bgµν
ation of the metric can be written as hµν ” δgµν “ BL` δL` ` BL´ δL´ . We can therefore compute the
infinitesimal surface charge using either formalism (1.112) or (1.136). In fact, the annoying supple-
mentary term E (1.114) identically vanishes in Fefferman-Graham coordinates so the surface charge
is uniquely defined! As we have done for the vectors themselves, we can develop the charges into
Fourier modes and we obtain
¿ ż 2π
p`q ` `
δLm “ kξ p`q rδg; gs “ dφ δL` px` qeimx ; (2.58)
m 8πG 0
S1
¿ ż 2π
p´q ` ´
δLn “ kξ p´q rδg; gs “ dφ δL´ px´ qeinx . (2.59)
n 8πG 0
S1

The integration is performed on any circle, either at infinity or at finite r. The charges are clearly
integrable,
ż 2π ż 2π
p`q ` imx` p´q ` ´
Lm “ dφ L` px qe `
, Lm “ dφ L´ px´ qeimx (2.60)
8πG 0 8πG 0

so the representation theorem holds! When we integrated the charges, we chose to define all charges
of the zero mass BTZ black hole (with L` “ L´ “ 0) as zero. Let us cross-check that these charges
form an algebra under the Poisson bracket
¿ ż 2π
!
p`q p`q
)
p`q
” ı ` `
Lm , Ln “ δξ p`q Lm “ kξ p`q δξ p`q g, g “ dφ δξ p`q L` px` qeimx . (2.61)
n m n 8πG 0 n
1
S8

The variation of the first component of the stress tensor L` can be deduced as follows. We can
compute the Lie derivative of gab on the flow of ξ p`q . Since this vector is a generator of the asymptotic
group, it must preserve the expansion (2.57) then at the linear level

Lξ p`q gab “ gab rL` ` δξ p`q L` , L´ ` δξ p`q L´ s ´ gab rL` , L´ s. (2.62)

Simply by inspecting this relation component by component, one finds that

1 3
δξ p`q L` “ V ` B` L` ` 2L` B` V` ´ B` V` ; (2.63)
2
δξ p`q L´ “ 0. (2.64)

The first relation implies that L` transforms as an element

! of the )coadjoint representation of the
p`q p´q
Witt algebra, while the second one indicates to us that Lm , Ln “ 0 @m, n P Z, so the chiral
fragmentation also holds at the level of the charge algebra. Let us compute explicitly the Poisson

Lecture 2. Three dimensional Einstein’s gravity 52

 2.2. Asymptotically anti-de Sitter phase space

bracket :
ż 2π ˆ ˙
!
p`q p`q
) i` imx` inx` inx` 1 3 inx`
i Lm , Ln “ dφ e e B` L` ` 2L` B` e ´ B` e (2.65)
8πG 0 2
ż 2π „ 
i` ipm`nqx` ipm`nqx` 1 3 ipm`nqx`
“ dφ ´ipm ` nqe L` ` 2inL` e ´ p´iqn e (2.66)
8πG 0 2
„ż 2π  3
` ` `m
“ pm ´ nq dφ L` eipm`nqx ` δm`n,0 (2.67)
8πG 0 8G
p`q `
“ pm ´ nqLm`n ` m3 δm`n,0 (2.68)
8G

where the second equality was obtained by performing an integration by parts on B` or equivalently
on Bφ since the charges are computed at t “ Cst. The third equality comes from the integral repre-
sentation of the discrete δ-function. We see that the representation theorem is obeyed, and the central
` c
charge is given by m3 δm`n,0 8G ” 3
12 m δm`n,0 where the dimensionless Brown-Henneaux central charge
is:
3`
c“ . (2.69)
2G
The central extension is obviously zero for the zero-mode m “ 0, but by shifting the zero mode of the
charges (2.60), it is possible to cancel it for m “ ´1, 0, `1 :

p`q p`q
!
p`q p`q
) ´
p`q
¯ c
L̃m “ Lm ` δm,0 N, N P R ñ i L̃m , L̃n “ pm ´ nq L̃m`n ´ δm`n,0 N ` m3 δm`n,0 . (2.70)
12

We choose N “ c{24 in order to get a centerless subalgebra of AdS3 exact symmetries:

!
p`q p`q
)
p`q c
i L̃m , L̃n “ pm ´ nqL̃m`n ` pm2 ´ 1qm δm`n,0 . (2.71)
12
! )
p`q p´q
We need to also shift the p´q sector. We also have L̃m , L̃n “ 0, so the check is achieved. The
shift c{24 of both sectors amounts to a shift of the mass of c{12 “ 1{p8Gq, which is nothing else than
the difference of mass between the zero mass BTZ black hole and global AdS3 , as shown in Figure
2.7. Everything fits in nicely!

This result is a strong hint that AdS3 gravity is deeply related to a 2d CFT. It took many years to
unravel a deeper connection, through one instance of Maldacena’s AdS{CFT correspondence [24],
which requires much more structure. AdS3 is in this case embedded in 10d supergravity itself the
low energy limit of string theory. It is remarkable that a relatively simple semi-classical analysis of
pure gravity already hints at a holographically dual 2d conformal field theory!

A more detailed connection with CFTs can be also made with what has been presented together with
one additional ingredient. The entropy of a high energy state in a 2d CFT is given by a universal
formula known as Cardy’s formula,
˜c c ¸
c L EL c R ER
SCFT “ 2π ` . (2.72)
6 6

Lecture 2. Three dimensional Einstein’s gravity 53

 2.3. Asymptotically flat phase space

Here, EL and ER are the eigenvalues in the high energy state of L0 and L̄0 , the zero-modes of the Vi-
rasoro algebra, and c L , c R are the left and right Virasoro central charges. Using the Brown-Henneaux
central charge for both the left and right moving Virasoro’s and the zero modes of the BTZ black
hole, you will find that SCFT is exactly the geometrical BTZ black hole entropy (2.24)! In exact in-
stances of the AdS{CFT correspondence, a black hole can be described as a high energy state in
a CFT and its microscopic entropy can be exactly computed from field theory degrees of freedom
without gravity! (Now this is not all: finding instances of AdS{CFT correspondences and finding the
exact microscopic degrees of freedom for non-supersymmetric black holes are hard problems, which
are equivalent to quantifying gravity! This has been done only in a few cases in string theory starting
from Maldacena’s work).

2.3 Asymptotically flat phase space

2.3.1 Flat limit and the BMS 3 group

Let us give some elements about the asymptotically flat case. Now the cosmological constant Λ
vanishes and Einstein’s equation in the vacuum are simply Rµν “ 0. The asymptotic structure of
Minkowski spacetime is very different than AdS spacetime. We refer the reader to the Penrose dia-
gram depicted in Figure 3.1, which also applies in 3d. We focus on spacetimes that are locally asymp-
totically flat, i.e. whose metric tensor reduce to ηµν when some null coordinate r reaches infinity. As
in the asymptotically AdS3 case, a set of boundary conditions can be formulated [35], and the most
general solution of Einstein’s equations can be exactly derived [36]. Instead of developing here the
full deduction of suitable boundary conditions and the analytical derivation of the solutions, we take
advantage of our knowledge about asymptotically AdS3 spacetimes, and we simply take the limit
` Ñ 8 (which is equivalent to Λ Ñ 0) [33]. Geometrically, this process rejects the boundary cylinder
to infinity and as the length scale become infinite, the entire bulk looks like the previous "center" of
AdS3 , and thus is locally flat. Expressed in terms of Bondi coordinates pu, r, φq, where u is the retarded
time, r is the luminosity distance, and φ the angle on the circle at infinity, the solution reads as
” u ı
ds2 “ Θpφqdu2 ´ 2dudr ` 2 Ξpφq ` Bφ Θpφq dudφ ` r2 dφ2 . (2.73)
2

The phase space is also parametrized by 2 arbitrary functions on the boundary (here the circle at
infinity) but here they depend only on φ because x˘ “ pt{`q ˘ φ Ñ φ when ` Ñ 8. One can check
that this metric is Ricci-flat, and clearly, we retrieve Minkowski spacetime for Θ “ ´1 and Ξ “ 0.
Again, instead of computing the asymptotic symmetry group that preserves the phase space, we can
directly take the“flat limit" of the asymptotic symmetry group of the AdS3 phase space. To do this,
we first define
p`q 1 p´q 1
ξ m “ p` Pm ` Jm q ; ξn “ p` P´n ´ J´n q. (2.74)
2 2
A straightforward computation shows that

1
irPm , Pn s “ pm ´ nqJm`n ; irJm , Jn s “ pm ´ nqJm`n ; irJm , Pn s “ pm ´ nqPm`n . (2.75)
`2

Lecture 2. Three dimensional Einstein’s gravity 54

 2.3. Asymptotically flat phase space

So when we take the flat limit ` Ñ 8, the Pm commute, the Jm form a Witt algebra, and act non-
trivially on the Pm .

irPm , Pn s “ 0 ; irJm , Jn s “ pm ´ nqJm`n ; irJm , Pn s “ pm ´ nqPm`n . (2.76)

These commutation relations define the so-called bms3 algebra ! Let us first note that it extends the
Poincaré algebra isop2, 1q, in the same way that the asymptotic symmetry algebra of the AdS3 phase
space extended the sop2, 2q algebra of exact symmetries of AdS3 : the first modes m, n “ ´1, 0, `1
form a subalgebra of bms3 containing (check it as an exercise) :

B 2 ` 1 translations : Bt “ P0 , Bx “ P`1 ` P´1 , By “ ipP`1 ´ P´1 q ;

B 1 rotation Bφ “ R0 and two boosts xBt ´ tBx “ J`1 ` J´1 , yBt ´ tBy “ ipJ`1 ´ J´1 q.

We see that the asymptotic symmetry group of asymptotically flat spacetimes is also larger than
the exact symmetry group of the flat spacetime itself ! Instead of 3 translations, we get an abelian
subalgebra of bms3 , usually denoted as vectpS1 qab , that contains the so-called supertranslations Pm . The
Lorentz algebra sop2, 1q is also enhanced into an infinite-dimensional algebra of diffeomorphisms on
the circle, vectpS1 q, which now contains the so-called superrotations Jm . The last commutation relations
tell us that the supertranslations form an ideal as do the translations in the Poincaré subalgebra. Since
vectpS1 q acts on vectpS1 qab as the adjoint representation, we can write

bms3 “ vectpS1 q iad vectpS1 qab . (2.77)

The asymptotic symmetry group is the integral version of this algebra, and reads as a semi-direct
product BMS3 “ DiffpS1 q ˙ Ad VectpS1 qab between the group of diffeomorphisms on the circle, and its
own Lie algebra, seen here as a abelian normal subgroup.

We can obtain by a fairly similar process the bms3 charge algebra. Before taking ` Ñ 8, we again
define
p`q 1 p´q 1
Lm “ p`Pm ` Jm q ; Ln “ p`P´n ´ J´n q. (2.78)
2 2
Recalling the Virasoro charge algebra (2.71), one can check that

1
irPm , Pn s “ pm ´ nqJm`n ; (2.79)
`2
irJm , Jn s “ pm ´ nqJm`n ; (2.80)
1
irJm , Pn s “ pm ´ nqPm`n ` mpm2 ´ 1qδm`n,0 . (2.81)
4G

The central charge is now free of any `. The charges remain finite in the flat limit, and their algebra
reads as

irPm , Pn s “ 0 ; (2.82)
irJm , Jn s “ pm ´ nqJm`n ; (2.83)
1
irJm , Pn s “ pm ´ nqPm`n ` mpm2 ´ 1qδm`n,0 . (2.84)
4G

Lecture 2. Three dimensional Einstein’s gravity 55

 2.3. Asymptotically flat phase space

The centerless part of this algebra forms an algebra isomorphic to isop2, 1q, so the lowest modes m, n “
´1, 0, `1 are the Poincaré charges (energy, linear and angular momentum and Lorentz charges). By
direct analogy, the Pm are called supermomenta while the Jm receive the name of super-Lorentz charges
or superrotation charges.

Again using the flat limit process, it is possible to compute the explicit form of the bms3 vectors. What
we have to do is simply take the expressions of ξ ` , ξ ´ , compute the transformation to pass into P, J
vectors, then take the flat limit and express the result in Bondi coordinates (see again [33]) :
$
u ´1
& ξ “ Tpφq ` uBφ Rpφq ` O pr q;
’
ξ Y,T “ ξ µ Bµ : ξ r “ ´rBφ Rpφq ` O pr0 q; (2.85)
’
ξ “ Rpφq ´ ur Bφ2 Rpφq ` O pr´1 q.
% φ

The generators depend on two arbitrary functions on the circle, Tpφq representing arbitrary super-
translations, and Rpφq representing arbitrary superrotations! We obtain Pm with pT “ eimφ , R “ 0q
and Jn with pT “ 0, R “ einφ q. The corresponding surface charges are
ż 2π
1
Pn “ dφpΘpφq ` 1qeinφ , (2.86)
16πG 0
ż 2π
1
Jn “ dφΞpφqeinφ . (2.87)
8πG 0

We can transform (2.73) along the flow of ξ T,R to deduce the transformation laws of the metric fields
Θ, Ξ :

δT,R Θ “ RBφ Θ ` 2Bφ RΘ ´ 2Bφ3 R, (2.88)

1
δT,R Ξ “ RBφ Ξ ` 2Bφ RΞ ` TBφ Θ ` Bφ TΘ ´ Bφ3 T. (2.89)
2

We can now infer the “boundary field content” of the phase space. By virtue of (2.88), Θ belongs to
the coadjoint representation of DiffpS1 q. We can introduce what we call the superrotation field Ψpφq
which is invariant under supertranslations and which transforms under superrotations as

δT,R Ψ “ RBφ Ψ ` Bφ R. (2.90)

The transformation (2.88) allows to recognize Θ “ pBφ Ψq2 ´ 2Bφ2 Ψ ` 8GMe2Ψ . When Ψ “ 0, one
is left with a zero mode which cannot be generated by a diffeomorphism Θ0 “ 8GM. The mass
M is recognized after computing the charges, as the canonical conjugated charge to P0 “ Bt . To
untangle the second transformation law (2.89) a second fundamental boundary field is necessary
which we call the supertranslation field Cpφq. After some algebra, we find convenient to define
Ξ “ ΘBφ C ´ Bφ3 C ` 4GJe2Ψ ` 21 Bφ ΘC. The transformation property (2.89) is reproduced from (2.90)
and

δT,R C “ T ` RBφ C ´ CBφ R. (2.91)

The zero mode Ξ0 “ 4GJ is recognized after computing the charges as determined by the angular

Lecture 2. Three dimensional Einstein’s gravity 56

 2.3. Asymptotically flat phase space

momentum conjugated to ´Bφ . In summary, the field space is parameterized by the supertranslation
field Cpφq, the superrotation field Ψpφq and the two zero modes M, J. This description is slightly
redundant, because not all the modes of C and Ψ lead to distinct metrics (for example the lowest 3
harmonics of C are annihilated by Bφ pBφ2 ` 1q and therefore they do not modify Ξ). Studying these
subtleties is called studying the orbits of the BMS3 group and it has been done in detail, we refer the
reader to [37]!

Let us finally discuss how Minkowski spacetime transforms under supertranslations and superro-
tations. First, it does not transform under the Poincaré subgroup since these are isometries. In the
language that we just developped, Θ “ ´1 and Ξ “ 0 are left invariant under Poincaré transfor-
mations, because Bφ pBφ2 ` 1qR “ Bφ pBφ2 ` 1qT “ 0. Now, acting with general supertranslations and
superrotations, the metric changes and the canonical charges also change. If one only acts with
supertranslations, the energy remains zero, so the vacuum is degenerate and parameterized by its
superrotation charge! The field Cpφq is precisely the Goldstone boson which comes from the spon-
taneous breaking of supertransation invariance. When also acting with superrotations, the energy
now changes and one finds new (classical) states with supertranslation charges, which are related to
Minkowski by the action of superrotations.

2.3.2 Constant representatives : spinning particles and flat cosmologies

To conclude our discussion of the flat case, let us focus on constant representatives of the phase space.
As we did for the asymptotically AdS3 phase space, let us reduce Θ, Ξ to their zero modes in terms of
M, J and represent the phase diagram. If we look at the previous phase space (Figure 2.7), we see that
the flat limit ` Ñ 8 cancels the slope of the extremal lines. A second effect is that the upper half plane
is also not anymore filled by black holes in the flat case: the reason is that by sending ` Ñ 8 we zoom
in the interior of the black holes so we are left with cosmological spacetimes without horizon! This is
consistent with a theorem due to Ida [38] that says that black holes cannot exist in 2 ` 1 dimensions
when the cosmological constant is not negative assuming reasonable matter (obeying the null energy
condition). Below the massless line, we find again spinning particles that we will describe a bit more
here !

For later convenience, let us note M̄ “ 8GM and J̄ “ 4GJ. For constant representatives, the metric
reads simply as
ds2 “ M̄du2 ´ 2dudr ` J̄dudφ ` r2 dφ2 . (2.92)

When M̄ ą 0 and J̄ is arbitrary, we get an expanding spacetime (or a contracting spacetime after time
reversal) enclosing an (unphysical) time machine hidden by a cosmological horizon! Let us define
α2 “ M̄ and r0 “ | J̄{α|. The spacetime is clearly divided in two distinct parts:

B If r ă r0 we can perform the change of coordinate r2 Ñ r̄2 “ pr02 ´ r2 q{α2 to get the line element
ds2 “ p´αdt ` r0 dφq2 ` dr̄2 ´ α2 r̄2 dφ2 . In this inner region, Bφ is always timelike, and gener-
ates closed timelike curves. This unphysical “time machine” is shielded from the rest of the
spacetime by the cosmological horizon r “ r0 [39, 40] ;

B In the outer region (r ą r0 ), one can define new coordinates T 2 “ pr2 ´ r02 q{α2 and X “ pαt{r0 q `

Lecture 2. Three dimensional Einstein’s gravity 57

 2.3. Asymptotically flat phase space

φ such that pX, φq „ pX ` 2π, φ ` 2πq. The metric becomes ds2 “ ´dT 2 ` r02 dX 2 ` α2 T 2 dφ2 ,
which describes a spacetime expanding with growing T with T “ 0 as the big bang. The closed
timelike curves of the inner region are enclosed in the “pre big-bang” era, which we need to cut
out. This kind of cosmological spacetime is called a flat cosmology !
?
Let us consider the subset of metrics for which M̄ ă 0. Let us denote α2 “ ´ M̄ ą 0 (α “ ´M) and
r0 “ J̄{α. The change of coordinate r̄2 “ pr2 ` r02 q{α2 puts the line element into the form
ˆ ˙2
2 4GJ
ds “ ´ αdt ´ dφ ` dr̄2 ` α2 r̄2 dφ2 (2.93)
α

which is nothing but a spinning particle found in 1984 by Deser, ’t Hooft and Jackiw [41]. We can give
a more geometrical interpretation of this line element by performing a second change of coordinates
t̄ “ αt ´ r0 φ, φ̄ “ αφ. In these coordinates, ds2 “ ´dt̄2 ` dr̄2 ` r̄2 dφ̄2 : it is now manifest that
the spacetime is locally flat ! But the difference between a spinning particle and the Minkowskian
vacuum lies in the periodic identification :
ˆ ˙
8πGJ
pt, φq „ pt, φ ` 2πq Ñ pt̄, φ̄q „ t̄ ´ , φ̄ ` 2πα . (2.94)
α

For J ‰ 0, there is a twist in the time identification, which leads to spin. Let us discuss the static case
J “ 0. The phase space is summarized in Figure 2.8.

Flat cosmologies

J
1
´ 8G Conical defects
Minkowski

Conical excesses

Figure 2.8: Constant representatives of the asymptotically flat phase space.

Figure adapted with permission from [33].Copyrighted by the American Physical Society.

B For M̄ “ ´1 or, equivalently, α “ 1, one has Minkowski spacetime.

B For ´1 ă M̄ ă 0 (α2 ă 1), we find a conical defect. The spacetime can be created from a cut
and paste procedure. Cut a wedge of angle 2πp1 ´ |α|q out of the plane and glue the remaining
edges. This is a conical defect.

Lecture 2. Three dimensional Einstein’s gravity 58

 2.4. Chern-Simons formulation

B When M̄ ă ´1 (α2 ą 1), we find a conical excess. This is equivalent to incise the plane along
a half-line and then introduce an excendentary section of angle 2πp|α| ´ 1q between the two
edges of the incision. In this case, the energy spectrum is not bounded from below, so we often
discard these solutions.

Relaxing the hypothesis of analyticity, multi-particle solutions can be found [41], and each of them
may carry a BMS representation! Such a general metric has not yet been described. As already men-
tioned, these particles will not attract, as we have already stressed before; the Newtonian potential
does not exist in 3d gravity ! On this remark, we close the presentation of the asymptotically flat
phase space, with the hope that through it, you got more intuition about the intriguing properties of
3d gravity !

2.4 Chern-Simons formulation

The last topic that we will discuss concerning 3d gravity is a reformulation of the theory. Einstein’s
theory with Λ ă 0 is equivalent (at least classically) to the difference of two Chern-Simons actions of
non-abelian gauge fields which both transform under the adjoint representation of SOp2, 1q. This is
the Chern-Simons formulation of 3d Einstein gravity, first discovered by Achúcarro and Townsend
in 1986 [27].

First, let us reset our notations. Spacetime indices will still be denoted as µ, ν, ... but now latin indices
will represent Lorentz indices in the local triad frame. We will denote the Levi-Civita connection
compatible with gµν as ∇µ and reserve D as the covariant derivative on objects that transform un-
der the gauge group! We will write forms in bold, except the triad and spin connection, which is
conventional.

2.4.1 Local Lorentz triad

At each point of the manifold M, we can find a local change of frame in which the metric is locally flat
(this is the equivalence principle). The natural basis of this frame is given by a triad of Lorentz vector-
valued 1-forms te a “ e aµ dx µ u which obey gµν dx µ dx ν “ ηab e a eb . Since the metric admits an inverse,
µ µ µ µ
we can also define the inverse of e aµ which we denote by ea (ea ebµ “ δab and ea e aν “ δν ). With respect
µ
to this orthonormal local basis, the connection coefficients are given by ea ∇µ eb ” ∇ a eb “ ec ω cab .
Recall that for a vector χ “ χα eα we have ∇µ χ “ pBµ χα ` Γαµν χν qeα so we can inverse the previous
relation to obtain the equation linking the connection coefficients Γαµν in the coordinate basis with the
connection coefficients ω cab in the orthonormal basis :

ω cab “ ecν ea pBµ ebν ` ebα Γνµα q “ ecν ea ∇µ ebν .

µ µ
(2.95)

We also define the spin connection ω ab fi ω abc ec which are tensor-valued 1-forms on M that belong to
the adjoint representation of the local Lorentz group. Indeed, under a Lorentz rotation of the triad,

ω 1ab “ Λ ac ω cd pΛ´1 qdb ` Λ ac pdΛ´1 qcb . (2.96)

We can also use the flat metric ηab and its inverse η ab to lower and raise local Lorentz indices,

Lecture 2. Three dimensional Einstein’s gravity 59

 2.4. Chern-Simons formulation

ωab “ ηac ω cb . The metric-compatibility of the connection (i.e. ∇α gµν “ 0) reads as ωab “ ´ωba
in this formalism, so the spin connection can be seen as antisymmetric 3 ˆ 3 matrices of one-forms.

In 3d spacetime, the special feature that leads to a further reformulation is that antisymmetric matrices
ωab are dual to vectors, since the completely antisymmetric tensor ε abc exists in 3d (with ε012 “ 1). So
we are able to construct a local Lorentz vector from the spin connection :

1 abc
ω a fi ε ωbc ô ωab “ ´ε abc ω c (2.97)
2

or, equivalently, a Lorentz one-form ωa “ ηab ω b . Therefore, it is possible to treat the spin connection
ωa and the triad ea on an equal footing!

2.4.2 Chern-Simons action

Using the existence of the cosmological length scale `, Λ “ ´1{`2 , we can introduce the dimension-
ally consistent connections
ea ea
Aa “ ω a ` ; Āa “ ω a ´ . (2.98)
` `
Let us now denote by Ja a set of matrices of sop2, 1q that obey the algebra rJa , Jb s “ ε abc η cd Jd ” ε abc Jc .
Since Aa carries one Lorentz index, it can be understood as the components of a sop2, 1q vector in the
matricial base tJa u. We will consider a matrix representation of the connection:

e a Ja e
A “ A a Ja “ ω a Ja ` ”ω` (2.99)
` `
and the same for the other connection Ā “ Āa Ja . We can say that A and Ā transform under the adjoint
representation of SOp2, 1q !

Since we are working with a local orthonormal triad, we can make profit of Cartan’s calculus. In
particular, the second Cartan structure equation links curvature with the spin connection:

Rab “ dω ab ` ω ac ^ ω cb . (2.100)

From R ab we can reconstruct the components of the Riemann tensor, given that

1 a 1
Rab “ R ec ^ ed “ R abµν dx µ ^ dx ν ñ Rα βµν “ eαa ebβ R abµν . (2.101)
2 bcd 2

We can peform a contraction between Rab and ε abc to get a vector in the Lorentz frame

1 abc 1
Ra fi ε Rbc “ dω a ` ε a bc ω b ^ ω c . (2.102)
2 2

Since Ra also transforms under the adjoint representation of the local Lorentz group, we can build a
matrix representation of it:

1
R fi Ra Ja “ dω ` rJc , J f sω c ^ ω f “ dω ` ω ^ ω. (2.103)
2

Lecture 2. Three dimensional Einstein’s gravity 60

 2.4. Chern-Simons formulation

The two-form R should not be confused with the Ricci scalar R.

Now that we have defined all the necessary geometrical objects, we are ready to show that the
Einstein-Hilbert action for 3d gravity is equivalent (up to a boundary term) to a couple of Chern-
Simons actions, one for each connection. The latter is build up from a 3-form, because the spacetime
is 3-dimensional. The scalar action is obtained by integration of this 3-form on M. The most simple
3-forms that one can construct from A are A ^ dA and A ^ A ^ A. These are the only two terms which
appear in the Chern-Simons form

2
IrAs “ A ^ dA ` A ^ A ^ A. (2.104)
3

After a straightforward computation that can be done without any subtlety, we obtain
„ 
2 2
trpIrAs ´ IrĀsq “ tr 2e ^ R ` 2 e ^ e ^ e ´ dpω ^ eq . (2.105)
` 3`

where the trace is taken on the representation of sop2, 1q to which belong the matrices A, R,... On the
2
sop2, 1q algebra we can define a Killing product such as kpJa , Jb q “ N trpJa Jb q ” ηab . The normalisation
N is representation-dependent, and as we will work with the natural 2d representation, we pick the
value N “ 1 which is the standard normalisation for this representation. So the explicit computation
of the trace gives
„  „ 
2 1 1
tr 2e ^ R ` 2 e ^ e ^ e ´ dpω ^ eq “ ε abc e a ^ Rbc ` 2 e a ^ eb ^ ec ´ dpω a ^ ea q. (2.106)
3` 2 3`
?
Recalling the definition of a determinant, one has detpeq “ detpeµa q “ ´g and furthermore

ε abc e a ^ Rbc “ ´g R d3 x ε abc e a ^ eb ^ ec “ 3! ´g d3 x

a a
; (2.107)

where R is here the Ricci scalar associated to gµν . Finally

ˆ ˙
2 1a 1a 2
trpIrAs ´ IrĀsq “ ´gR ` 2 ´g d3 x ´ dpω a ^ ea q. (2.108)
` 2 ` `

We have thus showed that the difference of two Chern-Simons actions:

ż
k `
SCS rA, Ās “ trpIrAs ´ IrĀsq , k fi (dimensionless) (2.109)
4π M 4G

is equivalent to the 3d Einstein-Hilbert action, up to a boundary term that has no effect on the equa-
tions of motion:
ż ż
1 1
3
ω a ^ ea “ SEH rgs ` Boundary term. (2.110)
a
SCS rA, Ās “ d x ´g pR ` 2Λq ´
16πG M 16πG BM

So we can use the Chern-Simons formulation of 3d gravity, which presents the very nice advantage
to be based on gauge fields which belong to the adjoint representation of sop2, 1q!

Lecture 2. Three dimensional Einstein’s gravity 61

 2.4. Chern-Simons formulation

It is a simple matter of tensorial calculus to get the equations of motion associated to (2.109). First
let us develop all exterior products in components. We are allowed to use spacetime indices to write
explicitly IrAs, IrĀs since all objects are tensorial...
„ ˆ ˙
k k 2
LCS rA, Ās “ trpIrAs ´ IrĀsq “ tr ´εµαβ Aµ Bα A β ` Aµ Aα A β ´ pĀ-partq. (2.111)
4π 4π 3

The Euler-Lagrange equations for the gauge field A then read as

δLCS BLCS BLCS

“ ´ Bα (2.112)
δAµ BAµ BBα Aµ
k ` ˘ k ´ βαµ ¯
“ ´ εµαβ Bα A β ` 2Aν A β ´ ´ε Bα A β (2.113)
4π 4π
k µαβ ` ˘
“´ ε Bα A β ` A α A β (2.114)
2π
k
“ ´ εµαβ Fαβ (2.115)
2π

where we have defined the curvature tensor F associated to the connection A as F “ dA ` A ^ A ñ

Fµν “ Bµ Aν ´ Bν Aµ ` rAµ , Aν s. Fµν is antisymmetric by construction. The equations of motion exactly
state that the connections A and Ā are flat !

F “ dA ` A ^ A “ 0 & F̄ “ dĀ ` Ā ^ Ā “ 0. (2.116)

Any flat solution is locally pure gauge, i.e. of the form

A “ g´1 dg, (2.117)

as we can check straighforwardly

dA ` A ^ A “ dg´1 ^ dg ` g´1 dg ^ g´1 dg (2.118)

´1 ´1 ´1 ´1
“ ´g dg ^ g dg ` g dg ^ g dg “ 0. (2.119)

2.4.3 General covariance and charges

We now know the relationship between Chern-Simons theory and 3d Einstein gravity. The latter the-
ory is by construction invariant under arbitrary diffeomorphisms, so a natural question is: how is a
diffeomorphism represented in terms of gauge transformations of the fundamental fields A and Ā ?
First recall that an infinitesimal gauge transformation with parameter λ “ λ a Ja acts on the connection
field as δλ Aµ “ Dµ λ where Dµ is the gauge-covariant derivative defined by Dµ λ “ Bµ λ ` rAµ , λs.
The second term is necessary for non-abelian gauge groups.

If we perform a gauge transformation with parameters λ a and λ̄ a , does it correspond to a diffeomor-

Lecture 2. Three dimensional Einstein’s gravity 62

 2.4. Chern-Simons formulation

phism of the metric obtained from the Chern-Simons dictionary? We need to compute

`
δeµ “ pδξ Aµ ´ δξ Āµ q (2.120)
2
`
“ pDµ λ ´ D̄µ λ̄q (2.121)
2
` 1
“ pBµ pλ ´ λ̄q ` rωµ , λ ´ λ̄sq ` reµ , λ ` λ̄s (2.122)
2 2
1
“ Bµ peν ξ ν q ` rωµ , eν ξ ν s ` reµ , λ ` λ̄s. (2.123)
2

In the last step, we introduced our ansatz for the diffeomorphism: eµa ξ µ “ 2` pλ a ´ λ̄ a q. We then have

1
δeµ “ eν Bµ ξ ν ` pBµ eν ` rωµ , eν sqξ ν ` reµ , λ ` λ̄s (2.124)
2
a ν a a b ν 1
“ Ja peν Bµ ξ ` pBµ eν ` ωµb eν qξ q ` reµ , λ ` λ̄s (2.125)
2
1
“ Ja peνa Bµ ξ ν ` eνa Γνµλ ξ λ q ` reµ , λ ` λ̄s (2.126)
2
1
“ eλ ∇µ ξ ` reµ , λ ` λ̄s.
λ
(2.127)
2

The first term is responsible for the transformation of the metric under the infinitesimal diffeomor-
phism parametrized by ξ µ . The second term is responsible for a local Lorentz transformation which
does not appear in the variation of the metric. We can check that it is so by evaluating
´ ¯
δgµν “ δ eµa eνb η ab (2.128)
´ ¯
“ δ eµa eνb trpJa Jb q (2.129)

“ trpδeµ eν ` eµ δeν q (2.130)

ˆ ˙ ˆ ˙
1 1
“ tr eλ ∇µ ξ λ ` reµ , λ ` λ̄s eν ` eµ eλ ∇ν ξ λ ` reν , λ ` λ̄sq (2.131)
2 2
“ trpeλ eν ∇µ ξ λ ` eµ eλ ∇ν ξ λ q (2.132)
“ ∇ν ξ µ ` ∇µ ξ ν . (2.133)

That proves it! Since we are not interested in the local Lorentz transformations, we can fix the ambi-
guity in our definition of ξ µ by selecting

λ a “ ξ µ Aµa , λ̄ a “ ξ µ Āµa . (2.134)

Using (2.98), we have λ a ´ λ̄ a “ 2` ξ µ e aµ which matches our definition.

Let us now quickly compute the surface charges associated to these diffeomorphisms. In the Chern-
Simons theory, we can obtain a general expression of these quantities (which are conserved if ξ µ is a
symmetry), which is furthermore very simple and elegant ! The path we take is the Barnich-Brandt
µ
method. We begin by computing the conserved current Sξ defined in the second Noether theorem

Lecture 2. Three dimensional Einstein’s gravity 63

 2.4. Chern-Simons formulation

(Result 5):

δLCS δLCS
dSξ “ a
δλ Aµa ` δ Ā a , pλ “ Aµ ξ µ , λ̄ “ Āµ ξ µ q (2.135)
δAµ δ Āµa λ̄ µ
δLCS
“ pBµ λ a ` rAµ , λsa q ` pbarred sectorq (2.136)
δAµa
˜ ¸
δLCS a δLCS δLCS
“ Bµ a
λ ´ Bµ a
λ` rAµ , λsa ` pbarred sectorq (2.137)
δAµ δAµ δAµa
˜ ¸
δLCS a
“ Bµ λ ` pbarred sectorq. (2.138)
δAµa

δLCS k µαβ a a “ 0 and εµαβ F a rA , λsa “

In the last step, we replaced δAµa “ ´ 2π ε Fαβ and used that εµαβ Bµ Fαβ αβ µ

trpεµαβ Fαβ rAµ , λsq “ trpεµαβ rFαβ , Aµ sλq “ 0 by cyclic property of the trace and antisymmetry. There-
fore,

µ k µαβ a a a a
Sξ “ ´ ε pFαβ λ ` F̄αβ λ̄ q. (2.139)
2π

A simple application of Anderson’s operator (1.121) separates us from the charge formula. Let us
µ
look only at the unbarred sector. Since there is only one derivative acting on the field Aµ in Sξ , there
is only one term to compute:

1 B B ´ µ 2 ¯
kξBB “ IδA
2
Sξ “ δAα Sξ pd xqµ (2.140)
2 BB A Bdx ν
„ ν α  „ 
1 B k µρσ
( B 1 β γ
“ δAα ´ tr ε pBρ Aσ ` Aρ Aσ qλ ε µβγ dx dx (2.141)
2 BBν Aα 2π Bdx ν 2
k ! ) k
“ ´ tr δAα εµρσ pδρν δασ ` 0qλ ε µνγ dx γ “ ´ trpδAα λqεµνα ε µνγ dx γ (2.142)
4π 4π
k k k
“ ´ trpδAα λqp´2δγα qdx γ “ trpδAα λqdx α “ trpλδAα qdx α . (2.143)
4π 2π 2π

The last equality is obtained thanks to the cyclicity of the trace. If we incorporate the contribution of
the Ā-part and substitute λ, λ̄ in terms of ξ µ , we have just found that

k “ ‰
kξ rδA, As “ tr piξ AqδA ´ piξ ĀqδĀ . (2.144)
2π

This is our final infinitesimal charge formula! Note that it is not clearly integrable, as in other non-
linear theories such as Einstein gravity: compare with (1.136)! One needs to specify either a specific
vector, or boundary conditions to deduce the (integrated) surface charge.

2.4.4 AdS 3 phase space in the Chern-Simons formalism

We wish to close this section, and at the same time this chapter by deriving the connections A, Ā that
describe the AdS3 phase space in the Chern-Simons formulation. Let us begin by the most simple
case which is global AdS3 itself. Recall that in global coordinates pρ, t, φq, the metric reads as
´ ¯
ds2 “ `2 dρ2 ´ cosh2 ρ dt2 ` sinh2 ρ dφ2 . (2.145)

Lecture 2. Three dimensional Einstein’s gravity 64

 2.4. Chern-Simons formulation

Since the line element is diagonal, one choice of the orthonormal basis is quite trivial:

e0 “ ` cosh ρ dt ; e1 “ ` sinh ρ dφ ; e2 “ `dρ. (2.146)

Consequently, the non-vanishing components of the spin connection are

+ #
ω20 “ sinh ρ dt ω 0 “ ω12 “ cosh ρ dφ
ùñ (2.147)
ω12 “ cosh ρ dφ ω 1 “ ω20 “ sinh ρ dt

together with the other components related by indicial symmetries. The components of the connec-
tions A “ Aa Ja , Ā “ Āa Ja evaluate to

A0 “ cosh ρ pdt ` dφq Ā0 “ ´ cosh ρ pdt ´ dφq;

A1 “ sinh ρ pdt ` dφq Ā1 “ sinh ρ pdt ´ dφq; (2.148)
A2 “ dρ Ā2 “ ´dρ.

We can clearly simplify the notations by introducing again lightcone coordinates x˘ “ t ˘ φ, so we

get : #
A “ pcosh ρ J0 ` sinh ρ J1 q dx` ` J2 dρ;
(2.149)
Ā “ p´ cosh ρ J0 ` sinh ρ J1 q dx´ ´ J2 dρ.
`J
We claim that the connections are locally pure gauge, A “ g´1 dg, with g “ e x 0 eρJ2 . Trivially we
get Aρ “ g´1 Bρ g “ J2 and A´ “ g´1 B´ g “ 0. The last component is more tricky to compute since
rJ0 , J2 s ‰ 0. To progress we have to invoke Hadamard’s lemma

1 1
AdjX Y “ eadjX Y ô e X Ye´X “ Y ` rX, Ys ` rX, rX, Yss ` rX, rX, rX, Ysss ` ¨ ¨ ¨ (2.150)
2! 3!

We find A` “ g´1 B` g “ e´ρJ2 J0 eρJ2 “ cosh ρ J0 ` sinh ρ J1 after summation of MacLaurin’s series of
´J
cosh, sinh. As an exercise, show that Ā is also pure gauge, and associated to ḡ “ e´x 0 e´ρJ2 . To make
connection with the more general form that we will derive in a few moments, let us write the gauge
fields in terms of the generators tL´1 , L0 , L`1 u defined by

1 1
J0 “ pL`1 ` L´1 q ; J1 “ pL`1 ´ L´1 q ; J2 “ L0 (2.151)
2 2

and which satisfy the slp2, Rq algebra (isomorphic to the Lorentz algebra, or the centerless part of the
Witt algebra):
rL`1 , L0 s “ `L`1 ; rL´1 , L0 s “ ´L´1 ; rL`1 , L´1 s “ 2L0 . (2.152)

The global AdS3 solution is thus perfectly reproduced by the Chern-Simons connections
#
A “ ` 12 peρ L`1 ` e´ρ L´1 q dx` ` L0 dρ,
(2.153)
Ā “ ´ 12 peρ L´1 ` e´ρ L`1 q dx´ ´ L0 dρ,

or, if we want to specify these in a manifestly pure gauge fashion A “ g´1 dg, Ā “ ḡ´1 d ḡ where

1 ` 1 ´
g “ e 2 pL`1 `L´1 qx eρL0 ; ḡ “ e´ 2 pL`1 `L´1 qx e´ρL0 . (2.154)

Lecture 2. Three dimensional Einstein’s gravity 65

 2.4. Chern-Simons formulation

As a side note, remember that the Killing symmetry algebra of AdS3 is sop2, 2q. The fact that the
Chern-Simons action contains two chiral terms makes manifest the chiral decomposition sop2, 2q “
sop2, 1q ‘ sop2, 1q...

Let us extend the discussion now to the entire AdS3 phase space. We recall that under the Brown-
Henneaux boundary conditions, any asymptotically AdS3 metric can be written in the Fefferman-
Graham gauge as (2.57). After setting r “ eρ and rewriting the null boundary fields as L˘ px˘ q Ñ
1 ˘
k L˘ px q we get
« ˜ ˆ ˙2 ¸ ff
1 1
ds2 “ `2 dρ2 ` L` pdx` q2 ` L´ pdx´ q2 ´ e2ρ `
` ˘
L` L´ e´2ρ dx` dx´ . (2.155)
k k

The Lorentz triad

ˆ ˙ ˆ ˙
1 1
epρq “ `dρ ; ep`q “ ` eρ dx` ´ L´ e´ρ dx´ ; ep´q “ ` eρ dx´ ´ L` e´ρ dx` (2.156)
k k

is chosen to bring the metric into the form ds2 “ epρq b epρq ´ 12 ep`q b ep´q ´ 21 ep´q b ep`q where in-
dices in brackets are the local Lorentz indices. We leave as an exercise to the reader to compute the
connections # `1
L`1 ´ 2k L` px` qe´ρ L´1 dx` ` L0 dρ,
˘
A “ ` eρ
` 21 ρ (2.157)
L´1 ´ 2k L´ px´ qe´ρ L`1 dx´ ´ L0 dρ.
˘
Ā “ ´ 2e

This is a generalization of global AdS3 spacetime, which is recovered when L` “ L´ “ ´k{4.

As a very last check, we can show that the Chern-Simons charge formula reproduces the charges that
we have derived in the metric formalism. Using (2.157), we get immediately

2 2
δA “ ´e´ρ δL` L´1 dx` and δĀ “ `e´ρ δL´ L`1 dx´ . (2.158)
k k

The charge formula is

ż
k
δHξ “ trrλδA ´ λ̄δĀs. (2.159)
2π S

Since δA „ L´1 , the trace will be non-zero only for λ „ L`1 , since trpL`1 L´1 q “ ´1 and otherwise
zero. With the same reasoning, we can concentrate on the λ̄ „ L´1 part only. The gauge parameters
which capture the n-th harmonic of the diffeomorphism ξ p`q defined in (2.47) are

p`qµ 1 ` p`qµ
λ “ ξn Aµ “ p¨ ¨ ¨ qL0 ` einx eρ L`1 and λ̄ “ ξ n Āµ “ p¨ ¨ ¨ qL0 ` 0. (2.160)
2

The charge formula then evaluates to

ż ˆ ˙
k 1 inx` ρ ´ρ 2
δHξ p`q “ trpL`1 L´1 q e e ´e δL` dx` (2.161)
n 2π S 2 k
ˆ ż ˙
1 inx` `
“δ e L` px qdφ (2.162)
2π S

Lecture 2. Three dimensional Einstein’s gravity 66

 2.4. Chern-Simons formulation

since dx` “ dφ on the boundary circle. The charge is clearly integrable and it reproduces our expec-
`
tations. The null boundary fields L˘ being related to the previous ones by L˘ “ kL˘ “ 4G L˘ , we
have in fact proven that ż
` `
Hξ p`q “ dφ L` einx (2.163)
n 8πG S

which is precisely the surface charge (2.60) obtained in the metric formalism. This last cross-check
ends up our trip into the marvellous world of 3d gravity!

References
Spinning particles as conical defects and excesses in 3d Minkowski spacetime and AdS3 were de-
scribed in 1984 by Deser, Jackiw and ’t Hooft [41, 42].

The Chern-Simons formulation of 3d gravity was found by Achúcarro and Townsend in 1986 [27]. In
this lectures, we followed the presentation of unpublished notes of Juan Jottar (partly published in
[43]) who we gratefully thank.

The same year Brown and Henneaux performed their asymptotic symmetry group analysis of AdS3
[23] using the Hamiltonian formalism. We followed here the equivalent covariant approach and used
the notation of [6].

The Bañados, Teitelboim and Zanelli BTZ black hole was found in 1992 [28]. We followed the sub-
sequent paper [29] in which Henneaux joined, where most of the geometrical properties of the BTZ
black hole were found.

The BMS3 asymptotic symmetry algebra was found at null infinity in [44, 35] and at spatial infinity
in [45]. Understanding the limit from AdS3 to Minkowski for the symmetry algebra was described in
[35], and for the full phase space in [33]. The flat spacetime cosmologies and their thermodynamical
properties were understood only in 2012 in [39, 40].

Lecture 2. Three dimensional Einstein’s gravity 67

 Lecture 3

Asymptotically flat spacetimes

For the next lecture, we go back to more realistic gravitational models: indeed we consider four-
dimensional asymptotically flat spacetimes, which are the solutions of General Relativity with lo-
calised energy-momentum sources. We are obviously not going to make an exhaustive overview
of this rich and deep topic. We will start with a review of the work of Penrose on the conformal
compactification of asymptotically flat spacetimes in order to get a global view on the asymptotic
structure. We will then concentrate on the properties of radiative fields by reviewing the work of
van der Burg, Bondi, Metzner and Sachs of 1962. One may think at first that the group of asymp-
totic symmetries of radiative spacetimes is the Poincaré group, but a larger group appears, the BMS
group which contains so-called supertranslations. Additional symmetries, known as superrotations,
also play a role and we shall briefly discuss them too.

This enhanced BMS symmetry group has been the focus of much recent work. In particular, it is re-
lated to the so-called displacement memory effect of General Relativity whose various facets where
independently discovered in the 70s, 80s and 90s and that we will review here, an independent sub-
leading spin memory effect, and to soft graviton theorems that we will not cover in these lectures.

Finally, we will give some comments on the scattering problem in General Relativity, show that the
extended asymptotic group gives conserved quantities once junction conditions are fixed at spatial
infinity. This analysis is still under development by the international community at the time of writ-
ing these lecture notes and brings fascinating insights into the infrared properties of gravity!

3.1 A definition of asymptotic flatness

3.1.1 Asymptotic structure of Minkowski spacetime

The easiest way to introduce the various notions of asymptotic infinities of Minkowski spacetime is
to introduce the Penrose compactification, which conformally maps the spacetime to another non-
physical Lorentzian manifold with finite extent and boundary that is differentiable almost everywhere.
We first recall fundamentals about the conformal compactification of Minkowski spacetime, and then
derive from it a geometric definition of asymptotic flatness. Even though the construction is simple,
it leads to a rich asymptotic structure with a lot of subtetlies on the order of limits and non-geometric
properties at the boundary of the unphysical spacetime, so let’s be careful!

Conformal compactification of Minkowski spacetime

Let us begin with a quick review of the conformal compactification of Minkowski spacetime, written
in spherical coordinates pt, r, x A q as ds2 “ ´dt2 ` dr2 ` r2 γ AB dx A dx B , where γ AB is the unit round
metric on the 2-sphere. To represent the whole spacetime on a finite portion of a sheet of paper, we
declare that every point on this sheet actually represent a 2-sphere (except the space origin which

Lecture 3. Asymptotically flat spacetimes 69

 3.1. A definition of asymptotic flatness

is a spacetime line), so we consider for the moment the 1 ` 1 dimensional metric ´dt2 ` dr2 . To
study causal motion on that spacetime, we introduce null coordinates u “ t ´ r and v “ t ` r such
that the metric becomes ´dudv and radial outgoing (resp. ingoing) null geodesics are simply la-
beled by constant values of u (resp. v). We compactify the support of these coordinates thanks to
a coordinate transformation u “ tan U, v “ tan V, whose action remains diagonal on pu, vq. Now
U, V Ps ´ π{2, π{2r, but the metric ´dUdV{pcos U cos Vq2 still diverges at “infinities” mapped on the
line segments |U| “ π{2 and |V| “ π{2. Let us now perform a Weyl transformation with confor-
mal factor ΩpU, Vq “ cos U cos V to delete the diverging prefactor. The spacetime does not obey
Einstein’s equations anymore and does not faithfully represent the physical distances but the causal
structure of the original metric is preserved. Reintroducing the angular contributions, we are left
with

pu ´ vq2 sin2 pU ´ Vq
ds2 “ ´dudv ` γ AB dx A dx B ùñ Ω2 ds2 “ ´dUdV ` γ AB dx A dx B (3.1)
4 4

and now we can extend the spacetime to |U| “ π{2 and |V| “ π{2. Going back to space- and timelike
coordinates T “ U ` V P r´π, πs and R “ V ´ U P r0, πs, the conformal metric reads as

sin2 R
Ω2 ds2 “ ´dT 2 ` dR2 ` γ AB dx A dx B (3.2)
4

On the sheet of paper, the Minkowski spacetime has been compressed into a triangle with finite
extent. Each point of this triangle, including the null boundaries but excluding the left vertical space-
time origin line and the boundary points i` , i0 , i´ , represents a 2-sphere whose radius varies with
R in a non-monotone way, since it equals sin R{2. At the end of the day, we get thus the following
diagram, on which the geometrical structure of infinity is highlighted.

i`

I`

U
T
r“0

i0
R

I´
i´

Figure 3.1: Penrose conformal diagram of Minkowski spacetime Rp3,1q . Radial null geodesics are represented
in blue, while a radial timelike geodesic is represented in red.

Lecture 3. Asymptotically flat spacetimes 70

 3.1. A definition of asymptotic flatness

We now detail this structure:

1. Past timelike infinity (i´ ) : at pR, Tq “ p0, ´πq, it represents the asymptotic sphere reached when
t Ñ ´8 while keeping r fixed. It is also the starting point of any (maximally extended) timelike
geodesic;

2. Future timelike infinity (i` ) : at pR, Tq “ p0, `πq, it represents the asymptotic sphere reached
when t Ñ `8 while keeping r fixed. It is also the ending point of any (maximally extended)
timelike geodesic;

3. Spacelike infinity (i0 ) : at the right-hand vertex of the triangle pR, Tq “ p`π, 0q, it represents the
asymptotic sphere reached when r Ñ `8 at fixed t;

4. Past null infinity (I ´ ) : the line segment R ` T “ π in the conformal diagram, it represents
the 3-surface formed by the starting points of ingoing null geodesics (the region reached when
r Ñ 8 and u is fixed);

5. Future null infinity (I ` ) : the line segment R ´ T “ π in the conformal diagram is the fu-
ture counterpart of I ´ , and contains the terminal points of outgoing null geodesics (the region
reached when r Ñ `8 and v is fixed).

What is fantastic is that we can easily read off the causal structure of spacetime in a glance! The radial
null geodesics (or equivalently the lightcones of observers) point in the directions `π{4 and ´π{4.
Such a (maximally prolonged) geodesic always starts at some point of I ´ , continues perpendicularly
to the center of spacetime, is “reflected” by the segment r “ 0 and ends its journey at some other
point of I ` . At the null infinities I ˘ the null direction leads to an induced metric ds2 “ 0du2 ` dΩ2
of zero determinant, while the topology is S2 ˆ R.

Singular points

Now, don’t get fooled. We didn’t solve the asymptotic structure of Minkowski. The points/spheres
at i0 , i` , i´ are singular in the conformal description. For example, fields propagating on Minkowski
get multivalued there! More precisely, a propagating field will get a different limit to either of these
points, depending on the order of limits between large distances and future or past. We therefore
need to resolve these singular points.

Resolving the structure around i0 amounts to introduce a foliation of spacetime around i0 with well
chosen 3-surfaces. A useful foliation is the hyperbolic one pτ, ρ, x A q where t “ ρ sinh τ, r “ ρ cosh τ.
By definition, τ is timelike, and ρ is spacelike. The Minkowski metric becomes
´ ¯
ds2 “ dρ2 ` ρ2 ´dτ 2 ` cosh2 τ γ AB dx A dx B (3.3)

outside the origin lightcone (centered at r “ t “ 0) which is the domain where this set of hyperbolic
coordinates is well-defined. It is clear that the spacetime is now foliated by dS3 (hyperboloids) of
constant ρ. Spatial infinity is now defined as ρ Ñ 8. The boundary hyperbolic metric is now a
smooth codimension 1 manifold which resolves i0 . It intersects null infinity at two spheres denoted
` ´
by I´ and I` which are respectively the past limit of future null infinity and the future limit of past null

Lecture 3. Asymptotically flat spacetimes 71

 3.1. A definition of asymptotic flatness

`
infinity. In the hyperbolic description, I´ coincides with the sphere at the future time τ Ñ 8 of the
´
boundary hyperboloid, and I` is the sphere at the past τ Ñ ´8 of the boundary hyperboloid.

One can also blow up the geometry around i` and i´ , by introducing a second set of coordinates
pτ̂, ρ̂, x A q, this time covering the patch inside the origin lightcone, with t “ τ̂ cosh ρ̂, r “ τ̂ sinh ρ̂. The
slicing will be realised on the timelike coordinate τ̂. Each hypersurface is conformal to the Euclidean
version of AdS3 , as we can directly see from the line element
´ ¯
ds2 “ ´dτ̂ 2 ` τ̂ 2 dρ̂2 ` sinh2 ρ̂ γ AB dx A dx B . (3.4)

The three foliations are represented on Figure 3.2. The union of the foliations inside and outside the
origin lightcone provides a manifold whose “corners” are smooth and have a differentiable structure.

i`

Euclidean AdS3 foliation

I`

dS3 foliation

i0 i0

dS3 foliation

I´

Euclidean AdS3 foliation

i´

Figure 3.2: Hyperbolic foliations which blow up the geometry near the singular points i` , i´ , i0 . The origin
lightcone (centered at r “ t “ 0) is drawn in green.

3.1.2 Gravity in Bondi gauge

Let us now introduce gravity. Let gµν be the metric. We would like to define a notion of asymptotic
flatness. There are two ways to do so:
1. Using covariant objects but involving unphysical fields such as a conformal factor (a scalar
field) used to do a Penrose compactification of spacetime;

2. Using an adapted coordinate system and specifying fall-off conditions.

We will follow the second route which more easily allows to analyse the details of the asymptotic
structure.

Lecture 3. Asymptotically flat spacetimes 72

 3.1. A definition of asymptotic flatness

We would like to define asymptotically flat spacetimes which approach a notion of future null infin-
ity I ` . Physically, this describes the so-called “radiation zone” where gravitational waves and other
null wave phenomena leave their imprint on spacetime far from the sources. This problem has been
addressed by Bondi, van der Burg, Metzner and Sachs in the 60ies, which we now review. We con-
sider a family of null hypersurfaces labeled by a constant u coordinate. The normal vector of these
hypersurfaces nµ “ gµν Bν u is null by construction, so we fix guu “ 0. We define angular coordinates
x A “ pθ, φq such that the directional derivative along the normal nµ is zero, nµ Bµ x A “ 0 ñ guA “ 0.
We finally select the radial coordinate r to be the luminosity distance, i.e. we fix Br detpg AB {r2 q “ 0.
The coordinates x µ “ pu, r, x A q so defined are known in the literature as the Bondi-Sachs coordinate
system or Bondi gauge. After lowering the indices, we find grr “ grA “ 0. The 4d line element takes
the form
ds2 “ gµν dx µ dx ν “ guu du2 ` 2gur dudr ` 2guA dudx A ` g AB dx A dx B . (3.5)

We can now define the notion of asymptotic flatness. We would like to obtain Minkowski spacetime
in the limit r Ñ 8 at constant u, x A , which is written in retarded coordinates as ds2 “ ´du2 ´ 2dudr `
r2 γ AB dx A dx B where γ AB is the unit round metric on the 2-sphere. Therefore, we demand

lim guu “ lim gur “ ´1 ; lim guA “ 0 ; lim g AB “ r2 γ AB . (3.6)

rÑ8 rÑ8 rÑ8 rÑ8

Boundary conditions are in fact more restrictive. We need to ensure that we define a phase space with
well defined charges. We cannot be too restrictive, since we need to keep all physical spacetimes, such
as black hole mergers for example. After analysis, it was proposed to consider

guu “ ´1 ` O pr´1 q, gur “ ´1 ` O pr´2 q, guA “ O pr0 q, g AB “ r2 γ AB ` O prq. (3.7)

The class of allowed metrics for these fall-off conditions can be derived:

ds2 “ ´du2 ´ 2dudr ` r2 γ AB dx A dx B pMinkowskiq

2m 2
` du ` rC AB dx A dx B ` D B C AB dudx A
r „ 
1 AB 1 4 1 1
` C AB C dudr ` pNA ` uB A m B q ´ B A pCBC C q dudx A ` γ AB CCD CCD dx A dx B
BC
16r2 r 3 8 4
` pSubleading termsq. (3.8)

Here all indices are raised with γ AB and γ AB C AB “ 0. In Bondi gauge, the metric defines a hierarchy
of several physically relevant fields which we now explicit:

B mpu, x A q is the Bondi mass aspect. It gives the angular density of energy of the spacetime as
measured from a point at I ` labeled by u and in the direction pointed out by the angles
x A . The Bondi mass is obtained after performing an integration of m on the sphere: Mpuq “
2 d Ω mpu, x q. One can show that Bu Mpuq ď 0 for pure gravity or gravity coupled to matter
ű 2 A
S8
obeying the null energy condition. Physically, radiation carried by gravitational waves or null
matter such as electromagnetic fields escapes through I ` and lowers the energy of spacetime
when the retarded time u evolves. At u Ñ ´8, the Bondi mass equates the ADM energy, or
total energy of a Cauchy slice of spacetime.

Lecture 3. Asymptotically flat spacetimes 73

 3.1. A definition of asymptotic flatness

B At the first subleading order, we find another field: the field C AB pu, x A q which is traceless
(γ AB C AB “ 0) and symmetric. It therefore contains two polarization modes. It contains all
the information about the gravitational radiation around I ` . Its retarded time variation is the
Bondi news tensor NAB “ Bu C AB . This is the analog of the Maxwell field for gravitational radia-
tion and its square is proportional to the energy flux across I ` as we will see a bit later.

B At second subleading order, one finds NA pu, x A q the angular momentum aspect. It is closely
related to the angular density of angular momentum with respect to the origin defined as the
zero luminosity distance r “ 0. Its integration on S2 contracted with the generator of rotations
is related to the total angular momentum of the spacetime, evaluated at I ` at retarded time u.

The metric as written so far does not obey Einstein’s equations. One finds two additional constraints
upon pluging this consistent ansatz into Einstein’s equations:

1 A B 1
Bu m “ D D NAB ´ Tuu with Tuu “ NAB N AB ` 4π lim pr2 Tuu
M
q (3.9)
4 8 rÑ8

and
ˆ ˙
1 ´ ¯ 1
Bu NA “ ´ D B DB DC C AC ´ D A DC CBC ` uB A Tuu ´ D B DC NBC ´ TuA (3.10)
4 4
M 1 1 1
with TuA “ 8π lim pr2 TuA q ´ B A pCBC N BC q ` DB pC BC NCA q ´ C AB DC N BC . (3.11)
rÑ8 4 4 2
M the stress tensor of matter, and D is the covariant derivative associated to
Here we denote by Tµν A
γ AB . Of course, since we are performing an expansion close to I ` , the only relevant matter is the
null matter. The gravitational wave contributions naturally add up to the null matter contributions.

Because of these constraints, a generic initial data on I ` is specified by m, C AB and NA at initial

retarded time and NAB at all retarded times, in addition of course with all the subleading fields that
we ignored so far. Given that there is an infinite tower of subleading multipoles at spatial infinity,
there will also be a tower of subleading terms around null infinity.

3.1.3 Initial and late data

` `
Let us now study in more details the initial and final data at I´ and I` , the early and late retarded
times of I ` at first and second subleading order, which are the relevant orders to study mass and
angular momentum conservation. This initial or late data depends upon the class of spacetimes that
we are studying. Let us restrict our analysis to solutions that start from the vacuum in the far past,
and revert to it in the far future (in particular, this assumption rules out black hole formation). Such
spaces have been defined in a rigorous way by Christodoulou and Klainerman [46] and subsequent
authors. They showed that it exists a class of Cauchy data which decays sufficiently fast at spatial
infinity such that the Cauchy problem leads to a smooth geodesically complete solution. In fact they
proved the non-linear stability of Minkowski spacetime. In such analyses, the Bondi news falls off as
´ ¯
NAB “ O |u|´p1`εq , pε ą 0q (3.12)

Lecture 3. Asymptotically flat spacetimes 74

 3.2. Asymptotic symmetries : the BMS 4 group

when u Ñ ˘8, while m and NA remain finite in the two limits. Now, even if black holes form in
the spacetime, we don’t expect that these quantities will behave differently since they don’t emit ra-
diation at early or late retarded times, so we just assume that all asymptotically flat spacetimes obey
these conditions.

`
Since Bu NA Ñ 0 and TuA Ñ 0 when one approaches I´ , the evolution equation for NA gives
ˇ
D B pDB DC C AC ´ D A DC CBC qˇ “0 (3.13)
ˇ
`
I´

which constraints the initial value of C AB . Note that the divergent term uB A p¨ ¨ ¨ q in (3.10) has can-
celled thanks to the fall off condition of the Bondi news. A symmetric traceless tensor on the 2-sphere
like C AB forms a representation of SOp3q and takes the general form

C AB “ ´2D A DB C ` γ AB D2 C ` ε CpA DBq DC Ψ. (3.14)

The first term defines a scalar field Cpu, x A q on the sphere. The second term is parity-violating,
and depends on a pseudo-scalar field, Ψpu, x A q. Now, the equation (3.13) implies that D2 pD2 `
2qΨpu, x A q “ 0 at I´
`
which implies Ψ “ 01 , but Cpu, x A q|I ` “ Cpx A q can be non-vanishing. We
´
will call Cpx A q the supertranslation memory field for reasons that will be clear in a few moments.

A similar construction can be performed at I ´ up to switching u into v and ` into ´, so we are left
with two sets of radiative data on both null infinities:
! ˇ ˇ ˇ )
B Radiation at past null infinity : Cpx A qˇI ´ , mpx A qˇI ´ , NA px A qˇI ´ , NAB pv, x A q, . . . ;
` ` `
! ˇ ˇ ˇ )
B Radiation at future null infinity : Cpx A qˇI ` , mpx A qˇI ` , NA px A qˇI ` , NAB pu, x A q, . . . .
´ ´ ´

These quantities form a set of initial data at null infinity at first and second subleading order in
the luminosity distance expansion since Einstein’s equations and the gauge conditions provide with
all the metric components from this set of data. This closes our discussion on the set of physical
solutions, and we now turn to the asymptotic symmetries of asymptotically flat spacetimes.

3.2 Asymptotic symmetries : the BMS 4 group

Let us discuss the vector fields that preserve the Bondi gauge and the boundary conditions. As
discussed in Lecture 2, such infinitesimal diffeomorphisms are either pure gauge, or belong to the
non-trivial set of asymptotic symmetries. This distinction requires to compute the conserved charges
associated with the infinitesimal diffeomorphisms. At this point, let us just enumerate the vector
fields that change the leading fields in the asymptotic expansion of the metric and therefore are
candidates to be asymptotic symmetries. Preserving the Bondi gauge requires

Lξ grr “ 0, Lξ grA “ 0, Lξ Br detpg AB {r2 q “ 0. (3.15)

1 Strictly speaking it implies that Ψ “ 0 up to the lowest l “ 0, 1 spherical harmonics. However, these harmonics are

exactly zero modes of the differential operator ε CpA DBq DC Ψ defined in C AB . Therefore we can set them to zero.

Lecture 3. Asymptotically flat spacetimes 75

 3.2. Asymptotic symmetries : the BMS 4 group

Preserving the boundary conditions (3.7) further requires

Lξ guu “ O pr´1 q, Lξ gur “ O pr´2 q, Lξ guA “ O pr0 q, Lξ g AB “ O prq. (3.16)

One first solves the constraints (3.15) exactly which allows to express the 4 components of ξ µ in terms
of 4 functions of u, x A . One can then solve (3.16) to reduce these 4 functions to only 3 functions on
the 2-sphere, namely Tpx A q and R A px B q. The resulting vector is
” u ı
ξ T,R “ TpxC q ` D A R A pxC q ` opr0 q Bu (3.17)
„ 2 
A C 1 A C ´1
` R px q ´ D Tpx q ` opr q B A (3.18)
r
„ 
r`u A C 1 A C 0
` ´ D A R px q ` D A D Tpx q ` opr q Br (3.19)
2 2

where Tpx A q is unconstrained. The last constraint in (3.16) imposes that R A obeys the conformal
Killing equation on the 2-sphere,

D A R B ` DB R A “ γ AB DC RC . (3.20)

These vectors are known as the BMS generators, in honour to Bondi, Metzner, Sachs and van der
Burg who were the pioneers in studying the asymptotic behaviour of the gravitational radiation
field. We will see later that the canonical charges associated with these generators are non-trivial and
therefore these diffeomorphisms acquire the name of asymptotic symmetries. Analogously to the 3d
case, we immediately see that the asymptotic algebra is larger than the Poincaré algebra! The explicit
computation of the algebra under the Lie bracket is not quite difficult and ends with
#
T̂ “ R A D A T 1 ` 21 TD A R1A ´ R1A D A T ´ 21 T 1 D A R A ;
ξ T̂,R̂ “ rξ T,R , ξ T1 ,R1 s ùñ (3.21)
R̂ A “ R B DB R1A ´ R1B DB R A .

These relations define the bms4 algebra. Trivial boundary diffeomorphisms T “ R A “ 0 form an ideal
of this algebra. Taking the quotient by this ideal, we are left with the asymptotic algebra of asymp-
totically flat spacetimes compatible with the Bondi-Sachs boundary conditions. The generators can
be divided into 2 categories, the vectors generated by T known as supertranslations and the vectors
generated by R A the Lorentz transformations or their extension: the superrotations.

3.2.1 Supertranslations
The relations (3.21) imply that ξ T,0 ” ξ T form an abelian ideal of the bms4 algebra. They generalize
the translations and receive for these reasons the name of supertranslations

1 1
ξ T “ TpxC qBu ´ D A TpxC qB A ` D A D A TpxC qBr ` ¨ ¨ ¨ (3.22)
r 2

Note that supersymmetry was found 10 years later, so this denomination has nothing to do with su-
persymmetry! Since Tpx A q is a completely arbitrary scalar field on the sphere, the exponentiation of
these vectors gives rise to an abelian subgroup S of the asymptotic symmetry group, which is infinite-

Lecture 3. Asymptotically flat spacetimes 76

 3.2. Asymptotic symmetries : the BMS 4 group

dimensional. S admits one unique normal finite subgroup that reproduces exactly the Poincaré trans-
lations. The associated generators ξ T are built from the first 4 spherical harmonics in the decomposi-
tion of T, which verify D A DB T ` γ AB D2 T “ 0, namely Tpx A q “ a0 Y00 px A q ` am`2 Y1m px A q, aµ P R, m P
t´1, 0, `1u. As an example, Bz amounts to T “ Y11 px A q. Indeed, we have Bz “ cos θBr ´ 1r sin θBθ in
spherical static coordinates pt, r, θ, φq, so in retarded coordinates Bz “ ´ cos θBu ` cos θBr ´ 1r sin θBθ .

I`
Tpθ, φqBu

u “ u0

2 pu q i0
S8 0

I´

Figure 3.3: BMS supertranslations

The associated conserved charges are the supermomenta. The infinitesimal canonical charges are
finite, non-vanishing, not integrable but their non-integrable piece is related to the flux passing
through null infinity,
„ ż  ż
1 ? 1 ?
δQ T “ δ d Ω
2
γTm ` d2 Ω γ N AB δC AB . (3.23)
4πG 32πG

As non-trivial diffeomorphisms, the supertranslations act on the asymptotically flat phase space,
transforming a geometry into another one, physically inequivalent. The Bondi news is transformed
following δT NAB “ L T NAB “ TBu NAB so supertranslations have a relationship with gravitational
radiation as we will see later. Other fields also vary non-trivially under ξ T , for example

LT C AB “ TBu C AB ´ 2D A DB T ` γ AB DC DC T, (3.24)
1 ” AB ı
LT m B “ TBu m B ` N D A DB T ` 2D A N AB DB T . (3.25)
4

From these relations, it is obvious that a supertranslation cannot create inertial mass, or gravita-
tional radiation. Indeed, if we apply a supertranslation on the Minkowski global vacuum m “
NAB “ C AB “ 0, we get L T m “ L T NAB “ 0. The only field that can be shifted is C AB , but
since the Bondi news remain zero, we are left with L T C AB “ ´2D A DB T ` γ AB D2 T. Recalling that
C AB “ ´2D A DB C ` γ AB D2 C in a non-radiative configuration, we deduce that the supertranslation
memory field C is shifted as

δT Cpx A q “ Tpx A q (3.26)

Lecture 3. Asymptotically flat spacetimes 77

 3.2. Asymptotic symmetries : the BMS 4 group

under a supertranslation. This explains half of the name of the field! The fixation of C is equivalent
to a spontaneous breaking of the supertranslation invariance between gravitational vacua, and as a
consequence C is the Goldstone boson which accompanies this breaking. It is noteworthy that the 4
Poincaré translations are not concerned by this breaking, because they consist in the 4 lowest spher-
ical harmonics of Tpx A q, which are annihilated by the differential operator ´2D A DB ` γ AB D2 . So C,
up to the first 4 harmonics, labels the various degeneracies of the gravitational field. Moreover, since
supertranslations commute with the time translation, their associated charges will commute with
the Hamiltonian which means that all these degenerate states have the same energy. This remarkable
feature of asymptotically flat gravity was only found in 2013 by Strominger [47]2 .

3.2.2 Lorentz algebra and its extensions

The second set of generators of the bms4 algebra is constituted by the generators with T “ 0, denoted
by ξ R and given asymptotically by

r`u u
ξ R “ R A pxC qB A ´ D A R A pxC qBr ` D A R A pxC qBu ` ¨ ¨ ¨ (3.27)
2 2

Remember that the fixation of the boundary sphere metric as a part of the boundary conditions im-
poses the conformal Killing equations (3.20). The easiest way to solve the conformal Killing equation
on S2 is to introduce complex stereographic coordinates on the sphere z “ eiφ cotpθ{2q, z̄ “ z˚ . The
unit round metric on S2 is simply the off-diagonal line element ds2S2 “ 4p1 ` zz̄q´2 dzdz̄. The pz, z̄q
component of D A R B ` DB R A “ DC RC γ AB is identically obeyed. The pz, zq and pz̄, z̄q components
simply reduce to the holomorphicity conditions Bz̄ Rz “ 0 and Bz Rz̄ “ 0. So Rz pzq is an holomorphic
function (and Rz̄ pz̄q is its antiholomorphic counterpart) which can be expanded in Laurent’s series,
and appears so as a sum of monomial terms Rz “ zk , k P Z. Considering vk “ zk Bz , we claim that
vk is globally well-defined only when k “ 0, 1, 2. Indeed, when k ă 0, zk is singular at the origin
z “ 0 (south pole θ “ π), and when k ą 2, zk is singular at the point at infinity z “ 8 (the north
pole θ “ 0), since under the transformation z Ñ w “ z´1 , vk becomes ´w2´k Bw . Three globally well-
defined (complex) vectors also come from the antiholomorphic part, so we are left with a subalgebra
of 6 well-defined conformal isometries of the sphere. The real part of this algebra gives 6 asymptotic
Killing vectors. These are nothing but the Lorentz generators. Indeed, the complexification of sop3, 1q
is isomorphic to the direct sum slp2, Rq ‘ slp2, Rq. In stereographic retarded coordinates, a simple
exercise can convince us that the Lorentz sop3, 1q vectors of Minkowski read as
´ u¯ z u u`r u
ξ R P sop3, 1q ðñ ξ R “ 1 ` R Bz ´ D z̄ Dz Rz Bz̄ ´ Dz Rz Br ` Dz Rz Bu ` c.c. (3.28)
2r 2r 2 2

with Rz P t1, z, z2 , i, iz, iz2 u. So the globally defined ξ R are simply the asymptotic Lorentz transforma-
tions. If we discard the singular generators, we finally get after exponentiation the historical form of
the BMS4 group:
BMS4 “ SOp3, 1q ˙ S “ Lorentz ˙ Supertranslations (3.29)
2 Similarly
to the construction of the AdS3 phase space of stationary field configurations with boundary fields turned
on that was described in (2.57) or for the asymptotically flat analogue (2.73), one can construct gravitational vacua and
Schwarzschild black holes that carry a supertranslation field, see [48, 49].

Lecture 3. Asymptotically flat spacetimes 78

 3.2. Asymptotic symmetries : the BMS 4 group

It reproduces the semi-direct structure of the Poincaré group: the Lorentz group acts non-trivially on
the abelian factor S as it does on the global translations. The only difference, and a crucial one, is that
the translational part is enhanced, which implies the degeneracy of the gravitational Poincaré vacua.

One can argue that there is no obvious reason to restrict Rz pzq to be a globally well-defined function
on the sphere. After all, conformal field theories exist on a two-sphere and singular local confor-
mal transformations play an important role. The proposition of Barnich and Troessaert [36] (and
before them of de Boer and Solodukhin [50]) is to allow the full range of k in the Laurent spec-
trum of Rz pzq, generalizing the global conformal transformations to meromorphic superrotations de-
fined from a meromorphic function with a finite set of poles on the sphere. The conformal Killing
vectors are now obeyed locally, except at the poles. For example, if we pick Rz “ pz ´ z0 q´1 , we get
Bz̄ Rz “ 2πδ2 pz ´ z0 q. The boundary conditions are therefore slightly enhanced, since singularities of
the boundary metric γ AB are now allowed on the two-sphere. Physically, one can interpret these sin-
gularities as cosmic strings that reach out to null infinity as recently emphasized by Strominger and
Zhiboedov [51] following an earlier construction of Penrose [52]. Several problems however arise:
if one insists in defining a consistent asymptotic symmetry algebra, the commutators (3.21) imply
that one needs to generalise the supertranslations to supertranslations with poles as well. Then, one can
show that all these singular supertranslations admit infinite conserved charges for the Kerr black hole
[36]. Another issue is that the standard definition of energy is not bounded from below if the phase
space is enhanced consistently with the action of meromorphic superrotations [48]. These problems
can be resolved, but at the cost of renormalizing the symplectic structure [53]. The meromorphic
superrotations are therefore not on the same footing as the supertranslation asymptotic symmetries.

Another asymptotic symmetry group was also proposed [54] with a distinct extension of the Lorentz
group, the entire group of diffeomorphisms on the 2-sphere, DiffpS2 q, leading to the total asymptotic
symmetry group
Extended BMS4 “ DiffpS2 q ˙ Supertranslations (3.30)

The argument is based on the equivalence [55] of the Ward identities of DiffpS2 q symmetry with the
subleading soft graviton theorem [56]. In this case, the leading order metric g AB is allowed to fluc-
tuate, except its determinant. A set of consistent boundary conditions which leads to the definition
of surface charges associated with these symmetries can be worked out [53] but it requires again a
renormalization procedure.

3.2.3 Gravitational memory effects

In this short section, we want to provide evidence that BMS symmetries are more than mathematical
artefacts acting on the phase space, but are intrinsically linked to physical phenomena, known as
gravitational memory effects.

Displacement memory effect

Let us consider a couple of inertial observers (that we will refer to as the "detector") travelling near
future null infinity I ` . The detector is localized in a region with no gravitational radiation or more

Lecture 3. Asymptotically flat spacetimes 79

 3.2. Asymptotic symmetries : the BMS 4 group

generally no null signal at both late and early (retarded) times. Let us declare that the radiation
is turned on at u “ ui and stops at u “ u f . For any value of retarded time excluded from the
interval rui , u f s, the Bondi news tensor and the matter stress-tensor are identically zero by hypothesis.
The detector, which moves on a timelike trajectory in the far region, experiences null radiation only
during the time interval ∆u “ u f ´ ui which we suppose ∆u ! r. We will show that their constant
separation will be permanently shifted due to the null radiation in a precise way. For the setup, see
Figure 3.4. The leading shift is an angular displacement which is not visible on the Penrose diagram.

i`
I`

u “ uf

u “ ui

Ra
di
i0
atio
n

I´
i´

Figure 3.4: Displacement memory effect.

The two inertial observers forming the detector follow a timelike geodesics in the vicinity of I ` ,
characterised by a 4-velocity vµ . Since their trajectories are located near I ` , we can admit that vµ Bµ “
Bu up to subleading corrections (necessary for vµ to verify vµ vµ “ ´1 ` O pr´2 q in Bondi gauge). The
separation between the two geodesic trajectories is given by the deviation vector sµ which is solution
of the famous equation of geodesic deviation :

µ
∇v ∇v sµ “ R αβγ V
α
V β sγ (3.31)

where ∇v “ vµ ∇µ is the directional derivative along vµ . We suppose that both detectors move on the
same celestial sphere so sr “ 0. We get

1 2
r2 γ AB Bu2 s B “ RuAuB s B ðñ γ AB Bu2 s B “ B C AB s B (3.32)
2r u

after using RuAuB “ ´ 2r Bu2 C AB ` O pr0 q as one can check using the metric in Bondi gauge. Let us
introduce the perturbation of the deviation vector as s B “ siB ` r´1 ssub
B . Integrating once in u we

obtain an integration constant, which we set to zero assuming that the velocity is zero if the news is
zero. Integrating once more, we get

1
γ AB ∆ssub
B
“ ∆C AB siB ` O pr´2 q . (3.33)
2r

Lecture 3. Asymptotically flat spacetimes 80

 3.2. Asymptotic symmetries : the BMS 4 group

A “ s A pu “ u q ´ s A pu “ u q and ∆C
where ∆ssub sub f sub i AB “ C AB pu “ u f q ´ C AB pu “ ui q. Therefore, if
the field ∆C AB is non-zero during the time interval rui , u f s, the deviation between the two inertial
observers of the detector will be irreversibly shifted : this is the displacement memory effect ! This is
a DC effect. Flashing a light between these two observers will measure the shift, which is therefore
detectable. This was first observed by Zeldovich and Polnarev in 1974 [57].

Let us now study the causes of such a displacement. Any process that can change the tensor C AB will
lead to the displacement. We integrate the variation of the Bondi mass aspect (3.9) between ui and u f
to get simply ż uf
1
∆m “ D A D B ∆C AB ´ du Tuu . (3.34)
4 ui

If the spacetime is stationary before ui and after u f :

∆C AB “ C AB, f ´ C AB,i “ ´2D A DB ∆C ` γ AB D2 ∆C. (3.35)

Injecting this in the previous relation, we obtain that the shift of supertranslation memory field ∆C
obeys a quartic elliptic equation which is sourced by 3 qualitatively distinct terms :
ż uf „ 
1 1
´ pD2 ` 2qD2 ∆C “ ∆m ` du AB 2 M
NAB N ` 4π lim pr Tuu q . (3.36)
4 ui 8 rÑ8

The displacement memory detector will trigger for each of the following causes:

B If the Bondi mass aspect varies between ui and u f . This is sometimes called ordinary memory,
but it is not ordinary to our common sense! For example, a single massive body containing
a string that suddently separate into two parts due to a trigger will modify the Bondi mass
aspect m because the mass will suddenly possess a strong dipolar component. What Einstein
gravity tells us is that a signal is sent at null infinity with that information, and the memory
effect follows.

B If null matter reaches I ` between ui and u f . This is sometimes called the null memory effect. For
example, electromagnetic radiation causes the displacement memory effect.

B And finally, if gravitational waves pass through I ` between ui and u f . This is sometimes
called the Christodoulou effect [58, 59], even though one could argue that it was found earlier by
Blanchet and Damour in the post-Newtonian formalism [60, 61].

The displacement memory effect was never observed at the time of writing, but it may be observed
in the close future by gravitational wave detectors [62] or pulsar timing arrays.

Spin memory effect

Remember the intimate relationship between the BMS supertranslation charge (the local energy on
the celestial sphere or Bondi mass) and the displacement memory effect: the change of Bondi mass
between initial and final stationary states sources the displacement memory effect – see equation
(3.36). Given the existence of a local angular momentum on the celestial sphere, the Bondi angular

Lecture 3. Asymptotically flat spacetimes 81

 3.2. Asymptotic symmetries : the BMS 4 group

momentum aspect, and the associated DiffpS2 q extended symmetry, Pasterski, Strominger and Zhi-
boedov raised and proposed an answer in 2015 [63] to the following question: is there a gravitational
effect sourced by the Bondi angular momentum aspect? Yes, they called it the spin memory effect.

The starting point is that in the decomposition of C AB (3.14), Ψ is zero at retarded times without
şu
radiation, but uif du Ψ can still be measured3 . Note that the polarisation mode Ψ is a pseudo-scalar
which flips under parity. In order to measure it, let’s consider the following thought experiment.
We consider light rays which orbit a circle C of radius L (like a toroid optic fiber, or a circular wall
of mirrors) in the vicinity of I ` . The photons are allowed to travel clockwise or counterclockwise
in this system. Again we assume that no null radiation passes through the system at early and late
times. The system experiences the passage of radiation which carries non-trivial angular momentum
(NA ‰ 0) during a finite range of retarded time rui , u f s, as before. Let us denote by ∆u the relative
time delay between the clockwise and counterclockwise light rays. For u ă ui , we set the system
such that ∆u “ 0. The spin memory effect resides in the fact the ∆u is no more zero at late time after
the null radiation has passed. After a computation explained in [63], one gets
ż ż ´
1 ¯
∆u “ du D A C AB dx B . (3.37)
2πL C

This integral is independent of C and therefore only depends upon Ψ. In order to see it, compute
D A C AB using the decomposition (3.14). The terms involving C lead to D A C AB “ ´2DB pD2 ` 2qC
which is a total derivative on the circle, and therefore leads to a vanishing integral on the closed cir-
şu
cle. Therefore, ∆u is only function of uif du Ψ.

The sources of spin memory can be obtained by integrating the conservation law of the angular mo-
mentum density (3.10) between ui and u f . The definite integral of the first term on the right-hand side
şu
does not depend upon C as one can check. It leads to some differential operator acting on uif du Ψ,
şu
symbolically denoted by D uif du Ψ. The rest of the right-hand side brings a total contribution of the
şu
null matter and gravitational wave sources: ´ uif du T̂uA . We therefore find
ż uf ż uf
D du Ψ “ ∆NA ` du T̂uA . (3.38)
ui ui

There is a decomposition of the origin of the spin memory effect into 3 qualitative classes similar to
the displacement memory effect: a change of angular momentum aspect, angular momentum flux
from null matter and angular momentum flux from gravitational waves.

The spin memory effect is subleading with respect to the displacement memory effect and is probably
not observable by the current 2G technology of gravitational wave detectors [64].

3 The attentive reader might already notice that such an observable is not clearly the difference between a quantity
şu
in the initial and final state. Instead uif du Ψ is non-local in retarded time! Therefore, though it bears analogy with the
displacement memory effect, the spin memory effect is not (yet?) proven to be a memory effect at the first place! For a
discussion of memory effects associated with DiffpS2 q symmetries, see [53].

Lecture 3. Asymptotically flat spacetimes 82

 3.3. Scattering problem and junction conditions

3.3 Scattering problem and junction conditions

We close this lecture about asymptotically flat spacetimes by a little glance at the very fundamental
scattering problem in General Relativity. Let us first consider scattering of null radiation (null matter
fields or gravitational waves) around Minkowski spacetime. We are interested in relating the out
states at I ` to the in states at I ´ . Are there universal constraints among these S -matrix elements?
This is the question that Strominger asked and answered in his 2013 paper [47].

So far we found an asymptotic symmetry group at I ` , the BMS group (with or with extension de-
pending upon the boundary conditions). From now on we will add a superscript BMS` to our
notation to remind us where this group is defined. It is not difficult to convince yourself that exactly
the same construction can be performed at I ´ upon switching retarded to advanced coordinates.
There is therefore a BMS´ asymptotic symmetry group. Now, acting with symmetries on the initial
state should be reflected on the final state. There cannot be independent symmetries acting on both
the initial and final states, otherwise the scattering problem would not be defined! The BMS group
is therefore defined as a diagonal subgroup of the product BMS` ˆ BMS´ .

But which diagonal subgroup? Or in other words, how to identify the generators at I ` with the
ones at I ´ ? The crucial clue is that propagating fields around Minkowski spacetime obey universal
antipodal matching conditions at spatial infinity. For each bulk field Φi , one can define its limit at I `
` ´
and then take u Ñ ´8 which defines the field at I´ . Similarly, one can define the field at I` . An
antipodal matching condition would mean that

Φi pθ, φq|I ` “ Φi pπ ´ θ, φ ` πq|I ´ (3.39)

´ `

Now it turns out that the electromagnetic Liénard-Wierchert field describing the retarded electro-
magnetic field of a uniformly moving source obeys these antipodal matching conditions. The metric
of the boosted Kerr black hole is also expected to obey these conditions. Moreover, these conditions
are CPT invariant and Lorentz invariant.

A complementary perspective comes from perturbative quantum gravity. The S -matrix should be
BMS invariant in the sense that acting with BMS charges on the “in” state and “out” states com-
mute: Q` ´
T,R S “ S Q T,R . These relationships are the Ward identities of supertranslations and Lorentz
transformations or their superrotation extension. Now, as shown in 2014 [65] it turns out that the
supertranslation Ward identities are identical, after a change of notation, to Weinberg’s leading soft
graviton theorems derived in 1965 [66]. At subleading order, it was also shown in 2014 [55] that
the DiffpS2 q superrotation Ward identities are identical, after rewriting, to the newly found Cachazo-
Strominger’s subleading soft graviton theorems [56].

The antipodal matching conditions are compatible with the soft theorems, which validates their
range of applicability around Minkowski spacetime. Yet, it has not been derived whether or not
the antipodal map is generally valid for any subleading field in the asymptotic expansion close to
null infinity, and for spacetimes with other causal structures, such as spacetimes containing a black

Lecture 3. Asymptotically flat spacetimes 83

 3.3. Scattering problem and junction conditions

hole formed from collapse. For the leading order fields, consistent boundary conditions exist which
admit antipodal matching boundary conditions both in 4d Einstein gravity [67, 68] and 3d Einstein
gravity [45].

Assuming that the antipodal matching conditions hold in generality directly leads to conservation
laws. Indeed, physical quantities depend upon the fields, so if all relevant fields are antipodally
` ´
matched, the conserved charges at I´ and I` (either supertranslations or superrotations) are related
by the antipodal map symbolically denoted by AntiPodMapp˝q,

Q|I ` “ AntiPodMappQq|I ´ (3.40)

´ `

Using the conservation laws of these charges, of the form Bu Q “ J u and Bv Q “ J v , we deduce by
integration along u and v the conservation laws
ż ˆż ˙
u v
du J ` Q|I ` “ AntiPodMap dv J ` Q|I ´ . (3.41)
` ´

These (supertranslation and superrotation) conservations laws are the conservation of energy and an-
gular momentum at each angle on the 2-sphere.

This topic is still under investigation, especially in relationship with black holes since it has been
conjectured to be relevant if not crucial to resolve the black hole information paradox [69, 70] (see
however [71, 72]) !

References
The BMS group of supertranslations and Lorentz transformations of 4-dimensional asymptotically
flat spacetimes was described in the founding papers of 1962 [73, 74]. The proposed extension by
meromorphic superrotations and supertranslations was studied in [50, 75, 36]. The proposed exten-
sion to diffeomorphisms on the 2-sphere was derived in [55, 54].

The displacement memory effect was independently discovered from the three qualitatively distinct
sources of displacement memory: the change of Bondi mass aspect [57], gravitational wave flux
[60, 61, 58, 59] and null matter radiation [76]. The spin memory effect was described in [63]. Experi-
mental prospects include [62, 64].

The renewal of the topic of BMS symmetries is largely due to the work on BMS invariance of scat-
tering [47] and the triangle relationship between BMS supertranslation symmetries, Weinberg’ soft
graviton theorem [65] and the displacement memory effect [77].

The latest proposal to solve the black hole information paradox using BMS soft hair [70] slightly dif-
fers from the first proposal [69] (briefly mentioned in [77]).

These lectures are also based on several reviews and papers [78, 79, 48, 49, 80].

Lecture 3. Asymptotically flat spacetimes 84

 Lecture 4

Rotating black holes

For this last lecture, we will focus on particular properties of astrophysically realistic 4d black holes,
which are rotating and uncharged. We will concentrate on stationary asymptotically flat black holes,
that admit by definition an asymptotically timelike Killing vector Bt . The rigidity theorem due to
Hawking states that “stationarity implies axisymmetry” so these solutions also possess an additional
axial Killing vector Bφ . Such black holes have necessarily spherical topology. They contain a sin-
gularity hidden by an event horizon which is also a Killing horizon. They are exactly described by
the Kerr black hole solution found in 1963. Quite remarkably, the Kerr black hole solution is also a
dynamical attractor: it is the final state of collapse of matter. It is therefore one of the most important
analytical solutions of Einstein’s equations due to its universality! First of all, we will review the
main features of this rich spacetime, before entering in more advanced considerations.

In a second step, we will study the maximally spinning limit of the Kerr solution, the extremal Kerr
black hole. The extremal Kerr black hole lies at the frontier between the regular Kerr black holes
and naked singularities which are thought to be unphysical (it would be proven unphysical if one
could prove Penrose’s cosmic censorship). The third law of black hole mechanics states that it is im-
possible to spin up a black hole beyond the maximal limit because the extremal black hole has zero
Hawking temperature and no physical process can reach absolute zero temperature. If one attempts
to send finely-tuned particles or waves into a near-extremal black hole in order to further approach
extremality, one realizes that there is a smaller and smaller window of parameters that allows one
to do so. On the other hand, if one starts with an extremal black hole, one can simply throw in a
massive particle to make the black hole non-extremal. In summary, extremal black holes are finely
tuned and (classically) unreachable black holes.

The extremal Kerr black hole is a very interesting solution because it admits special near horizon lim-
its. Such limits admit enhanced conformal symmetry that shares features with anti-de Sitter space-
times where holography and therefore quantum gravity is most understood. The attempts (with
successes and failures) to describe the extremal Kerr black hole with holographic techniques is called
the Kerr/CFT correspondence and will be briefly reviewed here.

The final part of these lectures will be devoted to the analysis of gravitational perturbations around
Kerr geometries. The Kerr black hole is currently under experimental tests by the LIGO/Virgo grav-
itational wave detectors. Indeed, the final stages of black hole mergers consist in a quasi-normal
mode ringing of the resulting black hole which is well-described by perturbation theory around the
Kerr black hole. Since the Kerr black hole only depends upon 2 parameters, namely the mass and
angular momentum, the resonance frequencies (the quasi-normal modes) of the black hole are char-
acteric signatures of Einstein gravity. The emerging experimental science of black hole spectroscopy
will soon test the limits of Einstein gravity and look for possible deviations!

Lecture 4. Rotating black holes 85

 4.1. The Kerr solution — Review of the main features

4.1 The Kerr solution — Review of the main features

4.1.1 Metric in Boyer-Lindquist coordinates

The Kerr metric describes the most general regular asymptotically flat, stationary and axisymmetric
spacetime in 4d Einstein gravity. In Boyer-Lindquist coordinates pt, r, θ, φq the metric reads as

∆` dr2 sin2 θ ` 2
ˆ ˙
˘2 ˘2
2
ds “ ´ dt ´ a sin2 θdφ ` Σ 2
` dθ ` pr ` a2 qdφ ´ adt (4.1)
Σ ∆ Σ

where
∆prq fi r2 ´ 2GMr ` a2 ,
(4.2)
Σpr, θq fi r2 ` a2 cos2 θ.
In the following, we will set G “ 1. The metric possesses 2 Killing vectors, Bt and Bφ , because its com-
ponents do not depend on these coordinates. Contrary to the Schwarzschild black holes, the Kerr
solution is not static, since the time reversal transformation t Ñ ´t does not preserve the cross-term
dtdφ. The mixed term between t and φ has the effect of “dragging” the spacetime along with the
rotating body, just as water being dragged along by the surface of an immersed spinning ball. This
effect of “frame dragging” is very important in the discussion of the physical properties of Kerr black
holes, and it can be shown that it is responsible of the gyroscopic precession of test-rotating bodies,
called the “Lense-Thirring effect”.

Kerr black holes form a 2-parameter family of solutions labeled by M and a. M is the surface charge
associated to Bt when evaluated on the sphere at infinity, so it’s clearly the total energy (or total mass
since we are in the rest frame) of the black hole. The second Killing vector ´Bφ gives rise to another
conserved surface charge which is J “ aM, the angular momentum, so the parameter a “ J{M is the
specific angular momentum.

4.1.2 Killing horizon and black hole thermodynamics

The metric admits a physical singularity at Σ “ 0 because the curvature invariant Rµναβ Rµναβ blows
up at that locus. For a ‰ 0, it can be shown to be a ring, known as the ring singularity of Kerr.
Assuming that Penrose’s cosmic censorship principle is true, this singularity must be shielded by a
event horizon.

The metric (4.1) admits several coordinate singularities. First at θ “ 0, π one has the familiar polar
coordinate singularities which can be removed by switching to local cartesian coordinates. Second,
at ∆prq “ 0 the first term in (4.1) vanishes while the dr2 term blows up. This occurs for 2 values of r,
?
denoted by r˘ “ M ˘ M2 ´ a2 , since ∆prq is a quadratic function of r. The outer value r “ r` is the
event horizon H` of the black hole, which we will prove later on. The existence of the event horizon
bounds the angular momentum as
´ M ď a ď M. (4.3)

If a ą M, there is no event horizon and the curvature singularity is naked, which is unphysical, so
we assume that the bound (4.3) holds in Nature.

Lecture 4. Rotating black holes 86

 4.1. The Kerr solution — Review of the main features

The horizon r “ r` is in fact a Killing horizon: its null rays are generated by a Killing vector, ξ “
Bt ` Ω H Bφ . One can check that Ω H “ a
2Mr` is such that ξ 2 “ 0 at H` and ∇ξ ξ µ “ κξ µ at H` where κ is
the surface gravity of the black hole (with respect to the unit normalized generator such that ξ 2 “ ´1
at spatial infinity). Thanks to the groundbreaking Hawking result of 1974 [81], the surface gravity
determines the Hawking temperature of the black hole to be

κ pr` ´ Mq
TH “ “ . (4.4)
2π 4πMr`

Due to the rotation, the Kerr black hole possesses a second Killing horizon, which is called the inner
horizon defined at the radius r´ ď r` . This horizon is a Cauchy horizon as we will discuss below.
The limiting case a “ M is called the extremal case. The extremal black hole only possesses one hori-
zon and the Hawking temperature is identically zero. The third law of thermodynamics prevents
a thermal system to reach zero temperature. Anagolously, no physical process exists that allows to
reach an exactly extremal black hole.

The area of the outer horizon divided by 4 (4Gh̄ in MKSA units) has the interpretation of microscopic
entropy of the black hole thanks to the famous 1973 Bardeen, Carter and Hawking result [82] on black
hole thermodynamics combined with Hawking’s 1974 result [81] on the identification of black hole
temperature, with earlier insights from Bekenstein [83]. The Bekenstein-Hawking entropy of the Kerr
black hole is
AH
SBH “ “ 2πMr` . (4.5)
4
For an extremal black hole, r` “ M “ a and Sext 2
BH “ 2πM “ 2π J.

One of the main challenges of a theory of quantum gravity theory is to account for this entropy! It
obeys the first law of black hole thermodynamics

TH δSBH “ δM ´ Ω H δJ (4.6)

and the second law of thermodynamics which states that the entropy of the outer universe plus the
black hole entropy always increases.

4.1.3 Ergoregion
The concept of energy as measured by an asymptotic observer is the conserved quantity associated
with Bt . An interesting feature of the Kerr solution is that Bt becomes spacelike beyond a specific
surface known as the ergosphere. The metric component gtt “ ´p∆ ´ a2 sin2 θq{Σ vanishes at the
?
ergosphere radius rerg “ M ` M2 ´ a2 cos2 θ. The ergosphere lies outside the event horizon and
therefore delimits a region called the ergoregion which is depicted in Figure 4.1. Since the energy
becomes qualitatively similar to a (spacelike) momentum in that region from the point of view of
an asymptotic observer, the energy can take a negative sign locally while remaining globally posi-
tive. This leads to many important phenomena of energy extraction from a Kerr black hole, either
by particles (the Penrose process), waves (the superradiant effect [84]) or magnetic accretion disks (the
Blandford-Znajek process [85]). . .

Lecture 4. Rotating black holes 87

 4.1. The Kerr solution — Review of the main features

ere
ph izo
n
or

os
Erg

h
θ

nt
Eve
r
Equatorial
plane

Figure 4.1: Ergoregion of the Kerr black hole (t fixed).

4.1.4 Event horizon and Cauchy horizon

In order to clarify the notions of event horizon and Cauchy horizon we need to clarify the causal
structure of the Kerr metric. For that purpose, let us draw the corresponding Penrose diagrams. We
only consider the non-extremal case r´ ă r` . Extremal diagrams are formally obtained by removing
the inner parts between the horizons r “ r` and r “ r´ and gluing them. We refer to the excellent
Carter lectures [86] for details. The maximal extension of the non-extremal Kerr black hole is obtained
by gluing up an infinite sequence of two Penrose diamonds that are depicted in Figures 4.2a and 4.2b.
The orange parts with same opacity are overlapping and must be glued.
r“

r“
r´

r`
r“

r“
r´

r`

i` i` i´ i´
r“

r“
r`

r´
`

`
I

I
Ring singularity

Ring singularity
`

`
r“

r“
r`

r´
I

i0 r “ rerg r “ rerg i0 i0 i0
r“

r“
r`

r´
´

´
I

I
´

´
r“

r“
r`

r´
I

i´ i´ i` i`
r“

r“
r´

r`
r“

r“
r´

r`

(a) Penrose diagram for r ą r´ (b) Penrose diagram for r ă r´

Figure 4.2: Penrose diagrams for Kerr spacetime.

Lecture 4. Rotating black holes 88

 4.1. The Kerr solution — Review of the main features

Note that these diagrams share some features with the non-extremal BTZ ones. The major difference
concerns the lateral shapes. For BTZ, we get straight vertical lines, which reflect the AdS3 asymptotic
geometry of BTZ, while for Kerr we get diamonds, which reflect the asymptotic flatness of Kerr!

The physical meaning of the inner horizon r “ r´ becomes clear on the Penrose conformal diagram
4.2a. We will actually show that the surface is a Cauchy horizon. Physics usually requires the existence
of a Cauchy surface Σ, which is an hypersurface that all light rays and massive particle trajectories
intersect exactly once. Then hyperbolic field equations will determine unambiguously the past and
future behaviour of their solutions when initial data is taken on Σ. Now, the point is that if the region
beyond r “ r´ is taken into account, no Cauchy hypersurface exists. Indeed, if we pick a point P1 at
radius larger than the inner horizon r “ r´ , the past lightcone of P1 entirely crosses Σ (see Figure 4.3),
so the physics at P1 is entirely determined by the initial data given on Σ. Let us now take a point P2
beyond r “ r´ . After drawing the past lightcone in this point, we see that it also requires information
from the other asymptotic region I ´ and so the initial data on Σ is not sufficient to determine the
event at P2 . This prevents the existence of a single Cauchy surface. The surface beyond which the
events are no more causally determined by Σ is the Cauchy horizon!
C
r´

au
r“

ch

P2
y
ho
riz

´
I
´

on

P1
i` i`
Ev

r`
e
`

r“
nt

I
`
I

ho
riz
on

i0 i0
Cauchy surface Σ

Figure 4.3: The inner horizon is a Cauchy horizon.

In fact, numerical analysis shows that any perturbation falling into a Kerr black hole leads to a phys-
ical curvature singularity around the Cauchy horizon [87]. It is therefore safe to assume that the
region beyond the inner horizon is unphysical. Science-fiction movies like to exploit the wormhole
which connects the asymptotic region I ´ of Figure 4.2b to the asymptotic region I ` of Figure 4.2b.
More precisely, one can imagine a spaceship falling into the black hole (crossing r “ r` ), passing
the Cauchy horizon, avoiding the ring singularity and emerging from the asymptotic flat region on
the right-hand side of Figure 4.2b. This scenario is however unphysical because the classical journey
will end at the Cauchy horizon where curvature blows up. Only quantum gravity can tell us what
happens then...

Lecture 4. Rotating black holes 89

 4.1. The Kerr solution — Review of the main features

4.1.5 Linear stability of black holes

The question of stability of a given spacetime is crucial if one looks at the relevant solutions that one
could observe in Nature. Indeed, if we consider a spacetime pM, gq whose metric tensor is perturbed
gµν Ñ gµν ` hµν , and if this spacetime is totally disrupted in the long run even by small perturbations,
it cannot be a long-lived solution of gravity which will be actually observed!

The most simple case to treat is obviously the flat spacetime, whose non-linear stability has been
proven by Christodoulou and Klainerman in the nineties [46] (see also the Lecture 3 on asymptoti-
cally flat spacetimes). By non-linear stability, we mean that, under some weak assumptions, any non-
linear perturbation of the Minkowski metric will lead to a spacetime that asymptotes to Minkowski
at late times. All perturbations will decay. For some large perturbations, the gravitational field may
become so important that an asymptotically flat black hole appears. There is an intermediate, critical
behavior, where the perturbation is at the onset of black hole formation known as the Choptuik criti-
cal collapse [88]. When a black hole has formed, the stability question can also be addressed but this
time in terms of perturbations of a stationary black hole.

The linear stability of the static Schwarzschild solution has been proven [89, 90]. It means that if one
imposes a small perturbation gµν Ñ gµν ` hµν where hµν obeys the linearised equations of motion, all
frequencies of the Fourier modes of hµν will have a negative imaginary part: all perturbations will
decay. A similar statement has been obtained for non-extremal Kerr black holes [91, 92]: it implies
that the Kerr black hole is relevant to describe the late stages of gravitational collapse.

Now, something special occurs for extremal black holes. The key point is that the event horizon co-
incides with the Cauchy horizon. We just saw that a non-extremal black hole has a distinct event
horizon located at r “ r` and Cauchy horizon located at r “ r´ . For an extremal black hole, the Pen-
rose diagram collapses such that r “ r` “ r´ is at the same time the event horizon and the Cauchy
horizon. We have discussed that the Cauchy horizon is linearly unstable to gravitational perturba-
tions. We can therefore expect that the event horizon will admit unstable modes along the horizon.

Indeed, it has been recently shown that rotating and charged black holes in their extremal regime are
linearly unstable under respectively gravitational or electromagnetic perturbations [93] ! A generic
perturbation of an extremal black hole will produce a non-extremal black hole, and all perturba-
tions will decay consistently with the linear stability results. Some perturbations keep the black hole
extremal, and those perturbations are the ones of interest. If one fine-tunes the perturbation to be ex-
tremal and leading to an unstable mode, the perturbation leads to infinite gradients at the location of
the event horizon, but no curvature singularity appears. The non-linear final state is not a stationary
extremal black hole, but a non-stationary extremal black hole [94].

Lecture 4. Rotating black holes 90

 4.2. Extremal rotating black holes

4.2 Extremal rotating black holes

There are two very different motivations to study the extremal Kerr black hole:

B From the point of view of quantum gravity. The extremal Kerr black hole is an intermediate case
between the physical but hard to study non-extremal Kerr black hole and unphysical but easier
to comprehend supersymmetric (and therefore extremal) black holes in string theory;

B From the point of view of observational science. This black hole is a limit for near-extremal Kerr
black holes with very specific observational signatures.

The first motivation led to the Kerr/CFT correspondence [95], which is an attempt to relate the extremal
Kerr black hole with a dual CFT, with partial successes and failures, see the review [96]. The second
motivation led to several analytical analyses of physical processes around black holes and identifica-
tions of “smoking gun” observational signatures for either electromagnetic or gravitational wave as-
tronomy. This topic is under active development at the time of writing these lectures (see [97, 98, 99]).

As a physicist, one should ask: are there nearly extremal black holes in Nature? By the third law of
black hole thermodynamics, we can never reach extremality. In 1974, Thorne gave a precise bound
on how high the spin can be using a specific thin disk accretion disk model [100]. The disk can spin
up the black hole up to
J ď 0.998 M2 (4.7)

where the absorption cross-section of retrograde photons emitted from the disk exceeds the cross-
section of prograde photons. If the black hole spins faster, it will accrete too many photons with
retrograde orbital motion which will spin down the black hole. However, this bound is only valid
for one specific model of disk accretion and it can be beaten!

Astronomical observations give encouraging results: the stellar mass black holes known as GRS
1905+105 and Cygnus X-1 have been claimed to admit J{M2 ą 98%. Also some supermassive black
holes at the center of galaxies MGC-6-30-15, or 1H 0707-495 have also been claimed to have a spin
ratio higher than 98%...

In theory, it is practical to use the parameter λ defined as

c
a2
λ“ 1´ (4.8)
M2

in order to measure how close one is from extremality. Schwarzschild has λ “ 1 and extremal Kerr
has λ “ 0. What we are after is near extremal black holes with λ “ 10´3 or even 10´6 where the near-
extremal features that we are going to describe really express themselves in physical phenomena.
A famous science-fiction example of nearly extremal Kerr black hole is “Gargantua” in Interstellar
which, according to Thorne [101], needs to have λ ă 10´7 in order to be consistent with key features
of the movie script. Let’s now go to the physics.

Lecture 4. Rotating black holes 91

 4.2. Extremal rotating black holes

4.2.1 Near horizon geometries

Let us use from now on the two parameters M, λ to denote a generic Kerr black hole. The event
horizon and Cauchy horizon lie at radii r˘ “ Mp1 ˘ λq. What happens in the limit λ Ñ 0? From the
point of view of the asymptotic observer, the geometry becomes the one of extremal Kerr. But from
the point of view of an observer close to the black hole horizon, something very different happens.

We start from the Boyer-Lindquist patch pt, r, θ, φq. For an observer close to the horizon, we switch to
a coordinate system corotating with the black hole by taking

t
Φ “ φ ´ Ωext
H t “ φ´ ` Opλq (4.9)
2M

up to terms small in the near-extremal limit λ ! 1. Moreover, we need to resolve the radius and time
which are not good coordinates close to the horizon so we define
#
t p
T “ 2Mκ λ ;
r´r` ´p (4.10)
R “ κ M λ ;

where 0 ă p ă 1 and κ is any real normalization. In the limit λ Ñ 0, we can write the Kerr metric as

dR2
„ 
ds “ 2M Γpθq ´R dT ` 2 ` dθ ` Λ pθqpdΦ ` RdTq ` O pλ p q
2 2 2 2 2 2 2
(4.11)
R

where Γpθq “ p1 ` cos2 θq{2 and Λpθq “ 2 sin θ{p1 ` cos2 θq are two geometrical factors. This metric is
known as the near horizon extremal Kerr geometry or NHEK geometry.

Choosing instead p “ 1, we zoom even closer to the horizon and the near-horizon limit changes to

dR2
„ 
ds2 “ 2M2 Γpθq ´RpR ` 2κqdT 2 ` ` dθ 2 ` Λ2 pθqpdΦ ` pR ` κqdTq2 ` O pλq. (4.12)
RpR ` 2κq

This is the so-called near-NHEK metric [102, 103]. For p ą 1 or p ă 0, the limit is not well-defined, so
that’s all we can do.

There are therefore 3 different extremal limits from Kerr depending on the type of observer that one
is considering! Moreover, these regions formally decouple in the extremal limit, but they never ex-
actly decouple for a nearly extremal black hole. They are summarized in Figure 4.4. Already an
interesting consequence of the time definition (4.10) is that the relative gravitational redshift between
the (near-)NHEK and extremal Kerr geometries is formally infinite in the limit λ Ñ 0. This leads
to very interesting phenomena of high energy collisions in the near-horizon region between near-
horizon waves and exterior waves entering the near-horizon region which are highly blueshifted in
the near-horizon frame [104].

Lecture 4. Rotating black holes 92

 4.2. Extremal rotating black holes

Extremal Kerr
Horizon

Near NHEK NHEK

Figure 4.4: The three asymptotically decouped regions of Kerr in the near-extremal regime.

Let’s now describe the geometry of (near-)NHEK. In fact, NHEK and near-NHEK are diffeomorphic
to each other so we only need to discuss the NHEK geometry. The explicit diffeomorphism is

1 κT a
R1 “ e RpR ` 2κq, (4.13)
κ
R`κ
T 1 “ ´e´κT a , (4.14)
RpR ` 2κq
1 R
Φ1 “ Φ ´ log (4.15)
2 R ` 2κ

as you can check! The NHEK geometry contains the line element ´R2 dT 2 ` dR2 {R2 which is exactly
AdS2 spacetime in Poincaré coordinates. AdS2 spacetime contains 3 Killing vectors which form a
SLp2, Rq algebra. This symmetry algebra is in fact lifted to a symmetry algebra of the entire NHEK
geometry, together with the Up1q with BΦ . The 4 Killing vectors are BT and BΦ together with the scale
transformation
ξ 3 “ TBT ´ RBR (4.16)

and another non-trivial exact symmetry

T2
ˆ ˙
1 1
ξ4 “ ` BT ´ TRBR ´ BΦ . (4.17)
2R2 2 R

The careful reader would have already derived the existence of the symmetry ξ 3 : rescaling T by κ
and R by κ ´1 does not change the metric as we saw in (4.10), which is the finite version of the Killing
symmetry! Since BΦ commutes with the 3 other Killing vectors, the symmetry algebra of the NHEK

Lecture 4. Rotating black holes 93

 4.2. Extremal rotating black holes

geometry is

SLp2, Rq ˆ Up1q. (4.18)

The existence of conformal SLp2, Rq symmetry in the near-horizon region of a nearly extremal Kerr
black hole implies that physics in the near-horizon region can be described in the language of critical
phenomena. This motivated the recent exploration of magnetospheres, electromagnetic emission,
accretion and gravitational wave emission from the NHEK region which all carry critical behavior
caused by approximate conformal invariance.

4.2.2 Extremal BTZ black holes and their dual CFT description
In 3d gravity, we have shown during the second lecture that it exists boundary conditions (Brown-
Henneaux [23]) such that asymptotically AdS3 spacetimes form a phase space whose asymptotic
symmetry group is the direct product of two copies of the Virasoro group (which is the infinite-
dimensional 2d conformal group). This shows that quantum gravity with Brown-Henneaux bound-
ary conditions, if it exists, is a conformal field theory, or in other words can be described in dual
terms in the language of a 2d CFT. (The existence of quantum gravity may require an embedding in
string theory.)

In particular, extremal BTZ black holes can be understood as particular states of the dual CFT, and
their entropy can be understood from a microscopic counting in the dual CFT. Since this situation is
very well understood, we will start by reviewing these results as a starter for describing the attempt
at describing the extremal Kerr black hole with a CFT.

Extremal BTZ geometry and near-horizon limit

The asymptotic boundary cylinder of AdS3 is naturally described by boundary lightcone coordinates
x˘ “ t{` ˘ φ. The extremal BTZ black hole in Fefferman-Graham coordinates r, x˘ is given by

dr2
ˆ ˙ˆ ˙
2 4L´ ´ 4L` `
ds “ `2 2 ´ rdx ´ ``
dx ´
rdx ´ ` dx (4.19)
r r r

where L˘ are constant and related to the physical charges by M` “ L` ` L´ , J “ L` ´ L´ . Remember

that the event horizon lies at r “ 0 in these coordinates. The extremal limit is given by |J| “ M`. Let
us choose the branch J “ `M`, or equivalently, L´ “ 0. The other extremal branch is similar with
L` exchanged with L´ . In the extremal limit, the line element simply reads as

dr2
ds2ext “ `2 ´ r2 dx` dx´ ` 4J `pdx` q2 . (4.20)
r2

What is the angular velocity in the extremal limit? Let us recall that the vector ξ “ Bt ` Ω H Bφ must
generate the event horizon. In null coordinates
ˆ ˙ ˆ ˙
1 1 1
ξ “ pB` ` B´ q ` Ω H pB` ´ B´ q “ ` Ω H B` ` ´ Ω H B´ . (4.21)
` ` `

Lecture 4. Rotating black holes 94

 4.2. Extremal rotating black holes

But by definition this vector must be null on the horizon (r “ 0), so

ˆ ˙2
1 1
ext µ ν ext
` Ωext “ 0 ùñ Ωext
ˇ
2ˇ
ξ r“0
“ gµν ξ ξ “ g`` pξ ` q2 “ 4J ` H H “´ . (4.22)
` `
2
The generator of the horizon is therefore ξ “ ` B´ . The near-horizon limit can be obtained from
the strict extremal solution by introducing a near-horizon coordinate system depending on a small
parameter λ running to zero. The adapted change of coordinates is found to be
$ a
T
& t “ λ? J `;
’
pt, r, φq Ñ pT, R, Φq : r “ ` λR; (4.23)
’ a
φ “ Φ ` Ωext T
J `.
%
H λ

In the limit λ Ñ 0, the metric becomes

˜ c ¸
dR 2 J 4J
ds2 “ `2 ´ 2R dTdΦ ` dΦ2 (4.24)
4R2 ` `
» ˜ d ¸2 fi
2
` – dR 2 16J `
“ 2
´ R2 dT 2 ` dΦ ´ RdT fl . (4.25)
4 R ` 16J

The structure of the line element is very similar to the NHEK geometry. Again, we recognize the
metric as a combination of AdS2 with the metric on the Up1q circle (with a non-trivial fibration on
AdS2 ). The exact symmetry group is again exactly SLp2, Rq ˆ Up1q. The Killing vectors BT and BΦ are
again enhanced with the vectors ξ 3 and ξ 4 !

Chiral zero temperature states and chiral sector of a CFT

Let us now turn to the dual CFT interpretation of extremal BTZ black holes. We remember that the
non-extremal BTZ black hole can be described as a thermal high energy ensemble in the dual CFT.
What about the extremal BTZ black hole? It admits L´ “ 0 as we saw (or L` “ 0 if we choose the
other chiral branch). It also has zero Hawking temperature which is dual to the temperature of the
thermal ensemble in the CFT. So the state that is dual to the extremal BTZ black hole is an ensemble
of chiral states (L´ “ 0) with zero temperature.

Let us see how it fits with the asymptotic symmetry group analysis. We saw that assuming Brown-
Henneaux boundary conditions, the asymptotic symmetry group consists of two copies of the Vira-
soro algebra. What happens when one takes the near-horizon limit of the extremal BTZ black hole?
The change of coordinates of the boundary lightcone coordinates is

t t 2T a
x` “ `φ “ Φ ; x´ “ ´φ “ J ` ´ Φ. (4.26)
` ` λ

In the limit λ Ñ 8, functions of x` are well-defined but functions of x´ are not. It implies that
right-movers are allowed in the near-horizon limit, but left-movers need to be set to the ground state.
Therefore, the phase space described in (2.57) does not admit a near-horizon limit. We need to set
L´ px´ q “ 0. After performing the near-horizon limit, the resulting chiral phase space is described by

Lecture 4. Rotating black holes 95

 4.2. Extremal rotating black holes

the metric » ˜ d ¸2 fi
`2 dR2 16L` pΦq `
ds2 “ – ´ R2 dT 2 ` dΦ ´ RdT fl . (4.27)
4 R2 ` 16L` pΦq

The exact symmetry group SLp2, Rq ˆ Up1q of the near-horizon BTZ geometry is therefore enhanced
to the asymptotic symmetry group

SLp2, Rq ˆ Up1qR Ñ SLp2, Rq L ˆ VirR (4.28)

since the Up1qR symmetry refers to BΦ which is enhanced. (In fact, the SLp2, Rq L symmetry is strictly
not present in the asymptotic symmetry group since its charges are all zero and therefore trivial, but
since it is a Killing symmetry the factor is usually kept.) The dual field theory corresponding to the
phase space of extremal geometries with excitations consistent with the near-horizon limit is there-
fore the original CFT but amputated to a chiral sector.

The chiral nature of the state dual to extremal BTZ black hole, and the corresponding chiral excita-
tions that still exist in the near-horizon limit are illustrated in Figure 4.5.

Ð L` Ð L`

L´ Ñ

L´ Ñ
Left-moving Right-moving Left-moving Right-moving
sector sector sector sector
(a) The non-extremal BTZ black hole is a thermal (b) The extremal BTZ black hole is a chiral ensemble.
ensemble.

Figure 4.5: CFT thermal and zero temperature ensembles associated to the BTZ black hole.

Chiral Cardy formula

The black hole microscopic entropy counting still works in the extremal limit. Cardy’s formula re-
stricted to a chiral sector gives on the CFT side:
c
c R L` a
SCFT “ 2π “ π ` L` (4.29)
6

which agrees with the Bekenstein-Hawking entropy of the extremal BTZ black hole,

πr` M a
Sext “ “ `π “ π ` L` (4.30)
2G G

Lecture 4. Rotating black holes 96

 4.2. Extremal rotating black holes

?
where r` “ r´ “ ` 4GM. Here it is crucial that a 2d CFT with two chiral sectors exists in order to
derive the formula. Another subtetly is that Cardy’s formula is strictly valid for L˘ " c, so we are
outside of its range of validity but the matching still works!

This “unreasonable validity of Cardy’s formula” motivated to look for possible extensions of his
range of validity. For example in [105] one uses the fact that there is a mass gap of 1{p8Gq “ c{12
between the AdS3 vacuuum and the first BTZ black hole where c “ c L “ c R “ 3`{p2Gq is the Brown-
Henneaux central charge as we described in the second lecture. In the dual CFT language it can
be translated to the existence of a “sparse light spectrum of states” in the energy range ´c{12 ă
L` ` L´ ă 0. For such CFTs, an extended range of applicability of Cardy’s formula exists. One first
use the standard thermodynamics relations to define a conjugate chemical potential to L` and L´ :
ˆ ˙ ˆ ˙
1 BS 1 BS
“ and “ . (4.31)
T´ BL´ L` T` BL` L´

The dimensionless chemical potentials T` , T´ are sometimes called by abuse of language “left and
π2 2 π2 2
right temperatures”. Using the full Cardy formula we get L` “ 6 c R T` , L´ “ 6 c L T´ and the
entropy can be written in the canonical ensemble as

π2
SCFT “ pc L TL ` c R TR q (4.32)
3

where the extended range of validity is now TL ą 1{p2πq, TR ą 1{p2πq [105]. For the extremal BTZ
black hole we have TR " 1 but TL “ 0 so we are still outside of the range of validity. More work is
yet needed to understand this matching of entropy!

4.2.3 The Kerr/CFT correspondence

After defining the near-horizon limit of nearly extremal Kerr and reviewing some relevant back-
ground material on the 3d BTZ black hole as a warm up, we are now ready to present the Kerr/CFT
correspondence. In 2009, Guica, Hartman, Song, and Strominger proposed [95] a new type of holo-
graphic duality between 4d extremal Kerr black holes and 2d conformal field theories analogous to
a duality between BTZ black holes and 2d CFTs that we briefly reviewed. The conjectured dual-
ity is based on properties of the near-horizon limit of the extremal Kerr black hole. Even though
the original conjecture turned out incorrect, an updated conjecture still holds; the work also gener-
ated very interesting developments in holography and still contains a mysterious entropy matching
that deserves further research. Let’s now review some of these developments following a viewpoint
enriched with the subsequent research work [96].

Virasoro symmetry

The basis of a 2-dimensional CFT is its symmetry group consisting of two copies of the Virasoro alge-
bra. But asking for two copies of the Virasoro algebra is asking too much. As we saw for the case of
the near-horizon limit of the extremal BTZ black hole we can only hope in the near-horizon limit for
only one Virasoro algebra extending the SLp2, Rq ˆ Up1q symmetry as SLp2, Rq ˆ Vir. The fundamen-
tal reason is that any non-extremal excitation prevents the existence of a decoupling geometry such

Lecture 4. Rotating black holes 97

 4.2. Extremal rotating black holes

as the near-horizon BTZ or the NHEK geometry. In the Kerr case, non extremal excitations lead to a
non-vanishing coupling between the NHEK region and the asymptotically flat region. This implies
that there is no 2d CFT describing the Kerr black hole, which invalidates the original Kerr/CFT con-
jecture. There are however some disturbing connections which suggest a relationship with at least
part of the structure of a 2d CFT [106, 107, 108, 109].

Since non-extremal physics does not exist in the near-horizon of an extremal black hole, there is classi-
cally no dynamics except non-trivial diffeomorphisms and topology. The NHEK geometry therefore
bears much resemblence with the near-horizon BTZ black hole in 3d gravity! The original approach
of [95] was inspired from the Brown-Henneaux analysis: one first imposes boundary conditions and
study which symmetries preserve the boundary conditions, one checks that the associated charges
are finite and integrable and one write the Poisson bracket to derive the charge algebra of asymptotic
symmetries. Now, since the boundary conditions contain nothing else than diffeomorphisms and
topology, one can just study these two features.

The ansatz for the generator of the sought-for Virasoro algebra is the following
ˆ ˙
1 b
χ “ εpΦqBΦ ´ ε pΦq RBR ` BT ` subleading terms (4.33)
R

which is built from an arbitrary function on the circle εpΦq. In the original Kerr/CFT paper b was set
to zero but then there is no smooth classical phase space. Instead b “ 1 leads to a smooth classical
phase space [110, 15]. Since Φ is 2π-periodic, the function εpΦq can be mode expanded. We define
Ln fi χrεpΦq “ ´e´inΦ s. It is straightforward to check that irLm , Ln s “ pm ´ nqLm`n which is the Witt
algebra on the circle. The signs are chosen for later convenience (in order to have a positive central
charge!).

Starting from this ansatz alone, it is possible to exponentiate this generator to generate the finite dif-
feomorphism of the NHEK geometry depending on one arbitrary function ψpΦq of Φ and thereby
construct a (small) phase space of asymptotically NHEK solutions [15]. One can then evaluate the
surface charge associated with any arbitrary generator (4.33) for infinitesimal variations on the phase
space, which solely amount to vary ψ. The resulting charges Lm are finite and integrable and there-
fore the generator (4.33) is promoted to an asymptotic symmetry. The Poisson algebra of charges is
isomorphic to the algebra of asymptotic symmetries up to central terms as follows from the general
results derived in the first lecture. One finds

c 3
itLm , Ln u “ pm ´ nqLm`n ` m δm`n,0 (4.34)
12

with central charge

c “ 12J, (4.35)

where J is the angular momentum. Moreover, one can check that the asymptotic symmetries act
everywhere in the bulk spacetime, which promotes them to symplectic symmetries, similarly to 3d Ein-
stein gravity. Because there is no finite energy excitation in NHEK, the SLp2, Rq Killing symmetries

Lecture 4. Rotating black holes 98

 4.2. Extremal rotating black holes

are associated with zero charges are therefore trivial. The asymptotic symmetry group therefore con-
sists of one copy of the Virasoro algebra with central charge c “ 12J.

Conjugated chemical potential and Cardy matching

What it remains to be known to use the (chiral version of the) canonical Cardy formula (4.32) is the
chemical potential associated with angular momentum. Usually, it is the angular velocity of the black
hole, but for an extremal black hole the mass is also function of the angular momentum, so we need
to use the definition
1 BSext
“ . (4.36)
TΦ BJ
1
For extremal Kerr, it gives TΦ “ 2π , which quite annoyingly just lies beyond the extended range of
applicability of Cardy’s formula. Yet, the entropy of extremal Kerr black hole matches with the chiral
canonical Cardy formula
π2
Sext Kerr “ 2π J “ c TΦ (4.37)
3
after using the definitions of c and TΦ . This is the remarkable entropy matching performed by the
Kerr/CFT correspondence! In all known examples of extremal black holes including black holes with
higher curvature corrections [18], this match was always shown to hold. However, there is no clear 2d
CFT here and therefore the main hypothesis to derive Cardy’s formula does not hold. Yet it matches.
This is the “unreasonable universality of Cardy matching” which is still a mystery today.

Frolov-Thorne vacuum

Let us close this section by interpreting the conjugated chemical potential TΦ . The interpretation can
be made in the context of quantum field theories in curved spacetimes. But first let us derive an
equivalent formula for TΦ . So far we defined TΦ δSext “ δJ. The first law is TH δS “ δM ´ Ω H δJ. Let
us specialize the first law to extremal variations where δJ “ δpM2 q or δM “ 1
2M δJ “ Ωext
H δJ. We find
Ωext
H ´Ω H
δSext “ TH δJ. Comparing with the definition of TΦ we find

1 Ωext ´ Ω H
“ H . (4.38)
TΦ TH

The main result of Hawking in 1974 is that the quantum field state at late times after a black hole has
κ
formed by collapse is described as a thermal ensemble at Hawking temperature TH “ 2π where κ is
the surface gravity of the black hole. The root of the effect can be attributed to pair creations and an-
nihilations in the quantum vacuum at the vicinity of the horizon. If one of the particles is trapped by
the horizon, its conjugated pair cannot annihilate and is emitted by the black hole. The derivation of
the thermal nature of the spectrum requires to use the properties of event horizons and the definition
of vacuum.

In curved spacetimes the definition of vacuum is not unique. It depends on the definition of pos-
itive frequency modes, which requires a timelike Killing vector. In Schwarzschild spacetime, there

Lecture 4. Rotating black holes 99

 4.3. Black hole spectroscopy

is a globally timelike Killing vector Bt which naturally defines the so-called Hartle-Hawking vac-
uum. This state is regular at the event horizon and thermal. It is described by a density matrix
ρ “ expp´ TωH q. For the Kerr black hole, there is no global timelike Killing vector, due to the ergo-
sphere. For an observer close to the horizon it is natural to define positive frequency modes with re-
spect to the generator of the horizon ξ “ Bt ` Ω H Bφ . This vector field is timelike in a region bounded
by the event horizon and an outer region known as the velocity of light surface where ξ becomes
null. The vacuum defined this way is known as the Frolov-Thorne vacuum. It is described by a density
Hm
matrix ρ “ expp´ ω´Ω
TH q where ω is the frequency and m the azimuthal number of the wave.

In the near-horizon limit, one can rewrite the density matrix in terms of variables adapted to the
NHEK coordinates. We first rewrite a wave as

Fpr, θq expp´iωt ` imφq “ FpR, θq expp´iΩT ` iMΦq. (4.39)

Using the change of coordinates (4.9)-(4.10) we find m “ M and ω “ mΩext

H ` λΩ. It implies that

ω ´ ΩH m Ωext
H ´ ΩH
ˆ ˙ ˆ ˙ ˆ ˙
m
ρ “ exp ´ “ exp ´m “ exp ´ . (4.40)
TH TH TΦ

This shows that the Frolov-Thorne vacuum is thermally populated in the extremal limit with a “tem-
perature” equal to TΦ .

There are however some caveats here that need to be pointed out. While the generator of the horizon
ξ is timelike outside the event horizon of a non-extremal Kerr black hole, a singular behavior occurs
in the near-horizon extremal limit. In NHEK the vector ξ “ BT is timelike only in a polar wedge
around the north and south poles, but it is spacelike around the equator (indeed, check the sign
of gTT by evaluating Λpθq in (4.11)). In fact, there is no globally timelike Killing vector in NHEK.
So there is no quantum vacuum in NHEK. One should understand the “NHEK vacuum” only as a
(singular) limit of the Frolov-Thorne vacuum of the near-extremal case.

4.3 Black hole spectroscopy

On September 14, 2015, gravitational waves were detected for the first time thanks to the twin detec-
tors of the Laser Interferometer Gravitational-wave Observatory (LIGO) [111]. This event launched
the new era of gravitational wave astronomy. The upcoming direct observations of gravitational
physics enable today to test Einstein’s theory of gravity in the strong field regime at a precision never
reached by other experiments. Third generation detectors, and in particular LISA planned for 2034,
will add new precision measurements to be compared with theoretical predictions. This is a unique
and very exciting time for gravitational physics.

In this lecture, we would like to present the late stages of black hole mergers, after the two bodies
have collapsed to form a black hole. The final state of the merger is a Kerr black hole with small
perturbations which can be approximated at late stages by linearized perturbations. The mathemat-
ical problem of solving linearized Einstein’s equations around the Kerr black hole benefited from a

Lecture 4. Rotating black holes 100

 4.3. Black hole spectroscopy

crucial contribution from Teukolsky in 1972 during his PhD studies with Kip Thorne. He found a de-
termining set of variables which separates [112]. This allows to reduce the complexity of the problem
to solving two coupled ordinary differential equations. In turn, Leaver showed in 1985 how to solve
these differential equations to arbitrary precision [113]. This allows to deduce the fundamental char-
acteristic decay frequencies (the quasi-normal mode frequencies) of the Kerr black hole numerically
up to arbitrary precision.

Analoguously to the spectral lines that the hydrogen atom can electromagnetically emit, the Kerr
quasi-normal modes tell us what the Kerr black hole can gravitationally radiate. At the time of writ-
ing, the observations do not yet allow to check the exact quasi-normal modes frequencies of the Kerr
black hole, but the experimental science of black hole spectroscopy will soon start and will allow to
check the validity of the Kerr metric!

In order to present the founding work of Teukolsky, it is necessary to first introduce the Newman-
Penrose formalism and Petrov’s classification of the Weyl tensor.

4.3.1 Fundamentals of the Newman-Penrose formalism

The metric field gµν is not the appropriate field to decribe fundamental physics which also involves
fermions. Coupling Einstein gravity to fermions requires to introduce the more fundamental basis
of tetrads eµa , a “ 0, 1, 2, 3, such that eµa is an invertible matrix and gµν “ ηab eµa eνb where ηab is the
Minkowski metric of the tangent space at each point. The formulation in terms of tetrads allows to
use the Cartan formalism where the fundamental ingredient are the one-forms e a “ eµa dx µ . We saw in
the second Lecture how this Cartan formalism led us to reformulate 3d Einstein gravity as a sum of
two Chern-Simons theories!

The Newman-Penrose formalism is a tetrad formalism with complex tetrads where the tangent space
Minkowski metric ηab is chosen at each point to be
¨ ˛
0 ´1 0 0
˚ ‹
˚ ´1 0 0 0 ‹
ηab “˚ ‹. (4.41)
˚ 0 0 0 1 ‹
˝ ‚
0 0 1 0

The tetrad frame is chosen to be a set of 4 null vectors lµ , nµ , mµ , m̄µ with

gµν “ ´lµ nν ´ nµ lν ` mµ m̄ν ` m̄µ mν . (4.42)

Because they is no complete real basis of null directions in a Lorentzian manifold, two of these vectors
have to be complex, and since the final metric is real these two complex tetrad are complex conjugated.
The two real tetrads label particular ingoing and outgoing null directions, so this formalism is well
adapted to describe geometrically the propagation of gravitational waves.

The parallel transport must be re-expressed in terms of the general Newman-Penrose tetrad basis, so

Lecture 4. Rotating black holes 101

 4.3. Black hole spectroscopy

instead of using the standard 4 connections ∇µ , we consider instead 4 locally-defined directional co-
variant derivatives on the flow of the tetrad, which are historically denoted by D, ∆, δ, δ̄ and straight-
forwardly defined as

D “ l µ ∇µ ; ∆ “ nµ ∇µ ; δ “ mµ ∇µ ; δ̄ “ m̄µ ∇µ . (4.43)

The Christoffel symbols have 4 ˆ 6 “ 24 real components, equivalent to 12 complex numbers. In the
Newman-Penrose formalism, one defines 12 complex spin coefficients which encode the same informa-
tion. These coefficients have an individual name. Here is the complete list so that you have a precise
idea:

κ “ ´mµ l ν ∇ν lµ ; σ “ ´mµ mν ∇ν lµ ; (4.44)

λ “ ´nµ m̄ν ∇ν m̄µ ; ν “ ´nµ nν ∇ν m̄µ ; (4.45)
ρ “ ´m m̄ ∇ν lµ ;
µ ν
µ “ ´n m ∇ν m̄µ ;
µ ν
(4.46)
τ “ ´m n ∇ν lµ ;
µ ν
v “ ´n l ∇ν m̄µ ; µ ν
(4.47)
1
e “ ´ pnµ l ν ∇ν lµ ` mµ l ν ∇ν m̄µ q ; (4.48)
2
1
γ “ ´ pnµ nν ∇ν lµ ` mµ nν ∇ν m̄µ q ; (4.49)
2
1 µ ν
α “ ´ pn m̄ ∇ν lµ ` mµ m̄ν ∇ν m̄µ q ; (4.50)
2
1 µ ν
β “ ´ pn m ∇ν lµ ` mµ mν ∇ν m̄µ q. (4.51)
2

Now let us examine the formulation of the curvature. In 4 dimensions, the Riemann tensor has 20
independent real components, while the Ricci tensor, being symmetric, admits 10 real components.
One can build the traceless part of the Riemann tensor, it is called the Weyl tensor, and it therefore
contains 10 real or 5 complex independent components. The Weyl tensor is more precisely defined as

1
Wµνρσ “ Rµνρσ ´ gµrρ Rσsν ` gνrρ Rσsµ ` Rgµrρ gσsν . (4.52)
3

The basis of 5 complex scalars Ψi , i “ 0, . . . , 4 that allows to label an arbitrary Weyl tensor are called
the Weyl-Newman-Penrose scalars and there are defined as

Ψ0 “ Wαβγδ l α m β l γ mδ , (4.53)
Ψ1 “ Wαβγδ l α n β l γ mδ , (4.54)
Ψ2 “ Wαβγδ l m m̄ n ,
α β γ δ
(4.55)
Ψ3 “ Wαβγδ l n m̄ n ,
α β γ δ
(4.56)
Ψ4 “ Wαβγδ nα m̄ β nγ m̄δ . (4.57)

These 5 scalars are invariant under diffeomorphisms but depend on the choice of the tetrad basis.
Let us define 3 antisymmetric bivectors

Xµν “ ´2nrµ m̄νs ; Yµν “ 2lrµ mνs ; Zµν “ 2mrµ m̄νs ´ 2lrµ nνs . (4.58)

Lecture 4. Rotating black holes 102

 4.3. Black hole spectroscopy

One can show as an exercise that the Weyl tensor can indeed be written as a combination of the 5
Weyl-Newman-Penrose scalars in the tetrad basis. The exact formula is

Wαβγδ “Ψ0 Xαβ Xγδ ` Ψ1 Xαβ Zγδ ` Zαβ Xγδ ` Ψ2 Yαβ Xγδ ` Xαβ Yγδ ` Zαβ Zγδ
` ˘ ` ˘

` Ψ3 Yαβ Zγδ ` Zαβ Yγδ ` Ψ4 Yαβ Yγδ ` c.c.

` ˘
(4.59)

Since the Weyl-Newman-Penrose scalars are not invariant under a change of tetrad, it is important
to discuss the exact ambiguity. One can perform a local Lorentz transformation at each spacetime
point, which rotates the tetrad. Such a transformation is labelled by 6 real numbers at each point. It
is useful to categorize these 6 local Lorentz rotations in 3 types:

• Rotations of type I which leave l µ unchanged (a P C);

l µ ÞÑ l µ , nµ ÞÑ nµ ` a˚ mµ ` am̄µ ` aa˚ l µ , mµ ÞÑ mµ ` al µ m̄µ ÞÑ m̄µ ` a˚ l µ . (4.60)

• Rotations of type II which leave nµ unchanged (b P C);

nµ ÞÑ nµ , l µ ÞÑ l µ ` b˚ mµ ` bm̄µ ` bb˚ nµ , mµ ÞÑ mµ ` bnµ , mµ ÞÑ mµ ` b˚ nµ . (4.61)

• Rotations of type III which leave the directions of l µ and nµ unchanged and rotate mµ by an
angle in the mµ , m̄µ plane (A, θ P R);

l µ ÞÑ A´1 l µ , nµ ÞÑ Anµ , mµ ÞÑ eiθ mµ m̄µ ÞÑ e´iθ m̄µ . (4.62)

You can check that under these transformations, the tetrads preserve the same orthonomality con-
ditions. In other words, the metric on the tangent space (4.41) is preserved. There are also discrete
flips, such as exchanging l µ Ô nµ . After some work involving the evaluation of the Weyl tensor
contracted with each of the possible combinations of tetrads, one can find the transformation laws of
the Weyl-Newman-Penrose scalars under each rotation of type I, II and III:

Ψ0 ÞÑ Ψ0 ,
Ψ1 ÞÑ Ψ1 ` a˚ Ψ0 ,
Ψ2 ÞÑ Ψ2 ` 2a˚ Ψ1 ` pa˚ q2 Ψ0 , (4.63)
Ψ3 ÞÑ Ψ3 ` 3a˚ Ψ2 ` 3pa˚ q2 Ψ1 ` pa˚ q3 Ψ0 ,
Ψ4 ÞÑ Ψ4 ` 4a˚ Ψ3 ` 6pa˚ q2 Ψ2 ` 4pa˚ q3 Ψ1 ` pa˚ q4 Ψ0 ;

Ψ0 ÞÑ Ψ0 ` 4bΨ1 ` 6b2 Ψ2 ` 4b3 Ψ3 ` b4 Ψ4 ,

Ψ1 ÞÑ Ψ1 ` 3bΨ2 ` 3b2 Ψ3 ` b3 Ψ4 ,
Ψ2 ÞÑ Ψ2 ` 2bΨ3 ` b2 Ψ4 , (4.64)
Ψ3 ÞÑ Ψ3 ` bΨ4 ,
Ψ4 ÞÑ Ψ4 ;

Lecture 4. Rotating black holes 103

 4.3. Black hole spectroscopy

Ψ0 ÞÑ A2 e´2iθ Ψ0 ,
Ψ1 ÞÑ A´1 eiθ Ψ1 ,
Ψ2 ÞÑ Ψ2 , (4.65)
Ψ3 ÞÑ Ae´iθ Ψ3 ,
Ψ4 ÞÑ A2 e´2iθ Ψ4 .

4.3.2 Fundamentals of Petrov’s classification

Let us now review the classification that Petrov obtained in 1954 [114]1 , using the Newman-Penrose
formalism. This classification is taylor made for 4 spacetime dimensions. A higher dimensional clas-
sification also exists but was established only 50 years later [115].

We would like to get an algebraic, coordinate-independent, classification of solutions to Einstein

field equations. The non-homogeneous solutions (Tµν ‰ 0) are dependent on the matter content
through the Ricci tensor Rµν which is the trace part of the Riemann tensor. The traceless part of the
Riemann tensor, the Weyl tensor, is left unconstrained by the matter fields and therefore represents
the purely gravitational field. For vacuum solutions (Tµν “ 0), the Ricci tensor is always zero even
though the gravitational field (the Weyl tensor) could be non-trivial. The classification of the Weyl
tensor is therefore useful to categorize both homogeneous and non-homogeneous solutions to Ein-
stein’s equations. Note that the Weyl tensor is conformally invariant and therefore may classify at
most conjugacy classes of metrics differing by a global conformal Weyl factor (for example AdS4 and
Minkowski will belong to the same class).

Petrov classified the Weyl tensor by the number of degenerate local eigenvalues and (antisymmetric)
eigenbivectors of the Weyl tensor. The eigenvalue equation reads as

µν αβ
W αβ X “ λ X µν . (4.66)

A non-trivial result due to Penrose in 1960 shows that solving this eigenvalue problem is equivalent
to classify spacetimes according to the degeneracy of principal null directions of the Weyl tensor. Such
directions are spanned by null vectors kµ obeying

krα Wβsγδrρ k σs kγ kδ “ 0. (4.67)

Yet another equivalent formulation of the classification is the following. We have just seen that with
respect to a chosen tetrad, the Weyl tensor is completely determined by the five Weyl-Newman-
Penrose scalars. The third formulation of the classification consists in determining how many of
these scalars can be made to vanish for a given spacetime by choosing a suitable orientation of the
tetrad frame.

Let us concentrate on this third formulation. Given a metric together with a Newman-Penrose frame
1 Asan anecdote on ULB connections, it is amusing to notice that this theory was also independently developed at ULB
by Géhéniau in 1957 (who supervised the PhD of M. Henneaux 23 years later).

Lecture 4. Rotating black holes 104

 4.3. Black hole spectroscopy

l µ , nµ , mµ , m̄µ , we can deduce the 5 Weyl components Ψi . Let us assume that Ψ4 ‰ 0. If this is not the
case, we can perform a rotation of the type I (4.63) to make it non zero, as long as the spacetime is not
conformally flat in which case all the Weyl tensor vanishes. Now consider a rotation of type II (4.64)
with complex parameter b. It is clear that Ψ0 can be made to vanish if b is a root of the equation

Ψ0 ` 4bΨ1 ` 6b2 Ψ2 ` 4b3 Ψ3 ` b4 Ψ4 “ 0. (4.68)

This equation has always exactly 4 roots, and the corresponding directions of l µ , namely l µ ` b˚ mµ `
bm̄µ ` bb˚ nµ are called the principal null directions of the Weyl tensor. Indeed, contracting the tetrad
decomposition of the Weyl tensor (4.59) with multiple instances of l µ , one obtains after using the
orthonormality condition (4.41) of the tetrad:

lrα Wβsγδrρ lσs l γ l δ “ Ψ0 lrα m̄ βs lrρ m̄σs ` Ψ˚0 lrα m βs lrρ mσs . (4.69)

After the type II frame rotation that we just defined by solving (4.68), we have a vanishing new
Ψ0 “ 0, and therefore the new null direction l µ is indeed a principal null direction in the sense of
Penrose. If one or two more roots coincide, the Weyl tensor is said to be algebraically special; other-
wise it is said to be algebraically general. The various ways in which the roots coincide or are distinct
lead to Petrov’s classification.

There are in total 6 types of Weyl tensors, which is summarized in Table 4.6(a). Type I is the most
general one with 4 distinct principal null directions, and type O is the totally degenerate case where
the Weyl tensor is vanishing. The Kerr spacetime is type D with two distinct but doubly degenerate
principal null directions. The so-called Goldberg-Sachs theorem implies that for any vacuum type D
spacetime such as the Kerr black hole, the principal null directions are shear-free geodesic congru-
ences. The fact that only Ψ2 is non-vanishing for Kerr and the fact that its null principal directions
are shear-free geodesic congruences make the Newman-Penrose formalism particularly adapted to
describe the gravitational physics of Kerr! Let us now turn on to this physics...

Lecture 4. Rotating black holes 105

 4.3. Black hole spectroscopy

Petrov type Multiplicity of p.n.d. Vanishing Weyl components Criterion on Wαρσβ

I p1, 1, 1, 1q Ψ0 “ 0 krγ Wαsρσrβ k δs kρ kσ “ 0

II p2, 1, 1q Ψ0 “ Ψ1 “ 0 krγ Wαsρσβ kρ kσ “ 0

D p2, 2q Ψ0 “ Ψ1 “ 0 krγ Wαsρσβ kρ kσ “ 0

III p3, 1q Ψ0 “ Ψ1 “ Ψ2 “ 0 krγ Wαsρσβ kρ “ 0

N p4q Ψ0 “ Ψ1 “ Ψ2 “ Ψ3 “ 0 Wαρσβ kα “ 0

O ∅ Ψi “ 0, @i Wαρσβ “ 0

(a) Characterisation of Petrov types. kµ is always the most degenerate principal null direction (p.n.d.). Adapted from
[116].

II D

III N O

(b) The Penrose graph summarizing the degeneracy growth

in Petrov’s classification. Each arrow indicates one
additional degeneracy. Adapted from [116].

Figure 4.6: Summary of Petrov’s algebraic classification.

Lecture 4. Rotating black holes 106

 4.3. Black hole spectroscopy

4.3.3 Quasi-normal mode ringing of Kerr

In this section, we will analyze the linear perturbations of the Kerr black hole. Generally, one can
consider a theory of matter fields ΦiM minimally coupled to Einstein gravity without cosmological
constant, whose action can be written as
ż ż
1
d4 x ´g R ` d4 x ´g LrΦiM , gµν s
a a
S“ (4.70)
16πG

where LrΦiM , gµν s is the lagrangian density of matter fields collectively denoted by ΦiM . The varia-
tion of S with respect to the metric field gµν gives rise to Einstein’s equations Gµν “ 8πGTµν , and the
variation of the matter fields supplements it by the equations of motion for ΦiM . We define the back-
ground field as the Kerr black hole solution of this system of differential equations, denoted by ḡµν and
Φ̄iM “ 0. We consider perturbations around that background: gµν “ ḡµν ` hµν , ΦiM “ Φ̄iM ` δΦiM . If
the deviations are small, we can approximate them by the linearized perturbations around the back-
ground field. The linearized field equations are a set of partial second-order differential equations
which depend upon the background geometry.

The Kerr black hole ḡµν admits two Killing isometries, Bt and Bφ in the traditional Boyer-Lindquist
coordinates. As a result, no explicit t or φ dependence is present in the equations, and one can Fourier
transform them to impose the equations on the Fourier modes of both metric and matter variables. In
other words, one Fourier expands all the metric and matter fields collectively denoted as Φi pt, r, θ, φq
as ż
1 ÿ
Φ pt, r, θ, φq “
i
dωe´iωt eimφ Fi pr, θq (4.71)
2π mPZ

and all equations reduce to equations in terms of Fi pr, θq which depend upon r, θ, but also M, J (the
Kerr black hole parameters) and m, ω (the perturbation parameters).

To complete the problem, we still need to impose the physical boundary conditions. By definition of
the event horizon, all modes need to be ingoing at r “ r` so we demand that

rÑr`
e´iωt`imφ Fpr, θq ÝÝÝÑ e´iωv‹ `imφ‹ Fpθq (4.72)

where r‹ is the tortoise coordinate, v‹ “ t ` r‹ the advanced time and φ‹ the angular coordinate which
define the regular ingoing Eddington-Finkelstein coordinates v‹ , r‹ , θ, φ‹ (or in other words, which
resolves the geometry near the horizon). For the Kerr black hole, one has

r 2 ` a2 a
dr‹ fi dr, dφ˚ fi dφ ` dr. (4.73)
∆ ∆

We also impose that there is no ingoing field from I ´ , which defines a “purely outgoing mode at null
infinity I ` ”. This boundary condition prevents the stimulated emission of the black hole, and selects
the intrinsic emission. It reads as

rÑ8, u fixed
e´iωt`imφ Fpr, θq ÝÝÝÝÝÝÝÑ e´iωu`imφ F̃pθq (4.74)

Lecture 4. Rotating black holes 107

 4.3. Black hole spectroscopy

where u “ t ´ r is the asymptotically flat retarded time.

Providing these boundary conditions, the solutions of the linearized equations of motion will turn
out to belong to discrete sets of solutions, since one can show that the frequency ω will be quantized
ωlmN in terms of 3 integer numbers. The numbers l, m are nothing else than (deformed) spherical
harmonic numbers and N “ 0, 1, 2, . . . is known as the overtone. Here the system is dissipative since
waves escape either at null infinity or inside the event horizon. The frequencies are therefore com-
plex: a positive imaginary part means a growing mode (an unstable mode) and a negative imaginary
part means a damping/decaying mode. The proof of linear stability of Kerr amounts to show that
all imaginary parts are negative! The presence of a family of discrete frequencies characterizing the
Kerr black hole is very similar to the spectral lines of the hydrogen atom and therefore the science of
quasi-normal modes is often called black hole spectroscopy, as announced in the introduction. Let us
now derive all this!

Separation of variables and Teukolsky master equation

The question is: how to deal with partial differential equations involving pr, θq in a Kerr background?
The answer is: find how to separate them! Let us first give a hint that the equations of spins 0, 1 and
2 are separable.

We consider a causal geodesic that travels outside the Kerr black hole x µ pλq. The tangent vector
µ
x9 µ fi dx
dλ obeys gµν x x “ 0 or ´1. Both stationarity and axisymmetry guarantee that the geodesic
9µ 9ν
motion has two dynamical invariants, which are simply the energy of the particle moving on the
geodesic e “ ´µkµ gµν x9 ν (k “ Bt , µ is the rest mass) and its orbital angular momentum j “ µkµ gµν x9 ν
(k “ Bφ ). In addition to this, Carter found in 1968 [117] that there is in fact another conserved quantity
on the flow of causal geodesics, quadratic in the momentum, called in his name the Carter constant
Q. It takes the following form in Boyer-Lindquist coordinates
« ˆ ˙2 ff
j
Q“ µ2 x9 θ2 2 2 2
` cos θ a pµ ´ e q ` 2
. (4.75)
sin θ

One often uses the related constant K “ Q ` pj ´ aeq2 which is always non-negative. The existence of
4 first order equations for x µ pλq allows to analytically solve the geodesic problem in Kerr, though in
terms of elliptic integrals [118]. The conservation of Q is related to a hidden symmetry, which is of
higher order than Killing isometries. For the Kerr background, one can construct a symmetric tensor
Kµν which satisfies a close cousin of the Killing equation

∇pλ Kµνq “ 0. (4.76)

For this reason, Kµν is said to be a Killing tensor, and is non-trivial in the sense that it is not simply the
tensor product of the two Killing vectors always discovered.

The contraction between Kµν or gµν with the 4-velocity of any particle gives a dynamical invariant,
so we can define an analogue operator to the Kerr d’Alembertian l “ gµν ∇µ ∇ν by performing the

Lecture 4. Rotating black holes 108

 4.3. Black hole spectroscopy

contraction lK “ K µν ∇µ ∇ν . We have the fundamental property that these two second-order differ-
ential operators on the manifold actually commute rl, lK s ” 0, as you can check as an exercise. As
shown by Carter [119], this property is at the origin of the separability of the scalar wave equation
in the Kerr background. While it is not known whether there is a similar symmetry-based structure
that allows the separability of the spin 2 perturbation (the linearized Einstein perturbation), it turns
out to hold as we will now show!

The proof of separability came from Teukolsky in 1972 [112] who found the right combination of the
metric perturbation which separates. He employed the Newman-Penrose formalism and Petrov’s
classification. Providing that the Kerr spacetime is Petrov type D, one can find 2 preferential null
orthogonal directions to build the Newman-Penrose basis, which are the degenerate principal null
directions of the Weyl tensor.

The two (double) principal null directions of the Weyl tensor are

1` 2 1 ` 2
r ` a2 , ∆, 0, a r ` a2 , ´∆, 0, a .
˘ ˘
lµ “ and nµ “ (4.77)
∆ 2Σ

One completes the basis with the complex vector

ˆ ˙
µ 1 i
m “? ia sin θ, 0, 1, (4.78)
2pr ` ia cos θq sin θ

to get the so-called Kinnersley tetrad [120]. With respect to this choice of null basis, the only non-
vanishing Weyl component is
M
Ψ2 “ ´ . (4.79)
pr ´ ia cos θq3
We therefore proved that the Kerr black hole is Type D, as stated in Section 4.3.2: all effects of gravita-
tion are caused by the single Weyl scalar Ψ2 in the Kinnersley tetrad. A modification gµν Ñ gµν ` hµν
of the Kerr metric amounts to perturb Ψi by δΨi . In fact, if one can solve only for Ψ0 or Ψ4 . The
other Weyl perturbations can be then deduced, up to the additional information on how M and J are
perturbed. The entire perturbation hµν could then be reconstructed from all variations δΨi .

The point is that the equation for pr ´ ia cos θq4 δΨ4 is separable (equal to a function of θ times a
function of r). Teukolsky realised later that the equation for δΨ0 is also separable. In fact, there is an
identity relating δΨ0 and δΨ4 so the equations are equivalent. Physically, the Weyl scalars δΨ0 and
δΨ4 both describe the two polarization modes of the gravitational waves. At the linear level, there
are only three possible perturbations: gravitational waves, changes of the gravitational potentials
(change of M, J)2 or changes of coordinates (which can belong to the asymptotic symmetry group and
therefore be non-trivial). It is thus easy to understand that we can reconstruct all δΨi from either δΨ0
or δΨ4 together with the knowledge of how M and J change since these variables are diffeomorphic
invariant. Since the perturbations of spin 0 or 1 also separate, it is convenient to write the master
2 Tobe precise, there are 2 additional gravitational potentials that are usually discarded because considered unphysical:
the NUT charge, a sort of magnetic analogue to the mass but which generates closed timelike curves, and the acceleration
parameter, which introduces conical wire singularities.

Lecture 4. Rotating black holes 109

 4.3. Black hole spectroscopy

equation which separates all spins in a single notation. This is called the Teukolsky master equation!

pr2 ` a2 q2
 2
B ψ 4Mar B 2 ψ
„ „ 2  2
a 1 B ψ
´ a2 sin2 θ ` ` ´
∆ Bt2 ∆ BtBφ ∆ sin2 θ Bφ2
ˆ ˙ ˆ ˙ „ 
´s B s`1 Bψ 1 B Bψ apr ´ Mq i cos θ Bψ
´∆ ∆ ´ sin θ ´ 2s ` (4.80)
Br Br sin θ Bθ Bθ ∆ sin2 θ Bφ
Mpr2 ´ a2 q
„ 
Bψ
´ 2s ´ r ´ ia cos θ ` ps2 cot2 θ ´ sqψ “ T
∆ Bt

where T is a source term if we couple gravity to matter. For the spin 0 case, ψ is just the original scalar
field. For the spin 2 case, ψ is either δΨ0 or pr ´ ia cos θq4 δΨ4 . The general solution of the Teukolsky
master equation takes the separable form

ż 8 ÿ
`l
1 ÿ
ψpt, r, θ, φq “ dωe ´iωt
eimφ Rslmω prqSlmω
s
pcos θq. (4.81)
2π
l“|s| m“´l

s
The equation for Slmω pcos θq is called the spin weighted spheroidal harmonic equation

m2 ` 2msx ` s2 s
„  „ 
d 2 d s 2 2 2 s
p1 ´ x q S pxq ` a ω x ´ 2aωsx ` Elmω ´ Slmω pxq “ 0 (4.82)
dx dx lmω 1 ´ x2
s
where x “ cos θ and Elmω is the separation constant. When a “ 0, the dependence in ω drops out and
s pcos θq reduce to spin-weighted spherical harmonics Y s pθ, φq “ Ss pcos θqeimφ after
the functions Slm lm lm
s
inclusion of the Fourier φ factor. In this case, the angular separation constants Elmω s are known
“ Elm
s “ lpl ` 1q.
analytically to be Elm

The radial equation is the radial Teukolsky equation:

B s`1 BRlmω
∆´s p∆ q ´ VprqRlmω prq “ Tlmω prq (4.83)
Br Br

with source Tlmω prq and potential

pKmω q2 ´ 2sipr ´ MqKmω

Vprq “ ´ ´ 4siωr ` λ`mω ,
∆
Kmω fi pr2 ` a2 qω ´ ma, (4.84)
λ`mω fi Elmω ´ 2amω ` a2 ω 2 ´ sps ` 1q.

When a “ 0, the m dependence drops out. This is a consequence of SOp3q symmetry.

Solving the angular equation

s
The solutions of (4.82) are the so-called spin-weighted spheroidal harmonics Slmω pxq which form, for each
value of the spin s, an orthogonal system of functions on the interval r´1, 1s (recall that x “ cos θ)
ż `1
s
dx Slmω pxqSls1 m1 ω1 pxq “ δll 1 δmm1 δpω ´ ω 1 q. (4.85)
´1

Lecture 4. Rotating black holes 110

 4.3. Black hole spectroscopy

They are only defined for values of integers parameters l, m in the range l ě |s| and |m| ď l, other-
wise they are simply identically zero. The case s “ 0 reduces to the spheroidal harmonics which are
well-known, even by symbolic computation softwares such as MathematicaTM .

An analytic representation of the solutions for s ą 0 was found for first time by Leaver [113], thanks to
the Frobenius method. Let us now review this result. The equation (4.82) has three singular points,
among which two regular singular points at x “ ˘1, and an irregular singularity at x “ 8 if we
extend analytically the domain of definition of the coordinate x. Boundary conditions are imposed
such that solutions to (4.82) are finite at the regular singular points. The local behaviour around
these points can be worked out easily. One finds that Slm pxq „ x k , with k “ ˘ 12 |m ` s| at x “ `1 and
k “ ˘ 21 |m ´ s| at x “ ´1. The boundary conditions allow us to discard the negative exponents. When
x runs to infinity, the equation can be integrated and the boundary condition is fixed analytically
from the other two singular points. It leads to Slm pxq „ exppiaωxq. So the Frobenius method gives a
natural ansatz: one can consider the product of the solutions at singular points times a Taylor series
around x “ ´1
`8
1 1 ÿ
iaωx
Slm pxq “ e p1 ` xq 2 |m`s| p1 ´ xq 2 |m´s| cn p1 ` xqn . (4.86)
n“0

Substituting this ansatz in the angular equation, and equating all coefficients multiplying the same
monomials x k , for any k P N, we find
#
α0 c1 ` β 0 c0 “ 0 ;
(4.87)
αn cn`1 ` β n cn ` γn cn´1 “ 0, @ n P N0 ,

which spans a 3-term recurrence on the coefficients cn . The coefficients tαn , β n , γn u are independent
of x (or θ) and their explicit forms can be found in [113]. If we define rn “ cn`1 {cn , the second equality
becomes αn rn ` β n ` γn {rn´1 “ 0 from which we can extract

´γn ´γn
rn´1 “ “ ¨¨¨ (4.88)
β n ` αn rn α p´γn`1 q
β n ` β n`1n `αn`1 rn`1

It is a continuous fraction which expresses the ratio rn in terms of the next one rn`1 . It converges if
nÑ8
rn ÝÝÝÑ 0, and it turns out to be the case! When a “ 0, ω completely disappears of all the equations,
so the spheroidal equation can be solved numerically to arbitrary precision, which gives at the same
s pcos θq and the constants E s . Otherwise the frequency impacts the recursion, so one has to
time Slm lm
solve simultaneously the spheroidal equation and the radial equation, whose quantized solutions will
give the proper frequencies of the system.

Solving the radial equation

The radial equation has been solved formally by means of a similar method (Frobenius expansion
and continued fraction) by Leaver, also in [113]. The singular points of the radial equation for Rlm prq
are the two roots of ∆, which are the radial position of the event horizon r “ r` and Cauchy hori-
zon r “ r´ , and r “ 8. The point r “ r` is a regular singularity, and the behaviour of the so-
lution in its neighbourhood is fixed by the ingoing boundary condition. On the other hand, the
point at infinity r “ 8 is an essential singularity, where we impose purely outgoing boundary con-

Lecture 4. Rotating black holes 111

 4.3. Black hole spectroscopy

ditions as a definition of quasi-normal modes. The problem consists at solving an ordinary differ-
ential equation together with boundary conditions are two separated locations: it is a boundary
value problem, which you might be familiar with in the context of the quantum description of the
hydrogen atom. Around r “ r` , one can integrate the radial equation to find the two solutions
?
Rlm prq „ riσ` and Rlm prq „ r´s´iσ` where σ` fi pωr` ´ amq{ 1 ´ 4a2 . The ingoing boundary con-
dition rules out the first behaviour. Similarly, we can proceed near infinity to get Rlm prq equal to a
combinaison of r´1´iω e´iωr which is ingoing and then forbidden by the boundary conditions, and
Rlm prq „ r´1´2s`iω eiωr which is outgoing and accepted. After working out the contribution at the
second regular singular point r “ r´ , the Frobenius ansatz can be written as

`8 ˆ ˙n
iωr ´1´s`iω`iσ` ´s´iσ`
ÿ r ´ r`
Rlm prq “ e pr ´ r´ q pr ´ r` q dn . (4.89)
n“0
r ´ r´

Injecting it into the radial Teukolsky equation gives a second set of recursion relations
#
α10 d1 ` β10 d0 “ 0 ;
(4.90)
α1n dn`1 ` β1n dn ` γn1 dn´1 “ 0, @ n P N0 .

The complex coefficients tα1n , β1n , γn1 u again do not depend on r, but well in ω, a, and the separation
s . For a given set of parameters a, m, l, s, the frequency ω is a root of the continued
constants Elmω
fraction equation
α10 γ11
β10 ´ α11 γ21
“ 0. (4.91)
β11 ´ β12 ´¨¨¨
s
Equations (4.88) and (4.91) are two equations for the unknown variables Elmω and ω. They may be
solved simultaneously by non-linear root-search algorithms. The solutions of coupled recurrences
are found to be a set a quantized modes of frequency ωlmN where N is the previously announced
overtone number (N P N). The proof of linear stability of the Kerr black hole in Einstein gravity is
then obtained after checking (numerically) that indeed ImpωlmN q ă 0 for all modes.

Geometric interpretation in the eikonal limit

Let us briefly mention that some analytical quasi-normal modes solutions have been derived, see
the review [121] and references therein. Although the spectrum can be numerically calculated to
arbitrary precision, these analytic solutions give us more insight. As as example, when the mul-
tipolar index l runs to infinity (the so-called eikonal regime or geometric optics approximation), it
is possible to reformulate the Teukolsky radial and angular equations to put them into the form
ε f 2 pzq ` Upzq f pzq “ 0, where ε ! 1 and z is either x, r. The solution of such an equation can be ap-
proximatively computed thanks to WKB expansions f pzq „ exppS0 {ε ` S1 ` εS2 ` ¨ ¨ ¨ q. This allows to
relate the corresponding wave solutions to null geodesics as pioneered by Press in 1971 [122] and de-
rived for Kerr much more recently [123]! Each quasi-normal mode solution pωlmN , l, mq corresponds
to a specific null geodesic with certain conserved quantities pe, j, Qq (remember the definition (4.75))
according to the following dictionary:

B RepωlmN q “ e. The wave frequency is the same as the energy of the corresponding null ray;

Lecture 4. Rotating black holes 112

 4.3. Black hole spectroscopy

B m “ j. The azimuthal number corresponds to the z-axis projection of the orbital momentum ~j
of the null ray. The latter is quantized to get a standing wave in the azimuthal direction φ;

B RepElmω q is related to a combination of Q and m2 . The separation constants Elmω are complex.
Their real part is linked to the Carter constant Q and the square of the axial orbital kinetic mo-
mentum j2 . The angular momentum eigenvalue in the colatitudinal direction also gets quan-
tized in order to get standing waves in the θ direction.

B The decay rate ImpωlmN q “ γL , where γL is a Lyapunov exponent of the orbit, which indicates a
rate of instability of the null geodesic congruence.

Schwarzschild spectroscopy

We present here some concrete numerical results. Let us begin by the Schwarzschild black hole (hence
with a “ 0), see Figure 4.7b.

(a) Quasi-normal modes frequencies for gravitational perturbations (b) Comparison of quasi-normal modes fundamental spectra l “ |s|
(s “ 2). for scalar, vector, and gravitational perturbations (s “ 0, 1, 2).

Figure 4.7: Quasi-normal modes for Schwarzschild black holes. Reproduced from online data [124] and private
communication of E. Berti. Original figure published in [121] © IOP Publishing. Reproduced with permission. All rights reserved.

The graphs are symmetric under real frequency reversal: each positive frequency Repωq mode is as-
sociated with a negative Repωq mode with the same imaginary part. All imaginary parts are negative
(this is the property of linear stability). There are several branches for each l, with increasing over-
tone number N. There is no dependency in m since it factors out of the radial equation (this is due to
SOp3q symmetry).

The first feature worth to notice is that for gravitational perturbations (s “ 2) of static black holes
there are algebraically special modes that have exactly Repωq “ 0. They are given by Mω « ˘ipl ´
1qlpl ` 1qpl ` 2q{12. Such real frequency modes do not exist for other spin s perturbations. For each
l, we will call the modes below these algebraically special modes as the weakly damped modes, and
above, the high damped ones. This qualitative distinction is peculiar to gravitational perturbations:
for other fields s ‰ 2, one approaches monotonically the asymptotic high-damped regime. Here are
some additional properties:

Lecture 4. Rotating black holes 113

 4.3. Black hole spectroscopy

B The most weakly damped modes are the most relevant ones for experimental detection since
they give the leading signal. ln black hole mergers, these modes appear in the ringdown at late
stages after the merger. They have not yet been detected at the time of writing. The dominant
s “ 2 mode is the l “ 2, N “ 0 mode with frequency Mω “ 0.3737 ´ 0.0890i.

B In the large overtone limit N Ñ 8, the modes are highly damped. The asymptotic expansion
of the frequency reads as ω « TH ln 3 ´ i2πTH pN ` 21 q ` O pN ´1{2 q for gravitational perturba-
tions. The appearence of Hawking’s temperature suggests a microscopic interpretation of these
modes.

B In the large l limit, one has the eikonal regime or geometric optics regime where the quasi-
normal modes can be mapped to null geodesics, as we already explained. The leading eikonal
approximation is ω “ Ωγ l ´ ipN ` 1{2qλγ , where Ωγ is the angular frequency of a photon
orbiting a geodesic with same conserved charges as the quasi-normal mode, and γL “ ´ipN `
1{2qλγ is the Lyapunov exponent related to the instability frequency λγ of the orbit.

Kerr spectroscopy

Let us finally discuss the rotating Kerr black hole. The quasi-normal mode spectrum has a rich and
complex structure. The SOp3q symmetry is now slightly broken, so the spectrum depends upon m
in addition to l. This leads to a sort of Zeeman effect (which is the splitting of quantum states of
electrons around atoms due to an external magnetic field) which splits the spectrum of frequencies.

The weakly damped modes are again the most relevant for gravitational wave detection. The N “ 0,
l “ 2, m “ 0 mode is the least damped with Mω020 « 0.4437 ´ 0.0739p1 ´ a{Mq0.3350 , as obtained by
numerical interpolation. There seems to be no known analytic formula in the highly damped regime.
The spacing of the imaginary part of the frequency is not the constant 2πTH but now grows with
a{M. The eikonal regime has been described earlier;

Let us conclude with a brief comment about quasi-normal modes around extremal rotating black
holes. It was shown recently [125] that the spectrum bifurcates in the near-extremal limit into the so-
called “zero-damped” and “damped” quasi-normal modes. The zero-damped quasi-normal modes
as their name indicates have Impωq Ñ 0 in the extremal limit while the damped ones keep a non-
zero imaginary part. This new physical feature arises from the occurence of the near-horizon region
of near-extremal Kerr black hole. The zero-damped quasi-normal modes are emitted from the near-
horizon region, while the damped ones from the “far” extremal Kerr region.

Here we conclude this journey into the marvelous world of rotating black holes. We hope that the
material will trigger the curiosity of the reader to dig in further into these very rich topic full of
discoveries!

Lecture 4. Rotating black holes 114

 4.3. Black hole spectroscopy

References
Many excellent references exist on the Kerr black hole, probably many of which I missed reading.
For a review of the coordinates patches of Kerr and its Penrose diagrams, I would recommend the
lecture notes of Carter of 1973 [86]. For the thermodynamics of Kerr, I particularly like the review of
Wald [126]. For clearly and concisely reviewing many classical features of Kerr and also the Hawking
radiation, I recommend the lecture notes of Townsend of 1997 [127]. A modern review of the Kerr
metric which includes a discussion of perturbation theory and hidden symmetries (together with a
historical review) is given by Teukolsky in 2014 [128]. It largely inspired this lecture.

The study of the near-horizon region of extremal Kerr black holes started by the work of Bardeen,
Teukolsky and Press in 1972 [129] but since it didn’t have direct astrophysical interest it was not until
1999 that the actual decoupled near-horizon region was discovered by Bardeen and Horowitz [130].
Recent interest for this limit is partly due to the Kerr/CFT correspondence in 2009 [95]. The lectures
are mostly based on my review (updated in Dec 2016) [96]. In particular, some of the developments
of [103, 131, 102, 132, 110, 15, 6] were covered.

The quasi-normal modes of the Kerr black hole are reviewed in [121, 133]. A classic reference on the
Newman-Penrose formalism is [134] but unfortunately it uses the overall minus signature convention
for the metric. I mostly followed the appendix of [135] that contains many useful expressions in
overall plus signature. For a review of Frolov’s classification, I would recommend the corresponding
introductory chapter of the book [116]. For more details on the mysterious relationship between
hidden symmetries and separability, see the reviews [136, 137]. A partial database of explicit values
of quasi-normal modes can be found in [124, 121]. For a recent high accuracy study, see [138].

Acknowledgments
G.C. thanks the organizers of the Amsterdam-Brussels-Geneva-Paris Doctoral School and the Inter-
national Solvay Institutes for the opportunity to give these lectures. This work was partly supported
by the ERC Starting Grant 335146 “HoloBHC". G.C. is a Research Associate and A.F. is Research
Fellow of the Fonds de la Recherche Scientifique F.R.S.-FNRS (Belgium).

Lecture 4. Rotating black holes 115

 Bibliography

[1] C. Misner, K. Thorne, and J. Wheeler, Gravitation. W. H. Freeman, 1973.

[2] R. Wald, General Relativity. University of Chicago Press, 1984.

[3] G. Barnich, F. Brandt, and M. Henneaux, “Local BRST cohomology in gauge theories”,
Phys.Rept. 338 (2000) 439–569, arXiv:hep-th/0002245 [hep-th].

[4] G. Compère and D. Marolf, “Setting the boundary free in AdS/CFT”, Class.Quant.Grav. 25
(2008) 195014, arXiv:0805.1902 [hep-th].

[5] G. Compère, L. Donnay, P.-H. Lambert, and W. Schulgin, “Liouville theory beyond the
cosmological horizon”, JHEP 1503 (2015) 158, arXiv:1411.7873 [hep-th].

[6] G. Compère, P. Mao, A. Seraj, and M. M. Sheikh-Jabbari, “Symplectic and Killing symmetries
of AdS3 gravity: holographic vs boundary gravitons”, JHEP 01 (2016) 080, arXiv:1511.06079
[hep-th].

[7] G. Barnich and G. Compère, “Surface charge algebra in gauge theories and thermodynamic
integrability”, J.Math.Phys. 49 (2008) 042901, arXiv:0708.2378 [gr-qc].

[8] V. Iyer and R. M. Wald, “Some properties of Noether charge and a proposal for dynamical
black hole entropy”, Phys.Rev. D50 (1994) 846–864, arXiv:gr-qc/9403028 [gr-qc].

[9] L. F. Abbott and S. Deser, “Stability of Gravity with a Cosmological Constant”, Nucl. Phys.
B195 (1982) 76–96.

[10] S. Deser and B. Tekin, “Gravitational energy in quadratic curvature gravities”, Phys. Rev. Lett.
89 (2002) 101101, arXiv:hep-th/0205318 [hep-th].

[11] S. Deser and B. Tekin, “Energy in generic higher curvature gravity theories”, Phys. Rev. D67
(2003) 084009, arXiv:hep-th/0212292 [hep-th].

[12] P. Di Francesco, P. Mathieu, and D. Senechal, Conformal Field Theory. Graduate Texts in
Contemporary Physics. Springer-Verlag, New York, 1997.
http://www-spires.fnal.gov/spires/find/books/www?cl=QC174.52.C66D5::1997.

[13] J. Lee and R. M. Wald, “Local symmetries and constraints”, J.Math.Phys. 31 (1990) 725–743.

[14] G. Compère, K. Murata, and T. Nishioka, “Central Charges in Extreme Black Hole/CFT
Correspondence”, JHEP 05 (2009) 077, arXiv:0902.1001 [hep-th].

[15] G. Compère, K. Hajian, A. Seraj, and M. M. Sheikh-Jabbari, “Wiggling Throat of Extremal

Black Holes”, JHEP 10 (2015) 093, arXiv:1506.07181 [hep-th]. [JHEP10,093(2015)].

Bibliography 117
 Bibliography

[16] G. Compère and F. Dehouck, “Relaxing the Parity Conditions of Asymptotically Flat Gravity”,
Class. Quant. Grav. 28 (2011) 245016, arXiv:1106.4045 [hep-th]. [Erratum: Class. Quant.
Grav.30,039501(2013)].

[17] G. Barnich and F. Brandt, “Covariant theory of asymptotic symmetries, conservation laws and
central charges”, Nucl.Phys. B633 (2002) 3–82, arXiv:hep-th/0111246 [hep-th].

[18] T. Azeyanagi, G. Compère, N. Ogawa, Y. Tachikawa, and S. Terashima, “Higher-Derivative

Corrections to the Asymptotic Virasoro Symmetry of 4d Extremal Black Holes”, Prog. Theor.
Phys. 122 (2009) 355–384, arXiv:0903.4176 [hep-th].

[19] G. Compère, Symmetries and conservation laws in Lagrangian gauge theories with applications to the
mechanics of black holes and to gravity in three dimensions. PhD thesis, Brussels U., 2007.
arXiv:0708.3153 [hep-th].
http://inspirehep.net/record/758992/files/arXiv:0708.3153.pdf.

[20] R. L. Arnowitt, S. Deser, and C. W. Misner, “Dynamical Structure and Definition of Energy in
General Relativity”, Phys. Rev. 116 (1959) 1322–1330.

[21] T. Regge and C. Teitelboim, “Role of Surface Integrals in the Hamiltonian Formulation of
General Relativity”, Annals Phys. 88 (1974) 286.

[22] J. D. Brown and M. Henneaux, “On the Poisson Brackets of Differentiable Generators in
Classical Field Theory”, J. Math. Phys. 27 (1986) 489–491.

[23] J. D. Brown and M. Henneaux, “Central Charges in the Canonical Realization of Asymptotic
Symmetries: An Example from Three-Dimensional Gravity”, Commun.Math.Phys. 104 (1986)
207–226.

[24] J. M. Maldacena, “The Large N limit of superconformal field theories and supergravity”, Int. J.
Theor. Phys. 38 (1999) 1113–1133, arXiv:hep-th/9711200 [hep-th]. [Adv. Theor. Math.
Phys.2,231(1998)].

[25] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri, and Y. Oz, “Large N field theories,
string theory and gravity”, Phys.Rept. 323 (2000) 183–386, arXiv:hep-th/9905111 [hep-th].

[26] E. Witten, “(2+1)-Dimensional Gravity as an Exactly Soluble System”, Nucl. Phys. B311 (1988)
46.

[27] A. Achucarro and P. K. Townsend, “A Chern-Simons Action for Three-Dimensional anti-De

Sitter Supergravity Theories”, Phys. Lett. B180 (1986) 89.

[28] M. Banados, C. Teitelboim, and J. Zanelli, “The Black hole in three-dimensional space-time”,
Phys.Rev.Lett. 69 (1992) 1849–1851, arXiv:hep-th/9204099 [hep-th].

[29] M. Banados, M. Henneaux, C. Teitelboim, and J. Zanelli, “Geometry of the (2+1) black hole”,
Phys.Rev. D48 no. 6, (1993) 1506–1525, arXiv:gr-qc/9302012 [gr-qc].

[30] A. Strominger, “Black hole entropy from near horizon microstates”, JHEP 02 (1998) 009,
arXiv:hep-th/9712251 [hep-th].

Bibliography 118
 Bibliography

[31] E. Witten, “Three-Dimensional Gravity Revisited”, arXiv:0706.3359 [hep-th].

[32] A. Maloney and E. Witten, “Quantum Gravity Partition Functions in Three Dimensions”,
JHEP 02 (2010) 029, arXiv:0712.0155 [hep-th].

[33] G. Barnich, A. Gomberoff, and H. A. Gonzalez, “The Flat limit of three dimensional
asymptotically anti-de Sitter spacetimes”, Phys. Rev. D86 (2012) 024020, arXiv:1204.3288
[gr-qc].

[34] G. Compère, W. Song, and A. Strominger, “New Boundary Conditions for AdS3”, JHEP 05
(2013) 152, arXiv:1303.2662 [hep-th].

[35] G. Barnich and G. Compère, “Classical central extension for asymptotic symmetries at null
infinity in three spacetime dimensions”, Class. Quant. Grav. 24 (2007) F15–F23,
arXiv:gr-qc/0610130 [gr-qc].

[36] G. Barnich and C. Troessaert, “Aspects of the BMS/CFT correspondence”, JHEP 1005 (2010)
062, arXiv:1001.1541 [hep-th].

[37] B. Oblak, BMS Particles in Three Dimensions. PhD thesis, Brussels U., 2016. arXiv:1610.08526
[hep-th]. http://inspirehep.net/record/1494790/files/arXiv:1610.08526.pdf.

[38] D. Ida, “No black hole theorem in three-dimensional gravity”, Phys. Rev. Lett. 85 (2000)
3758–3760, arXiv:gr-qc/0005129 [gr-qc].

[39] G. Barnich, “Entropy of three-dimensional asymptotically flat cosmological solutions”, JHEP

10 (2012) 095, arXiv:1208.4371 [hep-th].

[40] A. Bagchi, S. Detournay, R. Fareghbal, and J. Simón, “Holography of 3D Flat Cosmological

Horizons”, Phys. Rev. Lett. 110 no. 14, (2013) 141302, arXiv:1208.4372 [hep-th].

[41] S. Deser, R. Jackiw, and G. ’t Hooft, “Three-Dimensional Einstein Gravity: Dynamics of Flat
Space”, Annals Phys. 152 (1984) 220.

[42] S. Deser and R. Jackiw, “Three-Dimensional Cosmological Gravity: Dynamics of Constant

Curvature”, Annals Phys. 153 (1984) 405–416.

[43] J. de Boer and J. I. Jottar, “Thermodynamics of higher spin black holes in AdS3 ”, JHEP 01
(2014) 023, arXiv:1302.0816 [hep-th].

[44] A. Ashtekar, J. Bicak, and B. G. Schmidt, “Asymptotic structure of symmetry reduced general
relativity”, Phys. Rev. D55 (1997) 669–686, arXiv:gr-qc/9608042 [gr-qc].

[45] G. Compère and A. Fiorucci, “Asymptotically flat spacetimes with BMS3 symmetry”, Class.
Quant. Grav. 34 no. 20, (2017) 204002, arXiv:1705.06217 [hep-th].

[46] D. Christodoulou and S. Klainerman, The Global nonlinear stability of the Minkowski space.
Princeton Legacy Library, 1993.

[47] A. Strominger, “On BMS Invariance of Gravitational Scattering”, JHEP 07 (2014) 152,
arXiv:1312.2229 [hep-th].

Bibliography 119
 Bibliography

[48] G. Compère and J. Long, “Vacua of the gravitational field”, JHEP 07 (2016) 137,
arXiv:1601.04958 [hep-th].

[49] G. Compère and J. Long, “Classical static final state of collapse with supertranslation
memory”, Class. Quant. Grav. 33 no. 19, (2016) 195001, arXiv:1602.05197 [gr-qc].

[50] J. de Boer and S. N. Solodukhin, “A Holographic reduction of Minkowski space-time”, Nucl.

Phys. B665 (2003) 545–593, arXiv:hep-th/0303006 [hep-th].

[51] A. Strominger and A. Zhiboedov, “Superrotations and Black Hole Pair Creation”, Class.
Quant. Grav. 34 no. 6, (2017) 064002, arXiv:1610.00639 [hep-th].

[52] R. Penrose, “The geometry of impulsive gravitational waves”, in General relativity: Papers in
honour of J.L. Synge, L. O’Raifeartaigh, ed., pp. 101–115. 1972.

[53] G. Compère, A. Fiorucci, and R. Ruzziconi, “Superboost transitions, refraction memory and
super-Lorentz charge algebra”, arXiv:1810.00377 [hep-th].

[54] M. Campiglia and A. Laddha, “New symmetries for the Gravitational S-matrix”, JHEP 04
(2015) 076, arXiv:1502.02318 [hep-th].

[55] M. Campiglia and A. Laddha, “Asymptotic symmetries and subleading soft graviton
theorem”, Phys. Rev. D90 no. 12, (2014) 124028, arXiv:1408.2228 [hep-th].

[56] F. Cachazo and A. Strominger, “Evidence for a New Soft Graviton Theorem”,
arXiv:1404.4091 [hep-th].

[57] Y. B. Zel’dovich and A. G. Polnarev, “Radiation of gravitational waves by a cluster of

superdense stars”, Soviet Astronomy 18 (Aug., 1974) 17.

[58] D. Christodoulou, “Nonlinear nature of gravitation and gravitational wave experiments”,

Phys. Rev. Lett. 67 (1991) 1486–1489.

[59] K. S. Thorne, “Gravitational-wave bursts with memory: The Christodoulou effect”, Phys. Rev.
D45 no. 2, (1992) 520–524.

[60] L. Blanchet and T. Damour, “Tail Transported Temporal Correlations in the Dynamics of a
Gravitating System”, Phys. Rev. D37 (1988) 1410.

[61] L. Blanchet and T. Damour, “Hereditary effects in gravitational radiation”, Phys. Rev. D46
(1992) 4304–4319.

[62] P. D. Lasky, E. Thrane, Y. Levin, J. Blackman, and Y. Chen, “Detecting gravitational-wave

memory with LIGO: implications of GW150914”, Phys. Rev. Lett. 117 no. 6, (2016) 061102,
arXiv:1605.01415 [astro-ph.HE].

[63] S. Pasterski, A. Strominger, and A. Zhiboedov, “New Gravitational Memories”, JHEP 12 (2016)
053, arXiv:1502.06120 [hep-th].

[64] D. A. Nichols, “Spin memory effect for compact binaries in the post-Newtonian
approximation”, Phys. Rev. D95 no. 8, (2017) 084048, arXiv:1702.03300 [gr-qc].

Bibliography 120
 Bibliography

[65] T. He, V. Lysov, P. Mitra, and A. Strominger, “BMS supertranslations and WeinbergÕs soft
graviton theorem”, JHEP 05 (2015) 151, arXiv:1401.7026 [hep-th].

[66] S. Weinberg, “Infrared photons and gravitons”, Phys. Rev. 140 (Oct, 1965) B516–B524.
https://link.aps.org/doi/10.1103/PhysRev.140.B516.

[67] C. Troessaert, “The BMS4 algebra at spatial infinity”, arXiv:1704.06223 [hep-th].

[68] M. Henneaux and C. Troessaert, “BMS Group at Spatial Infinity: the Hamiltonian (ADM)
approach”, JHEP 03 (2018) 147, arXiv:1801.03718 [gr-qc].

[69] S. W. Hawking, M. J. Perry, and A. Strominger, “Soft Hair on Black Holes”, Phys. Rev. Lett. 116
no. 23, (2016) 231301, arXiv:1601.00921 [hep-th].

[70] A. Strominger, “Black Hole Information Revisited”, arXiv:1706.07143 [hep-th].

[71] M. Mirbabayi and M. Porrati, “Dressed Hard States and Black Hole Soft Hair”, Phys. Rev. Lett.
117 no. 21, (2016) 211301, arXiv:1607.03120 [hep-th].

[72] R. Bousso and M. Porrati, “Soft Hair as a Soft Wig”, Class. Quant. Grav. 34 no. 20, (2017)
204001, arXiv:1706.00436 [hep-th].

[73] H. Bondi, M. G. J. van der Burg, and A. W. K. Metzner, “Gravitational waves in general
relativity. 7. Waves from axisymmetric isolated systems”, Proc. Roy. Soc. Lond. A269 (1962)
21–52.

[74] R. K. Sachs, “Gravitational waves in general relativity. 8. Waves in asymptotically flat

space-times”, Proc. Roy. Soc. Lond. A270 (1962) 103–126.

[75] G. Barnich and C. Troessaert, “Symmetries of asymptotically flat 4 dimensional spacetimes at

null infinity revisited”, Phys. Rev. Lett. 105 (2010) 111103, arXiv:0909.2617 [gr-qc].

[76] L. Bieri, P. Chen, and S.-T. Yau, “Null Asymptotics of Solutions of the Einstein-Maxwell
Equations in General Relativity and Gravitational Radiation”, Adv. Theor. Math. Phys. 15 no. 4,
(2011) 1085–1113, arXiv:1011.2267 [math.DG].

[77] A. Strominger and A. Zhiboedov, “Gravitational Memory, BMS Supertranslations and Soft
Theorems”, JHEP 01 (2016) 086, arXiv:1411.5745 [hep-th].

[78] G. Barnich and C. Troessaert, “BMS charge algebra”, JHEP 12 (2011) 105, arXiv:1106.0213
[hep-th].

[79] E. E. Flanagan and D. A. Nichols, “Conserved charges of the extended Bondi-Metzner-Sachs

algebra”, Phys. Rev. D95 no. 4, (2017) 044002, arXiv:1510.03386 [hep-th].

[80] A. Strominger, “Lectures on the Infrared Structure of Gravity and Gauge Theory”,
arXiv:1703.05448 [hep-th].

[81] S. Hawking, “Particle Creation by Black Holes”, Commun.Math.Phys. 43 (1975) 199–220.

Bibliography 121
 Bibliography

[82] J. M. Bardeen, B. Carter, and S. W. Hawking, “The four laws of black hole mechanics”,
Communications in Mathematical Physics 31 (June, 1973) 161–170.

[83] J. D. Bekenstein, “Black holes and the second law.”, Nuovo Cimento Lettere 4 (1972) 737–740.

[84] Y. B. Zel’Dovich, “Generation of Waves by a Rotating Body”, Soviet Journal of Experimental and
Theoretical Physics Letters 14 (Aug., 1971) 180.

[85] R. D. Blandford and R. L. Znajek, “Electromagnetic extraction of energy from Kerr black
holes”, Monthly Notices of the Royal Astron. Soc. 179 (May, 1977) 433–456.

[86] B. Carter, “Black hole equilibrium states.”, in Black Holes (Les Astres Occlus), C. Dewitt and
B. S. Dewitt, eds., pp. 57–214. 1973.

[87] A. Ori, “Evolution of linear gravitational and electromagnetic perturbations inside a Kerr
black hole”, Phys.Rev.D. 61 no. 2, (Jan., 2000) 024001.

[88] M. W. Choptuik, “Universality and scaling in gravitational collapse of a massless scalar field”,
Phys. Rev. Lett. 70 (1993) 9–12.

[89] T. Regge and J. A. Wheeler, “Stability of a schwarzschild singularity”, Phys. Rev. 108 (Nov,
1957) 1063–1069. https://link.aps.org/doi/10.1103/PhysRev.108.1063.

[90] C. V. Vishveshwara, “Stability of the schwarzschild metric”, Phys. Rev. D1 (1970) 2870–2879.

[91] B. F. Whiting, “Mode Stability of the Kerr Black Hole”, J. Math. Phys. 30 (1989) 1301.

[92] M. Dafermos, I. Rodnianski, and Y. Shlapentokh-Rothman, “Decay for solutions of the wave
equation on Kerr exterior spacetimes III: The full subextremal case |a| < M”,
arXiv:1402.7034 [gr-qc].

[93] S. Aretakis, “The Wave Equation on Extreme Reissner-Nordstrom Black Hole Spacetimes:
Stability and Instability Results”, arXiv:1006.0283 [math.AP].

[94] K. Murata, H. S. Reall, and N. Tanahashi, “What happens at the horizon(s) of an extreme black
hole?”, Class. Quant. Grav. 30 (2013) 235007, arXiv:1307.6800 [gr-qc].

[95] M. Guica, T. Hartman, W. Song, and A. Strominger, “The Kerr/CFT Correspondence”, Phys.
Rev. D80 (2009) 124008, arXiv:0809.4266 [hep-th].

[96] G. Compère, “The Kerr/CFT correspondence and its extensions: a comprehensive review”,
Living Rev. Rel. 15 (2012) 11, arXiv:1203.3561 [hep-th].

[97] S. E. Gralla, A. Lupsasca, and A. Strominger, “Observational Signature of High Spin at the
Event Horizon Telescope”, arXiv:1710.11112 [astro-ph.HE].

[98] G. Compère, K. Fransen, T. Hertog, and J. Long, “Gravitational waves from plunges into
Gargantua”, arXiv:1712.07130 [gr-qc].

[99] A. Lupsasca, A. P. Porfyriadis, and Y. Shi, “Critical Emission from a High-Spin Black Hole”,
arXiv:1712.10182 [gr-qc].

Bibliography 122
 Bibliography

[100] K. S. Thorne, “Disk accretion onto a black hole. 2. Evolution of the hole.”, Astrophys. J. 191
(1974) 507–520.

[101] K. Thorne and C. Nolan, The Science of Interstellar. W. W. Norton, 2014.

https://books.google.be/books?id=PbWYBAAAQBAJ.

[102] A. J. Amsel, G. T. Horowitz, D. Marolf, and M. M. Roberts, “No Dynamics in the Extremal
Kerr Throat”, JHEP 09 (2009) 044, arXiv:0906.2376 [hep-th].

[103] I. Bredberg, T. Hartman, W. Song, and A. Strominger, “Black Hole Superradiance From
Kerr/CFT”, JHEP 04 (2010) 019, arXiv:0907.3477 [hep-th].

[104] T. Piran, J. Shaham, and J. Katz, “High Efficiency of the Penrose Mechanism for Particle
Collisions”, Astrophysical Journal Letters 196 (Mar., 1975) L107.

[105] T. Hartman, C. A. Keller, and B. Stoica, “Universal Spectrum of 2d Conformal Field Theory in
the Large c Limit”, JHEP 09 (2014) 118, arXiv:1405.5137 [hep-th].

[106] A. Castro, A. Maloney, and A. Strominger, “Hidden Conformal Symmetry of the Kerr Black
Hole”, Phys. Rev. D82 (2010) 024008, arXiv:1004.0996 [hep-th].

[107] M. Cvetic and F. Larsen, “Conformal Symmetry for Black Holes in Four Dimensions”, JHEP 09
(2012) 076, arXiv:1112.4846 [hep-th].

[108] S. Pasterski, S.-H. Shao, and A. Strominger, “Flat Space Amplitudes and Conformal Symmetry
of the Celestial Sphere”, Phys. Rev. D96 no. 6, (2017) 065026, arXiv:1701.00049 [hep-th].

[109] D. Kapec, P. Mitra, A.-M. Raclariu, and A. Strominger, “2D Stress Tensor for 4D Gravity”,
Phys. Rev. Lett. 119 no. 12, (2017) 121601, arXiv:1609.00282 [hep-th].

[110] G. Compère, K. Hajian, A. Seraj, and M. Sheikh-Jabbari, “Extremal Rotating Black Holes in the
Near-Horizon Limit: Phase Space and Symmetry Algebra”, arXiv:1503.07861 [hep-th].

[111] Virgo, LIGO Scientific Collaboration, B. P. Abbott et al., “Observation of Gravitational Waves
from a Binary Black Hole Merger”, Phys. Rev. Lett. 116 no. 6, (2016) 061102, arXiv:1602.03837
[gr-qc].

[112] S. A. Teukolsky, “Rotating black holes - separable wave equations for gravitational and
electromagnetic perturbations”, Phys. Rev. Lett. 29 (1972) 1114–1118.

[113] E. W. Leaver, “An Analytic representation for the quasi normal modes of Kerr black holes”,
Proc. Roy. Soc. Lond. A402 (1985) 285–298.

[114] A. Z. Petrov, “The classification of spaces defining gravitational fields”, Uchenye Zapiski
Kazanskogo Gosudarstvennogo Universiteta im. V. I. Ulyanovicha-Lenina [Scientific Proceedings of
Kazan State University, named after V.I. Ulyanov-Lenin], 114, (8) 55-69 (1954)). Jubilee (1804-1954)
Collection. 114 (1954) .

[115] A. Coley, R. Milson, V. Pravda, and A. Pravdova, “Classification of the Weyl tensor in higher
dimensions”, Class. Quant. Grav. 21 (2004) L35–L42, arXiv:gr-qc/0401008 [gr-qc].

Bibliography 123
 Bibliography

[116] H. Stephani, D. Kramer, M. A. H. MacCallum, C. Hoenselaers, and E. Herlt, Exact solutions of

Einstein’s field equations. Cambridge University Press, 2004. http://www.cambridge.org/gb/
academic/subjects/physics/theoretical-physics-and-mathematical-physics/
exact-solutions-einsteins-field-equations-2nd-edition?format=PB.

[117] B. Carter, “Global structure of the Kerr family of gravitational fields”, Phys. Rev. 174 (1968)
1559–1571.

[118] R. Fujita and W. Hikida, “Analytical solutions of bound timelike geodesic orbits in Kerr
spacetime”, Class. Quant. Grav. 26 (2009) 135002, arXiv:0906.1420 [gr-qc].

[119] B. Carter, “Killing Tensor Quantum Numbers and Conserved Currents in Curved Space”,
Phys. Rev. D16 (1977) 3395–3414.

[120] W. Kinnersley, “Type D Vacuum Metrics”, J. Math. Phys. 10 (1969) 1195–1203.

[121] E. Berti, V. Cardoso, and A. O. Starinets, “Quasinormal modes of black holes and black
branes”, Class. Quant. Grav. 26 (2009) 163001, arXiv:0905.2975 [gr-qc].

[122] W. H. Press, “Long Wave Trains of Gravitational Waves from a Vibrating Black Hole”,
Astrophys. J. 170 (Dec., 1971) L105.

[123] H. Yang, D. A. Nichols, F. Zhang, A. Zimmerman, Z. Zhang, and Y. Chen,

“Quasinormal-mode spectrum of Kerr black holes and its geometric interpretation”, Phys. Rev.
D86 (2012) 104006, arXiv:1207.4253 [gr-qc].

[124] “E. Berti’s homepage, Ringdown.” https://pages.jh.edu/~eberti2/ringdown/. Accessed:

2018-09-05.

[125] H. Yang, A. Zimmerman, A. Zenginoglu, F. Zhang, E. Berti, and Y. Chen, “Quasinormal

modes of nearly extremal Kerr spacetimes: spectrum bifurcation and power-law ringdown”,
Phys. Rev. D88 no. 4, (2013) 044047, arXiv:1307.8086 [gr-qc]. [Phys. Rev.D88,044047(2013)].

[126] R. M. Wald, “The thermodynamics of black holes”, Living Rev. Rel. 4 (2001) 6,
arXiv:gr-qc/9912119 [gr-qc].

[127] P. K. Townsend, “Black holes: Lecture notes”, arXiv:gr-qc/9707012 [gr-qc].

[128] S. A. Teukolsky, “The Kerr Metric”, Class. Quant. Grav. 32 no. 12, (2015) 124006,
arXiv:1410.2130 [gr-qc].

[129] J. M. Bardeen, W. H. Press, and S. A. Teukolsky, “Rotating black holes: Locally nonrotating
frames, energy extraction, and scalar synchrotron radiation”, Astrophys. J. 178 (1972) 347.

[130] J. M. Bardeen and G. T. Horowitz, “The Extreme Kerr throat geometry: A Vacuum analog of
AdS(2) x S2 ”, Phys.Rev. D60 (1999) 104030, arXiv:hep-th/9905099 [hep-th].

[131] V. Balasubramanian, J. de Boer, M. M. Sheikh-Jabbari, and J. Simon, “What is a chiral 2d CFT?

And what does it have to do with extremal black holes?”, JHEP 02 (2010) 017,
arXiv:0906.3272 [hep-th].

Bibliography 124
 Bibliography

[132] O. J. C. Dias, H. S. Reall, and J. E. Santos, “Kerr-CFT and gravitational perturbations”, JHEP 08
(2009) 101, arXiv:0906.2380 [hep-th].

[133] R. A. Konoplya and A. Zhidenko, “Quasinormal modes of black holes: From astrophysics to
string theory”, Rev. Mod. Phys. 83 (2011) 793–836, arXiv:1102.4014 [gr-qc].

[134] S. Chandrasekhar, “The mathematical theory of black holes”, in Oxford, UK: Clarendon (1992)
646 p., OXFORD, UK: CLARENDON (1985) 646 P. 1985.

[135] C. Merlin, A. Ori, L. Barack, A. Pound, and M. van de Meent, “Completion of metric
reconstruction for a particle orbiting a Kerr black hole”, Phys. Rev. D94 no. 10, (2016) 104066,
arXiv:1609.01227 [gr-qc].

[136] E. G. Kalnins, W. M. Jr., and G. C. Williams, “Separability of Wave Equations”, Fundam. Theor.
Phys. 100 (1999) 33–52.

[137] V. Frolov, P. Krtous, and D. Kubiznak, “Black holes, hidden symmetries, and complete
integrability”, Living Rev. Rel. 20 no. 1, (2017) 6, arXiv:1705.05482 [gr-qc].

[138] G. B. Cook and M. Zalutskiy, “Gravitational perturbations of the Kerr geometry:

High-accuracy study”, Phys. Rev. D90 no. 12, (2014) 124021, arXiv:1410.7698 [gr-qc].

Bibliography 125