General Relativity, Cosmology and Astrophysics Perspectives 100 Years After Einstein's Stay in Prague

Fundamental Theories of Physics 177
Jiří Bičák
Tomáš Ledvinka Editors
General
Relativity,
Cosmology and
Astrophysics
Perspectives 100 years after Einstein‘s
stay in Prague
Fundamental Theories of Physics
Volume 177
Series editors
Henk van Beijeren, Utrecht, The Netherlands
Philippe Blanchard, Bielefeld, Germany
Paul Busch, York, UK
Bob Coecke, Oxford, UK
Dennis Dieks, Utrecht, The Netherlands
Detlef Dürr, München, Germany
Roman Frigg, London, UK
Christopher A. Fuchs, Waterloo, Canada
Giancarlo Ghirardi, Trieste, Italy
Domenico Giulini, Hannover, Germany
Gregg Jaeger, Boston, USA
Claus Kiefer, Cologne, Germany
Klaas Landsman, Nijmegen, The Netherlands
Christian Maes, Leuven, Belgium
Hermann Nicolai, Golm, Germany
Vesselin Petkov, Montreal, Canada
Alwyn van der Merwe, Denver, USA
Rainer Verch, Leipzig, Germany
Reinhard Werner, Hannover, Germany
Christian Wuthrich, La Jolla, USA
For further volumes:

http://www.springer.com/series/6001
Jiří Bičák Tomáš Ledvinka
•
Editors
General Relativity,
Cosmology and Astrophysics
Perspectives 100 years after Einstein’s stay
in Prague
123
Editors
Jiří Bičák Tomáš Ledvinka
Faculty of Mathematics and Physics Faculty of Mathematics and Physics
Charles University Charles University
Praha 8 Praha 8
Czech Republic Czech Republic
ISBN 978-3-319-06348-5 ISBN 978-3-319-06349-2 (eBook)

DOI 10.1007/978-3-319-06349-2
Springer Cham Heidelberg New York Dordrecht London
Library of Congress Control Number:2014938217
Springer International Publishing Switzerland 2014

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must
always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
A few days before leaving Prague, after 15 months spent at the German part of
Charles-Ferdinand University, Albert Einstein submitted a paper titled Relativity
and Gravitation. Reply to a Comment by M. Abraham. It was received by Annalen
der Physik on July 4, 1912. Here Einstein summarized the contemporary state of
his relativistic theory of gravitation and, remarkably, anticipated what a future
theory of gravity should look like.1
‘‘Relativity and Gravitation: 100 years after Einstein in Prague,’’ was the name
of the conference held in Prague on June 25–29, 2012, inspired by the title, date,
and significance of this last of Einstein’s Prague papers. The aim of the conference
was twofold. First, it was to review the present status of the general theory of
relativity (both classical and quantum) and its applications in cosmology and
astrophysics from a broad perspective. The second aim was to present the newest
results in each of these fields. This volume is based on the invited plenary lectures
at the conference. In another volume, ‘‘Relativity and Gravitation: 100 years after
Einstein’s stay in Prague,’’ appearing in the ‘‘Springer Proceedings in Physics,’’
articles based on contributed talks and posters are included; more on cultural and
other events associated with the conference is recalled therein.
The articles included in this volume represent a broad and highly qualified view
on the present state of general relativity, quantum gravity, and their cosmological
and astrophysical implications. As such, it may serve as a valuable source of
knowledge and inspiration for experts in these fields, as well as an advanced
source of information for young researchers.
The contents is divided into four broad parts: (i) Gravity and Prague, (ii)
Classical General Relativity, (iii) Cosmology and Quantum Gravity, and (iv)
Numerical Relativity and Relativistic Astrophysics.
1
See the contribution by J. Bičák in this Volume.
v
vi Preface
Gravity and Prague
In the first contribution, Julian Barbour ‘‘honors’’ Kepler and Mach for their
work and fundamental discoveries made in Prague. Barbour starts with Greek
planetary astronomy, continues with Copernicus, and comes to a profound
description of the work of Kepler, ‘‘the true discoverer of heliocentricity.’’ After a
short intermezzo about Doppler, he comes to Ernst Mach. Mach profoundly
influenced several generations of experimentalists (not only) in Prague. However,
within the context of the conference, the most important issue was Mach’s
influence on Einstein in the formulation of general relativity. At the same time,
though, Einstein sometimes ‘‘distorted’’ Mach. A novel conception of Mach’s
principle within geometrodynamics, called by Barbour the ‘‘shape dynamics,’’
concludes Barbour’s excursion through almost 2,200 years.
Jiří Bičák describes how and why Einstein was invited to Prague, and what his
influence was on Czech science and culture. The main themes that occupied him
then were the principle of equivalence, the bending of light, gravitational lensing,
gravitational redshift, and frame dragging effects. In the article, these topics are
discussed from a present-day perspective. Perhaps most importantly, just before
leaving Prague, Einstein summarized his views on what basic features a new
theory of gravity should possess, including the invariance with respect to a larger
group than the Lorentz group, the local significance of the equivalence principle,
or the nonlinearity of the field equation for gravity.
Classical General Relativity
The problem of measurement, of precise definition of observers and observables

they employ, has been one of the most distinct features of both special and general
relativity since their birth. Donato Bini summarizes the results obtained by his
colleagues and him during the last two decades on the ‘‘measurement process.’’
The process involves clearly defined geometrical and physical quantities arising
from an identification of ‘‘space’’ and ‘‘time’’ relative to a given observer within a
congruence of timelike worldlines.
A novel theme, unimaginable to be discussed 100 or even 20 years ago, is
reviewed in Gary Gibbons’ article: The Role of General Relativity in Other Parts
of Physics. Gibbons concentrates on a number of specific problems in which
geometrical ideas are employed: the description of shallow water waves is anal-
ogous to the behavior of the rays in the Schwarzschild metric or around straight
cosmic strings. In optics, examples of a left-handed light moving in a medium with
a negative refractive index can be described with the help of hyperbolic or
Lobachevsky space. Zermelo’s problem of minimizing the travel time of a boat
moving with a fixed speed in a Riemannian metric under the presence of a ‘‘wind’’
is discussed using Finsler geometry. Other problems include: invisibility cloaks,
Preface vii
hyperbolic metamaterials, gravitational kinks, Bloch walls, liquid crystal droplets,

and last but not least the popular subject of graphene.
A very old theme in general relativity is the N-body problem. For over 30 years
a leading figure in this field has been Thibault Damour. He first reviews ongoing
post-Newtonian calculations, continuing to still higher orders in v/c, and combined
with formalisms yielding gravitational waveforms. Next he describes the effective
field theory approach employing diagrammatic methods of quantum field theory,
numerical relativity simulations, and gravitational self-force theory. The main
attention is then paid to the ‘‘Effective One Body’’ (EOB), formalism that Damour
and his collaborators started to develop at the end of the 1990s. The goal of the
EOB formalism is to obtain an analytical description of the motion and radiation
of coalescing binaries during the whole process, from inspiral to the final
black-hole ring-down. The results obtained by employing the EOB formalism are
compared with those coming from numerical relativity and self-force computa-
tions. The ways in which EOB theory may progress are indicated and the
conclusion is taken from Henri Poincaré: ‘‘There are no solved problems, there are
only more-or- less solved problems... .’’
The article by Leor Barack on the gravitational self-force follows. Originally,
the self-force theory was developed for so-called ‘‘Extreme Mass Ratio Inspirals’’
(EMRIs), in which a test particle (say a 10M black hole) moves along a geodesic
of a stationary background geometry of a large mass (say a 106M black hole).
If the mass of the particle is taken into account, the effect of the perturbation of the
background geometry on the particle gives rise to a gravitational self-force.
To calculate the resulting deviation from the geodesic motion is a formidable task.
Barack, a principal author of a powerful, practical method for doing this for EMRI
orbits gives a brief but comprehensive review of both successes and difficulties.
As an illustration, important cases of the self-force effects are discussed: the
induced shift in the frequency of the innermost stable circular orbit or the
correction to the periastron shift.
For a number of years, the problem of motion has been studied by Gerhard
Schäfer and his collaborators concentrating on the use of the Hamiltonian treat-
ments. The Hamiltonian formalisms of Arnowitt-Deser and Misner (ADM), Dirac
and Schwinger are compared. The results based on the ADM approach are
combined with the post-Newtonian/post-Minkowskian approximations. Lastly,
Schäfer tells how the complicated problem of incorporating the spin of the
particles into the formalism has been recently analyzed.
The article by Marc Mars focuses on geometric inequalities involving physical
quantities like mass, charge, area, or angular momentum. This work is interwoven
by definitions, propositions, lemmas, and theorems, though it also contains helpful
intuitive remarks. The best known inequalities are the famous positive mass the-
orem and the Penrose inequality (the theorem under the presence of trapped sur-
faces). More recently, with new concepts of Marginally Outer Trapped Surfaces
(MOTS) and dynamical horizons, a great interest arose in inequalities considering
area and angular momentum. Other inequalities involving charge, the cosmolog-
ical constant, and the topology of MOTS’s, are concisely reviewed.
viii Preface
It has been a long-standing problem whether the spin–spin interaction between

two black holes can balance their gravitational attraction. The article by Gernot
Neugebauer and Jörg Hennig, based on Neugebauer’s talk in Prague, provides a
survey of the numerous papers addressing this problem in the past. It also describes
the rigorous non-existence proof based on the formulation of a boundary value
problem for the nonlinear Ernst equation under the assumption of the existence of
two disconnected Killing horizons.
Robert Wald’s article discusses a recently obtained dynamical stability crite-
rion for black holes in D C4 spacetime dimensions. This relates stability with
respect to axisymmetric perturbations to the positivity of a certain canonical
energy. This energy is determined by second variations in the mass, angular
momentum, and horizon area like those appearing in the first law of black hole
thermodynamics. One consequence is that black branes, corresponding to ther-
modynamically unstable black holes, are dynamically also unstable. Wald con-
cludes that ‘‘The remarkable relationship between the laws of black hole physics
and the laws of thermodynamics […] extends to dynamical stability.’’
A brief contribution by Piotr Bizoń and Andrzej Rostworowski, based on the
talk given by Rostworowski, summarizes their numerical and perturbative work on
the instability of anti-de Sitter spacetime based on numerical and perturbative
calculations. Results for spherically symmetric massless scalar fields strongly
indicate that anti-de Sitter space is unstable to arbitrarily small perturbations—
eventually forming black holes.
Higher dimensional black holes within classical general relativity in higher
dimensions are discussed in the last two articles of this part. Harvey Reall ana-
lyzes stationary vacuum solutions within (i) Kaluza–Klein theory and (ii) general
relativity. In addition to Myers-Perry black holes, solutions include the famous
black rings and black Saturns found by Reall and Roberto Emparan. Generaliza-
tions of these are mentioned, as well as the problem of instabilities and the search
for perturbative solutions.
The next article is based on the talk by Valeri Frolov. It concentrates on black
holes with topologically spherical horizons but non-vanishing cosmological con-
stant and NUT parameter. Crucial is the existence of the principal conformal
Killing-Yano tensor which is admitted by Kerr-NUT-AdS metrics in four and
higher dimensions. This object enabled Carter in 1968 to separate geodesic and
wave equations in the four-dimensional Kerr geometry. The principal conformal
Killing-Yano tensor allows similar results to be derived in higher dimensions as
well.
Cosmology and Quantum Gravity
Lars Andersson’s article deals with cosmological models and their stability. The
principle of equivalence and Mach’s principle are mentioned and shown to have
roles different from the ‘‘hierarchy of cosmological principles.’’ A brief discussion
Preface ix
follows of cosmological models and issues such as the coincidence problem and
the role of inhomogeneities. The second part has a more technical character:
The asymptotics and nonlinear stability of various models is surveyed, including
de Sitter space, the deformed Milne model, and generalized anisotropic Kasner
models. The present status of the Belinskii, Khalatnikov, Lifshitz proposal is also
discussed.
The contribution by Misao Sasaki discusses inflation and the birth of cosmo-
logical perturbations. Inflationary universe, despite various objections, is being
accepted by more and more active cosmologists not only because it helps to
explain various problems of the standard big-bang theory, but also appears to be in
agreement with new observational data on the anisotropy of the microwave
background. Sasaki describes slow-roll inflation, the curvature perturbations
arising from vacuum fluctuations of the inflation field, and tensor perturbations of
the metric. Possible primordial non-Gaussianities, if observed, could significantly
constrain cosmological models. The review concludes with a brief description of a
powerful formalism which allows curvature perturbations to be calculated on
superhorizon scales.
What preceded inflation? This question is asked and an answer is suggested in
the article on Loop Quantum Cosmology (LQC) by Abhay Ashtekar. His dis-
covery of new variables in 1986 initiated the birth of the Loop Quantum Gravity
(LQG). Ashtekar suggests that ‘‘even though we are far from a complete theory [of
quantum gravity], advances can occur by focusing on specific physical problems.’’
One ‘‘grand’’ problem is the LQC based on a truncated LQG. It concentrates on (i)
the resolution of the ‘‘initial’’ singularity, (ii) the formulation of effective LQC
dynamics which leads to inflation, and (iii) the extension of cosmological
perturbation theory to the Planck regime. Ashtekar reviews progress in these
directions. He also indicates how LQC relates initial conditions (at a bounce) with
observations.
One open aspect of inflation is the issue of how quantum fluctuations in the
inflaton field transmute into observed classical inhomogeneities. This problem is
addressed by Daniel Sudarsky. In the standard ‘‘philosophy’’ of quantum theory
there exists a collapse of the wave function during a measuring process. How can
this happen in the Universe? Some believe in Everett’s many world interpretation.
More radical ideas require ‘‘novel physics,’’ such as gravity-induced collapse of
the wave function. Such an approach is adopted by Sudarsky. He investigates it
within a semiclassical treatment of gravity interacting with quantum fields during
inflation.
A broad view on the state of quantum gravity (QG) from a particle physics
perspective is given by Hermann Nicolai. It covers a large territory in a fairly
nontechnical style. There are difficulties in both general relativity and quantum
field theory associated with the use of a continuum at and below Planck-scale
distances. The principle differences in the approaches to QG between two main
candidates, string theory and LQG, are elucidated and considered to be a sign that
‘‘we are probably still very far from the correct answer!’’ A number of other issues
that should be answered by a future theory of QG include the divergence problem
x Preface
of perturbative general relativity, the hierarchy problem (the smallness of G), and
the question of whether the Standard Model of particle physics remains true up to
the Planck scale. The incompleteness of the Standard Model is considered as ‘‘one
of the strongest arguments in favor of quantizing gravity and searching for new
concepts replacing classical notions of space and time.’’
Numerical Relativity and Relativistic Astrophysics
Relativistic astrophysics celebrated its 50th birthday in December 2013. Seven

articles included in this part demonstrate remarkably well the enormous progress
this field experienced in the last half century.
Luciano Rezzolla illustrates the progress in numerical relativity in three
examples: (1) numerical calculations of a binary merger of two neutron stars
producing an extremely strong magnetic field, (2) numerical calculations leading
to a black hole due to the collision of two self-gravitating fluids moving toward
each other with ultra-relativistic velocities, (3) numerical study of the recoil
dynamics of a black hole formed from head-on collision of two black holes with
different masses. The presence of an ‘‘anti-kick’’ is studied from various points of
view, including instructive figures, observational data, and issues of interest in
mathematical relativity.
A comprehensive review on instabilities of relativistic stars is given by John
Friedman and Nikolaos Stergioulas. It represents an extension of the talk given
by John Friedman in Prague and is partially based on the book Rotating
Relativistic Stars published by the authors in 2013. An action for perturbations
leads to the canonical energy and momentum and to the criterion for stability.
Various types of instabilities have been analyzed: convective instability, axi-
symmetric instability, nonaxisymmetric instabilities leading to the formation of
bar modes, and instabilities driven by gravitational waves like the Chandrasekhar-
Friedman-Schutz (CFS) instability. The CFS instability is primarily analyzed and
questions of how it can be influenced by the complex physics in neutron stars are
discussed.
In ‘‘Gravity talks: observing the universe with gravitational waves’’ (GW),
Bernard Schutz indicates that, indeed, the detectors are like microphones, not
pointed in some direction; and phases of the waves are more important than the
amplitudes. The network of six GW detectors is portrayed eloquently. In this field,
data analysis is most important. Its use at present is demonstrated by Ein-
stein@Home platform which led to the discovery of new pulsars. The information
we can obtain from detection (expected to occur by 2017) not only yields masses
and spins of colliding black holes or neutron stars but also, for example, the
distance to the sources. The properties of likely sources (neutron star or black hole
binary coalescence, neutron stars interiors and pulsars) are discussed. The article
concludes with information about LISA, substituted by somewhat restricted
Preface xi
eLISA, which should be launched if the LISA Pathfinder is successful. The

enormous advantages of detectors in space are highlighted.
Gerhard Heinzel (talking in Prague), and Karsten Danzmann, write about the
LISA and eLISA missions in greater detail, focusing also on financial problems. It
became clear in March 2011 that the originally designed joint ESA/NASA Mission
cannot rely on support from NASA. The redesigned eLISA (‘‘evolved’’ LISA)
could be supported by ESA alone. Despite the cost reduction, thousands of
compact white dwarfs binaries and hundreds of black hole binaries inspirals could
still be observed. The authors conclude that ‘‘the technology is well developed, the
team is strong and convinced that LISA must fly in the early 2020s… .’’
A wide range of relativistic effects, including the strong-field regime, can be
investigated by observing pulsars. Michael Kramer wrote a readable and com-
prehensive article on the current state-of-the art experiments involving these
‘‘cosmic lighthouses.’’ Going from individual pulsars, he shows how about 10 % of
about 2,000 known pulsars are in binary systems. These enable tests of general
relativity and alternative theories with extraordinary precision. The effects include
the precession of periastron, gravitational redshift, Shapiro delay, GW, and spin-
orbit coupling; at the same time, the basic principles of the theory are tested. The
detailed properties of the famous Hulse-Taylor pulsar are summarized and, in
particular, of the double pulsar. It provides tests of the GW quadrupole formula far
below the 0.1 % level. The masses and the orbital and relativistic spin-precession
are measured with extremely high accuracy. There are good prospects of detecting
GW by the ‘‘Pulsar Timing Array’’ method and a great challenge exists to test
fundamental predictions like the ‘‘no-hair theorem’’ if a pulsar orbiting the black
hole in the center of our Galaxy is discovered with the future Square Kilometer
Array.
In the last two articles, based on talks by Marek Abramowicz and by Ramesh
Narayan (with co-authors Jeffrey McClintock and Alexander Tchekhovskoy),
regions of extreme gravity effects are discussed. Three instruments planned for
missions in the near future (e.g. ‘‘The event horizon telescope’’) will provide
angular and time resolution that will enable us to investigate the immediate
neighborhoods of event horizons, ergospheres, innermost stable circular orbits
(ISCO), and circular photon orbits around black holes. All these features of black
holes are discussed in Abramowicz’s review, the main attention being paid to
ISCO because here the standard paradigm might have to be modified by the
magneto-rotational instability leading to turbulence around ISCO.
In the last contribution, the authors deal with a long-standing, complex, and
fascinating astrophysical problem: how jets from both stellar-mass and super-
massive black holes form, and how are they powered. Results of recent computer
simulations of black hole accretion and jets, in which magnetic fields twisted by
the rotating black hole play an essential role, combined with recent observations,
imply that (i) jets are powered from the black hole energy (rather than from a
surrounding disk), (ii) ‘‘the first observational evidence for a correlation between
jet power and black hole spin has finally been obtained.’’
xii Preface
The conference was organized2 under the auspices of the Rector of Charles
University, the oldest university north of the Alps (founded in 1348). It was
sponsored by the Faculty of Mathematics and Physics of Charles University, and,
in particular, by the Karel Janeček Foundation.3 We also acknowledge a contin-
uous support of the Czech Science Foundation, now under the grant No. 14-
37086G (Albert Einstein Centre).
Prague, December 2013 Jiří Bičák

Tomáš Ledvinka
2
The Scientific Organizing Committee included: M. Abramowicz, L. Andersson, A. Ashtekar, J.
Barbour, J. Bičák, R. Blandford, B. Brügmann, P. Chruściel, T. Damour, K. Danzmann, F. de
Felice, G. Ellis, J. Friedman, H. Friedrich, V. Frolov, G. Gibbons, G. Horowitz, J. Katz, K.
Kuchař, J. Lewandowski, G. Neugebauer, H. Nicolai, I. Novikov, M. Rees, O. Reula, L. Rezzolla,
M. Sasaki, G. Schäfer, B. Schmidt, A. Starobinsky, P. Tod, R. Wald, and C. Will.
The local organizing committee included: J. Bičák, M. Bursa, P. Hadrava, D. Heyrovský, V.
Karas, D. Kofroň, P. Krtouš, J. Langer, T. Ledvinka, J. Podolský, V. Pravda, O. Semerák, Z.
Stuchlík, O. Svítek, V. Špička, and M. Žofka, all of them associated (now or in the past) with the
Institute of Theoretical Physics of the Faculty of Mathematics and Physics of the Charles
University in Prague.
3
We would also like to express our gratitude to Frank Schulz and Vera Oswald from the Living
Reviews in Relativity, based in the Albert Einstein Institute in Golm, for their help with
references.
Contents
Part I Gravity and Prague
Kepler and Mach’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Julian Barbour
Einstein in Prague: Relativity Then and Now . . . . . . . . . . . . . . . . . . . 33

Jiří Bičák
Part II Classical General Relativity
Observers, Observables and Measurements in General Relativity . . . . 67

Donato Bini
Some Links Between General Relativity and Other

Parts of Physics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Gary W. Gibbons
The General Relativistic Two Body Problem and the Effective

One Body Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Thibault Damour
Gravitational Self-Force: Orbital Mechanics Beyond

Geodesic Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Leor Barack
Hamiltonian Formalism for Spinning Black Holes

in General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
Gerhard Schäfer
Stability of Marginally Outer Trapped Surfaces and Geometric

Inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Marc Mars
xiii
xiv Contents
Stationary Black-Hole Binaries: A Non-existence Proof . . . . . . . . . . . 209

Gernot Neugebauer and Jörg Hennig
Dynamic and Thermodynamic Stability of Black Holes

and Black Branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
Robert M. Wald
Instability of Anti-de Sitter Spacetime . . . . . . . . . . . . . . . . . . . . . . . . 239

Piotr Bizoń and Andrzej Rostworowski
Higher-Dimensional Black Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

Harvey S. Reall
Black Holes, Hidden Symmetry and Complete Integrability:

Brief Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Valeri P. Frolov
Part III Cosmology and Quantum Gravity
Cosmological Models and Stability. . . . . . . . . . . . . . . . . . . . . . . . . . . 277

Lars Andersson
Inflation and Birth of Cosmological Perturbations . . . . . . . . . . . . . . . 305

Misao Sasaki
Loop Quantum Gravity and the Planck Regime of Cosmology . . . . . . 323

Abhay Ashtekar
The Inflationary Origin of the Seeds of Cosmic Structure:

Quantum Theory and the Need for Novel Physics. . . . . . . . . . . . . . . . 349
Daniel Sudarsky
Quantum Gravity: The View From Particle Physics . . . . . . . . . . . . . . 369

Hermann Nicolai
Part IV Numerical Relativity and Relativistic Astrophysics
Three Little Pieces for Computer and Relativity. . . . . . . . . . . . . . . . . 391

Luciano Rezzolla
Instabilities of Relativistic Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

John L. Friedman and Nikolaos Stergioulas
Contents xv
Gravity Talks: Observing the Universe with Gravitational Waves. . . . 459

Bernard F. Schutz
LISA in 2012 and Beyond: 20 Years After the First Proposal . . . . . . . 477
Gerhard Heinzel and Karsten Danzmann
Einstein’s Gravity as Seen by a Cosmic Lighthouse Keeper . . . . . . . . 483

Michael Kramer
The Astrophysical Signatures of Black Holes: The Horizon,

The ISCO, The Ergosphere and The Light Circle. . . . . . . . . . . . . . . . 501
Marek A. Abramowicz
Energy Extraction from Spinning Black Holes Via Relativistic Jets. . . 523
Ramesh Narayan, Jeffrey E. McClintock and Alexander Tchekhovskoy
Part I
Gravity and Prague
Kepler and Mach’s Principle
Julian Barbour
Abstract The definitive ideas that led to the creation of general relativity crystallized
in Einstein’s thinking during 1912 while he was in Prague. At the centenary meeting
held there to mark the breakthrough, I was asked to talk about earlier great work
of relevance to dynamics done at Prague, above all by Kepler and Mach. The main
topics covered in this chapter are: some little known but basic facts about the planetary
motions; the conceptual framework and most important discoveries of Ptolemy and
Copernicus; the complete change of concepts that Kepler introduced and their role
in his discoveries; the significance of them in Newton’s work; Mach’s realization
that Kepler’s conceptual revolution needed further development to free Newton’s
conceptual world of the last vestiges of the purely geometrical Ptolemaic world
view; and the precise formulation of Mach’s principle required to place GR correctly
in the line of conceptual and technical evolution that began with the ancient Greek
astronomers.
1 Introduction
Some of the most important advances in science are associated with Prague.
The meeting at which the talk on which this chapter is based celebrated Einstein’s
breakthrough to the key ideas of general relativity (GR) in 1912 near the end of
his time in the Bohemian capital. In this chapter, I wish to honour Kepler and his
discovery of the laws of planetary motion and Mach’s critique of Newton’s concepts
of absolute space and time. The creation of GR is unthinkable without them. I also
wish to give what I believe is the correct formulation of Mach’s principle. I believe
that misunderstanding about this, ironically due to Einstein, may well be holding
back both cosmology and the discovery of the quantum law of the universe.
J. Barbour (B)
Department of Physics, University of Oxford, Oxford, UK
e-mail: julian.barbour@physics.ox.ac.uk
J. Bičák and T. Ledvinka (eds.), General Relativity, Cosmology and Astrophysics, 3

Fundamental Theories of Physics 177, DOI: 10.1007/978-3-319-06349-2_1,
© Springer International Publishing Switzerland 2014
4 J. Barbour
Fig. 1 The eccentricity e is E

OS/OB, the ellipticity ε =
DE/OE is ε = e2 /2 and
very small for the naked-eye P
planets D
O
B A
S S’
I begin with some background to Kepler’s discovery of the laws of planetary

motion and then discuss the key intuitive ideas that enabled him to find them. We shall
see that Kepler’s reaction to the conceptual framework he inherited from Ptolemy,
Copernicus and all previous astronomers was a clear anticipation of Mach’s reaction
to Newton’s absolute space and time. In the broadest terms, one can see the creation
of GR proceeding along a line of conceptual and technical development associated
with six scientists: Ptolemy, Copernicus, Kepler, Newton, Mach, Einstein.
The main topics covered in this chapter are listed in the abstract, so I turn directly
to their presentation.
2 Some Important Facts of Planetary Motion
Everyone knows Kepler’s three laws: 1. The planets move in ellipses with the Sun at
one focus. 2. The radius vector from the Sun to the planet sweeps out equal areas in
equal times. 3. The period of each planet is proportional to a 3/2 , where a is the major
axis of its ellipse. The first two laws were discovered in 1605, the third followed in
1618.
However, what really counts for understanding the history of planetary astron-
omy1 up to Kepler’s discovery of his first two laws is the form they take when the
eccentricity of the ellipse is relatively small, as it is for all the planets. We must start
with basic facts about ellipses (Fig. 1).
Because the ellipticity ε is half the square of the eccentricity e, the magnitude of
ε is small for all the planets. The planet with the largest eccentricity, e ≈ 1/5, is
Mercury; then comes Mars with e ≈ 1/11; Jupiter and Saturn have e ≈ 1/20; the
1 See [1] for a detailed discussion of the history.

Kepler and Mach’s Principle 5
Fig. 2 The orbit of Mercury is circular to one part in 50. The figure shows the foci and centre of
the orbit and the circle that most closely approximates the ellipse. The circle is inside the ellipse
along the line of the apsides joining the foci and outside at the quadrants. For Mars, the planet for
which Kepler discovered ellipticity, the gap between the circle and ellipse is 4.5 times less. The
figure shows clearly that the effects of eccentricity are far more readily observable than those of
ellipticity
Earth (e ≈ 1/60) and Venus (e ≈ 1/140) have very small eccentricities and their
orbits are wonderfully circular. This is crucial for the effects that even conscientious
observers, who could use only the naked eye until 1610, were likely to find. The
Sun and Moon subtend about 30 arc minutes on the sky. The accuracy of Ptolemy’s
observations was about a third of that, 10⊗ . Tycho Brahe’s heroic observations, mostly
in the period 1576–1597, pushed the accuracy to 2⊗ , as Brahe claimed, or 4⊗ according
to Kepler’s more sober estimate.
What these facts about the accuracy of naked-eye astronomy mean is that the
effects due to the orbit eccentricity, typically with a magnitude of degrees, were
readily observable, while those due to the ellipticity were virtually undetectable. The
only planet for which this is not strictly true is Mercury, but it is close to the Sun
and seldom well seen, so it played no significant role in the discovery of Kepler’s
laws.2 In one of several flukes in astronomy—the nearly equal apparent diameters
of the Sun and Moon and the advance of the perihelion of Mercury included—it just
so happens that among the remaining planets Mars was the most readily observable
and has an ellipticity just large enough for Kepler’s genius to espy it in the multitude
of Brahe’s observations.
To get an idea of Brahe and Kepler’s achievement in the discovery of the ellip-
ticity, Fig. 2 shows the orbit of Mercury. To the eye, it is a circle. One needs the
2 It did help the belated recognition of Kepler’s laws. His Rudolphine Tables (1627) led to the
correct prediction and observation of the transit of Mercury across the Sun in 1631, a year after
Kepler’s death. The vastly superior accuracy of the Tables compared with the rivals, and the laws
on which they were based, could no longer be denied.
6 J. Barbour
Fig. 3 Illustration of the

empty-focus effect for
Mercury. The Sun is left
of the centre of the orbit. An
observer at the empty focus on
the right sees the planet move
round a great circle with near
perfect uniformity
circumscribing circle to see the difference. Fapp—for all practical purposes, to use
John Bell’s acronym—the planetary orbits are circles as far as naked-eye astronomy
is concerned.
If the near circularity of the orbits is little known even among many astronomers,
another remarkable fact is virtually unknown. It relates to a property of the empty
focus of the planet’s ellipse. If you could hover in a spacecraft just above the Sun’s
surface and watch a planet on the celestial sphere, you would see it move in a
great circle with a decidedly non-uniform motion: first because, in accordance with
Kepler’s 2nd law, its physical speed in space does change, and, second, because
the Sun is displaced from the centre of the orbit. This geometrical effect doubles the
nonuniformity of the observed angular speed in the small-eccentricity approximation
appropriate for the planets. If you then fly to the centre of the orbit and hover there,
the geometrical distortion is eliminated, and the observed angular speed reflects the
true variable speed. But a miracle happens if you journey on to the empty focus: the
geometrical effect that enhanced the non-uniformity above the Sun is now reversed:
you see the planet move round its great circle with near perfect uniformity. Figure 3
illustrates the combined effect of the circularity and empty-focus effect.
The way to understand the actual process of discovery of the laws of planetary
motions is through approximations to Kepler’s laws. If the eccentricity e is zero, the
orbits are circles and the speed uniform; if e is small, the orbits are still effectively
circles but eccentric, and the speed on the circle is nonuniform though seen from
the empty focus it is amazingly uniform. The major advances in the early history of
planetary astronomy were largely due to these two effects. They are, respectively,
very good approximations to Kepler’s first two laws. It is no exaggeration to say that
without them celestial dynamics could not have begun. However, they later became a
source of great confusion: Copernicus and Brahe were literally going round in circles
trying to make sense of the circles they imagined really were there in the sky. That’s
the story to which we now turn, beginning with the Greek astronomy.
3 Greek Planetary Astronomy
Let’s start with what is readily observable: the Sun. Aristotle already knew that its
motion around the ecliptic is non-uniform. Sometime around 150 BC (or 150 BCE
as we should now say), Hipparchus, the first great astronomer of antiquity, made
accurate measurements and a model to explain the non-uniformity. He supposed that
the Sun moves on a perfect circle and with perfect uniformity and that the observed
non-uniformity of its motion was due to a displacement D of the centre of the circle
from the centre of the Earth, assumed to be the centre of the Universe. The ratio of
D to R, the radius of the circle, was its eccentricity, the origin of both the technical
and non-technical meanings of the word.
Hipparchus needed only two observations, the times taken by the Sun from the
vernal equinox to the summer solstice and from there to the autumnal equinox, 94.5
and 92.5 days, respectively. These two data were enough to fix the two unknown
parameters of his model: the magnitude of e = D/R and the direction of the eccentric
centre. This defined the line of the apsides, which joins apogee and perigee (or
aphelion and perihelion in heliocentric astronomy). Hipparchus found e = 1/24,
at that epoch more than twice the eccentricity of the Earth’s heliocentric, or Sun’s
geocentric, orbit, which was then ≈1/57.
Part of the inaccuracy was due to observational error, but the major contribution
was a flaw in the theory. Having not the remotest reason to suspect non-uniformity of
the motion, Hipparchus had inadvertently doubled the eccentricity. What was truly
remarkable was that, when the observational accuracy had been pushed to its naked-
eye limit by Islamic astronomers and Brahe, Hipparchus’s incorrect model proved
to be amazingly accurate in its predictions. In fact, it gives deviations from the true
positions never greater than 3/4 of an arc minute, way below the detection level of
naked-eye observations.3 For this reason, Hipparchus’s model, converted appropri-
ately to heliocentric motion of the Earth, passed unscathed through the Copernican
revolution. One of the gems in Kepler’s work was, as we shall see, the dethronement
of the Hipparchan model. It had reigned supreme for over 1700 years.
We now move on 300 years to the next great astronomer of antiquity: Claudius
Ptolemy. He worked in Alexandria and had access there to, among much else, Baby-
lonian observations made nearly a millennium earlier. He embarked on what was
surely the first rationally and comprehensively planned scientific-research project in
human history—the theoretical explanation through uniform circular motions of all
3 This remarkable accuracy is due to the very small eccentricity of the Earth’s orbit, currently about
1/60. A solar theory developed by a Martian Hipparchus would not have survived for long because
Mars has eccentricity ≈ 1/11 (crucial for Kepler’s discoveries). It is worth mentioning that the
theory-independent quantity most immediately observable is always twice the eccentricity. For the
Sun observed from the Earth, this is currently 1/30 of a radian or about 2◦ (four apparent solar
diameters), half of which comes from the relative geometrical displacement of 1/60 and half from
the physical non-uniformity described by Kepler’s 2nd law. These equal contributions to observable
effects are very important for understanding the history of ancient astronomy. By comparison, the
observable ellipticity effects in the solar motion are about 120 times smaller at 1/30 of the apparent
solar diameter, i.e., 1⊗ .
8 J. Barbour
Fig. 4 The retrograde motions of Jupiter. Against the backdrop of the fixed stars, Jupiter (like all
the planets) generally moves eastwards, never straying far from the ecliptic (the great circle on
which the Sun moves). But about once every 13 months it executes its retrograde motions, in which
the eastward motion comes to a stop and reverses into a westward motion until that halts and the
eastward motion is resumed. Any attentive observer who keeps reasonably good records is likely
to try to describe the complicated observed motion as a superposition of two simpler motions
the seven planets. Meaning wanderers, these included the Sun and Moon as well
as Mercury, Venus, Mars, Jupiter and Saturn. Ptolemy’s astronomical compendium,
known through its Arabic title as The Almagest and written around 150 CE, was the
handbook of astronomy for close on 1500 years. His epicycles tend to be mocked
today as the paradigm of poor ad hoc science, often by scientists who should know
better,4 but they were one of the great contributions to the advance of science. On
top of that, Ptolemy made what is arguably the first great discovery in the history of
dynamics. To that I now turn.
Compared with the observed motion of a planet, that of the Sun is simplicity itself,
being just the mirror image of the purely periodic motion of a single planet, Mother
Earth, around the Sun. But the motion of a planet seen from the Earth is a compound
of two incommensurate periodic motions. This compounding leads to the famous
observed retrograde motions (Fig. 4).
Following earlier proposals, which were probably only qualitative and may
have been suggested by the great mathematician Apollonius (circa 255–170 BCE),
Ptolemy attempted to describe them with the epicycle–deferent model (Fig. 5), which
I describe in the caption only for the simpler case of the three outer planets.
In heliocentric terms, the motion of the guide point D is the actual motion of the
planet, while the epicyclic motion is the reflection of the Earth’s motion seen through
the relative motion of the planet against the stars. The specific eccentricities of the
various planetary orbits played a crucial role in the details of Ptolemy’s theory. The
key thing to understand is that the Earth’s eccentricity, eE , is significantly smaller
4 I suspect some of them trust the scientifically very ignorant account in Koestler’s Sleepwalkers,
in which it is stated that “There is something profoundly distasteful about Ptolemy’s universe; it is
the work of a pedant with much patience and little originality, doggedly piling ‘orb in orb’.” The
Almagest use the bare minimum of epicycles.
Fig. 5 Ptolemy sought to explain the regular eastward motion of, say, Jupiter, through uniform
eastward motion of an invisible guide point D on a circle called the deferent. Around D a spoke of
length less than the deferent radius rotated on an epicycle with perfect uniformity carrying the planet
P at its tip. The centre C of the deferent is displaced from the position of the terrestrial observer O
in order to explain the fact that even without the epicyclic motion the general eastward motion of
Jupiter is manifestly non-uniform like that of the Sun as described by Hipparchus’s model
that that of the three outer planets, eP , which are moreover further from the Sun.
On the sky, the observable effect O due to the orbital eccentricities is, to a first
approximation,
eP aE
O= ,
eE aP
where aE and aP are the Earth’s and the planet’s semi-major axes, respectively. For
Saturn, Jupiter and Mars O ≈ 1/30, 1/15, 1/8, respectively. This meant that the
nonuniformity in the planet’s motion, represented in Ptolemy’s theoretical model
by the motion of the invisible guide point D, was readily observable and could
not be ignored. In a first attempt to describe it, Ptolemy copied the Hipparchan
solar model exactly by a simple displacement of the centre of the deferent from the
terrestrial observer. For the epicycle motion, he assumed perfect uniformity around
D. In heliocentric terms, this corresponds to an exactly zero-eccentricity circular
orbit of the Earth. Note that the error due to circularity is tiny, around one part in
3,600 and unobservable; the error due to the zero eccentricity is only 1/60. However,
both of these are reduced by the ratio aE /aP and escaped Ptolemy. They would in any
case have been very difficult for him, with his rudimentary mathematics, to model.
What is extremely interesting is the way Ptolemy fixed the parameters of the
deferent. The epicyclic motion being simply the reflection of the Earth’s motion,
the epicycle always points in the direction of the Sun. Ptolemy knew this; it did not
10 J. Barbour
prompt him to heliocentricity, but it did help him to fix the deferent parameters and
to make that great discovery I mentioned.
His task was to fix the position of the invisible guide point: mission impossible you
might think. But no; when the Earth is exactly between the planet and the Sun (so that
the planet, in opposition, is due south at midnight), the epicycle, with the planet on
its tip, points simultaneously towards the Sun and the Earth. Moreover, seen from the
Earth, the guide point D is exactly behind the planet. Using such observations, spread
necessarily over many years, Ptolemy could mimic what Hipparchus had done for the
Sun. He was able to determine the eccentricity and line of the apsides of the deferent.
But this was simultaneously the heliocentric orbit of the planet! However, because
Ptolemy directly copied Hipparchus and did not suspect physical nonuniformity in
the motion of D, he too found double the actual eccentricity.
Now in the solar motion there was no possibility of detecting the error. But
Ptolemy’s model was not yet complete. He had to fix the ratio of the epicycle and
deferent radii. For this he needed just one more observation, of necessity made when
the planet is not in opposition. There is an almost poetic touch worth mentioning
here. When the planet is in opposition and due south a midnight, it rises at sun-
set, acronychal in Greek.5 Ptolemy needed just one non-acronychal observation to
determine the length of the epicycle.
With the model complete, Ptolemy—good scientist that he was—tested it using
further non-acronychal observations. Dismay: the model failed to predict them cor-
rectly. After a long period of trial and error that, as Ptolemy admitted, had no prin-
cipled basis except fidelity to observation—and hence truth—he found a deferent
model that worked very well.
He discovered that it was necessary to halve his previous deferent eccentricity and
introduce an ‘equalizing point’, or equant as it is now called. It lay on the other side of
the centre of the orbit from the observer along the line of the apsides, which remained
unchanged. Around the equant one had to imagine a spoke that rotated with perfectly
uniform (hence equalizing) angular velocity and cut the deferent circle in its new
position at the point when the guide point D must be. As before, the planet-carrying
epicycle rotated with perfect uniformity about D.
With this model (somewhat modified for Venus and Mercury), Ptolemy found
he could describe and predict the motion of all the planets with surprisingly good
accuracy. What, in the long run, was truly significant for astronomy and dynamics,
was that he had found a wonderfully good approximation to what Kepler’s second
law predicts. For, in heliocentric terms, Ptolemy’s equant is none other than the empty
focus of the planet’s orbit—and I have already explained what a superb approximation
that is. Because he was also working with eccentric, perfectly circular orbits, he also
had an excellent approximation to Kepler’s first law.
If we discount the barely observable and hence ‘thankless’ Mercury, Ptolemy’s
theory was correct for all the other planets to excellent accuracy. For Mars, with the
largest eccentricity, the maximal deviation from Kepler’s laws was only one part in
225. That is the measure of his achievement.
5 Acronychal and non-acronychal observations were still vital in Kepler’s work.

Final comments before we move on to Copernicus and Kepler. First, all of

Ptolemy’s work was based on measurement of angles between objects that could
be seen. Second, none of his work or anything really accurate in astronomy could
have been done up to the invention of truly accurate clocks in the 20th century with-
out the diurnal revolution of the stars, aka the Earth’s rotation. It was the one and
only clock that could be used. Time was also read off it by measurement of angles
between visible objects.
4 Copernicus
It is ironic that Copernicus stumbled on his revolutionary idea by trying to undo

Ptolemy’s greatest discovery: the equant. Despite great admiration for Ptolemy’s
technical skill and achievements, Copernicus strongly disliked the equant’s viola-
tion of the literally sacred principle that all the divine objects in the heavens must
move with perfect uniformity in perfect circles. Ptolemy had maintained the circles
(with good reason—they worked) but had discarded uniformity for the sake of truth.
Copernicus, like at least one Islamic astronomer before him, sought to replace the
equant device by a combination of uniform circular motions that, of necessity, was
more complicated than Ptolemy’s solution if fidelity to observational facts was to be
maintained. While working on this project, he realized that all the retrograde motions
of the planets could be understood as effects of relative motion if one assumes that
the Earth is not at rest but moves in a circle.
There is an important point here that needs to be emphasized: Copernicus proposed
a theory of terrestrial mobility, not heliocentricity. This was still the main point
for Galileo, as shown by his famous retort “Eppur si muove.” All that Copernicus
needed, and said, was that the Sun must be near the centre of the circle in which
the Earth moved.6 For Copernicus, the Sun and its precise position had no physical
significance. He said the Sun had a worthy place in the heavens, placed to illuminate
the dance of the planets. As we shall see, the true discoverer of heliocentricity was
Kepler.
Copernicus made four great contributions: first, he unavoidably, though without
having an inkling of its significance, drew attention to the Sun, which was very
important for Kepler; second, he explained the retrograde motions; third, as an under-
appreciated consequence of that, he brought to planetary astronomy a unity entirely
lacking in Ptolemy’s universe. Fourth, his arrangement of the solar system and the
absence of observed parallax of any of the stars required the stars to be immensely
farther away than Saturn.
The third contribution needs a little elaboration. Since Ptolemy worked with
angles, he had no way of determining any distances. He therefore set all defer-
ent radii equal to the nominal value unity and found the epicycle radius as a ratio to
unity. Copernicus realized that this ratio, different for each planet, simply reflected
6 Copernicus actually thought that the Sun might have some slow motion of its own.
12 J. Barbour
the ratio of the radii of his various circles: the Earth’s to that of the planets. He could
use the radius of the Earth’s orbit as a trigonometric base line to determine the dis-
tances to the planets. This immediately gave him a very good overall picture of the
solar system. He worked out the correct order, distances and average speeds of all
the planets, measured of course in terms of the radius of the Earth’s orbit (now, of
course, the astronomical unit) and the terrestrial day. He obtained a qualitative form
of Kepler’s third law and facts that Kepler subsequently found highly suggestive.
It is often said that the Copernican and Ptolemaic arrangements are kinematically
identical. This is not strictly true and does not do justice to Copernicus. Geometrical
dispositions are part of kinematics. Ptolemy’s Almagest did not have them or the
lower bound on the distance to the stars.
A nice way to compare the respective achievements is this: when Ptolemy died,
he could predict what the sky would look like—where the planets would be—as
seen from Alexandria centuries after his death, but he had no idea what it would
look like from Mars. When Copernicus died in 1543, he did know or, at least, knew
how to calculate the positions of the Sun and planets as seen from Mars. In fact, the
possibility was only literally confirmed in the space age.
Copernicus did great things, but, from a modern point of view, he bequeathed a
most odd solar system to posterity. I have already mentioned the Sun’s role as a mere
lantern. Really strange was the location of the ‘centre of the Copernican universe’.
Ptolemy had discovered an equant in the deferents of all the planets, essentially
because the Earth, unbeknown to him, was a spaceship that allowed him to look at
the planets’ positions from a whole circle in the solar system and not just from the
Sun’s position (as in the acronychal observations). But because the Sun’s motion is
merely the Earth’s 180◦ out of phase, there was no way observation could force an
equant on the solar motion. Ptolemy left the Hipparchan model unchanged.
And so did Copernicus. Even though he made Mother Earth a planet like the others
and contrived makeshift substitutes for their equants, it never occurred to him that the
Earth should get anything equivalent, so he simply inverted the Hipparchan model
and gave the Earth an eccentricity twice what it should have. That simultaneously
singled out the empty focus of the Earth’s orbit as a special point. Moreover, because
of the fluke of the Earth’s eccentricity being so small Copernicus was misled into
thinking that the lines of the apsides of the planets all converged, not on the Sun,
but at the very same point that Kepler was later to identify as the Earth’s empty
focus. Figure 6 shows how small the mismatch was—but also that Copernicus did
not propose a truly heliocentric system.
In fact, the clearest evidence of that is in the diagrams which Copernicus drew to
show the orbits of the planets. They do not show the Sun. It was in no way an integral
part of his scheme.
There were many other oddities, some very bizarre, in the Copernican cosmos,
most of which arose because Copernicus simply inverted the Ptolemaic models. When
De Revolutionibus was published in the same year 1543 that he died, Copernicus knew
he had made a monumental discovery, but, like Ptolemy, his insights and methods
were purely geometrical and kinematical. Kepler commented “he was unaware of
his riches”.
Fig. 6 Copernicus believed that the lines of the apsides of the three outer planets converged on
a void point near the Sun that was actually the empty focus of the Earth’s heliocentric orbit. The
true Sun is the black disc on the left, where the solid lines of the apsides converge. The dashed
lines show Copernicus’s belief. The mismatch is small but shows that Copernicus thought solely in
geometrical terms, for which such an arrangement with the Sun playing no physical role is perfectly
acceptable
5 Kepler
In this section, I want to concentrate on the huge conceptual change that Kepler
(1571–1630), for whom a magnificent portrait (Fig. 7) survives, introduced and how
he anticipated Mach’s attitude to dynamics. I am firmly of the belief that more is still
to come of it. I can only pick out the highlights. The details, which are absorbing,
can be found in [1].
The best place to start is Brahe’s observations of the comet of 1577, which estab-
lished its distance as interplanetary. For Kepler, the supreme importance of the obser-
vations were that they ‘destroyed’ the crystal spheres widely believed to carry the
planets. In his fascinating account of how he mastered the motion of Mars, the
Astronomia Nova published in 1609, he repeatedly pointed out that Brahe’s observa-
tions proved that the planets were not carried by spheres. The comet had passed clean
through the solar system without crashing into them. They could not be there. In one
of the great intuitive insights in the history of science, he proclaimed: “Henceforth
the planets must find their way through the void like the birds through the air. We
must philosophize about these things differently.”
Crucial questions then arose. What moves the planets? How do they find their
way? What if anything is directing them? Let us start with the second and third
questions, which reveal Kepler’s affinity with Mach—or better Mach’s with Kepler.
Birds find their way around the world by reference to features in the terrain and sky.
But, according to the astronomy Kepler inherited from Ptolemy, Copernicus and all
previous astronomers including Tycho Brahe, literally everything was controlled and
directed by void points, above all the equants that Ptolemy had discovered, in empty
14 J. Barbour
Fig. 7 Johannes Kepler. The

epitaph that he composed
for himself read “I used to
measure the Heavens, now I
measure the shadows of Earth.
The mind belonged to Heaven,
the body’s shadow lies here.”
His grave and tombstone in
Regensburg, where he died in
1630, have been lost
featureless space. The difference from birds was blatant. Like them, the planets must
use visible objects. Motion is relative to things you can observe. The only significant
visible things that the planets could be guided by were the Sun and the distant stars.
The guiding and determining role of visible matter is exactly what Mach was insisting
on two and a half centuries later and led him to argue so persuasively against Newton’s
absolute space.
Moreover, if crystals spheres do not carry the planets, whence comes their motion?
The planets must either have inherent motive force or be subject to it. This was a
veritable change of mindset. A prominent part of the immensely long subtitle to the
Astronomia Nova proclaimed it to be Celestial Physics. Kepler introduced forces
into the heavens. True, they were Aristotelian, with the force assumed to determine
the velocity it imparts and not the acceleration as in Newtonian dynamics. The
astronomical data could give Kepler no hints in that direction, ironically for the same
reason that Einstein three centuries later was able to subsume gravitational forces
and inertia into a single geodesic law. Indeed, until very late in his work, Kepler
believed the planets moved in circles. What more perfect and self-contained motion
exists than that?
What was really important about Kepler’s forces was not their mode of action
but their conjectured source: the Sun and the planets themselves. In this key respect,
Kepler’s forces correctly prefigured Newton’s. Their sources and controlling power
resided in physical bodies, not void points in empty space. Here too, in identifying
motion-controlling power with bodies and not space, Kepler anticipated Mach, who
insisted that apparently force-free inertial motion was nothing of the sort but the
Fig. 8 This diagram

encapsulates the difference
between Kepler and all his
predecessors. The arrival of a
new mind on the astronomical
scene is demonstrated
nowhere more clearly than
in the comparison of the dia-
grams in De Revolutionibus
and the Astronomia Nova.
The Sun is prominent by its
absence in Copernicus’s; in
Kepler’s, as here, it takes
pride of place, controlling the
motion of the planet through
physical forces
outcome of an as yet unknown physical effect of all the matter in the Universe. Let
me here quote Mach [2, p. 296]: “The natural investigator must feel the need of
further insight—of knowledge of the immediate connections, say, of the masses of
the Universe. There will hover before him as an ideal an insight into the principles
of the whole matter, from which accelerated and inertial motion result in the same
way. The progress from Kepler’s discovery to Newton’s law of gravitation, and the
impetus given by this to the finding of a physical understanding of the attraction
in the manner in which electrical actions at a distance have been treated, may here
serve as a model.” That Mach sensed an affinity between himself and Kepler comes
through in this quotation.
Let us return to details. Kepler had to explain two different kinds of motion: the
eccentric circular motion around the Sun and the motion towards and away from the
Sun during its course. To explain the circular motion, Kepler conjectured (nearly a
decade before Galileo observed it!) that the Sun rotates about an axis perpendicular
to the ecliptic and that what one might called ethereal ‘spokes’, rotating with the
Sun, protruded from its equator. These, he assumed, swept the planets along in their
circular motion, their strength diminishing with increasing distance from the Sun
in order to explain why the more distant planets moved slower. As for the motion
towards and away from the Sun, he conjectured that that it housed a powerful magnet
and each planet a lesser one. The alignment of the magnetic poles would pull the
planet towards the Sun on one side of the orbit and repel it on the other side.7 These
ideas are illustrated in Kepler’s diagram shown in Fig. 8.
By modern standards, Kepler’s forces were rather primitive and could not have
survived detailed quantitative testing. What was decisive was that they focussed all
of Kepler’s interest on the Sun and its preciselocation. He was firmly convinced that
7 William Gilbert’s influential book on magnetism, published in 1600, strongly influenced Kepler’s
thinking.
16 J. Barbour
the centre of the solar system did not lie at that mysterious void point that, post his
discoveries, we recognize as the empty focus of the Earth’s orbit, but at the entre
of the relatively nearby mighty physical Sun (the distance between the Sun and the
empty focus is 1/30 of the Earth’s semimajor axis).
There were two ways to confirm this: first, to show that the lines of the apsides all
converged exactly on the Sun, not the void point relatively close to it. Second, to show
that the speed in orbit was not controlled by the void equant, as it was in Ptolemy’s
and, de facto despite his intense dislike of it, in Copernicus’s astronomy, but by the
Sun. I shall come to the crucial steps through which Kepler eventually came to his
area law, which governs the speed, in a moment. First, I want to make clear why
Kepler has the credit for heliocentricity in a way that Copernicus does not. Modest
as the move from the empty to the occupied focus as centre of the solar system may
appear, it was a small step that anticipated and made possible Newton’s giant leap
of understanding in the workings of the world. It identified the turning point.
Kepler’s conceptual ideas drove all the technical work done at Brahe’s behest in
Prague—to establish the precise motion of Mars to the same accuracy that the Dane’s
incomparable observations allowed. His primary tool was trigonometry, which he put
to use like no one before him. Kepler was the first man who could roam truly freely
in imagination through the solar system. Appropriately, he also wrote almost the first
work of science fiction: a dream of a journey to the Moon. The entire thrust of his
trigonometric work was to establish heliocentricity beyond gainsaying, above all to
show that the knitting-needle lines of the apsides all converged bang in the middle
of the Sun and not as in the Copernican scheme (Fig. 6) at the nearby void point.8
There is even a sense in which Kepler anticipated gauge theory: he knew perfectly
well that, kinematically, all of his precise geometrical results could be expressed just
as adequately in the Tychonic 9 or geocentric Ptolemaic schemes as in the Copernican
arrangement, but triumphantly pointed out that in all three the lines of the apsides
meet at one point in the Sun. That was the ‘gauge-invariant’ content of his discoveries.
The area law, the discovery of which we have still to discuss, had the same status.
Let us go through the most important technical advances to which Kepler was
led by his intuition. I said that Ptolemy?s acronychal observations were made at
opposition, when the observed planet is due south at midnight and the Sun is directly
behind the terrestrial observer. This is not quite true; to facilitate computations and
very likely because he did not realize the importance of the difference, Ptolemy’s
actual acronychal observations were made not when the true Sun was behind the
observer but a substitute, a mathematically defined ‘mean Sun’ that moved around
the ecliptic with perfectly uniform speed, coinciding with the true Sun only at the
equinoxes. The angular distance between the true and mean Sun could be as much
8 Kepler’s work was actually a much more logically consistent and definitive proof of the Copernican
cosmology than Galileo was able to muster. In fact, magnificent as his many achievements were,
the Tuscan completely failed to recognize or begin to comprehend Kepler’s achievement.
9 Tycho Brahe could not believe in the immensity of the Universe that, given the absence of any
observable stellar parallax, followed from Copernicus’s proposal and therefore proposed that the
Sun goes round the Earth while Mercury and Venus orbit it.
as those 2◦ corresponding to twice the Earth’s eccentricity. Brahe had continued

the Ptolemaic practice of using the mean Sun; Kepler meticulously corrected his
observations by interpolation to make them correspond to his beloved true Sun. This
was a first useful sharpening of accuracy.
Kepler was also the first person to understand how to take into account correctly
the fact that the planets do not all move in the ecliptic. This gives rise to significant
effects, causing the retrogression loops to have an out-of-ecliptic component (Fig. 4);
without proper understanding of and correction for the effect Kepler could never have
found the ellipticity of Mars’s orbit. In fact, Kepler’s first major result is what one
might call his zeroth law: each planet moves in plane that passes through the Sun
and is fixed in the frame defined by the stars.
But the real gem in Kepler’s work that prepared the ground for his greatest discov-
eries was his finding of the true location of the Earth. All motion is relative. If you
are trying to determine the position and motion of a distant object, you will surely
make errors if you are mistaken about your own position and motion.10 This is what
Kepler understood perfectly—and he had good reason to be concerned. According
to Copernicus and Brahe, the Earth’s orbit was sui generis: unlike those of the other
planets, it had no equant. Kepler’s sense of the uniformity of nature told him that could
not be true. The Earth had to have an equant. If so, that would mean it would have
only half the eccentricity attributed to it by Copernicus and Brahe. Existing theory
must be putting the Earth in the wrong place and thereby distorting the interpretation
of the observations of all the extraterrestrial bodies.
In an article written in 1930 to mark the 300th anniversary of Kepler’s death,
Einstein described—with good reason—Kepler’s halving of the Earth’s eccentricity
as one of the most beautiful things in all of science. Kepler knew that too. He had
the finest diagram in his book engraved to show the way it was done (Fig. 9).
His stroke of genius was to use a trigonometric base line formed by the Sun and
Mars at times when he knew Mars was at exactly the same point in its orbit. One
of the great clarifications due to Copernicus’s insights was that the planets traced
out invariable orbits in the space defined by the Sun and fixed stars, returning to the
same orbital position after completion of one orbit. Now among all data, heliocentric
periods were the easiest to determine accurately; that of Mars was known to be
685 terrestrial days. Kepler searched among Brahe’s 21-year treasury for Martian
observations that by chance were separated by multiples of 685 days. At them, Mars
must be at the same point in space. Kepler found three such observations. Acronychal
observations of Mars and the theory of them, which he could trust, told Kepler the
direction of Mars as seen from the Sun at all times. The direction to the Earth was
also known, so Kepler could determine the angle between Mars and the Earth seen
from the Sun. Brahe’s observations gave him the angle between the Sun and Mars as
seen from the Earth. Kepler had the one fixed Sun–Mars side of the triangle and two
angles of the Sun–Mars–Earth triangle. Three such observations gave three positions
10 I was told some years ago that the largest uncertainty in many high-precision tests of GR was the
uncertainty in the Earth’s position that results from the perturbing influence of the asteroids, whose
size is known but not, to sufficient accuracy, their densities.
18 J. Barbour
Fig. 9 This diagram shows the halving of the Earth’s eccentricity in the Copernican cosmology
and of the Sun’s in the Ptolemaic and Tychonic. In the Copernican scheme, Mars is at the point x on
three occasions. Knowing the relevant angles, Kepler could determine the corresponding positions
of the Earth at the three points on the dashed circle. They established the true position of the Earth’s
orbit and that it must have half the eccentricity assumed by Copernicus
of the Earth. But one only needs three known points to fix the position and size of a
circle.11 Kepler’s determination of the position of the Earth’s orbit finally revealed
the error in Hipparchus’s solar theory. The halving of the Earth’s eccentricity created
a firm foundation for astronomy. It was hugely important.
Having properly located the Earth, and still believing in circular orbits, Kepler set
about fitting the parameters of Mars’s orbit to Brahe’s observations. He still made
use of the Ptolemaic equant even though convinced the speed in orbit must somehow
be determined relative to the Sun and not the void equant. It was a wonderful piece of
work and a posthumous triumph for Ptolemy’s circles and equant. But the merry-go-
round just would not get everything right. Kepler tweaked here, he tweaked there, but
whatever he did an occasional error of up to just 8 min of arc in the position of Mars
would show up. He could never have done it without Brahe’s observations, which
were both as accurate as they could be and also, most importantly, comprehensive.12
This is the place to quote Kepler:
“We, whom God in his goodness has given such a careful observer in Tycho
Brahe, and whose observations reveal the 8⊗ error of Ptolemy’s calculations, should
thankfully recognize the goodness of God and make use of it. That is, we should
make the effort (supported by the arguments for the falsity of our assumptions) to
find at last the true form of the celestial motions... These 8⊗ alone reveal the need for
reformulation of the whole of astronomy; they become the material of a great part
of my work.”
Now it is time to talk about the area law. Along with establishing where the lines
of the apsides meet all in one place, this was the other great bonus of the ‘Machian’
shifting from a void point to the Sun. For the apside adjustment, the shift was from
the Earth’s empty focus. In the case of the area law, it was from Mars’s empty focus.
Kepler was keenly aware of the value of the equant phenomenon: mathemati-
cally, in pre-calculus days, anything that involved uniform motion (angular veloc-
ity about the equant in this case) was a significant plus. Non-uniform motion was
barely tractable. For this reason astronomers were still using the equant to calculate
ephemerides in Newton’s time three quarters of a century after Kepler, absolutely
insistent on a physical interpretation of celestial motions, had done away with it.
The area law was one of the most serendipitous discoveries—of which there are
so many—in science. Kepler was looking for a law, governed by the Sun, that would
determine the speed in orbit of each planet. His physical intuition told him the Sun
11 Keplers’ work on Mars began under the assumption that the Earth has an exactly circular orbit.
Because the Earth’s eccentricity is ≈1/60, this assumption is accurate to better than one part in
7000. Even when Kepler knew the Earth’s orbit could not be a perfect circle, he could assume it to
be so for his work on Mars with its far larger eccentricity ≈1/11. In the story of the discovery of the
laws of their motion, the planets were like the characters in a good novel. Each had an individual
personality determined by its eccentricity and semimajor axis. The interaction of these personalities,
reflected in the observational data, is what makes the discovery of the laws of the planets’ motion
such an absorbing story.
12 Ptolemy had made and used relatively few observations obtained at times and orbital positions he
expected to be especially valuable for construction and testing of his observations. Brahe believed
in blanket coverage of the orbits: who could know what would be relevant and revealing? This was
truly prescient and of immense value to Kepler.
20 J. Barbour
Fig. 10 The exact area law and the speed law that follows from the empty-focus effect both predict
that the orbital speeds at aphelion and perihelion are inversely proportional to the distance from
the Sun
must exert a force on the planet that would be stronger, the closer the planet was to
the Sun. In fact, that was clearly indicated by observations, which showed that the
planets moved fastest when closest to the Sun. Plausibly enough, Kepler guessed
that the speed would be inversely proportional to the distance from the Sun. Strong
support for this came from the fact that the equant law showed the orbital speeds at
aphelion and perihelion to be exactly in inverse proportion (Fig. 10).
But the mathematics of the putative law, applied to the eccentric orbit of Mars,
proved to be beyond Kepler’s abilities. He therefore decided to replace what he
regarded as the exact law by an approximation in the form of the area law! He was
encouraged to this by his recollection of the way Archimedes had estimated the area
of a circle by dividing it into ever smaller segments. As his work progressed and he
gained an increasing number of accurate locations of Mars through his application
of trigonometry to Brahe’s observations, always under the key assumption that the
Sun was the centre of the solar system and the controller of planetary motions, he
came to realize that the area law did actually govern the speed in orbit.
Kepler’s final, very tortuous breakthrough to the joint discovery of ellipticity of
the orbit and the area law was in fact somewhat delayed by his enthusiasm for theory.
The moment he found unambiguous evidence that Mars’s orbit could not be circular,
he started to speculate and initially guessed an egg-shaped orbit, i.e., fatter at one
end than the other. Slowly, as he accumulated more and more accurate locations, the
egg was abandoned. A chance glance at a table of logarithmic tables was what finally
led him to the ellipse—more serendipity. The full truth at last came to him around
Easter 1605.
6 Kepler’s Significance
It would be a futile counterfactual exercise to ask how science would have developed
without Brahe and Kepler’s extraordinary efforts. However, it is entirely possible
that, even without the discovery of the telescope and the possibility that gave for
more accurate observations, decades could have passed before the discovery of the
laws of planetary motion. What is absolutely certain is that Newton’s Principia is
inconceivable without Kepler’s discoveries. All three of Kepler’s laws were impor-
tant: from the third, Newton deduced the 1/r 2 force law for gravity; from the first, that
the planet’s elliptical motions could be understood as the outcome of two competing
tendencies—rectilinear inertial motion and gravitationally induced deflection from

it along the direction to the Sun. In many ways, Kepler’s second law was actually
the most important. Newton recognized this by making it the subject of his very first
proposition in the Principia. It will be worth saying something about this.
By 1670 at the latest, Newton had most of the elements of a rudimentary dynamics,
above all the notion of inertial motion that would persist forever were it not changed
by the action of other bodies. He understood elastic collisions and the nature of
centrifugal force. What hindered a full blossoming of dynamics was the prevail-
ing mechanistic conception of the world due above all to Descartes. According to
this view, all mechanical action took place through direct contact: collisions. The
Cartesian cosmos was a terribly crowded world crammed full of pieces of matter in
continual collision. Newton basically subscribed to this view. Although he had laws
to describe collisions, there was little he could do with them.
The real advance almost certainly came in 1679, when Robert Hooke, newly
appointed as secretary of the Royal Society, pressed Newton hard to confirm his
(Hooke’s) proposal “of compounding the celestial motions of the planets of a direct
motion by the tangent & an attractive motion towards the central body.” Calculations
of Newton that he may well have made as a result of Hooke’s letters of 1679 have
survived and include the key result that became Proposition 1 in the Principia:
Proposition 1 The areas which revolving bodies describe by radii drawn to an

immovable centre of force do lie in the same immovable planes, and are proportional
to the times in which they are described.
This is the theoretical explanation of Kepler’s area law. It had far-reaching impli-
cations, for it told Newton that nature should be described, mathematically at least,
by forces that act over distances. Huygens had coined the expression centrifugal
force; in explicit imitation, Newton called his new forces centripetal. He was well
aware of the revolutionary nature of what he was doing; he was proposing to give
universally despised occult forces a decisive role in physics. He was very cautious
about this and emphasized, in hypotheses non fingo, that he was not making any
assumptions about the physical mode of action of the forces he introduced. What he
did stress was that such forces, introduced mathematically, could explain at a stroke
a vast number of diverse phenomena. They opened up a whole field for exploration
that is still ongoing. Kepler’s 8⊗ led to more than the reformulation of astronomy.
7 Intermezzo: Christian Doppler
Before we move on to Mach and Einstein, brief mention should be made of Christian
Doppler (Fig. 11) and the important eponymous effect that he predicted in 1842 while
a professor at the Czech Polytechnic in Prague. Ironically, Doppler was seeking an
explanation of the different colours of binary stars; the effect he proposed to explain
the difference was physically correct but completely wrong in his application. Binary
22 J. Barbour
Fig. 11 Christian Doppler

(1803–1853)
stars have different colours because, in the first place, their temperatures (and to some
extent their chemical compositions) are different. Doppler suggested, without at that
time any experimental support, that the observed frequency of the light emitted by
the stars depended on their orbital speeds, which are different (and epoch dependent).
This is true, but the effect was far too small to explain the colour differences.
The Dutch meteorologist Buys Ballot (1817–1890) made the first experimental
confirmation of the Doppler effect in 1845 by getting a group of musicians to play a
calibrated note on a train on the line between Utrecht and Amsterdam. Of course, in
those days there were no police sirens that make the effect so evident today. Despite
this confirmation of the effect for sound, Doppler’s proposal of a dependence of the
observed frequency of light on the speed of the source remained controversial for a
surprisingly long time—decades. One person who helped to establish it was Ernst
Mach.
8 Mach and Kinematic Residues in Dynamics
Mach (Fig. 12) was one of the great experimentalists of all times and a man of wide
interests. His name is associated with three very diverse things in science: Mach
bands in psychology, the Mach number in aerodynamics, and Mach’s principle in
the theory of gravity and inertia. Mach was twice nominated for the Nobel Prize for
his discovery of shock waves (Fig. 13), but so many exciting discoveries were being
made in the early 20th century that he missed the honour he deserved.
Fig. 12 Ernst Mach

(1838–1916)
Fig. 13 Mach’s flash

photograph of a supersonic
bullet and the shock wave
generated by it
Before we discuss Mach’s critique of Newton’s absolute space and time and
Einstein’s reaction to it, it is worth reproducing the diagram (Fig. 14) in Blackmore’s
valuable informative biography of Mach [3]. This shows the instrument Mach devised
soon after completing his doctorate to demonstrate the Doppler effect for sound. A
vertical tube AA rotates in the plane perpendicular to the page. Air forced through
the column creates sound in the whistle at G. A person standing in the plane of the
rotating tube hears a clearly modulated pitch of the whistle as it rotates at the end of
the tube, while someone standing some distance way at right angles to the plane of
24 J. Barbour
Fig. 14 Mach’s device to confirm the Doppler effect for sound, reproduced from [3]
rotation hears a constant pitch. The Doppler effect for sound is demonstrated in this
simple way. The apparatus “became a frequently used class demonstration device
throughout Central Europe for many years” [3, p. 19].
To return to our topic, visible and physical markers were key to Kepler’s discov-
eries. The Sun had a dual role: it defined and created motion of the planets. As for
the stars, Kepler, like Copernicus, declared them the ultimate frame of reference, by
definition at rest. His laws did not contradict this, and the stars did not exhibit any
relative motion.
The star-studded shell of this closed world retained the Sun’s warmth generated at
its centre, or focus. Kepler introduced the Latin word for hearth into scientific usage,
first in optics and then in astronomy. Descartes (1596–1650) shattered Kepler’s cosy
‘home’ when he introduced the mechanical philosophy. He, above all, marks the
transition from the closed world to the infinite universe. In it, all bodies, including
the stars and their constituents, move relative to each other.
Descartes actually had two diametrically opposed concepts of motion: absolute
and relative. The origins of both are worth retelling (for more details, see [1]). Let
us start with the first. One day, lying on his bed, he is said to have spotted a fly on
the ceiling and saw he could fix its position by its two distances from the walls. The
story is ben trovato, apposite even if invented. Cartesian coordinates, so convenient
for defining straight lines, were born. The idea of rectilinear inertial motion (not yet
named so) was already in circulation; in a book, Le Monde, ready for publication in
1632, Descartes made it the foundation of mechanics long before Newton.
Implicit here is an unchanging reified space like the ceiling on which the fly
crawled: Newton’s absolute frame in all but name. No longer do the Sun and stars
define motion. Space does. Whereas pre-Kepler void points governed the planets’
motions, now invisible space controls all motion. This is occult forces in spades!
Descartes was just about to publish Le Monde, which assumed correctness of
the Copernican cosmology, when he heard about Galileo’s condemnation by the
Inquisition. In alarm—piety he claimed—Descartes hurriedly withdrew his book
and thought long and hard how he could save his mechanical philosophy. Eventually
he introduced a quite different definition of place and motion in his Principles of
Philosophy (1644). He declared all motion to be relative. Now any one body has
infinitely many positions and motions according to which bodies are used to define
them. However, he did grant the existence of ‘one true philosophical definition of
position’, according to which the position of any one body is defined by its envelope,
i.e., the immediately adjacent matter that surrounds it. The reason for this definition,
actually a throwback to the Aristotelian notion of topos, is to be found in Descartes’
contention that the Earth is carried around the Sun by a vortex, which is thus its imme-
diate envelope. The point then is that the Earth does not move relative to the vortex
and therefore does not move in accordance with the true definition. Since terres-
trial mobility (and not heliocentricity) was the Inquisition’s objection to Copernicus,
Descartes felt he had secured his position and explicitly stated that in accordance
with his proposal the Earth does not move.
But after this avowal of pure relationalism, Descartes, failing to note the contra-
diction, reverted to uniform rectilinear motion as the first principle of mechanics.
This made no sense in a world with position and motion defined relatively in either
way. It required an implicit absolute space.
Newton studied Descartes’ book closely and did see the contradiction. Knowing
what could be done with the law of inertia, he recoiled from the virtual impossibility
of expressing it rigourously in Descartes’ shifting cosmos. The prominence given
to absolute space and time in the Scholium at the start of the Principia are a covert
dismissal of Descartes, even though Newton does grant the great difficulty of distin-
guishing ”the true motions of particular bodies from the apparent; because the parts
of that immovable space in which those motions are performed do by no means come
under the observation of our senses.”
In fact, as Mach was later to remark, Newton’s laws were never verified relative
to absolute space and time but to exactly the same referents that Kepler had used: the
effectively fixed stars and the time-measuring clock supplied by the diurnal revolution
of the stars.
Descartes’ absolute and relative are the origin of the reductionistic–holistic
dichotomy. Mach the holist reacted to Newton the reductionist when he spoke of
‘immediate connections’ and the ideal that hovers before the natural investigator as
‘an insight into the principles of the whole matter’. The essence of reductionism is
threefold: simple objects, atoms, that move in accordance with simple laws, primarily
the law of inertia, in a simple background: absolute space. But if position is relative,
only the totality of separations between objects is real: the world is held together by
an indissoluble network of relations, and history is nothing but their evolution. As
26 J. Barbour
Mach said: “The universe is not twice given, with an earth at rest and an earth in
motion; but only once, with its relative motions alone determinable” (Mach, p.284).
But Mach went much further than this epistemological verity. Kepler had given
the Sun a dynamical role. Mach extended it to the stars or, rather, the totality of
masses of the Universe. They should not only define but also control motion. That
Mach envisaged this done by some as yet unknown physical mechanism is confirmed
by his famous refutation of Newton’s bucket argument 13 for absolute motion:
“Newton’s experiment with the rotating vessel of water simply informs us that
the relative rotation of the water with respect to the sides of the vessel produces
no noticeable centrifugal forces, but that such forces are produced by its relative
rotation with respect to the mass of the earth and the other celestial bodies. No one
is competent to say how the experiment would turn out if the sides of the vessel
increased in thickness and mass till they were ultimately several leagues thick.”
It is well known that Mach’s critique made a powerful impression on the late
teenage Einstein. It was the main stimulus to his attempt to eliminate all trace of
Newton’s absolute space through the creation of general relativity. The idea that
the totality of the masses of the universe ‘work together’ to create the local inertial
frames in which force-free bodies move rectilinearly and uniformly is what Einstein
called Mach’s principle. For a variety of reasons, this has had a tangled history, for
which several factors are responsible.
9 Einstein’s Reaction to Mach
Let me start with Einstein’s strange confusion of two distinct meanings of inertia:
there is inertial motion, as defined in Newton’s first law, and inertial mass. That they
are distinct is evident: the concept of inertial mass does not enter into the statement
of the first law. Mach gave a much admired operational definition of inertial mass,
which he defined through the accelerations bodies impart to each other when they
interact. These are inversely proportional to their intrinsic inertial masses. Mach’s
disagreement with Newton on this score was not about substance but proper for-
mulation. What really concerned Mach was the origin of inertial motion: Newton
believed absolute space governed it, Mach the totality of masses in the universe.
Reading Einstein’s various comments about Mach and inertia I am forced to
conclude he was the victim of semantic confusion. He does not seem to have seen
any difference between the two meanings of the word inertia. His most egregious
distortion of Mach is in the 1917 paper in which he laid the foundation of modern
relativistic cosmology. He claims:
13 Newton introduced the bucket to make a serious scientific argument but simultaneously a fool of
Descartes, whose mechanical philosophy relied heavily on centrifugal force. Many people writing
on the absolute–relative debate and unaware of the background to the bucket argument have been
misled into thinking the issue is about the difference between linear and circular motion, which is
not true. In fact, I increasingly think Newton confused himself.
Einsteinian
time Each observer makes
a different split
into space and time
Machian
In each instant the positions
of objects are defined
relative to each other
Fig. 15 The two quite different meanings of relativity. Relativity as defined by Minkowski and
Einstein refers to the ambiguity in the splitting of four-dimensional spacetime into time and space.
Relativity as defined by Mach means that the position of any one body is defined at a given instant by
its distances to all the other bodies in the Universe at that instant. The conflict of concepts is evident:
Machian relativity makes complete sense if there exists a distinguished notion of simultaneity, but
that is denied as the first principle of Einsteinian relativity
“In a consistent theory of relativity there can be no inertia relatively to ‘space’,

but only an inertia of masses relatively to one another. If, therefore, I have a mass
at a sufficient distance from all other masses in the universe, its inertia must fall to
zero.”
Mach would have dismissed this comment as a gross distortion of his ideas;
Einstein is clearly substituting a bogus issue about inertial mass for Mach’s proper
concern with inertial motion. Unfortunately, Einstein’s 1917 comment led to several
misguided attempts to implemented a Mach’s principle along inappropriate lines.
However, the complexity of the Machian issue has a much more solid basis and
raises a real dilemma, which is illustrated in Fig. 15. What is at stake is the very
meaning of the word relativity. Einstein and Minkowski had in mind the observer
dependence of the split of spacetime into space and time and, more generally, to the
complete freedom to lay down coordinates on spacetime in any suitably continuous
way. Einstein spoke of general covariance; today one speaks of four-dimensional dif-
feomorphism invariance. That may be called Einsteinian relativity. The most impor-
tant aspect of it is the denial of simultaneity as a physically significant concept.
In contrast, relativity as originally formulated by Mach makes no sense without
an underlying notion of simultaneity: it asserts that the position of any given object
at a given instant is defined by its distance to all the other objects in the universe in
that instant. Most relativists today would say that this is a hopelessly obsolete notion
because Einstein and Minkowski showed that our intuitive notion of simultaneity
has no counterpart in the physical world. Does this mean that Mach’s principle is a
dead duck?
Not necessarily. In his only article not devoted to quantum mechanics, John Bell
wrote on special relativity and sought “to drive home the lesson that the laws of
28 J. Barbour
physics in any one reference frame account for all physical phenomena, including
the observations of moving observers” [4, p. 77]. Attempts to find a generically
distinguished frame in Minkowski space are doomed to fail on account of its high
symmetry, but the case is altered in GR: gravity brings structure into spacetime. I
want to use the remainder of this chapter to give what I believe is the correct definition
of Mach’s principle and to show how Einsteinian gravity is much more compatible
with Machian relativity than one might imagine. I shall even suggest that Machian
relativity is the deeper principle.
However, I must first briefly recount how Einstein set out to implement the idea that
Mach had espoused: that inertial motion should not be governed by absolute space
but the totality of masses in the Universe (a more detailed account is given in [5]).
What was decisive for Einstein was his discovery of special relativity in 1905. This
arose from his successful reconciliation of Maxwell’s electrodynamics, with its only
apparent need for an ether, and Galilean relativity applied to all physical phenomena.
The lesson Einstein drew from his success was that uniform motion through Newton’s
absolute space could not be defined—it was impossible to associate any actual speed
with it. Although he said nothing explicit at the time, this result already suggested to
Einstein the way to implement Mach’s idea: to show that the alleged absolute space
had no observable effects at all, for then one could argue that it does not exist. The
impossibility of determining a speed of uniform motion through space was the first
step in that direction.
The decisive idea that set in motion Einstein’s long search for a new theory
of gravity was the equivalence principle, that ‘happiest thought’ of his life which
occurred to Einstein in 1907. Its importance for Einstein was not so much that gravity
and inertia are identical in essence but the possibility “that the principle of relativity
is also satisfied for systems moving relatively to each other with acceleration” [6].
The equivalence principle suggested that this could be done for at least uniform
accelerations.
Einstein’s strategy from then on was clear and settled. He would attempt to extend
the relativity principle ever further. The next step, clearly suggested by Mach’s retort
to Newton’s bucket argument, called for extension to uniform circular motion, for
which the magnitude of the acceleration, as in the equivalence-principle elevator,
is constant in magnitude but its direction changes. From there, the logical step to
complete relativity of motion was not too difficult. Einstein advanced the principle
of general covariance as the physical foundation of the new theory of gravity he was
seeking.
Two aspects of Einstein’s approach should be noted. His principle of relativity did
not in any way directly address the way in which the Universe itself behaved. It merely
said that the description of its behaviour should be the same in whatever coordinate
system one cared to describe it. Einstein’s immediate acceptance of Kretschmann’s
objection that general covariance in itself had no physical content but was merely
a requirement of mathematical consistency was a remarkable volte-face that has
generated much argument and confusion about the foundations of general relativity
and, in particular, Mach’s principle. The only conclusion I wish to draw from this
brief discussion is that Einstein did not attempt a direct implementation of Mach’s
ideas but attacked the problem indirectly. This comes out especially clearly in a
comment he made in 1918 [7]:
“We want to distinguish more clearly between quantities that belong to a physical
system as such ... and quantities that depend on the coordinate system. Ones initial
reaction would be to require that physics should introduce in its laws only quantities of
the first kind. However, the scientific development has not confirmed this conjecture.
It cannot dispense with coordinate systems.”
There is a clear anticipation here of the distinction, now commonplace due to
developments in gauge theory, between so-called true degrees of freedom and redun-
dant degrees of freedom. What I want to question is whether Einstein had correctly
identified what are the “quantities that belong to a physical system as such”. In my
final section before brief conclusions, I wish to suggest that he may have made the
incorrect identification.
10 The Machian Approach: Shape Dynamics
There is no doubt what Mach regarded as the true physical quantities: bodies that
possess intrinsic mass and distances (in Euclidean space) between them. He most
certainly did not think time had any ontological reality [2, p. 273]: “It is utterly
beyond our power to measure the changes of things by time. Quite the contrary, time
is an abstraction at which we arrive from the changes of things.” But Minkowski,
followed by Einstein, had given time the same ontological status as space. This
led Einstein to identify the “quantities that belong to a physical system as such”
with four-dimensional spacetime intervals, whereas Mach had identified them with
exclusively three-dimensional spatial entities. Let us see where such a standpoint
takes us.
Let us start with one thing on which we can be sure Mach and Einstein would
have agreed: if the local frames in which force-free particles move inertially are
determined by the universe, there must be a sense in which the universe is a closed
dynamical system, for otherwise one could never close the circle and say the whole
determines the parts: local inertial frames. This underlying sense is implicit in Mach14
and explicit in Einstein’s 1917 cosmological model.
Let us then allow the notion of simultaneity and assume that the universe is
a closed dynamical system. We can consider two models: an island universe of
N point particles, which matches the ontology of Mach’s original proposal, and a
three-dimensional Riemannian geometry closed up on itself,which corresponds to
14 See his comment (Mach, p. 287) “Nature does not begin with elements, as we are obliged to
begin with them. It is certainly fortunate for us that we can, from time to time, turn aside our eyes
from the overpowering unity of the All, and allow them to rest on individual details. But we should
not omit, ultimately to complete and correct our views by a thorough consideration of the things
which for the time being we left out of account.” How can completion come without a definite sense
in which the universe is closed?
30 J. Barbour
closed-space vacuum GR. In the point-particle case, the difference between Newton
and Mach is easily expressed in terms of configuration spaces. Newton’s is R 3N , three
coordinates for each particles. But Mach said only the inter-particle separations are
real. We need to quotient R 3N by the Euclidean translations and rotations to obtain the
3N -6-dimensional Machian relative configuration space (RCS). Absolute position
and orientation are removed from the RCS. In fact, although Mach did not recognize
the need, one must go a step further since distance presupposes an absolute scale.
We also need to quotient by dilatations; this takes us to the 3N -7-dimensional shape
space S . I would say that an instantaneous shape of the universe matches Mach’s
requirement that we grasp the ‘immediate connections’.
However, that does not yet mean that we have gained ‘an insight into the principles
of the whole matter’. Machian histories of the universe will be curves in S . The issue
now is this: what determines these curves? The fact is that any Newtonian history
can be represented as a curve in S : one simply plots the representative points of the
successive shapes. In what way would a Machian history be distinguished from an
arbitrary Newtonian one plotted in S ?
A problem with Mach is that he tended to speak in general intuitive terms. It
is here that a penetrating analysis by Poincaré [8], who analyzed the problem in
much more precise terms, provides the guide. Poincaré asked: what defect, if any,
arises from Newton’s use of absolute space? His answer was that a true believer
in relationalism, convinced that only inter-particle separations rab have physical
significance, would pose the initial-value problem of particle dynamics in these
terms: rab , ṙab should determine the evolution rab (t), a, b = 1, 2, ..., N , uniquely.
This matches the formulation in terms of the particle coordinates and velocities;
in accordance with Laplacian determinism, r a , ṙ a , a = 1, 2, ..., N , determine the
evolution.
Poincaré pointed out that the rather natural transfer of this requirement from R 3N
to the RCS fails. The reason is that the data rab , ṙab contain no information about
the angular momentum L in the system, whereas this information is encoded in
r a , ṙ a (under the assumption that the masses are known). Although the presence or
absence of L is undetectable in rab , ṙab initial data, the curves that result do encode
information about L. Poincaré said that this fact, reflected in the manifest presence
of angular momentum in the solar system, was the true evidence for the existence of
a dynamically active agent in addition to the separations rab and their rates of change
ṙab . As a convinced believer that only relative motions should have dynamical effect,
Poincaré said he found this state of affairs repugnant but that it was necessary to
accept the empirical evidence.
It is strange that Poincaré did not consider a Machian resolution to the problem,
namely that for for a dynamically closed universe as a whole the relative data do
determine the future uniquely. One can then attribute the failure of this requirement
in subsystems of the universe precisely to the fact that the masses of the universe
do determine local inertial frames of reference. Poincaré formulated his ideas in the
context of the RCS, but they can be directly extended to and made more stringent
in shape space. This leads me to the formulation of the Mach-Poincaré principle for
particle dynamics in these terms.
Mach–Poincarè Principle. Specification of a point and direction (strong form)

or point and tangent vector (weak form) in shape space S should determine the
evolution in S uniquely.15
It is necessary to allow for the weaker form if one is to model expansion of the
Universe. For discussion of this delicate issue, see my introduction to shape dynamics
[9]. The extension to dynamical geometry is relatively obvious; the shape space in this
case is conformal superspace, which provides the natural framework for describing
the dynamics of three-dimensional conformal geometries. I cannot describe in detail
this work; the most important chapters are [10–12] (see also Koslowski’s contribution
to this conference proceedings).
What one can say is that, if the Universe is spatially closed, there is a well-defined
sense in which GR implements Mach–Poincaré principle in the weak form very
well, indeed perfectly if there is no cosmological constant. For all the details I must
refer the reader to the references already cited, but the key conclusions do need to
be stated, at least for vacuum gravity. First, by virtue of its clearly formulated first
principles shape dynamics introduces of necessity a notion of simultaneity into GR
and insists that the physical entity which is evolving is the conformal three-geometry
on successive leaves of a foliation of spacetime by hypersurfaces of constant-mean-
(extrinsic)-curvature (CMC). Second, the spacetime in which these hypersurfaces are
embedded is completely determined by specification of a point and tangent vector
in conformal superspace. This fact was first demonstrated in [13].
If the ideas of shape dynamics, which do follow very naturally from Mach’s
ideas, are vindicated, it will be incorrect to view shape dynamics as a rule to select
certain special solutions—those that are globally hyperbolic and CMC foliable—
from among the full set allowed by GR. Rather GR might have to be seen as an
extension of shape dynamics beyond its physical domain. I suspect we shall have to
await the quantum theory of gravity to see if this view is justified.
11 Conclusions
Galileo said “He that attempts natural philosophy without geometry is lost.” He
meant of course three-dimensional geometry, which was still Euclidean in his day,
though I am sure he would have greeted Riemann’s generalization with enthusiasm.
15 In my mind, the great virtue of Poincaré’s analysis is that he formulates requirements on the
form of a dynamical theory of the Universe in terms of the initial data that one regards as belonging
“to the physical system as such”. This allows a much more precise formulation than Einstein’s
requirement that all coordinate systems should be on an equal footing, which is actually void of
content, or that the action should satisfy certain symmetry requirements, which is also amenable
to adjustment, as one sees with the passage from standard Newtonian dynamics to parametrized
particle dynamics, which adds reparametrization invariance as a symmetry without changing the
physical content of the theory. In contrast, implementation of Mach’s ideas boils down (in the
case of the weak Mach–Poincaré principle) to identification of the true (configurational) degrees of
freedom and construction of a theory in which they and their velocities wrt an independent variable
uniquely determine the evolution of the true degrees of freedom.
32 J. Barbour
The first step in the still ongoing creation of the dynamical theory of the Universe
was Hipparchus’s theory of the Sun’s motion. It is important that all the great work
in astronomy reviewed in this chapter studied the evolution of intrinsic shapes, of
which the fixed stars formed part. Absolute position, orientation and size played
role at all; I have already emphasized that every conclusion drawn in astronomy
was based on measurement of angles between observed physical objects. These
included measurements of what was called time but was actually the diurnal rotation
of the stars. Even now, with geometry curved and made dynamical, the irreducible
epistemological basis of science is observed angles. We now see the Universe as
almost infinitely flexible, but we cannot do without angles.
A conformal geometry supplies the ‘immediate connections’ that Mach exhorted
us to grasp. As regards “the principles of the whole matter”, I would say that as far
as classical physics is concerned they are encapsulated in the weak Mach–Poincaré
principle applied to a closed Universe whose possible spatial configurations are
defined by conformal geometry.
Let me end with my sincere thanks to the organizers, above all Jiří Bičák, for the
invitation to speak at the wonderful conference in Prague.
References
1. Barbour, J.B.: Absolute or Relative Motion?: The Discovery of Dynamics. Absolute or Rel-
ative Motion?: A Study from a Machian Point of View of the Discovery and the Structure of
Dynamical Theories. The Discovery of Dynamics. Cambridge University Press, Cambridge
(1989)
2. Mach, E.: The Science of Mechanics. Open Court, LaSalle (1960)
3. Blackmore, J.T.: Ernst Mach. His Life, Work, and Influence. University of California Press,
Berkeley (1972)
4. Bell, J.S.: Speakable and Unspeakable in Quantum Mechanic. Cambridge University Press,
Cambridge (1987)
5. Barbour, J.B.: The Part Played by Mach’s Principle in the Genesis of Relativistic Cosmology,
p. 47. Cambridge University Press, Cambridge (1990)
6. Einstein, A.: Berichtigungen zu der Arbeit: “Über das Relativitätsprinzip und die aus demselben
gezogenen Folgerungen”. Jahrbuch der Radioaktivität und Elektronik 5, 98 (1908)
7. Einstein, A.: Prinzipielles zur allgemeinen Relativitätstheorie. Annalen der Physik 360, 241
(1918). doi:10.1002/andp.19183600402
8. Poincaré, H.: Science and Hypothesis. Walter Scott, London (1905)
9. Barbour, J.B.: Shape Dynamics: An Introduction. Birkäuser, Berlin (2012)
10. Anderson, E., Barbour, J., Foster, B.Z., Kelleher, B., Murchadha, N.Ó.: The physical gravita-
tional degrees of freedom. Class. Quantum Gravity 22, 1795 (2005). doi:10.1088/0264-9381/
22/9/020
11. Gomes, H., Gryb, S., Koslowski, T.: Einstein gravity as a 3D conformally invariant theory.
Class. Quantum Gravity 28(4), 045005 (2011). doi:10.1088/0264-9381/28/4/045005
12. Gomes, H., Koslowski, T.: The link between general relativity and shape dynamics. Class.
Quantum Gravity 29(7), 075009 (2012). doi:10.1088/0264-9381/29/7/075009
13. Barbour, J., Murchadha, N.Ó.: Conformal superspace: the configuration space of general rel-
ativity. ArXiv e-prints arxiv:1009.3559 [gr-qc] (2010)
Einstein in Prague: Relativity
Then and Now
Jiří Bičák
Abstract It was during his stay in Prague that Einstein started in earnest to develop
his ideas about general relativity. I will recall those days in 1911 and 1912, discuss
Einstein’s papers on gravitation from that period and emphasize which new concepts
and ideas he introduced. I also want to indicate how the main themes that preoccupied
him then, the principle of equivalence, bending of light, gravitational redshift and
frame dragging effects, are alive in contemporary relativity.
1 Introduction
I would like to start as I did in my talk at the conference, quoting what Einstein
wrote soon after his arrival in Prague on April 3, 1911: “The city of Prague is very
fine, so beautiful that it is worth a long journey for itself” (from the letter to his
friend M. Besso on May 13, 1911); or, “I have a magnificent institute here in which
I work very comfortably. . . By the way, Czechs are much more harmless than one
thinks” (from the letter to M. Grossmann on April 27, 1911). I hope the conference
participants, 100 years after Einstein in Prague, had a similar impression.
The quote from Hesiod’s “Works and Days” from the seventh century BC—The
price of achievement is toil [‘Schinderei’]; and the gods have ruled that you must
pay in advance—enables me to give a brief summary: Einstein paid much in Prague.
The days: April 1911–July 1912. The works: principle of equivalence, light bending,
dragging of inertial frames; features of a future theory of gravity.
It was not until 1911, only after his arrival in Prague, that Einstein’s interest in
quantum theory started to diminish and his systematic concentration on the problems
of a new theory of gravity began. There were specific issues which Einstein analyzed
J. Bičák (B)
Institute of Theoretical Physics, Charles University in Prague, V Holešovičkách 2,
180 00 Prague, Czech Republic
e-mail: bicak.troja@gmail.com

34 J. Bičák
in Prague, for example that the bending of light due to the Sun’s gravitational field
is observable. However, as we shall see later, just before leaving Prague, on July 4,
1912, Einstein submitted a short paper to Annalen der Physik in which a number of
fundamental features of the final general relativity were anticipated. Concerning new
ideas and liberating oneself from old views, especially in this respect, “Einstein paid
much in Prague”. Before we turn to Einstein’s work on gravity in Prague we shall
recall his days here. We cannot describe the atmosphere in Prague of those days,
with tensions between three main groups of its inhabitants, Czechs, Germans and
Jews, on one hand, and with many intellectually inspiring aspects and interactions
between them on the other hand, in any detail, but literature of a special interest will
be mentioned.
2 Why and How He was Invited to Prague
Charles University in Prague, founded in 1348 as the first university “beyond the
Alps”, was originally one educational center for Czechs, Germans, Poles and various
south-European nations. Nationalism led to its division in 1882 into the German and
Czech parts.1 In 1911 the Czech part had 4 432 students, the German part 1 844
students but, for example, Mach’s lectures in the German part were visited by a
number of Czech students. In the German part, the head of mathematical physics,
Professor F. Lippich, was due to retire and the German university decided to have the
Institute of Theoretical Physics that existed in the Czech University already. The chief
advocate of the proposal to appoint Albert Einstein was Anton Lampa, professor of
experimental physics, a great “Machian”, a Czech by origin but an ardent supporter
of Germanization who hoped that Einstein would develop further Mach’s ideas.2
Another member of the commission for choosing a candidate for theoretical physics
was Georg Pick, a mathematician of broad interests, with whom Einstein made later,
during his stay in Prague, a close friendship. In his recommendation letter from 1910,
Max Planck wrote that if Einstein’s theories were to be confirmed, Einstein would
be considered the Copernicus of the twentieth century.
1 In 1879/80 and 1883/84 Ernst Mach was elected Rector of the University. However, during his
second Rectorship he resigned since he disapproved of the division of Prague University into Czech
and German parts.
2 Anton Lampa (1968–1938) was an experimental physicist, interested also in philosophy, history
and cultural aspects of physics. Interestingly, he was the first to publish a paper on the appearance of
a moving rod according to special relativity (Zeits. f. Phys., 1924). For the life and work of Lampa,
see [1].
Einstein in Prague: Relativity Then and Now 35
3 Days in Prague
No view of Einstein in Prague can be more original and immediate than that of
Philipp Frank,3 Einstein’s successor as head of the Institute of Theoretical Physics in
the German university in Prague until 1938. In 1929, Frank organized the famous first
meeting on the Epistemology of the Exact Sciences in Prague. From 1939 till 1941
he wrote the biography of Einstein [3], which Einstein endorsed. Its whole Chapter
IV, “Einstein at Prague”, besides information on Einstein and the theory of relativity,
paints amusingly also the atmosphere in Prague at the beginning of the twentieth
century. Frank, when writing Einstein’s biography, did not have manuscripts and other
documents at his disposal, so for an earnest Einstein biographer, other sources are
important. Nevertheless, Frank tells various entertaining stories which are probably
not far from the truth. So, for example, it was the custom for a newly appointed
professor to pay a visit to his colleagues. Einstein started to pay these visits and
decided to go first to romantic old parts of Prague. But later he found that these calls,
which numbered around forty, were really a waste of time and he stopped his visits.
The professors whom he had not visited felt puzzled but in fact the main reason was
that they did not live in interesting parts of the city or their name was far back in the
alphabet. Another nice story supports our point that it was in Prague where Einstein
for the first time started to concentrate on gravity rather than on quantum physics.
Frank recalls how during his first visit of Einstein in Prague, Einstein took him to
the window of his office from which they could overlook a large garden behind the
wall (see Fig. 1, with the wall on the right). Einstein told Frank that he saw people
there in deep meditation but also groups very vividly discussing. Only quite later he
learned that the park belonged to the insane asylum of Bohemia. Einstein pointed to
people walking there, turned to Frank and said: “Those are the madmen who do not
occupy themselves with the quantum theory”. This is not true today. The park and
gardens still belong to the mental hospital, however, it is open to the public during
the day. And physicists from our Faculty nearby quite often walk across it to get to
the center of the city.
In Prague, Einstein associated himself with a group of Jewish intellectuals who
gathered in the evenings at Berta Fanta’s salon in the house “At the Unicorn” in the
Old Town Square. There philosophy was discussed and music played. Einstein met
an ardent Zionist, Hugo Bergmann, the son-in-law of Berta Fanta, but he was then not
able to arouse Einstein’s interest in his ideas. In the letter to Hedwig Born in 1916,
Einstein wrote that Zionists in Prague are “a small troop of unrealistic people, harking
back to the Middle Ages”. Still later, after Bergmann became a Hebrew University
professor in Jerusalem, Einstein called him “the serious saint from Prague”. During
3 P. Frank (1884, Vienna—1966, Cambridge, Mass.) was a theoretical physicist and logical positivist,
a member of the Vienna Circle. His work in relativity is summarized in detail in the comprehensive
article by Havas [2]. With R. von Mises, Frank published a book on differential and integral equations
in physics; later he wrote several books on the philosophy of science. His booklet Relativity–a richer
truth, with a foreword by Albert Einstein on the “Laws of science and the laws of ethics”, published
in 1951, is less known. It touches on a number of philosophical and ethical issues.
36 J. Bičák
Fig. 1 The building of the former Faculty of Philosophy of the German University on Viničná 7
wherein Einstein had “an excellent Institute with a beautiful library”
the evenings Einstein perhaps met Franz Kafka, although Kafka, in fact, did not like
to go there, but he certainly was acquainted with Max Brod, a writer and journalist
who later, after Czechoslovakia was founded in 1918, played a significant role in
promoting the Czech culture. Almost all of Einstein biographers, inspired by Philipp
Frank, make the point that Max Brod drew on Einstein’s character for his portrait of
Kepler in his novel Tycho Brahe’s Way to God and some even suggest that in this
way certain egocentric features of Einstein’s personality were disclosed. However,
when reading Brod’s autobiography Streitbares Leben, we discover that he was quite
unhappy with Frank’s interpretation and even wrote a letter to Einstein to explain
that he never noticed any egocentric features in his behavior and, in any case, that it
was rather the poet and writer Franz Werfel, Brod’s friend, who contributed to the
portrait of Kepler.
Einstein visited Prague once again in 1921 when he accepted an invitation from
Urania, Prague’s German Society, to give a lecture on the theory of relativity. Accom-
panied by Frank, he visited the Physical Institute of the Czech part of Charles Univer-
sity. “By this visit Einstein wanted to express his sympathy for the new Czechoslo-
vak Republic and its democratic policy under Masaryk’s leadership”, writes Frank.
Abraham Pais in his celebrated biography of Einstein [4] gives the list of people Ein-
stein suggested for the Nobel Prize. Masaryk is among them, proposed by Einstein
for the Peace Prize. Einstein was later in correspondence with Masaryk regarding
the fate of a pacifist Přemysl Pitter.4
4 Copies of their letters are available in “Einstein Archives Online”—see http://www.alberteinstein.

info/
4 The Czech Culture and Science Responding

to Einstein’s Work
Karel Čapek (1890–1938), a humanist with encyclopedic knowledge, one of the

best known writers and journalists of the Masaryk era, who was a close friend of
Masaryk, wrote “philosophical” novels, charming detective stories, anti-war dra-
mas, and also was a pioneer of the Czech science fiction. Alan J. Friedman and
Carol C. Donley in their book Einstein as myth and muse (Cambridge University
Press 1985) start their section “Approaches to relativity in fiction” emphasizing that
from the use of the profoundly wrong aphorism “everything is relative”, various
authors explore intricate possibilities of Einstein and his theories. And they con-
tinue: “A remarkable early exposition of the possibilities appeared in 1924, with
Karel Čapek’s novel Krakatit. Čapek’s awareness of science and technology was
indicated by mentions. . . of the leading scientists of the day, including Einstein,
Rutherford, Planck, Bohr and Millikan. The plot concerns an inventor who has dis-
covered a way to release atomic energy. . . The technical details are as accurate as
they could be in the early 1920s, and atomic energy is correctly seen as a possi-
bility emerging from the radioactivity work of Becquerel and Rutherford, and not
from Einstein’s theories. . . The inventor, Prokop, is torn in the traditional struggle
between God and the devil. . . Prokop’s bewilderment, in the literal form of a fever,
is described by the first metaphor from relativity: . . . It appeared he was moving with
velocity approaching velocity of light; in some way his heart was compressed. But
that was only Fitzgerald-Lorentz contraction. . . ” Prokop then finds himself in the
closed Einstein universe. . . Curiously, Einstein’s closed universe attracted also one
of the most sophisticated Czech art critics and writers, F. X. Šalda. In 1928 already,
he makes analogies between Einstein’s conception of the finite, closed Universe and
the conception of space in paintings by Cézanne.
We cannot continue here with more examples of the inspiring role which
Einstein’s theories exerted on the Czech culture. Let us just look at Fig. 2 where
the portrait of Einstein by well-known Czech oil painter and graphic artist Max
Švabinský (1873–1962) is shown. In a recent interview in the University magazine
Babylon, Švabinský’s son-in-law, originally a mathematician, declared that he sent
one of the copies of this lithography to Robert Oppenheimer and it was hanging on
the wall of his office in the Institute of Advanced Studies in Princeton.
4.1 Impact on Czech Physics and Astronomy
Relativity theory was popularized and even taught quite soon by the Czech physicists
and astronomers. The first papers were written by A. Dittrich and A. Žáček in 1912.
One of the main protagonists of Einstein’s theories was professor of theoretical
physics at the Czech part of the Charles University, František Záviška (1879–1945).
38 J. Bičák
Fig. 2 Einstein’s portrait from 1955 by Max Švabinský
In 1925, his semi-popular book Einstein’s principle of relativity and theory of gravity
including basic principles of general relativity appeared. He was in good relations
with Philipp Frank and translated Frank’s book Das Ende der mechanischen Physik
(The end of mechanistic Physics), published in 1935, into Czech. It is a criticism of
the totalitarian (Nazi) philosophy from the point of view of the theory of knowledge
and philosophy of science. An interesting impact of Einstein’s prediction of light
bending on a Czech astronomer, F. Link, will be discussed below. At present there
is a rather extensive literature on Einstein’s influence on culture and science in the
Czech lands between World Wars I and II available, and on Einstein’s work done
during the Prague stay.5
5 To give some examples, we quote a booklet [5] published on the occasion of Einstein’s centenary
in 1979 (in Czech and partially in German), the 3rd number of the Czechoslovak Journal of Physics
from the same year dedicated to Einstein and the comprehensive article on his route to general
relativity, concentrated primarily on the Prague period [6]. Two articles in English [7, 8] are texts
of talks about Einstein’s Prague papers on gravity given at the Conference of the European Physical
Society in Prague in 1984 and at the Marcel Grossmann meeting in Perth in 1988. And a very
recent detailed work [9] by Těšínská, a historian of science (containing 57 references) concentrates
After World War II Miroslav Brdička (1912–2007) wrote one of the first original
papers on general relativity (“On gravitational waves”, Proc. Roy. Irish Acad., 1951)
after his stay 1948–1949 as a scholar in the Dublin Institute for Advanced Studies.
It has been curious to see this work quoted and Brdička’s name observed in the title
of the very recent paper by G. Gibbons and C. Rugina.6 The communist upheaval in
Czechoslovakia in 1948 led to a large wave of emigration. Within this there was also
a geometer Václav Hlavatý who became professor at Indiana University and started
to work extensively in relativity, in particular on Einstein’s unified field theories. In
another wave of emigration, after August 1968, Karel Kuchař came first to Princeton,
following the invitation of John Wheeler, and then became professor in Salt Lake
City. His influential work in the quantum theory of covariant systems, canonical
quantum gravity and the issue of time is well known to contemporary relativists,
including some of participants of this conference.
Finally, I am glad to say that at present there are several groups active in relativity,
relativistic astrophysics and cosmology in the Czech Republic—at the Faculty of
Mathematics and Physics (Theoretical Physics, Astronomy), in the Academy of Sci-
ences (Institute of Astronomy, Mathematical Institute) in Prague, and at the Silesia
University in Opava and Masaryk University in Brno. A volume Gravitation: follow-
ing the Prague inspiration [10] contains comprehensive essays by 14 Czechoslovak
relativists and astrophysicists about their work. Some most recent results are included
in the Proceedings [11] of this conference, containing contributions based on oral
and poster presentations.
Before I finish this “Czech intermezzo” let me present the Fig. 3 from a two-day
celebratory meeting on the occasion of Einstein’s centenary which took place in the
Carolinum–the same place as our conference–on February 26 and 27, 1979. Among
more than 200 participants there were several distinguished guests from abroad,
including two associated directly with Einstein—P. G. Bergmann and J. A. Wheeler.
Peter Bergmann met his wife Margot when they both were students of Physics at
Prague German University.7
5 Lectures, Seminars and Papers of Albert Einstein in Prague
Below in Table 1 we see that the lectures Einstein gave in Prague were on classical
subjects like mechanics, thermodynamics and molecular theory of heat. It is not
known whether in the seminars he organized more modern topics were included.
(Footnote 5 continued)
on Einstein’s call to Prague and on development of theoretical physics at the Czech university in
Prague in relation to Einstein’s work.
6 See Coryacher-Chaplygin, Kovalevskaya, and Brdička-Eardley-Nappi-Witten pp-waves space-
times with higher rank Stäckel-Killing tensor, J. Math. Phys 52, 122901 (2011).
7 Unfortunately they did not recall where Einstein’s or even Frank’s office was. However, their faces
turned into a fine smile when we went through the big door to see the Einstein memorial tablet in
the entrance hall inside because they could clearly identify the door into the building in Fig. 1.
40 J. Bičák
Fig. 3 During Einstein’s centenary celebrations outside Carolinum in February 1979: John
Archibald Wheeler, Andrzej Trautman, Mrs. Melcher, Ernst Schmutzer, Jiří Langer, Margot
Bergmann, Peter Bergmann, and Horst Melcher (from left to right). The present conference took
place in the same place... (Photograph taken by the author)
Table 1 Einstein’s lectures and seminars in Prague

Period Title No. of students
20.4.1911–31.7.1911 Mechanik diskreter Massenpunkte (3h) 13
Thermodynamik (2h) 12
Seminar 6
19.10.1911–27.3.1912 Mechanik diskreter Massenpunkte (3h) 12
Wärmelehre (2h) 13
Seminar 7
12.4.1912–31.7.1912 Mechanik der Kontinua (2h) 10
Molekulartheorie der Wärme (3h) 11
Seminar 7
During the first year, the lectures were given in Klementinum in the centre of the
historical part of Prague, then they moved to Viničná 7.
Concerning the Prague papers of Albert Einstein, five were on thermodynamics,
radiation theory and quantum theory, among them also just brief notes. An exception
was the review on the problem of specific heats which Einstein presented at the
first Solvay Congress in November 1911 where he and Friedrich Hasenöhrl from
Vienna were the only representatives of the Austro-Hungarian Empire. Since we are
concerned primarily with gravity, we shall not analyze those papers, but we give now
the complete list of the works of Albert Einstein in Prague on the theory of relativity
and gravitation.8 The first three papers appeared in 1911, and the following four in
1912. The titles are here given in English, the exact German title and bibliographical
details are given in the References.
1. The Theory of Relativity [12].
2. On the Ehrenfest Paradox. Comment on Varičak’s Paper [13].
3. On the Influence of Gravitation on the Propagation of Light [14].
4. The Speed of Light and the Statics of the Gravitational Field [15].
5. On the Theory of the Static Gravitational Field and Note Added in Proof [16].
6. Is There a Gravitational Effect Which Is Analogous to Electrodynamic Induc-
tion? [17].
7. Relativity and Gravitation. Reply to a Comment by M. Abraham [18].
The first paper, assigning Prague as Einstein’s address already, is based on a
lecture given in Zurich in January 16, 1911, before Einstein’s arrival in Prague. The
lecture, in which Einstein used the term “Relativity Theory” in a title for the first
time, was followed by the discussion published in April 1912 (see “The collected
papers of Albert Einstein”, Vol. 3). The second paper is a short note on a paradox
involving moving measuring rigid rods in special relativity. Our main attention will
be paid to five papers on the development of a new theory of gravity. Before we
turn to them in detail, we shall first present a document, valuable in connection
with Einstein’s stay in Prague, though largely unknown. As a document of historical
importance it was first published and commented upon in 1979 in [5], then quoted in
the biography of Pais [4], and later also elsewhere. The document concerns the 1923
Czech translation of Einstein’s little book About the Special and General Theory of
Relativity in Plain Terms.9 Einstein wrote a special foreword to the Czech edition;
this appeared in both the German original and the Czech translation. In the foreword
he recalls what he did during his Prague stay: “I am pleased that this small book, in
which the main ideas of the theory of relativity are explained without mathematical
elaboration, should now appear in the native language of the country in which
I found the necessary concentration for developing the basic idea of the general
theory of relativity which I had already conceived in 1908. In the quiet rooms of the
Institute of Theoretical Physics of Prague’s German University in Viničná Street, I
discovered that the principle of equivalence implies the deflection of light rays near
the Sun by an observable amount, without at that time knowing that a similar result
8 Einstein’s long review on the Special Theory of Relativity for the Handbuch der Radiologie was
started in Prague at the beginning of 1912, and continued after Einstein’s move to Zurich (the quality
of ink and paper improved after the move). The First World War interrupted the publication. Einstein
was later unwilling to add material on general relativity or even revise the existing manuscript.
Still, it is an extraordinarily precious document since it is the earliest and most significant of the
surviving scientific manuscripts written by Einstein before World War I. A fine facsimile was
published in 1996 by George Braziller, Inc., in association with J. Safra Foundation and the Israel
Museum, Jerusalem. For more details on the manuscript, see the Collected papers of Albert Einstein,
Volume 4.
9 The book was translated by an excellent physics teacher at a highly regarded Czech Gymnasium
in Prague, V. Štíbr. Incidentally, after World War II Štíbr was the physics teacher of Karel Kuchař.
42 J. Bičák
Fig. 4 The cover of the Czech edition of Einstein’s popular book “About the Special and General
Theory of Relativity in Plain Terms” from 1923 and the foreword in original German which Einstein
wrote for the Czech edition
had been derived from Newton’s mechanics and his corpuscular theory of light. In
Prague I also discovered the shift of spectral lines towards the red which is not
yet completely confirmed. However, the decisive idea of the analogy between the
mathematical formulation of the theory and the Gaussian theory of surfaces came
to me only in 1912 after my return to Zürich, without being aware at that time of
the work of Riemann, Ricci, and Levi-Civita. This was first brought to my attention
by my friend Grossmann when I posed to him the problem of looking for generally
covariant tensors whose components depend only on derivatives of the coefficients
of the quadratic fundamental invariant. It now appears that it is already possible
to evaluate the achievements and limitations of the whole theory. It gives a deep
knowledge of the physical nature of space, time, matter and gravity; however, it does
not provide sufficient means for solving the problems of quanta and of the atomic
constitution of the elementary electric units of which matter is composed.”
In Fig. 4, the cover and the German foreword are displayed.
During talks about Einstein’s Prague period a question often arises whether the
idea of using Riemannian geometry in building a new theory of gravity emerged
during the Prague stay already, or only after his return to Zurich when he started to
collaborate with his friend Marcel Grossmann. In the foreword above, Einstein men-
tions that the work of Riemann, Ricci, and Levi-Civita was brought to his attention
Fig. 5 All physical phenomena will proceed in the same way in the systems K and K ≈ —principle
of equivalence
first by Grossmann. Nevertheless, Einstein did not care much about giving precise
historical statements; for example, he writes that in Prague he discovered the shift
of spectral lines towards red although this effect is contained in his paper from 1907
already in which he tackled the problem of gravity for the first time and then left
it until the beginning of the Prague period. That the idea of using the Riemannian
geometry was conceived in Prague already as a result of discussions with his friend
mathematician Georg Pick is stated in the biography by Frank [3], as well as in the
“Personal Reminiscence” by Reinhold Fürth who studied at the German University
of Prague during 1912–1916 and became then an important member of the Faculty.10
6 The Principle of Equivalence
At the beginning of his first paper on gravity [14] the equivalence of the systems K
and K ≈ indicated in Fig. 5 is discussed. To paraphrase Einstein’s words in English
10 Fürth says that “Pick, professor of mathematics at Prague and a fellow violin player, drew
his [Einstein’s] attention to the Italian, Levi-Civita, and his absolute differential calculus. . . (see
Einstein—the first hundred years, ed. by M. Goldsmith et al. Pergamon Press 1980). Fürth became
a member of Max Born’s group in Edinburgh in 1939 but still in October 1938, after the Munich
agreement, he was in Prague and published the text “Der Streit um die Deutung der Relativitaet-
stheorie” (The struggle about the meaning of the relativity theory) in which he defends relativity
against philosophers like Prof. Kraus in Prague who wanted to find inconsistencies of the theory by
“metaphysical” arguments.
44 J. Bičák
Fig. 6 The principle of equivalence implies the gravitational redshift
he clearly states that the equivalence of both systems is certain as far as we restrict
ourselves to purely mechanical phenomena, but it will acquire a deeper meaning if
we extend the equivalence to all laws of nature. Then we have a principle which has
a great heuristic meaning. In his biography A. Einstein: Creator and Rebel Banesh
Hoffmann, Einstein’s direct collaborator, writes: “In the paper of 1907. . . Einstein
had already begun his attack on the problem of acceleration, and he returned to it in
his Prague paper of 1911. His arguments, particularly in its 1911 form, must rank as
one of the most remarkable in the history of science.” The principle of equivalence
is then applied to derive the gravitational redshift—for the first time in a beautifully
pedagogical way (see Fig. 6). As Mark Twain writes: “The nice thing about Science
is that one gets such wholesale returns of conjecture from such a trifling investment
of fact”. . .
What is the present-day formulation of the (weak) equivalence principle? Employ-
ing Cliff Will’s formulation from his Living Reviews article [19]:
• Test bodies fall with the same acceleration independently of their structure or
composition.
• The outcome of any local non-gravitational experiment is independent of: (a) the
velocity of the local inertial frame in which it is performed, (b) where and when
in the universe it is performed.
From the time of Newton and Eötvös it has been a continuing effort to measure a
possible violation of the first item. After the 1950s, it is connected with the names of
Dicke in Princeton, Braginskij in Moscow and, most recently, with the group at the
University of Washington which used a torsion balance tray to study the accelerations
Fig. 7 Dicke used the field of the Sun to put the limits on difference in acceleration of various
materials. (The suspension of the torsion balance is perpendicular to the picture)
of various materials toward local masses, towards the Sun (as Dicke did—see Fig. 7)
and even towards the Galaxy.
The present best limits on the fractional difference in acceleration of various
materials/bodies A and B are

aA − aB (0.3 ± 1.8) × 10−13 , Eöt-Wash
ε= 1 (1)
(−1.0 ± 1.4) × 10−13 , LLR
2 (a A + a B )
where the first result is given by experiments at the University of Washington, whereas
the second comes from Lunar laser ranging. See [19] for more details.
An endeavor to improve the limit on ε will undoubtedly continue since some theo-
ries, inspired primarily by string theory, predict the violation of the weak equivalence
principle due to the presence of dilatons. That is why a Satellite Test of Equivalence
Principle (“STEP”) has been conceived to improve the limit on ε by about five orders
of magnitude to ε = 10−17 − 10−18 . It is a drag-free satellite consisting of an outer
shell around an inner test mass [20].
6.1 Gravitational Redshift Today
Concerning the gravitational redshift, the first reliable experiment was the Pound-
Rebka-Snider experiment in 1960 using the Mössbauer effect to measure precisely
the frequency shift of γ -rays in a 22.6 m tower at Harvard. These experiments, yield-
ing an accuracy of the order of 10 %, were followed by clocks placed in an aircraft
and a rocket; the clock rates were compared with the same clocks on the ground. An
accuracy of 7 × 10−5 in determining ν was obtained by using a hydrogen maser
clock in the rocket by Vessot et al. in 1980. Only 30 years later this accuracy was
claimed to have been improved upon by laboratory experiments based on quantum
46 J. Bičák
interference of atoms [21] which yielded an accuracy of 7 × 10−9 . However, the

interpretation was criticized in the work [22] which considers the experiment as a
test of the universality of free fall but not the redshift effect. An intriguing idea for
producing and measuring a “force-free gravitational redshift” (a gravitational ana-
logue of the Aharonov-Bohm effect) imagines two atomic matter waves which serve
as clocks at different gravitational potentials and the redshift can be caused just by
a potential difference although a force vanishes. This suggestion was published 100
years after Einstein was going to leave Prague, just two weeks before our conference
in [23]. Therein the authors address also the criticism raised in [22]. Anyway, the
best known “common” proof of gravitational redshift we currently have available is,
of course, the daily successful operation of the Global Positioning System (GPS).
What is the present situation concerning the gravitational shift of the spectral lines
from the surface of the Sun relative to the corresponding laboratory lines which, in his
Prague paper [12], Einstein predicted to be λ = 2.1×10−6 λ, which in the velocity
scale is approximately 600 ms−1 ? The measurements are complicated because there
are various sources of wavelength shifts, such as radial currents different at different
levels, convection, inner asymmetries of spectral lines etc. The first convincing mea-
surements appeared only in 1991 by using the infrared oxygen triplet [24]. Converted
to velocities, the three chromospheric oxygen lines yielded 627 ± 10 ms−1 which is
0.99 ± 0.02 of the value predicted by the principle of equivalence. Precise measure-
ments of the solar redshift are planned [25] which should reach an accuracy of 10−6
and so test the second order relativistic effects. For the most recent pioneering work
on the gravitational redshift of galaxies in clusters, see [26].
7 Bending of Light
After the formulation of the principle of equivalence and its application in deriving
the formula for redshift, Einstein turned in the same paper [14] to the problem of
light propagation in a gravitational field. The equivalence of the systems K ≈ and K
in Fig. 6 implies the frequency shift, i.e., different rates of clocks located at different
gravitational potentials Φ in the case of a homogeneous gravitational field. However,
typically for his grasp of the laws of nature, Einstein assumes that the same effect takes
place in a general static field.11 As a consequence, the velocity of light, measured by
clocks influenced by a gravitational potential, depends on the value of the potential.
Denoting the velocity of light at origin of coordinates by c0 , then at a place in which
gravitational potential with respect to the origin is Φ, the velocity of light is given
by the relation c = c0 (1 + Φ/c2 ). Employing a simple picture based on Huygens
11 Einstein notices the effect of gravity on the light propagation in his paper from 1907 already in
which, for the first time, the gravitational field and equivalence principle are mentioned. However,
only a homogeneous field is analyzed and the effect considered in the field of the Earth is found to
be unobservable.
principle, Einstein then derives the formula for an angle α by which a light ray will
be deflected in the direction n ≈ along any trajectory s:

1 ∂Φ
α=− 2 ds. (2)
c ∂n ≈
According to this formula a ray propagating around a celestial body with mass M
deflects towards the body by the value given in Fig. 8—the picture of the final page of
Einstein’s best known paper from the Prague period. Under the formula the resulting
value of 0.83 arcseconds for a light ray passing close to the Sun is given, half of the
value following from the final form of general relativity. Einstein promptly contacted
various observatories in the world and tried to encourage observation of the effect
during a total eclipse.
The only response came from Erwin Freundlich (1885–1965), an assistant in the
Royal Observatory in Berlin.12 Freundlich took part in a solar eclipse expedition in
1914 in Crimea but World War I broke and he was interned in Russia. It was only
just after the war, in 1919, when the effect was measured for the first time by the
British expeditions led by Eddington and Dyson. There is much literature on these
observations, some of which attributed the confirmation of the relativistic value to
Eddington’s high regard of general relativity and also his wish, of a pacifist, to show
how despite the war English astronomers confirm a theory by a German physicist. It is
true, difficulties during the expeditions appeared, but in a recent article, D. Kennefick
carefully re-analyzed the case and showed that Eddington et al. had good reasons for
the claim that general relativity was confirmed, and Newtonian theory was not (see
Fig. 9).
The light deflection is one of the finest examples of the continuing success of
general relativity as the best gravity theory available. When one allows a more gen-
eral form of a static spherically symmetric vacuum metric as it can follow from an
alternative theory of gravity by inserting parameters β and γ into the Schwarzschild
metric (see Fig. 10), one finds that it is γ which enters the formula for the deflection
of waves.
If general relativity is correct, γ = 1. Let us have a look at Fig. 11 from Cliff Will’s
Living Reviews in Relativity article [19] on “The Confrontation between General
Relativity and Experiment”.13 A 2004 analysis of ⊗ 2 million VLBI observations
of 541 radio sources at 87 VLBI sites imply γ − 1 = (−1.7 ± 4.5) × 10−4 !
12 For a detailed account on Freundlich and tests of relativity theory, see [27]. In fact, the first attempt
to measure light deflection was already made during an eclipse in October 1912 by C. D. Perrine from
Cordoba, Argentina, who was inspired by Freundlich. However, rain frustrated all efforts. (I thank
Jorge Pullin for pointing this out in the discussion after my lecture in the conference.) Freundlich
later wrote one of the first books on Einstein’s gravitation theory (Springer 1916, Cambridge 1920).
From the “Prague perspective” it is interesting to notice that in 1937 he was appointed professor
of astronomy at the Charles University in Prague, however, was forced to leave because of Hitler’s
policies towards Czechoslovakia in January 1939.
13 Will’s reviews on the verification of Einstein’s theory are well known. Perhaps less recognized
is his “Resource Letter” on the tests of gravity [28] which gives many references on both current
literature as well on some of historical papers.
48 J. Bičák
Fig. 8 The last page of Einstein’s most famous Prague paper on deflection of light
For example, scalar-tensor theories must have parameter ω > 40000 (which their
proponents like R. Dicke considered to be around 6), to be compatible with the light
deflection observations.
Fig. 9 Portrait of Erwin Freundlich by Max Pechstein from 1918. The heading of a recent article
which showed undoubtedly that Eddington’s and Dyson’s expeditions of 1919 to measure light
bending, despite difficulties they encountered, could state safely that their observations ruled out
Newton theory and confirmed general relativity
A fundamental consequence of the influence of gravity on the light propagation

is the existence of black holes formed by gravitational collapse. I dare to state this
commonly known fact only because I recall how in Erice in 1974 Roger Penrose
drew a picture similar to Fig. 12. He only “forgot” to write “Prague 1911” beneath
the left picture.
8 Gravitational Lensing
In contemporary astronomy, nowhere is light deflection so extensively and profitably

used as in the phenomenon of gravitational lensing. Going now from the present to the
past we may look at just two well-known examples: the images of two galaxy clusters
after collision (Fig. 13) indicating the existence of dark matter in the clusters; it moves
further after the collision (its distribution, shown by green counters, being known
thanks to gravitational weak-lensing) than ionized gas (yellow-red in the right part
observed in X-rays); and a Horseshoe Einstein Ring from Hubble telescope (Fig. 14).
Here I wish to make just a few remarks on the history of discovering the effect,
in which Prague played a significant role, unrecognized generally yet. It was long
thought that the effect was first described by Einstein in a paper published in Science
on December 4, 1936, after his interaction with a Czech amateur scientist Rudi
Mandl. Later Einstein’s Scratch Notebook from 1912 (when he was still in Prague)
was found and it became evident that the effect was found by him then already. The
complete history of this finding and its relation to the 1936 paper is comprehensively
described in several publications by Renn and Sauer; see, for example, [30].
It is well-known that the effect was also discussed by Eddington in 1920 and
by Chwolson in 1924. However, the role of a Czech astronomer, an expert in the
eclipse phenomena of the Moon and planets, František Link (1906–1984), remains
50 J. Bičák
Fig. 10 The generalization of the Schwarzschild metric to include alternative theories of gravity
apparently unknown outside his country.14 In Fig. 15 it is seen that Link published
his first paper on using Einstein’s bending in photometry in Comptes Rendus in
March 1936, more than eight months before Einstein’s
√ paper appeared in Science.
In this paper already, his expression for ρ◦ = K kα1 corresponds to the angular
size of the Einstein ring, usually denoted as Θ E . A much more detailed work on the
“Photometric consequences of the Einstein deflection” by Link appeared in Bulletin
Astronomique in 1937. In that time Link did not notice Einstein’s paper in Science
14 On the occasion of the centenary of Link’s birth there was a seminar organized by astronomers
and historians of science at Charles University on November 29, 2006, where gravitational lensing
was extensively discussed and the role of Link, the first director of the Astronomical Institute of
the Academy, recalled. Incidentally, after my talk was prepared, including the part on lensing and
Link, a preprint appeared in the arXiv:1206.1165v1 [physics.hist-ph] by D. Valls-Gabaud in which
the pioneering role of Link in the origins of gravitational lensing is emphasized.
Fig. 11 Limits on parameter γ from measurements of the light deflection. γ = 1 in general

relativity. Taken from [19]
but he gave a very detailed account of the history of lensing and many mathematical
details later in his monograph [31].
9 Prague Works on Gravitation from 1912
These are four papers: two on the static gravitational field, [15, 16], including non-
linear field equations and the motion of test particles in a given field, the first work
on a new effect–on linear gravitational dragging [17], and last but not least, the reply
to a comment by Abraham [18] in which Einstein outlines the main features a future
theory of gravity should possess.
In the first paper, an inertial frame I0 is first considered and a uniformly
accelerated frame K with an acceleration a with respect to I0 (Fig. 16). Denot-
ing coordinates in these frames by I0 (τ, ξ, ε, ζ ) and K (t, x, y, z), Einstein
assumed the following approximate
form of the transformationbetween
the frames:
ξ = λ(x) + α(x)t 2 + O t 3 , τ = β(x) + γ (x)t + δ(x)t 2 + O t 3 , ε = y, ζ = z,
where functions λ(x), α(x), . . . are to be determined as follows. First assume that at
t = 0 the origins coincide, ξ = 0, x = 0. Consider the line elements in I0 and K
to have the form
52 J. Bičák
Fig. 12 In the strong-field case matter can act on light so that a black hole is formed. This is
indicated in the spacetime diagram of gravitational collapse on the right
Fig. 13 “The bullet cluster”—the collision of two galaxy clusters (1E0657–558) provides the best
current evidence for the nature of dark matter (taken from [29]). See the text for details
2
dsI = −c∗2 dτ 2 + dξ 2 + dε2 + dζ 2 , c∗ = 1
2
dsK = −c2 (x)dt 2 + dx 2 + dy 2 + dz 2 .
Requiring then that dsI2 = 0 ⇔ ds 2 = 0, one finds functions λ(x), α(x), . . .

0 K
in the transformation and arrives at the following final form of the transformation
between the inertial and uniformly accelerated frames:
ξ=x+ 1
2 act 2 , τ = ct, c = c0 + ax. (3)
Fig. 14 A Horseshoe Einstein Ring (Credit: ESA / Hubble & NASA)
Fig. 15 The first publication on gravitational lensing: Link F., Comptes Rendus 202 (16 Mar 1936),
917–919. Notice that the paper was communicated by C. Fabry
9.1 Intermezzo
Today we know that the transformation to the rigid uniformly accelerated (“Rindler”)
frame reads
ξ = 1
a (cosh at − 1) + x cosh at, (4)
τ = 1
a sinh at + x sinh at. (5)

For small t (neglecting O t 3 ) Einstein’s “Prague transformation” above immedi-
ately follows from these two relations. The spacetime orbits of uniformly accelerated
particles are hyperbolas; hyperbolas (timelike at z 2 > t 2 , spacelike at z 2 < t 2 ) are
54 J. Bičák
Fig. 16 The system K , considered by Einstein in [15], is “uniformly accelerated” (in a relativis-
tic sense) with respect to inertial frame I0 ; and a schematic spacetime diagram of the C-metric
representing two black holes uniformly accelerated in opposite directions
orbits of the boost Killing vector. Spacetimes with axial/rotational symmetry and
boost symmetry, so called boost-rotation symmetric spacetimes, play an important
role in general relativity: they are radiative, with a non-vanishing news function,
have a plausible Newtonian limit, admit global null infinity; they have been used as
test-beds in numerical relativity. One of the best known examples is the C-metric, rep-
resenting two uniformly accelerated black holes in opposite directions—a schematic
diagram is on the right in Fig. 16. In the Rindler-type coordinates (which can be
introduced in the quadrants in which sources occur and the boost Killing vector is
timelike) the spacetime is static but in the other two quadrants it is dynamical (locally
of the Einstein-Rosen type).15
Returning back to 1912, we notice that the velocity of light in a uniformly acceler-
ated frame, Eq. 3, satisfies Laplace’s equation. Now invoking the equivalence princi-
ple, Einstein postulates that the field equation even for a general static gravitational
field will read
c = 0 (in vacuum), c = kcρ (in matter).
However, he soonrealized that one gets contradictions with conservation of energy

and momentum ( f dV ≤= 0, with f = −ρ grad c). Hence, in a shortly published
following paper [16] he avoided the inconsistency by modifying the field equation
into
15I dared to make this intermezzo since our group in Prague has been devoting quite an effort to
understand these spacetimes (see, e.g., contribution by Bičák and Kofroň in the Proceedings [11]
and the literature quoted therein).

1 grad2 c
c = k cρ + .
2k c
The field equation becomes nonlinear; and the additional term on the right-hand side
describes the energy density of the gravitational field which thus itself contributes
as a source to the field. (An interesting discussion of this theory of gravity from
a purely Newtonian point of view was presented in this conference by Domenico
Giulini—see [11].) In the same paper, a local (“pocket”) temperature is introduced
and shown to depend on c, i.e., on a local gravitational potential; and in the Appendix
the equations of motion for a test particle in a given gravitational field are derived
from the variational/Hamilton principle–the foreshadow of a variational principle
for geodesics. Still more remarkably, Einstein introduces, for the first time, a “local
view” on the equivalence principle since the nonlinear field equation above is not
satisfied by c = c0 + ax. The local transformation between the two frames in a small
neighborhood is generalized to
1 dc 2
ξ=x+ c t , c(x) arbitrary.
2 dx
9.2 Dragging of Inertial Frames
The following paper from the Prague period is not directly connected to a construction
of a new theory of gravity. However, it fits well to the Prague physics since it is
related to Ernst Mach–“the fact is that Mach exercised a great influence upon our
generation through his historical-critical writings. . . ”, wrote Einstein in 1916. By
using equations of motion derived in the preceding paper, Einstein considers a shell
of matter and its influence on a mass point placed in its center; as the shell starts
to accelerate, a force on the point mass appears and its inertial mass increases (see
Fig. 17).
In the “Machian spirit” Einstein puts forward an idea that perhaps the inertia of
a point mass is fully determined by the action of all other masses. We know today
that this does not arise in general relativity, however, it is so that inertial frames
are, under suitable conditions, determined by averages over matter distributions.
However, Einstein’s considerations from Prague show, for the first time, that an
inertial frame will be dragged by moving masses. The effect was soon exhibited also
for rotating masses by Einstein and Thirring and has been studied extensively until
the present, including recently the dragging effects due to gravitational waves (see
contributions by Pfister, by Lynden-Bell and Katz, and by Bičák, Katz, Ledvinka and
Lynden-Bell in the Proceedings [11] of this conference).
56 J. Bičák
Fig. 17 From his “Prague theory of gravity” Einstein discovered a (linear) dragging effect
The planning and eventually launching of the Stanford gyroscopic experiment,

later called Gravity Probe B, started in 1959 and ended in 2011.16 The spacecraft
with gyros were orbiting the Earth during a mission that ran from April 2004 to
September 2005; however, many simulations had to be performed after the measure-
ments finished. There were eventually four superconducting gyros in the satellite
and both the geodetic effect (cf. the “left” gyroscope in Fig. 18) and (much smaller)
dragging effect due to the rotation of the Earth were measured. In an article pub-
lished in Physical Review Letters from May 31, 2011, “the authors reported analysis
16 It started thus contemporaneously with my studies at the Faculty of Mathematics and Physics in
Prague and closed close to my retirement; but I enjoyed this long time span because the experiment
is interesting and well explainable to a general public.
Fig. 18 Stanford gyro experiment–“Gravity Probe B”–measured both a geodetic precession due
to curvature around Earth (gyro 1) and a dragging effect due to the rotation of Earth (gyro 2)
of the data from all four gyroscopes results in a geodetic drift rate of −6601.8 ± 18.3
milliarcsecond/year (≡ mas/yr; 1 mas = 4.84 × 10−9 rad) and a frame-dragging drift
rate of −37.2 ± 7.2 mas/yr, to be compared with the GR predictions of −6606.1
mas/yr and −39.2 mas/yr, respectively”. An extensive literature on science, technol-
ogy, sociology and politics associated with the experiment is available and interesting
to read (see, e.g., http://physics.aps.org/articles/v4/43). Principally, however, despite
the doubts about the initial results (burdened by some unexpected phenomena like,
for example, random patches of electrostatic potentials on both gyros and their hous-
ing) it is now firmly established that dragging effects, or as many relativists like to
say—gravitomagnetic effects, were measured and so general relativity is confirmed
by using stationary rather than just static gravitational fields. Another confirmation
of the dragging (Lense-Thirring) effect came from the laser tracking of two Earth-
orbiting LAGEOS satellites, analyzed in a number of papers by I. Ciufolini and his
collaborators.
58 J. Bičák
9.3 Relativität and Gravitation. Erwiderung. . .
In his last paper from the Prague period “Relativity and Gravitation. Reply to a
Comment by M. Abraham” received by Annalen der Physik on July 4, 1912, just
a few days before he left Prague, Einstein was inspired by Abraham’s criticism to
summarize the views on the contemporary state of the theory of relativity and its
description of gravity. From Einstein’s summary it is evident that he learned many
fundamental issues during his Prague stay. This was also the reason we decided to
give the name to the conference following this work. The main points which Einstein
expressed in his summary are as follows:
• Local significance of equivalence principle.

• Equations of motion for point masses (variational principle).
• Equations of the electromagnetic field when gravity is present.
• Nonlinear field equation for gravity (energy density of gravitational field itself as
a source).
• All equations must be form invariant with respect to a larger group than the Lorentz
group.
• Spacetime coordinates lose their simple physical meaning.
Let us quote the last point in more detail, in Einstein’s own words: “Man
sieht schon aus dem bisher behandelten, daß die Raum-Zeit-Koordinaten ihre ein-
fache physikalische Deutung einbüßen werden, und es ist noch nicht abzusehen,
welche Form die allgemeinen raumzeitlichen Transformationsgleichungen haben
könnten. . .”
In Prague it was thus for the first time that Einstein realized that spacetime coor-
dinates need not determine the distances between spacetime points. Later, in his
Autobiographical Notes,17 he comments “Why were another seven years required
for the construction of the general theory of relativity? The main reason lies in the
fact that it is not so easy to free oneself from the idea that coordinates must have
an immediate metrical meaning”. At the end of his Prague stay, Einstein was on
the right track towards general relativity. Important features of the new theory were
understood, but gravitation was described wholly by one function only, the variable
velocity of light.
As we discussed above, Einstein might have learned about Riemannian geometry
in Prague already from Georg Pick, however, this does not seem to be of great
importance since, in any case, a few months after he left Prague for Zurich, he
started to collaborate with Marcel Grossmann and, with some exaggeration perhaps,
one may say that they had general relativity almost in hand:
17 In Albert Einstein-Philospher and Scientist, The Library of the Living Philosphers Vol. VII, ed.
P. A. Schild, Open Court Publ. Co. 1949.
• Gravitation described by ten components of the metric tensor gμν .

• Correct variational principle for the geodesic equation.
• Maxwell equations in general gravitational field covariant under general coordinate
transformations.
• The source of gravity T μν covariantly conserved, ∇ν T μν = 0.
• The field equations μν = κ Tμν where μν is formed from the metric tensor and
its derivatives.
Although in 1913 they first considered the Ricci tensor at the left-hand side of
the field equations (on the right-hand side of which was a symmetrical energy-
momentum tensor of matter T μν ), they wrongly concluded that they do not get
a correct Newtonian limit, i.e., in vacuum the Laplace equation for the Newtonian
potential.18 And Einstein went astray from his route to a correct theory by assuming
that the field equations, in order to give a meaningful Newtonian limit, cannot be
generally covariant. . . The world had to wait another three years to see the equations
Rμν − 21 Rgμν = κ Tμν . About their Genesis, see the monumental survey quoted
in [32].
9.4 Coda
On Saturday, May 15, 2004, the Prague daily newspaper “Lidové Noviny” published
an article “Einstein through the eyes of Johanna from Czechia”; see Fig. 19. It was
just in those days when Princeton University published the diary of Johanna Fanta,
the daughter-in-law of Berta Fanta whose salon Einstein liked to visit during his
stay in Prague in 1911–1912. Einstein and Johanna knew each other from Einstein’s
Berlin period already, but became quite close friends, spending often time on the Lake
Carnegie in a boat, when Einstein was in Princeton and Johanna was employed in
the library of the University. During the last two years Johanna made regular notes of
what Einstein said or wrote. This is most interesting, often touching and sad reading.
Let us quote the notes from just three days. April 13, 1954: “Expresses annoyance at
Oppenheimer for letting the McCarthy and Atomic Energy Commission affairs bother
him so much. Already told the press that he has great respect for Oppenheimer, both
as a human being and as a scientist.” October 24, 1954: “He calculated like crazy
again today but accomplished nothing.” April 10, 1955, 8 days before Einstein’s
death: “He tried all day to compose a radio message on behalf of Israel and did not
succeed in finishing it. He claims he is totally stupid—that he has always thought so,
and that only once in a while was he able to accomplish something.”
In a rather poor, hilly part of south-western Moravia, close to the place where Gus-
tav Mahler was born, there lived various “country-type” sages interested in literature,
art, music, religion, and also science. In the left side of the last Fig. 20 there is the
18For a detailed analysis of the “Zurich notebook” which Einstein began to write soon after leaving
Prague, see [32].
60 J. Bičák
Fig. 19 An article about Einstein and his friend Johanna Fanta in Lidové noviny, May 2004
Fig. 20 Covers of two books: Einstein’s course published in Moravian village Stará Říše in
September 1926, containing the translation of parts of Eddington’s popular book Space, Time and
Gravitation and the correspondence of Otokar Březina, poet and essayist, who created almost all
of his works in a nearby village Nová Říše
cover of the “Einsteinian course” published in September 1926 by Josef Florian in

the village Stará Říše. It contains a very fine translation of Eddington’s famous book
with conversation between an experimental physicist, a pure mathematician. and a
relativist who advocates the newer conceptions of time and space in physics. On the
cover one can see symbols of Noah, Moses, Jesus, etc but, at the left low corner, there
is also the correct expression for the variational principle for geodesics! On the right
side there is the cover of the second volume of the correspondence of Otokar Březina,
the top poet of Czech symbolism who lived and taught in a neighboring village Nová
Říše (cf. Wikipedia). In his correspondence he mentions Planck’s quantum theory,
Bohr, Minkowski and Einstein in the 1920s. In a letter to the professor of classical
literature, František Novotný in Brno, who sent him his monumental work about
Plato, Březina wrote encouraging words which I used both at the end of my talk in
June and in [5] on the occasion of our Einstein centenary celebrations in 1979 in
Karolinum as the present conference took place:
“There will always be minds who, by the united power of knowledge and dreams,
science and poetry, will strive for a unified picture of all phenomena in the Universe,
an image that in equal measure corresponds both to the eternal longing of the human
spirit for harmony and beauty and to the thirst of the heart for justice.” Is not then
the example of Albert Einstein a reason for an enduring optimism?
Acknowledgments I am grateful to David Kofroň for the help with the manuscript, especially
with the figures, and for his suggestions improving the text. I also thank the Albert Einstein
Institute, Golm, where most of this text was written after the conference, for the kind hospital-
ity, Lars Andersson for comments and Ian Hinder for correcting some of my Czechisms. The
partial support from the Czech Science Foundation No 14-37086G (A. Einstein Center) is also
acknowledged.
References
1. Kleinert, A.: Anton Lampa und Albert Einstein-Die Neubesetzung der physikalischen
Lehrstühle an der deutschen Univesrität Prag 1909 und 1910. Gesnerus 32, 285 (1975)
2. Havas, P.: Einstein, relativity and gravitation research in Vienna before 1938. In: Goenner, H.,
Renn, J., Ritter, J., Sauer, T. (eds.) The Expanding Worlds of General Relativity, pp. 161–206.
Birkhauser, Boston-Basel-Berlin (1999)
3. Frank, P.: Einstein, his Life and Times. Alfred A. Knopf, New York (1947)
4. Pais, A.: ‘Subtle is the Lord..’ The Science and the Life of Albert Einstein. Clarendon Press,
Oxford (1982)
5. Bičák, J.: Einstein a Praha (in Czech and German). Prometeus, Prague (1979)
6. Bičák, J.: Einstein’s route to the general theory of relativity (in Czech). Československý Časopis
pro Fyziku A29, 222 (1979)
7. Bičák, J.: Einstein’s Prague articles on gravitation. In: Blair, D.G., Buckingham, M.J. (eds.)
Proceedings of the 5th M. Grossmann Meeting on General Relativity, pp. 1325–1334. World
Scientific, Singapore (1989)
62 J. Bičák
8. Bičák, J.: Einstein’s Prague ideas on gravitation: the history and the present. In: Janta, J.,
Pantoflíček, J. (eds.) Trends in Physics, pp. 65–75. Union of Czechoslovak Mathematics and
Physics, Prometheus, Prague (1985)
9. Těšínská, E.: Theoretical physics at the Prague University and its connections with Prague
stay of A. Einstein 100 years ago (in Czech), Pokroky Matematiky, Fyziky a Astronomie 57,
146–168 (2012)
10. Semerák, O., Podolský, J., Žofka, M. (eds.): Gravitation: Following the Prague Inspiration.
World Scientific, New Jersey (2002)
11. Bičák, J., Ledvinka, T. (eds.): Relativity and Gravitation: 100 Years After Einstein in Prague.
Springer, Switzerland (2014)
12. Einstein, A.: Die Relativitäts-Theorie. Vierteljahrsschrift der Naturforschenden Gesellschaft
in Zürich 56, 1 (1911)
13. Einstein, A.: Zum Ehrenfestschen Paradoxon. Bemerkung zu V. Varičaks Aufsatz. Phys. Z. 12,
509 (1911)
14. Einstein, A.: Über den Einfluss der Schwerkraft auf die Ausbreitung des Lichtes. Ann. d. Phys.
35, 898 (1911)
15. Einstein, A.: Lichtgeschwindigkeit und Statik des Gravitationsfeldes. Ann. d. Phys. 38, 355
(1912)
16. Einstein, A.: Zur theorie des statischen Gravitationsfeldes und Nachtrag zur Korrektur. Ann.
d. Phys. 38, 443 (1912)
17. Einstein, A.: Gibt es eine Gravitationswirkung, die der elektrodynamischen Induktionswirkung
analog ist? Vierteljahrsschrift für gerichtliche Medizin und öffentliches Sanitätswesen 44, 37
(1912)
18. Einstein, A.: Relativität und Gravitation: Erwiderung auf eine Bemerkung von M. Abraham.
Ann. d. Phys. 38, 1059 (1912)
19. Will, C.M.: The confrontation between general relativity and experiment. Living Rev. Relativ.
9(3), (2006). http://www.livingreviews.org/lrr-2006-3
20. Sumner, T.J., et al.: Satellite test of equivalence principle. Adv. Space Res. 39, 254 (2007).
doi:10.1016/j.asr.2006.09.019
21. Müller, H., Peters, A., Chu, S.: A precision measurement of the gravitational redshift by the
interference of matter waves. Nature 463, 926 (2010). doi:10.1038/nature08776
22. Wolf, P., et al.: Does an atom intorferometer test the gravitational redshift at the Compton
frequency? Class. Quantum Grav. 28, 145017 (2011). doi:10.1088/0264-9381/28/14/145017
23. Hohensee, M.A., Estey, B., Hamilton, P., Zeilinger, A., Müller, H.: Force-free gravitational
redshift: proposed gravitational Aharonov-Bohm experiment. Phys. Rev. Lett. 108, 230404 (5
pages) (2012). doi:10.1103/PhysRevLett.108.230404
24. LoPresto, J.C., Schrader, C., Pierce, A.K.: Solar gravitational redshift from the infrared oxygen
triplet. Astrophys. J. 376, 757 (1991). doi:10.1086/170323
25. Cacciani, A., Briguglio, R., Massa, F., Rapex, P.: Precise measurement of the solar
gravitational redshift. Celest. Mech. and Dyn. Astrometry 95, 425 (2006). doi:10.1007/s10569-
006-9014-0
26. Wojtak, R., Hansen, S.H., Hjorth, J.: Gravitational redshift of galaxies in clusters as predicted
by general relativity. Nature 477, 567 (2011). doi:10.1038/nature10445
27. Hentschel, K.: Erwin Finlay Freundlich and testing Einstein’s theory of relativity. Arch. Hist.
Exact Sci. 47, 143 (1994)
28. Will, C.M.: Resource letter PTG-1: precision tests of gravity. Amer. J. Phys. 78, 1240 (2010).
doi:10.1119/1.3481700
29. Clowe, D., et al.: A direct empirical proof of the existence of dark matter. Astrophys. J. Lett.
648, L109 (2006)
30. Renn, J., Sauer, T.: Eclipses of the stars-Mandl, Einstein, and the early history of gravitational
lensing. In: Ashtekar, A., Cohen, R., Howard, D., et al. (eds.) Revisiting the Foundations of
Relativistic Physics—Festschrift in Honour of John Stachel, pp. 69–92. Kluwer Academic
Publishers, Dordrecht (2003)
31. Link, F.: Eclipse Phenomena in Astronomy. Springer, New York (1969)
32. Renn, J., et al.: Einstein’s Zurich notebook: gravitation in the twilight of classical physics. In:
Renn J. et al. (eds.) The Genesis of General Relativity, Boston Studies in the Philosophy of
Science, vol. 1–4. Springer, Dordrecht (2007)
Part II
Classical General Relativity
Observers, Observables and Measurements
in General Relativity
Donato Bini
Abstract To perform any physical measurement it is necessary to identify in a non

ambiguous way both the observer and the observable. A given observable can be
then the target of different observers: a suitable algorithm to compare among their
measurements should necessarily be developed, either formally or operationally. This
is the task of what we call “theory of measurement,” which we discuss here in the
framework of general relativity.
1 Introduction
The spacetime (or absolute) point of view constitutes a unified scenario for quantities
which, in the pre-relativistic physics, were associated with distinct notions: time and
space themselves, energy and momentum, mechanical power and force, electric and
magnetic fields and so on. In every day experience, however, our intuition is still
compatible with the perception of a three-dimensional space and a one-dimensional
time and therefore any physical measurement requires a local recovery of the pre-
relativistic type of separation between space and time. To this purpose we need
some prescription in order to perform the required splitting, and hence identifying a
“space” and a “time” relative to any given observer. Any such prescription requires
a congruence of timelike world lines with a future-pointing unit tangent vector field
u (i.e., the local time direction) which we interpret as the world lines of a family of
(test) observers with associated 4-velocity u.
The splitting of the tangent space at each point of the congruence into the local
time direction u and the local rest space spanned by vectors orthogonal to u (hereafter
L RSu ), allows one to decompose all spacetime tensors, including tensorial operators,
and tensor equations into their spatial and temporal components. One may ask then
D. Bini (B)
Istituto per le Applicazioni del Calcolo “M. Picone,” CNR, 00185 Rome, Italy
e-mail: binid@icra.it

68 D. Bini
if there exist natural or special observer families in a given spacetime. This is clearly
the case of a stationary spacetime where a special observer (timelike) congruence
is associated with the timelike Killing vector field. Also, any spacetime admitting a
spacelike foliation has naturally associated with it a timelike congruence, namely that
of the normal directions to the slicing itself. It is known that any spacetime admitting
separable geodesics (e.g., as a consequence of the existence of a Killing tensor of
rank 2) also admits a foliation. For example, Kerr spacetime with the metric written
in standard Boyer-Lindquist coordinates summarizes these three conditions: it has
the family of static observers (whose world lines are aligned with the coordinate time
lines, i.e., with unit tangent vector parallel to the timelike Killing vector εt ), the family
of locally nonrotating observers or ZAMOs (whose world lines are orthogonal to the
t = constant hypersurfaces) and finally, the families of Painlevé-Gullstrand observers
who follow geodesic timelike lines since the latter have a separable dependence from
the coordinates. A review of the essential splitting formalism follows below. For more
details one can refer for instance to Refs. [1–6] (and references therein).
2 Orthogonal Decompositions
Let g be the four-dimensional spacetime metric with signature +2 and components

gγβ (γ, β = 0, 1, 2, 3), ≈ its associated covariant derivative operator and ν the unit
volume 4-form which assures spacetime orientation. Let u be a future-pointing unit
timelike vector field which identifies an observer, u · u = −1. The local splitting of
the tangent space into orthogonal sub-spaces uniquely related to the given observer
u, is accomplished by a temporal projection operator T (u) (along u) and a spatial
projection operator P(u) (generating the L RSu ). These operators, in mixed form,
are defined as follows
T (u) = −u λ ⊗ u Φ P(u) = I + u λ ⊗ u Φ (1)
where I ◦ α γ β is the identity on the tangent spaces of the manifold and the sym-
bols λ and Φ identify the fully contravariant and covariant representation of tensors,
respectively. In terms of components the above relations write
T (u)γ β = −u γ u β , P(u)γ β = α γ β + u γ u β . (2)

Given a qp -tensor S, let us denote as [P(u)S] its fully spatial projection obtained
by acting with the operator P(u) on all of its indices,
∂ ...
[P(u)S]γ... γ α
β... = P(u) ∂ · · · P(u) β · · · S α... (3)
The splitting of S relative to a given observer is the set of tensors which arise from
the spatial and temporal projection of each of its indices as we are going to discuss.
Observers, Observables and Measurements in General Relativity 69
This observer-dependent set of tensors represent S and it is termed as its (geometrical)

measurement by the observer u.
1. Splitting of a vector
If S is a vector field then its splitting gives rise to a scalar field and a spatial vector
field
S √ {u · S, [P(u)S]}. (4)
In terms of components they read
Sγ √ {u ∂ S ∂ , P(u)γ ∂ S ∂ }. (5)
In fact, with respect to the observer u, the vector S admits then the following
representation
S γ = [T (u)S]γ + [P(u)S]γ = −(u ∂ S ∂ )u γ + P(u)γ ∂ S ∂ . (6)

2. Splitting of a 11 -tensor

If S is a mixed 11 -tensor field, then its splitting consists of a scalar field, a spatial

vector field, a spatial 1-form and a spatial 11 -tensor field, namely
S γ β √ {u α u ∂ S ∂ α , P(u)γ ∂ u α S ∂ α , P(u)α γ u ∂ S ∂ α , P(u)γ ∂ P(u)α β S ∂ α }.
In terms of these fields, the tensor S admits the following representation with
respect to the observer u
S γ β = [T (u)γ ∂ + P(u)γ ∂ ][T (u)α β + P(u)α β ]S ∂ α

= (u α u ∂ S ∂ α )u γ u β − u γ u ∂ P(u)α β S ∂ α
−u α u β P(u)γ ∂ S ∂ α + [P(u)S]γ β . (7)
p
The local spatial and temporal projections of a q -tensor is easily generalized.
For example, the metric tensor gγβ has the (trivial) representation
gγβ = P(u)γβ + T (u)γβ .
3. Splitting of p-forms
Given a p-form
1
S = S[γ1 ...γ p ] βγ1 ⊗ · · · ⊗ βγ p ◦ Sγ ...γ βγ1 ∗ · · · ∗ βγ p (8)
p! 1 p
70 D. Bini
we define the electric part of S relative to the observer u the quantity

S (E) (u) = −u ω Sω γ1 ...γ p−1 (9)
γ1 ...γ p−1
or in a more compact form S (E) (u) = −u S. Similarly we define as the magnetic

part of S the quantity

S (M) (u) = P(u)β1 γ1 . . . P(u)β p γ p Sβ1 ...β p (10)
γ1 ...γ p
or, in a compact form, S (M) (u) = P(u)S. From the above definitions we deduce
the following representation of S
S = u Φ ∗ S (E) (u) + S (M) (u), (11)
or in components
Sγ1 ...γ p = p!u [γ1 [S (E) (u)]γ2 ...γ p ] + [S (M) (u)]γ1 ...γ p . (12)
For example, the splitting of the unit volume 4-form ν gives rise to the following
representation
ν = −u Φ ∗ ν(u), (13)

that is [u Φ ∗ ν(u)]γβ∂ α = 2u [γ ν(u)β]∂ α + 2u [∂ ν(u)α]γβ , where the spatial unit
volume 3-form
ν(u)γβ∂ = u α ναγβ∂ (14)
is the only nontrivial spatial field which arises from the splitting of the volume
4-form. Using the spacetime (Hodge) duality operation (⇔ ), one can associate
with any p-form S (with 0 ≤ p ≤ 4) a (4 − p)-form. Similarly a spatial duality
operation (⇔(u) ) is defined for a spatial p-form S (u S = 0) replacing ν with
ν(u), namely
⇔(u) 1
Sγ1 ...γ3− p = Sβ ...β ν(u)β1 ...β p γ1 ...γ3− p . (15)
p! 1 p
For example, given a spatial 2-form S, its spatial dual is
1
[⇔(u) S]γ = ν(u)γβ∂ Sβ∂ . (16)
2
This operation satisfies the property ⇔(u) ⇔(u) S = S. Let us now consider the splitting
of ⇔ S where S is given by (11). We have
⇔
S = u Φ ∗ [⇔ S](E) (u) + [⇔ S](M) (u)
= ⇔ [u Φ ∗ S (E) (u) + S (M) (u)]
= ⇔(u) S (E) (u) + ⇔ [⇔(u) [⇔(u) [S (M) (u)]]]
= ⇔(u) S (E) (u) + (−1) p−1 u Φ ∗ ⇔(u) [S (M) (u)]. (17)
Comparing the first and the last line we have
[⇔ S](E) (u) = (−1) p−1⇔(u) [S (M) (u)], [⇔ S](M) (u) = ⇔(u) S (E) (u). (18)
4. Splitting of differential operators

In general relativity one has several spacetime tensorial differential operators
which act on tensor fields. Let us recall them: if T is a tensor field of any rank,
we have
a. The Lie derivative of T along the direction of a given vector field X : [£ X T ].
b. The covariant derivative of T : ≈T .
c. The absolute derivative of T along a curve with unit tangent vector X and
parameterized by s: ≈ X T ◦ DT /ds.
d. The Fermi–Walker derivative of T along a non-null curve with unit tangent
vector X and parameterized by s defined by
D(fw,X ) T γ β DT γ β
= ± [a(X ) ∗ X ]γ ∂ T ∂ β − [a(X ) ∗ X ]∂ β T∂γ ,
ds ds
where ± refer to transport along timelike or spacelike curves, respectively.
Finally if S is a p-form, one has
5. The exterior derivative of S: d S.
Application of the spatial projection into the L RSu of a family of observers u to
the spacetime derivatives (i)–(v), yields new operators which can be more easily
confronted with those defined in a three-dimensional Euclidean space. Given a
tensor field T of components T γ... β... we have in fact
a. The spatially projected Lie derivative along a vector field X
[£(u) X T ]γ... β... ◦ P(u)γ ω . . . P(u)ρ β . . . [£ X T ]ω ... ρ... ; (19)
when X = u we use also the notation
≈(u)(lie) T ◦ £(u)u T, (20)
and this operation will be termed “spatial-Lie temporal derivative”.

b. The spatially projected covariant derivative along any e∂ frame direction
72 D. Bini
≈(u)∂ T ◦ P(u)≈∂ T, (21)
namely
[≈(u)∂ T ]γ... β... = P(u)γ γ1 . . . P(u)β1 β . . . P(u)ω ∂ ≈ω T γ1 ... β1 ... .
c. The spatially projected absolute derivative along a curve with unit tangent
vector X
[P(u)≈ X T ]γ... β... = P(u)γ γ1 . . . P(u)β1 β . . . [≈ X T ]γ1 ... β1 ... (22)
d. The spatially projected “Fermi–Walker derivative” along a curve with unit

tangent vector X and parameterized by s
γ... ω ...
D(fw,X ) T D(fw,X ) T
P(u) β... = P(u)γ ω . . . P(u)ρ β . . . ρ... .
ds ds
e. the spatially projected exterior derivative of a p-form S
d(u)S ◦ P(u)d S, (23)
namely [d(u)S]γ1 ...γ p β = P(u)β1 γ1 . . . P(u)ω β [d S]β1 ...ω .

Note that all these spatial differential operators are well defined since they arise
from the spatial projection of spacetime differential operators. From their defini-
tions it is clear that both the Fermi–Walker and the Lie derivatives of the vector
field u along itself vanish identically (and so do the projections orthogonal to u
of these derivatives). The only derivative of u along itself which is meaningful
being different than zero is the covariant derivative
P(u)≈u u = ≈u u = a(u). (24)
3 Three-Dimensional Notation
Let u be a given a family of observers and X a spatial vector with respect to u. It is

then convenient to introduce the 3-dimensional vector notation for the spatial inner
product and the spatial cross product of two spatial vector fields X and Y . The spatial
inner product is defined as
X ·u Y = P(u)γβ X γ Y β (25)
while the spatial cross product is

[X ×u Y ]γ = ν(u)γ β∂ X β Y ∂ , (26)
where ν(u)γ β∂ = u ω νω γ β∂ as stated.

In terms of the above definitions we can define spatial gradient, curl and divergence
operators of functions f and spatial vector fields X as
gradu f = ≈(u) f , curlu X = ≈(u) ×u X , divu X = ≈(u) ·u X. (27)
In components these relations read
[gradu f ]γ = ≈(u)γ f = P(u)γβ eβ ( f ),

[curlu X ]γ = ν(u)γβ∂ ≈(u)β X ∂ = u ω νω γβ∂ ≈β X ∂ ,
[divu X ] = ≈(u)γ X γ = P(u)γβ ≈γ X β . (28)
It is useful to extend the above definitions to

1. the spatial cross product of a vector X by a symmetric tensor A,
[X ×u A]γβ = ν(u) ∂ α(γ X ∂ Aβ) α , (29)
2. the spatial cross product of two symmetric spatial tensors A and B,
[A ×u B]γ = ν(u)γβ∂ Aβ α B α∂ , (30)
3. the spatial inner product of two symmetric spatial tensors A and B,
[A ·u B]γ β = Aγ∂ B ∂β , (31)
4. the trace of the above tensor product
Tr[A ·u B]Aγβ B γβ , (32)
5. the spatial divergence of a spatial tensor
[divu X ]γ...β = ≈(u)ω X ω γ...β , X = P(u)X. (33)
4 Kinematics of the Observer’s Congruence

Let us consider now the splitting of the covariant derivative ≈β u γ . This operation
generates two spatial fields namely the acceleration vector field a(u) and the kine-
matical tensor field k(u)defined as
a(u) = P(u)≈u u, k(u) = −≈(u)u = β(u) − Θ (u). (34)

74 D. Bini
where
[β(u)]γβ = −P(u)μ τ
γ P(u)β ≈[μ u τ] ,
1
[Θ (u)]γβ = P(u)μ τ
γ P(u)β ≈(μ u τ) = [£(u)u P(u)]γβ , (35)
2
are the components of tensor fields β(u) and Θ (u) having the meaning respectively
of vorticity (whose sign depends on convention1 ) and expansion. From the above
definitions, the tensor field ≈β u γ can be written as
≈β u γ = −a(u)γ u β − k(u)γ β . (36)
The expansion tensor field Θ (u) may itself be decomposed into its trace-free and pure
trace parts
1
Θ (u) = ω (u) + ξ(u)P(u), (37)
3
where the trace-free tensor field ω (u) (ω (u)γ γ = 0) is termed shear and the scalar
ξ(u) = ≈γ u γ (38)
is termed volumetric (or isotropic) scalar expansion.

Define also the vorticity vector field β(u) = 1/2 curlu u as the spatial dual of the
spatial rotation tensor, and given by
1 1
β(u)γ = ν(u)γβ∂ β(u)β∂ = νω γβ∂ u ω ≈β u ∂ . (39)
2 2
Although we use the same symbol for the vorticity tensor and the associated vector
they can be easily distinguished by the context.
5 Adapted Frames
Given a field of observers u, a frame {eγ } with γ = 0, 1, 2, 3 (with dual βγ ) is termed
adapted to u if e0 = u and ea with a = 1, 2, 3 are orthogonal to u, namely u · ea = 0.
From this it follows that β0 = −u Φ . In this section all indices denote components
relative to the frame {eγ }. The evolution of the frame vectors along the world lines of
u is governed by the relations ≈u eγ = eω ζ ω γ0 and one can express the connection
coefficients in terms of the kinematical quantities of the observer congruence u. The
result is the following
1 We have adopted the ≈-convention differently from a ;-convention also widely used.
ζ a 00 = a(u)a , ζ 0 a0 = a(u)a , ζ b a0 = C(fw) b a ,

(40)
ζ b 0a = −k(u)b a , ζ 0 ba = −k(u)ba ,
where the Fermi–Walker structure functions C(fw)ba are introduced so that
P(u)≈u ea = C(fw) b a eb . (41)
They can also we written as
C(fw) b a = C(lie) b a − k(u)b a , C(lie) b a ◦ βb (£(u)u ea ), (42)
implying
P(u)£u ea = £(u)u ea = C(lie) b a eb . (43)
Similarly, it is straightforward to express the structure functions in terms of kinemat-

ical quantities. In fact, from the definition
eγ C γ β∂ = [eβ , e∂ ] = ≈eβ e∂ − ≈e∂ eβ (44)
we have
eγ C γ 0b = ≈u eb − ≈eb u = a(u)b u + [C(fw) c b + k(u)c b ]ec

= a(u)b u + C(lie) c b ec , (45)
so that C 0 0b = a(u)b and C c 0b = C(lie) c b . Similarly
eγ C γ bc = ≈eb ec − ≈ec eb = 2β(u)bc u + 2ζ d [cb] ed , (46)
so that C 0 bc = 2β(u)bc and C d bc = 2ζ d [cb] .

Finally, the structure functions satisfy the Jacobi identities which can also be given
a 3 + 1 form.
5.1 Spatial-Fermi–Walker and Spatial-Lie Temporal

Derivatives
We have introduced the Fermi–Walker structure functions C(fw) b a ,
P(u)≈u ea = C(fw) b a eb , (47)
as well as the Lie structure functions C(lie) b a entering the projected Lie derivative
along u (which we also termed as “spatial-Lie temporal derivative”, see Eq. (20))
76 D. Bini
£(u)u ea = P(u)£u ea = C(lie) b a eb = C b 0a eb . (48)
It is then useful to handle both these operations with a unified notation [1]
≈(u)(tem) ea = C(tem) b a eb , tem = fw, lie, (49)
where we define
≈(u)(fw) ea ◦ P(u)≈u ea , ≈(u)(lie) ea ◦ P(u)£u ea = £(u)u ea . (50)
Therefore if X is a vector field orthogonal to u, i.e. X · u = 0, we have
d Xa
≈(u)(tem) X = ≈(u)(tem) (X a ea ) = ea + X a C(tem) b a eb
dδu
d Xb
= + X a C(tem) b a eb = (≈(u)(tem) X b )eb . (51)
dδu
The operation ≈(u)(fw) = P(u)≈u is termed “spatial-Fermi–Walker temporal deriv-

ative.” It can be extended to non-spatial fields. If we apply this operation to the vector
field u itself we have
≈(u)(fw) u = P(u)≈u u = a(u). (52)
Hence the temporal derivatives so defined through their action on purely spatial and
purely temporal fields can now act on any spacetime field.
5.2 Frame Components of the Riemann Tensor
From the definition
eγ R γ β∂ α = [≈e∂ , ≈eα ]eβ − C ω ∂ α ≈eω eβ , (53)
we have
eγ R γ 0b0 = [≈eb , ≈u ]u − C ω b0 ≈eω u (54)

= [≈(u)b + a(u)b ]a(u) + ≈(u)(fw) k(u) b − [k(u) ] b ec ,
c c 2 c
so that
R c 0b0 = [≈(u)b + a(u)b ]a(u)c + ≈(u)(fw) k(u)c b − [k(u)2 ]c b , (55)

where
≈(u)(fw) k(u)c b = ≈u k(u)c b + C(fw) c f k(u) f b − C(fw) f b k(u)c f . (56)
Similarly one obtains
R 0 bcd = −2[≈(u)[c k(u)|b|d] + β(u)cd a(u)b ] (57)
and
R f bcd = R(fw) f bcd − 2k(u)b[c k(u) f d] , (58)
where

R(fw) f bcd = 2e[c ζ f
|b|d] + 2ζ s b[c ζ f
|s|d] − C s cd ζ f
bs
−2β(u)cd C(fw) f b . (59)
This tensor is termed “Fermi–Walker spatial Riemann tensor”2 ; it can be written in

invariant form as follows
R(fw) (u)(X, Y )Z = {[≈(u) X , ≈(u)Y ]Z − ≈(u)[X,Y ] }Z

−2β(u)(X, Y )≈(u)(fw) Z , (60)
where X , Y and Z are spatial fields with respect to u and we note that
[X, Y ] = P(u)[X, Y ] − 2βΦ (X, Y ) u. (61)
The Fermi–Walker spatial Riemann tensor has not all the symmetries of a three
dimensional Riemann tensor. For instance it does not satisfy the Ricci identities. In
fact we have
0 = R f [bcd] = R(fw) f [bcd] − 2k(u)[bc k(u) f d] , (62)
and hence
R(fw) f [bcd] = 2k(u) f [b β(u)cd] . (63)
From the latter one can construct a new Riemann tensor with all the necessary
symmetries. Ferrarese [2] has shown that the symmetry-obeying Riemann tensor,
denoted as R(sym) ab cd , is related to the Fermi–Walker Riemann tensor (60) by
2 A Lie spatial Riemann tensor can be defined similarly, replacing the Fermi–Walker structure
functions C(fw) f b with the corresponding Lie structure functions C(lie) f b according to Eq. (42).
78 D. Bini
R(sym) ab cd = R(fw) ab cd − 2β(u)ab β(u)cd − 4Θ (u)[a [c β(u)b] d]

= R ab cd + 2k(u)b [c k(u)a d] − 2β(u)ab β(u)cd
−4Θ (u)[a [c β(u)b] d] . (64)
Together with the spatial symmetric Riemann tensor R(sym) ab cd we also intro-
duce the spatial symmetric Ricci tensor, R(sym) a b = R(sym) ca cb as well as the the
associated scalar R(sym) = R(sym) a a .
6 Comparing Families of Observers
Let u and U be two unitary timelike vector fields. Define the relative spatial velocity
of U with respect to u from the splitting relations
U = ∂ (U, u)[u + τ(U, u)] = ∂ (U, u)[u + ||τ(U, u)||τ̂(U, u)], (65)
and
u = ∂ (u, U )[U + τ(u, U )] = ∂ (u, U )[U + ||τ(u, U )||τ̂(u, U )]. (66)
where τ̂(U, u) is the unitary vector giving the direction of τ(U, u) in the rest frame
of u. Both the spatial relative velocity vectors have the same magnitude
||τ(U, u)|| = [τ(U, u)γ τ(U, u)γ ]1/2 = [τ(u, U )γ τ(u, U )γ ]1/2 .
The common gamma factor is related to that magnitude by
∂ (U, u) = ∂ (u, U ) = [1 − ||τ(U, u)||2 ]−1/2 = −Uγ u γ (67)
hence we recognize it as the relative Lorentz factor. It is convenient to abbreviate

∂ (U, u) by ∂ and ||τ(U, u|| by τ when their meaning is clear from the context and
there are no more than two observers involved.
Let us notice here that by substituting Eq. (65) into Eq. (66) we obtain the following
relation
−τ̂(u, U ) = ∂ [τ̂(U, u) + τu], (68)
which together with U = ∂ [u + τ τ̂(U, u)] yields the relative boost B(U, u) from u
to U , namely
B(U, u)u = U = ∂ [u + τ τ̂(U, u)]

B(U, u)τ̂(U, u) = −τ̂(u, U ) = ∂ [τ̂(U, u) + τu]. (69)
The inverse relations hold by interchanging U with u. The boost acts as the identity
on the intersection of their local rest spaces L RSu ≡ L RSU .
1. Maps between LRSs
The spatial measurements of two observers in relative motion can be compared
only relating their respective LRSs. Let U and u be two such observers and
L RSU and L RSu their LRSs. There exists several maps between these LRSs; for
example, by combining the projection operators P(U ) and P(u) one can form
the following “mixed projection” maps:
a. P(U, u) from the L RSu into L RSU , defined as
P(U, u) = P(U )P(u) : L RSu ∇ L RSU , (70)
with inverse: P(U, u)−1 : L RSU ∇ L RSu ;

b. P(u, U ) from the L RSU into L RSu , defined as
P(u, U ) = P(u)P(U ) : L RSU ∇ L RSu , (71)
with inverse: P(u, U )−1 : L RSu ∇ L RSU .

Note that P(U, u) ⇒= P(u, U )−1 as it follows from their representations
P(U, u) = P(u) + ∂ τU ⊗ τ̂(U, u),

P(U, u)−1 = P(U ) + τU ⊗ τ̂(u, U ),
P(u, U ) = P(U ) + ∂ τu ⊗ τ̂(u, U ),
−1
P(u, U ) = P(u) + τu ⊗ τ̂(U, u).
One can then show that
P(U, u)τ̂(U, u) = −∂ τ̂(u, U ),

1
P(u, U )−1 τ̂(U, u) = − τ̂(u, U ). (72)
∂
Note that
P(U, u) = P(U ) P(u) = P(U ) P(U, u) = P(U, u) P(u). (73)
Moreover the following relations hold
P(U ) = P(U, u) P(U, u)−1 , P(u) = P(U, u)−1 P(U, u). (74)
80 D. Bini
2. The boost maps
Similarly to what we have done combining the projection maps, also the boost
B(U, u) induces an invertible map between the local rest spaces of the given
observers defined as
B(lrs) (U, u) ◦ P(U )B(U, u)P(u) : L RSu ∇ L RSU . (75)
It acts as the identity on the intersection of their subspaces L RSU ≡ L S Ru . Being

the boost an isometry, exchanging the role of U and u in (75) leads to the inverse
boost
B(lrs) (U, u)−1 ◦ B(lrs) (u, U ) : L RSU ∇ L RSu . (76)
The representations of the boost and its inverse can be given in terms of the
associated tensors
B(lrs)u (U, u), B(lrs)U (U, u), B(lrs)u (u, U ), B(lrs)U (u, U ), (77)
defined by:
B(lrs)u (U, u) = P(U, u)−1 B(lrs) (U, u),

B(lrs)U (U, u) = B(lrs) (U, u) P(U, u)−1 , (78)
with the corresponding expressions for the inverse boost obtained simply by
exchanging the role of U and u and with
B(lrs) (U, u) = B(lrs)U (U, u) P(U, u) = P(U, u) B(lrs)u (U, u). (79)
The explicit expression of B(lrs)u (U, u), for example, is given by
1−∂
B(lrs)u (U, u) = P(u) + τ̂(U, u) ⊗ τ̂(U, u). (80)
∂
This can be shown as follows. Let X ∈ L RSu then
B(lrs)u (U, u)X = P(U, u)−1 [B(lrs) (U, u)X ]

= P(U, u)−1 [B(lrs) (U, u)X || τ̂(U, u) + X ⊥ ], (81)
where X || = X · τ̂(U, u) and X ⊥ = X − X || τ̂(U, u), that is
X = X || τ̂(U, u) + X ⊥ , (82)
and we have used the fact that the boost reduces to the identity for vectors not
belonging to the boost plane, as X ⊥ . In this case the boost plane is spanned by
the vectors u and τ̂(U, u). Taking into account (69), namely
B(lrs) (U, u)τ̂(U, u) = −τ̂(u, U ), (83)
as well as the linearity of the boost map, we have
B(lrs)u (U, u)X = P(U, u)−1 [X − X || τ̂(u, U ) − X || τ̂(U, u)]

1 − ∂ ||
=X+ X τ̂(U, u), (84)
∂
where P(U, u)−1 X = X because X ∈ L RSu :
P(U, u)−1 X = P(U, u)−1 P(u)X = P(U, u)−1 P(U )P(u)X

= P(U, u)−1 P(U, u)X = P(u)X = X ; (85)
hence P(U, u)−1 τ̂(U, u) = τ̂(U, u) because τ(U, u) belongs to L RSu . More-
over, from (72), by exchanging the roles of U and u, we find
1 1
P(U, u)−1 τ̂(u, U ) = − τ̂(U, u) = B(u, U )τ̂(u, U )
∂ ∂
1 −1
= B(U, u) τ̂(u, U ). (86)
∂
Therefore:
1
B(lrs)u (U, u)X = X − X || − + 1 τ̂(U, u)
∂
∂ −1
= P(u) − τ̂(U, u) ⊗ τ̂(U, u)Φ X (87)
∂
which is equivalent to (80).

Similarly, for the inverse boost B(lrs) (u, U ) one has
∂ −1
B(lrs)u (u, U ) = P(u) − τ̂(U, u) ⊗ τ̂(U, u)Φ ,
∂
∂ −1
B(lrs)U (u, U ) = P(U ) − τ̂(u, U ) ⊗ τ̂(u, U )Φ . (88)
∂
Thus, if S is a vector field such that S ∈ L RSU , then its inverse boost is the vector
belonging to L RSu
B(lrs) (u, U )S = [P(u) − ∂ (∂ + 1)−1 τ(U, u) ⊗ τ(U, u)Φ ] P(u, U )S. (89)
82 D. Bini
7 Splitting of Derivatives Along a Timelike Curve
Consider a congruence of curves CU with tangent vector field U and proper time δU
as parameter. We know, at this stage, that the evolution along CU of any tensor field
can be specified by one of the following spacetime derivatives:
1. the absolute derivative along CU : D/dδU = ≈U ,
2. the Fermi–Walker derivative along CU : D(fw,U ) /dδU ,
3. the spacetime Lie derivative along CU : £U , for which we use also the notation
D(lie,U ) /dδU = £U .
The action of the Fermi–Walker and Lie derivatives on a vector field X is related to
the absolute derivative as follows
D(fw,U ) X
= ≈U X + a(U )(U · X ) − U (a(U ) · X )
dδU
= P(U )≈U X − U ≈U (X · U ) + a(U )(X · U ) ,
D(lie,U ) X
= [U, X ] = ≈U X + a(U )(U · X ) − k(U ) X , (90)
dδU
where k(U ) = β(U ) − Θ (U ) is the kinematical tensor of the congruence CU defined

in (34). For X = U we have
D(fw,U ) U D(lie,U ) U
= 0, = 0, (91)
dδU dδU
whereas DU /dδU = a(U ).

If X is spatial with respect to U , namely X · U = 0, we have instead
D(fw,U ) X
= P(U )≈U X
dδU
D(lie,U ) X
= ≈U X − k(U ) X
dδU
= P(U )≈U X − U (a(U ) · X ) − k(U ) X. (92)
The projection orthogonal to U of D(lie,U ) X/dδU as in (92) gives
D(lie,U ) X D(fw,U )
P(U ) = X − k(U ) X. (93)
dδU dδU
Let u be another family of observers whose world lines have as parameter the proper
time δu . One can introduce on the congruence CU whose unit tangent vector field
can be written as
U = ∂ (U, u)[u + τ(U, u)], (94)

two new parametrizations δ(U,u) and (U,u) as follows
dδ(U,u) d(U,u)
= ∂ (U, u), = ∂ (U, u)||τ(U, u)||, (95)
dδU dδU
where δ(U,u) corresponds to the proper times of the observers u when their curves are
crossed by a given curve of CU and (U,u) corresponds to the proper length on CU .
The projection orthogonal to u of the absolute derivative along U is expressed by
D
P(u) = P(u)≈U = ∂ [P(u)≈u + P(u)≈τ(U,u) ]
dδU
= ∂ [P(u)≈u + ≈(u)τ(U,u) ]. (96)
We note that in the above equation the derivative operation P(u)≈u is just what we
have termed spatial-Fermi–Walker temporal derivative, i.e. ≈(u)(fw) , in (50). For a
vector field X we can write then
D(fw,U ) X D(fw,U,u) X
P(u) ◦
dδU dδU
DX
= P(u) + P(u, U )a(U )(U · X )
dδU
− ∂ τ(U, u)(a(U ) · X ), (97)
D(lie,U ) X D(lie,U,u) X
P(u) ◦
dδU dδU
DX
= P(u) + P(u, U )a(U )(U · X )
dδU
− P(u)[k(U ) X ]. (98)
We shall now examine the projected absolute derivative in detail.
7.1 Projected Absolute Derivative
Consider the absolute derivative of u along U , namely ≈U u. Since u is unitary, then

u · Du/dδU = 0 and we can write
Du Du
= P(u) = ∂ [P(u)≈u u + P(u)≈τ(U,u) u]
dδU dδU
= ∂ [≈(u)(fw) u + P(u)≈τ(U,u) u]
= ∂ [a(u) + β(u) ×u τ(U, u) + Θ (u) τ(U, u)]. (99)
84 D. Bini
Let us denote the above quantity as (minus) Fermi–Walker gravitational force, namely
(G) Du
F(fw,U,u) = − = −∂ [a(u) + β(u) ×u τ(U, u) + Θ (u) τ(U, u)] .
dδU
(G)
It should be stressed here that, although F(fw,U,u) is referred to as a gravitational
force, it contains contributions by true gravity and by inertial forces.
Consider now the case of X orthogonal to u, i.e. X · u = 0. The projection onto
L RSu of the absolute derivative of X along U gives
DX D(fw,U,u) X
P(u) = ∂ [P(u)≈u X + ≈(u)τ(U,u) X ] ◦ . (100)
dδU dδU
This differential operator plays an important role since both Fermi–Walker and Lie
derivatives along U can be expressed in terms of it. In terms of (adapted) frame
components the above expression reads

DX d Xb
P(u) = + ∂ X a C(fw) b a + τ(U, u)c ζ b ac eb , (101)
dδU dδU
where we set X = X a ea . Introducing the relative standard time parametrization

δ (U, u) defined in (95), we have
a
DX D(fw,U,u) X D(fw,U,u) X
P(u) = = ea (102)
dδ(U,u) dδ(U,u) dδ(U,u)
or in components
D(fw,U,u) X b
d Xb
= + X a C(fw) b a + τ(U, u)c ζ (u)b ac . (103)
dδ(U,u) dδ(U,u)
A particular vector field which is orthogonal to u and is defined all along CU is the
field of relative velocities, τ(U, u). We introduce the acceleration of U relative to
u by
D(fw,U,u) D
a(fw,U,u) = τ(U, u) = ∂ P(u) τ(U, u). (104)
dδ(U,u) dδU
Considering instead the unit vector τ̂(U, u), this quantity can be written as
D
a(fw,U,u) = P(u) [τ τ̂(U, u)], (105)
dδ(U,u)
where τ = ||τ(U, u)||. Finally we have

dτ D
a(fw,U,u) = τ̂(U, u) + τ P(u) τ̂(U, u). (106)
dδ(U,u) dδ(U,u)
It is therefore quite natural to denote the first term as a tangential Fermi–Walker

(T )
acceleration, a(fw,U,u) of U relative to u and the second as centripetal Fermi–Walker
(C)
acceleration a(fw,U,u) of U relative to u:
(T ) (C)
a(fw,U,u) = a(fw,U,u) + a(fw,U,u) , (107)
where
(T ) dτ
a(fw,U,u) = τ̂(U, u) ,
dδ(U,u)
(C) D D(fw,U,u)
a(fw,U,u) = τ P(u) τ̂(U, u) = τ τ̂(U, u). (108)
dδ(U,u) dδ(U,u)
To generalize the classical mechanics notion of centripetal acceleration we need to

convert the relative standard time parametrization into an analogous relative standard
length parametrization3 :
d(U,u) = τdδ(U,u) . (109)
With this parametrization we have
(C) D D(fw,U,u)
a(fw,U,u) = τ 2 P(u) τ̂(U, u) = τ τ̂(U, u)
d(U,u) dδ(U,u)
τ2
= ν̂(fw,U,u) = τ 2 k(fw,U,u) ν̂(fw,U,u) , (110)
R(fw,U,u)
where ν̂(fw,U,u) is a unit spacelike vector orthogonal to τ̂(U, u), k(fw,U,u) is the
Fermi–Walker relative curvature and R(fw,U,u) is the curvature radius of the curve
such that
ν̂(fw,U,u) D
k(fw,U,u) ν̂(fw,U,u) = = P(u) τ̂(U, u). (111)
R(fw,U,u) d(U,u)
Clearly, if geometrically or physically motivated, one can replace the Spatial-Fermi–

Walker temporal derivative with the Spatial-Lie temporal derivative defining the
corresponding quantities. Doing this one really understands the power of the notation
used. For example and for later use one can define Lie relative curvature of a curve
and the associated curvature radius
3 In fact the Euclidean space definition involves spatial orbits parameterized by the (spatial) curvi-
linear abscissa.
86 D. Bini
ν̂(lie,U,u) D(lie,U,u)
k(lie,U,u) ν̂(lie,U,u) = = τ̂(U, u). (112)
R(lie,U,u) d(U,u)
Difficult to think of other efficient relativistic generalizations of the classical concepts

of inertial forces besides this one.
8 Preferred Slicing in Spacetimes Admitting

Separable Geodesics
After providing the general framework of spacetime splitting techniques let us briefly
review now some recent results concerning the existence of preferred slicing in those
spacetimes admitting separable geodesics (see [7] and references therein).
Let the coordinates x γ (γ = 0 . . . 3, with x 0 = t) be such that the geodesic
equations are separable in the metric ds 2 = gγβ d x γ d x β . Using the Hamilton–
Jacobi formalism we can write the tangent vector U γ = d x γ (κ)/dκ to the affinely
parametrized timelike geodesics as the gradient of the fundamental action function
S = S(x γ , κ), Uγ = εγ S, satisfying the Hamilton–Jacobi equation
εS
− = H (x γ , εγ S), (113)
εκ
with κ an affine parameter for the integral curves of U and the Hamiltonian
1 γβ 1
H= g εγ Sεβ S = − μ2 = const, (114)
2 2
the latter identity following from the normalization condition U γ Uγ = −μ2 for
timelike geodesics. Assume that S can be separated in its dependence on the variables
x γ and κ, namely
1 2
S= μ κ + St (t) + S1 (x 1 ) + S2 (x 2 ) + S3 (x 3 ). (115)
2
Thus we have for the 1-form U Φ ◦ Uγ d x γ = εγ Sd x γ = d(S − 21 μ2 κ), where

here d stands for the spacetime differential only. Moreover, since in this case U is a
gradient it is also necessarily vorticity-free: dU Φ = 0, and there exists a distribution
of constant action hypersurfaces T ◦ −S + 21 μ2 κ = const with
− dT = Uγ d x γ , (116)
such that U γ is the associated unit normal vector field. When one sets μ = 1, then the
time function T measures the proper time along the geodesics and the corresponding
lapse function has the fixed value N = 1. For a stationary spacetime in which t is
taken to be a Killing time coordinate, then Ut = −E is a constant interpreted as a

conserved energy, with St (t) = −Et, and the metric is independent of t. One then
has
− dT = −Edt + Ua d x a . (117)
8.1 Static Spherically Symmetric Spacetimes
Consider the case of static spherically symmetric spacetimes. The metric written in
standard spherical-like coordinates is
ds 2 = −eτ dt 2 + eκ dr 2 + r 2 (dΘ 2 + sin2 Θ dφ 2 ), (118)
where the functions τ and κ depend only on the radial coordinate. Then L = Uφ is
an additional Killing constant associated with the conserved angular momentum so
Uγ d x γ = −Edt + (εr Sr )dr + (εΘ SΘ )dΘ + Ldφ, (119)
and the corresponding Hamilton–Jacobi equation
1 L2
− e−τ E 2 + e−κ (εr Sr )2 + (ε Θ SΘ ) 2
+ = −μ2 (120)
r2 sin2 Θ
can be easily separated in its dependence on the coordinates by setting the square
bracket expression to a separation constant K , leading to

d Sr K + μ2 r 2 d SΘ L2
= ψr eκ/2 E 2 e−τ − , = ψΘ K − , (121)
dr r2 dΘ sin2 Θ
where |r | = 1 = |ψΘ |.

As stated above, we can set μ = 1 to characterize a new foliation by a new
temporal coordinate T measuring proper time along the orthogonal geodesics. We
are left to specify E, L, and K to obtain a specific family of timelike geodesics
covering the spacetime. The simplest choice would be a spherically symmetric 4-
velocity field involving only radial motion of the geodesics relative to the original
coordinates. We can achieve this in two steps. First we can require that this family
of geodesics be tangent to the equatorial plane Θ = π/2, which requires K = L 2
to make UΘ = 0, resulting in
L2
Ur2 = eκ−τ E 2 − eτ 1 + . (122)
r2
We then impose the radial condition L = 0, so that

88 D. Bini

Ur = ψr e(κ−τ)/2 E 2 − eτ , (123)
leaving finally the choice of the energy constant E. For spatially asymptotically flat
spacetimes where eτ < 1 approaches 1 as r ∇ ⇐, to have a choice which works even
at spatial infinity, we must have E ≥ 1, in which case the value may be interpreted as
the energy of the radially moving geodesics at spatial infinity. Of course one could
choose E < 1 but this would limit the slicing to the interior of a cylinder in spacetime
inside the radial turning point of the geodesic motion.
The new time differential is then
dT = Edt − Ur dr. (124)
A new global coordinate system for static spacetimes is given by (X γ ) = (T, R, Θ, φ)

with R = r and Θ and φ unchanged and T = Et + f (r ) given by integrating the
differential equation f (r ) = −Ur . This leads to
Ur
εT = E −1 εt , ε R = εr + εt , (125)
E
and the transformed metric is
ds 2 = −dT 2 + ∂ab (d X a + N a dT )(d X b + N b dT ), (126)
with unit lapse function and the shift vector field aligned with the new radial direction,
i.e.,

N a = −α aR U r = −α aR e−κ Ur = −α aR r e−(κ+τ)/2 E 2 − eτ . (127)
The 3-metric induced on the T = const hypersurfaces is then given by
(3) eκ+τ 2
ds 2 = dr + r 2 (dΘ 2 + sin2 Θ dφ 2 ). (128)
E2
In the case of vacuum as well as in the presence of a nonzero cosmological constant
one has κ + τ = 0, so that the induced metric is then
(3) dr 2
ds 2 = + r 2 (dΘ 2 + sin2 Θ dφ 2 ), (129)
E2
whose only nonvanishing component of the spatial Riemann curvature tensor and
the spatial curvature scalar are
(3) 1 − E2 1
R Θφ Θφ = = (3) R, (130)
r2 2
with positive or negative curvature respectively for 0 < E < 1 (bound geodesics)
or E > 1 (unbound geodesics). The choice E = 1 leads to a flat 3-geometry. The
additional sign choice r = −1 corresponds to the radially infalling geodesics which
start at rest at spatial infinity. This is the case for the Schwarzschild spacetime where
the Painlevé-Gullstrand coordinates were originally found [8, 9].
9 Discussion
The results summarized above have been developed over a period of about 20 years
(starting from the 1990s), initially motivated by the necessity to correctly define iner-
tial forces in general relativity. As a consequence, the whole “measurement process”
was reformulated, paying much attention to involve only quantities with a clear
geometrical and physical meaning.
Actually, the relativistic generalization of well known classical quantities neces-
sitated the introduction of the so called observer’s viewpoint and, formally, the sys-
tematic use of “1 + 3” spacetime splitting techniques. Relative Frenet–Serret frames,
for instance, were perhaps the most suited tools to explain how inertial forces could
enter the general relativistic dynamics of test particles, in full similarity with the
classical situation.
A lot of progress in this field (which was not at all a newborn field) was possible
because of the international competition started in analyzing explicit applications of
formalism to test particle motion in black hole spacetimes. The original enthusiasm
was swept away when a satisfactory understanding of the problem was obtained. Nev-
ertheless, many aspects of the formalism developed in this field may be exported in
different contexts and hence one should wait for another wave of splitting formalism
when new applications will be taken into account.
Acknowledgments I’m indebted to my “teachers,” Profs. R.T. Jantzen and F. de Felice, for a more
than 20 years of collaboration and friendship. I also acknowledge the numerous useful discussions
with Dr. A. Geralico. Finally, I warmly thank the organizers of this wonderful meeting in Prague
for all their work.
References
1. R.T. Jantzen, P. Carini, D. Bini, The many faces of gravitoelectromagnetism, Ann. Physics 215,
1 (1992). doi:10.1016/0003-4916(92)90297-Y
2. Ferrarese, G.: Proprietà di secondo ordine di un generico riferimento fisico in Relatività generale.
Rend. Mat. Roma 24, 57 (1965)
3. D. Bini, P. Carini, R.T. Jantzen, The intrinsic derivative and centrifugal forces. I: Theoretical
foundations, Int. J. Mod. Phys. D 6, 1 (1997). doi:10.1142/S0218271897000029
4. D. Bini, P. Carini, R.T. Jantzen, The intrinsic derivative and centrifugal forces. II: Applications
to some familiar stationary axisymmetric spacetimes, Int. J. Mod. Phys. D 6, 143 (1997). doi:10.
1142/S021827189700011X
90 D. Bini
5. D. Bini, F. de Felice, R.T. Jantzen, Absolute and relative Frenet-Serret frames and Fermi-Walker
transport, Class. Quantum Grav. 16, 2105 (1999). doi:10.1088/0264-9381/16/6/333
6. de Felice, F., Bini, D.: Classical Measurements in Curved Space-Times. Cambridge Monographs
on Mathematical Physics (Cambridge University Press, Cambridge; New York (2010)
7. D. Bini, A. Geralico, R.T. Jantzen, Separable geodesic action slicing in stationary spacetimes,
Gen. Rel. Grav. 44, 603 (2012). doi:10.1007/s10714-011-1295-2
8. Painlevé, P.: La mécanique classique et la theorie de la relativité. C. R. Acad. Sci. 173, 677
(1921)
9. Gullstrand, A.: Allgemeine Lösung des statischen Einkörperproblems in der Einsteinschen Grav-
itationstheorie. Ark. Mat. Astron. Fys. 16(8), 1 (1922)
Some Links Between General Relativity
and Other Parts of Physics
Gary W. Gibbons
Abstract Now that General Relativity has become such a central part of modern
physics, its geometrical formalism being taught as part of almost all undergraduate
physics courses, it is natural to ask: how can its basic concepts and techniques be used
to illuminate areas of physics which have no connection with gravity? Another way
of asking this question is: are the analogues situations to those occurring in General
Relativity? The search for such analogues is of course an old one, but recently,
because of advances in technology, these questions have become more topical. In
this talk I will illustrate this theme by examples drawn from optics, acoustics, liquid
crystals, graphene and the currently popular topic of cloaking.
1 Introduction
General Relativity, its mathematical techniques and conceptual framework are by

now part of the tool-kit of (almost) all theoretical physicists and at least some pure
mathematicians. They have become part of the natural language of physics. Indeed,
parts of the subject are passing into mathematics departments. It is natural therefore
to ask to what extent can they illuminate other (non-relativistic) areas of physics.
It is also the case that the relentless onward progress of technology makes possible
analogue experiments illustrating basic ideas in General Relativity. In this talk I will
illustrate this ongoing process of unification.
G. W. Gibbons(B)
DAMTP, Centre for Mathematical Sciences, Wilberforce Rd, Cambridge CB3 0WA, UK
e-mail: gwg1@damtp.cam.ac.uk

92 G. W. Gibbons
1.1 Dynamical Casimir Effect
As a topical example of the relentless progress of technology November 2011 saw

the demonstration in the laboratory [1] some 40 years after the original prediction
[2, 3] of a very basic mechanism in semi-classical General Relativity: amplifica-
tion of vacuum fluctuations in a time-dependent environment. This is the basis of
all we believe about inflationary perturbations, Hawking evaporation, Black Hole
information “Paradox” and much of AdS/CFT correspondence etc.
1.2 Some Previous Work
The idea of finding analogue models for general relativistic effects is not new, but
the pace has quickened of late. Some important early work was done on cosmic
strings modelled by vortices in superfluid helium 4 and by Volovik [4], who noted
that the order parameter of some phases of superfluid helium 3 is a triad ei such that
ei · e j = εi j . More recently, the emphasis has shifted to the optics of metamaterials
and most recently to graphene. There are also interesting analogies in liquid crystals.
2 Shallow Water Waves
Let’s start with a very simple example which will illustrate some basic ideas. If
γ = γ(t, x, y) is the height of the water above its level when no waves are present
and h = h(x, y) the depth of the water, then shallow water waves satisfy the non-
dispersive wave equation (this is the analogue of the Einstein Equivalence Principle)
(ag hγx )x + (ag hγ y ) y = γtt ,
where ag is the acceleration due to gravity. From now on we adopt units in which
ag = 1. The wave operator coincides with the covariant d’Alembertian
1 ≈
≈ ∂μ ( −g g μν ∂ν γ) = 0,
−g
with respect to the 2 + 1 dimensional spacetime metric
ds 2 = −h 2 dt 2 + h(dx 2 + dy 2 ).
Applying ray theory and geometrical optics, one writes
γ = Ae−iλ(t−W (x,y)) ,
Some Links Between General Relativity and Other Parts of Physics 93
where A(x, y) is slowly varying. To lowest order W satisfies the Hamilton-Jacobi

equation
2 2
∂W ∂W 1
+ = ,
∂x ∂y h
and the rays are solutions of
dx ∂W
=h .
dt ∂x
Given any static spacetime metric
ds 2 = −V 2 dt 2 + gi j dx i dx j ,
the projection x i = x i (t) of light rays, that is characteristic curves of the covariant
wave equation or the Maxwell or the Dirac equations, onto the spatial sections are
geodesics of the Fermat or optical metric given by
gi j
dso2 = dx i dx j .
V2
In the special case of shallow water waves, the rays are easily seen to be geodesics
of the metric
dx 2 + dy 2
dso2 = .
h
For a linearly shelving beach,
h⊗y y > 0,
the rays are cycloids, and all ray’s strike the shore, i.e. y = 0, orthogonally. For a
quadratically shelving beach,
h ⊗ y2 y > 0,
the rays are circles centred on the shore at y = 0, and again every ray intersects the
shore at right angle. In fact the optical metric in this case is
dx 2 + dy 2
dso2 = ,
y2
which is Poincaré’s metric of constant curvature on the upper half plane. If x is

periodically identified, one obtains the metric induced on a tractrix of revolution in
E3 , sometimes called the Beltrami trumpet (i.e. H 2 /Z).
94 G. W. Gibbons
For an embedded surface of revolution the induced metric is
h i j dx i dx j = dΦ 2 + C 2 (Φ) dα 2 , 0 ◦ α < 2∂,

C (Φ) = x + y = R ,
2 2 2 2
dΦ = d R 2 + dz 2 ,
2
√√
with Gauss curvature K = − CC .
For Beltrami’s trumpet we have Φ ∗ 0 and thus
Φ 1
C(Φ) = a exp − , K =− ,
a a2
and if we denote
Φ
w = aα + ia exp ,
a
then the metric is
a 2 |dw|2
h i j dx i dx j = .
(⇔w)2
Note that (dz/dΦ)2 > 0 and thus the embedding can never reach the conformal
boundary at y = 0. This will be significant later.
The optical time for rays to reach the shore in the second example above is infinite.
This reminds one of the behaviour of event horizons. In fact there is a rather precise
correspondence. The Droste-Schwarzschild metric in isotropic coordinates (setting
G = 1 = c) is
⎧ ⎨
1− m 2
m 4
ds 2 = − ⎧ 2r
⎨ dt 2 + 1 + (dx 2 + dy 2 + dz 2 ),
m 2 2r
1+ 2r
⎩
with r = x 2 + y 2 + z 2 . The isotropic radial coordinate r is related to the Schwarz-
schild radial coordinate R by
m 2
R =r 1+ .
2r
The event horizon is at R = 2m, r = m

2. If we restrict the Schwarzschild metric to
the equatorial plane z = 0 we obtain
⎧ ⎨
1− m 2
m 4
ds = − ⎧
2 2r
⎨ dt 2 + 1 + (dx 2 + dy 2 ).
m 2 2r
1+ 2r
The optical metric is
(1 + 2r )
m 6
dso2 = (dx 2 + dy 2 ),
(1 − 2r )
m 2
and
(r − 2)
m 2
h= r 4.
(r + 2)
m 6
We get the analogue of a black hole: a circularly symmetric island whose edge is at
r = m2 and away from which the beach shelves initially in a quadratic fashion and
ultimately levels out as r ≤ ≡. Since
1 dh 2 4 6
= + − > 0,
h dr r− m
2 r r+ m
2
the beach shelves monotonically.

To obtain a cosmic strings for which the optical metric is a flat cone with deficit
angle ε = 2∂ p
p+1 one needs a submerged mountain with
p
h ⊗ (x 2 + y 2 ) p+1 .
As p ≤ ≡, we get a parabola of revolution and the optical metric approaches that

of an infinitely long cylinder. If p = 1 the mountain is conical, like a submerged
volcano. In physical coordinates x, y, the rays are bent, but one may introduce
coordinates in which they are flat:
2∂
ds 2 = dr̃ 2 + r̃ 2 dα̃ 2 , 0 ◦ α̃ ◦ .
p+1
In these coordinates the rays are straight lines. One could multiply these examples
to cover such things as cosmic strings, moving water and vortices. To take into account
the fact that the Earth is round we replace E2 by S 2
dx 2 + dy 2 ≤ dβ 2 + sin2 β dα 2 ,
which gives Einstein’s static universe in 2 + 1 dimensions. To take into account that
it is rotating, we replace the static, i.e. time-reversal invariant metric, by a stationary
metric
dβ 2 + sin2 β dα 2 ≤ dβ 2 + sin2 β (dα − ω dt)2 .

96 G. W. Gibbons
3 Optics and Maxwell’s Equations
Maxwell’s source-free equations in a medium are
∂B
curl E = − , div B = 0,
∂t
∂D
curl H = + , div D = 0,
∂t
or if (ρi jk = ±1, 0)
1
F = −E i dt ∇ dx i + ρi jk Bi dx j ∇ dx k ,
2
1
G= Hi dt ∇ dx i + ρi jk Di dx j ∇ dx k ,
2
dF = 0 = dG.
In what follows it will be important to realise that these equations hold in any
coordinate system and they do not require the introduction of a spacetime metric.
However to “close the system”, one must relate F to G by means of a “constitutive
equation”. If the medium is assumed to be static and linear, then
Di = ρi j E j , Bi = μi j H j ,
where ρi j is the dielectric permittivity tensor and μi j the magnetic permeability

⎧
tensor. If they are assumed symmetric: ρi j = ρ ji , μi j = μ ji , then E = 21 E i Di +
⎨
Hi Bi may be regarded as the energy density and S = E × H the energy current or
Poynting vector since Maxwell’s equations imply
∂E
div S + = 0.
∂t
“In olden days a glimpse of stocking was thought of as something shocking” and
certainly μi j and ρi j were assume positive definite “but now”, with the advent of
nanotechnology and the construction of metamaterials “anything goes”.
3.1 Left-handed Light
As long ago as 1964, V.G. Vestilago pointed out that isotropic substances with
μi j = μεi j , ρi j = ρεi j and for which
μ < 0, ρ < 0,
give rise to left-handed light moving in a medium with a negative refractive index.
In 2001, Shelby et al. [5] produced this effect for microwave frequencies. In 2002,
Smith et al. [6] appeared to have produced this effect in the laboratory.
Assuming a spacetime dependence proportional to an arbitrary function of
k · x − λt, with λ > 0, one finds
k × E = λ B, k × H = − λ D,
k × E = μλ H, k × H = − ρ λ E.
It is always the case that (E, H, S) form a right-handed orthogonal triad but if both
μ and ρ are negative then (E, H, k) form a left-handed orthogonal triad and so S and
k are anti-parallel rather than parallel as is usually the case. Since the wave vector k
must be continuous across a junction between a conventional medium and an exotic
medium with μ < 0, ρ < 0, this gives rise to backward bending light.
The speed of propagation v = n1 , where n is the refractive index, is given by
λ2 1
v2 = = ;
k 2 μρ
it is natural to take the negative square root to get the refractive index
1
n = −≈ .
μρ
Given a spacetime metric gμν one has a natural way of specifying a constitutive
relation:
G = Θg F,
where Θg denotes the Hodge dual with respect to the spacetime metric g such that
Θg Θg = −1. If
ds 2 = −V 2 (x k ) dt 2 + gi j (x k ) dx i dx j ,
Tamm [7], Skrotskii, [8] and Plebański [9] showed that

⎫
det glm i j
μi j = ρi j = g .
V2
A medium with μi j = ρi j is said to be impedance matched. A similar result holds

for resistivity problems such as that Calderon [10] encountered oil prospecting
∇ · j = 0, E = −∇α, ji = τi j E j ,
98 G. W. Gibbons
⎧ ⎨ 1 ≈
∂i τi j ∂ j α = 0 ⇒ ∈g2 α = ≈ αi g g i j ∂ j α = 0,
g
with
≈
τi j = g gi j , gi j = (det τi j )Φi j .
If
1
τi j = εi j ,
z
we get Poincaré metric on upper half space model of hyperbolic or Lobachevsky

space H 2 :
dx 2 + dy 2 + dz 2
ds 2 = .
z2
The conformal boundary is a perfect conductor.

In physics we may choose either the West Coast signature convention (−, +, +,
+), so that gtt < 0 and gi j is positive definite or the East Coast convention
(+, −, −, −) for which gtt > 0 and gi j is negative definite. By Sylvester’s law
of inertia the signature is locally constant, however running between the East Coast
and the West coast there must be a curve on which the spacetime signature flips (as
originally suggested in a different context by Arthur Eddington in 1922). Clearly,
light passing from Coast to Coast will get bent back.
By Fermat’s principle electromagnetic waves move along geodesics of the optical
metric
dso2 = V −2 gi j dx i dx j ,
but this is invariant under signature change.
3.2 Zermelo-Randers-Finsler Geometry
If time reversal symmetry is broken a stationary metric may be cast in three different
forms [11]:
ds 2 = −U (dt + λi dx i )2 + ξi j dx i dx j
⎬ ⎭
= U −(dt − bi dx i )2 + ai j dx i dx j
U ⎬ ⎭
= −dt 2
+ h i j (dx i
− W i
dt)(dx j
− W j
dt) .
1 − hi j W i W j
Fermat’s principle for light rays now generalises to Zermelo’s problem: minimize the
travel time of a boat moving with fixed speed with respect to a Riemannian metric
h i j in the presence of a “wind” W i .
One may also think of the problem as one of a particular type of Finsler geometry
considered first by Randers with a Finsler function of homogeneous degree one in
i
velocity v i = dx dζ defining a line element ds = Fdζ, given by
F= ai j v i v j + bi v i .
Alternatively, one may think of a charged particle of unit mass and unit charge,
moving on a Riemannian manifold with metric ai j and magnetic field Bi j = ∂i b j −
∂i b j . In General Relativity, this is gravito-magnetism verified recently by the GP-B
satellite experiment.
In the absence of time reversal symmetry there is a magneto-electric effect first
predicted by L. Landau and E.M. Lifshitz in 1956 and exhibited for instance by
Cr 2 O3 :
Bi = μi j H j + δ ji E j , Di = ρi j E j + δi j H j ,
1 1
E = μi j Hi H j + δ E i H j + ρi j E i E j .
2 2
If we take as constitutive relation G = Θg F, then μi j , ρi j and δi j may be read off
from the spacetime metric.
In a moving medium, a typical sound or light wave satisfies

(∂t − W i ∂i )2 − h i j ∂i ∂ j u = 0.
The rays solve the Zermelo problem with wind W i . For sound waves this is known to
explain the curious (and irritating) propagation of traffic noise. The rays behave like
charged magnetic particles, the magnetic field being given by the vertical gradient of
the horizontal wind. Of course a vertical gradient in temperature and hence refractive
index will also provide an anti-mirage effect. This produces a curved metric h i j .
Claude Warnick and I have recently modelled this by a charged particle moving in a
magnetic field on the upper half plane [12].
For black or white holes Zermelo picture is equivalent to the use of Painlevé-
Gullstrand coordinates. Here is a low-tech example involving just a kitchen sink
[13]. The ripples are surface tension ripples (Fig. 1).
100 G. W. Gibbons
Fig. 1 Here is the Mach cone
3.3 Invisibility Cloaks
Designing invisibility cloaks, analogue black holes, etc. using metamaterials and
transformation optics. The basic idea is to start with a metric and read off ρi j and
μi j . The metric could even be flat and obtained by a local diffeomorphism from the flat
metric by which a beam or pencil of parallel straight lines in Cartesian coordinates
are taken to the desired set of light rays in an impedance matched metamaterial
medium. This technique has been much exploited by Pendry, Leonhardt and their
collaborators and followers recently.
As pointed out by Uhlmann and others, similar problems arise in Calderon’s
inverse problem: given a measurement of E and α on the boundary of some domain,
can you determine uniquely the conductivity in the interior or can a reservoir of oil
be invisible to the prospector?
In general one needs anisotropic materials.
To obtain an isotropic metamaterial medium the local diffeomorphisms should
be conformal. The oldest and best known example of this is Maxwell’s fish eye lens
which makes use use of Hipparchus’s stereographic projection. This is the basis of
the Luneburg lens [14].
A variant due to Minano [15] pulls back the round metric on S 2 (β, α) to R 2 (x, y)
using
1 1
1 − sin β p α 1 − sin β p α
x= cos , y= sin ,
cos β p cos β p
to get
r p−1
dso2 = dβ 2 + cos2 β dα 2 = n 2 (dx 2 + dy 2 ), n = 2 p2 .
r2p + 1
To get a black hole start again with the Droste-Schwarzschild metric in isotropic
coordinates
⎧ ⎨
1− m 2
m 4 2
ds = − ⎧
2 2r
⎨ dt 2 + 1 + dx + dy 2 + dz 2 ,
m 2 2r
1+ 2r
m 3 m −1
n =μ=ρ = 1+ 1− .
2r 2r
The original cloak construction by Uhlmann works like this. We consider a spher-
ical shell or solid annulus a < r < 2a in r, β, α space and map it onto the punctured
disc 0 < r̃ < 2a by
r̃ = 2(r − a), β̃ = β, α̃ = α.
The map is the identity: r = r̃ for r > 2a, r̃ > 2a. Now pull back the flat metric
dr̃ 2 + r̃ 2 (dβ̃ 2 + sin2 β̃ dα̃ 2 ) and straight lines in r̃ , β̃ , α̃ space:
ds 2 = 4 dr 2 + 4(r − a)2 (dβ 2 + sin2 β dα 2 ),

2
ρ = μ = diag 2(r − a) sin β, 2 sin β,
2
.
sin β
No light ray (or electric current) enters the solid ball r < a.
The construction just given is strikingly similar to that used in the hole argu-
ment which played a big part in Einstein’s understanding of the concept of general
covariance and his search for covariant field equations in the years from 1913 to 1915
[16].
It is remarkable that what hitherto has been of interest almost exclusively to
philosophers and historians of science is now at the centre of a new technology!
3.4 Hyperbolic Metamaterials and Two-Time Physics
Another possibility are hyperbolic metamaterials for which ρi j is an indefinite matrix.

The dispersion relation for a bi-refringent medium with μi j = εi j is a quartic cone
of two sheets:

k x2 k 2y k z2 λ2 k x2 k 2y k z2 λ2
+ 2+ 2 − 2 + 2 + 2 − 2 = 0,
n 2o no no c n 2e ne no c
with n 2o = ρz , n 2e = ρx = ρ y . Exceptional electromagnetic waves in a uniaxial

medium thus obey
102 G. W. Gibbons

1 ∂2 E 1 ∂2 E 1 ∂2 E ∂2 E
= + + .
c2 ∂t 2 ρ1 ∂z 2 ρ2 ∂x2 ∂ y2
The idea is [17] that dipole moments in some crystals such as δ quartz interact
with lattice vibrations to form phonon-polariton modes called restrahlen bands in
the mid infra-red region for which both ρ1 and ρ2 can become negative. Moreover
because of crystal anisotropy ρ1 and ρ2 change sign at slightly different temperatures.
This would allow an effective two-time physics.
In a model in a layered composite dielectric material
ρm ρd
ρ2 = n m + (1 − n m )ρd , ρ1 = ,
(1 − n m )ρm + n m ρd
where the subscripts d and m stand for dielectric and metal respectively and ρm is
frequency dependent and can be negative; n m is the volume fraction of metal. In a
simple Drude model
λ2p
ρm = 1 − ,
λ2 + iλξ
ξ
with λp being small. If n m ⊥ 1 we have
n m λ2p
ρ2 ⇐ ρd − , ρ1 ⇐ ρd .
λ2 + iλξ
4 Chiral Nematics
Rather than consider artificial impedance matched or hyperbolic metamaterials, we

may consider realistic substances such as chiral nematics in their helical phase. Up
to a divergence the Frank-Oseen free energy is

1
F= |∈ q n|2 − ζ(n · n − 1) d3 x,
2
q
∈i n j = ∂ j n j + q ρi jk n k ,
where ∈ q is Euclidean metric preserving connection with torsion. The free energy
q
density would vanish if n were covariantly constant with respect to ∈ q , i.e. ∈i n j = 0.
But rather like an anti-ferromagnet it is frustrated since
q q q q
(∈i ∈ j − ∈ j ∈i )n k ≥= 0.
The substance may adopt a compromise configuration called the helical phase
which satisfies the second order equations but not the first order Bogomolnyi type
equation
n = (cos( pz), sin( pz), 0).
Optics in a nematic liquid crystal is governed by Fermat’s principle using the Joets-
Ribotta metric
dso2 = n 2e dx2 + (n 2o − n 2e )(n · dx)2 ,
where n o is the refractive index of the ordinary ray and n e that of the extra-ordinary
ray.
Introducing tree one-forms with Maurer-Cartan relations
ζ1 = cos( pz) dx + sin( pz) dy, dζ1 = ζ3 ∇ ζ2 ,

ζ2 = cos( pz) dx − sin( pz) dy, dζ2 = ζ3 ∇ ζ1 ,
ζ3 = p dz, dζ3 = 0,
we find the Joets-Ribotta metric to be
n 2e 3 2
dso2 = n 2o (ζ1 )2 + n 2e (ζ2 )2 + (ζ ) .
p2
This is a left-invariant metric on Ẽ(2), the universal cover of the two-dimensional

Euclidean group E(2) whose Lie algebra e(2) is of Type V I I0 in Bianchi’s classifi-
cation.
Thus the helical phase of chiral nematic crystals gives rise to a static Bianchi
V I I0 cosmology:
n 2e 3 2
ds 2 = −dt 2 + n 2o (ζ1 )2 + n 2e (ζ2 )2 + (ζ ) ,
p2
and one may, and we did, use all the standard tools of general relativistic cosmology
to describe its optical and electromagnetic properties, including solving Maxwell’s
equations, applying the Floquet-Bloch theorem and the associated Mathieu-Hill
equation.
5 Gravitational Kinks
The topology of a Lorentzian metric may be (partially) captured by a direction field

n i . Given a Riemannian metric giRj , and a unit direction field n i such that giRj n i n j = 1,
we may construct a Lorentzian metric giLj via
104 G. W. Gibbons
1 ij ij 1
giLj = giRj − ni n j , gL = g R − ni ni , n i = giRj n j .
sin2 δ cos2 δ
Conversely, given giLj and giRj we may reconstruct n i up to a sign. Fixing the sign
amounts to fixing a time orientation. In what follows we will choose giRj to be the
usual flat Euclidean metric:
ij 1
ds L2 = g L dx i dx j = dx2 − (n · dx)2 .
cos2 δ
Given a closed surface enclosing a domain D, Finkelstein and Misner quantified
the notion of tumbling light cones– the light cone tips over on = ∂ D – by introduc-
ing a kink number which counts how many times the light cone tips over on . The
outward unit normal ν gives a 2-dimensional cross section of the four-dimensional
bundle S() of unit 3-vectors over . In the orientable case, the director field gives
another 2-dimensional cross section of S(). The kink number, kink(, g L ), is
number of intersections of these two sections with attention paid to signs. In the
non-orientable case, one considers the bundle of directions. If the Lorentzian metric
is non-singular, we have
κ (D) = kink(∂ D, g L ).
For planar domains kink(∂ D, g L ) is the obvious winding number.

Disclination line is defined by
y
n = (cos(sα), sin(sα), 0), α = arctan ,
x
where s ∈ Z ∪ Z + 21 . If s is half integral, then we just have a direction field, not a

vector field.
n · dx = cos((s − 1)α) dr + sin((s − 1)α)r dα.

∂
For δ = 2 we get

ds L2 = giLj dx i dx j = − cos(2(s − 1)α) dr 2 − r 2 dα 2 ) − 2 sin(2(s − 1)α)r dr dα.
Moving around a circle r = constant, the radial coordinate is timelike and the
angular coordinate spacelike or vice versa depending upon the sign of cos(2(s −1)α)
(tumbling light cones). The metric components giLj are finite and det giLj = −r 2 ⇒
metric non-singular if r > 0.
5.1 Bloch Walls
If parity symmetry holds then a typical free energy functional takes the form

1 ⎧ ⎨
F[M] = dx δi j ∂i M · ∂ j M + φi j Mi M j .
2
In the uniaxial case with the easy direction along the third direction δi j = diag(δ1 , δ1 ,
δ2 ), φi j = diag(φ, φ, 0). For a domain wall separating a region x ⊥ −1 and with
M pointing along the positive 3rd direction, from the region x +1 where it points
along the negative 3rd direction, we have
M = M (0, sin β (x), cos β (x)), M = constant,
δ1
and find that β must satisfy the quadrantal pendulum equation, l = φ ,
1
β2 − sin2 β = constant √ .
l2
If we impose the boundary condition that β ≤ 0 as x ≤ −≡ and β ≤ ∂ as
x ≤ +≡, then constant √ = 0 and
x
cos β = − tanh .
l
∂
The Lorentzian metric (if δ = 2) is
⎧ ⎨
ds 2 = giLj dx i dx j = dx 2 + cos(2β ) dy 2 − dz 2 − 2 sin 2β dzdy.
This closely resembles our previous examples and clearly exhibits the phenomenon
of tumbling light cones. We note, en passant, that in principle the tensor δi j could
itself vary with position. If so, we might interpret it in terms of an effective metric
gi j with inverse g i j and g = det gi j obeying
≈
δi j = g gi j .
5.2 Liquid Crystal Droplets
The normal νi = ∂i S to the surface S = constant of a droplet of anisotropic nematic

phase inside a domain D with unit outward normal ν surrounded by an isotropic
phase satisfies the constant angle condition
106 G. W. Gibbons
n · ν = cos δ = constant.
That is
1 ij ij
ν·ν− (ν · n)(ν · n) = 0 = g L νi ν j = g L ∂i S ∂ j S.
cos2 δ
The surface ∂ D of the droplet ∂ D is a null-hypersurface or wave surface (a solution
of the zero-rest-mass Hamilton-Jacobi equation).
Taking the z-coordinate as time so time runs vertically upwards and making the
ansatz
z
S= + W (x, y), ∈W · ∈W = 1.
sin δ
Simple solutions of this Eikonal equation are given by sandpiles with ∂2 −δ the angle
of repose.
These describe Bitter domains in a ferromagnetic film with n = |M|M
with normal
ν and boundary condition M · ν = 0:
∇ · M = 0, |M| = constant;
∇·n =0 ⇒ n x = ∂ y ψ, n y = −∂x ψ |∈ψ| = 1.
The axisymmetric solution is the spiral wave surface swept out by the involute of
a circle, a helical developable:
⎛ ⎛ ⎞⎞
z r 2 r 2
S=± ± a ⎝ 2 − 1 − arctan ⎝ 2 − 1⎠⎠ ± aα.
sin δ a a
5.3 Helical Phase
We make the ansatz
S = F(z) + x cos β + y sin β,
where F(z) solves the quadrantal pendulum equation

2
dF
cos (β − pz) − cos δ = cos δ
2 2
.
dz
Thus

1
F= dz cos2 (β − pz) − cos2 δ.
cos δ
The surface is ruled by horizontal straight lines making a constant angle β with
the x-axis and is bounded by | pz − (β + n∂ )| < δ, n ∈ Z. In other words it
is horizontal cylinder or tube. The angle that the director n makes with the fixed
direction (cos β, sin β, 0) cannot be less than δ.
6 Graphene
The hexagonal Graphene “lattice” in x-space has a hexagonal Brillouin zone in the
dual p-space and is the sum of two triangular (true) lattices, A and B in x-space. Each
lattice has a Fermi surface in p-space and these two Fermi surfaces, governing the
conduction and valence bands, touch in two conical Dirac points inside a Brillouin
zone. Thus the dispersion relation for small p is
E = ±|p|.
Low energy excitations are governed by
E = σ · p ,
where the two-component has two pseudo-spin states.

But this is the massless Dirac equation! (cf. [18])
On a curved graphene sheet it becomes the Dirac equation on a curved surface
⊂ E3 in Euclidean 3-space with metric
ds 2 = −dt 2 + h i j dx i dx j , i, j = 1, 2,
where h i j is the induced metric.

Since the massless Dirac equation is conformally invariant we may think of this
metric on R × as the optical metric of a static metric with gtt ≥ = constant.
If is a Beltrami trumpet with metric of constant negative curvature, we have
the optical metric of identified Rindler spacetime. This is also near horizon optical
geometry of a general 2-dimensional black hole. Unfortunately, we cannot find an
isometric embedding of H 2 /Z into E3 all the way down to y = 0, the horizon.
This is a general problem: a global theorem of Hilbert forbids isometric embed-
dings of complete surface of constant negative curvature into Euclidean space E3 .
More generally, we may consider a BTZ black hole [19]:
108 G. W. Gibbons
2
dr 2 J
2
ds BT Z = −π dt + 2
+ r dα − 2 dt ,
2
π 2r
r2 J2
π(r ) = − M + .
l2 4r 2
Zermelo metric is
dr 2 r 2 2
h i j dx i dx j = + dα ,
π2 π
Wind
J
W i ∂i = ∂α .
2r 2
In the near horizon limit h i j is of Beltrami trumpet form.
The massless Dirac equation in the Zermelo frame is

1 C√ 1 1
ξ 1 ∂Φ + + ξ 2 ∂α + ξ 0 (∂t + W ∂α ) + ξ 0 ξ 1 ξ 2 C W √ = 0,
2C C 4
ξ 0 = iτ2 , ξ 1 = τ1 , ξ 2 = τ3 . ξ 0 ξ 1 ξ 2 = 1,
and we get a position dependent “mass-like” term and a connection term. If ⊗

e−iλt+imα we have that
−ie A0 = imW, ⇒ e A0 = −mW.
A stationary Zermelo metric induces in the Dirac equation an effective, position

dependent radial electric field.
We could have done this calculation in the Randers frame. The detailed form of the
Randers metric is considerably more complicated. The embedding is qualitatively
similar, but different.
More interestingly, because now the roles of t and α have essentially been inter-
changed, we now find that there is an effective magnetic vector potential in the Dirac
equation. Therefore, the magnetic vector potential in the Randers frame appears as
an electric potential in the Zermelo one.
Since the Zermelo and Randers frames are in relative motion, this is just a man-
ifestation of the fact that under boosts magnetic and electric fields transform into
themselves. In either case, these effects could be mimicked by applying external
electric (Zermelo) or magnetic (Randers) fields to the two different graphene sheets.
6.1 Cold Atoms
Instead of graphene one may consider, and people have discussed, metrics in the
context of cold atoms in Bose-Einstein condensates [20].
7 Conclusion and Propects
In this talk I have described on some areas of non-gravitational physics where ana-
logues of basic ideas in general relativity come into play. They include
1. Dynamic Casimir effect
2. Water and sound waves
3. Cloaking and other devices using metamaterials
4. Nematic liquid crystals
5. Graphene
Other areas not covered include
1. Bose-Einstein condensate
2. Dirac metals
3. Smectic and blue phases in liquid crystals
References
1. Wilson, C.M., Johansson, G., Pourkabirian, A., et al.: Observation of the dynamical Casimir
effect in a superconducting circuit. Nature 479, 376 (2011). doi:10.1038/nature10561
2. Moore, G.T.: Quantum theory of the electromagnetic field in a variable-length one-dimensional
cavity. J. Math. Phys. 11, 2679 (1970). doi:10.1063/1.1665432
3. Fulling, S.A., Davies, P.C.W.: Radiation from a moving mirror in two dimensional space-
time—Conformal anomaly. Proc. R. Soc. London Ser. A 348, 393 (1976). doi:10.1098/rspa.
1976.0045
4. Volovik, G.E.: The Universe in a Helium Droplet, International Series of Monographs on
Physics, vol. 117. Oxford University Press, Oxford, New York (2003)
5. Shelby, R.A., Smith, D.R., Schutz, S.: Experimental verification of a negative index of refrac-
tion. Science 292, 77 (2001). doi:10.1126/science.1058847
6. Smith, D.R., Schurig, D., Pendry, J.B.: Negative refraction of modulated electromagnetic
waves. Appl. Phys. Lett. 81, 2713 (2002). doi:10.1063/1.1512828
7. Tamm, I.E.: Èlektrodinamika anizotropnoj sredy v special’noj teorii otnositel’nosti. Zh. Rus.
Fiz.-Khim. Obshchestva, Otd. Fiz 56, 248 (1924).
8. Skrotskii, G.V.: On the influence of gravity on the light propagation. Dokl. Akad. Nauk. SSSR
114, 73 (1957). Soviet Physics Doklady 2, 226 (1957).
9. Plebański, J.: Electromagnetic waves in gravitational fields. Phys. Rev. 118, 1396 (1960).
doi:10.1103/PhysRev.118.1396
10. Calderón, A.P.: On an inverse boundary value problem. In: Meyer, W.H., Raupp, M.A. (eds.)
Seminar in Numerical Analysis and its Applications to Continuum Physics, pp. 65–73. So-
ciedade Brasileira de Matemática, Rio de Janeiro (1980)
110 G. W. Gibbons
11. Gibbons, G.W., Herdeiro, C.A.R., Warnick, C.M., Werner, M.C.: Stationary metrics and optical
Zermelo-Randers-Finsler geometry. Phys. Rev. D 79, 044022 (2009). doi:10.1103/PhysRevD.
79.044022
12. Gibbons, G.W., Warnick, C.M.: The geometry of sound rays in a wind. Contemp. Phys. 52,
197 (2011). doi:10.1080/00107514.2011.563515
13. Jannes, G., Rousseaux, G.: The circular jump as a hydrodynamic white hole. ArXiv e-prints.
arXiv:1203.6505 [gr-qc] (2012)
14. Luneberg, R.K.: Mathematical Theory of Optics. Brown University, Providence, RI (1944)
15. Miñano, J.C.: Perfect imaging in a homogeneous threedimensional region. Opt. Exp. 14, 9627
(2006). doi:10.1364/OE.14.009627
16. Norton, J.D.: The hole argument. In: Zalta, E.N. (ed.) The Stanford Encyclopedia of Philos-
ophy, Fall 2011 edn. Stanford University, Stanford (2011) http://plato.stanford.edu/archives/
fall2011/entries/spacetime
17. Smolyaninov, I.I.: Critical opalescence in hyperbolic metamaterials. J. Opt. 13, 125101 (2011).
doi:10.1088/2040-8978/13/12/125101
18. Semenoff, G.W.: Condensed-matter simulation of a three-dimensional anomaly. Phys. Rev.
Lett. 53, 2449 (1984). doi:10.1103/PhysRevLett.53.2449
19. Cvetič, M., Gibbons, G.W.: Graphene and the zermelo optical metric of the BTZ black hole.
Ann. Phys. (N.Y.) 327, 2617 (2012). doi:10.1016/j.aop.2012.05.013
20. Boada, O., Celi, A., Latorre, J.I., Lewenstein, M.: Dirac equation for cold atoms in artificial
curved spacetimes. New J. Phys. 13, 035002 (2011). doi:10.1088/1367-2630/13/3/035002
The General Relativistic Two Body Problem
and the Effective One Body Formalism
Thibault Damour
Abstract A new analytical approach to the motion and radiation of (comparable

mass) binary systems has been introduced in 1999 under the name of Effective
One Body (EOB) formalism. We review the basic elements of this formalism, and
discuss some of its recent developments. Several recent comparisons between EOB
predictions and Numerical Relativity (NR) simulations have shown the aptitude of
the EOB formalism to provide accurate descriptions of the dynamics and radiation
of various binary systems (comprising black holes or neutron stars) in regimes that
are inaccessible to other analytical approaches (such as the last orbits and the merger
of comparable mass black holes). In synergy with NR simulations, post-Newtonian
(PN) theory and Gravitational Self-Force (GSF) computations, the EOB formalism
is likely to provide an efficient way of computing the very many accurate template
waveforms that are needed for Gravitational Wave (GW) data analysis purposes.
1 Introduction
The general relativistic N -body problem has been investigated from the early days
of Einstein’s gravitation theory (and even earlier, because it was already tack-
led by Johannes Droste within the framework of the 1913 Einstein-Grossmann
“Entwurf” theory). Here, we shall focus on the general relativistic two-body problem.
This problem has been the subject of many investigations within the post-Newtonian
(PN) formalism, since the pioneering works of Einstein (1915; when m 1 ≈ m 2 ),
Lorentz and Droste (1917), Levi-Civita (1937) and Einstein, Infeld and Hoffmann
(1938). [See, e.g., [1] for a review and references to the early literature.] For many
years, the first post-Newtonian (1PN) approximation (i.e. the inclusion of the leading-
order relativistic corrections, proportional to (v/c)2 or GM/c2 r , to the Newtonian
T. Damour (B)
Institut des Hautes Études Scientifiques, 35 route de Chartres, 91440 Bures-sur-Yvette, France
e-mail: damour@ihes.fr
112 T. Damour
equations of motion) appeared as being accurate enough for applying Einstein’s

theory to known binary systems. The situation changed in the mid 1970s with the
discovery of the Hulse-Taylor binary pulsar PSR 1913+16. The need to compare the
accurate observations of this system (by Taylor and collaborators) to the predictions
of Einstein’s theory motivated the development of improved relativistic theories of
binary systems, applicable to strongly self-gravitating bodies, and including terms
up to the 2.5PN approximation (i.e. O[(v/c)5 ] beyond Newton). [See [2] and ref-
erences therein.] The situation has again changed recently with the development of
interferometric gravitational wave (GW) detectors, and the prospect of detecting the
GW’s emitted during the last orbits and the coalescence of binary systems made
of black holes or neutron stars. The latter prospect motivated the development (or
improvement) of several different methods of computing the motion and radiation
of binary systems.
First, this motivated pushing PN calculations of the dynamics of binary systems
to the 3PN level [3–7], with inclusion of 3.5PN radiation-reaction terms [8–10]. Sec-
ond, this motivated the development of new, accurate GW generation formalisms,
notably the Blanchet-Damour-Iyer (matched) “multipolar post-Minkowskian” for-
malism [11–16] and the “direct integration of the relaxed Einstein’s equations” for-
malism of Will and collaborators [17–19], which extended previous work by Epstein
and Wagoner [20] and Thorne [21]. These GW generation formalisms allowed one
to compute emitted gravitational waveforms with an unprecedented PN accuracy.1
After the 1PN correction to the waveform [12, 20, 22], there is a 1.5PN “tail”
(i.e. hereditary) correction [15, 23, 24], then a “direct” 2PN term [17, 25, 26], fol-
lowed by higher-order corrections [27–35]. [See [36] for a detailed account and
more references.] Parallely to these improved PN computations of the GW emission
of comparable-mass systems (with m 1 ⊗ m 2 ), other authors developed the analyti-
cal theory of the GW emission of extreme mass-ratio systems (with m 1 ≈ m 2 ); see
Refs. [37–40] and the review of Sasaki and Tagoshi [41].
Some of the PN calculations of the dynamics, and/or GW emission, of comparable-
mass systems have been recently (re)done (e.g. the 3PN dynamics [42]) by using a
somewhat different formalism, dubbed “effective field theory” [43]. Let us, how-
ever, note that most of the technical aspects of the effective-field-theory approach
had already been introduced and used before. For instance: (i) Ref. [44] discussed
the (Fokker) two-body effective action due to the exchange of a linear field (of spin
s = 0, 1 and 2) ; (ii) Ref. [45] explicitly discussed the representation (and computa-
tion) of the (nonlinear) effective two-body action in terms of Feynman-like diagrams
(made of concatenated propagators and vertices); (iii) The appendix A of Ref. [46]
discussed finite-size effects in terms of nonminimal worldline couplings in the effec-
tive action; (iv) The (quantum field theory) technique of dimensional regularization
(together with a diagrammatic analysis of ultraviolet divergences) was crucially used
1 For gravitational waveforms, one conventionally defines the PN accuracy as the fractional PN
accuracy with respect to the leading-order, O(c−5 ), quadrupolar emission. E.g., a 1PN-accurate
waveform retains next-to-leading order terms, i.e. terms smaller than the quadrupolar waveform by
a factor O(c−2 ).
The General Relativistic Two Body Problem and the Effective One Body Formalism 113
to derive the 3PN dynamics in Refs. [5, 6], and 3PN radiation in Ref. [31]; and (v)
The exponential parametrization of the metric (which suppresses the leading-order
gravitational cubic vertex) had been introduced in Ref. [12] and then standardly used
in many PN works. It is, however, possible that the more systematic (and automa-
tized) implementation of such diagrammatic methods, together with the tapping of
standard techniques for computing Feynman graphs (as exemplified in [42]) may
allow one to be more efficient in computing higher-order processes, or, at least, to
open new ways of understanding them (see, in this respect, Ref. [47]).
Separately from these purely analytical approaches to the motion and radiation
of binary systems, which have been developed since the early days of Einstein’s
theory, Numerical Relativity (NR) simulations of Einstein’s equations have relatively
recently (2005) succeeded (after more than thirty years of developmental progress)
to stably evolve binary systems made of comparable mass black holes [48–51].
This has led to an explosion of works exploring many different aspects of strong-
field dynamics in General Relativity, such as spin effects, recoil, relaxation of the
deformed horizon formed during the coalescence of two black holes to a stationary
Kerr black hole, high-velocity encounters, etc.; see [52] for a review. In addition,
recently developed codes now allow one to accurately study the orbital dynamics,
and the coalescence of binary neutron stars. Much physics remains to be explored
in these systems, especially during and after the merger of the neutron stars (which
involves a much more complex physics than the pure-gravity merger of two black
holes).
Recently, a new source of information on the general relativistic two-body prob-
lem has opened: gravitational self-force (GSF) theory. This approach goes one step
beyond the test-particle approximation (already used by Einstein in 1915) by taking
into account self-field effects that modify the leading-order geodetic motion of a
small mass m 1 moving in the background geometry generated by a large mass m 2 .
After some ground work (notably by DeWitt and Brehme) in the 1960s, GSF the-
ory has recently undergone rapid developments (mixing theoretical and numerical
methods) and can now yield numerical results that yield access to new information
on strong-field dynamics in the extreme mass-ratio limit m 1 ≈ m 2 . See Ref. [53]
for a review (see also the chapter by L. Barack in this volume).
Each of the approaches to the two-body problem mentioned so far, PN theory,
NR simulations and GSF theory, have their advantages and their drawbacks. It has
become recently clear that the best way to meet the challenge of accurately computing
the gravitational waveforms (depending on several continuous parameters) that are
needed for a successful detection and data analysis of GW signals in the upcoming
LIGO/Virgo/GEO/. . . network of GW detectors is to combine knowledge from all
the available approximation methods: PN, NR and GSF. Several ways of doing so
are a priori possible. For instance, one could try to directly combine PN-computed
waveforms (approximately valid for large enough separations, say r 10 G(m 1 +
m 2 )/c2 ) with NR waveforms (computed with initial separations r0 > 10 G(m 1 +
m 2 )/c2 and evolved up to merger and ringdown). However, this method still requires
too much computational time, and is likely to lead to waveforms of rather poor
accuracy, see, e.g., [54–56].
114 T. Damour
On the other hand, five years before NR succeeded in simulating the late inspiral
and the coalescence of binary black holes, a new approach to the two-body problem
was proposed: the Effective One Body (EOB) formalism [57–60]. The basic aim
of the EOB formalism is to provide an analytical description of both the motion
and the radiation of coalescing binary systems over the entire merger process, from
the early inspiral, right through the plunge, merger and final ringdown. As early
as in 2000 [58], this method made several quantitative and qualitative predictions
concerning the dynamics of the coalescence, and the corresponding GW radiation,
notably: (i) a blurred transition from inspiral to a ‘plunge’ that is just a smooth
continuation of the inspiral, (ii) a sharp transition, around the merger of the black
holes, between a continued inspiral and a ring-down signal, and (iii) estimates of
the radiated energy and of the spin of the final black hole. In addition, the effects of
the individual spins of the black holes were investigated within the EOB [60, 61]
and were shown to lead to a larger energy release for spins parallel to the orbital
angular momentum, and to a dimensionless rotation parameter J/E 2 always smaller
than unity at the end of the inspiral (so that a Kerr black hole can form right after
the inspiral phase). All those predictions have been broadly confirmed by the results
of the recent numerical simulations performed by several independent groups (for
a review of numerical relativity results and references see [52]). Note that, in spite
of the high computer power used in NR simulations, the calculation, checking and
processing of one sufficiently long waveform (corresponding to specific values of
the many continuous parameters describing the two arbitrary masses, the initial spin
vectors, and other initial data) takes on the order of one month. This is a very strong
argument for developing analytical models of waveforms.
2 EOB Description of the Conservative Dynamics of Two Body

Systems
Before reviewing some of the technical aspects of the EOB method, let us indicate
the historical roots of this method. First, we note that the EOB approach comprises
three, rather separate, ingredients:
1. a description of the conservative (Hamiltonian) part of the dynamics of two
bodies;
2. an expression for the radiation-reaction part of the dynamics;
3. a description of the GW waveform emitted by a coalescing binary system.
For each one of these ingredients, the essential inputs that are used in EOB works
are high-order post-Newtonian (PN) expanded results which have been obtained by
many years of work, by many researchers (see references above). However, one of
the key ideas in the EOB philosophy is to avoid using PN results in their original
“Taylor-expanded” form (i.e. c0 + c1 v/c + c2 v 2 /c2 + c3 v 3 /c3 + · · · + cn v n /cn ),
but to use them instead in some resummed form (i.e. some non-polynomial function
of v/c, defined so as to incorporate some of the expected non-perturbative features
of the exact result). The basic ideas and techniques for resumming each ingredient
of the EOB are different and have different historical roots.
Concerning the first ingredient, i.e. the EOB Hamiltonian, it was inspired by an
approach to electromagnetically interacting quantum two-body systems introduced
by Brézin et al. [62].
The resummation of the second ingredient, i.e. the EOB radiation-reaction force
F , was initially inspired by the Padé resummation of the flux function introduced
by Damour et al. [63]. More recently, a new and more sophisticated resummation
technique for the radiation reaction force F has been introduced by Damour and
Nagar [64, 65].
As for the third ingredient, i.e. the EOB description of the waveform emitted by a
coalescing black hole binary, it was mainly inspired by the work of Davis et al. [66]
which discovered the transition between the plunge signal and a ringing tail when a
particle falls into a black hole. [Additional motivation for the EOB treatment of the
transition from plunge to ring-down came from work on the, so-called, “close limit
approximation” [67].] In addition, a very efficient resummation of the waveform has
been introduced by Damour et al. [64, 68, 69]. It will be discussed in detail below.
Within the usual PN formalism, the conservative dynamics of a two-body sys-
tem is currently fully known up to the 3PN level [3–7, 42] (see below for the par-
tial knowledge beyond the 3PN level). Going to the center of mass of the system
( p1 + p2 = 0), the 3PN-accurate Hamiltonian (in Arnowitt-Deser-Misner-type co-
ordinates) describing the relative motion, q = q 1 − q 2 , p = p1 = − p2 , has the
structure
1 1 1
relative
H3PN (q, p) = H0 (q, p) + H2 (q, p) + 4 H4 (q, p) + 6 H6 (q, p) , (1)
c2 c c
where
1 2 G Mμ
H0 (q, p) = p − , (2)
2μ |q|
with
M ◦ m 1 + m 2 and μ ◦ m 1 m 2 /M , (3)
corresponds to the Newtonian approximation to the relative motion, while H2 de-

scribes 1PN corrections, H4 2PN ones and H6 3PN ones. In terms of the rescaled
variables q √ ◦ q/G M, p√ ◦ p/μ, the explicit form (after dropping the primes for
◦ H/μ reads [5, 70, 71]
readability) of the 3PN-accurate rescaled Hamiltonian H
116 T. Damour
2
HN (q, p) = p − 1 , (4)
2 q

1PN (q, p) = 1 (3ε − 1)( p2 )2 − 1 (3 + ε) p2 + ε(n · p)2 1 + 1 ,
H (5)
8 2 q 2q 2
2PN (q, p) = 1 (1 − 5ε + 5ε 2 )( p2 )3
H
16
1 1
+ (5 − 20ε − 3ε 2 )( p2 )2 − 2ε 2 (n · p)2 p2 − 3ε 2 (n · p)4
8 q
1 1 1 1
+ (5 + 8ε) p2 + 3ε(n · p)2 2 − (1 + 3ε) 3 , (6)
2 q 4 q
3PN (q, p) = 1 (−5 + 35ε − 70ε 2 + 35ε 3 )( p2 )4
H
128
1
+ (−7 + 42ε − 53ε 2 − 5ε 3 )( p2 )3 + (2 − 3ε)ε 2 (n · p)2 ( p2 )2
16
1
+ 3(1 − ε)ε 2 (n · p)4 p2 − 5ε 3 (n · p)6
q

1 1
+ (−27 + 136ε + 109ε 2 )( p2 )2 + (17 + 30ε)ε(n · p)2 p2
16 16
⎜
1 1
+ (5 + 43ε)ε(n · p)4 2
12 q
⎟ ⎩ ⎫ ⎜
25 1 2 335 23 2 2
+ − + γ − ε− ε p
8 64 48 8
⎩ ⎫ ⎬
85 3 2 7 1
+ − − γ − ε ε(n · p) 2
16 64 4 q3
⎩ ⎫ ⎜
1 109 21 2 1
+ + − γ ε 4. (7)
8 12 32 q
In these formulas ε denotes the symmetric mass ratio:

μ m1 m2
ε◦ ◦ . (8)
M (m 1 + m 2 )2
The dimensionless parameter ε varies between 0 (extreme mass ratio case) and 41
(equal mass case) and plays the rôle of a deformation parameter away from the
test-mass limit.
It is well known that, at the Newtonian approximation, H0 (q, p) can be thought of
as describing a ‘test particle’ of mass μ orbiting around an ‘external mass’ G M. The
EOB approach is a general relativistic generalization of this fact. It consists in looking
for an ‘effective external spacetime geometry’ gμεeff (x λ ; G M, ε) such that the geodesic
dynamics of a ‘test particle’ of mass μ within gμε eff (x λ ; G M, ε) is equivalent (when
expanded in powers of 1/c2 ) to the original, relative PN-expanded dynamics (1).

Let us explain the idea, proposed in [57], for establishing a ‘dictionary’ between
the real relative-motion dynamics, (1), and the dynamics of an ‘effective’ particle
of mass μ moving in gμε eff (x λ , G M, ε). The idea consists in ‘thinking quantum me-
chanically’.2 Instead of thinking in terms of a classical Hamiltonian, H (q, p) (such

relative , Eq. (1)), and of its classical bound orbits, we can think in terms of the
as H3PN
quantized energy levels E(n, ν) of the quantum bound states of the Hamiltonian
operator H (q̂, p̂). These energy levels will depend on two (integer valued) quantum
numbers n and ν. Here (for a spherically symmetric interaction, as appropriate to
H relative ), ν parametrizes the total orbital angular momentum (L 2 = ν(ν + 1) 2 ),
while n represents the ‘principal quantum number’ n = ν + nr + 1, where nr (the
‘radial quantum number’) denotes the number of nodes in the radial wave function.
The third ‘magnetic quantum number’ m (with −ν ∗ m ∗ ν) does not enter the
energy levels because of the spherical symmetry of the two-body interaction (in the
center of mass frame). For instance, the non-relativistic Newton interaction, Eq. (2),
gives rise to the well-known result
⎩ ⎫
1 G Mμ 2
E 0 (n, ν) = − μ , (9)
2 n
which depends only on n (this is the famous Coulomb degeneracy). When considering
the PN corrections to H0 , as in Eq. (1), one gets a more complicated expression of
the form

1 λ2 λ 2 ⎭ c11 c20 λ 4 ⎭ c13 c22 c31 c40
relative
E 3PN (n, ν) = − μ 2 1 + 2 + 2 + 4 + 2 2 + 3 + 4
2 n c nν n c nν3 n ν n ν n
⎭ ⎜
λ 6 c15 c60
+ 6 + ··· + 6 , (10)
c nν5 n
where we have set λ ◦ G Mμ/ = G m 1 m 2 /, and where we consider, for

simplicity, the (quasi-classical) limit where n and ν are large numbers. The 2PN-
accurate version of Eq. (10) had been derived by Damour and Schäfer [72] as early
as 1988 while its 3PN-accurate version was derived by Damour, Jaranowski and
Schäfer in 1999 [70]. The dimensionless coefficients c pq are functions of the sym-
metric mass ratio ε ◦ μ/M, for instance c40 = 18 (145 − 15ε + ε 2 ). In classical
mechanics (i.e. for large n and ν), it is called the ‘Delaunay Hamiltonian’, i.e. the
Hamiltonian expressed in terms of the action variables3 J = ν = 2γ 1
pΦ dΦ, and
N = n = Ir + J , with Ir = 2γ pr dr .
1
The energy levels (10) encode, in a gauge-invariant way, the 3PN-accurate relative
dynamics of a ‘real’ binary. Let us now consider an auxiliary problem: the ‘effective’
dynamics of one body, of mass μ, following (modulo the Q term discussed below) a
2 This is related to an idea emphasized many times by John Archibald Wheeler: quantum mechanics
can often help us in going to the essence of classical mechanics.
3 We consider, for simplicity, ‘equatorial’ motions with m = ν, i.e., classically, α = γ .
2
118 T. Damour
geodesic in some ε-dependent ‘effective external’ (spherically symmetric) metric4

eff
gμε d x μ d x ε = −A(R; ε) c2 dT 2 + B(R; ε) d R 2 + R 2 (dα 2 + sin2 α dΦ 2 ) . (11)
Here, the a priori unknown metric functions A(R; ε) and B(R; ε) will be constructed
in the form of expansions in GM/c2 R:
⎩ ⎫2 ⎩ ⎫3 ⎩ ⎫4
GM GM GM GM
A(R; ε) = 1 +
a1 +
a2 +
a3 +
a4 + ··· ;
c2 R c2 R c2 R c2 R
⎩ ⎫2 ⎩ ⎫3
GM GM GM
B(R; ε) = 1 +
b1 2 +
b2 + b3 + ··· , (12)
c R c2 R c2 R
where the dimensionless coefficients an ,

bn depend on ε. From the Newtonian limit,
a1 = −2. In addition, as ε can be viewed as a deformation
it is clear that we should set
parameter away from the test-mass limit, we require that the effective metric (11)
tend to the Schwarzschild metric (of mass M) as ε ⇔ 0, i.e. that
A(R; ε = 0) = 1 − 2G M/c2 R = B −1 (R; ε = 0) .
Let us now require that the dynamics of the “one body” μ within the effective
eff be described by an “effective” mass-shell condition of the form
metric gμε
με
geff pμeff pεeff + μ2 c2 + Q( pμeff ) = 0 ,
where Q( p) is (at least) quartic in p. Then by solving (by separation of variables)

the corresponding ‘effective’ Hamilton-Jacobi equation
⎩ ⎫
με ∂ Seff ∂ Seff ∂S
geff + μ2 c2 + Q = 0,
∂xμ ∂xε ∂xμ
Seff = −Eeff t + Jeff Φ + Seff (R) , (13)
one can straightforwardly compute (in the quasi-classical, large quantum numbers
limit) the effective Delaunay Hamiltonian Eeff (Neff , Jeff ), with Neff = n eff , Jeff =
νeff (where Neff = Jeff + I Reff , with I Reff = 2γ
1
R dR, PR = dSeff (R)/dR).
p eff eff
This yields a result of the form
4 It is convenient to write the ‘effective metric’ in Schwarzschild-like coordinates. Note that the
effective radial coordinate R differs from the two-body ADM-coordinate relative distance R ADM =
|q|. The transformation between the two coordinate systems has been determined in Refs. [57, 59].

1 λ2 λ2 eff
c11 eff
c20
Eeff (n eff , νeff ) = μc − μ 2 1 + 2
2
+ 2
2 n eff c n eff νeff n eff

λ4 eff
c13 eff
c22 eff
c31 eff
c40
+ 4 + 2 2 + 3 + 4
c n eff ν3eff n eff νeff n eff νeff n eff
⎜
λ6 eff
c15 eff
c60
+ 6 + . . . + , (14)
c n eff ν5eff n 6eff
where the dimensionless coefficients ceffpq are now functions of the unknown coeffi-
an ,
cients bn entering the looked for ‘external’ metric coefficients (12).
At this stage, one needs to define a ‘dictionary’ between the real (relative) two-
body dynamics, summarized in Eq. (10), and the effective one-body one, summarized
in Eq. (14). As, on both sides, quantum mechanics tells us that the action variables
are quantized in integers (Nreal = n, Neff = n eff , etc.), it is most natural to
identify n = n eff and ν = νeff . One then still needs a rule for relating the two
relative and E . Buonanno and Damour [57] proposed to look
different energies E real eff
for a general map between the real energy levels and the effective ones (which, as
seen when comparing (10) and (14), cannot be directly identified because they do
not include the same rest-mass contribution5 ), namely

relative ⎩ relative 2
Eeff relative
E real E real relative
E real E real
−1= f = 1 + λ1 + λ2
μc2 μc2 μc2 μc2 μc2
3 ⎫
relative
E real
+ λ3 + ... . (15)
μc2
The ‘correspondence’ between the real and effective energy levels is illustrated in
Fig. 1.
Finally, identifying Eeff (n, ν)/μc2 to 1 + f (E real relative (n, ν)/μc2 ) yields a system
of equations for determining the unknown EOB coefficients an ,

bn , λn , as well as
the three coefficients z 1 , z 2 , z 3 parametrizing a general 3PN-level quartic mass-shell
deformation:
⎩ ⎫2
1 1 GM
Q 3PN ( p) = 6 2 z 1 p4 + z 2 p2 (n · p)2 + z 3 (n · p)4 .
c μ R
[The need for introducing a quartic mass-shell deformation Q only arises at the 3PN
level.]
The above system of equations for an ,
bn , λn (and z i at 3PN) was studied at
the 2PN level in Ref. [57], and at the 3PN level in Ref. [59]. At the 2PN level it
total = Mc2 + E relative = Mc2 + Newtonian terms + 1PN/c2 + · · · , while E

effective =
5Indeed E real real
μc + N + 1PN/c2 + · · · .
2
120 T. Damour
Fig. 1 Sketch of the correspondence between the quantized energy levels of the real and effec-
tive conservative dynamics. n denotes the ‘principal quantum number’ (n = nr + ν + 1, with
nr = 0, 1, . . . denoting the number of nodes in the radial function), while ν denotes the (relative)
orbital angular momentum (L 2 = ν(ν + 1) 2 ). Though the EOB method is purely classical, it is
conceptually useful to think in terms of the underlying (Bohr-Sommerfeld) quantization conditions
of the action variables I R and J to motivate the identification between n and ν in the two dynamics
was found that, if one further imposes the natural condition b1 = +2 (so that the
linearized effective metric coincides with the linearized Schwarzschild metric with
mass M = m 1 + m 2 ), there exists a unique solution for the remaining five unknown
a2 ,
coefficients a3 ,
b2 , λ1 and λ2 . This solution is very simple:
ε
a3 = 2ε,
a2 = 0, b2 = 4 − 6ε, λ1 = , λ2 = 0 . (16)
2
At the 3PN level, it was found that the system of equations is consistent, and under-
determined in that the general solution can be parametrized by the arbitrary values of
z 1 and z 2 . It was then argued that it is natural to impose the simplifying requirements
z 1 = 0 = z 2 , so that Q is proportional to the fourth power of the (effective) radial
momentum pr . With these conditions, the solution is unique at the 3PN level, and is
still remarkably simple, namely
a4 = a4 ε, d3 = 2(3ε − 26)ε, λ3 = 0, z 3 = 2(4 − 3ε)ε.

Here, a4 denotes the number
94 41 2
a4 = − γ ≤ 18.6879027, (17)
3 32
while d3 denotes the coefficient of (GM/c2 R)3 in the PN expansion of the combined
metric coefficient
D(R) ◦ A(R) B(R).

Replacing B(R) by D(R) is convenient because (as was mentioned above), in the
test-mass limit ε ⇔ 0, the effective metric must reduce to the Schwarzschild metric,
namely
⎩ ⎫
−1 GM
A(R; ε = 0) = B (R; ε = 0) = 1 − 2 2 ,
c R
so that
D(R; ε = 0) = 1.
The final result is that the three EOB potentials A, D, Q describing the 3PN two-
body dynamics are given by the following very simple results. In terms of the EOB
“gravitational potential”
GM
u◦ 2 ,
c R
A3PN (R) = 1 − 2u + 2 ε u 3 + a4 ε u 4 , (18)
D3PN (R) ◦ (A(R)B(R))3PN = 1 − 6εu 2 + 2(3ε − 26)εu 3 , (19)
1 4
2 pr
Q 3PN (q, p) = 2(4 − 3ε)ε u . (20)
c2 μ2
In addition, the map between the (real) center-of-mass energy of the binary system
relative = H relative = E tot
relative − Mc and the effective one Eeff is found to have the
E real 2
very simple (but non trivial) form

Eeff relative
E real ε E real
relative s − m 21 c4 − m 22 c4
= 1 + 1+ = , (21)
μc2 μc2 2 μc2 2 m 1 m 2 c4
where s = (Erealtot )2 ◦ (Mc2 + E relative )2 is Mandelstam’s invariant s = −( p + p )2 .

real 1 2
It is truly remarkable that the EOB formalism succeeds in condensing the com-
plicated, original 3PN Hamiltonian, Eqs. (4)–(2), into the very simple potentials
A, D and Q displayed above, together with the simple energy map Eq. (21). For
instance, at the 1PN level, the already somewhat involved Lorentz-Droste-Einstein-
Infeld-Hoffmann 1PN dynamics (Eqs. (4) and (5)) is simply described, within the
EOB formalism, as a test particle of mass μ moving in an external Schwarzschild
background of mass M = m 1 + m 2 , together with the (crucial but quite simple)
energy transformation (21). [Indeed, the ε-dependent corrections to A and D start
only at the 2PN level.] At the 2PN level, the seven rather complicated ε-dependent
coefficients of H 2PN (q, p), Eq. (6), get condensed into the two very simple additional
contributions + 2εu 3 in A(u), and − 6εu 2 in D(u). At the 3PN level, the eleven quite
complicated ε-dependent coefficients of H 3PN , Eq. (2), get condensed into only three
122 T. Damour
simple contributions: + a4 εu 4 in A(u), + 2(3ε − 26)εu 3 in D(u), and Q 3PN given

by Eq. (20). This simplicity of the EOB results is not only due to the reformulation
of the PN-expanded Hamiltonian into an effective dynamics. Indeed, the A-potential
happens to be much simpler than it could a priori have been: (i) as already noted
it is not modified at the 1PN level, while one would a priori expect to have found
a 1PN potential A1PN (u) = 1 − 2u + εa2 u 2 with some non zero a2 ; and (ii) there
are striking cancellations taking place in the calculation of the 2PN and 3PN coef-
ficients a2 (ε) and a2 (ε) = a2 ε + a2√ ε 2 , and
a3 (ε), which were a priori of the form
a3 (ε) = a3 ε + a3 ε + a3 ε , but for which the ε-nonlinear contributions a2√ ε 2 , a3√ ε 2
√ 2 √√ 3
and a3√√ ε 3 precisely cancelled out.

The fact that the 3PN coefficient a4 in the crucial ‘effective radial potential’
A3PN (R), Eq. (18), is rather large and positive indicates that the ε-dependent non-
linear gravitational effects lead, for comparable masses (ε ⊗ 14 ), to a last stable
(circular) orbit (LSO) which has a higher frequency and a larger binding energy than
what a naive scaling from the test-particle limit (ε ⇔ 0) would suggest. Actually,
the PN-expanded form (18) of A3PN (R) does not seem to be a good representation
of the (unknown) exact function AEOB (R) when the (Schwarzschild-like) relative
coordinate R becomes smaller than about 6GM/c2 (which is the radius of the LSO
in the test-mass limit). By continuity with the test-mass case, one a priori expects
that A3PN (R) always exhibits a simple zero defining an EOB “effective horizon” that
is smoothly connected to the Schwarzschild event horizon at R = 2GM/c2 when
ε ⇔ 0. However, the large value of the a4 coefficient does actually prevent A3PN
to have this property when ε is too large, and in particular when ε = 1/4. It was
therefore suggested [59] to further resum6 A3PN (R) by replacing it by a suitable
Padé (P) approximant. For instance, the replacement of A3PN (R) by7
1 + n1u
A13 (R) ◦ P31 [A3PN (R)] = (22)
1 + d1 u + d2 u 2 + d3 u 3
ensures that the ε = 41 case is smoothly connected with the ε = 0 limit.

The same kind of ε-continuity argument, discussed so far for the A function, needs
to be applied also to the D3PN (R) function defined in Eq. (19). A straightforward
way to ensure that the D function stays positive when R decreases (since it is D = 1
when ε ⇔ 0) is to replace D3PN (R) by D30 (R) ◦ P30 [D3PN (R)], where P30 indicates
the (0, 3) Padé approximant and explicitly reads
1
D30 (R) = . (23)
1 + 6εu 2 − 2(3ε − 26)εu 3
6 The PN-expanded EOB building blocks A3PN (R), B3PN (R), . . . already represent a resummation
of the PN dynamics in the sense that they have “condensed” the many terms of the original PN-
expanded Hamiltonian within a very concise format. But one should not refrain to further resum
the EOB building blocks themselves, if this is physically motivated.
7 We recall that the coefficients n and (d , d , d ) of the (1, 3) Padé approximant P 1 [A
1 1 2 3 3 3PN (u)]
are determined by the condition that the first four terms of the Taylor expansion of A13 in powers of
u = GM/(c2 R) coincide with A3PN .
3 EOB Description of Radiation Reaction and of the Emitted

Waveform During Inspiral
In the previous Section we have described how the EOB method encodes the conser-
vative part of the relative orbital dynamics into the dynamics of an ’effective’ particle.
Let us now briefly discuss how to complete the EOB dynamics by defining some
resummed expressions describing radiation reaction effects, and the corresponding
waveform emitted at infinity. One is interested in circularized binaries, which have
lost their initial eccentricity under the influence of radiation reaction. For such sys-
tems, it is enough (in first approximation [58]; see, however, the recent results of Bini
and Damour [73]) to include a radiation reaction force in the PΦ equation of motion
only. More precisely, we are using phase space variables R, PR , Φ, PΦ associated to
polar coordinates (in the equatorial plane α = γ2 ). Actually it is convenient to replace
the radial momentum
PR by the momentum conjugate to the ‘tortoise’ radial coor-
dinate R≡ = dR(B/A)1/2 , i.e. PR≡ = (A/B)1/2 P∇R . The real EOB Hamiltonian is
obtained by first solving Eq. (21) to get Hrealtotal = s in terms of Eeff , and then by
solving the effective Hamilton-Jacobi equation to get Eeff in terms of the effective
phase space coordinates q eff and peff . The result is given by two nested square roots
(we henceforth set c = 1):
real ⎛
HEOB 1
ĤEOB (r, pr≡ , Φ) = = 1 + 2ε ( Ĥeff − 1) , (24)
μ ε
where
⎝

pΦ2 pr4≡
Ĥeff ⎞
= pr≡ + A(r ) 1 + 2 + z 3 2 ,
2 (25)
r r
with z 3 = 2ε (4−3ε). Here, we are using suitably rescaled dimensionless (effective)

variables: r = R/GM, pr≡ = PR≡ /μ, pΦ = PΦ /μ GM, as well as a rescaled time
t = T /GM. This leads to equations of motion for (r, Φ, pr≡ , pΦ ) of the form
dΦ ∂ ĤEOB
= ◦ β, (26)
dt ∂ pΦ
⎩ ⎫1/2
dr A ∂ ĤEOB
= , (27)
dt B ∂ pr≡
dpΦ
= FˆΦ , (28)
dt
⎩ ⎫1/2
dpr≡ A ∂ ĤEOB
=− , (29)
dt B ∂r
which explicitly read

124 T. Damour
dΦ ApΦ
= ◦β, (30)
dt εr 2 Ĥ Ĥeff
⎩ ⎫1/2 ⎩ ⎫
dr A 1 2A 3
= pr≡ + z 3 2 pr≡ , (31)
dt B ε Ĥ Ĥeff r
dpΦ
= FˆΦ , (32)
dt
⎩ ⎫ ⎠ ⎩ ⎫ ⎩ √ ⎫
d pr≡ A 1/2 1 √
pΦ2 √ 2A A 2A 4
=− A + 2 A − + z3 − 3 pr≡ , (33)
dt B 2ε Ĥ Ĥeff r r r2 r
where A√ = dA/dr. As explained above the EOB metric function A(r ) is defined
by Padé resumming the Taylor-expanded result (12) obtained from the matching
between the real and effective energy levels (as we were mentioning, one uses a
similar Padé resumming for D(r ) ◦ A(r ) B(r )). One similarly needs to resum FˆΦ ,
i.e., the Φ component of the radiation reaction which has been introduced on the r.h.s.
of Eq. (28).
Several methods have been tried during the development of the EOB formalism to
resum the radiation reaction FˆΦ (starting from the high-order PN-expanded results
that have been obtained in the literature; see references in the Introduction above).
Here, we shall briefly explain the new, parameter-free resummation technique for the
multipolar waveform (and thus for the energy flux) introduced in Ref. [68, 69] and
perfected in [64]. To be precise, the new results discussed in Ref. [64] are twofold: on
the one hand, that work generalized the ν = m = 2 resummed factorized waveform
of [68, 69] to higher multipoles by using the most accurate currently known PN-
expanded results [33–35] as well as the higher PN terms which are known in the test-
mass limit [39, 40]; on the other hand, it introduced a further resummation procedure
which consists in considering a new theoretical quantity, denoted as ωνm (x), which
enters the (ν, m) waveform (together with other building blocks, see below) only
through its ν-th power: h νm ⇒ (ωνm (x))ν . Here, and below, x denotes the invariant
PN-ordering parameter given during inspiral by x ◦ (GMβ/c3 )2/3 .
The main novelty introduced by Refs. [64, 68, 69] is to write the (ν, m) multipolar
waveform emitted by a circular nonspinning compact binary as the product of several
factors, namely
⎭
(ρ) GMε (ρ) (ν+ρ)/2 ν−ρ,−m γ (ρ) ν
h νm = n νm cl+ρ (ε)x Y , Θ Ŝeff Tνm eiτνm ωνm . (34)
c2 R 2
Here ρ denotes the parity of ν + m (ρ = γ(ν + m)), i.e. ρ = 0 for “even-parity”

(mass-generated) multipoles (ν + m even), and ρ = 1 for “odd-parity” (current-
(ρ) (ρ)
generated) ones (ν + m odd); n νm and cl+ρ (ε) are numerical coefficients; Ŝeff is a
μ-normalized effective source (whose definition comes from the EOB formalism);
Tνm is a resummed version [68, 69] of an infinite number of “leading logarithms”
entering the tail effects [15, 27]; τνm is a supplementary phase (which corrects the
phase effects not included in the complex tail factor Tνm ), and, finally, (ωνm )ν denotes
the ν-th power of the quantity ωνm which is the new building block introduced in [64].
Note that in previous papers [68, 69] the quantity (ωνm )ν was denoted as f νm and we
will often use this notation below. Before introducing explicitly the various elements
entering the waveform (34) it is convenient to decompose h νm as
(ρ) (N ,ρ) (ρ)

h νm = h νm ĥ νm , (35)
(N ,ρ)
where h νm is the Newtonian contribution (i.e. the product of the first five factors
in Eq. (34)) and
(ρ) (ρ)
ĥ νm ◦ Ŝeff Tνm eiτνm f νm (36)
represents a resummed version of all the PN corrections. The PN correcting factor

(ρ) (ρ)
ĥ νm , as well as all its building blocks, has the structure ĥ νm = 1 + O(x).
The reader will find in Ref. [64] the definitions of the quantities entering the
“Newtonian” waveform h (N ,ρ)
νm , as well as the precise definition of the effective source
(ρ) (ρ)
factor Ŝeff , which constitutes the first factor in the PN-correcting factor ĥ νm . Let us
(ρ)
only note here that the definition of Ŝeff makes use of EOB-defined quantities. For
(0)
instance, for even-parity waves (ρ = 0) Ŝeff is defined as the μ-scaled effective
(ρ)
energy Eeff /μc2 . [We use the “J -factorization” definition of Ŝeff when ρ = 1, i.e.
for odd parity waves.]
The second building block in the factorized decomposition is the “tail factor” Tνm
(introduced in Refs. [68, 69]). As mentioned above, Tνm is a resummed version of
an infinite number of “leading logarithms” entering the transfer function between
the near-zone multipolar wave and the far-zone one, due to tail effects linked to its
propagation in a Schwarzschild background of mass MADM = HEOB real . Its explicit
expression reads
ˆ
ξ(ν + 1 − 2ik̂) ˆ ˆ
Tνm = eγ k̂ e2ik̂ log(2kr0 ) , (37)
ξ(ν + 1)
∇
where r0 = 2GM/ e and k̂ˆ ◦ G HEOB real mβ and k ◦ mβ. Note that k̂ˆ differs from
k by a rescaling involving the real (rather than the effective) EOB Hamiltonian,
computed at this stage along the sequence of circular orbits.
The tail factor Tνm is a complex number which already takes into account some
of the dephasing of the partial waves as they propagate out from the near zone to
infinity. However, as the tail factor only takes into account the leading logarithms, one
needs to correct it by a complementary dephasing term, eiτνm , linked to subleading
logarithms and other effects. This subleading phase correction can be computed as
(ρ)
being the phase τνm of the complex ratio between the PN-expanded ĥ νm and the
above defined source and tail factors. In the comparable-mass case (ε ∈= 0), the 3PN
τ22 phase correction to the leading quadrupolar wave was originally computed in
Ref. [69] (see also Ref. [68] for the ε = 0 limit). Full results for the subleading
126 T. Damour
partial waves to the highest possible PN-accuracy by starting from the currently
known 3PN-accurate ε-dependent waveform [35] have been obtained in [64]. For
higher-order test-mass (ε ⇔ 0) contributions, see [74, 75]. For extensions of the
(non spinning) factorized waveform of [64], see [76–78].
The last factor in the multiplicative decomposition of the multipolar waveform
can be computed as being the modulus f νm of the complex ratio between the PN-
(ρ)
expanded ĥ νm and the above defined source and tail factors. In the comparable mass
case (ε ∈= 0), the f 22 modulus correction to the leading quadrupolar wave was
computed in Ref. [69] (see also Ref. [68] for the ε = 0 limit). For the subleading
partial waves, Ref. [64] explicitly computed the other f νm ’s to the highest possible
PN-accuracy by starting from the currently known 3PN-accurate ε-dependent wave-
form [35]. In addition, as originally proposed in Ref. [69], to reach greater accuracy
the f νm (x; ε)’s extracted from the 3PN-accurate ε ∈= 0 results are completed by
adding higher order contributions coming from the ε = 0 results [39, 40]. In the
particular f 22 case discussed in [69], this amounted to adding 4PN and 5PN ε = 0
terms. This “hybridization” procedure was then systematically pursued for all the
other multipoles, using the 5.5PN accurate calculation of the multipolar decomposi-
tion of the gravitational wave energy flux of Refs. [39, 40].
(ρ)
The decomposition of the total PN-correction factor ĥ νm into several factors is
in itself a resummation procedure which already improves the convergence of the
PN series one has to deal with: indeed, one can see that the coefficients entering
increasing powers of x in the PN expansion of the f νm ’s tend to be systematically
(ρ)
smaller than the coefficients appearing in the usual PN expansion of ĥ νm . The reason
for this is essentially twofold: (i) the factorization of Tνm has absorbed powers of
(ρ)
mγ which contributed to make large coefficients in ĥ νm , and (ii) the factorization
of either Ĥeff or ĵ has (in the ε = 0 case) removed the presence of an inverse
square-root singularity located at x = 1/3 which caused the coefficient of x n in any
PN-expanded quantity to grow as 3n as n ⇔ ⊥.
To further improve the convergence of the waveform several resummations
of the factor f νm (x) = 1 + c1νm x + c2νm x 2 + · · · have been suggested. First,
Refs. [68, 69] proposed to further resum the f 22 (x) function via a Padé (3,2) approx-
imant, P23 { f 22 (x; ε)}, so as to improve its behavior in the strong-field-fast-motion
regime. Such a resummation gave an excellent agreement with numerically computed
waveforms, near the end of the inspiral and during the beginning of the plunge, for
different mass ratios [68, 79, 80]. As we were mentioning above, a new route for
resumming f νm was explored in Ref. [64]. It is based on replacing f νm by its ν-th
root, say
ωνm (x; ε) = [ f νm (x; ε)]1/ν . (38)
The basic motivation for replacing f νm by ωνm is the following: the leading
(ρ) ν
“Newtonian-level” contribution to the waveform h νm contains a factor ζν rharm vρ ,
where rharm is the harmonic radial coordinate used in the MPM formalism [12, 14].
When computing the PN expansion of this factor one has to insert the PN expansion of
the (dimensionless) harmonic radial coordinate rharm , rharm = x −1 (1+c1 x +O(x 2 )),
as a function of the gauge-independent frequency parameter x. The PN re-expansion

of [rharm (x)]ν then generates terms of the type x −ν (1+νc1 x+· · · ). This is one (though
not the only one) of the origins of 1PN corrections in h νm and f νm whose coefficients
grow linearly with ν. The study of [64] has pointed out that these ν-growing terms
are problematic for the accuracy of the PN-expansions. The replacement of f νm by
ωνm is a cure for this problem.
Several studies, both in the test-mass limit, ε ⇔ 0 (see Fig. 1 in [64]) and in the
comparable-mass case (see notably Fig. 4 in [65]), have shown that the resummed
factorized (inspiral) EOB waveforms defined above provided remarkably accurate
analytical approximations to the “exact” inspiral waveforms computed by numerical
simulations. These resummed multipolar EOB waveforms are much closer (espe-
cially during late inspiral) to the exact ones than the standard PN-expanded wave-
forms given by Eq. (35) with a PN-correction factor of the usual “Taylor-expanded”
form
h (ρ)PN

νm = 1 + c1νm x + c3/2
νm 3/2
x + c2νm x 2 + · · ·
See Fig. 1 in [64], and slide 29 in my (June 2012) Prague presentation.

Finally, one uses the newly resummed multipolar waveforms (34) to define a
resummation of the radiation reaction force FΦ as
1 (νmax )
FΦ ◦ − F , (39)
β
where the (instantaneous, circular) GW flux F (νmax ) is defined as
νmax
ν
2
F (νmax ) ◦ (mβ)2 |Rh νm |2 . (40)
16γ G
ν=2 m=1
Summarizing: Eqs. (34) and (39), (40) define resummed EOB versions of the
waveform h νm , and of the radiation reaction FˆΦ , during inspiral. A crucial point is that
these resummed expressions are parameter-free. Given some current approximation
to the conservative EOB dynamics (i.e. some expressions for the A, D, Q potentials),
they complete the EOB formalism by giving explicit predictions for the radiation
reaction (thereby completing the dynamics, see Eqs. (26)–(29)), and for the emitted
inspiral waveform.
4 EOB Description of the Merger of Binary Black Holes

and of the Ringdown of the Final Black Hole
Up to now we have reviewed how the EOB formalism, starting only from analytical
information obtained from PN theory, and adding extra resummation requirements
128 T. Damour
(both for the EOB conservative potentials A, Eq. (22), and D, Eq. (23), and for the
waveform, Eq. (34), and its associated radiation reaction force, Eqs. (39), (40)) makes
specific predictions, both for the motion and the radiation of binary black holes. The
analytical calculations underlying such an EOB description are essentially based on
skeletonizing the two black holes as two, sufficiently separated point masses, and
therefore seem unable to describe the merger of the two black holes, and the subse-
quent ringdown of the final, single black hole formed during the merger. However, as
early as 2000 [58], the EOB formalism went one step further and proposed a specific
strategy for describing the complete waveform emitted during the entire coalescence
process, covering inspiral, merger and ringdown. This EOB proposal is somewhat
crude. However, the predictions it has made (years before NR simulations could
accurately describe the late inspiral and merger of binary black holes) have been
broadly confirmed by subsequent NR simulations. [See the Introduction for a list of
EOB predictions.] The original EOB proposal (which was motivated partly by the
closeness between the 2PN-accurate effective metric gμε eff [57] and the Schwarzschild
metric, and by the results of Refs. [66] and [67]) consists of:
(i) defining, within EOB theory, the instant of (effective) “merger” of the two black
holes as the (dynamical) EOB time tm where the orbital frequency β(t) reaches
its maximum;
(ii) describing (for t ∗ tm ) the inspiral-plus-plunge (or simply insplunge) waveform,
h insplunge (t), by using the inspiral EOB dynamics and waveform reviewed in the
previous Section; and
(iii) describing (for t ⇐ tm ) the merger-plus-ringdown waveform as a superposition
of several quasi-normal-mode (QNM) complex frequencies of a final Kerr black
hole (of mass M f and spin parameter a f , self-consistency estimated within the
EOB formalism), say
⎩ ⎫
Rc2 +
C N+ e−δ N (t−tm ) ,
ringdown
h νm (t) = (41)
GM
N
with δ N+ = λ N + i ζ N , and where the label N refers to indices (ν, ν√ , m, n),

with (ν, m) being the Schwarzschild-background multipolarity of the considered
(metric) waveform h νm , with n = 0, 1, 2 . . . being the ‘overtone number’ of
the considered Kerr-background Quasi-Normal-Mode, and ν√ the degree of its
associated spheroidal harmonics Sν√ m (aδ, α );
(iv) determining the excitation coefficients C N+ of the QNM’s in Eq. (41) by us-
ing a simplified representation of the transition between plunge and ring-down
obtained by smoothly matching (following Ref. [68]), on a (2 p + 1)-toothed
“comb” (tm − pτ, . . . , tm − τ, tm , tm + τ, . . . , tm + pτ) centered around the
merger (and matching) time tm , the inspiral-plus-plunge waveform to the above
ring-down waveform.
Finally, one defines a complete, quasi-analytical EOB waveform (covering the

full process from inspiral to ring-down) as:
insplunge ringdown
νm (t) = α (tm − t) h νm
h EOB (t) + α (t − tm ) h νm (t) , (42)
where α (t) denotes Heaviside’s step function. The final result is a waveform that
essentially depends only on the choice of a resummed EOB A(u) potential, and, less
importantly, on the choice of resummation of the main waveform amplitude factor
f 22 = (ω22 )2 .
We have emphasized here that the EOB formalism is able, in principle, start-
ing only from the best currently known analytical information, to predict the full
waveform emitted by coalescing binary black holes. The early comparisons between
3PN-accurate EOB predicted waveforms8 and NR-computed waveforms showed a
satisfactory agreement between the two, within the (then relatively large) NR uncer-
tainties [81, 82]. Moreover, as we shall discuss below, it has been recently shown
that the currently known Padé-resummed 3PN-accurate A(u) potential is able, as is,
to describe with remarkable accuracy several aspects of the dynamics of coalescing
binary black holes [83, 84].
On the other hand, when NR started delivering high-accuracy waveforms, it be-
came clear that the 3PN-level analytical knowledge incorporated in EOB theory was
not accurate enough for providing waveforms agreeing with NR ones within the high-
accuracy needed for detection, and data analysis of upcoming GW signals. [See, e.g.,
the discussion in Sect. II of Ref. [77].] At that point, one made use of the natural
flexibility of the EOB formalism. Indeed, as already emphasized in early EOB work
[60, 85], we know from the analytical point of view that there are (yet uncalculated)
further terms in the u-expansions of the EOB potentials A(u), D(u), . . . (and in the
x-expansion of the waveform), so that these terms can be introduced either as “free
parameter(s) in constructing a bank of templates, and [one should] wait until” GW
observations determine their value(s) [60], or as “fitting parameters and adjusted so
as to reproduce other information one has about the exact results” (to quote Ref. [85]).
For instance, modulo logarithmic corrections that will be further discussed below,
the Taylor expansion in powers of u of the main EOB potential A(u) reads
ATaylor (u; ε) = 1 − 2u +
a3 (ε)u 3 +
a4 (ε)u 4 +
a5 (ε)u 5 +
a6 (ε)u 6 + · · · ,
where the 2PN and 3PN coefficients a3 (ε) = 2ε and a4 (ε) = a4 ε are known, but
where the 4PN, 5PN, . . . coefficients, a5 (ε), a6 (ε), . . . have not yet been calculated
(see, however, below). A first attempt was made in [85] to use numerical data (on
circular orbits of corotating black holes) to fit for the value of a (single, effective) 4PN
parameter of the simple form a5 (ε) = a5 ε entering a Padé-resummed 4PN-level A
potential, i.e.
8 The new, resummed EOB waveform discussed above was not available at the time, so that these
(N ,ρ)
comparisons employed the coarser “Newtonian-level” EOB waveform h 22 (x).
130 T. Damour

A14 (u; a5 , ε) = P41 A3PN (u) + εa5 u 5 . (43)
This strategy was pursued in Refs. [69, 86] and many subsequent works. It was
a6 (ε) = a6 ε,
pointed out in Ref. [65] that the introduction of a further 5PN coefficient
entering a Padé-resummed 5PN-level A potential, i.e.

A15 (u; a5 , a6 , ε) = P51 A3PN (u) + εa5 u 5 + εa6 u 6 , (44)
helped in having a closer agreement with accurate NR waveforms.

In addition, Refs. [68, 69] introduced another type of flexibility parameters of the
EOB formalism: the non quasi-circular (NQC) parameters accounting for uncalcu-
lated modifications of the quasi-circular inspiral waveform presented above, linked
to deviations from an adiabatic quasi-circular motion. These NQC parameters are of
various types, and subsequent works [65, 77, 79, 80, 87–89] have explored several
ways of introducing them. They enter the EOB waveform in two separate ways. First,
through an explicit, additional complex factor multiplying h νm , e.g.
= (1 + a1νm n 1 + a2νm n 2 ) exp[i(a3νm n 3 + a4νm n 4 )] ,

NQC
f νm
where the n i ’s are dynamical functions that vanish in the quasi-circular limit (with
n 1 , n 2 being time-even, and n 3 , n 4 time-odd). For instance, one usually takes n 1 =
( pr≡ /r β)2 . Second, through the (discrete) choice of the argument used during the
plunge to replace the variable x of the quasi-circular inspiral argument: e.g. either
xβ ◦ (GMβ)2/3 , or (following [90]) xΦ ◦ vΦ2 = (rζ β)2 , where vΦ ◦ β rζ , and
rζ ◦ r [ψ(r, pΦ )]1/3 is a modified EOB radius, with ψ being defined as
⎡ ⎝
⎤
⎩ ⎫−1
2 dA(r ) p 2
⎣1 + 2ε ⎞ A(r ) 1 + Φ − 1⎦ .
ψ(r, pΦ ) = 2 (45)
r dr r2
For a given value of the symmetric mass ratio, and given values of the A-flexibility
parameters a5 (ε),a6 (ε), one can determine the values of the NQC parameters aiνm ’s
from accurate NR simulations of binary black hole coalescence (with mass ratio ε)
by imposing, say, that the complex EOB waveform h EOB νm (t
EOB ; a6 ; aiνm ) oscu-
a5 ,
lates the corresponding NR one h νm (t ) at their respective instants of “merger”,
NR NR
EOB ◦ t EOB was defined above (maximum of βEOB (t)), while t NR
where tmerger m merger is de-
fined, say, as the (retarded) NR time where the modulus |h 22 (t)| of the quadrupolar
NR
waveform reaches its maximum. The order of osculation that one requires between
νm (t) and h νm (t) (or, separately, between their moduli and their phases or frequen-
h EOB NR
cies) depends on the number of NQC parameters aiνm . For instance, a1νm and a2νm
νm and allow one to match both |h νm | and its first time
affect only the modulus of h EOB EOB
νm νm
derivative, at merger, to their NR counterparts, while a3 , a4 affect only the phase
of the EOB waveform, and allow one to match the GW frequency ζνm EOB (t) and its
first time derivative, at merger, to their NR counterparts. The above EOB/NR match-
ing scheme has been developed and declined in various versions in Refs. [65, 77, 79,
80, 87–89, 91, 92]. One has also extracted the needed matching data from accurate
NR simulations, and provided explicit, analytical ε-dependent fitting formulas for
them [65, 77, 92].
Having so “calibrated” the values of the NQC parameters by extracting non-
perturbative information from a sample of NR simulations, one can then, for any
choice of the A-flexibility parameters, compute a full EOB waveform (from early
inspiral to late ringdown). The comparison of the latter NQC-completed EOB wave-
form to the results of NR simulations is discussed in the next Section.
5 EOB Versus NR
There have been several different types of comparison between EOB and NR. For
instance, the early work [81] pioneered the comparison between a purely analytical
EOB waveform (uncalibrated to any NR information) and a NR waveform, while
the early work [93] compared the predictions for the final spin of a coalescing black
hole binary made by EOB, completed by the knowledge of the energy and angular
momentum lost during ringdown by an extreme mass ratio binary (computed by the
test-mass NR code of [94]), to comparable-mass NR simulations [95]. Since then,
many other EOB/NR comparisons have been performed, both in the comparable-
mass case [65, 69, 79, 80, 82, 86, 87], and in the small-mass-ratio case [68, 88,
89, 96, 97]. Note in this respect that the numerical simulations of the GW emis-
sion by extreme mass-ratio binaries have provided (and still provide) a very useful
“laboratory” for learning about the motion and radiation of binary systems, and their
description within the EOB formalism.
Here we shall discuss only two recent examples of EOB/NR comparisons, which
illustrate different facets of this comparison.
5.1 EOB[NR] Waveforms Versus NR Ones
We explained above how one could complete the EOB formalism by calibrating some
of the natural EOB flexibility parameters against NR data. First, for any given mass
a5 (ε),
ratio ε and any given values of the A-flexibility parameters a6 (ε), one can use
NR data to uniquely determine the NQC flexibility parameters ai ’s. In other words,
we have (for a given ε)
ai = ai [NR data; a5 , a6 ] ,
where we defined a5 and a6 so that a5 (ε) = a5 ε,a6 (ε) = a6 ε. [We allow for some
residual ε-dependence in a5 and a6 .] Inserting these values in the (analytical) EOB
132 T. Damour
waveform then defines an NR-completed EOB waveform which still depends on the
two unknown flexibility parameters a5 and a6 .
In Ref. [65] the (a5 , a6 )-dependent predictions made by such a NR-completed
EOB formalism were compared to the high-accuracy waveform from an equal-mass
binary black hole (ε = 1/4) computed by the Caltech-Cornell-CITA group [98],
(and then made available on the web). It was found that there is a strong degeneracy
between a5 and a6 in the sense that there is an excellent EOB-NR agreement for an
extended region in the (a5 , a6 )-plane. More precisely, the phase difference between
the EOB (metric) waveform and the Caltech-Cornell-CITA one, considered between
GW frequencies MζL = 0.047 and MζR = 0.31 (i.e., the last 16 GW cycles
before merger), stays smaller than 0.02 radians within a long and thin banana-like
region in the (a5 , a6 )-plane. This “good region” approximately extends between
the points (a5 , a6 ) = (0, −20) and (a5 , a6 ) = (−36, +520). As an example (which
actually lies on the boundary of the “good region”), we shall consider here (following
Ref. [99]) the specific values a5 = 0, a6 = −20 (to which correspond, when ε = 1/4,
a1 = −0.036347, a2 = 1.2468). [Damour and Nagar [65] did not make use of the
NQC phase flexibility; i.e. it took a3 = a4 = 0. In addition, it used n 2 = r̈ /r β2 and
NQC
introduced a (real) modulus NQC factor f νm only for the dominant quadrupolar
wave ν = 2 = m.] We henceforth use M as time unit. This result relies on the proper
comparison between NR and EOB time series, which is a delicate subject. In fact, to
compare the NR and EOB phase time-series κ22 NR (t ) and κ EOB (t
NR 22 EOB ) one needs
to shift, by additive constants, both one of the time variables, and one of the phases.
In other words, we need to determine φ and λ such that the “shifted” EOB quantities
√ √
tEOB = tEOB + φ , κ22EOB = κ22
EOB
+λ (46)
“best fit” the NR ones. One convenient way to do so is first to “pinch” (i.e. constrain
to vanish) the EOB/NR phase difference at two different instants (corresponding to
two different frequencies ζ1 and ζ2 ). Having so related the EOB time and phase
variables to the NR ones we can straigthforwardly compare the EOB time series to
its NR correspondant. In particular, we can compute the (shifted) EOB–NR phase
difference
√
ψζ1 ,ζ2 κ22
EOBNR
(tNR ) ◦ κ22EOB (t √EOB ) − κ22 (t ).
NR NR
(47)
Figure 2 compares9 (the real part of) the analytical EOB metric quadrupolar wave-
form 22EOB /ε to the corresponding (Caltech-Cornell-CITA) NR metric waveform
22 /ε. [Here, 22 denotes the Zerilli-normalized asymptotic quadrupolar wave-
NR
∇
form, i.e. 22 ◦ Rh 22 / 24 with R = Rc2 /GM.] This NR metric waveform has
been obtained by a double time-integration (following the procedure of Ref. [80])
from the original, publicly available, curvature waveform ψ422 [98]. Such a curvature
waveform has been extrapolated both in resolution and in extraction radius. The
agreement between the analytical prediction and the NR result is striking, even around
9 The two “pinching” frequencies used for this comparison are Mζ1 = 0.047 and Mζ2 = 0.31.
0.3 Numerical Relativity (Caltech-Cornell)

EOB ( a 5 = 0; a 6 =-20)
0.2
0.1
[Ψ22]/ν
−0.1
−0.2
1:1 mass ratio
−0.3
500 1000 1500 2000 2500 3000 3500 4000
t
Fig. 2 This figure illustrates the comparison (made in Refs. [65, 99]) between the (NR-completed)
EOB waveform (Zerilli-normalized quadrupolar (ν = m = 2) metric waveform (42) with
parameter-free radiation reaction (39) and with a5 = 0, a6 = −20) and one of the most accurate
numerical relativity waveform (equal-mass case) nowadays available [98]. The phase difference
between the two is ψκ ∗ ±0.01 radians during the entire inspiral and plunge, which is at the level
of the numerical error
0.3
0.2
0.1
[Ψ22 ]/ν
−0.1
−0.2
−0.3 Merger time
3800 3820 3840 3860 3880 3900 3920 3940 3960 3980 4000
t
Fig. 3 Close up around merger of the waveforms of Fig. 2. Note the excellent agreement between
both modulus and phasing also during the ringdown phase
the merger. See Fig. 3 which closes up on the merger. The vertical line indicates the
location of the EOB-merger time, i.e., the location of the maximum of the orbital
frequency.
The phasing agreement between the waveforms is excellent over the full time
span of the simulation (which covers 32 cycles of inspiral and about 6 cycles of
134 T. Damour
ringdown), while the modulus agreement is excellent over the full span, apart from
two cycles after merger where one can notice a difference. More precisely, the phase
EOB − κ NR , remains remarkably small (⊗ ±0.02 radians)
difference, ψκ = κmetric metric
during the entire inspiral and plunge (ζ2 = 0.31 being quite near the merger). By
comparison, the root-sum of the various numerical errors on the phase (numerical
truncation, outer boundary, extrapolation to infinity) is about 0.023 radians during
the inspiral [98]. At the merger, and during the ringdown, ψκ takes somewhat larger
values (⊗ ±0.1 radians), but it oscillates around zero, so that, on average, it stays
very well in phase with the NR waveform whose error rises to ±0.05 radians during
ringdown. In addition, Ref. [65] compared the EOB waveform to accurate numerical
relativity data (obtained by the Jena group [80]) on the coalescence of unequal mass-
ratio black-hole binaries. Again, the agreement was good, and within the numerical
error bars.
This type of high-accuracy comparison between NR waveforms and EOB[NR]
ones (where EOB[NR] denotes a EOB formalism which has been completed by fit-
ting some EOB-flexibility parameters to NR data) has been pursued and extended
in Ref. [77]. The latter reference used the “improved” EOB formalism of Ref. [65]
with some variations (e.g. a third modulus NQC coefficient ai , two phase NQC co-
efficients, the argument xβ = (Mβ)2/3 in (ωνm (x))ν , eight QNM modes) and
Taylor
calibrated it to NR simulations of mass ratios q = m 2 /m 1 = 1, 2, 3, 4 and 6 per-

formed by the Caltech-Cornell-CITA group [56, 100]. They considered not only the
leading (ν, m) = (2, 2) GW mode, but the subleading ones (2, 1), (3, 3), (4, 4) and
(5, 5). They found that, for this large range of mass ratios, EOB[NR] (with suitably
fitted, ε-dependent values of a5 and a6 ) was able to describe the NR waveforms
essentially within the NR errors. This confirms the usefulness of the EOB formalism
in helping the detection and analysis of upcoming GW signals.
Here, having in view GW observations from ground-based interferometric detec-
tors, we focussed on comparable-mass systems. The EOB formalism has also been
compared to NR results in the extreme mass-ratio limit ε ≈ 1. In particular, Ref. [88]
found an excellent agreement between the analytical and numerical results.
5.2 EOB[3PN] Dynamics Versus NR One
Let us also mention other types of EOB/NR comparisons. Recently, two examples of
EOB/NR comparisons have been performed directly at the level of the dynamics of a
binary black hole, rather than at the level of the waveform. Moreover, contrary to the
waveform comparisons of the previous subsection which involved an NR-completed
EOB formalism (“EOB[NR]”), the dynamical comparisons we are going to discuss
involve the purely analytical 3PN-accurate EOB formalism (“EOB[3PN]”), without
any NR-based improvement.
First, Le Tiec et al. [83] have extracted from accurate NR simulations of slightly
eccentric binary black-hole systems (for several mass ratios q = m 1 /m 2 between
1/8 and 1) the function relating the periastron-advance parameter
ψΘ
K =1+ ,
2γ
(where ψΘ is the periastron advance per radial period) to the dimensionless averaged
angular frequency MβΦ (with M = m 1 + m 2 as above). Then they compared the
NR-estimate of the mass-ratio dependent functional relation
K = K (MβΦ ; ε) ,
where ε = q/(1+q)2 , to the predictions of various analytic approximation schemes:

PN theory, EOB theory and two different ways of using GSF theory. Let us only
mention here that the prediction from the purely analytical EOB[3PN] formalism for
K (MβΦ ; ε) [101] agreed remarkably well (essentially within numerical errors) with
its NR estimate for all mass ratios, while, by contrast, the PN-expanded prediction
for K (MβΦ ; ε) [70] showed a much poorer agreement, especially as q moved away
from 1.
Second, Damour et al. [84] have recently extracted from accurate NR simulations
of black-hole binaries (with mass ratios q = m 2 /m 1 = 1, 2 and 3) the gauge-
invariant relation between the (reduced) binding energy E = (E tot − M)/μ and the
(reduced) angular momentum j = J/(GμM) of the system. Then they compared
the NR-estimate of the mass-ratio dependent functional relation
E = E( j; ε)
to the predictions of various analytic approximation schemes: PN theory and various

versions of EOB theory (some of these versions were NR-completed). Let us only
mention here that the prediction from the purely analytical, 3PN-accurate EOB[3PN]
for E( j; ε) agreed remarkably well with its NR estimate (for all mass ratios) essen-
tially down to the merger. This is illustrated in Fig. 4 for the q = 1 case. By contrast,
the 3PN expansion in (powers of 1/c2 ) of the function E( j; ε) showed a much poorer
agreement (for all mass ratios).
6 Other Developments
6.1 EOB with Spinning Bodies
We lack space here for discussing the extension of the EOB formalism to bi-
nary systems made of spinning bodies. Let us start by mentioning that the spin-
extension of the EOB formalism was initiated in Ref. [60], that the first EOB-based
analytical calculation of a complete waveform from a spinning binary was performed
in Ref. [61], and that the first attempt at calibrating a spinning EOB model to accu-
136 T. Damour
−0.03
NR
3PN-Taylor
−0.04 EOB 3PN
EOB adiabatic
3PN
−0.05
−0.06
−0.0375
−0.07
E
−0.038
−0.08
−0.09 −0.0385
−0.1 −0.039
−0.11 −0.0395
−0.12
−0.04
3.92 3.93 3.94 3.95 3.96 3.97 3.98
−0.13
2.8 3 3.2 3.4 3.6 3.8 4
j
Fig. 4 Comparison (made in [84]) between various analytical estimates of the energy-angular
momentum functional relation and its numerical-relativity estimate (equal-mass case). The standard
“Taylor-expanded” 3PN E( j) curve shows the largest deviation from NR results, especially at low
j’s, while the two (adiabatic and nonadiabatic) 3PN-accurate, non-NR-calibrated EOB E( j) curves
agree remarkably well with the NR one
rate NR simulations of spinning (non precessing) black-hole binaries was presented

in [102]. In addition, several formal aspects related to the inclusion of spins in the
EOB formalism have been discussed in Refs. [103–107] (see references within these
papers for PN works dealing with spin effects) and a generalization of the factor-
ized multipolar waveform of Ref. [64] to spinning, non-precessing binaries has been
constructed in Refs. [76, 78].
6.2 EOB with Tidally Deformed Bodies

In binary systems comprising neutron stars, rather than black holes, the tidal
deformation of the neutron star(s) will significantly modify the phasing of the emit-
ted gravitational waveform during the late inspiral. As GW’s from binary neutron
stars are expected sources for upcoming ground-based GW detectors, it is important
to extend the EOB formalism by including tidal effects (see [108] and references
therein). This extension has been defined in Refs. [109, 110]. The comparison be-
tween this tidal-extended EOB and state-of-the-art NR simulations of neutron-star
binaries has been discussed in Refs. [111, 112]. It appears from these comparisons
that the tidal-extended EOB formalism is able to describe the motion and radiation
of neutron-star binaries within NR errors. More accurate simulations will be needed
to ascertain whether one needs to calibrate some higher-order flexibility parameters
of the tidal-EOB formalism, or whether the currently known analytic accuracy is

sufficient.
6.3 EOB and GSF
We mentioned in the Introduction that GSF theory has recently opened a new source
of information on the general relativistic two-body problem. Let us briefly mention
here that there has been, recently, a quite useful transfer of information from GSF
theory to EOB theory. The program of using GSF theory to improve EOB theory was
first highlighted in Ref. [101]. That work pointed to several concrete gauge-invariant
calculations (within GSF theory) that would provide accurate information about
the O(ε) contributions to several EOB potentials. More precisely, let us define the
functions a(u) and d̄(u) as the ε-linear contributions to the EOB potentials A(u; ε)
and D(u; ε) ◦ D −1 (u; ε):
A(u; ε) = 1 − 2u + ε a(u) + O(ε 2 ) ,
D(u; ε) = (AB)−1 = 1 + ε d̄(u) + O(ε 2 ) .
Reference [101] has shown that a computation of the GSF-induced correction to

the periastron advance of slightly eccentric orbits would allow one to compute the
following combination of EOB functions
1
ω̄(u) = a(u) + u a √ (u) + u(1 − 2u) a √√ (u) + (1 − 6u) d̄(u) .
2
The GSF-calculation of the EOB function ω̄(u) was then performed in Ref. [113] (in
the range 0 ∗ u ∗ 16 ).
More recently, a series of works by Le Tiec and collaborators [114–116] have
(through an indirect route) shown how GSF calculations could be used to compute the
EOB ε-linear a(u) function separately from the d̄(u) one. Reference [116] then gave
a fitting formula for a(u) over the interval 0 ∗ u ∗ 15 as well as accurate estimates
of the coefficients of the Taylor expansion of a(u) around u = 0 (corresponding to
the knowledge of the PN expansion of a(u) to a very high PN order). Very recently,
Ackay et al. [117] succeeded in accurately computing (through GSF theory) the
EOB a(u) function over the larger interval 0 ∗ u ∗ 13 . It was (surprisingly) found
! "−
that a(u) diverges like a(u) ≥ 0.25(1 − 3u)−1/2 at the light-ring limit u ⇔ 13 .
The meaning for EOB theory of this singular behavior of a(u) at the light-ring is
discussed in detail in Ref. [117].
138 T. Damour
6.4 Toward Further Improvements to EOB
Let us finally mention some avenues for further progress in EOB theory.
Logarithmic contributions to the A(u) and D(u) functions have been recently
computed at the 4PN level [101, 118] and even the 5PN one [116, 119].10 They have
been incorporated in a recent, improved implementation of the EOB formalism [92].
Two groups have embarked on a calculation of the (full) conservative dynamics at
the 4PN level [120, 121]. If they succeed, it will be important to translate their gauge-
dependent results in the gauge-invariant form used in EOB theory. [Remember
that EOB theory is essentially based on the gauge-invariant Delaunay Hamiltonian
H (Ia ).]
More generally, let us emphasize that the EOB formalism provides a convenient,
gauge-invariant way of packaging both the conservative dynamics and the multipolar
waveform. This EOB packaging has often turned out to be very economical. We
recommend that authors computing high-order PN corrections to either the dynamics
or the waveform reexpress their results in terms of the EOB building blocks.
For instance, Jaranowski and Schaëfer [121] have recently given a partial result at
4PN, expressed in terms of the (gauge-invariant) function E(MβΦ ; ε). In terms of
this function, the 4PN contribution is a polynomial of the fourth degree in ε, namely,
with x ◦ (MβΦ )2/3 and
1
E(x; ε) = − μc2 x(1 + e1P N (ε)x + e2P N (ε)x 2 + e3P N (ε)x 3
2
+ e4P N (ε; ln x)x 4 + O(x 5 ln x)),
they found
3969 301 3 77 4 448

e4P N (ε; ln x) = − + c1 ε + c2 ε 2 + ε + ε + ε ln x, (48)
128 1728 31104 15
where they could not compute the values of the coefficients c1 and c2 of the terms
linear and quadratic in ε, but only the contributions cubic and quartic in ε. We
wish to point out that their result is re-expressed in a more economical (and more
informative) way in terms of the basic EOB potential A(u; ε). Indeed, in terms of
the PN expansion, of A(u; ε),
ATaylor (u; ε) = 1 − 2u +
a3 (ε)u 3 +
a4 (ε)u 4 +
a5 (ε; ln u)u 5 +
a6 (ε; ln u)u 6 + · · ·
the information contained in the above result can be entirely re-expressed in terms
of the 4PN-level coefficient a5 (ε; ln u). When doing this re-expression, one then
finds that the information content of Eq. (48) is that the 4PN-level EOB coefficient
a5 (ε; ln u) is no more than quadratic in ε, i.e.

10See Ref. [113], which quoted and used some combinations of the logarithmic contributions to
a(u) and d̄(u). given in Ref. [119].
64

a5 (ε; ln u) = (a5 + ln u)ε + a5√ ε 2 ,
5
without contributions of degree ε 3 and ε 4 . We recall that similar cancellations of

higher ε n terms were found at lower PN orders in the EOB A(u; ε) function. Namely,
they were found to contain only terms linear in ε, while a3 (ε) could a priori have
been quadratic in ε, and a4 (ε) could a priori have been cubic in ε. The fact that sim-
ilar remarkable cancellations still hold, according to the result of [121], at the 4PN
level, is a clear indication that the EOB packaging of information of the dynamics
in the A(u; nu) potential is quite compact. Indeed, it says that the two complicated
ε + 31104
301 3
terms 1728 77
ε 4 in the energy function are already encoded in the structure
of the EOB formalism. Finally, note that the full gauge-invariant content of a 4PN
computation of the dynamics, when interpreted within the EOB formalism, is de-
scribed by only three EOB terms: the coefficient a5 (ε; ln u) in A(u; ε), an analogous

coefficient d̄ 4 (ε; ln u) in D̄(u; ε), and an additional contribution to Q( p).
Regarding the waveform, let us mention another recent example where it would
have been useful and clarifying to use the EOB packaging. Namely, when re-
expressing it in terms of the factorized EOB waveform, the new content of the
recent 3.5PN level computation by Faye et al. [122] of the PN-expanded quadrupo-
lar waveform h 22 , is entirely contained in an additional 3.5PN-level contribution
to the supplementary phase, namely τ22 = (30995/1134 ε + 962/135 ε 2 ) x 7/2 . In-
deed, the 3.5PN-level contributions to the modulus computed in [122] were already
included in the factorized EOB waveform of Ref. [65].
7 Conclusions
We hope that this brief review has made it clear that:
1. There is a complementarity between the various current approaches to the general
relativistic two-body problem: post-Newtonian, Effective One Body, gravita-
tional self-force and numerical relativity simulations (of both comparable-mass
and extreme-mass-ratio systems).
2. The effective one body formalism offers a convenient framework for combining,
in a synergetic manner, information coming from the other approaches. This
formalism seems to constitute an efficient way to analytically describe the motion
and radiation of circularized11 binaries, and to provide accurate gravitational
wave templates for detection and data analysis.
3. The general relativistic two-body problem is more lively than ever. It provides
an example of Poincaré’s sentence: “Il n’y a pas de problèmes résolus, il y a
seulement des problèmes plus ou moins résolus.” [“There are no (definitely)
solved problems, there are only more-or-less solved problems.”]
11 See [73] for a recent extension of the EOB formalism to non-circular (ellipticlike or
hyperboliclike) motions.
140 T. Damour
References
1. Damour, T.: The problem of motion in Newtonian and Einsteinian gravity. In: Hawking, S.W.,
Israel, W. (eds.) Three Hundred Years of Gravitation, pp. 128–198. Cambridge University
Press, Cambridge (1987)
2. Damour, T.: Gravitational radiation and the motion of compact bodies. In: Deruelle, N., Piran,
T. (eds.) Gravitational Radiation, pp. 59–144. North-Holland, Amsterdam (1983)
3. Jaranowski, P., Schäfer, G.: Third post-Newtonian higher order ADM Hamilton dynamics
for two-body point-mass systems. Phys. Rev. D 57, 7224 (1998). doi:10.1103/PhysRevD.57.
7274 [Erratum: ibid. 63, 029902 (2001)]
4. Blanchet, L., Faye, G.: General relativistic dynamics of compact binaries at the third post-
Newtonian order. Phys. Rev. D 63, 062005 (2001). doi:10.1103/PhysRevD.63.062005
5. Damour, T., Jaranowski, P., Schäfer, G.: Dimensional regularization of the gravitational in-
teraction of point masses. Phys. Lett. B 513, 147 (2001)
6. Blanchet, L., Damour, T., Esposito-Farèse, G.: Dimensional regularization of the third post-
Newtonian dynamics of point particles in harmonic coordinates. Phys. Rev. D 69, 124007
(2004). doi:10.1103/PhysRevD.69.124007
7. Itoh, Y., Futamase, T.: New derivation of a third post-Newtonian equation of motion for
relativistic compact binaries without ambiguity. Phys. Rev. D 68, 121501 (2003). doi:10.
1103/PhysRevD.68.121501
8. Pati, M.E., Will, C.M.: Post-Newtonian gravitational radiation and equations of motion via
direct integration of the relaxed Einstein equations. II: Two-body equations of motion to
second post-Newtonian order, and radiation-reaction to 3.5 post-Newton. Phys. Rev. D 65,
104008 (2002). doi:10.1103/PhysRevD.65.104008
9. Konigsdorffer, C., Faye, G., Schäfer, G.: The binary black-hole dynamics at the third-and-a-
half post-Newtonian order in the ADM-formalism. Phys. Rev. D 68, 044004 (2003). doi:10.
1103/PhysRevD.68.044004
10. Nissanke, S., Blanchet, L.: Gravitational radiation reaction in the equations of motion of
compact binaries to 3.5 post-Newtonian order. Class. Quantum Gravity 22, 1007 (2005).
doi:10.1088/0264-9381/22/6/008
11. Blanchet, L., Damour, T.: Radiative gravitational fields in general relativity I. General structure
of the field outside the source. Philos. Trans. R. Soc. Lond. Ser. A 320, 379 (1986)
12. Blanchet, L., Damour, T.: Post-Newtonian generation of gravitational waves. Ann. Inst. Henri
Poincare A 50, 377 (1989)
13. Damour, T., Iyer, B.R.: Multipole analysis for electromagnetism and linearized gravity with
irreducible cartesian tensors. Phys. Rev. D 43, 3259 (1991). doi:10.1103/PhysRevD.43.3259
14. T. Damour, B.R. Iyer, Post-Newtonian generation of gravitational waves. II. The spin mo-
ments. Ann. Inst. Henri Poincare A 54, 115 (1991)
15. Blanchet, L., Damour, T.: Hereditary effects in gravitational radiation. Phys. Rev. D 46, 4304
(1992). doi:10.1103/PhysRevD.46.4304
16. Blanchet, L.: Second-post-Newtonian generation of gravitational radiation. Phys. Rev. D 51,
2559 (1995). doi:10.1103/PhysRevD.51.2559
17. Will, C.M., Wiseman, A.G.: Gravitational radiation from compact binary systems: gravita-
tional waveforms and energy loss to second post-Newtonian order. Phys. Rev. D 54, 4813
(1996). doi:10.1103/PhysRevD.54.4813
18. Will, C.M.: Generation of post-Newtonian gravitational radiation via direct integration of the
relaxed Einstein equations. Prog. Theor. Phys. Suppl. 136, 158 (1999). doi:10.1143/PTPS.
136.158
19. Pati, M.E., Will, C.M.: Post-Newtonian gravitational radiation and equations of motion via
direct integration of the relaxed Einstein equations. I: Foundations. Phys. Rev. D 62, 124015
(2000). doi:10.1103/PhysRevD.62.124015
20. Epstein, R., Wagoner, R.V.: Post-Newtonian generation of gravitational waves. Astrophys. J.
197, 717 (1975). doi:10.1086/153561
21. Thorne, K.S.: Multipole expansions of gravitational radiation. Rev. Mod. Phys. 52, 299 (1980).
doi:10.1103/RevModPhys.52.299
22. Wagoner, R.V., Will, C.M.: Post-Newtonian gravitational radiation from orbiting point masses.
Astrophys. J. 210, 764 (1976). doi:10.1086/154886
5
23. Wiseman, A.G.: Coalescing binary systems of compact objects to (post) 2 -Newtonian order.
IV. The gravitational wave tail. Phys. Rev. D 48, 4757 (1993). doi:10.1103/PhysRevD.48.
4757
24. Blanchet, L., Schäfer, G.: Gravitational wave tails and binary star systems. Class. Quantum
Gravity 10, 2699 (1993). doi:10.1088/0264-9381/10/12/026
25. Blanchet, L., Damour, T., Iyer, B.R., Will, C.M., Wiseman, A.G.: Gravitational-radiation
damping of compact binary systems to second post-Newtonian order. Phys. Rev. Lett. 74,
3515 (1995). doi:10.1103/PhysRevLett.74.3515
26. Blanchet, L., Damour, T., Iyer, B.R.: Gravitational waves from inspiralling compact binaries:
energy loss and waveform to second-post-Newtonian order. Phys. Rev. D 51, 5360 (1995).
doi:10.1103/PhysRevD.51.5360 [Erratum: ibid. 54, 1860 (1996)]
27. Blanchet, L.: Gravitational-wave tails of tails, Class. Quantum Gravity 15, 113 (1998). [Er-
ratum ibid. 22, 3381 (2005)]
28. Blanchet, L., Iyer, B.R., Joguet, B.: Gravitational waves from inspiralling compact binaries:
Energy flux to third post-Newtonian order. Phys. Rev. D 65, 064005 (2002). [Erratum: ibid.
71, 129903 (2005)]
29. Blanchet, L., Iyer, B.R.: Hadamard regularization of the third post-Newtonian gravitational
wave generation of two point masses. Phys. Rev. D 71, 024004 (2005)
30. Blanchet, L., Damour, T., Esposito-Farèse, G., Iyer, B.R.: Gravitational radiation from inspi-
ralling compact binaries completed at the third post-Newtonian order. Phys. Rev. Lett. 93,
091101 (2004). doi:10.1103/PhysRevLett.93.091101
31. Blanchet, L., Damour, T., Esposito-Farèse, G., Iyer, B.R.: Dimensional regularization of the
third post-Newtonian gravitational wave generation from two point masses. Phys. Rev. D 71,
124004 (2005)
32. Blanchet, L.: Quadrupole-quadrupole gravitational waves. Class. Quantum Gravity 15, 89
(1998). doi:10.1088/0264-9381/15/1/008
33. Berti, E., Cardoso, V., Gonzalez, J.A., et al.: Inspiral, merger and ringdown of unequal mass
black hole binaries: a multipolar analysis. Phys. Rev. D 76, 064034 (2007). doi:10.1103/
PhysRevD.76.064034
34. Kidder, L.E.: Using full information when computing modes of post-Newtonian waveforms
from inspiralling compact binaries in circular orbit. Phys. Rev. D 77, 044016 (2008). doi:10.
1103/PhysRevD.77.044016
35. Blanchet, L., Faye, G., Iyer, B.R., Sinha, S.: The third post-Newtonian gravitational wave
polarisations and associated spherical harmonic modes for inspiralling compact binaries in
quasi-circular orbits. Class. Quantum Gravity 25, 165003 (2008). doi:10.1088/0264-9381/
25/16/165003
36. Blanchet, L.: Gravitational radiation from post-Newtonian sources and inspiralling compact
binaries. Living Rev. Relativ. 5, lrr-2002-3 (2002). http://www.livingreviews.org/lrr-2002-3
37. Poisson, E.: Gravitational radiation from a particle in circular orbit around a black hole. I.
Analytic results for the nonrotating case. Phys. Rev. D 47, 1497 (1993)
38. Sasaki, M.: Post-Newtonian expansion of the ingoing-wave Regge-Wheeler function. Progress
Theoret. Phys. 92, 17 (1994). doi:10.1143/PTP.92.17
39. Tagoshi, H., Sasaki, M.: Post-Newtonian expansion of gravitational waves from a particle in
circular orbit around a Schwarzschild black hole. Progress Theoret. Phys. 92, 745 (1994).
doi:10.1143/PTP.92.745
40. Tanaka, T., Tagoshi, H., Sasaki, M.: Gravitational waves by a particle in circular orbit around
a Schwarzschild black hole. Progress Theoret. Phys. 96, 1087 (1996). doi:10.1143/PTP.96.
1087
41. Sasaki, M., Tagoshi, H.: Analytic black hole perturbation approach to gravitational radiation.
Living Rev. Relativ. 6, lrr-2003-6 (2003). http://www.livingreviews.org/lrr-2003-6
142 T. Damour
42. Foffa, S., Sturani, R.: Effective field theory calculation of conservative binary dynamics at
third post-Newtonian order. Phys. Rev. D 84, 044031 (2011). doi:10.1103/PhysRevD.84.
044031
43. Goldberger, W.D., Rothstein, I.Z.: An effective field theory of gravity for extended objects.
Phys. Rev. D 73, 104029 (2006). doi:10.1103/PhysRevD.73.104029
44. Damour, T., Esposito-Farèse, G.: Tensor multiscalar theories of gravitation. Class. Quantum
Gravity 9, 2093 (1992). doi:10.1088/0264-9381/9/9/015
45. Damour, T., Esposito-Farèse, G.: Testing gravity to second post-Newtonian order: a field
theory approach. Phys. Rev. D 53, 5541 (1996). doi:10.1103/PhysRevD.53.5541
46. Damour, T., Esposito-Farèse, G.: Gravitational-wave versus binary-pulsar tests of strong-field
gravity. Phys. Rev. D 58, 042001 (1998). doi:10.1103/PhysRevD.58.042001
47. Goldberger, W.D., Ross, A.: Gravitational radiative corrections from effective field theory.
48. Pretorius, F.: Evolution of binary black hole spacetimes. Phys. Rev. Lett. 95, 121101 (2005).
doi:10.1103/PhysRevLett.95.121101
49. Campanelli, M., Lousto, C.O., Marronetti, P., Zlochower, Y.: Accurate evolutions of orbit-
ing black-hole binaries without excision. Phys. Rev. Lett. 96, 111101 (2006). doi:10.1103/
PhysRevLett.96.111101
50. Baker, J.G., Centrella, J., Choi, D.I., Koppitz, M., van Meter, J.: Gravitational wave extraction
from an inspiraling configuration of merging black holes. Phys. Rev. Lett. 96, 111102 (2006).
51. Boyle, M., Brown, D.A., Kidder, L.E., et al.: High-accuracy comparison of numerical relativity
simulations with post-Newtonian expansions. Phys. Rev. D 76, 124038 (2007). doi:10.1103/
PhysRevD.76.124038
52. Pretorius, F.: Binary black hole coalescence. In: Colpi, M., et al. (eds.) Physics of Relativistic
Objects in Compact Binaries: from Birth to Coalescense, Astrophysics and Space Science
Library, vol. 359, pp. 305–370. Springer/Canopus, Dordrecht (2009)
53. Barack, L.: Gravitational self force in extreme mass-ratio inspirals. Class. Quantum Gravity
26, 213001 (2009). doi:10.1088/0264-9381/26/21/213001
54. Hannam, M., Husa, S., Sperhake, U., Bruegmann, B., Gonzalez, J.A.: Where post-Newtonian
and numerical-relativity waveforms meet. Phys. Rev. D 77, 044020 (2008)
55. Hannam, M., Husa, S., Bruegmann, B., Gopakumar, A.: Comparison between numerical-
relativity and post-Newtonian waveforms from spinning binaries: the orbital hang-up case.
Phys. Rev. D 78, 104007 (2008)
56. MacDonald, I., Mroué, A.H., Pfeiffer, H.P., et al.: Suitability of hybrid gravitational wave-
forms for unequal-mass binaries. Phys. Rev. D 87, 024009 (2013). doi:10.1103/PhysRevD.
87.024009
57. Buonanno, A., Damour, T.: Effective one-body approach to general relativistic two-body
dynamics. Phys. Rev. D 59, 084006 (1999). doi:10.1103/PhysRevD.59.084006
58. Buonanno, A., Damour, T.: Transition from inspiral to plunge in binary black hole coales-
cences. Phys. Rev. D 62, 064015 (2000). doi:10.1103/PhysRevD.62.064015
59. Damour, T., Jaranowski, P., Schäfer, G.: On the determination of the last stable orbit for
circular general relativistic binaries at the third post-Newtonian approximation. Phys. Rev. D
62, 084011 (2000)
60. Damour, T.: Coalescence of two spinning black holes: an effective one-body approach. Phys.
Rev. D 64, 124013 (2001). doi:10.1103/PhysRevD.64.124013
61. Buonanno, A., Chen, Y., Damour, T.: Transition from inspiral to plunge in precessing binaries
of spinning black holes. Phys. Rev. D 74, 104005 (2006). doi:10.1103/PhysRevD.74.104005
62. Brézin, E., Itzykson, C., Zinn-Justin, J.: Relativistic balmer formula including recoil effects.
63. Damour, T., Iyer, B.R., Sathyaprakash, B.S.: Improved filters for gravitational waves from
inspiralling compact binaries. Phys. Rev. D 57, 885 (1998). doi:10.1103/PhysRevD.57.885
64. Damour, T., Iyer, B.R., Nagar, A.: Improved resummation of post-Newtonian multipolar
waveforms from circularized compact binaries. Phys. Rev. D 79, 064004 (2009). doi:10.
1103/PhysRevD.79.064004
65. Damour, T., Nagar, A.: An improved analytical description of inspiralling and coalescing
black-hole binaries. Phys. Rev. D 79, 081503 (2009). doi:10.1103/PhysRevD.79.081503
66. Davis, M., Ruffini, R., Tiomno, J.: Pulses of gravitational radiation of a particle falling radially
into a Schwarzschild black hole. Phys. Rev. D 5, 2932 (1972). doi:10.1103/PhysRevD.5.2932
67. Price, R.H., Pullin, J.: Colliding black holes: the close limit. Phys. Rev. Lett. 72, 3297 (1994).
68. Damour, T., Nagar, A.: Faithful effective-one-body waveforms of small-mass-ratio coalescing
black-hole binaries. Phys. Rev. D 76, 064028 (2007)
69. Damour, T., Nagar, A.: Comparing effective-one-body gravitational waveforms to accurate
numerical data. Phys. Rev. D 77, 024043 (2008)
70. Damour, T., Jaranowski, P., Schäfer, G.: Dynamical invariants for general relativistic two-
body systems at the third post-Newtonian approximation. Phys. Rev. D 62, 044024 (2000).
doi:10.1103/PhysRevD.62.044024
71. Damour, T., Jaranowski, P., Schäfer, G.: Poincaré invariance in the ADM Hamiltonian ap-
proach to the general relativistic two-body problem. Phys. Rev. D 62, 021501 (2000). doi:10.
1103/PhysRevD.62.021501 [Erratum: ibid. D 63, 029903 (2000)]
72. Damour, T., Schäfer, G.: Higher order relativistic periastron advances and binary pulsars.
Nuovo Cimento B 101, 127 (1988). doi:10.1007/BF02828697
73. Bini, D., Damour, T.: Gravitational radiation reaction along general orbits in the effective
one-body formalism. Phys. Rev. D 86, 124012 (2012). doi:10.1103/PhysRevD.86.124012
74. Fujita, R., Iyer, B.R.: Spherical harmonic modes of 5.5 post-Newtonian gravitational wave
polarisations and associated factorised resummed waveforms for a particle in circular orbit
around a Schwarzschild black hole. Phys. Rev. D 82, 044051 (2010). doi:10.1103/PhysRevD.
82.044051
75. Fujita, R.: Gravitational radiation for extreme mass ratio inspirals to the 14th post-Newtonian
order. Progress Theoret. Phys. 127, 583 (2012). doi:10.1143/PTP.127.583
76. Pan, Y., Buonanno, A., Fujita, R., Racine, E., Tagoshi, H.: Post-Newtonian factorized multi-
polar waveforms for spinning, non-precessing black-hole binaries. Phys. Rev. D 83, 064003
(2011). doi:10.1103/PhysRevD.83.064003
77. Pan, Y., Buonanno, A., Boyle, M., et al.: Inspiral-merger-ringdown multipolar waveforms of
nonspinning black-hole binaries using the effective-one-body formalism. Phys. Rev. D 84,
124052 (2011). doi:10.1103/PhysRevD.84.124052
78. Taracchini, A., Pan, Y., Buonanno, A., et al.: Prototype effective-one-body model for non-
precessing spinning inspiral-merger-ringdown waveforms. Phys. Rev. D 86, 024011 (2012).
doi:10.1103/PhysRevD.86.024011
79. Damour, T., Nagar, A., Dorband, E.N., Pollney, D., Rezzolla, L.: Faithful effective-one-body
waveforms of equal-mass coalescing black-hole binaries. Phys. Rev. D 77, 084017 (2008).
doi:10.1103/PhysRevD.77.084017
80. Damour, T., Nagar, A., Hannam, M., Husa, S., Bruegmann, B.: Accurate effective-one-body
waveforms of inspiralling and coalescing black-hole binaries. Phys. Rev. D 78, 044039 (2008)
81. Buonanno, A., Cook, G.B., Pretorius, F.: Inspiral, merger and ring-down of equal-mass black-
hole binaries. Phys. Rev. D 75, 124018 (2007). doi:10.1103/PhysRevD.75.124018
82. Pan, Y., Buonanno, A., Baker, J.G., et al.: A data-analysis driven comparison of analytic and
numerical coalescing binary waveforms: nonspinning case. Phys. Rev. D 77, 024014 (2008).
doi:10.1103/PhysRevD.77.024014
83. Le Tiec, A., Mroue, A.H., Barack, L., et al.: Periastron advance in black hole binaries. Phys.
Rev. Lett. 107, 141101 (2011). doi:10.1103/PhysRevLett.107.141101
84. Damour, T., Nagar, A., Pollney, D., Reisswig, C.: Energy versus angular momentum in black
hole binaries. Phys. Rev. Lett. 108, 131101 (2012). doi:10.1103/PhysRevLett.108.131101
85. Damour, T., Gourgoulhon, E., Grandclément, P.: Circular orbits of corotating binary black
holes: comparison between analytical and numerical results. Phys. Rev. D 66, 024007 (2002).
doi:10.1103/PhysRevD.66.024007
86. Buonanno, A., Pan, Y., Baker, J.G., et al.: Toward faithful templates for non-spinning binary
black holes using the effective-one-body approach. Phys. Rev. D 76, 104049 (2007). doi:10.
1103/PhysRevD.76.104049
144 T. Damour
87. Buonanno, A., Pan, Y., Pfeiffer, H.P., et al.: Effective-one-body waveforms calibrated to
numerical relativity simulations: coalescence of non-spinning, equal-mass black holes. Phys.
88. Bernuzzi, S., Nagar, A., Zenginoglu, A.: Binary black hole coalescence in the extreme-mass-
ratio limit: testing and improving the effective-one-body multipolar waveform. Phys. Rev. D
83, 064010 (2011). doi:10.1103/PhysRevD.83.064010
89. Barausse, E., Buonanno, A., Hughes, S.A., et al.: Modeling multipolar gravitational-wave
emission from small mass-ratio mergers. Phys. Rev. D 85, 024046 (2012). doi:10.1103/
PhysRevD.85.024046
90. Damour, T., Gopakumar, A.: Gravitational recoil during binary black hole coalescence using
the effective one body approach. Phys. Rev. D 73, 124006 (2006). doi:10.1103/PhysRevD.
73.124006
91. Bernuzzi, S., Nagar, A., Zenginoglu, A.: Binary black hole coalescence in the large-mass-ratio
limit: the hyperboloidal layer method and waveforms at null infinity. Phys. Rev. D 84, 084026
(2011). doi:10.1103/PhysRevD.84.084026
92. Damour, T., Nagar, A., Bernuzzi, S.: Improved effective-one-body description of coalesc-
ing nonspinning black-hole binaries and its numerical-relativity completion. ArXiv e-prints
arxiv:1212.4357 [gr-qc] (2012)
93. Damour, T., Nagar, A.: Final spin of a coalescing black-hole binary: an effective-one-body
approach. Phys. Rev. D 76, 044003 (2007)
94. Damour, T., Nagar, A., Tartaglia, A.: Binary black hole merger in the extreme mass ratio limit.
Class. Quantum Gravity 24, S109 (2007)
95. Gonzalez, J.A., Sperhake, U., Bruegmann, B., Hannam, M., Husa, S.: Total recoil: the maxi-
mum kick from nonspinning black-hole binary inspiral. Phys. Rev. Lett. 98, 091101 (2007).
96. Yunes, N., Buonanno, A., Hughes, S.A., Miller, M.C., Pan, Y.: Modeling extreme mass ratio
inspirals within the effective-one-body approach. Phys. Rev. Lett. 104, 091102 (2010). doi:10.
1103/PhysRevLett.104.091102
97. Yunes, N., Buonanno, A., Hughes, S.A., et al.: Extreme mass-ratio inspirals in the effective-
one-body approach: Quasi-circular, equatorial orbits around a spinning black hole. Phys. Rev.
D 83, 044044 (2011). doi:10.1103/PhysRevD.83.044044
98. Scheel, M.A., Boyle, M., Chu, T., et al.: High-accuracy waveforms for binary black hole
inspiral, merger, and ringdown. Phys. Rev. D 79, 024003 (2009). doi:10.1103/PhysRevD.79.
024003
99. Damour, T., Nagar, A.: The effective one-body description of the two-body problem. In:
Blanchet, L., Spallicci, A., Whiting, B. (eds.) Mass and Motion in General Relativity, Fun-
damental Theories of Physics, vol. 162, pp. 211–252. Springer, Berlin (2011). doi:10.1007/
978-90-481-3015-3_7
100. Buchman, L.T., Pfeiffer, H.P., Scheel, M.A., Szilágyi, B.: Simulations of non-equal mass
black hole binaries with spectral methods. Phys. Rev. D 86, 084033 (2012). doi:10.1103/
PhysRevD.86.084033
101. Damour, T.: Gravitational self force in a Schwarzschild background and the effective one
body formalism. Phys. Rev. D 81, 024017 (2010). doi:10.1103/PhysRevD.81.024017
102. Pan, Y., Buonanno, A., Buchman, L.T., et al.: Effective-one-body waveforms calibrated to
numerical relativity simulations: coalescence of non-precessing, spinning, equal-mass black
holes. Phys. Rev. D 81, 084041 (2010). doi:10.1103/PhysRevD.81.084041
103. Damour, T., Jaranowski, P., Schäfer, G.: Effective one body approach to the dynamics of two
spinning black holes with next-to-leading order spin-orbit coupling. Phys. Rev. D 78, 024009
(2008)
104. Barausse, E., Racine, E., Buonanno, A.: Hamiltonian of a spinning test-particle in curved
spacetime. Phys. Rev. D 80, 104025 (2009). doi:10.1103/PhysRevD.80.104025 [Erratum:
ibid. D 85, 069904 (2012)]
105. Barausse, E., Buonanno, A.: An improved effective-one-body Hamiltonian for spinning black-
106. Nagar, A.: Effective one body Hamiltonian of two spinning black-holes with next-to-next-to-
leading order spin-orbit coupling. Phys. Rev. D 84, 084028 (2011). doi:10.1103/PhysRevD.
84.084028
107. Barausse, E., Buonanno, A.: Extending the effective-one-body Hamiltonian of black-hole
binaries to include next-to-next-to-leading spin-orbit couplings. Phys. Rev. D 84, 104027
(2011). doi:10.1103/PhysRevD.84.104027
108. Damour, T., Nagar, A., Villain, L.: Measurability of the tidal polarizability of neutron stars
in late-inspiral gravitational-wave signals. Phys. Rev. D 85, 123007 (2012). doi:10.1103/
PhysRevD.85.123007
109. Damour, T., Nagar, A.: Effective one body description of tidal effects in inspiralling compact
binaries. Phys. Rev. D 81, 084016 (2010). doi:10.1103/PhysRevD.81.084016
110. Bini, D., Damour, T., Faye, G.: Effective action approach to higher-order relativistic tidal
interactions in binary systems and their effective one body description. Phys. Rev. D 85,
124034 (2012). doi:10.1103/PhysRevD.85.124034
111. Baiotti, L., Damour, T., Giacomazzo, B., Nagar, A., Rezzolla, L.: Analytic modelling of tidal
effects in the relativistic inspiral of binary neutron stars. Phys. Rev. Lett. 105, 261101 (2010).
112. Baiotti, L., Damour, T., Giacomazzo, B., Nagar, A., Rezzolla, L.: Accurate numerical sim-
ulations of inspiralling binary neutron stars and their comparison with effective-one-body
analytical models. Phys. Rev. D 84, 024017 (2011). doi:10.1103/PhysRevD.84.024017
113. Barack, L., Damour, T., Sago, N.: Precession effect of the gravitational self-force in a Schwarz-
schild spacetime and the effective one-body formalism. Phys. Rev. D 82, 084036 (2010).
doi:10.1103/PhysRevD.82.084036
114. Le Tiec, A., Blanchet, L., Whiting, B.F.: The first law of binary black hole mechanics in
general relativity and post-Newtonian theory. Phys. Rev. D 85, 064039 (2012). doi:10.1103/
PhysRevD.85.064039
115. Le Tiec, A., Barausse, E., Buonanno, A.: Gravitational self-force correction to the bind-
ing energy of compact binary systems. Phys. Rev. Lett. 108, 131103 (2012). doi:10.1103/
116. Barausse, E., Buonanno, A., Le Tiec, A.: The complete non-spinning effective-one-body met-
ric at linear order in the mass ratio. Phys. Rev. D 85, 064010 (2012). doi:10.1103/PhysRevD.
85.064010
117. Akcay, S., Barack, L., Damour, T., Sago, N.: Gravitational self-force and the effective-one-
body formalism between the innermost stable circular orbit and the light ring. Phys. Rev. D
86, 104041 (2012). doi:10.1103/PhysRevD.86.104041
118. Blanchet, L., Detweiler, S.L., Le Tiec, A., Whiting, B.F.: High-order post-Newtonian fit of
the gravitational self-force for circular orbits in the Schwarzschild geometry. Phys. Rev. D
81, 084033 (2010). doi:10.1103/PhysRevD.81.084033
119. Damour, T.: The 5PN contributions to EOB potentials. (unpublished) (2010)
120. Foffa, S., Sturani, R.: The dynamics of the gravitational two-body problem at fourth post-
Newtonian order and at quadratic order in the Newton constant. ArXiv e-prints 1206.7087
[gr-qc] (2012)
121. Jaranowski, P., Schäfer, G.: Towards the 4th post-Newtonian Hamiltonian for two-point-mass
systems. Phys. Rev. D 86, 061503 (2012). doi:10.1103/PhysRevD.86.061503
122. Faye, G., Marsat, S., Blanchet, L., Iyer, B.R.: The third and a half post-Newtonian gravita-
tional wave quadrupole mode for quasi-circular inspiralling compact binaries. Class. Quantum
Gravity 29, 175004 (2012). doi:10.1088/0264-9381/29/17/175004
Gravitational Self-Force: Orbital Mechanics
Beyond Geodesic Motion
Leor Barack
Abstract The question of motion in a gravitationally bound two-body system

is a longstanding open problem of General Relativity. When the mass ratio η is
small, the problem lends itself to a perturbative treatment, wherein corrections to the
geodesic motion of the smaller object (due to radiation reaction, internal structure,
etc.) are accounted for order by order in η, using the language of an effective grav-
itational self-force. The prospect for observing gravitational waves from compact
objects inspiralling into massive black holes in the foreseeable future has in the past
15 years motivated a program to obtain a rigorous formulation of the self-force and
compute it for astrophysically interesting systems. I will give a brief survey of this
activity and its achievements so far, and will identify the challenges that lie ahead.
As concrete examples, I will discuss recent calculations of certain conservative post-
geodesic effects of the self-force, including the O(η) correction to the precession
rate of the periastron. I will highlight the way in which such calculations allow us to
make a fruitful contact with other approaches to the two-body problem.
1 Background: The Self-Force Domain of the Two-Body

Problem and Astrophysical Relevance
In Newtonian gravity, the dynamics of a two-body system is extremely simple:

an isolated system of two gravitationally-bound point masses admits two conserved
integrals—the energy and angular momentum—and the resulting motion is precisely
periodic. The corresponding general-relativistic problem is radically more difficult.
In General Relativity (GR), the orbits in a bound binary are never periodic: gravi-
tational radiation removes energy and angular momentum from the system, and the
radiation back-reaction gradually drives the two objects tighter together until they
L. Barack (B)
School of Mathematics, University of Southampton, Southampton SO17 1BJ, UK
e-mail: leor@soton.ac.uk
148 L. Barack
eventually merge. It is testimony to the intricacy of the problem, that this qualitative
picture of radiative damping has been fiercely debated in the literature until well into
the 1960s (with some results famously predicting an energy gain in binary systems
emitting waves!). Work by Hermann Bondi and others in the early 1960s eventually
settled the dispute, and the prevailing view was later fully vindicated with the 1975
Nobel-Prize observation by Hulse and Taylor of radiative decay in the binary pulsar
PSR 1913+16 [1], consistent with Einstein’s quadrupole formula.
The dynamics in the Hulse-Taylor binary is very nearly Newtonian, due to the
relatively large separation between the two components. We now know that the
universe around us abounds with much more strongly gravitating systems, from
pairs of inspiralling white dwarfs in nearby galaxies (likely progenitors of most
Type Ia supernovae) to the dramatic coalescence of supermassive black holes at
cosmological distances—by far the brightest events in the universe by energy output.
In the coming years, direct observations of gravitational waves (GWs) will open
a new window on the universe and allow us to peer deeply into these strong-field
processes, which are largely invisible in electromagnetic spectrum. This is an exciting
prospect, because GWs from coalescing compact binaries and black holes will probe
the hitherto unexplored strong-field, highly-dynamical sector of Einstein’s theory,
where a variety of exotic nonlinear effects manifest themselves. With this prospect
comes the need to have an accurate theoretical model of the two-body dynamics in the
strong-field regime, and a prediction of the emitted gravitational waveforms. These
waveforms are needed not only to allow interpretation of the signals and facilitate
precision tests of GR theory, but also to enable the very extraction of some of the
weaker signals from the noisy detector output.
In general, the description of the nonlinear strong-field dynamics in the binary
system entails a full Numerical-Relativistic (NR) treatment, whereby the Einstein
field equations are formulated as an initial/boundary problem and solved numerically.
Efforts to obtain numerical solutions for (in particular) black hole binaries date back
to work in the 1960s and 1970s [2, 3], but it was not until 2005 that first successful
simulations were performed [4–6]. Today NR codes are capable of tracking the
complicated nonlinear evolution of a spacetime containing two (spinning) black
holes during the final stages of the merger. But NR methods have a limited utility in
situations where the two black holes are far apart, or when one of the components is
much heavier than the other. Each of these two regimes of the two-body problem (see
Fig. 1) is characterized by two greatly separate lengthscales (the distance between
the objects vs. their individual radii of curvature in the former case; the radius of
curvature of the larger black hole vs. that of the smaller one in the latter case),
which is difficult to accommodate in an NR framework due to the high resolution
requirements and long evolution time.
Fortunately, the presence of two separate lengthscales also means that the problem
becomes amenable to a simpler, perturbative treatment. In the first of the above
regimes—at sufficiently large separations—the dynamics is best analyzed using the
tools of post-Newtonian (PN) theory [7], whose roots go as far back as 1938, to the
classical paper by Einstein, Infeld and Hoffman [8]. In PN theory, corrections to
the Newtonian dynamics are incorporated into the equations of motion (essentially)
Gravitational Self-Force: Orbital Mechanics Beyond Geodesic Motion 149
Fig. 1 Schematic representation of the three domains of the binary black-hole problem, with the
corresponding natural computational frameworks. Numerical solutions of the full Einstein equations
are effectual for very close binaries of comparable masses. Widely separated binaries are treated
most efficiently using the tools of post-Newtonian theory (less so when the mass ratio is extreme).
Strongly gravitating binaries of large mass ratios are the realm of self-force theory. Much current
study focuses on comparing the predictions of the different approaches where they interface (dashed
lines)
order by order in the binary separation. PN results are useful (a priori) only when
the binary separation is not too small, but no assumption is made about the masses
of the two objects, which can be kept arbitrary.
The second, so-called extreme mass-ratio regime, is most naturally explored
within the framework of black hole perturbation theory. Here the “zeroth-order”
configuration is that of a test particle moving along a geodesic of the fixed, station-
ary background spacetime of the larger object (say, a black hole). This can then serve
as a basis for a perturbative scheme, whereby corrections due to the finite mass of the
small object (and due also, ultimately, to its internal structure) are included order by
order in the small mass-ratio η. At O(η), the gravitational field of the small object
is a linear perturbation of the background geometry. The back-reaction from this
perturbation gives rise to an effective gravitational self-force (GSF) that gradually
diverts the small object from its geodesic motion. In this picture, it is the GSF that is
responsible for the radiative decay of the orbit. It is also responsible for a variety of
conservative effects arising from finite-η corrections to the background gravitational
potential. The GSF description is useful (a priori) when η is sufficiently small, but
the separation between the two objects need not be large. Indeed, GSF theory covers
precisely the domain accessible to neither NR nor PN: strongly gravitating binaries
of small η.
The basic notion of a “self-force” (aka back-reaction force or braking force)
in a radiating system is an old one, dating many decades back, in the context of
electrodynamics, to the classical works by Lorentz [9] and Dirac [10] on the electron’s
equation of motion. In 1960 DeWitt and Brehme generalized this idea to GR by
150 L. Barack
deriving an equation of motion for an electric charge moving in curved space [11].
Their method involved a careful application of Gauss’s theorem on a thin worldtube
surrounding the particle’s worldline, with an imposition of local energy-momentum
conservation. This work has a fundamental importance also in that it develops the
essential mathematical toolkit underpinning contemporary GSF theory: the calculus
of bi-tensors in curved spacetime.
The failure of the Huygens principle in 3 + 1D curved spacetime means that the
self-interaction effect in GR is fundamentally nonlocal: the self-acceleration for-
mally depends on the entire past history of the particle in question. This represents
a significant departure from the flat-space case, and it brings about significant com-
plications, technical as well as conceptual. The application of the self-force idea to
the motion of a mass particle in GR presents several more complications. Not least
among these is the fact that the very notion of a point mass is ill-defined in GR [12].
A rigorous derivation of the GSF equations of motion cannot, therefore, rely on a
point-mass assumption, as conveniently done in a linear theory like electrodynamics.
For many years, a robust formulation of the GSF remained an open problem in GR
theory.
Interest in the problem grew in the mid 1990s, when it was proposed that the
planned space-based GW detector Laser Interferometer Space Antenna (LISA [13])
could observe signals from the inspiral of compact objects into massive black holes
in galactic nuclei. (The inspiraling objects need to be compact—white dwarfs, neu-
tron stars of stellar-mass black holes—because main-sequence stars will be tidally
destroyed before they can produce an interesting GW signal.) Later work confirmed
that a LISA-like mission should be able to detect hundreds of events, out to cosmo-
logical distances (z ≈ 1) [14]. Dubbed Extreme Mass Ratio Inspirals (EMRIs), these
sources have a unique facility as precision probes of strong-field gravity. In a typ-
ical LISA-band EMRI (a ≈ 10M⊗ /106 M⊗ system), the inspiralling object spends
the last few years of inspiral in a very tight orbit around the massive hole, emitting
some 105 –106 gravitational wave cycles. The inspiral trajectories show extreme ver-
sions of periastron precession, Lense-Thirring precession of the orbital plane, and
other strong-field effects. This complex dynamics is encoded in the GWs, which
then carry a detailed map of the spacetime geometry around the massive hole. It
was calculated, for example, that LISA will be able to measure fractional deviations
as small as 1:1000 in the quadrupole moment of the black hole metric [15], allow-
ing precision tests of GR and setting tight bounds on the parameters of alternative
theories of gravity.
It is a crucial prerequisite for realizing this science potential that accurate the-
oretical templates of the inspiral waveforms are at hand. This, in turn, requires a
detailed understanding of the radiative evolution. In a typical LISA-band EMRI, the
GSF drives the orbital decay over a timescale of months, and it dephases the orbit
over mere hours. A useful model of the long-term orbital phase evolution therefore
ought to account properly for GSF effects, which one must be able to calculate for
generic (eccentric, arbitrarily inclined) strong-field orbits around a Kerr black hole
of arbitrary spin.
2 The GSF Program: Foundation
The EMRI problem provided an important impetus to rapid progress in GSF research.
In 1997 Mino et al. [16] derived (what has since been “canonized” as) the formal
expression for the O(m 2 ) GSF [that is, the O(m) self-acceleration] acting on a
particle of mass m moving in an arbitrary vacuum spacetime of characteristic radius
of curvature R ◦ m. The expression was derived using two methods (a third,
axiomatic method was presented around the same time by Quinn and Wald [17],
leading to the same result). The first method was a direct application of DeWitt and
Brehme’s analysis to the gravitational case, assuming from the onset that the particle
can be represented as a point source to the perturbed Einstein equations. The second
method removed this assumption: the small object was taken to be a (Schwarzschild)
black hole, and its representative “worldline” on the background spacetime was
defined and derived using the procedure of matched asymptotic expansions.
In its general form, matched asymptotic expansions is a common tool in physics
for studying problem involving multiple scales (most relevantly, it has been applied
in PN theory [18]). In the particular implementation of [16], the equation of motion is
obtained by matching together two series representations of the metric: a “far field”
expansion in m/r (where r is a suitable measure of spatial distance from the small
black hole), and a “near” field expansion in r/R. The first expansion treats the field of
the small black hole at r ◦ m as a small perturbation on the external geometry, and
the second expansion accounts for the background curvature at r √ R via the small
tidal deformations it induces on the metric of the small black hole. The assumption
m √ R means there is a “buffer zone” m √ r √ R where both descriptions
apply, and demanding that the descriptions agree in this zone constrains the motion
of the small black hole. Matching at leading order shows that the O(m 0 ) motion
is a geodesic on the background geometry. Matching at the next order gives the
O(m) acceleration of the small black hole on the background geometry, interpreted
as a GSF effect. The accelerated “worldline” is defined from a far-field point of
view, via a suitable limiting process. In subsequent work [19–22], this procedure
was generalized and put on a more mathematically firm footing. For example, the
small object was allowed to possess spin and consist of any form of matter (not
necessarily a black hole). The most elegant and rigorous derivation was presented
by Gralla and Wald [20], whose analysis derives both near and far zones as certain
limits of a single global metric. For a thorough and self-contained review of these
theoretical developments, see Ref. [19].
At first post-geodesic order in the GSF approximation, the equation of motion has
the form
β
mu α ∗α u β = Fself , (1)
where u α is the particle’s four-velocity on the background spacetime, ∗α is a covariant

β
derivative on the background spacetime, and Fself (⇔ m 2 ) is the GSF. The above
β
foundational work gives an expression for Fself in terms of the “tail” part of the
152 L. Barack
physical metric perturbation associated with the point particle—the part arising from
the piece of the Green’s function supported inside (rather than on) the past light-cone
of the source. Roughly speaking, it establishes that it is the back-reaction from the
tail part of the self-field (which is finite) that is responsible for the self-acceleration.
Two comments are in order. First, as is easily seen, the GSF itself is a gauge-
dependent notion, and so is the accelerated trajectory in the background geometry:
an O(m) gauge transformation in the perturbed geometry results in a physically dis-
tinct accelerated trajectory. Thus, a meaningful information about the post-geodesic
motion is contained only in the combination of the GSF and the metric perturbation
(in a particular gauge). The above foundational derivations of the GSF all involve a
specific gauge choice—the so called Lorenz gauge (in which the divergence of the
trace-reversed metric perturbation is set to zero). This is a convenient choice because
(i) it preserves the local isotropic nature of the particle singularity, and (ii) the per-
turbation equations in the Lorenz gauge are fully hyperbolic and admit a well-posed
initial-value formulation. It should not be assumed without a careful examination
that the GSF is meaningful or well defined in any other given gauge (this has been a
source of much confusion and debate in the GSF literature). A gauge transformation
formula for the GSF was derived in Ref. [23], which also proposed some criteria for
admissible GSF gauges. The topic is further developed in Ref. [24].
A second comment is that Eq. (1) is only guaranteed to hold momentarily at each
point along the trajectory. It is quite a separate task to formulate a scheme that
faithfully accounts for the long-term evolution of the orbit. A subtlety is that the
Lorenz gauge condition cannot be imposed consistently when the source’s worldline
is accelerating. Ref. [20] suggested a scheme where the Lorenz-gauge perturbation
equations and the equation of motion (1) are solved as a coupled set in a self-consistent
manner, without actively imposing the Lorenz gauge conditions (a similar scheme of
“gauge relaxation” has been used in PN theory); the gauge violations which would
then occur at O(m 2 ) will presumably be accounted for within a consistent second-
order GSF formulation once this becomes available. In a more recent work, Pound
[25] has used techniques from singular and multiple-scale perturbation theories in
attempt to put the idea of gauge relaxation on sound mathematical footing, but it
seems the issue remains somewhat controversial for now. Stronger consensus is
likely to be reached soon, with the advent of the second-order GSF formulations (see
β
Sect. 6). In any case, we note that a computation of the local GSF Fself will constitute
a necessary input for any ultimate scheme for the long-term evolution of the orbit.
In a 2003 paper [26] Detweiler and Whiting proposed an appealing reinterpretation
of Eq. (1) in terms of geodesic motion in a smooth perturbed spacetime. They showed
β
that the GSF Fself can be interpreted as the back-reaction force from a certain smooth
metric perturbation h αβ R , which, unlike the “tail” field mentioned above, is a vacuum
solution of the linearized Einstein equations. The particle can be thought to be moving
along a geodesic of gαβ + h αβ R , where g
αβ is the background metric. The equation of
motion is reformulated as
m ũ α ∗˜ α ũ β = 0, (2)
where ũ α and ∗˜ are the four-velocity and covariant derivative in gαβ + h αβ R .
A pedagogical review of this construction is given in Ref. [19]. It should be empha-

R is not the particle’s physical metric perturbation (for
sized that the perturbation h αβ
example, it is not causal), but rather a mathematical construct that serves as an effec-
tive potential for the motion. The two alternative formulations of the post-geodesic
motion, Eqs. (1) and (2), are equivalent and both are useful; workers in the field often
flip between the two points of view as necessary to highlight different aspects of the
problem.
3 The GSF Program: Computation
3.1 Mode-Sum Regularization
The above formulation is directly applicable to the EMRI problem, where the “large”
scale R is provided by the mass M of the large black hole. A practical method
for calculating the GSF for EMRI orbits, known as mode-sum regularization was
introduced in 2000 [27], and subsequently became the main working framework
for GSF calculations in black hole spacetimes. The method is an implementation of
the robust formulation discussed above (no extra regularization is introduced), and
we shall give a schematic description of it here. A detailed review can be found in
Ref. [28].
As mentioned above, the GSF can be interpreted as the effective force due to the
α = m∗ αβγ h R , where ∗ αβγ is a suitable deriva-
Detweiler-Whiting R-field: Fself βγ
tive operator defined along the particle’s worldline in the background metric. (The
original mode-sum scheme was formulated in terms of the tail field but we shall
use here the equivalent R-field formulation for simplicity.) The R-field itself can
be obtained from the subtraction h βγ R = h full − h S , where the “full” field is the
βγ βγ
physical (retarded) solution of the linearized Einstein equation sourced by the parti-
cle’s energy-momentum, and the “S”-field (for singular field) is a particular solution
prescribed by Detweiler and Whiting [26]. The fields h full S
βγ and h βγ have the same
R , is a smooth
singular structure near the moving particle, so that their difference, h βγ
≤
(C ) function.
In the mode-sum scheme one essentially performs the above subtraction mode-
by-mode in a multipole expansion, and the GSF is then reconstructed from a sum
over multipole contributions:
≤

α
Fself = m∗ αβγ (h full
βγ )l
− (h S l
βγ ) . (3)
l=0
Here a superscript ‘l’ denotes the l-multipole of the corresponding field (defined, as
usual in black hole perturbation theory, via integrals over two-spheres surrounding
154 L. Barack
the large black hole), summed over azimuthal (‘m’) modes. The advantage of the
multipole decomposition is twofold: First, numerical methods in black-hole perturba-
tion theory are usually based on multipole expansions, so that numerical calculations
normally output individual modal contributions anyway. Second, and more crucially,
βγ ) [or (h βγ ) ] is finite and (piecewise)
each of the individual modal contributions (h full l S l
differentiable at the particle’s location, which makes the subtraction more manage-
able in practice.
Now, as first suggested in Ref. [27], Eq. (3) can be put into a more useful form using
some analytic input. One can analytically study the large-l behavior of the S-field
modes (h βγS )l and their derivatives at the particle, and it turns out that (generically)
the derivatives admit a large-l expansion in 1/l, whose leading term is of O(l). (The
last statement depends somewhat on the gauge, but here we shall ignore this subtlety
for simplicity.) If the first few terms in this expansion are known, one can rewrite
Eq. (3) in the form
≤

α
Fself =m ∗ αβγ (h full
βγ )l
− A α
l − B α
− C α
/l , (4)
l=0
where Aα , B α and C α (“regularization parameters”) depend on the particle’s location

and velocity (and on the background spacetime) but not on l; importantly, it was
≤ αβγ α α α
shown [28, 29] that the residue l=0 ∗ (h βγ ) − A l − B − C /l vanishes
S l
along any geodesic orbit in Kerr spacetime. The regularization parameters were
calculated analytically for generic orbits in Schwarzschild [30] and later for generic
orbits in Kerr [28, 29]. With the regularization parameters given analytically, the
task of computing the GSF along a given (pre-specified) orbit reduces to that of
βγ ) to serve as input in the mode-sum formula (4).
obtaining the full modes (h full l
This is usually done numerically, by solving the suitable set of mode-decomposed

perturbation equations with retarded boundary conditions, sourced by the particle
orbit in question.
Numerical implementations of the mode-sum formula (4) are reviewed in Ref. [28].
Typically, the particle is taken to move on a fixed geodesic orbit, and the perturbation
equations are solved for the corresponding source (the back-reaction effect of the
GSF on the orbit has only recently been accounted for in a numerical simulation—
see Sect. 4 below). The most advanced implementations of mode-sum regularization
are capable of computing the GSF along any (bound) geodesic in Schwarzschild
spacetime—these codes where presented in Refs. [31, 32] (time-domain version)
and [33, 34] (frequency-domain version). There are also calculations in Kerr space-
time [35, 36], but for now they are restricted to the toy model of a scalar-field
self-force. (Shah et al. recently used mode-sum regularization to compute a certain
GSF-related effect on circular equatorial orbit in Kerr spacetime [37], but they have
not computed the GSF itself.) Further advance in mode-sum calculations is repre-
sented by the recent analytic derivation of higher-order regularization parameters
[terms of O(l −2 ) and higher in the large-l decomposition of the singular field at the
particle] by Heffernan et al. [38, 39]. This now helps to accelerate the convergence
of the mode-sum in numerical implementations, leading to much improved precision
in GSF calculations.
3.2 Puncture Method
This alternative computation method has been in development since 2007 [40–42].
The idea here is to “regularize” the field equations themselves, rather than (as in
the mode-mode method) their solutions. This method works best with time-domain
numerical implementations in 2+1 or 3+1 dimensional, and can benefit from recent
advances in numerical method for time-domain evolution of hyperbolic equations
in GR. Other advantages: the method offers a direct route to the Kerr problem (still
a challenge for mode-sum regularization), and it offers a convenient framework
for studies of the orbital evolution under the GSF effect. Following is a schematic
description of the method as applied to a scalar-field analogue model; a fuller review
can be found in Ref. [28].
Let us write the scalar-field equation in the schematic form φ full = S, where
is a suitable wave operator (depending on the scalar-field theory), S is a source term
corresponding to a point particle of scalar charge q, and φ full is the sought-for retarded
solution. Let φ R and φ S be the scalar-field analogues of Detweiler–Whiting’s R and
α = q∗
S fields, respectively, so that Fself ˜ α φ R is the scalar-field self-force, with ∗˜ α
a suitable gradient operator. The implementation of the puncture scheme begins
with finding an analytic approximation to φ S , denoted φ P (the “puncture”), with the
property that φ P − φ S and ∗˜ α (φ P − φ S ) both vanish along the particle’s worldline
(the field φ S can be extended globally as convenient). Then the self-force can be
computed via Fself α = q∗ ˜ α φ Res (evaluated at the particle), where the “residual” field
is φ Res := φ − φ . The latter satisfies the “punctured” equation
full P
φ Res = S − φ P := Seff , (5)
where the “effective source” Seff no longer contains a delta function. The residual
non-smoothness of Seff arises from the fact that φ P is only a finite approximation
to φ S (a full expression for φ S is not known in explicit form); one can improve
the smoothness of Seff by designing a “higher-order” input puncture φ P , for which
higher-order derivatives of (φ P −φ S ) also vanish at the particle. High order punctures
for the GSF, and corresponding effective sources, were derived in Refs. [43, 44].
The task of computing the self-force now reduces to solving the field equation
φ Res = Seff with suitable boundary conditions. One usually truncates (or otherwise
attenuates) the support of φ P far from the particle, so that the necessary boundary
conditions are the usual “retarded” ones. Several groups are now engaged in code
development for self-force calculations in the puncture approach. Most work so
far has been confined to the toy model of a scalar field (as a platform for code
development and testing) [43–46], but a first implementation of the GSF has very
156 L. Barack
recently been presented [47] for circular orbits around a Schwarzschild black hole.
Existing puncture codes use a 2+1 (or 3+1)-dimensional grid, which avoids the issue
of separability of the field equations in Kerr spacetime. This makes the extension of
Schwarzschild codes to Kerr rather straightforward. Indeed, following on from [47],
a first implementation of the GSF in Kerr will be presented shortly in a forthcoming
paper [48].
As mentioned, time-domain codes based on the puncture method provide a nat-
ural platform for studying the orbital evolution under the self-force: rather than
pre-specifying the orbit, one can compute the self-acceleration at any time slice,
then modify the source accordingly “in real time” so as to compute the evolving
orbit in a self-consistent manner (say, using the idea of Lorenz-gauge relaxation).
Unfortunately, current codes are not sufficiently efficient computationally to track
the evolution in an EMRI-relevant system over many orbits. (A first self-consistent
evolution simulation was presented recently for the scalar-field self-force [49], but
computational cost restricts the ability of the code to compute more than a handful
of orbital cycles.) In the past few years, this computational challenge has attracted
some interest from experts in Numerical Relativity, leading to several programs to
develop custom-built advanced numerical techniques for integrating the perturbation
equations with pointlike sources. These include methods based on finite elements
[50], adaptive mesh refinement [51], and hyperboloidal slicing [52]. It is hoped that
this activity will lead to a dramatic improvement in the computational performance
of time-domain GSF codes.
4 Orbital Evolution Under the GSF Effect
An important milestone in the GSF program was reached last year, with a first
computation of the long-term orbital evolution under the full (first order) GSF [34].
This computation was based on a frequency-domain implementation of the mode-
sum method in Schwarzschild spacetime, developed in [33, 35, 36]. Rather than
evolving the orbit in a fully self-consistent manner as described above (which is not
easily achievable in a frequency-domain framework), an approximation was used, in
which the value of the GSF at each point along the evolving orbit is taken to be that
computed along a fixed geodesic tangent to the orbit at that point. This approximation
is a good one in situations where the timescale on which the orbit evolves is much
longer than the effective “memory” time associated with the tail field that produces
the GSF. In an EMRI-relevant strong-field system with η √ 1, the former is expected
to be larger than the latter by an amount of order 1/η ◦ 1 over much of the inspiral;
the approximation will cease to be useful only very near the last stable orbit, where
the adiabatic inspiral transits to a direct plunge and the orbit evolves quickly. Let us
briefly review the method of [34], then present some results for illustration.
The calculation in [34] is a general-relativistic adaptation of the standard method
of “variation of parameters”, or “osculating orbits”, used in celestial mechanics.
In this osculating geodesics approach, the inspiral motion is reconstructed from
a smooth sequence of tangent geodesics. In practice, this amounts to solving evolution

equations for all the orbital elements that characterize the geodesic motion (principal
as well as positional), with the driving force provided by the GSF. The necessary
GSF information is prepared in advance, in the form of a global interpolation formula
based on a dense data grid over the relevant phase space.
Let us give some more detail. Bound geodesics of the Schwarzschild geometry
can be parametrized by their semilatus rectum pM and eccentricity e, defined via
r± = pM/(1 ≡ e), where r = r+ and r = r− are the apastron and periastron
radii, respectively [hereafter (t, r, θ, ϕ) are standard Schwarzschild coordinates on
the background spacetime]. The geodesic motion of a test particle is then described
by [53]
pM
r = r g (t; p, e, χ0 ) = , (6)
1 + e cos[χ (t) − χ0 ]
χ (t)
p 1/2 dχ ∇
ϕ = ϕg (t; p, e, χ0 ) = , (7)
p − 6 − 2e cos(χ ∇ − χ0 )
χ (0)
where χ (t) is a monotonically increasing parameter along the orbit (a relativistic

generalization of mean anomaly), related to t via dχ /dt = ( p − 2 − 2e cos v)(1 +
e cos v)2 ( p−6−2e cos v)1/2 [( p−2)2 −4e2 ]−1/2 /(M p 2 ), with v := χ −χ0 . Without
loss of generality we have assumed that the motion takes place in the equatorial plane
(θ = π/2), and took t (χ0 ) = ϕ(χ0 ) = 0 (i.e., at t = 0 the particle is at periastron
at ϕ = 0); p and e are principal elements, which determine the “shape” of the
orbit, χ0 is a positional element, which describes the orientation of the major axis.
Both principal and positional elements evolve secularly under the effect of the GSF;
the secular evolution of p and e is dissipative, while that of χ0 is conservative—it
describes the precession effect of the GSF. Both principal and positional elements
also exhibit quasi-periodic oscillations.
In the osculating geodesics approach, the inspiral motion is described by r =
r g (t; p(t), e(t), χ0 (t)) and ϕ = ϕg (t; p(t), e(t), χ0 (t)), where p(t), e(t), χ0 (t) are
called osculating elements. The rate of change of these elements is determined from
α per unit m) of the tangent geodesic. Evolution
the local self-acceleration (i.e., Fself
formulas for the osculating elements, given the GSF, where obtained in Refs. [54]
(Schwarzschild case) and [55] (Kerr case). These formulas require as input the func-
α (χ −χ ; p, e). In the implementation of Ref. [34] this function was obtained
tion Fself 0
from numerical GSF data computed along a sample of 1,100 geodesics covering the
p, e parameter space. A suitable interpolation model was derived, based on a Fourier
representation of the χ -dependence and a power-law series ansatz for the p, e depen-
dence. With this GSF input at hand, the evolution equations for { p(t), e(t), χ0 (t)}
were then solved numerically starting from some initial values. An example is shown
in Fig. 2.
158 L. Barack
Fig. 2 Evolution of the osculating elements in a sample case with m = 10M⊗ and M = 106 M⊗ .
We show the eccentricity e (red, left axis) and periastron phase χ0 (blue, right axis) as functions of
semi-latus rectum p, as the binary inspirals from ( p, e) = (12, 0.2) down to the innermost stable
orbit (dashed curve). Marks along the curves count down (from right to left) 500 days, 100 days,
10 days, 1 day and 1 h to the onset of plunge. Note the orbit initially circularizes, but near the plunge
the eccentricity begins to increase. Note also the phase χ0 decreases monotonically, implying that the
conservative GSF acts to reduce the rate of relativistic precession. The upper inset is an enlargement
of the near-plunge region; the manifest oscillatory behavior is due to the variation of the GSF with
the radial phase. The lower inset shows the magnitude of the adiabaticity parameter α := ⇒| ṗ/ p|∈T
(the average is over a radial period T ) versus the distance ε = p − 6 − 2e to the innermost stable
orbit, confirming that the evolution is strongly adiabatic until very near the end. (Graphics from
Ref. [34].)
5 Gauge Invariant Conservative Effects and Comparison

with Other Methods
In the last few years GSF results have been used in a variety of applications going
beyond the original EMRI program. GSF data can be used to compute gauge-invariant
“observables” that describe post-geodesic corrections to the gravitational potential in
the two-body system. These can then be utilized as reference points for comparison
with the predications of PN theory and with results from full NR simulations. Our
current knowledge allows us to go only one order beyond the geodesic approximation
[i.e., to O(η)], but at this order the computed corrections are exact, and they give
us a direct and hitherto unavailable handle on the very-strong-field conservative
dynamics.
What do we mean by “conservative” dynamics? The precession effect already
mentioned is an example. More generally, the GSF can be split in a unique way into
dissipative and conservative bits. The dissipative bit is obtained, for example, by
replacing h full
βγ (the retarded metric perturbation) in Eq. (4) with the “half retarded
minus half advanced” metric perturbation, and the conservative piece is similarly
obtained from the “half retarded plus half advanced” perturbation. (This decompo-
sition bears on time-symmetry and not on secularity: in general, both dissipative
and conservative pieces of the GSF would have secular effects on the orbit.) In the
simple case of (quasi)circular motion in the equatorial plane of a Kerr black hole,
the dissipative piece is given by the coordinate components Ftself and Fϕself (related
to dissipation rate of energy and angular momentum, respectively), and the con-
servative piece is given by Fselfr . In more general cases it is still straightforward to
construct the two pieces of the GSF separately in practice, either by obtaining both
retarded and advanced solutions of the perturbation equations, or (more economi-
cally, as described in [28]) by exploiting the time-symmetry of bound geodesics in
Kerr spacetime. Indeed, what makes the communication between GSF and PN so
natural is the fact that in both approaches (and unlike in NR) the conservative and
dissipative aspects of the dynamics can each be studied easily in isolation (this is true
at least at first post-geodesic order in the GSF approximation, and through several
orders in the PN expansion). In any case, to return to our question, what we mean by
conservative dynamics is described by solutions to the equation of motion (1), with
the GSF on the right-hand side replaced by its conservative piece (or, equivalently,
with the dissipative piece “turned off”).
A first gauge-invariant “observable” was proposed by Detweiler in 2005 [56].
Considering the effect of the conservative GSF on circular orbits in Schwarzschild
spacetime, it is easy to see that both components u t and u ϕ of the particle’s four-
velocity u α (on the background spacetime) are invariant through O(m) under O(m)
gauge transformations that respect the helical symmetry of the perturbed spacetime.
The combination Ω := u ϕ /u t , which is the “observable” t-frequency of the per-
turbed orbit, is obviously also invariant. Detweiler proposed to utilize the O(m)
piece of the function u t (Ω)—let us denote it u t1 (Ω)—as a concrete gauge-invariant
measure of the post-geodesic effect. [A simple calculation shows that u t1 (Ω) does
not actually involve the GSF itself; rather, it is constructed from the scalar contrac-
R u α u β , where h R , recall, is the Detweiler-Whiting R field.] Detweiler also
tion h αβ αβ
suggested an interpretation of u t1 (Ω) as a measure of the GSF correction to the grav-
itational redshift along a line of sight perpendicular to the orbital plane (but note this
interpretation is subtle: the actual redshift from the point particle obviously diverges;
rather, it is the redshift defined in the nonphysical effective metric gαβ + h αβ R that
this interpretation alludes to). Detweiler used u t1 (Ω) as a first contact point with PN
theory, showing that the predictions from perturbation theory agree with PN formulas
in the weak field [57]. This impressive comparison was pushed to higher-order in the
PN expansion in subsequent work [58], also showing how by fitting to GSF data one
can derive numerical values for [the O(m) pieces of] higher-order, yet unknown PN
coefficients. The quantity u t1 (Ω) also served a reference point in a first comparison
of GSF calculations carried out in different gauges [59]. Finally, in [60] the notion of
redshift variable was generalized to eccentric orbits (using certain orbital averages);
preliminary comparison with PN calculations in the eccentric case show a very good
agreement [61].
160 L. Barack
In 2009 Barack and Sago computed the conservative GSF-induced shift in the
frequency of the ISCO of a Schwarzschild black hole, using GSF analysis of slightly
perturbed circular orbits [62]. They found, in fractional terms,

ΔΩ
= (0.25101546 ± 0.00000005) × η, (8)
Ω isco
where the uncertainty is due to the finite numerical accuracy of the GSF computa-
tion. (We cite here the higher precision value obtained more recently in [63], and we
account for a certain gauge correction introduced by Damour in [64], which “regular-
izes” the Lorenz-gauge time coordinate, known to be otherwise non-asymptotically-
flat [65].) This was arguably a first concrete physical result, with a clear physical
interpretation, to have emerged from the GSF program. Its importance was in that
it provided a long-sought benchmark in the strong field. The ISCO shift result was
immediately used as an accurate reference point in an exhaustive study of the per-
formance of various PN methods [66]. It was also used to inform an “empirical”
formula (based also on results from NR and PN) for the remnant masses and spins in
binary black hole mergers [67], and to constrain some of the analytical parameters
of the Effective One Body (EOB) potential [64].
The latter work, especially, highlighted the promise of a synergy between the GSF
and other approaches. EOB was introduced by Buonnano and Damour in 1999 [68] as
an analytical framework for modelling the two-body dynamics across all mass ratios
(see T. Damour’s contribution in this volume). At the heart of EOB is an effective
one-body Hamiltonian, whose form is chosen to reproduce the known results at the
test-particle limit, as well as all known PN results. The EOB Hamiltonian includes
a number of “calibration” functions that can be adjusted by available NR data—and
now also based of GSF information as it becomes available. In [64] Damour made
the point that GSF results are particularly useful for calibrating EOB theory (even
more so than NR data) given their accuracy, cleanness, and the fact that conservative
effects can be computed separately. In this way, GSF calculations, whose validity is
a priori restricted to the extreme mass ratio regime, can indirectly contribute to the
development of a universal model of the two-body dynamics across all mass ratios.
In more recent work, Barack and Sago computed the GSF correction to the peri-
astron precession of eccentric orbits around a Schwarzschild black hole [60]. In
the limit of zero eccentricity the result is gauge invariant, and can be used to test
the GSF prediction against that of PN theory in the weak field regime. This indeed
was done in Ref. [69], where the precession results where also used to improve the
calibration of EOB. In a recent culmination of this effort, a four-way collaboration
between groups working on NR, PN, EOB and GSF presented a complete compari-
son between the predictions of all these methods, using the relativistic precession as
a reference point [70]. The results of this study suggest, remarkably, that GSF theory
may be applicable well beyond its natural extreme-mass-ratio domain. See Figs. 3
and 4 for an illustration.
In parallel, there has been progress in utilizing Detweiler’s redshift variable u α1
for EOB studies. This followed from a new formulation by Le Tiec and collaborators
Fig. 3 The relativistic periastron advance per orbit, K , for a close binary of mass ratio η =
m:M = 1:8. As independent variable we use the invariant frequency Ωϕ , given here in units of
inverse-total-mass. (Note the periapsis advances by as much as half a cycle per radial period in this
extreme regime of GR, corresponding to separations of just a few Schwarzschild radii.) The results,
extracted from Ref. [70], show a comparison between the predictions of all methods available today.
The shaded region comes from full NR simulations (with error margins), and the lower panel shows
the relative difference between the predictions of each approximation method and the NR data. The
curve labelled ‘Schw’ is the test-particle (geodetic) result, given for reference. ‘GSFη’ refers to the
standard GSF prediction, whereas in ‘GSFν’ the mass ratio η has been replaced with the symmetric
mass ratio ν ⊥ m M/(m + M)2 [this replacement is “allowed” since ν = η through O(η)]. ‘PN’
is the best available (3PN) PN result, and ‘EOB’ is a certain EOB model (see [70] for details). The
GSF approximation, with the replacement η ⇐ ν, seems to perform remarkably well even though
the mass ratio is not very extreme
[71] of a general “first law of binary black hole mechanics”, relating infinitesimal
variations of the total (invariant) energy and angular momentum of the binary system
to variations of the individual rest masses—a relation which turned out to involve
the redshift variable. Further work [72] then related the redshift variable to the EOB
potentials, in a way that established a new useful link between GSF data and the EOB
functions, leading to a complete determination of two of the main EOB functions at
O(m). Most recently [63], this analysis was extended to the very strong-field regime
below the ISCO and down to the “light-ring” at r = 3M, revealing interesting new
features of the EOB potentials.
6 Outlook
The primary ambition of the EMRI program (in the “experimental” context of low-
frequency gravitational-wave astronomy) is to obtain a faithful model of the long-
term orbital evolution in—and emitted gravitational waves from—compact-object
162 L. Barack
Fig. 4 Same as in Fig. 3, now for η = 1 (equal masses). Even though this system is a priori well
outside the reach of perturbation theory, the GSFν prediction does extremely well, outperforming
all other approximation methods. It remains to be seen whether this remarkable agreement is merely
fortuitous or representative of a more general behavior
binaries of small mass ratios (η ≈ 10−4 –10−9 ), allowing both objects to spin. We
are still quite far from achieving this goal. We have accomplished the important
prerequisite of being able to compute the leading-order self-acceleration of orbits
which are not evolving, in the Schwarzschild background case. But we are yet to learn
how to extend this to orbits in Kerr, how to consistently evolve the orbit under the
GSF effect, and how to incorporate the second-order GSF and the small object’s spin
in our calculations (both will be necessary in order to achieve the phase accuracy
needed for LISA applications). At the same time, we have learned to appreciate
how GSF calculations provide us with new, “high-fidelity” quantitative information
about the dynamics in two-body systems, and we are beginning to learn how this
can be used to test the faithfulness of PN calculations and inform the development
of a universal EOB model. In that, GSF calculations are proving useful far beyond
their original motivation. As the field matures, a wider range of applications become
apparent. There has been recent work to explore the role of GSF in high-energy black
hole scattering (of hypothetical relevance in experimental particle physics) [73, 74],
and other applications are foreseeable.
Below we give a brief summary of (what we consider to be) the main challenges
that lie ahead in the GSF field.
Foundational issues:—The first-order GSF is well understood at the foundational
level, but the situation is less clear at second order. There currently exist at least two
independent formulations of the second-order equation of motion, one by Pound
[75, 76] and another by Gralla [77]. Both use (variants of) matched asymptotic
expansions but each chooses to represent the motion in a rather different way,
making a direct comparison nontrivial. It is important to understand the relation

between the two formulations, and whether they are compatible. It is also impor-
tant to translate (either or both) of the second-order formulations into a practical
computational scheme for the GSF in Kerr geometry—perhaps akin to mode-sum
regularization or to the puncture scheme. There is some initial work in this direction,
but this problem is likely to remain an important research front for the coming years.
Tightly coupled to the question of a valid second-order formulation is the issue of
long-term orbital evolution. Two-timescale expansion methods have been invoked
[25, 78] to suggest an evolution scheme and control its error, but it is not yet clear
how they would preform in practice (for example, how “runaway” self-acceleration
terms would behave in a self-consistent evolution).
Generic inspirals in Kerr experience resonant episodes, where the ratio of the
radial and longitudinal orbital frequencies crosses a low-order rational value, and the
otherwise ergodic orbit becomes (quasi-)periodic. During resonant epochs radiation
reaction acts on quite a different timescale (because the usual “averaging” effect of
ergodicity is lost), leading effectively to a sudden jump in the values of the principal
orbital elements, and a “resetting” of the orbital phases. If the goal is to obtain
accurate phase-coherent waveforms for EMRI systems, it is important to derive an
accurate model of the resonant crossing. Some recent work began to address this
problem [79–81].
Computational issues:—A high priority task for the community is to extend
existing methods and working codes for GSF calculations in Schwarzschild to a
Kerr background. There are two main avenues of approach to the Kerr problem. In
the time-domain approach one computes the metric perturbation due to the particle
by evolving the linearized Einstein equations in the Lorenz gauge on a time-spatial
grid using a finite-difference scheme. The 1 + 1D (time-radial) treatment that works
so well in the Schwarzschild case is no longer useful in Kerr, because the Lorenz-
gauge perturbation equations in Kerr are not separable (in any known way) into
individual multipole modes in the time-domain. Instead, one has to work in 2 + 1D
or 3 + 1D. This can be done using the puncture method described above, but so
far implementations have been restricted to Schwarzschild (refraining from a 1+1D
decomposition for the sake of preparing the ground for a Kerr implementation), or
to a scalar field. One of the major technical obstacles in moving on to Kerr is the
treatment of the “non-radiative” piece of the metric perturbation (the piece which
reduces to the monopole and dipole modes in the Schwarzschild case). This piece
has a numerically important contribution to the GSF, but so far attempts to compute
it via time-domain evolution have been futile due to numerical instabilities. A few
ad-hoc solutions to this problem were suggested recently in Ref. [47] (including the
use of a judiciously chosen generalized Lorenz-gauge in an intermediate step), but
they are yet to be tested in the Kerr case.
The alternative approach is based on a frequency domain treatment, in which the
metric perturbation is solved for mode-by-mode in a multipole-Fourier decomposi-
tion. The advantage is that one now only deals with ordinary differential equations,
but the method is only applicable for bound orbits of sufficiently small eccentric-
ities. The problem of seperability in Kerr remains if one insists on working in the
164 L. Barack
Lorenz gauge, but there are suggestions to overcome this using a specially designed
gauge in which (1) the metric perturbation is reconstructable from curvature scalars,
which obey fully separable equations in Kerr (Teukolsky’s formalism); and (2) the
GSF is still mathematically well-defined and physically meaningful, as it is in the
Lorenz gauge. A proposal of this kind was put forward long ago in [23], and a variant
thereof is being under active development by Shah and collaborators [37, 82, 83].
This approach offers a computationally efficient route to the GSF, but much further
development is needed.
Synergy with other methods:—The initial work described in Sect. 5 bears the
promise of much further fruitful exchange between GSF and other computational
approaches to the two-body problem, exploiting new GSF data as they become
available. One of the challenges is to devise computable gauge-invariant quanti-
ties to facilitate a common language between the various approaches. Second-order
GSF results, when at hand, will allow us to refine our comparisons and constrain the
EOB potentials with exquisite accuracy. It is important to understand if and why the
“GSFν” always provides a good (how good?) approximation even when the mass
ratio is not extreme. What are the aspects of the dynamics in equal-mass binaries
that can be modelled faithfully using purely perturbative methods?
GSF/PN comparison so far has been limited to circular or slightly eccentric
orbits. There is merit in extending this comparison to fully eccentric orbits, which
can be done using the available GSF results in Schwarzschild. Here are some spe-
cific invariant quantities that should be accessible (at least in principle) to existing
Schwarzschild GSF codes, and could serve at contact points with—and strong-field
benchmarks for—other approaches: (1) the GSF correction to the angular momentum
and azimuthal frequency of a marginally-bound orbit on the capture threshold [64];
(2) the GSF correction to the Lyapunov exponent of unstable circular orbits below
the ISCO; (3) the GSF correction to the function relating the azimuthal and radial
frequencies on the “singular curve” identified in Appendix A of Ref. [60], where
the transformation to orbital frequencies as system parameters becomes singular.
Once GSF results in Kerr are available, one could compute the GSF correction to
the Lense–Thirring precession of the orbital plane, and other spin-related effects of
the GSF.
Acknowledgments This work was supported by the European Research Council under grant No.
304978; and by STFC in the UK through grant number PP/E001025/1.
References
1. Hulse, R.A., Taylor, J.H.: Discovery of a pulsar in a binary system. Astrophys. J. 195, L51
(1975). doi:10.1086/181708
2. Hahn, S.G., Lindquist, R.W.: The two-body problem in geometrodynamics. Ann. Phys. (N.Y.)
29, 304 (1964). doi:10.1016/0003-4916(64)90223-4
3. Smarr, L.: Space-time generated by computers: black holes with gravitational radiation. Ann.
N.Y. Acad. Sci. 302, 569 (1977). doi:10.1111/j.1749-6632.1977.tb37076.x
4. Pretorius, F.: Evolution of binary black-hole spacetimes. Phys. Rev. Lett. 95, 121101 (2005).
6. Baker, J.G., Centrella, J., Choi, D.I., Koppitz, M., van Meter, J.: Gravitational-wave extraction
7. Blanchet, L., Gravitational radiation from post-Newtonian sources and inspiralling compact
binaries. Living Rev. Relativ. 9, lrr-2006-4 (2006). http://www.livingreviews.org/lrr-2006-4
8. Einstein, A., Infeld, L., Hoffmann, B.: The gravitational equations and the problem of motion.
Ann. Math. 39, 65 (1938). doi:10.2307/1968714
9. Lorentz, H.A.: Theory of Electrons (1915) 2nd ed. Dover, New York (1952)
10. Dirac, P.A.M.: Classical theory of radiating electrons. Proc. R. Soc. Lond. Ser. A 167, 148
(1938). doi:10.1098/rspa.1938.0124
11. DeWitt, B.S., Brehme, R.W.: Radiation damping in a gravitational field. Anna. Phys. 9, 220
(1960). doi:10.1016/0003-4916(60)90030-0
12. Geroch, R., Traschen, J.: Strings and other distributional sources in general relativity. Phys.
13. LISA Project Office. http://lisa.nasa.gov
14. Gair, J.R., Barack, L., Creighton, T., et al.: Event rate estimates for lisa extreme mass ratio
capture sources. Class. Quantum Grav. 21, S1595 (2004). doi:10.1088/0264-9381/21/20/003
15. Barack, L., Cutler, C.: Using lisa extreme-mass-ratio inspiral sources to test off-kerr devia-
tions in the geometry of massive black holes. Phys. Rev. D 75, 042003 (2007). doi:10.1103/
PhysRevD.75.042003
16. Mino, Y., Sasaki, M., Tanaka, T.: Gravitational radiation reaction to a particle motion. Phys.
17. Quinn, T.C., Wald, R.M.: Axiomatic approach to electromagnetic and gravitational radiation
reaction of particles in curved spacetime. Phys. Rev. D 56, 3381 (1997). doi:10.1103/PhysRevD.
56.3381
18. Blanchet, L., Damour, T.: Hereditary effects in gravitational radiation. Phys. Rev. D 46, 4304
(1992). doi:10.1103/PhysRevD.46.4304
19. Poisson, E., Pound, A., Vega, I.: The motion of point particles in curved spacetime. Living Rev.
Relativ. 14(7), lrr-2011-7 (2011). http://www.livingreviews.org/lrr-2011-7
20. Gralla, E., Wald, R.M.: A rigorous derivation of gravitational self-force. Class. Quantum Grav.
25, 205009 (2008). doi:10.1088/0264-9381/25/20/205009
21. Pound, A.: Self-consistent gravitational self-force. Phys. Rev. D 81, 024023 (2010). doi:10.
1103/PhysRevD.81.024023
22. Harte, A.I.: Mechanics of extended masses in general relativity. Class. Quantum Grav. 29,
055012 (2012). doi:10.1088/0264-9381/29/5/055012
23. Barack, L., Ori, A.: Gravitational self-force and gauge transformations. Phys. Rev. D 64, 124003
(2001). doi:10.1103/PhysRevD.64.124003
24. Gralla, S.E.: Gauge and averaging in gravitational self-force. Phys. Rev. D 84, 084050 (2011).
doi:10.1103/PhysRevD.84.084050
25. Pound, A.: Singular perturbation techniques in the gravitational self-force problem. Phys. Rev.
D 81, 124009 (2010). doi:10.1103/PhysRevD.81.124009
26. Detweiler, S., Whiting, B.F.: Self-force via a green’s function decomposition. Phys. Rev. D 67,
024025 (2003). doi:10.1103/PhysRevD.67.024025
27. Barack, L., Ori, A.: Mode sum regularization approach for the self-force in black hole spacetime.
28. Barack, L.: Gravitational self-force in extreme mass-ratio inspirals. Class. Quantum Grav. 26,
213001 (2009). doi:10.1088/0264-9381/26/21/213001
29. Barack, L., Ori, A.: Gravitational self-force on a particle orbiting a Kerr black hole. Phys. Rev.
166 L. Barack
30. Barack, L., Mino, Y., Nakano, H., Ori, A., Sasaki, M.: Calculating the gravitational self-force
in Schwarzschild spacetime. Phys. Rev. Lett. 88, 091101 (2002). doi:10.1103/PhysRevLett.
88.091101
31. Barack, L., Sago, N.: Gravitational self-force on a particle in circular orbit around a Schwarz-
schild black hole. Phys. Rev. D 75, 064021 (2007). doi:10.1103/PhysRevD.75.064021
32. Barack, L., Sago, N.: Gravitational self-force on a particle in eccentric orbit around a Schwarz-
schild black hole. Phys. Rev. D 81, 084021 (2010). doi:10.1103/PhysRevD.81.084021
33. Akcay, S.: Fast frequency-domain algorithm for gravitational self-force: circular orbits in
Schwarzschild spacetime. Phys. Rev. D 83, 124026 (2011). doi:10.1103/PhysRevD.83.124026
34. Warburton, N., Akcay, S., Barack, L., Gair, J.R., Sago, N.: Evolution of inspiral orbits around a
Schwarzschild black hole. Phys. Rev. D 85, 061501 (2012). doi:10.1103/PhysRevD.85.061501
35. Warburton, N., Barack, L.: Self-force on a scalar charge in Kerr spacetime: circular equatorial
orbits. Phys. Rev. D 81, 084039 (2010). doi:10.1103/PhysRevD.81.084039
36. Warburton, N., Barack, L.: Self-force on a scalar charge in Kerr spacetime: eccentric equatorial
orbits. Phys. Rev. D 83, 124038 (2011). doi:10.1103/PhysRevD.83.124038
37. Shah, A.G., Friedman, J.L., Keidl, T.S.: Extreme-mass-ratio inspiral corrections to the angular
velocity and redshift factor of a mass in circular orbit about a Kerr black hole. Phys. Rev. D
86, 084059 (2012). doi:10.1103/PhysRevD.86.084059
38. Heffernan, A., Ottewill, A., Wardell, B.: High-order expansions of the Detweiler-Whiting
singular field in Schwarzschild spacetime. Phys. Rev. D 86, 104023 (2012). doi:10.1103/
PhysRevD.86.104023
39. Heffernan, A., Ottewill, A., Wardell, B.: High-order expansions of the Detweiler-Whiting
singular field in Kerr spacetime. Phys. Rev. D 89, 024030 (2014). doi:10.1103/PhysRevD.89.
024030
40. Barack, L., Golbourn, D.A.: Scalar-field perturbations from a particle orbiting a black hole
using numerical evolution in 2 + 1 dimensions. Phys. Rev. D 76, 044020 (2007). doi:10.1103/
PhysRevD.76.044020
41. Barack, L., Golbourn, D.A., Sago, N.: m-mode regularization scheme for the self-force in Kerr
spacetime. Phys. Rev. D 76, 124036 (2007). doi:10.1103/PhysRevD.76.124036
42. Vega, I., Detweiler, S.: Regularization of fields for self-force problems in curved spacetime:
foundations and a time-domain application. Phys. Rev. D 77, 084008 (2008). doi:10.1103/
PhysRevD.77.084008
43. Dolan, S.R., Barack, L., Wardell, B.: Self-force via m-mode regularization and 2+1D evolution.
II. Scalar-field implementation on Kerr spacetime. Phys. Rev. D 84, 084001 (2011). doi:10.
1103/PhysRevD.84.084001
44. Wardell, B., Vega, I., Thornburg, J., Diener, P.: Generic effective source for scalar self-force
calculations. Phys. Rev. D 85, 104044 (2012). doi:10.1103/PhysRevD.85.104044
45. Vega, I., Diener, P., Tichy, W., Detweiler, S.: Self-force with (3+1) codes: a primer for numerical
relativists. Phys. Rev. D 80, 084021 (2009). doi:10.1103/PhysRevD.80.084021
46. Dolan, S.R., Barack, L.: Self-force via m -mode regularization and 2 + 1D evolution: Founda-
tions and a scalar-field implementation on Schwarzschild spacetime. Phys. Rev. D 83, 024019
(2011). doi:10.1103/PhysRevD.83.024019
47. Dolan, S.R., Barack, L.: Self-force via m-mode regularization and 2+1D evolution: III. Grav-
itational field on Schwarzschild spacetime, ArXiv e-prints arXiv:1211.4586 [gr-qc] (2012)
48. Dolan, S., Barack, L.: Self-force via m-mode regularization and 2+1D evolution: IV. Gravita-
tional field on Kerr spacetime. (In preparation)
49. Diener, P., Vega, I., Wardell, B., Detweiler, S.: Self-consistent orbital evolution of a parti-
cle around a Schwarzschild black hole. Phys. Rev. Lett. 108, 191102 (2012). doi:10.1103/
50. Cañizares, P., Sopuerta, C.F.: Efficient pseudospectral method for the computation of the self-
force on a charged particle: circular geodesics around a Schwarzschild black hole. Phys. Rev.
D 79, 084020 (2009). doi:10.1103/PhysRevD.79.084020
51. Thornburg, J.: Adaptive mesh refinement for characteristic grids. Gen. Rel. Grav. 43, 1211
(2011). doi:10.1007/s10714-010-1096-z
52. Zenginoglu, A.: Hyperboloidal layers for hyperbolic equations on unbounded domains. J.
Comput. Phys. 230, 2286 (2011). doi:10.1016/j.jcp.2010.12.016
53. Darwin, C.: The gravity field of a particle. Proc. R. Soc. Lond. Ser. A 249, 180 (1959). doi:10.
1098/rspa.1959.0015
54. Pound, A., Poisson, E.: Multiscale analysis of the electromagnetic self-force in a weak gravi-
tational field. Phys. Rev. D 77, 044012 (2008). doi:10.1103/PhysRevD.77.044012
55. Gair, J.R., Flanagan, É., Drasco, S., Hinderer, T., Babak, S.: Forced motion near black holes.
56. Detweiler, S.L.: Perspective on gravitational self-force analyses. Class. Quantum Grav. 22,
S681 (2005). doi:10.1088/0264-9381/22/15/006
57. Detweiler, S.: Consequence of the gravitational self-force for circular orbits of the Schwarz-
schild geometry. Phys. Rev. D 77, 124026 (2008). doi:10.1103/PhysRevD.77.124026
58. Blanchet, L., Detweiler, S., Le Tiec, A., Whiting, B.F.: Post-newtonian and numerical calcu-
lations of the gravitational self-force for circular orbits in the Schwarzschild geometry. Phys.
Rev. D 81(064004), 2010 (2010). doi:10.1103/PhysRevD.81.064004.Erratum:ibid.81,084033
59. Sago, N., Barack, L., Detweiler, S.: Two approaches for the gravitational self-force in black
hole spacetime: comparison of numerical results. Phys. Rev. D 78, 124024 (2008). doi:10.
1103/PhysRevD.78.124024
60. Barack, L., Sago, N.: Beyond the geodesic approximation: conservative effects of the grav-
itational self-force in eccentric orbits around a Schwarzschild black hole. Phys. Rev. D 83,
084023 (2011). doi:10.1103/PhysRevD.83.084023
61. Barack, L., Le Tiec, A., Sago, N.: In progress
62. Barack, L., Sago, N.: Gravitational self-force correction to the innermost stable circular orbit
of a Schwarzschild black hole. Phys. Rev. Lett. 102, 191101 (2009). doi:10.1103/PhysRevLett.
102.191101
63. Akcay, S., Barack, L., Damour, T., Sago, N.: Gravitational self-force and the effective-one-
body formalism between the innermost stable circular orbit and the light ring. Phys. Rev. D
86, 104041 (2012). doi:10.1103/PhysRevD.86.104041
64. Damour, T.: Gravitational self-force in a Schwarzschild background and the effective one-body
formalism. Phys. Rev. D 81, 024017 (2010). doi:10.1103/PhysRevD.81.024017
65. Barack, L., Lousto, C.O.: Perturbations of Schwarzschild black holes in the Lorenz gauge:
formulation and numerical implementation. Phys. Rev. D 72, 104026 (2005). doi:10.1103/
PhysRevD.72.104026
66. Favata, M.: Conservative self-force correction to the innermost stable circular orbit: comparison
with multiple post-Newtonian-based methods. Phys. Rev. D 83, 024027 (2011). doi:10.1103/
PhysRevD.83.024027
67. Lousto, C.O., Nakano, H., Zlochower, Y., Campanelli, M.: Statistical studies of spinning black-
68. Buonanno, A., Damour, T.: Effective one-body approach to general relativistic two-body
dynamics. Phys. Rev. D 59, 084006 (1999). doi:10.1103/PhysRevD.59.084006
69. Barack, L., Damour, T., Sago, N.: Precession effect of the gravitational self-force in a Schwarz-
schild spacetime and the effective one-body formalism. Phys. Rev. D 82, 084036 (2010). doi:10.
1103/PhysRevD.82.084036
70. Le Tiec, A., Mroué, A.H., Barack, L., et al.: Periastron advance in black-hole binaries. Phys.
71. Le Tiec, A., Blanchet, L., Whiting, B.F.: First law of binary black hole mechanics in general
relativity and post-Newtonian theory. Phys. Rev. D 85, 064039 (2012). doi:10.1103/PhysRevD.
85.064039
72. Barausse, E., Buonanno, A., Le Tiec, A.: Complete nonspinning effective-one-body metric at
linear order in the mass ratio. Phys. Rev. D 85, 064010 (2012). doi:10.1103/PhysRevD.85.
064010
73. Barausse, E., Cardoso, V., Khanna, G.: Testing the cosmic censorship conjecture with point
particles: the effect of radiation reaction and the self-force. Phys. Rev. D 84, 104006 (2011).
doi:10.1103/PhysRevD.84.104006
168 L. Barack
74. Gundlach, C., Akcay, S., Barack, L., Nagar, A.: Critical phenomena at the threshold of immedi-
ate merger in binary black hole systems: the extreme mass ratio case. Phys. Rev. D 86, 084022
(2012). doi:10.1103/PhysRevD.86.084022
75. Pound, A.: Second-order gravitational self-force. Phys. Rev. Lett. 109, 051101 (2012). doi:10.
76. Pound, A.: Nonlinear gravitational self-force: field outside a small body. Phys. Rev. D 86,
084019 (2012). doi:10.1103/PhysRevD.86.084019
77. Gralla, S.E.: Second-order gravitational self-force. Phys. Rev. D 85, 124011 (2012). doi:10.
1103/PhysRevD.85.124011
78. Hinderer, T., Flanagan, É.: Two-timescale analysis of extreme mass ratio inspirals in Kerr
spacetime: orbital motion. Phys. Rev. D 78, 064028 (2008). doi:10.1103/PhysRevD.78.064028
79. Gair, J., Yunes, N., Bender, C.M.: Resonances in extreme mass-ratio inspirals: asymptotic and
hyperasymptotic analysis. J. Math. Phys. 53, 032503 (2012). doi:10.1063/1.3691226
80. Flanagan, É., Hinderer, T.: Transient resonances in the inspirals of point particles into black
holes. Phys. Rev. Lett. 109, 071102 (2012). doi:10.1103/PhysRevLett.109.071102
81. Flanagan, É., Hughes, S.A., Ruangsri, U.: Resonantly enhanced and diminished strong-field
gravitational-wave fluxes. ArXiv e-prints arXiv:1208.3906 [gr-qc] (2012)
82. Keidl, T.S., Shah, A.G., Friedman, J.L., Kim, D.H., Price, L.R.: Gravitational self-force in a
radiation gauge. Phys. Rev. D 82, 124012 (2010). doi:10.1103/PhysRevD.82.124012
83. Shah, A.G., Keidl, T.S., Friedman, J.L., Kim, D.H., Price, L.R.: Conservative, gravitational
self-force for a particle in circular orbit around a Schwarzschild black hole in a radiation
gauge. Phys. Rev. D 83, 064018 (2011). doi:10.1103/PhysRevD.83.064018
Hamiltonian Formalism for Spinning Black
Holes in General Relativity
Gerhard Schäfer
Abstract A Hamiltonian treatment of gravitationally interacting spinning black

holes is presented based on a tetrad generalization of the Arnowitt-Deser-Misner
(ADM) canonical formalism of general relativity. The formalism is valid through
linear order in the single spins. For binary systems, higher-order post-Newtonian
Hamiltonians are given in explicit analytic forms. A next-to-leading order in spin
generalization is presented, others are mentioned. Comparisons between the Hamil-
tonian formalisms by ADM, Dirac, and Schwinger are made.
1 Introduction
About half a century after Einstein’s invention of general relativity, Hamiltonian

formulations of the theory became available. Dirac developed his formalism in the
years 1958–1959 [1–3], Arnowitt, Deser, and Misner (ADM) in the years 1959–1960
[4–6], for a summary see [7], and Schwinger in 1963 [8]. The three formalisms came
up from quite different action functionals: Dirac used the Einstein action, ADM
the Einstein-Hilbert action in Palatini form, and Schwinger the tetrad-generalized
Palatini-based Einstein-Hilbert action. In the following, the three approaches are
summarized and compared. Often in the article, the units 16ε G = 1 and c = 1 are
applied.
In the context of his constraint dynamics formalism with various types of con-
straints, weakly and strongly vanishing ones, Dirac gave the full Hamiltonian, i.e.
before applying the constraint equations and coordinate conditions, in the form, with
γi denoting partial space-coordinate derivative and i 0 spatial infinity,
G. Schäfer (B)
Friedrich-Schiller-Universität Jena, Theoretisch-Physikalisches Institut, Max-Wien-Platz 1,
07743 Jena, Germany
e-mail: gos@tpi.uni-jena.de
170 G. Schäfer

HD = − d 3 xγi (g −1/2 γ j (gγ i j )) + d 3 x(N H − N i Hi )

−1/2
= − d Si g γ j (gγ ) + d 3 x(N H − N i Hi )
ij
(1)
i0
(correctness of this form within the full non-linear theory was shown only in 1974,
by Regge and Teitelboim [9]), whereas ADM and Schwinger gave the Hamiltonian
in fully reduced forms, i.e. after applying constraint equations and appropriate coor-
dinate conditions,

HADM = d xγi γ j (gi j − νi j gkk ) = d Si γ j (gi j − νi j gkk ),
3
(2)
i0

HS = − d 3 xγi γ j (gγ i j ) = − d Si γ j (gγ i j ). (3)
i0
Though derived under specific coordinate conditions, here the form of the three-
dimensional metric gi j , (i, j = 1, 2, 3), with inverse metric γ i j , is still left general to
better illuminate the general expressions behind. All three surface integrals coincide
under the assumptions made: asymptotic flat spacetimes with coordinates of the
form N = 1 + O(1/r ), gi j = νi j + O(1/r ), and N i = O(1/r ) at spacelike infinity
(r ≈ ⊗) including the first derivatives of the metric functions to decay as 1/r 2 . The
Lagrangian multipliers N ◦ (−g 00 )−1/2 and N i ◦ γ i j g0i are respectively coined
“lapse” and “shift” functions by Wheeler [10]. The clear identification of HADM
with the total energy of the system is one of the merits of ADM in the Hamiltonian
approach to general relativity. The constraint equations read
H = 0 and Hi = 0, (4)
with the Hamilton density of weight one, used by ADM and Dirac,

1 1
H ◦ −g 1/2 R + g g
ik jl ε ε
i j kl
− (gij ε )
ij 2
+ H M (ADM)
g 1/2 2

1 1
= B + γi (g −1/2 γ j (gγ i j )) + 1/2 gik g jl ε i j ε kl − (gi j ε i j )2 + H M (D),
g 2
(5)
where
1 1/2
B◦ g gi j,k glm,n [(γ il γ jm − γ i j γ lm )γ kn + 2(γ ik γ lm − γ il γ mk )γ jn ], (6)
4
or with Schwinger’s weight-two density,
Hamiltonian Formalism for Spinning Black Holes in General Relativity 171
g 1/2 H = Q + γi γ j q i j + q ik q jl λi j λkl − (q i j λi j )2 + g 1/2 H M (S), (7)
where
1 1 1
Q ◦ − q mn γm q kl γn q kl − qln γm q kl γk q mn − q kl γk ln(q 1/2 )γl ln(q 1/2 ), (8)
4 2 2
with λi j = −g −1 (εi j − 21 εgi j ) and q i j = gγ i j , q = g 2 , where g denotes the deter-

minant of gi j and lowering of indices is with gi j , ε = gi j ε i j . The Dirac and ADM
canonical field momentum is given by εi j = −g 1/2 (K i j − K gi j ), with K = γ i j K i j ,
where K i j = −N Φi0j is the extrinsic curvature of a spacelike hypersurface defined
through constant-in-time slice, t = x 0 = const. Φi0j denote Christoffel symbols.
Schwinger’s canonical field momentum λi j is just g −1/2 K i j . The intrinsic curvature
scalar reads R. The expressions gi j d x i d x j and K i j d x i d x j are respectively called
first and second fundamental form of a hypersurface. Both tensors gi j and K i j are
symmetric.
The momentum density of weight one takes the forms
Hi ◦ 2gi j Dk ε jk + H Mi (ADM and D)

= −λlm γi q lm + γi (2λlm q lm ) − γl (2λim q lm ) + H Mi (S), (9)
where Di denotes the three-dimensional covariant space derivative. The given densi-
ties are densities with respect to three-dimensional coordinate transformations. H M
and H Mi are matter densities.
In 1967, DeWitt [11] and more refined later on in 1974, Regge and Teitelboim
[9] explicitly showed that in asymptotically flat spacetimes the Hamiltonian, before
applying constraint equations and coordinate conditions, takes the form

H= d Si γ j (gi j − νi j gkk ) + d 3 x(N H − N i Hi ). (10)
i0
This general Hamiltonian—not yet to be identified with the energy of the system—
delivers all field equations also including those for the lapse and shift functions after
imposing appropriate coordinate conditions.
Undoubtedly, originating from the very useful coordinate choice made by ADM—
one may call it “maximal isotropic” —,
ε ii = 0, 3γ j gi j − γi g j j = 0 or gij = ανij + hijTT , (11)
the ADM formalism became the most often applied canonical formalism. The inde-
ij
pendent field variables therein are εTT and h iTT
j . Both variables are traceless and
tracefree (TT). Already in 1961 Kimura [12] used this formalism for applications.
The Poisson bracket reads
172 G. Schäfer
⎜ ⎟
ν F νG νG ν F
{F, G} = νiTTkl
j − TT kl , (12)
νh i j νεTT
TT kl νh i j νεTT
with
1 γi γ j
νiTTkl
j = (Pil P jk + Pik P jl − Pkl Pi j ), Pi j = νi j − , (13)
2 √2
where 1/√ 2 denotes the inverse Laplacian. The nonlocality of the TT-operator νiTTkl
j
is just the gravitational analogue of the well-known nonlocality of the Coulomb
gauge in the electrodynamics. If Schwinger would have chosen coordinate condi-
tions corresponding to those introduced above (ADM also introduced another set of
coordinate conditions to which Schwinger adjusted),
ij
λii = 0, q i j = ∂νi j + f TT , (14)
a similar simple technical formalism for detailed calculations would have resulted
ij
with the independent field variables λiTT j and f TT . To our best knowledge, only
the paper by Kibble [13] delivers an application of Schwinger’s formalism, apart
from Schwinger himself, namely a Hamilton formulation of the Dirac spinor field
in gravity. Much later in 1978, Nelson and Teitelboim [14] completed the same
task within the tetrad-generalized Dirac formalism [15]. The Poisson bracket in the
Schwinger formalism resembles very much the ADM one.
In terms of the ADM variables, Schwinger’s fundamental field components take
the quite simple form, [16],
1
q i j = (α 2 − h TT h TT )νi j − αh iTT
j + h ik h k j .
TT TT
(15)
2 kl kl
Dirac on the other side had chosen the following coordinate system or gauge, called
“maximal slicing” because of the field momentum condition,
ε = 0, γ j (g 1/3 γ i j ) = 0. (16)
The corresponding independent field variables are
1
ε̃ i j = (ε i j − γ i j ε )g 1/3 , g̃i j = g −1/3 gi j . (17)
3
To leading order linear in the metric functions, the Dirac gauge coincides with the
ADM gauge. The full reduction of the Dirac-form dynamics to the independent
degrees of freedom has been performed by Regge and Teitelboim [9]. The Poisson
bracket reads

ν F νG νG ν F 1 ν F νG
{F, G} = ν̃iklj − + (ε̃ i j g̃ kl − ε̃ kl g̃ i j ) i j , (18)
ν g̃i j ν ε̃ kl ν g̃i j ν ε̃ kl 3 ν ε̃ ν ε̃ kl
with
1 k l 1
ν̃iklj = (ν ν + νil ν kj ) − g̃i j g̃ kl , g̃i j g̃ jl = νil . (19)
2 i j 3
The following quoted results, apart from the last one, are based on the ADM
approach. The first post-Newtonian (1PN) equations of motion, usually called
Einstein-Infeld-Hoffmann equations of motion, were rederived by Kimura in 1961
[12]. In 1974, Ohta et al. [17] derived the 2PN binary equations with shortcomings
corrected by Damour and Schäfer [18] only much later. About the same time, in 1985,
Schäfer succeeded with the dissipative 2.5PN level, [16]. Using for the first time in
post-Newtonian calculations the technique of dimensional regularization, the 3PN
binary Hamiltonian was obtained in 2001 by Damour et al. [19]. Much simpler to
derive was the dissipative 3.5PN level [20]. Based on a post-linear paper by Schäfer
[21], the first post-Minkowskian (1PM) n-body Hamiltonian was achieved in closed
form by Ledvinka, Schäfer, and Bičák in 2008 [22]. The general relativistic Hamilton
dynamics of compact objects with spin (proper rotation) has found several explicit
results. Counting the order of the spin as 1/c (notice the spin of a miximally rotating
black hole reading G M 2 /c), the leading order spin-orbit coupling is decribed by
an 1.5PN Hamiltonian [23], the next to leading order one by a 2.5PN Hamiltonian,
[24, 25], and the next-to-next to leading order one by a 3.5PN Hamiltonain [26].
The leading order radiation damping from spin-orbit coupling is decribed by an 4PN
Hamiltonian [27]. In case of spin(1)–spin(2) coupling, the conservative 2PN Hamil-
tonian is given in [23], the 3PN Hamiltonian by [25, 28], and the 4PN one by [29].
The 4.5PN Hamiltonian of the leading order radiation damping from spin(1)–spin(2)
coupling is given in [30]. Results on the spin(1)–spin(1) coupling are the 2PN and
3PN Hamiltonians for black holes by respectively [23] and [31]. Several leading
higher-order-in-spin Hamiltonians were obtained from Kerr metric considerations,
[32]. Spinning test-particles in the Kerr metric have been treated by Barausse, Racine,
and Buonanno in 2009, [33] using Dirac’s constraint dynamics formalism.
2 Analytic Representation of Binary Black Holes:

The Brill-Lindquist Initial Value Solution
The model used in this article to describe compact objects are Dirac delta functions.
In this section it will be shown that Dirac delta functions can indeed be used to
describe black holes in interaction with each other.
An isolated black hole with mass m is represented through the line element
174 G. Schäfer
⎜ ⎟2
1− Gm
2r c2 Gm 4
ds = −
2
c dt + 1 +
2 2
νi j d x i d x j
1+ Gm
2r c2
2r c2
⎜ ⎟2
1− Gm
2Rc2 Gm 4
=− c dt + 1 +
2 2
νi j d X i d X j , (20)
1+ Gm
2Rc2
2Rc2
when isotropic coordinates x i with r 2 = x i x i and X i with R 2 = X i X i are employed.

⎩ ⎫2
The two coordinate systems are related through the inversion map R = Gm 2c2
/r .
Obviously, the line element shows isometry under this map. For binary black holes,
initially at rest, the metric at that instant of time may have the form
⎬ 2
1− β1 G
− β2 G 4
ds 2 = − ⎭
2r1 c2 2r2 c2  c2 dt 2 + 1 + ω1 G + ω2 G dx2 , (21)
ω1 G ω2 G 2r1 c2 2r2 c2
1+ 2r1 c2
+ 2r2 c2
as shown by Brill and Lindquist in 1963, [34], for the space part of the given metric.
The time part of the given metric has been derived later by Jaranowski and Schäfer
[35]. The coefficients ωa and βa , a = 1, 2 do depend on the masses and the relative
coordinate distance of the two black holes r122 = (x i − x i )(x i − x i ). Furthermore,
1 2 1 2
ra = (x − xa )(x − xa ), where xa are the position vectors of the black holes. Notice,
2 i i i i i
our variables are living in an euclidean space x i which is conformally related with
the physical one. The geometrical interpretation of Brill-Lindquist black holes are
two Einstein-Rosen bridges, both starting in the same physical space but ending in
two different other ones. The energy of the Brill-Lindquist black hole configuration
reads
c4
E AD M = − d Si γi ρ = (ω1 + ω2 )c2 , (22)
2ε G
i0
ω1 G ω2 G
where ρ = 1 + 2r1 c2
+ 2r2 c2
, where Θρ = 0 for x i ∗= x1i , x2i with Θ ◦ √ 2 .
The inversion map of the three-metric of black hole, say 1, at its throat reads r1⇔i =
r1i ω12 G 2 /4c4 r12 , where r1⇔i = x ⇔i − x1i , r1i = x i − x1i , r1 = |x i − x1i |. The
three-metric line element dl 2 takes the forms

ω1 G ω2 G 4 2
dl = ρ dx = 1 +
2 4 2
+ dx
2r1 c2 2r2 c2
4
ω1 G ω2 G
= ρ ⇔4 dx⇔ = 1 + ⇔ 2 1 + dx⇔ ,
2 2
(23)
2r1 c 2r2 c2
ω1 G ω1 ω2 G 2 ω12 G 2 r1⇔i
with ρ ⇔ = 1 + 2r1⇔ c2
+ 4r2 r1⇔ c4
and r2i = 4c4 r1⇔2
+ r12
i , ri = ri − ri = xi − xi .
12 1 2 2 1
Hereof, by definition, the rest-mass of black hole 1 comes out to read

c2 ω1 ω2 G
m1 ◦ − d Si⇔ γi⇔ ρ ⇔ = ω1 + . (24)
2ε G 2r12 c2
i0
The calculations presented above treat the metric functions geometrically as pure
vacuum solution without sources. No divergences occur. Infinities only mean infinite
distances. In the following it will be shown how to reconstruct the Brill-Lindquist
initial value solution with the aid of Dirac delta functions.
Let us have a look at the constraint equations with point-mass sources,

1 j 1 j 16ε G ⎩ 2 2 ⎫1/2
g 1/2 R − ε ij εi − εii ε j = m a c + γ i j pai pa j νa (25)
g 1/2 2 c 3
a
≤ 16ε G ≤
(identical with 2 −gG 00 = c4
−gT 00 in standard notation),
j 16ε G
− 2γ j εi + ε kl γi gkl = pai νa , (26)
c3 a
≤ G≤
(2 −gG i0 = 16ε c4
−gTi0 ). νa is an abbreviation of ν(x i − xai ) with xai the position
vector of mass a ( d 3 xνa = 1). Using the ADM gauge in the form

1 4
gi j = 1 + τ νi j + h iTT j , [3γ j gi j − γi g j j = 0],
8
2 ij
ε ii = 0, or ε ij = γi ε j + γj ε i − νij γk ε k + εTT , (27)
3
ij
j = εTT =
the constraint equations simplify to the following equation, imposing h iTT
pai = 0,

1 16ε G
− 1 + τ Θτ = m a νa . (28)
8 c2 a
With the aid of the ansatz

4G ω1 ω2
τ= + , (29)
c2 r1 r2
after Hadamard partie finie regularization, the constraint equation yields,
ωa ωb G
m a = ωa + , a ∗= b, (30)
2rab c2
176 G. Schäfer
⎬ 

ma + mb c2 rab ⎭ ma + mb ma − mb 2
ωa = m a − + 1+ 2 + − 1 . (31)
2 G c rab /G 2c2 rab /G
The Hamiltonian clearly results in

ω1 ω2
HBL = (ω1 + ω2 ) c2 = (m 1 + m 2 ) c2 − G . (32)
r12
The d-dimensional generalization of the above treatment runs as follows. The

d-metric functions read,
4 1d −2
gi j = ρ d−2 νi j , ρ = 1 + τ,
4d −1
⎜ ⎟
4G Φ ( d−2 2 ) ω1 ω2
τ= 2 d−2
+ d−2 . (33)
c ε d−2 2 r1 r2
The d-dimensional inverse Laplacian Θ−1 takes the form,
Φ ((d − 2)/2) 2−d

− Θ−1 νa = ra , (34)
4ε d/2
where Φ denotes the Euler gamma function. The ansatz for ρ thus reads
⎜ ⎟
G(d − 2)Φ ((d − 2)/2) ω1 ω2
ρ =1+ + (35)
c2 (d − 1)ε (d−2)/2 r1d−2 r2d−2
and the constraint equation takes the form

⎜ ⎜ ⎟⎟
G(d − 2)Φ ((d − 2)/2) ω1 ω2
1+ + ωa νa = m a νa . (36)
c2 (d − 1)ε (d−2)/2 r1d−2 r2d−2
Chosing 1 < d < 2, a fully finite result comes in the form,

⎜ ⎟
G(d − 2)Φ ((d − 2)/2) ωb
1+ ωa νa = m a νa , a ∗= b. (37)
c2 (d − 1)ε (d−2)/2 r12
d−2
Analytic continuation through d = 3 can now be performed without facing diver-

gences.
3 Spin in Minkowski Space
Before entering into gravity in asymptotically flat spacetimes, a discussion of spin in

Minkowski spacetime seems most convenient. Using canonical variables, the total
angular momentum J = (J i ) = (Ji ) of a particle with spin vector Ŝ reads
J = X̂ × P + Ŝ, (38)
where X̂ denotes the canonical position vector and P its canonical linear momentum.
The Lorentz boost is given by
1
K ◦ −tP + G = −tP + HX̂ − Ŝ × P, (39)
H +m
≤
with the center-of-energy vector G. The free-particle Hamiltonian H = m 2 + P2 .
The center-of-energy position vector is given by
1
X̄ = X̂ − Ŝ × P, (40)
(H + m)H
thus, G = H X̄. The center-of-spin vector, or Newton-Wigner position vector,

is defined by X̂. The Poisson brackets of its components vanish { X̂ i , X̂ j } = 0.
The center-of-energy position vector X̄ = X̂ − (H +m)H 1
Ŝ × P has non-vanishing
Poisson brackets of its components and the center-of-inertia position vector X =
X̂ + (H +m)m1
Ŝ × P as well. Using the center-of-inertia position vector, in four-
dimensional language, S μξ Pξ = 0 holds, for the center-of-energy position vector
S̄ μξ n ξ = 0, n μ = (−1, 0, 0, 0) is valid, and for the center-of-spin position vector
m Ŝ μξ n ξ + Ŝ μξ Pξ = 0 happens. The reason for those different spin-supplementary
conditions is rooted in the invariance of the total angular momentum against shift of
coordinates.
For isolated systems, the Poincaré algebra is valid,
{Pi , H} = {Ji , H } = 0, {Ji , P j } = ζi jk Pk ,

{Ji , J j } = ζi jk Jk , {Ji , G j } = ζi jk G k , {G i , H } = Pi ,
1 1
{G i , P j } = 2 H νi j , {G i , G j } = − 2 ζi jk Jk . (41)
c c
G i is not a constant of motion, but K i is,
dK/dt = γK/γt + {K, H } = −P + {G, H } = 0. (42)
For many-particle systems with interaction, it generally holds,

178 G. Schäfer

P= pa , J = (ra × pa + sa ), (43)
a a

M◦ H 2 − P2 , H = M 2 + P2 , (44)
G 1 G
X̂ = + (J − × P) × P, (45)
H M(H + M) H
with
{ X̂ i , X̂ j } = {P i , P j } = 0, { X̂ i , P j } = ν i j ,
{M, X̂ j } = {M, P j } = {M, H } = 0. (46)
⎛
For free particles with spin, one has, H = h a , with h a = m a2 + pa2 , and
⎛ a
G = a (h a ra − h a +m
1
a
sa × pa ).
4 Spin and Gravity: Asymptotic Flat Spacetimes

The treatment of spin in gravity is most conveniently achieved by the introduction
μ μ
of a tetrad field ea (μ = 0, 1, 2, 3; a = 0, 1, 2, 3) having the properties ea ebμ =
δab and eaμ ebξ δ = gμξ = gξμ . Local Lorentz transformations are defined by
ab
⇔μ μ
ea = L ba eb with L ac δab L bd = δcd . The condition of homogeneous transformation
of the spacetime derivative of a physical object τ under local Lorentz transformations
introduces a linear connection ωμ ab ,
1 ab γ
Dμ τ ◦ γμ τ + ωμ G [ab] τ, γμ ◦ μ , (47)
2 γx
with transformation property ωμ ⇔ab = L a L b ωcd + L a γ L bd . The object G

c d μ d μ [ab] is
defined through infinitesimal local Lorentz transformations of τ, ντ = −νκ [ab]
G [ab] τ, with infinitesimal group parameters νκ [ab] .
The curvature tensor Rμξ ab is defined by
Dμ Dξ τ − Dξ Dμ τ = Rμξ
ab
G [ab] τ, (48)
ab
Rμξ = γμ ωξab − γξ ωμ
ab
+ ωξac ωμ
bd
δcd − ωμ ωξ δcd .
ac bd
(49)
The simplest Lagrangian density (apart from the herein trivial cosmological constant)
for the gravitational field reads,
LG = det(eγc )eaμ ebξ Rμξ

ab
(ω) + γμ C μ , (50)
where the exact divergence γμ C μ is needed to render the variational principle valid
for variations with prescribed properties at the boundary of the four-dimensional
integration area,
The vacuum field equations read
νLG ab φ a cb ψ φ
0= μ = det(eγ )(2Rμφ eb − eμ Rψφ ec eb ),
c
(51)
νea
νLG
0= ≡ ωμ
ab
= ωμ
ab
(e, γξ e). (52)
νωμ
ab
The latter equation reduces the gravitational field to the Einsteinian one with sup-
pressed torsion. This property will be kept in the following when treating spinning
sources of the gravitational field.

The matter action W M = d 4 x[L M + LC ] for a spinning classical object can
be put into the form, e.g. [36], where the Lagrangian density of the dynamical part
reads,
⎝ μ
1 dz 1 dπ ab
L M = dτ pμ − Sab ωμab + Sab ν(4) (53)
2 dτ 2 dτ
and where the constraints part is given by

⎝
[i]a λ3 2
LC = dτ λa1 p b Sab + λ2[i] Λ pa − ( p + m ) ν(4) .
2
(54)
2

ν(4) is a four-dimensional Dirac delta function, ν(4) = ν(x μ − z μ ), ( d 4 xν(4) = 1), τ
is a proper time variable, pμ and z μ denote respectively the four-dimensional kinetic
momentum form and position vector of the particle and Sab is its spin tensor. The
angle variables π ab with dπ ab = ΛCa dΛCb = −dπ ba are anholonomic ones. Capital
indices refer to body-fixed Lorentz-frame coordinates. λ1 , λ2 , and λ3 are Lagrangian
multipliers.
The equations of motion resulting hereof reads,
DSab
= 0, (55)
Dτ
Dpμ 1 dz μ
= − Rμψab u ψ S ab , u μ ◦ = λ3 p μ . (56)
Dτ 2 dτ
They completely correspond to the divergence freeness or dynamics of the Tulczyjew
stress-energy tensor for pole-dipole particles,
⎞ ⎠
≤ μξ μ ξ (μ ξ)ω
N gT = dτ λ3 p p ν(4) + u S ν(4) . (57)
||ω
180 G. Schäfer
The index symbol || denotes covariant four-dimensional derivative.
4.1 Hamiltonian for Self-Gravitating Spinning Compact Objects
Variation of the matter action with respect to the Lagrangian multipliers λ1 , λ2 , and
λ3 results in the relations, respectively,
pk γ k j S ji
nSi ◦ n μ Sμi = = gi j nS j , (58)
np
p(i) pa
Λ[ j](0) = Λ[ j](i) , Λ[0]a
= − , (59)
p (0) m

np ◦ n μ pμ = − m 2 + γ i j pi p j , γ ik gk j = ν ij , (60)
where n μ = (1, −N i )/N , n μ = (−N , 0, 0, 0).

To fix the tetrad field, the so-called time gauge (Schwinger’s coining, but intro-
duced by Dirac earlier) proves extremely useful,
μ 1 Ni
e(0) = n μ , i.e. e(0)
0
= , e(0)
i
=− . (61)
N N
Then
(m)
gi j = ei e(m) j . (62)
The matter Lagrangian density is split into three parts, a kinetic part with time
derivative of the matter and field variables, respectively denoted L M K and LG K ,
and a constraint part, L MC ,
L M K , LG K , L MC = −N H matter + N i Hi matter . (63)
The matter energy and momentum densities read,
pi nS j
H matter = −npν − K i j ν − (nS k ν);k , (64)
np

1 mk pk nSl
Hi matter
= ( pi + K i j nS )ν +
j
γ Sik ν + νi(k γ l)m ν . (65)
2 np ;m
The semicolon denotes covariant three-dimensional derivative; later on, the comma
will denote partial derivative. The transformation to canonical matter variables, indi-
cated by hats, is given by
nS i pk γ k j Ŝ ji
z i = ẑ i − , nSi = − , (66)
m − np m
pi nS j p j nSi
Si j = Ŝi j − + , (67)
m − np m − np

[i]( j) [i](k) p(k) p ( j)
Λ = Λ̂ νk j + , (68)
m(m − np)

( j) 1 p(k nS j) kj
pi = p̂i − K i j nS − Â j kl
e( j)k el,i + Sk j + Φ i, (69)
2 np
where
1 mp(i nS j)
gik g jl Âkl = Ŝi j + (70)
2 np(m − np)
and
S ab Sab = Ŝ(i)( j) Ŝ(i)( j) = 2 Ŝ(i) Ŝ(i) = 2s 2 = const, (71)
Λ̂(i)
[k] Λ̂
[k]( j)
= νi j , (72)
(i)
d π̂ (i)( j) ◦ Λ̂[k] d Λ̂[k]( j) = −d π̂ ( j)(i) . (73)
Putting L M K + LG K = LˆM K + LˆG K + (td), one finds,
(i)( j)
LˆM K = p̂i ẑ˙ ν + Ŝ(i)( j) π̂˙
i 1
ν, (74)
2
(k)
LˆG K = Âi j e(k)i e j,0 ν. (75)
Adding the Lagrangian of gravity, LG , results in a new Lagrangian for gravity, also
see Deser and Isham [37],
182 G. Schäfer
(k)
LˆG K + LG = [2ε i j + Âi j ν]e(k)i e j,0 + LGC − Ei,i (76)
with
LGC = −N H field + N i Hi field (77)
and
⎝
1 1⎩ ⎫2
H field = − ≤ gR + gi j ε i j − gi j gkl ε ik ε jl , (78)
g 2
jk
Hi field = 2gi j ε ;k , (79)
as well as,
Ei = gi j, j − g j j,i . (80)
The application of the crucial spatially symmetric time gauge for the tetrads intro-
duced by Kibble, [13],
e(i) j = ei j = e ji , (81)

ei j e jk = gik , ei j = (gkl ) (matrix root!), (82)
reduces the tetrads to the metric functions. The partial derivatives of the tetrads result
in the expressions
(k) 1 (kl)
e(k)i e j,μ = Biklj gkl,μ + gi j,μ , Biklj = B[i j] , (83)
2
where (··) and [··] denote symmetrization and antisymmetrization, respectively, and
γem j γemi
2Biklj = emi − em j . (84)
γgkl γgkl
From the new gravity action the canonical field momentum is easily read off to be
1 (i j)
Â ν + Bkl Â[kl] ν.
ij
εcan
ij
= εij + (85)
2
The ADM spacetime coordinate conditions do take now the following forms,
3gi j, j − g j j,i = 0, εcan

ii
= 0, (86)
gi j = ρ 4 νi j + h iTT
j , εcan = ε̃can + εcan ,
ij ij i jTT
(87)
i jTT
j and εcan ,
with the transverse-traceless objects h iTT
i jTT
h iiTT = εcan
iiTT
= h iTT
j, j = εcan, j = 0, (88)
ij
and longitudinal one ε̃can ,
j 2
ε̃can
ij
= Vcan,
i
j + Vcan,i − νi j Vcan,k .
k
(89)
3
The constraint equations read
H field + H matter = 0, Hi field + Hi matter = 0. (90)
Finally, the total action in canonical form is given by, [36],

⎝
dt p̂i ẑ˙ i + Ŝ(i)( j) π̂˙ (i)( j) − E ,
1
W = d 4 x εcan h i j,0 +
i jTT TT
(91)
2

with E = d Si Ei , and the Hamiltonian reads,
⎡ ⎣
E ◦ HADM = −8 d 3 x Θρ ẑ i , p̂i , Ŝ(i)( j) , h iTT
j , ε i jTT
can , (92)
with Poisson bracket commutation relations
{ẑ i , p̂ j } = νi j , { Ŝ(i) , Ŝ( j) } = i jk Ŝ(k) ,

klTT ⇔ ⇔
j (x, t), εcan (x , t)} = νi j ν(x − x ).
{h iTT TTkl
(93)
5 Post-Newtonian (PN) Hamiltonians

In this section, the spin will be counted of order (1/c)0 and not 1/c as in the Intro-
duction. For non-spinning compact objects, the binary dynamics is known up to the
3.5PN order,
H (t) = m 1 c2 + m 2 c2 + H N + H1P N
+ H2P N + H3P N + · · ·
+ H2.5P N (t) + H3.5P N (t) + · · · , (94)
where the 2.5P N and 3.5P N Hamiltonians are non-autonomous dissipative ones,
[20]. Introducing the following quantities, Ĥ = (H − Mc2 )/μ, μ = m 1 m 2 /M,
184 G. Schäfer
M = m 1 + m 2 , ξ = μ/M with 0 ∇ ξ ∇ 1/4 (test particle case ξ = 0, equal mass

case ξ = 1/4), p = p1 /μ, r = r12 = |x1 − x2 |, pr = (n · p), q = (x1 − x2 )/G M,
and n = n12 = q/|q|, in the center-of-mass frame, p1 + p2 = 0, the following
expressions hold,
p2 1
Ĥ N = − , (95)
2 q
1 1 1 1
c2 Ĥ1P N = (3ξ − 1) p 4 − [(3 + ξ) p 2 + ξpr2 ] + 2 , (96)
8 2 q 2q
1
c4 Ĥ2P N = (1 − 5ξ + 5ξ 2 ) p 6
16
1 1
+ [(5 − 20ξ − 3ξ 2 ) p 4 − 2ξ 2 pr2 p 2 − 3ξ 2 pr4 ]
8 q
1 1 1 1
+ [(5 + 8ξ) p 2 + 3ξpr2 ] 2 − (1 + 3ξ) 3 , (97)
2 q 4 q
1
c6 Ĥ3P N = (−5 + 35ξ − 70ξ 2 + 35ξ 3 ) p 8
128 ⎝
1
+ (−7 + 42ξ − 53ξ 2 − 5ξ 3 ) p 6 + (2 − 3ξ)ξ 2 pr2 p 4
16

1
+ 3(1 − ξ)ξ 2 pr4 p 2 − 5ξ 3 pr6
q
⎝
1 1
+ (−27 + 136ξ + 109ξ 2 ) p 4 + (17 + 30ξ)ξpr2 p 2
16 16

1 1
+ (5 + 43ξ)ξpr4 2
12 q
⎝
25 1 2 335 23 2
+ − + ε − ξ − ξ p2
8 64 48 8

85 3 2 7 1
+ − − ε − ξ ξpr2 3
16 64 4 q
⎝
1 109 21 2 1
+ + − ε ξ 4. (98)
8 12 32 q
To save space, from the dissipative Hamiltonians only the leading one is given, [16],
⎝
2 n i n j d 3 Q̂ i j (tˆ)
c5 Ĥ2.5P N (tˆ) = pi p j − . (99)
5 q d tˆ3
Here Q̂ i j (tˆ) = ξ(q ⇔i q ⇔ j − νi j q ⇔2 /3) and its time derivatives (tˆ = t/G M) are allowed
to be eliminated using the equations of motion. Only after the performance of the
phase-space derivatives, the primed variables are allowed to be identified with the
unprimed ones.
For convenience, the spin-gravity interaction Hamiltonians are given in the non-
center-of-mass frame. To simplify notation, S ◦ Ŝ will be put. The leading order
spin-orbit Hamiltonian reads, spin not counted in terms of 1/c,
⎝
G 1 3m b
1P N
HSO = 2 2
(Sa × nab ) · pa − 2pb . (100)
c a rab 2m a
b∗=a
The leading order spin(1)–spin(2) Hamiltonian takes the form,
G 1
HS1P N
1 S2
= 3
[3(Sa · nab )(Sb · nab ) − (Sa · Sb )] (101)
c2 a 2rabb∗=a
and the leading order spin(1)–spin(1) dynamics is given by, going beyond linear
order in spin,
G m2
HS1P N
1 S1
= 3
[3(S1 · n12 )(S1 · n12 ) − (S1 · S1 )] . (102)
c2 2m 1r12
The next-to-leading order spin-orbit Hamiltonian reads,

⎝ ⎝
G 5m 2 p21 3(p1 · p2 )
HS2P
O
N
= 4 2 − ((p1 × S1 ) · n12 ) 3
+
c r 8m 1 4m 21

3p22 3(p1 · n12 )(p2 · n12 ) 3(p2 · n12 )2
− + +
4m 1 m 2 4m 21 2m 1 m 2
⎝
(p1 · p2 ) 3(p1 · n12 )(p2 · n12 )
+ ((p2 × S1 ) · n12 ) + (103)
m1m2 m1m2
⎞ ⎠
2(p2 · n12 ) 3(p1 · n12 )
+ ((p1 × S1 ) · p2 ) −
m1m2 4m 21
⎝ ⎞ ⎠
G2 11m 2 5m 22
+ 4 3 − ((p1 × S1 ) · n12 ) +
c r 2 m1
⎝
15m 2
+ ((p2 × S1 ) · n12 ) 6m 1 + + (1 ⇒ 2)
2
and the next-to-leading order spin(1)–spin(2) Hamiltonian is given by

186 G. Schäfer
HS2P N
1 S2
= (G/2m 1 m 2 c4 r 3 )[3((p1 × S1 ) · n12 )((p2 × S2 ) · n12 )/2
+ 6((p2 × S1 ) · n12 )((p1 × S2 ) · n12 )
− 15(S1 · n12 )(S2 · n12 )(p1 · n12 )(p2 · n12 )
− 3(S1 · n12 )(S2 · n12 )(p1 · p2 ) + 3(S1 · p2 )(S2 · n12 )(p1 · n12 )
+ 3(S2 · p1 )(S1 · n12 )(p2 · n12 ) + 3(S1 · p1 )(S2 · n12 )(p2 · n12 )
+ 3(S2 · p2 )(S1 · n12 )(p1 · n12 ) − 3(S1 · S2 )(p1 · n12 )(p2 · n12 )
+ (S1 · p1 )(S2 · p2 ) − (S1 · p2 )(S2 · p1 )/2 + (S1 · S2 )(p1 · p2 )/2]
+ (3/2m 21 r 3 )[−((p1 × S1 ) · n12 )((p1 × S2 ) · n12 )
+ (S1 · S2 )(p1 · n12 )2 − (S1 · n12 )(S2 · p1 )(p1 · n12 )]
+ (3/2m 22 r 3 )[−((p2 × S2 ) · n12 )((p2 × S1 ) · n12 )
+ (S1 · S2 )(p2 · n12 )2 − (S2 · n12 )(S1 · p2 )(p2 · n12 )]
+ (6G 2 (m 1 + m 2 )/c4 r 4 )[(S1 · S2 ) − 2(S1 · n12 )(S2 · n12 )]. (104)
Finally, the next-to-leading order spin(1)–spin(1) dynamics reads,

⎝
G m2 3m 2
HS2P N
1 S1
= 3
(p1 · S1 )2 + (p1 · n)2 S21
4 3
c r 4m 1 8m 31
3m 2 2 3m 2 3
− 3
p1 (S1 · n)2 − 3
(p1 · n) (S1 · n) (p1 · S1 ) − p22 S21
8m 1 4m 1 4m 1m2
9 3 9
+ p2 (S1 · n)2 + (p1 · p2 ) S21 − (p1 · p2 ) (S1 · n)2
4m 1 m 2 2 4m 21 4m 21
3
− (p1 · n) (p2 · S1 ) (S1 · n)
2m 21
3 3
+ 2 (p2 · n) (p1 · S1 ) (S1 · n) + (p1 · n) (p2 · n) S21
m1 4m 21

15
− (p 1 · n) (p 2 · n) (S 1 · n)2
4m 21
⎝
G2m2 6m 2   2m 2
− 4 4 5 1+ (S1 · n) − S1 + 4 1 +
2 2
(S1 · n) .2
2c r 5m 1 m1
(105)
Also this Hamiltonian goes beyond linear order in spin. For its derivation an extension
of the Tulczyjew stress-energy tensor for pole-dipole particles was needed, [31].
As further examples, the next-to-leading order spin-orbit center-of-energy vector is
given,
pa2
2P N
GSO =− (pa × Sa )
a 8c4 m 3
a
⎝
mb G 5xa + xb
+ ((pa × Sa ) · nab ) − 5(pa × Sa )
a b∗=a 4c4 m a rab rab
⎝
G 3 1
+ (pb × Sa ) − (nab × Sa )(pb · nab )
a b∗=a c4 rab 2 2

xa + xb
− ((pa × Sa ) · nab ) , (106)
rab
as well as the spin(1)–spin(2) one,

 
G xa Sa
2P N
GSS = [3(Sa · nab )(Sb · nab ) − (Sa · Sb )] 3
+ (Sb · nab 2 .
)
2c4 a rab rab
b∗=a
(107)
To summarize, for binary systems, and in part for many-body systems too, the fol-
lowing PN Hamiltonians are known fully explicitly,
H = H N + H1P N + H2P N + H2.5P N + H3P N + H3.5P N

+ HS1P
O + HS O + HS O + HS O
N 2P N 3P N 3.5P N
+ HS1P N
1 S2
+ HS2P N
1 S2
+ HS3P N
1 S2
+ HS3.5P
1 S2
N
+ HS1P N
1 S1
+ HS1P N
2 S2
+ HS2P N
1 S1
+ HS2P N
2 S2
+ H p1 S 3 + H p2 S 3 + H p1 S1 S 2 + H p2 S2 S 2
2 1 2 1
+ HS1 S 3 + HS 3 S2 + HS 2 S 2 . (108)
2 1 1 2
Also the n-body Hamiltonian through linear order in Newton’s gravitational constant
G is known H = H1P M , as well as the test-spin Hamiltonian in the Kerr metric.
For selfgravitating objects, the Hamiltonians primarily come out in the form
H = H [ p, q, h TT , εTT ]. The transition to a Routhian description of the type
H = H [ p, q, h TT , ḣ TT ] then allows the derivation of an autonomous Hamiltonian for
the conservative dynamics in the form H = H [ p, q, h TT (x; p, q), ḣ TT (x; p, q)] =
H ( p, q) as well as a non-autonomous one given by H (t) = H [ p, q, h TT (x; p ⇔ , q ⇔ ),
ḣ TT (x; p ⇔ , q ⇔ )] = H ( p, q; p ⇔ , q ⇔ ). The presented dynamical systems have found
derivations with other methods too. Particularly the Effective Field Theory method
by Goldberger and Rothstein, [38], has proven very powerful. For details the reader
is referred to the literature.
Acknowledgments The author thanks Stanley Deser for useful discussions.

188 G. Schäfer
References
1. Dirac, P.A.M.: The theory of gravitation in Hamiltonian form. Proc. R. Soc. Lond. A 246, 333
(1958). doi:10.1098/rspa.1958.0142
2. Dirac, P.A.M.: Fixation of coordinates in the hamiltonian theory of gravitation. Phys. Rev. 114,
924 (1959). doi:10.1103/PhysRev.114.924
3. Dirac, P.A.M.: Energy of the gravitational field. Phys. Rev. Lett. 2, 368 (1959). doi:10.1103/
PhysRevLett.2.368
4. Arnowitt, R., Deser, S.: Quantum theory of gravitation: general formulation and linearized
theory. Phys. Rev. 113, 745 (1959). doi:10.1103/PhysRev.113.745
5. Arnowitt, R., Deser, S., Misner, C.M.: Canonical variables in general relativity. Phys. Rev. 117,
1959 (1960). doi:10.1103/PhysRev.117.1595
6. Arnowitt, R., Deser, S., Misner, C.M.: Consistency of the canonical reduction of general rela-
tivity. J. Math. Phys. 1, 434 (1960). doi:10.1063/1.1703677
7. Arnowitt, R., Deser, S., Misner, C.W.: The dynamics of general relativity. In: Witten L. (ed.)
Gravitation: An Introduction to Current Research, pp. 227–265. Wiley, New York (1962).
doi:10.1007/s10714-008-0661-1
8. Schwinger, J.: Quantized gravitational field. Phys. Rev. 130, 1253 (1963)
9. Regge, T., Teitelboim, C.: Role of surface integrals in the hamiltonian formulation of general
relativity. Ann. Phys. (N.Y.) 88, 286 (1974). doi:10.1016/0003-4916(74)90404-7
10. Wheeler, J.A.: Geometrodynamics and the issue of the final state. In: DeWitt, C., DeWitt, B.S.
(eds.) Relativity, Groups, and Topology, pp. 315–520. Gordon and Breach, New York (1964)
11. DeWitt, B.S.: Quantum theory of gravity. I. The canonical theory. Phys. Rev. 160, 1113 (1967).
doi:10.1103/PhysRev.160.1113
12. Kimura, T.: Fixation of physical space-time coordinates and equation of motion of two-body
problem. Prog. Theor. Phys. 26, 157 (1961). doi:10.1143/PTP.26.157
13. Kibble, T.W.B.: Canonical variables for the interacting gravitational and Dirac fields. J. Math.
Phys. 4, 1433 (1963). doi:10.1063/1.1703923
14. Nelson, J.E., Teitelboim, C.: Hamiltonian formulation of the theory of interacting gravitational
and electron fields. Ann. Phys. (N.Y.) 116, 86 (1978). doi:10.1016/0003-4916(78)90005-2
15. Dirac, P.A.M.: Interacting gravitational and spinor fields. In: Recent Developments in General
Relativity, pp. 191–207. Pergamon Press, Oxford (1962)
16. Schäfer, G.: The gravitational quadrupole radiation-reaction force and the canonical formalism
of ADM. Ann. Phys. (N.Y.) 161, 81 (1985). doi:10.1016/0003-4916(85)90337-9
17. Ohta, T., Okamura, H., Kimura, T., Hiida, K.: Coordinate condition and higher order gravita-
tional potential in canonical formalism. Prog. Theor. Phys. 51, 1598 (1974). doi:10.1143/PTP.
51.1598
18. Damour, T., Schäfer, G.: Lagrangians for n point masses at the second post-Newtonian approx-
imation of general relativity. Gen. Relativ. Gravit. 17, 879 (1985). doi:10.1007/BF00773685
19. Damour, T., Jaranowski, P., Schäfer, G.: Dimensional regularization of the gravita-
tional interaction of point masses. Phys. Lett. B 513, 147 (2001). doi:10.1016/S0370-
2693(01)00642-6
20. Jaranowski, P., Schäfer, G.: Radiative 3.5 post-Newtonian ADM Hamiltonian for many-body
point-mass systems. Phys. Rev. D 55, 4712 (1997). doi:10.1103/PhysRevD.55.4712
21. Schäfer, G.: The ADM Hamiltonian at the postlinear approximation. Gen. Relativ. Gravit. 18,
255 (1986). doi:10.1007/BF00765886
22. Ledvinka, T., Schäfer, G., Bičák, J.: Relativistic closed-form Hamiltonian for many-body grav-
itating systems in the post-Minkowskian approximation. Phys. Rev. Lett. 100, 251101 (2008).
23. Damour, T.: Coalescence of two spinning black holes: an effective one-body approach. Phys.
24. Damour, T., Jaranowski, P., Schäfer, G.: Hamiltonian of two spinning compact bodies with
next-to-leading order gravitational spin-orbit coupling. Phys. Rev. D 77, 064032 (2008). doi:10.
1103/PhysRevD.77.064032
25. Hartung, J., Steinhoff, J.: Next-to-leading order spin-orbit and spin(a)-spin(b) Hamiltonians
for n gravitating spinning compact objects. Phys. Rev. D 83, 044008 (2011). doi:10.1103/
PhysRevD.83.044008
26. Hartung, J., Steinhoff, J.: Next-to-next-to-leading order post-Newtonian spin-orbit Hamiltonian
for self-gravitating binaries. Ann. Phys. (Berlin) 523, 783 (2011). doi:10.1002/andp.201100094
27. Steinhoff, J., Wang, H.: Canonical formulation of gravitating spinning objects at 3.5 post-
Newtonian order. Phys. Rev. D 81, 024022 (2010). doi:10.1103/PhysRevD.81.024022
28. Steinhoff, J., Hergt, S., Schäfer, G.: The next-to-leading order gravitational spin(1)-spin(2)
dynamics in Hamiltonian form. Phys. Rev. D 77, 081501 (2008). doi:10.1103/PhysRevD.77.
081501
29. Hartung, J., Steinhoff, J.: Next-to-next-to-leading order post-Newtonian spin(1)-spin(2) Hamil-
tonian for self-gravitating binaries. Ann. Phys. (Berlin) 523, 919 (2011). doi:10.1002/andp.
201100163
30. Wang, H., Steinhoff, J., Zeng, J., Schäfer, G.: Leading-order spin-orbit and spin(1)-spin(2)
radiation-reaction Hamiltonians. Phys. Rev. D 84, 124005 (2011). doi:10.1103/PhysRevD.84.
124005
31. Steinhoff, J., Hergt, S., Schäfer, G.: Spin-squared Hamiltonian of next-to-leading order gravi-
tational interaction. Phys. Rev. D 78, 101503 (2008). doi:10.1103/PhysRevD.78.101503
32. Hergt, S., Schäfer, G.: Higher-order-in-spin interaction Hamiltonians for binary black holes
from source terms of the Kerr geometry in approximate ADM coordinates. Phys. Rev. D 77,
104001 (2008). doi:10.1103/PhysRevD.77.104001
33. Barausse, E., Racine, E., Buonanno, A.: Hamiltonian of a spinning test particle in curved
spacetime. Phys. Rev. D 80, 104025 (2009). doi:10.1103/PhysRevD.80.104025
34. Brill, D.R., Lindquist, R.W.: Interation energy in geometrostatics. Phys. Rev. 131, 471 (1963).
doi:10.1103/PhysRev.131.471
35. Jaranowski, P., Schäfer, G.: Lapse function for maximally sliced Brill-Lindquist initial data.
36. Steinhoff, J., Schäfer, G.: Canonical formulation of self-gravitating spinning-object systems.
Europhys. Lett. 87, 50004 (2009). doi:10.1209/0295-5075/87/50004
37. Deser, S., Isham, C.J.: Canonical vierbein form of general relativity. Phys. Rev. D 14, 2505
(1976). doi:10.1103/PhysRevD.14.2505
38. Goldberger, W.D., Rothstein, I.Z.: An effective field theory of gravity for extended objects.
Stability of Marginally Outer Trapped
Surfaces and Geometric Inequalities
Marc Mars
Abstract Marginally outer trapped surfaces (MOTS) admit a notion of stability

that in many respects generalizes a similar notion for minimal hypersurfaces. Sta-
ble MOTS play an interesting role in a number of geometric inequalities involving
physical parameters such as area, mass, charge or, in the axially symmetric case,
angular momentum. Some of those inequalities are global in nature while others are
local, with interesting relationships between them. In this lecture the notion of sta-
ble MOTS will be reviewed and some of the geometric inequalities involving stable
MOTS will be described.
1 Introduction
Geometric inequalities play a fundamental role in gravitation because they provide

information on the relationship between physically relevant quantities in a robust
way, independently of the details of the particular spacetime under consideration.
It is often the case that not even field equations are necessary for the validity of
such inequalities and that only energy conditions are required, which make their
range of validity very broad and transverse to several theories of gravity. One of the
most fundamental geometric inequalities in gravitation is the Positive Mass Theorem
which, as is well-known establishes that the ADM mass is non-negative for any
asymptotically flat spacetime satisfying the dominant energy condition (DEC), i.e.
such that the Einstein tensor contracted with any pair of future directed causal vectors
gives a non-positive quantity. Another very important geometric inequality, which
so far has been proved only in special circumstances is the Penrose inequality [1].
M. Mars (B)
Dpto. Física Fundamental, University of Salamanca, Pl. de la Merced s/n,
37008 Salamanca, Spain
e-mail: marc@usal.es
192 M. Mars
In the asymptotically flat case, this inequality conjectures that the total ADM mass
M satisfies the inequality
|Smin | ≈ 16ε M 2 .
where |Smin | is the minimal area needed to enclose a given weakly future trapped
surface S, i.e. a closed, spacelike, codimension-two surface with future directed,
causal mean curvature vector. The notion of “enclosing” also needs a definition, see
e.g. [2] for details. The Penrose inequality is a strengthening of the positive mass
theorem when there are trapped surfaces present. In turn, the Penrose inequality can
be strengthened when the spacetime is charged and the total charge Q cannot be
radiated away (e.g. in electrovacuum, or when all matter present in the spacetime is
electrically neutral). In this case, the conjectured inequality becomes [3]

Q2
|Smin | ≈ 8ε M −
2
+ M2 M2 − Q2 .
2
It is clear that, in order for this inequality to even make sense, it is necessary that
the total ADM mass satisfies the bound M ⊗ |Q|. This inequality was proved in
[4] (see also [5] for a mathematically complete argument) and provides a direct
strengthening of the positive mass theorem in charged spacetimes, irrespectively of
whether a weakly future trapped surface is present in the spacetime or not.
Another global charge in asymptotically flat spacetimes is the ADM angular
momentum J . In general, angular momentum can be radiated away by gravitational
waves, so in general no strengthening of the Penrose inequality involving angular
momentum should be expected (this is because the physical argument leading to the
Penrose inequality involves, on the one hand, the weak cosmic censorship hypothesis
and, on the other, the asymptotic values of mass, charge and angular momentum of
the black hole that forms during the collapse). There is one interesting case, how-
ever, when angular momentum cannot be radiated away, namely when the spacetime
is axially symmetric. Thus, in this case (and assuming again that electric charge is
either absent or cannot be radiated away) the Penrose inequality can be strengthened
to (see [6, 7] for a discussion)

Q2 4
|Smin | ≈ 8ε M2 − + M − Q2 M 2 − J 2 := FQ,J (M), (1)
2
where the last equality defines a function of M for each choice of Q and J . As before,
in order for this inequality to make sense, it becomes necessary that M 4 − Q 2 M 2 −
J 2 ⊗ 0 or, equivalently,
1 2 4
M2 ⊗ (Q + Q + 4J 2 ).
2
Stability of Marginally Outer Trapped Surfaces and Geometric Inequalities 193
This is a strengthening of the positive mass theorem that should be valid for any
asymptotically flat, axially symmetric spacetime with conserved total electric charge.
This result has been proved whenever the spacetime admits a maximal (i.e. with sec-
ond fundamental form of vanishing trace), axially symmetric slice with two asymp-
totically flat ends, first in vacuum [8, 9] and then in electrovacuum [10, 11]. The
assumption of having more than one asymptotically flat end is made because, in
electrovacuum, the total electric charge and the total angular momentum can only
be non-zero in the presence of non-trivial topology. In the case of vacuum with an
arbitrary number of asymptotically flat ends, the inequality is not yet settled but a
closely related inequality has been proved in [12], which possibly reduces to the
previous one depending on the values of a certain functional defined on stationary
and axially symmetric asymptotically flat spacetimes (see the review [13] for many
more details).
All the geometric inequalities discussed so far are global in nature because they
involve the total ADM mass, and the charge and angular momentum are also global
quantities defined at infinity. Even the area term in the Penrose inequality is global
quantity because of the need of taking the minimal area enclosure of the weakly future
trapped surface. It is straightforward
to check that FQ,J is an increasing function of
M. Its minimum value is 4ε Q 4 + 4J 2 . It therefore follows that, whenever the
minimal area enclosure |Smin | satisfies |Smin | ≈ 4ε Q 4 + 4J 2 then the Penrose
inequality (1) is satisfied automatically. It follows that this inequality is non-trivial
only if

|Smin | ⊗ 4ε Q 4 + 4J 2 .
In particular, when the total charge is non-conserved or vanishes identically, this

inequality becomes
|Smin | ⊗ 8ε |J |
and when the total angular momentum is non-conserved, the inequality reads
|Smin | ⊗ 4ε Q 2 .
These inequalities are still of global nature, but now it makes sense to try and see
whether a local version of them is still valid. Indeed, one may think of replacing the
minimal area enclosure |Smin | by the area |S| of the weakly future trapped surface
itself. Moreover, the notion of total charge enclosed by a closed, orientable surface
makes sense, and in the axially symmetric case, the presence of a Killing vector
allows one to define the Komar angular momentum of any closed, orientable surface.
Moreover, in electrovacuum there is a modification [14] of the Komar definition of
angular momentum involving the electromagnetic field which provides a conserved
quantity in the sense that it gives an object depending only on the homology class of
the surface under consideration.
194 M. Mars
There has been very interesting and remarkable progress in recent years towards
the proof of local inequalities of this type on certain surfaces. They were first found in
the case of stationary and axially symmetric black hole horizons admitting arbitrary
matter outside the horizon but such that a neighbourhood of the horizon itself is
vacuum or electrovacuum. The first result along those lines was for the degenerate
case (i.e. when the surface gravity of the horizon vanishes) [15]. The non-degenerate
case for vacuum horizons was dealt with in [16]. Finally, the inequality in the charged,
rotating case and for electrovacuum horizons was solved in [17]. These results had
important implications in the problem of non-existence of stationary and axially
symmetric two-black hole configurations [18–21] (see also G. Neugebauer’s and
J. Hennig’s contribution to this volume).
Remarkably, a purely local version of this inequality where only properties of
suitable spacelike two-surfaces are used has also been obtained recently. The first
case, proved by Dain and Reiris [22] involved stable minimal, axially symmetric
surfaces embedded in maximal, axially symmetric hypersurfaces in a vacuum space-
time. In this setting, the universal inequality |S| ⊗ 8ε |J | was proved, where J is the
Komar angular momentum. This inequality was then extended [23] to arbitrary, axi-
ally symmetric, stable marginally outer trapped surfaces (defined below) embedded
in a spacetime with arbitrary matter contents as long as the dominant energy condi-
tion is satisfied. The case with electric charge (and no angular momentum so that no
need to restrict oneself to axially symmetric situations) was analyzed in [24] where
the inequality |S| ⊗ 4ε Q 2 was proved for suitable surfaces. The case involving both
charge and angular momentum has been proved recently in [25].
The key underlying property of the local versions of the inequality is the notion of
stability, both for minimal hypersurfaces and for marginally outer trapped surfaces.
The aim of this lecture is to review the notion of stability for marginally outer trapped
surfaces and discuss some of its consequences. Then, I will present in more detail the
various inequalities and explain how does stability enter into the arguments. The final
aim will be to relate the black hole-type inequalities to the purely local inequalities
by summarizing recent results [26] on the stability properties of Killing horizons.
2 Basics on the Geometry of Spacelike Surfaces
Our framework will be a four-dimensional spacetime (M, g), which we will take to be
oriented and time-oriented. Scalar product with the spacetime metric will be denoted
by ◦ , √ and S will refer to a closed (i.e. compact without boundary), spacelike, two-
dimensional, orientable, connected, embedded surface in (M, g) (simply surface
from now on). The normal space to S at any of its points is a Lorentzian vector
space. The collection of all normal spaces is a vector bundle over S that admits two
global, smooth, nowhere zero cross-sections {γ, k} satisfying ◦γ, γ√ = 0, ◦k, k√ = 0
and ◦γ, k√ = −1. We always take γ (and hence k) to be future directed. These sections
are defined uniquely up to the usual boost freedom γ ∗ Fγ, k ∗ F −1 k, where F is a
smooth, positive scalar function on S. The (positive definite) induced metric on S will
be denoted by h and the corresponding covariant derivative by D. The null extrinsic

curvatures are defined, as usual, by χγ AB ⇔ ◦e A , ≤e B γ√, χk AB ⇔ ◦e A , ≤e B k√, where
{e A } is a basis of the tangent space of S. The null expansions are the traces of these
tensors, i.e. νγ = trh (χγ ), νk = trh (χk ). A relevant geometric object in the following is
the mean curvature vector, defined in terms of the null expansions by H = −νγ k−νk γ.
It is well-known (and straightforward) that this vector is independent of the choice
of null basis {γ, k}. Finally, the normal bundle admits a canonical connection with
connection one-form given by s A ⇔ −◦k, ≤e A γ√ in the basis {γ, k}.
The mean curvature is fundamental, among other things, because it contains full
information on how the area of the surface changes to first order under general
deformations. Denoting by λΦ |S| the first order variation of area along a deformation
vector Φ , the following identity holds

λΦ |S| = ◦H, Φ √αh ,
S
where αh is the metric volume form of (S, h). Hence, when H is future causal then
the area of S does not increase for any future causal deformation Φ . This is generally
taken as a clear signal of the presence of a strong gravitational field. A surface S with
this property is called weakly future trapped.
3 Marginally Outer Trapped Surfaces and Stability
As just mentioned, the future causal character for the mean curvature is sufficient
to signal the presence of a strong gravitational field. However, if the surface admits
a well-defined notion of exterior (for instance when S is contained in a spacelike,
asymptotically flat hypersurface and separates this hypersurface into an asymptot-
ically flat exterior and a compact domain), then a non-increase of area along the
exterior future null cone may also be taken as convincing indication that the gravi-
tation field on the surface is strong. Such surfaces are defined by the property that
νγ ≈ 0, where γ is the future null normal pointing into the exterior domain.
The borderline case is given by surfaces satisfying the equality case νγ = 0. It
turns out that studying this borderline case is interesting even when there is no clearly
distinguished notion of exterior. This leads to the following standard definition:
Definition 1 A marginally outer trapped surface (MOTS) is a surface which satisfies

either νγ = 0 or νk = 0 everywhere (after renaming we always take νγ = 0).
From the general expression for the first variation of area it follows that MOTS are
stationary points for the area functional with respect to arbitrary variations tangent to
γ. Stationary points of any functional call immediately for analyzing their behaviour
under second order variations. In this case, the result is a direct consequence of the
Raychauduri equation and gives
196 M. Mars

λ∂γ
2
|S| = − ∂ 2 G(γ, γ) + |χγ |2 α S ,
S
where G is the Einstein tensor of the spacetime and | · | is the norm of tensors on S
with the metric h. In contrast to the case of minimal hypersurfaces in Riemannian
ambient manifolds, the second order variation does not define a differential operator
acting on ∂. Moreover, it is always non-positive definite, provided the spacetime
satisfies the null energy condition NEC (G(γ≡ , γ≡ ) ⊗ 0 for any null vector γ≡ ). Thus,
the second variation of area along the direction γ for MOTS does not provide with any
useful notion of stability. A suitable alternative is to use first variations of νγ (which
vanishes for a MOTS) along arbitrary normal directions not tangent to γ. To be more
precise, select a normal direction to S nowhere tangent to γ. This defines uniquely
a vector v satisfying ◦γ, v√ = 1. Consequently, it can be decomposed uniquely as
v = −k + V γ where V is a scalar function on S. In other words, we parametrize the
collection of directions orthogonal to S and nowhere tangent to γ by a real function
V : S ∗ R via the generator v defined above.
Note that V may a priori have any value. The sign of V at any given point p ∇ S is
directly tied to the causal character of the normal direction selected at p. If V > 0 then
this direction is spacelike, if V = 0 the direction is null and if V < 0 the direction
is timelike. It is also convenient to define the vector v β := k + V γ, which defines
a second normal direction. This direction is linearly independent to the previous
one except when v is null, in which case they coincide. In any other case the causal
character of v β is always opposite (in the sense of spacelike vs. timelike) to the causal
character of v. In particular, v β is future causal if and only if v is “achronal” (i.e.
spacelike or null or, equivalently, V ⊗ 0).
Given a fixed direction defined by v we can perform variations restricted to this
direction. Any such variation is defined by a vector Φ = ∂v, where ∂ : S ∗ R.
The first order variation of νγ along ∂v was first computed by Newman [27] (the
calculation was performed assuming implicitly that ∂ ⇒= 0 everywhere, but the result
is generally valid, c.f. [28]). The resulting expression gives a differential operator L v
acting ∂ via the definition L v (∂) ⇔ λ∂v νγ . Its explicit expression is
R(h)
L v (∂) = −ωh ∂ + 2 s A D A ∂ + − G(γ, v β ) − V |χγ |2 − s A s A + D A s A ∂,
2
(2)
where ωh is the Laplacian of (S, h) and R(h) is the curvature scalar of the metric h.
As discussed in detail in [28] this operator is elliptic and, in general, not self-
adjoint. However, any elliptic operator on a compact manifold (or on a bounded
domain with Dirichlet boundary conditions) admits a principal eigenvalue ρv .
This is a real eigenvalue (i.e. such that there exists a real function Θv satisfying
L v (Θv ) = ρv Θv ) with the property that any other eigenvalue (which will be complex
in general) satisfies Re(ρ) > ρv . Moreover, as in the self-adjoint case, the principal
eigenfunction Θv does not change sign and hence can be taken to be positive every-
where. For self-adjoint operators, the principal eigenvalue admits a characterization
in terms of the so-called Rayleigh-Ritz quotient which is very useful, firstly because
it provides upper bounds for the principal eigenvalue and secondly, and even more
importantly, because it gives lower bounds for certain integral functionals acting on
arbitrary functions. This latter property is the key for translating sign conditions on
the principal eigenvalues into useful analytic inequalities that, in turn, can be used
to derive geometric properties of the surface. This is crucial, e.g., for studying stable
minimal hypersurfaces in Riemannian manifolds. Such surfaces have the property
that their own stability operator, which is now self-adjoint, has a non-negative prin-
cipal eigenvalue.
The Rayleigh-Ritz characterization is no longer true for non self-adjoint operators.
Given its importance, it is reasonable to ask whether there exists any analogue to this
characterization valid for any elliptic operator. Donsker and Varadhan [29] found
a number of characterizations of the principal eigenvalue of the min-max type, i.e.
involving a minimization of a certain class of suprema. Such characterizations are, in
general, difficult to work with. In [28] one of these characterizations was elaborated
further and a Rayleigh-Ritz type characterization for the principal eigenvalue of any
elliptic second order operator was found. In order to describe it, recall that any one-
form in a compact Riemannian manifold without boundary can be decomposed as
the sum of the differential of a function f plus a divergence-free one-form. Such
decomposition is usually called Helmholz decomposition in the physics literature.
Recall that this decomposition is unique up to an additive constant in f provided the
manifold is connected.
In the case of the stability operator L v , the characterization of ρv obtained in [28]
reads as follows.
Proposition 1 Let L v be the stability operator (2). Decompose the normal connec-
tion one-form s A according to the Helmoltz decomposition as s A = D A f + z A ,
where z A is divergence-free. Then the principal eigenvalue of L v is given by
S |Du|2 + R(h)
2 − G(γ, v β ) − V |χγ |2 u 2 − |dτu + z|2 u 2 αh
ρv = inf , (3)
u>0 S u 2 αh
where τu is any solution of
2 2
− ωh τu − D A τu D A u = z A D A u. (4)
u u
It is straightforward to check that, given any positive function u on S, the partial
differential equation (4) always admits a solution, which is unique up to an additive
constant.
An immediate consequence of this result is the so-called “symmetrized inequality
along v” proved by Galloway and Schoen [30] using explicit estimates,

1
|Du| + R(h)u αh ⊗
2 2
ρv + G(γ, v β ) + V |χγ |2 u 2 αh . (5)
S 2 S
198 M. Mars
Indeed, by dropping the term |dτu + z|2 u 2 in (3) the right-hand side is never
decreased, and hence so does infimum, which leads immediately to (5).
The stability operator allows us to define a notion of stability for MOTS [28],
which generalizes a similar notion for minimal hypersurfaces.
Definition 2 A MOTS S is stable along v if and only if the principal eigenvalue ρv

of the stability operator satisfies ρv ⊗ 0. S is strictly stable along v if and only if
ρv > 0.
One of the consequences of this definition is given by the following lemma, which
links the stability of a MOTS with the existence of suitable variations which increase
the value of the outer null expansion.
Lemma 1 A MOTS S is stable along v if there exists an outward variation ∂v
(∂ ⊗ 0, ∂ ⇒⇔ 0) such that λ∂v νγ ⊗ 0. S is strictly stable if, in addition, λ∂v νγ > 0.
Although, by construction, there is a stability operator (and a notion of stability) for
each direction v, the dependence of L v on v is very simple, namely L v = L −k −W V ,
where W is defined as W := (G(γ, γ) + |χγ |2 ). Note that under the NEC W ⊗ 0 and
hence stability improves when v is tilted away from γ (i.e. when V is made larger at
every point). In fact, when W ⇒⇔ 0, there always exists a direction sufficiently tilted
away from γ for which the principal eigenvalue is positive, and hence the MOTS is
strictly stable along this direction. However, it turns out that the stability of the MOTS
along a direction v gives, in general, useful information only when the direction v is
achronal. This is because the terms G(γ, v β ) and V |χγ |2 are non-negative (under the
NEC) only when V ⊗ 0, i.e. when v is spacelike or null at every point. This leads
to the following definition, spelled out in [23] and closely related to the notion of
future outer trapping horizon defined by Hayward [31].
Definition 3 A MOTS S is spacetime stable if it is stable along an achronal direc-
tion v.
It is now straightforward to check that, under the NEC, S is spacetime stable if and
only if it is stable along −k. Since the null direction −k is privileged as a transverse
direction for the MOTS, we will write ρ−k simply as ρ− .
4 Area-Charge-Topology Inequalities for MOTS
As discussed in the Introduction, a class of local inequalities exists for suitable sur-
faces relating the area and the total charge enclosed by the surface. If the surface
has non-trivial topology, its genus (recall that our surfaces are two-dimensional, con-
nected and orientable, so that their topology is uniquely determined by their genus)
also enters into the inequality. The basic assumption made in this context is that
the energy-momentum contents of the spacetime splits into an electromagnetic field
and the rest of matter in such a way that this rest satisfies the dominant energy
condition. Allowing also a cosmological constant ξ, the Einstein tensor takes the
form (assuming we are in General Relativity) G μζ + ξgμζ = Tμζ EM + T mat , where
μζ
EM = 2(F F δ − 1 g F F δβ ) is the energy-momentum tensor of the electro-
Tμζ μδ ζ 4 μζ δβ
magnetic field Fμζ and T mat is arbitrary except for the condition of satisfying DEC.
The electric and magnetic charges on S are defined as

1 1
QE = F(γ, k)αh , QM = F β (γ, k)αh ,
4ε S 4ε S
where F β is the Hodge dual of F. We now assume that S is a MOTS. If we insert

u = 1 in the symmetrized inequality (5) along −k (i.e. with V = 0) it follows

4ε(1 − g) ⊗ (ρ + ξ) |S| + (F(γ, k))2 + (F β (γ, k))2 αh . (6)
S
Now, following [24], the Hölder inequality implies the inequality S f 2 ⊗ |S|−1
( s f )2 valid for any L 2 function f . Thus, (6) implies
4ε(1 − g) ⊗ (ρ + ξ)|S| + 16ε 2 |S|−1 (Q 2E + Q 2M ). (7)
This inequality including the cosmological constant, arbitrary topology of S and the
principal eigenvalue has been obtained in [32], where a number of consequences
have also been derived. Two immediate consequences are the following:
Assume that S is spacetime stable. Since, under the assumptions above, the space-
time satisfies the NEC (irrespectively of the sign of ξ) it follows that S is stable along
the direction −k, i.e. ρ− ⊗ 0. If we assume further that ξ ⊗ 0, then it follows imme-
diately that S must be of spherical or toroidal topology. This recovers a well-known
theorem on the topology of MOTS due to Hawking [33]. This theorem has been
generalized to higher dimensions in [30] where it was shown that the Yamabe type
of any stable MOTS must be non-negative. The case of vanishing Yamabe type (i.e.
toroidal topology in the case of four dimensions) has been shown to be very rigid in
[34] and to be excluded when the MOTS satisfies suitable barrier properties.
If S is spacetime stable and there is a negative cosmological constant, then (7)
implies an upper bound on the genus of S, namely g ≈ 1 − |S|ξ 4ε . This bound was
first obtained in [35].
For the area-charge inequality, (7) immediately implies that, as long as S is space-
time stable and ξ ⊗ 0, then |S| ⊗ 4ε(Q 2E + Q 2M ), as first proved in [24]. This
inequality, in turn, is a generalization of a previous result by Gibbons which estab-
lishes this inequality in the case of minimal surfaces embedded a time-symmetric
slice [36].
200 M. Mars
5 Axially Symmetric MOTS and Angular Momentum
We next discuss local geometric inequalities involving angular momentum.

As mentioned in the Introduction, this case requires the surfaces to be axially sym-
metric in order to have a proper definition of angular momentum. In principle, one
may think that this entails restricting oneself to axially symmetric spacetimes. In
fact, fewer requirements are needed and the following definition, essentially put for-
ward in [23], is sufficient for the purpose of writing down and proving the desired
inequalities:
Definition 4 A MOTS S is axially symmetric if there exists a vector α tangent to S

satisfying
1. Lα h = 0.
2. Lα s = 0, for some choice of basis {γ, k}.
3. α commutes with the stability operator L v for some choice of v.
It is clear that if the spacetime (M, g) admits an axial Killing vector α and α is tangent
to S, then the S is also axially symmetric according to this definition.
The angular momentum J of an axially symmetric MOTS is then defined by

1
J (S) := s(α)α S .
8ε S
It is straightforward to check that this definition is independent of the choice of basis

{γ, k}. Nevertheless, for some expressions below, it is necessary to restrict the basis
to satisfy point 2 in definition 4. We will do so from now on.
The following theorem due to Jaramillo et al. [23] establishes a remarkable
inequality involving the area and the angular momentum of an axially symmetric
MOTS. The only requirement is that the MOTS is spacetime stable and that energy-
momentum contents of the spacetime satisfies the dominant energy condition. It
is therefore a very general and robust inequality which reveals a deep connection
between the rotation and the shape of quasi-local black holes in four spacetime
dimensions.
Theorem 1 Let (S, h) be an axially symmetric, two-dimensional MOTS, stable with

respect to an achronal direction v in a spacetime satisfying DEC. Then
|S| ⊗ 8ε |J |. (8)
Moreover, equality can only happen if the following five conditions are simultane-
ously satisfied:
(i) S is marginally stable,
4|J | sin2 ν 2
(ii) h = |J | 1 + cos2 ν dν 2 + dκ ,
1 + cos2 ν
(iii) G(γ, k) = 0 on S,
2J sin2 ν
(iv) z = 2 dκ,
|J | 1 + cos2 ν
(v) If v is spacelike then G(γ, γ) = 0 and χγ = 0.
The equality case corresponds to the geometry of the extreme Kerr horizon, i.e.
the induced metric on any spacelike section of the degenerate horizon of the Kerr
2
metric satisfying MADM = |JADM |. Previous to Theorem 1, the same inequality had
been proved by Dain and Reiris [22] for minimal surfaces embedded in maximal
slices of a vacuum spacetime
In the following I will describe the basic steps involved in the proof of this result.
The argument has two parts. The first one consists in finding an inequality valid for
arbitrary functions on S and applying it to an appropriate function defined in terms of
the geometry of S. The second step consists in showing that the resulting inequality
can be related to the angular momentum, to conclude finally that (8) holds.
Step 1 was accomplished in [23] as a consequence of the spacetime stability of
S by performing suitable direct estimates. However, as mentioned before, stable
MOTS satisfy general inequalities given by the Rayleigh-Ritz type characterization
described in Proposition 1. This allows us to describe step one as follows.
First, using u = 1 in the general inequality (3) (notice that for u = 1, the solution
to equation (4) is τu = const) and the Gauss-Bonnet theorem shows that the genus
of the surface must be positive, or else z A = 0, which would imply J = 0 and hence
a trivial area-angular momentum inequality. So, only the spherically symmetric case
needs to be considered. In this setting axisymmetry implies that the divergence-
free one form z A in the Helmholz decomposition of the connection one-form s A is
proportional to the Killing vector α (with its indexes lowered with the metric h).
Consequently, the source term in equation (4) vanishes for any choice of axially
symmetric u. The corresponding solution is again τu = const. Since restricting the
class of functions u in (3) to being axially symmetric cannot decrease the infimum,
it follows from stability that

R(h) 2
|Du|2 + u αh ⊗ G(γ, v β ) + V |χγ |2 + |z|2 u 2 αh
S 2 S

⊗ |z|2 u 2 αh , (9)
S

where the second inequality follows because G(γ, v β ) + V |χγ |2 u 2 is non-negative
under the assumptions of the theorem. Now, it is simple to see that the metric h can
be written in the form (c.f. [22])
1
h= e2c e−ψ (φ ) dφ 2 + eψ (φ ) dκ 2 , −∈ < φ < ∈, 0 < Θ ≈ 2ε, (10)
cosh2 φ
where c is a constant related to the total area by |S| = 4ε ec . Regularity on the axis
of symmetry (i.e. where φ ∗ ±∈) imposes the following asymptotic behaviour on
ψ (φ ),
202 M. Mars
dψ
lim ψ = c, lim = 0.
φ ∗±∈ φ ∗±∈ dφ
ψ
Now, a key insight of Dain and Reiris [22] was to use the function u = e− 2 in the
analytic inequality (9). This gives an inequality which involves only ψ and z A . The
second part of the proof consists in showing that this leads to an inequality involving
only J and the area |S| in such a way that all details of the function ψ (φ ) and z A (φ )
disappear. To discuss this second part we use a simplified version of an argument
due to [21], which, in turn, is a simplification of the original argument in [37] where
the absolute minimum of a renormalized energy of a harmonic map was computed.
First of all we define a function Y (φ ) (up to an arbitrary additive constant) by the
equation z(α)αh = 21 dYdφ dφ ⊥ dκ. From the definition of angular momentum we have
8J = lim (Y (φ ) − Y (−φ )).

φ ∗∈
ψ
Now, inserting u = e− 2 and the expression for z A in terms of Y (φ ) in (9) and
computing explicitly the curvature scalar for the metric (10) it is straightforward to
see that the stability inequality becomes
2 2
1 dX dY
0⊗ + − 4 dφ dΘ, (11)
S X2 dφ dφ
where X (φ ) has been defined as X = |α|2 (i.e. X = eψ cosh−2 (φ )). Now we can view
the first term in the integrand as the energy-density of a path γ (φ ) ⇔ {X (φ ), Y (φ )}
in the hyperbolic space (H2 , gH2 = d X X+dY
2 2
2 ). Note that, since the variable φ takes
values on the whole real line and X (φ ) ∗ 0 when |φ | ∗ ∈ the total energy of the
path diverges (this simply reflects the fact that while the function in the right-hand
side of (11) is integrable on S, this property is obviously not true for the constant term
in the expression). Rewriting the total integral as a definite integral from φ = −L
to φ = L and sending L to infinity, inequality (11) becomes, after performing the
trivial angular integration,
L
0 ⊗ lim gH2 (γ̇ , γ̇ ) dφ − 8L ,
L∗+∈ −L
where dot means derivative with respect to φ . We now apply once again the Hölder
L L
inequality in the form −L gH2 (γ̇ , γ̇ ) dφ ⊗ 2L 1
( −L gH2 (γ̇ , γ̇ )dφ )2 and use the
obvious property that the total length of any curve is never smaller than the distance
between its initial and final points to rewrite the stability inequality as

1
0 ⊗ lim distH2 [γ (L), γ (−L)] − 8L .
2
(12)
L∗+∈ 2L
Now, the distance function between two arbitrary points (X 1 , Y1 ) and (X 2 , Y2 ) in

the Poincaré upper-half plane is well-known to be [38]

(Y2 − Y1 )2 + (X 2 − X 1 )2
distH2 [(X 1 , Y1 ), (X 2 , Y2 )] = arccosh 1 + .
2X 1 X 2
The limit L ∗ +∈ in (12) is straightforward to obtain after using ψ (±L) ∗ c and

Y (L) − Y (−L) ∗ 8J and gives

8ε |J |
0 ⊗ 8 ln ⇐≥ |S| ⊗ 8ε |J |,
|S|
which proves the inequality. The statements regarding the equality case are also
straightforward in this framework and follow by imposing that all inequalities along
the process become equalities. In particular, the curve γ (φ ) must be a parametrized
geodesic in (H2 , gH2 ). This fixes the metric h and z A to be those in the extreme Kerr
geometry. This proves points (ii) and (v) in the theorem. The rest follows directly
from the non-negative terms discarded in the stability inequality.
6 Area-Angular Momentum Inequality for Black Holes
As mentioned in the Introduction, the first examples of area-angular momentum

inequalities were obtained for black holes. The assumptions in this setting were
that the spacetime is stationary and axially symmetric and that it contains a Killing
horizon. Moreover, the spacetime was assumed to be vacuum in a neighbourhood of
the horizon, but matter was allowed outside the black hole. The first case treated in the
literature assumed the Killing horizon to be degenerate, i.e. with vanishing surface
gravity. In this setting, Ansorg and Pfister [15] were able to show that the inequality
|S| = 8ε |J | was universally valid. Here |S| is the area of any axially symmetric
spacelike section of the Killing horizon. The non-degenerate case was considered in
[16] where the inequality |S| > 8ε |J | was proved under the additional condition that
the black hole is subextremal. The definition of subextremal is due to [39] and requires
the existence of a future directed null vector k transverse to the Killing horizon for
which the inequality λk νΦ < 0 holds, where Φ is the Killing vector which generates
the Killing horizon. It is clear from the definition of strict stability of MOTS that a
Killing horizon is subextremal if and only if all of its spacelike sections are strictly
stable along the transverse direction k. The condition of subextremality was used in
[16] by working in an Eddington-Finkelstein advanced extension of the following
metric in Boyer-Lindquist type coordinates

d R2 4
ds 2 = μ̂ + dν 2 + û sin2 ν (dκ − τdt)2 − R 2 − rh2 dt 2 ,
R − rh
2 2 û
204 M. Mars
where the Killing horizon is located at R = rh . Imposing the condition λm νΦ < 0,

where m is proportional to π R in Eddington-Finkelstein coordinates, and discarding
a number of positive terms, the actual ”subextremality” assumption made by the
authors was

π R (û μ̂)| R=rh sin ν dν > 0,
S0
where S0 is a section of the Killing horizon corresponding to a constant coordinate

time tˆ in the Eddington-Finkelstein advanced coordinate system. This equality is
certainly implied by the geometric subextremality condition by Booth and Fairhurst,
but it is not equivalent to it and, in principle, it is weaker. The purely local inequality
in Theorem 1 requires stability, but not strict stability, so the connection between the
two inequalities is not immediately obvious. The proofs are also quite different, so
it became a problem of interest to try and relate the purely local and the black hole
versions of the area-angular momentum inequality. A first step along this direction
was made in [21] where the comparison was focused on the relationship between
the proof in the minimal surface case in [22] and the behaviour of the Killing hori-
zon on its bifurcation surface. However, no geometrically clear reason of why both
inequalities work in seemingly different regimes was given. In [26] a detailed study
of the stability properties of Killing horizons was performed. As a by product, a
clear relationship between the two types of area-angular momentum inequality was
obtained. The next section is devoted to reviewing these results briefly.
7 MOTS and Killing Horizons

Recall that, in a spacetime (M, g) admitting a Killing vector Φ , a Killing horizon H
is a null hypersurface where the Killing vector Φ is null, tangent and nowhere zero.
The integral lines of Φ are null geodesics on H and the surface gravity is the scalar
H
function on H defined by κ: ≤Φ Φ = κΦ . It is well-known that κ is constant if (M, g)
satisfies the DEC. In the following we will assume that H has topology S × R with
S closed and that the R factor is tangent to the integral lines of Φ .
The Killing equations imply immediately that all spacelike sections S0 in H are
MOTS and, in fact, with vanishing second fundamental form along Φ , i.e. χΦ (S0 ) = 0.
H
The Raychaudhuri equation then implies G(Φ, Φ ) = 0. These two properties say that
the function W introduced before vanishes identically in this case and, hence, that
the stability operator of S0 is independent of the transverse direction v. Both the
stability operator and its principal eigenvalue are therefore properties of S0 alone.
We will denote them simply by L S0 and ρ S0 respectively. A natural question is then
whether the stability is a property of the horizon itself or whether it depends on the
choice of section S0 . To address this issue it is convenient to obtain first an explicit
form of the stability operator. This was done in [26] for general totally geodesic null
hypersurfaces. Here we restrict ourselves to Killing horizons for definiteness.
Theorem 2 Let S0 be a spacelike section of a Killing horizon H of topology S × R.

Let k be a null vector field on S0 , orthogonal to S0 and satisfying ◦Φ, k√ = −1. The
stability operator L S0 takes the form

∂
L S0 (∂) = −D A u D A
+ 2z A D A ∂ − κνk ∂,
u
where the positive scalar function u and the one-form z A are defined via the Helmholz
decomposition of the normal connection one-form s A as s = z + du 2u .
Notice that u is defined up to an arbitrary multiplicative constant, which obviously

has no effect in the expression for L S0 . Combining this theorem with the behaviour
of s A and νk under a change of section the following theorem follows [26].
Theorem 3 If H has constant surface gravity, then ρ S0 is independent of S0 . More-
over there exist Killing horizons H with non-constant κ for which ρ S0 depends
on S0 .
Regarding the analysis of the area-angular momentum inequality for Killing hori-
zons, it turns out that the stability of S0 has implications on certain integral of the
transverse null expansion νk on S0 . More precisely [26].
Proposition 2 Let H be a Killing horizon with topology S × R and let S0 any

spacelike section of H . Assume that z A and du in the decomposition s = z + du 2u
are orthogonal (this occurs automatically if H is axially symmetric and S0 respects
the axial symmetry and has spherical topology). Then the following holds
• If S0 is stable then S0 κνk u ≈ 0.
• If S0 is strictly stable then S0 κνk u < 0.
• If S0 is stable and κ is constant and non-zero, then S0 κνk u = 0 ⇐≥ νk = 0.
This Proposition is the key property that allows one to find the link between the
proof of the area-angular momentum inequality in the black hole case [16] and the
purely local inequality in [23]. This relationship is given in the following result [26].
Theorem 4 Assume that the spacetime (M, g) satisfies the DEC and admits a Killing
vector Φ with a Killing horizon H . Assume that H S2 × R is axially symmetric
with κ ⇒= 0 constant. Write the metric of spacelike sections of H in the form
h = e2c e−ψ dν 2 + eψ sin2 ν dΘ 2 . (13)
Then there exists a section S1 for which s = z − 21 dψ . Moreover, if S1 κνk e−ψ ≈ 0,

then |S| ⊗ 8ε |J |.
Notice that S1 is not assumed to be stable in this theorem. The inequality involving νk
is sufficient for proving the area-angular momentum inequality. Proposition 2 shows
that stability of the Killing horizon (or equivalently stability of any of its spacelike
206 M. Mars
sections) is sufficient to imply the validity of the integral inequality assumed in the
theorem. It turns out that the surface S0 used by [16] is precisely the surface S1 in
this theorem. Moreover, their inequality S0 π R (û μ̂)| R=rh sin ν dν > 0 is precisely
the geometric inequality S κνk e−ψ < 0, and the link between the two approaches
becomes clear. A similar clarification for the degenerate case has been recently
obtained in [40].
8 Further Results and Open Problems
As mentioned in the Introduction, when the spacetime has an electromagnetic

field, the area-energy momentum inequality can be strengthened to include the
electric charge. This was first done for stationary and axially symmetric degen-
erate black holes in [15]. The main assumption was that the spacetime is elec-
trovacuum in a neighbourhood of the horizon and it was shown that the equality
|S| = 4ε Q 4 + 4J 2 always holds. In the same setting, but allowing non-degenerate
subextremal black holes, the strict inequality |S| > 4ε Q 4 + 4J 2 was proved in
[17]. The local version of this inequality (i.e. for stable, axially symmetric MOTS)
has been proved in [25].
Everything we have said so far involves four-dimensional spacetimes. Recent
work by Hollands [41] establishes the following generalization to arbitrary dimen-
sion.
Theorem 5 Let (M n+1 , g) be a vacuum spacetime with a non-negative cosmolog-
ical constant ξ, n ⊗ 3 and which admits a stable Killing horizon H of topology
S × R with S closed. Assume further that (M n+1 , g) admits an additional isometry
group U (1)n−2 leaving H invariant and consider a section S0 of H respecting the
U (1)n−2 symmetries. Let α± be the Killing vectors which vanish, respectively, at the
“north” and “south” poles of S0 . Then
|S0 | ⊗ 8ε |J (α+ )J (α− )|1/2 ,
where J (α) := 1
8ε S0 s(α)α S0
(for the precise definition of the vectors α± , see in [41]).
Before concluding this lecture, let me present a brief list of open problems. The
first one refers to the higher dimensional case. The statement above requires that
the spacetime contains a Killing horizon and that the spacetime is vacuum, possibly
with a positive cosmological constant. It would be of interest to relax this and admit
a general spacetime satisfying the DEC. It is also of interest to prove the statement
directly at the local level, i.e. for stable MOTS. Another interesting problem in this
context is whether the symmetry assumptions can be relaxed to a small number of
linearly independent Killing vectors.
In relation to the structure of the proof of the area-angular momentum inequality,
ψ
it would be of interest to find a deeper reason of why the choice of u = e− 2 , where ψ
is defined as a metric coefficient, is the appropriate choice and what is the underlying
reason for the role played by the hyperbolic space (and other highly symmetric spaces
in the charged case) in the proof of the inequality.
Another interesting problem is to understand why the surface t˜ = const in
Eddington-Finkesltein coordinates for black holes is precisely the surface S1 in the
area-angular momentum inequality for Killing horizons, i.e. why for these coor-
dinates, the normal connection one-form is linked to the induced metric via the
condition spelled out in Theorem 4.
References
1. Penrose, R.: Naked singularities. Ann. N.Y. Acad. Sci. 224, 125 (1973). doi:10.1111/j.1749-
6632.1973.tb41447.x
2. Mars, M.: Present status of the penrose inequality. Class. Quantum Grav. 26, 193001 (2009).
doi:10.1088/0264-9381/26/19/193001
3. Gibbons, G.W.: The isoperimetric and bogomolny inequalities for black holes in global rie-
mannian geometry. In: Willmore, T.J., Hitchin, N.J. (eds.) Ellis Horwood Series in Mathematics
and Its Applications, pp. 194–202. Ellis Horwood, Chichester (1984)
4. Gibbons, G.W., Hawking, S.W., Horowitz, G.T., Perry, M.J.: Positive mass theorems for black
holes. Commun. Math. Phys. 88, 295 (1983). doi:10.1007/BF01213209
5. Herzlich, M.: The positive mass theorem for black holes revisited. J. Geom. Phys. 26, 97 (1998).
doi:10.1016/S0393-0440(97)00040-5
6. Dain, S., Lousto, C.O., Takahashi, R.: New conformally flat initial data for spinning black
holes. Phys. Rev. D 66, 104038 (2002). doi:10.1103/PhysRevD.65.104038
7. Szabados L.B.: Quasi-local energy-momentum and angular momentum in GR: a review article.
Living Rev. Relativity 7, lrr-2004-4 (2004). http://www.livingreviews.org/lrr-2004-4
8. Dain, S.: The inequality between mass and angular momentum for axially symmetric black
holes. Int. J. Mod. Phys. D 17, 519 (2008). doi:10.1142/S021827180801219X
9. Dain, S.: Proof of the angular momentum-mass inequality for axisym-metric black holes. J.
Differ. Geom. 79, 33 (2008)
10. Chruściel P.T., Lopes Costa J.: Mass, angular-momentum, and charge inequalities for axisym-
metric initial data. Class. Quant. Grav. 26, 235013 (2009). doi:10.1088/0264-9381/26/23/
235013
11. Costa, J.L.: Proof of a Dain inequality with charge. J. Phys. A: Math. Theor. 43, 285202 (2010).
doi:10.1088/1751-8113/43/28/285202
12. Chruściel P.T., Li Y., Weinstein G.: Mass and angular-momentum inequalities for axi-symmetric
initial data sets: II. Angular-momentum. Ann. Phys. (N.Y.) 323, 2591 (2008). doi:10.1016/j.
aop.2007.12.011
13. Dain, S.: Geometric inequalities for axially symmetric black holes. Class. Quantum Grav. 29,
073001 (2012). doi:10.1088/0264-9381/29/7/073001
14. Simon, W.: Gravitational field strength and generalized komar integral. Gen. Relativ. Gravit.
17, 439 (1985). doi:10.1007/BF00761903
15. Ansorg, M., Pfister, H.: A universal constraint between charge and rotation rate for degenerate
black holes surrounded by matter. Class. Quantum Grav. 25, 035009 (2008). doi:10.1088/0264-
9381/25/3/035009
16. Hennig, J., Ansorg, M., Cederbaum, C.: A universal inequality between the angular momentum
and the horizon area for axisymmetric and stationary back holes with surrounding matter. Class.
Quantum Grav. 25, 162002 (2008). doi:10.1088/0264-9381/25/16/162002
208 M. Mars
17. Hennig, J., Cederbaum, C., Ansorg, M.: A universal inequality for axisymmetric and stationary
black holes with surrounding matter in the einstein-maxwell theory. Commun. Math. Phys. 293,
449 (2010). doi:10.1007/s00220-009-0889-y
18. Neugebauer, G., Hennig, J.: Non-existence of stationary two-black-hole configurations. Gen.
Relativ. Gravit. 41, 2113 (2009). doi:10.1007/s10714-009-0840-8
19. Hennig, J., Neugebauer, G.: Non-existence of stationary two-black-hole configurations: the
degenerate case. Gen. Relativ. Gravit. 43, 3139 (2011). doi:10.1007/s10714-011-1228-0
20. Hennig, J., Neugebauer, G.: Stationary two-black-hole configurations: a non-existence proof.
J. Geom. Phys. 62, 613 (2012). doi:10.1016/j.geomphys.2011.05.008
21. Chruściel, P.T., Eckstein, M., Nguyen, L., Szybka, S.: Existence of singularities in two-kerr
black holes. Class. Quantum Grav. 28, 245017 (2011). doi:10.1088/0264-9381/28/24/245017
22. Dain, S., Reiris, M.: Area—Angular momentum inequality for axisymmetric black holes. Phys.
23. Jaramillo, J.L., Reiris, M., Dain, S.: Black hole area-angular momentum inequality in non-
vacuum spacetimes. Phys. Rev. D 84, 121503 (2011). doi:10.1103/PhysRevD.84.121503
24. Dain, S., Jaramillo, J.L., Reiris, M.: Area-charge inequality for black holes. Class. Quantum
Grav. 29, 035013 (2012)
25. Gabach-Clement M.E., Jaramillo J.L., Reiris M.: Proof of the area-angular momentum-charge
inequality for axisymmetric black holes, ArXiv e-prints 1207.6761 [gr-qc] (2012)
26. Mars, M.: Stability of mots in totally geodesic null horizons. Class. Quantum Grav. 29, 145019
(2012). doi:10.1088/0264-9381/29/14/145019
27. Newman, R.P.A.C.: Topology and stability of marginal 2-surfaces. Class. Quantum Grav. 4,
277 (1987). doi:10.1088/0264-9381/4/2/011
28. Andersson, L., Mars, M., Simon, W.: Stability of marginally outer trapped surfaces and exis-
tence of marginally outer trapped tubes. Adv. Theor. Math. Phys. 12, 853 (2008)
29. Donsker, M.D., Varadhan, S.R.S.: On the principal eigenvalue of second order elliptic differ-
ential operators. Commun. Pure Appl. Math. 29, 591 (1976). doi:10.1002/cpa.3160290606
30. Galloway, G.J., Schoen, R.: A generalization of hawking’s black hole topology theorem to
higher dimensions. Commun. Math. Phys. 266, 571 (2006). doi:10.1007/s00220-006-0019-z
31. Hayward, S.: General laws of black-hole dynamics. Phys. Rev. D 49, 6467 (1994). doi:10.
1103/PhysRevD.49.6467
32. Simon, W.: Bounds on area and charge for marginally trapped surfaces with a cosmological
constant. Class. Quantum Grav. 29, 062001 (2012). doi:10.1088/0264-9381/29/6/062001
33. Hawking, S.W.: The event horizon. In: DeWitt, C., DeWitt, B.S. (eds.) Black Holes, pp. 1–56.
Gordon and Breach, New York (1973)
34. Galloway, G.J.: Rigidity of marginally trapped surfaces and the topology of black holes. Comm.
Anal. Geom. 16, 217 (2008)
35. Woolgar, E.: Bounded area theorems for higher-genus black holes. Class. Quantum Grav. 16,
3005 (1999). doi:10.1088/0264-9381/16/9/316
36. Gibbons, G.W.: Some comments on gravitational entropy and the inverse mean curvature flow.
Class. Quantum Grav. 16, 1677 (1999). doi:10.1088/0264-9381/16/6/302
37. Aceña, A., Dain, S., Gabach Clément, M.E.: Horizon area-angular momentum inequality for a
class of axially symmetric black holes. Class. Quantum Grav. 28, 105014 (2011). doi:10.1088/
0264-9381/28/10/105014
38. Anderson, J.W.: Hyperbolic Geometry. Springer Undergraduate Mathematics Series. Springer,
London (2007)
39. Booth, I., Fairhurst, S.: Extremality conditions for isolated and dynamical horizons. Phys. Rev.
D 77, 084005 (2008). doi:10.1103/PhysRevD.77.084005
40. Jaramillo, J.L.: A note on degeneracy, marginal stability and extremality of black hole horizons.
41. Hollands, S.: Horizon area-angular momentum inequality in higher dimensional spacetimes.
Stationary Black-Hole Binaries:
A Non-existence Proof
Gernot Neugebauer and Jörg Hennig
Abstract We resume former discussions of the question, whether the spin-spin

repulsion and the gravitational attraction of two aligned black holes can balance
each other. Based on the solution of a boundary problem for disconnected (Killing)
horizons and the resulting violation of characteristic black hole properties, we present
a non-existence proof for the equilibrium configuration in question. From a mathe-
matical point of view, this result is a further example for the efficiency of the inverse
(“scattering”) method in non-linear theories.
1 Introduction
The examination of time-independent two-body systems dates back to the early days
of General Relativity. In a 1922 paper, Bach and Weyl [1] discussed the superposi-
tion of two exterior Schwarzschild solutions in Weyl coordinates as a characteristic
example for an equilibrium configuration consisting of two “sphere-like” bodies at
rest. Bach noted that this static solution develops a singularity on the portion of the
symmetry axis between the two bodies, which violates the elementary flatness on this
interval. In a supplement to Bach’s contribution, Weyl focused on the interpretation
of this type of singularity and used stress components of the energy-momentum ten-
sor to define a non-gravitational repulsion between the bodies which compensates
the gravitational attraction. Weyl’s result is based on some artificial assumptions but
G. Neugebauer
Theoretisch-Physikalisches Institut, Friedrich-Schiller-Universität Jena, Max-Wien-Platz 1,
07749 Jena, Germany
e-mail: neugebauer@tpi.uni-jena.de
J. Hennig (B)
Department of Mathematics and Statistics, University of Otago, P.O. Box 56,
Dunedin 9054, New Zealand
e-mail: jhennig@maths.otago.ac.nz
210 G. Neugebauer and J. Hennig
Fig. 1 Illustration of two J1

aligned rotating black holes
with horizon areas A1 , A2 and
angular momenta J1 , J2
A1
J2
A2
implies an interesting question: Are there repulsive effects of gravitational origin

which could counterbalance the omnipresent mass attraction?
Post-Newtonian approximations tell us that the interaction of the angular momenta
of rotating bodies (“spin-spin interaction”) could indeed generate repulsive effects.
This is a good motivation to study, in a rigorous way, stationary two-body problems.
In this contribution we shall summarize the results that we obtained for a sta-
tionary two-black-hole system consisting of two aligned rotating black holes with
parallel (or anti-parallel) spins, see Fig. 1. The representation is based on three recent
papers which contain the details of the analysis [2–4]. The idea of our non-existence
proof is to construct the exterior gravitational fields of two disconnected Killing
horizons, see Fig. 2, via a boundary problem for the nonlinear Ernst equation, which
is essentially equivalent to the vacuum Einstein equations. Fortunately, this equa-
tion belongs to a class of completely integrable differential equations, which can be
mapped to linear structures (“Linear Problems”). This fact is the source of power-
ful solution generating methods such as Bäcklund transformations. It can be shown
that a single Bäcklund transformation [5, 6] applied to Minkowski space creates a
Kerr-NUT spacetime which includes the spacetime of the rotating black hole. Since
iterative Bäcklund transformations act as a “nonlinear superposition principle”, the
double-Kerr-NUT solution [5, 7] was considered to be a good candidate for the so-
lution of the two-horizon problem and extensively discussed in the literature [7–17].
However, there was no argument ensuring that this particular solution be the only can-
didate. We have removed this objection and shown that the solution procedure for the
boundary problem necessarily leads to a subclass of the double-Kerr-NUT solution.
Stationary Black-Hole Binaries 211
Fig. 2 A two-black-hole
equilibrium configuration
in Weyl-Lewis-Papapetrou
coordinates (Adapted from
[2])
K1
K2
K3
K4
Thus we could make use of results derived for the double-Kerr-NUT solution. The
result is in line with a theorem of Varzugin [18, 19], which says that the 2N -soliton
solution by Belinskiĭ and Zakharov [20, 21] contains all possible solutions (if any
existed) corresponding to an equilibrium configuration of black holes. The subclass
is characterized by a set of restrictions for the parameters of the general double-
Kerr-NUT solution. These restrictions, first derived and discussed by Tomimatsu and
Kihara [11, 16], ensure the regularity of the double-Kerr-NUT solution on the axis of
symmetry and on the horizons. An elegant reformulation of the Tomimatsu-Kihara
regularity conditions by Manko et al. [15] made it possible to express black hole
quantities such as mass, angular momentum and surface area in terms of independent
parameters. We have made use of these results. Another condition to be satisfied is
the positivity of the total mass. Combining the restrictions with symmetry arguments,
Hoenselaers and Dietz [8, 10] and Krenzer [13] could show that the double-Kerr-
NUT solution cannot describe a configuration consisting of two identical black holes.
Manko and Ruiz [14] generalized this result by showing that the regularity conditions
imply that at least one of the two horizons has a negative Komar mass. They argued,
without giving any explanation, that this peculiarity casts out the double-Kerr-NUT
solution. Remarkably, Ansorg and Petroff [22], who described an equilibrium con-
figuration with a positive total mass and a component which has a negative Komar
mass, came to an opposite interpretation. However, considerations like these stim-
ulated us to examine further black hole inequalities. Fundamental local “state vari-
ables” of a rotating black hole are its area A and its angular momentum J . Indeed,
for a single black hole these quantities are restricted by the inequality 8π |J | ≈ A.
Based on results of Ansorg and Pfister [23], who examined extremal black holes,
Hennig et al. [24], who, following Booth and Fairhurst [25], studied sub-extremal
black holes defined by existence of trapped surfaces (surfaces with a negative expan-
sion of outgoing null rays) in every sufficiently small interior neighbourhood of the
event horizon, and Chrusćiel et al. [26], we can assume that each of the two black
holes has to satisfy the angular momentum-area inequality individually. Surprisingly,
Dain and Reiris [27] were able to extend its range of application to non-stationary
black holes, see also the overview article by Dain [28] and references therein.
2 Mathematical Tools
2.1 Metric and Horizons
The exterior vacuum gravitational field of axially symmetric and stationary gravi-
tational sources can be described in cylindrical Weyl-Lewis-Papapetrou coordinates
(ρ, ζ, ϕ, t),1 in which the line element takes the form

ds 2 = e−2U e2k (dρ 2 + dζ 2 ) + ρ 2 dϕ 2 − e2U (dt + a dϕ)2 , (1)
where the “Newtonian” gravitational potential U , the gravitomagnetic potential a

the “superpotential”
and k are functions of ρ and ζ alone. At large distances r =
ρ 2 + ζ 2 ⊗ ◦ from isolated sources located around the origin of the coordinate
system, r = 0, the spacetime has to be Minkowskian,
r ⊗ ◦ : ds 2 = dρ 2 + dζ 2 + ρ 2 dϕ 2 − dt 2 . (2)
Metric (1) admits an Abelian group of motions G 2 with the generators (Killing
vectors)
ξ i = δti , (stationarity), (3)

η =
i
δϕi , (axisymmetry), (4)
where the Kronecker symbols δti , δϕi indicate that ξ i has only a time t-component
whereas ηi points in the azimuthal ϕ-direction. ηi has closed compact trajectories
about the axis of symmetry and is therefore spacelike off the axis (and the horizons).
ξ i is timelike sufficiently far from the black holes but can become spacelike inside
ergoregions. Obviously,
e2U = −ξ i ξi , a = −e−2U ηi ξ i (5)
1 In the following, we also use the complex coordinates z = ρ + iζ and z̄ = ρ − iζ . t is the time
coordinate.
is a coordinate-free representation of the two relativistic gravitational potentials U

and a.
In stationary (and axisymmetric) spacetimes, the event horizon as the central black
hole property is a local concept. Consider the Killing vector ξ √ ,
ξ √ = ξ + Ωη (6)
with the norm

e2V = −(ξ √ , ξ √ ) = e2U (1 + Ωa)2 − ρ 2 Ω 2 e−4U , (7)
where Ω is a real constant. A connected component of the set of points with e2V = 0,
which is a null hypersurface, (de2V , de2V ) = 0, is called a Killing horizon H (ξ √ ),
H (ξ √ ) : e2V = −(ξ √ , ξ √ ) = 0, (de2V , de2V ) = 0. (8)
Since the Lie derivative Lξ √ of e2V vanishes, we have (ξ √ , de2V ) = 0. Being null
vectors on H (ξ √ ), ξ √ and de2V are proportional to each other,
H (ξ √ ) : de2V = −2κξ √ . (9)
Using the (vacuum) field equations one can show that the surface gravity κ is a
constant on H (ξ √ ). Ω is the angular velocity of the horizon. In black hole thermo-
dynamics, κ and Ω are conjugate to the extensive quantities A (area) and J (angular
momentum), respectively.
In the ρ-ζ plane (t = constant, ϕ = constant) of the Weyl-Lewis-Papapetrou
coordinate system (1), the horizons cover a finite portion on the ζ -axis (ρ = 0), see
Fig. 2, or shrink to a point [29]. It turns out that extended horizons (“sub-extremal
horizons”) and point-like horizons (“degenerate horizons”) require different con-
siderations. Note that a Killing horizon is always a two-surface in the time slice
t = constant. The degeneracy to a line or a point is a peculiarity of the special
coordinate system.
In this paper, we explain the non-existence proof for extended (sub-extremal)
horizons and end up with a brief comment on degenerate horizons.
The dashed line in Fig. 2 sketches the boundaries of the vacuum region: A + , A 0 ,
A are the vacuum parts of the ζ -axis (axis of symmetry), H (1) and H (2) denote
−
the two Killing horizons, which are located in the intervals [K 2 , K 1 ] and [K 4 , K 3 ]
on the ζ -axis, and C stands for spatial infinity. The gravitational fields a, k, U have
to satisfy the following boundary conditions
A ± , A 0 : a = 0, k = 0, (10)
H (i) : 1 + Ωi a = 0, i = 1, 2, (11)
C : U ⊗ 0, a ⊗ 0, k ⊗ 0, (12)
where Ω1 and Ω2 are the angular velocities of the two horizons. Equations (10)
characterize the axis of symmetry (rotation axis). The first relation originates from
the second equation in (5), since the compact trajectories of η with the standard
periodicity 2π become infinitesimal circles with the consequence η ⊗ 0. The second
relation is a necessary condition for elementary flatness (Lorentzian geometry in the
vicinity of the rotation axis). Equation (11) is a reformulation of Eqs. (8) (e2V = 0)
and (7) since the horizons are located on the ζ -axis (ρ = 0), see Fig. 2. Finally,
Eq. (12) ensures the asymptotic flatness of the metric (1), see (2).
2.2 Field Equations and Linear Problem
The stationary and axisymmetric vacuum Einstein equations for the metric potentials
U and a are equivalent to the Ernst equation
1
(∗ f ) f ,ρρ + f ,ζ ζ + f ,ρ = f ,ρ2 + f ,ζ2 (13)
ρ
for the complex function
f (ρ, ζ ) = e2U (ρ,ζ ) + ib(ρ, ζ ), (14)
where the twist potential b is defined by
a,ρ = ρe−4U b,ζ , a,ζ = −ρe−4U b,ρ . (15)
The potential k can be calculated from

1
k,ρ = ρ U,ρ2
− U,ζ2 + e−4U (b,ρ 2
− b,ζ
2
) , (16)
4
1 −4U
k,ζ = 2ρ U,ρ U,ζ + e b,ρ b,ζ . (17)
4
As a consequence of the Ernst equation (13), the integrability conditions a,ρζ =
a,ζρ and k,ρζ = k,ζρ are satisfied such that the metric potentials a and k may be
calculated via line integration from the Ernst potential f . Since e2U = ∗ f , all
metric coefficients in (1) can uniquely be determined from f . Thus the boundary
problem for the vacuum Einstein equations reduces to a boundary problem for the
Ernst equation. However, we have to cope with non-local boundary conditions (10)–
(12), (15)–(17) for the Ernst potential f . Fortunately, these boundary conditions are
well-adapted to the “inverse method”, which we applied to solve the boundary value
problem.
The Ernst equation is the integrability condition Φ ,z z̄ = Φ ,z̄z of the Linear Prob-
lem (LP) [30, 31]
N 0 0 N
Φ ,z = +λ Φ, (18)
0 M M 0
M̄ 0 1 0 M̄
Φ ,z̄ = + Φ, (19)
0 N̄ λ N̄ 0
where the pseudopotential Φ(z, z̄, λ) is a 2 × 2 matrix depending on the spectral

parameter
K − iz̄
λ= (20)
K + iz
as well as on the complex coordinates
z = ρ + iζ, z̄ = ρ − iζ, (21)
whereas M, N and the complex conjugate quantities M̄, N̄ are functions of z, z̄

(or ρ, ζ ) alone and do not depend on the constant complex parameter K . Since the
integrability condition must hold identical in λ (or K ) it yields a system of first order
differential equations for N and M which is equivalent to the Ernst equation. The
first order system has the “first integrals”
f ,z f¯,z
M= , N= . (22)
f + f¯ f + f¯
Vice versa, any solution f of the Ernst equation admits the unique determination of
the pseudopotential Φ up to constants of integration. Thus the Ernst equation (13)
and the LP (18), (19) are equivalent to each other.
Multiplying (18) by dz and adding (19) multiplied by dz̄ one obtains the refor-
mulation dΦ = (· · · )Φ of the LP in the form of a system of (overdetermined) total
differential equations.
Without loss of generality we choose the standard representation
ψ(ρ, ζ, λ) ψ(ρ, ζ, −λ)

Φ= (23)
χ (ρ, ζ, λ) −χ (ρ, ζ, −λ)
where
1
ψ̄ ρ, ζ, = χ (ρ, ζ, λ) (24)
λ̄
due to the special structure of the coefficient matrices of the LP. For K ⊗ ◦ and
λ ⊗ −1, the functions ψ, χ can be normalized by
ψ(ρ, ζ, −1) = χ (ρ, ζ, −1) = 1. (25)

As a consequence of the LP, the Ernst potential and the gravitomagnetic potential
can be read off from the pseudopotential Φ at K ⊗ ◦ and λ ⊗ +1,
f (ρ, ζ ) = χ (ρ, ζ, 1) = Φ 21 (ρ, ζ, 1), (26)

∂
a(ρ, ζ ) = ie −2U
K2
[χ (−λ) − ψ(−λ)] − C, (27)
∂K λ=1
K ⊗◦
where C is an arbitrary constant. The idea of the inverse (scattering) method is to

discuss Φ, for fixed but arbitrary values of ρ, ζ (z, z̄) as a holomorphic function
of λ or K . In the latter case, Φ “lives” on the two sheets of the Riemann surface
associated with (20). As this mapping depends on the parameters ρ, ζ , the position
of the branch points K B = iz̄, K̄ B = −iz and the branch cut between them changes
with the coordinates.
3 Non-existence Proof
3.1 Integration of the Linear Problem
In order to solve the boundary problem (13), (10)–(12), we shall integrate (“solve”)
the LP along the dashed line in Fig. 2 which marks, in the ρ-ζ plane, the boundary
of the vacuum domain outside the horizons. Starting from and returning to any axis
point, say ρ = 0, ζ ⇔ A + , we shall make use of the boundary conditions and
finally arrive at a representation of the Ernst potential on the axis of symmetry. It
turns out that this representation is sufficient to express all black hole quantities
(such as areas A1 , A2 and angular momenta J1 , J2 of the black holes) in terms of
three independent real parameters (plus two additional scaling parameters) and to
establish the equations of state of the black hole thermodynamics of the equilibrium
configuration under discussion. Furthermore, the axis values f (ζ ) = f (ρ = 0, ζ )
fix the solution f (ρ, ζ ) of the Ernst equation uniquely [32].
Since λ(K ) as defined in (20) degenerates on the ζ -axis, λ = ±1, the LP dΦ =
(· · · ) Φ can easily be integrated. For λ = +1 one obtains
f¯ 1 A(K ) B(K )
A ± , A 0 , H (i) : Φ = L, L = . (28)
f −1 C(K ) D(K )
The representation for λ = −1 follows from (23) by exchanging column elements.

Remarkably, the Φ-matrix separates. The first factor depends on the path of in-
tegration whereas L representing the “integration constants” is a function of the
spectral parameter K alone. There is a difference between the case of two ex-
tended horizons and that of one or two degenerate horizons. In the first case one
can parametrize the dashed curve by the coordinate ζ everywhere on the ζ -axis, i.e.
f = f (ρ = 0, ζ )= f (ζ ) on A ± , A 0 , H (i) . This is clearly impossible if the dashed

curve runs around a point-like horizon. A path like this can be described by an in-
finitesimal semicircle which brings (local) polar coordinates (R, θ ) into play [33].
Then we have f = f (R ⊗ 0, θ ) = f (θ ), θ ⇔ [0, π ] in (28).
To exploit characteristic properties of the horizons such as (8) and (9), it is helpful
to introduce corotating frames of reference defined by
ρ √ = ρ, ζ √ = ζ, ϕ √ = ϕ − Ωt, t √ = t, (29)
where Ω = Ω1 , Ω2 . This coordinate transformation induces transformations of the

gravitational potentials in (1) such that
√
e2U ≤ e2V = e2U [(1 + Ωa)2 − Ω 2 ρ 2 e−4U ], (30)
2U √
(1 − Ωa √ )e = (1 + Ωa)e2U , (31)
where a prime denotes “corotating” quantities. To determine the corotating Ernst

√
potential f √ one has to apply (15) to a √ and e2U . Finally, using the Eq. (22) for
N √ , M √ , one obtains the corotating pseudopotential [31]
Φ √ = TΩ Φ, (32)
where
1 + Ωa − Ωρe−2U 0
TΩ =
0 1 + Ωa + Ωρe−2U
−1 −λ
+ i(K + iz)Ωe−2U . (33)
λ 1
The validity of the Ernst equations of the non-rotating and corotating system at
the points of intersection A /H (axis of symmetry/extended horizon) and A /C
(axis/circle at infinity), see Fig. 2, implies that Φ and Φ √ must be continuous there
as well. By way of example let us consider the continuity of Φ in (28) and Φ √ in (32)
at the point ρ = 0, ζ = K 1 . It immediately leads to a connection of the horizon and
axis values of the “integration constants” L(1) and L+ , cf. (28),
F1
L(1) = 1 − L+ , (34)
2iΩ (1) (K − K 1 )
where F1 is is a special case of Fi as needed later,
− fi 1
Fi := , f i = f (K i ), i = 1, . . . , 4. (35)
− f i2 f i
Note that f¯1 = − f 1 since e2V and e2U are continuous at the points of intersection
and e2V = 0 from the side of the horizon. If one continues the interlinking procedure
along the closed contour, one returns to the starting point with the result
01
L+ (L+ )−1 = R+ , (36)
10
where

4
Fi 01
R+ := 1 − (−1)i (i)
(37)
2iΩ (K − Ki ) 10
i=1
with Ω (1) = Ω (2) = Ω1 , Ω (3) = Ω (4) = Ω2 .

Point-like (degenerate) horizons can be involved without any difficulty by setting
K 1 = K 2 or/and K 3 = K 4 in these equations [33].
We shall show that (36) with (37) as the result of the integration of the LP along
the closed (dashed) contour in Fig. 2 yields the Ernst potential on the axis.
3.2 Ernst Potential on the Axis
At the branch points K = K B = iz̄, K = K̄ B = −iz of the Riemann K -surface,

where λ = 0, λ = ◦, respectively, one finds from (23) Φ 11 = Φ 12 , Φ 21 = −Φ 22 ,
01 1 0
K = KB : Φ Φ −1 = . (38)
10 0 −1
On the ζ -axis one has confluent branch points, K B = K̄ B = ζ . For this choice one
obtains from (38), (28) and (36) in terms of the Ernst potential f + on A +
−1
f¯+ (ζ ) 1 1 0 f¯+ (ζ ) 1
R+ (ζ ) = (39)
f + (ζ ) −1 0 −1 f + (ζ ) −1
and so
1
[R+ (ζ ) − 1] =0 (40)
f + (ζ )
for R+ (ζ ) := R+ (K = ζ ), cf. (37). Note that (40) is equivalent to (39): a second

column resulting from (39) and (40) are complex conjugate.
We shall now discuss properties of the Ernst potential on the axis which can be
derived from the eigenvalue Eq. (40). First of all, let us point out that similar equations
can be derived for all intervals A , H by the interlinking procedure as explained
in (34). At the first glance, f + (ζ ) seems to be a quotient of two polynomials of
fourth degree in ζ . However, attention must be paid to the fact that the characteristic
determinant has to vanish,
|R+ (ζ ) − 1| = 0. (41)
This condition tells us that the numerator and the denominator of f + (ζ ) must have
two common zeros such that the axis potential is a quotient of two (normalized)
polynomials of second degree in ζ ,
n 2 (ζ ) ζ 2 + qζ + r
f + (ζ ) = = 2 , (42)
d2 (ζ ) ζ + sζ + t
where q, r , s, t are complex constants which can be expressed in terms of f i , K i and

Ω (i) , (i = 1, . . . , 4). For extended (sub-extremal) horizons the following reparame-
trization is useful:
Defining
d̄2 (K i ) n̄ 2 (K i )
αi := , αi ᾱi = 1, βi := , βi β̄i = 1 (43)
d2 (K i ) n 2 (K i )
and using
f + (K i ) = − f + (K i ), i = 1, . . . , 4, (44)
one obtains
βi = −αi . (45)
The Eqs. (43) form a linear algebraic system for the parameters q, r , s, t. Eliminating
them in (42) one arrives at a determinant representation for the Ernst potential on
the axis A + , f + (ζ ), which can be written in the form

s12 − 1 s14 − 1

s23 − 1 s34 − 1 αi (ζ − K i ) − α j (ζ − K j )
+
f (ζ ) = , si j := , (46)
s12 + 1 s14 + 1 Ki j

s23 + 1 s34 + 1
where
K i j := K i − K j , i, j = 1, . . . , 4. (47)
The continuation of f + (ζ ) to all space is unique [32] and leads to the representation

R12 − 1 R14 − 1

R23 − 1 R34 − 1 αi ri − α j r j
f (ρ, ζ ) = , Ri j := , (48)
R12 + 1 R14 + 1 Ki j

R23 + 1 R34 + 1
where

ri := (ζ − K i )2 + ρ 2 ≡ 0, i, j = 1, . . . , 4. (49)
f (ρ, ζ ) is the Ernst potential of the double-Kerr-NUT solution which was originally
generated by a two-fold Bäcklund transformation of Minkowski space [5, 7] in the
form of a quotient of two 5 × 5 determinants. According to a rule of Yamazaki [17],
this type of determinants can be reduced to 2×2 determinants as used in (48). Making
use of (15) [or (27)], (16), (17) and e2U = ∗ f , one finds determinant representations
for all metric coefficients, i.e. for a, k, e2U in (1), see [4].
As a condition identical in K , Eq. (41) yields four constraints among the parame-
ters Ω1 , Ω2 ; K 1 − K 2 , K 2 − K 3 , K 3 − K 4 ; f 1 , . . ., f 4 . It can be shown [4] that this
system of algebraic equations guarantees that a = 0 on A ± , A 0 and C . As a con-
sequence, it eliminates NUT parameters from the Ernst potential. In consideration
of (40) and introducing dimensionless coordinates ρ̃, ζ̃ via
ρ ζ − K1
ρ̃ = , ζ̃ = (50)
K 23 K 23
one realizes that f + (ζ ) from (40) and therefore f (ρ, ζ ) depend on four real para-
meters.
3.3 Weyl-Bach Force Between the Black Holes
So far we have examined the Ernst equation as a classical field equation. It is question-
able whether the parameter conditions (41) alone could rule out the Ernst potential
under discussion and lead to a non-existence proof. Consider the static Ernst equa-
tion which is an axisymmetric Laplace equation for the “Newtonian” gravitational
potential U in Weyl coordinates. The superposition of two solutions with aligned rod-
shaped sources (as “classical” precursors of horizons) is regular outside the sources.
It is the gravitational interaction (“force”) between the rods that forbids equilibrium.
Bach, who examined this example in the already mentioned Bach and Weyl paper [1],
noted that the metric function k cannot vanish on the portion of the axis between the
two sources and that this fact violates the regularity of the solution. Weyl’s remarks
(published as a supplement to Bach’s paper) focused on the interpretation of this type
of singularity. He used fictitious stresses described by an energy-momentum tensor
to define a force of attraction between the sources. This “Weyl-Bach force” turned
out to be proportional to a constant value of k on the portion of the axis between the
two sources (in our notation k 0 ) provided that k + = 0, k − = 0, which is a possible

gauge. Note that k as defined in (16), (17) has constant values on all intervals of
the ζ -axis. One of them, considered to be the arbitrary integration constant, can be
chosen so that, say, k + = 0. Integration of (16), (17) along C in Fig. 2 results in
k − = 0.
Equipped with that physical as well as geometrical interpretation of the superpo-
tential k 0 (“attractive force” that violates the Lorentzian geometry in the vicinity of
the rotation axis) we shall examine the boundary condition
k0 = 0 (51)
for the gauge
k + = k − = 0. (52)
Our discussion is based on the original parametrization of the double Kerr-NUT

solution, see (43), (45), (46), and Kramer’s representation of e2k [12]. (In principle,
one could determine the axis values of this gravitational potential from the axis
values of the Ernst potential by integrating the Eqs. (16), (17) along the axis, i.e. by
operations on the dashed contour in Fig. 2.) It turns out that the boundary conditions
for non-overlapping extended horizons (51), (52) are satisfied by only one parameter
condition:
α1 α2 + α3 α4 = 0. (53)
Two of the four conditions (41) can be used to eliminate Ω1 and Ω2 . The two
remaining equations turn out to be equivalent to the equations

(1 − α4 )2 2 (1 − α3 )2 K 14 K 24
w = , w := ⇔ [1, ◦), (54)
α4 α3 K 13 K 23

(1 + α2 )2 √2 (1 + α1 )2 K 23 K 24
w = , w √ := ⇔ (0, 1]. (55)
α2 α1 K 13 K 14
The three restrictions (53)–(55) are nothing else but a reformulation of the original
Tomimatsu-Kihara conditions [11, 16]. This reformulation is due to essential ex-
aminations of the double-Kerr-NUT solution by Manko, Ruiz and Sanabria-Gómez
[14, 15]. As was particularly shown in [14], the restrictions (53)–(55) can be solved
to express the parameters α1 , . . . , α4 in terms of the three real parameters w, w √ and
φ:
w √ α 2 + iεα α 2 + iw √ εα
α1 = √
, α2 = , (56)
w − iεα 1 − iw √ εα
Fig. 3 The singular Ernst potential for a particular example configuration. Parameters: φ = −0.1,
w = 1.3, w √ = 0.5, K 1 = 2, K 23 = 2, ε = 1
wα 2 − α α 2 − wα
α3 = , α4 = , (57)
w−α 1 − wα
where ε = ±1 and
α3 α4 = −α1 α2 ≤ α 2 (α = eiφ , φ ⇔ [0, 2π )). (58)
Now we have arrived at the final form of the solution of the boundary problem
(10)–(12) for the Ernst equation (13). Eliminating the αi in favour of w, w √ , φ and
introducing dimensionless coordinates (50), f becomes a function of two coordinates
and three real parameters,2
f = f (ρ̃, ζ̃ ; w, w √ , φ). (59)
Note that the relative horizon “lengths” K 12 /K 23 , K 34 /K 23 can be expressed in

terms of w, w√ as well. At this point we cannot guarantee that this Ernst potential is
well-behaved. Computer experiments show that the regularity of the Ernst potential
on the axis of symmetry must be “paid of” in the form of singular rings outside the
horizons. Figure 3 conveys an impression of the structure of this type of singularity.
Irregularities on the horizons could also rule out the solution. Since area A and
angular momentum J are characteristic parameters for each axisymmetric black hole,
2 From this point of view, the Ernst potential of the Kerr solution depends on one real parameter
and two dimensionless coordinates.

it was obvious to examine restrictions of these parameters. Based on the literature

as commented on in the introduction, we could take for granted that the inequality
8π |J | < A has to hold on both horizons of a regular spacetime with two sub-extremal
black holes.
3.4 Angular Momentum-Area Inequality and Non-existence Proof
In order to examine the inequalities
H (i) : 8π |Ji | < Ai , i = 1, 2, (60)
we calculate the quantities
8π Ji
pi := , i = 1, 2, (61)
Ai
in terms of the parameters w, w√ and φ ⇔ [0, 2π ). Note that this computation can be
performed with the aid of the axis values of the Ernst potential alone, see [2]. The
result is
1 + Φw √ w(w − Φ)
p1 = ε , p2 = ε (62)
w √ (Φ + w √ ) 1 − wΦ
where
Φ := cos φ + ε sin φ, ε = ±1. (63)
From this we have
w √2 + 2Φw √ + 1
p12 − 1 = (1 − w √2 ) <0 (64)
w √2 (Φ + w √ )2
and
w 2 − 2Φw + 1
p22 − 1 = (w 2 − 1) < 0. (65)
(wΦ − 1)2
For the allowed parameter ranges w ⇔ [1, ◦), w√ ⇔ (0, 1], which follow from the
definitions (54), (55) of w and w √ , these inequalities can only hold if
w √2 + 2Φw √ + 1 < 0 and w 2 − 2Φw + 1 < 0. (66)
This, however, implies Φw√ < 0 and Φw > 0 in contradiction to w√ > 0 and
w > 0. Thus we have proved the non-existence of stationary and axisymmetric
Non-existence of
dΦ = (... )Φ + 8π |J| < A
two-black-hole equilibrium
Fig. 4 Summary of the non-existence proof for two sub-extremal black holes
configurations consisting of two aligned sub-extremal black holes and we conclude

that the spin-spin repulsion cannot compensate for the gravitational attraction.
4 Summary
As a characteristic example for the ongoing discussion about existence or non-

existence of stationary equilibrium configurations within the theory of General Rel-
ativity, we have studied the question whether two aligned, sub-extremal black holes
can be in equilibrium. The result of our above analysis, whose details can be found in
[2–4], is negative: there are no two-black-hole equilibrium configurations! The idea
of the non-existence proof is illustrated in Fig. 4 and can be summarized as follows.
Equilibrium configurations with two aligned rotating black holes, if any existed,
can be described by a boundary value problem for two separate (Killing-) hori-
zons. Remarkably, this problem can be solved by integrating the Linear Problem
dΦ = (· · · )Φ along the dashed contour as sketched in Fig. 2 or Fig. 4. Thus we ar-
rive necessarily at particular Kerr-NUT solutions which have two horizons [at least
according to the definitions (8) and (9)] and show the correct regular behaviour at in-
finity and on the symmetry axis, whereas regularity off the axis is not guaranteed. On
the contrary, we find that all candidate solutions indeed do suffer from irregularities.
One of them is the violation of the angular momentum-area inequality 8π |J | < A,
which must hold for any regular sub-extremal black hole. We could show that there is
no choice of parameters for which angular momentum and area of the two horizons
jointly satisfy the inequality. Hence, there exists no regular solution of the vacuum
field equations for stationary two-black-hole configurations. For brief comments on
the extension of the non-existence proof to degenerate black holes, see the following
supplement and Fig. 5a.
(a) (b) (c)
ζ ζ
Einstein-Maxwell
equations
3 × 3 Linear Problem
Fig. 5 Illustration of further BVPs that can be solved with the inverse scattering method: a two-
horizon problem with one degenerate black hole, b rigidly rotating disk of dust, c constructive
uniqueness proof for the Kerr-Newman black hole
5 Supplement
5.1 Degenerate Horizons
The analysis as presented above only applies to configurations with two sub-extremal
(“extended”) horizons. But we have already indicated several times that it is possible
to extend the proof to configurations containing one or even two degenerate black
holes (with “point-like” horizons, where K 1 = K 2 or/and K 3 = K 4 ), see Fig. 5a.
In the first part of this supplement we give an outline of this generalisation. For details
we refer to [3].
As already mentioned, the degeneracy of a Killing horizon to a point is merely a
peculiarity of the Weyl-Lewis-Papapetrou coordinates. In order to resolve the internal
structure of the horizon, we have discussed the LP in polar coordinates centred at
the point-like horizon. Integrating the LP, we found that all possible equilibrium
configurations with degenerate black holes can be obtained as particular limits of
the double-Kerr-NUT family of solutions. It turned out that there are two families of
(two-parametric) solutions describing configurations with one degenerate and one
extended horizon and three (one-parametric) solution families for configurations with
two degenerate horizons. In order to exclude these families as acceptable equilibrium
configurations, we showed that they suffer from unphysical singularities.
In the case of one degenerate and one sub-extremal black hole, the angular
momentum-area inequality becomes an equality for the degenerate horizon (which
turned out to be satisfied identically and, therefore, did not provide any new infor-
mation). The inequality for the sub-extremal black hole restricts the parameters, but
does not yet exclude the possibility of a regular equilibrium configuration. Hence an
additional ingredient was required for the desired non-existence proof, which was
the positivity of the total mass (ADM mass) of the spacetime. As first shown by
Schoen and Yau [34, 35], the mass of a regular, asymptotically flat spacetime sat-
isfying the dominant energy condition (which is certainly satisfied for a black hole
vacuum spacetime) is positive. However, for the configuration in question, we found
that the entire parameter range in which the sub-extremal black hole satisfies the
angular momentum-area inequality has a negative mass. Thus we could conclude
that the solutions with one degenerate and one sub-extremal horizon are singular.
Finally, we studied the three solution branches for configurations with two degen-
erate horizons. One branch could be excluded since it has a negative ADM mass for
all possible parameter values. The other two branches have negative masses for most
parameter values, but there are small parameter regions with positive mass. Fortu-
nately, these solution branches are relatively simple and, by studying the solutions
of a certain quartic equation, it was possible to demonstrate explicitly that singular-
ities (singular rings around the ζ -axis) are present even in the parameter range with
positive ADM mass.
Hence we could extend the non-existence proof to all forms of horizons.
5.2 Further Applications of the Inverse Method
In the following we briefly comment on some other applications of the inverse (scat-
tering) method to rotating objects. The integration of the Linear Problem of the Ernst
equation dΦ = (. . . )Φ along a suitable closed contour was first practised to deter-
mine the gravitational field of a rigidly rotating disk of dust [36–38], see Fig. 5b.
Among other things, the solution (expressed in terms of theta functions) describes a
parametric collapse of the disk with a final phase transition to an extreme black hole.
Obviously, the integration method under discussion can be used to construct the
Kerr solution as the unique solution of the one-horizon boundary problem [39]. This
corresponds to the methods of electrostatics and is an alternative to “complex tricks”
and other formal derivations of the Kerr metric in the textbooks.
Extending the vacuum examinations to Einstein-Maxwell fields, Meinel [40] has
recently constructed the Kerr-Newman solution by integrating a 3 × 3 electrovacuum
LP, see Fig. 5c and thus proved the uniqueness of the solution.
Acknowledgments We would like to thank Reinhard Meinel, Marcus Ansorg and Andreas
Kleinwächter for many valuable discussions and Ben Whale for commenting on the manuscript.
References
1. Bach, R., Weyl, H.: Neue Lösungen der Einsteinschen Gravitationsgleichungen. Mathemat. Z.
13, 134 (1922). [Republication in English: Gen. Relativ. Gravit. 44, 817, (2012)]
2. Neugebauer, G., Hennig, J.: Non-existence of stationary two-black-hole configurations. Gen.
Relativ. Gravit. 41, 2113 (2009). doi:10.1007/s10714-009-0840-8
3. Hennig, J., Neugebauer, G.: Non-existence of stationary two-black-hole configurations: the
degenerate case. Gen. Relativ. Gravit. 43, 3139 (2011). doi:10.1007/s10714-011-1228-0
4. Neugebauer, G., Hennig, J.: Stationary two-black-hole configurations: a non-existence proof.
J. Geom. Phys. 62, 613 (2012). doi:10.1016/j.geomphys.2011.05.008
5. Neugebauer, G.: A general integral of the axially symmetric stationary Einstein equations. J.
Phys. A 13, L19 (1980). doi:10.1088/0305-4470/13/2/003
6. Neugebauer, G.: Gravitostatics and rotating bodies. In: Hall, G.S., Pulham J.R. (eds.) Pro-
ceedings of 46th Scottish Universities Summer School in Physics (Aberdeen), Copublished by
SUSSP Publications, Edinburgh, and Institute of Physics Publishing, London (1996)
7. Kramer, D., Neugebauer, G.: The superposition of two Kerr solutions. Phys. Lett. A 75, 259
(1980). doi:10.1016/0375-9601(80)90556-3
8. Dietz, W., Hoenselaers, C.: Two mass solution of Einstein’s vacuum equations: the double Kerr
solution. Ann. Phys. 165, 319 (1985). doi:10.1016/0003-4916(85)90301-X
9. Hoenselaers, C.: Remarks on the double-Kerr-solution. Prog. Theor. Phys. 72, 761 (1984).
doi:10.1143/PTP.72.761
10. Hoenselaers, C., Dietz, W.: Talk given at the GR10 meeting, Padova (1983)
11. Kihara, M., Tomimatsu, A.: Some properties of the symmetry axis in a superposition of two
Kerr solutions. Prog. Theor. Phys. 67, 349 (1982). doi:10.1143/PTP.67.349
12. Kramer, D.: Two Kerr-NUT constituents in equilibrium. Gen. Relativ. Gravit. 18, 497 (1980).
doi:10.1007/BF00770465
13. Krenzer, G.: Schwarze Löcher als Randwertprobleme der axialsymmetrisch-stationären
Einstein-Gleichungen, PhD Thesis, University of Jena (2000)
14. Manko, V.S., Ruiz, E.: Exact solution of the double-Kerr equilibrium problem. Class. Quantum
Grav. 18, L11 (2001). doi:10.1088/0264-9381/18/2/102
15. Manko, V.S., Ruiz, E., Sanabria-Gómez, J.D.: Extended multi-soliton solutions of the Einstein
field equations: II. Two comments on the existence of equilibrium states. Class. Quantum Grav.
17, 3881 (2000). doi:10.1088/0264-9381/18/2/102
16. Tomimatsu, A., Kihara, M.: Conditions for regularity on the symmetry axis in a superposition
of two Kerr-NUT solutions. Prog. Theor. Phys. 67, 1406 (1982). doi:10.1143/PTP.67.1406
17. Yamazaki, M.: Stationary line of N Kerr masses kept apart by gravitational spin-spin interaction.
Phys. Rev. Lett. 50, 1027 (1983). doi:10.1103/PhysRevLett.50.1027
18. Varzugin, G.: Equilibrium configuration of black holes and the inverse scattering method.
Theoret. Math. Phys. 111, 667 (1997). doi:10.1007/BF02634055
19. Varzugin, G.: The interaction force between rotating black holes in equilibrium. Theoret. Math.
Phys. 116, 1024 (1998). doi:10.1007/BF02557144
20. Belinskiı̆, V.A., Zakharov, V.E.: Integration of the Einstein equations by means of the inverse
scattering problem technique and construction of exact soliton solutions. Pis’ma Zh. Eksp.
Teor. Fiz. (in Russian) 75, 1955 (1978). [English translation: Sov. Phys. JETP 48, 985 (1978)]
21. Belinskiı̆, V.A., Zakharov, V.E.: Stationary gravitational solitons with axial symmetry. Pis’ma
Zh. Eksp. Teor. Fiz. (in Russian) 77, 3 (1979). [English translation: Sov. Phys. JETP 50, 1
(1979)]
22. Ansorg, M., Petroff, D.: Negative Komar mass of single objects in regular, asymptotically flat
spacetimes. Class. Quantum Grav. 23, L81 (2006). doi:10.1088/0264-9381/23/24/L01
23. Ansorg, M., Pfister, H.: A universal constraint between charge and rotation rate for degenerate
black holes surrounded by matter. Class. Quantum Grav. 25, 035009 (2008). doi:10.1088/0264-
9381/25/3/035009
24. Hennig, J., Ansorg, M., Cederbaum, C.: A universal inequality between the angular momentum
and horizon area for axisymmetric and stationary black holes with surrounding matter. Class.
Quantum Grav. 25, 162002 (2008). doi:10.1088/0264-9381/25/16/162002
25. Booth, I., Fairhurst, S.: Extremality conditions for isolated and dynamical horizons. Phys. Rev.
D 77, 084005 (2008). doi:10.1103/PhysRevD.77.084005
26. Chruściel, P.T., Eckstein, M., Nguyen, L., Szybka, S.J.: Existence of singularities in two-Kerr
black holes. Class. Quantum Grav. 28, 245017 (2011). doi:10.1088/0264-9381/28/24/245017
27. Dain, S., Reiris, M.: Area-angular momentum inequality for axisymmetric black holes. Phys.
28. Dain, S.: Geometric inequalities for axially symmetric black holes. Class. Quantum Grav. 29,
073001 (2012). doi:10.1088/0264-9381/29/7/073001
29. Carter, B.: Black hole equilibrium states, In: deWitt, C., deWitt, B. (eds.) Black Holes (Les
Houches). Gordon and Breach, London (1973)
30. Neugebauer, G.: Recursive calculation of axially symmetric stationary Einstein fields. J. Phys.
A 13, 1737 (1980). doi:10.1088/0305-4470/13/5/031
31. Neugebauer, G., Meinel, R.: Progress in relativistic gravitational theory using the inverse scat-
tering method. J. Math. Phys. 44, 3407 (2003). doi:10.1063/1.1590419
32. Hauser, I., Ernst, F.J.: Proof of a Geroch conjecture. J. Math. Phys. 22, 1051 (1981). doi:10.
1063/1.525012
33. Meinel, R., Ansorg, M., Kleinwächter, A., Neugebauer, G., Petroff, D.: Relativistic Figures of
Equilibrium. Cambridge University Press, Cambridge (2008)
34. Schoen, R.M., Yau, S.-T.: On the proof of the positive mass conjecture in general relativity.
Commun. Math. Phys. 65, 45 (1979). doi:10.1007/BF01940959
35. Schoen, R., Yau, S.T.: Proof of the positive mass theorem. II. Commun. Math. Phys. 79, 231
(1981). doi:10.1007/BF01942062
36. Neugebauer, G., Meinel, R.: The Einsteinian gravitational field of a rigidly rotating disk of
dust. Astrophys. J. 414, L97 (1993). doi:10.1086/187005
37. Neugebauer, G., Meinel, R.: General relativistic gravitational field of a rigidly rotating disk of
dust: axis potential, disk metric and surface mass density. Phys. Rev. Lett. 73, 2166 (1994).
38. Neugebauer, G., Meinel, R.: General relativistic gravitational field of a rigidly rotating disk of
dust: solution in terms of ultraelliptic functions. Phys. Rev. Lett. 75, 3046 (1995). doi:10.1103/
PhysRevLett.75.3046
39. Neugebauer, G.: Rotating bodies as a boundary value problems. Ann. Phys. (Leipzig) 9, 342
(2000)
40. Meinel, R.: Constructive proof of the Kerr-Newman black hole uniqueness including the ex-
treme case. Class. Quantum Grav. 29, 035004 (2012). doi:10.1088/0264-9381/29/3/035004
Dynamic and Thermodynamic Stability
of Black Holes and Black Branes
Robert M. Wald
Abstract I describe recent work with Stefan Hollands that establishes a new crite-
rion for the dynamical stability of black holes in D ≈ 4 spacetime dimensions in
general relativity with respect to axisymmetric perturbations: Dynamical stability is
equivalent to the positivity of the canonical energy, E , on a subspace of linearized
solutions that have vanishing linearized ADM mass, momentum, and angular mo-
mentum at infinity and satisfy certain gauge conditions at the horizon. We further
show that E is related to the second order variations of mass, angular momentum,
and horizon area by E = δ 2 M − i Ωi δ 2 Ji − (κ/8π )δ 2 A, thereby establishing a
close connection between dynamical stability and thermodynamic stability. Thermo-
dynamic instability of a family of black holes need not imply dynamical instability
because the perturbations towards other members of the family will not, in gen-
eral, have vanishing linearized ADM mass and/or angular momentum. However, we
prove that all black branes corresponding to thermodynamically unstable black holes
are dynamically unstable, as conjectured by Gubser and Mitra. We also prove that
positivity of E is equivalent to the satisfaction of a “local Penrose inequality,” thus
showing that satisfaction of this local Penrose inequality is necessary and sufficient
for dynamical stability.
1 Introduction
It is of considerable interest to determine the linear stability of black holes in (D-

dimensional) general relativity. It is also of interest to determine the linear stability of
the corresponding black branes in (D + p)-dimensions, i.e., spacetimes with metric
of the form
R. M. Wald (B)
Department of Physics, Enrico Fermi Institute, The University of Chicago,
Chicago, IL 60637, USA
e-mail: rmwa@uchicago.edu
230 R. M. Wald

p
p = ds D + dz i2 ,
2 2
d s̃ D+ (1)
i=1
where ds D2 is a black hole metric. In this paper, I will describe some recent general
results, obtained in collaboration with Stefan Hollands, on the stability of black holes
and black branes. In our work, we restrict consideration to vacuum general relativity
without a cosmological constant, but our methods are applicable to general theories
of gravity derived from a diffeomorphism covariant Lagrangian. A full account of
our results can be found in [1].
One can analyze the stability of a black hole or black brane by writing out the
linearized Einstein equation of the black hole or black brane background spacetime.
One can establish linear stability by finding a positive definite conserved norm for
perturbations. Linear instability can be established by finding a solution with (gauge
independent) unbounded growth in time. However, even in the very simplest cases—
such as the Schwarzschild black hole [2, 3] and the Schwarzschild black string [4]—it
is quite nontrivial to carry out the decoupling of equations and the fixing of gauge
needed to determine stability or instability directly from the equations of motion.
Furthermore, since this analysis depends on the details of the equations of motion,
it must be done on a case-by-case basis. Thus, it would be useful to have a much
simpler criterion for stability that can be applied to any black hole or black brane.
2 Dynamic and Thermodynamic Stability of Black

Holes and Black Branes
In ordinary thermodynamics, one has a very simple and general criterion for ther-
modynamic instability of a homogeneous system in thermal equilibrium. Consider
such a system, whose entropy, S, is a function of energy, E, and other extensive state
parameters X i , so that S = S(E, X i ). The condition for thermodynamic instability
is that the Hessian matrix
 2 ⎜
∂ S
2 ∂ S
∂ E2 ∂ Xi ∂ E
HS =  ⎟. (2)
∂2 S ∂2 S
∂ E∂ X i ∂ X i ∂ X j
admit a positive eigenvalue.1 This criterion arises from the fact that if the Hessian
had a positive eigenvalue, then one could increase total entropy by exchanging E
and/or X i between different parts of the system. To see this more explicitly, let
ξ0 = (E 0 , X i0 ) denote the parameter values of a particular thermodynamic state,
let v be an arbitrary vector in the thermodynamic state space, and consider the one-
1Note that for the case where E is the only state parameter, this criterion is equivalent to the system
having a negative heat capacity.
Dynamic and Thermodynamic Stability of Black Holes and Black Branes 231
parameter family ξ(λ) = ξ0 + λv of thermodynamic states. It is obvious that for this

family, we have d 2 E/dλ2 = d 2 X i /dλ2 = 0, whereas d 2 S/dλ2 |0 = Hess S |ξ0 (v, v).
Suppose now that, for our homogeneous system, we change the state parameters by
λv in one part of the system and compensate for this by changing the state parameters
by −λv in a different part of the system (of the same “size”). To first order in λ, there
will be no change in the total entropy. To second order in λ, the change in total entropy
will be proportional to Hess S |ξ0 (v, v). Thus, if H S admits a positive eigenvalue, one
can choose v so as to find a state of higher entropy at fixed (E, X i ) arbitrarily close
to the original thermal equilibrium state.
An equivalent statement of the criterion for thermodynamic instability of the state
ξ0 = (E 0 , X i0 ) is (assuming T > 0) that one can find a one parameter family ξ(λ)
of thermal equilibrium states with ξ(0) = ξ0 such that

δ2 E − T δ2 S − Yi δ 2 X i < 0, (3)
i
where Yi ⊗ (∂ E/∂ X i ) S and δ 2 denotes d 2 /dλ2 evaluated at λ = 0. To see this,

we note that for a one parameter family of the form ξ0 + λv, the left side is just
−T Hess S |ξ0 (v, v). However, by the first law of thermodynamics (i.e., the definitions
of Yi and T −1 ⊗ (∂ S/∂ E) X i ), the left side does not depend on the second order
change in the state, so Eq. (3) holds for the family ξ(λ) if and only if it holds for the
family ξ0 + λv with v = dξ/dλ|0 .
Black holes are thermodynamic systems, with
E ◦ M,
A
S◦ , (4)
4
X i ◦ Ji , Q i .
Thus, in the vacuum case (Q i = 0) a black hole would be said to be thermodynam-

ically unstable if the Hessian matrix
 ⎜
∂2 A ∂2 A
∂ 2 ∂ Ji ∂ M
HA =  2 M ⎟ (5)
∂ A ∂2 A
∂ M∂ Ji ∂ Ji ∂ J j
admits a positive eigenvalue. This is equivalent to finding a perturbation for which

κ 2
δ2 M − δ A− Ωi δ 2 Ji < 0. (6)
8π
i
One might expect that this condition for thermodynamic instability might imply
dynamical instability. However, this is clearly false: The Schwarzschild black hole
has negative heat capacity (A = 16π M 2 , so ∂ 2 A/∂ M 2 = 32π > 0) but is well
232 R. M. Wald
known to be dynamically stable. A black hole is not “homogeneous” in a manner

that would allow one to borrow energy and/or angular momentum from one part of
it and give it to another part in such a way as to increase the total entropy (area)
at fixed total energy and angular momentum in the manner described above for
thermodynamic systems.
However, a black brane is potentially homogeneous in this sense, and the Schwarz-
schild black string is known to be unstable [4]. The Gubser-Mitra conjecture [5, 6]
states that the above thermodynamic criterion for instability is a valid criterion dy-
namical instability for black branes. As described further below, our work provides
a proof of the Gubser-Mitra conjecture, which follows as a consequence of a more
fundamental stability criterion that we shall establish.
Another simple possible stability criterion that is applicable to black holes is the
“local Penrose inequality,” discussed in [7]. We reformulate this criterion as follows:
Suppose one has a family of stationary, axisymmetric black holes parametrized by
M and angular momenta J1 , . . . , J N . Consider a one-parameter family gab (λ) of
axisymmetric spacetimes, with gab (0) being a member of this family with surface
gravity κ > 0. Consider initial data on a hypersurface Σ passing through the bifurca-
tion surface B. By the linearized Raychaudhuri equation, to first order in λ, the event
horizon coincides with the apparent horizon on Σ. They need not coincide to second
order in λ, but since B is an extremal surface in the background spacetime, their
areas must agree to second order. Let A (λ) denotes the area of the apparent horizon
of gab (λ), and let Ā(λ) denote the area of the event horizon of the stationary black
hole in the family with the same mass and angular momentum as gab (λ). Suppose
that to second order, we have
δ 2 A > δ 2 Ā.
Since (i) the area of the event horizon can only increase with time (by cosmic cen-
sorship), (ii) the final mass of the black hole cannot be larger than the initial total
mass (by positivity of Bondi flux), (iii) its final angular momenta must equal the ini-
tial angular momenta (by axisymmetry), and (iv) Ā(M, J1 , . . . , J N ) is an increasing
function of M at fixed Ji (by the first law of black hole mechanics with κ > 0), it
follows that there would be a contradiction if the perturbed black hole solution were
to settle down to a stationary black hole in the family. This implies that satisfaction of
this inequality implies instablity—although it does not imply stability if δ 2 A √ δ 2 Ā
always holds. As discussed further below, our more fundamental stability criterion
implies that satisfaction of δ 2 A √ δ 2 Ā is necessary and sufficient for the dynamical
stability of black holes with respect to axisymmetric perturbations.
Our results are based upon identities arising from the Lagrangian formulation
of general relativity. Although we restrict consideration here to vacuum general
relativity, these formulas can be generalized to allow for the presence of matter fields
and, indeed, they can be generalized to an arbitrary diffeomorphism covariant theory
of gravity [8], provided only that the field equations are derived from a Lagrangian.
The key identities we use are obtained as follows:
The Lagrangian D-form for vacuum general relativity in D dimensions is
1
L a1 ···a D = R εa1 ···a D . (7)
16π
Its first variation yields
δL = E · δg + dθ, (8)
where E = 0 is the vacuum Einstein field equation and the (D − 1)-form θ (g, δg) is
the “boundary term” that is usually discarded when the variation of L is performed
under an integral sign. Explicitly, we have
1 ac bd
θa1 ···a D−1 = g g (∗d δgbc − ∗c δgbd )εaa1 ···a D−1 . (9)
16π
The symplectic current (D − 1)-form is defined by
ω(g; δ1 g, δ2 g) = δ1 θ (g; δ2 g) − δ2 θ (g; δ1 g). (10)
The symplectic form, WΣ (g; δ1 g, δ2 g), is obtained by integrating ω over a Cauchy

surface Σ ⎩
WΣ (g; δ1 g, δ2 g) ⊗ ω(g; δ1 g, δ2 g). (11)
Σ
It can be shown to be given by [9]

⎩
1
WΣ (g; δ1 g, δ2 g) = − (δ1 h ab δ2 pab − δ2 h ab δ1 pab ), (12)
32π
Σ
where
pab ⊗ h 1/2 (K ab − h ab K ), (13)
where K ab is the extrinsic curvature of Σ.

The Lagrangian L, Eq. (7), is diffeomorphism covariant. An arbitrary vector field
X a is the generator of an infinitesimal diffeomorphism, and associated to X a is a
conserved Noether current (D − 1)-form, defined by
J X ⊗ θ (g, L X g) − X · L , (14)
where X · L denotes the (D − 1)-form X a L aa1 ···a D−1 . It can be shown quite generally
[10] that J X always can be written in the form
JX = X · C + d Q X , (15)
where C = 0 are the constraint equations [11] of the theory, and where the (D − 2)-
form Q X is called the Noether charge.
234 R. M. Wald
We now take the first variation of J X , using Eqs. (14) and (15) as well as Eqs. (8)
and (10). We thereby obtain the following fundamental variational identity:
ω(g; δg, L X g) = X · [E(g) · δg] + X · δC + d [δ Q X (g) − X · θ (g; δg)] . (16)
It should be emphasized that Eq. (16) holds for an arbitrary metric gab (not necessarily
a solution to the field equations), an arbitrary metric perturbation δgab (not necessarily
a solution to the linearized field equations) and an arbitrary vector field X a .
By definition, a Hamiltonian, h X , for the “time evolution” generated by X a is a
function on phase space whose first variation satisfies
δh X = WΣ (g; δg, L X g) (17)
if and only if gab satisfies the equations of motion E = 0. By Eq. (16), if a Hamil-
tonian h X conjugate to X a exists, its first variation must satisfy
⎩
δh X = (X · δC + d [δ Q X (g) − X · θ (g; δg)]) . (18)
Σ
For asymptotically flat spacetimes, this motivates the definition of the ADM con-
served quantity, H X , associated with an asymptotic symmetry X a , as the quantitiy
defined for solutions whose first variation is given by
⎩
δ HX = [δ Q X (g) − X · θ (g; δg)]. (19)
⇔
Now consider a stationary black hole solution (E = 0) with surface gravity κ > 0,
so the event horizon is of “bifurcate type,” with bifurcation surface B. Let Σ be a
Cauchy surface for the exterior region, so that it extends from spatial infinity to B.
We choose X to be the horizon Killing field

K a = ta + Ωi φia . (20)
Finally, let γ = δg satisfy the linearized constraint equations δC = 0. Integration of

the fundamental identity (16) over Σ—using L X g = 0, E = 0, and δC = 0—then
yields the first law of black hole mechanics [8]
κ
0 = δM − Ωi δ Ji − δ A. (21)
8π
i
To proceed further, we impose two gauge conditions at B on our perturbation

γ = δg. The first condition,
δϑ| B = 0, (22)
ensures that the location of the horizon does not change to first order. (Here ϑ| B
denotes the expansion of the outgoing null geodesics from B.) The second condition
is
δA
δε| B = ε, (23)
A
where ε| B denotes the surface area element on B. The imposition of these condi-
tions does not involve any loss of generality, i.e., they can be imposed for arbitrary
perturbations [1].
We define the canonical energy of a perturbation γ by
E ⊗ WΣ (g; γ , Lt γ ) . (24)
The second variation of our fundamental identity (16) then yields (for axisymmetric
perturbations)
κ 2
E = δ2 M − Ωi δ 2 Ji − δ A. (25)
8π
i
Thus positivity of E for all perturbations γ is equivalent to thermodynamic stability

(see Eq. (6)).
Our results on dynamical stability follow from various properties of E . To establish
these properties, it is useful to view E as a quadratic form on perturbations:
E (γ1 , γ2 ) = WΣ (g; γ1 , Lt γ2 ). (26)
In [1], we proved that E satisfies the following properties:
• E is conserved, i.e., it takes the same value if evaluated on another Cauchy surface
Σ ≤ extending from spatial infinity to B.
• E is symmetric, E (γ1 , γ2 ) = E (γ2 , γ1 ).
• When restricted to perturbations for which δ A = 0 and δ Pi = 0 (where Pi is the
ADM linear momentum), E is gauge invariant.
• When restricted to the subspace, V , of perturbations for which δ M = δ Ji =
δ Pi = 0 (and hence, by the first law of black hole mechanics δ A = 0), we have
E (γ ≤ , γ ) = 0 for all γ ≤ ≡ V if and only if γ is a perturbation towards another
stationary and axisymmetric black hole.
Thus, if we restrict to perturbations in the subspace, V ≤ , of perturbations in V

modulo perturbations towards other stationary black holes, then E is a non-degenerate
quadratic form. Consequently, on V ≤ , either (a) E is positive definite or (b) there is
a ψ ≡ V ≤ such that E (ψ) < 0. If (a) holds, then E provides a positive definite
conserved norm on perturbations. Thus, if (a) holds, we have stability.
To analyze case (b), we must consider the flux of E through null infinity, I + ,
and through the black hole horizon, H . Let δ Nab denote the perturbed Bondi news
tensor at null infinity and let δσab denote the perturbed shear on the horizon. If the
perturbed black hole were to “settle down” to another stationary black hole at late
236 R. M. Wald
times, then δ Nab ∇ 0 and δσab ∇ 0 at late times. In [1], we showed that—for
axisymmetric perturbations—the change in canonical energy is then given by
⎩ ⎩
1 1
ΔE = − δ Ñcd δ Ñ cd − (K a ∗a u) δσcd δσ cd √ 0. (27)
16π 4π
I H
Thus, E can only decrease. Therefore if one has a perturbation ψ ≡ V such that
E (ψ) < 0, then ψ cannot “settle down” to a stationary solution at late times because
E = 0 for stationary perturbations with δ M = δ Ji = δ Pi = 0. Thus, in case (b) we
have instability.
The above results show that the necessary and sufficient condition for stability of
a black hole (or black brane) with respect to axisymmetric perturbations is positivity
of E on a Hilbert space, V , of perturbations with vanishing perturbed mass, angular
momentum, and linear momentum, δ M = δ Ji = δ Pi = 0. This is our fundamental
criterion for the dynamical stability of black holes and black branes. In view of
Eqs. (6) and (25), it follows that dynamical stability is equivalent to thermodynamic
stability on the subspace of perturbations that satisfy δ M = δ Ji = δ Pi = 0.
The restriction that δ M = δ Ji = δ Pi = 0 can be removed for the case of black
branes as a consequence of the following theorem [1]:
Theorem 1 Suppose a family of black holes parametrized by (M, Ji ) is thermo-

dynamically unstable at (M0 , J0i ), i.e., there exists a perturbation within the black
hole family for which E < 0. Then, for any black brane corresponding to (M0 , J0i )
one can find a sufficiently long wavelength perturbation for which E˜ < 0 and
δ M̃ = δ J˜i = δ P̃i = δ Ã = δ T̃i = 0 (where T̃i denotes the momenta conjugate
to the translational symmetries of the brane).
This theorem is proven by starting with the initial data for the perturbation to
another black hole with E < 0, multiplying it by exp(ikz)—where “z” denotes a
brane coordinate as in Eq. (1)—and then re-adjusting the initial data so that it satisfies
the constraints. The new data will automatically satisfy δ M̃ = δ J˜A = δ P̃i = δ Ã =
δ T̃i = 0 because of the exp(ikz) factor. For sufficiently small k, it can be shown to
satisfy E˜ < 0.
The above theorem, together with our fundamental criterion for dynamical stabil-
ity, proves the Gubser-Mitra conjecture. To illustrate the nature of this result, consider
the one-parameter family of Schwarzschild black holes, parametrized by mass M. It
is easily seen that the “change of mass” perturbation has E < 0. However, this tells
one nothing about the stability of Schwarzschild black holes because, obviously, for
these perturbations we have δ M ⇒= 0, so they do not “count” for testing stability of
the Schwarzschild black hole. However, the fact that E < 0 for this “change of mass”
perturbation proves the instability of Schwarzschild black branes to sufficiently long
wavelength perturbations.
The equivalence of the satisfaction of the local Penrose inequality to our fun-
damental stability criterion for black holes can be seen as follows. As above, let
ḡab (M, Ji ) be a family of stationary, axisymmetric, and asymptotically flat black
hole metrics on M. Let gab (λ) be a one-parameter family of axisymmetric met-

rics such that gab (0) = ḡab (M0 , J0i ). Let M(λ), Ji (λ) denote the mass and angu-
lar momenta of gab (λ) and let A (λ) denote the area of its apparent horizon. Let
ḡab (λ) = ḡab (M(λ), Ji (λ)) denote the one-parameter family of stationary black
holes with the same mass and angular momenta as gab (λ). We have the following
result:
Theorem 2 There exists a one-parameter family gab (λ) for which
A (λ) > A¯(λ) (28)
≤ of ḡ (M , (J )
to second order in λ if and only if there exists a perturbation γab ab 0 0i
≤
with δ M = δ Ji = δ Pi = 0 such that E (γ ) < 0.
Proof The first law of black hole mechanics implies A (λ) = A¯(λ) to first order in
λ, so what counts are the second order variations. Since the families have the same
mass and angular momenta, we have
⎫ ⎬
κ d 2A d 2 A¯
(0) − (0) = E (γ̄ , γ̄ ) − E (γ , γ )
8π dλ2 dλ2
= −E (γ ≤ , γ ≤ ) + 2E (γ ≤ , γ̄ )
= −E (γ ≤ , γ ≤ ),
where γ ≤ = γ̄ − γ . ⊥
∈
In summary, the remarkable relationship between the laws of black hole physics
and the laws of thermodynamics has been shown to extend to dynamical stability.
This research was sponsored in part by NSF grants PHY-0854807 and PHY-
1202718 to the University of Chicago.
References
1. Hollands, S, Wald, R.M.: Stability of black holes and black branes. Commun. Math. Phys. 313,
257–290 (2012). doi:10.1007/s00220-012-1638-1
2. Regge, T, Wheeler, J.A.: Stability of a Schwarzschild singularity. Phys. Rev. 108, 1063 (1957).
doi:10.1103/PhysRev.108.1063
3. Zerilli, F.J.: Effective potential for even-parity Regge-Wheeler gravitational perturbation equa-
tions. Phys. Rev. Lett. 24, 737 (1970). doi:10.1103/PhysRevLett.24.737
4. Gregory, R., Laflamme, R.: Black strings and p-branes are unstable. Phys. Rev. Lett. 70, 2837
(1993). doi:10.1103/PhysRevLett.70.2837
5. Gubser, S.S., Mitra, I.: The Evolution of unstable black holes in anti-de Sitter space. J. High
Energy Phys. 2001(08), 018 (2001). doi:10.1088/1126-6708/2001/08/018
6. Gubser, S.S., Mitra, I.: Instability of charged black holes in anti-de Sitter space. ArXiv e-prints
arxiv:hep-th/0009126 (2000)
238 R. M. Wald
7. Figueras, P., Murata, K., Reall, H.S.: Black hole instabilities and local Penrose inequalities.
Class. Quant. Grav. 28, 225030 (2011). doi:10.1088/0264-9381/28/22/225030
8. Iyer, V., Wald, R.M.: Some properties of the Noether charge and a proposal for dynamical
black hole entropy. Phys. Rev. D 50, 846 (1994). doi:10.1103/PhysRevD.50.846
9. Burnett, G.A., Wald, R.M.: A conserved current for perturbations of Einstein-Maxwell space-
times. Proc. R. Soc. Lond. A 430, 57 (1990). doi:10.1098/rspa.1990.0080
10. Iyer, V., Wald, R.M.: Comparison of the Noether charge and Euclidean methods for computing
the entropy of stationary black holes. Phys. Rev. D 52, 4430 (1995). doi:10.1103/PhysRevD.
52.4430
11. Seifert, M.D., Wald, R.M.: General variational principle for spherically symmetric perturba-
tions in diffeomorphism covariant theories. Phys. Rev. D 75, 084029 (2007). doi:10.1103/
PhysRevD.75.084029
Instability of Anti-de Sitter Spacetime
Piotr Bizoń and Andrzej Rostworowski
Abstract In this talk we summarize our recent numerical and perturbative

calculations which indicate that AdS spacetime is unstable. Namely, we study spheri-
cally symmetric Einstein-massless-scalar field equations with negative cosmological
constant and show that this system is unstable against black hole formation for a large
class of initial data arbitrarily close to the AdS solution. We conjecture that this insta-
bility is triggered by a resonant mode mixing which gives rise to diffusion of energy
from low to high frequencies.
1 Introduction
Anti-de Sitter (AdS) spacetime is the unique maximally symmetric solution of the
vacuum Einstein equations G εγ + Λgεγ = 0 with negative cosmological constant
Λ. In d + 1 dimensions the AdS metric in global dimensionless coordinates (t ∈
R, x ∈ [0, ν/2), λ ∈ S d−1 ) reads
Φ2
ds 2 = −dt 2
+ d x 2
+ sin 2
x dλ 2
S d−1 ,
cos2 x
where Φ2 = −d(d − 1)/2Λ sets the length scale. Conformal infinity x = ν/2 is
the timelike cylinder I = R × S d−1 with the boundary metric dsI 2 = −dt 2 +
2 2
sin x dλ S d−1 .
Asymptotically AdS spacetimes (that is spacetimes which share the conformal
boundary with AdS but may be very different in the bulk, in particular may contain
P. Bizoń · A. Rostworowski (B)

Institute of Physics, Jagiellonian University, Kraków, Poland
e-mail: arostwor@th.if.uj.edu.pl
P. Bizoń
Max Planck Institute for Gravitational Physics (Albert Einstein Institute), Golm, Germany
240 P. Bizoń and A. Rostworowski
horizons) have come to play a central role in theoretical physics, prominently due
to the AdS/CFT correspondence which conjectures a duality between gravity in the
AdS bulk and a quantum field theory on the conformal boundary at infinity. By the
positive energy theorem, AdS spacetime is a ground state among asymptotically AdS
spacetimes, much as Minkowski spacetime is a ground state among asymptotically
flat spacetimes. However, the evolutions of small perturbations of these ground states
are different. In the case of Minkowski, small perturbations disperse to infinity and
the spacetime is asymptotically stable [1]. In contrast, asymptotic stability of AdS is
precluded because the conformal boundary acts like a mirror at which perturbations
propagating outwards bounce off and return to the bulk that results in complex
nonlinear wave interactions in an effectively bounded domain. Understanding of
these interactions is the key to the problem of stability of AdS spacetime.
2 Instability of Anti-de Sitter Spacetime
In this talk we summarize our recent numerical and perturbative calculations

[2, 3] which indicate that AdS spacetime is unstable. To make the problem tractable
we assume spherical symmetry. Since by Birkhoff’s theorem spherically symmetric
vacuum solutions are static, we need to add matter to generate dynamics. A simple
matter model is the minimally coupled massless scalar field

1
G εγ + Λgεγ = 8ν G αε ∂ αγ ∂ − gεγ (α∂)2 , g εγ ∇ε ∇γ ∂ = 0 . (1)
2
Let us recall that in the asymptotically flat case (Λ = 0) this model has led to impor-
tant insights, such as the proof of the weak cosmic censorship by Christodoulou
[4, 5] and the discovery of critical phenomena at the threshold for black hole forma-
tion by Choptuik [6].
We use the following parametrization of asymptotically AdS spacetimes
Φ2 −2β 2 −1

ds 2 = −Ae dt + A d x 2
+ sin 2
x dλ 2
S d−1
,
cos2 x
where A and β are functions of (t, x). As our results are qualitatively the same for
all d ≥ 3, for concreteness we set d = 3 herefater. Under the above assumptions
the Einstein-massless scalar field equations (1) reduce to the quasilinear hyperbolic-
elliptic system consisting of the scalar wave equation
1
αt A−1 eβ αt ∂ = α x tan 2
x A e −β
α x ∂ (2)
tan2 x
and two constraint equations (we set 4ν G = 1)
Instability of Anti-de Sitter Spacetime 241
1 + 2 sin2 x
αx A = (1 − A) − sin x cos x A ω, αx β = − sin x cos x ω , (3)
sin x cos x
where
ω = A−2 e2β (αt ∂)2 + (αx ∂)2
is the scalar field energy density. This system has a one-parameter family of static
solutions (∂ = 0, β = 0, A = 1 − M cos3 x/ sin x) which are Schwarzschild-AdS
black holes for M > 0 and the pure AdS for M = 0.
We restrict our attention to smooth solutions with finite mass
ν/2
1
M := Aω tan2 x d x.
2
0
It follows that near x = ν/2 we must have (using y = ν/2 − x)

∂(t, x) = f ∞ (t) y 3 +O y 5 , β(t, x) = β∞ (t)+O y 6 , A(t, x) = 1−2M y 3 +O y 6 .
The local well-posedness of the above initial-boundary value problem was proved
in [7].
The dynamics of solutions starting from small initial data
˙ |t=0 = (ρ f (x), ρg(x))

(∂, ∂)
can be approximated using weakly nonlinear perturbation analysis. To this end we

expand the solution in the perturbation series
∂ = ρ∂1 + ρ3 ∂3 + · · · , β = ρ2 β2 + ρ4 β4 + · · · , 1 − A = ρ2 A2 + ρ4 A4 + · · · ,
where (∂1 , ∂˙ 1 )|t=0 = ( f (x), g(x)) and (∂ j , ∂˙ j )|t=0 = (0, 0) for j > 1. Inserting
this expansion into the field equations (2) and (3) and collecting terms of the
same order in ρ, we obtain a hierarchy of linear equations which can be solved
order-by-order. At the first order we get the linear wave equation
∂¨1 + L∂1 = 0,
where
1
L=− α x tan 2
x α x
tan2 x
is an essentially self-adjoint operator on L 2 ([0, ν/2], tan2 x d x). The eigenvalues of

L are λ2j = (3 + 2 j)2 ( j = 0, 1, . . . ) which implies that AdS is linearly stable.
The corresponding orthonormal eigenfunctions are
( 1 , 23 )
e j (x) = d j cos3 x P j 2 (cos 2x) ,
where d j is a normalization factor. Thus, at the linear level the solution is
∞

∂1 (t, x) = a j cos(λ j t + γ j ) e j (x),
j=0
where amplitudes a j and phases γ j are determined by the initial data. Using this
solution at the second order we get perturbations of the metric functions A2 and β2
(so called backreaction) and at the third order we obtain an inhomogeneous linear
wave equation of the form ∂¨ 3 + L∂3 = S(∂1 , A2 , β2 ). A calculation shows that in
general ∂3 contains secular terms that grow linearly in time. They are due to four-wave
resonances present in the Fourier decomposition of the source S. We interpret this
breakdown of the perturbation analysis as indicating the onset of instability at time
of order O(ρ−2 ). We believe that the secular terms appearing in ∂3 are progenitors of
the higher-order resonant mode mixing which shifts the energy spectrum to higher
frequencies. This heuristics is corroborated by numerical simulations which show
that, indeed, generic perturbations start to grow rapidly after a time that scales as
ρ−2 . This growth eventually leads to the formation of a horizon.
To demonstrate the transfer of energy to higher frequencies we define the Fourier
coefficients
Θ j := (A1/2 αx ∂, αx e j ) and τ j := (A−1/2 eβ αt ∂, e j )
and express the mass as the Parseval sum

∞

M= E j (t),
j=0
where
E j := τ 2j + λ−2
j Θj
2
is the j-mode energy. The evolution of the energy spectrum, that is the distribution
of mass among the modes, is depicted in Fig. 1 for Gaussian initial data. Initially,
the energy is concentrated in low modes; the exponential cutoff of the spectrum
expresses the smoothness of initial data. During the evolution the range of excited
modes increases and the spectrum becomes broader. Just before horizon formation
the spectrum exhibits the power-law scaling E j ∼ j −ε with exponent ε ≈ 1.2.
This value seems to be universal, i.e., the same for all initial data (but it changes
with dimension d). Note that the formation of a black hole provides a cutoff for
Instability of Anti-de Sitter Spacetime 243
0.1
t=0
t=1100
t=1495
0.01
Ej /M
0.001
0.0001
10 100 1000
j
Fig. 1 Log-log plot of the energy spectrum at three moments of time: initial, intermediate, and just
before collapse. The fit of the power law E j ∼ j −ε at time t = 1495 gives the slope ε ≈ 1.2
the turbulent energy cascade (in amusing analogy to viscosity for the turbulent cas-
cade in fluids). Clearly, the formation of the power-law spectrum reflects the loss of
smoothness of the solution during collapse; it would be very interesting compute ε
analytically.
To summarize, our numerical simulations and formal nonlinear perturbation
analysis lead us to conjecture that anti-de Sitter space is unstable against the forma-
tion of a black hole under arbitrarily small generic perturbations. We wish to stress the
genericity condition in the above conjecture: we do not claim that all perturbed solu-
tions end up as black holes. On the contrary, in [2] we gave evidence for the existence
of non-generic solutions that remain non-singular for very long (possibly infinite)
time. In particular, preliminary calculations based on the Poincaré-Lindstedt method
indicate the existence of time periodic solutions. A similar conjecture (existence of
geons) was put forward by Dias et al. [8] for the vacuum Einstein equations.
The results described above have opened up new and unexpected research paths
lying at the interface of classical general relativity and turbulence theory. Explo-
ration of these paths will hopefully lead to better understanding of the dynamics of
asymptotically AdS spacetimes, which in turn may have interesting implications in
gauge/gravity dualities.
References
1. Christodoulou, D., Klainerman, S.: The Global Nonlinear Stability of the Minkowski Space,
Princeton Mathematical Series, vol. 41. Princeton University Press, Princeton (1993)
2. Bizoń, P., Rostworowski, A.: Weakly turbulent instability of anti-de Sitter spacetime. Phys. Rev.
3. Jałmużna, J., Rostworowski, A., Bizoń, P.: AdS collapse of a scalar field in higher dimensions.
4. Christodoulou, D.: The problem of a self-gravitating scalar field. Commun. Math. Phys. 105,
337 (1986). doi:10.1007/BF01205930
5. Christodoulou, D.: A mathematical theory of gravitational collapse. Commun. Math. Phys. 109,
613 (1987). doi:10.1007/BF01208960
6. Choptuik, M.W.: Universality and scaling in gravitational collapse of a massless scalar field.
7. Holzegel, G., Smulevici, J.: Self-gravitating Klein-Gordon fields in asymptotically anti-de-Sitter
spacetimes. Ann. Henri Poincaré 13, 991 (2012). doi:10.1007/s00023-011-0146-8
8. Dias, Ó.J.C., Horowitz, G.T., Santos, J.E.: Gravitational turbulent instability of anti-de Sitter
space. Class. Quantum Gravity 29(19), 194002 (2012). doi:10.1088/0264-9381/29/19/194002
Higher-Dimensional Black Holes
Harvey S. Reall
Abstract This article reviews black hole solutions of higher-dimensional General

Relativity. The focus is on stationary vacuum solutions and recent work on instabil-
ities of such solutions.
1 Introduction
General Relativity (GR) in D > 4 spacetime dimensions has been actively inves-
tigated for more than a decade. There are several reasons for this interest in higher
dimensions, and higher-dimensional black holes in particular.
1. Statistical calculation of black hole entropy using string theory. This was first
done for certain D = 5 black holes [1]. Each entropy calculation is a check on the
theory, irrespective of the dimension. Hence the study of higher-dimensional black
holes is a worthwhile contribution to developing a theory of quantum gravity.
2. The gauge/gravity correspondence [2] relates the properties of black holes in D
dimensions to strongly coupled, finite temperature, quantum field theory in D − 1
dimensions. This provides a way of calculating certain field theory quantities
which are very hard to determine by more traditional methods.
3. The possibility of producing tiny higher-dimensional black holes at colliders in
certain “brane-world” scenarios [3].
4. Higher-dimensional black hole spacetimes might have useful mathematical prop-
erties. For example, analytically continued versions of black hole solutions have
been used to obtain explicit metrics on compact Sasaki-Einstein spaces [4].
H. S. Reall (B)
Department of Applied Mathematics and Theoretical Physics, University of Cambridge,
Wilberforce Road, Cambridge CB3 0WA, UK
e-mail: hsr1000@cam.ac.uk
246 H. S. Reall
5. An explicit higher-dimensional solution might provide a clean example of some

important effect in GR. A nice example of this is the frame-dragging effect exhib-
ited by the “black Saturn” solution (see below).
6. Progress in quantum field theory has been made by considering D different from
4, and fields different from those of the Standard Model. In the same spirit, perhaps
we will arrive at a better understanding of GR by allowing the parameter D to
take values other than 4 [5].
This article is a brief, selective, review of higher-dimensional black hole solu-

tions. The scope is limited to solutions of the vacuum Einstein equation without
cosmological constant. I shall discuss two types of black holes. (i) Black hole solu-
tions of Kaluza-Klein theory that are static in the higher-dimensional sense. This
includes black string solutions. (ii) Asymptotically flat black hole solutions. In each
case, there has been recent progress in demonstrating the existence of instabilities of
certain solutions and so special attention is given to this topic.
We close this introduction by presenting the simplest higher-dimensional black
hole, the D-dimensional Schwarzschild solution:
dr 2 r D−3
+
ds 2 = − f dt 2 + + r 2 dΩ D−2
2
, f =1− (1)
f r
2
where dΩ D−2 is the line-element on a unit round S D−2 and the event horizon is at
r = r+ .
2 Black Holes in Kaluza-Klein Theory
Consider vacuum GR with a compact Kaluza-Klein circle. We are interested in

black hole solutions of this theory which are asymptotically flat in a Kaluza-Klein
sense, which means that, at large distance in the non-compact directions, the metric
approaches that of Minkowski space with a compact circle of circumference L:
ds 2 ≈ −dt 2 + dr 2 + r 2 dΩ D−2
2
+ dz 2 , z∼z+L (2)
I shall describe the known static black hole solutions of this theory. For a more
detailed recent review see [6].
The simplest such black hole solution is the product of the (D − 1)-dimensional
Schwarzschild solution with a circle of circumference L. This gives a black string
solution for which cross-sections of the event horizon have topology S 1 × S D−3 . In
the decompactified limit L → ∞ it gives a black string of infinite length.
Another solution of this theory describes a black hole of topology S D−2 localized
on the Kaluza-Klein (KK) circle. Such solutions are not known explicitly. Solutions
describing black holes with radius much smaller than L have been constructed per-
turbatively [7, 8]. Larger black hole solutions have been constructed numerically for
Higher-Dimensional Black Holes 247
D = 5, 6 [9, 10]. They cannot become arbitrarily large: there is an upper bound
on their mass determined by L. In fact, as one moves along this family of solutions,
starting from a small black hole, the mass increases to a maximum and then decreases
[10, 11].
The higher-dimensional Schwarzschild solution is stable against linearized gravi-
tational perturbations [12]. However, black strings suffer from the Gregory-Laflamme
(GL) instability [13]. If r+ /L is less than a certain D-dependent critical value then
there exist linearized gravitational perturbations which grow exponentially with time.
These perturbations break the translational symmetry around the KK circle. If r+ /L
exceeds the critical value then the string is believed to be stable. So “thin” strings are
unstable and “fat” strings are stable. Taking the limit L → ∞ shows that uncom-
pactified black strings are unstable.
Since its discovery, the nonlinear evolution of this instability has been the subject
of considerable debate. It was first proposed that the black string would “pinch off” to
form a localized KK black hole or holes [13]. This cannot happen classically owing
to the result that black holes cannot bifurcate. Instead, a singularity would have to
form at the horizon at the moment of the pinching off, but perhaps this is resolved
by quantum effects.
Recent numerical simulations (with D = 5) support this picture [14]. These start
from a perturbation which is sinusoidal along the flat direction. When evolved, the
perturbation becomes more inhomogeneous, with the configuration reaching a tran-
sient state resembling a line of localized black holes connected by thin threads of black
string. However, these threads are in turn unstable and suffer the Gregory-Laflamme
instability, but on a shorter timescale, leading to smaller black holes connected by
even thinner strings. The process appears to continue on smaller and smaller scales,
in a self-similar manner, but in a finite total time as measured by an observer far from
the string. The curvature of (parts of) the horizon becomes large in this process, so
it seems that a naked singularity does indeed form. This is strong evidence against
the validity of the cosmic censorship hypothesis in higher dimensions.
The black strings described above are invariant under translations around the
KK circle. For this reason they are called uniform black strings. Some time ago, it
was conjectured that there should exist a 1-parameter family of static nonuniform
black strings which lack this translational symmetry, and bifurcate from the uniform
black string family at the critical value of r+ /L discussed above [15]. Such solutions
were subsequently constructed perturbatively, for infinitesimal non-uniformity [16].
Fully nonlinear solutions have been constructed numerically [17–20]. Just like the
localized KK black holes, there is an upper bound on their mass determined by L.
Kol [21] conjectured that the families of localized KK black hole solutions and
non-uniform black strings should merge at a common limiting solution. The idea is
that, moving along the family of localized black holes, they increase in size until
they fill the KK circle, and then transition to a black string of high non-uniformity.
Numerical evidence supporting this suggestion was obtained in [9–11].
The perturbative construction of non-uniform black strings reveals that, for
D ≤ 13, infinitesimally non-uniform black strings have lower horizon area than a
uniform black string of the same mass, whereas for D > 13, they have greater hori-
248 H. S. Reall
D−2
A/L
localized
black holes
merger
point
uniform
black strings
non-uniform
black strings
GL point
D− 3
M/L
Fig. 1 Schematic plot of horizon area against mass for Kaluza-Klein black holes/strings with
D = 5, 6 based on results of Kudoh and Wiseman [10], Headrick et al. [11]. The dashed curve is
the uniform black string branch, the solid curve the non-uniform string branch and the dotted curve
the localized black hole branch. It seems likely that this will be the qualitative behaviour for all
D ≤ 11 (Plot reproduced from [20])
zon area [22]. This suggests that infinitesimally non-uniform black strings should
be classically unstable for D ≤ 13 but stable for D > 13. It also suggests that a
non-uniform black string could be an endpoint of the GL instability for D > 13.
Very recently, Figueras et al. [20] has performed the first study of the stability
of non-uniform black strings with finite non-uniformity. The results confirm the
perturbative results for infinitesimal non-uniformity. It was found that the instability
persists to large non-uniformity for D ≤ 11. Strong evidence that non-uniform
strings with D > 13 are all stable was presented. Solutions with D = 12, 13 were
constructed for the first time in [20]. It was found that there is a maximum mass
as one moves along the family of non-uniform black strings. Solutions before the
maximum are unstable whereas solutions after the maximum appear to be stable,
and can have greater horizon area than a uniform string of the same mass.
Properties of these different solutions can be summarized in a plot of horizon area
against mass, for fixed L. This is useful in understanding possible time evolution of
instabilities of the various solutions. Horizon area increases, and energy decreases
(via emission of gravitational waves) so the final state of an instability must lie “up
and left” of the initial state on such a diagram.
Figure 1 shows the qualitative form of a plot of horizon area against mass for the
cases D = 5, 6 (see [6] for a more quantitative plot). Recall that there is a maximum
mass solution along the localized black hole branch of solutions. By the first law, an
extremum of the mass must also be an extremum of horizon area, and this results in
the cusp shown in the figure.
Fig. 2 Schematic plot of D−2

horizon area against mass A/L
for Kaluza-Klein black
holes/strings with D = 12, 13 non-uniform
based on results of Figueras black strings
et al. [20]. The localized black
hole curve is conjectural (Plot
reproduced from [20])
localized
black holes
merger
point GL point
uniform
black strings
D −3
M/L
The stability of localized black holes has not been studied. This could be done
using the method of Figueras et al. [20]. This method shows that an instability appears
as one moves through a maximum of mass along a branch of solutions. Since small
localized black holes are expected to be stable, it seems likely that, moving along the
branch of solutions, they will be stable until the maximum is reached and unstable
thereafter. Hence, in Fig. 1, unstable solutions lie on the part of the uniform string
curve extending from the origin to the GL point, along the non-uniform string curve
from the GL point to the merger point, and then along the localized black hole curve
from the merger point to the cusp.
For D = 12, 13, the change in behaviour of the non-uniform string branch implies
that the diagram must change to that of Fig. 2 where the cusp now appears along the
non-uniform black string branch. (Note that localized black hole solutions have not
been constructed for D > 6 so this part of the is conjectural.) In this case, unstable
solutions lie on the part of the uniform string curve extending from the origin to the
GL point and along the non-uniform string curve from the GL point to the cusp.
The results of Figueras et al. [20] suggest that there is no maximum mass non-
uniform black string for D = 11. Hence it seems likely that Fig. 1 gives the behaviour
for all D ≤ 11.
For D > 13, the results for non-uniform black strings, and the simplest guess for
the behaviour of localized black holes, results in the phase diagram shown in Fig. 3.
Unstable solutions lie on the part of the uniform string curve extending from the
origin to the GL point.
250 H. S. Reall
D−2
A/L
GL point
non-uniform
black strings
merger
localized
point
black holes
uniform
black strings
D−3
M/L
Fig. 3 Schematic plot of horizon area against mass for Kaluza-Klein black holes/strings with
D > 13 based on results of Figueras et al. [20]. The localized black hole curve is conjectural (Plot
reproduced from [20])
3 Asymptotically Flat Black Holes
3.1 Introduction
For a comprehensive 2008 review of higher-dimensional black holes, see [5]. For
more recent reviews, see [23, 24].
Recall that angular momentum is defined in terms of an antisymmetric matrix
Ji j where i, j run over the D − 1 spatial dimensions (e.g. for a particle, Ji j =
xi p j − x j pi ). For D = 4, Ji j is equivalent to a vector Ji and so, by choosing z-
axis aligned with this vector, one can write Ji j in terms of a single component J .
For D > 4 dimensions, the best one can achieve by a choice of axes is a block-
diagonal form for Ji j where each block is a 2 × 2 antisymmetric matrix specified by
a component J I , where I = 1, . . . , N = [(D − 1)/2].
There are two families of explicit black hole solutions of the vacuum Einstein
equation in D > 4 spacetime dimensions: Myers-Perry black holes [25] and black
rings [26, 27].
3.2 Myers-Perry Black Holes
The Myers-Perry solution is the generalization of the Kerr solution to D spacetime

dimensions. See [28] for a more detailed review. Myers-Perry black holes share many
properties with the Kerr solution. Cross-sections of the event horizon have spherical
topology S D−2 . The solution is uniquely parameterized by its mass M and its angular
(a) (b)
Fig. 4 Parameter space of Myers-Perry black holes for a D = 5 and b D = 6. The axes are
dimensionless angular momentum j I ∼ J I M −(D−2)/(D−3) . Non-extreme black holes correspond
to the shaded region. The boundary of this region corresponds to extreme black holes, except for
the vertices of the square, which describe singular solutions (Plot reproduced from [5])
momenta J I . The general Myers-Perry solution has N = [(D − 1)/2] commuting

rotational Killing vector fields ∂/∂φ I . The Myers-Perry solution with all J I = 0
reduces to the D-dimensional Schwarzschild solution.
Recall that, for given M, the Kerr solution has an upper bound on its angular
momentum |J | ≤ M 2 and saturating this bound gives the extreme Kerr solution
with a regular, but degenerate horizon. For D = 5, there is a similar upper bound
on the angular momenta of the MP solution: for given M, regular black holes have
J1 , J2 confined to a square region centred on the origin in the (J1 , J2 ) plane, see
Fig. 4. Saturating this bound gives a black hole with a degenerate horizon except
when one of the angular momenta vanishes (the vertices of the square), in which
case the spacetime is singular with no horizon.
For D > 5, there is a qualitative difference between MP and Kerr. It is possible
for the angular momenta to be arbitrarily large for fixed M. A “singly spinning”
black hole, i.e., one with J2 = J3 = · · · = J N = 0, has no upper bound on J1 .
More generally, it is possible for some of the angular momenta to be very large if
others are small. See Fig. 4 for the D = 6 case. Emparan and Myers [29] studied the
geometry of singly spinning MP black holes in the “ultraspinning” limit of very large
rotation. It was found that the black hole becomes flattened into the plane of rotation,
so that it resembles a rotating pancake. The geometry near the intersection of the
axis of rotation with the horizon approaches that of a black membrane: the product
of a (D − 2)-dimensional Schwarzschild solution with two flat directions. Black
membranes suffer from the Gregory-Laflamme instability. Hence it was conjectured
in [29] that rapidly rotating D > 5 MP black holes are classically unstable.
Confirmation of this conjecture required a study of linearized gravitational pertur-
bations of MP solutions. For D = 4, the study of gravitational perturbations of a Kerr
black hole is simplified by the remarkable “decoupling” phenomenon discovered by
Teukolsky [30], which reduces the problem to a PDE for a single scalar quantity.
Unfortunately, decoupling does not occur for D > 4 [31] and so one has to solve a
large set of coupled PDEs instead. This was done in [32], which studied a class of
linearized perturbations of a singly spinning MP solution, restricting to perturbations
252 H. S. Reall
that preserve the symmetries of the MP solution, i.e., stationarity and the rotational
symmetries. For fixed M, it was found (numerically) that there exists a critical value
of J1 for which a singly spinning D > 5 MP solution admits a non-trivial stationary
linearized gravitational perturbation. This was interpreted as the “threshold mode”
indicating the onset of the ultraspinning instability of Emparan and Myers [29], i.e.,
black holes with larger J1 should be unstable.
There is a more symmetrical class of MP solutions: those with odd D and J1 =
J2 = · · · = J N ≡ J . Such solutions are “cohomogeneity-1”: they depend non-
trivially only on the radial coordinate. This implies that the equations governing
gravitational perturbations are ODEs rather than PDEs [33]. However, in this case,
there is an upper (extremality) bound on J for given M, i.e., there is no reason to
expect an ultraspinning instability. The D = 5 case was studied in [34] and no
evidence of any instability was found. However, [35] showed that, for D = 9, with J
close to the upper bound, there are linearized gravitational perturbations which grow
exponentially with time, i.e., an instability. This was extended to D = 7 in [36],
which also considered a class of MP solutions interpolating between singly spinning
and cohomogeneity-1 and determined the threshold of instability in this case.
Although decoupling of perturbations does not occur for higher-dimensional black
holes, it does occur for the near-horizon geometry of an extreme vacuum black hole
[31]. Durkee and Reall [37] used this to argue that the instability of near-extreme
cohomogeneity-1 MP solutions can be predicted analytically, thereby extending the
result to any odd D > 5. Tanahashi and Murata [38] used the same approach
to show that MP solutions with even D and J1 = J2 = · · · = J N (which are
cohomogeneity-2) also are unstable near extremality.
The perturbations discussed so far are invariant under the Killing vector field
Ω I ∂/∂φ I where Ω I are the angular velocities of the horizon.1 This makes the result-
ing equations easier to solve. In the singly spinning case, this means that the per-
turbations are axisymmetric. Nonaxisymmetric perturbations of singly spinning MP
have been studied using full-blown numerical relativity [39, 40].
Shibata and Yoshino [39] studied the D = 5 case and found that, for J1 near to
the upper bound, an initially small non-axisymmetric perturbation grows in ampli-
tude. It was not possible to evolve the system long enough to determine the end-
point of this instability. The corresponding problem for D = 6, 7, 8 was studied
in [40]. An instability was found for large enough dimensionless angular momen-
tum j1 ≡ J1 M −(D−2)/(D−3) . This instability appears at a lower value of j1 than
the axisymmetric instability discussed above, i.e., the nonaxisymmetric instability
is the first one to appear as the angular momentum is increased. In this case, it was
possible to follow the long time evolution of the instability. It was found that the
perturbed black hole emits gravitional waves, which carry away angular momen-
tum (and energy), and the black hole finally settles down to a (presumably stable)
Myers-Perry black hole with a lower value of j1 . It is not clear whether this also hap-
1 The exception is the analysis in [34] of the cohomogeneity-1 D = 5 case, i.e., J1 = J2 . The
results are consistent with stability of this solution.
pens for D = 5 or whether the evolution of the instability is qualitatively different

in this case.
Finally we should note that there is an instability that afflicts extreme black holes,
including extreme Myers-Perry. Consider a massless scalar field in the background
of a Kerr black hole. In the non-extreme case, it has been proved that, for any
initial data (decaying at infinity), the scalar field and all its derivatives decay on, and
outside, the horizon [41]. However, the extreme case is qualitatively different. For
axisymmetric initial data, the scalar field decays on, and outside, the event horizon
[42]. However, a transverse derivative of the scalar field at the horizon generically
does not decay (hence the energy-momentum tensor does not decay), and higher
transverse derivatives at the horizon blow up, i.e., they become large at late time
[43]. Therefore the scalar field is stable in the background of an arbitrarily non-
extreme black hole, but unstable in the background of an exactly extreme black hole.
Note that the instability involves power-law, rather than exponential, growth in time.
The scalar field here should be regarded as a toy model for linearized gravita-
tional perturbations, which suggests that an extreme Kerr black hole should suffer a
gravitational instability. Lucietti and Reall [44] showed that this is indeed the case.
This reference also showed that similar non-decay and blow-up results hold for a
massless scalar field in any extreme black hole spacetime. Hence extreme black holes
generically are unstable, as conjectured in [45].
3.3 Black Rings
A black ring is an asymptotically flat black hole for which horizon cross-sections have
topology S 1 × S D−3 . There is a heuristic argument for the existence of such objects in
vacuum gravity. Take a finite segment of (uniform) black string and imagine forming
it into a loop. The loop would collapse under its own gravity and tension. However, if
it rotates then Newtonian arguments suggest that the resulting centrifugal repulsion
can balance the gravitational and tension forces, resulting in a stationary black ring.
This heuristic argument for the existence of black rings is confirmed by explicit
solutions, which are known only for the special case D = 5 [26, 27]. These solutions
are the first examples of asymptotically flat black holes with horizon cross-sections
of non-spherical topology. They form a 3-parameter family and have 2 commuting
rotational symmetries. Unlike the Myers-Perry solutions, black rings are not uniquely
labelled by M and J I : it is possible for there to be two black rings with the same
values for these quantities. Furthermore, a black ring can have the same value for
M, J I as a Myers-Perry black hole. See Fig. 5. Hence the existence of black rings
shows that black hole uniqueness cannot be straightforwardly extended to higher
dimensions.
An important difference between black rings and D = 5 MP solutions is that, for
given M, there is a lower bound on the angular momentum J1 , and (for J2 = 0) no
upper bound. Black rings do not admit a regular static limit, as expected from the
heuristic argument for their existence.
254 H. S. Reall
Fig. 5 Phase space of D = 5

Myers-Perry black holes and
black rings. See Fig. 4 for
notation. For each point of the
light grey regions there exists
a “thin” black ring. For each
point of the mid-grey region
there exists a MP black hole.
For each point of the dark grey
region there exists a MP black
hole, a fat black ring and a thin
black ring (Plot reproduced
from [5])
It is convenient to divide black rings into two subclasses according to the sign
of the heat capacity at constant angular momenta c J . Black rings with c J < 0 are
called “thin”; and those with c J > 0 are called “fat”. The terminology arises from
the geometry of the horizon. Thin rings look more like hula-hoops and fat rings more
like bagels. Rings of each type are uniquely labelled by M, J I .
Heuristic arguments indicate that fat black rings probably are unstable. Arcioni
and Lozano-Tellechea [46] used the Poincaré turning point method to argue that fat
black rings with small c J should have one more “unstable mode” than thin rings.
Elvang et al. [47] considered certain singular deformations of the black ring solution
to determine an effective potential for radial deformations of the ring. It was found
that thin rings sit at a local minimum of the potential but fat rings correspond to a
local maximum, suggesting instability.
These results were confirmed by the analysis of Figueras et al. [48], which intro-
duced a new method for studying black hole stability. If a black hole is stable then
any small perturbation of it must eventually settle down to a black hole belonging
to the same family, with a small change in its parameters. If one restricts to rota-
tionally symmetric perturbations, so that angular momentum is conserved, then, by
using increase of horizon area and decrease of (Bondi) energy, one can deduce that
the initial data describing the perturbed black hole must satisfy a certain inequality
relating its mass, angular momenta and horizon area. If one can find an initial per-
turbation that violates this inequality then the black hole cannot be stable. The nice
thing about this method is that it requires only the construction of initial data, rather
than determining the full time-evolution of the perturbation. Using this approach, it
was shown that fat black rings suffer from a rotationally symmetric instability.
Rings with large J1 (for given M) are very thin. As J1 → ∞, the geometry near a
section of the ring approaches that of a boosted uniform black string. Since the latter
suffers from the Gregory-Laflamme instability, it seems very likely that black rings
with large J1 will be classically unstable. This instability breaks rotational symmetry
so it cannot be studied using the approach of Figueras et al. [48]. Demonstrating the
existence of this instability will require a study of linearized gravitational perturba-
tions of black rings (or full-blown numerical GR). This has not yet been attempted.
In summary, fat black rings are known to be unstable and very thin black rings
are believed to be unstable. But it is not known whether all black rings are unstable
or whether some thin, but not too thin, rings are stable.
So far we have been discussing D = 5 black rings. Explicit black ring solutions
are not known for D > 5. However, approximate solutions describing very thin
black rings have been constructed using the perturbative “blackfold” approach to be
described below [49]. Recently, [50] reported a breakthrough in determining D > 5
dimensional black ring solutions by numerical solution of the Einstein equation.
Results were presented for black rings with D = 6, 7 with a single non-vanishing
angular momentum J1 . Their properties appear similar to those of D = 5 black rings,
and agree with the predictions of the blackfold approach when the radius becomes
large.
3.4 Black Saturn and Generalizations
Given the existence of black rings and Myers-Perry black holes, it is natural to ask
whether one can “superpose” these solutions to construct a “Black Saturn” describing
a MP black hole with a concentric black ring. Of course the Einstein equation is
nonlinear so this is highly non-trivial. Nevertheless, solution generating techniques
have been used to construct such a solution with D = 5 [51].
Black Saturn is the first example of an explicit stationary, asymptotically flat,
vacuum, regular, multi-black hole solution. It provides an interesting demonstration
of the frame-dragging effect: if one sets the (Komar) angular momentum of the MP
black hole to zero then its angular velocity is non-zero because the horizon generators
are dragged around by the rotation of the black ring.
Solution generating techniques have also been used to construct solutions with
multiple concentric black rings. For example the di-ring of Iguchi and Mishima [52],
Evslin and Krishnan [53] describes a pair of concentric black rings lying in the same
plane. Elvang and Rodriguez [54] gave a solution describing a pair of concentric
black rings lying in orthogonal planes.
3.5 Classification of Asymptotically Flat Black Holes
For D = 4, the black hole uniqueness theorem provides a complete classification

of stationary vacuum black holes. For D > 4, the known solutions show that the
situation is much more complicated and the uniqueness theorem does not generalize
in a simple way. However, it is useful to explore whether some aspects of the D = 4
theorem can be generalized to D > 4.
256 H. S. Reall
Uniqueness of static vacuum black holes turns out to generalize straightforwardly

to D > 4: the only asymptotically flat static vacuum black holes solution is the
Schwarzschild solution, for any D ≥ 4 [55].
For non-static black holes, the first logical step in the D = 4 uniqueness theorem
is Hawking’s topology theorem [56], stating that horizon cross-sections must have
S 2 topology. This result has been generalized to D > 4 dimensions, with the result
that horizon cross-sections must admit a metric of positive scalar curvature [57].
This is a topological restriction on the horizon. For D = 4 it reduces to Hawking’s
result. For D = 5 it implies that the horizon cross-section must be either S 3 (or a
quotient), S 1 × S 2 , or a connected sum of these. It is striking that S 3 and S 1 × S 2 are
precisely the topologies realized by the known Myers-Perry and black ring solutions.
However, the possibility of taking quotients and connected sums implies that there
are infinitely many topologies consistent with this theorem. For D > 6, there is no
simple description of the possible topologies.
The next step in the D = 4 uniqueness proof for non-static black holes is
Hawking’s rigidity theorem [56], the statement that a stationary, rotating, analytic,
black hole solution must be axisymmetric. This has been generalised to D > 4: a
stationary, rotating, analytic, black hole solution must admit a rotational symmetry
(a U (1) isometry acting as a rotation at infinity) [58, 59]. However, this theorem
guarantees only one rotational symmetry whereas the known explicit solutions have
N > 1 commuting rotational symmetries, i.e., more symmetry than guaranteed by
the theorem. This suggests that there may exist other D > 4 black hole solutions
which have less symmetry than the known solutions. We will discuss this further in
the next section.
Progress with classifying black holes can be made if one restricts attention to
the case of multiple rotational symmetries. The classification of D = 5 stationary
rotating vacuum black holes with two rotational symmetries was studied in [60]
(extending [61]). It was shown that two such solutions are isometric if, and only if,
they have the same mass, angular momenta, and rod structure. The latter (introduced
in [62, 63]) encodes the angular velocity of the horizon and the nature of “axes of
rotation” in the spacetime. The assumed symmetries imply that the horizon topology
must be S 3 , S 1 × S 2 or a lens space (a quotient of S 3 ). It is not known whether
(asymptotically flat) solutions of the latter kind exist.
3.6 Perturbative Solutions
Given the difficulty in finding explicit solutions of the Einstein equation, perturbative
techniques have been used to obtain some insight into what other solutions might
exist. Of particular interest are the possible topologies of higher-dimensional black
holes, and the question of whether there exist higher-dimensional black holes with
just one rotational symmetry. Two techniques have been used to investigate these
questions.
The “blackfold” technique is a method for constructing black hole solutions whose
horizons exhibit a large hierarchy of length scales. An example is a black ring for
which the S 1 radius is much greater than the S D−3 radius [49]. Such large-radius
black ring solutions are constructed perturbatively as an expansion in the ratio of these
radii. More generally, the blackfold method has been used to construct approximate
higher-dimensional solutions with topologies of the form S p1 × S p2 × · · ·× S p K × s q
where p1 , . . . p K are odd, q ≥ 2, S p1 , . . . S pk are large radius spheres and s q is a
small radius sphere [64].
The blackfold approach has also been used to construct solutions with just one
rotational symmetry. Emparan et al. [65] presented a perturbative solution describing
a D = 5 “helical” black ring. This solutions can be visualised by imagining a spring
formed into a loop.
Another perturbative approach is to study linearized perturbations of an explicit
solution. Non-uniform black strings provide a nice example of this. A uniform
black string admits a time-independent linearized perturbation corresponding to the
“threshold mode” of the Gregory-Laflamme instability. This time-independent per-
turbation exists precisely at the point where the non-uniform black string family
bifurcates from the uniform string family. It corresponds to a non-uniform string
solution with infinitesimal non-uniformity. Thus by studying linearized perturba-
tions of the uniform string one can infer the existence of the non-uniform string
family.
This method has been applied to perturbations of Myers-Perry black holes. Con-
sider a singly spinning MP black hole with D > 5. As explained above, [32] showed
that there is a critical J1 (for given M) at which such a solution admits a non-trivial
time-independent linearized perturbation, corresponding to the threshold mode of
the ultraspinning instability. Earlier, [29] had suggested that such a perturbation
should be interpreted as evidence for a new family of black holes that bifurcates
from the MP family. This new family would possess the same symmetries as the MP
solution but with a slightly deformed horizon, corresponding to a depression at the
poles of the sphere. It was suggested that moving along this new branch of solutions,
the depression would increase, corresponding to an increasingly “pinched” sphere.
Eventually, the sphere is expected to “pinch off” completely, and merge with the
black ring family [49].
The same strategy has been applied to perturbations of cohomogeneity-1 MP
black holes (odd D with J1 = J2 = · · · = J N ). Recall that [35, 36] demonstrated
an instability of such solutions near extremality. Again there is a time-independent
threshold mode for this instability and so this was interpreted as evidence for a
new branch of solutions bifurcating from the MP family. However, in this case, the
threshold mode does not preserve the symmetries of the background geometry. In
general it preserves only the rotational symmetry whose existence is guaranteed by
the rigidity theorem. Hence the new family of solutions should have just one rotational
symmetry. Note that these black holes would have spherical topology. Furthermore,
in this case, there is not just one threshold mode but a multi-parameter set of them. If
each of these extends to the new branch of solutions then these new solutions would
258 H. S. Reall
have many more parameters than the MP family. For example, in D = 9 they would
have 70 parameters [35], many more than the 5 parameters of the MP solution.
Acknowledgments I would like to thank Pau Figueras and Keiju Murata for permission to repro-
duce figures from [20] and Roberto Emparan for permission to reproduce figures from [5]. I am
supported by a Royal Society University Research Fellowship and by European Research Council
grant no. ERC-2011-StG 279363-HiDGR.
References
1. Strominger, A., Vafa, C.: Microscopic origin of the bekenstein-hawking entropy. Phys. Lett.
B379, 99 (1996). doi:10.1016/0370-2693(96)00345-0
2. Maldacena, J.M.: The large n limit of superconformal field theories and supergravity. Adv.
Theor. Math. Phys. 2, 231 (1998)
3. Kanti, P.: Black holes at the LHC. In: Papantonopoulos, E. (eds.) Physics of Black Holes: A
Guided Tour, Lecture notes in physics, vol. 769, pp. 387–423. Springer, Berlin (2009). doi:10.
1007/978-3-540-88460-6_10
4. Cvetic, M., Lü, H., Page, D.N., Pope, C.N.: New Einstein-Sasaki and Einstein spaces from
Kerr-de Sitter. J. High Energy Phys. 2009(07), 082 (2009). doi:10.1088/1126-6708/2009/07/
082
5. Emparan, R., Reall, H.S.: Black holes in higher dimensions. Living Rev. Relativ. 11(6), lrr-
2008-6 (2008). http://www.livingreviews.org/lrr-2008-6
6. Horowitz, G.T., Wiseman, T.: General black holes in Kaluza-Klein theory. In: Horowitz, G.T.
(ed.) Black Holes in Higher Dimensions, pp. 69–98. Cambridge University Press, Cambridge
(2012)
7. Harmark, T.: Small black holes on cylinders. Phys. Rev. D 69, 104015 (2004). doi:10.1103/
PhysRevD.69.104015
8. Gorbonos, D., Kol, B.: A dialogue of multipoles: matched asymptotic expansion for caged
black holes. J. High Energy Phys. 2004(06), 053 (2004). doi:10.1088/1126-6708/2004/06/053
9. Sorkin, E., Kol, B., Piran, T.: Caged black holes: black holes in compactified spacetimes. ii.
5d numerical implementation. Phys. Rev. D 69, 064032 (2004). doi:10.1103/PhysRevD.69.
064032
10. Kudoh, H., Wiseman, T.: Connecting black holes and black strings. Phys. Rev. Lett. 94, 161102
(2005). doi:10.1103/PhysRevLett.94.161102
11. Headrick, M., Kitchen, S., Wiseman, T.: A new approach to static numerical relativity, and
its application to kaluza-klein black holes. Class. Quantum Grav. 27, 035002 (2010). doi:10.
1088/0264-9381/27/3/035002
12. Ishibashi, A., Kodama, H.: Stability of higher dimensional Schwarzschild black holes. Prog.
Theor. Phys. 110, 901 (2003). doi:10.1143/PTP.110.901
13. Gregory, R., Laflamme, R.: Black strings and p-branes are unstable. Phys. Rev. Lett. 70, 2837
(1993). doi:10.1103/PhysRevLett.70.2837
14. Lehner, L., Pretorius, F.: Black strings, low viscosity fluids, and violation of cosmic censorship.
15. Horowitz, G.T., Maeda, K.: Fate of the black string instability. Phys. Rev. Lett. 87, 131301
(2001). doi:10.1103/PhysRevLett.87.131301
16. Gubser, S.S.: On non-uniform black branes. Class. Quantum Grav. 19, 4825 (2002). doi:10.
1088/0264-9381/19/19/303
17. Wiseman, T.: Static axisymmetric vacuum solutions and nonuniform black strings. Class. Quan-
tum Grav. 20, 1137 (2003). doi:10.1088/0264-9381/20/6/308
18. Kleihaus, B., Kunz, J., Radu, E.: New nonuniform black string solutions. J. High Energy Phys.
2006(06), 016 (2006). doi:10.1088/1126-6708/2006/06/016
19. Sorkin, E.: Non-uniform black strings in various dimensions. Phys. Rev. D 74, 104027 (2006).
doi:10.1103/PhysRevD.74.104027
20. Figueras, P., Murata, K., Reall, H.S.: Stable non-uniform black strings below the critical dimen-
sion. J. High Energy Phys. 2012(11), 071 (2012). doi:10.1007/JHEP11(2012)071
21. Kol, B.: Topology change in general relativity, and the black hole black string transition. J.
High Energy Phys. 2005(10), 049 (2005). doi:10.1088/1126-6708/2005/10/049
22. Sorkin, E.: Critical dimension in the black-string phase transition. Phys. Rev. Lett. 93, 031601
(2004). doi:10.1103/PhysRevLett.93.031601
23. Horowitz, G.T. (ed.): Black Holes in Higher Dimensions. Cambridge University Press, Cam-
bridge (2012)
24. Maeda, K., Shiromizu, T., Tanaka T. (eds.): Higher Dimesional Black Holes, Progress of
Theoretical Physics Supplement, vol. 189, Oxford University Press, Oxford (2011). http://
ptps.oxfordjournals.org/content/189.toc
25. Myers, R.C., Perry, M.J.: Black holes in higher dimensional space-times. Ann. Phys. (N.Y.)
172, 304 (1986). doi:10.1016/0003-4916(86)90186-7
26. Emparan, R., Reall, H.S.: A rotating black ring solution in five dimensions. Phys. Rev. Lett.
88, 101101 (2002). doi:10.1103/PhysRevLett.88.101101
27. Pomeransky, A.A., Sen’kov, R.A.: Black ring with two angular momenta. ArXiv e-prints
arXiv:0612005 [hep-th] (2006)
28. Myers, R.C.: Myers-Perry black holes. In: Horowitz, G.T. (ed.) Black Holes in Higher Dimen-
sions, pp. 101–133. Cambridge University Press, Cambridge (2012)
29. Emparan, R., Myers, R.C.: Instability of ultra-spinning black holes. J. High Energy Phys.
2003(09), 025 (2003). doi:10.1088/1126-6708/2003/09/025
30. Teukolsky, S.A.: Rotating black holes: separable wave equations for gravitational and electro-
magnetic perturbations. Phys. Rev. Lett. 29, 1114 (1972). doi:10.1103/PhysRevLett.29.1114
31. Durkee, M., Reall, H.S.: Perturbations of higher-dimensional spacetimes. Class. Quantum Grav.
28, 035011 (2011). doi:10.1088/0264-9381/28/3/035011
32. Dias, O.J.C., Figueras, P., Monteiro, R., Santos, J.E., Emparan, R.: Instability and new phases
of higher-dimensional rotating black holes. Phys. Rev. D 80, 111701(R) (2009). doi:10.1103/
PhysRevD.80.111701
33. Kunduri, H.K., Lucietti, J., Reall, H.S.: Gravitational perturbations of higher dimensional rotat-
ing black holes: tensor perturbations. Phys. Rev. D 74, 084021 (2006). doi:10.1103/PhysRevD.
74.084021
34. Murata, K., Soda, J.: Stability of five-dimensional Myers-Perry black holes with equal angular
momenta. Prog. Theor. Phys. 120, 561 (2008). doi:10.1143/PTP.120.561
35. Dias, Ó.J.C., Figueras, P., Monteiro, R., Reall, H.S., Santos, J.E.: An instability of higher-
dimensional rotating black holes. J. High Energy Phys. 2010(05), 076 (2010). doi:10.1007/
JHEP05(2010)076
36. Dias, Ó.J.C., Monteiro, R., Santos, J.E.: Ultraspinning instability: the missing link. J. High
Energy Phys. 2011(08), 139 (2011). doi:10.1007/JHEP08(2011)139
37. Durkee, M., Reall, H.S.: Perturbations of near-horizon geometries and instabilities of Myers-
Perry black holes. Phys. Rev. D 83, 104044 (2011). doi:10.1103/PhysRevD.83.104044
38. Tanahashi, N., Murata, K.: Instability in near-horizon geometries of even-dimensional Myers-
Perry black holes. Class. Quantum Grav. 29, 235002 (2012). doi:10.1088/0264-9381/29/23/
235002
39. Shibata, M., Yoshino, H.: Nonaxisymmetric instability of rapidly rotating black hole in five
dimensions. Phys. Rev. D 81, 021501 (2010). doi:10.1103/PhysRevD.81.021501
40. Shibata, M., Yoshino, H.: Bar-mode instability of rapidly spinning black hole in higher dimen-
sions: numerical simulation in general relativity. Phys. Rev. D 81, 104035 (2010). doi:10.1103/
PhysRevD.81.104035
41. Dafermos, M., Rodnianski, I.: The black hole stability problem for linear scalar perturbations.
In: Damour, T., Jantzen, R.T., Ruffini R. (eds.) The Twelfth Marcel Grossmann Meeting On
Recent Developments in Theoretical and Experimental General Relativity, Astrophysics and
Relativistic Field Theories, pp. 132–189. World Scientific, Singapore (2012). doi:10.1142/
9789814374552_0008
260 H. S. Reall
42. Aretakis, S.: Decay of axisymmetric solutions of the wave equation on extreme Kerr back-
grounds. ArXiv e-prints arxiv:1110.2006 [gr-qc] (2011)
43. Aretakis, S.: Horizon instability of extremal black holes. ArXiv e-prints arxiv:1206.6598 [gr-
qc] (2012)
44. Lucietti, J., Reall, H.S.: Gravitational instability of an extreme Kerr black hole. Phys. Rev. D
86, 104030 (2012). doi:10.1103/PhysRevD.86.104030
45. Marolf, D.: The dangers of extremes. Gen. Relativ. Gravit. 42, 2337 (2010). doi:10.1007/
s10714-010-1027-z
46. Arcioni, G., Lozano-Tellechea, E.: Stability and critical phenomena of black holes and black
rings. Phys. Rev. D 72, 104021 (2005). doi:10.1103/PhysRevD.72.104021
47. Elvang, H., Emparan, R., Virmani, A.: Dynamics and stability of black rings. J. High Energy
Phys. 2006(12), 074 (2006). doi:10.1088/1126-6708/2006/12/074
48. Figueras, P., Murata, K., Reall, H.S.: Black hole instabilities and local penrose inequalities.
49. Emparan, R., Harmark, T., Niarchos, V., Obers, N.A., Rodriguez, M.J.: The phase structure of
higher-dimensional black rings and black holes. J. High Energy Phys. 2007(10), 110 (2007).
doi:10.1088/1126-6708/2007/10/110
50. Kleihaus, B., Kunz, J., Radu, E.: Black rings in six dimensions. Phys. Lett. B 718, 1073 (2013).
doi:10.1016/j.physletb.2012.11.015
51. Elvang, H., Figueras, P.: Black saturn. J. High Energy Phys. 2007(05), 050 (2007). doi:10.
1088/1126-6708/2007/05/050
52. Iguchi, H., Mishima, T.: Black di-ring and infinite nonuniqueness. Phys. Rev. D 75(064018),
2008 (2007). doi:10.1103/PhysRevD.75.064018 [Erratum: ibid. 78, 069903]
53. Evslin, J., Krishnan, C.: The black di-ring: an inverse scattering construction. Class. Quant.
Grav. 26, 125018 (2009). doi:10.1088/0264-9381/26/12/125018
54. Elvang, H., Rodriguez, M.J.: Bicycling black rings. J. High Energy Phys. 2008(04), 045 (2008).
doi:10.1088/1126-6708/2008/04/045
55. Gibbons, G.W., Ida, D., Shiromizu, T.: Uniqueness and nonuniqueness of static black holes in
higher dimensions. Phys. Rev. Lett. 89, 041101 (2002). doi:10.1103/PhysRevLett.89.041101
56. Hawking, S.W., Ellis, G.F.R.: The Large Scale Structure of Space-Time. Cambridge Mono-
graphs on Mathematical Physics. Cambridge University Press, Cambridge (1973)
57. Galloway, G.J., Schoen, R.: A generalization of hawking’s black hole topology theorem to
higher dimensions. Commun. Math. Phys. 266, 571 (2006). doi:10.1007/s00220-006-0019-z
58. Hollands, S., Ishibashi, A., Wald, R.M.: A higher dimensional stationary rotating black
hole must be axisymmetric. Commun. Math. Phys. 271, 699 (2007). doi:10.1007/s00220-
007-0216-4
59. Moncrief, V., Isenberg, J.: Symmetries of higher dimensional black holes. Class. Quantum
Grav. 25, 195015 (2008). doi:10.1088/0264-9381/25/19/195015
60. Hollands, S., Yazadjiev, S.: Uniqueness theorem for 5-dimensional black holes with two axial
killing fields. Commun. Math. Phys. 283, 749 (2008). doi:10.1007/s00220-008-0516-3
61. Morisawa, Y., Ida, D.: A boundary value problem for the five-dimensional stationary rotating
black holes. Phys. Rev. D 69, 124005 (2004). doi:10.1103/PhysRevD.69.124005
62. Emparan, R., Reall, H.S.: Generalized Weyl solutions. Phys. Rev. D 65, 084025 (2002). doi:10.
1103/PhysRevD.65.084025
63. Harmark, T.: Stationary and axisymmetric solutions of higher-dimensional general relativity.
64. Emparan, R., Harmark, T., Niarchos, V., Obers, N.A.: World-volume effective theory for higher-
dimensional black holes. Phys. Rev. Lett. 102, 191301 (2009). doi:10.1103/PhysRevLett.102.
191301
65. Emparan, R., Harmark, T., Niarchos, V., Obers, N.A.: New horizons for black holes and branes.
J. High Energy Phys. 2010(04), 046 (2010). doi:10.1007/JHEP04(2010)046
Black Holes, Hidden Symmetry and Complete
Integrability: Brief Review
Valeri P. Frolov
Abstract This chapter contains a brief review of the remarkable properties of higher
dimensional rotating black holes with the spherical topology of the horizon. We
demonstrate that these properties are connected with and generated by a special geo-
metrical object, the Principal Conformal Killing-Yano tensor (PCKYT). The most
general solution, describing such black holes, Kerr-NUT-ADS metric, admits this
structure. Moreover a solution of the Einstein Equations with (or without) a cos-
mological constant which possesses PCKYT is the Kerr-NUT-ADS metric. This
object (PCKYT) is responsible for such remarkable properties of higher dimen-
sional rotating black holes as: (i) complete integrability of geodesic equations and
(ii) complete separation of variables of the important field equations.
1 Introduction
Main motivations for study higher dimensional black holes are nicely summarized
in the chapter by Harvey Reall. String theory, brane world models and ADS/CFT
correspondence naturally involve higher dimensional gravity, and black holes play
role of natural probes of extra dimensions. Possibility of creation of microscopic
black holes in high energy colliders was (and still is) a subject which attracts a lot
of intension. Besides these ‘rather technical’ reasons there exists another one of
more general nature. The driving force of our scientific knowledge is quite often our
curiosity. We keep asking ourselves questions: “What happens if ...”. From time to
time by answering such questions one discovers interesting and non-trivial results.
V. P. Frolov (B)
Theoretical Physics Institute, University of Alberta, Edmonton, AB T6G 2E1, Canada
e-mail: vfrolov@ualberta.ca
262 V. P. Frolov
Fig. 1 Stationary 5D black holes
By answering the question “Do black hole exist in higher dimensional gravity
and what are their properties?” one understands better which of the properties of
four-dimensional black holes are generic (valid in any number of dimensions) and
which are specific in only our four-dimensional world. One of the most surprising
discoveries was that stationary black holes in higher dimensions can have large
variety of the horizon topology. The first of such 5-dimensional black objects with
toroidal topology of the horizon S 1 ×S 2 , called black ring, was found by Emparan and
Reall [1]. By now there are known many other stationary exact solutions describing
vacuum black objects in 5D, the horizon of which is schematically shown at Fig. 1
(For general discussion see the chapter by Reall in this volume and a nice review
[2]).
Let us emphasize, that according to the general definition all such black objects,
strictly speaking, are black holes, since the region beyond their event horizon is
causally separated from the future null infinity. However, in order to distinguish black
objects with different horizon topology they received special nicknames (black ring,
black Saturn, etc), while the name of a black hole is reserved for the black object
with spherical topology of the horizon. In this talk we shall follow this tradition.
This means that from now on any time when we speak about a black hole we have in
mind the object presented at the Fig. 1a. To be more concrete we focus our attention
on isolated higher dimensional rotating black holes with spherical topology of the
horizon, obeying equations
Rμν = Λgμν . (1)
The corresponding spacetime is asymptotically either flat (Λ = 0) or (anti) DeSitter

one.
Black Holes, Hidden Symmetry and Complete Integrability 263
2 Four Dimensional Kerr-NUT-ADS Metric and Its Higher

Dimensional Generalization
2.1 Four Dimensional Kerr-NUT-ADS Spacetime
The simplest example of a vacuum static black hole in 4D is the Schwarzschild metric
ds 2 = − f dt 2 + dr 2 / f + r 2 dΩ22 , f = 1 − r0 /r. (2)
It contains only one parameter, the mass M of the black hole, which determines the
gravitational radius (size of the horizon) r0 = 2M. Here and later we denote by dΩm2
the metric on a unit m-dimensional round sphere.
A vacuum stationary black hole is characterized by one more parameter, J . Such
a black hole is rotating and J is the value of its angular momentum. In fact the angular
momentum (measured at infinity) is described by 3 × 3 antisymmetric matrix Ji j .
By rigid 3-dimensional rotations this matrix can be put in a standard form
 
0J 0
J =  −J 0 0  . (3)
00 0
Kerr metric (with J ≈ M 2 ) is the most general vacuum stationary solution describing
a regular black hole in an asymptotically flat spacetime.
In 1963 Newman, Tamburino and Unti [3] found another generalization of the
Schwarzschild solution which besides mass M contained another parameter N . This
parameter N , called NUT-parameter, describes “gravitomagnetic monopole” [4]. The
corresponding NUT solution contains an analogue of the Dirac strings for a mag-
netic monopole, but in the gravitational case it cannot be excluded and does affect
spacetime. The most general stationary black hole solution of the Einstein equa-
tion (1) in an asymptotically (anti)deSitter spacetime contains all these parameters
(M, J, N , Λ). It was obtained in [5, 6] and it is called Kerr-NUT-ADS metric.
2.2 Higher Dimensional Solutions
Let us discuss now higher dimensional black hole solutions of the Einstein equation
(1). We denote by D the total number of (space and time) dimensions. Since some
of relations have slightly different form in odd and even dimensions, we write
D = 2n + ε. (4)
264 V. P. Frolov
It is quite easy to generalize the Schwarzschild solution to any number of spacetime

2
dimensions. It is sufficient to substitute in (2) dΩ D−2 by the line element on (D−2)−
dimensional sphere S D−2 , and put f = 1 − (r0 /r ) D−3 . This metric is known as
Tangherlini solution [7]. The cosmological constant can also be easily included by
using f in the form
Λ 2
f = 1 − (r0 /r ) D−3 − r . (5)
D−1
Before discussing higher dimensional rotating black holes let us make the following
remark. In the asymptotically flat spacetime the total angular momentum of the
objects, as measured at infinity, it described by an antisymmetric tensor Ji j , where
i and j are spatial indices. By suitable rigid rotations of the spatial coordinates, this
(D − 1) × (D − 1)-matrix can be transformed into the following canonical form:
 
0 J1 0 0 ...
⎜ −J1 0 0 0 ...⎟
⎜ ⎟
J=⎜
⎜ 0 0 0 J2 ...⎟⎟. (6)
 0 0 −J2 0 ...
... ... ... ... ...
It is easy to see that the total number of independent 2 × 2 blocks is equal to n −

1 + ε. This means that there exist n − 1 + ε independent components of the angular
momentum Ji , associated with n − 1 + ε asymptotic independent two-dimensional
spatial planes of rotation. The most general solution for a vacuum stationary rotating
black hole was found in 1986 [8]. It contains n + ε independent constants: massM
and components of the angular momentum Ji .
It took 20 years to discover the most general higher dimensional black hole solu-
tion, which besides mass and angular momentum contains also the cosmological
constant and generalization of the NUT parameters [9]. This solution is called a
‘general Kerr-NUT-ADS metric’. It is a natural generalization of the 4D Kerr-NUT-
ADS metric. It contains D − ε arbitrary parameters, which include mass, (n − 1 + ε)
rotation parameters, cosmological constant and (n − 1 − ε) NUT parameters.
3 Principal Conformal Killing-Yano Tensor
3.1 Definition
The remarkable fact is: Properties of higher dimensional rotating black holes and
their four dimensional ‘cousins’ are very similar. There exist a very deep geomet-
rical reason for this similarity. All these metrics admit a special geometric object,
the Principal Conformal Killing-Yano tensor (PCKYT), which is a generator of a
complete set of explicit and hidden symmetries, that uniquely specifies the solution
up to some constants. This solution coincides with Kerr-NUT-ADS metric.
In fact this object is special case of a closed conformal Killing-Yano tensor. We
shall give a general definition of Killing-Yano tensors later. We shall explain also
in more detail what are the hidden symmetries and how are they generated. At the
moment, we focus on some remarkable properties of PCKYT.
Let us first give a definition of the PCKYT. Consider a rank 2 antisymmetric
tensor (2-form) h which obeys the equation
⊗c h ab = gca ξb − gcb ξa . (7)
By contracting this equation one finds
1
ξa = ⊗ b h ba . (8)
D−1
If one antisymmetrizes the indices a, b, c in (7), the right-hand side of this equation
vanishes. This means that h is closed form, and (at least locally) can be presented in
the form
h = d b, (9)
where b is a potential one-form.

Another important relation can be obtained if one takes the covariant derivative of
(8) and symmetrises it. Using an expression for commutator of covariant derivatives,
and the Einstein equation (1) one proves that
ξ(a;b) = 0. (10)
In other words a spacetime admitting the 2-form h which is a solution of (7) always
has a Killing vector. We call it a primary Killing vector.
We call h obeying (7) the Principal Conformal Killing-Yano tensor (PCKYT) if
it is non-degenerate. More concretely this means the following:
• The matrix rank of (D × D)-antisymmetric matrix h ab is the largest possible, that
is equal to 2n.
• Consider eigenvalue problem for a matrix H ab = h ac h cb
H ab e(i)
b
= xi e(i)
a
. (11)
It is easy to see that h ab e(i)

b is again an eigenvector with the same eigenvalue x .
(i)
We assume that H has largest possible number, n, of different eigenvalues, and
hence n linearly independent eigen 2-planes.
266 V. P. Frolov
3.2 Remarkable Properties of PCKYT
The most general higher dimensional black hole metric, Kerr-NUT-ADS solution,
admits the Principal Conformal Killing-Yano tensor. This result was first proved in
[10] for Myers-Perry metrics, and later in [11] it was proved for the general Kerr-
NUT-ADS spacetimes.
Moreover, a solution of the Einstein equation (1) in any number of dimensions
(D ◦ 4), which admits the Principal Conformal Killing-Yano tensor, is the Kerr-
NUT-ADS spacetime. This result under special assumptions was first proved in [12].
A general proof was given in [13, 14].
In other words, the existence of PCKYT is a characteristic property of higher
dimensional rotating black holes, with the spherical topology of horizon. Namely
the existence of this object explains many remarkable properties of these spaces
which sometimes are called ‘miraculous’. Before discussing this subject let us briefly
remind some properties of completely integrable dynamical systems.
4 Complete Integrability
4.1 Liouville Theorem
Particle and light motion in a curved spacetime is described by geodesic equations.

These equations are of the second order. Let x a (τ ) be a trajectory. By introducing
a momentum pa = gab ẋ b as an independent variable, it is possible to rewrite the
geodesic equations in the first order form. These equations has the Hamiltonian
form. This means that the general theory of dynamical systems can be applied for
this problem. This approach is well known and its tools are very useful. Let us
demonstrate this for the special problem: motion of a particle in a spacetime of a
higher dimensional rotating black hole.
Consider a dynamical system. It is described by an even dimensional phase space
M 2m with a closed non-degenerate symplectic 2-form Ω and a Hamiltonian H , which
is a scalar function on the phase space. If z A are coordinates, then the dynamical
equation is
ż A = Ω AB H,B . (12)
Poisson bracket for two functions F and G on the phase space is defined as
{F, G} = −Ω AB F,A G ,B . (13)
These two functions are called to be in involution if their Poisson bracket vanishes.
A scalar function F(z A ) on the phase space is a first integral of motion if its Poisson
bracket with the Hamiltonian vanishes {F, H } = 0 .
Liouville proved the following theorem: If a system with a Hamiltonian H in

the 2m dimensional phase space has m independent first integrals in involution,
F1 = H, F2 , . . . , Fm , then the system can be integrated by quadratures. Such a
system is called completely integrable.
4.2 Relativistic Particle as Dynamical System
Motion of a particle in a curved spacetime is special case of a dynamical system. If

D is the dimension of the spacetime and its coordinates are x a , a particle trajectory
is a line x a (τ ). The canonical coordinates in the corresponding phase space are
(x a , pa = gab ẋ b ). The canonical symplectic form and the Hamiltonian are
1 ab
Ω = d x a √ dpa , H= g pa pb . (14)
2
It is easy to show that the Hamilton equations of motion in the phase space are
equivalent to geodesic equations in the spacetime.
Let us assume that the Hamilton equations have an integral of motion of the form
K = K a1 ...aq (x) pa1 . . . paq , then
K (a1 ...aq ; aq+1 ) = 0. (15)
Such a tensor in the spacetime is called a Killing tensor. The Killing tensor of the
rank 1, ξa , is a Killing vector. The metric gab is a trivial Killing tensor.
It is well known that Killing vectors generate symmetry transformation on the
spacetime manifold with metric gab . Usually this symmetry is called an explicit sym-
metry. The Killing tensors of higher rank more directly connected with the properties
of the phase space, and their meaning is not so transparent in the spacetime itself.
For this reason, they are often called hidden symmetries.
Poisson brackets in the phase space after reduction to the spacetime determine the
Schouten-Nijenhuis brackets. When two integrals of motion are in the involution and
each of them is a monomial in momentum, the Schouten-Nijenhuis bracket for the
corresponding Killing tensors vanish. In such a case we simply say that the Killing
tensors commute.
The Liouville theorem being applied to relativistic particles implies that the geo-
desic equations are completely integrable (can be solved in quadratures) if there exist
D independent commuting Killing tensors. Some of them can be Killing vectors. D
dimensional Kerr-NUT-ADS admits n +ε Killing vectors. For complete integrability
there must exist additionally n Killing tensors. One of them is trivial, gab .
268 V. P. Frolov
4.3 Page’s Proposal
Let u a = d x a /dτ be a unit tangent vector to a particle trajectory, then Pba = δba +u a u b
is a projector to a plane, orthogonal to u a . Using the definition of the PCKYT (Eq. (7))
it is easy to show that Fab = Pac h cd Pbd is parallel transported along the geodesic.
Hence any scalar invariant constructed from this tensor is a constant of motion. To
obtain such scalars it is sufficient to take a trace of powers of the matrix Fab . In [15] it
was shown that these integrals of motion are independent and commuting, and their
total number is sufficient for the complete integrability of the geodesic equations in
the Kerr-NUT-ADS spacetime. However these set of integral of motion is reducible.
In other words most of these integrals are of high rank,while there exist a complete
set of lower rank integrals of motion. In fact, in a general case in the presence of the
PCKYT the irreducible set of integrals of motion is determined by (n + ε) Killing
vectors and n Killing tensors of the second rank. To demonstrate this we need to
consider Killing and Killing-Yano tensors in more detail.
5 PCKYT and Killing-Yano Tower
5.1 Killing-Yano Tensors
Let us first introduce two objects. The Killing-Yano (KY) tensor ka1 ...aq is an anti-
symmetric q−form on the spacetime, which obeys the equation
⊗a ka1 ...aq = ⊗[a ka1 ...aq ] . (16)
On the other hand, the closed conformal Killing-Yano (CCKY) tensor h a1 ...aq is
an antisymmetric q—form the covariant derivative of which is determined by its
divergence
⊗a h a1 ...aq = qga[a1 ξa2 ...aq ] , (17)
1
ξa2 ...aq = ⊗b h ba2 ...aq . (18)
D −r +1
KY and CCKY tensors are related to each other through the Hodge duality: the
Hodge dual of a KY form is a CCKY tensor, and vice versa. It is easy to check that
if ka1 ...aq is a Killing-Yano tensor, then
a ...aq
K ab = kaa2 ...aq kb 2 (19)
is a Killing tensor. We shall use the following schematic notation for this operation
K = k · k.
CCKY tensors possess the following remarkable property: An external product of

two CCKY tensors is again a CCKY tensor. This property was at first proved in the
tensorial form in [16]. Slightly later a simple direct proof of this result was obtained
in [17] using the formalism of differential forms.
5.2 Killing-Yano Tower
Let us return to our main ‘hero’—Principal Conformal Killing-Yano tensor. This is

a special case of CCKY tensor. Its additional properties are that it has tensor rank 2
and it is non-degenerate. Namely these properties make it so useful. If the spacetime
admits a PCKYT h it also has a whole set of other objects, which we call Killing-Yano
tower. First of all one can define a set of external powers of h
h√( j) = ⎩h √ .⎫⎬
. . √ h⎭ . (20)
j times
All of these objects are CCKY tensors of different matrix rank, starting from 2 for
j = 1 till 2(n − 1) for j = n − 1. Taking Hodge dual tensors for each of h√( j) with
j = 1, . . . , n − 1 one obtains a set n − 1 Killing-Yano tensors k( j) , and their squares
K( j) = k( j) · k( j) (21)
determine n − 1 Killing tensors of the rank 2.

For j > n h√( j) = 0. The case j = n is special. In the even dimensional case for
j = n one obtains an absolutely antisymmetric object. Its Hodge dual is a scalar. For
the odd dimensional case and j = n, the Hodge dual object is a Killing vector. One
can also show that action of the Killing tensors K( j) on the primary Killing vector
gives a new independent (secondary) Killing vectors.
As a result of this construction starting with the PCKYT h one obtains (n − 1)
second rank Killing tensors and (n + ε) Killing vectors. This, together with metric
gives D = 2n + ε integrals of motion. Additional check shows that they are inde-
pendent and in involution. This proves that the geodesic motion in the spacetime
with PCKYT (and hence in the most general Kerr-NUT-ADS metric) is completely
integrable.
Complete integrability of physically interesting finite dimensional dynamical sys-
tems is quite rare property. Examples of such systems include: motion in Euclidean
space under central potential; motion in the two Newtonian fixed centers; geodesics
on an ellipsoid; Jacobi motion of a rigid body about a fixed point (Euler, Lagrange
and Kowalevski); Neumann model (for more details and examples see e.g. the mono-
graph [18]). Characteristic property of a completely integrable dynamical system is
that its trajectories are regular and can be used to construct regular foliations of the
phase space. The opposite case are chaotic dynamical systems. It should be empha-
270 V. P. Frolov
sized that the study of geodesic motion in higher dimensional black holes provides
one with new wide class completely integrable dynamical systems, which might have
interesting applications both in the theoretical and mathematical physics.
6 Separation of Variables
6.1 General Remarks
The complete integrability of geodesic equations in the higher dimensional black

holes is closely related to the property of the complete separation of variables in
some field equations in the same spacetime.
Consider a Hamiltonian H (p, q), where p = ( p1 , . . . , pm ) and q = (q 1 , . . . , q m ).
The Hamilton-Jacobi equation for this Hamiltonian is a first order partial differential
equation for function S(q) of the form
H (S,q , q) = 0. (22)
Here S,q = (S,q 1 , . . . , S,q m ). Suppose q 1 and S,q 1 enter this equation only in a special
combination Φ1 (S,q 1 , q 1 ). Then the variable q 1 can be separated and a solution S
can be written in the form
S = S1 (q 1 , C1 ) + S ∗ (q 2 , . . . , q m ) , Φ1 (S,q 1 , q 1 ) = C1 , (23)
and the new function S ∗ (q 2 , . . . , q m ) obeys a reduced Hamilton-Jacobi

∗
H1 (S,q 2 , . . . , q , . . .) = 0.
2
(24)
Complete separation of variables implies that the solution S(q) can be written in the
form
S = S1 (q 1 , C1 ) + S2 (q 2 , C1 , C2 ) + . . . + Sm (q m , C1 , . . . , Cm ). (25)
The constants Ci generate first integrals on the phase space. When these integrals are
independent and in involution, the system is integrable in the Liouville sense (see
e.g. [19]).
6.2 Complete Separation of Variables in Kerr-NUT-ADS Spacetime
Complete integrability of geodesic equations in the four-dimensional Kerr metric was

discovered by Carter [20] who succeeded to separate variables in the corresponding
Hamilton-Jacobi equation. Similar approach does work also in five dimensional
case for the Myers-Perry metric, written in a similar Boyer-Lindquist coordinates

[21]. However numerous attempts to separate variable in the higher dimensional
Myers-Perry metric were not successful, except some special cases when additional
restrictions were imposed on the rotating parameters.
Nevertheless, the general Kerr-NUT-ADS spacetime with arbitrary number of
dimensions allows complete separation of variables for main field equations. The
reason is the following. The very property of the separation of variables implies
that there exist such a special coordinate system, in which this property is valid.
For example, the Boyer-Lindquist in the Myers-Perry spacetime with D > 5 are
not good for this purpose. The separation is possible only if one makes additional
restrictions of the parameters of the solution. The existence coordinates in which the
separability takes place in the Kerr-NUT-ADS in a general case is connected with
presence of the Principal Conformal Killing-Yano tensor.
Let us describe these special coordinates. We already mention that eigen-values
xi , determined by Eq. (11), are independent, and they can be used as n coordinates (at
least in some spacetime domain). Moreover the spacetime with PCKYT has (n + ε)
independent commuting Killing vectors ξ( j) , j = 0, . . . , n −1+ε. Consider integral
lines of these vector fields
dy a
= ξ(aj) . (26)
dψ j
It can be shown that the set of (D = 2n + ε)−quantities (xi , ψ j ) can be used as

coordinates. Namely in these coordinates the complete separation of variables takes
place. This was first demonstrated for Klein-Gordon and Hamilton-Jacobi equations
in the higher dimensional Kerr-NUT-ADS in [22]. The massive Dirac equation has a
similar property [23]. In fact these three different types of equations are closely related
to on another. The Hamilton-Jacobi equation for S can be obtained as the eikonal
equation in the lowest order of WKB approximation, by substituting ϕ ⇔ exp(i S)
into the Klein-Gordon equation
( − m 2 )ϕ = 0. (27)
On the other hand, the massive Dirac equation is just ‘a square root’ of the Klein-
Gordon equation.
6.3 Complete Integrability and Separation of Variables in Weakly

Charged Black Holes
Till now we discussed black hole solutions of the Einstein equation (1). It is interesting
that their nice properties still remain valid for a wider class of the black holes which
are slightly charged. We assume that the cosmological constant vanishes, so that a
Killing vector ξ a field obeys the equation
272 V. P. Frolov
ξa = 0. (28)
This equation coincides with the Maxwell equation for the electromagnetic field
potential Aa in the Lorentz gauge Aa;a = 0. Hence, one can consider ξ a as a test
electromagnetic field on a given Kerr-NUT background and include the interaction of
charged particles with it. For the primary Killing vector ξ(0) such a system describes
a weakly charged black hole. The secondary fields can be used to describe weakly
magnetized ones.
It is interesting that equations of motion of charged particles in weakly charged
higher dimensional black holes are completely integrable and the corresponding
Hamilton-Jacobi and Klein-Gordon equations are completely separable [24]. More-
over, the complete separability also takes place for charged Dirac equations in a
weakly charged black hole [25].
In these notes I focused on the remarkable properties of higher dimensional rotat-
ing black holes with the spherical topology of the horizon. It was demonstrated that
many of their properties are quite similar to the properties of their 4 dimensional
Kerr-NUT-ADS ‘cousins’. The reason of this is the existence of the Principal Con-
formal Killing-Yano tensor, which determines quite ‘rigid’ structure of the solutions
and serves as a ‘seed’ generating their explicit and hidden symmetries. Many recent
interesting generalization of this approach and its application is discussed in the
review article of Marco Cariglia included in this volume. It also contains plenty of
important references. I need also to mention three general reviews on this subject
[26–28] which contain a lot of additional information and references.
Acknowledgments The author thanks David Kubiznak for useful remarks. The author thanks the
Natural Sciences and Engineering Research Council of Canada and the Killam Trust for the financial
support. He also appreciate the financial support from CNRS and the hospitality of the University
of Tours, where this chapter was written.
References
1. Emparan, R., Reall, S.: A rotating black ring solution in five dimensions. Phys. Rev. Lett. 88,
101101 (2002). doi:10.1103/PhysRevLett.88.101101
2. Emparan, R., Reall, S.: Black holes in higher dimensions. Living Rev. Relativ. 11(6), lrr-2008-6
(2008). http://www.livingreviews.org/lrr-2008-6
3. Newman, E., Tamburino, L., Unti, T.: Empty-space generalization of the Schwarzschild metric.
J. Math. Phys. 4, 915 (1963). doi:10.1063/1.1704018
4. Bonnor, W.B.: Physical interpretation of vacuum solutions of Einstein’s equations. Part I.
Time-independent solutions. Gen. Relativ. Gravit. 24, 551 (1992). doi:10.1007/BF00760137
5. Carter, B.: Hamilton-Jacobi and Schrödinger separable solutions of Einstein’s equations. Com-
mun. Math. Phys. 10, 280 (1968)
6. Carter, B., A new family of Einstein spaces, Physics Letters A, 26, 399 (1968). doi:10.1016/
0375-9601(68)90240-5
7. Tangherlini, F.R.: The flatness problem and the pulsating universe. Nuovo Cimento B
108(1253), 1995 (1993). doi:10.1007/BF02741278. Erratum: ibid. 110, 1505
8. Myers, R.C., Perry, M.J.: Black holes in higher dimensional space-times. Ann. Phys. (N.Y.)
172, 304 (1986). doi:10.1016/0003-4916(86)90186-7
9. Chen, W., Lü, H., Pope, C.N.: General Kerr-NUT-Ads metrics in all dimensions. Class. Quan-
tum Grav. 23, 5323 (2006). doi:10.1088/0264-9381/23/17/013
10. Frolov, V.P., Kubiznák, D.: Hidden symmetries of higher-dimensional rotating black holes.
11. Kubizňák, D., Frolov, V.P.: The hidden symmetry of higher dimensional kerr-nut-ads space-
times. Class. Quantum Grav. 24, F1 (2007). doi:10.1088/0264-9381/24/3/F01
12. Houri, T., Oota, T., Yasui, Y.: Closed conformal Killing-Yano tensor and Kerr-NUT-de Sitter
space-time uniqueness. Phys. Lett. B 656, 214 (2007). doi:10.1016/j.physletb.2007.09.034
13. Krtouš, P., Frolov, V.P., Kubizňák, D.: Hidden symmetries of higher-dimensional black holes
and uniqueness of the Kerr-NUT-(A)ds spacetime. Phys. Rev. D 78, 064022 (2008). doi:10.
1103/PhysRevD.78.064022
14. Houri, T., Oota, T., Yasui, Y.: Closed conformal Killing-Yano tensor and the uniqueness of
generalized Kerr-Nut-de Sitter spacetime. Class. Quantum Grav. 26, 045015 (2009). doi:10.
1088/0264-9381/26/4/045015
15. Page, D.N., Kubizňák, D., Vasudevan, M., Krtouš, P.: Complete integrability of geodesic motion
in general higher-dimensional rotating black-hole spacetimes. Phys. Rev. Lett. 98, 061102
(2007). doi:10.1103/PhysRevLett.98.061102
16. Krtouš, P., Kubizňák, D., Page, D.N., Frolov, V.P.: Killing-Yano tensors, rank-2 Killing tensors,
and conserved quantities in higher dimensions. J. High Energy Phys. 2007(02), 004 (2007).
doi:10.1088/1126-6708/2007/02/004
17. Frolov, V.P.: Hidden symmetries of higher-dimensional black hole spacetimes. Prog. Theor.
Phys. Suppl. 172, 210 (2008). doi:10.1143/PTPS.172.210
18. Babelon, O., Bernard, D., Talon, M.: Introduction to Classical Integrable Systems. Cambridge
Monographs on Mathematical Physics. Cambridge University Press, Cambridge (2003)
19. Arnold, V.I.: Mathematical Methods of Classical Mechanics, Graduate Texts in Methematics,
vol. 60. Springer, New York (1989)
20. Carter, B.: Global structure of the Kerr family of gravitational fields. Phys. Rev. 174, 1559
(1968). doi:10.1103/PhysRev.174.1559
21. Frolov, V.P., Stojkovic, D.: Quantum radiation from a 5-dimensional rotating black hole. Phys.
Rev. D 67, 084004 (2003). doi:10.1103/PhysRevD.67.084004 (2 in 1, see FS2: Phys. Rev. D
68, 064011)
22. Frolov, V.P., Krtouš, P., Kubizňák, D.: Separability of Hamilton-Jacobi and Klein-Gordon
equations in general Kerr-NUT-Ads spacetimes. J. High Energy Phys. 2007(02), 005 (2007).
doi:10.1088/1126-6708/2007/02/005
23. Oota, T., Yasui, Y.: Separability of Dirac equation in higher dimensional Kerr-NUT-de Sitter
spacetime. Phys. Lett. B 659, 688 (2008). doi:10.1016/j.physletb.2007.11.057
24. Frolov, P., Krtouš, P.: Charged particle in higher dimensional weakly charged rotating black
hole spacetime. Phys. Rev. D 83, 024016 (2011). doi:10.1103/PhysRevD.83.024016
25. Cariglia, M., Frolov, V.P., Krtouš, P., Kubizňák, D.: Electron in higher-dimensional weakly
charged rotating black hole spacetimes, ArXiv e-prints. arXiv:1211.4631 [gr-qc] (2012)
26. Frolov, V.P., Kubizňák, D.: Higher-dimensional black holes: hidden symmetries and separation
of variables. Class. Quantum Grav. 25, 154005 (2008). doi:10.1088/0264-9381/25/15/154005
27. Cariglia, M., Krtouš, P., Kubizňák, D.: Hidden symmetries and integrability in higher dimen-
sional rotating black hole spacetimes. Fortschr. Phys. 60, 947 (2012). doi:10.1002/prop.
201200005
28. Yasui, Y., Houri, T.: Chapter 5. Hidden symmetry and exact solutions in Einstein gravity. Prog.
Theor. Phys. Suppl. 189, 126 (2011). doi:10.1143/PTPS.189.126
Part III
Cosmology and Quantum Gravity
Cosmological Models and Stability
Lars Andersson
I would already have concluded my researches about world

harmony, had not Tycho’s astronomy so shackled me that I
nearly went out of my mind.
Johannes Kepler
Letter to Herwart, quoted in [1, p. 127]
Abstract Principles in the form of heuristic guidelines or generally accepted dogma

play an important role in the development of physical theories. In particular, philo-
sophical considerations and principles figure prominently in the work of Albert
Einstein. As mentioned in the talk by Jiří Bičák at this conference, Einstein for-
mulated the equivalence principle, an essential step on the road to general relativity,
during his time in Prague 1911–1912. In this talk, I would like to discuss some aspects
of cosmological models. As cosmology is an area of physics where “principles” such
as the “cosmological principle” or the “Copernican principle” play a prominent role
in motivating the class of models which form part of the current standard model, I
will start by comparing the role of the equivalence principle to that of the principles
used in cosmology. I will then briefly describe the standard model of cosmology to
give a perspective on some mathematical problems and conjectures on cosmological
models, which are discussed in the later part of this paper.
1 Introduction
As stated by Einstein in his paper from 1912 [2], submitted just before his departure
from Prague, the equivalence principle is “eine Natürliche Extrapolation einer der
allgemeinsten Erfahrungssätze der Physik”,1 and can consequently be claimed to
1 “A natural extrapolation of one of the most general empirical propositions of physics”
L. Andersson (B)
Albert Einstein Institute, Am Mühlenberg 1, 14476 Potsdam, Germany
e-mail: laan@aei.mpg.de
278 L. Andersson
be exactly valid on all scales. Since the equivalence principle is compatible with
Einstein’s relativity principle of 1905 only in the limit of constant gravitational
potential, accepting the principle of equivalence meant that a new foundation for the
theory of gravitation must be sought. The challenge of doing so, which Einstein in
his 1912 paper poses to his colleagues: “Ich möchte alle Fachgenossen bitten, sich
an diesem wichtigen Problem zu versuchen!”, is one that he himself devoted the
coming years to, finally arriving at the 1915 theory of general relativity.
General Relativity describes the universe as a 4-manifold M with a metric gαβ of
Lorentzian signature. The Einstein equations,
1
Rαβ − Rgαβ + Λgαβ = 8π GTαβ , (1)
2
originally given in [3], relate the geometry of spacetime (M, gαβ ) to matter fields with
energy-momentum tensor Tαβ . By the correspondence principle, the stress energy
tensor Tαβ should correspond to the stress energy tensor of a special relativistic
matter model, and in particular be divergence free. For “ordinary matter” one expects
Tαβ to satisfy energy conditions such as the dominant energy conditon. Here I have
included the “cosmological constant term” Λgαβ in (1), which was not present in the
equations given in [3]. The left hand side of (1), where Rαβ is the Ricci tensor, R is
the Ricci scalar and Λ is a constant, is the most general covariant tensor expression
of vanishing divergence, depending on gαβ and its derivatives up to second order, and
linear in second derivatives. Further, its left hand side is the most general second order
Euler-Lagrange equation, derived by varying a covariant Lagrange density defined
in gαβ and its first two derivatives, see [4, 5] and references therein. The covariance
of the equations of general relativity under spacetime diffeomorphisms, makes the
theory compatible with the strong version of the equivalence principle.
Since it can be claimed to be exactly valid, the equivalence principle is subject to
empirical tests and there is a long history of experiments testing various versions of
the (weak or strong) equivalence principles, see e.g. [6], see also [7] in this volume.
Until the present, the equivalence principle has survived all experimental tests, and
an experiment clearly demonstrating a deviation from the predictions based on the
equivalence principle would necessitate a revision of the foundations of modern
physics.
The arguments of the physicist and philosopher Ernst Mach played an important
role in the development of Einstein’s ideas leading up to general relativity, including
the formulation of the equivalence principle. The fact that in general relativity, matter
influences the motion of test particles via its effect on spacetime curvature means
that in contrast to Newtonian gravity, the “action at a distance” which was criticized
by Mach is not present in general relativity, which hence agrees with the guiding idea
which Einstein referred to as “Mach’s principle”, i.e. loosely speaking the idea that
the distribution of matter in the universe determines local frames of inertia, see [8], see
also [9]. The role of Mach’s principle in the context of cosmology is discussed in [10].
This played a central role in Einstein’s development of general relativity, and also in
his discussion of general relativistic cosmology, but it appears difficult to formulate
Cosmological Models and Stability 279
Fig. 1 Kepler’s model of the solar system based on platonic solids, from Mysterium Cosmograph-
icum (1596)
experimentally testable consequences, cf. [11], although Mach’s principle has of

course been brought up in connection with “Newton’s bucket” and frame dragging.
The book [12] gives an excellent overview of issues related to Mach’s principle.
However, the principles which are most relevant for the present discussion are the
hierarchy of “cosmological principles”, for example the cosmological principle of
Einstein and the perfect cosmological principle of Bondi, Gold and Hoyle. See [13,
Sect. 2.1] for an overview of the cosmological principles. These principles play a
role which is fundamentally different from that of the equivalence principle, in the
sense that they do not make predictions which are expected to be exactly true at all
scales. At best, they can be viewed as simplifying assumptions that enable one to
construct testable physical models.
The work of Kepler, who is perhaps more intimately connected with Prague than
Einstein, provides an interesting illustration of the relationship between theoretical
principle and observation. In the time of Kepler, the world-model of Copernicus had
placed the sun at the center of the universe and described the planets as moving on
circular orbits around it. Not long before his move to Prague in 1600, Kepler believed
himself to have completed the Copernican world-model based on the mathematical
perfection of circles, by adding to it an element of equal perfection and beauty, namely
the geometry of the Platonic solids, which according to Kepler’s expectations would
determine the sizes of the planetary orbits (Fig. 1).
280 L. Andersson
Fortunately, it was possible for Kepler to use Tycho Brahe’s observational data
to test the predictions of his model. To his deep consternation Kepler realized that
the planets do not, after all, move on circular orbits. The beautiful principles which
had inspired Kepler to laboriously analyze the observational data of Tycho had to
be discarded. In analyzing the data, Kepler not only discovered his three laws of
planetary motion but also came close to introducing the notion of force which became
fully clear only through the work of Newton. One could say that through the work
of Kepler and later Newton, one set of “a priori” principles (those of Copernicus
and Kepler) were replaced by a model based on the dynamical laws of Newtonian
gravity.
1.1 The Cosmological Principle
Although Newtonian ideas continued to dominate physics throughout the 19th cen-
tury, there were well known anomalies of a theoretical as well as observational
nature, and these served as a guide for the developments of the early 20th century.
The conflict between the covariance of Maxwell theory under the Lorentz group and
the more restricted invariance properties of the Newtonian laws led to the introduc-
tion of special relativity. Similarly, as discussed above, the incompatibility of special
relativity and gravitation led to the development of general relativity. The expla-
nation of the anomalous precession of the perihelion of Mercury [14]2 by general
relativity was, together with its new prediction for the deflection of light by the sun,
confirmed by subsequent observations [17], were among the factors which led to its
rapid acceptance.
Among the main paradoxes of Newtonian physics and world view in applications
to cosmology were Olbers’ paradox and the incompatibility of Newtonian gravity
with infinitely extended homogenous matter distributions, which had prevented the
construction of a cosmological model consistent with Newtonian ideas. This latter
fact, which had been elucidated by von Seeliger and others, see [18] for discussion
and references, played an important role in Einstein’s reasoning about cosmological
models in his 1917 paper [19], in particular in motivating the introduction of the
cosmological constant in that paper.
As has already been mentioned, the philosophy of Mach, albeit firmly based in
Newtonian physics, was an important source of inspiration for Einstein. However,
incorporating Machian ideas in a general relativistic cosmology presented serious
difficulties. After some early attempts had been discarded, Einstein in [19] adopted
a spatially homogenous model of the universe as a means of making a general rel-
ativistic cosmology compatible with Machian ideas. Introducing a “cosmological
constant” term Λgαβ in the field equation of general relativity, which Einstein first
2 From the perspective of the current situation in physics, it is amusing to recall that attempts had
been made in the 19th century to explain the observed precession of Mercury both by dark matter
[15] (the planet Vulcan hypothesis) as well as modifications of gravity [16].
Fig. 2 Hubble’s original 1929 graph [21]
motivated through a discussion of homogenous matter distributions in Newtonian

gravity, and assuming that there is a family of observers who see the same matter
density everywhere, led to a static universe filled with a homogenous and isotropic
matter distribution. The spacetime of the Einstein model is a Lorentzian cylinder.
The line element takes, up to a rescaling, the form
ds 2 = gαβ d x α d x β = −dt 2 + g S 3 .
This give a solution to (1) with positive Λ, and with matter consisting of a pressureless
fluid with everywhere constant energy density.
Shortly after Einstein’s initial work on a static general relativistic cosmology,
Friedmann [20] proposed a model of an expanding universe
ds 2 = −dt 2 + a 2 (t)gκ (2)
where a(t) is a scale factor, gκ for κ = +1, 0, −1 is the sperical, flat or hyperbolic
metric. Line elements of the form (2) are also called Robertson-Walker line elements,
see below. During the 1920s, Lemaître and Hubble showed, based on observational
work of Slipher, Humason and others, that redshift increases with distance leading
to the Hubble law, see Fig. 2, which fits with the expanding Friedmann models. In
the context of the expanding Friedmann models, Olbers’ paradox can be resolved.
Expanding Friedmann models containing ordinary matter have a ≈ 0 at some time
in the past, where spacetime curvature and matter densities diverge. These models
led, via the work of Lemaître, Gamow, Hoyle and others, to the hot big bang model
which is the basis for the cosmological models in use today.
Milne criticized the big bang models on the basis that they introduced an extra-
neous “cosmic time” and also that they lacked explanatory power (e.g. the sign of
282 L. Andersson
the spatial curvature is a priori undetermined). Instead, he proposed an extension of

what he termed “Einstein’s cosmological principle”, to the effect that “The universe
must appear the same to all observers” [22]. Milne added to this the postulate that
observations are interpreted by each observer according to the principles of special
relativity and argued that this “extended relativity principle” led to an essentially
unique cosmological model.
The derivation of the general form of the line element compatible with the isotropy
of the universe, and also with Einstein’s cosmological principle in the sense discussed
by Milne was given by Robertson [23] and Walker [24] around the same time, and
found to be of the same form as that used by Friedmann and Lemaître in their
cosmological models. As pointed out by Robertson [25], the general relativistic line
element compatible with Milne’s cosmology is a special case of (2), namely the
empty κ = −1 universe, which is locally isometric to Minkowski space. This is
therefore known as the Milne model.
It was a similar dissatisfaction with the lack of predictivity of general relativistic
cosmology that led Bondi and Gold [26] and Hoyle [27] to introduce the “perfect
cosmological principle”, which is essentially a version of the postulate of Milne, but
viewed from the perspective of general relativity. By allowing for creation of matter,
they showed that it is possible to construct an expanding cosmological model satisfy-
ing this principle. However, the perfect cosmological principle tightly constrains the
possible models of the universe and the resulting steady state model is considered to
be incompatible with observations. The book of Kragh [28] contains an interesting
discussion of the conflict between the steady state model and the now-standard “big
bang” cosmology.
From the current perspective, it may be said that the introduction of what Milne
called Einstein’s cosmological principle led to a class of general relativistic cos-
mological models. By introducing a collection of perfect fluids, a much simplified
version of the problem of cosmological modelling reduces to the problem of fitting a
relatively small number of parameters to observational data, which could be said to
put cosmology on a similar footing as high energy particle physics. Indeed, as men-
tioned by Peebles [29, Chap. 1], it was Weinberg [30] who introduced the notion,
borrowed from high energy particle physics, of a “standard model” into cosmology.
At present, with the tremendous influx of data from observations of many differ-
ent types and at many different wavelengths, including observations of the cosmic
microwave background and galaxy surveys, it is often stated that we are entering an
era of precision cosmology. However, the widening range of observational methods
makes the process from observations to parameter estimation increasingly complex.
In particular, the prominent role of simplifying assumptions or principles in the
formulation of cosmological models and the model dependence in the analysis of
astronomical data, makes it important to keep in mind the difference between a model
which fits data to a high degree of precision and a model which accurately describes
the actual universe [31].
2 Cosmological Models
For a Friedmann model, with line element of the form (2), the stress energy tensor
has the form
Tαβ = ρuα uβ + p(gαβ + uα uβ ) ,
which is compatible with perfect fluid matter. Here u α is the unit timelike normal to
the t level sets, which in the special case of the Friedmann model coincides with the
normalized 4-velocity of the fluid particles, ρ is the energy density of the matter and
p is the pressure. We consider matter and radiation as described by a collection of
fluids, indexed by i, with linear equations of state,
pi = ωi ρi .
The Hubble constant (i.e. up to a constant factor the mean curvature of the t level
sets) is
H = ȧ/a
In the special case of a Friedmann model, the contribution of the curvature of the t
level sets in the Einstein equations can be described in terms of a fluid with equation
of state p = −ρ/3, while the effect of the cosmological constant can be described
by a fluid satisfying p = −ρ. Thus if we consider a simple model containing a fluid
with pressure zero (dust), and with a cosmological constant Λ, this can be described
by introducing the dimensionless density parameters
8π
Ωm = ρm , “Matter”: ω = 0,
3H 2
κ
Ωκ = − 2 2 , “Curvature”: ω = −1/3,
a H
8π
ΩΛ = ρΛ , “Vacuum”: ω = −1.
3H 2
The model can be parametrized by the present values
Ωm0 , Ωκ0 , ΩΛ0 ,
of the density parameters. The conservation of matter and equation of state implies
that the fluid densities ρi depend only on the scale factor
ρi ⊗ a −3(1+ωi ) . (3)
The Hamiltonian constraint (i.e. the projection of the Einstein equations (1) on u α )
takes the form
Ωm + Ωκ + ΩΛ = 1, (4)
284 L. Andersson
Ωm = 0 Ωm = 1
M Bb
κ = −1 κ=0 κ=1
Fig. 3 The dynamics of Friedmann dust models for Λ = 0
which, using (3), can be written as
H2 a 3 a 2
0 0
= Ω0m + Ω0Λ + Ω0κ . (5)
H02 a a
Here H0 , a0 are the present value of the Hubble constant and of the scale factor
respectively. Due to the uncertainty in the value of H0 , it is usually given in terms of
a dimensionless parameter h as
H0 = 100 hkm s−1 Mpc−1 .
Equation (5) can be integrated to relate observable quantities, e.g. redshift and lumi-
nosity distance, for given values of the parameters H0 , Ωm0 , Ωκ0 , ΩΛ0 .
It is convenient to study the global behavior of Friedmann models in terms of the
dimensionless density parameters. This analysis is explained in [32, Chap. 2], see
also [33, 34]. Due to the Hamiltonian constraint (4), we have Ωκ = 1 − Ωm − ΩΛ .
The fixed points of the dynamical system in the (Ωm , ΩΛ ) plane are the Einstein-
de Sitter big-bang model Bb = (1, 0) and the spatially flat de Sitter model dS =
(0, 1), as well as the empty κ = −1 Milne model M = (0, 0). One finds that Bb
is a source and dS is a sink, while M is a saddle point. The static Einstein universe
has H = 0, so the dimensionless parameters Ωm and ΩΛ are ill-defined, but this
point may be represented in an extended phase space as E = (◦, ◦). This point
is unstable, but is connected to the source Bb by an exceptional trajectory, which
separates the models which recollapse from those which expand forever.
Restricting to Λ = 0, the only fixed points are Bb and M, with Bb a source and
M a sink, see Fig. 3. The unstable Einstein-de Sitter universe Bb has slow volume
growth a √ t 2/3 , while the stable Milne universe M has volume growth a √ t. In
fact, this growth rate is maximal among Λ = 0 models. This indicates that rapid
volume growth goes together with stability.
Now we can give an extremely simplified description of the current situation in
cosmology by saying that the laws of general relativity together with the cosmological
principle and observations leads to the “standard model” with the cosmological
parameters
Ωκ0 √ 0, Ωm0 √ 0.3, ΩΛ0 √ 0.7, h √ 0.7.
The standard model is a big bang model. There is an initial singularity, a ≈ 0 as

t ≈ 0 and the universe expands indefinitely to the future, a ∗ ◦ as t ∗ ◦. The
Fig. 4 This figure shows some orbits for Friedmann cosmologies with dust and dark energy (Λ)
in the (Ωm , ΩΛ ) plane. The Einstein-de Sitter point Bb = (1, 0) is a source, the Milne point
M = (0, 0) is a saddle node, and the de Sitter point dS = (0, 1) is a sink. See [33] for background
model predicts a hot big bang, which leads to the prediction of cosmic background
radiation [35, 36]. The observation of a highly homogenous cosmic background
radiation with a spectrum close to that of a black body is a major success of the big
bang models of cosmology.
Most of the energy density in the standard model consists at present of as yet
unknown “dark matter” (accounting for approximately 85 % of the matter density)
and “dark energy” in the form of the cosmological constant. Dark matter, which
for a long time has been broadly accepted in astronomy and cosmology, cf. [37], is
distinguished from dark energy by the fact that its existence is motivated by studies of
the dynamics of galaxy clusters and galactic rotation curves, which are independent of
the Friedmann model which forms the basis of the standard model in cosmology. On
the other hand, the cosmological constant was deemed unacceptable on philosophical
grounds and entered the standard model fairly recently, shortly before the year 2000;
the effects of dark energy being seen only indirectly via cosmological models and
eg. studies of structure formation in the universe (Fig. 4).
The acceptance of Λ came about only after the observation of the dimming of type
Ia supernovae. The observations are interpreted as saying that the rate of expansion
is accelerated, i.e. ä > 0, which is incompatible with a Friedmann model filled
with ordinary matter and Λ = 0. Figure 5 shows the supernova data compared to
the standard model and Einstein-de Sitter. The horizontal axis is the Milne model
(Ωm = ΩΛ = 0).
286 L. Andersson
Fig. 5 Magnitude residual for SNe Ia Gold data [38] (dots) relative to the Milne model, plotted
against redshift z. The black, solid curve is the standard model, while the green, dashed curve is
the Einstein-de Sitter model. The horizontal axis is the Milne model
2.1 Cosmological Problems
One of the important arguments against introducing the cosmological constant (apart
from the difficulty of explaining the value Λ which appears motivated by cosmology
from the point of view of particle physics) has been the coincidence problem, which
might also be termed the “why now” problem. Figure 6 shows the time evolution of
the dark energy density ΩΛ . We see that it is only close to the present epoch that ΩΛ
becomes significant, and in the later universe it will dominate the dynamics. Due to
the different scaling behavior of the matter and Λ densities in view of (3), the fact
that these are both of order unity at the present epoch is a coincidence that could be
argued to be contrary to the idea that we are not “special observers”. In contrast, in
the Einstein-de Sitter model the matter density is time independent.
A related problem is the flatness problem. Roughly speaking, this is the question
why Ωκ √ 0 at present. In case Λ = 0 this can be seen to be problematic simply
from Fig. 6. Since Bb is unstable, fine tuning of the initial conditions is required in
order to have Ωκ √ 0 at present. Lake [33] argues, using the presence of a conserved
quantity for the dynamics in the (Ωm , ΩΛ )-plane, that fine tuning is not needed to
have Ωκ √ 0 throughout the history of the universe.
The universe is not exactly homogenous or isotropic; this holds at best in an
approximate sense on sufficiently large scales. This raises the problem of whether
Fig. 6 The time evolution of the dark energy density, see [39] for discussion
it is possible to determine from observations, which are necessarily restricted to our

past light cone, to what extent, and at what scales, the assumption of homogeneity
and isotropy is valid. A problem here is that local isotropy (i.e. isotropy around the
world line of one observer) does not imply global homogeneity (Fig. 7).
The Ehlers-Gehren-Sachs theorem gives conditions under which it is possible to
conclude from exact isotropy of the cosmic microwave background that the universe
is exactly isotropic. However, this result can fail in several ways. For example, there
are homogenous but non-isotropic models where the CMB is exactly isotropic at one
instant in time. Extensions of the EGS theorem to situations where only approximate
isotropy of the CMB holds are problematic, see [40–42] and references therein.
This raises the problem of determining to what degree observations of the actual
universe can be modelled and analyzed in the framework of Friedmann models (and
perturbations thereof). One aspect of this problem is the question whether there is
a scale at which (statistical) homogeneity and isotropy can be said to hold. Current
estimates place this scale at approximately 150 h −1 Mpc, see e.g. [43], see also [44].
However, recent observations indicate the existence of inhomogenous structures of
a dimension which may be in conflict with isotropy at this scale, see [45]. It is
conceivable that observations which extend to ever higher redshifts continue to yield
evidence of structures in the universe of a size comparable to the homogeneity scale.
Some aspects of inhomogeneity in cosmology were recently surveyed in a focus
issue of CQG, see [46] and references therein.
288 L. Andersson
Fig. 7 Sloan digital sky survey galaxy map, from www.sdss.org
The question of how the potential effects of large scale inhomogeneities on obser-
vations should be analyzed raises several important issues. Ellis has formulated the
“fitting problem”, see [47] and references therein, which asks about the effect of ana-
lyzing observations from an inhomogenous universe via a Friedmann model which
is in some sense the “best fit” to the actual universe. The effect on observations of
the fact that the model universe used to analyze data is only an approximation of the
actual universe is sometimes referred to as “backreaction”. An important question
here is whether perturbation theory can be applied to take into account the deviation
of the model from the actual universe. Kolb et al. have argued [48] that this analysis
should take into account the peculiar velocities due to the different expansion rate
in the model and the actual universe. Another effect of inhomogeneities which also
sometimes is referred to as backreaction, is the dynamical effect of the inhomo-
geneities on the expansion of the universe. A possible approach is to use averaging
[49] or coarse-graining [50] to derive a set of effective equations modelling the uni-
verse. In order to carry out such a scheme, one must introduce closure relations which
allow one to extract an autonomous system. It is here worth mentioning the ideas on
multi scale averaging, see e.g. [51, 52]. In particular, Wiltshire [53] argues that one
should consider modifying the Copernican principle to take into account the idea that
we reside in a gravitationally bound structure in a universe which has both bound
systems and voids.
It is apparent that the matter distribution in the universe is “lumpy” due to the
matter concentrations in stars, galaxies and other structures, and inhomogenous due
to the presence of large scale voids and bound structures, and the effect of these
must be taken into account when analyzing observations, see Clarkson et al. [54] for
discussion. The optical properties of the universe are, in the Friedmann models which
form the basis for the standard model of cosmology, calculated using the properties
of a fluid which is used to approximate the actual matter distribution. Thus it is
necessary to analyze whether the optical properties of a lumpy matter distribution
differ in a significant way from the optical properties of a fluid. Light from distant
stars passes through the gravitational wells of bound objects as well as voids on the
way to the observer, and the effect of this process must be analyzed and compared
to light passing through the fluid in a Friedmann model. This problem has been
studied by among others Clifton et al. [55], see also [56]. In this context, we also
mention the so-called swiss cheese models, in which one attempts to analyze the
optical effect of voids and structure in the universe by introducing under-densities
in a background Friedmann model, see e.g. [57] and references therein. The swiss
cheese models generally suffer from the limitation that the over-all expansion of the
model is determined by the chosen background Friedmann geometry.
In this situation one may contemplate introducing weaker cosmological principles,
incorporating ideas of statistical homogeneity, or weakening the Copernican principle
by restricting to matter bound observers as suggested by Wiltshire.
As we have seen, the standard cosmological model is not located at a fixed point
for the dynamical system governing the evolution of the dimensionless parameters
Ωm , ΩΛ , rather it is close to the spatially flat orbit connecting the source Bb to the
sink dS. Further, in that orbit, Ωm /ΩΛ takes on all positive real values. Thus, we
as observers are not in an asymptotic regime, but rather, as mentioned above, at a
special moment where Ωm and ΩΛ are both of order unity. Thus, from this point
of view, we are neither in the “early universe” or the “late universe” and we cannot
argue that our current universe is singled out as the asymptotic state of the evolution
of the universe.
This makes the situation in cosmology rather different from the situation in many
branches of physics where asymptotically stable objects are those which one expects
to find in nature. As an example, the Kerr black hole solution is expected to be the
unique stationary, asymptotically flat black hole spacetime. In order to establish the
astrophysical significance of this solution, it is essential to prove that it is stable. This
leads to the black hole stability problem, one of the central open problems in general
relativity. The problem of determining from observations whether or not for example
the supermassive black holes expected to be found at the center of most galaxies are
Kerr black holes or not is being actively studied.
As was just mentioned, from the point of view of the current standard model in
cosmology, questions about the asymptotics of cosmological models do not appear
to be the right ones to ask. Nevertheless, such questions give rise to interesting
mathematical problems which we shall discuss in the rest of this paper. The questions
about the asymptotic behavior of cosmological models include the structure of the
big-bang singularity and questions about the behavior in the expanding direction. In
particular we can ask: What does an observer in the late universe see?
290 L. Andersson
3 Asymptotics of Cosmological Models
In this section we will describe a scenario for the asymptotic future behavior of
cosmological models with vanishing cosmological constant. Recall that the Milne
model with line element
ds 2 = −dt 2 + t 2 gH3
where gH3 is the hyperbolic 3-metric with sectional curvature −1, is isometric to the
flat interior of the lightcone in Minkowski space. The Milne universe may be viewed
as the future of O, the origin in Minkowski space. This point represents the big bang
singularity in the Milne universe and is in the past of all spacetime points (i.e. all
observers). The cosmological time at a point P is the proper time elapsed from the
origin to P. The level surfaces of cosmological time are simply the hyperboloids. We
next consider a flat, but non-isotropic model, which may be viewed as a deformation
of Milne. Let I be a spacelike interval in Minkowski space and consider the future
of I . The resulting spacetime can be constructed by cutting the Milne spacetime by
a timelike hyperplane through O and gluing in a spacetime of the form R2+1 × I
with line element
−dt 2 + t 2 gH2 + dz 2 .
The deformed Milne spacetime has a big-bang singularity given by the interval I ,
and defining the cosmological time at P as the maximal proper time of any past
inextendible geodesic starting at P the level sets of cosmological time are as in
Fig. 8; it is flat and empty, but not homogenous and isotropic. Measuring the volume
of co-moving regions in the deformed Milne universe we see that in the deformed
regions, the volume of the cosmic time levels grows asymptotically as t 2/3 , i.e. the
growth rate of the Einstein-de Sitter universe, while in the undeformed regions,
the growth rate is asymptotically as t. The behavior is similar for the level sets of
the Hubble (mean curvature) time. On the other hand, asymptotically as t ∗ ◦, the
volume fraction in the undeformed region tends to 1, while in the asymptotic past
(near the big bang) these regions have a negligible volume fraction.
More general flat spacetimes may be constructed as the future of sets (e.g. frac-
tals) in Minkowski space, and quotients of these by the action of discrete groups
of isometries. Flat, or more generally, constant curvature spacetimes are examples
of G-structures and such spacetimes admitting compact Cauchy surfaces have been
completely analyzed, starting with the work of Mess [58], see also [59], who analyzed
the class of constant curvature 2 + 1 dimensional spacetimes admitting a compact
Cauchy surface. For example, one may show that the space of flat 2 + 1 dimensional
spacetimes with Cauchy surface of genus g > 1 is isomorphic to ∂M × M , where
M is Teichmuller space of surfaces of genus g and ∂M is the Thurston boundary.
The particular case of constant curvature spacetimes with compact Cauchy surface
has been analyzed in [60]. In particular, it was shown there that such flat spacetimes
can be globally foliated by Cauchy surfaces of constant mean curvature (i.e. constant
Hubble time).
a(t) ∼ t2/3 a(t) ∼ t
Fig. 8 A flat cosmological spacetime not isometric to Milne. A level set of cosmological time t is
shown. The vertical lines indicate the flat wedge which has been glued in
Neck region – slow volume growth
Hyperbolic regions
Fig. 9 Qualitative shape of Hubble level set
The level sets of Hubble time can be related to the level sets of the cosmological
time by an application of a maximum principle, and one may show that the volume
growth of these level sets is comparable to that of the level sets of the cosmological
time. This leads to a generalization of the statements made above for the simple
deformed Milne universe, see [61].
In view of the above mentioned work, these generalized Milne spacetimes may
have a very complex (e.g. fractal) big bang type initial singularity. In some cases their
future asymptotics can be analyzed, see [62]. One finds that the level sets of Hubble
time decompose into “neck regions” with slow volume growth, and “hyperbolic
regions” with fast volume growth. The scale free geometry of these level sets may
may be depicted as in Fig. 9.
In particular, one finds that in the asymptotically expanding direction, the volume
fraction of asymptotically, hyperbolic (thick) regions dominate while the neck regions
292 L. Andersson
T = −∞
T =∞
Fig. 10 The Einstein flow in the 2 + 1 dimensional case
(thin) become insignificant. Therefore, a “typical” (volume averaged) observer at late

time lives in a thick region.
It is interesting to compare the relation between the thin and thick regions to
the overdense and void regions in an inhomogenous universe containing matter, in
particular in view of the fact that the thin regions have volume growth approximating
that of Einstein-de Sitter universe which has critical matter density.
We now consider the generalization of the above picture to the case of general,
inhomogenous universes. We start by noting that the Lorentzian Einstein equations
define a flow on the space of (scale free) geometries. By analogy with the Ricci flow
of Riemannian geometries, this may be termed the Einstein flow.
For simplicity, we consider spacetimes (M, gab ) of dimension D = d + 1 which
are vacuum, i.e. with
Rαβ = 0 .
Suppose M admits a foliation by Cauchy surfaces of constant mean curvature H .

Introduce the dimensionless logarithmic constant mean curvature (Hubble) time T =
− ln(H/H0 ), and consider the evolution of the scale free geometry [g] = H 2 g. The
Lorentzian Einstein equations define a flow T ⇔≤ [g](T ), on the space of scale
free geometries. In particular, in the 2 + 1 dimensional case, the Einstein equations
correspond to a time dependent Hamiltonian system on Teichmüller space [63],
and each universe corresponds to a curve connecting a point on the boundary of
Teichmuller space to an interior point, see Fig. 10.
One arrives at the following heuristic scenario [64, 65]. Consider spacetimes with
Cauchy surface M. The non-collapsing case corresponds to the case where M has
negative Yamabe type. For T ∗ ◦, (M, [g]) decomposes into hyperbolic pieces and
Seyfert fibered pieces, and this decomposition corresponds to a (weak) geometriza-
tion, cf. [65]. The Einstein flow in CMC time results in a thick/thin decomposition of
M, where the thick (hyperbolic) pieces have full volume growth. As a consequence
we have that in the far future, the hyperbolic pieces represent most of the volume
of M, cf. Fig. 11. Proving statements along the lines described above appears to be
very difficult, and one must therefore start by considering sub-problems.
Fig. 11 The collapse of necks in the Einstein flow
4 Results on Nonlinear Stability
To give some perspective on the nonlinear stability problems introduced above, we

discuss some results on other stability problems in general relativity. These are orga-
nized according to the asymptotic model spacetime. The black hole stability problem,
cf. [66] for discussion and references, is not mentioned here. In the following, we
mention only the cases with conformally flat background spacetimes.
4.1 Minkowski
First we consider the nonlinear stability of Minkowski space, i.e. R4 with line element
ds 2 = −dt 2 + d x 2 + dy 2 + dz 2 .
The conformal type of Minkowski space is that of the Minkowski diamond, see
Fig. 12. In this causal diagram, each interior point represents a 2-sphere.
Nonlinear stability holds, in the sense that for Cauchy data near Minkowski data,
the maximal development is geodesically complete and asymptotically Minkowskian.
A key fact is that radiation carries energy through the conformal boundary I . Due
to the fact that the nonlinearity in the Einstein equations is quadratic, it is necessary
to exploit a cancellation in the equations in order to prove stability.
The first result in this direction is due to Friedrich [67], who proved that for data
close to the data induced on a hyperboloid in Minkowski space, one has nonlin-
ear stability to the future, and with suitable asymptotic regularity for the data, the
maximal development has a regular I + to the future of the initial slice. The full
nonlinear stability result was proved by Christodoulou and Klainerman [68]. This
work was extended to include the full peeling at I by Klainerman and Nicolo [69].
294 L. Andersson
I+
i0
I−
Fig. 12 Conformal diagram of Minkowski space
A simpler proof of nonlinear stability, using wave coordinates (spacetime harmonic

coordinates) gauge was given by Lindblad and Rodnianski [70]. Using both of these
methods, the proof of nonlinear stability can be readily adapted to the Einstein-matter
system, provided that the matter fields do not destroy the conformal properties of the
Einstein equations. Examples include a massless scalar field, which was included in
the work of Lindblad and Rodnianski, and a Maxwell field, see [71].
4.2 de Sitter
Next we consider cosmological models with positive Λ. The canonical example is

de Sitter space with line element
ds 2 = −dt 2 + cosh2 (t)g S 3 .
This is conformal to a finite cylinder with spacelike conformal boundary, and hence
one has future horizons and “locality” at I + (Fig. 13). Due to this fact, topology does
not matter for the future dynamics (but cf. [72]). Due to the locality at I + , we have
that a suitable notion for smallness in the stability argument can be defined locally
in space. We mention some results in this setting. Friedrich proved global nonlinear
stability of de Sitter space for the Einstein-Yang-Mills system with positive cosmo-
logical constant [73]. Ringström proved a “local in space” small data global existence
results for the Einstein-Λ-scalar field system [74, 75]. The case of fluid matter was
considered in this situation by Rodnianski and Speck [76] for the irrotational case,
see Speck [77] for the Einstein-Euler system. Finally, the Einstein-Λ-Vlasov system
has been studied by Ringström [78] (Fig. 13).
I+
de Sitter spacetime: conformal to a Milne spacetime: conformal to an

finite cylinder. infinite cylinder.
Fig. 13 Conformal diagrams of de Sitter and Milne spacetimes
4.3 Milne
Finally we consider the stability problem for a cosmological models with Λ = 0.
Here, the only general results are for the vacuum case. By passing to a quotient of
the Milne spacetime, we may consider a flat spacetime which has a Cauchy surface
isometric to a compact hyperbolic 3-manifold. The line element is
ds 2 = −dt 2 + t 2 gH3
(κ = −1 empty Friedmann) and the spacetime is conformal to an infinite cylinder
−dτ 2 + gH3
In this case topology does matter, in the sense that an observer is able to see the
whole past of his spacetime. Since there is no future conformal boundary, it is not
possible to localize the future evolution problem.
Future stability for Milne with compact Cauchy surface as described above was
proven by the author in collaboration with Moncrief for spacetime dimension d + 1,
d ≡ 3, cf. [79, 80], see also [81, 82]. For the 2 + 1 dimensional case, see [63].
Concerning the stability problem for the Einstein-matter systems in this setting,
much less is known than in the case with positive Λ. Some sub-problems have been
considered for the Einstein-Vlasov system in Bianchi symmetry (spacetimes with
a 3-dimensional Lie group acting by isometries on Cauchy surfaces), see [83–85].
Finally, we mention the work concerning test fluids on Friedmann backgrounds by
Speck [86].
The case of vacuum spacetimes with U (1) symmetry leads after a Kaluza-Klein
reduction to 2 + 1 dimesional gravity with wave maps matter. The nonlinear stability
of the flat cones over surfaces of genus g > 1 in this setting has been studied by
studied by Choquet-Bruhat and Moncrief, see [87, 88].
296 L. Andersson
5 Generalized Kasner Spacetimes
In Sect. 4.3 we discussed a stability theorem for the future of a Cauchy surface in a
class of spacetimes. The background spacetime in that case is a Lorentz cone over
a compact Einstein space with negative scalar curvature, i.e. a generalized Milne
space. In particular these are warped products of the line with an Einstein space. In
this section we shall discuss a class of double warped product spacetimes, with two
scale factors. These spacetimes which were considered in [89] may be viewed as
generalized Kasner spacetimes. They have the form
M√
=R×M × N,
with (M, g), (N , h), compact negative Einstein spaces of dimensions m, n, respec-
tively. The dimension of M is D = d + 1 = m + n + 1. We assume Ricg =
−(m + n − 1)g, Rich = −(m + n − 1)h. and consider a line element on M of the
form
ds 2 = −dt 2 + a 2 (t)g + b2 (t)h .
Let p = −ȧ/a, q = −ḃ/b, and introduce the scale invariant variables
1 1
P = p/H, Q = q/H, A= , B= .
aH bH
The Einstein equations imply an autonomous system for (P, Q, A, B) with 2 con-
straints. A dynamical systems analysis shows that the generic orbit has generalized
Kasner behavior, i.e. a √ t p , b √ t q at singularity, and is asymptotically Friedmann
(in fact asymptotic to a Lorentz cone spacetime) in the expanding direction
∇
a, b = t + O(t 1−λ ), λ∇ > 0 .
Friedmann is a stable node only in spacetime dimension D ≡ 11 (Fig. 14).
5.1 From α to ω
Belinskiı̆ et al. [90] argued that a generic cosmological singularities in 3 + 1 dimen-

sions in spacetimes with ordinary matter is oscillatory. The picture developed by
Belinskiı̆ et al. is often referred to as the BKL proposal. BKL type behavior has been
proved rigorously so far only for the Bianchi VIII and IX models, see [91], where
also strong cosmic censorship for this class of models was shown. On the other hand,
Belinskiı̆ and Khalatnikov [92] pointed out that cosmological singularities in space-
times containing stiff fluid or scalar field can be non-oscillatory, or quiescent. The
heuristic analysis of Belinskiı̆ and Khalatnikov was extended to the higher dimen-
sional case by Demaret et al. [93] who showed that quiescent behavior at singularity
(F1 ) (F1 )
A=
A=
0
0
(FB ) (FB )
(FA ) (FA )
B
B
=0
=0
(F2 ) (F2 )
D < 10 D > 10
Fig. 14 The dynamics of the generalized Kasner models [89]. The arrows point in the past direction.
There are five fixed points, one of which is the Friedmann point in the interior of the phase space.
The Friedmann point is a past unstable node for D > 10 and a unstable spiral point for D < 10.
The past stable fixed points F1,2 satisfy condition (6) for D > 10. This implies quiescent behavior
at the singularity for inhomogenous deformations of the generalized Kasner models in D > 10
in D = d + 1 dimensions holds if the condition
1 + p1 − pd − pd−1 > 0 (6)
holds, where pa are the generalized Kasner exponents at the singularity. This heuristic
analysis shows that (6) holds in vacuum only if D ≡ 11, and hence one expects that
generic vacuum, D < 11 spacetimes have oscillatory singularity, while generic
vacuum, D ≡ 11 spacetime have quiescent singularity. It was shown in [89, Sect. 4]
that (6) holds for generalized Kasner spacetimes if D ≡ 11, in agreement with the
result of Demarat et al. [93].
As a step towards making this heuristic scenario rigorous, the author showed
with Rendall [94] that generic D = 4 spacetime with scalar field has quiescent
singularity. In that paper we constructed a full parameter family of Einstein-scalar
field and Einstein-stiff fluid spacetimes with quiescent singularity using Fuchsian
analysis. This work was extended to the case of D ≡ 11 vacuum spacetimes by
Damour et al. [95], again using a Fuchsian analysis.
One may use the techniques discussed above to prove that a type of global non-
linear stability holds for a class of generalized Kasner spacetimes. It was shown in
[96] that for generalized Kasner spacetimes as above, with D ≡ 11, satisfying the
additional condition that the moduli space of negative Einstein metrics on M, N is
integrable (which is expected to hold in general), there is a full-parameter family
of C ω Cauchy data on M × N , such that the maximal Cauchy development (M, g)
has a global CMC time function, and has quiescent, crushing singularity. Further
298 L. Andersson
(M, g) is future causally complete and is asymptotically Friedmann to the future,

with g(T ) ≤ γ◦ M + γ N , as T ≤ ◦, where γ M and γ N are negative Einstein
◦ ◦ ◦
metrics on M, N , respectively. This applies to a large variety of factors M, N , and
can easily be generalized to multiple factors.
6 Concluding Remarks
In this paper we have given brief overview of some of the ideas underlying the
general relativistic cosmological models which form the core of the standard model of
cosmology, and pointed out the need for an improved analysis, both from the physical
and mathematical point of view, of the effect of deviations from homogeneity and
isotropy in the dynamics of cosmological models, and consequently in the analysis
of cosmological data. Motivated by this, we have discussed some results on nonlinear
stability for cosmological models. We end by listing some open problems.
The exponential expansion caused by the presence of the cosmological constant
in the case Λ > 0 and also in the presence of certain self-gravitating scalar field
models for inflation makes the large data future behavior of these models tractable
and here there are several results which do not require any symmetry assumptions,
see Sect. 4.2.
For the case Λ = 0 and ordinary matter, the situation is more delicate. The global
behavior of cosmological models is well understood in highly symmetric cases,
including the 3 + 1 dimensional Friedmann, Bianchi, Gowdy (spatial T 2 symmetric,
with symmetry action generated by hypersurface orthogonal Killing fields) and so-
called surface symmetric cases, see [97] and references therein. For the Bianchi
case, see the remarks in Sect. 5.1 and [98, 99], and for the Gowdy case see [100] and
references therein. However, for large data, the asymptotic behavior of the general
T 2 , U (1) (circle symmetric) and the full 3 + 1 case are mostly open. Similarly, future
stability is open in the 3 + 1 dimensional case for Einstein-matter models without
symmetry assumptions in the case Λ = 0. As an example, one would like to prove
nonlinear stability of Milne for Einstein-Vlasov. This is work in progress by the
author with Fajman.
Our understanding of the behavior of cosmological models in the direction of
the initial singularity is also limited. The BKL proposal provides a heuristic scenario
which has been verified only in the Bianchi case, where also strong cosmic censorship
has been shown to hold, see above. However, in spite of some recent progress [101–
103], even the question whether the singularity in generic Bianchi models is local,
is open. See [99, 104] for references and discussion. For Gowdy models with T 3
Cauchy surface, Ringström has proved that strong cosmic censorship holds, see [100]
for an overview, while for Gowdy with Cauchy surfaces diffeomorphic to S 3 or
S 2 × S 1 , and the general T 2 symmetric case (dropping the condition on hypersurface
orthogonality) the situation is much more complicated and cosmic censorship is open.
In particular, in the T 2 symmetric case, one has the new phenomenon of dynamical
spikes, see [105, 106].
The work by the author and Rendall, and by Damour et al. on quiescent
singularities, see Sect. 5.1 opens up the problem of proving quiescent behavior at the
singularity as well as global nonlinear stability for an open set of Cauchy data (in a
suitable topology). This is work in progress by the author and Ringström. Work on
this type of stability problem for the Friedmann case was mentioned in a recent talk
by Speck [107]. For the case D < 11 one may consider suitable Einstein-scalar field
models and for D ≡ 11 one may formulate the global nonlinear stability problem for
the generalized Kasner backgrounds as discussed in Sect. 5. Here it should be pointed
out that the global stability result mentioned there relies on Fuchsian methods and
therefore suffers from the same weakness as the work by the author and Rendall,
and Damour et al. on quiescent singularities. It would be interesting to prove a true
nonlinear stability result, stating that for an open set of Cauchy data close to the gen-
eralized Kasner background data, the maximal development is geodesically complete
to the future, asymptotically Friedmann, and with crushing singularity with geometry
close to, in a suitable sense, the singularity in the generalized Kasner spacetime.
For the near future, I expect that numerical studies of cosmological models in GR,
with less symmetry than the 2 Killing field models including LTB, T 2 and spherically
symmetric models studied in detail so far, will play an important role in exploring
the future behavior of cosmological models. One can expect that such investigations
will have an impact on both physical cosmology and the mathematical analysis of
cosmological models.
Acknowledgments I would like to thank Jiří Bičák and Bernd Schmidt for their comments on an
early version of the paper.
References
1. Caspar, M.: Kepler. Dover Publications, New York (1993)

2. Einstein, A.: Relativität und Gravitation. Erwiderung auf eine Bemerkung von M. Abraham.
Ann. Phys. 343, 1059 (1912). doi:10.1002/andp.19123431014
3. Einstein, A.: Näherungsweise Integration der Feldgleichungen der Gravitation, Sitzungs-
berichte der Königlich Preußischen Akademie der Wissenschaften, (Berlin), 844–847 (1915)
4. Lovelock, D.: The uniqueness of the Einstein field equations in a four-dimensional space.
Arch. Ration Mech. Anal. 33, 54 (1969). doi:10.1007/BF00248156
5. Lovelock, D.: Divergence-free tensorial concomitants. Aequationes Math. 4, 127 (1970)
6. Haugan, M.P., Lämmerzahl, C.: Principles of equivalence: their role in gravitation physics
and experiments that test them. In: Lämmerzahl, C., Everitt, C.W.F., Hehl, F.W. (eds.) Gyros,
Clocks, Interferometers: Testing Relativistic Gravity in Space. Lecture Notes in Physics, vol.
562, p. 195. Springer, Berlin (2001)
7. Bičák, J.: Einstein in Prague: relativity then and now. In: Bičák, J., Ledvinka, T. (eds.) General
Relativity, Cosmology and Astrophysics: Perspectives 100 Years After Einstein in Prague,
pp. 271–297. Springer, Berlin (2014)
8. Einstein, A.: Prinzipielles zur allgemeinen Relativitätstheorie. Ann. Phys. 360, 241 (1918).
doi:10.1002/andp.19183600402
9. Barbour, J.B.: The part played by Mach’s Principle in the genesis of relativistic cosmology.
p. 47, Cambridge University Press, Cambridge (1990)
300 L. Andersson
10. Bičák, J., Katz, J., Lynden-Bell, D.: Cosmological perturbation theory, instantaneous gauges,
and local inertial frames. Phys. Rev. D 76(6), 063501 (2007). doi:10.1103/PhysRevD.76.
063501
11. Will, C.: Testing Machian effects in laboratory and space experiments. In: Barbour, J.B.,
Pfister, H. (eds.) Mach’s Principle: From Newton’s Bucket to Quantum Gravity, p. 365. (1995)
12. Barbour, J.B., Pfister, H. (eds.): Mach’s Principle: From Newton’s Bucket to Quantum Gravity.
Birkhäuser, Basel (1995)
13. Lahav, O., Suto, Y.: Measuring our universe from galaxy redshift surveys. Living Rev. Relativ.
7, 8 (2004)
14. Le Verrier, U.J.: Theorie du mouvement de mercure. Ann. l’Obs. Paris 5, 1 (1859)
15. Lescarbault, M., Le Verrier, U.J.: Passage d’une planete sur le disque du soleil. Ann. l’Obs.
Paris 5, 394 (1860)
16. Hall, A.: A suggestion in the theory of mercury. Astron. J. 14, 49 (1894). doi:10.1086/102055
17. Dyson, F.W., Eddington, A.S., Davidson, C.: A determination of the deflection of light by the
sun’s gravitational field, from observations made at the total eclipse of may 29, 1919. R. Soc.
Lond. Philos. Trans. Ser. A 220, 291 (1920). doi:10.1098/rsta.1920.0009
18. Norton, J.D.: The Cosmological Woes of Newtonian Gravitation Theory, p. 271. Birkhäuser,
Boston (1999)
19. Einstein, A.: Kosmologische Betrachtungen zur allgemeinen Relativitätstheorie, Sitzungs-
berichte der Königlich Preußischen Akademie der Wissenschaften (Berlin). 142–152 (1917)
20. Friedman, A.: Über die Krümmung des Raumes. Z. Phys. 10, 377 (1922). doi:10.1007/
BF01332580
21. Hubble, E.: A relation between distance and radial velocity among extra-galactic nebulae.
Proc. Nat. Acad. Sci. 15, 168 (1929). doi:10.1073/pnas.15.3.168
22. Milne, E.A.: World-structure and the expansion of the universe. Mit 6 Abbildungen. Z. Angew.
Phys. 6, 1 (1933)
23. Robertson, H.P.: Relativistic cosmology. Rev. Mod. Phys. 5, 62 (1933). doi:10.1103/
RevModPhys.5.62
24. Walker, A.G.: On Riemanntan spaces with spherical symmetry about a line, and the conditions
for isotropy in Genj relativity. Q. J. Math. 6, 81 (1935). doi:10.1093/qmath/os-6.1.81
25. Robertson, H.P.: On E. A. Milne’s theory of world structure. Z. Angew. Phys. 7, 153 (1933)
26. Bondi, H., Gold, T.: The steady-state theory of the expanding universe. Mon. Not. R. Astron.
Soc. 108, 252 (1948)
27. Hoyle, F.: A new model for the expanding universe. Mon. Not. R. Astron. Soc. 108, 372
(1948)
28. Kragh, H.: Cosmology and Controversy: The Historical Development of Two Theories of the
Universe. Princeton University Press, Princeton (1996)
29. Peebles, P.J.E.: Principles of Physical Cosmology. Princeton University Press, Princeton
(1993)
30. Weinberg, S.: Gravitation and Cosmology: Principles and Applications of the General Theory
of Relativity. Wiley, New York (1972)
31. Peebles, P.J.E.: From Precision Cosmology to Accurate Cosmology. ArXiv e-prints
astro-ph/0208037 (2002)
32. Wainwright, J., Ellis, G.F.R.: Dynamical Systems in Cosmology. Cambridge University Press,
Cambridge (2005)
33. Lake, K.: The flatness problem and λ. Phys. Rev. Lett. 94(20), 201102 (2005). doi:10.1103/
34. Helbig, P.: Is there a flatness problem in classical cosmology? Mon. Not. R. Astron. Soc. 421,
561 (2012). doi:10.1111/j.1365-2966.2011.20334.x
35. Alpher, R.A., Bethe, H., Gamow, G.: The origin of chemical elements. Phys. Rev. 73, 803
(1948). doi:10.1103/PhysRev.73.803
36. Alpher, R.A., Herman, R.: Evolution of the universe. Nature 162, 774 (1948). doi:10.1038/
162774b0
37. van den Bergh, S.: The early history of dark matter. Publ. Astron. Soc. Pac. 111, 657 (1999).
doi:10.1086/316369
38. Riess, A.G., Strolger, L.G., Casertano, S., et al.: New Hubble Space Telescope discoveries
of type Ia supernovae at z ≡ 1: narrowing constraints on the early behavior of dark energy.
Astrophys. J. 659, 98 (2007). doi:10.1086/510378
39. Carroll, S.M.: The cosmological constant. Living Rev. Relativ. 4, 1 (2001). doi:10.12942/lrr-
2001-1
40. Clarkson, C.: Establishing homogeneity of the universe in the shadow of dark energy. C. R.
Phys. 13, 682 (2012). doi:10.1016/j.crhy.2012.04.005
41. Maartens, R.: Is the universe homogeneous? R. Soc. Lond. Philos. Trans. Ser. A 369, 5115
(2011). doi:10.1098/rsta.2011.0289
42. Räsänen, S.: Relation between the isotropy of the cmb and the geometry of the universe. Phys.
Rev. D 79(12), 123522 (2009). doi:10.1103/PhysRevD.79.123522
43. Marinoni, C., Bel, J., Buzzi, A.: The scale of cosmic isotropy. J. Cosmol. Astropart. Phys. 10,
036 (2012). doi:10.1088/1475-7516/2012/10/036
44. Sylos, F.: Labini. Inhomogen. Universe Class. Quantum Gravity 28(16), 164003 (2011).
doi:10.1088/0264-9381/28/16/164003
45. Clowes, R.G., Harris, K.A., Raghunathan, S., et al.: A structure in the early universe at
z√1.3 that exceeds the homogeneity scale of the R-W concordance cosmology. ArXiv e-
prints arXiv:1211.6256 (2012)
46. Andersson, L., Coley, A.: Inhomogeneous cosmological models and averaging in cosmology:
overview. Class. Quantum Gravity 28(16), 160301 (2011). doi:10.1088/0264-9381/28/16/
160301
47. Ellis, G.F.R.: Inhomogeneity effects in cosmology. Class. Quantum Gravity 28(16), 164001
(2011). doi:10.1088/0264-9381/28/16/164001
48. Kolb, E.W., Marra, V., Matarrese, S.: Cosmological background solutions and cosmological
backreactions. Gen. Rel. Grav. 42, 1399 (2010). doi:10.1007/s10714-009-0913-8
49. Buchert, T.: Toward physical cosmology: focus on inhomogeneous geometry and its non-
perturbative effects. Class. Quantum Gravity 28(16), 164007 (2011). doi:10.1088/0264-9381/
28/16/164007
50. Korzyński, M.: Covariant coarse graining of inhomogeneous dust flow in general relativity.
Class. Quantum Gravity 27(10), 105015 (2010). doi:10.1088/0264-9381/27/10/105015
51. Räsänen, S.: Evaluating backreaction with the peak model of structure formation. J. Cosmol.
Astropart. Phys. 4, 026 (2008). doi:10.1088/1475-7516/2008/04/026
52. Wiegand, A., Buchert, T.: Multiscale cosmology and structure-emerging dark energy: a plau-
sibility analysis. Phys. Rev. D 82(2), 023523 (2010). doi:10.1103/PhysRevD.82.023523
53. Wiltshire, D.L.: Exact solution to the averaging problem in cosmology. Phys. Rev. Lett. 99(25),
251101 (2007). doi:10.1103/PhysRevLett.99.251101
54. Clarkson, C., Ellis, G.F.R., Faltenbacher, A., et al.: (mis)interpreting supernovae observations
in a lumpy universe. Mon. Not. R. Astron. Soc. 426, 1121 (2012). doi:10.1111/j.1365-2966.
2012.21750.x
55. Clifton, T., Rosquist, K., Tavakol, R.: An exact quantification of backreaction in relativistic
cosmology. Phys. Rev. D 86(4), 043506 (2012). doi:10.1103/PhysRevD.86.043506
56. Bentivegna, E., Korzyński, M.: Evolution of a periodic eight-black-hole lattice in numerical
relativity. Class. Quantum Gravity 29(16), 165007 (2012). doi:10.1088/0264-9381/29/16/
165007
57. Marra, V., Kolb, E.W., Matarrese, S.: Light-cone averages in a swiss-cheese universe. Phys.
58. Mess, G.: Lorentz spacetimes of constant curvature. Technical Report IHES/M/90/28. Institute
des Hautes Etudes Scientifiques (1990)
59. Andersson, L., Barbot, T., Benedetti, R., et al.: Notes on: “Lorentz spacetimes of constant
curvature” [geom. dedicata 126 (2007), 3–45] by G. Mess. Geom. Dedicata 126, 47 (2007).
doi:10.1007/s10711-007-9164-6
302 L. Andersson
60. Andersson, L., Barbot, T., Beguin, F., Zeghib, A.: Cosmological time versus cmc time. Asian
J. Math. 16, 37 (2012)
61. Andersson, L.: Constant mean curvature foliations of flat space-times. Comm. Anal. Geom.
10(5), 1125 (2002)
62. Andersson, L.: Constant mean curvature foliations of simplicial flat spacetimes. Comm. Anal.
Geom. 13(5), 963 (2005). http://projecteuclid.org/getRecord?id=euclid.cag/1144438303
63. Andersson, L., Moncrief, V., Tromba, A.J.: On the global evolution problem in 2 + 1 gravity.
J. Geom. Phys. 23(3–4), 191 (1997)
64. Fischer, A.E., Moncrief, V.: Hamiltonian reduction of Einstein’s equations and the geometriza-
tion of three-manifolds. In: International Conference on Differential Equations, vol. 1, 2, pp.
279–282. World Scientific Publishing, River Edge (2000) (Berlin, 1999)
65. Anderson, M.T.: On long-time evolution in general relativity and geometrization of 3-
manifolds. Comm. Math. Phys. 222(3), 533 (2001). doi:10.1007/s002200100527
66. Andersson, L., Blue, P.: Hidden symmetries and decay for the wave equation on the Kerr
spacetime. ArXiv e-prints arXiv:0908.2265 (2009)
67. Friedrich, H.: On the existence of n-geodesically complete or future complete solutions of
Einstein’s field equations with smooth asymptotic structure. Comm. Math. Phys. 107(4), 587
(1986)
68. Christodoulou, D., Klainerman, S.: The Global Nonlinear Stability of the Minkowski Space.
Princeton University Press, Princeton (1993)
69. Klainerman, S., Nicolò, F.: Peeling properties of asymptotically flat solutions to the Einstein
vacuum equations. Class. Quantum Gravity 20, 3215 (2003). doi:10.1088/0264-9381/20/14/
319
70. Lindblad, H., Rodnianski, I.: The global stability of Minkowski space-time in harmonic gauge.
Ann. Math. 171(2–3), 1401 (2010). doi:10.4007/annals.2010.171.1401
71. Bieri, L., Zipser, N.: AMS/IP Studies in Advanced Mathematics. Extensions of the stability
theorem of the Minkowski space in general relativity, vol. 45. American Mathematical Society,
Providence (2009)
72. Andersson, L., Galloway, G.J.: Ds/cft and spacetime topology. Adv. Theor. Math. Phys. 6(2),
307 (2002)
73. Friedrich, H.: On the global existence and the asymptotic behavior of solutions to the Einstein-
Maxwell-Yang-Mills equations. J. Differ. Geom. 34(2), 275 (1991)
74. Ringström, H.: Future stability of the Einstein-non-linear scalar field system. Invent. Math.
173, 123 (2008). doi:10.1007/s00222-008-0117-y
75. Ringström, H.: Power law inflation. Commun. Math. Phys. 290, 155 (2009). doi:10.1007/
s00220-009-0812-6
76. Rodnianski, I., Speck, J.: The stability of the irrotational Euler-Einstein system with a positive
cosmological constant. ArXiv e-prints arXiv:0911.5501 (2009)
77. Speck, J.: The nonlinear future-stability of the flrw family of solutions to the Euler-Einstein
system with a positive cosmological constant. ArXiv e-prints arXiv:1102.1501 [math.AP]
(2011)
78. Ringström, H.: On the topology and future stability of models of the universe. Oxford Uni-
versity Press, Oxford, New York (2013)
79. Andersson, L., Moncrief, V.: Future complete vacuum spacetimes. In: The Einstein equations
and the large scale behavior of gravitational fields, pp. 299–330. Birkhäuser, Basel (2004)
80. Andersson, L., Moncrief, V.: Einstein spaces as attractors for the Einstein flow. J. Differ.
Geom. 89(1), 1 (2011). http://projecteuclid.org/getRecord?id=euclid.jdg/1324476750
81. Reiris, M.: Aspects of the long time evolution in general relativity and geometrizations of
three-manifolds (ProQuest LLC, Ann Arbor, MI). Ph. D. Thesis. State University of New
York at Stony Brook http://gateway.proquest.com/openurl?urlver=Z39.88-2004&rftvalfmt=
info:ofi/fmt:kev:mtx:dissertation&resdat=xri:pqdiss&rftdat=xri:pqdiss:3206447 (2005)
82. Reiris, M.: The ground state and the long-time evolution in the cmc Einstein flow. Ann. Henri
Poincaré 10(8), 1559 (2010). doi:10.1007/s00023-010-0027-6
83. Rendall, A.D., Tod, K.P.: Dynamics of spatially homogeneous solutions of the Einstein-Vlasov
equations which are locally rotationally symmetric. Class. Quantum Gravity 16, 1705 (1999).
doi:10.1088/0264-9381/16/6/305
84. Heinzle, J.M., Uggla, C.: Dynamics of the spatially homogeneous Bianchi type I Einstein-
Vlasov equations. Class. Quantum Gravity 23, 3463 (2006). doi:10.1088/0264-9381/23/10/
016
85. Nungesser, E.: Late-time behaviour of the Einstein-Vlasov system with Bianchi I symmetry.
J. Phys. Conf. Ser. 314(1), 012097 (2011). doi:10.1088/1742-6596/314/1/012097
86. Speck, J.: The stabilizing effect of spacetime expansion on relativistic fluids with sharp results
for the radiation equation of state. ArXiv e-prints arXiv:1201.1963 (2012)
87. Choquet-Bruhat, Y., Moncrief, V.: Future global in time Einsteinian spacetimes with u(1)
isometry group. Ann. Henri Poincaré 2(6), 1007 (2001)
88. Choquet-Bruhat, Y.: Future complete U(1) symmetric Einsteinian spacetimes, the unpolarized
case. In: The Einstein equations and the large scale behavior of gravitational fields, pp. 251–
298. Birkhäuser, Basel (2004)
89. Andersson, L., Heinzle, J.M.: Eternal acceleration from M-theory. Adv. Theor. Math. Phys.
11(3), 371 (2007). http://projecteuclid.org/getRecord?id=euclid.atmp/1185303966
90. Belinskiı̆, V.A., Khalatnikov, I.M., Lifshitz, E.M.: Oscillatory approach to a singular point in
the relativistic cosmology. Adv. Phys. 19, 525 (1970). doi:10.1080/00018737000101171
91. Ringström, H.: Curvature blow up in bianchi viii and ix vacuum spacetimes. Class. Quantum
Gravity 17, 713 (2000). doi:10.1088/0264-9381/17/4/301
92. Belinskiı̆, V.A., Khalatnikov, I.M.: Effect of scalar and vector fields on the nature of the
cosmological singularity. Soviet J. Exp. Theor. Phys. 36, 591 (1973)
93. Demaret, J., Henneaux, M., Spindel, P.: Nonoscillatory behaviour in vacuum Kaluza-Klein
cosmologies. Phys. Lett. B 164(1–3), 27 (1985)
94. Andersson, L., Rendall, A.D.: Quiescent cosmological singularities. Comm. Math. Phys.
218(3), 479 (2001)
95. Damour, T., Henneaux, M., Rendall, A.D., Weaver, M.: Kasner-like behaviour for subcritical
einstein-matter systems. NASA STI/Recon Technical Report N 2, 87809 (2002)
96. Andersson, L.: Stability of doubly warped product spacetimes. In: Sidoravicius, V. (ed.) New
Trends Math. Phys., pp. 23–32. Springer, Dordrecht (2009)
97. Andersson, L.: The global existence problem in general relativity. In: The Einstein equations
and the large scale behavior of gravitational fields, pp. 71–120. Birkhäuser, Basel (2004)
98. Ringström, H.: The future asymptotics of Bianchi VIII vacuum solutions. Class. Quantum
Gravity 18, 3791 (2001). doi:10.1088/0264-9381/18/18/302
99. Heinzle, J.M., Ringström, H.: Future asymptotics of vacuum Bianchi type VI0 solutions.
Class. Quantum Gravity 26, 145001 (2009). doi:10.1088/0264-9381/26/14/145001
100. Ringström, H.: Cosmic censorship for Gowdy spacetimes. Living Rev. Relativ. 13, 2 (2010)
101. Béguin, F.: Aperiodic oscillatory asymptotic behavior for some Bianchi spacetimes. Class.
Quantum Gravity 27(18), 185005 (2010). doi:10.1088/0264-9381/27/18/185005
102. Liebscher, S., Härterich, J., Webster, K., Georgi, M.: Ancient dynamics in Bianchi models:
approach to periodic cycles. Commun. Math. Phys. 305, 59 (2011). doi:10.1007/s00220-011-
1248-3
103. Reiterer, M., Trubowitz, E.: The BKL conjectures for spatially homogeneous spacetimes.
ArXiv e-prints arXiv:1005.4908 (2010)
104. Heinzle, J.M., Uggla, C.: Mixmaster: fact and belief. Class. Quantum Gravity 26(7), 075016
(2009). doi:10.1088/0264-9381/26/7/075016
105. Andersson, L., van Elst, H., Lim, W.C., Uggla, C.: Asymptotic silence of generic cosmological
singularities. Phys. Rev. Lett. 94(5), 051101 (2005). doi:10.1103/PhysRevLett.94.051101
106. Heinzle, J.M., Uggla, C., Lim, W.C.: Spike oscillations. Phys. Rev. D 86(10), 104049 (2012).
doi:10.1103/PhysRevD.86.104049
107. Speck, J.: On big bang spacetimes. In Dafermos, M. et al (eds.) Mathematical Aspects of
General Relativity, Oberwolfach Reports, vol. 37, p. 2308. (2012). doi:10.4171/OWR/2012/
37
Inflation and Birth of Cosmological
Perturbations
Misao Sasaki
Abstract We review recent developments in the theory of inflation and cosmological

perturbations produced from inflation. After a brief introduction of the standard,
single-field slow-roll inflation, and the curvature and tensor perturbations produced
from it, we discuss possible sources of nonlinear, non-Gaussian perturbations in
other models of inflation. Then we describe the so-called δ N formalism, which is
a powerful tool for evaluating nonlinear curvature perturbations on super Hubble
scales.
1 Introduction
One of the most successful applications of the theory of general relativity is cos-
mology. Over the past half century the big-bang theory of the universe, that the
universe was born in an extremely hot and dense state, expanded explosively and
cooled down to the present state, was observationally tested from various aspects
and it is now firmly established. According to the big-bang theory, our universe is
about 14 Giga years old, and the universe was radiation-dominated in the beginning.
It became matter-dominated when the universe was about 100,000 years old, which
happens to be about the same time when the photons decoupled from baryons, and
started to travel freely until today, which are observed as the cosmic microwave
background (CMB) radiation. The epoch when the CMB photons were scattered last
before they reach us forms a 3-dimensional hypersurface, and it is called the last
scattering surface (LSS).
Despite its tremendous success, there are still a couple of very basic problems that
the big-bang theory cannot explain. One of them is the horizon problem or perhaps
better to be called the causality problem, and the other the flatness problem or the
entropy problem.
M. Sasaki (B)
Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan
e-mail: misao@yukawa.kyoto-u.ac.jp
306 M. Sasaki
1.1 Horizon Problem
Let us first consider the horizon problem. The big-bang theory assumes an
homogeneous and isotropic universe on large scales. So the metric is assumed to
be in the form
ds 2 = −dt 2 + a 2 (t)dσ(3)
2
, (1)
2 is the 3-metric of a constant curvature space with K being the curvature,

where dσ(3)
(3) R i j j
= K (δki δm − δm
j
i δ ). A coordinate system that spans dσ 2 is said to be
km k (3)
comoving because an observer staying at a fixed point on the 3-space is comoving
with the expansion of the universe. In this spacetime, the time-time component of
the Einstein equations, the Friedmann equation, is
ρ K ȧ
H2 = 2
− 2; H≈ , (2)
3M pl a a
where M 2pl = (8πG)−1 in the units = c = 1, and the trace of the space-space
components of the Einstein equations gives
ä ρ + 3P
=− , (3)
a 3M 2pl
where ρ is the energy density and P is the pressure in the universe. This latter
equation shows that the expansion of the universe is always decelerating as long
as ρ + 3P > 0, which holds for both radiation P = ρ/3 and matter P = 0. For
simplicity, if we assume a simple equation of state P/ρ = w = constant and K = 0
(which should be a good approximation in the early universe when w = 1/3 since
ρ ⊗ a −3(1+w) = a −4 ), one finds
2 1
a ⊗ tn; n = < 1 for w > − . (4)
3(1 + w) 3
This result may be regarded as a consequence of the attractive nature of the gravita-
tional force.
Now we introduce the conformal time dη = dt/a(t), and rewrite the metric as
ds 2 = a 2 (η)d ŝ 2 ; d ŝ 2 = −dη 2 + dσ(3)

2
. (5)
Since the conformal transformation of the metric does not change the causal structure,
the static metric d ŝ 2 perfectly describes the causal structure of the universe. If the
range of η were infinite to the past, there would be no horizon problem. The problem
is that the conformal time is finite in the past if w > −1/3 or ρ + 3P > 0, because
Inflation and Birth of Cosmological Perturbations 307
now
last scattering
surface
x
Fig. 1 Horizon problem. The conformal time of the last scattering surface η L SS from η = 0 is
about 1/30 of that of today η0
t
dt ◦ t dt ◦ 2
η= ⊗ ; n= . (6)
a(t ◦ ) 0 t ◦n 3(1 + w)
0
This implies that the size of lightcone emanating from a point at the beginning of
the universe when η = 0 will cover only a finite fraction of spacetime. Since the
comoving distance traveled by light is equal to the corresponding conformal time
interval, the comoving radius of the causally connected region on the LSS is equal
to its conformal time η L SS . From the fact that the LSS is located at redshift z √ 103
and the universe is approximately matter-dominated since then, one finds that this
region will cover only a tiny fraction (about 10−3 sr) of the sky. This is the horizon
problem (see Fig. 1).
The solution is clear: The horizon problem disappears if the conformal time is
either infinite in the past or the beginning of the universe η = 0 is extended sufficiently
back in time to cover the whole visible universe. Since the comoving radius of the
visible universe on the LSS is η0 − η L SS , where η0 is the conformal time today,
the problem is solved if η L SS > η0 − η L SS . In Einstein gravity, this means that
the equation of state must be w < −1/3 or the expansion of the universe must be
accelerating (ä > 0) for a sufficient lapse of time in the very early universe.
Here we should note that solving the horizon problem does not mean explaining
the homogeneity and isotropy of the universe. As it is clear from the above argument,
we had to assume the homogeneity and isotropy of the universe to pose the horizon
problem. This point is very often misunderstood in the literature.
1.2 Flatness Problem

Again we assume a spatially homogeneous and isotropic universe, Eq. (1). The
Friedmann equation (2) tells us that the curvature term K /a 2 is completely negligible
in the early universe when ρ ⊗ a −4 . Conversely, if the curvature term was of the
same order of magnitude as the density at an epoch in the early universe, the universe
308 M. Sasaki
must have either collapsed (if K > 0) or become completely empty (if K < 0) by
now.
Alternatively, since the energy density is dominated by radiation in the early
universe and so is the entropy of the universe, the problem may be rephrased as the
existence of huge entropy within the curvature radius of the universe,
3 3
a a0
S = T3 ∗ ⇔ T03 ∗ > T03 H03 ⇔ 1087 , (7)
|K | |K |
where T0 ⇔ 2.7 K is the CMB temperature today [1] and H0 ⇔ 72 km/s/Mpc is the
Hubble constant [2]. Hence the flatness problem may be called the entropy problem.
It is then apparent that the solution to the flatness problem needs huge entropy
production at a sufficiently early stage of the universe.
1.3 Inflation as a Solution to Horizon and Flatness Problems
A simple and perhaps the best solution to the horizon and flatness problems is given
by the inflationary universe [3, 4]. Let us assume that the universe was dominated
by a spatially homogeneous scalar field. For a minimally coupled canonical scalar
field φ, we have
1 2 1 2
ρ= φ̇ + V (φ), P= φ̇ − V (φ), (8)
2 2

so ρ+3P = 2 φ̇2 −V (φ) . Hence if φ̇2 < V (φ), we may have accelerated expansion.
In particular, if the energy density is dominated by the potential energy, φ̇2 ≤ V (φ),
the motion of the scalar field can be ignored within a few expansion times √ H −1 ,
and the universe expands almost exponentially,
ρ
H2 ⇔ ⇔ constant. (9)
3M 2pl
The curvature term K /a 2 becomes completely negligible.

Thus if the universe is dominated by the potential energy, or the vacuum energy,
and the potential energy is converted to radiation after a sufficient lapse of time of
such a stage, a huge entropy is produced and the horizon and flatness problems are
solved simultaneously.
2 Slow-Roll Inflation and Vacuum Fluctuations
There have been a number of proposals for inflationary models. Among others, a
simplest class of models, and which explains the observational data almost perfectly,
is the slow-roll inflation [5–7]. The field equation for φ and the Friedmann equation
are

◦ 1 1 2
φ̈ + 3H φ̇ + V (φ) = 0, H = 2
φ̇ + V (φ) , (10)
3M 2pl 2
where we have justifiably neglected the curvature term.

The standard slow-roll condition consists of two assumptions. One is that φ̈ is
negligible compared to 3H φ̇ in the field equation, that is, the equation of motion is
friction-dominated. The other is that the kinetic term φ̇2 /2 is negligible compared to
the potential term V in the energy density. Under this condition we have
V ◦ (φ) V
φ̇ = − ; H2 = . (11)
3H 3M 2pl
Then the potential energy dominance implies
3 2
Ḣ φ̇ 3φ̇2 M 2pl V ◦2
ε≈− 2 = 2 ⇔ ⇔ ≈ εV ≤ 1, (12)
H 1 2 2V 2 V2
φ̇ + V
2
that is, the universe is expanding almost exponentially, and the friction-dominated
equation of motion |φ̈/(3H φ̇)| ≤ 1 implies
V ◦◦ 2 V
◦◦
⇔ M pl ≈ ηV ≤ 1. (13)
3H 2 V
The single-field slow-roll inflation satisfies these conditions.
The important property of slow-roll inflation is that Eq. (11) is completely inte-
grable since H is a function of φ. In particular, there is one-to-one correspondence
between φ and t. So instead of the cosmic time t we may measure the time in terms
of the value of the scalar field.
Here we introduce a quantity which plays a very important role in the dynamics
of slow-roll inflation, namely the number of e-folds counted backward in time, say
from the end of inflation to an epoch during inflation,
tend
a(tend )
= exp[N (t ≡ tend )] ≡ N = N (φ) = H dt. (14)
a(t)
t (φ)
310 M. Sasaki
Fig. 2 The Hubble radius L = H −1 and the length scale L = a/k of a comoving wavenumber k
in the inflationary cosmology, and the definition of the number of e-folds N (φ)
Its important property is that by definition it does not depend on how and when the
inflation began. As shown in Fig. 2, N is uniquely determined in terms of the value of
the scalar field (up to a constant which depends on the choice of an epoch from which
N is computed), and one can associate N with the time at which a given comoving
wavenumber k crossed the Hubble radius, k = a H , at which the value of the scalar
field was φk ; N = N (φk ). As we shall see below, this turns out to be an essential
quantity for the evaluation of the curvature perturbation from inflation.
2.1 Curvature Perturbation

Let us now consider the curvature perturbation produced from inflation. It arises from
the quantum vacuum fluctuations of the inflaton field φ. Since a rigorous derivation
would take too much space, here we give an intuitive, rather hand-waving derivation.
We caution that it could well lead to an incorrect result if used blindly.
The vacuum fluctuations of the inflaton field with a comoving wavenumber k
is given simply by its positive frequency function, ϕk . Because of the condition
V ◦◦ /H 2 ≤ 1, on scales k/a ∇ H , the inflaton field fluctuation behaves like a
minimally coupled massless scalar. Hence we have
1 k
|⇒δφ|k∈|2 = |ϕk |2 , ϕk √ ∗ e−iωk t ; ωk = ∇ H. (15)
a 3/2 2ωk a
As the universe expands the physical wavenumber decreases exponentially and be-
comes smaller than the Hubble parameter, k/a < H , or the physical wavelength
exceed the Hubble radius. Then the oscillations of ϕk are frozen. This could be re-
garded as “classicalization” of the quantum fluctuations. Note that this is merely an
interpretation. In a more rigorous sense, freezing of the mode function is a process
toward infinite squeezing of the vacuum state.
Setting a = k/H in Eq. (15) gives
H k
ϕk √ ∗ ; ≤ H. (16)
2k 3 a
Therefore the mean square amplitude in unit logarithmic interval of k is

2
4πk 3 H
⇒δφ ∈k ≈
2
|ϕk |2 ⇔ . (17)
(2π)3 2π k/a=H
Inclusion of the non-trivial evolution of the background spacetime and the coupling
of the scalar field fluctuation with the metric fluctuation do not change the above
estimate if we interpret δφ in the above as those evaluated on the flat slicing, that is,
on hypersurfaces on which the spatial scalar curvature remains unperturbed.
It is known that the curvature perturbation on the comoving hypersurface Rc is
conserved if the perturbation is adiabatic [8]. The comoving hypersurface is defined
as a surface of uniform φ. Then the gauge transformation from the flat slicing to the
comoving slicing gives the relation between Rc and δφ,
H
Rc = − δφ. (18)
φ̇
Since this is conserved for k/a < H , the spectrum of the comoving curvature
perturbation in unit logarithmic interval of k is given by
2
4πk 3 H2
PR (k) ≈ |Rk |2 ⇔ . (19)
(2π)3 2π φ̇ k/a=H
A rigorous, first-principle derivation of the above result was first done in [9, 10].
The important relation of the above result with the number of e-folds was first
pointed out in [11]: If we rewrite Eq. (14) as
tend φ
end
H
N= H dt = dφ, (20)
φ̇
t φ
we find

∂N H
δ N (φk ) = δφ = − δφ = Rc , (21)
∂φ k/a=H φ̇ k/a=H
provided that we identify δφ with the scalar field fluctuation evaluated on the flat
hypersurface. This is called the δ N formula.
312 M. Sasaki
The δ N formula implies that we only need the knowledge of the background
evolution to obtain the power spectrum of the comoving curvature perturbation,
once we know the amplitude of the quantum fluctuations of the scalar field at the
horizon crossing (i.e. when k/a = H ). It is quite generally given by H/(2π) in
slow-roll inflation. With careful geometrical considerations, the δ N formula can be
extended to general multi-field inflation [12],
2
H ∂N ∂N
PR (k) = ||⊥ N ||2 ; ||⊥ N ||2 ≈ G ab (φ) , (22)
2π ∂φa ∂φb
where G ab is the field space metric and it is assumed that the vacuum expectation
values are given by
2
H
⇒δφ δφ ∈ = G
a b ab
. (23)
2π
The nonlinear generalization of the δ N formalism will be discussed in Sect. 4.
2.2 Tensor Perturbation
There are not only vacuum fluctuations of the inflaton field but also those of the
transverse-traceless part of the metric, ∂ i h iTj T = δ i j h iTj T = 0, that is, the tensor
perturbation or gravitational wave degrees of freedom. If we construct the second-
order action for h iTj T , we find
M 2pl ∗
S√ d 4 x −g(ḣ iTj T )2 + · · · . (24)
8
To quantize h iTj T it is convenient to normalize the kinetic term to the canonical form.
This gives

1 ∗ M pl T T
S√ d 4 x −g(φ̇i j )2 + · · · ; φi j ≈ h . (25)
2 2 ij
If one writes down the field equation for φi j , one finds its mode function φk obeys
exactly the same equation as the one for a minimally coupled massless scalar field,
k2
φ̈k + 3H φ̇k + φk = 0. (26)
a2
Since there are two independent degrees of freedom in φi j , the power spectrum of
the tensor perturbation h iTj T is obtained as
4 4πk 3 8H 2
PT (k) = ×2× |φ k | 2
= . (27)
M pl (2π)3 (2π)2 M 2pl
Taking the ratio of the tensor spectrum to the curvature perturbation spectrum, we
find [12]
PT Ḣ
r≈ ⇐ 8|n T | = −2 2 , (28)
PR H
where n T is the tensor spectral index, n T = d ln PT (k)/d ln k, and the equality holds
for the case of single-field slow-roll inflation. This is a consistency relation in general
slow-roll inflation. As a prototype example, if we consider chaotic inflation [7], we
expect to have r √ 0.1.
The important point to be kept in mind is that the existence of the vacuum fluctu-
ations of the tensor part of the metric is a proof of the existence of quantum gravity.
These fluctuations exist in any theory of gravity that respects general covariance,
apart from possible inessential modifications of the spectrum. Thus a clear detection
of the tensor spectrum will be a confirmation of not only the inflationary universe
but also of quantum gravity.
3 Origin of Non-Gaussianity
The standard, single-field slow-roll inflation predicts that the curvature perturbation
is a Gaussian random field and it has an almost scale-invariant spectrum. This seems
to fit the current observational data quite well [13], it is quite possible that the actual
model turns out to be non-standard. Maybe it is multi-field, maybe non-slow-roll
and/or non-canonical. In such a case, the curvature perturbation may become non-
Gaussian. Search for possible non-Gaussian signatures in the primordial curvature
perturbation has become one of the important directions in observations in recent
years [14, 15].
Here we consider possible origins of non-Gaussianity in the curvature perturba-
tion. Essentially one can classify the origins into three categories: (1) Self-interactions
of the inflaton field and/or non-trivial vacua, (2) multi-field dynamics, and (3) non-
linearity in gravity.
The non-Gaussianities of the first category are generated on subhorizon scales
during inflation, hence they are of quantum field theoretical origin. Those of the
second category are usually generated on superhorizon scales either during or after
inflation, and they are due to nonlinear coupling of the scalar field to gravity. Since
they are generated on superhorizon scales, they are of classical origin. Finally those
of the third category are due to nonlinear dynamics in general relativity. Hence they
are generated after the scale of interest re-enters the Hubble horizon. Since the last
category is not really primordial in nature, let us focus on the first two categories.
314 M. Sasaki
3.1 Non-Gaussianity From Self-Interaction/Non-Trivial Vacuum
It is known that conventional self-interactions by the potential are ineffective [16].

This can be seen by considering chaotic inflation, for example. In the simplest case
of a quadratic potential, V = m 2 φ2 /2, the inflaton is actually a free field apart from
the interaction through gravitation perturbations. But the gravitational interaction is
Planck-suppressed, i.e., it is always suppressed by a factor O(M −2 pl ). In the case of a
quartic potential, V = λφ , it is known that λ should be extremely small λ √ 10−15
4
in order for it to be consistent with observation.

Thus some kind of unconventional self-interaction is necessary. A popular ex-
ample is the case of a scalar field with a non-canonical kinetic term such as DBI
inflation [17]. In this case the kinetic term takes the form,
K √ f −1 (φ) 1 − f (φ)φ̇2 ≈ f −1 γ −1 . (29)
If we expand this perturbatively,
K = K 0 + δ1 K + δ2 K + δ3 K + · · · , (30)
we will find
δ2 K ⊗ γ 3 , δ3 K ⊗ γ 3+2 , (31)
since δγ = γ 3 δ X where X ≈ f φ̇2 /2. If we regard the third order part as the interac-
tion, the above implies that the scalar field fluctuation will be expressed qualitatively
as
δφ √ δφ0 + γ 2 δφ20 + · · · , (32)
where δφ0 is the free, Gaussian fluctuation. Thus the non-Gaussianity in δφ may
become large if γ, which mimics the Lorentz factor, is large [18].
A non-trivial vacuum state is another source of non-Gaussianity. If the universe
were a pure de Sitter spacetime, gravitational interaction would be totally negligible
in vacuum, except for the effect due to graviton (tensor mode) loops. This may be
regarded as due to the maximally symmetric nature of the de Sitter space, S O(4, 1),
which has the same number of degrees of symmetry as the Poincare (Minkowski)
symmetry. In slow-roll inflation, the de Sitter symmetry is slightly broken. Neverthe-
less the effect induced by this symmetry breaking is small because it is suppressed
by the slow-roll parameter ε = − Ḣ /H 2 .
However, if the vacuum state does not respect the de Sitter symmetry, there can be
a large non-Gaussianity. Such a deviation from the quasi-de Sitter vacuum, usually
called the Bunch-Davies vacuum, may occur in various situations, studied e.g. in
[19, 20].
Fig. 3 An illustration of the

energy density configuration
in the multi-field case. The
density of the A-matter/field
ρ A may vary nonlinearly
without significantly affecting
the total energy density
3.2 Non-Gaussianity From Multi-Field Dynamics
Non-Gaussianity may appear if the energy momentum tensor depends nonlinearly

on the scalar field even if the fluctuation of the scalar field itself is Gaussian. This
effect is generally important when the fluctuations are on superhorizon scales, i.e., the
characteristic wavelength is larger than the Hubble radius. It is small in single-field
slow-roll models because the linear approximation is valid to high accuracy [21],
generically suppressed by the slow-roll parameter ηV defined in Eq. (13).
For multi-field models, however, the contribution to the energy momentum tensor
from some of the fields can be highly nonlinear as depicted in Fig. 3. The important
property of non-Gaussianity in this case is that it is always of the spatially local
type. Namely, to second order in nonlinearity, the curvature perturbation will take
the form [22],
3 local 2
Rc (x) = Rc,0 (x) + f Rc,0 (x), (33)
5 NL
where Rc,0 is the Gaussian random field and f Nlocal L is a constant representing the
amplitude of non-Gaussianity. The factor 3/5 in front of f Nlocal
L is due to a historical
reason. The reason why it is of local type is simply causality: No information can
propagate over a length scale greater than the Hubble horizon scale.
Observationally, this type of non-Gaussianity can be tested by using the so-called
squeezed type templates where one of the wavenumbers, say k1 , in the bispectrum
B(k1 , k2 , k3 ) is much smaller than the other two, k1 ≤ k2 ⇔ k3 [14], and there are
a few observational indications that f Nlocal
L is actually non-vanishing. For example,
the WMAP 7 year data analysis gave a one-sigma bound 11 < f Nlocal L < 53 (68 %
CL) [13].
4 δ N Formalism
As mentioned in Sect. 2, the δ N formalism is a powerful tool to evaluate the comov-
ing curvature perturbation on superhorizon scales. It then turned out that it can be
easily extended to the evaluation of nonlinear, non-Gaussian curvature perturbations
[23, 24]. Let us recapitulate its definition and properties:
316 M. Sasaki
(1) δ N is the perturbation in the number of e-folds counted backward in time from
a fixed final time, say t = t f , to some initial time t = ti .
(2) The final time t f should be chosen such that the evolution of the universe has
become unique by that time, i.e., the universe has reached the adiabatic limit.
Then the hypersurface t = t f should be identified with a comoving (or uniform
density) slice, and the initial hypersurface t = ti should be identified with a flat
slice.
(3) δ N is equal to the conserved (nonlinear) comoving curvature perturbation on
superhorizon scales at t > t f .
(4) By definition, it is nonlocal in time. However, because of its purely geometrical
definition, it is valid independent of which theory of gravity one considers,
provided that the adiabatic limit is reached by t = t f .
There are various kinds of sources that generate δ N . They may be classified into
three types, as depicted in Fig. 4. The left one describes a perturbation along the
evolutionary trajectory of the universe. This case is the same as that of single-field
slow-roll inflation, in which the comoving curvature perturbation is conserved all
the way until it re-enters the horizon. The middle one is the case when a small
difference in the initial data develops into a substantial difference in δ N . Typically
this is realized when there is some instability orthogonal to the trajectory, like the
case when the scalar field moves along a ridge. This type of sources of δ N usually
induces a feature in the spectrum and/or bispectrum of the curvature perturbation.
The right one represents the case when the perturbation orthogonal to the trajectory
does not contribute to the curvature perturbation until or after the end of inflation, but
δ N is generated due to a sudden transition that brings the universe into an adiabatic
stage. Typical examples are curvaton models [25–27] and multi-brid inflation models
[28, 29].
Here, for the sake of completeness, let us present the precise definition of the
nonlinear δ N formula. See Fig. 5. It is based on the leading order approximation in
the spatial gradient expansion or the separate universe approach [23], where spatial
derivatives are assumed to be negligible in comparison with time derivatives. At lead-
ing order of the spatial gradient expansion, if we express the spatial volume element
as (3) γ = a 3 (t) exp[3R(t, x)] where a(t) is the scale factor of a fiducial homoge-
neous and isotropic universe, we easily find that the perturbation in the number of
e-folds along a comoving trajectory between two hypersurfaces t = t1 and t = t2 is
given by
δ N (t2 , t1 ; x i ) = R(t2 , x i ) − R(t1 , x i ), (34)
where x i are the comoving coordinates. Here we note that this is purely a geometrical
relation. It has nothing to do with any equations of motion.
First we fix the final hypersurface t = t2 . It should be taken at the stage when the
evolution of the universe has become unique. That is, there exists no isocurvature
perturbation any longer that could develop into an adiabatic perturbation at later
epochs. Thus the comoving curvature perturbation is conserved at t > t2 . In the
Fig. 4 Three different types of δ N . The field space (φ1 , φ2 ) in the figure represents the degrees
of freedom in the initial condition of the universe. The adiabatic limit is defined to be the stage by
which all the trajectories converge to a unique one
Fig. 5 Definition of nonlinear δ N . It is defined as the perturbation in the number of e-folds from
an initial flat slice to a final comoving slice
context of the concordance εCDM model of the universe, this corresponds to the
final radiation-dominated stage of the universe.
Next we choose the initial slice t = t1 . It should be chosen to be flat. Here ‘flat’
means that the perturbation in the spatial volume element vanishes. Namely, the flat
slice is defined as a hypersurface on which R = 0. We note that despite its name,
the scalar curvature vanishes only in the linear theory limit: It is non-vanishing in
general in the nonlinear case.
Applying the above choice of the initial and final hypersurfaces to Eq. (34), it is
trivial to see that we have
δ N (t2 , t1 ; x i ) = Rc (t2 , x i ). (35)

318 M. Sasaki
Now by assumption Rc is conserved at t > t2 . So it is the quantity we want to

evaluate. This completes the derivation of the nonlinear δ N formula.
As mentioned above, since Eq. (34) is a pure geometrical relation, so is the non-
linear δ N formula (35). This is the reason why it can be applied to any theory of
gravity as long as it is a geometrical (i.e. general covariant) theory.
Of course, the above definition tells us nothing about how to evaluate it in practice.
In this respect, we have a very fortunate situation in the case of inflationary cosmol-
ogy. It is the fact that the evaluation of the quantum fluctuations of the inflaton field,
whether it is single- or multi-component, can be most easily done in a gauge in which
the time slicing is chosen to be flat [12]. Thus we can choose the initial slice to be
an epoch when the scale of our interest has just exited the horizon during inflation.
Let the fluctuations of a multi-component scalar field on the flat slice at t = t1 be
δφa . Then assuming that the values of the scalar field determine the evolution of the
universe completely, which is the case for slow-roll inflation, the nonlinear δ N can
be simply evaluated as
δ N = N (φa + δφa ) − N (φa ), (36)
where N (φa ) is the e-folding number of the fiducial background. In particular, to

second order in δφa , we obtain
∂ N a 1 ∂2 N
Rc = δ N = δφ + δφa δφb + · · · . (37)
∂φa 2 ∂φa ∂φb
Comparing this with Eq. (33), we see that the curvature perturbation takes a bit more
complicated form that the simplest form. Nevertheless if we consider the bispec-
trum, i.e., the Fourier component of the three-point function ⇒Rc (x1 )Rc (x2 )Rc (x3 )∈,
we find there is a quantity that exactly corresponds to f Nlocal
L defined in Eq. (33).
Namely [24],
3 local G ab G cd Na Nbc Nd ∂N ∂2 N
fN L = 2 ; Na ≈ , Nab ≈ . (38)
5 2 ||⊥ N ||2 ∂φa ∂φa ∂φb
Before concluding this section, we mention the fact that the δ N formalism does
not require the scalar field fluctuations to be Gaussian. In fact, except for the last
equation in the above, Eq. (38) which assumes the Gaussianity of δφa , the general δ N
formula (36) or its second order version (37) can be used for non-Gaussian δφa [30].
Such a case may happen, for example, in multi-field DBI inflation.
5 Summary
It has been about 30 years since the inflationary universe was first proposed, and
there is increasing observational evidence that inflation did take place in the very
early universe. Among others, the measured CMB temperature anisotropy is fully
consistent with the predictions of inflation that the primordial curvature perturbation
spectrum is almost scale-invariant and it is statistically Gaussian.
Inflation also predicts a scale-invariant tensor spectrum, and if the energy scale
of inflation is high enough as in the case of chaotic inflation, the tensor-scalar ratio r
can be as large as 0.1. If this is the case, the tensor perturbation will be detected in the
near future, and it will confirm not only the inflationary universe but also quantum
gravity.
Even if the tensor perturbation will not be detected, there may be other interesting
signatures of inflation. Non-Gaussianity from inflation is attracting attention as one
of those signatures that can distinguish or constrain models of inflation significantly.
We discussed that the origins of primordial non-Gaussianities may be classified
into three categories, according to different length scales on which different mecha-
nisms are effective:
(1) Quantum theoretical origin on subhorizon scales during inflation.
(2) Classical nonlinear scalar field dynamics on superhorizon scales during or after
inflation.
(3) Nonlinear gravitational dynamics after the horizon re-entry.
In particular we argued that non-Gaussianities in the second case are always of spa-
tially local type. We then mentioned that there are three different kinds of situations
in which such local non-Gaussianities can be generated, and described in some detail
a very efficient method to compute them, namely, the δ N formalism.
Apparently identifying properties of primordial non-Gaussianities in the observa-
tional data is extremely important for understanding the physics of the early universe.
Here we mentioned only the bispectrum or the 3-point function. But if it is detected,
higher order n-point functions may become important as a model discriminator. Other
types of non-Gaussianity discriminators may also become necessary.
What is important is that we are now beginning to test observationally the physics
of the very early universe, the physics at an energy scale closer to the Planck scale,
at a scale that can never be attained in high energy accelerator experiments.
Cosmology has become not only a precision science, but now it constitutes a
truly indispensable part of fundamental physics. General relativity is the backbone
of cosmology. I wonder what Einstein would say if he were here in this very exciting
era—100 years after he visited Prague.
Acknowledgments I am very grateful to the organizers of the conference, “Relativity and Grav-
itation, 100 years after Einstein in Prague”, particularly to Jiří Bičák, who kindly invited me to
this meeting, and who accorded me a warm hospitality. I am also grateful to Laila Alabidi for
careful reading of the manuscript and very useful comments. This work was supported in part by
320 M. Sasaki
JSPS Grant-in-Aid for Scientific Research (A) No. 21244033, and by Monbukagaku-sho Grant-in-
Aid for the Global COE programs, “The Next Generation of Physics, Spun from Universality and
Emergence” at Kyoto University.
References
1. Mather, J.C., Cheng, E.S., Cottingham, D.A., et al.: Measurement of the cosmic microwave
background spectrum by the COBE FIRAS instrument. Astrophys. J. 420, 439 (1994). doi:10.
1086/173574
2. Freedman, W.L., Madore, B.F.: The Hubble constant. Annu. Rev. Astron. Astrophys. 48, 673
(2010). doi:10.1146/annurev-astro-082708-101829
3. Sato, K.: First-order phase transition of a vacuum and the expansion of the universe. Mon. Not.
R. Astron. Soc. 195, 467 (1981)
4. Guth, A.H.: Inflationary universe: a possible solution to the horizon and flatness problems.
5. Linde, A.D.: A new inflationary universe scenario: a possible solution of the horizon, flatness,
homogeneity, isotropy and primordial monopole problems. Phys. Lett. B 108, 389 (1982).
doi:10.1016/0370-2693(82)91219-9
6. Albrecht, A., Steinhardt, P.J.: Cosmology for grand unified theories with radiatively induced
symmetry breaking. Phys. Rev. Lett. 48, 1220 (1982). doi:10.1103/PhysRevLett.48.1220
7. Linde, A.D.: Chaotic inflation. Phys. Lett. B 129, 177 (1983). doi:10.1016/0370-
2693(83)90837-7
8. Kodama, H., Sasaki, M.: Cosmological perturbation theory. Prog. Theor. Phys. Suppl. 78, 1
(1984). doi:10.1143/PTPS.78.1
9. Mukhanov, V.F.: Gravitational instability of the universe filled with a scalar field. JETP Lett.
41, 493 (1985)
10. Sasaki, M.: Large scale quantum fluctuations in the inflationary universe. Prog. Theor. Phys.
76, 1036 (1986). doi:10.1143/PTP.76.1036
11. Starobinskii, A.A.: Multicomponent de Sitter (inflationary) stages and the generation of per-
turbations. JETP Lett. 42, 152 (1985)
12. Sasaki, M., Stewart, E.D.: A general analytic formula for the spectral index of the density
perturbations produced during inflation. Prog. Theor. Phys. 95, 71 (1996). doi:10.1143/PTP.
95.71
13. Komatsu, E., Smith, K.M., Dunkley, J., et al.: Seven-year Wilkinson microwave anisotropy
probe (WMAP) observations: cosmological interpretation. Astrophys. J. Suppl. Ser. 192, 18
(2011). doi:10.1088/0067-0049/192/2/18
14. Komatsu, E., Afshordi, N., Bartolo, N et al.: Non-Gaussianity as a probe of the physics of the
primordial universe and the astrophysics of the low redshift universe. In: Proceedings of Astro
2010: The Astronomy and Astrophysics Decadal Survey—Science White Papers, National
Academy of Sciences (2009). http://sites.nationalacademies.org/bpa/bpa_050603
15. Sasaki, M., Wands, D.: Focus section editorial: Non-linear and non-Gaussian cosmological
perturbations, Class. Quantum Gravity 27(12) 120301 (2010). doi:10.1088/0264-9381/27/12/
120301
16. Maldacena, J.M.: Non-Gaussian features of primordial fluctuations in single field inflationary
models. J. High Energy Phys. 2003(05), 013 (2003). doi:10.1088/1126-6708/2003/05/013
17. Alishahiha, M., Silverstein, E., Tong, D.: DBI in the sky: non-Gaussianity from inflation with
a speed limit. Phys. Rev. D 70, 123505 (2004). doi:10.1103/PhysRevD.70.123505
18. Mizuno, S., Arroja, F., Koyama, K., Tanaka, T.: Lorentz boost and non-Gaussianity in multifield
DBI inflation. Phys. Rev. D 80, 023530 (2009). doi:10.1103/PhysRevD.80.023530
19. Chen, X., Easther, R., Lim, E.A.: Generation and characterization of large non-Gaussianities
in single field inflation. J. Cosmol. Astropart. Phys. 2008(04), 010 (2008). doi:10.1088/1475-
7516/2008/04/010
20. Flauger, R., McAllister, L., Pajer, E., Westphal, A., Xu, G.: Oscillations in the CMB from axion
monodromy inflation. J. Cosmol. Astropart. Phys. 2010(06), 009 (2010). doi:10.1088/1475-
7516/2010/06/009
21. Salopek, D.S., Bond, J.R.: Nonlinear evolution of long-wavelength metric fluctuations in in-
flationary models. Phys. Rev. D 42, 3936 (1990). doi:10.1103/PhysRevD.42.3936
22. Komatsu, E., Spergel, D.N.: Acoustic signatures in the primary microwave background bis-
pectrum. Phys. Rev. D 63, 063002 (2001). doi:10.1103/PhysRevD.63.063002
23. Lyth, D.H., Malik, K.A., Sasaki, M.: A general proof of the conservation of the curvature
perturbation. J. Cosmol. Astropart. Phys. 2005(05), 004 (2005). doi:10.1088/1475-7516/2005/
05/004
24. Lyth, D.H., Rodrǵuez, Y.: Inflationary prediction for primordial non-Gaussianity. Phys. Rev.
25. Lyth, D.H., Wands, D.: Generating the curvature perturbation without an inflaton. Phys. Lett.
B 524, 5 (2002). doi:10.1016/S0370-2693(01)01366-1
26. Moroi, T., Takahashi, T.: Effects of cosmological moduli fields on cosmic microwave back-
ground. Phys. Lett. B 522(215), 2002 (2001). doi:10.1016/S0370-2693(01)01295-3.Erratum:
ibid.539,303
27. Sasaki, M., Valiviita, J., Wands, D.: Non-Gaussianity of the primordial perturbation in the
curvaton model. Phys. Rev. D 74, 103003 (2006). doi:10.1103/PhysRevD.74.103003
28. Sasaki, M.: Multi-brid inflation and non-Gaussianity. Prog. Theor. Phys. 120, 159 (2008).
doi:10.1143/PTP.120.159
29. Naruko, A., Sasaki, M.: Large non-Gaussianity from multi-brid inflation. Prog. Theor. Phys.
121, 193 (2009). doi:10.1143/PTP.121.193
30. Byrnes, C.T., Koyama, K., Sasaki, M., Wands, D.: Diagrammatic approach to non-Gaussianity
from inflation. J. Cosmol. Astropart. Phys. 2007(11), 027 (2007). doi:10.1088/1475-7516/
2007/11/027
Loop Quantum Gravity and the Planck
Regime of Cosmology
Abhay Ashtekar
Abstract The very early universe provides the best arena we currently have to test
quantum gravity theories. The success of the inflationary paradigm in accounting for
the observed inhomogeneities in the cosmic microwave background already illus-
trates this point to a certain extent because the paradigm is based on quantum field
theory on the curved cosmological space-times. However, this analysis excludes the
Planck era because the background space-time satisfies Einstein’s equations all the
way back to the big bang singularity. Using techniques from loop quantum gravity,
the paradigm has now been extended to a self-consistent theory from the Planck
regime to the onset of inflation, covering some 11 orders of magnitude in curvature.
In addition, for a narrow window of initial conditions, there are departures from
the standard paradigm, with novel effects, such as a modification of the consistency
relation involving the scalar and tensor power spectra and a new source for non-
Gaussianities. The genesis of the large scale structure of the universe can be traced
back to quantum gravity fluctuations in the Planck regime. This report provides a
bird’s eye view of these developments for the general relativity community.
1 Introduction
In this conference, Professor Bičák and others described the ideas that Einstein
developed in Prague during 1911–12. From then until 1915 he worked largely by
himself on the grand problem of extending the reach of special relativity to encompass
gravity. Finally, in November 1915, he provided us with the finished theory. For
almost a century, the relativity community has been engaged in understanding the
astonishingly rich physics the theory contains, testing it ever more accurately, and
A. Ashtekar (B)
Institute for Gravitation and the Cosmos and Physics Department, Penn State, University
Park, PA16802, USA
e-mail: Ashtekar@gravity.psu.edu
324 A. Ashtekar
applying it to greater and greater domains of astrophysics and cosmology. The theory
has so many marvelous features. Amazingly, the field equations turned out to provide
an elliptic-hyperbolic system with a well-posed initial value problem. After many
decades, we realized that the total mass of an isolated system is a well defined
geometric invariant and, furthermore, positive if the local energy density of matter is
positive. The theory naturally admits cosmological solutions in which the universe
is expanding, just as the observations tell us. It admits black hole solutions that serve
as engines for the most energetic phenomena seen in the universe. None of these
fascinating features that we now regard as fundamental consequences were part of
Einstein’s motivation during his quest which he described as “one of the most exciting
and exacting times of my life” [1]. He essentially handed to us the finished product
on a platter. We have been engaged in uncovering the numerous hidden treasures
it contains by working out the philosophical, mathematical, physical, astronomical
and cosmological consequences of the new paradigm.
But we know that the theory is incomplete. Indeed, it exhibits its own fundamen-
tal limitations through singularities where space-time ends and general relativistic
physics comes to a halt. We also understand that this occurs because general relativity
ignores quantum physics. Perhaps the most outstanding example is the prediction of
the big bang. If we go back in time, much before we reach the singularity, matter den-
sities exceed the nuclear density, ≈1014 –1015 gms/cc, where we definitely know that
quantum properties of matter dominate. Since gravity couples to matter, the concep-
tual paradigm of general relativity becomes inadequate. If we go further back in time,
general relativity presents us with an epoch in which densities reach ≈1094 gms/cc.
This is the Planck scale and now physics of general relativity becomes inadequate
not only conceptually but also in practice. In this regime we expect gross departures
from Einstein’s theory. Just as it is totally inadequate to use Newtonian mechanics
to explore physics near the horizon of a solar mass black hole, it is incorrect to trust
general relativity once the matter density and space-time curvature enter the Planck
regime. Thus, big bang is a prediction of general relativity in a domain in which it
is simply invalid. Normally physicists do not advertise such predictions of theories.
But unfortunately they often seem to make an exception for the big bang. One hears
statements like ‘the cosmic microwave background (CMB) is a fingerprint of the
big bang’. But in the standard scenario, CMB refers to a time some 380,000 years
after the putative big bang. Existence or even the detailed features of CMB have
no bearing on whether the big bang with infinite matter density and curvature ever
occurred. Indeed, as we will see, loop quantum cosmology (LQC) has no big bang
singularity and yet reproduces these features. What about inflation? In the standard
scenario, it is supposed to have commenced ‘only’ 107 Planck seconds after the big
bang. Does its success not imply that there was a big bang? It does not because the
matter density and curvature at the onset of inflation are only 10−11 –10−12 times the
Planck scale. Indeed, this is why one can use Einstein’s equations and quantum field
theory (QFT) on Friedmann, Lemaître, Robertson, Walker (FLRW) solutions in the
analysis of inflation. Inflationary physics by itself cannot say what really happened in
the Planck regime and, again, as we will see, is compatible with the LQC prediction
that there was no big bang singularity.
Loop Quantum Gravity and the Planck Regime of Cosmology 325
Thus, to know what really happened in the Planck regime and go beyond the
singularities predicted by general relativity, we need a viable quantum theory of
gravity. Since the search for this theory has been ongoing for decades, justifiably,
there is sometimes a sentiment of pessimism in the general relativity circles. In
my view, this is largely because one judges progress using the criterion of general
relativity. In a masterful stroke, Einstein gave us the final theory and we have been
happily engaged in investigating its content. It seems disappointing that this has not
happened with quantum gravity. But progress of physical theories has more often
mimicked the development of quantum theory rather than general relativity. More
than a century has passed since Planck’s discovery that launched the quantum. Yet,
the theory is incomplete. We do not have a satisfactory grasp of the foundational
issues, often called the ‘measurement problem’, nor do we have a single example
of an interacting QFT in 4 dimensions. A far cry from what Einstein offered us in
1915! Yet, no one would deny that quantum theory has been extremely successful;
indeed, much more so than general relativity.
Thus, while it is tempting to wait for another masterful stroke like Einstein’s to
deliver us a finished quantum gravity theory, it is more appropriate to draw lessons
from quantum theory. There, progress occurred by focussing not on the ‘final, fin-
ished’ theory, but on concrete physical problems where quantum effects were im-
portant. It would be more fruitful to follow this path in quantum gravity. Indeed,
even though we are far from a complete theory, advances can occur by focusing on
specific physical problems and challenges.
Over the last several years, research in loop quantum gravity (LQG) has been
driven by this general philosophy. In addition to seeking a completion of the general
program based on connection variables, spin networks and spin foams, more and
more effort is now focused on specific physical problems where quantum gravity
effects are expected to be important. The idea behind this research is to first truncate
general relativity (with matter) to sectors tailored to specific physical problems, and
then pass to quantum theory using the background independent methods based on the
specific quantum geometry that underlies LQG. This strategy of focusing on specific
problems of quantum gravity also distinguishes LQG from string theory in terms of
their main trust in the last few years. In string theory, the focus has shifted to using
the well-understood parts of gravity to explore other areas of physics—use of the
AdS/CFT hypothesis to understand the strong coupling regime of QCD, to gain new
insights into hydrodynamics and tackle the strong coupling problems in mathematical
physics to better understand condensed matter systems such as high temperature
super-conductivity. The LQG community, on the other hand, has continued to tackle
the long standing problems of quantum gravity per se—absence of a space-time in the
background, the problem of time, fate of cosmological singularities in the quantum
theory, quantum geometry of horizons, and derivation of the graviton propagator in
a background independent setting.
The goal of my talk was to report the advances in the cosmology of the very early
universe that have resulted from a continued application of the truncation strategy
in LQG. Of course, both the talk and this report can only provide a bird’s eye view
of these developments. The results I reported are based largely on joint work with
326 A. Ashtekar
Corichi, Pawlowski and Singh [2–6] on the singularity resolution in cosmology; with
David Sloan [7, 8] on effective LQC dynamics tailored to inflation, with Kaminski
and Lewandowski [9] on QFT on quantum space-times; and especially with Ivan
Agullo and William Nelson on extension of the cosmological perturbation theory
to the Planck regime and its application to inflation [10–12]. Therefore, there is a
large overlap with the material covered in these original references. Finally, by now
there are well over a 1000 papers on LQC which include several investigations of
inflationary dynamics. What I can cover constitutes only a very small fraction of
what is known. For reviews on results until about a year ago, see, e.g. [13, 14].
2 Setting the Stage
Perhaps the most significant reason behind the rapid and spectacular success of
quantum theory, especially in its early stage, is the fact that there was already a
significant accumulation of relevant experimental data, and further experiments to
weed out ideas could be performed on an ongoing basis. Unfortunately this is not the
case for quantum gravity simply because theory has raced far ahead of technology.
Indeed, even in the classical regime, we still lack detailed tests of general relativity
in the strong field regime!
Currently, the early universe offers by far the best arena to test various ideas on
quantum gravity. Most scenarios assume that the early universe is well described by a
FLRW solution to Einstein’s equations with suitable matter, together with first order
perturbations. The background is treated classically, as in general relativity, and the
perturbations are described by quantum fields. Thus, the main theoretical ingredient
in the analysis are: cosmological perturbation theory and QFT on FLRW space-
times. It is fair to say that among the current scenarios, the inflationary paradigm has
emerged as the leading candidate. In addition to the common assumption described
above, this scenario posits:
• Sometime in its early history, the universe underwent a phase of rapid expansion.
This was driven by the slow roll of a scalar field in a suitable potential causing the
Hubble parameter to be nearly constant.
• Fourier modes of the quantum fields representing perturbations were initially in a
specific state, called the Bunch-Davies (BD) vacuum, for a certain set of co-moving
wave numbers (ko , 2000ko ) where the physical wave length of the mode ko equals
the radius RLS of the observable universe at the surface of last scattering.1
1 Strictly speaking, the BD vacuum refers to deSitter space; it is the unique ‘regular’ state which is
invariant under the full deSitter isometry group. During slow roll, the background FLRW geometry
is only approximately deSitter whence there is some ambiguity in what one means by the BD
vacuum. One typically assumes that all the relevant modes are in the BD state (tailored to) a few
e-foldings before the mode ko leaves the Hubble horizon. Throughout this report, by BD vacuum I
mean this state.
• Soon after any mode exits the Hubble radius, its quantum fluctuation can be
regarded as a classical perturbation and evolved via linearized Einstein’s
equations.
One then evolves the perturbations from the onset of the slow roll till the end of
inflation using QFT on FLRW space-times and calculates the power spectrum (see,
e.g., [15–19]). When combined with standard techniques from astrophysics to further
evolve the results to the surface of last scattering, one finds that they are in excellent
agreement with the inhomogeneities seen in the CMB. Supercomputer simulations
have shown that these inhomogeneities serve as seeds for the large scale structure
in the universe. Thus, in a precise sense, the origin of the qualitative features of the
observed large scale structure can be traced back to the fluctuations in the quantum
vacuum at the onset of inflation. This is both intriguing and very impressive.
Over the years, the inflationary paradigm has witnessed criticisms from the
relativity community, most eloquently expressed by Roger Penrose (see, e.g., [20]).
However, these criticisms refer to the motivations that were originally used by the
proponents, rather than to the methodology underlying its success in accounting for
the CMB inhomogeneities. There are plenty of examples in fundamental physics
where the original motivations turned out not to be justifiable but the idea was highly
successful. I share the view that, while the basic assumptions, listed above, have
not been justified from first principles, the success of the inflationary paradigm with
CMB measurements is impressive because one ‘gets much more out than what one
puts in’.
In spite of this success, however, the inflationary scenario is conceptually
incomplete in several respects. (For a cosmology perspective on these limitations
see e.g. [21, 22].) In particular, as Borde, Guth and Vilenkin [23] showed, inflation-
ary space-times inherit the big-bang singularity in spite of the fact that the inflation
violates the standard energy conditions used in the original singularity theorems [24].
As we discussed in Sect. 1, this occurs because one continues to use general relativ-
ity even in the Planck regime in which it is simply not applicable. One expects new
physics to play a dominant role in this regime, thereby resolving the singularity and
significantly changing the very early history of the universe. One is therefore led to
ask: Will inflation arise naturally in the resulting deeper theory? Or, more modestly,
can one at least obtain a consistent quantum gravity extension of this scenario?
The open-ended nature of the inflationary paradigm has three facets. First, there
are issues whose origin lies in particle physics. Where does the inflaton come from?
How does potential arise? Is there a single inflaton or many? If many, what are the
interactions between them? Since the required mass of the inflaton is very high,
above 1012 Gev, the fact that we have not seen it at CERN does not mean it cannot
exist. But in the inflationary scenario this is the only matter field in the early universe
and particles of the standard model are supposed to be created during ‘reheating’ at
the end of inflation when the inflaton is expected to roll back and forth around its
minimum. However, how this happens is not at all well-understood. What are the
admissible interactions between the inflaton and the standard model particles which
328 A. Ashtekar
causes this decay? Does the decay produce the correct abundance of the standard
model particles? These questions with origin in particle physics are wide open.
The second issue is the quantum to classical transition referred to in the last
assumption of standard inflation. In practice one calculates the expectation values of
perturbations and the two point function at the end of inflation and assumes one can
replace the actual quantum state of perturbations with a Gaussian statistical distri-
bution of classical perturbations with the mean and variance given by the quantum
expectation value and the 2-point function. As a calculational devise this strategy
works very well. However, what happens physically? While this issue has drawn
attention, we do not yet have a clear consensus on the actual, detailed physics that is
being approximated in the last assumption.
The third set of issues have their origin in quantum gravity. In the standard
inflationary scenario, one specifies initial conditions at the onset of inflation and
then evolves the quantum perturbations. As a practical strategy, something like this
is unavoidable within general relativity. Ideally one would like to specify the ini-
tial conditions at ‘the beginning’, but one simply cannot do this because the big
bang is singular. Furthermore, since the curvature at the onset of inflation is some
10−11 –10−12 times the Planck scale, by starting calculations there, one bypasses the
issue of the correct Planck scale physics. But this is just an astute stopgap measure.
Given any candidate quantum gravity theory, one can and has to ask whether one
can do better. Can one meaningfully specify initial conditions in the Planck regime?
In a viable quantum gravity theory, this should be possible because there would be
no singularity and the Planck scale physics would be well-controlled. If so, in the
systematic evolution from there, does a slow roll phase compatible with the 7 year
WMAP data [25] arise generically or is an enormous fine tuning needed? One could
argue that it is acceptable to use fine tuning because, after all, the initial state is very
spatial. If so, can one provide physical principles that select this special state? In
the standard inflationary scenario, if we evolve the modes of interest back in time,
they become trans-Planckian. Is there a QFT on quantum cosmological space-times
needed to adequately handle physics at that stage? Can one arrive at the BD vacuum
(at the onset of the WMAP slow roll) starting from the initial conditions at the Planck
scale?
In this report, I will not address the first two sets of issues. Rather, the focus will
be on the incompleteness related to the third set, i.e., on quantum gravity. Systematic
advances within LQC over the past six years have provided a viable extension of
the inflationary scenario all the way to the Planck regime. This extension enables
us to answer in detail most of the specific questions posed above. To arrive at a
coherent extension, LQC had to develop a conceptual framework, mathematical
tools and high precision numerical simulations because the issues are so diverse:
The meaning of time in the Planck regime; the nature of quantum geometry in the
cosmological context; QFT on quantum cosmological space-times; renormalization
and regularization of composite operators needed to compute stress energy and back
reaction; and, relation between theory and the WMAP data.
A consistent theoretical framework to deal with cosmological perturbations on
quantum FLRW space-times now exists [11]. Starting with ‘natural’ initial conditions
in the Planck regime, one can evolve the quantum perturbations on quantum FLRW
backgrounds and study in detail the pre-inflationary dynamics [10, 12]. Detailed
numerical simulations have shown that the predictions are in agreement with the
power spectrum and the spectral index reported in the 7 year WMAP data. However,
there is also a small window in the parameter space where the initial state at the onset
of inflation differs sufficiently from the BD vacuum assumed in standard inflation
to give rise to new effects. These are the prototype observable signatures of pre-
inflationary dynamics. In this sense, LQC offers the possibility of extending the reach
of cosmological observations to the deep Planck regime of the early universe.
3 Why Pre-inflationary Dynamics Matters
It is often claimed that pre-inflationary dynamics will not change the observable
predictions of the standard inflationary scenario. Indeed, this belief is invoked to
justify why one starts the analysis just before the onset of the slow roll. The belief
stems from the following argument, sketched in the left panel of Fig. 1. If one evolves
the modes that are seen in the CMB back in time starting from the onset of slow
roll, their physical wave lengths εphy continue to remain within the Hubble radius
1/HGR all the way to the big bang. Therefore, one argues, they would not experience
curvature and their dynamics would be trivial all the way from the big bang to the
onset of inflation; because they are not ‘exited’, all these modes would be in the BD
vacuum at the onset of inflation. However, this argument is flawed on two accounts.
First, if one examines the equation governing the evolution
⊗ of these modes, one finds
that what matters is the curvature radius Rcurv = 6/R determined by the Ricci
scalar R, and not the Hubble radius. The two scales are equivalent only during slow
roll on which much of the intuition in inflation is based. However, in general they are
quite different from one another. Thus we should compare εphy with Rcurv in the pre-
inflationary epoch. The second and more important point is that the pre-inflationary
evolution should not be computed using general relativity, as is done in the argument
given above. One has to use an appropriate quantum gravity theory since the two
evolutions are expected to be very different in the Planck epoch. Then modes that
are seen in the CMB could well have εphy Rcurv in the pre-inflationary phase. If
this happens, these modes would be excited and the quantum state at the onset of the
slow roll could be quite different from the BD vacuum. Indeed, the difference could
well be so large that the amplitude of the power spectrum and the spectral index
are incompatible with WMAP observations. In this case, that particular quantum
gravity scenario would be ruled out. On the other hand, the differences could be
more subtle: the new power spectrum for scalar modes could be compatible with
observations but there may be departures from the standard predictions that involve
tensor modes or higher order correlation functions of scalar modes, changing the
standard conclusions on non-Gaussianities [26–30]. In this case, the quantum gravity
theory would have interesting predictions for future observational missions. Thus,
330 A. Ashtekar
Fig. 1 Schematic time evolution of the Hubble radius (red solid line on the right in each panel)
and of wave lengths of three modes seen in the CMB (three solid blue lines in each panel). Credits:
W. Nelson. Left Panel: General relativity. The modes of interest have wave lengths less than the
Hubble radius 1/HGR all the way from the big bang (tsing ) until after the onset of slow roll. Right
Panel: LQC. The Hubble radius diverges at the big bounce (tBoun ), decreases rapidly to reach its
minimum in the deep Planck era and then increases monotonically. Because of this, modes seen in
the CMB can have wave lengths larger than the Hubble radius 1/HLQC in the very early universe.
Detailed analysis shows that what really matters is the curvature radius Rcurv shown schematically
by the dashed red line rather than the Hubble radius 1/HLQC . But again the modes can exit the
curvature radius in the Planck regime and, if they do, they are excited during the pre-inflationary
evolution. They will not be in the BD vacuum at the onset of slow roll inflation
pre-inflationary dynamics can provide an avenue to confront quantum gravity theories

with observations.
These are not just abstract possibilities. The right panel of Fig. 1 shows
schematically the situation in LQC. (For the precise behavior obtained from nu-
merical simulations, see [12, Fig. 1].) The wave lengths of some of the observable
modes can exit the curvature radius during pre-inflationary dynamics, whence there
are departures from the standard predictions (which turn out to be of the second type
in the discussion above).
So far we have focused only on why a common argument suggesting that
pre-inflationary dynamics cannot have observational consequences is fallacious. At
a deeper level, pre-inflationary dynamics matters because of a much more general
reason: It is important to know if inflationary paradigm is part of a conceptually
coherent framework encompassing the quantum gravity regime. Can one trust the
standard scenario in spite of the fact that the modes it focuses on become trans-
Planckian in the pre-inflationary epoch? Does one have to artificially fine-tune initial
conditions in the Planck regime to arrive at the BD vacuum? Do initial conditions for
the background in the Planck regime naturally give rise to solutions that encounter
the desired inflationary phase some time in the future evolution? To investigate any
one of these issues, one needs a reliable theory for pre-inflationary dynamics and
also good control on its predictions.
4 The LQG Strategy
LQG offers an attractive framework to investigate pre-inflationary dynamics because

its underlying quantum geometry becomes important at the Planck scale and leads to
the resolution of singularities in a variety of cosmological models. In particular the
following cosmologies have been investigated in detail: the k = 0 and k = 1 FLRW
models are discussed in [2–5, 31–35]; a non-zero cosmological constant is included
in [6, 36, 37]; anisotropic are discussed via Bianchi I, II and IX models in [1, 38–
40]; and the inhomogeneous Gowdy models, which have been analyzed in detail in
classical general relativity, have been studied in [41–46]. In all cases, the big bang
singularity is resolved and replaced by quantum bounces. It is therefore natural to use
LQC as the point of departure for extending the cosmological perturbation theory.
In the standard perturbation theory, one begins with linearized solutions of Ein-
stein’s equations on a FLRW background. Unfortunately, we cannot mimic this pro-
cedure because in LQG we do not yet have the analog of full Einstein’s equations
that one would have to perturb. But one can adopt the truncation strategy discussed
in Sect. 1. Thus, one starts with a truncation γTrun of the phase space γ of general
relativity, tailored to the linear perturbations off FLRW backgrounds. Furthermore
since we are interested in the issue of whether the inflationary framework admits
a quantum gravity extension, the matter source will be just a scalar field φ with
the simplest, i.e. quadratic, potential V (φ) = (1/2)m 2 φ 2 . Thus, γTrun is given by
γTrun = γo × γ1 where γo is the 4-dimensional FLRW phase space, with the scale
factor a and the homogeneous inflaton φ as configuration variables, and γ1 is the
phase space of gauge invariant first order perturbations consisting of a scalar mode
and two tensor modes. Since the background fields are homogeneous, it is simplest
to assume that the perturbations are purely inhomogeneous. Thus, regarded as a
sub-manifold of the full phase space γ, γTrun is the normal bundle over γo .
As usual, for perturbations one can freely pass between real space and momen-
tum space using Fourier transforms of fields in co-moving coordinates. For pre-
inflationary dynamics, we work with the Mukhanov–Sasaki variables, denoted by
Qk , because they are well-defined all the way from the bounce to the onset of slow
roll.2 We denote the two tensor modes collectively by Tk . This structure is the same
as that used in standard inflation [47].
New features appear in the next step: In the passage to quantum theory, we work
with the combined system, i.e., with all of γTrun . Therefore, we are naturally led
a theory in which not only the perturbations but even the background geometry is
quantum. Rather than having quantum fields Qˆ and Tˆ propagating on a classical
FLRW space-time, they now propagate on a quantum FLRW geometry.
2 The curvature perturbations Rk fail to be well-defined at the ‘turning point’ where φ̇ = 0, which
occurs during pre-inflationary dynamics. However, they are much more convenient for relating the
spectrum of perturbations at the end of inflation with the CMB temperature fluctuations. Therefore,
we first calculate the power spectrum PQ for Mukhanov–Sasaki variable Qk and then convert it
to PR , reported in Fig. 3.
332 A. Ashtekar
Thus, the strategy to truncate the classical phase space and then pass to quantum
theory using LQG techniques leads to a novel quantum theory. The total Hilbert
space is a tensor product, H = Ho ◦ H1 , where Ho is the space of wave functions
νo describing a quantum FLRW geometry and H1 is the space of quantum states λ
of perturbations. The first task is to construct the Hilbert space Ho of physical states
νo (a, φ), by imposing the Hamiltonian constraint on the homogeneous sector γo .
The second task is to study quantum dynamics of fields Qˆ and Tˆ on the quantum
geometry encapsulated in νo (a, φ). In particular we have to introduce the Hilbert
space H1 of wave functions λ(Qk , Tk ) of perturbations and develop techniques to
calculate the 2-point functions on H1 that are needed to obtain the scalar and the
tensor power spectra. The final task is to check the self-consistency of the truncation
strategy with which we began. Already in the classical theory, the truncated phase
space γTrun is useful only so long as the back reaction can be neglected. Therefore, in
the quantum theory, we have to check that the H admits solutions νo ◦ λ in which
the energy density of perturbations is negligible compared to that in the background
all the way from the LQC bounce to the onset of slow roll. On the analytical side, this
requires the introduction of suitable regularization and renormalization techniques
for quantum fields Qˆ and Tˆ propagating on the quantum background νo . On the
numerical side, one has to devise accurate numerical methods to calculate the energy
density in perturbations with sufficient precision during the evolution all the way
from the bounce to the onset of inflation, as the background energy density falls by
some 11 orders of magnitude.
These tasks have been carried out in [10–12] using earlier results obtained in
[4–9]. The next two sections provide a flavor of this analysis.
5 Analytical Aspects
Background Quantum Geometry In the classical theory, dynamics on γo is gener-

ated by the single, homogeneous, Hamiltonian constraint, Co = 0. Each dynamical
trajectory on γo represents a classical FLRW space-time. In quantum theory, physical
states are represented by wave functions νo (a, φ) satisfying the quantum constraint
Ĉo νo = 0. Each of these solutions represents a quantum FLRW geometry.
We are interested in those solutions νo which remain sharply peaked on classical
FLRW solutions at late times. In the sector of the theory that turns out to be physi-
cally most interesting [12], these states remain sharply peaked all the way up to the
bounce but in the Planck regime they follow certain effective trajectories which in-
clude quantum corrections [14]. In particular, rather than converging on the big bang
singularity, as classical FLRW solutions do, they exhibit a bounce when the density
reaches Φmax √ 0.41ΦPl (see Fig. 2). It turns out that each (physically distinct) effec-
tive solution is completely characterized by the value φB that the inflaton assumes at
the bounce. This value turns out to be the key free parameter of the theory. Finally,
we need full quantum evolution from the bounce only until the density and curvature
3.5
3
2.5
2
1.5
φ 1
0.5
0
-0.5
-1
-1.5
1 1e+10 1e+20 1e+30 1e+40 1e+50 1e+60 1e+70 1e+80 1e+90 1e+100
v
Fig. 2 An effective LQC trajectory in presence of an inflation with a quadratic potential (1/2)m 2 φ 2 ,
where the value m = 6.1 × 10−6 m Pl of the mass is calculated from the 7 year WMAP data (source).
Here V ≈ a 3 is the volume of a fixed fiducial region. The long (blue) sloping line at the top depicts
slow roll inflation. As V decreases (right to left), we go back in time and the inflaton φ first climbs up
the potential, then turns around and starts going descending. In classical general relativity, volume
would continue to decrease until it becomes zero, signalling the big bang singularity. In LQC, the
trajectory bounces at φ ≈ 0.95 and volume never reaches zero; the entire evolution is non-singular
fall by a factor of, say, 10−6 –10−7 . After that, the background can be taken to follow
the general relativity trajectory to a truly excellent approximation.3 (For details, see
[4, 5]).
Dynamics of Perturbations There is an important subtlety which is often over-
looked in the quantum gravity literature: Dynamics of perturbations is not gener-
ated by a constraint, or, indeed by any Hamiltonian. On the truncated phase space
γTrun , the dynamical trajectories are tangential to a vector field X α of the form
αβ αβ
X α = ∂o ωβ Co + ∂1 ωβ C∗2 where ∂o and ∂1 are the symplectic structures on γo
and γ1 , and C∗2 is the part of the second order Hamiltonian constraint function in
which only terms that are quadratic in the first order perturbations are kept (ignoring
terms which are linear in the second order perturbations). X α fails to be Hamiltonian
on γTrun because C∗2 depends not only on perturbations but also background quanti-
ties. However, given a dynamical trajectory ρo (t) on γo and a perturbation at a point
thereon, X α provides a canonical lift of ρo (t) to the total space γTrun , describing the
evolution of that perturbation along ρo (t).
Therefore, in the quantum theory, dynamics of the combined system cannot be ob-
tained by simply imposing a quantum constraint on the wave functions νo ◦ λ
3 During this phase, the scalar field is monotonic in time in the effective trajectory. Therefore we can
use the scalar field as an ‘internal’ or ‘relational’ time variable with respect to which the background
scale factor (and curvature) as well as perturbations evolve. This interpretation is not essential but
very helpful in practice because of the form of the Hamiltonian constraint Ĉo νo = 0 (for details,
see e.g. [14]).
334 A. Ashtekar
of the combined system. One has to follow a procedure similar to what is done in
the classical theory. Thus, one first obtains a background quantum geometry νo by
solving Ĉo νo (a, φ) = 0, specifies the quantum state λ(Qk , Tk ) of the perturba-
tion at, say, the bounce time, and evolves it using the operator Ĉ∗2 . The resulting
state ν (a, Qk , Tk , φ) describes the evolution of the quantum perturbations λ on
the quantum geometry νo in the Schrödiger picture. (For details, see [11]).
Trans-Planckian Issues Quantum perturbations Q, ˆ Tˆ propagate on quantum
geometries νo which are all regular, free of singularities. Thus, the the framework is
tailored to cover the Planck regime. What is the status of the ‘trans-Planckian prob-
lems’ which are associated with modes of trans-Planckian frequencies in heuristic
discussions? To probe this issue one has to first note that the quantum Riemannian
geometry underlying LQG is quite subtle [48–50]: in particular, while there is a min-
imum non-zero eigenvalue of the area operators, the area gap, there is no volume
gap, even though their eigenvalues are also discrete [51, 52].4 As a consequence,
there is no fundamental obstacle preventing the existence of trans-Planckian modes
of perturbations in our truncated theory. Indeed, in the homogeneous LQC models
that have been analyzed in detail, the momentum p(φ) of the scalar field φ is gener-
ally huge in Planck units. This poses no problem and, in particular, on the physical
Hilbert space the total energy density is still guaranteed to be bounded by Φmax (see,
e.g. [14]). Similarly, perturbations Q,ˆ Tˆ of our truncated theory are permitted to
acquire trans-Planckian momenta. The real danger is rather that, in presence of such
modes, the energy density in perturbations may fail to be negligible compared to that
in the quantum background geometry. This issue is extremely non-trivial, especially
in the Planck regime. If the energy density does become comparable to that in the
background, then we would not be able to neglect the back-reaction and our trunca-
tion would fail to be self-consistent.5 This is the trans-Planckian problem we face in
our theory of quantum perturbations on inflationary quantum geometries. To address
it we need regularization and renormalization methods to compute energy density
for quantum fields on quantum FLRW geometries. (For details, see [11, 12]).
An Unforeseen Simplification As we just noted, the underlying FLRW quantum
geometry provides the necessary control on calculations in the deep Planck regime.
However, it confronts us with a new challenge of developing the mathematical theory
of quantum fields on quantum geometries. At first this problem seems formidable.
But fortunately there is a key simplification within the test field approximation we
are using in the truncated theory [9, 11]: Mathematically the evolution of Q, ˆ Tˆ
on any one of our quantum geometries νo is completely equivalent to that of these
4 Properties of the eigenvalues of length operators [53–55] have not been analyzed in comparable
detail. But since their definitions involve volume operators, it is expected that there would be no
‘length gap’.
5 Of course, this would not imply that the inflationary scenario does not admit an extension to the
Planck regime. But to obtain it one would then have to await the completion of a full quantum
gravity theory.
fields propagating on a dressed, effective metric g̃ab constructed from νo .6 Note that
g̃ab contains quantum corrections and does not satisfy Einstein’s equation. Indeed, it
does not even satisfy the effective equations of LQC because, whereas the effective
trajectories follow the ‘peak of νo ’, g̃ab also knows about certain fluctuations encoded
in νo .7 Nonetheless, since g̃ab is a smooth metric with FLRW symmetries, it is now
possible to use the rich machinery of QFT on cosmological space-times to analyze
the dynamics of Q, ˆ Tˆ in detail. In addition, one can now make use of the powerful
technique of adiabatic regularization that has been developed over some three decades
[56–61]. In particular, by restricting ourselves to states λ of perturbations which
are of fourth adiabatic order, one can compute the expectation values of energy
density. This provides a clear avenue to face the true trans-Planckian problem, i.e.,
to systematically test if the truncation approximation is valid.
This remarkable simplification occurs because the dynamics of test quantum fields
is not sensitive to all the details of the probability amplitude for various FLRW
metrics encapsulated in νo ; it experiences only to a few moments of this distribution.
The phenomenon is analogous to the propagation of light in a medium where all the
complicated interactions of the Maxwell field with the atoms in the medium can be
captured just in a few parameters such as the refractive index. (For details, see [9,
11, 12]).
Initial Conditions In the Schrödinger picture, the above simplification enables us
to evolve the quantum state λ of perturbations. But we still have to specify the
initial conditions. Since the big bang of general relativity is replaced by the big
bounce in LQC, it is natural to specify them at the bounce. Now, in the truncation
approximation, perturbation is treated as a test field. Therefore, it is appropriate to
assume that the initial state has the form νo ◦ λ at the bounce. Furthermore this
tensor product form will be preserved under dynamics so long as the back reaction
due to the perturbation remains negligible.
Let us begin with νo . In the effective theory, phase space variables are subject to
certain constraints at the bounce. We assume that νo is sharply peaked at a point
on this constraint surface (with small fluctuations in each of the two ‘conjugate’
variables). At the bounce, the allowed range of φ is finite but large, |φB | ⇔ (0, 7.47×
105 ) in Planck units, and a detailed analysis of effective solutions has shown that
unless |φB | < 0.93, the effective trajectory necessarily encounters a slow roll phase
compatible with WMAP sometime in the future [8]. Thus, the peak of initial νo is
almost unconstrained. However, the requirement that νo be peaked is very strong
and makes the initial state of background geometry very special.
For perturbations, we assume the following three conditions on λ at the bounce: (i)
Symmetry: λ should be invariant under the FLRW isometry group, i.e., under spatial
6 For scalar modes, the classical equation of motion involves also ‘an external potential’ A. This
has also to be replaced by a dressed effective potential Ã. For details, see [12].
7 While this difference is conceptually important, because the states ν of interest are so sharply
o
peaked, in practice the deviations from effective trajectories are small even in the Planck regime.
Of course the deviations from classical solutions are enormous in the Planck regime because g̃ab is
non-singular.
336 A. Ashtekar
translations and rotations. This condition is natural because these are the symmetries
of the background νo and hence also of g̃ab it determines; (ii) Regularity: λ should be
of forth adiabatic order so that the Hamiltonian operator has a well-defined action on
it; and, (iii) The initial renormalized energy density ≤λ| Φ̂ |λ≡ren in the perturbation
should be negligible compared to the energy density Φmax in the background. We
have an explicit example, |λ≡ = |0obv ≡, of such a state called the ‘obvious vacuum
of forth adiabatic order’ which has several attractive properties [12]. Furthermore we
also know that, given a state satisfying these properties, there are ‘infinitely many’
such states in its neighborhood. Thus, the existence of the desired states is assured.
However, in view of the large freedom that remains, it is desirable to develop clear-
cut physical criteria to cut down this freedom significantly. This is an open issue,
currently under investigation. (For details, see [10–12]).
Let us summarize the analytical framework. The initial condition for the quantum
state νo ◦ λ of the combined system is specified at the bounce in such a manner
that a slow roll inflation compatible with the 7 year WMAP data is guaranteed in
the background geometry. Thanks to an unforeseen simplification, we can use tech-
niques from QFT on cosmological space-times to evolve the perturbations Qˆ and Tˆ
on the quantum background geometry νo . Finally, the initial conditions guarantee
that the truncation approximation does hold at the bounce: λ can be regarded as a
perturbation whose back reaction on νo is negligible initially. Furthermore, states
are sufficiently regular to enable us to calculate the energy density in the background
and in the perturbation at all times. Therefore, one can carry out the entire evolu-
tion numerically, calculate the power spectra and spectral indices and check if the
truncation approximation continues to hold under evolution all the way from the
bounce to the onset of the slow roll.
As discussed in Sect.3, a priori there are several possible outcomes. Pre-inflationary
dynamics could have such a strong effect that the power spectra and the spectral
indices that result from these calculations are incompatible with the WMAP obser-
vations. In this case, the LQC extension would be ruled out by observations. It is also
possible that the Planck scale dynamics is such that the back reaction ceases to be
negligible very soon after the bounce making the truncation strategy inconsistent. In
this case, our truncation strategy would fail to be self-consistent. Finally, even if these
possibilities do not occur, we may find that, for observable modes, the state at the
onset of inflation is sufficiently different from the BD vacuum that there are depar-
tures from the standard inflationary predictions for future observations. One needs
explicit numerical simulations to find out which of these various a priori possibilities
are realized.
6 Numerical Aspects, Observations and Self-Consistency
In this section, numerical values of all physical quantities will be given in natural
Planck units c = = G = 1 (as opposed to the reduced Planck units used in the
cosmology literature where one sets 8Θ G = 1). We will use both the conformal time
τ̃ and the proper (or cosmic) time t˜ determined by the dressed effective metric g̃ab
via d s̃ 2 := g̃ab d x a d x b = a 2 (−d τ̃2 +dx2 ) = −d t˜2 +a 2 dx2 (where, as usual, x a are
the co-moving coordinates). This is because the cosmology literature generally uses
conformal time but comparison with general relativity can be made more transparent
in cosmic time by setting it equal to zero at the big bang in general relativity and at
the big bounce in LQC.
6.1 WMAP Phenomenology
The 7 year WMAP data [25] uses a reference mode kξ √ 8.58ko where, as before,
ko is the co-moving wave number of the mode whose physical wave length equals
the radius of the observable universe at the surface of last scattering. The WMAP
analysis provides us with the amplitude PR (kξ ) of the power spectrum and the
spectral index n s (kξ ) which encodes the small deviation from scale invariance, both
for the scalar perturbations. The values are given by
PR (kξ ) = (2.430 ± 0.091) × 10−9 and n s (kξ ) = 0.968 ± 0.012 . (1)
For the quadratic potential considered here, these observational data provide the fol-
lowing values of the Hubble parameter H and the slow roll parameter ζ = − Ḣ /H 2 :
H (τ̃(kξ )) = 7.83 × 10−6 and ζ(τ̃(k∇ )) = 8 × 10−3 , (2)
where τ̃(kξ ) is the conformal time in our dressed effective metric g̃ab at which the
mode kξ exited the Hubble radius and the ‘dot’ refers to the derivative w.r.t. t˜. Since the
physical wave length of the mode ko is 8.58 times larger, it must have left the Hubble
radius ≈2 e-foldings before τ̃(kξ ). Onset of slow roll inflation is taken to commence
a little before the ko exits its Hubble horizon. The value of the Hubble parameter
at this time is so low that the total energy density is less than 10−11 ΦPl . Therefore
throughout the inflationary era general relativity is an excellent approximation to
LQC. Equations of general relativity (or, LQC) determine the mass m of the inflaton
as well as values of the inflaton φ at τ̃(kξ ):
m = 1.21 × 10−6 and φ(t˜(kξ )) = ±3.15 . (3)
Because of the observational error bars, these quantities are uncertain by about 2 %.
In the numerical simulations we use the value of m given in (3). (For details, see [8]).
338 A. Ashtekar
6.2 Evolution of the Background
So far numerical evolutions of the background wave function νo are feasible only
for kinetic dominated bounces, i.e., bounces for which φB is small. This is because
the required time over which one has to integrate to arrive in the general relativ-
ity regime increases rapidly with φB . Fortunately, as we will see below, this is the
most interesting portion of the allowed values of φB . These simulations show that
νo remains sharply peaked on an effective trajectory. Since there is no obvious rea-
son why this should not continue for higher φ B values, it is instructive to examine
all effective trajectories without restricting ourselves to kinetic energy dominated
bounces. The trajectory would be compatible with the 7 year WMAP data only if
at the point at which H takes the value 7.83 × 10−6 , within the margin given by
observational errors, ζ = 8 × 10−3 , and φ = 3.15. A surprising result is that
this is in fact the case under a very mild condition: φB ⇒ 0.93 [8]. Note that this
result is much stronger than the qualitative ‘attractor behavior’ of inflationary trajec-
tories because it is quantitative and tuned to the details of the WMAP observations.
(For details, see [8]).
To make contact with the WMAP observations, we need to find kξ and the time
τ̃(kξ ) at which the mode with co-moving wave number kξ exits the Hubble horizon
during inflation. For this, it is simplest to fix the scale factor at the bounce and we will
choose the convention aB = 1. (Note that this is very different from atoday = 1 often
used in cosmology.) Then, along each dynamical trajectory one locates the point at
which the Hubble parameter takes the value H = 7.83 × 10−6 (and makes sure that
at this time ζ and φ are given by (2) and (3) within observational errors). One calls
the conformal time at which this occurs τ̃(kξ ) and numerically calculates the scale
factor a(τ̃(kξ )) at this time. Then, the value of the co-moving momentum kξ of this
mode is determined by the fact that this mode exits the Hubble radius at time τ̃(kξ ).
Thus, one asks that the physical wave number of this mode should equal the Hubble
parameter: k/a(τ̃(kξ )) = H (τ̃(kξ )). Table 1 shows the values of kξ , the physical
wave length of the mode at the bounce time, the proper time t˜(kξ ) at which the mode
exits the Hubble horizon, and the number of e-foldings between the bounce and time
t˜(kξ ) for a range of values of φB which turns out to be physically most interesting.
(For details, see [12]).
6.3 Evolution of Perturbations
Preliminary numerical simulations were first carried out using four different states λ
at the bounce, satisfying the initial conditions discussed in Sect. 3. They showed that
the results are essentially insensitive to the choice. Then detailed and much higher
precision simulations were carried out using |λ≡ = |0obv ≡, the ‘obvious vacuum of
fourth adiabatic order’, at the bounce because, as mentioned before, this state has a
number of attractive properties. These simulations revealed an unforeseen behavior:
Table 1 This table from [12] shows the value of the reference co-moving momentum kξ used in the
WMAP data, the corresponding physical wavelength εξ (t˜B ) at the bounce, the time t˜(kξ ) at which
the mode kξ exits the Hubble radius during inflation, and ln[a(t˜(kξ ))/a(t˜B )], the number of e-folds
of expansion between the bounce and t˜(kξ )
φ(t B ) k∇ ln k∇ ε∇ (t B ) tk ∇ ln[a(tk∇ )/a(t B )]
0.934 0.0016 −6.4 4008 1.8 × 105 5.2
1 0.024 −3.7 261 5.2 × 105 8.0
1.05 0.17 −1.8 37.1 7.6 × 105 10
1.1 1.2 0.2 5.1 1.0 × 106 12
1.15 9.17 2.83 0.63 1.25 × 106 13.9
1.2 70.7 4.2 0.09 1.48 × 106 16
1.3 4.58 × 103 8.43 1.36 × 10−3 1.97 × 106 20.2
1.5 2.7 × 107 17.1 2.3 × 10−7 2.9 × 106 28.9
We focus on the range for φB that is relevant to explore whether pre-inflationary dynamics can lead
to deviations from the BD vacuum at the onset of the slow roll
2.4
2.2
2
1.8
1.6
PLQC
PBD
1.4
1.2
1
0.8
0.6
0.4
6 7 8 9 10 11 12 13 14 15
k
Fig. 3 Ratio of the LQC power spectrum for curvature perturbations in the scalar modes to that
predicted by standard inflation (source [12]). For small k, the ratio oscillates very rapidly. The (red)
solid curve shows averages over (co-moving) bins with width 0.5 δPl −1
the power spectra for scalar and tensor perturbations are largely insensitive to the
value of φB . However, recall that there is finite window (ko , 2000ko ) of co-moving
modes that can be seen in the CMB. Because of the pre-inflationary dynamics, the
value of kξ —and hence of ko —does depend on φ B and rapidly increases with φB .
(See Table 1.) Therefore, the window of observable modes is sensitive to the value
of φB and moves steadily to the right as φB increases.
LQC
Figure 3 shows the plot of the ratio PR /PR BD of the LQC power spectrum
to the standard inflationary one for curvature perturbations R of the scalar modes.
The (blue) circles are the data points. The LQC power spectrum has very rapid
oscillations (whose amplitudes decay quickly with k) which descend to the ratio that
340 A. Ashtekar
is plotted. Since observations have only a finite resolution, to compare with data it is
simplest to average over small bins. We used bins which, at the bounce, correspond
−1
to a band-width in physical wave numbers of 0.5tPl . The result is the solid (red)
line. We see that the two power spectra agree for k 6.5 but LQC predicts an
enhancement for k 6.5. We will now comment on these features.
Let us first note that the LQC power spectrum in this plot uses the value φB = 1.15.
As Table 1 shows, the corresponding kξ is 9.17. At this value, the two power spectra
are identical, whence the amplitude and the spectral index obtained from the LQC
evolution at k = kξ agrees with the values (1) observed by WMAP. However, as we
remarked, for k 6.5, the LQC prediction departs from that of standard inflation.
These low k values correspond to δ 22 in the angular decomposition used by
WMAP for which the error bars are quite large. Therefore, although the LQG power
spectrum differs from the standard one in this range, both are admissible as far as
the current observations are concerned.
What is the physics behind the enhancement of the LQC power spectrum for
k 6.5? And where does this specific scale come from? This enhancement is due to
pre-inflationary dynamics. At the bounce, the scalar curvature has a universal value
in LQC which sets a scale kLQC √ 3.21. Modes with k ∈ kLQC experience negligible
curvature during their pre-inflationary evolution while those with k comparable to
kLQC or less do experience curvature and therefore get excited. These are general
physical arguments and one needs numerical simulations to determine exactly what
‘much greater than’ and ‘comparable to’ means. The simulations show that modes
with k 2kLQC already satisfy the ‘much greater than’ criteria. They are not excited
and for them the LQC state λ at the onset of inflation is virtually indistinguishable
from the BD vacuum. That is why the two power spectra are essentially the same for
k 2kLQC . But for modes with k 2kLQC the LQC state λ has excitations over the
BD vacuum whence there is an enhancement of the power spectrum.
What happens if we change φB ? As we remarked above, the prediction of the
LQC power spectrum is pretty insensitive to the value of φB but the window in the
k space spanned by modes which are observable in the CMB changes, moving to
the right as φB increases. Now, as Table 1 shows, if φB > 1.2, we have ko > 6.5,
whence none of the observable modes would be excited during the pre-inflationary
evolution. In this case, at the onset of the slow roll, the LQC sate λ would be
indistinguishable from the BD vacuum, whence all LQC predictions would agree
with those of standard inflation. Thus, there is a narrow window, 0.93 ⊥ φB ⊥ 1.2 for
which the background νo admits the desired slow roll phase and yet LQC predictions
for future observations can differ from the standard ones. One example is given by
a consistency relation r = −8n t in standard inflation, where r = 2PT BD /P is the
R
tensor to scalar ratio and n t is the spectral index for tensor modes. This relation is
significant because it does not depend on the form of inflationary potential. It turns
out that r does not change in LQC but n t does, whence this standard consistency
relation is modified. Future observations would be able to test for such departures.
There is also a systematic study of the effect that excitations over the BD vacuum
can have on non-Gaussianities [26–30]. Furthermore, it has been recently pointed
out that these non-Gaussianities could be seen in the galaxy correlation functions and
also in certain distortions in the CMB [62–64]. Thus, there are concrete directions
in which cosmological observations could soon start probing effects that originate at
the Planck scale. (For further details, see [12]).
6.4 Self Consistency
Finally, let us discuss the issue of self-consistency of the truncation scheme, i.e.,
the issue of whether the test field approximation continues to hold under evolution.
This issue is quite intricate and had remained unexplored because of two different
issues. The first issue is conceptual: it was not clear how to compute the renormalized
energy density for the quantum fields Q, ˆ Tˆ in a manner that is meaningful in the
Planck regime. As discussed in Sect. 5, we were able to construct this framework
by ‘lifting’ the adiabatic renormalization theory on classical cosmological space-
times to that on quantum geometries νo . The second set of difficulties comes from
numerics: one requires very high accuracy and numerical precision. This is because
(i) the rapid oscillations of integrand of ≤λ| Φ̂ |λ≡ren in the k space make it difficult to
evaluate the exact value of the renormalized energy density; and, (ii) the background
energy density itself decreases from Planck scale to 10−11 times that scale. Indeed,
so far we have only managed to find an upper bound on the energy density in the
perturbations, shown in Fig. 4. But this suffices to show that, for φ B > 1.22, our initial
conditions at the bounce do give rise to a self-consistent solution νo ◦ λ throughout
the evolution from the big bounce to the onset of slow roll. These solutions provide a
viable extension of the standard inflationary scenario all the way to the Planck scale.
The issue of whether one can push the value of φB to include the interesting domain
φB < 1.2 is still under investigation. (There are several aspects to this problem,
including a better handling of the infrared regime, briefly discussed in [12].)
7 Summary and Discussion
I began in Sect. 1 by making some suggestions: (i) Progress in quantum gravity

should be gauged by the degree to which an approach succeeds in overcoming lim-
itations of general relativity; (ii) The development of quantum theory, rather than
general relativity, offers a better example to emulate in this endeavor; and, (iii) As in
quantum theory, it may be more fruitful to resolve concrete physical problems at the
interface of gravity and quantum theory rather than focusing all efforts on obtaining a
complete quantum gravity theory in one stroke. In Sects. 2 and 3 we saw that the very
early universe offers an obvious arena for this task for both conceptual and practical
reasons. Conceptually, the big bang is a prediction of general relativity in a regime in
which the theory is not applicable, whence it is important to find out what really hap-
pened in the Planck regime. In practical terms, currently the early universe offers the
342 A. Ashtekar
1
0.01
0.0001
1e-06
1e-08
ρ
1e-10
1e-12
1e-14
1e-16
1e-18
0.0001 0.01 1 100 10000
tPl
Fig. 4 For φB = 1.23m Pl , energy density in the background (upper curve) and an upper bound
on the energy density in perturbations (lower curve) are plotted against time from the bounce to
the onset of slow roll using Planck units (source [10]). The test field approximation holds across a
change of 11 orders of magnitude in both quantities
best hope to confront quantum gravity theories with observations. In particular, we

saw that the inflationary paradigm has been highly successful in accounting for the
inhomogeneities in the CMB—and hence accounting for the large scale structure of
the universe—but it has several limitations. In Sects. 4–6, I summarized how the lim-
itations related to the Planck scale physics are being addressed in LQG. Specifically,
by using the truncation strategy of LQG, over the last 6 years it has been possible to
extend the inflationary paradigm all the way to the deep Planck regime. (For other
treatments of pre-inflationary dynamics within LQG, see e.g. [65, 66].)
The first finding is that the big bang singularity is resolved in LQC and replaced by
the big bounce. Since quantum physics—including quantum geometry—is regular
at the big bounce, it is natural to specify initial conditions for the quantum state
νo that encodes the background, homogeneous quantum geometry, as well as for λ
that describes the quantum state of perturbations. Physically, the initial conditions
amount to assuming that the state νo ◦ λ at the bounce should satisfy ‘quantum
homogeneity’. More precisely, at the bounce one focuses just on that region which
expands to become the observable universe and demands that it be homogeneous
except for the inevitable quantum fluctuations that one cannot get rid of even in
principle. Now, because of the pre-inflationary and inflationary expansion, the region
of interest has a radius smaller than ≈10δPl at the bounce. But as has been emphasized
in the relativity literature, this creates a huge fine tuning problem. For, to account
for the impressive fact that inhomogeneities in the CMB are really tiny—just one
part in 105 —the required homogeneity at the bounce has to be truly extraordinary.
The standard inflationary paradigm is not really applicable at the Planck scale and,
even if one were to ignore this fact, it does not have a natural mechanism to achieve
this degree of homogeneity. In LQC, on the other hand, the big bang singularity is
resolved precisely because there is an in-built repulsive force with its origin in the
specific quantum geometry that underlies LQG. While this force is negligible when
curvature is less than, say, 10−6 in Planck units, it rises spectacularly in the Planck
regime, overcomes the huge classical gravitational attraction and prevents the big
bang singularity. In more general models referred to in Sect. 4, one finds a pattern:
every time a curvature scalar enters the Planck regime, this repulsive force becomes
dominant and dilutes that curvature scalar, preventing a singularity (see e.g. [14]).
This opens the possibility that the ‘dilution effect’ of the repulsive force may be
sufficient to create the required degree of homogeneity on the scale of about 10δPl ,
thereby accounting for the assumed ‘quantum homogeneity’. If this idea could to be
developed in detail, dynamics of the pre-bounce universe will leave no observable
effects, providing a clear-cut case for specifying initial conditions at the bounce. Of
course, the pre-bounce dynamics will still lead to inhomogeneities at larger scales on
the bounce surface but they would have Fourier modes whose physical wave length
is much larger than the radius of the observable universe. Therefore, they would not
be in the observable range; in the truncated theory considered here, they would be
absorbed in the quantum geometry of the homogeneous background. This ‘dilution
mechanism’ and other issues related to initial conditions are likely to be a center of
activity in the coming years.
As we saw in Sects. 5 and 6, we now have a conceptual framework and numerical
tools to evolve these initial conditions all the way from the bounce to the onset of
slow roll. The result depends on where one is in the parameter space that is labeled
by the value φB of the inflaton at the bounce. For a very large portion of the parameter
space we obtain the following three features: (i) Some time in its future evolution, the
background geometry encounters a slow roll phase that is compatible with the 7 year
WMAP observations; (ii) At the onset of this slow roll, the state λ of perturbations is
essentially indistinguishable from the BD vacuum used in standard inflation; and (iii)
the back reaction due to perturbations remains negligible throughout pre-inflationary
dynamics in which the background curvature falls by some 11 orders of magnitude,
justifying the underlying ‘truncation approximation’. Thus, for this portion of the
parameter space, we have a self-consistent extension of the standard inflationary
paradigm.
There is, however, a small window in the parameter space for which the feature
(i) is realized but the initial state at the onset of inflation contains an appreciable
number of BD excitations. This number is within the current observational limits.
But the presence of these excitations signals new effects such as a departure from the
inflationary ‘consistency relation’ involving both scalar and tensor modes and a new
source of non-Gaussianities. These could be seen in future observational missions
[62–64]. The physical origin of these effects can be traced back to a new energy scale
kLQC defined by the universal value of the scalar curvature at the bounce. Excitations
with k 2kLQC are created in the Planck regime near the bounce. It turns out that if
the number N of e-foldings in the scale factor a between the bounce and τ̃ = τ̃k ξ is
less than 15, then the modes which are excited would be seen in the CMB. This occurs
only in the small window of parameter space referred to above. Since the window
is very small, the ‘a priori probability’ that one of these values of φB is realized in
344 A. Ashtekar
Nature would seem to be tiny. However, one can turn this argument around. Should
these effects be seen, the parameter space would be narrowed down so much that very
detailed calculations would become feasible. In either case, it is rather exciting that
the analysis relates initial conditions and Planck scale dynamics with observations,
thereby expanding the reach of cosmology to the earliest moment in the deep Planck
regime.
Even when a self-consistent solution νo ◦ λ to the truncated theory exists, how
would it fit in full LQG? Recall the situation in classical general relativity. In cos-
mology as well as black hole physics, one routinely expects first order perturbations
whose back reaction is negligible to provide excellent approximations to the phenom-
enological predictions of the exact theory. I see no obvious reason why the situation
would be different in quantum gravity. As a simple example to illustrate the general
viewpoint, consider the Dirac solution of the hydrogen atom. Since one assumes
spherical symmetry prior to quantization, this truncation excludes photons from the
beginning. Therefore, at a conceptual level, the Dirac description is very incomplete.
Yet, as far as experiments are concerned, it provides excellent approximations to
answers provided by full QED until one achieves the accuracy needed to detect the
Lamb shift. I expect the situation to be similar for our truncated theory: Conceptually
it is surely quite incomplete vis a vis full LQG, but the full theory will provide only
small corrections to the observable effects.
To conclude, let me emphasize that there was no a priori reason to anticipate
either of the two main conclusions—the extension of standard inflation to the
Planck regime for much of the parameter space and deviations from some of its
predictions in a narrow window. Indeed, it would not have been surprising if the
pre-inflationary dynamics of LQC was such that the predicted power spectra were
observationally ruled out for the ‘natural’ initial conditions we used at the bounce, or,
if the self-consistency of truncation had failed quite generally because of the Planck
scale dynamics. Indeed, this could well occur in generic bouncing scenarios, e.g.
in situations in which the expansion between the bounce and the surface of large
scattering is not sufficiently large for the modes observed in the CMB to have wave
lengths smaller than the curvature radius throughout this evolution.
Acknowledgments This report is based on joint work with Ivan Agullo, Alejandro Corichi, Woj-
ciech Kaminski, Jerzy Lewandowski, William Nelson, Tomasz Pawlowski, Parampreet Singh and
David Sloan over the past six years. I am most grateful for this collaboration. I am also indebted
to a very large number of colleagues especially in the LQG community for discussions, comments,
questions and criticisms. This work was supported by the NSF grant PHY-1205388 and the Eberly
research funds of Penn state.
References
1. Einstein A.: Letter to Arnold Sommerfeld, dated November 28th, 1915, reporting his discovery
of general relativity (1915)
2. Ashtekar, A., Pawlowski, T., Singh, P.: Quantum nature of the big bang. Phys. Rev. Lett. 96,
141301 (2006). doi:10.1103/PhysRevLett.96.141301
3. Ashtekar, A., Pawlowski, T., Singh, P.: Quantum nature of the big bang: An analytical and
numerical investigation I. Phys. Rev. D 73, 124038 (2006). doi:10.1103/PhysRevD.73.124038
4. Ashtekar, A., Pawlowski, T., Singh, P.: Quantum nature of the big bang: improved dynamics.
Phys. Rev. D 74, 084003 (2006)
5. Ashtekar, A., Corichi, A., Singh, P.: Robustness of predictions of loop quantum cosmology.
6. Pawlowski, T., Ashtekar, A.: Positive cosmological constant in loop quantum cosmology. Phys.
7. Ashtekar, A., Sloan, D.: Loop quantum cosmology and slow roll inflation. Phys. Lett. B 694,
108 (2010)
8. Ashtekar, A., Sloan, D.: Probability of inflation in loop quantum cosmology. Gen. Relativ.
Gravit. 43, 3619 (2011). doi:10.1007/s10714-011-1246-y
9. Ashtekar, A., Kaminski, W., Lewandowski, J.: Quantum field theory on a cosmological, quan-
tum space-time. Phys. Rev. D 79, 064030 (2009). doi:10.1103/PhysRevD.79.064030
10. Agullo, I., Ashtekar, A., Nelson, W.: A quantum gravity extension of the inflationary scenario.
11. Agullo, I., Ashtekar, A., Nelson, W.: An extension of the quantum theory of cosmological
perturbations to the Planck era. ArXiv e-prints arXiv:1211.1354 [gr-qc] (2012)
12. Agullo, I., Ashtekar, A., Nelson, W.: The pre-inflationary dynamics of loop quantum cosmol-
ogy: confronting quantum gravity with observations. Class. Quant. Grav. 30, 085014 (2013).
doi:10.1088/0264-9381/30/8/085014
13. Bojowald, M.: Loop quantum cosmology. Living Rev. Relativ. 8, 11. http://www.livingreviews.
org/lrr-2005-11 (2005)
14. Ashtekar, A., Singh, P.: Loop quantum cosmology: a status report. Class. Quantum Grav. 28,
213001 (2011). doi:10.1088/0264-9381/28/21/213001
15. Liddle, A.R., Lyth, D.H.: Cosmological Inflation and Large-Scale Structure. Cambridge Uni-
versity Press, Cambridge (2000)
16. Dodelson, S.: Modern Cosmology. Academic Press, Amsterdam (2003)
17. Mukhanov, V.: Physical Foundations of Cosmology. Cambridge University Press, Cambridge
(2005)
18. Weinberg, S.: Cosmology. Oxford University Press, Oxford (2008)
19. Gorbunov, D.S., Rubokov, V.A.: Introduction to the Theory of the Early Universe: Cosmological
Perturbations and Inflationary Theory. World Scientific, Singapore (2011)
20. Penrose, R.: Faith, Fashion and Fantasy in the New Physics of the Universe
(2003). http://lectures.princeton.edu/2006/roger-penrose/, http://hulk03.princeton.edu:8080/
WebMedia/lectures/
21. Brandenberger, R.H.: Introduction to early universe cosmology. ArXiv e-prints
arXiv:1103:2271 [astro-ph.CO], (PoS(ICFI 2010)001) (2011)
22. Brandenberger, R.H., Martin, J.: Trans-Planckian issues for inflationary cosmology,
arXiv:1211.6753
23. Borde, A., Guth, A., Vilenkin, A.: Inflationary space-times are not past-complete. Phys. Rev.
24. Hawking, S.W., Ellis, G.F.R.: Large Scale Structure of Space-Time. Cambridge University
25. Komatsu, E., et al.: Seven-year Wilkinson microwave anisotropy probe (WMAP) observations:
cosmological interpretation. Astrophys. J. Suppl. Ser. 192, 18 (2011). doi:10.1088/0067-0049/
192/2/18
26. Holman, R., Tolley, A.: Enhanced non-Gaussianity from excited states. J. Cosmol. Astropart.
Phys. 2008(05), 001 (2008). doi:10.1088/1475-7516/2008/05/001
27. Agullo, I., Parker, L.: Non-gaussianities and the stimulated creation of quanta in the inflationary
universe. Phys. Rev. D 83, 063526 (2011a). doi:10.1103/PhysRevD.83.063526
28. Agullo, I., Parker, L.: Stimulated creation of quanta during inflation and the observable universe
Gen. Relativ. Gravit. 43, 2541 (2011b). doi:10.1007/s10714-011-1220-8
346 A. Ashtekar
29. Ganc, J.: Calculating the local-type fNL for slow-roll inflation with a non-vacuum initial state.
30. Agullo, I., Navarro-Salas, J., Parker, L.: Enhanced local-type inflationary trispectrum from a
non-vacuum initial state. J. Cosmol. Astropart. Phys. 2012(05), 019 (2012). doi:10.1088/1475-
7516/2012/05/019
31. Bojowald, M.: Absence of singularity in loop quantum cosmology. Phys. Rev. Lett. 86, 5227
(2001). doi:10.1103/PhysRevLett.86.5227
32. Ashtekar, A., Bojowald, M., Lewandowski, J.: Mathematical structure of loop quantum cos-
mology. Adv. Theor. Math. Phys. 7, 233 (2003)
33. Singh, P.: Are loop quantum cosmologies never singular? Class. Quantum Grav. 26, 125005
(2009). doi:10.1088/0264-9381/26/12/125005
34. Ashtekar, A., Pawlowski, T., Singh, P., Vandersloot, K.: Loop quantum cosmology of k=1 FRW
models. Phys. Rev. D 75, 0240035 (2006)
35. Szulc, L., Kaminski, W., Lewandowski, J.: Closed FRW model in loop quantum cosmology.
Class. Quantum Gravity 24, 2621 (2007)
36. Bentivegna, E., Pawlowski, T.: Anti-deSitter universe dynamics in LQC. Phys. Rev. D 77
124025 (2008). doi:10.1103/PhysRevD.77.124025
37. Kaminski, W., Pawlowski, T.: The LQC evolution operator of FRW universe with positive
cosmological constant. Phys. Rev. D 81, 024014 (2010)
38. Ashtekar, A., Wilson-Ewing, E.: Loop quantum cosmology of Bianchi type I models, Phys.
39. Martin-Benito, M., Mena Marugan, G.A., Pawlowski, T.: Loop quantization of vacuum Bianchi
I cosmology. Phys. Rev. D 78, 064008 (2008). doi:10.1103/PhysRevD.78.064008
40. Wilson-Ewing, E.: Loop quantum cosmology of Bianchi type IX models. Phys. Rev. D 82,
043508 (2010). doi:10.1103/PhysRevD.82.043508
41. Martin-Benito, M., Garay, L.J., Mena, G.A.: Hybrid quantum Gowdy cosmology: combining
loop and Fock quantizations. Phys. Rev. D 78, 083516 (2008). doi:10.1103/PhysRevD.78.
083516
42. Garay, L.J., Martn-Benito, M., Mena Marugan, G.A.: Inhomogeneous loop quantum cosmol-
ogy: Hybrid quantization of the Gowdy model. Phys. Rev. D 82 044048 (2010). doi:10.1103/
PhysRevD.82.044048
43. Brizuela, D., Mena Marugan, G.A., Pawlowski, T.: Big bounce and inhomogeneities. Class.
Quantum Grav. 27, 052001 (2010). doi:10.1088/0264-9381/27/5/052001
44. Martin-Benito, M., Mena Marugan, G.A., Wilson-Ewing, E.: Hybrid quantization: from
Bianchi I to the Gowdy model. Phys. Rev. D 82, 084012 (2010). doi:10.1103/PhysRevD.
82.084012
45. Martn-Benito, M., Martin-de Blas, D., Mena Marugan, G.A.: Matter in inhomogeneous loop
quantum cosmology: the Gowdy T 3 model. ArXiv e-prints arXiv:1012.2324 [gr-qc] (2010)
46. Brizuela, D., Mena Marugan, G.A., Pawlowski, T.: Effective dynamics of the hybrid quantiza-
tion of the Gowdy T3 universe. ArXiv e-prints arXiv:1106.3793 [gr-qc] (2011)
47. Langlois, D.: Hamiltonian formalism and gauge invariance for linear perturbations in inflation.
Class. Quant. Grav. 11, 389 (1994). doi:10.1088/0264-9381/11/2/011
48. Ashtekar, A., Lewandowski, J.: Background independent quantum gravity: a status report.
Class. Quantum Grav. 21, R53 (2004). doi:10.1088/0264-9381/21/15/R01
49. Rovelli, C.: Quantum Gravity. Cambridge University Press, Cambridge (2004)
50. Thiemann, T.: Introduction to Modern Canonical Quantum General Relativity. Cambridge
University Press, Cambridge (2007)
51. Rovelli, C., Smolin, L.: Discreteness of area and volume in quantum gravity. Nucl. Phys. B
442(593), 1995 (1995). doi:10.1016/0550-3213(95)00150-Q. Erratum: ibid. 456, 753
52. Ashtekar, A., Lewandowski, J.: Quantum theory of geometry I: area operators. Class. Quantum
Grav. 14, A55 (1997). doi:10.1088/0264-9381/14/1A/006
53. Thiemann, T.: A length operator for canonical quantum gravity. J. Math. Phys. 39, 3372 (1998)
54. Bianchi, E.: The length operator in loop quantum gravity. Nucl. Phys. B 807, 591 (2009).
doi:10.1016/j.nuclphysb.2008.08.013
55. Ma, Y., Soo, C., Yang, J.: New length operator for loop quantum gravity. Phys. Rev. D 81,
124026 (2010). doi:10.1103/PhysRevD.81.124026
56. Parker, L.: The creation of particles in an expanding universe. Ph.D. thesis, Harvard University,
Harvard (1966)
57. Parker, L., Fulling, S.A.: Adiabatic regularization of the energy-momentum tensor of a quan-
tized field in homogeneous spaces. Phys. Rev. D 9, 341 (1974). doi:10.1103/PhysRevD.9.
341
58. Birrell, N.D.: The application of adiabatic regularization to calculations of cosmological inter-
est. Proc. R. Soc. Lond. Ser. A 361, 513 (1978)
59. Anderson, P.R., Parker, L.: Adiabatic regularization in closed Rebertson–Walker universes.
Phys Rev. D 36, 2963 (1987). doi:10.1103/PhysRevD.36.2963
60. Fulling, S.: Aspects of Quantum Field Theory in Curved Space-Times. Cambridge University
61. Parker, L., Toms, D.: Quantum Field Theory in Curved Space-Time. Cambridge University
62. Ganc, J., Komatsu, E.: Scale dependent bias of galaxies and μ-type disruption of the cosmic
microwave background spectrum from a single field inflation with a modified initial state. Phys.
63. Agullo, I., Shandera, S.: Large non-Gaussian halo bias from single field inflation. J. Cosmol.
Astropart. Phys. 2012(09), 007 (2012). doi:10.1088/1475-7516/2012/09/007
64. Schmidt, F., Hui, L.: CMB power asymmetry from Gaussian modulation. ArXiv e-prints
arXiv:1210.2965 [astro-ph.CO] (2012)
65. Linsefors, L., Cailleteau, T., Barrau, A., Grain, J.: Primordial tensor power spectrum in
holonomy corrected Omega-LQC. ArXiv e-prints arXiv:1212.2852 [gr-qc] (2012)
66. Fernandez-Mendez, M., Mena Marugan, G.A., Olmedo, J.: Hybrid quantization of an infla-
tionary universe. Phys. Rev. D 86, 024003 (2012). doi:10.1103/PhysRevD.86.024003
The Inflationary Origin of the Seeds of Cosmic
Structure: Quantum Theory and the Need for
Novel Physics
Daniel Sudarsky
Abstract The Inflationary account for the emerging of the seeds of cosmic structure
from quantum fluctuations is a central part of our current views of cosmology. It is, on
the one hand, extremely successful at the phenomenological level, and yet, it retains
an aspect that is generally regarded as controversial: The exact mechanism by which
quantum fluctuations transmute into actual inhomogeneities. We will review the
considerations that lead us to conclude that the fully satisfactory resolution of the
issue requires novel physics, and we will discuss an option we have been considering
in this regard.
1 Introduction
This conference commemorates the time spent by Einstein in Prague, which was
instrumental in his development of General Relativity, a subject which has been the
focus of the majority of the other presentations. I will be touching on the other great
question that preoccupied Einstein at the time: Quantum Theory. We note that, the
subject of this manuscript; inflation, represents the only generally accepted example
of an instance in which General Relativity, Quantum Theory and observations come
together. It is, therefore, quite remarkable that it is precisely here where we must
confront the conceptual difficulties of quantum theory itself. In fact, the ideas I will
be exploring are strongly motivated by the arguments that Penrose [1] and Diósi
[2] have been advancing regarding the collapse of the wave function as a dynamical
process to be incorporated in a modified Schrödinger’s equation, and the role that
gravity might play in this.
D. Sudarsky (B)
Institute for Nuclear Sciences, National Autonomous University of Mexico, Apdo.
Postal 70-543, 04510 Mexico, D.F., Mexico
e-mail: sudarsky@nucleares.unam.mx
350 D. Sudarsky
In the present manuscript, and in contrast to other works [3, 4] where I have
focused on the difficulties or shortcomings of the postures advocated in standard
treatments of the emergence of the seeds of cosmic structure, I will be focusing on
aspects that would be encountered independently of the conceptual approach one
takes, as long as one attempts to provide a specific characterization of the various
stages in the cosmological evolution. Moreover, I will show the connection, some-
times not easily recognized, between the actual treatment we have been using, and
other approaches to deal with the characterization of space-time when the matter
content is taken to be described in terms of quantum fields.
2 The General Setting
The first simplification we will be using is justified by the fact that, despite the many
important and sometimes spectacular advances, at this point in time we still do not
have a fully workable and completely satisfactory theory of quantum gravity. For
instance, we do not know how to construct a quantum state representing Minkowski
space-time. In fact, it is well known that any canonical approach to quantum gravity
inevitably leads to a timeless theory where the recovery of fully covariant space-
time notions becomes, by itself, a nontrivial task. Those approaches usually require
selecting a physical observable to play the role of a clock, and while this can be
achieved quite satisfactorily in certain cases, the resolution of the problem in full
generality is not available.
Therefore, although we will adhere to the view that the fundamental description
of everything, including space-time, ought to be always quantum mechanical, we
will be using a classical description of the space-time metric.
In general, we will view the so called classical regimes in connection with some
physical variable as those where such quantities can be described to a sufficient accu-
racy by their classical counterparts representing the corresponding quantum expec-
tation values. The paradigmatic example here is provided by the coherent states of
a harmonic oscillator which correspond to minimal wave-packets with expectation
values of position and momentum following the classical equations of motion. In the
specific case of the space-time, we will accept that, at the fundamental level, it would
have a quantum description in terms of some unspecified variables (they might be
those of loop quantum gravity (LQG), the “causal set” approach, or the “dynamical
triangulations” approach, etc) but we will be characterizing them, according to this
view, using effectively classical terms. We might consider such description in analogy
with the hydrodynamical characterization of a fluid: as representing a good enough
description at certain scales, but having a radically different description at a more
fundamental level. Einstein’s equations would correspond to the Navier-Stokes equa-
tions, the space-time metric to the fluid velocity and density fields, and the atomic
and molecular characterization of the matter making the fluid would correspond to
the fundamental degrees of freedom of quantum gravity.
The Inflationary Origin of the Seeds of Cosmic Structure 351
According to this view, Einstein’s equations would be of limited validity and

there would be conditions where they will not hold. In fact, in such situations one
can expect a more general failure of the characterization of the situation in terms
of a space-time metric, just as, in the case of a fluid, one might expect not only the
violation of Navier-Stokes equations but also of the hydrodynamic characterization
of matter, when something like a phase transition from liquid to gas is taking place.
There is one example coming from LQG where one faces precisely this kind of
breakdown: It corresponds to the situation where a collapsing star forms a black hole
that eventually evaporates completely [5]. In that work, it is argued that, although
the classical singularity would be resolved in terms of the LQG degrees of freedom,
the region of space-time corresponding to the singularity is characterized, in terms
of the fundamental variables, by a situation that has no metric counterpart.
On the other hand, the situations I want to consider are those that require a full
quantum treatment of the matter fields. For this, we will rely on the standard quantum
field theory in curved space-time treatments, such as described for instance in [6].
The setting is therefore that of semiclassical gravity where Einstein’s equations read:
G με = 8γ G≈T̂με ⊗ (1)
and matter is described in terms of states of a quantum field, which, in our case, will
be the scalar field of inflationary cosmology.
The validity of such semiclassical treatment is, as we have indicated, of limited
scope, and would require, among other things, the scalar curvature of space-time to
be small compared to l −2p (l p is the Planck length). The inflationary regime is thought
to be associated to scales that are way below the Planck mass and thus this part of
the requirement should be easily satisfied in the case we want to consider.
3 The Stern-Gerlach Analogy

The general issue we want to consider is the emergence of the seeds of quantum
structure from the quantum fluctuations of the inflaton vacuum. The aspect that will
be guiding our inquest is the required change in symmetry between the conditions
characterized by a classical and unperturbed Robertson Walker space-time, where
the inflaton field’s zero mode is slowly rolling down the potential, and where the other
modes are in the the Bunch-Davies vacuum, to a stage characterized by a slightly
perturbed RW space-time containing small anisotropies and inhomogeneities, and a
quantum state of the inflaton where the expectation value of the energy momentum
tensor has the corresponding anisotropies and inhomogeneities.
The question is how to characterize the evolution in time (because, after all,
emergence is a word that has very clear time connotations1 ), from a situation corre-
sponding to a homogeneous and isotropic (H&I) background and a quantum aspect
characterized by a H&I state, to a stage lacking such symmetries. This issue has
1 Something emerges when it is not present at a certain time but it is present at a later time.
352 D. Sudarsky
been the central focus of the discussion in previous works: Is there a measurement
involved? Can we account for it using just decoherence? Should we rely on the many
worlds interpretation (MWI) or must we call upon a novel gravity-induced collapse
of the wave function following the ideas of Penrose and Diósi? I have extensively
argued that among those the only option is the last one, however, here I will focus
on aspects that should be dealt with, even if one is intent on sticking with some of
the alternative views mentioned above.
In order to clarify the issue, I shall consider a much simpler problem, and point
to the parallels with the inflationary problem, as well as to those aspects where the
analogy breaks down.
Consider a standard Stern-Gerlach experiment: The setting involves an electron
that has been prepared moving along the x axis from the x < 0 region, towards a
magnet placed at the origin of coordinates. The inhomogeneous magnetic field points
along the y axis so that the electron will be diverted towards the +y or −y directions
depending on whether the spin state of the electron is |+⊗ or |−⊗ (we are taking the
basis to be that of eigen-states of the spin along the y axis). If the spin was prepared
initially in the direction +x (eigenstate of the x component of the spin) we know that
there is a 50 % probability that the electron will be diverted towards the +y direction.
Let us imagine for a moment that we do not fully understand the theory, that
there are aspects of the electromagnetic interaction that still elude us (the allegory
here alludes to the quantum theory of gravity, of course). Now, let us consider the
theoretical analysis of the said experiment: If we do not invoke any sort of reduction,
or collapse of the wave function, the result of the unitary evolution will be a state
that corresponds to a superposition of the electron going up and the electron going
down. However, suppose we want to investigate in depth what happens when the
electron is deflected: Say, we want to understand exactly the details of the momentum
transferred from the magnetic field to the electron. We could, for instance, find that
the momentum transfer has components predominantly in the Y direction (depending
on the deflection, with a sign that depends on the alternative) but is accompanied by
momentum components in the X direction (of a specific and correlated magnitude),
and use this to study the change in the kinetic energy. We can even inquire about
the rate of transfer as the electron moves along the X direction (by considering
an appropriate wave packet characterization of the electron and, say, following the
expectation value of the center of mass). How can we do that if we maintain that the
electron is, even after the passing through the magnet, in the superposition of moving
up and moving down? In that case, if we try to compute the momentum transfer from
the EM field to the electron we will find that it is zero.
Suppose we want to further inquire about the back reaction of the electron on the
EM field. It would seem very difficult to do so without incorporating the collapse.
Note that, if we are so inclined, we could even adopt the many worlds interpretation
(MWI) but still concern ourselves with one of the realizations of the electron’s path,
that which corresponds to “our branch of the many worlds”. Now, suppose we wanted
to do this before we had a fully workable quantum theory of the electromagnetic field,
but instead we had very refined experimental data about the back reaction acting on
the magnet as a result of the electron scattering. Could we not hope to investigate
some of the properties of the quantum electromagnetic field using a combination

of the data, some rough classical characterization of the EM field, taken to be only
valid to a certain degree (evidently not in the full description of the back reaction,
but, perhaps, as it applies to the “after” and “before” state of the EM field)? Could we
hope to do that if we had never even been able to consider the back reaction, as a result
of our failure to acknowledge that the full superposition (in which the expectation
value of the momentum transfer was zero) was not the appropriate description? It
seems we could start considering something like the dispersion of the momentum
transfer but, that would be very difficult due to our lacking of a workable quantum
theory of Maxwell’s field.
The situation we face regarding the problem of the emergence of seeds of structure
in inflationary cosmology is, in a sense, analogous to the one above: the symmetry
of homogeneity and isotropy in the cosmological case has in the example above a
simple counterpart: the symmetry y ◦ −y.
The most important aspect where the analogy breaks down is the fact that in
contrast with the Stern-Gerlach example above, in cosmology we can not call upon
external observers, and, what is even worse, the emergence of structure, which is
what we want to explain, is a prerequisite for the subsequent emergence of anything
one can consider as an “observer”.
The issue of symmetry is brought in because, as we all know, when dealing with
complicated problems, symmetry arguments are often one of the few paths available
to arrive to clear and definite conclusions, and thus they provide the only hope to
make progress.
4 A Word About Collapse Theories
The idea of modifying quantum theory by adding to it a mechanism for explicit

dynamical reduction has a long history. The existing work in this direction includes:
GRW [7], Pearle [8], Diósi [2], Penrose [1], Bassi [9] (where its worthwhile noting
some recent advances towards making collapse theories compatible with special
relativity [10, 11]), and recently Weinberg [12].
As an example, let us consider the modification of the Schrödinger equation that
underlies the Continuous Spontaneous Localization (CSL) theory, developed in [13]:
ν2
d|ψ⊗ = −{[i Ĥ − ( Â − ≈ψ| Â|ψ⊗)2 ]dt + ν( Â − ≈ψ| Â|ψ⊗)dWt }|ψ⊗, (2)
2
where Wt is a Wiener process (Wt2 = t).

Its merit is that it includes the unitary Schrödinger evolution U and the nonde-
terministic, non-unitary reduction process R (for measuring Â) in a unified fashion.
The proposal for particles assumes that the fundamental localization takes place in
354 D. Sudarsky
position or configuration space, thus Â = X̂. The value of the parameter ν is taken
to be small enough so that particle physics is not strongly affected, but large enough
so that it leads to a rapid localization of macroscopic objects.
In the reminder of this manuscript, we will not follow any of those proposals,
but take from them only some essential aspects. What we want to do, at this point,
is to present a generic formalism capable of treating the cosmological problem that
motivates this line of research, within the context of semiclassical gravity with a
full quantum treatment for the inflaton field (including its zero mode), incorporating
a collapse process. On the other hand, we should mention the work [14], where
precisely a version of CSL has been adopted to this problem, as well as an ongoing
research project by our group in which a different implementation of those ideas is
studied [15].
5 The Self Consistent Semiclassical Configurations
In this section, we will describe the precise formalism that we consider as appropriate
to characterize, at the desired level, the situation we will be studying. As we said, the
setting is that of semiclassical gravity and quantum field theory in curved space-time.
Such setting is often considered as a context in which the space time is fixed and
given, and where the quantum fields can, at most, produce some small modifications
which are referred to as the back reaction of space-time to the effects of the quantum
fields. That point of view will not be sufficient for our purposes, as we want to be
able to, in principle, treat the case where the only matter content is represented by
the inflation field, and where the space-time is fundamentally tied to its properties:
Inflation is supposed to be the result of the non vanishing value of the inflation
potential. In such situation, the field is generically treated at a classical level, and
only its perturbations are quantized. We want to be able to explore the setting in
which the classical quantum partition in the description is, in principle, not tied to a
perturbative treatment.2
We have formalized these ideas in [16] based on the notion of the “Self Consistent
Semiclassical Configurations” (SSC) provided by the following,
Definition 1 The set gμε (x), λ̂(x), γ̂ (x), Hˆ , |Φ ⊗ √ Hˆ represents a SSC if and
only if λ̂(x) and γ̂ (x) correspond to quantum field operators over the Hilbert space
and Hˆ is constructed acceding to the standard QFT over the curved space-time with
metric gμε (x) (as described in, say [6]), and the state |Φ ⊗ in Hˆ is such that:
G με [g(x)] = 8γ G≈Φ |T̂με [g(x), λ̂(x), γ̂ (x)]|Φ ⊗ (3)
for all the points x in the space-time manifold.
2 The point is that a perturbative treatment should be considered as an approximation to something

else, and it is very useful when one can establish explicitly what is that which the perturbative
treatment is approximately trying to describe.
It is, in a sense, the GR version of Schrödinger-Newton equation [17] where one

considers the Schrödinger equation for the wave function ψ of a particle subject to
the gravitational interaction described in terms of a Newtonian potential α N :
∂ψ 1 2
i =− ∗ ψ + Mα N ψ, (4)
∂t 2M
where the wave function of the particle is taken as a gravitating mass distribution,
therefore
∗ 2 α N = 4γ G M|ψ|2 . (5)
The non linearity implied by these equations is known to lead to interesting and
suggestive behavior [18].
6 Collapse
The point, however, is that this setting will not, by itself, be enough to describe the
situations involving a collapse of the wave function. As we have argued, in order to
be able to describe the evolution in time from an early inflationary era characterized
by an H&I situation to a later regime characterized by a situation that is not H&I, we
will be relying on the collapse of the wave function, represented here in the simplest
fashion: an instantaneous jump in the state of the quantum field.
The collapse process reflecting some remanent signature from a fundamental
quantum gravity regime (as suggested by Penrose and Doisi’s ideas) will be described
here as an instantaneous jump: i.e., besides the standard smooth unitary evolution of a
quantum field characterized by the Schrödinger’s dynamics (the so called U process,
which in the QFT theory setting we are considering here, is usually incorporated
using the “Heisenberg picture”), there are, sometimes, spontaneous jumps in the
quantum state:
. . . |0⊗k1 ⇔ |0⊗k2 ⇔ |0⊗k3 ⇔ . . . . ◦ . . . .|β ⊗k1 ⇔ |0⊗k2 ⇔ |0⊗k3 ⇔ . . . . (6)
We might view the collapse as triggered by an aspect of the dynamics which is not
susceptible of description in standard Hamiltonian terms, but which is nonetheless
taken as an interaction treated here in the interaction picture. That is, we take the
Hamiltonian part of the evolution and absorb it in the quantum field operators as in
the standard Heisenberg picture, but we leave the reminder, viewed as an interaction,
to be treated using the interaction picture.
356 D. Sudarsky
7 Relation to Other Approaches
This setting might seem very novel and unusual, however, the fact is that it can be
seen to lie, unsuspectedly, underneath some more conventional approaches, such as
the stochastic gravity formalism [19].
In order to see this, let us consider one of such jumps or collapses: |ψ(t)⊗ =
ω (t0 − t)|0⊗ + ω (t − t0 )|Φ ⊗, and its gravitational effects.
Now Einstein’s semiclassical equations read:
G με = 8γ G≈ψ(t)|T̂με |ψ(t)⊗ (7)
which we can write as:
G με = 8γ G≈0|T̂με |0⊗ + 8γ GΦμε , (8)
where
Φμε ≤ ω (t − t0 )(≈Φ |T̂με |Φ ⊗ − ≈0|T̂με |0⊗) (9)
might be seen as corresponding to an individual stochastic step. Stochastic gravity

might correspond to a continuous version of dynamical collapses (like CSL).
Note: the equation can not be valid on the jump, but might well be so before
and after. We take the view, motivated in part by the black hole singularity example
in LQG, that during the jump the degrees of freedom of the quantum space-time
are excited. In the fluid analogy, this might be thought as corresponding to some
chemical reaction or phase transition occurring in the fluid. It is clear that during
such processes, which generally involve energy flux between the atomic or molecular
degrees of freedom to the macroscopic degrees of freedom characterized in terms
of the fluid variables, the Navier-Stokes equations can not be valid. If, however, the
phase transition takes place rapidly, one can assume such equation to be valid before
and after the chemical reaction or phase transition.
Next, consider the inflationary problem at hand, and assume one adopts one of
the more popular postures regarding the emergence of classicality, or more precisely,
the generation of primordial inhomogeneities and anisotropies. These include, for
instance (i) the notion that after a given mode exits the horizon (its physical wave-
length as seen in a co-moving frame, becomes larger than the Hubble radius) the
fluctuation corresponding to that mode becomes classical, or (ii) that, due to some
decoherence effect, we can at a certain point adopt the Many Worlds Interpretation
of quantum theory, and consider the state of the quantum field as characterizing not
our universe, but an ensemble of universes of which ours is just a typical element.
Now, let us say that in case (i) we want to produce a description (even an approximate
one) of our universe concentrating, for simplicity, on a single mode, but we want
a description that is valid before the mode exits the horizon and afterwards. In that
case, the approach I will present, seems to be the best one can do, as long as we do not
have a workable theory of quantum gravity which allows us to characterize space-

time in a full quantum language. In case (ii) we might also be interested in putting
together the characterization of our space-time before the decoherence is taken to
be effective, and the one describing the particular branch of the many worlds, or the
particular element of the enabled universes in which we happen to find ourselves.
Again, in that situation the analysis I will present would offer perhaps the furthest
one can go in achieving the said goal, given the present stage of the development of
candidate theories of quantum gravity.
8 Application to Inflation
As we discussed in the previous sections, space-time will be treated as classical. In

the case of interest, working in a specific gauge, and ignoring the tensor perturbations
the metric is taken as:

ds 2 = a 2 (ρ) −(1 + 2ψ)dρ2 + (1 − 2ψ)Θi j d x i d x j , ψ(ρ, x) ≡ 1 (10)
with a(ρ) the scale factor and ψ(ρ, x) representing (to the extent that it is nonzero)
a possible slight departure from homogeneity and isotropy of the space-time, the so
called Newtonian potential. We will use the notation H ≤ a −1 da/dρ (not to be
confused with the hatted quantity that stands for a Hilbert space).
Also, as explained before, the
scalar field, which we take here to be described
by the simple action S = 1/2 d 4 x(∗μ τ∗ μ τ − m 2 τ 2 ), including the zero mode
(which in standard discussions of inflation is usually treated at a classical level) is
treated here using quantum field theory on curved space times, so we write:

τ̂(x) = âξ u ξ (x) + âξ† u ∇ξ (x) , (11)
ξ
with the functions u ξ (x) a complete set of normal modes, orthonormal w.r.t. the
symplectic product:

((τ1 , γ1 ), (τ2 , γ2 ))Sympl ≤ −i τ1 γ2∇ − γ1 τ2∇ d 3 x. (12)
ζ
For simplicity, we set the problem in a co-moving coordinate box of size L.

Finally, one constructs the state such that Einstein semiclassical equations hold.
This is nontrivial, but ts is a well defined problem. In what follows, the discussion will
omit some complications that are required for the rigorous analysis, and discussed
in detail in [16], but which are not central to the issue at hand. They have to do with
the hermiticity of the operators that play a central role in the collapse, an issue that
will be overlooked here to avoid nonessential complications in the presentation.
358 D. Sudarsky
8.1 The Homogeneous and Isotropic Case: SSC I
We assume an almost de Sitter slow-roll expansion characterized by the parameters

H0(I ) and δ (I ) (using standard inflationary notation [20]).
The quantum field theory construction requires a complete set of modes, which
(I) (I)
we take to be of the form u k (x) = vk (ρ)eik·x /L 3/2 .
The field equation of motion then leads to:

(I) (I) (I)
v̈k + 2H (I) v̇k + k 2 + a 2(I) m 2 vk = 0, (13)
for modes, which must be normalized according to
= ia −2(I) .
(I) (I)∇ (I) (I)∇
vk v̇k − v̇k vk (14)
For the modes with k ⇒= 0, the most general solution to the evolution equation is a
(1) (2)
linear combination of the functions: ρ3/2 Hε (−kρ) and ρ3/2 Hε (−kρ), (the Hankel
(I)
functions of first and second kind), with (ε (I) )2 = (9/4)−(m/H0 )2 . However, these
functions are not well behaved at the origin and thus the zero mode is not included.
For k = 0 the general solution to the equation is a linear combination of the functions
ρ(3−2ε)/2 and ρ(3+2ε)/2 . The choice can be made arbitrarily provided it has a positive
symplectic norm. We take:
(I) i
(I) 3
2 2(I)
(I) m /3H0
v0 (ρ) = (I)
1− −H0 ρ −H0 ρ . (15)
H0 6
For the k ⇒= 0 modes, we make the Bunch-Davies choice: i.e., we use modes that,
in the asymptotic past, behave as purely “positive frequency solutions”. This fixes
Hˆ (I) as the Fock space of the SSC-I construction.
To complete the SSC construction we still need to find a state|Φ (I) ⊗ √ Hˆ (I) such
that its expectation value for the energy-momentum tensor leads to the desired nearly
de Sitter expansion. Consider a state in which all the modes with k ⇒= 0 are in their
vacuum state, while the zero mode is excited in a coherent state:
(I) (I)†
|Φ (I) ⊗ = ceΦ0 â0
|0(I) ⊗, (16)
Using Einstein’s equations for the metric with a vanishing Newtonian potential,
and the fact that for a coherent state we have
≈Φ (I) | : (τ (I) )2 : |Φ (I) ⊗ = (≈Φ (I) |(τ (I) )|Φ (I) ⊗)2 , (17)
one finds that the expectation value of the field should satisfy:

(I)
δ (I) m 2 /3(H0 )2
≈Φ (I) |(τ (I) )|Φ (I) ⊗ ∈ ρ . (18)
(I)
On the other hand, taking the parameter Φ0 as real we find:
m 2 /3H 2(I)
(I)
2Φ0 (I) 0
≈Φ (I) |τ̂ (I) (x)|Φ (I) ⊗ = (I)
−H0 ρ . (19)
L 3/2 H0
That is, we will have compatibility if we set:
(I)
m2 (I) (Φ0 )2
δ (I) = 2(I)
, H0 = 16γ Gδ (I) . (20)
3H0 L3
This completes the explicit SSC -I construction representing an H&I state and space-
time metric, corresponding to the early stages of inflation.
Next, we want to consider a situation where the universe is no longer H&I but
has been excited in the k0 mode: We will denote this new SSC by SSC-II.
8.2 A Simple Inhomogeneous and Anisotropic Case: SSC II
(II)
It will be characterized by the parameters H0 and δ (II) (which might, in principle,
differ slightly from those corresponding to the SSC-I discussed in the previous
section), and a Newtonian potential described by an (in principle) arbitrary function
ψ(ρ, x) = ε P(ρ)Cos(k0 .x), where P(ρ) is an (in principle) arbitrary function, and
ε is a small (expansion) parameter (please do not confuse with δ).
The strategy: We first construct the “generic” Hilbert space assuming that P(ρ)
is given. Then, make an “educated” guess for the form of the quantum state, and by
requiring that our construction be a SSC we will find what the function P(ρ) ought
to be.
The first step is to find the complete set of modes, which we write as:
1
[v (ρ) eik·x + ε(Θvk (ρ) ei(k−k0 )·x + Θvk (ρ) ei(k+k0 )·x )]
(II) (II)0 (II)− (II)+
u k (x) =
L 3/2 k
(21)
to the zeroth order in ε, the evolution equation is given by

(II)0 (II)0 (II)0
v̈k + 2H (II) v̇k + k 2 + a 2(II) m 2 vk = 0, (22)
with normalization condition
= ia −2(I) ,
(II)0 (II)0∇ (II)0 (II)0∇
vk v̇k − v̇k vk (23)
360 D. Sudarsky
(II)0
which is identical to the construction we have already done. Thus, we take the vk (ρ)
as before.
At first order in ε the corresponding evolution equation takes the form

(II)± (II)± (II)±
Θ v̈k + 2H (II) Θ v̇k + (k ± k0 )2 + a 2(II) m 2 Θvk =Fk (ρ) (24)
where

(II)0 (II)0
Fk (ρ)≤ 4v̇k Ṗ − 2 2k 2 + a 2(II) m 2 vk P. (25)
The normalization condition (needed only at one time) is:
(II)0∇ (II)+ (II)0∇ (II)+ (II)0 (II)−∇ (II)0 (II)−∇

v̇k+k0 Θvk − vk+k0 Θ v̇k − v̇k Θvk+k0 + vk Θ v̇k+k0 (26)

(II)0 (II)0∇ (II)0 (II)0∇
= 4 vk v̇k+k0 − v̇k vk+k0 P. (27)
If we had P(ρ) and the initial conditions for the Θvk , the equation above would define
a unique solution. As we said, we will assume that P(ρ) is given and take the initial
conditions to be
(II)± (II)± (II)0
Θ v̇k (ρc ) = 0, Θvk (ρc ) = 4vk (ρc )P(ρc ). (28)
This finishes the generic (i.e. for arbitrary P) construction of Hˆ (II) .

Next, we need to find the state |κ (II) ⊗ √ Hˆ (II) that completes the SSC-II construc-
tion. The symmetries of the space-time led us to consider the “ansatz”:
(II) (II) (II) (II) (II)

|κ (II) ⊗ = . . . |κ−2k0 ⊗ ⇔ |κ−k0 ⊗ ⇔ |κ0 ⊗ ⇔ |κk0 ⊗ ⇔ |κ2k0 ⊗ . . .. (29)
The vector in Fock space is characterized by the specific modes that are excited (all
other modes are assumed to be in the vacuum of the corresponding oscillator) and
(II)
the parameters κk indicate the coherent state for the mode k.
The expectation value of the field operator in such a state is given by

τκ (x) = τκ,0 (ρ) + Θτκ,k0 (ρ)eik0 ·x + Θτκ,2k0 (ρ)ei2k0 ·x + · · ·.
(II) (II) (II) (II)
(30)
(II)
We note that the coefficients Θτκ,nk0 (ρ) have a contribution from the modes nk,
(II)
(n − 1)k and (n + 1)k. We set Θτκ,nk0 (ρ) = 0 for all n ⊥ 2, simply by imposing the
(II) (II) (II)
required relations between the parameters κ±k0 , κ±2k0 , κ±3k0 , etc. It is easy to see
(II) (II)
that |κ±nk0 | ⇐ εn |κ0 |.
The conditions above ensure that there are no terms in e±ink0 ·x (with n ⊥ 2)
appearing in the expectation value of the energy-momentum tensor. That is necessary
for the compatibility of our state ansatz with the semiclassical Einstein’s equations.
We studied these in detail up to to the first order in ε.
The zero order equations are identical to those we found in constructing the SSC-I.
They fix the construction of SSC to the lowest order, i.e. they determine the relation
(II)
between a (II) and κ0 .
Considering the next order one obtains after a lengthy calculation the key result,
that enables us to carry out the construction in a complete manner: That the equations
can be combined into a simple dynamical equation for the Newtonian potential, which
is independent of the first order quantities and where, at the level of precision we are
working at, the equation above becomes simply:

P̈ + δ (II) H (II) Ṗ + k02 − δ (II) H 2(II) P = 0. (31)
The general solution

1 1
P(ρ) = C1 ρ 2 [1+δ
(II) ]
Jξ (−kρ) + C2 ρ 2 [1+δ
(II) ]
Yξ (−kρ), (32)
where Jξ (−kρ) and Yξ (−kρ) are the Bessel functions of the first and second kind,
ξ = [1 + 3δ (II) ]/2.
Einstein’s equations lead, as is well known, to constraints which, at this order pro-
vide relations involving the initial values that would determine the specific solution
P(ρ):
≥
P 4γ Gδ (II) H (II)
ε = 2
Ṗ k0 − H 2(II) δ (II)
(II)
(3H (II) − am 3/δ (II) ) 1 Θτκ,k
× · (II)
0 .
(am 3/δ (II) H (II) − k02 + (δ (II) − 3)H 2(II) ) −H (II) Θ τ̇κ,k
0
(33)
(II) (II)
Thus, given Θτκ,k0 (ρc ) and Θ τ̇κ,k0 (ρc ), we have a completely determined space-
time metric. In particular, we have a completely determined function P(ρ) and thus,
as discussed around Eq. (26), a completely specified set of mode functions for the
expansion of the field operator. Furthermore, those determine the state parameters
κk0 (and thus the rest as well). Thus, we have a complete SSC-II (to this order in ε).
8.3 The Collapse: Joining SSC-I and SSC-II
Next, we want to consider a space-time that includes a collapse. That is, a space-time
that results from the matching of the two constructions. We will consider here that the
collapse corresponds to a hypersurface that is matched to the hypersurfaces ρ = ρc
362 D. Sudarsky
of SSC-I and SSC-II. Note that this gives such hypersurface ζc a preferred status
in the resulting space-time, and is not something to be thought as related to a gauge
freedom: To the past of that hypersurface ζc the space-time is H&I, and to the future
it is not. We will assume here the induced metric is continuous on ζc . This requires
P(ρc ) = 0 and thus

(II) (II)
(3H (II) − am 3/δ (II) )Θτκ,k0 (ρc ) + Θ τ̇κ,k0 (ρc ) = 0, (34)
and therefore

(II)
ε Ṗ = − 4γ Gδ (II) H (II) Θτκ,k0 (ρc ), (35)
which indicates a discontinuity in the extrinsic curvature of the hypersurface ζc .

Assume that the collapse is characterized by a loose analogy with “an imprecise
(I)
measurement” (of the operators τ̂k0 (ρ)) in standard QT: Before the collapse, the
(I)
operator had zero expectation value but an uncertainty φτ̂k0 (ρc ), and thus we assume
that after the collapse, the new expectation value will fall in that range. We thus
consider what would be the energy momentum tensor computed using a state that
results from such measurement (on the SSC-I side of ζc ) and demand that the state
on the SSC-II side be such that the energy momentum tensor on the SSC-II side of
ζc be exactly that. The final result is then:

2
(II) (I) (I) (I) −1
Θτκt ,k0 (ρc ) = x k0 ≈0k0 | φτ̂k0 (ρc ) |0k0 ⊗ ≈ xk0 a(ρc )
2k
with xk0 a random number taken from a distribution characterized by a Gaussian

function centered at zero with unit-spread. A choice of the random number xk0 then
determines the SSC-II.
Thus, we have a well defined framework where one could, in principle, carry out
all the analysis of the collapse approach to the inflationary origin of the seeds of
cosmic structure.
9 Phenomenological Studies
In a realistic situation, we need to consider a collapse of not just one, but of all
the modes. Thus, in contrast with the previous analysis, we have adopted for this
purpose a simplified treatment that makes the realistic problem manageable. We
note that considering simultaneously all modes is required if we want to compare
the theory with observations.
In this simplified treatment, one avoids the complications of the previous treatment
by ignoring the multiplicity of Hilbert spaces and considering the quantum states
that result from collapses to be elements of the Hilbert space based on the initial
homogeneous and isotropic space-time (as is done in standard treatments). We thus

split the treatment into that of a classical homogeneous (“background”) part, and an
in-homogeneous part (“fluctuation”), i.e. g = g0 + Θg, τ = τ0 + Θτ.
The background is taken again to be Friedmann-Robertson universe (with vanish-
ing Newtonian potential), and the homogeneous scalar field τ0 (ρ). In the previous,
more precise treatment this would have corresponded to the zero mode of the quantum
field.
The main difference, with respect to the ordinary approach, will be in the spatially
dependent perturbations. Here, our approach indicates we should quantize the scalar
field but not the metric perturbation.
We will set a = 1 at the “present cosmological time”, and assume that the infla-
tionary regime ends at a value of ρ = ρ0 , negative and very small in absolute terms.
Again, in our case the semiclassical Einstein’s equations, at lowest order lead to
∗ 2 ψ = 4γ G τ̇0 ≈Θ τ̇⊗ = s≈Θ τ̇⊗, (36)
where s ≤ 4γ G τ̇0 .
Consider the quantum theory of the field Θτ. In this practical treatment it is conve-
nient to work with the rescaled field variable y = aΘτ and its conjugate momentum
γ = Θ τ̇/a. We decompose the field and momentum operators as:
1 ik·x 1 ik·x
y(ρ, x) = e ŷk (ρ), γ y (ρ, x) = e γ̂k (ρ), (37)
L3 L3
k k
+ †
where ŷk (ρ) ≤ yk (ρ)âk + ȳk (ρ)â−k and γ̂k (ρ) ≤ gk (ρ)âk + ḡk (ρ)â−k . The usual

choice of modes yk (ρ) = ≥1 1 − ρki exp(−ikρ), gk (ρ) = −i k2 exp(−ikρ),
2k
which leads to what is known as the Bunch-Davies vacuum: the state defined by
âk |0⊗ = 0 . At this point it is worthwhile to remind the reader that this state is
translationally and rotationally invariant, as can be easily checked by applying the
corresponding rotation and displacement operators to it. Note also that ≈0| ŷk (ρ)|0⊗ =
0 and ≈0|γ̂k (ρ)|0⊗ = 0 . The collapse will modify the state and thus expectation
values of the operators ŷk (ρ) and γ̂k (ρ).
Next, we specify the rules according to which the collapse happens, and thus the
state |Θ⊗ after the collapse. We assume that after the collapse, the expectation values
of the field and momentum operators in each mode will be related to the uncertainties
of the pre-collapse state (these quantities for the vacuum are not zero).
In the vacuum state, ŷk and γ̂k are characterized by Gaussian wave functions
centered at 0 with spread φyk and φγ y k , respectively.
We will want to consider various possibilities for the detailed form of this collapse.
Thus, for their generic form, associated with the ideas above, we assume that at time
ρkc the part of the state corresponding to the mode k undergoes a sudden jump so
that, immediately afterwards, the state describing the system is such that
364 D. Sudarsky

y
≈ ŷk (ρkc )⊗Θ = xk,1 φ ŷk , ≈γ̂k (ρkc )⊗Θ = xk,2 φγ̂k , (38)
where xk,1 , xk,2 are (single specific values) selected randomly from within a Gaussian
distribution centered at zero with spread one.
Finally, using the evolution equations for the expectation values (i.e. using Ehren-
fest’s Theorem), we obtain ≈ ŷk (ρ)⊗ and ≈γ̂k (ρ)⊗ for the state that resulted from the
collapse for all later times.
9.1 Analysis of the Phenomenology
The semi-classical version of the perturbed Einstein’s equation that, in our case, leads
to ∗ 2 ψ = 4γ G τ̇0 ≈Θ τ̇⊗ indicates that the Fourier components at the conformal time
ρ are given by:
ψk (ρ) = −(s/ak 2 )≈γ̂k (ρ)⊗. (39)
Prior to the collapse, the state is the BD vacuum, and it is easy to see that
≈0|γ̂k (ρ)|0⊗ = 0, so in that situation we would have ψk (ρ) = 0. However, after
the collapse has occurred, we have instead: ψk (ρ) = −(s/ak 2 )≈Θ|γ̂k (ρ)|Θ⊗ ⇒= 0.
From those quantities, we can reconstruct the Newtonian potential (for times after
the collapse):

1 ik·x sU (k) k 1
ψ (ρ, x) = 3 e ψk (ρ) = F(k)eik·x , (40)
L k2 L 3 2a
k k
where F(k) contains, besides the random quantities xk,i , i = 1, 2, the information
about the time at which the collapse of the wave function for the mode k occurs.
We now focus our attention on the “Newtonian potential” on the surface of last
scattering: ψ (ρ D , x D ), where ρ D is the conformal time at decoupling and x D are
co-moving coordinates of points on the last scattering surface corresponding to us as
observers. The quantity is identified with the temperature fluctuations on the surface
of last scattering. Thus:

∇ 2
ξlm = ψ (ρ D , x D )Ylm d π. (41)
The factor U (k) is called the transfer function and represents known physics like
the acoustic oscillations of the plasma. Now, putting all this together we find,
≥
1 U (k) k
ξlm = s F(k)4γi l jl (|k|R D )Ylm (k̂), (42)
L 3 2a k2
k
where jl (x) is the spherical Bessel function of the first kind, R D ≤ ||xD ||, and
k̂ indicates the direction of the vector k. Note that in the usual approaches it is
impossible to produce an explicit expression for this quantity, other than zero.
Thus ξlm is the sum of complex contributions from all the modes, i.e., the equiv-
alent to a two dimensional random walk, whose total displacement corresponds to
the observational quantity. We then evaluate the most likely value of such quantity,
and then pass to the continuum obtaining:

s2 U (k)2 C(k) 2
|ξlm |2M.L . = jl (|k|R D ) k 3 dk. (43)
2γa 2 k4
The function C(k) encodes information contained in F(k). For each model of collapse
it has a slightly different functional form.
It turns out that in order to get a reasonable spectrum, we have one single simple
option: z k must be almost independent of k. That is: ρkc = z/k.
This result shows that the details of the collapse have observational consequences!
In fact, we have
2 1
C(k) = 1 + 2
sin2 φk + sin(2φk ), (44)
zk zk
where φk = kρ−z k , z k = ρkc k with ρ representing the conformal time of observation,

and ρkc the conformal time of collapse of the mode k.
If z k is independent of k, this will not modify the form of the spectrum because
these functions become constants. We can consider simple departures from the pat-
tern ρkc = z/k, say, assuming ρkc = A/k + B. These can now be compared with
observations! We have carried out a preliminary exploration [21] considering the
departures from the HZ spectrum, and a more detailed analysis [22] incorporating the
well understood late time physics (acoustic oscillations, etc.) and comparing directly
with the observational data. Those represent the first limits on a collapse model
coming from the CMB observations. This analysis can now be used to constrain
a more specific version of the collapse theories, particularly those schemes which
indicate specific ranges for the collapse times and the specific operators involved in
the collapse process.
10 Conclusions
We have argued elsewhere extensively about the need to deal with the fact that the
standard accounts for the generation of the primordial seeds of cosmic structure from
quantum fluctuations during inflation are not completely satisfactory. In the present
work, I have focused on the formal implementation of such ideas, and on the fact
that, even if one wanted to ignore the conceptual issues that we have pointed out
in previous works, but at the same time, one wanted to consider the possibility of
366 D. Sudarsky
a maximal characterization of our universe, one would be making a similar kind of

description as that used in the collapse approach we favor.
We have presented a formalism that allows the incorporation of a collapse process
within a semiclassical treatment of gravity interacting with quantum fields. Finally,
I have made a brief overview of the phenomenological analysis that relied on a
simplified treatment, and which made it manageable to consider the realistic situation
involving collapses in all modes of the quantum field.
There is, in fact, an interesting possibility of connection of the ideas presented
here to some appearing in the context of the singularity resolution LQG, where one
expects a failure of the approach to lead to even an approximate characterization, in
terms of classical geometry. In this case, we have found a specific kind of breakdown
of the space-time description at the collapsing hypersurface: A discontinuity in the
extrinsic curvature of such hypersurface. The amount of such discontinuity is related
to the details of the collapse. The investigation of this issue in the context of candidate
theories for quantum space-time would be very interesting.
We believe that one of the best ways to inquire about the interface of quantum and
gravitation is by pushing our attempts to describe space-time in the context where
quantum effects become important. The inflationary situation offers us a unique
opportunity. In order for us to be able to take full advantage of such window into the
unknown, we need to start by recognizing the shortcomings in our current treatments.
We trust that the program here outlined represents the first steps in that direction.
Acknowledgments This work is supported in part by the CONACYT grant No 101712. and by
UNAM-PAPIIT IN107412-3 grant. The author thanks the organizers for the kind hospitality during
the conference in Prague.
References
1. Penrose, R.: On gravity’s role in quantum state reduction. Gen. Relativ. Gravit. 28, 581 (1996).
doi:10.1007/BF02105068
2. Diósi, L.: A universal master equation for the gravitational violation of quantum mechanics.
Phys. Lett. A 120, 337 (1987). doi:10.1016/0375-9601(87)90681-5
3. Perez, A., Sahlmann, H., Sudarsky, D.: On the quantum origin of the seeds of cosmic structure.
4. Sudarsky, D.: Shortcomings in the understanding of why cosmological perturbations look
classical. Int. J. Mod. Phys. D 20, 509 (2011). doi:10.1142/S0218271811018937
5. Ashtekar, A., Bojowald, M.: Quantum geometry and the Schwarzschild singularity. Class.
Quantum Grav. 23, 391 (2006). doi:10.1088/0264-9381/23/2/008
6. Wald, R.M.: Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics.
Chicago Lectures in Physics. University of Chicago Press, Chicago (1994)
7. Ghirardi, G.C., Rimini, A., Weber, T.: Unified dynamics for microscopic and macroscopic
systems. Phys. Rev. D 34, 470 (1986). doi:10.1103/PhysRevD.34.470
8. Pearle, P.: Reduction of the state vector by a nonlinear Schrödinger equation. Phys. Rev. D. 13,
857 (1976). doi:10.1103/PhysRevD.13.857
9. Bassi, A., Ghirardi, G.: Dynamical reduction models. Phys. Rep. 379, 257 (2003). doi:10.1016/
S0370-1573(03)00103-0
10. Tumulka, R.: On spontaneous wave function collapse and quantum field theory. Proc. R. Soc.
London, Ser. A 462, 1897 (2006). doi:10.1098/rspa.2005.1636
11. Bedingham, D.J.: Relativistic state reduction dynamics. Found. Phys. 41, 686 (2011). doi:10.
1007/s10701-010-9510-7
12. Weinberg, S.: Collapse of the state vector. Phys. Rev. A 85, 062116 (2012). doi:10.1103/
PhysRevA.85.062116
13. Pearle, P.: Combining stochastic dynamical state-vector reduction with spontaneous localiza-
tion. Phys. Rev. A 39, 2277 (1989). doi:10.1103/PhysRevA.39.2277
14. Martin, J., Vennin, V., Peter, P.: Cosmological inflation and the quantum measurement problem.
15. Cañate, P., Pearle, P., Sudarsky, D.: CSL wave function collapse model as a mechanism for the
emergence of cosmological asymmetries in inflation. ArXiv e-prints arXiv:1211.3463 [gr-qc]
(2012)
16. Diez-Tejedor, A., Sudarsky, D.: Towards a formal description of the collapse approach to the
inflationary origin of the seeds of cosmic structure. J. Cosmol. Astropart. Phys. 2012(07), 045
(2012). doi:10.1088/1475-7516/2012/07/045
17. Diósi, L.: Gravitation and quantum-mechanical localization of macro-objects. Phys. Lett. A
105, 199 (1984). doi:10.1016/0375-9601(84)90397-9
18. van Meter, J.R.: Schrödinger-Newton ‘collapse’ of the wavefunction. Class. Quantum Grav.
28, 215013 (2011). doi:10.1088/0264-9381/28/21/215013
19. Hu, B.L., Verdaguer, E.: Stochastic gravity: theory and applications. Living Rev. Relativ. 11(3),
lrr-2008-3 (2008). http://www.livingreviews.org/lrr-2004-3
20. Mukhanov, V.: Physical Foundations of Cosmology. Cambridge University Press, Cambridge
(2005)
21. de Unánue, A., Sudarsky, D.: Phenomenological analysis of quantum collapse as source of the
seeds of cosmic structure. Phys. Rev. D 16, 043510 (2008). doi:10.1103/PhysRevD.78.043510
22. Landau, S.J., Scóccola, C.G., Sudarsky, D.: Cosmological constraints on nonstandard infla-
tionary quantum collapse models. Phys. Rev. D 85, 123001 (2012). doi:10.1103/PhysRevD.
85.123001
Quantum Gravity: The View From Particle
Physics
Hermann Nicolai
Abstract This lecture reviews aspects of and prospects for progress towards a theory
of quantum gravity from a particle physics perspective, also paying attention to recent
findings of the LHC experiments at CERN.
1 Introduction
First of all I would like to thank Jiří Bičák for inviting me to this prestigious conference
in commemoration of Einstein’s stay in Prague a 100 years ago. Although it was only
a short stay, as Einstein left Prague again after little more than 1 year, it was here
that he made major progress towards the final version of General Relativity, and
surely the beauty of this city must have played an important inspirational part in this
endeavor.
In view of the more general nature of this conference, I have decided not to give a
technical talk on my current work, but rather to present some thoughts on the state of
quantum gravity from the point of view of a particle physicist, but with an audience
of general relativists in mind. Taking such a point of view is quite appropriate, as
LHC is about to end its first phase of experiments, with the solid evidence for a scalar
boson that has all the requisite properties of a Higgs boson as the main outcome so
far. This boson was the final missing link in the Standard Model of Particle Physics
(or SM, for short), and therefore its discovery represents the final step in a story that
has been unfolding for almost 50 years. Equally important, as the CERN experiments
continue to confirm the Standard Model to ever higher precision, with (so far) no
indications of ‘new physics’, it is also a good time to ask whether these results can
H. Nicolai (B)
Max Planck Institute for Gravitational Physics (Albert Einstein Institute),
Am Mühlenberg 1, 14476 Golm, Germany
e-mail: nicolai@aei.mpg.de
370 H. Nicolai
possibly offer any insights into quantum gravity. So my main message will be that we
should not ignore the hints from particle physics in our search for quantum gravity!
I do not think I need to tell you why a theory of quantum gravity is needed, as some
of the key arguments were already reviewed in other talks at this conference. There
is now ample evidence that both General Relativity (GR) and Quantum Field Theory
(QFT) are incomplete theories, and both are expected to break down at sufficiently
small distances. The generic occurrence of space-time singularities in GR is an
unavoidable feature of the theory, indicating that classical concepts of space and
time must be abandoned at distances of the order of the Planck scale. Likewise, there
are indications of a breakdown of conventional QFT in this regime. Accordingly,
and in line with the title of this lecture, I would therefore like to concentrate on the
lessons from particle physics pointing beyond QFT and conventional concepts of
space and time.
In its current incarnation, QFT mainly relies on perturbation theory. The ultraviolet
(UV) divergences that inevitably appear in higher order Feynman diagrams require
a carefully crafted procedure for their removal, if one is to arrive at testable pre-
dictions. This renormalization prescription in essence amounts to an order by order
tuning of a finite number of parameters by infinite factors. Although mathematically
on very shaky grounds, this procedure has produced results in stunning agreement
with experimental findings, with a precision unmatched by any other scheme in the
physical sciences. The most famous example is, of course, the QED prediction of
the anomalous magnetic moment of the electron, but the agreement between very re-
cent precision measurements at LHC and the theoretical predictions of the Standard
Model is now equally impressive. Yet, in spite of this extraordinary success there is
good reason to believe that neither the SM in its present form nor any of its quantum
field theoretic extensions (such as the supersymmetric versions of the SM) are likely
to exist in a strict mathematical sense. The ineluctable conclusion therefore seems to
be that the UV completion of the SM requires something beyond QFT as we know it.
The difficulties in both GR and conventional QFT have a common origin. In both
frameworks space-time is assumed to be a continuum, that is, a differentiable mani-
fold. As a consequence, there should exist no obstacle of principle in going to arbi-
trarily small distances if either of these theories were universally valid. Nevertheless,
the very nature of quantum mechanics suggests that its principles should ultimately
also apply to space-time itself, whence one would expect the emergence of a grainy
structure at the Planck scale. Indeed, and in spite of their disagreements, almost all
approaches to quantum gravity1 are united in their expectation that something dra-
matic must happen to space-time at Planck scale distances, where the continuum
should thus give way to some kind of discretuum.
A second, and related, source of difficulties is the assumption that elementary
particles are to be treated as point-like excitations. And indeed, there is not a shred of
a hint so far that would point to an extended structure of the fundamental constituents
1 With the possible exception of the Asymptotic Safety program [1].

Quantum Gravity: The View From Particle Physics 371
of matter (quarks, leptons and gauge bosons), so this assumption seems well sup-
ported by experimental facts. Nevertheless, it is at the root of the ultraviolet infinities
in QFT. Moreover, it is very hard to do away with, because the point-likeness of
particles and their interactions seems to be required by both relativistic invariance
and locality/causality—building a (quantum) theory of relativistic extended objects
is not an easy task! In classical GR, the very notion of a point-particle is problematic
as well, because any exactly point-like mass would have to be a mini black hole
surrounded by a tiny horizon, and thus the putative point particle at the center would
move on a space-like rather than a time-like trajectory. Again, one is led to the con-
clusion that these concepts must be replaced by more suitable ones in order to resolve
the inconsistencies of GR and QFT.
Current approaches to quantum gravity can be roughly put into one of the two
following categories (for a general overviews see e.g. [2–4]2 ).
• According to the first hypothesis quantum gravity in essence is nothing but the
non-perturbative quantization of Einstein Gravity (in metric/connection/loop or
discrete formalism). Thus GR, suitably treated and eventually complemented by
the Standard Model of Particle Physics or one of its possible extensions, should
correctly describe the physical degrees of freedom also at the very smallest dis-
tances. The first attempt of quantizing gravity relied on canonical quantization,
with the spatial metric components and their conjugate momenta as the canonical
variables, and the Wheeler-DeWitt equation governing the dynamics [3]. Super-
imposing Schrödinger-type wave mechanics on classical GR, this scheme was still
rather close to classical concepts of space and time. By contrast, modern versions of
this approach look quite different, even though their starting point is still the stan-
dard Einstein-Hilbert action in four dimensions: for instance, the discrete structure
that emerges from the loop quantum gravity program relies on holonomies and
fluxes as the basic variables, leading to a discretuum made of spin networks or
spin foams [5, 6].
• According to the opposite hypothesis (most prominently represented by string the-
ory [7–9]) GR is merely an effective (low energy) theory arising at large distances
from a more fundamental Planck scale theory whose basic degrees of freedom
and whose dynamics are very different from either GR or conventional QFT, and
as yet unknown. In this view, classical geometry and space-time itself, as well as
all matter degrees of freedom are assumed to be ‘emergent’, in analogy with the
emergence of classical macroscopic physics from the completely different quan-
tum world of atoms and molecules. Likewise, concepts such as general covariance
and even background independence might only emerge in the large distance limit
and not necessarily be features of the underlying theory. Consequently, attempts
to unravel the quantum structure of space and time by directly quantizing Ein-
stein’s theory would seem as futile as trying to derive microscopic physics by
2 As there is a vast literature on this subject, I here take the liberty of citing only a few representative
introductory texts, where more references can be found.
372 H. Nicolai
Fig. 1 The steady progress of quantum gravity?
applying canonical quantization procedures to, say, the Navier-Stokes equation.

The fundamental reality might then be something like the abstract space of all
conformal field theories, only a small subset of which would admit a geometrical
interpretation. The occasional ‘condensation’ of a classical space-time out of this
pre-geometrical framework would then appear as a rare event.
Pursuing different and independent ideas is certainly a good strategy as long as

we do not know the final answer, but it is a bit worrisome (at least to me) that
the proponents of the different approaches not only base their approaches on very
different assumptions, but continue to speak languages that are foreign to one an-
other. Surely, when zeroing in on the ‘correct’ theory there should be a convergence
of ideas and concepts: when Schrödinger proposed wave mechanics and Heisenberg
formulated matrix mechanics, these were initially regarded as very different, but it
did not take long before it became clear that they were just equivalent descriptions of
the same theory. Unfortunately, at this time there is no such convergence in existing
approaches to quantum gravity—a sign that we are probably still very far from the
correct answer! So let us hope that our noble search will not end like the historic
event in the Breughel painting shown in Fig. 1.
2 The Divergence Problem
From the point of view of perturbative QFT the basic difference between gravity
and matter interactions is the non-renormalizability of perturbatively treated GR.
For instance, at two loops the Einstein-Hilbert action must be supplemented by the
following counterterm cubic in the Weyl tensor [10, 11] 3

(2) 1 209 1 μν
εdiv = d V CμνλΦ C λΦ α∂ Cα∂ , (1)
γ 2880 (16π 2 )2
if the calculations are to produce finite predictions for graviton scattering at this order
(the parameter γ here is the deviation from four dimensions in dimensional regular-
ization, and must be taken to zero at the end of the calculation). At higher orders
there will arise similar infinities that likewise must be cancelled by counterterms of
higher and higher order in the Riemann tensor. Because one thus has to introduce
an unlimited number of counterterms in order to make predictions at arbitrary loop
orders and therefore has to fix an infinite number of coupling constants, the theory
looses all predictive power.
From the non-renormalizability of perturbatively quantized gravity, one can draw
quite different conclusions, in particular reflecting the two opposite points of view
cited above. According to the string/supergravity ‘philosophy’, a consistent quan-
tization of gravity necessarily requires a modification of Einstein’s theory at short
distances, in order to cancel the infinities. This entails the necessity of (possibly
supersymmetric) matter and in particular fermions, thus furnishing a possible raison
d’être for the existence of matter in the world. It was originally thought that the UV
finiteness requirements might single out the unique maximally supersymmetric field
theory—maximal N = 8 supergravity—as the prime candidate for a unified theory
of quantum gravity, but that theory was eventually abandoned in favor of superstring
theory as it became clear that maximal supersymmetry by itself may not suffice to
rule out all possible counterterms. Superstring theory gets rid of the divergences in a
different way, by resolving the point-like interactions of QFT into extended vertices,
relying not only on supersymmetry, but also on a specifically ‘stringy’ symmetry,
modular invariance. Nevertheless, very recent developments [12] have rekindled
the debate whether N = 8 supergravity could, after all, be a purely field theoretic
extension of Einstein’s theory that is UV finite to all orders.
On the other side, one can argue that the UV divergences of perturbative quantum
gravity are merely an artifact of the perturbative treatment, and will disappear upon a
proper non-perturbative quantization of Einstein’s theory. In this view, perturbative
quantization is tantamount to ‘steamrollering Einstein’s beautiful theory into flatness
and linearity’ (R. Penrose): by giving up the core features of GR, namely general
covariance and background independence, one cannot expect to get any sensible
3There is no need here to distinguish between the Riemann tensor and the Weyl tensor, as all terms
containing the Ricci scalar or the Ricci tensor can be absorbed into (possibly divergent) redefinitions
of the metric.
374 H. Nicolai
complete answer. This is the point of view adopted by most of the ‘non-string’
approaches, see e.g. [5, 6]. The concrete technical implementation of this proposal
invokes unusual properties which are very different from familiar QFT concepts; for
instance, the finiteness properties of canonical loop quantum gravity hinge on the
non-separability of the kinematical Hilbert space.4 These features are at the origin
of the difficulties that this approach encounters in recovering a proper semi-classical
limit, and make it difficult to link up with established QFT results. Also for this
reason there is so far no clue from non-perturbative quantization techniques as to
what the detailed mechanism is that could dispose of the divergence (1).
There is a third (and more conservative) possibility that has lately received consid-
erable attention, namely asymptotic safety [1, 14]. This is the proposal that the non-
renormalizability of quantum gravity can be resolved by a kind of non-perturbative
renormalizability, in the sense that there might exist a non-trivial fixed point to which
the theory flows in the UV. Such a behavior would be similar to QCD, which flows to
an asymptotically free theory in the UV, but the UV fixed point action for Einstein’s
theory would not be free, but rather characterized by higher order contributions in the
Riemann tensor. In this case there would be no such thing as a ‘smallest distance’, and
space-time would remain a continuum below the Planck scale. If it works, asymptotic
safety is probably the only way to tame the divergences of perturbatively quantized
gravity without resorting to the cancellation mechanisms invoked by supergravity
and superstring theory. The hypothetical non-perturbative renormalizability of grav-
ity would also have to come to the rescue to resolve the inconsistencies of standard
QFT.
However, independently of which point of view one prefers, it should be clear that
no approach to quantum gravity can claim complete success that does not explain
in full and convincing detail the ultimate fate of the divergences of perturbative
quantum gravity.
3 The Role of Matter
A main point of disagreement between the different approaches concerns the role
of matter degrees of freedom. At least up to now, in modern loop and spin foam
quantum gravity or other discrete approaches ‘matter does not matter’, in the sense
that matter degrees of freedom are usually treated as more of an accessory that can be
added at will once the quantization of pure gravity has been achieved. By contrast, to
a supergravity/string practitioner the matter content of the world must play a key role
in the search for quantum gravity. String theory goes even further, positing that the
graviton is but one excitation (although a very distinguished one) among an infinite
tower of quantized vibrational modes that should also include all the constituents of
4 The non-separability of the kinematical Hilbert space is also a crucial ingredient in proposals to
resolve space-time singularities in loop quantum cosmology [13].
matter, and that all these degrees of freedom are required for the consistency of the
theory.
Perhaps it is fitting at this point to recall what Einstein himself remarked on the
different character of the two sides of his field equations: the left hand side is pure
geometry and beautifully unique, thus made of marble, whereas the right hand side
has no share in this beauty:
1
Rμν − gμν R = β Tμν (2)
2

Timber?
Marble
Indeed, the question that occupied Einstein until the end of his life was this: can we
understand the right hand side geometrically, thereby removing its arbitrariness? Put
differently, is there a way of massaging the right hand side and moving it to the left
hand side, in such a way that everything can be understood as coming from some
sort of generalized geometry?
Over the last 90 years there has been some remarkable progress in this direc-
tion (see e.g. the reprint volume [15]), but we still do not know whether these
ideas really pan out. Already in 1921, T. Kaluza noticed that electromagnetism
(Maxwell’s theory) can be understood as originating from a five-dimensional theory
of pure gravity; later O. Klein extended this proposal to non-abelian gauge interac-
tions. The idea of higher dimensions and of finding a geometrical explanation for
the existence of matter continues to hold fascination to this day, most recently with
the idea of large extra dimensions (whereas the original Kaluza-Klein proposal as-
sumed the extra dimensions to be of Planck size in extension). Supersymmetry and
supergravity may likewise be viewed as variants of the Kaluza-Klein program: they
generalize ordinary geometry by including fermionic dimensions. This leads to the
replacement of ordinary space-time by a superspace consisting of bosonic (even) and
fermionic (odd) coordinates, thus incorporating fermionic matter into the geometry
[16]. Accordingly, the possible discovery of supersymmetric particles at LHC could
be interpreted as evidence of new dimensions of space and time.
There is not so much discussion of such ideas in the ‘non-string’ context, where
neither unification nor extra dimensions feature prominently (the Ashtekar variables
exist only in three and four space-time dimensions), and the focus is on Einstein
gravity in four dimensions. Although loop and spin foam quantum gravity are thus
very much tuned to four dimensions, there have nevertheless been attempts to extend
the framework to higher dimensions, specifically by replacing the groups SU (2)
(for loop quantum gravity) or S O(4) or S O(3, 1) (for spin foam models) by bigger
groups, with higher dimensional analogues of the Ashtekar variables, but it is not
clear whether one can arrive in this way at a unification properly incorporating the
SM degrees of freedom.
376 H. Nicolai
4 The Hierarchy Problem
A problem that is not so much in the focus of the GR community, but much discussed
in the particle physics community concerns the question of scales and hierarchies. The
gravitational force is much weaker than the other forces (as one can see immediately
by comparing the gravitational attraction between the electron and the nucleus in an
atom with the electric Coulomb force, which differ by a factor 10−40 ). The so-called
hierarchy problem, then, is the question whether this huge difference in scales can be
‘naturally’ understood and explained.5 In particle physics the problem is reflected in
the mass hierarchies of elementary particles. Already by itself, the observed particle
spectrum covers a large range of mass values: light neutrinos have masses of less
than 1 eV, the lightest quarks have masses of a few MeV while the top quark, which
is the heaviest quark discovered so far, has a mass of around 173 GeV, so even quark
masses differ by factors on the order of 105 , presenting a ‘little hierarchy problem’.
But all these mass values are still tiny in comparison with the Planck scale, which is
at 1019 GeV! This, then, is the distance that theory has to bridge: at the lower end it
is the electroweak scale that is now being explored at LHC, while at the higher end
it is Planck scale quantum gravity.
A much advertized, but very QFT specific, indication of the problem is the occur-
rence of quadratic divergences in radiative corrections to the scalar (Higgs) boson
mass, which require an enormous fine-tuning to keep the observed value so small
in comparison with the Planck mass, the ‘natural’ value. The absence of quadratic
divergences in supersymmetric theories, where divergences are at most logarithmic,
is widely considered as a strong argument for low energy supersymmetry, and the
prediction that each SM particle should be accompanied by a supersymmetric partner.
The smallness of the gravitational coupling in comparison with the other couplings
in nature is the main obstacle towards the verification or falsification of any proposed
model of quantum gravity. Unless there is a dramatic evolution of the strength of
the gravitational coupling over experimentally reachable energy scales there is no
hope of ‘seeing’ quantum gravity effects in the laboratory. So one needs to find ways
and means to reason indirectly in order to identify low energy hints of Planck scale
physics. One possibility might be to look for signatures of quantum gravity in the
detailed structure of CMB fluctuations. The other possibility (which is more in line
with this talk) is to try to read the signs and hints from the observed structure of
the low energy world. In the final consequence, this would require a more or less
unique prediction for low energy physics and the observed matter content of the
world.6 A more exotic possibility, advocated by proponents of large extra dimension
scenarios, could be an (enormous) increase in the gravitational coupling strength in
5 Of course, the biggest and most puzzling hierarchy problem concerns the smallness of the observed
cosmological constant.
6 This option is not very popular with aficionados of the multiverse or the anthropic principle but,
interestingly, the hope for a unique path from quantum gravity to the SM is also prominently visible
in the very first papers on the heterotic superstring [17, 18].
the TeV range that would make quantum gravity and quantum string effects directly
accessible to experiment (‘TeV scale quantum gravity’).
At any rate, it remains a key challenge for any proposed theory of quantum gravity
to offer quantifiable criteria for its confirmation or falsification. And the emphasis
here is on ‘quantitative’, not on qualitative features that might be shared by very
different approaches and thus may not suffice to discriminate between them (for
instance, I would suspect this to be the case for specific properties of the CMB
fluctuations, which may not contain enough information for us to ‘read off’ quantum
gravity). So the challenge is to come up with criteria that allow to unambiguously
discriminate a given proposal against alternative ones!
5 From the Standard Model to the Planck Scale
By now, the SM of particle physics is an extremely well tested theory. It is based on the
(Yang-Mills) gauge principle with Yang-Mills group G S M = SU (3)c × SU (2)w ×
U (1)Y . Forces are mediated by spin-one gauge bosons. Matter is made up of spin- 21
fermions: at this time, we know of 48 fundamental fermions which are grouped into
three generations (families) of 16 quarks and leptons each (including right-chiral
neutrinos). There is no evidence so far from LHC of any new fundamental fermions.
However, after decades of theoretical research, we still do not know what dis-
tinguishes the SM gauge group from other possible choices. Apart from anomaly
cancellations (see below), the same ignorance prevails with regard to the observed
matter content of the SM. Why does Nature repeat itself with three generations of
quarks and leptons (another fact confirmed by CERN experiments, as well as cos-
mological observations)? What causes symmetry breaking and what is the origin
of mass? And, returning to the question of hierarchies, what keeps the electroweak
scale stable with regard to the Planck scale? (More on this below...) And, finally,
why do we live in four space-time dimensions? To tackle these questions, numerous
proposals have been put forward for physics beyond the Standard Model (or ‘BSM
physics’, for short): Grand Unification (or GUTs, for short), technicolor, low energy
supersymmetry, large extra dimensions, TeV scale gravity, excited gauge bosons,
and so on.
The main recent progress is the discovery of a scalar boson by LHC and the strong
evidence that this particle has all the requisite properties of the Higgs boson, espe-
cially with the most recent data indicating that it has indeed spin zero and even parity
(a remaining uncertainty concerns the coupling to the SM fermions, which must be
proportional to their masses). In addition the symmetry breaking mechanism giving
mass to gauge bosons (‘Brout-Englert-Higgs mechanism’) has now been confirmed.
However, much to the dismay of many of my colleagues, no signs of ‘new physics’
have shown up so far at LHC. It is therefore not excluded that there may be nothing
more than the SM, augmented by right-chiral neutrinos, right up to the Planck scale,
a scenario that is usually referred to as the ‘Grand Desert’.
378 H. Nicolai
Fig. 2 Can the SM survive up

to the Planck scale? The upper
envelope enforces avoidance
of Landau pole for the scalar
self-coupling, while the lower
envelope ensures avoidance of
vacuum instability [19] (with
an assumed top quark mass of
175 GeV this plot is not quite
up to date, but this does not
affect our main conclusions)
Since the Higgs boson is partly responsible7 for the generation of mass, and mass
measures the strength of gravitational coupling, one can reasonably ask whether these
data contain indications of Planck scale physics reaching down to the electroweak
scale. It is here that the question of stability of the electroweak scale comes into
play. Actually, the stability of the Standard Model is under menace from two sides.
Following the renormalization group evolution of the scalar self-coupling up to the
Planck scale, one danger is the Landau pole where the scalar self-coupling diverges
(as happens for IR free theories with scalar fields) and the theory breaks down. The
other danger is the potential instability caused by the negative contribution to the
effective potential from the top quark (the effective potential includes perturbatively
computable quantum corrections to the classical potential). In Homer’s tale, Ulysses
has to maneuver his ship between two formidable obstacles, Skylla and Charybdis:
on the one side he must steer it away from the rock against which it will crash, and
on the other side must avoid the sea monster that will swallow the ship. The plot of
Fig. 2 illustrates the situation. The vertical direction is the Higgs mass, which grows
proportionally with the Higgs self-coupling. Through the RG evolution the scalar
self-coupling will eventually hit the Landau pole, at which point the theory crashes.
Exactly where this happens depends very delicately on the Higgs mass. The upper
curve (the rock) in Fig. 2 [19] relates the location of the Landau pole directly to the
Higgs mass. The lower curve (the sea monster) is the constraint from the negative
contribution to the effective Higgs potential. If this contribution is too negative, it
will make the potential unbounded from below. For the SM, the negative contribution
is mainly due to the top quark, and it is a danger precisely because the top quark mass
is so large. As a result, if you want to salvage the SM up to the Planck scale, there
remains only a very narrow strip for the SM parameters (masses and couplings). The
recent results and data from LHC indicate that Nature might indeed avail itself of
this possibility: with a Higgs mass of about 125 GeV, the Landau pole can safely
7 Only partly, as for instance the larger part of the proton mass is due to non-perturbative QCD
effects!
hide behind the Planck scale, but this value is so low that the SM hovers on the
brink of instability! See also [20, 21] for interesting alternative interpretations of this
value from the points of view of asymptotic safety, and non-commutative geometry,
respectively.
To be sure, the potential instability of the effective potential is the worse of the
two dangers. Namely, the occurrence of a Landau pole can always be interpreted
as signalling the onset of ‘new physics’ where new degrees of freedom open up
and thereby cure the problem. A well known example of this phenomenon is the
old Fermi theory of weak interactions, where the non-renormalizable four-fermion
vertex is dissolved at sufficiently high energies by new degrees of freedom (W and Z
bosons) into a renormalizable and unitary theory. Another example would be the (still
conjectural) appearance of supersymmetry in the TeV range, which would remove
the Landau pole and also ensure full stability, as the effective potential in a globally
supersymmetric theory is always bounded from below (this is no longer true for local
supersymmetry).
If we find out whether or not there are genuine new degrees of freedom in the
TeV range of energies, we may also get closer to answering the old question of
the ultimate divisibility of matter, namely the question whether the known particles
possess further substructures, sub-substructures, and so on, as we probe smaller
and smaller distances. Translated into the UV, the question can be rephrased as the
question whether there are any ‘screens’ (≈ scales of ‘new physics’) between the
electroweak scale and the Planck scale. The more of such screens there were between
the electroweak scale and the Planck scale, the less one would be be able to ‘see’ of
Planck scale physics. On the other hand, the fewer there are, the harder becomes the
challenge of explaining low energy physics from Planck scale physics.
LHC is now testing a large number of ‘BSM’ proposals, and actually eliminating
many of them.8 Figure 3 shows the latest exclusion plot from May 2013 on the search
for various signatures of supersymmetry. Figure 4 shows a similar plot from May
2013 for ‘exotica’ such as large extra dimensions, mini black holes, excited W and Z
bosons, quark substructure, and so on, with some exclusions already reaching up to
10 TeV. As you can see, even to refute only a representative subset of the proposals
on the market and to keep up with the flood of theoretical ideas is a painstaking effort
for the experimentalists, requiring teams of thousands of people and thousands of
computers!
There are theoretical indications that LHC may not reveal much new beyond the
SM Higgs boson, and thus no screens between the electroweak scale and the Planck
scale. It is a remarkable fact that the SM Lagrangian is classically conformally
invariant except for a single term, the explicit mass term in the Higgs potential.
But in a classically conformal theory mass terms can in principle be generated by
the conformal anomaly, the quantum mechanical breaking of conformal invariance,
which could also trigger spontaneous symmetry breaking [23]. Maybe no explicit
mass terms are needed in the SM Lagrangian, and the mechanism stabilizing the
8 The two plots shown below have been downloaded from the CERN website [22] where also a
summary of many further results can be found.
380 H. Nicolai
ATLAS SUSY Searches* - 95% CL Lower Limits
*Only a selection of the available mass limits on new states or phenomena is shown. All limits quoted are observed minus 1 theoretical signal cross section uncertainty.
Fig. 3 Low energy supersymmetry? [22]
electroweak scale is conformal symmetry rather than low energy supersymmetry

[24, 25]? A further hint in this direction comes from the flows of the SM couplings
under the renormalization group: it almost looks like these couplings could ‘keep
each other under control’ so as to prevent both Landau poles and instabilities right
up to the Planck scale! This is because bosons and fermions contribute with opposite
signs to the corresponding ω-functions. The scalar self-coupling would normally
blow up under the flow, but is kept under control by the top quark contribution which
delays the appearance of the Landau pole until after the Planck scale. The same
mechanism is at work for the top quark (Yukawa) coupling which is asymptotically
not free either: it, too, would blow up, but is kept under control by the strong coupling
ρs , again shifting the Landau pole beyond the Planck scale. Finally ρs itself is kept
under control in the UV by asymptotic freedom.
In summary, it could just be that the mass patterns and the couplings in the
Standard Model precisely conspire to make the theory survive to the Planck scale.
In this case there would be no ‘new physics’ beyond the electroweak scale and the
theory would have to be embedded directly as is into a Planck scale theory of quantum
gravity. In my opinion, this may actually be our best chance to gain direct access to
the Planck scale, both theoretically and experimentally!
Fig. 4 Low energy exotics? [22]
6 Anomalies
There is another remarkable property of the SM which may be interpreted as a hint

of how Planck scale physics could affect low energy physics, and this is the complete
cancellation of gauge anomalies (see [26] for an introduction and many references
to the original work). Anomalies occur generically when a classical Lagrangian is
invariant under a symmetry, but that symmetry cannot be preserved by the regular-
ization that quantization requires. When the regulator is removed there is a finite
remnant, and this is referred to as the anomaly, an O() violation of a classical
conservation laws. The classic example of such a symmetry is chiral invariance that
explains the (near-)masslessness of fermions, but cannot be regulated, leading to the
famous axial anomaly in QED that accounts for the decay of the π 0 meson.
When anomalous currents are coupled to gauge fields, the anomaly can deal a fatal
blow to the theory. Recall that the coupling of a gauge field Aμ to charged matter
generally takes the Noether form ⊗ Aμ J μ , where J μ is the classically conserved
matter current. In the presence of an anomaly the variation of this term would give

μ μ
Θ Aμ J = τμ ξ J = − ξτμ J μ ⊗ O() ◦= 0 . (3)
382 H. Nicolai
Gauge invariance would thus no longer hold, and this violation would destroy the
renormalizability of the SM and thereby its predictivity. To verify that all gauge
anomalies and gravitational anomalies cancel in the Standard Model requires the
computation of various triangle diagrams with chiral fermions circulating in the loop,
and involves traces of the form Tr T a {T b , T c }, where T a belong to the Lie algebra
of the Standard Model gauge group. More specifically, the calculation reduces to the
evaluation of

± Tr Y Y Y = ± Tr ttY = ± Tr Y = 0, (4)
where Y is the electroweak hypercharge, and t denotes any generator of SU (2)w

or SU (3)c ; the sum runs over all SM fermions, with ‘+’ for positive and ‘−’ for
negative chirality fermions. If you work through the whole list of such diagrams you
will find that they all ‘miraculously’ sum up to zero [26]. From (4) it is obvious that
the cancellation would be trivial if the SM were a vector-like theory with no preferred
handedness or chirality. Remarkably, Nature prefers to break parity invariance, and
to do so subtly in a way that maintains the renormalizability, hence consistency.
In fact, the anomaly cancellations fix the fermion content almost uniquely to what
it is, separately for each generation. Therefore, despite its ‘messy’ appearance the
Standard Model is surprisingly unique, and also surprisingly economical for what
it does!
There are two crucial features that must be emphasized here. The first is that
a proper anomaly does not and must not depend on how the theory is regulated.
Secondly, anomalies are often regarded as a perturbative phenomenon, but this is not
strictly true. The famous Adler-Bardeen theorem asserts that the anomaly is entirely
due to the one-loop contribution, and that there are thus no further contributions
beyond one loop. In other words, the one-loop result is exact to all orders, hence
non-perturbative!
Anomalies should also be expected to play a role in quantum gravity, and in
determining whether a specific proposal is ultimately consistent or not. For instance,
the classical constraint algebra of General Relativity in the Hamiltonian formulation
has the schematic form
{D, D} √ D , {D, H } √ H , {H, H } √ D, (5)
where D and H , respectively denote the diffeomorphism constraints and the Hamil-
tonian constraint, and this algebra is expected to be modified by quantum correc-
tions. This expectation is borne out by the simplest example, matter-coupled quantum
gravity in two space-time dimensions. Here the most general form of the space-time
diffeomorphism algebra including anomalies is known to take the form

[T±± (x), T±± (y)] = Θ ∗ (x, y) T±± (x) + T±± (y) + cΘ ∗∗∗ (x, y) , (6)
where x, y ⇔ R, T±± := H ± D and c is the central charge. As is well known,

virtually all of string theory hinges on the non-zero value of the central charge c!
Unfortunately in higher dimensions, there exists neither an analogous uniqueness
result, nor even a classification of what the anomalies may be. The main difficulty
here is that higher-dimensional diffeomorphism algebras are ‘soft’, which means that
Lie algebra structure ‘constants’ are not really constant, but field dependent.
7 Outlook
So where do we stand? At this time there is a growing array of proposals for quantum
gravity, based on a variety of different and even mutually contradictory assumptions
and hypotheses. The following is a selection of current approaches (to which you
may add your own favorite):
• Supergravity, Superstrings and M-Theory

• AdS/CFT and Holography
• Path integrals: Euclidean, Lorentzian, matrix models, ...
• Canonical Quantization (metric formalism)
• Loop Quantum Gravity (with either connections or holonomies)
• Discrete Quantum Gravity: Regge calculus, (causal) dynamical triangulations
• Discrete Quantum Gravity: spin foams, group field theory
• Non-commutative geometry and space-time
• Asymptotic Safety and RG Fixed Points
• Emergent (quantum) gravity from thermodynamics
• Causal Sets
• Cellular Automata (‘computing quantum space-time’)
Among these string theory remains the leading contender, not least because it
naturally incorporates (and even requires) matter degrees of freedom. Nevertheless,
we still do not have a single hint from experiment and observation (for instance, in the
form of supersymmetric partners to the known elementary particles) that it is indeed
the right theory. Perhaps it is thus not so surprising that ‘non-string approaches’ have
been gaining in popularity over the past few years.
Having grown out of particle physics and being modeled on its basic concepts,
string theory has no problem of principle in connecting to low energy physics; being
a perturbative approach, it also has no difficulties in reproducing the correct semi-
classical limit and the Einstein field equations. But after more than two decades of
effort, string theory is still struggling to reproduce the Standard Model as is, that
is, without the heavy extra baggage that comes with (for instance) the supersym-
metric extensions of the SM referred to as ‘MSSM’, ‘CMSSM’ or ‘NMSSM’, and
so on. Moreover, it has considerable difficulties in incorporating a positive cosmo-
logical constant—in fact, like supergravity, superstring theory has an overwhelming
preference for negative ζ! String theory, as originally formulated, is a background
384 H. Nicolai
dependent and perturbative theory. However, there have been important advances
and recent developments, especially in connection with the AdS/CFT correspon-
dence and gauge/gravity or weak/strong dualities, that transcend perturbation theory
and have provided important insights into the non-perturbative functioning of the the-
ory (see e.g. [27] for a recent update). Nevertheless, in its present form string theory
does not offer a convincing scenario for the resolution of (cosmological) space-time
singularities, and so far cannot tell us what really ‘happens’ to space-time at the
Planck scale.
I have already mentioned the impressive recent advances in perturbative QFT
techniques [12], yielding evidence that N = 8 supergravity may be finite to all orders,
contrary to expectations held for more than 30 years. If this theory could be shown
to be a purely quantum field theoretic extension of Einstein’s theory without UV
singularities, this would partially undermine one of string theory’s chief arguments
why QFT must be abandoned. Of course, this would not relieve us of the task of
working towards a non-perturbative understanding of physics at the very shortest
distances, as the putative finiteness by itself would not tell us why and how the space-
time continuum is dissolved at the Planck scale. And even if the theory turned out to
be UV finite, many would doubt whether N = 8 supergravity has anything to do with
‘real world physics’. Yet, there is a curious coincidence here: when supersymmetry
is completely broken, eight spin- 21 fermions are converted into Goldstinos in order
render the eight gravitinos massive, leaving us with 48 spin- 21 fermions, exactly the
right number! Most likely a mirage, but who knows?9
In contrast to string theory the non-perturbative approaches put the main emphasis
on GR concepts from the very beginning, to wit, (spatial) background independence
and diffeomorphism invariance. Following this avenue has led to intriguing new
ideas and proposals as to what a quantum space-time might actually ‘look like’.
Nevertheless, it is hard to see how such ideas could ever be put to a real test (other
than internal consistency checks). A main criticism from the point of view taken
here is that these approaches have not incorporated essential insights from particle
physics up to now, such as the restrictions from anomaly cancellations. Furthermore,
the ambiguities related to quantization and the incorporation of matter couplings
have not been resolved in a satisfactory fashion in my opinion, and the recovery of
the proper semi-classical limit remains an outstanding challenge.
To conclude let me restate my main worry. In one form or another the existing
approaches to quantum gravity suffer from a very large number of ambiguities, so far
preventing any kind of prediction with which the theory will stand or fall. Even at the
risk of sounding polemical, I would put this ambiguity at 10500 (or even more)—in
any case a number too large to cut down for any conceivable kind of experimental
or observational advance.
• Superstring theory predicts the existence of myriads of ‘consistent’ vacua, all of
which are supposed to be realized somewhere in the multiverse (or ‘megaverse’)—
leading to the conclusion that essentially anything goes when it comes to answering
9 On this point, see also [28].

the questions raised at the beginning of Sect. 5 (most notably, it is claimed that the
multiverse also ‘solves’ the cosmological constant problem).
• Loop quantum gravity and related approaches are compatible with many ‘consis-
tent’ Hamiltonians (or spin foam models), and with an essentially arbitrary menu
of matter fields. Even disregarding technical issues such as quantization ambi-
guities, it looks again like almost anything goes. Idem for models of lattice and
discrete quantum gravity.
• Asymptotic Safety is an assumption that, according to its proponents, works
almost generically—that is, independently of the specific ‘initial’ conditions for
the RG flows, of the matter content and even the number of space-time dimensions
(if that number is not extremely large), leaving us with numerous ‘consistent’ RG
flows.
In my view the real question is this: if there are all these ‘consistent’ (according to
your definition) ansätze, does Nature simply pick the ‘right’ answer at random from
a huge variety of possibilities, or are there criteria to narrow down the number of
choices? Being exposed to many talks from the different ‘quantum gravity camps’ I
am invariably struck by the success stories I keep hearing, and the implicit or explicit
claims that ‘we are almost there’. I, for one, would much prefer to hear once in a
while that something does not work, and to see some indications of inconsistencies
that might enable us to discriminate between a rapidly growing number of diverging
ideas on quantum gravity [29, 30]. If, however, the plethora of theory ambiguities
were to stay with us I would conclude that our search for an ultimate explanation,
and with it the search for quantum gravity, may come to an ignominious end (like in
Breughel’s painting). I cannot imagine that this is what Einstein had in mind during
his stay in Prague, nor in the later years of his life when he was striving to figure
out “the old one’s tricks” (or, in the original German, “dem Alten auf die Schliche
kommen”).
So let me repeat my main message: the incompleteness of the Standard Model is
one of the strongest arguments in favor of quantizing gravity and searching for new
concepts replacing classical notions of space and time. The observed features of SM
may contain important hints of its possible UV completion and Planck scale physics,
and these hints should be given due consideration in the search for a consistent theory
of quantum gravity.
Acknowledgments I would like to thank Jianwei Mei for his help in turning my talk into a
(hopefully) readable text and Krzysztof Meissner for many enjoyable and illuminating discussions
on the state of the art.
References
1. Weinberg, S.: Living with Infinities, ArXiv e-prints 0903.0568 [hep-th] (2009)
2. Hawking, S.W., Israel, W. (eds.): General Relativity: An Einstein Centenary Survey. Cam-
bridge University Press, Cambridge (1979)
386 H. Nicolai
3. Kiefer, C.: Quantum gravity, International Series of Monographs on Physics. Clarendon Press,
Oxford (2004)
4. Ashtekar, A. (ed.): 100 Years of Relativity. Space-Time Structure: Einstein and Beyond. World
Scientific, Singapore (2005)
5. Rovelli, C.: Quantum Gravity, Cambridge University Press, Cambridge (2004)
6. Thiemann, T.: Modern Canonical Quantum General Relativity. Cambridge Monographs on
Mathematical Physics. Cambridge University Press, Cambridge (2007)
7. Green, M.B., Schwarz, J.H., Witten, E.: Superstring Theory, vol. 1 & 2. Cambridge Mono-
graphs on Mathematical Physics. Cambridge University Press, Cambridge (1987)
8. Polchinski, J.: String Theory, vol. 1 & 2. Cambridge Monographs on Mathematical Physics.
Cambridge University Press, Cambridge (1998)
9. Blumenhagen, R., Lüst, D., Theisen, S.: Basic Concepts of String Theory. Theoretical and
Mathematical Physics. Springer, Berlin (2012)
10. Goroff, M.H., Sagnotti, A.: The ultraviolet behavior of Einstein gravity. Nucl. Phys. B 266,
709 (1986). doi:10.1016/0550-3213(86)90193-8
11. van de Ven, A.: Two-loop quantum gravity. Nucl. Phys. B 378, 309 (1992). doi:10.1016/0550-
3213(92)90011-Y
12. Bern, Z., Carrasco, J.J., Dixon, L.J., Johansson, H., Roiban, R.: Amplitudes and ultraviolet
behavior of N = 8 supergravity. Fortschr. Phys. 59, 561 (2011). doi:10.1002/prop.201100037
13. Bojowald, M.: Absence of a singularity in loop quantum cosmology. Phys. Rev. Lett. 86, 5227
(2001). doi:10.1103/PhysRevLett.86.5227
14. Reuter, M., Saueressig, F.: Quantum Einstein gravity. New J. Phys. 14, 055022 (2012). doi:10.
1088/1367-2630/14/5/055022
15. Applequist, T., Chodos, A., Freund, P.G.O. (eds.): Modern Kaluza-Klein Theories, Frontiers
in Physics. Addison-Wesley, Reading (1987)
16. Wess, J., Bagger, J.: Supersymmetry and Supergravity. Princeton University Press, Princeton
(1983)
17. Gross, D.J., Harvey, J.A., Martinec, E.J., Rohm, R.: Heterotic string. Phys. Rev. Lett. 54, 502
(1985). doi:10.1103/PhysRevLett.54.502
18. Candelas, P., Horowitz, G.T., Strominger, A., Witten, E.: Vacuum configurations for super-
strings. Nucl. Phys. B B258, 46 (1985). doi:10.1016/0550-3213(85)90602-9
19. Hambye, T., Riesselmann, K.: Matching conditions and Higgs boson mass upper bounds reex-
amined. Phys. Rev. D 55, 7255 (1997). doi:10.1103/PhysRevD.55.7255 See also Proceedings
of the ‘ECFA/DESY Study on Physics and Detectors for the Linear Collider’, DESY 97-123E,
ed. R. Settles
20. Shaposhnikov, M., Wetterich, C.: Asymptotic safety of gravity and the Higgs boson mass.
Phys. Lett. B 683, 196 (2010). doi:10.1016/j.physletb.2009.12.022
21. Chamseddine, A.H., Connes, A.: Resilience of the spectral standard model. J. High Energy
Phys. 9, 104 (2012). doi:10.1007/JHEP09(2012)104
22. ATLAS Physics Summary Plots. (ATLAS Experiment 2013 CERN). URL https://twiki.cern.
ch/twiki/bin/view/AtlasPublic/CombinedSummaryPlots
23. Coleman, S., Weinberg, E.: Radiative corrections as the origin of spontaneous symmetry
breaking. Phys. Rev. D 7, 1888 (1973). doi:10.1103/PhysRevD.7.1888
24. Bardeen, W.A.: On naturalness in the standard model. In: FERMILAB-CONF-95-391-T, C95–
08-27.3 (1995)
25. Meissner, K.A., Nicolai, H.: Conformal symmetry and the standard model. Phys. Lett. B 648,
312 (2007). doi:10.1016/j.physletb.2007.03.023
26. Bertlmann, R.A.: Anomalies in Quantum Field Theory, International Series of Monographs
on Physics. Clarendon Press, Oxford (1996)
27. Blau, M., Theisen, S.: String theory as a theory of quantum gravity: a status report. Gen.
Relativ. Gravit. 41, 743 (2009). doi:10.1007/s10714-008-0752-z
28. Nicolai, H., Warner, N.P.: The SU(3)×U(1) invariant breaking of gauged N = 8 supergravity.
Nucl. Phys. B 259, 412 (1985). doi:10.1016/0550-3213(85)90643-1
29. Nicolai, H., Peeters, K., Zamaklar, M.: Loop quantum gravity: an outside view. Class. Quantum
Gravity 22, R193 (2005). doi:10.1088/0264-9381/22/19/R01
30. Alexandrov, S., Roche, P.: Critical overview of loops and foams. Phys. Rep. 506, 41 (2011).
doi:10.1016/j.physrep.2011.05.002
Part IV
Numerical Relativity and Relativistic
Astrophysics
Three Little Pieces for Computer
and Relativity
Luciano Rezzolla
Abstract Numerical relativity has made big strides over the last decade. A number
of problems that have plagued the field for years have now been mostly solved. This
progress has transformed numerical relativity into a powerful tool to explore funda-
mental problems in physics and astrophysics, and I present here three representative
examples. These “three little pieces” reflect a personal choice and describe work that
I am particularly familiar with. However, many more examples could be made.
1 Introduction
Numerical relativity has hardly seen better times before. Over the last few years, in
fact, a truly remarkable development has shaken the field. Starting from the first
simulations showing that black-hole binaries could be evolved for a few orbits
[1–3], or that black-hole formation could be followed stably using simple gauges
and without excision [4], new results, some awaited for decades, have been obtained
steadily. As a direct consequence of this “Renaissance”, it is now possible to simulate
binary black holes [5] and binary neutron stars [6] accurately for dozens of orbits,
from the weak-field inspiral, down to the final black-hole ringdown (see also [7, 8]
for recent reviews).
There are several reasons behind this rapid progress. These include the use of more
advanced and accurate numerical techniques [9, 10], the availability of larger compu-
tational facilities, but also the development of formulations of the Einstein equations
and gauges that are particularly well-suited for numerical evolutions [11–21]. The
phase transition that has taken during this year has radically changed numerical rel-
L. Rezzolla (B)
Max-Planck-Institut für Gravitationsphysik, Albert-Einstein-Institut, Potsdam, Germany
e-mail: rezzolla@th.physik.uni-frankfurt.de
L. Rezzolla
Institut für Theoretische Physik, Frankfurt, Germany
392 L. Rezzolla
ativity, freeing it from the corner of idealised investigations. Most importantly, it has
transformed numerical relativity into a research area where long-standing problems
can found a quantitative and accurate solution, and into a tool by means of which it
is possible to explore fundamental aspects of physics and astrophysics.
Numerous examples could be given to testify this transformation, although I will
report here only those that I am particular familiar with. More specifically, in what
follows I will discuss: (i) how numerical simulations of magnetised neutron stars
provide convincing evidence that this process leads to the conditions that are expected
behind the phenomenology of short gamma ray burst; (ii) how numerical simulations
of the head-on collision of selfgravitating fluids boosted at relativistic speeds can be
used to understand the conditions leading to the formation of a black hole and provide
a dynamical version of the hoop conjecture; (iii) how the study of the local properties
of apparent horizons can be used to explain bizarre behaviours in binary black-
hole simulations and can be effectively correlated with a portion of the spacetime
infinitely far away: J + . This selection is by no means comprehensive, but rather
a very personal one, and I apologise in advance for not discussing all the excellent
work that cannot find space in this contribution.
2 First Piece: From Neutrons Star to Gamma-Ray Bursts
The numerical investigation of the inspiral and merger of binary neutron stars in
full general relativity has seen enormous progress made in recent years. Crucial
improvements in the formulation of the equations and numerical methods, along
with increased computational resources, have extended the scope of early simu-
lations. These developments have made it possible to compute the full evolution,
from large binary-separations up to black-hole formation, without and with mag-
netic fields [6, 22–26], and with idealised or realistic equations-of-state [27, 28].
This tremendous advancement is also providing information about the entire grav-
itational waveform, from the early inspiral up to the ringing of the black hole (see,
e.g., [29–31]). Advanced interferometric detectors starting from 2014 are expected
to observe these sources at a rate of ≈40–400 events per year [32].
These simulations also probe whether the end-product of mergers can serve as the
“central engine” of short gamma-ray bursts (SGRBs) [33–35]. The prevalent scenario
invoked to explain SGRBs involves the coalescence of a binary system of compact
objects, e.g., a black hole and a neutron star or two neutron stars [36–39]. After the
coalescence, the merged object is expected to collapse to a black hole surrounded
by an accretion torus. An essential ingredient in this scenario is the formation of a
central engine, which is required to launch a relativistic outflow with an energy of
≈1048 –1050 erg on a timescale of ≈0.1–1 s [38, 39].
The qualitative scenario described above is generally supported by the associ-
ation of SGRBs with old stellar populations, distinct from the young massive star
associations for long GRBs [40, 41]. It is also supported to a good extent by fully
general-relativistic simulations, which show that the formation of a torus of mass
Three Little Pieces for Computer and Relativity 393
Mtor 0.4 M⊗ around a black hole with spin J/M 2 ◦ 0.7 − 0.8, is inevitable [27].
In addition, recent simulations have also provided the first evidence that the merger of
a binary of modestly magnetised neutron stars naturally forms many of the conditions
needed to produce a jet of ultrastrong magnetic field, with properties that are broadly
consistent with SGRB observations. This missing link between the astrophysical
phenomenology of GRBs and the theoretical expectations is a genuine example of
the new potential of numerical relativity and I will discuss it in detail below.1
2.1 The Numerical Setup
It is not useful to discuss here in detail the numerical setup and the technical details of
the numerical codes used in these calculations. These details can be found in Refs. [26,
43], while a description of the physical initial data was presented in Ref. [42]. It is
sufficient to recall here that the evolution of the spacetime is obtained using a three-
dimensional finite-differencing code providing the solution of a conformal traceless
formulation of the Einstein equations [44] (i.e., the CCATIE code). The equations of
general-relativistic magnetohydrodynamics (GRMHD) in the ideal-MHD limit are
instead solved using a code code [43, 45, 46] which adopts a flux-conservative for-
mulation of the equations as presented in [47] and high-resolution shock-capturing
schemes (i.e., the Whisky code). In order to guarantee the divergence-free charac-
ter of the MHD equations the flux-CD approach described in [48] was employed,
although with the difference that the vector potential is used as evolution variable
rather than the magnetic field. Both the Einstein and the GRMHD equations are
solved using the vertex-centred adaptive mesh-refinement (AMR) approach pro-
vided by the Carpet driver [49]. In essence, the highest-resolution refinement level
is centred around the peak in the rest-mass density of each star and in moving the
“boxes” following the position of this maximum as the stars orbit. The boxes are
evolved as a single refinement level when they overlap. The calculations were car-
ried out using six levels of mesh refinement with the finest level having a resolution
of ε = 0.1500 M⊗ ◦ 221 m.
From a physical point of view, the initial data is given by a configuration that
could represent the properties of a neutron star-binary a few orbits before their coa-
lescence. More specifically, we simulate two equal-mass neutron stars, each with a
gravitational mass of 1.5 M⊗ (i.e., sufficiently large to produce a black hole soon
after the merger), an equatorial radius of 13.6 km, and on a circular orbit with initial
separation of ◦45 km between the centres (all lengthscales are coordinate scales).
Confined in each star is a poloidal magnetic field with a maximum strength of 1012 G.
At this separation, the binary loses energy and angular momentum via emission of
gravitational waves, thus rapidly proceeding on tighter orbits as it evolves.
1 Much of what follows is taken from the discussion presented in Ref. [42].
394 L. Rezzolla
2.2 The Basic Dynamics
After about 8 ms (≈3 orbits) the two neutron stars merge forming a hypermassive neu-
tron star (HMNS), namely, a rapidly and differentially-rotating neutron star, whose
mass, 3.0 M⊗ , is above the maximum mass, 2.1 M⊗ , allowed with uniform rotation
by an ideal-fluid equation of state (EOS),2 p = (γ − 1)ρε, where ρ is the baryonic
density, the specific internal energy, and γ = 2 with an adiabatic index of 2. Being
metastable, a HMNS can exist as long as it is able to resist against collapse via a
suitable redistribution of angular momentum (e.g., deforming into a “bar” shape [6,
22]), or through the increased pressure-support coming from the large temperature-
increase produced by the merger. However, because the HMNS is also losing angu-
lar momentum through gravitational waves, its lifetime is limited to a few ms, after
which it collapses to a black hole with mass M = 2.91 M⊗ and spin J/M 2 = 0.81,
surrounded by a hot and dense torus with mass Mtor = 0.063 M⊗ [26].
These stages of the evolution can be seen in Fig. 1, which shows snapshots of
the density colour-coded between 109 and 1010 gr/cm3 , and of the magnetic field
lines (green on the equatorial plane and white outside the torus). Soon after the
black hole formation the torus reaches a quasi-stationary regime, during which the
density has maximum values of ≈1011 g/cm3 , while the accretion rate settles to
Ṁ ≈ 0.2 M⊗ /s. Using the measured values of the torus mass and of the accretion
rate, and assuming the latter will not change significantly, such a regime could last
for taccr ◦ Mtor / Ṁ ◦ 0.3 s, after which the torus is fully accreted; furthermore, if
the two neutron stars have unequal masses, tidal tails are produced which provide
additional late-time accretion [27]. This accretion timescale is close to the typical
observed SGRB durations [38, 50]. It is also long enough for the neutrinos produced
in the torus to escape and annihilate in its neighbourhood; estimates of the associated
energy deposition rate range from ≈1048 erg/s [51] to ≈1050 erg/s [52], thus leading
to a total energy deposition between a few 1047 erg and a few 1049 erg over a fraction
of a second. This energy would be sufficient to launch a relativistic fireball, but
because radiative losses are not yet accounted for the large reservoir of thermal
energy in the torus cannot be extracted in these simulations.
The gravitational wave signal of the whole process is shown in the top part of
the left panel in Fig. 2, while the bottom part exhibits the evolution of the MHD
luminosity, L MHD , as computed from the integrated Poynting flux (solid line) and of
the corresponding energy, E MHD , (dashed line). Clearly, the MHD emission starts
only at the time of merger and increases exponentially after black-hole formation,
when the gravitational wave signal essentially shuts off. Assuming that the quasi-
stationary MHD luminosity is ≈4 × 1048 erg/s, the total MHD energy released
during the lifetime of the torus is ≈1.2 × 1048 erg, which, if spread over an opening
half-angle of ≈30√ (see discussion below), suggests a lower limit to the isotropic
equivalent energy in the outflow of ≈9 × 1048 erg. While this is at the low end of the
observed distribution of gamma-ray energies for SGRBs, larger MHD luminosities
2 The use of a simplified EOS does not influence particularly the results besides determining the precise time when the
HMNS collapses to a black hole.
Fig. 1 Snapshots at representative times of the evolution of the binary and of the formation of a
large-scale ordered magnetic field. Shown with a colour-code map is the density, over which the
magnetic-field lines are superposed. The panels in the upper row refer to the binary during the
merger (t = 7.4 ms) and before the collapse to black hole (t = 13.8 ms), while those in the lower
row to the evolution after the formation of the black hole (t = 15.26 ms, t = 26.5 ms). Green
lines sample the magnetic field in the torus and on the equatorial plane, while white lines show the
magnetic field outside the torus and near the black hole spin axis. The inner/outer part of the torus
has a size of ≈90/170 km, while the horizon has a diameter of ◦9 km
are expected either through the additional growth of the magnetic field via the winding
of the field lines in the differentially-rotating disk (the simulation covers only one
tenth of taccr ), or when magnetic reconnection (which cannot take place within an
ideal-MHD approach), is also accounted for (which may also increase the gamma-ray
efficiency, e.g., [53]).
The last two panels of Fig. 1 offer views of the accreting torus after the black-hole
formation. Although the matter dynamics is quasi-stationary, the last two panels
clearly show that the magnetic-field is not and instead evolves significantly. It is
only when the system is followed well after the formation of a black hole, that
MHD instabilities develop and generate the central, low-density, poloidal-field fun-
nel. This regime, which was not accessible to previous simulations [23, 24, 54], is
essential for the jet formation [55, 56]. Because the strongly magnetised matter in the
torus is highly conductive, it shears the magnetic-field lines via differential rotation.
396 L. Rezzolla
Fig. 2 Left panel gravitational wave signal shown through the = 2, m = 2 mode of the
+polarization, (h + )22 , (top part) and of the MHD luminosity, L MHD , (bottom part) as computed
from the integrated Poynting flux and shown with a solid line. The corresponding energy, E MHD ,
is shown with a dashed line. The dotted and dashed vertical lines show the times of merger (as
deduced from the first peak in the evolution of the gravitational wave amplitude) and black-hole
formation, respectively. Right panel Evolution of the maximum of the magnetic field in its poloidal
(red solid line) and toroidal (blue dashed line) components. The bottom panel shows the maximum
local fluid energy indicating that an unbound outflow (i.e., E loc > 1) develops and is sustained after
black-hole formation
A measurement of the angular velocity in the torus indicates that it is essentially

Keplerian and thus unstable to the magneto-rotational instability [57], which devel-
ops ◦5 ms after black-hole formation and amplifies exponentially both the poloidal
and the toroidal magnetic fields; the e-folding time of the instability is ◦2.5 ms and
in good agreement with the one expected in the outer parts of the torus [57]. Because
of this exponential growth, the final value of the magnetic field is largely insensitive
to the initial strength and thus a robust feature of the dynamics (see also [58] for a
similar behaviour recently computed in a HMNS).
A quantitative view of the magnetic-field growth is shown in the right panel
of Fig. 2, which shows the evolution of the maximum values in the poloidal and
toroidal components. Note that the latter is negligibly small before the merger, reaches
equipartition with the poloidal field as a result of a Kelvin-Helmholtz instability
triggered by the shearing of the stellar surfaces at merger [54, 59], and finally grows
to ◦ 1015 G by the end of the simulation. At later times (t 22 ms), when the
instability is suppressed, the further growth of the field is due to the shearing of
the field lines and it increases only as a power-law with exponent 3.5 (4.5) for the
poloidal (toroidal) component. Although the magnetic-field growth essentially stalls
after t ◦ 35 ms, further slower growths are possible [60], yielding correspondingly
larger Poynting fluxes. Indeed, when the ratio between the magnetic flux across the
horizon and the mass accretion rate becomes sufficiently large, a Blandford-Znajek
mechanism [61] may be ignited [62]; such conditions are not met over the timescale
Fig. 3 Magnetic-field structure in the HMNS (first panel) and after the collapse to black hole (last
three panels). Green refers to magnetic-field lines inside the torus and on the equatorial plane,
while white refers to magnetic-field lines outside the torus and near the axis. The highly turbulent,
predominantly poloidal magnetic-field structure in the HMNS (t = 13.8 ms) changes systematically
as the black hole is produced (t = 15.26 ms), leading to the formation of a predominantly toroidal
magnetic field in the torus (t = 21.2 ms). All panels have the same linear scale, with the horizon’s
diameter being of ◦9 km
of the simulations, but could develop over longer timescales. Also shown in the
right panel of Fig. 2 is the maximum local fluid energy, highlighting that an unbound
outflow (i.e., E loc > 1) develops after black-hole formation along the outer walls of
the torus and persists for the whole duration of the simulation.
Finally, Fig. 3 provides a summary of the magnetic-field dynamics. It shows the
magnetic field in the HMNS formed after the merger and its structure and dynamics
after the collapse to black hole. In particular, in the last three panels it shows the
magnetic-field structure inside the torus and on the equatorial plane (green), and
outside the torus and near the axis (white). It is apparent that the highly turbulent
magnetic field in the HMNS (t = 13.8 ms) changes systematically as the black hole
is produced (t = 15.26 ms), leading to the formation of a toroidal magnetic field in
the torus (t = 21.2 ms).3 As the MRI sets in, the magnetic field is not only amplified,
3 Turbulence in relativistic flows is an extremely challenging problem that is also essentially unexplored. Also in this
case, the first relativistic simulations have been performed only recently [63, 64].
398 L. Rezzolla
but also organises itself into a dual structure, which is mostly toroidal in the accretion
torus with Btor ≈2 × 1015 G, but predominantly poloidal and jet-like along the black-
hole spin axis, with Bpol ≈ 8 × 1014 G (t = 26.5 ms).4 Note that the generation of
an ordered large-scale field is far from trivial and a nonlinear dynamo may explain
why the MRI brings a magnetic field self-organization, as it has been also suggested
in case of MRI-mediated growth of the magnetic field in the conditions met in the
collapse of massive stellar cores [60, 66]. However, the jet-like structure produced
in the simulation is not yet the highly collimated ultrarelativistic outflow expected
in SGRBs (see also below).
The magnetic hollow jet-like structure has an opening half-angle of ≈30√ , which
sets an upper limit for the opening half-angle of any potential outflow, either produced
by neutrino energy deposition [55] or by electromagnetic processes [56]. In these
simulations most of the outflow develops along the edges of the jet-like structure, via
a turbulent layer of electromagnetic driven matter, which shields the central funnel
from excessive baryonic pollution. It is reasonable to expect that such a layer is
crucial to set the opening angle of any ultrarelativistic jet, to shape both the radial
and transverse structure of the jet, as well as to determine its stability properties.
The Lorentz factors of the outflow measured in these simulations are not very high
(γ 4), but can potentially be amplified by several orders of magnitude in the
inner baryon-poor regions through special-relativistic effects [67], the variability
of the flow [68], or when resistive-MHD effects are taken into account [69]. Such
accelerations will be produced as a more realistic and general-relativistic treatment
of the radiative losses will become computationally affordable.
2.3 Comparison with Observations
Below I briefly discuss how the results presented above broadly match the properties
of the central engine as deduced from the observations.
Duration: The observed duration of the prompt gamma-ray emission GRBs is energy
dependent and is usually determined through Tx , the time over which x % of the total
counts are observed, between the (100 − x)/2 and (100 + x)/2 percentiles. The most
common intervals used are T90 (or T50 ), initially defined [50] between 20 keV and
2 MeV. The GRB duration distribution is bimodal [50], where the durations of SGRBs
(approximately 25 % of GRBs) are well-fit by a fairly wide log-normal distribution
centred around T90 ∗ 0.8 s with a FWHM of 1.4 dex [38]. The typical redshifts of the
SGRBs observed with Swift are in the range z ≈ 0.3 − 1, suggesting a central value
of the intrinsic duration distribution of ∗ ⇔1 + z≤−1 0.8 s ≈ 0.5 s, and a comparably
wide distribution around this value. This is in close agreement with our accretion
time of ≈ 0.3 s.
4 A similar magnetic-field configuration has been recently reproduced also when simulating the merger of a magnetised
neutron star onto a black hole [65].
Energy: The isotropic equivalent energy output in the prompt gamma-ray emission of
SGRBs, E γ,iso , spans a wide range, from (2.7± 1) × 1048 erg (in the observed energy
range 15–350 keV) for GRB 050509B at a redshift of z = 0.225 [70], up to (1.08 ±
0.06) × 1053 erg (in the observed energy range 10 keV–30 GeV) for GRB 090510
at z = 0.903 [71]. However, the most typical values are in the range E γ,iso ≈
1049 − 1051 erg [38]. In this model, the highly relativistic outflow may be powered
either by neutrino-anti neutrino annihilation, or by the Blandford-Znajek mechanism.
For the former one might expect a total energy release between a few 1047 erg and
≈1049 erg [72, 73], into a bipolar relativistic jet of opening half-angle θjet ≈ 8 −
30√ , corresponding to a fraction f b ≈ 0.01 − 0.13 of the total solid angle, and
isotropic equivalent energies, E ν ν̄,iso , between a few 1048 erg and ≈1051 erg. For the
latter mechanism, instead, and if the magnetisation near the event horizon becomes
sufficiently high, the jet power for these values for the black-hole mass and spin
is [74]
2
f rel B
L BZ ≈ 3.0 × 10 50
erg/s, (1)
0.1 2 × 1015 G
where f rel is the fraction of the total Blandford-Znajek power that is channelled
into the resulting relativistic jet (and f rel ≈ 0.1 might be expected for ejecta with
asymptotic Lorentz factors above 100). This relativistic outflow is launched over a
timescale of ≈0.2 s and corresponds to
−1 2
f rel fb B
E BZ,iso ≈ 1.2 × 10 51
erg . (2)
0.1 0.05 2 × 1015 G
Comparing the X-ray afterglow luminosity (after 10 or 11 h) and E γ,iso suggests that
the efficiency of the prompt gamma-ray emission in SGRBs is typically high [70], and
similar to that of long GRBs [75], with E γ,iso ≈ (0.1 − 0.9)E iso , radiating between
≈10 and ≈90 % of the initial energy of the ultrarelativistic outflow. Therefore, this
model is able to accommodate the observed E γ,iso values.
Lorentz factor: The Fermi Gamma-Ray Space Telescope has detected GeV emission
from SGRBs [76], suggesting typical lower limits of γmin ≈ 102 − 103 . In particular,
γmin ∗ 1200 was obtained for GRB 090510 [71]. However, a more realistic model
[77] results in γmin values lower by a factor of ≈3. Therefore, the central engine
should be capable of producing outflow Lorentz factors of at least a few hundred. The
fact that our simulation produces a strongly magnetised mildly relativistic outflow
at angles near ≈30√ from the black-hole spin axis would help shield the inner region
near the spin axis from excessive baryon loading, and thus assist in achieving high
asymptotic Lorentz factors at large distance from the source, after the outflow in this
region is triggered by neutrinos and/or the Blandford-Znajek mechanism.
Jet angular structure: This is poorly constrained by observations (even more so than
for long GRBs). The only compelling case for a jet break in the afterglow light-curve
is for GRB 090510 [78], which occurred very early on (after ≈1,400 s), and would
thus imply an extremely narrow jet (θjet ≈ 0.2 − 0.4√ ) and modest true energy output
400 L. Rezzolla
in gamma-rays (≈1048 erg). If this is indeed a jet break, it might correspond to a

line of sight near a very narrow and bright core of a jet, which also has significantly
wider wings. Observers with lines of sight along these wings would then see a
much dimmer and more typical SGRB [79, 80]; without such wings, however, the
observations would suggest a very large intrinsic and beaming-corrected event rate
per unit volume. In most cases there are only lower limits on a possible jet break
time [38], resulting in typical limits of f b 10−2 or θjet 8√ . This is consistent with
the expectation of θjet ≈ 8 − 30√ for the ultrarelativistic ejecta capable of producing
a SGRB (which would also imply a reasonable SGRBs intrinsic event rate per unit
volume).
2.4 Summary
The calculations reported above demonstrate that a binary merger of two neutron stars
inevitably leads to the formation of a relativistic jet-like and ultrastrong magnetic
field, which could serve as a central engine for SGRBs. Because the magnetic-
field growth is exponential, the picture emerging from these simulations is rather
general and applies equally even to mildly magnetised neutron stars. Overall, this first
“little piece” of numerical relativity removes a significant uncertainty as to whether
such binary mergers can indeed produce the central engines of SGRBs. While the
electromagnetic energy release is already broadly compatible with the observations,
the simulations discussed above lack a proper treatment of the energy losses via
photons and neutrinos or resistive dissipation, which can provide a fundamental
contribution to the energy-input necessary to launch the fireball and cool the torus [51,
52]. This additional energy input, whose self-consistent inclusion in general relativity
remains extremely challenging, may help to launch an ultrarelativistic outflow very
early after the black hole forms and complete the picture of the central engine of a
SGRB.
3 Second Piece: A Dynamical Hoop Conjecture
The second “little piece” of numerical relativity that I will discuss aims at address-
ing the issue of necessary conditions for the formation of a black hole, which still
represents one of the most intriguing and fascinating predictions of classical general
relativity. There is abundant astronomical evidence that black holes exist, and a num-
ber of considerations supporting the idea that under suitable conditions gravitational
collapse is inevitable [81]. In addition, there is overwhelming numerical evidence
that black-hole formation does take place in a variety of environments [42]. Yet, a
rigorous definition of the sufficient conditions for black-hole formation is still lack-
ing. Hence, it is not possible to predict whether the collision of two compact objects,
either stars or elementary particles, will lead to the formation of a black hole.
The hoop conjecture proposed by Thorne in the ’70s, provides some reasonable
and intuitive guidelines [82]. I recall that the conjecture states that a black hole is
formed if an amount of “mass-energy” E can be compressed to fit within a hoop
with radius equal or smaller than the corresponding Schwarzschild radius, i.e., if
Rhoop ≡ Rs = 2G E/c4 , where G is gravitational constant and c the speed of light.
Even though it can be made precise under particular circumstances [83], the hoop
conjecture is not meant to be a precise mathematical statement and, in fact, it is
difficult to predict if the above-mentioned collision will compress matter sufficiently
to fit within the limiting hoop. Loosely speaking, what is difficult is to determine
which part of the “kinetic energy” of the system can be accounted to fit within the
hoop. Since at the collision the conversion of kinetic energy into internal energy is a
highly nonlinear process, any quantitative prediction becomes rapidly inaccurate as
the speeds involved approach that of light.
As stated above, the hoop conjecture is purely classical. A quantum-mechanical
equivalent is not difficult to formulate, although not very stringent, as it simply
implies that a black hole will be formed at Planck-energy scales. The predicting
power does not improve significantly when considering the conditions of black-hole
formation in higher-dimensional theories of gravity (see, e.g., [84–86]). In these
frameworks, the energy required for black-hole formation might be significantly
smaller [84], thus providing the possibility of producing them in the Large Hadron
Collider (LHC) [87], but no firm conclusion has been reached yet.
Clearly, although numerical simulations represent a realistic route to shed some
light on this issue (see, e.g., [88–90]), even the simplest scenario of the collision
of two compact objects at ultrarelativistic speeds is far from being simple and it
is actually very challenging. A first step was taken by Eardley and Giddings [91],
who have studied the formation of a black hole from the head-on collision of two
plane-fronted gravitational waves with nonzero impact parameter (previous work of
D’Eath and Payne [92–94] using different methods had considered a zero impact
parameter). In all of these analyses each incoming particle is modelled as a point
particle accompanied by a plane-fronted gravitational shock wave corresponding to
the Lorentz-contracted longitudinal gravitational field of the particle. At the instant
of collision the two shock waves pass through one another and interact through a
nonlinear focusing and shearing. As a result of their investigation, a lower bound was
set on the cross-section for black-hole production, i.e., σ > 32.5(G E/2c4 )2 , where
E is the centre-of-mass (lab) energy. More recently, and in a framework which is
closer to the one considered here, this problem has been investigated by Choptuik
and Pretorius [95], who studied the collision of two classical spherical solitons, with
a total energy ofthe system in the lab frame E = 2γb m 0 c2 , where m 0 is the “rest-
mass”, γb ∇ 1/ 1 − vb2 /c2 and vb the boost velocity. They were then able to show
that for collisions with sufficiently high boost, i.e., γb 2.9, a black hole can be
formed.
In what follows I discuss what has been recently reported on the first calculations of
black-hole production from the collision of two compact, selfgravitating, fluid objects
402 L. Rezzolla
boosted at ultrarelativistic speeds5 (A similar investigation by East and Pretorius [97]

has also appeared at about the same time).
I start by pointing out that there are several important differences with the previous
investigations in [91–95]. Differently from [91–94], in fact, I will consider colliding
objects that are not in vacuum and are not treated as point particles. Rather, they
are relativistic stars, which obviously extended and selfgravitating objects, thus with
a behaviour that is intrinsically different. Also, differently from [95], these objects
are not described as scalar fields, but as fluids and thus represent a more realistic
description of baryonic matter, such as the one employed when simulating relativistic
heavy-ion collisions [98]. These intrinsic differences also make the comparison with
the works of [91–94] very hard if possible at all. On the other hand, many analo-
gies exist with the collision of bosons stars considered in [95], and that, as I will
discuss below, can be interpreted within the more general description of black-hole
production from ultrarelativistic collisions.
Overall, the most important and distinguishing feature in the collision of two
selfgravitating stars is that a black hole can be produced even from zero initial
velocities if the initial masses are large enough; this behaviour is clearly absent in
all previous results, where instead a critical initial boost is necessary [91–95]. In
addition, for each value of the effective Lorentz factor, ⇔γ≤, a critical initial mass
exists, Mc , above which a black hole is formed and below which matter, at least in
part, selfgravitates. More importantly, both Mc follows a simple scaling with ⇔γ≤,
thus allowing to extrapolate the results to the masses and energies of modern particle
accelerators and to deduce that black-hole production is unlikely at LHC scales.
3.1 The Numerical Setup
The numerical setup employed in the simulations is the same presented in [99], and
it uses an axisymmetric code to solve in two spatial dimensions, (x, z), the set of
the Einstein and of the relativistic-hydrodynamic equations [100]. The axisymmetry
of the spacetime is imposed exploiting the “cartoon” technique, while the hydrody-
namics equations are written explicitly in cylindrical coordinates. All the simulations
use an ideal-fluid EOS with γ = 2. The initial configurations consist of spherical
stars, constructed as in [99, 101] after specifying the central density, ρc , where the
latter also serves as parameter to determine the critical model. The stars have an ini-
tial separation D and are boosted along the z-direction via a Lorentz transformation
with boost vb /c. To limit the initial violation in the constraints, D is chosen to be
sufficiently large, i.e., D = 240 M⊗ , and an optimal superposition of the two isolated-
star solutions that will be presented in a longer paper. The grid has uniform spacing
ε = 0.08(0.06) M⊗ with extents x/M⊗ ⇒ [0, 80] and z/M⊗ ⇒ [0, 150(200)],
where the round brackets refer to the more demanding high-boost cases. Reflection
boundary conditions are applied on the z = 0 plane, while outgoing conditions are
used elsewhere.
5 Much of what follows is taken from the discussion presented in Ref. [96].
Fig. 4 Representative snapshots of the rest-mass density, ρ in units where c = 1 = M⊗ (top row),
of the Lorentz factor, γ (middle row), and of the local fluid energy, −u 0 (bottom row), for subcritical
models with an initial small boost vb /c = 0.3 (left panel) or a large one vb /c = 0.8 (right panel).
Note that the post-collision flow is essentially jet-like for the low-boost case (left panel), while
essentially spherical for the high-boost case (right panel); in this latter case, most of the matter is
unbound
3.2 The Basic Dynamics
The dynamics of the process is rather simple. As the two stars approach each other,
the initial boost velocity increases as a result of the gravitational attraction, leading
to a strong shock as the surfaces of the stars collide. In the case of supercritical initial
data, i.e., of stars with masses above a critical value, Mc , a black hole is promptly
produced and most of the matter is accreted. Conversely, in the case of subcritical
initial data, i.e., of stars with masses below Mc , the product of the collision is a hot
and extended object with large-amplitude oscillations. Part of the stellar matter is
unbound and leaves the numerical grid as the product of the collision reaches an
equilibrium.
Figure 4 shows snapshots at representative times of the rest-mass density, ρ (top
row), of the Lorentz factor, γ ∇ (1−vi vi /c2 )−1/2 (middle row), and of the local fluid
energy, −u 0 (bottom row), for two subcritical models. The left panel, in particular,
refers to a binary boosted at vb /c = 0.3. Note that the stars are strongly compressed by
the collision, with the rest-mass density increasing exponentially. The merged object
expands in a jet-like fashion along the z-direction, with the bulk of the matter being
accelerated up to γ ≈ 16, or equivalently, v/c ≈ 0.998, but then settling on much
slower flows with γ 2.1. Furthermore, the front of the jet has −u 0 > 1 indicating
that part of the shocked matter has sufficient energy to have become gravitationally
unbound. As a result, the rest-mass density at the center of the merged object is
404 L. Rezzolla
smaller than the maximum density of the initial configuration, although the origin
still represents the region where the density is the largest. The right panel, on the
other hand, refers to a highly-boosted binary, i.e., with vb /c = 0.8, with each star
being initially highly distorted by the Lorentz contraction. Also in this case, the
stars are strongly compressed by the collision, but the merged object expands in
a spherical blast-wave fashion, with an almost spherical distribution of matter and
bulk Lorentz factor. The latter reaches values as large as γ ≈ 30, or equivalently,
v/c ≈ 0.999, which, in contrast with the low-boost case, do not decrease in time.
As a comparison, the typical bulk Lorentz factors obtained in the merger of binary
neutron stars in quasi circular orbits is γ ≈ 1.03 [27]. The very large kinetic energies
involved in the collision are sufficient to make a very large portion of the stellar
matter unbound, as clearly shown by the bottom-right panel of Fig. 4, which reports
the local fluid energy. The rest-mass density distribution in the expanding blast wave
has a minimum at the origin, where a large rarefaction is produced by the matter
expanding as an ultrarelativistic thick shell.
The marked transition from a jet-like outflow, not too dissimilar from the simple
Bjorken flow used to model the very early states of relativistic ion-collisions [102],
to a shell-like structure, not too dissimilar from “transverse expansion” modelled in
the subsequent stages of relativistic ion-collisions (see [103] and references therein),
signals that it is not unreasonable to extrapolate some of the results presented here
also to the collision of ultrarelativistic elementary particles.
The transition from the two qualitatively-different regimes discussed above is
further confirmed by the evolution of the rest-mass normalized to the initial value
M0 . The simulations in fact reveal that the unbound fraction is just a few percent of
the total rest-mass in the case of a low-boost collision, with most of the matter being
confined in the selfgravitating “star”. This is to be contrasted with what happens for
a high-boost collision, where the unbound fraction is ≈100 % of the total rest-mass.
This behaviour provides a strong indication that, at least for subcritical collisions, the
role played by gravitational forces is a minor one as the kinetic energy is increased.
This is what happens in the collision of two particles at ultrarelativistic speeds, where
all of the matter is obviously unbound.
3.3 Critical Behaviour and Scaling
A remarkable property of the head-on collision of compact stars is the existence

of type-I critical behaviour, which was first pointed out in [104] and subsequently
extended in [100]. In essence, in these works it was found that when considering
stars with initial zero velocity at infinity, it is possible to fine-tune the initial central
density ρc (and hence the mass) near a critical value, ρνc , so that stars with ρc > ρνc
would collapse eventually to a black hole, while the models with ρc < ρνc would
eventually lead to a stable stellar configuration. As a result, the head-on collision
of two neutron stars near the critical threshold can be seen as a transition in the
space of configurations from an initial stable solution over to a critical metastable
Fig. 5 Critical line as a

function of the average
Lorentz factor, with circles
indicating black holes and
triangles selfgravitating
objects. The inset shows
the regimes explored at LHC
and measured in UHECR
one which can either migrate to a stable solution or collapse to a black hole [101].
As the critical limit is approached, the survival time of the metastable object, τeq ,
increases as τeq = −λ ln |ρc − ρνc |, with λ ≈ 10 [100, 104].
Although the free-fall velocities considered in [100, 104] were very small, the
critical behaviour continues to hold also when the stars are boosted to ultrarelativistic
velocities. Interestingly, the threshold ρνc becomes now a function of the initial effec-
tive boost. Determining ρνc becomes especially challenging as the Lorentz factor is
increased and the dynamics of the matter becomes extremely violent, with very strong
shocks and rarefaction waves. However, it was possible to determine the threshold
for all the range of initial boosts considered, i.e., vb /c ⇒ [0, 0.95], γb ⇒ [1, 3.2],
and even to a reasonable accuracy, e.g., ρνc = (3.288023 ± 0.000003) × 1014 g/cm3 ,
for the initial boost of vb /c = 0.3.
The existence of critical behaviour near which the details of the initial conditions
become irrelevant and which is the same at different boosts, i.e., λ does not depend on
γ nor on ρc (Refs. [104, 105] have shown there is “universality” when varying γ and
fixing ρc ), gives us a wonderful tool to explore the conditions of black-hole formation
also far away from the masses and Lorentz factors considered in this paper. This is
illustrated in Fig. 5, which reports the gravitational mass of the isolated spherical star
as a function of the effective initial Lorentz factor

d V Tμν n μ n ν
⇔γ≤ ∇ , (3)
d V Tμν n μ n ν 0
where Tμν is the stress-energy tensor, n μ is the unit normal to the spatial hyperspace
with proper volume element d V , and the index 0 refers to quantities measured in the
initial unboosted frame. I should stress that the definition of the effective Lorentz
factor (3) is necessary because the stars are extended and thus the Lorentz factor will
be different in different parts of the star. Expression (3), on the other hand, can be
406 L. Rezzolla
taken as ratio of the energies measured in the boosted and unboosted frames, and
hence a generalisation of the Lorentz factor for a point particle (Indeed ⇔γ≤ ∈ 1
for vb ∈ 1). Of course, other parametrizations are possible, still leading to scaling
laws, but with slightly different exponents. Filled circles indicate initial data leading
to a black hole, while triangles indicate initial data leading to a “star”, whereby I
mean an object which is at least in part selfgravitating (orange errorbars provide an
approximate upper limit of ≈8 % to the error in the measurements). Also indicated
as a blue solid line is the critical line separating the two regions of black hole and
star formation (the latter is shown as a shaded region). Clearly, the numerical results
provide a tight fit of the critical line with a power law
Mc 1 1
=K ∗ 0.92 . (4)
M⊗ ⇔γ≤n ⇔γ≤1.03
Expression (4) offers itself to a number of considerations. First, it essentially

expresses the conservation of energy. Second, in the limit of zero initial velocities,
⇔γ≤ ∈ 1, one obtains that Mc ◦ 0.92 M⊗ , so that the corresponding total mass, 2Mc ,
is only ≈ 12 % larger than the maximum mass of the relative spherical-star sequence,
i.e., Mmax = 1.637 M⊗ . Third, in the opposite limit of ⇔γ≤ ∈ ⊥, expression (4)
predicts that the critical mass will go zero. This is indeed what one would expect: as
the kinetic energy diverges, no room is left for selfgravitating matter, which will all
be ejected but for an infinitesimal amount which will go into building the zero-mass
critical black hole. Fourthly, (4) is also in agreement with the results in [95, 97],
whereby one can recognize the black-hole formation as the crossing of the critical
line when moving to larger Lorentz factors while keeping the rest-mass constant.
Finally, using (4) it is possible to probe whether the kinetic energies achieved by
modern particle accelerators, such as the LHC, are sufficient to produce micro black
holes from the collision of two ultrarelativistic particles. Using the results reported
in Ref. [106], the expected energies achieved by LHC in the next couple of years will
be 4 − 7 TeV, so that a proton, whose mass is ≈ 938 MeV ≈ 8.41 × 10−58 M⊗ , can
be accelerated up to γ ≈ 7.5 × 103 . I have therefore reported the range of masses and
Lorentz factors accessible to LHC in the inset in Fig. 5, where it appears as a small
magenta box. Note that the calculations reported here do not intend to be a realistic
approximation of the dynamics of ultrarelativistic particle collisions. However, these
calculations and the presence of a critical behaviour can be used to deduce that the
ranges reachable at the LHC are well below the critical line and thus in the region
where a partially-confined collided object is expected.
Of course, this line of arguments wildly extrapolates our results by almost 60
orders of magnitude in mass (11 in Lorentz factor) and neglects quantum effects and
extra-dimension effects that might be important at Planck-energy scales. Bearing
in mind these caveats, our calculations suggest that the production of micro black
holes at LHC will be unlikely. An additional confirmation that our estimates are not
unreasonable comes from considering the corresponding energy and Lorentz factors
for the observed ultra-high energy cosmic rays (UHECR), that are observed with
energies up to ≈1020 eV (i.e., γ ≈ 1011 ) and for which there is no evidence of black-
hole formation when interacting with the atmosphere [107]. Also in this case, the
relevant range of masses and Lorentz factors is shown in the inset and falls in the
region where no black holes should be produced.
As a final remark I note that the scaling relation (4) can be expressed equivalently
in terms of the original stellar compactness, M/R as

M 1 1
= K⇐ ⇐ ∗ 0.08 . (5)
R c ⇔γ≤n ⇔γ≤1.13
Since Mlab ∇ ⇔γ≤M is the mass in the lab frame, and since R is the largest dimension
in that frame being the transverse one to the motion, the ratio

Mlab 1 1
= K⇐ ⇐ ≈ K ⇐ 0.13 , (6)
R c ⇔γ≤n −1 ⇔γ≤
provides the condition for the amount of energy that, when confined in a hoop of
radius R, would lead to a black hole. Hence, expression (6) extends the spirit of
the hoop conjecture to the case in which a kinetic energy is present. Note that the
limiting value ⇔γ≤ = 1 does not corresponds to a static configuration (as in the hoop
conjecture) but to a binary that is at rest at infinity. This explains why in this limit
(Mlab /R)c = (M/R)c ◦ 0.08, which is considerably smaller than the value 1/2
predicted by the hoop conjecture.
3.4 Summary
The calculations reported above demonstrate that it is possible to find a criterion for
the conditions leading to black-hole formation in the collision of two selfgravitating
fluids moving at ultrarelativistic velocities. The Lorentz factors reached in these
simulations are considerably larger than those encountered in merging neutron-star
binaries, especially if the inspiral is along quasi-circular orbits. The properties of the
flow after the collision change with Lorentz factor, with most of the matter being
ejected in a spherical blast wave for large boosts. Interestingly, the collided object
exhibits a critical behaviour of type I, which is found to persist also as the initial boost
is increased. This allows one to derive a simple scaling law and extrapolate these
results to the energies of elementary particles at LHC and conclude that black-hole
production is unlikely in that case.
4 Third Piece: Horizons as Probes of Black-Hole Dynamics
The third and last “little piece” of numerical relativity that I will discuss is instead
about calculations in vacuum spacetimes and focuses on the merger of two black
holes. This process, which represents one of the most important source of gravitational
408 L. Rezzolla
waves, is generally accompanied by the recoil of the final black hole as a result of
anisotropic gravitational wave emission. While this scenario has been investigated
for decades [108] and first estimates have been made using approximated and semi-
analytical methods such as a particle approximation [109], post-Newtonian meth-
ods [110] and the close-limit approximation (CLA) [111], it is only thanks to the
recent progress in numerical relativity that accurate values for the recoil velocity
have been computed [112–120].
Besides being a genuine nonlinear effect of general relativity, the generation of
a large recoil velocity during the merger of two black holes has a direct impact
in astrophysics. Depending on its size and its variation with the mass ratio and
spin, in fact, it can play an important role in the growth of supermassive black
holes via mergers of galaxies and on the number of galaxies containing black holes.
Numerical-relativity simulations of black holes inspiralling on quasi-circular orbits
have already revealed many of the most important features of this process showing,
for instance, that asymmetries in the mass can lead to recoil velocities vk 175 km/s
[112, 113], while asymmetries in the spins can lead respectively to vk 450 km/s
or vk 4000 km/s if the spins are aligned [115, 116, 118] or perpendicular to the
orbital angular momentum [114, 121, 122] (see [123] for a review and [124] for the
most recent results).
At the same time, however, there are a number of aspects of the nonlinear processes
leading to the recoil that are far from being clarified even though interesting work
has been recently carried out to investigate such aspects [125–127]. One of these
features, and possibly the most puzzling one, is the generic presence of an “anti-
kick”, namely, of one (or more) decelerations experienced by the recoiling black
hole. Such anti-kicks take place after a single apparent horizon has been found and
have been reported in essentially all of the mergers simulated so far (see Fig. 8 of
Ref. [118] for some examples).
What follows discusses a phenomenological framework which provides a novel
description of the stages during which the anti-kick is generated, and that can be
used to formulate a simple and qualitative interpretation of the physics underlying
this process. I will focus on the head-on collision of two nonspinning black holes
with different mass. Although this is the simplest scenario for a black-hole merger,
it contains all the important aspects that can be encountered in more generic condi-
tions.6
4.1 The Basic Picture
I will start by presenting a qualitative interpretation of the antikick by considering

the simple head-on collision of two Schwarzschild black holes with unequal masses.
This qualitative picture will be made quantitative and gauge-invariant by studying the
logical equivalent of this process in the evolution of a Robinson-Trautman spacetime,
6 Much of what follows is taken from the discussion presented in Refs. [128–130].
Fig. 6 Cartoon of the generation of the anti-kick in the head-on collision of two unequal-mass
Schwarzschild black holes. Initially the smaller black hole moves faster and linear momentum is
radiated mostly downwards, thus leading to an upwards recoil of the system [stage (1)]. At the
merger the curvature is higher in the upper hemisphere of the distorted black hole (cf., red-blue
shading) and linear momentum is radiated mostly upwards leading to the anti-kick [stage (2)]. The
black hole decelerates till a uniform curvature is restored on the horizon [stage (3)]
with measurements of the recoil made at future null infinity. The insight gained with
this spacetime will be valuable to explain the anti-kick under generic conditions and
to contribute to the understanding of nonlinear black-hole physics.
Figure 6 illustrates the dynamics of the head-on collision using a schematic cartoon
where I am considering a coordinate system centred in the total centre of mass of the
system and where the smaller black hole is initially on the positive z-axis, while the
larger one is on the negative axis. As the two black holes free-fall towards each other,
the smaller one will move faster and will be more efficient in “forward-beaming” its
gravitational wave emission [110]. As a result, the linear momentum will be radiated
mostly downwards, thus leading to an upwards recoil of the black hole binary [cf.,
stage (1) in Fig. 6]. At the merger, the black-hole velocities will be the largest and so
will also be the anisotropic gravitational wave emission and the corresponding recoil
of the system. However, when a single apparent horizon is formed comprising the two
black holes, the curvature distribution on this 2-surface will be highly anisotropic,
being higher in the upper hemisphere (cf., red-blue shading in stage (2) of Fig. 6).
Because the newly formed black hole will want to radiate all of its deviations away
from the final Schwarzschild configuration, it will do so more effectively there where
the curvature is larger, thus with a stronger emission of gravitational waves from the
northern hemisphere. As a result, after the merger the linear momentum will be
emitted mostly upwards and this sudden change in sign will lead to the anti-kick.
The anisotropic gravitational wave emission will decay exponentially as the curvature
410 L. Rezzolla
gradients are erased and the quiescent black hole reaches its final and decelerated
recoil velocity [cf., stage (3)].7
Although this picture refers to a head-on collision, it is supported by the findings
in the CLA (where the direction of the ringdown kick is approximately opposite to
that of the accumulated inspiral plus plunge kick) [127] and it can be generalized
to a situation in which the black holes have different masses, different spins and
are merging through an inspiral. Also in a more generic case, the newly-formed
apparent horizon will have a complicated but globally anisotropic distribution of
the curvature, determining the direction (which is in general varying in time) along
which the gravitational waves will be emitted. Hence, the geometric properties in a
dynamical horizon (of a black or white hole) determine its global dynamics. I next
use the Robinson-Trautman spacetime to validate this picture.
4.2 A Useful Playground
The Robinson-Trautman spacetime represents a class of vacuum solutions admitting

a congruence of null geodesics which are twist and shear-free [133], with a future
stationary horizon and a dynamical past (outer trapping) horizon [134–138] (past
apparent horizon hereafter). A Robinson-Trautman spacetime is thus regarded as an
isolated nonspherical white hole emitting gravitational waves, where the evolution
of the apparent horizon curvature-anisotropies and the total spacetime momentum
dynamics can be related unambiguously. The metric is given by [139]

2M⊥ 2r ∂u Q r2
ds = − K −
2
− du 2 − 2dudr + dλ2 , (7)
r Q Q2
where Q = Q(u, λ), u is the standard null coordinate, r is the affine parameter of
the outgoing null geodesics, and λ = {θ, φ} are the angular coordinates on the unit
sphere S 2 . Here M⊥ is a constant and is related to the asymptotic mass, while the
function K (u, λ) is the Gaussian curvature of the surface corresponding to r = 1
and u = constant, K (u, λ) ∇ Q 2 (1 + ≥λ 2 ln Q), where ≥ 2 is the Laplacian on S 2 .
λ
The Einstein equations then lead to the evolution equation
∂u Q(u, λ) = −Q 3 ≥λ
2
K (u, λ)/(12M⊥ ). (8)
Any regular initial data Q = Q(0, λ) will smoothly evolve according to (8) until
it achieves a stationary configuration corresponding to a Schwarzschild black hole
at rest or moving with a constant speed [140]. Equation (8) implies the existence of
7 I should remark that other explanations have also been suggested. One of them makes use of the Landau-Lifshitz
pseudotensor and explains the recoil in terms of the cancellation of large and opposite fluxes of momentum, part of
which are “swallowed” by the black hole [131]. Another one is even more essential and explains the antikick is in terms
of the spectral features of the signal at large distances, quite independently of the presence of a black-hole horizon [132].
All of these views serve the scope of providing an intuitive description and are in my view equally valid and useful.

the constant of motion A ∇ S 2 dλ/Q 2 , which clearly represents the area of the
surface u = const., r = const. and can be used to normalise Q so that A = 4π. All
the physically relevant information is contained in the function Q(u, λ), and this
includes the gravitational radiation, which can be extracted by relating Q(u, λ) to
the radiative part of the Riemann tensor [141, 142].
The past apparent horizon radius R(u, λ) is given by the vanishing expansion of
the future ingoing null geodesics, satisfying [134, 135]
Q 2 ≥λ
2
ln R = K − 2M⊥ /R. (9)
The mass and momentum of the black hole are computed at future null infinity
using the Bondi four-momentum [139]

M⊥ ηα
P α (u) ∇ dλ , (10)
4π Q3
S2
with (η α ) = (1, sin θ cos φ, sin θ sin φ, cos θ). Given smooth initial data, the space-
time will evolve to a stationary non-radiative
∪ solution which, in axisymmetry, has
the form Q(⊥, θ) = (1 ∓ vx)/ 1 − v2 , with x ∇ cos θ [139]. The Bondi four-
momentum associated to Q(⊥, θ) has components
(P(⊥))α = M⊥ / 1 − v2 (1, 0, 0, ±v) , (11)
so that the parameter v in Q(⊥, θ) can be interpreted as the velocity of the Schwarz-
schild black hole in the z-direction.
One of the difficulties with Robinson-Trautman spacetimes is the definition of
physically meaningful initial data. Although this is meant more as a proof-of-
principle than a realistic configuration, it is possible to adopt the prescription sug-
gested in Ref. [142]
−2
1 q
Q(0, θ) = Q 0 ∪ +∪ , (12)
1 − wx 1 + wx
and which was interpreted to represent the final stages (i.e., after a common apparent
horizon is formed) of a head-on collision of two boosted black holes with opposite
velocities w and mass ratio q [142]. In practice, to reproduce the situation shown in
Fig. 6, it is sufficient to choose w < 0 and take q ⇒ [0, 1]; a more general class of
initial data and the corresponding phenomenology can be found in [129, 143]. Note
that Q 0 is chosen so that to A = 4π and that in general the deformed black hole will
not be initially at rest. As a result, given the initial velocity v0 ∇ P 3 (0)/P 0 (0), a boost
α 3
is performed transformation P = Φαβ (v0 )P β so that P (0) = 0 by construction.
The numerical solution of Eq. (8) with initial data (12) is performed using a Galerkin
decomposition as discussed in detail in [139].
412 L. Rezzolla
Fig. 7 Typical evolution

of a Robinson-Trautman
spacetime. Shown in the
lower panel is the evolution
of the curvature K AH at the
north (x = 1) and south pole
(x = −1). Shown in the
upper panel is the evolution
of the recoil, which stops
decreasing when the curvature
difference is erased by the
emitted radiation (dotted
line). Note that the curvature
decays exponentially to that
of a Schwarzschild black hole
(inset)
Figure 7 reports the typical evolution of a Robinson-Trautman spacetime with

the lower panel showing the evolution of the curvature of the past apparent horizon
K AH ∇ 2M⊥ /R 3 (x) at the north (x = 1) and south pole (x = −1), and with the
upper panel showing the evolution of the recoil velocity. Note that the two local
curvatures are different initially, with the one in the upper hemisphere being larger
than the one in the lower hemisphere (cf., Fig. 6). However, as the gravitational
radiation is emitted, this difference is erased. When this happens, the deceleration
stops and the black hole attains its asymptotic recoil velocity. The inset reports the
curvature difference relative to the asymptotic Schwarzschild one, K AH − 1, whose
exponentially decaying behaviour is the one expected in a ringing black hole (see
also Fig. 7 of Ref. [129]).
As mentioned before, the one shown in Fig. 7 is a typical evolution of a Robinson-
Trautman spacetime and is not specific of the initial data (12). By varying the values
of w, in fact, it is possible to increase or decreases the final recoil and a sign change
in w simply inverts the curvature at the poles so that, for instance, initial data with
w > 0 would yield a black hole accelerating in the positive z-direction. Interestingly,
it is even possible to fine-tune the parameter w so that the recoil produced for a
Robinson-Trautman spacetime mimics the anti-kick produced by the quasi-circular
inspiral of nonspinning binaries. This is shown in Fig. 8, which reports the recoil
as a function of the symmetric mass ratio ν ∇ q/(1 + q)2 , and where the dashed
line refers to the anti-kick for the inspiral of nonspinning binaries in the CLA [127]
(the parameters chosen, i.e., w = −0.425 and r12 = 2 M, are those minimising the
differences). Considering that the two curves are related only logically and that the
CLA one contains all the information about inspiralling black holes, including the
orbital rotation, the match is surprisingly good.
Fig. 8 Recoil velocity shown as a function of the symmetric mass ratio ν when w = −0.425, with
the dashed line refers to the anti-kick from the inspiral of nonspinning binaries in the CLA [127].
Note that the curve can be thought as composed of two different branches
It is also suggestive to think that the curve in Fig. 8 is actually composed of two
different branches, one of which is characterized by large curvature gradients across
the apparent horizon but small values of the curvature (this is the low-ν branch and is
indicated with squares), while the other is characterized by small curvature gradients
and large values of the curvature (this is the high-ν branch and is indicated with
circles). The same recoil velocity can then be produced by two different values of ν,
for which the effects of large curvature gradients and small local curvatures are the
same as those produced by small curvature gradients but large local curvatures.
To go from this intuition to a mathematically well-defined measure one can com-
pute the mass multipoles of the intrinsic curvature of the initial data using the for-
malism developed in [144] for dynamical horizons. Namely, it is possible to compute
the mass moments as (the mass-current are obviously zero)

Pn (x̃)
Mn ∇ dλ , (13)
Q (θ)R(θ)
2
where Pn (x̃) is the Legendre polynomial in terms of the coordinate x̃(θ) which obeys
∂θ x̃ = − sin θ R(θ)2 /(RAH
2 Q(θ)2 ), with R
AH ∇ AAH /(4π) and x̃(0) = 1. Using
these multipoles it is possible to construct an effective-curvature parameter K eff
that represents a measure of the global curvature properties of the initial data and
414 L. Rezzolla
Fig. 9 Recoil velocity shown as a function of the effective curvature. In contrast with Fig. 8, which
uses the same symbols employed here, the relation between the curvature and the recoil is now
injective
from which the recoil depends in an injective way. Because this effective-curvature
parameter has to contain the contribution from the even and odd multipoles, the
expression

K eff = M2 M2n+1 /3n−1 , (14)
n=1
was found to reproduce exactly what expected (note M1 = 0 to machine precision).

This is shown in Fig. 9, which reports the recoil velocity as a function of K eff . As
predicted, and in contrast with Fig. 8, the relation between the curvature and the recoil
is now injective, with the maximum recoil velocity being given by the maximum
value of K eff (see inset), and with the two branches coinciding. The expression (14)
suggested above for K eff is not unique and indeed a more generic one will have
to include also the mass-current multipoles to account for the spin contributions
(see discussion below). However, lacking a rigorous mathematical guidance, the
phenomenological K eff is a reasonable, intuitive approximation.
4.3 A More General View
Despite the valuable insight, the treatment summarised above and presented
in Ref. [128] had obvious limitations. First, the Ansatz (14) for K eff , i.e., K eff =
f even (M2 ) × f odd (M2+1 ) is not straightforwardly generalizable to the
non-axisymmetric case. Second, the functions f even and f odd can be written in the
simplest
possible form, i.e., as a linear expansion in M ’s, i.e., K eff = (a2 M2 + a4 M4 + · · · )×
(a3 M3 + a5 M5 + · · · ), where the phenomenological coefficients a ’s depend on
the details of the employed initial data. Finally, the white-hole horizon analysis
in Robinson-Trautman spacetimes needs to be extended to the genuine black-hole
horizon case.
While the focus in what discussed above (and presented in Ref. [128]) was on
expressing the difference between the final kick velocity v⊥ and the instantaneous
kick velocity vk (u) at an (initial) given time u, in terms of the geometry of the
common apparent horizon at that time u, it is possible to derive a more generic
view based on geometric quantities that are evaluated at a given time during the
evolution. More specifically, it is possible to consider the variation of the Bondi
linear momentum vector in time (d PiB /du)(u) as the relevant geometric quantity to
monitor at null infinity J + . This quantity can then be correlated with a counterpart
on the black-hole horizon H + , e.g., a vector K̃ eff i (v) (function of an advanced time
v), which represents an extension of the effective curvature introduced in the previous
section.8
In the case of a Robinson-Trautman spacetime, the causal relation between the
white-hole horizon H − and null infinity J + made possible to establish an explicit
functional relation between dvk /du and K eff ⇐ (u). In the case of a generic black-hole
horizon, such a direct causal relation between the inner horizon and J + is lost.
However, since the corresponding causal pasts of J + and H − coincide in part,
non-trivial correlations are still possible and expected. These correlations can be
measured by comparing geometric quantities h inn (v) at H + and h out (u) at J + ,
i (v)
both considered here as two timeseries.9 In particular, it is reasonable to take K̃ eff
as h inn (v) and (d PiB /du)(u) as h out (u).
This approach resembles therefore the methodology adopted in scattering exper-
iments. Gravitational dynamics in a given spacetime region affects the geometry
of appropriately-chosen outer and inner hypersurfaces of the black-hole spacetime.
These hypersurfaces are then understood as test screens on which suitable geometric
quantities must be constructed. The correlations between the two encode geomet-
ric information about the dynamics in the bulk, providing information useful for
an inverse-scattering approach to the near-horizon dynamics. In asymptotically flat
black-hole spacetimes, null infinity J + and the (event) black-hole horizon H +
provide natural choices for the outer and inner screens. This is summarised in the
Carter-Penrose diagram in Fig. 10, which illustrates the cross-correlation approach to
near-horizon gravitational dynamics. The event horizon H + and null infinity J +
provide spacetime screens on which geometric quantities, accounting respectively
for horizon deformations and wave emission, are measured. Their cross-correlation
encodes information about the bulk spacetime dynamics.
8 Another appealing approach that has a similar goal of correlating strong-fields effects with (the visualization of)
spacetime curvature has been proposed recently by the group in Caltech [145, 146].
9 Note that the meaningful definition of timeseries cross-correlations requires the introduction of a (gauge-dependent)
relation between advanced and retarded time coordinates v and u . In an initial value problem this is naturally provided
by the 3 + 1 spacetime slicing by time t .
416 L. Rezzolla
r=0 i+
u
i0
r=0 v
i−
Fig. 10 Carter-Penrose diagram illustrating the scattering approach to near-horizon gravitational

dynamics in a generic spherically symmetric collapse. The event horizon H + and null infinity J +
provide spacetime canonical screens on which geometric quantities, respectively accounting for
horizon deformations and wave emission, are defined. Their cross-correlation encodes nontrivially
information about the bulk spacetime dynamics
The picture offered by Fig. 10 can be easily adapted to the 3+1 approach commonly
adopted in numerical relativity. Since neither the black-hole event horizon nor null
infinity are in general available during the evolution,10 it is possible to adopt as inner
and outer screens a dynamical horizon H + (future outer trapping horizon [147–149])
and a timelike tube B at large spatial distances, respectively. In this case, the time
function t associated with the 3 + 1 spacetime slicing provides a (gauge) mapping
between the retarded and advanced times u and v, so that cross-correlations between
geometric quantities at H + and B can be calculated as standard timeseries h inn (t)
and h out (t). This is summarised in the Carter-Penrose diagram in Fig. 11, which is
the same as in Fig. 10, but where the 3 + 1 slicing sets an in-built common time t for
cross-correlations between the dynamical horizon H + (i.e., the inner screen) and a
large-distance timelike hypersurface B (i.e., the outer screen).
Within this conceptual framework it is then possible to define a phenomenolog-
ical curvature vector K̃ ieff (t) in terms of the mass multipoles of the Ricci scalar
curvature 2R at H + and show that this is closely correlated with a geometric quan-
tities (d PiB /dt)(t), representing the variation of the Bondi linear momentum time
on J + . How to do this in practice for a black-hole spacetime requires much more
space that I can take in this contribution and therefore refer the interested reader to
Refs. [129, 130], where this is discussed in great detail.
10 The latter would properly require either characteristic or a hyperboloidal evolution approach.
r=0 i+
+
t4 H
t3 u
t2 i0
t1
B
r=0 v
i−
Fig. 11 Carter-Penrose diagram for the scattering picture in a Cauchy initial value approach. The
dynamical horizon H + and a large-distance timelike hypersurface B provide inner and outer
screens. Note that the dynamical horizon is split in two portions: outer and inner (solid and dashed
blue lines, respectively) and that the 3+1 slicing sets a common time t for cross-correlations
4.4 Summary
The discussion reported above demonstrates that qualitative aspects of the post-
merger recoil dynamics at infinity can be understood in terms of the evolution of the
geometry of the common horizon of the resulting black hole. Moreover, suitably-built
quantities defined on inner and outer worldtubes (represented either by dynamical
horizons or by timelike boundaries) can act as test screens responding to the space-
time geometry in the bulk, thus opening the way to a cross-correlation approach
to probe the dynamics of spacetime. This picture was shown to hold both for a
simple Robinson-Trautman spacetime, but also for more generic binary black-hole
spacetimes. In this latter case, this is possible through the construction of a phe-
nomenological vector K̃ ieff (t) from the Ricci curvature scalar 2R on the dynamical
horizon sections, which then captures the global properties of the flux of Bondi linear
momentum (d PiB /dt)(t) at infinity, namely the acceleration of the BH.
A geometric framework looking at the horizon’s properties offers a number of
connections with the literature developing around the use of horizons to study the
dynamics of black holes, as well as with the interpretations of such dynamics in
terms of a viscous-hydrodynamics analogy. Much of the machinery developed using
dynamical trapping horizons as inner screens can be extended also when a common
horizon is not formed (as in the calculations reported in Ref. [150]). While in such
cases the identification of an appropriate hypersurface for the inner screen can be
more difficult, once this is found its geometrical properties can be used along the
lines of the cross-correlation approach discussed here for dynamical horizons.
418 L. Rezzolla
5 Conclusions
The “three little piece” for numerical computer and relativity presented in the sec-
tions above ought to provide a reasonable idea of the “Renaissance” that numerical
relativity is now experiencing. More importantly, they should be able to convey the
enormous potential that numerical-relativity simulations have in revealing aspects of
the theory that cannot be handled analytically, or in exploring nonlinear regimes that
cannot be investigated through perturbative approaches. As remarked repeatedly, the
examples brought represent only a personal (and biased) selection of the intense work
carried out recently and surely are not exhaustive in terms of the physical scenarios
that can be explored. Much more can be said about this and surely it will not have to
wait for the bicentenary of Einstein’s stay in Prague.
Acknowledgments The work discussed here has been carried out in collaboration with
M. A. Aloy, L. Baiotti, B. Giacomazzo, J. Granot, J. L. Jaramillo, C. Kouveliotou, R. P. Macedo
K. Takami, whom I am indebted with. My thanks go also to the numerical-relativity group of the
AEI for providing such a stimulating and productive environment. Support comes through the DFG
grant SFB/Trans-regio 7 and “CompStar”, a Research Networking Programme of the ESF. The
calculations have been performed on the clusters at the AEI.
References
1. Pretorius, F.: Evolution of binary black hole spacetimes. Phys. Rev. Lett. 95, 121101 (2005).
3. Baker, J.G., Centrella, J., Choi, D.I., Koppitz, M., van Meter, J.: Gravitational wave extraction
4. Baiotti, L., Rezzolla, L.: Challenging the paradigm of singularity excision in gravitational
collapse. Phys. Rev. Lett. 97, 141101 (2006). doi:10.1103/PhysRevLett.97.141101
5. Chu, T., Pfeiffer, H.P., Scheel, M.A.: High accuracy simulations of black hole binaries: spins
anti-aligned with the orbital angular momentum. Phys. Rev. D 80(12), 124051 (2009). doi:10.
1103/PhysRevD.80.124051
6. Baiotti, L., Giacomazzo, B., Rezzolla, L.: Accurate evolutions of inspiralling neutron-star
binaries: prompt and delayed collapse to a black hole. Phys. Rev. D 78(8), 084033 (2008).
doi:10.1103/PhysRevD.78.084033
7. Centrella, J., Baker, J.G., Kelly, B.J., van Meter, J.R.: Black-hole binaries, gravitational waves,
and numerical relativity. Rev. Mod. Phys. 82, 3069 (2010). doi:10.1103/RevModPhys.82.3069
8. Pfeiffer, H.P.: Numerical simulations of compact object binaries. Class. Quant. Grav. 29(12),
124004 (2012). doi:10.1088/0264-9381/29/12/124004
9. Radice, D., Rezzolla, L.: Discontinuous Galerkin methods for general-relativistic hydrody-
namics: formulation and application to spherically symmetric spacetimes. Phys. Rev. D 84(2),
024010 (2011). doi:10.1103/PhysRevD.84.024010
10. Radice, D., Rezzolla, L.: THC: a new high-order finite-difference high-resolution shock-
capturing code for special-relativistic hydrodynamics. Astron. Astrophys. 547, A26 (2012).
doi:10.1051/0004-6361/201219735
11. Nakamura, T., Oohara, K., Kojima, Y.: General relativistic collapse to black holes and gravi-
tational waves from black holes. Prog. Theor. Phys. Suppl. 90, 1 (1987)
12. Bona, C., Massó, J., Seidel, E., Stela, J.: New formalism for numerical relativity. Phys. Rev.
13. Shibata, M., Nakamura, T.: Evolution of three-dimensional gravitational waves: harmonic
slicing case. Phys. Rev. D 52, 5428 (1995). doi:10.1103/PhysRevD.52.5428
14. Baumgarte, T.W., Shapiro, S.L.: On the numerical integration of Einstein’s field equations.
15. Alcubierre, M., Brügmann, B., Diener, P., et al.: Gauge conditions for long-term numeri-
cal black hole evolutions without excision. Phys. Rev. D 67, 084023 (2003). doi:10.1103/
PhysRevD.67.084023
16. Pretorius, F.: Numerical relativity using a generalized harmonic decomposition. Class. Quant.
Grav. 22, 425 (2005). doi:10.1088/0264-9381/22/2/014
17. Alic, D., Bona, C., Bona-Casas, C.: Towards a gauge-polyvalent numerical relativity code.
Phys. Rev. D 79(4), 044026 (2009). doi:10.1103/PhysRevD.79.044026
18. Bernuzzi, S., Hilditch, D.: Constraint violation in free evolution schemes: comparing BSS-
NOK with a conformal decomposition of Z4. Phys. Rev. D 81, 084003 (2010). doi:10.1103/
PhysRevD.81.084003
19. Müller, D., Grigsby, J., Brügmann, B.: Dynamical shift condition for unequal mass black hole
binaries. Phys. Rev. D 82, 064004 (2010). doi:10.1103/PhysRevD.82.064004
20. Schnetter, E.: Time step size limitation introduced by the BSSN Gamma driver. Class. Quant.
Grav. 27, 167001 (2010). doi:10.1088/0264-9381/27/16/167001
21. Alic, D., Bona-Casas, C., Bona, C., Rezzolla, L., Palenzuela, C.: Conformal and covariant
formulation of the Z4 system with constraint-violation damping. Phys. Rev. D 85, 064040
(2012)
22. Shibata, M., Taniguchi, K.: Merger of binary neutron stars to a black hole: disk mass, short
gamma-ray bursts, and quasinormal mode ringing. Phys. Rev. D 73, 064027 (2006). doi:10.
1103/PhysRevD.73.064027
23. Anderson, M., Hirschmann, E.W., Lehner, L., et al.: Magnetized neutron star mergers and
gravitational wave signals. Phys. Rev. Lett. 100, 191101 (2008)
24. Liu, Y.T., Shapiro, S.L., Etienne, Z.B., Taniguchi, K.: General relativistic simulations of mag-
netized binary neutron star mergers. Phys. Rev. D 78, 024012 (2008). doi:10.1103/PhysRevD.
78.024012
25. Giacomazzo, B., Rezzolla, L., Baiotti, L.: Can magnetic fields be detected during the inspiral
of binary neutron stars? Mon. Not. R. Astron. Soc. 399, L164 (2009). doi:10.1111/j.1745-
3933.2009.00745.x
26. Giacomazzo, B., Rezzolla, L., Stergioulas, N.: Collapse of differentially rotating neutron
stars and cosmic censorship. Phys. Rev. D 84(2), 024022 (2011). doi:10.1103/PhysRevD.84.
024022
27. Rezzolla, L., Baiotti, L., Giacomazzo, B., Link, D., Font, J.A.: Accurate evolutions of unequal-
mass neutron-star binaries: properties of the torus and short GRB engines. Class. Quant. Grav.
27(11), 114105 (2010). doi:10.1088/0264-9381/27/11/114105
28. Kiuchi, K., Sekiguchi, Y., Shibata, M., Taniguchi, K.: Exploring binary-neutron-star-merger
scenario of short-gamma-ray bursts by gravitational-wave observation. Phys. Rev. Lett.
104(14), 141101 (2010). doi:10.1103/PhysRevLett.104.141101
29. Baiotti, L., Damour, T., Giacomazzo, B., Nagar, A., Rezzolla, L.: Analytic modelling of tidal
effects in the relativistic inspiral of binary neutron stars. Phys. Rev. Lett. 105, 261101 (2010).
30. Bernuzzi, S., Nagar, A., Thierfelder, M., Brügmann, B.: Tidal effects in binary neutron star
coalescence. Phys. Rev. D 86(4), 044030 (2012). doi:10.1103/PhysRevD.86.044030
31. Hotokezaka, K., Kyutoku, K., Shibata, M.: Exploring tidal effects of coalescing binary neutron
stars in numerical relativity. Phys. Rev. D 87(4), 044001 (2013). doi:10.1103/PhysRevD.87.
044001
420 L. Rezzolla
32. Abadie, J., Abbott, B.P., Abbott, R., et al.: Topical review: predictions for the rates of compact
binary coalescences observable by ground-based gravitational-wave detectors. Class. Quant.
Grav. 27(17), 173001 (2010). doi:10.1088/0264-9381/27/17/173001
33. Paczynski, B.: Gamma-ray bursters at cosmological distances. Astrophys. J. Lett. 308, L43
(1986). doi:10.1086/184740
34. Eichler, D., Livio, M., Piran, T., Schramm, D.N.: Nucleosynthesis, neutrino bursts and gamma-
rays from coalescing neutron stars. Nature 340, 126 (1989). doi:10.1038/340126a0
35. Narayan, R., Paczynski, B., Piran, T.: Gamma-ray bursts as the death throes of massive binary
stars. Astrophys. J. 395, L83 (1992)
36. Ruffert, M., Janka, H.T.: Gamma-ray bursts from accreting black holes in neutron star mergers.
Astron. Astrophys. 344, 573 (1999)
37. Rosswog, S., Ramirez-Ruiz, E., Davies, M.B.: High-resolution calculations of merging neu-
tron stars - III. Gamma-ray bursts. Mon. Not. R. Astron. Soc. 345, 1077 (2003). doi:10.1046/
j.1365-2966.2003.07032.x
38. Nakar, E.: Short-hard gamma-ray bursts. Phys. Rep. 442, 166 (2007). doi:10.1016/j.physrep.
2007.02.005
39. Lee, W.H., Ramirez-Ruiz, E.: The progenitors of short gamma-ray bursts. New J. Phys. 9, 17
(2007). doi:10.1088/1367-2630/9/1/017
40. Fox, D.B., Frail, D.A., Price, P.A., et al.: The afterglow of GRB050709 and the nature of the
short-hard gamma-ray bursts. Nature 437, 845 (2005). doi:10.1038/nature04189
41. Prochaska, J.X., Bloom, J.S., Chen, H.W., et al.: The galaxy hosts and large-scale environments
of short-hard gamma-ray bursts. Astrophys. J. 642, 989 (2006). doi:10.1086/501160
42. Rezzolla, L., Giacomazzo, B., Baiotti, L., et al.: The missing link: merging neutron stars
naturally produce jet-like structures and can power short gamma-ray bursts. Astrophys. J.
732(11), L6 (2011). doi:10.1088/2041-8205/732/1/L6
43. Giacomazzo, B., Rezzolla, L.: WhiskyMHD: a new numerical code for general relativistic
magnetohydrodynamics. Class. Quant. Grav. 24, S235 (2007). doi:10.1088/0264-9381/24/
12/S16
44. Pollney, D., Reisswig, C., Rezzolla, L., et al.: Recoil velocities from equal-mass binary black-
hole mergers: a systematic investigation of spin-orbit aligned configurations. Phys. Rev. D
76, 124002 (2007)
45. Baiotti, L., Hawke, I., Montero, P., Rezzolla, L.: A new three-dimensional general-relativistic
hydrodynamics code. In: Capuzzo-Dolcetta R. (ed.) Computational Astrophysics in Italy:
Methods and Tools, vol. 1, p. 210. MSAIt, Trieste (2003)
46. Baiotti, L., Hawke, I., Montero, P.J., et al.: Three-dimensional relativistic simulations of
rotating neutron star collapse to a Kerr black hole. Phys. Rev. D 71, 024035 (2005)
47. Antón, L., Zanotti, O., Miralles, J.A., et al.: Numerical 3+1 general relativistic magnetohy-
drodynamics: a local characteristic approach. Astrophys. J. 637, 296 (2006). doi:10.1086/
498238
48. Toth, G.: The div B = 0 constraint in shock-capturing magnetohydrodynamics codes. J.
Comput. Phys. 161, 605 (2000). doi:10.1006/jcph.2000.6519
49. Schnetter, E., Hawley, S.H., Hawke, I.: Evolutions in 3D numerical relativity using fixed mesh
refinement. Class. Quant. Grav. 21(6), 1465 (2004). doi:10.1088/0264-9381/21/6/014
50. Kouveliotou, C., Meegan, C.A., Fishman, G.J., et al.: Identification of two classes of gamma-
ray bursts. Astrophys. J. 413, L101 (1993). doi:10.1086/186969
51. Dessart, L., Ott, C.D., Burrows, A., Rosswog, S., Livne, E.: Neutrino signatures and the
neutrino-driven wind in binary neutron star mergers. Astrophys. J. 690, 1681 (2009). doi:10.
1088/0004-637X/690/2/1681
52. Setiawan, S., Ruffert, M., Janka, H.: Non-stationary hyperaccretion of stellar-mass black holes
in three dimensions: torus evolution and neutrino emission. Mon. Not. R. Astron. Soc. 352,
753 (2004). doi:10.1111/j.1365-2966.2004.07974.x
53. McKinney, J.C., Uzdensky, D.A.: A reconnection switch to trigger gamma-ray burst jet dis-
sipation. Mon. Not. R. Astron. Soc. 419, 573 (2012). doi:10.1111/j.1365-2966.2011.19721.
x
54. Price, R.H., Rosswog, S.: Producing ultrastrong magnetic fields in neutron star mergers.
Science 312, 719 (2006). doi:10.1126/science.1125201
55. Aloy, M.A., Janka, H., Müller, E.: Relativistic outflows from remnants of compact object
mergers and their viability for short gamma-ray bursts. Astron. Astrophys. 436, 273 (2005).
doi:10.1051/0004-6361:20041865
56. Komissarov, S.S., Vlahakis, N., Königl, A., Barkov, M.V.: Magnetic acceleration of ultrarela-
tivistic jets in gamma-ray burst sources. Mon. Not. R. Astron. Soc. 394, 1182 (2009). doi:10.
1111/j.1365-2966.2009.14410.x
57. Balbus, S.A., Hawley, J.F.: Instability, turbulence, and enhanced transport in accretion disks.
Rev. Mod. Phys. 70, 1 (1998). doi:10.1103/RevModPhys.70.1
58. Siegel, D.M., Ciolfi, R., Harte, A.I., Rezzolla, L.: On the magnetorotational instability in
relativistic hypermassive neutron stars. ArXiv e-prints [arXiv:1302.4368[gr-qc]] (2013)
59. Giacomazzo, B., Rezzolla, L., Baiotti, L.: Accurate evolutions of inspiralling and magne-
tized neutron stars: equal-mass binaries. Phys. Rev. D 83(4), 044014 (2011). doi:10.1103/
PhysRevD.83.044014
60. Obergaulinger, M., Cerdá-Durán, P., Müller, E., Aloy, M.A.: Semi-global simulations of
the magneto-rotational instability in core collapse supernovae. Astron. Astrophys. 498, 241
(2009). doi:10.1051/0004-6361/200811323
61. Blandford, R.D., Znajek, R.L.: Electromagnetic extraction of energy from Kerr black holes.
Mon. Not. R. Astron. Soc. 179, 433 (1977)
62. Komissarov, S.S., Barkov, M.V.: Activation of the Blandford-Znajek mechanism in collapsing
stars. Mon. Not. R. Astron. Soc. 397, 1153 (2009). doi:10.1111/j.1365-2966.2009.14831.x
63. Radice, D., Rezzolla, L.: Universality and intermittency in relativistic turbulent flows of a hot
plasma. Astrophys. J. 766, L10 (2013). doi:10.1088/2041-8205/766/1/L10
64. Zrake, J., MacFadyen, A.I.: Spectral and intermittency properties of relativistic turbulence.
Astrophys. J. 763, L12 (2013). doi:10.1088/2041-8205/763/1/L12
65. Etienne, Z.B., Paschalidis, V., Shapiro, S.L.: General-relativistic simulations of black-hole-
neutron-star mergers: effects of tilted magnetic fields. Phys. Rev. D 86(8), 084026 (2012).
doi:10.1103/PhysRevD.86.084026
66. Lesur, G., Ogilvie, G.I.: Localized magnetorotational instability and its role in the accretion
disc dynamo. Mon. Not. R. Astron. Soc. 391, 1437 (2008). doi:10.1111/j.1365-2966.2008.
13993.x
67. Aloy, M.A., Rezzolla, L.: A powerful hydrodynamic booster for relativistic jets. Astrophys.
J. 640, L115 (2006). doi:10.1086/503608
68. Granot, J., Komissarov, S.S., Spitkovsky, A.: Impulsive acceleration of strongly magnetized
relativistic flows. Mon. Not. R. Astron. Soc. 411, 1323 (2011). doi:10.1111/j.1365-2966.
2010.17770.x
69. Dionysopoulou, K., Alic, D., Palenzuela, C., Rezzolla, L., Giacomazzo, B.: General-
relativistic resistive magnetohydrodynamics in three dimensions: formulation and tests. ArXiv
e-prints 1208.3487 [gr-qc] (2012)
70. Bloom, J.S., Prochaska, J.X., Pooley, D., et al.: Closing in on a short-hard burst progenitor:
constraints from early-time optical imaging and spectroscopy of a possible host galaxy of
GRB 050509b. Astrophys. J. 638, 354 (2006). doi:10.1086/498107
71. Ackermann, M., Asano, K., Atwood, W.B., et al.: Fermi observations of GRB 090510: a
short-hard gamma-ray burst with an additional, hard power-law component from 10 keV TO
GeV energies. Astrophys. J. 716, 1178 (2010). doi:10.1088/0004-637X/716/2/1178
72. Oechslin, R., Janka, H.T.: Torus formation in neutron star mergers and well-localized short
gamma-ray burst. Mon. Not. R. Astron. Soc. 368, 1489 (2006)
73. Birkl, R., Aloy, M.A., Janka, H.T., Müller, E.: Neutrino pair annihilation near accreting, stellar-
mass black holes. Astron. Astrophys. 463, 51 (2007). doi:10.1051/0004-6361:20066293
74. Lee, H.K., Wijers, R.A.M.J., Brown, G.E.: The Blandford-Znajek process as a central engine
for a gamma-ray burst. Phys. Rep. 325, 83 (2000). doi:10.1016/S0370-1573(99)00084-8
75. Granot, J., Königl, A., Piran, T.: Implications of the early X-ray afterglow light curves of
Swift gamma-ray bursts. Mon. Not. R. Astron. Soc. 370, 1946 (2006). doi:10.1111/j.1365-
2966.2006.10621.x
422 L. Rezzolla
76. Abdo, A.A., Ackermann, M., Asano, K., et al.: Fermi observations of high-energy gamma-ray
emission from GRB 080825C. Astrophys. J. 707, 580 (2009). doi:10.1088/0004-637X/707/
1/580
77. Granot, J., Cohen-Tanugi, J., do Couto e Silva, E.: Opacity buildup in impulsive relativistic
sources. Astrophys. J. 677, 92 (2008). doi:10.1086/526414
78. De Pasquale, M., Schady, P., Kuin, N.P.M., et al.: Swift and Fermi observations of the early
afterglow of the short gamma-ray burst 090510. Astrophys. J. 709, L146 (2010). doi:10.1088/
2041-8205/709/2/L146
79. Rossi, E., Lazzati, D., Rees, M.J.: Afterglow light curves, viewing angle and the jet structure
of γ-ray bursts. Mon. Not. R. Astron. Soc. 332, 945 (2002). doi:10.1046/j.1365-8711.2002.
05363.x
80. Peng, F., Königl, A., Granot, J.: Two-component jet models of gamma-ray burst sources.
Astrophys. J. 626, 966 (2005). doi:10.1086/430045
81. Wald, R.M.: General Relativity. The University of Chicago Press, Chicago (1984)
82. Thorne, K.: Nonspherical gravitational collapse: a short review. In: Klauder, J. (ed.) Magic
Without Magic: John Archibald Wheeler, p. 231. Freeman, San Francisco (1972)
83. Senovilla, J.M.M.: A reformulation of the hoop conjecture. Europhys. Lett. 81, 20004 (2008).
doi:10.1209/0295-5075/81/20004
84. Argyres, P.C., Dimopoulos, S., March-Russell, J.: Black holes and sub-millimeter dimensions.
Phys. Lett. B 441, 96 (1998). doi:10.1016/S0370-2693(98)01184-8
85. Yoshino, H., Nambu, Y.: Black hole formation in the grazing collision of high-energy particles.
86. Yoo, C.M., Ishihara, H., Kimura, M., Tanzawa, S.: Hoop conjecture and the horizon formation
cross section in Kaluza-Klein spacetimes. Phys. Rev. D 81(2), 024020 (2010). doi:10.1103/
PhysRevD.81.024020
87. Dimopoulos, S., Landsberg, G.: Black holes at the Large Hadron Collider. Phys. Rev. Lett.
87(16), 161602 (2001). doi:10.1103/PhysRevLett.87.161602
88. Sperhake, U., Cardoso, V., Pretorius, F., Berti, E., Gonzalez, J.A.: The high-energy collision of
two black holes. Phys. Rev. Lett. 101, 161101 (2008). doi:10.1103/PhysRevLett.101.161101
89. Shibata, M., Okawa, H., Yamamoto, T.: High-velocity collision of two black holes. Phys. Rev.
D 78(10), 101501 (2008). doi:10.1103/PhysRevD.78.101501
90. Sperhake, U., Cardoso, V., Pretorius, F., et al.: Cross section, final spin, and zoom-whirl behav-
ior in high-energy black-hole collisions. Phys. Rev. Lett. 103(13), 131102 (2009). doi:10.
91. Eardley, D.M., Giddings, S.B.: Classical black hole production in high-energy collisions.
92. D’Eath, P.D., Payne, P.N.: Gravitational radiation in black-hole collisions at the speed of light.
I. Perturbation treatment of the axisymmetric collision. Phys. Rev. D 46, 658 (1992). doi:10.
1103/PhysRevD.46.658
II. Reduction to two independent variables and calculation of the second-order news function.
III. Results and conclusions. Phys. Rev. D 46, 694 (1992). doi:10.1103/PhysRevD.46.694
95. Choptuik, M.W., Pretorius, F.: Ultrarelativistic particle collisions. Phys. Rev. Lett. 104(11),
111101 (2010). doi:10.1103/PhysRevLett.104.111101
96. Rezzolla, L., Takami, K.: Black-hole production from ultrarelativistic collisions. Class. Quant.
Grav. 30(1), 012001 (2013). doi:10.1088/0264-9381/30/1/012001
97. East, W.E., Pretorius, F.: Ultrarelativistic black hole formation. ArXiv e-prints 1210.0443
[gr-qc] (2012)
98. Rischke, D.H., Bernard, S., Maruhn, J.A.: Relativistic hydrodynamics for heavy-ion collisions.
I. General aspects and expansion into vacuum. Nucl. Phys. A 595, 346 (1995). doi:10.1016/
0375-9474(95)00355-1
99. Kellerman, T., Rezzolla, L., Radice, D.: Critical phenomena in neutron stars: II. Head-on
collisions. Class. Quant. Grav. 27(23), 235016 (2010). doi:10.1088/0264-9381/27/23/235016
100. Kellerman, T., Baiotti, L., Giacomazzo, B., Rezzolla, L.: An improved formulation of the
relativistic hydrodynamics equations in 2D Cartesian coordinates. Class. Quant. Grav. 25(22),
225007 (2008). doi:10.1088/0264-9381/25/22/225007
101. Radice, D., Rezzolla, L., Kellerman, T.: Critical phenomena in neutron stars: I. Linearly
unstable nonrotating models. Class. Quant. Grav. 27(23), 235015 (2010). doi:10.1088/0264-
9381/27/23/235015
102. Bjorken, J.D.: Highly relativistic nucleus-nucleus collisions: the central rapidity region. Phys.
103. Huovinen, P., Ruuskanen, P.V.: Hydrodynamic models for heavy ion collisions. Annu. Rev.
Nucl. Part. Sci. 56, 163 (2006). doi:10.1146/annurev.nucl.54.070103.181236
104. Jin, K.J., Suen, W.M., et al.: Critical phenomena in head-on collision of neutron stars. Phys.
Rev. Lett. 98, 131101 (2007)
105. Wan, M.B.: Universality and properties of neutron star type I critical collapses. Class. Quant.
Grav. 28(15), 155002 (2011). doi:10.1088/0264-9381/28/15/155002
106. Heuer, R.D.: News from CERN, LHC status and strategy for linear colliders. ArXiv e-prints
[arXiv:1202.5860 physics.acc-ph] (2012)
107. Blasi, P.: On the origin of very high energy cosmic rays. Mod. Phys. Lett. A20, 3055 (2005)
108. Peres, A.: Classical radiation recoil. Phys. Rev. 128, 2471 (1962). doi:10.1103/PhysRev.128.
2471
109. Fitchett, M.J., Detweiler, S.: Linear momentum and gravitational waves - circular orbits around
a Schwarzschild black hole. Mon. Not. R. Astron. Soc. 211, 933 (1984)
110. Wiseman, A.G.: Coalescing binary systems of compact objects to (post)5/2 Newtonian order.
2. Higher order wave forms and radiation recoil. Phys. Rev. D 46, 1517 (1992). doi:10.1103/
PhysRevD.46.1517
111. Andrade, Z., Price, R.H.: Head-on collisions of unequal mass black holes: close-limit predic-
tions. Phys. Rev. D 56, 6336 (1997). doi:10.1103/PhysRevD.56.6336
112. Baker, J.G., Centrella, J., Choi, D.I., et al.: Getting a kick out of numerical relativity. Astrophys.
J. 653, L93 (2006). doi:10.1086/510448
113. Gonzalez, J.A., Sperhake, U., Bruegmann, B., Hannam, M., Husa, S.: Total recoil: the maxi-
mum kick from nonspinning black-hole binary inspiral. Phys. Rev. Lett. 98, 091101 (2007)
114. Campanelli, M., Lousto, C.O., Zlochower, Y., Merritt, D.: Maximum gravitational recoil.
115. Herrmann, F., Hinder, I., Shoemaker, D., Laguna, P., Matzner, R.A.: Gravitational recoil from
spinning binary black hole mergers. Astrophys. J. 661, 430 (2007). doi:10.1086/513603
116. Koppitz, M., et al.: Recoil velocities from equal-mass binary-black-hole mergers. Phys. Rev.
117. Lousto, C.O., Zlochower, Y.: Further insight into gravitational recoil. Phys. Rev. D 77, 044028
(2008). doi:10.1103/PhysRevD.77.044028
118. Pollney, D., et al.: Recoil velocities from equal-mass binary black-hole mergers: a systematic
investigation of spin-orbit aligned configurations. Phys. Rev. D 76, 124002 (2007). doi:10.
1103/PhysRevD.76.124002
119. Healy, J., et al.: Superkicks in hyperbolic encounters of binary black holes. Phys. Rev. Lett.
102, 041101 (2009). doi:10.1103/PhysRevLett.102.041101
120. Lousto, C.O., Zlochower, Y.: Hangup kicks: still larger recoils by partial spin-orbit alignment
of black-hole binaries. Phys. Rev. Lett. 107(23), 231102 (2011). doi:10.1103/PhysRevLett.
107.231102
121. Campanelli, M., Lousto, C.O., Zlochower, Y., Merritt, D.: Large merger recoils and spin flips
from generic black-hole binaries. Astrophys. J. 659, L5 (2007)
122. Gonzalez, J.A., Hannam, M.D., Sperhake, U., Bruegmann, B., Husa, S.: Supermassive kicks
for spinning black holes. Phys. Rev. Lett. 98, 231101 (2007). doi:10.1103/PhysRevLett.98.
231101
424 L. Rezzolla
123. Rezzolla, L.: Modelling the final state from binary black-hole coalescences. Class. Quant.
Grav. 26, 094023 (2009). doi:10.1088/0264-9381/26/9/094023
124. Lousto, C.O., Zlochower, Y.: Nonlinear gravitational recoil from the mergers of precessing
black-hole binaries. ArXiv e-prints 1211.7099 [gr-qc] (2012)
125. Schnittman, J.D., Buonanno, A., van Meter, J.R., et al.: Anatomy of the binary black hole
recoil: a multipolar analysis. Phys. Rev. D 77(4), 044031 (2008). doi:10.1103/PhysRevD.77.
044031
126. Mino, Y., Brink, J.: Gravitational radiation from plunging orbits—perturbative study. Phys.
127. Le Tiec, A., Blanchet, L., Will, C.M.: Gravitational-wave recoil from the ringdown phase of
coalescing black hole binaries. Class. Quant. Grav. 27, 012001 (2010). doi:10.1088/0264-
9381/27/1/012001
128. Rezzolla, L., Macedo, R.P., Jaramillo, J.L.: Understanding the ’anti-kick’ in the merger
of binary black holes. Phys. Rev. Lett. 104, 221101 (2010). doi:10.1103/PhysRevLett.104.
221101
129. Jaramillo, J.L., Macedo, R.P., Moesta, P., Rezzolla, L.: Black-hole horizons as probes of
black-hole dynamics. I. Post-merger recoil in head-on collisions. Phys. Rev. D 85, 084030
(2012). doi:10.1103/PhysRevD.85.084030
130. Jaramillo, J.L., Macedo, R.P., Moesta, P., Rezzolla, L.: Black-hole horizons as probes of
black-hole dynamics. II. Geometrical insights. Phys. Rev. D 85, 084031 (2012). doi:10.1103/
PhysRevD.85.084031
131. Lovelace, G., et al.: Momentum flow in black-hole binaries: II. Numerical simulations of
equal-mass, head-on mergers with antiparallel spins. Phys. Rev. D 82, 064031 (2010). doi:10.
1103/PhysRevD.82.064031
132. Price, R.H., Khanna, G., Hughes, S.A.: Systematics of black hole binary inspiral kicks and
the slowness approximation. Phys. Rev. D 83(12), 124002 (2011). doi:10.1103/PhysRevD.
83.124002
133. Robinson, I., Trautman, A.: Some spherical gravitational waves in general relativity. Proc.
Roy. Soc. Lond. A265, 463 (1962). doi:10.1098/rspa.1962.0036
134. Penrose, R.: Naked singularities. Ann. N.Y. Acad. Sci. 224, 125 (1973). doi:10.1111/j.1749-
6632.1973.tb41447.x
135. Tod, K.P.: Analogue of the past horizon in the Robinson-Trautman metrics. Class. Quant.
Grav. 6, 1159 (1989). doi:10.1088/0264-9381/6/8/015
136. Chow, E.W.M., Lun, A.W.C.: Apparent horizons in vacuum Robinson-Trautman space-times.
J. Austral. Math. Soc. Serv. B 41, 217 (1999)
137. Natorf, W., Tafel, J.: Horizons in Robinson-Trautman space-times. Class. Quant. Grav. 25,
195012 (2008). doi:10.1088/0264-9381/25/19/195012
138. Podolsky, J., Svitek, O.: Past horizons in Robinson-Trautman spacetimes with a cosmological
constant. Phys. Rev. D 80, 124042 (2009). doi:10.1103/PhysRevD.80.124042
139. Macedo, R.P., Saa, A.: Gravitational wave recoil in Robinson-Trautman spacetimes. Phys.
140. Chrusciel, P.T.: Semi-global existence and convergence of solutions of the Robinson-Trautman
(2-dimensional Calabi) equation. Commun. Math. Phys. 137, 289 (1991). doi:10.1007/
BF02431882
141. de Oliveira, H.P., Damiao, I.: Soares, gravitational wave emission from a bounded source: the
nonlinear regime. Phys. Rev. D 70, 084041 (2004). doi:10.1103/PhysRevD.70.084041
142. Aranha, R.F., de Oliveira, H.P., Soares, I.D., Tonini, E.V.: The efficiency of gravitational
bremsstrahlung production in the collision of two Schwarzschild black holes. Int. J. Mod.
Phys. D 17, 2049 (2008). doi:10.1142/S0218271808013704
143. Aranha, R.F., Soares, I.D., Tonini, E.V.: Mass-energy and momentum extraction by gravita-
tional wave emission in the merger of two colliding black holes: the non-head-on case. Phys.
144. Ashtekar, A., Engle, J., Pawlowski, T., Van Den Broeck, C.: Multipole moments of isolated
horizons. Class. Quant. Grav. 21, 2549 (2004). doi:10.1088/0264-9381/21/11/003
145. Owen, R., Brink, J., Chen, Y., et al.: Frame-dragging vortexes and tidal tendexes attached
to colliding black holes: visualizing the curvature of spacetime. Phys. Rev. Lett. 106(15),
151101 (2011). doi:10.1103/PhysRevLett.106.151101
146. Nichols, D.A., Owen, R., Zhang, F., et al.: Visualizing spacetime curvature via frame-drag
vortexes and tidal tendexes: general theory and weak-gravity applications. Phys. Rev. D 84,
124014 (2011). doi:10.1103/PhysRevD.84.124014
147. Hayward, S.A.: General laws of black hole dynamics. Phys. Rev. D 49(12), 6467 (1994).
doi:10.1103/PhysRevD.49.6467
148. Ashtekar, A., Krishnan, B.: Dynamical horizons and their properties. Phys. Rev. D 68, 104030
(2003). doi:10.1103/PhysRevD.68.104030
149. Ashtekar, A., Krishnan, B.: Isolated and dynamical horizons and their applications. Living
Rev. Relativ. 7, 10 (2004)
150. Sperhake, U., Berti, E., Cardoso, V., Pretorius, F., Yunes, N.: Superkicks in ultrarelativistic
encounters of spinning black holes. Phys. Rev. D 83, 024037 (2011). doi:10.1103/PhysRevD.
83.024037
Instabilities of Relativistic Stars
John L. Friedman and Nikolaos Stergioulas
Abstract Stable relativistic stars in uniform rotation form a two-parameter family,

parametrized by mass and angular velocity. Limits on each of these quantities are
associated with relativistic instabilities. A radial instability to gravitational collapse
or explosion sets upper and lower limits on their mass, and an instability driven by
gravitational waves may set an upper limit on their spin. Our summary of relativistic
stability theory given here is based on and includes excerpts from the book Rotating
Relativistic Stars, by the present authors.
1 Introduction
A neutron star in equilibrium is accurately approximated by a stationary self-

gravitating perfect fluid.1 The character of its oscillations and their stability, however,
depend on bulk and shear viscosity, on the superfluid nature of its interior, and—for
modes near the surface—on the properties of the crust and the strength of its magnetic
field.
1 Departures from the local isotropy of a perfect fluid are associated with the crust; with
magnetic fields that are thought to be confined to flux tubes in the superfluid interior; and
with a velocity field whose vorticity is similarly confined to vortex tubes. Departures from
perfect fluid equilibrium due to a solid crust are expected to be smaller than one part in ≈10−3 ,
corresponding to the maximum strain that an electromagnetic lattice can support. The vortex
tubes are closely spaced; but the velocity field averaged over meter scales is that of a uniformly
rotating configuration. Finally, the magnetic field contributes negligibly to the pressure support
of the star, even in magnetars with fields of 1015 G.
J. L. Friedman (B)
Department of Physics, University of Wisconsin-Milwaukee, Milwaukee, WI 53211, USA
e-mail: friedman@uwm.edu
N. Stergioulas
Department of Physics, Aristotle University of Thessaloniki, 56124Thessaloniki, Greece
428 J. L. Friedman and N. Stergioulas
The stability of a rotating star is governed by the sign of the energy of its
perturbations; and the amplitude of an oscillation that is damped or driven by gravi-
tational radiation is governed by the rate at which its energy and angular momentum
are radiated. Noether’s theorem relates the stationarity and axisymmetry of the equi-
librium star to conserved currents constructed from the perturbed metric and fluid
variables. Their integrals, the canonical energy and angular momentum on a hyper-
surface can each be written as a functional quadratic in the perturbation, and the
conservation laws express their change in terms of the flux of gravitational waves
radiated to null infinity.
We begin with an action for perturbations of a rotating star from which these
conserved quantities are obtained [1–6]. We next review local stability to convec-
tion and to differential rotation. A spherical star that is stable against convection is
stable to all nonradial perturbations: Only the radial instability to collapse (or explo-
sion) can remain. Instability to collapse sets upper and lower limits on the masses
of stable relativistic stars, the analog for neutron stars of the Chandrasekhar limit.
A turning-point criterion governs this axisymmetric instability of spherical stars
against collapse and provides a sufficient condition for instability of rotating stars.
Finally, we consider the additional instabilities of rotating stars. These are nonax-
isymmetric instabilities that radiate gravitational waves. They may set an upper limit
on the spin of old neutron stars spun up by accretion and on nascent stars that form
with rapid enough rotation.
2 Action and Canonical Energy
The equations governing a perfect fluid are the Einstein equation and the equation
of motion of the fluid,
G εγ = 8π Tεγ , ⊗γ T εγ = 0, (1)
together with an equation of state. We denote by p, ν, λ and u ε the fluid’s pressure,

energy density, rest-mass density and 4-velocity, respectively, and define a tensor
q ε γ = Φγε + u ε u γ (2)
that is the spatial projection operator, the projection orthogonal to u ε . The stress-
energy tensor then has the form
T εγ = νu ε u γ + pq εγ .
Because the spatial projection q ε α ⊗γ T γα = 0 is the relativistic Euler equation,
q εγ ⊗γ p
u γ ⊗γ u ε = − , (3)
ν+ p
we call Eq. (1) the Einstein-Euler equations.
Instabilities of Relativistic Stars 429
Fig. 1 For small s, a

Lagrangian displacement
∂ ε can be regarded as a
vector for which s∂ ε joins the
position x of a fluid element in
an initial fluid flow to its
position βs (x) in the
perturbed fluid flow
One can obtain an action for stellar perturbations by introducing a Lagrangian

displacement ∂ ε joining each unperturbed fluid trajectory (the unperturbed worldline
of a fluid element) to the corresponding trajectory of the perturbed fluid, as shown
in Fig. 1.
The perturbative description is made precise by introducing a family of (time
dependent) solutions

Q(ω) = gεγ (ω), u ε (ω), λ(ω), s(ω) , (4)
and comparing to first order in ω the perturbed variables Q(ω) with their equilibrium
values Q(0). Eulerian and Lagrangian changes in the fluid variables are defined by

d
ΦQ := Q(ω) , ρQ = (Φ + Lξ )Q, (5)
dω ω=0
with Lξ the Lie derivative along ∂ ε .

Because oscillations of a neutron star proceed on a dynamical timescale, a
timescale faster than that of heat flow, one requires that the Lagrangian change ρs
in the entropy per unit rest mass vanishes. With this condition, ∂ ε and h εγ := Φgεγ
completely specify a perturbation of a perfect-fluid spacetime with an equation of
state of the form ν = ν(λ, s), p = p(λ, s). Perturbations of u ε , λ and ν are given by
1 ε γ α 1 1
ρu ε = u u u ρgγα , ρλ = − λq εγ ρgεγ , ρν = − (ν + p)q εγ ρgεγ ,
2 2 2
(6)
with ρgεγ = h εγ + ⊗ε ∂γ + ⊗γ ∂ε . Our restriction to adiabatic perturbations means
that the Lagrangian change in the pressure is given by
ρp ρλ 1
= Θ1 = − Θ1 q εγ ρgεγ , (7)
p λ 2
where the adiabatic index Θ1 is defined by
τ log p(λ, s) ν + p τ p(ν, s)

Θ1 = = . (8)
τ log λ p τν
The perturbed Einstein-Euler equations,
Φ(G εγ − 8π T εγ ) = 0, Φ(⊗γ T εγ ) = 0, (9)
are self-adjoint in a weak and 4-dimensional sense that they are a symmetric system
∂ ε,
up to a total divergence: For any pairs (∂ ε , h εγ ) and ( h εγ ), the symmetry relation
has the form
1

∂γ Φ(⊗α T γα |g|) + ∂,
h γα Φ (G γα − 8π T γα ) |g| = − 2L ( h; ∂, h)
16π
+ ⊗γ ξ γ , (10)
where L is symmetric under the interchange of (∂, h) and ( ∂ , ĥ). A symmetry

relation of the form (10) implies that L (2) (∂, h) := 21 L (∂, h; ∂, h) is a Lagrangian
density and
I (2) = d 4 xL (2) (11)
is an action for the perturbed system.

The conserved canonical energy associated with the timelike Killing vector is the
Hamiltonian of the perturbation, expressed in terms of configuration space variables,
Ec = d 3 x (ζ ε Lt ∂ε + π εγ Lt h εγ − L (2) ), (12)
S
where ζ ε and π εγ are the momenta conjugate to ∂ ε and h εγ . On a spacelike hyper-

surface with future pointing unit normal n ε = −ε⊗ ε t (where ε is the lapse), the
canonical momenta conjugate to ∂ ε and h εγ are given by
ζ ε = ζ α ε ⊗α t, π εγ = π α εγ ⊗α t, (13)
with
1 τL (∂, h; ∂, h)
ζ εγ = , (14)
2 τ⊗ε ∂γ
1 τL (∂, h; ∂, h)
π εγα = . (15)
2 τ⊗ε h γα
The corresponding canonical momentum has the form
Jc = − d 3 x (ζ ε Lφ ∂ε + π εγ Lφ h εγ ). (16)
S
If one foliates the background spacetime by a family of spacelike but asymp-

totically null hypersurfaces, the difference E 2 − E 1 in E c from one hypersurface
to another to its future is the energy radiated in gravitational waves to future null
infinity. Because this energy is positive definite, E c can only decrease. This suggests
that a condition for stability is that E c be positive for all initial data.
This is, in fact, an appropriate stability criterion, but there is a subtlety, associated
with a gauge freedom in choosing a Lagrangian displacement: There is a class of
trivial displacements, for which the Eulerian changes in all fluid variables vanish.
For a one (two) parameter equation of state, these correspond to rearranging fluid
elements with the same value of λ (and s).2 For a trivial displacement δε , the same
physical perturbation is described by the pairs h εγ , ∂ ε and h εγ , ∂ ε + δε , but, for
nonaxisymmetric perturbations, the canonical energy is not invariant under addition
of a trivial displacement, and its sign depends on this kind of gauge freedom. There
is, however, a preferred class of canonical displacements, the displacements ∂ ε that
are orthogonal to all trivial displacements, with respect to the symplectic product of
two perturbations,
∂,
W ( h; ∂, h) := ε ∂ ε +
(ζ ∂ a − π εγ
π εγ h εγ − ζε h εγ )d 3 x. (17)
Σ
The criterion for stability can then be phrased as follows:

1. If E c < 0 for some canonical data on Σ, then the configuration is unstable or
marginally stable: There exist perturbations on a family of asymptotically null
hypersurfaces Σu that do not die away in time.
2. If E c > 0 for all canonical data on Σ, the magnitude of E c is bounded in time
and only finite energy can be radiated.
The trivial displacements are relabelings of fluid elements with the same baryon
density and entropy per baryon. They are Noether-related to conservation of circula-
tion in surfaces of constant entropy per baryon [7–9], and canonical displacements
2 This is not the gauge freedom associated with infinitesimal diffeos of the metric and matter, but a
redundancy in the Lagrangian- displacement description of perturbations that is already present in
a Newtonian context.
are displacements that preserve the circulation of each fluid ring—for which the
Lagrangian change in the circulation vanishes.
For perturbations that are not spherical, stable perturbations have positive energy
and die away in time; unstable perturbations have negative canonical energy and
radiate negative energy to infinity, implying that E c becomes increasingly negative.
One would like to show that when E c < 0 a perfect-fluid configuration is strictly
unstable, that within the linearized theory the time-evolved data radiates infinite
energy and that |E c | becomes infinite along a family Σu of asymptotically null
hypersurfaces. There is no proof of this conjecture, but it is easy to see that if E c < 0,
the time derivatives ∂˙ ε and ḣ εγ must remain finitely large. Thus a configuration with
E c < 0 will be strictly unstable unless it admits perturbations that are time dependent
but nonradiative.
For spherical stars, radial perturbations have this property, but in that case, the rel-
ativistic Euler equation has the form of a Sturm-Liouville equation, and perturbations
with E c < 0 are in fact strictly unstable.
The symplectic form provides an alternate form of the canonical energy, used
in Wald’s article in this volume. Because of the quadratic structure of the second-
order Lagrangian, when the field equations are satisfied, Eq. (12) is equivalent to the
expression
1
Ec = W (L t ∂ ε , L t h εγ , ∂ ε , h εγ )
2
1
= (ζε L t ∂ ε + π εγ L t h εγ − L t ζε ∂ ε − L t π εγ h εγ )d 3 x. (18)
2
Σ
From this relation and Eq. (16), one has an immediate relation between Jc and E c
for a real-frequency mode with behavior ei(mκ+φt) :
m
Jc = − Ec . (19)
φ
3 Local Stability
3.1 Convective Instability
The criterion for the stability of a spherical star against convection is easy to under-
stand. When a fluid element is displaced upward, if its density decreases more rapidly
than the density of the surrounding fluid, then the element will be buoyed upward
and the star will be unstable. If, on the other hand, the fluid element expands less
than its surroundings it will fall back, and the star will be stable to convection.
As this argument suggests, criteria for convective stability are local, involving
perturbations restricted to an arbitrarily small region of the star or, for axisymmetric
perturbations, to an arbitrarily thin ring. For local perturbations, the Cowling approx-
imation is valid: The change in the gravitational field can be ignored. The argument
is this: A perturbation in density of order Φν/ν that is restricted to a region of volume
V ◦ R 3 can be regarded as adding or subtracting from the source a mass Φm of
order ΦνV . Then
Φm V Φν Φν
≈ 3 ◦ . (20)
M R ν ν
The change in the metric is then also smaller than Φν/ν by a factor V /R 3 , arbitrarily
small when the support of the matter perturbation is arbitrarily small. Note that,
because the metric perturbation is gauge-dependent, this statement about the small-
ness of the perturbed metric is also gauge-dependent. A more precise way of stating
this property of a local perturbation is that a gauge can be chosen in which the metric
perturbation is smaller than the density perturbation by a factor of order V /R 3 .
Convective instability of spherical relativistic stars was discussed by Thorne [10]
and subsequently, with greater rigor, by Kovetz [11] and Schutz [12]. An initial
heuristic treatment by Bardeen [13] of convective instability of differentially rotating
stars was made more precise and extended to models with heat flow and viscosity by
Seguin [14].
Consider a fluid element displaced radially outward from an initial position with
radial coordinate r to r + ∂ . The fluid element expands (or, if displaced inward,
contracts), with its pressure adjusting immediately—in sound travel time across the
fluid element—to the pressure outside:
dp
ρp = ξ · ⊗ p = ∂. (21)
dr
Heat diffuses more slowly, and the analysis assumes that the motion is faster than the
time for heat to flow into or out of the fluid element: The perturbation is adiabatic:
τν
ρν = ρp
τp s
τν dp ν + p dp
= ∂ = Θ1 ∂, (22)
τp s dr p dr
where we have used the adiabatic conditions (6) and (7).

The difference ρψ ν in the density of the surrounding star between r and r + ∂ is
given by
dν
ρψ ν = ∂ . (23)
dr
The displaced fluid element falls back if |ρν| < |ρψ ν|—if, that is, the fluid element’s
density decreases more slowly than the star’s density:

τν dp dν
∂ < ∂ . (24)
τp dr dr
s
The star is then stable against convection if the inequality,
dp dp/dr τp
:= < (25)
dν ψ dν/dr τν s
is satisfied, unstable if the inequality is in the opposite direction.

In particular, in a homentropic star with no composition gradient, the adiabatic
value of dp/dν coincides with its value in the equilibrium star,
τν dp
= , (26)
τp s dν ψ
implying that the star is marginally unstable.

For spherical stars Detweiler and Ipser [15] (generalizing a Newtonian result due
to Lebovitz [16]), argue that, apart from local instability to convection, one need
only consider radial perturbations: If a nonrotating star is stable to radial oscilla-
tions and stable against convection, the star is stable. The Detweiler-Ipser argument,
however, relies on completeness of normal modes and the assumption that all modes
are continuously joined to modes of a nearly Newtonian star, for which the Lebovitz
result should imply that all modes are stable. Although the result is almost certainly
true, the assumptions are not: There are outgoing modes—the w-modes—analogous
to the outgoing modes of black holes, that have no Newtonian counterparts.
3.1.1 Research Problem
Prove that perturbations of spherical stars are stable if they are stable against con-
vection and against radial perturbations.
This can be done by showing that, with reasonable assumptions about the EOS,
the canonical energy of a nonradial perturbation is negative only if the Schwarzschild
criterion is violated. The result may follow from an integral inequality (associated
with Eq. (42) of [15]), that is central to the Detweiler-Ipser argument.
Within minutes after their birth, neutron stars cool to a temperature below the
Fermi energy per nucleon, below 1012 K. Their neutrons are then degenerate, with a
nearly isentropic equation of state: Convectively stable, but with convective modes
having frequencies below 100 Hz, much lower than the kHz frequencies of the
f - and p-modes.
3.2 Convective Instability due to Differential Rotation:

The Solberg Criterion
Differentially rotating stars have one additional kind of convective (local) instability
[17]. If the angular momentum per unit rest mass, j = hu ε κ ε , decreases outward
from the axis of symmetry, the star is unstable to perturbations that change the
differential rotation law.
The criterion is easy to understand in a Newtonian context. Consider a ring of fluid
in the star’s equatorial plane that is displaced outward from r to r + ∂ , conserving
angular momentum and mass. Again the displaced ring immediately adjusts its
pressure to that of the surrounding star. If the ring’s centripetal acceleration is larger
that the net restoring force from gravity and the surrounding pressure gradient, it
will continue to move outward. Now in the unperturbed star, the centripetal accel-
eration is equal to the restoring force. As in the discussion of convective instability,
the displaced fluid element encounters the pressure gradient and gravitational field
of the unperturbed star at its new position, and the restoring force is the restoring
force on a fluid element at r + ∂ in the unperturbed star. Thus, if the displaced fluid
ring has the same value of v2 /r as the surrounding fluid it will be in equilibrium,
and the star will be marginally stable. If a displaced fluid ring has larger v2 /r than
its surrounding fluid the star will be unstable.
d
The difference in acceleration for the background star is ρψ (v2 /r ) = ∂ r (v2 /r ),
dr
and stability then requires
d v2 v2
∂r −ρ > 0, (27)
dr r r
for ∂ r > 0.
Because ρj = 0 and v( j, r ) = j (r )/r , we have
v2 j2 d 1
ρ = ρ 3 = j 2∂ r , (28)
r r dr r 3
while
v2 d j2
ρψ = ∂r , (29)
r dr r 3
implying
v2 v2 1 d j2
ρψ − ρ = ∂r 3 ; (30)
r r r dr
dj
and the star is stable only if> 0 in the equatorial plane (for j > 0), or, equiva-
dr
lently, only if τ ( π) > 0.
2
Bardeen [13] gives a heuristic argument for a restricted version of this criterion,
and a subsequent comprehensive and more precise treatment, including heat flow
and viscosity, is due to Seguin [14]. Abramowicz [18] gives the relativistic version
of the Newtonian argument summarized above; a presentation in [19] corrects some
misprints and also relates the criterion to the sign of the canonical energy.
Here, for the relativistic case, we present Bardeen’s simple argument. The rel-
ν+ p
ativistic angular momentum per unit rest mass is j = u ε κ ε , with κ ε the
λ
rotational Killing vector. The first law of thermodynamics for relativistic stars has
the form
T g
ΦM = t
ρd S + t ρd M0 + πρd J , (31)
u u
Σ
where d M0 = λd V, d J = jd M0 and g is the specific Gibbs free energy. If, in a

homentropic, differentially rotating star, j has an extremum as a function of radius
in the the equatorial plane, then there are perturbations that conserve baryon number
and that lower the energy of the system—for which Φ M < 0. The argument, for a
homentropic star, is this: On opposite sides of the extremum, there are two rings, 1 and
2, with the same value of j and with π2 > π1 . A perturbation that transfers matter
with baryon mass Φ M0 from ring 2 to ring 1 then gives Φ M = (π1 − π2 ) j Φ M0 < 0.
That is, unless j is a monotonic function, one can always find a perturbation with
negative energy.
This is a simplest example of the turning-point criterion governing axisymmetric
stability: A point of marginal stability along a sequence of circular orbits of a particle
is a point at which j is an extremum. The turning-point condition can be rephrased
in terms of the particle’s energy. For a particle of fixed rest mass, the difference in
energy of adjacent orbits is related to the difference in its angular momentum by
Φ E = πΦ J.
Then a point of marginal stability along a sequence of circular orbits of a particle of

fixed baryon mass is a point at which its energy is an extremum.
4 Axisymmetric Instability and Turning Points
For spherical stars in Newtonian gravity, instability sets in when the matter becomes
relativistic, when the adiabatic index Θ1 (more precisely, its pressure-weighted aver-
age Θ¯1 ) reaches the value 4/3 characteristic of zero rest-mass particles. This can be
seen from the Newtonian limit of the canonical energy,
1 ˙ 2 2 √ 2 Θ1 p 2 √ 2
Ec = λ ∂ + p ∂ + 4 (r ∂ ) d V. (32)
2 r 2r
Choosing as initial data ∂ = r , ∂˙ = 0, gives
9 9 4
Ec = 2r p √ + Θ1 pr 2 d V = Θ1 − p d V, (33)
2 2 3
implying instability for Θ¯1 < 4/3. This shows only that Θ1 < 4/3 is a sufficient
condition for instability, but spherical Newtonian polytropes with Θ1 > 4/3 are
stable.
By, in effect, deriving the relativistic canonical energy,
R
1 ω+ν 4 √ p √2
Ec = e p − + 8π p(ν + p) ∂ 2
2 r ν+ p
0
e3ω−ν 2
−ν 2 √
+ Θ 1 p (e r ∂ ) r 2 dr, (34)
r4
Chandrasekhar [20, 21] showed that the stronger gravity of general relativity implies
an earlier onset of instability: Even models with the stiffest equation of state must
be unstable to collapse for some value of compactness M/R < 9/8, the value for
the most compact uniform density model. The more stringent relativistic constraint
on Θ1 for a star to be stable against radial perturbations has the form
4 M
Θ1 < +K , (35)
3 R
where K is positive and of order unity [20]. Because a gas of photons has Θ1 = 4/3
and massive stars are radiation dominated, the instability can be important for stars
with M/R ◦ 1.
4.1 Turning Point Instability
The best-known instability result in general relativity is the statement that instability
to collapse sets in at a point of maximum mass, along a sequence of spherical
barotropic models. The configuration with maximum mass is called a turning point
along the sequence, and it is also the configuration with maximum baryon mass.
A similar result holds for uniformly rotating stars [22]: Instability to collapse is
implied by a point of maximum mass and maximum baryon mass, along a sequence
of uniformly rotating barotropic models with fixed angular momentum. As in the
spherical case, stars with higher central density than that of the maximum-mass con-
figuration are unstable. For rotating stars, however, the turning point is a sufficient
but not a necessary condition for instability: The onset of instability is at a config-
uration with slightly lower central density (for fixed angular momentum) than that
of the maximum-mass star. A formal symmetry in the way baryon mass and angular
momentum occur in the first law implies that the line of turning points is also the
line of extrema of angular momentum along sequences of fixed baryon mass.
For dynamical oscillations of neutron stars the adiabatic index does not coincide
d log p(r )/dr
with the polytropic index, Θ1 ∗= Θ := , and the turning point criterion
d log λ(r )/dr
implies secular instability—an instability whose growth time is long compared to the
typical dynamical time of stellar oscillations. For spherical stars, the turning-point
instability proceeds on a time scale slow enough to accommodate the nuclear
reactions and energy transfer that accompany the change to a nearby equilibrium.
For rotating stars, the time scale must also be long enough to accommodate a transfer
of angular momentum from one fluid ring to another. That is, the growth rate of the
instability is limited by the time required for viscosity to redistribute the star’s angu-
lar momentum. For neutron stars, this is expected to be short, probably comparable
to the spin-up time following a glitch, and certainly short compared to the lifetime
of a pulsar or an accreting neutron star. For this reason, it is the secular instability,
that sets the upper and lower limits on the mass of spherical and uniformly rotating
neutron stars.
One can easily understand why the instability sets in at an extremum of the mass
by looking at a radial mode of oscillation of a nonrotating star with an equation
of state p = p(λ), ν = ν(λ). Along the sequence of spherical equilibria, a radial
mode changes from stability to instability when its frequency Λ changes from real
to imaginary, with Λ = 0 at the point of marginal stability. Now a zero-frequency
mode is just a time-independent solution to the linearized Einstein-Euler equations -
a perturbation from one equilibrium configuration to a nearby equilibrium with the
same baryon number. From the first law of thermodynamics (31), a perturbation that
keeps the star in equilibrium satisfies
g
ΦM = Φ M0 . (36)
ut
The relation implies that, for a zero frequency perturbation involving no change in
baryon number, the change Φ M in mass must vanish. This is the requirement that the
mass is an extremum along the sequence of equilibria. Models on the high-density
side of the maximum-mass instability point are unstable: Because the turning point
is a star with maximum baryon number as well as maximum mass, there are models
on opposite sides of the turning point with the same baryon number. Because g/u t
is a decreasing function of central density, the model on the high-density side of the
turning point has greater mass than the corresponding model with smaller central
density.
At the minimum mass, it is the low-density side that is unstable: Because the mass
is a minimum, the model on the low-density side of the turning point has greater mass
than the corresponding model with the same baryon number on the high-density side.
The precise statement of the turning-point criterion is the following result:
Theorem 1 (Friedman et al. [22]): Consider a continuous sequence of uniformly

rotating stellar models based on an equation of state of the form p = p(ν). Let ω be
the sequence parameter and denote the derivative d/dω along the sequence by (˙).
(i) Suppose that the total angular momentum is constant along the sequence and
that there is a point ω0 where Ṁ = 0 and where E > 0, (E˙ Ṁ) ∗ = 0. Then the
part of the sequence for which E˙ Ṁ > 0 is unstable for ω near ω0 .
(ii) Suppose that the total baryon mass M0 is constant along the sequence and that
there is a point ω0 where Ṁ = 0 and where π > 0, (π̇ Ṁ) ∗ = 0. Then the part
of the sequence for which π̇ Ṁ > 0 is unstable for ω near ω0 .
In form (ii) of the theorem, the first law implies that the turning point is an
extremum of angular momentum J along a sequence of constant rest mass. Friedman
et al. [22] point out the symmetry between M0 and J that implies this maximum-J
form of the theorem, and Cook et al. [23] first use the theorem in this form. For
rotating stars, the turning point criterion is a sufficient condition for secular instabil-
ity to collapse. In general, however, collapse can be expected to involve differential
rotation, and the turning point identifies only nearby uniformly rotating configura-
tions with lower energy. Rotating stars are therefore likely to be secularly unstable to
collapse at densities slightly lower than the turning point density. The onset of sec-
ular instability to collapse is at or before the onset of dynamical instability along a
sequence of uniformly rotating stars of fixed angular momentum, and recent work by
Takami et al. [24] appears to show that rapidly rotating stars can also be dynamically
unstable to collapse just prior to the turning point.
As illustrated in Fig. 2, they find a dynamical instability line that coincides with
the turning-point line for spherical stars and that, for rapid uniform rotation, has
a central density about 5 % below that of the turning point. This result somewhat
overstates the difference between the two lines, because it ignores the difference
Θ − Θ1 between the indices governing dynamical oscillations and the equilibrium
equation of state. The actual dynamical instability line begins at a spherical star with
higher density than the marginally unstable turning point star and probably crosses
the turning-point line to lower density at some angular velocity less than π K .
The greater significance of the Takami et al. result, however, is that stars along
the line determined by using the equilibrium equation of state are guaranteed to be
secularly unstable, because the diagnosed instability guarantees that the configu-
rations have lower energy than equilibria with the same baryon mass and angular
momentum. This means that the line of secular instability runs through rapidly rotat-
ing configurations with central densities more than 5 % below those along the line
of turning points.
5 Nonaxisymmetric Instabilities
Rapidly rotating stars and drops of water are unstable to a bar mode that leads to
fission in the water droplets; and a similar nonaxisymmetric instability is likely
Fig. 2 Stability lines in a (λc , M) diagram. The two solid black lines mark sequences with either
zero (lower line) or mass-shedding angular momentum (upper line), with the filled symbols marking
the corresponding maximum masses. The solid grey line is the neutral-stability line, “thickened”
by the error bar (dot-dashed lines). The grey dashed line is instead the turning-point criterion for
secular stability. Marked with empty or filled circles are representative models with constant angular
velocity O1, O2, O3, or constant initial central rest-mass density R1, R2, R3. (Figure from Takami
et al. [24]. Reproduced by permission of John Wiley and Sons.)
to be the reason most stars in the universe are in close binary systems. Galac-
tic disks are unstable to nonaxisymmetric perturbations that lead to bars and to
spiral structure. And a related instability of a variety of nonaxisymmetric modes,
driven by gravitational waves, the Chandrasekhar-Friedman-Schutz (CFS) instability
[8, 25]), may limit the rotation of neutron stars. The existence of the CFS-instability
in rotating stars was first found by Chandrasekhar [25] in the case of the l = 2 mode
in uniformly rotating, uniform density Maclaurin spheroids. Subsequently, Friedman
and Schutz [8, 26] showed that this instability also appears in compressible stars and
that all rotating, self-gravitating perfect fluid configurations are generically unsta-
ble to the emission of gravitational waves. We have seen that, along a sequence of
stellar models, a mode changes from stable to unstable when its frequency vanishes.
The generic-instability result means that zero-frequency nonaxisymmetric modes of
rotating perfect-fluid stellar models are marginally stable.
Whereas axisymmetric instability to collapse sets in at points that are nearly
independent of the magnitude of viscosity or the strength of gravitational waves,
the opposite is true for the nonaxisymmetric case. Gravitational radiation drives
a nonaxisymmetric instability that, if no other dissipation is present, makes every
rotating star unstable. Viscosity can drive a nonaxisymmetric instability in rapidly
rotating stars for which gravitational radiation is negligible. For slowly rotating stars,
however (and nearly all neutron stars rotate slowly compared to the Kepler limit),
viscosity simply damps out the gravitational-wave driven instability. That is, for
slow rotation, we will see that the timescale of the CFS instability is longer than the
timescale for viscous damping. On the other hand, for rapidly rotating neutron stars,
the instability’s timescale may be short enough that it limits the rotation of young
neutron stars and of old neutron stars spun up by accretion.
This review begins with a discussion of the CFS instability for perfect-fluid models
and then outlines the work that has been done to decide whether the instability is
present in young neutron stars and in old neutron stars spun up by accretion. For
very rapid rotation and for slower but highly differential rotation, nonaxisymmetric
modes can be dynamically unstable, with growth times comparable to the period
of a star’s fundamental modes, and the review ends with a brief discussion of these
related dynamical instabilities.
To understand the way the CFS instability arises, consider first a stable spherical
star. All its modes have positive energy, and the sign of a mode’s angular momentum
Jc about an axis depends on whether the mode moves clockwise or counterclockwise
around the star. That is, a mode with angular and time dependence of the form
cos(mκ + φ0 t)e−ε0 t , has positive angular momentum Jc about the z-axis if and
φ0
only if the mode moves in a positive direction: The pattern speed, − , is positive.
m
Because the wave moves in a positive direction relative to an observer at infinity,
the star radiates positive angular momentum to infinity, and the mode is damped.
Similarly, a mode with negative angular momentum has negative pattern speed,
φ0
− < 0, and radiates negative angular momentum to infinity; and the mode is
m
again damped.
Now consider a slowly rotating star with a backward-moving mode, a mode that
moves in a direction opposite to the star’s rotation. Because a short-wavelength fluid
mode (a mode with a Newtonian counterpart, not a w-mode) is essentially a wave in
the fluid, the wave moves with nearly the same speed relative to a rotating observer
that it had in the spherical star. That means that an observer at infinity sees the mode
dragged forward by the fluid. The real part φr of the frequency seen in a rotating
frame is the frequency associated with the κ coordinate κr = κ − πt of a rotating
observer. Then
mκ + φt = mκr + (φ + mπ)t = mκr + φr t,
implying that the frequency seen by the rotating observer is
φr = φ + mπ. (37)
For a slowly rotating star, φr ⇔ φ0 . When the star rotates with an angular velocity
greater than |φr /m|, the backward-going mode is dragged forward relative to an
observer at infinity, and φr and φ have opposite signs:
φr φ < 0. (38)
Because the pattern speed is now positive, the mode radiates positive angu-
lar momentum to infinity. But the canonical angular momentum is still negative,
because the mode is moving backward relative to the fluid: The angular momentum
of the perturbed star is smaller than the angular momentum of the star without the
backward-going mode. As the star radiates positive angular momentum to infinity,
Jc becomes increasingly negative, implying that the amplitude of the mode grows in
time: Gravitational radiation now drives the mode instead of damping it.
For large m or small φ0 , the pattern speed will be positive when π ⇔ |φ0 /m|. This
relation suggests two classes of modes that are unstable for arbitrarily slow rotation:
Backward-moving modes with large values of m and modes with any m whose
frequency is zero in a spherical star. Both classes of perturbations exist. The usual
p-modes and g-modes have finite frequencies for a spherical star and are unstable for
π |φ0 |/m; and r -modes, which have zero frequency for a non-rotating barotropic
star, are unstable for all values of m and π (that is, those r-modes are unstable that
are backward-moving in the rotating frame of a slowly rotating star).
We have so far not mentioned the canonical energy, but our key criterion for the
onset of instability is a negative E c . If we ignore the imaginary part of the frequency,
the change in the sign of E c follows immediately from the relation (19), Jc = − m φ Ec .
To take the imaginary part ImΛ = ε ∗= 0 of the frequency into account, we need to
... 2
use the fact that energy is lost at a rate Ė c ≤ Q ≤ Λ 6 for quadrupole radiation, with
Ė c proportional to higher powers of Λ for radiation into higher multipoles. Because
E c is quadratic in the perturbation, it is proportional to e−2εt , implying ε ≤ Λ 6 .
Thus ε/Λ ≡ 0 as Λ ≡ 0, implying that for a normal mode E c changes sign when
φ changes sign.
Although the argument we have given so far is heuristic, there is a precise form
of the statement that a stable, backward-moving mode becomes unstable when it is
dragged forward relative to an inertial observer (see Friedman and Schultz [26] and
Friedman and Stergioulas [19]).
Theorem 2 Consider an outgoing mode (h εγ (ω), ∂ ε (ω)), that varies smoothly along
a family of uniformly rotating perfect-fluid equilibria, labeled by ω. Assume that it has
t and κ dependence of the form ei(mκ+Λ t) , that φ = Re{Λ } satisfies φ/m + π > 0
for all ω, and that the sign of φ/m is positive for ω < ω0 and negative for ω > ω0 .
Then in a neighborhood of ω0 , ε := Im{Λ } ∇ 0; and if the mode has at least one
nonzero asymptotic multipole moment with l ⇒ 2 at future null infinity, the mode is
unstable (ε < 0) for ω > ω0 .
A corresponding result that does not rely on existence or completeness of normal

modes is the statement that one can always choose canonical initial data to make
E c < 0 [8, 19].
The growth
time τG R of the instability of a perfect fluid star is governed by the
d E
rate at which energy is radiated in gravitational waves:
dt GR

1 1 d E
=− , (39)
τGR 2E c dt GR
where (Thorne [27])
dE
=− φi2l+2 Nl |Φ Dlm |2 + |Φ Jlm |2 . (40)
dt
l⇒m
Here Dlm and Jlm are the asymptotically defined mass and current multipole moments
4π(l + 1)(l + 2)
of the perturbation and Nl = is, for low l, a constant of order
l(l − 1)[(2l + 1)!!]2
unity. In the Newtonian limit,
Φ Dlm = Φλ r l Ylm d 3 x. (41)
For a star to be unstable, the growth time τG R must be shorter than the viscous
damping time τviscosity of the mode, and the implications of this are discussed below.
In particular because the growth time is longer for larger l, only low multipoles can
be unstable in neutron stars.
Modes with Polar and Axial Parity The spherical symmetry of a nonrotating star
and its spacetime implies that perturbations can be labeled by fixed values l, m label-
ing an angular harmonic: The quantities h εγ , ∂ ε , Φλ, Φν, Φp, Φs that describe a pertur-
bation are all proportional to scalar, vector and tensor spherical harmonics constructed
from Ylm , and perturbations with different l, m values decouple. Similarly, because
spherical stars are invariant under parity (a map of each point P of spacetime to the
diametrically opposite
point on the symmetry sphere through P), perturbations with different parity decou-
ple, the parity of a perturbation is conserved, and normal modes have definite
parity. Perturbations associated with an l, m angular harmonic are said to have polar
parity if they have the same parity as the function Ylm , (−1)l . Perturbations having
parity (−1)l+1 , opposite to that of Ylm have axial parity. In the Newtonian literature,
modes of a rotating star that are continuously related to polar modes of a spherical
star are commonly called spheroidal; while modes whose spherical limit is axial are
called toroidal.
Every rotational scalar—ν, p, λ, and the components in the t-r subspace of the
perturbed metric h εγ and the perturbed fluid velocity Φu ε —can be expressed as a
superposition of scalar spherical harmonics Ym . As a result, modes of spherical
stars that involve changes in any scalar are polar. On the other hand, the angular
components of velocity perturbations can have either polar parity, with
Φv = f (r )⊗Ylm (42)
or axial parity, with Newtonian form
Φv = f (r )r × ⊗Ylm , (43)
and the relativistic form Φu ε ≤ νεγα Φ ⊗γ t⊗α r ⊗Φ Ylm .

There are two families of polar modes of perfect-fluid Newtonian stars, p-modes
(pressure modes) and g-modes (gravity modes). For short wavelengths, the p-modes
are sound waves, with pressure providing the restoring force and frequencies
Λ = cs k, (44)
where k is the wavenumber and cs is the speed of sound. The short-wavelength

g-modes are modes whose restoring force is buoyancy, and their frequencies are
proportional to the Brunt-Väisälä frequency, related to the difference between dp/dν
in the star and cs2 = τ p(ν, s)/τν. The fundamental modes of oscillation of a star
( f -modes), with no radial nodes, can be regarded as a bridge between g-modes and
p-modes.
Because axial perturbations of a spherical star involve no change in density or
pressure, there is no restoring force in the linearized Euler equation, and the linear
perturbation is a time-independent velocity field—a zero-frequency mode.3 In a
rotating star, the axial modes acquire a nonzero frequency proportional to the star’s
angular velocity π, a frequency whose Newtonian limit has the simple form
(l − 1)(l + 2)
Λ =− mπ, (45)
l(l + 1)
where the harmonic time and angular dependence of the mode is ei(mκ−Λ t) . These
modes are called r -modes, their name derived from the Rossby waves of oceans and
planetary atmospheres. The term r -mode can be usefully regarded as a mnemonic for
a rotationally restored mode. Eq. (37) implies that the r -mode associated with every
nonaxisymmetric multipole obeys the instability condition for every value of π: It
is forward moving in an inertial frame and backwards moving relative to a rotating
observer:
2m
Λr = π, (46)
l(l + 1)
with sign opposite to that of Λ . Because the rate at which energy is radiated is
greatest for the r -mode with l = m = 2, that is the mode whose instability grows
most quickly and which determines whether an axial-parity instability can outpace
viscous damping.
The instability of low-multipole r -modes for arbitrarily slow rotation is strikingly
different from the behavior of the low-multipole f - and p-modes, which are unstable
only for large values of π. The reason is that the frequencies of f - and p-modes are
high, and, from Eq. (38), a correspondingly high angular velocity is needed before
a mode that moves backward relative to the star is dragged forward relative to an
inertial observer at infinity. Of the polar modes, f -modes with l = m have the fastest
3 Axial perturbations of the spacetime of a spherical star include both axial perturbations of the
fluid and gravitational waves with axial parity. The axial-parity waves do not couple to the fluid
perturbation, which is stationary in the sense that τt Φu ε = 0.
N=1.0
1.0
0.95 m=2
m=3
0.9
0.85
m=4
0.8
K
0.75 m=5
/c
0.7
0.65
0.6
0.55
0.5 -2 -1
10 2 5 10 2
_
c
Fig. 3 Critical angular velocity π/π K vs. the dimensionless central energy density ν̄c for the
m = 2, 3, 4 and 5 neutral modes of N = 1.0 polytropes. The filled circles on the vertical axis are
the Newtonian values of the neutral points for each mode. (Reproduced from [28].)
growth rates; their instability points for uniformly rotating relativistic stars, found by
Stergioulas and Friedman [28], are shown in Fig. 3. (Work on these stability points
of relativistic stars is also reported in [29–32].)
The figure shows that, for uniform rotation, the l = m = 2 f -mode is unstable
only for stars with relatively high central density or high mass. For tabulated EOSs,
this practically applies to all neutron stars with masses greater than 1.3 M∈ and
T /|W | > 0.06 [33]. Because neutron stars rotate differentially at birth, the l = 2
mode, as well as higher modes, could be initially unstable for a larger range of
parameters.
Implications of the Instability The nonaxisymmetric instability may limit the rota-
tion of nascent neutron stars and of old neutron stars spun up by accretion; and the
gravitational waves emitted by unstable modes may be observable by gravitational
wave detectors. Whether a limit on spin is in fact enforced depends on whether
the instability of perfect-fluid models implies an instability of neutron stars; and
the observability of gravitational waves also requires a minimum amplitude and
persistence of an unstable mode. We briefly review observational support for an
instability-enforced upper limit on spin and then turn to the open theoretical issues.
Evidence for an upper limit on neutron-star spin smaller than the Keplerian fre-
quency π K comes from nearly 30 years of observations of neutron stars with mil-
lisecond periods, seen as pulsars and as X-ray binaries. The observations reveal
rotational frequencies ranging upward to 716 Hz and densely populating a range of
frequencies below that. Selection biases against detection of the fastest millisecond
radio pulsars have made conclusions about an upper limit on spin uncertain, but
Chakrabarty argues that the class of sources whose pulses are seen in nuclear bursts
Fig. 4 Highest observed neutron-star spin frequencies
(nuclear powered accreting millisecond X-ray pulsars) constitute a sample without

significant bias [34]. Their distribution of spins, together with the spins of other
millisecond pulsars, is shown in Fig. 4.
A magnetic field of order 108 G can limit the spin of an accreting millisecond
pulsar. Because matter within the magnetosphere corotates with the star, only matter
that accretes from outside the magnetosphere can spin up the star, leading to an
equilibrium period given approximately by Ghosh and Lamb [35]4
6/7 −3/7
B Ṁ
Peq ≈ 2 × 10−3 s . (47)
108 G 10 M∈ yr −1
−10
Because this period depends on the magnetic field, a sharp cutoff in the frequency
of accreting stars is not an obvious prediction of magnetically limited spins. For a
magnetically set maximum rotation rate of order 700–800 Hz the range of magnetic
fields would need to have a corresponding minimum cutoff value of about 108 G;
and the highest observed spin rates should correspond to the lowest magnetic fields.
The required cutoff and a fairly narrow range of observed frequencies has made
gravitational-wave limited spin a competitive possibility for accreting neutron stars.
Arguments for and against this based on available observations are given by White
and Zhang [37] and by Patruno et al. [38], respectively.
4 Shapiro and Teukolsky [36] give a clear, simplified version, and Eq. (47) is their Eq. (15.2.22),
with M = 1.4 M∈ , R = 10 km, and a ratio φs of the angular velocity to π K at the inner edge of
the disk set to 1.
Under what circumstances the CFS instability could limit the spin of recycled
pulsars has now been studied in a large number of papers. References to this work
can be found in the treatment in FS, on which the present review is based and in
comprehensive earlier discussions by Stergioulas [39], by Andersson and Kokkotas
[40], and by Kokkotas and Ruoff [41], while briefer reviews of more recent work
are given in [42, 43]. References in the present review are generally limited to initial
work and to a late paper that contains intervening references.
Whether the instability survives the complex physics of a real neutron star has
been the focus of most recent work, but it remains an open question. Studies have
focused on:
• Dissipation from bulk and shear viscosity and mutual friction in a superfluid
interior;
• magnetic field wind-up;
• nonlinear evolution and the saturation amplitude; and
• the possibility that a continuous spectrum replaces r -modes in relativistic stars.
We discuss these in turn and then summarize their implications for nascent, rapidly
rotating stars and for old stars spun up by accretions.
Viscosity When viscosity is included, the growth-time or damping time τ of an
oscillation has the form
1 1 1 1
= + + , (48)
τ τG R τb τs
with τb and τs the damping times due to bulk and shear viscosity. Bulk viscosity is
large at high temperatures, shear viscosity at low temperatures. This leaves a window
of opportunity in which a star with large enough angular velocity can be unstable.
The window for the l = m = 2 r -mode is shown in Fig. 5, for a representative
computation of viscosity. The highest solid curves on left and right mark the critical
angular velocity πc above which the l = m = 2 r -mode is unstable. The curves on the
left show the effect of shear viscosity at low temperature, allowing instability when
π < π K only for T > 106 K; the curve on the right shows the corresponding effect
of bulk viscosity, cutting off the instability at temperatures above about 4 × 1010 K.
There is substantial uncertainty in the positions of both of these curves.
Bulk viscosity arises from nuclear reactions driven by the changing density of
an oscillating fluid element, with neutrons decaying, n ≡ p + e + ν̄e , as the fluid
element expands and protons capturing electrons, p + e ≡ n + νe , as it contracts.
The neutrinos leave the star, draining energy from the mode. The rates of these URCA
reactions increase rapidly with temperature and are fast enough to be important above
about 109 K, with an expected damping time τb given by
1 1
= ζ (Φθ )2 d 3 x, (49)
τb 2E c
where θ = ⊗ε u ε is the divergence of the fluid velocity and the coefficient of bulk
viscosity ζ is given by [45]
1.0
0.9
0.8
0.7
K 0.6
0.5
c/
0.4 n-n e-e

0.3
0.2
0.1
0.0
5 6 7 8 9 10 11
10 10 10 10 10 10 10
T(K)
Fig. 5 Critical angular velocity for the onset of the r -mode instability as a function of temperature
(for a 1.5 M∈ neutron star model). The solid line corresponds to the O(π 2 ) result using electron-
electron shear viscosity, and modified URCA bulk viscosity. The dashed line corresponds to the
case of neutron-neutron shear viscosity. Dotted lines are O(π) approximations. (Reproduced from
[44].)
φ −2
g cm−1 s−1 ,
r
ζ = 6 × 1025 λ15
2 6
T9 (50)
1Hz
where T9 = T /(109 K). With these values, bulk viscosity suppresses the instability
in all modes above a few times 1010 K (see also Ipser and Lindblom [46, 47] and
Yoshida and Eriguchi [48]).5
These equations and Fig. 5 assume that only modified URCA reactions can occur,
that the URCA reactions require a collision to conserve four-momentum, and this will
be true when the proton fraction x p is less than about 1/9. Should the equation of state
be unexpectedly soft (and if the mass is large enough), direct URCA reactions would
be allowed, suppressing the instability for uniformly rotating stars at roughly 109 K
[52]. A soft equation of state would also more likely lead to stars with hyperons
in their core with an additional set of nuclear reactions that dissipate energy and
increase the bulk viscosity [53–57] or quarks [58–62]. However, the observation of
a 2 M∈ pulsar makes the existence of hyperons or quarks in the core of 1.4 M∈
neutron stars less probable.
In contrast to bulk viscosity, shear viscosity increases as the temperature drops. In
α
terms of the shear tensor Λεγ = (Φε +u ε u α )(ΦγΦ +u γ u Φ )(⊗α u Φ +⊗Φ u α − 23 gα Φ ⊗ν u ν ),
the damping time is given by
5 At temperatures above roughly 1010 K, another complication appears: neutrino absorption

increases with increasing temperature [49, 50], and the modified URCA bulk viscosity no longer
rises, but is reduced by an order of magnitude between 1010 K and 1011 K, allowing the instability
to operate in very hot proto-neutron stars [51].
1 1
= δΦΛ εγ ΦΛεγ d 3 x, (51)
τs Ec
where δ is the coefficient of shear viscosity. For nascent neutron stars hotter than
the superfluid transition temperature (about 109 K), a first estimate of the neutron-
neutron shear viscosity coefficient is [63]
δn = 2 × 1018 λ15 T9−2 g cm−1 s−1 ,

9/4
(52)
where λ15 = λ/(1015 g cm−3 ). Below the superfluid transition temperature, electron-
electron scattering determines the shear viscosity in the superfluid core, giving [64]
x 2
−5/3
g cm−1 s−1 .
p
δe = 2.5 × 1018 λ15 T9 (53)
0.1
Shear viscosity may be greatly enhanced after formation of the crust in a boundary
layer (Ekman layer) between crust and core [65–69]. The enhancement depends
on the extent to which the core participates in the oscillation, parametrized by the
slippage at the boundary. The uncertainty in this slippage appears to be the greatest
current uncertainty in dissipation of the mode by shear viscosity, and it significantly
affects the critical angular velocity of the r -mode instability in accreting neutron stars.
For f -modes, the part of the instability window in Fig. 5 to the left of 109 K is
thought to be removed by another dissipative mechanism that comes into play below
the superfluid transition temperature. Called mutual friction, it arises from the scat-
tering of electrons off magnetized neutron vortices. Work by Lindblom and Mendell
[70] shows that mutual friction in the superfluid core completely suppresses f - and
p-mode instabilities below the transition temperature. For the r -mode instability,
subsequent work by the same authors [71] finds that the mutual friction is much
smaller, with a damping time of order 104 s, too long to be important.
In a recent paper, Gaertig et al. [32] point out the possibility of an interaction
between vortices and quantized flux tubes that would result in a much smaller value
for the mutual friction. They argue that the resulting uncertainty is great enough that
shear viscosity could be the dominant dissipative mechanism for f -modes as well
as r -modes.
Magnetic Field Windup At second order in the perturbation, the nonlinear evolu-
tion of an unstable mode includes an axisymmetric part that describes a growing
differential rotation. Because differential rotation will wind up magnetic field lines,
the mode’s energy could be transferred to the star’s magnetic field [72–76]. Again
there is large uncertainty about the strength of a toroidal magnetic field that will be
generated by the differential rotation, what magnetic instabilities will arise, and what
the effective dissipation will be. Apart from the studies cited here (all of which deal
with r -modes) nearly all the remaining work on the evolution of unstable modes
ignores magnetic fields.
Relativistic r -Modes and a Possible Continuous Spectrum Relativistic r -modes
have been computed by a number of authors [41, 77–88]. Where the Newtonian
approximation has purely axial l = m r -modes for barotropic stars at lowest order in
π, in the full theory all rotationally restored modes include a polar part. The change
in the structure of the computed r -modes are small, but that may not be the end of
the story.
For non-barotropic stars Kojima found a single second-order eigenvalue equation
for the frequency, to lowest nonvanishing order in π. The coefficient of the highest
derivative term in that equation vanishes at some value of the radial coordinate r ,
for typical candidate neutron-star equations of state, and that singular behavior gives
a continuous spectrum. Lockitch et al. [83] consider the question of the continuous
spectrum and the existence of r -modes in some detail. They argue that the singularity
in the Kojima equation is an artifact of the slow-rotation approximation and is not
present if one includes terms of order π 2 . Their work is a strong argument for the
existence of r -modes in non-barotropic models.
Showing the existence of the mode, however, does not decide the question of
whether a continuous spectrum is also present or whether the existence of a con-
tinuous or nearly continuous spectrum significantly alters the evolution of an initial
perturbation.
Nonlinear Evolution Linear perturbation theory is valid only for small-amplitude
oscillations; as the amplitude of an unstable mode grows, couplings to other modes
become increasingly important, and the mode ultimately reaches a saturation ampli-
tude or is disrupted, losing coherence. The first nonlinear studies of the r -mode insta-
bility involved fully nonlinear 3+1 evolutions by Stergioulas and Font [89], in which
the r -mode was set at a large initial amplitude or Newtonian evolutions by Lindblom
et al. [90, 91] in which the r -mode was driven to large amplitude by an artificially
large gravitational-radiation reaction term. On a few tens of dynamical timescales,
saturation was seen only at an amplitude of order unity. Subsequently, simulations
on longer timescales showed a coupling to daughter modes [92, 93], suggesting that
the actual saturation amplitude of the r -mode is smaller than the amplitude at which
gravitational radiation reaction was switched off in the short-timescale simulations.
The grid resolution of 3+1 simulations, however, is currently too low to see cou-
plings to short-wavelength modes, and they cannot run for a time long enough to see
the growth from a realistic radiation reaction term. The alternative is to examine the
nonlinear evolution in the context of higher-order perturbation theory. To do this, the
Cornell group (initially with S. Morsink) [94–96] constructed a second-order pertur-
bation theory for rotating Newtonian stars, and then used the formalism to study the
nonlinear evolution of an unstable r -mode. Their series of papers leaves little doubt
that nonlinear couplings sharply limit the amplitude of an unstable r -mode, with a
possible range of 10−1 –10−5 (see Bondarescu et al. [97, 98] and references therein).
The nonlinear development of the f -mode instability has been modeled in three-
dimensional, hydrodynamical simulations (in a Newtonian framework) by Ou et al.
[99] and by Shibata and Karino [100], essentially confirming previous approximate
results obtained by Lai and Shapiro [49]. Kastaun et al. [101] report an initial non-
linear study of f -modes in general relativity. In the framework of a 3+1 simulation
in a Cowling approximation (a fixed background metric of the unperturbed rotating
star), they find limits on the amplitude of less than 0.1, set by wave-breaking and by
coupling to inertial modes. This can be regarded as an upper limit on the amplitude,
with second-order perturbative computations still to be done.
Instability Scenarios in Nascent Neutron Stars and in Old Accreting Stars Both
r -modes and f -modes may be unstable in nascent neutron stars that are rapidly
rotating at birth. Recent work on f -modes in relativistic models [32, 102] finds
growth times substantially shorter than previously computed Newtonian values. In
a particular model (where the l = m = 2 mode becomes unstable only very near
the mass-shedding limit), the l = m = 3 and l = m = 4 f -modes have growth
times of 103 –105 s for π near π K . In a typical scenario, a star with rotation near the
Kepler limit becomes unstable within a minute of formation, when the temperature
has dropped below 1011 K. As the temperature drops further, the instability grows to
saturation amplitude in days or weeks. Loss of angular momentum to gravitational
waves spins down the star until the critical angular velocity is reached below which
the star is stable, at or before the time at which the core becomes a superfluid.
The l = m = 3 mode (or the l = m = 2 mode in models with different masses
or equations of state than the one studied above) could be a source of observable
gravitational waves for supernovae in or near the Galaxy (but with an uncertain
event rate).
The time over which the instability is active depends on the saturation amplitude,
the cooling rate, and the superfluid transition temperature, and all of these have large
uncertainties. The time at which a superfluid transition occurs could be shorter than
a year, but recent analyses of the cooling of a neutron star in Cassiopeia [103, 104]
suggest a superfluid transition time for that star of order 100 years.
The scenario for the l = m = 2 r -mode instability of a nascent star is similar.
The r -mode instability itself was pointed out by Andersson [78], with a mode-
independent proof for relativistic stars given by Friedman and Morsink [105]. First
computations of the growth and evolution were reported by Lindblom et al. [106] and
by Andersson et al. [107], with effects of a crust discussed by Lindblom et al. [66].
Intervening work is referred to by Bondarescu et al. [98]; the simulations reported
by Bondarescu et al. include nonlinear couplings that saturate the amplitude and the
alternative possibilities for viscosity that we have discussed above. The r -mode’s
saturation amplitude is likely to be lower than that of the f -modes, and it is likely
to persist longer because of its low mutual friction.
As mentioned above, the r -mode instability of neutron stars spun up by accretion
has been more intensively studied in connection with the observed spins of LMXBs.
Papaloizou and Pringle [108] suggested the possibility of accretion spinning up a
star until it becomes unstable to the emission of gravitational waves and reaches a
steady state, with the angular momentum gained by accretion equal to the angular
momentum lost to gravitational waves. Following the discovery of the first mil-
lisecond pulsar, Wagoner examined the mechanism in detail for CFS unstable f -
modes [109]. Although mutual friction appears to rule out the steady-state picture for
f -modes, it remains a possibility for r -modes [67, 110–112]. Levin [113] and (inde-
pendently) Spruit [72], however, pointed out that viscous heating of the neutron star
by its unstable oscillations will lower the shear viscosity and so increase the mode’s
growth rate, leading to a runaway instability. The resulting scenario is a cycle in
which a cold, stable neutron star is spun up over a few million years until it becomes
unstable; the star then heats up, the instability grows, and the star spins down until it
is again stable, all within a few months; the star then cools, and the cycle repeats.
This scenario would rule out r -modes in LMXBs as a source of detectable gravi-
tational waves because the stars would radiate for only a small fraction of the cycle.
A small saturation amplitude, however, lengthens the time spent in the cycle, possibly
allowing observability [114]. The steady state itself remains a possible alternative in
stars whose core contains hyperons or free quarks (or if the “neutron stars” are really
strange quark stars) [54, 56, 57, 60, 112, 115]. Heating the core increases the bulk
viscosity, and with an exotic core, this growth in the bulk viscosity is large enough
to prevent the thermal runaway and allow a steady state. In Bondarescu et al. [97]
the nonlinear evolutions (restricted to 3 coupled modes) include neutrino cooling,
shear viscosity, hyperon bulk viscosity and dissipation at the core-crust boundary
layer, with parameters to span a range of uncertainty in these various quantities.
They display the regions of parameter space associated with the alternative scenar-
ios just outlined—steady state, cycle, and fast and slow runaways. In all cases, the
r -mode amplitude remains very small (≈10−5 ), but because of the long duration of
the instability, such systems are still good candidates for gravitational wave detection
by advanced LIGO class interferometers [43, 97, 116].
Dynamical Nonaxisymmetric Instability Work on dynamical nonaxisymmetric
instabilities is largely outside the scope of this review. They are most likely to be
relevant to proto-neutron stars and to the short-lived hypermassive neutron stars that
form in the merger of a double neutron star system. Unless the star has unusually
high differential rotation, instability requires a large value of the ratio T /|W | of
rotational kinetic energy to gravitational binding energy: comparable to the value
T /|W | = 0.27 that marks the dynamical instability of the l = m = 2 mode of
uniformly rotating uniform density Newtonian models (the Maclauring spheroids).
This bar instability, if present, will emit strong gravitational waves with frequencies
in the kHz regime. The development of the instability and the resulting waveform
have been computed numerically in the context of both Newtonian gravity and in
full general relativity (see [117–120] for representative studies).
Uniformly rotating neutron stars have maximum values of T /|W | smaller than
0.14, apparently precluding dynamical nonaxisymmetric instability. For highly dif-
ferential rotation, however, Centrella et al. [121] found a one-armed (m = 1) insta-
bility for smaller rotation, for T /|W | ≈ 0.14, but for a polytropic index of N = 3
which is not representative for neutron stars. Remarkably, Shibata et al. [122, 123]
then found found an m = 2 instability for T /|W | as low as 0.01, for models with
polytropic index N = 1, representing a stiffness appropriate to neutron stars. These
instabilities appear to be related to the existence of corotation points, where the
pattern speed of the mode matches the star’s angular velocity [124, 125]; Ou and
Tohline [126] tie the growth of the instability to a resonant cavity associated with a
minimum in the vorticity to density ratio (the so-called vortensity) . Collapsing cores
in supernovae are differentially rotating, and these instabilities of proto-neutron stars
arise in simulations of rotating core collapse [127, 128]. Because they can radiate
more energy in gravitational waves than the post-bounce burst signal itself, interest
in these dynamical instabilities is strong.
Acknowledgments J.F.’s work is supported in part by NSF Grant PHY 1001515. N.S.’s is sup-
ported in part by an Excellence Grant for Basic Research (Research Committee of the Aristotle
University of Thessaloniki—87896) and by an IKY–DAAD exchange grant (IKYDA 2012).
References
1. Taub, A.H.: General relativistic variational principal for perfect fluids. Phys. Rev. 94, 1468
(1954). doi:10.1103/PhysRev.94.1468
2. Taub, A.H.: Stability of general relativistic gaseous masses and variational principles. Com-
mun. Math. Phys. 15, 235 (1969). doi:10.1007/BF01645677
3. Schutz, B.F.: Perfect fluids in general relativity: velocity potentials and a variational principle.
4. Schutz Jr, B.F.: Linear pulsations and stability of differentially rotating stellar models. II.
General-relativistic analysis. Astrophys. J. 24, 343 (1972). doi:10.1086/190258
5. Carter, B.: Elastic perturbation theory in general relativity and variational principle for a
rotating solid star. Commun. Math. Phys. 30, 261 (1973). doi:10.1007/BF01645505
6. Schutz, B.F., Sorkin, R.D.: Variational aspects of relativistic field theories with applications
to perfect fluids. Ann. Phys. 107, 1 (1977). doi:10.1016/0003-4916(77)90200-7
7. Calkin, M.G.: An action principle for magnetohydrodynamics. Can. J. Phys. 41, 2241 (1963).
doi:10.1139/p63-216
8. Friedman, J.L.: Generic instability of rotating relativistic stars. Commun. Math. Phys. 62, 247
(1978)
9. Friedman, J.L., Schutz, B.F.: Lagrangian perturbation theory of nonrelativistic fluids. Astro-
phys. J. 221, 937 (1978). doi:10.1086/156098
10. Thorne, K.S.: Validity in general relativity of the Schwarzschild criterion for convection.
Astrophys. J. 144, 201 (1966). doi:10.1086/148595
11. Kovetz, A.: Schwarzschild’s criterion for convective instability in general relativity. Z. Astro-
phys. 66, 446 (1967)
12. Schutz Jr, B.F.: Taylor instabilities in relativistic stars. Astrophys. J. 161, 1173 (1970). doi:10.
1086/150620
13. Bardeen, J.M.: A variational principle for rotating stars in general relativity. Astrophys. J.
162, 71 (1970). doi:10.1086/150635
14. Seguin, F.H.: The stability of nonuniform rotation in relativistic stars. Astrophys. J. 197, 745
(1975). doi:10.1086/153563
15. Detweiler, S.L., Ipser, J.R.: A variational principle and a stability criterion for the non-radial
modes of pulsation of stellar models in general relativity. Astrophys. J. 185, 685 (1973).
doi:10.1086/152447
16. Lebovitz, N.R.: On Schwarzschild’s criterion for the stability of gaseous masses. Astrophys.
J. 142, 229 (1965). doi:10.1086/148279
17. Solberg, H.: Le mouvement d’inertie de l’atmosphere stable et son role dans la theorie des
cyclones, in Proces Verbaux de l’ Association de Météorologie. International Union of Geo-
desy and Geophysics. 6th General Assembly (Edinburgh), vol. 2, pp. 66–82. International
Union of Geodesy and Geophysics, Edinburgh (1936)
18. Abramowicz, M.A.: Rayleigh and Solberg criteria reversal near black holes: the optical geom-
etry explanation. ArXiv:astro-ph/0411718. (2004)
19. Friedman, J.L., Stergioulas, N.: Rotating Relativistic Stars. Cambridge University Press
(2013)
20. Chandrasekhar, S.: The dynamical instability of gaseous masses approaching the Schwarz-
schild limit in general relativity. Astrophys. J. 140, 417 (1964). doi:10.1086/147938
21. Fowler, W.A.: The stability of supermassive stars. Astrophys. J. 144, 180 (1966). doi:10.1086/
148594
22. Friedman, J.L., Ipser, J.R., Sorkin, R.D.: Turning-point method for axisymmetric stability of
rotating relativistic stars. Astrophys. J. 325, 722 (1988). doi:10.1086/166043
23. Cook, G.B., Shapiro, S.L., Teukolsky, S.A.: Spin-up of a rapidly rotating star by angular
momentum loss—Effects of general relativity. Astrophys. J. 398, 203 (1992). doi:10.1086/
171849
24. Takami, K., Rezzolla, L., Yoshida, S.: A quasi-radial stability criterion for rotating relativistic
stars. Mon. Not. R. Astron. Soc. 416, L1 (2011). doi:10.1111/j.1745-3933.2011.01085.x
25. Chandrasekhar, S.: Solutions of two problems in the theory of gravitational radiation. Phys.
26. Friedman, J.L., Schutz, B.F.: Secular instability of rotating Newtonian stars. Astrophys. J.
222, 281 (1978). doi:10.1086/156143
27. Thorne, K.: Multipole expansions of gravitational radiation. Rev. Mod. Phys. 52, 299 (1980).
doi:10.1103/RevModPhys.52.299
28. Stergioulas, N., Friedman, J.L.: Nonaxisymmetric neutral modes in rotating relativistic stars.
Astrophys. J. 492, 301 (1998). doi:10.1086/305030
29. Yoshida, S., Eriguchi, Y.: Neutral points of oscillation modes along equilibrium sequences of
rapidly rotating polytropes in general relativity: application of the Cowling approximation.
Astrophys. J. 490, 779 (1997)
30. Yoshida, S., Eriguchi, Y.: A numerical study of normal modes of rotating neutron star models
by the Cowling approximation. Astrophys. J. 515, 414 (1999). doi:10.1086/307012
31. Zink, B., Korobkin, O., Schnetter, E., Stergioulas, N.: Frequency band of the f-mode
Chandrasekhar-Friedman-Schutz instability. Phys. Rev. D 81(8), 084055 (2010). doi:10.1103/
PhysRevD.81.084055
32. Gaertig, E., Glampedakis, K., Kokkotas, K.D., Zink, B.: The f-mode instability in relativistic
neutron stars. Phys. Rev. Lett. 107, 101102 (2011). doi:10.1007/s10714-010-1059-4
33. Morsink, S., Stergioulas, N., Blattnig, S.: Quasi-normal modes of rotating relativistic stars
- neutral modes for realistic equations of state. Astrophys. J. 510, 854 (1999). doi:10.1086/
306630
34. Chakrabarty, D.: The spin distribution of millisecond X-ray pulsars. In: Wijnands, R. (ed.)
A Decade of Accreting Millisecond X-Ray Pulsars, AIP Conference Proceedings, vol. 1068,
pp. 67–74. (2008)
35. Ghosh, P., Lamb, F.K.: Accretion by rotating magnetic neutron stars. II–Radial and vertical
structure of the transition zone in disk accretion. Astrophys. J. 232, 259 (1979). doi:10.1086/
157285
36. Shapiro, S.L., Teukolsky, S.A.: Black Holes, White Dwarfs and Neutron Stars. Wiley, New
York (1983)
37. White, N.E., Zhang, W.: Millisecond X-ray pulsars in low-mass X-ray binaries. Astrophys.
J. 490, L87 (1997). doi:10.1086/311018
38. Patruno, A., Haskell, B., D’Angelo, C.: Gravitational waves and the maximum spin frequency
of neutron stars. Astrophys. J. 746, 9 (2012). doi:10.1088/0004-637X/746/1/9
39. Stergioulas, N.: Rotating stars in relativity. Living Rev. Relativ. 6, 3 (2003)
40. Andersson, N., Kokkotas, K.D.: The r-mode instability in rotating neutron stars. Int. J. Mod.
Phys. D 10, 381 (2001). doi:10.1142/S0218271801001062
41. Kokkotas, K.D., Ruoff, J.: Instabilities of relativistic stars. In: Proceedings of 2001: A Rela-
tivistic Spacetime Odyssey 2002, Firenze (2001)
42. Andersson, N., Ferrari, V., Jones, D.I., et al.: Gravitational waves from neutron stars: promises
and challenges. Gen. Relativ. Grav. 43, 409 (2011). doi:10.1007/s10714-010-1059-4
43. Owen, B.J.: How to adapt broad-band gravitational-wave searches for r-modes. Phys. Rev. D
82, 104002 (2010). doi:10.1103/PhysRevD.82.104002
44. Kokkotas, K.D., Stergioulas, N.: Analytic desctription of the r-mode instability in uniform
density stars. Astron. Astrophys. 341, 110 (1999)
45. Cutler, C., Lindblom, L., Splinter, R.J.: Damping times for neutron star oscillations. Astrophys.
J. 363, 603 (1990). doi:10.1086/169370
46. Ipser, J.R., Lindblom, L.: On the adiabatic pulsations of accretion disks and rotating stars.
Astrophys. J. 379, 285 (1991)
47. Ipser, J.R., Lindblom, L.: The oscillations of rapidly rotating Newtonian stellar models. II—
dissipative effects. Astrophys. J. 373, 213 (1991). doi:10.1086/170039
48. Yoshida, S., Eriguchi, Y.: Gravitational radiation driven secular instability of rotating poly-
tropes. Astrophys. J. 438, 830 (1995). doi:10.1086/175126
49. Lai, D., Shapiro, S.L.: Gravitational radiation from rapidly rotating nascent neutron stars.
Astrophys. J. 442, 259 (1995). doi:10.1086/175438
50. Bonazzola, S., Frieben, J., Gourgoulhon, E.: Spontaneous symmetry breaking of rapidly rotat-
ing stars in general relativity. Astrophys. J. 460, 379 (1996). doi:10.1086/176977
51. Lai, D.: Secular bar-mode evolution and gravitational waves from neutron stars. Astrophysical
sources for ground-based gravitational wave detectors. In: Proceedings of AIP Conference,
vol. 575, pp. 246–257. (2001)
52. Zdunik, J.L.: Damping of GRR instability by direct URCA reactions. Astron. Astrophys. 308,
828 (1996)
53. Jones, P.B.: Comment on Gravitational radiation instability in hot young neutron stars. Phys.
54. Lindblom, L., Owen, B.J.: Effect of hyperon bulk viscosity on neutron-star r-modes. Phys.
55. Haensel, P., Levenfish, K.P., Yakovlev, D.G.: Bulk viscosity in superfluid neutron star cores.
III. Effects of Σ − hyperons. Astron. Astrophys. 381, 1080 (2002). doi:10.1051/0004-6361:
20011532
56. Nayyar, M., Owen, B.J.: R-modes of accreting hyperon stars as persistent sources of gravita-
tional waves. Phys. Rev. D 73, 084001 (2006). doi:10.1103/PhysRevD.73.084001
57. Haskell, B., Andersson, N.: Superfluid hyperon bulk viscosity and the r-mode instability of
rotating neutron stars. Mon. Not. R. Astron. Soc. 408, 1897 (2010)
58. Madsen, J.: How to identify a strange star. Phys. Rev. Lett. 81, 3311 (1998). doi:10.1103/
PhysRevLett.81.3311
59. Madsen, J.: Probing strange stars and color superconductivity by r-mode instabilities in mil-
lisecond pulsars. Phys. Rev. Lett. 85, 10 (2000). doi:10.1103/PhysRevLett.85.10
60. Andersson, N., Jones, D.I., Kokkotas, K.D.: Strange stars as persistent sources of gravitational
waves. Mon. Not. R. Astron. Soc. 337, 1224 (2002). doi:10.1046/j.1365-8711.2002.05837.x
61. Jaikumar, P., Rupak, G., Steiner, A.W.: Viscous damping of r-mode oscillations in compact
stars with quark matter. Phys. Rev. D 78, 123007 (2008). doi:10.1103/PhysRevD.78.123007
62. Rupak, G., Jaikumar, P.: Constraining phases of quark matter with studies of r-mode damping
in compact stars. Phys. Rev. C 82, 055806 (2010). doi:10.1103/PhysRevC.82.055806
63. Flowers, E., Itoh, N.: Transport properties of dense matter. Astrophys. J. 206, 218 (1976).
doi:10.1086/154375
64. Shternin, P.S., Yakovlev, D.G.: Shear viscosity in neutron star cores. Phys. Rev. D 78, 063006
(2008). doi:10.1103/PhysRevD.78.063006
65. Bildsten, L., Ushomirsky, G.: Viscous boundary-layer damping of r-modes in neutron stars.
Astrophys. J. 529, L33 (2000). doi:10.1086/312454
66. Lindblom, L., Owen, B.J., Ushomirsky, G.: Effect of a neutron-star crust on the r-mode
instability. Phys. Rev. D 62, 084030 (2000). doi:10.1103/PhysRevD.62.084030
67. Andersson, N., Jones, D.I., Kokkotas, K.D., Stergioulas, N.: R-mode runaway and rapidly
rotating neutron stars. Astrophys. J. 534, L75 (2000). doi:10.1086/312643
68. Glampedakis, K., Andersson, N.: Crust-core coupling in rotating neutron stars. Phys. Rev. D
74, 044040 (2006). doi:10.1103/PhysRevD.74.044040
69. Glampedakis, K., Andersson, N.: Ekman layer damping of r modes revisited. Mon. Not. R.
Astron. Soc. 371, 1311 (2006). doi:10.1111/j.1365-2966.2006.10749.x
70. Lindblom, L., Mendell, G.: Does gravitational radiation limit the angular velocities of super-
fluid neutron stars. Astrophys. J. 444, 804 (1995). doi:10.1086/175653
71. Lindblom, L., Mendell, G.: R-modes in superfluid neutron stars. Phys. Rev. D 61, 104003
(2000). doi:10.1103/PhysRevD.61.104003
72. Spruit, H.C.: Gamma-ray bursts from X-ray binaries. Astron. Astrophys. 341, L1 (1999)
73. Rezzolla, L., Lamb, F.K., Shapiro, S.L.: R-mode oscillations in rotating magnetic neutron
stars. Astrophys. J. 531, L139 (2000). doi:10.1086/312539
74. Rezzolla, L., Lamb, F.K., Marković, D., Shapiro, S.L.: Properties of r modes in rotating
magnetic neutron stars. I. Kinematic secular effects and magnetic evolution equations. Phys.
75. Rezzolla, L., Lamb, F.K., Marković, D., Shapiro, S.L.: Properties of r modes in rotating
magnetic neutron stars. II. Evolution of the r modes and stellar magnetic field. Phys. Rev. D
64, 104014 (2001). doi:10.1103/PhysRevD.64.104014
76. Cuofano, C., Drago, A.: Magnetic fields generated by r-modes in accreting millisecond pulsars.
77. Kojima, Y.: Quasi-toroidal oscillations in rotating relativistic stars. Mon. Not. R. Astron. Soc.
293, 49 (1998). doi:10.1046/j.1365-8711.1998.01119.x
78. Andersson, N.: A new class of unstable modes of rotating relativistic stars. Astrophys. J. 502,
708 (1998). doi:10.1086/305919
79. Kojima, Y., Hosonuma, M.: The r-mode oscillations in relativistic rotating stars. Astrophys.
J. 520, 788 (1999). doi:10.1086/307481
80. Kojima, Y., Hosonuma, M.: Approximate equation relevant to axial oscillations on slowly
rotating relativistic stars. Phys. Rev. D 62, 044006 (2000). doi:10.1103/PhysRevD.62.044006
81. Lockitch, K.H., Andersson, N., Friedman, J.L.: Rotational modes of relativistic stars: Analytic
results. Phys. Rev. D 63, 024019 (2001). doi:10.1103/PhysRevD.63.024019
82. Lockitch, K.H., Friedman, J.L., Andersson, N.: Rotational modes of relativistic stars: Numer-
ical results. Phys. Rev. D 68, 124010 (2003). doi:10.1103/PhysRevD.68.124010
83. Lockitch, K.H., Andersson, N., Watts, A.L.: Regularizing the r-mode problem for non-
barotropic relativistic stars. Class. Quant. Grav. 21, 4661 (2004). doi:10.1088/0264-9381/
21/19/012
84. Ruoff, J., Kokkotas, K.D.: On the r-mode spectrum of relativistic stars in the low-frequency
approximation. Mon. Not. R. Astron. Soc. 328, 678 (2001). doi:10.1046/j.1365-8711.2001.
04909.x
85. Ruoff, J., Kokkotas, K.D.: On the r-mode spectrum of relativistic stars: the inclusion of the
radiation reaction. Mon. Not. R. Astron. Soc. 330, 1027 (2002). doi:10.1046/j.1365-8711.
2002.05169.x
86. Ruoff, J., Stavridis, A., Kokkotas, K.D.: Inertial modes of slowly rotating relativistic stars
in the Cowling approximation. Mon. Not. R. Astron. Soc. 339, 1170 (2003). doi:10.1046/j.
1365-8711.2003.06267.x
87. Yoshida, S., Lee, U.: Relativistic r-modes in slowly rotating neutron stars: numerical analysis
in the Cowling approximation. Astrophys. J. 567, 1112 (2002). doi:10.1086/338663
88. Kastaun, W.: Inertial modes of rigidly rotating neutron stars in Cowling approximation. Phys.
89. Stergioulas, N., Font, J.A.: Nonlinear r-modes in rapidly rotating relativistic stars. Phys. Rev.
90. Lindblom, L., Tohline, J.E., Vallisneri, M.: Non-linear evolution of the r-modes in neutron
stars. Phys. Rev. Lett. 86, 1152 (2001). doi:10.1103/PhysRevLett.86.1152
91. Lindblom, L., Tohline, J.E., Vallisneri, M.: Numerical evolutions of nonlinear r-modes in
neutron stars. Phys. Rev. D 65, 084039 (2002). doi:10.1103/PhysRevD.65.084039
92. Gressman, P., Lin, L.M., Suen, W.M., Stergioulas, N., Friedman, J.L.: Nonlinear r-modes in
neutron stars: instability of an unstable mode. Phys. Rev. D 66, 041303 (2002). doi:10.1103/
PhysRevD.66.041303
93. Lin, L.M., Suen, W.M.: Non-linear r-modes in neutron stars: a hydrodynamical limitation on
r-mode amplitudes. Mon. Not. R. Astron. Soc. 370, 1295 (2006). doi:10.1111/j.1365-2966.
2006.10536.x
94. Schenk, A.K., Arras, P., Flanagan, E.E., Teukolsky, S.A., Wasserman, I.: Nonlinear mode
coupling in rotating stars and the r-mode instability in neutron stars. Phys. Rev. D 65, 024001
(2002). doi:10.1103/PhysRevD.65.024001
95. Arras, P., Flanagan, E.E., Morsink, S.M., et al.: Saturation of the r-mode instability. Astrophys.
J. 591, 1129 (2003). doi:10.1086/374657
96. Morsink, S.M.: Relativistic precession around rotating neutron stars: effects due to frame-
dragging and stellar oblateness. Astrophys. J. 571, 435 (2002)
97. Bondarescu, R., Teukolsky, S.A., Wasserman, I.: Spin evolution of accreting neutron stars:
nonlinear development of the r-mode instability. Phys. Rev. D 76, 064019 (2007). doi:10.
1103/PhysRevD.76.064019
98. Bondarescu, R., Teukolsky, S.A., Wasserman, I.: Spinning down newborn neutron stars: non-
linear development of the r-mode instability. Phys. Rev. D 79, 104003 (2009). doi:10.1103/
PhysRevD.79.104003
99. Ou, S., Tohline, J.E., Lindblom, L.: Nonlinear development of the secular bar-mode instability
in rotating neutron stars. Astrophys. J. 617, 490 (2004). doi:10.1086/425296
100. Shibata, M., Karino, S.: Numerical evolution of secular bar-mode instability induced by the
gravitational radiation reaction in rapidly rotating neutron stars. Phys. Rev. D 70, 084022
(2004). doi:10.1103/PhysRevD.70.084022
101. Kastaun, W., Willburger, B., Kokkotas, K.D.: Saturation amplitude of the f-mode instability.
102. Gaertig, E., Kokkotas, K.D.: Gravitational wave asteroseismology with fast rotating neutron
stars. Phys. Rev. D 83, 064031 (2011). doi:10.1103/PhysRevD.83.064031
103. Page, D., Prakash, M., Lattimer, J.M., Steiner, A.W.: Rapid cooling of the neutron star in
Cassiopeia a triggered by neutron superfluidity in dense matter. Phys. Rev. Lett. 106, 081101
(2011). doi:10.1103/PhysRevLett.106.081101
104. Shternin, P.S., Yakovlev, D.G., Heinke, C.O., Ho, W.C.G., Patnaude, D.J.: Cooling neutron
star in the Cassiopeia a supernova remnant: evidence for superfluidity in the core. Mon. Not.
R. Astron. Soc. 412, L108 (2011). doi:10.1111/j.1745-3933.2011.01015.x
105. Friedman, J.L., Morsink, S.M.: Axial instability of rotating relativistic stars. Astrophys. J.
502, 714 (1998). doi:10.1086/305920
106. Lindblom, L., Owen, B.J., Morsink, S.M.: Gravitational radiation instability in hot young
neutron stars. Phys. Rev. Lett. 80, 4843 (1998). doi:10.1103/PhysRevLett.80.4843
107. Andersson, N., Kokkotas, K.D., Schutz, B.F.: Gravitational radiation limit on the spin of
young neutron stars. Astrophys. J. 510, 846 (1999). doi:10.1086/306625
108. Papaloizou, J., Pringle, J.E.: Gravitational radiation and the stability of rotating stars. Mon.
Not. R. Astron. Soc. 184, 501 (1978)
109. Wagoner, R.V.: Gravitational radiation from accreting neutron stars. Astrophys. J. 278, 345
(1984). doi:10.1086/161798
110. Bildsten, L.: Gravitational radiation and rotation of accreting neutron stars. Astrophys. J. 501,
L89 (1998). doi:10.1086/311440
111. Andersson, N., Kokkotas, K.D., Stergioulas, N.: On the relevance of the r-mode instability for
accreting neutron stars and white dwarfs. Astrophys. J. 516, 307 (1999). doi:10.1086/307082
112. Wagoner, R.V.: Conditions for steady gravitational radiation from accreting neutron stars.
Astrophys. J. 578, L63 (2002). doi:10.1086/344502
113. Levin, Y.: Runaway heating by r-modes of neutron stars in low-mass X-ray binaries. Astro-
phys. J. 517, 328 (1999). doi:10.1086/307196
114. Heyl, J.: Low-mass X-ray binaries may be important laser interferometer gravitational-wave
observatory sources after all. Astrophys. J. 574, L57 (2002). doi:10.1086/342263
115. Reisenegger, A., Bonacić, A.: Millisecond pulsars with r-modes as steady gravitational radi-
ators. Phys. Rev. Lett. 91, 201103 (2003). doi:10.1103/PhysRevLett.91.201103
116. Watts, A.L., Krishnan, B.: Detecting gravitational waves from accreting neutron stars. Adv.
Space Res. 43, 1049 (2009). doi:10.1016/j.asr.2009.01.006
117. Houser, J.L., Centrella, J.M., Smith, S.C.: Gravitational radiation from nonaxisymmetric
instability in a rotating star. Phys. Rev. Lett. 72, 1314 (1994). doi:10.1103/PhysRevLett.
72.1314
118. Tohline, J.E., Durisen, R.H., McCollough, M.: The linear and nonlinear dynamic stability of
rotating n = 3/2 polytropes. Astrophys. J. 298, 220 (1985). doi:10.1086/163600
119. Shibata, M.: Axisymmetric simulations of rotating stellar collapse in full general relativity—
criteria for prompt collapse to black holes. Progress Theoret. Phys. 104, 325 (2000). doi:10.
1143/PTP.104.325
120. Manca, G.M., Baiotti, L., De Pietri, R., Rezzolla, L.: Dynamical non-axisymmetric insta-
bilities in rotating relativistic stars. Class. Quant. Grav. 24, S171 (2007). doi:10.1088/0264-
9381/24/12/S12
121. Centrella, J.M., New, K.C.B., Lowe, L.L., Brown, J.D.: Dynamical rotational instability at
low T/W. Astrophys. J. 550, L193 (2001). doi:10.1086/319634
122. Shibata, M., Karino, S., Eriguchi, Y.: Dynamical instability of differentially rotating stars.
Mon. Not. R. Astron. Soc. 334, L27 (2002). doi:10.1046/j.1365-8711.2002.05724.x
123. Shibata, M., Karino, S., Eriguchi, Y.: Dynamical bar-mode instability of differentially rotating
stars: effects of equations of state and velocity profiles. Mon. Not. R. Astron. Soc. 343, 619
(2003). doi:10.1046/j.1365-8711.2003.06699.x
124. Watts, A.L., Andersson, N., Jones, D.I.: The nature of low T /|W | dynamical instabilities in
differentially rotating stars. Astrophys. J. 618, L37 (2005). doi:10.1086/427653
125. Saijo, M., Yoshida, S.: Low T /|W | dynamical instability in differentially rotating stars: diag-
nosis with canonical angular momentum. Mon. Not. R. Astron. Soc. 368, 1429 (2006). doi:10.
1111/j.1365-2966.2006.10229.x
126. Ou, S., Tohline, J.E.: Unexpected dynamical instabilities in differentially rotating neutron
stars. Astrophys. J. 651, 1068 (2006). doi:10.1086/507597
127. Ott, C.D., Ou, S., Tohline, J.E., Burrows, A.: One-armed spiral instability in a low-T /|W |
postbounce supernova core. Astrophys. J. 625, L119 (2005). doi:10.1086/431305
128. Ott, C.D.: The gravitational-wave signature of core-collapse supernovae. Class. Quant. Grav.
26, 063001 (2009). doi:10.1088/0264-9381/26/6/063001
Gravity Talks: Observing the Universe
with Gravitational Waves
Bernard F. Schutz
Abstract When the current upgrade of the large ground-based gravitational wave
detectors LIGO and VIRGO is completed, the new science of gravitational wave
astronomy will begin. In this overview I review the current status of the detector
projects on the ground and in space (LISA), the kinds of signals and sources they
expect to observe, and the science returns that are anticipated.
1 Introduction
The effort to detect gravitational waves has been one of the most remarkable examples
of sustained technology development in the history of physics and astronomy. Like
most kinds of instrumentation, gravitational wave detectors have been developed
through many steps, each one bringing a significant improvement in performance.
Unlike all other kinds of instrumentation that I am aware of, improvements in
the sensitivity of gravitational wave detectors have not yet—after over 50 years of
improvements—led to a single detection. In all other areas of physics and astronomy,
early instruments have made at least some measurements or detections of interest,
and these have stimulated and justified the next improvements in performance. This
cycle of observation/improvement/observation/improvement is clearly evident in the
histories of optical, radio, X-ray, and cosmic-ray astronomy during the last 50 years.
For gravitational wave detectors, evolving over the same period of time, the cycle was
simply improvement/improvement/improvement. The effort has been sustained, not
by the success of serendipitous observation, but by the dedication of those working
in the field and by the deep conviction (shared, of course, by the science-funding
bodies) not only that gravitational waves of a certain amplitude are certainly passing
through the Earth regularly, but that detecting these waves will dramatically change
our view of the universe.
B. F. Schutz(B)
Max Planck Institute for Gravitational Physics (Albert Einstein Institute), Potsdam, Germany
e-mail: bernard.schutz@aei.mpg.de
460 B. F. Schutz
The 50-year effort has not been without other rewards for those involved. On the
experiment side, many fascinating new technologies have been developed to allow
us to make the most sensitive measurements of distance changes ever accomplished.
Detector sensitivity after the current upgrades will be something like 107 times better
than the original instruments built in the 1960s [1]. This maps directly into range: our
detectors will be able to detect signals from sources 107 times further away than the
early detectors could. On the theory side, the possibility of observing gravitational
waves has stimulated many important developments, not least the ability to solve
Einstein’s equations on computers and to simulate systems that cannot be studied
analytically. But the big goal of direct detections has not yet been reached.
This will not be the case for much longer. The large ground-based detectors
LIGO, VIRGO, and GEO600 have already taken data with such high sensitivity that
detections could have happened if we had been a bit lucky with how close the nearest
event would happen; that no detections were registered was not a surprise, even if a
slight disappointment. But the next phase of data-taking with LIGO and VIRGO will
be different: if the instruments now being upgraded perform to specification, and if
the astrophysical event rate is close to the estimates made by the collaboration [2],
then the first detection is likely by around 2017, and it is very unlikely that nothing
will be registered before, say, 2020.
Unlike in the other areas of astronomy that have had the normal observa-
tion/improvement cycle, we believe we actually have a pretty good idea of what
the first detected gravitational wave signals will be: they will be from binary sys-
tems in which compact objects (neutron stars and/or black holes) spiral together
and merge. This expectation is based on theory—extensive studies of the motion
of compact objects in fully general-relativistic binaries, and of the astrophysics of
these objects—which was needed during the last 50 years in order to justify the effort
being put into the technology development, but which has even more importantly be-
come an essential part of the detection chain. The accurate predictions we now have
of what the gravitational waveforms from binaries and other systems (for example,
gravitational wave pulsars) should look like permit us to dig deep into instrumental
noise and detect with confidence signals that might otherwise have to wait for yet
another cycle of instrument improvement. Good data analysis based on theoretical
waveform predictions probably improves the sensitivity (equivalently, the distance
reach) of our detectors by factors of 10–30 for binaries, and up to several thousand
for pulsars.
Theoretical studies have led also to the development during the last 10 years of
a new paradigm for the kind of observing that will be done by gravitational wave
detectors. We now think of it as “listening” to the universe, rather than “watching”
it, as one does with normal telescopes. Our detectors are not pointed, but rather
omni-directional, like microphones. The data stream is one-dimensional, like an
audio stream. The waves are detected coherently, like audio waves, rather than just
bolometrically, like photons; in fact, the phase evolution of the waves contains more
information about the sources than the amplitude does. All these analogies help to
understand how we will extract information from our arrays of detectors. And in the
same way that hearing complements vision in animals, gravitational wave observing
Gravity Talks: Observing the Universe 461
will add a qualitatively new “sense” to our ability to record the universe around us:
relativistic systems we don’t know about, or are hidden, or are unable to radiate
light can be discovered, located, and studied using our network of microphone-like
detectors.
With first observations not far in the future, this meeting is a good time to look
forward toward the coming gravitational wave astronomy: what the instruments are,
what the sources are that are most likely to be detected, and what the kinds of
information are that we are likely to be able to infer when we are finally able to hear
the universe speaking to us through the medium of gravity.
2 Light Deflection and Gravitational Wave Detection

Since this meeting marks Einstein’s work in Prague, it seems appropriate to draw a
link between his work and the detection of gravitational waves. One of the subjects
that drew Einstein’s attention in Prague in 1911-12 was the deflection of light by
gravity. It is perhaps amusing that our present method of detecting gravitational
waves by interferometry also relies on the action of gravity on light.
Einstein’s work in Prague [3] was his famous demonstration that light, on passing
the Sun or another body, will be deflected. Because he relied essentially only on
the equivalence principle (the curvature of time) and did not yet have a theory that
included the curvature of space, he got only half of the right value; nevertheless it
was an important advance in his own thinking.
In principle he could have predicted also that the propagation time of light would
change due to the deflection, something we now call the Shapiro Effect. But as there
was no way to measure this with light from a distant star, he would not have given the
idea much thought. Nevertheless it is precisely these propagation-time changes that
we use in detectors today. The gravitational wave makes a time-dependent alteration
in the time it takes light to move up and back along the arms of a detector, and by
interferometry we can compare this to the propagation time in the perpendicular
arm. The signature of a gravitational wave is a difference in the propagation time
alterations between the two arms.
3 The Global Interferometer Network

The worldwide network of gravitational wave detectors, illustrated in Fig. 1, con-
sists of three large instruments (two LIGO and VIRGO), one medium-sized detector
(GEO600), and two projects: KAGRA in Japan and LIGO-India. The existing de-
tectors have successfully reached their first-stage sensitivity goals and demonstrated
that they could operate reliably and produce data. The data have been extensively
analyzed, leading to over 70 papers reporting methods and results [4], but no gravi-
tational wave signals were found in the data. This was not a surprise, since the first
sensitivity goals were modest relative to the expected event rates, so now the three
large instruments are upgrading their sensitivities by a factor of about 10, at which
level it would be very surprising if no detections were made. During this upgrade, the
462 B. F. Schutz
Fig. 1 The six large interferometers currently operating, upgrading, or planned
GEO600 detector is spending about 70 % of its time in “Astrowatch” mode, meaning

it is taking data just in case an interesting astronomical event occurs nearby, such as
a supernova in our galactic neighborhood. The rest of the time it is also upgrading
its sensitivity. The LIGO and GEO600 detectors are jointly managed and developed
within the LIGO Scientific Collaboration (LSC).
GEO600 has functioned as a technology development platform as well as a detec-
tor, and a number of the important technologies that are being used in the upgrades
of the larger instruments were first developed and/or tested in GEO600: high laser
powers, monolithic suspensions for controlling thermal noise, signal recycling, and
squeezed light. The basic installation that will turn LIGO into Advanced LIGO will
be completed in 2015, and there will follow a series of commissioning periods alter-
nating with observing runs, as the sensitivity is improved to the final goal. By 2017 it
seems reasonable to expect the first detections to have occurred. Advanced VIRGO
is on the same trajectory, perhaps 6 months to a year later.
KAGRA (previously LCGT) is a 3-km scale instrument with very ambitious tech-
nology: underground and cryogenic. It is now digging its tunnels in the Kamiokande
mine in Japan, and should be producing data with a sensitivity comparable to that of
Advanced LIGO and VIRGO before the end of the decade. The newest development
is LIGO-India, in which LIGO will install a detector into a vacuum system built by
India. The project has been endorsed in the US by the National Science Board, and
is awaiting final funding approval at cabinet level in India.
In the longer-term future, scientists have studied so-called third-generation
detectors and their technology. The most complete study was for the European
Einstein Telescope [5, 6], which will probably bid for first funding toward the end
of the current decade.
All of these instruments share data (or in the case of KAGRA and LIGO-India, plan
to do so) and publish results jointly, with more than 600 authors on each paper. When
one thinks about it, this degree of cooperation is unusual among physicists. Large
collaborations are not uncommon, but normally there are two or more collaborations
that compete with one another. In gravitational waves, all the detectors cooperate.
Such harmony is driven by the science: there are good reasons why the science
benefits from pooling data. First, one could not reliably claim a detection of a short
burst of gravitational radiation on the basis of the data of a single detector, because
the signals are so weak that it would be easy for some un-modeled noise in a de-
tector to masquerade as signal. Second, detectors are almost omni-directional, so to
get direction information and be able to solve for the polarization of a short burst
signal one needs three or more detectors, triangulating the position on the sky from
time-delays among the detectors. That means that the global network itself forms
an interferometer in the same sense that radio telescopes around the globe join in
VLBI networks. And with four or more detectors the science gets even better: since
instruments have duty cycles (for first-stage LIGO it was around 80 %), more detec-
tors means that there is better coverage of the sky in both direction and time. And
with more observed signals it is possible to extract better waveform information by
averaging over the (independent) noise in the various detectors.
These advantages can be made quantitative in terms of some figures of merit for
different networks, introduced in [7]. The Triple Detection Rate is a number that
reflects the sensitivity of a network (as measured by the spatial volume inside its
antenna pattern) and the time-coverage for an assumed duty cycle. For networks of
more than three detectors it weights the product of time and volume for all three-
detector sub-networks, on the grounds that one needs at least three detectors to get a
reasonable amount of science from a detection. The Sky Resolution figure of merit
measures the inverse of the typical area of the error box on the sky for locating
an observed event. Both of these measures are given for three detector networks in
Table 1. The networks are: the originally planned LIGO-VIRGO network, with two
detectors at LIGO Hanford in Washington, one at LIGO Livingston in Louisiana,
and the VIRGO detector (HHLV); VIRGO with the extended LIGO network with the
second Hanford detector moved to its new home in India (HILV); and the ultimate
worldwide network including Japan (HIJLV). The numbers are used only to compare
networks, so we do not give scalings that allow one to go from a figure of merit to
a measurable quantity like the area of the error box. But the relative values are
significant: HIJLV produces sky location error boxes that are a factor of 7.6 smaller
in area than HHLV, for example. This and the improvement in location going from
HHLV to HILV are the main motivation for building the two detectors in India and
Japan. However, one also sees that the added sensitivity and time-coverage if one
assumes 80 % duty cycle for the new detectors will practically double the number of
events that are detected with three or more detectors, comparing HHLV to HIJLV.
Anyone interested in the sociology of the international gravitational wave col-
laboration, about how scientists manage to cooperate and reach decisions without
having a strong hierarchical organization, should read the study [8]. The author,
464 B. F. Schutz
Table 1 Figures of merit for three networks (see text for explanation). Data from [7]
Network Triple rate Resolution
HHLV 4.86 0.65
HILV 5.94 2.96
HIJLV 8.62 4.60
H Collins, has privileged access to the inner discussions of the collaboration, which
he uses for his sociological study.
4 Data Analysis
Data analysis is an important part of any physics experiment or astronomical obser-
vation, but it plays a particularly important role in gravitational waves detection. In
fact, detection with the ground-based instruments relies on three key components: the
detectors’ sensitivity, the accuracy of the source modeling (e.g. predicting waveforms
from black hole mergers), and the sophistication of the data analysis.
The signals we expect will not be easily visible in the data streams: they will
emerge only after processing the data through intelligent filters that remove noise
and enhance signal. This so-called matched filtering relies, in turn, on good predicted
waveforms. So, unlike in most branches of astronomy, where the modeling of the
source comes after the data have been analyzed, here we need the modeling first as an
input to the data analysis. And because noise can always masquerade as signal, any
detection statement will be a statistical one. The LSC and VIRGO collaborations have
agreed that their first claim to detection will need to have a “5σ” level of confidence,
that there is less than one chance in three million that noise could have created the
claimed signal at any time during the entire data run in which the detection was
made. This will need to take into account the fact that searches are done over large
parameter spaces of possible signals (different masses, spins, sky locations, etc), so
that there is a large “trials factor” for all the different independent filters that have
been employed, which additionally discounts the significance of a detection.
Because the noise is not an ideal Gaussian distribution, but rather has a population
of random instrumental “glitches” that have to be discriminated from signals, and
because the number of filters is so large that the computational demands sometimes
exceed the available resources, it is generally not possible to do fully optimum signal
analysis. Instead, there is a premium on clever data analysis algorithms that get closer
to the theoretical optimum sensitivity. This means that if it is possible to develop
an algorithm that digs deeper into the noise by a factor of 2, then that algorithm
effectively improves the sensitivity of the detectors to that kind of signal by a factor
of 2. Developing such algorithms is usually a lot less expensive than upgrading the
hardware to achieve a sensitivity change by a factor of 2. That is why, alongside the
ongoing hardware upgrades, there is a large amount of work going into algorithm
development.
But besides algorithms, accurate input from the wider research community on
source and waveform models can be similarly effective in improving sensitivity.
The NINJA [9] and NRAR [10] collaborations are working hard to bring numerical
relativity results into the data analysis systems.
To cope with the unusually large demand for computing power of some of the
searches (particularly for low-amplitude continuous signals from pulsars) the grav-
itational wave community has developed the Einstein@Home platform [11]. This
is a screen saver that performs data analysis when a computer is otherwise idle. So
many people have downloaded and participate in the search that the platform cur-
rently delivers almost 700 Tflops of performance continuously! It has also been used
recently to search through radio and gamma-ray data for pulsars, with great success,
discovering systems that had been missed by previous data analysis. This validates
the methods and approach of the gravitational wave community in the search for
long-duration signals.
Because the data analysis involves such a large amount of preparation and re-
source, there has been considerable discussion about the value of public data releases.
Current policy by the LSC is to release small segments of data that contain confirmed
signals as soon as possible. Full data releases will occur after a period of perhaps
2 years, once the full data analysis apparatus of the collaboration has been applied
and the results understood. The collaboration is also preparing software interfaces to
allow scientists not experienced in the projects to perform their own analyses. One
unknown is how much support the collaboration will be funded to provide for outside
analysis: much of the complication of the analysis has to do with characteristics of
the detectors (un-modeled noise, etc) that may prove difficult for non-members to
deal with.
5 Observables
The data analysis ultimately determines the values of the parameters that are the
unknowns in the family of signals that are used as the template of the search, within
the uncertainties of the observations. For the most commonly expected signals, from
binary coalescence, the principal parameters are:
• Location on the sky. Fundamentally this comes from the differences in the arrival
times of the signals at the various detectors. Actually the arrival time is not well-
defined for a long-duration signal like that from a binary, so some fiducial time
has to be defined. One such time would be the expected arrival time of the final
coalescence signal of the system, if it consisted of two point particles. This can be
predicted from the inspiral waveform. The actual merger will be more complicated,
but the (fictional) coalescence arrival time provides a well defined time-parameter
determined by fitting the theoretical waveform to the data. Three such times of
arrival are enough to define two possible locations on the sky. With four detectors
this ambiguity is resolved.
466 B. F. Schutz
• Polarization. Once the direction to the source is known, two detector signals suffice
to determine the two polarizations. The polarization tells us the inclination of the
plane of the binary orbit to the line of sight. If the polarization changes (say through
orbit precession) then only one of the two possible directions determined by three
detectors will remain fixed in the sky. If there are three or more detectors and
the location of the signal is known, then the two polarization signals are over-
determined, at least if general relativity is valid, and that redundancy is useful:
there is a linear combination of outputs that contains no gravitational wave signal:
this is called a null stream, and it can be used, for example, to veto spurious signals
of instrumental origin. For N > 2 detectors there are N −2 null streams. With three
or more detectors it is possible to test the general-relativity model of polarization;
this could reveal new gravitational fields, which are a feature of many unified field
theories for quantum gravity.
• Masses and spins of the component masses. For signals that can be well modeled,
the phase/frequency information tells us about the masses. For a binary, fitting a
post-Newtonian waveform description to the orbit is enough to determine the indi-
vidual masses and spins, provided the signal is strong enough. The parameter that
3/5 3/5
is most reliably determined is the so-called chirp mass M = m 1 m 2 /M 1/5 =
μ3/5 M, where m k is the mass of the kth star, M = m 1 + m 2 the total mass, and
μ = m 1 m 2 /M 2 the symmetric mass ratio [12].
• Distance to the source. Remarkably, binary signals from systems whose orbital
frequency changes during the observation due to gravitational radiation reaction
contain enough information to determine the distance to the binary system: they
are standard candles, or (more appropriately) standard sirens [12]. To understand
how this is possible, consider first the three observables that one can measure for a
circular binary with equal masses: the frequency f of the signal, its time-scale for
changing τc = f / f˙ (called the chirp time), and the intrinsic amplitude of the wave
h (which is only known if the location of the signal on the sky and its polarization
have also been measured). These three depend on the mass m of both components,
their orbital radius a, and the distance d to the source. With three measured values,
each of these three unknowns can be determined, and in fact to within factors of
order unity
d ≈ c/( f 2 τc h).
If the system has unequal mass components, then this formula still applies, because
only the chirp mass M is needed. If the system is eccentric, then there is enough
information in the phase of the signal to determine the eccentricity and come back
to the same formula. And if the system is at cosmological distances, then the
distance measured is the luminosity distance D L . We expect this to be a powerful
added tool in astronomy, by checking the standard astronomical distance ladder
(which this method is completely independent of), by helping to identify binary
sources and their host galaxies, and even by providing a local value of the Hubble
constant [12, 13].
It is of course important to understand that not all parameters are determined

equally well; in fact the errors in some of them can have strong covariances with
one another. This issue is studied in detail by the search teams as they do their data
analysis. The location on the sky can have strong covariances with polarization, for
example. Determining the position accurately is a key for doing Multimessenger
Astronomy: finding correlated signals in the gravitational wave and electromagnetic
spectra. Conversely, an independently determined position (from an electromagnetic
observation) will reduce the position errors and thereby improve the determination
of other parameters.
6 Gravitational Waves from Neutron Stars and Black Holes
6.1 Neutron Star Binary Coalescence
The in-spiral and coalescence of two neutron stars is the top candidate for the first-
ever detection by the ground-based network. This is because the rate of such events
in the universe is fairly well understood [2] and the signal template is very accurately
modeled. The best estimate is that a three-detector network of interferometers at the
advanced level of sensitivity (which we expect to have from 2019 onwards) should
detect something like 40 events per year. The first event could come in 2017, earlier if
we are fortunate. With an enlarged network that includes LIGO-India and KAGRA,
the event rate could go up to 100–150 events per year. The maximum distance for
such events will be in the range 400–600 Mpc. The enlarged network does not have
a greatly increased reach. Instead, it covers the sky more isotropically, so it does not
miss very many events out to this distance [7].
These events will bring a great deal of exciting science. We will have a much
better sampling of the binary pulsar population, leading to mass distributions that
could give clues to the prior evolution of such systems. The strongest (closest) event
in any year should have a high enough signal-to-noise ratio (SNR) to provide a strong
constraint on the neutron-star equation of state. These mergers are likely to produce
gamma-ray bursts [14]; see Fig. 2. Because such bursts are strongly beamed, they
are not likely to be seen with each gravitational wave detection, but there should be
a few coincidences per year. The combined gamma-ray and gravitational-wave data
could also provide further insight into the physics of neutron stars [15].
By comparing the time of arrival of gravitational waves and gamma rays, one
can constrain the difference in the speed of electromagnetic and gravitational waves;
again, this would be a fundamental test of general relativity [16]. And by comparing
the arrival times of the left- and right-handed circular polarization components of the
waves (which will be possible by detailed fitting to the signal template) one can look
for birefringence in the propagation of gravitational waves, something again that is
possible in modified theories of gravity [17].
468 B. F. Schutz
Fig. 2 A poloidal magnetic field structure that arises naturally from binary neutron-star coalescence,
as computed using fully general-relativistic MHD [14]
The distance measures to binary systems described above, when combined with
galaxy surveys and other information, could determine the local value of the Hubble
constant to uncertainties less than 1 % within the first 3 years [13]. When combined
with accurate measurements of cosmic-microwave background measurements (as
expected from the Planck mission), these coincidences might even determine the
dark energy parameter w to accuracies of 1 % (D Holz, private communication).
6.2 Neutron Star Interiors
Neutron stars may be the most complex and extreme physical systems we know of,
and because of that there is still considerable uncertainty about the physics that goes
on inside them. The nuclear physics that underlies their equation of state is not ac-
cessible experimentally, and there are consequently considerable differences among
the various proposed theoretical model equations. Pulsar observations tell us about
strong magnetic fields, but their origin and interior distribution and strength are basi-
cally still a mystery. Therefore any information that gravitational wave observations
can shed on these objects will be welcome.
Modeling binary coalescence will certainly be one way of learning about them,
as mentioned in the previous section. Another avenue will be to detect gravitational
radiation from neutron star pulsation modes that may be excited by outbursts on
their surfaces, which occur when newly accreted matter undergoes a thermonuclear
explosion. The amount of energy that might be converted into gravitational waves is
not known, but searches in coincidence with X-ray and gamma-ray observations are
planned.
6.3 Black Hole Binary Coalescence
Like neutron stars, black holes are expected to be found in binary systems, some
of which will decay and lead to the merger of the black holes. However, we cannot
observe a population of binary black holes so we have less reliable statistics on the
rates we can expect. The best estimates suggest that the observed rate of coalescence
will be similar to that of neutron stars: a sparser population will be detected to greater
distances due to the larger masses [2]. Within the uncertainties it is perfectly possible
that the first detection will be a binary black hole merger.
These observations will enormously increase our understanding of the black
hole population: masses, spins, binary mass ratios. By comparing with numerical-
relativity simulations, which are now very accurate, it may be possible to test general
relativity, particularly cosmic censorship, which is the hypothesis (still unproven) that
a merger will always lead to a black hole and not a naked singularity. The proper-
ties of the final black hole can be inferred from its ringdown pulsation spectrum, if
observed.
Black holes probably also form binaries with neutron stars, although as yet no
pulsar has been detected in orbit around a black hole. Observations of these mergers
with gravitational waves would be very interesting from the point of view of binary
evolution theory and also to constrain the mechanisms for producing gamma-ray
bursts.
6.4 Gravitational Wave Pulsars
Spinning isolated neutron stars will emit gravitational waves if they are asymmetric.
The asymmetries must be small, of course, or known spinning pulsars would by now
have lost their spin to the emission of gravitational waves. This consideration sets an
upper limit on the expected amplitude (the spindown limit). For two pulsars, the Crab
and Vela, current gravitational wave observations have constrained the amplitudes
more tightly than the spindown limit [18, 19]. Advanced detectors may detect such
radiation or constrain many more systems.
The search for such pulsars is very demanding computationally, and blind searches
(for systems not already known as radio pulsars) are only possible with the volunteer-
computing platform Einstein@Home [20]. In fact, the power of this system is already
being used to find weak pulsar signals in radio telescope data [21], as mentioned
earlier.
470 B. F. Schutz
7 Other Gravitational Wave Sources
Supernova explosions motivated the first bar gravitational wave detectors [1]; the
expected kHz frequencies matched what bars could detect, and it seemed reasonable
to think that these explosions were asymmetric enough to produce strong radiation.
Recent studies, however, suggest that the amplitudes to be expected are rather small.
This has to do with the difficulty that stars have in producing supernovae in the
first place: the collapsed interior hangs up at large densities for a relatively long time
before finally acquiring enough energy to explode. During this time, any asymmetries
from the initial collapse are reduced, and the final explosion is dominated by outflow
rather than oscillation, although it is possibly very turbulent (which can lead to
stochastic gravitational wave emission). See, e.g. [22], for a recent study in this
rapidly developing field.
Another signal that is certain to be present, but is expected to be very weak, is
a stochastic gravitational wave background. There are many sources of such back-
grounds, including all the binary systems in the Galaxy. At nanohertz frequencies,
where pulsar timing arrays operate, this is expected to be the dominant signal and to
be detectable within the next 10 years [23]. But in the ground-based frequency band
these signals are likely to be weaker. A big prize would be to detect a background
generated by the Big Bang, but current estimates suggest it is rather weak.
8 Detecting Gravitational Waves from Space
8.1 LISA and eLISA
So far this review has focussed on the imminent detection of gravitational waves
by ground-based interferometers. But many of the people involved in the ground-
based activity are also helping to develop the project to put a detector into space.
The proposal has long been known as LISA, although events have recently led to
a descoped version called eLISA, described below. I shall use the term LISA as a
generic descriptor of a concept that is at least the long-term science goal of the field,
of which eLISA will be perhaps its first realization.
Gravitational waves at frequencies below about 1 Hz will be very difficult or
impossible to detect from the Earth because the Earth’s gravity is noisy at these fre-
quencies, and no detector could be made to respond to cosmic gravitational waves
and not also to terrestrial Newtonian gravity fluctuations. Seismic density distur-
bances, air pressure/density changes, and a host of other phenomena with timescales
between 1 s and 1 h would produce responses in detectors that are larger than those
expected from cosmic sources. The only way to observe in this frequency band is to
get away from the Earth, into space. The LISA proposal has been developed since
1995 within ESA, and then since 1998 jointly with NASA. Along with this proposal,
ESA has been developing the LISA Pathfinder (LPF) mission, whose purpose is to
demonstrate and prove the novel aspects of the LISA measurement technology that
cannot be tested on the ground. LPF must be flown before the final go-ahead can be
given to build a LISA mission. NASA is a minor partner in LPF.
Unfortunately, in 2011 NASA, beset with cost overruns on other missions, with-
drew from its partnership with ESA on LISA, and also on two other proposed large
missions. ESA then asked the three proposals to descope and enter a competition for
an ESA-only large mission launch around 2021. This competition was won in mid-
2012 by JUICE, a mission to the moons of Jupiter. Interestingly, eLISA/NGO, the
descoped gravitational wave mission, seems to have received the highest scientific
rating by the ESA committees, but was nevertheless not selected, primarily because
LPF has not yet launched. Since LISA Pathfinder is expected to be launched by 2015,
it is not unreasonable to think that the chances of success for eLISA in the selection
for the next large mission will be very good indeed. The eLISA team is therefore
working hard to understand better the science capabilities of the instrument and at
the same time to explore the possibility of a partnership with one or more other
agencies (including NASA) that might bring in enough resources to restore some
of the features of the original LISA proposal. The science document of the 2012
eLISA/NGO proposal is on the arXiv [24] and a summary appeared as a conference
proceeding [25].
The LISA concept is to put a long-baseline interferometer into space, with arm-
lengths not much shorter than the wavelengths of the gravitational waves being
detected. This requires three spacecraft in an L-shaped configuration with separations
of order 106 km. In the original LISA concept these spacecraft form an equilateral
triangle, and laser light is used along all three arms to monitor the small changes
in light travel-time along the arms produced by gravitational waves, as illustrated
in Fig. 3. In this way one can construct three different interferometers by using, in
turn, each spacecraft as the central hub of a two-arm interferometer. These three
interferometers measure different polarizations of the incoming radiation. If general
relativity is correct then the output signals will be linearly dependent; checking this
is a good test of the general relativity model for gravitational waves.
If the array is placed with its center on the Earth’s orbit, some 20⊗ or so away
from the Earth, and if the plane of the array is tilted by 60⊗ to the plane of the
Earth’s orbit, then when the spacecraft are given suitable initial velocities they will
remain in an equilateral triangle, which rotates counter to the direction of the orbit,
without any need to maintain the positions of the spacecraft: they simply follow their
Newtonian orbits around the Sun. The spacecraft do need to carry thrusters, but only
to resist external perturbations, such as fluctuations in solar radiation pressure. The
ends of the interferometers’ arms are defined by small cubes (called test masses)
that fall freely in cavities inside the spacecraft (not attached to the spacecraft). The
spacecraft effectively act as shields protecting the cubes from external perturbations.
In this way the cubes follow geodesics of the gravitational field closely enough to
track the disturbances produced by gravitational waves.
The LPF mission consists of a single spacecraft within which are two experiment
packages like those that mark the ends of the LISA arms. Therefore, LPF contains
a single LISA arm, only very short: half a meter instead of several million meters.
472 B. F. Schutz
Fig. 3 A schematic illustration of the LISA concept of three equidistant spacecraft in orbit around
the Sun, forming three interferometers. With an appropriately chosen orbit, the array remains an
equilateral triangle even as the spacecraft orbit the Sun freely
This means that LPF can prove the measurement technology of LISA (by measuring
changes δL in the proof-mass separation to the accuracy needed by LISA) but cannot
measure gravitational waves because the baseline L is so short that the smallest
strain that can be measured, δL/L, is far larger than any expected from astronomical
systems.
The eLISA proposal was descoped to save launch weight and some component
costs. It must still have three spacecraft, but only two of the arms are illuminated
with laser light, so there is only one interferometer signal. Although this reduces the
information available from the mission, eLISA would still do spectacular science. In
the following I describe this science without distinguishing too much between LISA
and eLISA; this is because events in the near future might lead to further changes in
the design of the mission that eventually is flown by ESA.
8.2 LISA Science
By virtue of its long arm lengths, any LISA-like mission will detect some events with
extraordinarily high signal-to-noise ratios, as compared with our expectations for
ground-based observing. Mergers of comparably sized massive black holes (106 M◦ )
can register SNR of up to 1000 in eLISA, and 5 or ten times higher in LISA. EMRI
events, in which a stellar-mass black hole is captured from a nearly circular orbit
by a massive black hole, can typically have SNR around 50–100. With such strong
signals, it is possible to do stringent tests of gravity theory. EMRI events will probe
whether the metric around central black holes in galaxies really is the Kerr metric.
Merger events will give a detailed picture of how horizons merge (which can be
compared with accurate numerical simulations) and will measure the spectrum of
the ring-down radiation emitted as the final black hole settles down. From this we
expect to measure the mass and spin of the hole, and thereby test cosmic censorship:
is it a “clothed” Kerr metric or is it a naked singularity with a > M?
The astrophysics return will be similarly impressive and important. Any LISA-
like mission will be able to detect comparable-mass black hole mergers at redshifts
of 15 and beyond, if there are any that happen in the frequency band of the instrument
(equivalently in the mass range 104 –105 M◦ ). These would be the earliest individual
astronomical systems ever detected and studied by any observing method. They
would tell us much about how galaxies formed, since black holes are believed to
be tracers, embedded in proto-galactic clouds, merging with one another when their
host clouds merge. A three-arm LISA would have enough information to measure
distances to individual events, which when combined with known cosmological
parameters would give the redshifts directly. With a two-arm eLISA mission, distance
measurements have large but finite errors, and when these are averaged over an
ensemble of detected events it will still be possible to discriminate among different
models of early galaxy and black hole evolution [24, 25].
The mergers of relatively nearby pairs of massive black holes (out to z = 2) are
likely to provide a link between mergers and their host galaxies. With positions
provided by the three-arm LISA configuration, it should be possible to identify
the galaxies since their morphology should be disturbed by having experienced the
gradual spiralling together of the holes over hundreds of millions of years. This will
be harder with eLISA data, but with both missions there will also be an intensive
search for time-linked counterparts to the merger: enhanced X-ray, optical, or radio
emission compared to observations of the same area of the sky in previous epochs.
Such studies should provide clues to the mechanisms of jet emission and other
phenomena associated with active galaxies.
The EMRI population will for the first time give us a relatively unbiased sample
of massive black holes in the centres of galaxies. This will determine the mass
distribution of these black holes and also help understand the nature of the star
clusters in their immediate neighbourhoods.
White dwarf binary systems in our Galaxy will actually dominate the LISA data
stream at low frequencies: there are enough systems in tight orbits to provide a
confusion background for LISA below 1 mHz, and a large number of these systems
will be directly detectable by both LISA and eLISA. This will again provide, for
the first time, an unbiased sampling of this important population, which represents
the endpoint of evolution of most binary star systems. The soon-to-be-launched Gaia
[26] mission will identify thousands of such systems, which can then be matched with
their gravitational wave counterparts by identifying systems with the same orbital
periods. Distances provided the gravitational wave measurements (for systems which
change their orbital period because of the emission of gravitational waves) will then
calibrate the absolute brightness of the components.
474 B. F. Schutz
9 Conclusion
The development of detectors has taken place over many decades, essentially because
most of the key technologies had to be developed to the required sensitivity level
within the field itself and could not be taken over from other areas of astronomy
of physics. But within a very few years, ground-based detectors should be making
observations with regularity, and as the network grows there could be events every
couple of days. Space-based detectors will take longer to realise, again because the
technology has to be developed and proved to the required sensitivity and degree of
reliability demanded by space missions.
Driving these proposals is the completely new science that detectors can return,
the new way of listening to the universe that provides unique information essentially
orthogonal to that provided by telescopes. We will see these science returns affecting
astronomy very soon, in the areas of neutron star physics, gamma-ray bursts, stellar
evolution, black hole studies, and the search for gravitational wave pulsars. The
science that can be delivered by a LISA-like space-based detector is recognised as
having a high priority, and I am confident that we will see the launch of such a
detector approved within the next few years, allowing us finally to listen to the the
low-frequency whispers of our universe.
References
1. Weber, J.: Gravitational radiation. Phys. Rev. Lett. 18, 498 (1967)
2. Abadie, J., Belcznski, K., et al. (LIGO Scientific Collaboration, V. Collaboration): Predic-
tions for the rates of compact binary coalescences observable by ground-based gravitational-
wave detectore. Class. Quantum Gravity 27(17), 173001 (2010). doi:10.1088/0264-9381/27/
17/173001
3. Einstein, A.: Über den Einfluß der Schwerkraft auf die Ausbreitung des Lichtes. Annalen der
Physik 35, 898 (1911)
4. LSC Publications. http://www.lsc-group.phys.uwm.edu/ppcomm/Papers.html
5. Punturo, M., Abernathy, M., Acernese, F., et al.: The third generation of gravitational wave
observatories and their science reach. Class. Quantum Gravity 27(8), 084007 (2010)
6. Sathyaprakash, B., Abernathy, M., Acernese, F., et al.: Scientific objectives of Einstein tele-
scope. Class. Quantum Gravity 29(12), 124013 (2012). doi:10.1088/0264-9381/29/12/124013
7. Schutz, B.F.: Networks of gravitational wave detectors and three figures of merit. Class. Quan-
tum Gravity 28(12), 125023 (2011). doi:10.1088/0264-9381/28/12/125023
8. Collins, H.: Gravity’s Shadow: The Search for Gravitational Waves. University of Chicago
Press, Chicago (2004)
9. Ajith, P., Boyle, M., Brown, D.A., et al.: The NINJA-2 catalog of hybrid post-
Newtonian/numerical-relativity waveforms for non-precessing black-hole binaries. Classical
and Quantum Gravity 29(12), 124001 (2012). doi:10.1088/0264-9381/29/12/124001
10. Home Page. https://www.ninja-project.org/doku.php?id=nrar:home
11. Einstein@Home Project Home Page. http://einstein.phys.uwm.edu/
12. Schutz, B.F.: Determining the Hubble constant from gravitational wave observations. Nature
323, 310 (1986)
13. Del Pozzo, W.: Inference of cosmological parameters from gravitational waves: applications to
second generation interferometers. Phys. Rev. D 86(4), 043011 (2012). doi:10.1103/PhysRevD.
86.043011
14. Rezzolla, L., Giacomazzo, B., Baiotti, L., et al.: The missing link: merging neutron stars
naturally produce jet-like structures and can power short gamma-ray bursts. Astrophys. J. Lett.
732, L6 (2011). doi:10.1088/2041-8205/732/1/L6
15. Nissanke, S., Holz, D.E., Hughes, S.A., Dalal, N., Sievers, J.L.: Exploring short gamma-ray
bursts as gravitational-wave standard sirens. Astrophys. J. 725, 496 (2010). doi:10.1088/0004-
637X/725/1/496
16. Will, C.M.: Bounding the mass of the graviton using gravitional-wave observations of inspi-
ralling compact binaries. Phys. Rev. D 57, 2061 (1998)
17. Alexander, S., Yunes, N.: Chern-Simons modified general relativity. Phys. Rep. 480, 1 (2009).
doi:10.1016/j.physrep.2009.07.002
18. Abbott, B.: Beating the spin-down limit on gravitational wave emission from the Crab pulsar.
Astrophys. J. Lett. 683, L45 (2008). doi:10.1086/591526
19. Abadie, J., Abbott, B.P., Abbott, R., et al.: Beating the spin-down limit on gravitational
wave emission from the Vela pulsar. Astrophys. J. 737, 93 (2011). doi:10.1088/0004-637X/
737/2/93
20. Aasi, J., Abadie, J., Abbott, B.P., et al.: Einstein@Home all-sky search for periodic gravitational
waves in LIGO S5 data, ArXiv e-prints arXiv:1207.7176 [gr-qc] (2012)
21. Knispel, B., Lazarus, P., Allen, B., et al.: Arecibo PALFA survey and Einstein@Home: binary
pulsar discovery by volunteer computing. Astrophys. J. Lett. 732, L1 (2011). doi:10.1088/
2041-8205/732/1/L1
22. Ott, C.D., Abdikamalov, E., Moesta, P., et al.: General-relativistic simulations of three-
dimensional core-collapse supernovae. ArXiv e-prints arXiv:1210.6674 [astro-ph.HE] (2012)
23. Sesana, A., Vecchio, A.: Gravitational waves and pulsar timing: stochastic background, indi-
vidual sources and parameter estimation. Class. Quant. Grav. 27, 084016 (2010). doi:10.1088/
0264-9381/27/8/084016
24. Amaro-Seoane, P., Aoudia, S., Babak, S., et al.: eLISA: astrophysics and cosmology in the
millihertz regime. ArXiv e-prints arXiv:1201.3621 [astro-ph.CO] (2012)
25. Amaro-Seoane, P., Aoudia, S., Babak, S., et al.: Low-frequency gravitational-wave science
with eLISA/NGO. Class. Quantum Gravity 29(12), 124016 (2012)
26. Turon, C., O’Flaherty, K.S., Perryman, M.A.C.: The three-dimensional universe with Gaia.
ESA Special Publication, SP-576 (2005). http://sci.esa.int/gaia/37100
LISA in 2012 and Beyond: 20 Years After
the First Proposal
Gerhard Heinzel and Karsten Danzmann
Abstract After 20 years of study as a joint ESA-NASA mission, LISA had to

be redesigned as an ESA-only mission in 2011/2012 to meet programmatic and
budgetary constraints of the space agencies. The result is a mission concept called
“eLISA” or “NGO” with two arms instead of three and one million km armlengths
instead of 5, which results in smaller launch mass but still provides revolution-
ary science. Nevertheless, even the reduced science is expected to be revolutionary
for the study of black holes and other astrophysical and cosmological questions.
“eLISA”/“NGO” was not selected in ESA’s call for the first (“L1”) large mission in
the Cosmic Vision program, but is a strong candidate for the L2 call, with possible
international contributions from the US and/or China.
1 Introduction
Gravitational Waves have been predicted in 1916 by Albert Einstein as a consequence

of GR. So far they have not yet been directly observed, despite the enormous efforts
invested in resonant detectors and ground-based interferometers since the 1960s,
culminating in the present LIGO, VIRGO, GEO600 and similar detectors. These
are ultra-sensitive Michelson-type laser interferometers with km-scale armlengths
and sophisticated vibration-isolation systems. Nevertheless, they are insensitive at
frequencies below about 10 Hz because of inevitable disturbances on Earth such as
seismic and gravity gradient noise. A first direct detection with these instruments
is expected in a few years when the next generation with yet another huge step in
sensitivity improvement goes online.
G. Heinzel (B) · K. Danzmann

Max-Planck-Institut für Gravitationsphysik (Albert-Einstein-Institut), Callinstrasse 38,
30167 Hannover, Germany
e-mail: gerhard.heinzel@aei.mpg.de
K. Danzmann
e-mail: karsten.danzmann@aei.mpg.de
478 G. Heinzel and K. Danzmann
The majority of expected sources, however, have frequencies well below 10 Hz

and are therefore only accessible from space. For their detection, space-based laser
interferometers have been studied since the 1980s. LISA was selected as an ESA
cornerstone mission in 1995, and the concept of a collaborative ESA/NASA mission
with 3 spacecraft in heliocentric orbits trailing the Earth and inclined by 60◦ against
the ecliptic first appeared in 1997. This basic mission concept is basically unchanged
until today.
The mission design was refined in great detail between 2004 and 2010 in an joint
ESA/NASA Mission Formulation study that included an ESA-sponsored industrial
study by EADS Astrium. Both the feasibility and scientific case were scrutinized in
numerous reviews both in Europe and the US, such as the NRC’s Beyond Einstein
Program Assessment Committee (BEPAC) in 2006–7 and the Astro2010 Astronomy
and Astrophysics Decadal Survey. All these reviews attested LISA a compelling and
convincing science case as well as technical feasibility. Literally thousands of papers
have been published on LISA sources, data analysis and instrumentation.
2 Planned ESA Cosmic Vision L1 Selection in 2011
In preparation for the call for proposals for the first “Large” mission (L1) in the ESA
Cosmic Vision program, an Assessment Study Report (“Yellow Book”) was prepared
in February 2011 [1] that summarizes the scientific objectives, mission design and
most important literature.
At that time three missions were in competition: LISA, the International X-ray
Observatory (IXO), and the Europa Jupiter System Mission (EJSM-Laplace), all of
which were conceived as ESA-NASA partnerships with about equal contributions.
In March 2011, however, ESA announced that due to budgetary constraints in
NASA related to (among other reasons) the James Webb Space Telescope, none of
these three missions could rely on the required NASA contribution [2]. Thus, the
L1 downselection was postponed by one year, and the three missions were given
the homework to modify their mission concepts such that they fit into an ESA-only
envelope.
The LISA team studied different options, supported by industry and a mission
concept study in ESA’s concurrent design facility (CDF) in June/July 2011, and
concluded that the “eLISA”/“NGO” concept described below would be the optimal
mission within the given constraints. A new version of the “Yellow Book” was
prepared and published in January 2012 [3].
3 The eLISA/NGO Mission Concept
Since the name LISA was considered to refer to the specific ESA/NASA mission
described in [1], with 6 arms of 5 million km length etc., it was required to find a
new name for the rescoped mission concept. The two names eLISA (“evolved LISA”)
LISA in 2012 and Beyond: 20 Years After the First Proposal 479
and NGO (“New Gravitational wave Observatory”) were chosen and used
interchangeably for this purpose. See, however, Sect. 6 below regarding current nam-
ing conventions. The given constraints were a total cost for ESA of 850 Me, and
member state contributions of about 200 Me. It quickly turned out that the most
efficient way to reduce the predicted cost while maintaining as much of the science
as possible is to reduce the launch mass and volume and the mission lifetime.
This was achieved by
• shrinking the telescope from 38 cm diameter to 20 cm, with subsequent reductions
also in length and height of the payload,
• shrinking the armlength from 5 to 1 million km, which requires less fuel for plane-
change maneuvers to leave the ecliptic plane and helps to restore the reduced light
power levels due to the smaller telescope,
• omitting the last maneuver of the cruise phase, resulting in a “drift-away” orbit
with ever increasing distance to the Earth,
• omitting the third arm, resulting in a ‘V’ shaped configuration instead of a full
triangle.
Especially the omission of the third arm obviously leads to further cost reductions
beyond the reduced launch mass, since less payload hardware is required. Other
factors that contributed to the cost saving are:
• reducing the required mission lifetime from 5 to 2 years, with associated savings
in on-ground testing and operations cost,
• providing the science instrument by an ESA memberstate consortium.
Apart from the above changes, the NGO payload is basically unchanged from LISA,
and would be mounted into 3 identical spacecraft buses based on the LISA Pathfinder
design, one ‘mother’ at the vertex of the ‘V’ and two simpler ‘daughter’ spacecrafts
at the ends. The two lighter daughter would together fit into one Soyuz launcher,
while a second launcher is used for the mother spacecraft.
These reductions come, of course, at a cost in instrument performance. Figure 1
shows the strain sensitivity in comparison.
A significant qualitative reduction in science output results from the omission of
the third arm: The NGO standard Michelson interferometer cannot instantaneously
disentangle two possible mechanisms for a reduction in signal amplitude: larger dis-
tance to the source or polarization mismatch to the antenna orientation. The LISA
triangle with 6 links, however, can simultaneously measure both polarizations and
thereby provide absolute distance measurements to sources whose brightness is
known, e.g., through the time evolution of the waveform. This shortcoming of NGO
is somewhat mitigated, however, for sources that are observed over a significant
fraction of a year, since the rotation of the constellation restores some of the polar-
ization sensitivity. NGO would observe thousands of Compact White Dwarf Binaries
(CWDB), hundreds of black hole binary inspirals and dozens of Extreme Mass Ratio
Inspirals (EMRI).
480 G. Heinzel and K. Danzmann
Fig. 1 Strain sensitivity of LISA (left) and NGO (right) in comparison (source [1, 3])
4 The eLISA Consortium
A significant programmatic change between LISA and NGO was the definition of an
‘instrument’ that would be delivered by the member states. That instrument consists
of most of the scientific payload, with the exception of the lasers and telescopes.
Hence, the instrument contains:
• the test masses in their Gravitational Reference Sensors, including front-end elec-
tronics and discharging equipment,
• the optical benches, including mechanisms and photoreceivers,
• the phasemeters and
• integration and testing tasks.
An important difference to the LISA planning is the phasemeter, which used to be a
NASA contribution, but for eLISA/NGO needs to come from Europe. An ESA tech-
nology development contract granted to a Danish/German team will be completed in
early 2013 and will produce a European phasemeter with all required functionality
and performance. A consortium consisting of institutes and space agencies in Den-
mark, France, Germany, Italy, Spain, Switzerland and UK was formed and committed
to develop and deliver the instrument.
5 The ESA L1 Decision and Its Aftermath
In early May 2012, the ESA Science Programme Committee (SPC) chose the Jupiter
moon mission JUICE (evolved from EJSM-Laplace) as L1 mission to be launched
in 2022. Reasons for not selecting eLISA/NGO in spite of its repeatedly praised
scientific value and technological readiness, as judged by ESA itself, were not clear;
possibly having to do with remaining concerns about the risk of such a revolutionary
new instrument and the desire to wait for the results of LISA Pathfinder, now planned
to be launched in 2015. At the same time ESA announced to continue the technology
development for eLISA/NGO.
LISA in 2012 and Beyond: 20 Years After the First Proposal 481
The 9th LISA Symposium was held in Paris from May 21–25, 2012, with wide
participation of scientists from Europe, the US and China. The community resolved
to push forward with the LISA concept and submit an unbeatable entry for the L2 call
for proposals in ESA’s Cosmic Vision program, expected to occur in 2013 or 2014.
The ‘instrument’ definition and consortium were maintained and are in function
now. The first meeting of the new eLISA consortium took place October 22–23,
2012, also in Paris, and it was decided to form several working groups for science
and technology. Both the consortium and ESA agree that third-party contributions
of up to 20 % of the mission cost are welcome, e.g. from the US or China, provided
they are non-essential, i.e. European alternative suppliers are available in case of
necessity.
6 Naming the Mission Concepts
At the 9th LISA Symposium in Paris it was decided to retain the name LISA for the
general concept of a gravitational wave detector in space, consisting of a triangle
in inclined heliocentric orbits, with drag-free operation, armlengths of some million
km and heterodyne laser interferometers along the arms. “eLISA”/“NGO” is one
specific incarnation of that concept under study in 2012.
7 Conclusion
Despite the drawbacks in 2011 and 2012, the scientific interest in LISA is stronger
than ever before, the technology is well developed, the team is strong and convinced
that LISA must fly in the early 2020s, and is committed to work hard to make that
happen.
References
1. LISA: Unveiling a hidden universe. Technical report ESA/SRE(2011) 3, Noordwijk, Paris

(2011). http://sci.esa.int/lisa/48364
2. Reich, E.S.: Europe makes do without NASA. Nature 471, 421 (2011). doi:10.1038/471421a
3. NGO: Revealing a hidden universe: opening a new chapter of discovery. Technical report
ESA/SRE(2011) 19, Noordwijk, Paris (2011). http://sci.esa.int/ngo/49839
Einstein’s Gravity as Seen by a Cosmic
Lighthouse Keeper
Michael Kramer
Abstract The last years have seen continuing activities in the exploration of our
understanding of gravity, motivated by results from precision cosmology and new
precision astrophysical experiments. At the centre of attention lies the question as to
whether general relativity is the correct theory of gravity. In answering this question,
we work not only towards correctly interpreting the phenomenon of “dark energy”
but also towards the goal of achieving a quantum theory of gravity. In these efforts,
the observations of pulsars, especially those in binary systems, play an important
role. Pulsars do not only provide the only evidence for the existence of gravitational
waves so far, but they also provide precision tests of general relativity and alternative
theories of gravity. This talk summarizes the current state-of-art in these experiments
and looks into the future.
1 Introduction
This conference celebrated Einstein’s time in Prague 100 years ago. As detailed
during the conference and in these proceedings, important groundwork for the later
theory of general relativity (GR) was achieved by Einstein during this time. Now, at
the time of writing, we are less than 3 years away from celebrating the centenary of
Einstein’s greatest achievement. And yet, we have also seen a lot of presentations
at this conference that addressed the properties and experimental consequences of
alternative theories of gravity. As a matter of fact, nearly a 100 years later, efforts in
testing GR and its concepts are still being made by many colleagues around the world,
using many different approaches. To date GR has passed all these experimental and
M. Kramer (B)
Max-Planck-Institut für Radioastronomie, Auf dem Hügel 69, 53121 Bonn, Germany
e-mail: mkramer@mpifr.de
M. Kramer
Jodrell Bank Centre for Astrophysics, University of Manchester, Oxford Road,
Manchester M13 9PL, UK
484 M. Kramer
observational tests with flying colours, but in light of recent progress in observational
cosmology in particular, the question of as to whether alternative theories of gravity
need to be considered is as topical as ever.
Many experiments are designed to achieve ever more stringent tests by either
increasing the precision of the tests or by testing different, new aspects. Some of the
most stringent tests are obtained by satellite experiments in the solar system, provid-
ing exciting limits on the validity of GR and alternative theories of gravity. However,
solar-system experiments are made in the gravitational weak-field regime, while de-
viations from GR may appear only near the Planck scale or in strong gravitational
fields.
We are all very much looking forward to the first direct detection of gravitational
waves with ground-based (and hopefully, eventually, space-based) detectors, which
will not only open a completely new window to the Universe but which will also
provide superb tests of GR under strong-field conditions. Meanwhile, it happens that
nature provides us with an almost perfect laboratory to test the strong-field regime—
in the form of binary radio pulsars.
1.1 Pulsars
Pulsars are rotating neutron stars that emit a radio beam that is eventually powered by
the pulsars’ rotational energy and that is centred on the magnetic axis of the neutron
star. As the magnetic axis and hence the beam are inclined to the rotation axis, the
pulsar acts as a cosmic lighthouse, and a pulsar appears as a pulsating radio source.
The moment of inertia and the stored rotational energy of pulsars are large, so that
in particular the fast rotating millisecond pulsars deliver a radio “tick” per rotation
with a precision that rivals the best atomic clocks on Earth. Corresponding pulse
(or spin) periods range from 1.4 ms to 8.5 s. As they concentrate an average of 1.4
solar masses on a diameter of only about 20 km, pulsars are exceedingly dense and
compact, representing the densest matter in the observable universe. The resulting
gravitational field near the surface is large, enabling strong-field tests of gravity.
For these tests, it is irrelevant how this emission is created, as long the lighthouse
effects sends us a regular beacon. That is useful, because after more than 40 years of
pulsar research, the details of the actual emission process still elude us. However, we
have a basic understanding that is sufficient to perform the experiments described
later. In our straw-man model, the high magnetic field of the rotating neutron star
(Bsurf ≈ 108 to 1014 Gauss) induces a huge electric quadrupole field and an elec-
tromagnetic force that exceeds gravity by 10–12 orders of magnitudes. Charges are
pulled out easily from the surface, and the result is a dense, magnetized plasma that
surrounds the pulsar. The strong magnetic field forces the plasma to co-rotate with
the pulsar like a rigid body. This co-rotating magnetosphere can only extend up to
a distance where the co-rotation velocity reaches the speed of light.1 This distance
defines the so-called light cylinder which separates the magnetic field lines into two
1 Strictly speaking, the Alfvén velocity will determine the co-rotational properties of the
magnetosphere.
Einstein’s Gravity as Seen by a Cosmic Lighthouse Keeper 485
distinct groups, i.e. open and closed field lines. Closed field lines are those which
close within the light cylinder, while open field lines would close outside. The plasma
on the closed field lines is trapped and will co-rotate with the pulsar forever. In con-
trast, plasma on the open field lines can reach highly relativistic velocities and can
leave the magnetosphere, creating the observed well-confined radio beam at a dis-
tance of a few tens to hundreds of km above the pulsar surface. It is this beam which
creates the pulsating signals once per rotation when it points towards Earth.
1.2 Pulsars and Their Companions
The idea behind the usage of pulsars for testing GR and alternative theories of gravity
is straightforward: if the pulsar is in orbit with a binary companion, we use the
measured variation in the arrival times of the received signal to determine and trace
the orbit of the pulsar about the common centre of mass as it moves in the curved
space-time of the companion.
This “pulsar timing” experiment is simultaneously clean, conceptually simple
and very precise. The latter is true since when measuring the exact arrival time of
pulses at our telescope on Earth, we do a ranging experiment that is vastly superior
in precision than a simple measurement of Doppler-shifts in the pulse period. This
is possible, since the pulsed nature of our signal links tightly and directly to the
rotation of the neutron star, allowing us to count every single rotation of a neutron
star. Furthermore, in this experiment we can consider the pulsar as a test mass that
has a precision clock attached to it.
While, strictly speaking, binary pulsars move in the weak gravitational field of a
companion, they do provide precision tests of the strong-field regime. This becomes
clear when considering that the majority of alternative theories predicts strong self-
field effects which would clearly affect the pulsars’ orbital motion. Hence, tracing
their fall in a gravitational potential, we can search for tiny deviations from GR,
providing us with unique precision strong-field tests of gravity.
As a result, a wide range of relativistic effects can be observed, identified and
studied. These include so far:
• Precession of periastron,
• Gravitational redshift,
• Shapiro delay due to curved space-time,
• Gravitational wave emission,
• Geodetic precession, relativistic spin-orbit coupling,
• Speed of gravity.
But we can also convert our observations in tests of concepts and principles deeply
embedded in theoretical frameworks, such as
• Strong Equivalence Principle (grav. Stark effect),
• Lorentz invariance of gravitational interaction,
486 M. Kramer
• Non-existence of preferred frames,

• Conservation of total momentum,
• Non-variation of gravitational constant,
which also leads to stringent limits on alternatives theories of gravity, e.g. tensor–
scalar theories, Tensor-Vector-Scalar (TeVeS) theories.
1.3 Nature is Kind...
The various effects or concepts to be tested require sometimes a rather different type
of laboratory. For instance, to test the violation of the Strong-Equivalence-Principle,
one would like to use a binary system that consists of different types of masses (i.e.
with different gravitational self-energy), rather than a system made of very similar
bodies. Fortunately, nature has been kind.
At the moment, we know about 2,000 pulsars, with about 10 % of these in binary
systems. The shortest orbital period is about 90 min while the longest period is 5.3
yr (e.g. [1]). We find different types of components, i.e. main-sequence stars, white
dwarfs (WD), neutron stars (NS) and even planets. Unfortunately, despite past and
on-going efforts, we have not yet found a pulsar about a stellar black hole companion
or about the supermassive black hole in the centre of our Galaxy [2]. Double neutron
star systems are rare but usually produce the largest observable relativistic effects in
their orbital motion and, as we will see, produce the best tests of general relativity
for strongly self-gravitating bodies. In comparison, pulsar—white dwarf systems
are much more common. Indeed, most pulsar companions are white dwarfs, with a
wide range of orbital periods, ranging from hours to days, weeks and months. Still,
many of them can be used for tests of gravitational theories where we utilize the fact
that white dwarfs and neutron stars differ very significantly in their structure and,
consequently, self-energies.
1.4 Precision Experiments
Using pulsar timing techniques, we make extremely precise measurements that al-
low us to probe gravitation with exquisite accuracy. Table 1 gives an idea about the
precision that we already achieve today. As discussed later in this contribution, with
future telescopes like the “Square Kilometre Array”, the precision will even be en-
hanced by at least two orders of magnitudes and should, for instance, allow us to
find a pulsar orbiting SGR A*, which would provide the mass of the central black
hole to a precision of an amazing 1M⊗ ! It would also allow us to measure the spin
of the black hole with a precision of 10−4 to 10−3 (enabling tests of the “cosmic
censorship conjecture”) and the quadrupole moment with a precision to 10−4 to 10−3
(thus enabling tests of the “no-hair theorem”). See later and [2] for more details.
Table 1 Examples of precision measurements using pulsar timing as a variation demonstration

what is possible today
Masses
Masses of neutron stars m 1 = 1.4398(2) M⊗ [3]
m 2 = 1.3886(2) M⊗ [3]
Mass of WD companion 0.207(2) M⊗ [4]
Mass of millisecond pulsar 1.67(2) M⊗ [5]
Main sequence star companion 1.029(8) M⊗ [5]
Mass of Jupiter and moons 9.547921(2) × 10−4 M⊗ [6]
Spin parameters
Period 5.757451924362137(2) ms [7]
Orbital parameters
Period 0.102251562479(8) day [8]
Eccentricity 3.5(1.1) × 10−7 [9]
Astrometry
Distance 157(1) pc [7]
Proper motion 140.915(1) mas yr−1 [7]
Tests of general relativity
Periastron advance 4.226598(5) deg yr−1 [3]
Shrinkage due to GW emission 7.152(8) mm/day [8]
GR validity (obs/exp) 1.0000(5) [8]
Constancy of grav. constant, Ġ/G −0.6(1.6) × 10−12 yr−1 [9]
The digit in bracket indicates the uncertainty in the last digit of each value. References are cited
2 The Hulse-Taylor Pulsar: Gravitational Wave Damping
The first binary pulsar to ever be discovered happened to be a rare double neutron star
system. It was discovered by Hulse and Taylor in 1974 [10]. The pulsar, B1913+16,
has a period of 59 ms and is in eccentric (e = 0.61) orbit around a unseen companion
with an orbital period of less than 8 h. It became soon clear that the pulsar does not
follow the movement expected from a simple Keplerian description of the binary
orbit, but that it shows the impact of relativistic effects. In order to describe the
relativistic effects in a theory-independent fashion, one introduces so-called “Post-
Keplerian” (PK) parameters that are included in a timing model to accurately describe
the measured pulse times-of-arrival (see e.g. [1] for more details).
For the Hulse-Taylor pulsar, it was soon measured that the system showed a
relativistic advance of its periastron, comparable to what is seen in the solar system
for Mercury, albeit with a much larger amplitude of ω̇ = 4.226598 ± 0.000005 deg/yr
[3]. GR predicts a value for the periastron advance that depends on the Keplerian
parameters and the masses of the pulsar and its companion:
−5/3
2/3 Pb 1
ω̇ = 3T⊗ (m p + m c )2/3 . (1)
2π 1 − e2
488 M. Kramer
Here, T⊗ is a constant, Pb the orbital period, e the eccentricity, and m p and m c the
masses of the pulsar and its companion. See [1] for further details.
The Hulse-Taylor pulsar also shows the effects of gravitational redshift (including
a contribution from a second-order Doppler effect) as the pulsar moves in its elliptical
orbit at varying distances from the companion and with varying speeds. The result
is a variation in the clock rate with an amplitude of γ = 4.2992 ± 0.0008 ms [3]. In
GR, the observed value is related to the Keplerian parameters and the masses as
1/3
2/3 Pb m c (m p + 2m c )
γ = T⊗ e . (2)
2π (m p + m c )4/3
We can now combine these measurements. We have two equations with a measured
left-hand side. On the right-hand side, we measured everything apart from two un-
known masses. We solve for those and obtain, m p = 1.4398 ± 0.0002 M⊗ and
m c = 1.3886 ± 0.0002 M⊗ [3]. These masses are correct if GR is the right theory of
gravity. If that is indeed the case, we can make use of the fact that (for point masses
with negligible spin contributions), the PK parameters in each theory should only be
functions of the a priori unknown masses of pulsar and companion, m p and m c , and
the easily measurable Keplerian parameters [11].2 With the two masses now being
determined using GR, we can compare any observed value of a third PK parameter
with the predicted value. A third such parameter is the observed decay of the orbit
which can be explained fully by the emission of gravitational waves. And indeed,
using the derived masses, and the prediction of general relativity, i.e.
−5/3
192π 5/3 Pb 1 + 24 e + 37
73 2
96 e
4
mpmc
Ṗb = − T⊗ , (3)
5 2π (1 − e )
2 7/2 (m p + m c )1/3
one finds an agreement with the observed value of Ṗbobs = (−2.423 ± 0.001)×10−12
[3]—however, only if a correction for a relative acceleration between the pulsar and
the solar system barycentre is taken into account. As the pulsar is located about 7 kpc
away from Earth, it experiences a different acceleration in the Galactic gravitational
potential than the solar system (see e.g. [1]). The precision of our knowledge to
correct for this effect eventually limits our ability to compare the GR prediction
to the observed value. Nevertheless, the agreement of observations and prediction,
today within a 0.2 % (systematic) uncertainty [3], represented the first evidence for
the existence of gravitational waves. Today we know many more binary pulsars where
we can detect gravitational wave emission. In one particular case, the measurement
uncertainties are not only more precise, but also the systematic uncertainties are
much smaller, as the system is much more nearby. This system is the Double Pulsar.
2 For alternative theories of gravity this statement may only be true for a given equation of state.
3 The Double Pulsar
The Double Pulsar was discovered in 2003 [12, 13]. It does not only show larger
relativistic effects and is much closer to Earth (about 1 kpc) than the Hulse-Taylor
pulsar, allowing us to largely neglect the relative acceleration effects, but the defining
unique property of the system is that it does not consist of one active pulsar and its
unseen companion, but that it harbours two active radio pulsars.
One pulsar is mildly recycled with a period of 22 ms (named “A”), while the other
pulsar is young with a period of 2.8 s (named “B”). Both orbit the common centre
of mass in only 147 min with orbital velocities of 1 Million km per hour. Being also
mildly eccentric (e = 0.09), the system is an ideal laboratory to study gravitational
physics and fundamental physics in general. A detailed account of the exploitation
for gravitational physics has been given, for instance, by [14–16]. An update on those
results is in preparation [8], with the largest improvement undoubtedly given by a
large increase in precision when measuring the orbital decay. Not even 10 years after
the discovery of the system, the Double Pulsar provides the best test for the accuracy
of the gravitational quadrupole emission prediction by GR far below the 0.1 % level.
In order to perform this test, we first determine the mass ratio of pulsar A and B
from their relative sizes of the orbit, i.e. R = x B /x A = m A /m B = 1.0714 ± 0.0011
[14]. Note that this value is theory-independent to the 1PN level [17]. The most
precise PK parameter that can be measured is a large orbital precession, i.e. ω̇ =
16.8991 ± 0.0001 deg/yr. Using Eq. (1), this measured value and the mass ratio,
we can determine the masses of the pulsars, assuming GR is correct, to be m A =
(1.3381 ± 0.0007) M⊗ and m B = (1.2489 ± 0.0007) M⊗ .
We can use these masses to compute the expected amplitude for the gravitational
redshift, γ , if GR is correct. Comparing the result with the observed value of γ =
383.9 ± 0.6 μs, we find that theory (GR) agrees with the observed value to a ratio
of 1.000 ± 0.002, as a first of five tests of GR in the Double Pulsar.
The Double Pulsar also has the interesting feature that the orbit is seen nearly
exactly edge-on. This leads to a 30 s long eclipse of pulsar A due to the blocking
magnetosphere of B that we discuss further below, but it also leads to a “Shapiro
delay”: whenever the pulse needs to propagate through curved space-time, it takes
a little longer than travelling through flat space-time. At superior conjunction, when
the signal of pulsar A passes the surface of B in only 20,000 km distance, the extra
path length due to the curvature of space-time around B leads to an extra time delay
of about 100 μs. The shape and amplitude of the corresponding Shapiro delay curve
yield two PK parameters, s and r , known as shape and range, allowing two further
tests of GR. s is measured to s = sin(i) = 0.99975 ± 0.00009 and is in agreement
with the GR prediction of
−2/3
−1/3 Pb (m A + m B )2/3
s= T⊗ x , (4)
2π mB
490 M. Kramer
(where x is the projected size of the semi-major axis measured in lt-s) within a ratio
of 1.0000 ± 0.0005. It corresponds to an orbital inclination angle of 88.7 ± 0.2◦ ,
which is indeed very close to 90◦ as suggested by the eclipses. r can be measured
with much less precision and yields an agreement with GR’s value given by
r = T⊗ m B , (5)
to within a factor of 0.98 ± 0.02.

A fourth test is given by comparing an observed orbital decay of 107.79 ±
0.11 ns/day to the GR prediction. Unlike the Hulse-Taylor pulsar, extrinsic effects are
negligible and the values agree with each other without correction to within a ratio
of 1.000 ± 0.001. This is already a better test for the existence of GW than possible
with the Hulse-Taylor pulsar and will continue to improve with time. Indeed, at the
time of writing the agreement has already surpassed the 0.1 % level significantly [8].
4 Relativistic Spin-Orbit Coupling
Apart from the Shapiro delay, the impact of curved space time is also immediately
measurable by its effect on the orientation of the pulsar spin in a gyroscope experi-
ment. This effect, known as geodetic precession or de Sitter precession, represents
the effect on a vector carried along with an orbiting body such that the vector points
in a different direction from its starting point (relative to a distant observer) after
a full orbit around the central object. Experimental verification has been achieved
by precision tests in the solar system, e.g. by Lunar Laser Ranging (LLR) measure-
ments, or recently by measurements with the Gravity Probe-B satellite mission (see
[18] for a review of experimental tests). However, these tests are done in the weak
field conditions of the solar system, so that pulsars provide the only access to the
strong-field regime.
In binary systems one can interpret the observations, depending on the reference
frame, as a mixture of different contributions to relativistic spin-orbit interaction. One
contribution comes from the motion of the first body around the centre of mass of
the system (de Sitter-Fokker precession), while the other comes from the dragging of
the inertial frame at the first body due to the translational motion of the companion
[19]. Hence, even though we loosely talk about geodetic precession, the result of
the spin-orbit coupling for binary pulsar is more general, and hence we will call
it relativistic spin-precession. The consequence of relativistic spin precession is a
precession of the pulsar spin about the total angular moment vector, changing the
orientation of the pulsar relative to Earth.
Since the orbital angular momentum is much larger than the angular momentum
of the pulsar, the orbital spin practically represents a fixed direction in space, defined
by the orbital plane of the binary system. Therefore, if the spin vector of the pulsar
is misaligned with the orbital spin, relativistic spin-precession leads to a change in
viewing geometry, as the pulsar spin precesses about the total angular momentum
Table 2 DNSs sorted according to the expected relativistic spin precession rate
PSR P (ms) Pb (d) x (lt-s) e Ωp (deg yr−1 )
J0737 − 3039A/B√ 22.7/2770 0.10 1.42/1.51 0.09 4.8/5.1
J1906 + 0746√ 144.1 0.17 1.42 0.09 2.2
B2127 + 11C√ 30.5 0.34 2.52 0.68 1.9
B1913 + 16√ 59.0 0.33 2.34 0.62 1.2
J1756 − 2251 28.5 0.32 2.76 0.18 0.8
B1534 + 12√ 37.9 0.42 3.73 0.27 0.5
J1829 + 2456 41.0 1.18 7.24 0.14 0.08
J1518 + 4904 40.9 8.64 20.0 0.25 –
J1753 − 2240 95.1 13.63 18.1 0.30 –
J1811 − 1736 104.2 18.8 34.8 0.83 –
J1141 − 6545√ 394.0 0.20 1.89 0.17 1.4
Also included is PSR J1141−6545 which is in a relativistic orbit about a white dwarf companion.
Pulsars marked with an asterisk have been identified of showing spin precession. For sources where
no precession rate is listed, the companion mass could not be accurately measured yet, indicating
however, that the precession rate is low
vector. Consequently, as many of the observed pulsar properties are determined by

the relative orientation of the pulsar axes towards the distant observer on Earth, we
should expect a modulation in the measured pulse profile properties, namely its shape
and polarisation characteristics [20]. The precession rate is another PK parameter
and given in GR by (e.g. [1])
5/3
2/3 2π m c (4m p + 3m c ) 1
Ωp = T⊗ × × × . (6)
Pb 2(m p + m c ) 4/3 1 − e2
In order to see a measurable effect in any binary pulsar, (a) the spin axis of the
pulsar needs to be misaligned with the total angular momentum vector and (b) the
precession rate must be sufficiently large compared to the available observing time
to detect a change in the emission properties. Table 2 lists the known Double Neutron
Star Systems which typically show the largest degree of relativistic effects due to
the often short eccentric binary orbits. However, the last entry in the table is PSR
J1141−6545 which is a relativistic system with a white dwarf companion. Those
pulsars that are marked with an asterisk have been identified as pulsars showing
relativistic spin precession. Note that the top 5 out of 8 sources (with a known
expected precession rate) indeed show the effect.
As the most relativistic binary system known to date, we expect a large amount of
spin precession in the Double Pulsar system. Despite careful studies, profile changes
for A have not been detected, suggesting that A’s misalignment angle is rather small
(e.g. Ferdman et al. in prep.). In contrast, changes in the light curve and pulse shape
on secular timescales [21] reveal that this is not the case for B. In fact, B had been
becoming progressively weaker and disappeared from our view in 2009 [22]. Making
the valid assumption that this disappearance is solely caused by relativistic spin
492 M. Kramer
precession, it will only be out of sight temporarily until it reappears later. Modelling
suggests that, depending on the beam shape, this will occur in about 2035 but an
earlier time cannot be excluded. The geometry that is derived from this modelling
is consistent with the results from complementary observations of spin precession,
visible via a rather unexpected effect described in the following.
The change in the orientation of B also changes the observed eclipse pattern in
the Double Pulsar, where we can see periodic bursts of emission of A during the
dark eclipse phases, with the period being the full- or half-period of B. As this
pattern is caused by the rotation of B’s blocking magnetospheric torus that allows
light to pass B when the torus rotates to be seen from the side, the resulting pattern
is determined by the three-dimensional orientation of the torus, which is centred
on the precessing pulsar spin. Eclipse monitoring over the course of several years
shows exactly the expected changes, allowing to determine the precession rate to
+0.66
Ωp,B = 4.77−0.65 deg/yr. This value is fully consistent with the value expected GR,
providing a fifth test [23]. This measurement also allows to test alternative theories of
gravity and their prediction for relativistic spin-precession in strongly self-gravitating
bodies for the first time (see [16] for details).
5 Alternative Theories
Despite the successes of GR, a range of observational data have fuelled the contin-
uous development of alternative theories of gravity. Such data include the apparent
observation of “dark matter” or the cosmological results interpreted in the form of
“inflation” and “dark energy”. Confronting alternative theories with data also in other
areas of the parameter space (away from the CMB or Galactic scales), requires that
these theories are developed sufficiently in order to make predictions. A particular
sensitive criterion is if the theory is able to make a statement about the existence and
type of gravitational waves. Most theories cannot do this (yet), but a class of theories
where this has been achieved is the class of tensor-scalar theories as discussed and
demonstrated by Damour and Esposito-Farèse in a series of works (e.g. [24]). For
corresponding tests, the choice of a double neutron star system is not ideal, as the
difference in scalar change (that would be relevant, for instance, for the emission
of gravitational dipole radiation) is small. The ideal laboratory would be a pulsar
orbiting a black hole, as the black hole would have zero scalar charge and the differ-
ence would be maximised. The next best laboratory is a pulsar-white dwarf system.
Indeed, such binary systems are able to provide constraints for alternative theories
of gravity that are equally good or even better than solar system limits [9].
A recently studied pulsar-white dwarf system turns out to be very exciting: PSR
J0348+0432 harbours a white dwarf whose composition and orbital motion can be
precisely derived from optical observations. The results allow us to measure the mass
of the neutron star, showing that it has a record-braking value of 2.01±0.04M⊗ ! This
is not only the most massive neutron star known (at least with reliable precision),
but the 39 ms pulsar and the white dwarf orbit each other in only 2.46 h, i.e. the
orbit is only tens of seconds longer than that of the Double Pulsar. Even though the
orbital motion is nearly circular, the effect of gravitational wave damping is clearly
measured. Thereby, the high pulsar mass and the compact orbit make this system a
sensitive laboratory of a previously untested strong-field gravity regime. Thus far, the
observed orbital decay agrees with GR, supporting its validity even for the extreme
conditions present in the system (Antoniadis et al. submitted).
6 Detecting Gravitational Waves with Pulsars
The observed orbital decay in binary pulsars detected via precision timing experi-
ments so far offers the only evidence for the existence of gravitational wave (GW)
emission. Intensive efforts are therefore on-going world-wide to make a direct de-
tection of gravitational waves that pass over the Earth. Ground-based detectors like
GEO600, VIRGO or LIGO use massive mirrors, the relative distance of which are
measured by a laser interferometer set-up, while the future space-based LISA de-
tector uses formation flying of three test-masses that are housed in satellites. The
change of the space-time metric around the Earth also influences the arrival times of
pulsar signals measured at the telescope, so that high-precision MSP timing can also
potentially directly detect GWs. Because pulsar timing requires the observations of a
pulsar for a full Earth orbit before the relative position between pulsar, Solar System
Barycentre and Earth can be precisely determined, only GWs with periods of more
than a year can usually be detected. In order to determine possible uncertainties in the
used atomic clocks, planetary ephemerides used, and also since GWs are expected to
produce a characteristic quadrupole signature on the sky, several pulsars are needed
to make a detection. The sensitivity of such a “Pulsar Timing Array” (PTA) increases
with the number of pulsars and should be able to detect pulsars in the nHz regime,
hence below the frequencies of LIGO (≈kHz and higher) and LISA (≈μHz) (see
Fig. 1).
A number of PTA experiments are ongoing, namely in Australia, Europe and
North America (see [25] for a summary). The currently derived upper limits on a
stochastic GW background (e.g. [26, 27]) are very close to the theoretical expectation
for a signal that originates from binary super-massive black holes expected from
the hierarchical galaxy evolution model [28, 29]. The best limit has recently been
published by the European Pulsar Timing Array (EPTA) that uses the telescopes
at Effelsberg, Jodrell Bank, Nancay and Westerbork [30]. Soon the Sardinia Radio
Telescope will be added to the EPTA once it is completed.
While the limits are close, it seems that “simply a bit of extra sensitivity is needed”
to make a first detection. This is the motivation for the Large European Array for
Pulsars (LEAP) project in Europe. It aims to phase-coherently connect Europe’s
largest radio telescopes to form an Arecibo-sized dish that can observe a large number
of millisecond pulsars with high sensitivity enabling high precision pulsar timing.
LEAP is part of the EPTA and also acts as a test-bed for technology development for
the SKA [27].
494 M. Kramer
-6 Current PTA
0.9K
PULSARS Blackbody
LIGO I Spectrum
LISA Advanced
Local
-8
strings LIGO
LEAP
1st Order Extended
EW Phase Inflation
MBH−MBH Transition
COBE Binaries
-10
LISA LIGO II
/VIRGO
1st Order
COBE EW Phase
Global Transition
-12 strings
SKA−PTA
-14 Slow-roll inflation - Upper Bound

CMB−POL
Inflation
-16 Inflation
-15 -10 -5 0 5 10
Fig. 1 Summary of the potential cosmological sources of a stochastic gravitational wave (GW)
background overlaid with bounds from COBE, current pulsar timing array (PTA) experiments and
the goals of CMB polarization experiments, LISA and Advanced LIGO. LEAP will improve on
the current best PTA limits by more than two orders of magnitude, enabling the detection of a GW
background caused by the merger of massive black holes (MBHs) in early galaxy formation. The
amplitude depends on the MBH mass function and merger rate, so that uncertainty is indicated by
the size of the shaded area. LEAP is the next logical step towards a PTA realized with the SKA
which will improve on the current sensitivity by about four orders of magnitude
Demonstrating the power of PTA experiments, Champion et al. [6] recently used
data of PTA observations to determine the mass of the Jovian system independently
of the spacecraft data obtained by fly-bys. Here, the idea is that an incorrectly known
planet mass will result in an incorrect model of the location of the Solar System
Barycentre relative to the Earth. However, as discussed the SSB is the reference
point for pulsar arrival time measurements, so that a mismatch between assumed
and actual position would lead to a periodic signal in the pulsar data with the period
being that of the planet with the ill-measured mass. This measurement technique is
sensitive to a mass difference of only 0.003 % of the mass of the Earth, and 10−7 -th
of Jupiter’s mass.
If LEAP or other experiments do not detect GWs in the next few years, a first
detection is virtually guaranteed with the more sensitive Phase I of the SKA. But the
science that can eventually be done with the full SKA goes far beyond simple GW
detection—a whole realm of astronomy and fundamental physics studies will become
possible. For instance, it will be possible to study the properties of gravitational
waves, such as their polarisation properties or the mass of the graviton [32, 33]. This
is achieved by measuring the degree of correlation in the arrival time variation of pairs
of pulsars separated by a certain angle on the sky. A positive correlation is expected for
pulsars in the same direction or 180◦ apart on the sky, while pulsars separated by 90◦
should be anti-correlated. The exact shape of this correlation curve obviously depends
on the GW polarisation properties [32] but also on the mass of the graviton [33]. The
latter becomes clear when we consider that a non-zero mass leads to a dispersion
relation and a cut-off frequency ωcut = m g c2 /, below which a propagation is not
possible anymore, affecting the degree of correlation possible between two pulsars.
With a 90 % probability, massless gravitons can be distinguished from gravitons
heavier than 3 × 10−22 eV (Compton wavelength λg = 4.1 × 1012 km), if bi-
weekly observation of 60 pulsars are performed for 5 years with pulsar RMS timing
accuracy of 100 ns. If 60 pulsars are observed for 10 years with the same accuracy,
the detectable graviton mass is reduced to 5 × 10−23 eV (λg = 2.5 × 1013 km) [33].
In addition to detecting a background of GW emission, the probability of detecting
a single GW source increases from a few percent now to well above 95 % with the
full SKA. We can, for instance, expect to find the signal of a single supermassive
black hole binary. Considering the case when the orbit is effectively not evolving
over the observing span, we can show that, by using information provided by the
“pulsar term” (i.e. the retarded effect of the GW acting on the pulsar’s surrounding
spacetime), we can achieve a rather astounding source localization. For a GW with an
amplitude exceeding 10−16 and PTA observations of 40 pulsars with weekly timing
to 30 ns precision, one can measure the GW source position to an accuracy of better
than ≈1 arcmin (Fig. 2, [31]). With such an error circle, an identification of the
GW source in the electromagnetic spectrum should be easily feasible. We note that
in order to achieve such a result, a precise distance measurement to the pulsars is
needed, which in turn can then be improved further during the fitting process that
determines the orbital parameters of the GW source. Fortunately, the SKA will be a
superb telescope to do astrometry with pulsars [34].
7 The Future and the Ultimate Laboratory
Essentially all upcoming and future telescopes will contribute in one way or the other
to advances in the field of pulsar astrophysics. LOFAR will find a large number of
neutron stars that are potential sources for tests of relativity or pulsar timing array
experiments that attempt to directly detect gravitational waves [35]. Even though
these experiments are performed at high frequencies, LOFAR can find appropriate
sources and also monitor the interstellar weather that needs to be corrected for in
high precision timing. The Chinese FAST telescope will have a collecting area that
will allow us to find and time pulsars that will be significantly better than currently
achieved with Arecibo [36]. However, the real big advance for pulsars and their
applications will be achieved with the Square Kilometre Array (SKA).
It is clear that the SKA will have a huge impact on the study of pulsars and their
applications, in particular in using them for our understanding gravity. Some of the
questions directly addressed with SKA pulsar studies are:
496 M. Kramer
σ σ
σ
δω
δ
σ
Fig. 2 Results of computations for the detection of a single GW source with the help of a PTA
consisting of 40 pulsars with an average timing precision of 300 ns at a typical distance of 200 pc
[31]. Top left Error in measuring the characteristic strain amplitude of a single GW source for
a variety of signal strengths. Top right Corresponding error in orbital inclination measurement,
bottom left positional error on the sky and bottom right error in determining the gravitational wave
frequency. See [31] for details
• What is the nature of gravity? Was Einstein right? Is gravity described by a tensor
field or are there additional scalar fields, as it is sometimes proposed to explain
Dark Energy?
• What are the properties of gravitational waves? Do gravitons have spin 2? What
is the mass of gravitons and hence the propagation speed of gravitational waves?
• What happens in strong gravitational fields, in conditions of extreme curvature
and near singularities? What are the properties of Black Holes? Do the no-hair
and cosmic censorships theorems hold?
Answering these questions requires a survey for pulsars and the high-precision timing
of a selected sample of those. In these searches, we can in particular expect to find
the first pulsar-black hole systems.
What makes a binary pulsar with a black-hole companion so interesting is that it
has the potential of providing a superb new probe of relativistic gravity. As pointed
out by Damour and Esposito-Farèse [37], the discriminating power of this probe
might supersede all its present and foreseeable competitors. The reason lies in the
fact that such a system would be very sensitive to strong gravitational self-field
effects, making it for instance an excellent probe for tensor-scalar theories.
But also for testing the black hole properties predicted by GR, a pulsar-BH system
will be superb laboratory. Wex and Kopeikin [38] showed that the measurement
Fig. 3 Timing residuals of a pulsar orbiting the supermassive BH in the Galactic Centre in a 0.1 year
orbit. Shown are three orbits with an eccentricity of e = 0.4 for an extreme Kerr BH. Even with a
timing precision of only 100 μs, the characteristic periodic residuals caused by the BH’s quadrupole
moment are clearly visible [2]
of classical and relativistic spin-orbit coupling in a pulsar-BH binary, in principle,

allows us to determine the spin and the quadrupole moment of the black hole. This
would test the “cosmic censorship conjecture” and the “no-hair theorem”. While [38]
showed that with current telescopes such an experiment would be almost impossible
to perform (with the possible exception of pulsars about the Galactic centre black
hole), Kramer et al. [39] pointed out that the SKA sensitivity should be sufficient.
Indeed, this experiment benefits from the SKA sensitivity in multiple ways. On one
hand, it provides the required timing precision but it also allows to perform the
Galactic Census which should eventually deliver the sample of pulsars with a BH
companion. As shown recently by Liu et al. [2] it should be “fairly easy” to measure
the spin of the GC black hole with a precision of 10−4 to 10−3 . Even for a pulsar
with a timing precision of only 100 μs, characteristic periodic residuals would enable
tests of the no-hair theorem with a precision of one percent or better (see Fig. 3)!
8 Summary and Conclusions

A variety of experiments and observational data exist that allow us to test our un-
derstanding of gravity with increased precision. So far, general relativity has passed
all tests with flying colours but the apparent existence of “Dark Energy” challenges
this simple picture. It is clear that the observations of pulsars will continue to play
an important part in testing general relativity and its alternatives. We expect to detect
498 M. Kramer
gravitational waves not only indirectly but also directly using pulsar observations,
and we have all reasons to believe that future searches will yield pulsars that can
probe the space-time around black holes. Combined with the results of other experi-
ments, namely the detection of gravitational waves with ground based detectors, we
can expect a bright future for our understanding of gravity.
I want to conclude with a quote from Einstein, which he made in a letter to Arnold
Sommerfeld on December 9th, 1915, a few days after he presented his final field
equations. Delighted by the fact that his theory could predict correctly the Mercury
perihelion advance, he wrote:
Wie kommt uns da die pedantische Genauigkeit der Astronomie zu Hilfe, über die ich mich
im Stillen früher oft lustig machte!
Unfortunately, Einstein died more than 10 years before pulsars were discovered.
But he once also said that if he were not a physicist, he would like to be a light-
house keeper. I dare to think that he would also have liked, perhaps, to be a pulsar
astronomer, i.e. a cosmic lighthouse keeper who can enjoy precision astronomy and
the science possible thanks to it.
References
1. Lorimer, D.R., KramerM.: Handbook of Pulsar Astronomy, Cambridge Observing Handbooks

for Research Astronomers, vol. 4, 1st edn. Cambridge University Press, New York (2005)
2. Liu, K., Wex, N., Kramer, M., Cordes, J.M., Lazio, T.J.W.: Prospects for probing the spacetime
of Sgr A* with pulsars. Astrophys. J. 747, 1 (2012). doi:10.1088/0004-637X/747/1/1
3. Weisberg, J.M., Nice, D.J., Taylor, J.H.: Timing measurements of the relativistic binary pulsar
PSR B1913+16. Astrophys. J. 722, 1030 (2010). doi:10.1088/0004-637X/722/2/1030
4. Hotan, A.W., Bailes, M., Ord, S.M.: High-precision baseband timing of 15 ms pulsars. Mon.
Not. R. Astron. Soc. 369, 1502 (2006). doi:10.1111/j.1365-2966.2006.10394.x
5. Freire, P.C.C., Bassa, C.G., Wex, N., et al.: On the nature and evolution of the unique binary
pulsar J1903+0327. Mon. Not. R. Astron. Soc. 412, 2763 (2011). doi:10.1111/j.1365-2966.
2010.18109.x
6. Champion, D.J., Hobbs, G.B., Manchester, R.N., et al.: Measuring the mass of solar system
planets using pulsar timing. Astrophys. J. 720, L201 (2010). doi:10.1088/2041-8205/720/2/
L201
7. Verbiest, J.P.W., Bailes, M., van Straten, W., et al.: Precision timing of PSR J0437–4715:
an accurate pulsar distance, a high pulsar mass, and a limit on the variation of Newton’s
gravitational constant. Astrophys. J. 679, 675 (2008). doi:10.1086/529576
8. Kramer, M., et al.: Tests of gravity from timing the Double Pulsar (2013). In preparation.
9. Freire, P.C.C., Wex, N., Esposito-Farèse, G., et al.: The relativistic pulsar-white dwarf binary
PSR J1738+0333—II. The most stringent test of scalar-tensor gravity. Mon. Not. R. Astron.
Soc. 423, 3328 (2012). doi:10.1111/j.1365-2966.2012.21253.x
10. Hulse, R.A., Taylor, J.H.: Discovery of a pulsar in a binary system. Astrophys. J. 195, L51
(1975). doi:10.1086/181708
11. Damour, T., Taylor, J.H.: Strong-field tests of relativistic gravity and binary pulsars. Phys. Rev.
D 45, 1840 (1992). doi:10.1103/PhysRevD.45.1840
12. Burgay, M., D’Amico, N., Possenti, A., et al.: An increased estimate of the merger rate of
double neutron stars from observations of a highly relativistic system. Nature 426, 531 (2003).
doi:10.1038/nature02124
13. Lyne, A.G., Burgay, M., Kramer, M., et al.: A double-pulsar system: a rare laboratory
for relativistic gravity and plasma physics. Science 303, 1153 (2004). doi:10.1126/science.
1094645
14. Kramer, M., Stairs, I.H., Manchester, R.N., et al.: Tests of general relativity from timing the
double pulsar. Science 314, 97 (2006). doi:10.1126/science.1132305
15. Kramer, M., Stairs, I.H.: The double pulsar. Annu. Rev. Astron. Astrophys. 46, 541 (2008).
doi:10.1146/annurev.astro.46.060407.145247
16. Kramer, M., Wex, N.: The double pulsar system: a unique laboratory for gravity. Class Quantum
Gravity 26, 073001 (2009). doi:10.1088/0264-9381/26/7/073001
17. Damour, T., Deruelle N.: General relativistic celestial mechanics of binary systems. II. The
Post-Newtonian timing formula, Ann. Inst. Henri Poincare (A) 44, 263 (1986)
18. Will, C.M.: The confrontation between general relativity and experiment. Living Rev. Relativity
9, lrr-2006-3 (2006). http://www.livingreviews.org/lrr-2006-3
19. Börner, G., Ehlers, J., Rudolph, E.: Relativistic spin precession in two-body systems. Astron.
Astrophys. 44, 417 (1975)
20. Damour, T., Ruffini, R.: Certain new verifications of general relativity made possible by the
discovery of a pulsar belonging to a binary system. C. R. Acad. Sci. Ser. A 279, 971 (1974)
21. Burgay, M., Possenti, A., Manchester, R.N., et al.: Long-term variations in the pulse emission
from PSR J0737–3039b. Astrophys. J. 624, L113 (2005). doi:10.1086/430668
22. Perera, B.B.P., McLaughlin, M.A., Kramer, M., et al.: The evolution of PSR J0737–3039b and
a model for relativistic spin precession. Astrophys. J. 721, 1193 (2010). doi:10.1088/0004-
637X/721/2/1193
23. Breton, R.P., Kaspi, V.M., Kramer, M., et al.: Relativistic spin precession in the double pulsar.
Science 321, 104 (2008). doi:10.1126/science.1159295
24. Damour, T., Esposito-Farèse, G.: Tensor-scalar gravity and binary-pulsar experiments. Phys.
25. Hobbs, G., Archibald, A., Arzoumanian, Z., et al.: The international pulsar timing array project:
using pulsars as a gravitational wave detector. Class. Quant. Grav. 27, 084013 (2010). doi:10.
1088/0264-9381/27/8/084013
26. Jenet, F.A., Hobbs, G.B., van Straten, W., et al.: Upper bounds on the low-frequency stochas-
tic gravitational wave background from pulsar timing observations: current limits and future
prospects. Astrophys. J. 653, 1571 (2006). doi:10.1086/508702
27. Ferdman, R.D., van Haasteren, R., Bassa, C.G., et al.: The European pulsar timing array: current
efforts and a leap toward the future. Class. Quant. Grav. 27, 084014 (2010). doi:10.1088/0264-
9381/27/8/084014
28. Sesana, A., Vecchio, A., Colacino, C.N.: The stochastic gravitational-wave background from
massive black hole binary systems: implications for observations with pulsar timing arrays.
Mon. Not. R. Astron. Soc. 390, 192 (2008). doi:10.1111/j.1365-2966.2008.13682.x
29. Sesana, A., Vecchio, A.: Gravitational waves and pulsar timing: stochastic background, indi-
vidual sources and parameter estimation. Class. Quant. Grav. 27, 084016 (2010). doi:10.1088/
0264-9381/27/8/084016
30. van Haasteren, R., Levin, Y., Janssen, G.H., et al.: Placing limits on the stochastic gravitational-
wave background using European pulsar timing array data. Mon. Not. R. Astron. Soc. 414,
3117 (2011). doi:10.1111/j.1365-2966.2011.18613.x
31. Lee, K.J., Wex, N., Kramer, M., et al.: Gravitational wave astronomy of single sources with
a pulsar timing array. Mon. Not. R. Astron. Soc. 414, 3251 (2011). doi:10.1111/j.1365-2966.
2011.18622.x
32. Lee, K.J., Jenet, F.A., Price, R.H.: Pulsar timing as a probe of Non-Einsteinian polarizations
of gravitational waves. Astrophys. J. 685, 1304 (2008). doi:10.1086/591080
33. Lee, K., Jenet, F.A., Price, R.H., Wex, N., Kramer, M.: Detecting massive gravitons using
pulsar timing arrays. Astrophys. J. 722, 1589 (2010). doi:10.1088/0004-637X/722/2/1589
34. Smits, R., Tingay, S.J., Wex, N., Kramer, M., Stappers, B.: Prospects for accurate distance
measurements of pulsars with the square kilometre array: enabling fundamental physics.
Astron. Astrophys. 528, A108 (2011). doi:10.1051/0004-6361/201016141
500 M. Kramer
35. Stappers, B.W., Hessels, J.W.T., Alexov, A., et al.: Observing pulsars and fast transients with
LOFAR. Astron. Astrophys. 530, A80 (2011). doi:10.1051/0004-6361/201116681
36. Smits, R., Lorimer, D.R., Kramer, M., et al.: Pulsar science with the five hundred metre aperture
spherical telescope. Astron. Astrophys. 505, 919 (2009). doi:10.1051/0004-6361/200911939
37. Damour, T., Esposito-Farèse, G.: Gravitational-wave versus binary-pulsar tests of strong-field
gravity. Phys. Rev. D 58, 042001 (1998). doi:10.1103/PhysRevD.58.042001
38. Wex, N., Kopeikin, S.: Frame dragging and other precessional effects in black hole-pulsar
binaries. Astrophys. J. 513, 388 (1999). doi:10.1086/306933
39. Kramer, M., Backer, D.C., Cordes, J.M., et al.: Strong-field tests of gravity using pulsars and
black holes. New Astron. Rev. 48, 993 (2004). doi:10.1016/j.newar.2004.09.020
The Astrophysical Signatures of Black Holes:
The Horizon, The ISCO, The Ergosphere
and The Light Circle
Marek A. Abramowicz
Abstract Three advanced instruments planned for a near future (LOFT, GRAVITY,
THE EVENT HORIZON TELESCOPE) provide unprecedented angular and time resolu-
tions, which allow to probe regions in the immediate vicinity of black holes. We
may soon be able to search for the signatures of the super-strong gravity that is
characteristic to black holes: the event horizon, the ergosphere, the innermost stable
circular orbit (ISCO), and the photon circle. This review discusses a few fundamental
problems concerning these theoretical concepts.
1 Introduction
Undoubtedly, the existence of black holes is one of a few most bizarre predictions
ever made in the whole acts of physics. The existence of massive objects with gravity
strong enough to prevent light from escaping their surfaces was anticipated already
in the XVIII century by John Mitchel, and later but independently by Pierre–Simon
Laplace. However, the fundamental properties of black holes have been discov-
ered, understood and described in terms of brilliant mathematical developments of
M. A. Abramowicz (B)
Nicolaus Copernicus Astronomical Center, ul. Bartycka 18, 00-716 Warszawa, Poland
e-mail: marek.abramowicz@physics.gu.se
M. A. Abramowicz
Physics Department, Institute of Astronomy (Prague), Silesian University, Opava,
The Czech Republic
M. A. Abramowicz
Department of Physics, University of Gothenburg, Gothenburg, Sweden
502 M. A. Abramowicz
Albert Einstein’s general theory of relativity, only much later—indeed in our times.1
Astrophysically, the most important of these fundamental properties are2 :
1. Event horizon: This may be imagined as a sphere (but not as a rigid surface)
with the “gravitational radius” r G ≈ G M/c2 surrounding the black hole of the
mass M, from within which nothing may emerge. This is a unique signature of
black holes.
2. Ergosphere: This is a region around a rotating black hole where spacetime itself
is dragged along in the direction of rotation at a speed greater than the local speed
of light in relation to the rest of the universe. In this region, negative energy states
are possible, which means that the rotational energy of the black hole can be
tapped through the “Penrose process”.
3. Innermost stable circular orbit (ISCO): This is the smallest circle (r = rms )
along which free particles may stably orbit around a black hole. No stable circular
motion is possible for r < rms . The presence of ISCO in the black hole case is
one of the most important features of the black hole accretion [4].
4. Circular photon orbit: At a specific radius, often called “the light circle”, photons
may circle freely around a black hole.
In the last three decades, robust detections were made of several astrophysical black
hole candidates within our Galaxy and in many others galaxies. However, no direct
and unambiguous observational signatures of the horizon, the ergosphere, the ISCO
and the light circle have been found. The obvious difficulty here is the high angular
resolution that is needed to observe the black hole signatures—the “angular size” in
the sky of a black hole at a distance D is ρ = r G /D. For SgrA*, the black hole at
the center of our Galaxy, this implies that the smallest observed images of accretion
structures around it have a size of a few μas. At the moment, they cannot be observed,
they are too small, but the angular resolution in this range will be reached by advanced
new detectors that are planned for the near future.
1. GRAVITY is planned for 2014 by the Max Planck Institute for the Extraterrestrial
Physics in Garching, Germany.3 It is the second-generation VLTI instrument for
precision narrow-angle astrometry and interferometric imaging, consisting of four
8m telescopes and a total collecting area of 200 m2 . It is the only interferometer
to allow direct imaging at high sensitivity and high image quality in a large
(≈2" ) field of view. It provides precision astrometry with resolution ≈10 μas
(microarcsecond), and imaging with resolution ≈4 mas (milliarcsecond). Note
1 The modern history of black holes started thanks to great discoveries of Schwarzschild [1],
Chandrasekhar [2] and Oppenheimer [3], and the follow-up work in the 1960s and the 1970s, done
mostly by collaborators and students of Dennis W. Sciama in Cambridge (Carter, Ellis, Gibbons,
Hawking, Penrose), John A. Wheeler in Princeton (Bekenstein, Ruffini, Thorne) and Yakov B.
Zel’dovich in Moscow (Novikov, Starobinsky) and other researchers (e.g. Israel, Damour, Kerr,
Kruskal, Wald).
2 Here, and in a few other places, I directly quote a Living Review devoted to the subject:
Abramowicz and Fragile, Foundations of Black Hole Accretion Disk Theory.

3 Homepage http://www.mpe.mpg.de/ir/gravity.
The Astrophysical Signatures of Black Holes 503
that the apparent size of the gravitational radius of SgrA*, the Galactic center
black hole, is 10 μ as.
2. THE EVENT HORIZON TELESCOPE (EHT) The EHT uses the technique of
Very Long Baseline Interferometry (VLBI) to synthesize an Earth-sized telescope
in order to achieve the highest resolution possible using ground-based instrumen-
tation. Several existing (or planned) radio telescopes are the part of the baseline:
ARO/SMT 10 m, APEX 12 m, ASTE 10 m CARMA array of six 10 m and nine of
6 m antennas, CSO 10 m, IRAM 30 m, JCMT 15 m, SMA array eight 6m anten-
nas. A few more telescopes and arrays will join. The target source is observed
simultaneously at all telescopes. The data are recorded at each of the sites and
later brought back to a processing facility where they are passed through a special
purpose supercomputer. It will be fully operational in 2015.
3. LOFT is a newly proposed space mission selected by ESA as one of the four
missions that will compete for a launch opportunity in the 2020s. The Large Area
Detector (LAD) on board of LOFT achieves an effective area of ≈10 m2 i.e. more
than an order of magnitude larger than current spaceborne X-ray detectors in the
2-30 keV range (up to 80 keV in expanded mode). LOFT will have improved
energy resolution (better than 260 eV) and will be able to investigate range from
submillisecond quasi-periodic oscillations (QPOs) to years long transient out-
bursts. LOFT will provide access (for the first time) to types of information in
these signals that are qualitatively new due to the capability to measure dynamical
timescale phenomena within their coherence time, where so far only statistical
averages of signals were accessible.
It is hoped that data from these new detectors will provide information sufficient to
answer several of the questions concerning horizon, ergosphere, ISCO and the light
circle that I describe in this review.
2 The Kerr Metric and Its Symmetries
The famous Kerr solution [5] describes the spacetime metric of a rotating, uncharged
black hole with the total mass M and the spin a = J/Mc (here J is the angular
momentum). In the Boyer–Lindquist coordinates the Kerr metric takes the form (in
the usual c = 1 = G units and + − −− signature),

μ ν 2Mr 2Mar sin2 θ
gμν d x d x = 1 − 2 dt 2 + 2 dtdφ
δ δ2

2Ma 2 r sin2 θ
− r 2 + a2 + sin2 θ dφ2
δ2
δ2
− dr 2 − δ2 dθ2 , (1)
ε
where ε = r 2 − 2 M r + a 2 and δ2 = r 2 + a 2 cos2 θ.

The Boyer-Lindquist coordinates are most widely used by astrophysicists. They

are singular at the horizon, which in these coordinates is given by ε = 0. The
Kerr metric is itself not singular at the horizon, which is apparent in the Kerr-Schild
coordinates. The Kerr-Schild version of the Kerr metric takes the form,

2Mr⊗ 4Mr⊗ 4Mr⊗
gμν d x μ d x ν = 1 − dt⊗2 − dt⊗ dr⊗ + sin2 θ⊗ dt⊗ dφ⊗
δ2⊗ δ2⊗ δ2⊗

2Ma 2 r⊗ sin2 θ⊗
− r⊗ + a +
2 2
sin2 θ⊗ dφ2⊗ − δ2⊗ dθ⊗2
δ2⊗

2Mr⊗ 2Mr⊗
− 1+ dr⊗ + 2a sin θ⊗ 1 +
2 2
dr⊗ dφ⊗ , (2)
δ2⊗ δ2⊗
where δ2⊗ = r⊗2 + a 2 cos2 θ⊗ .

The Kerr metric in the Boyer–Lindquist coordinates does not depend on t, φ. In
the Kerr–Schild coordinates it does not depend on t⊗ , φ⊗ . This is a sign of the Killing
symmetries. The Killing vectors defined by (in the Boyer–Lindquist coordinates)
μ μ
ημ = ρ t , ξμ = ρ φ , (3)
obey the Killing equation and commute,
◦ν ημ + ◦μ ην = 0, ◦ν ξμ + ◦μ ξν = 0, (4)
ξ ν ◦ν ημ = η ν ◦ν ξμ . (5)
For commuting Killing vectors one may write convenient relations,
1
ξ ν ◦ν ημ = η ν ◦ν ξμ = − ◦ν (ξ μ ημ ),
2
1
η ν ◦ν ημ = − ◦ν (η μ ημ ),
2
1
ξ ν ◦ν ξμ = − ◦ν (ξ μ ξμ ), (6)
2
which often allow to keep Christoffel’s symbols out of calculations.4

The existence of the Killing vectors η μ , ξ μ , implies conservation of energy E and
angular momentum L in geodesic motion,
⎧ μ ⎫
μ ⎨ u ◦μ E = 0 for E ∗ η μ u μ ⎬
u ◦μ u ν = 0 √ (7)
⎩ μ ⎭
u ◦μ L = 0 for L ∗ −ξ μ u μ
4 At http://www.physics.uci.edu/~etolleru/KerrOrbitProject.pdf the Christoffel symbols are given

(by Tollerud 2007) in the form of a Mathematica package. Unfortunately, there is an error in the
Kerr metric: the gtφ metric component is (consistently everywhere) factor of 2 too big. See also [6].
Fig. 1 Eddington-scaled luminosities (0.5–10 keV) of BH transients (filled circles) and NS tran-
sients (open circles) versus the orbital period (see [7, 8]). Only the lowest quiescent detections or
Chandra/XMM upper limits are shown. The plot shows all systems with known orbital periods,
which have optical counterparts and good distance estimates. The diagonal hatched areas delineate
the regions occupied by the two classes of sources and indicate the observed dependence of lumi-
nosity on orbital period. Note that the BH systems are on average nearly 3 orders of magnitude
fainter than the NS systems with similar orbital periods. Figure and caption from [9]
The specific angular momentum γ ∗ L /E is obviously also a constant of geodesic

motion. In a general, not necessarily geodesic case, the circular orbits are given
by u μ = A(η μ + ξ μ ), where ν is the angular velocity, connected to the specific
angular momentum by
r̃ 2
γ = ( − ω) , (8)
1 + ω r̃ 2
ξ μ ημ
ω=− = (frame dragging), (9)
ξ ν ξν
ξ ν ξν
r̃ 2 = − μ = (gyration radius)2 . (10)
η ημ
From (3) it follows that in the Boyer-Lindquist coordinates these quantities are
expressed by
uφ uφ dφ
E = u t , L = −u φ , γ = − , = t = . (11)
ut u dt
3 The Horizon
A coordinate independent condition for the Kerr black hole horizon could be
expressed by
(η ν ξν )2 − (η ν ην )(ξ μ ξμ ) = 0. (12)
The presence of a horizon is most often argued from observational estimates of the
“compactness parameter” ξ ≈ RG /R. Here R is the measured size of the black hole
candidate, and the gravitational radius RG is known from a mass measurement. Mass
measurements are very accurate in a few cases (based on Kepler’s laws and precise
orbit measurements), but in most cases they are not accurate (orbits are unobserved
directly). Size measurements are not yet accurate, and in addition, the measured size
is only an upper limit. In most of the considered cases, one estimates not the size of
the horizon but the size of ISCO or the size of the light circle radius.
Narayan [9–11] presented arguments that point to the horizon presence more
directly. When a central object accretes matter through an accretion disk, a part of its
radiation, L disk , originates at the disk, and a part L sur f at the surface of the object.
The surface may radiate itself, or reflect (re-radiate) a part of the disk radiation. The
total is therefore L = L disk + L sur f . The horizon of the black hole cannot radiate or
reflect L sur f = 0, and therefore in the black hole case, L = L disk . Thus, black holes
should look dimmer than non-black-hole objects with the similar accretion disks.
This indeed is observed in the case of accreting neutron stars vs. accreting black
holes, in the galactic X-ray binaries, as explained in Fig. 1. In the figure luminosities
of the binary black hole sources are compared with luminosities of the binary neutron
star sources with the same orbital binary period. In these binaries, the compact object
(a black hole or a neutron star) has a close companion, which is a low mass “normal”
star. The companion loses mass which is accreted onto the compact object via the
accretion disk. The luminosity of the accretion disk depends on the accretion rate,
which is determined by the mass loss rate from the companion. The accretion rate
cannot be directly measured, but the precisely and directly observed orbital period is a
rather accurate indicator of the accretion rate—sources with the same orbital periods
should have the same accretion rates. In Fig. 1, luminosities are given in terms of the
Eddington luminosity, L Edd = 1038 M/M⇔ [erg/s] which is proportional to the mass
of the compact object M. One should note that the neutron star sources are order
of magnitudes more luminous than the black hole sources in binaries with the same
binary periods. This difference may be indeed due to the (already mentioned) fact
that in the neutron star sources, matter that arrives at the surface may emit radiation,
while in the black hole sources matter is lost without a trace inside the black hole.
However, it was pointed out by several authors, e.g. [12, 13], for various reasons,
rigid surfaces of strange stars may also absorb all matter without a trace.
Narayan’s argument based on the evidence of some radiation lost inside the black
hole works, in a different version, also for the super-massive black holes, in particular
for SgrA*, the accreting black hole in the center of our Galaxy, for which the mass,
M = 4 × 106 M⇔ , was accurately measured from the analysis of stellar orbits in
Fig. 2 The four solid lines show independent upper limits on the mass accretion rate at the surface
of Sgr A* (assuming the source has a surface) as a function of the surface radius R. Each limiting
curve is derived from a limit on the quiescent flux of Sgr A* in an infrared band. The hatched area
at the top labeled “Typical RIAF Range” corresponds to the mass accretion rate in typical ADAF
models of Sgr A* (e.g. [14]). The horizontal dashed line represents the minimum accretion rate
needed to power the bolometric luminosity of Sgr A*. Figure and caption from [9], after [15]
its vicinity. The accreting black hole has a very low luminosity. As Fig. 2 explains,
a hypothesis that SgrA* is a dark object with a rigid surface is inconsistent with
observations and the standard accretion theory of radiatively inefficient accretion
flows (RIAF).
4 The Ergosphere
The surface of ergosphere in Kerr geometry is given by the covariant condition,5
η μ ημ = 0. (13)
In the Boyer–Lindquist coordinates this is equivalent to gtt = 0. Thus, inside the

ergosphere, the Killing vector η μ is spacelike, and therefore it is possible that, for a
timelike four-velocity u μ of a free particle, the conserved energy is negative,
5 In this Section I quote extensively from a paper in preparation: Abramowicz, Gourgoulhon,
Lasota, Narayan and Tchekhovskoy (2013), Blandford-Znajek mechanism as the Penrose process.
Application to Magnetically Arrested Disks.
u μ ημ = E < 0. (14)
4.1 The Penrose Process
Penrose [16] considered6 a freely falling particle with energy Ein , which enters the
ergosphere of a rotating black hole. He imagined that the particle disintegrates there
into two particles, with energies E(−) < 0 and E(+) > 0. Then, the particle with
negative energy E(−) falls into the black hole, and the other one escapes to infinity with
positive energy, Eout = E(+) . Since energy conservation implies that E(−) + E(+) =
Ein , one deduces that Eout > Ein , so there is a gain of energy “at infinity”. The negative
energy particle absorbed by the black hole has also a negative angular momentum,
L(−) < 0. The source for the gain of energy at infinity is therefore the rotational
energy of the black hole.
Soon after Penrose’s discovery that rotating black holes may be energy sources,
several authors suggested that the Penrose process may power relativistic jets
observed in quasars (and later in microquasars). However, a careful analysis by
[18–22] and others shown that it is unlikely that negative energy states, necessary
for the Penrose process to work, may be achieved through the particles disintegra-
tion inside ergosphere. The same conclusion was reached more recently by [23] for
high-energy collision of particles. The reason is that in the case of disintegration,
for a negative energy state to occur, velocities of fragments measured in their center
of mass frame should be very relativistic, v > c/2. In the case of collisions, the
particles with positive energies cannot escape, as they must have large and negative
radial momenta. Thus, they are captured (together with the negative energy particles)
by the black hole.
In the context of the super-energetic collisions there is a disagreement between
two opinions, based on recently obtained results. Firstly, Silk and his collaborators
[24] claimed that in the center-of-mass frame, the energy of particular types of colli-
sions may be arbitrary large and that this may lead to astrophysically important and
interesting consequences. Secondly, Bejger and collaborators [23] claimed that con-
sequences of such collisions are unobservable. Both results have been published and
attracted a considerable attention. There is a vigorous follow-up going on—see [25]
for the most recent significant result, which confirms and expands results obtained
by [23]: black holes are neither particle accelerators nor dark matter probes. This
leaves magnetic processes as the only astrophysically realistic way to extract rota-
tional energy from a rotating black hole.
6 See also [17].

4.2 The Blandford-Znajek Mechanism
The presence of ergosphere is today discussed mostly in two contexts: the origin of
relativistic jets and super-energetic collisions of particles deep inside the ergosphere.
In the jet context, Tchekhovskoy, Narayan and their collaborators [26–28] have
found that the famous Blandford–Znajek mechanism [29] works for “magnetically
arrested disks”, i.e. a special type of magnetized black hole accretion [30]. There are
several issues here that need to be studied and explained.
5 The ISCO
The existence of ISCO is probably the single most important strong-gravity effect
in the whole black hole accretion disk theory. There is, however, a controversy con-
cerning the ISCO. According to one view [31, 32], for small accretion rates, location
of ISCO determines the “inner edge” of the disk which separates the part of the disk
where matter rotates on almost Keplerian, almost circular orbits, from the plunging-in
region, where matter falls in into the black hole almost freely. There is no (signifi-
cant) radiation coming from the plunging-in region, and stresses there are negligible.
Numerous well-known and widely used results in accretion theory depend on the
assumption that ISCO is the sharp boundary between the two different accretion
regimes. In particular, works on the black hole spin estimate based on spectral fitting
adopt this assumption [33, 34]. According to the opposite view [35], most recently
eloquently summarized by Balbus [36], ISCO is not an important feature of black
hole accretion even for small accretion rates, because the magnetohydrodynamical
MRI instability makes the flow unstable and turbulent on both sides of ISCO.
It is obvious that because the ISCO appears as an important ingredient in numerous
specific (and important) results of the black hole accretion disk theory, it is one of
its few pillars on which the theory rests. For all these reasons, the ISCO receives
more attention in my review than the other three black hole accretion disk signatures
(horizon, ergosphere, circular photon orbit).
5.1 ISCO for Keplerian, Circular Orbits (Circular Geodesics)
In this section we use the Boyer–Lindquist coordinates (1). Let us consider a test
particle with mass m which moves on the equatorial plane, θ = π/2. For such a
motion, the polar components of the four velocity vanish, u θ = 0 = u θ , and from
u μ u ν g μν = 1 it follows that,
(u t )2 g tt + 2u t u φ g tφ + (u φ )2 g φφ + (u r )2 grr = 1. (15)
Let us now assume that the motion is almost circular in the sense that r (s) = r0 +
ρr (s) and |ρr (s)| ≤ r0 for some constant positive r0 , which corresponds to the radius
of the circle in the Boyer–Lindquist coordinates (s is the spacetime length along the
trajectory). Using the Boyer–Lindquist expressions (11) for the conserved energy E
and the conserved specific angular momentum γ, one may write equation (15) as
1 d(ρr )
|grr |(ρṙ)2 = E − Ueff (r, γ), (ρṙ ) ∗ , (16)
2 ds
1
Ueff = − ln g tt − 2γg tφ + γ2 g φφ , (17)
2
where E = ln E may also be rightly called the energy because in the Newtonian
limit E ≡ E Newton , and E ≡ E Newton + 1. The effective potential Ueff (r, γ) equals
its Newtonian counterpart in the Newtonian limit. Except of |grr | ≡ 1, Eq. (16)
has its form identical with the corresponding Newtonian one, very familiar from the
classical mechanics. Thus, using the same procedure as in the Newtonian case, one
may derive equations that govern the strictly circular geodesic motion ρṙ = 0, and a
slightly perturbed ρṙ ∇= 0 one,

∂Ueff
= 0 √ ∂r g tt − 2γ ∂r g tφ + γ2 ∂r g φφ = 0, (18)
∂r γ

1 ∂ 2 Ueff
= κ2 and ρr̈ + κ2 ρr = 0 . (19)
|grr | ∂r 2 γ
As the Kerr metric components g μν = g μν (r ) are known functions of the Boyer-

Lindquist coordinate r on the equatorial plane, the quadratic equation in (18) may be
easily solved. Its solution γ = γ K (r ) represents the Keplerian angular momentum
distribution,
M 1/2 (r 2 − 2a M 1/2 r 1/2 + a 2 )

γK = , (20)
r 3/2 − 2Mr 1/2 + a M 1/2
GM
( K )2 = 2 , (21)
r + a M 1/2
3/2
κ 1/2
ωr = = K 1 − 6x −1 + 8a⊗ x −3/2 − 3a⊗2 x −2 . (22)
A
Here a⊗ = a/M. In the second line (21) we give the Keplerian angular velocity
K , and in the third line (22) the radial epicyclic frequency, which is the redshifted
eigenvalue κ. In the Boyer–Lindquist coordinates the redshift factor is A = u t , and
covariantly,
1
A= ⎛ ⎝1/2 . (23)
(η μ ημ ) + 2(ξ μ ημ ) + 2 (ξ μ ξμ )
10.0 rms
7.0
rmb
5.0
rph
rM
3.0
rmb rms
rph
r0 θ 2
2.0
rH
1.5 direct
retrograde
1.0
0.0 0.2 0.4 0.6 0.8 1.0
a
Fig. 3 The location of ISCO for different Kerr black hole spins, rms (a), together with locations of
other characteristic radii (in terms of the Boyer–Lindquist coordinate r ). Adapted from [18]
From the simple harmonic oscillator equation in (19) it follows that for κ2 < 0 the
circular Keplerian orbits are unstable. The Innermost Stable Circular Orbit (ISCO)
is located therefore at the radius rms defined by equation κ2 (rms ) = 0. Its solution
yields
⎞
ISCO location : rms = M 3 + Z 2 − [(3 − Z 1 )(3 + Z 1 + 2Z 2 )]1/2 ,
Z 1 = 1 + (1 − a⊗2 )1/3 [(1 + a⊗ )1/3 + (1 − a⊗ )1/3 ],
Z 2 = (3a⊗2 + Z 12 )1/2 . (24)
The location of ISCO for different Kerr black hole spins, rms (a), is shown in Fig. 3,
together with locations of other characteristic radii (in terms of the Boyer–Lindquist
coordinate r ),
⎠
2
photon rph = 2M
1 + cos cos−1 (a⊗ ) , (25)
3
a⊗
bound rmb = 2M 1 − + 1 − a⊗ , (26)
2

horizon rH = M 1 + 1 − a⊗ , 2 (27)

ergosphere r0 = M 1 + 1 − a⊗ cos θ .
2 2 (28)
5.2 The Shakura–Sunyaev ISCO Paradigm for Thin

Accretion Disks
Many basic results of the black hole accretion theory are a direct consequence of the
existence of the ISCO.7 They are reviewed and discussed e.g. in [4]. The “standard”
black hole stationary, non-magnetized, accretion disks disks are characterized by
small (very sub-Eddington) accretion rates, Ṁ ≤ ṀEdd , and small vertical thickness,
h ∗ H/r ≤ 1. Their mathematical model was developed (in Newton’s theory) by
Shakura and Sunyaev in the most influential and important paper in the accretion
disk theory [37], hereafter SS73. Its general relativistic version, proper for the Kerr
black holes, was found by Novikov and Thorne [38], hereafter NT73.
For such disks, the ISCO behaves like a physical boundary, separating the disk
proper from the plunging region. In the disk proper (we will call it simply “disk”),
i.e. in the region r > rms , matter moves on nearly circular, nearly Keplerian orbits.
Consequently, the angular momentum of the disk is nearly Keplerian, γ(r ) ⇒ γ K (r ).
Dissipation of orbital energy and angular momentum is significant, and most (≈98 %)
of the accretion flow radiation comes from there. In the plunging region, r < rms ,
matters nearly falls freely. Stresses are ineffective in dissipating energy and trans-
porting angular momentum. Consequently, very little radiation (≈2 %) comes from
this region, and the angular momentum distribution there is almost constant, γ(r ) ⇒
const.
The SS73 and NT73 models were developed two decades before Balbus and
Hawley [39] made the seminal discovery that the torque needed for accretion discs to
operate originates from the turbulence caused by the Magneto Rotational Instability
(MRI) which weakly magnetized, differentially rotating fluids suffer. The view that
the MRI is crucial in providing that necessary torque, was supported by numerous
follow-up works, mostly based on magnetohydrodynamical numerical simulations.
Today it is generally accepted by a vast majority of the black hole accretion disk
researchers.8 For the review of successes of the hypothesis of the MRI induced
torque, see [42] and [43].
With no knowledge of the true nature of viscus torques in accretion disks, the SS73
and NT73 models adopted a heuristic viscosity prescription based on the assumption
that the torque is proportional to the total pressure,
T = α p, (29)
with 0 < α < 1 being a parameter. The SS73 and NT73 models have also adopted
other simplifying assumptions, among them that at the ISCO the torque is zero,
7 In this Section I quote in extenso a few paragraphs from an unfinished draft of an unpublished
paper by Abramowicz, Horák and Kluźniak, The MRI in the plunge-in region: the Shakura–Sunyaev
ISCO paradigm confirmed, 2013, in preparation.
8 However, [40, 41] and others pointed to some difficulties with the MRI concept and also with its
description in the shearing box simulations.

T (rms ) = 0, (30)
and that there is no radial (advective) heat transport in the disk. These assumptions
introduced an artificial singularity at the ISCO, not relevant for calculating spectra,
but obviously calling for a more accurate treatment of the flow near ISCO. This more
accurate treatment was initiated in works [44] and [45]. It eventually maturated as
the slim disk model [46].
Slim disks assume the alpha-viscosity prescription (29), but do not assume the
zero-torque condition at ISCO (30). Instead, they consistently solve (numerically) the
full set of the (relativistic, in the Kerr geometry) Navier–Stokes equations, ◦ν T νμ =
0, expanded in terms of the small vertical disk thickness h, up to the quadratic
order O(h 2 ). This includes solving the vertical radiative transfer in the diffusion
approximation [47]. Mathematically, the slim disk equations form an eigen-value
problem, with the eigen-value being the angular momentum of matter crossing the
black hole horizon. The physical reason for the eigen-value nature of the problem is
that the black hole accretion must necessarily be transonic, with the sonic point r S
being the critical (saddle) point of the slim disk differential equations. The regularity
conditions at r S assure that its location is very close to ISCO, r S = rms − A2 h 2 ,
where A depends on the equation of state, with the the sound speed C S ≤ c, which
is always true for very thin disks, h ≤ 1 [48].
The slim disk calculations fully confirmed that, for Ṁ ≤ Ṁedd and h ≤ 1, the
ISCO was a (quite sharp) boundary between disk and plunging region, and that the
stress at ISCO was indeed small. Despite that, the ISCO paradigm was challenged
mostly by Krolik [35], but also by others, on the ground that the alpha viscosity pre-
scription (29) used in the slim disk calculations was not adequate, because (according
to their view) it cannot properly describe the MRI torque across the ISCO. Paczynski
[31] and Afshordi and Paczynski [32] argued that independently of the physical nature
of the torque, for the thin (h ≤ 1) disks the small stress at ISCO is a consequence
of the angular momentum conservation. However, Paczyński’s clear arguments have
not been accepted (or understood) by a few opponents of the standard ISCO para-
digm. It was obvious that the opponents could be convinced only by calculations or
arguments that specifically include the MRI induced turbulence.
Fortunately for the validity of the standard ISCO paradigm, such calculations
and arguments are now available. Firstly, Hirose et al. [49] convincingly demon-
strated in terms of MHD shear box simulations (which included vertical stratification
and radiative cooling) that the standard Shakura-Sunyaev viscosity prescription (29)
adequately describes the MRI torque (see also [50]). Secondly, Shafee et al. [51],
Reynolds and Fabian [52] and Penna et al. [53] performed recently MHD global
simulations of the black hole accretion disks, concluding that for geometrically thin
accretion disks used in their simulations (i.e. with h ≈ 0.05 − 0.1):
1. The angular momentum distribution in the flow is characterized by γ(r ) ≈ γ K (r )
outside ISCO (i.e. for r > rms ) and γ(r ) ≈ const inside ISCO (i.e. for r < rms ),
see Fig. 4. Correspondingly to this (as ρ E ≈ ργ), there is very small energy
dissipation inside ISCO, and therefore very little radiation may originate there.
plunge-in disk
= plunge-in disk
(c rG )
ξturb
= flow
= const
√
(v r )2 vr 2
0 ξturb = vr
r/rG r/rG
Fig. 4 Left: Recent three-dimensional MHD simulations by Shafee et al. [51], Reynolds and Fabian
[52] and Penna etal. [53] proved that the angular momentum distribution, γ f low , in the vertically thin
accretion flow (h < 0.1) agrees within ≈2 % with that predicted by the standard ISCO paradigm
in SS73, NT73 and the slim disk models, i.e. that γ = γ K in the “disk” region outside ISCO, and
γ = const in the “plunge-in” region inside ISCO. The vertical dotted line indicates the ISCO. The
horizontal dotted line corresponds to the total angular momentum transport, divided by accretion
rate, γ0 = J˙/ Ṁ. Its constancy over the large range of radii proves that the steady state has been
achieved in the simulations. Figure adopted from [51]. Right: Time averaged profile of ξtur b , which
measures the relative magnitude of turbulent fluctuations in the accreting gas. The fluid becomes
mostly laminar inside the ISCO. Figure taken from [51]
2. Turbulent activity is pronounced at radii beyond about 10r G . Flow is nearly lam-
inar inside the ISCO, see Fig. 4. This agrees with γ(r ) ≈ const there, and also
suggests that the bulk of the dissipation occurs in the disk outside the ISCO, but
not in the plunging region.
The above mentioned new results could be taken as the final proof of the validity
of the standard ISCO paradigm if not one remaining subtle point. Most of the present
understanding of the properties of MIR is based on the MHD shear box simulations,
that are usually done with two particular assumptions—that the radial component of
the flow is negligible, and that the radial profile of angular momentum corresponds
to the Newtonian-Keplerian one,
Vr = 0, (31)
γ(r ) ≈ r s , with s = 1/2. (32)
In the disk proper these conditions are satisfied. However, in the plunging region
they are not fulfilled. Instead of Vr = 0 one has Vr ∈ C S (indeed Vr ≈ c), and
instead of s = 1/2, one has s = 0. These differences could very significantly affect
the strength of the MRI, as argued by [54].
However, works on slim disks and MHD simulations that support the standard
ISCO paradigm have adopted assumption (directly or not) that the MRI induced
torque is not qualitatively different on both sides of ISCO. For example, in terms of
the alpha prescription this corresponds to α(r ) ⇒ const.9
If one could argue that the MRI properties on both sides of ISCO are indeed
qualitatively the same (despite the differences in the behavior of Vr and s), then the
validity of the ISCO paradigm would be finally proven.
Most recently such arguments have been put forward by [36], who stated in
the Abstract of his paper that: the defining properties of the MRI—its maximum
growth rate and the direction of the associated eigenvector displacement—remain
unchanged as the Rayleigh discriminant passes from positive to negative values. In
other words, because the “Rayleigh discriminant” is simply κ2 , i.e. the square of
the radial epicyclic frequency which changes its sign at the ISCO, the properties
of MRI are the same on both sides of the ISCO. Thus, if the arguments in [36] are
correct, they provide the last missing point to complete the proof of the validity of
the standard ISCO paradigm.
5.3 Leaving the ISCO
Abramowicz et al. [4] revisited Krolik and Hawley’s [56] discussion of the location
of the “inner edge” rin of accretion disks around black holes, and expanded it to
include disks with around Eddington accretion rates. The concept of the inner edge
may be introduced by several empirical definitions of the accretion disk inner edge,
each serving a different practical purpose:
1. The potential spout edge rin = r pot , where the effective potential Ueff , see
equation (17) forms a self-crossing Roche lobe, and accretion is governed by the
Roche lobe overflow.
2. The sonic edge rin = rson , where the transition from subsonic to transonic accre-
tion occurs. Hydrodynamical disturbances do not propagate upstream a super-
sonic flow, and therefore the subsonic part of the flow is “causally” disconnected
from the supersonic part.
3. The variability edge rin = rvar , the smallest radius where orbital motion of
coherent spots may produce quasi periodic variability.
4. The stress edge rin = rstr , the outermost radius where the Reynolds stress is
small, and plunging matter has no dynamical contact with the outer accretion
flow.
5. The radiation edge rin = rrad , the innermost place from which significant
luminosity emerges.
6. The reflection edge rin = rr e f , the smallest radius capable of producing signifi-
cant fluorescent iron line.
They found [4] that: for black hole accretion disks with very sub-Eddington lumi-
nosities all these inner edges locate at ISCO. Thus, in this case, one may rightly
9 But see [55].

consider ISCO as the unique inner edge of the black hole accretion disk. However,
even for moderately higher luminosities, there is no such unique inner edge as differ-
ently defined edges locate at different places. Several of them are significantly closer
to the black hole than ISCO. The differences grow with the increasing luminosity.
For nearly Eddington luminosities, they are so huge that the notion of the inner edge
loses all practical significance.
5.4 Evidence for ISCO from the Observed Variability
Quasi periodic oscillations (called QPOs) are high frequency (about kHz) oscillations
in the X-ray fluxes from neutron star and black hole sources in the Galactic X-ray
sources (they have been also observerd in a few extra-Galactic sources and in SgrA*).
In many sources they appear as a pair of oscillations, and are called the “twin-peak”
QPOs. Before 2000, twin peak kHz QPOs have been observed only in the neutron
star sources. It was believed that they must be connected to the neutron star rigid
surfaces, and cannot occur in the black hole sources. Kluźniak and Abramowicz
suggested (see [57]) that the twin peak kHz QPOs are due to a non-linear resonance
in accretion disks oscillations, and for this reason their frequencies should have
ratios close to those of small natural numbers, for example 3:2. Their prediction that
the twin peak QPOs should also appear in the BH sources was soon confirmed by
Strohmayer [58], who observed a twin peak QPOs in the BH candidate GRO J1655-
40. The 3:2 non-linear resonance explanation is now generally accepted, but despite
its successes in finding general signatures of the resonance in the observational data,
several questions remain unanswered, in particular the behavior of the quality factor
Q of the QPOs. Barret et al. (see [59]) found that Q of the lower-frequency QPO in
the neutron-star sources increase with increasing QPO frequency up to Q ∈ 200 and
then it drops. The high-frequency QPO has Q ≈ 10 and does not follow the same
pattern. There is a consensus that this rules out any kinematical model of QPOs as
orbiting clumps or spots. The drop in Q in the lower-frequency QPO is attributed
to the existence of the ISCO (see Fig. 5), but this interpretation is not generally
accepted.
6 The Light Circle
The light circle issue is an emerging topic [61] that is connected to improving obser-
vational potentials in the sub-milliarcsecond radio imaging of the black hole sources,
in particular SgrA*, i.e. the black hole in the center of our Galaxy.10
The shortest timescale that may be in principle observed in the accreting black
hole sources is not connected to the accretion flow, but to the strong gravity itself
10Based on a lecture by M. Bursa given at the 9 RAGtime Workshop in Opava, 19–21 September,
2007 [61].
Fig. 5 A sudden drop in the quality factor Q of the observed twin peak QPO (lower frequency) was
suggested as an evidence for the existence of the ISCO. The data presented here is from the Rossi
X-ray Timing Explorer satellite observations of the neutron star binary source 4U 1636-536. The
source shows quasi-periodic oscillations with varying frequencies in the range 650 Hz–900 Hz. The
sharp drop in the quality factor (bottom panel) was recorded at the frequency ≈S870 Hz. Figure
from [60]
Fig. 6 Trajectories of direct and looped photons emerging from a flare on the surface of an accretion
disk. Figure reproduced from [61]
which close to the horizon loops photon trajectories around the black hole (see Fig. 6).
Signals from some transient events in accretion disk, e.g., random short-lived flares,
may therefore reach the observer repeatedly with delays corresponding to the travel
time around the circular photon orbit. The looped signals will introduce a correlation
in the variability data, with the timescale T photon shorter than the timescale at ICSO,
TI SC O . For a non-rotating black hole it is, (Fig. 7)
T photon = 32.6M, TI SC O = 92.3M. (33)
If the conditions are right, signal from some transient events in accretion disk,
e.g., random short-lived flares, may reach the observer repeatedly with delays corre-
50 50
y (mas)
y (mas)
0 0
−50 −50
−50 0 50 −50 0 50
x (mas) x (mas)
50 50
y (mas)
y (mas)
0 0
−50 −50
−50 0 50 −50 0 50
x (mas) x (mas)
Fig. 7 The circular light trajectory shows up in the theoretically calculated images of SgrA*.
Images taken from [62]
sponding to the travel time around the photon orbit and still with a sufficient intensity
to be practically detected. If the delay in arrival time from the vicinity of a black hole
between “direct” and “looped” photons could indeed be found in the light curves
of AGNs or microquasars, it would not only provide an excellent tool to measure
the mass and spin of the black hole, but it would also provide direct evidence for the
existence of (nearly) circular photon orbits. In this way one would demonstrate the
validity of an important prediction of general relativity in the regime of extremely
strong gravitational field, see Table 1.
Table 1 Light circle timescales for different black hole sources: AGN, i.e. supermassive black
holes in galactic centers, ULX i.e. hypothetical “intermediate mass” black holes postulated as an
explanation of the Ultra Luminous X-ray sources, and GBH, i.e. microquasars in the Galactic black
holes (X-ray binaries)
Luminosity [erg/s] Distance [kpc] Timescale T photon [s]
AGN 1041 − 1043 103 − 104 102 − 104
ULX 1039 − 1041 103 − 104 10−2 − 10−1
GBH 1036 − 1038 100 − 101 10−4 − 10−3
7 Conclusions
Today we have at hand strong observational arguments that, for all practical purposes,
prove that the compact objects detected in the Galactic X-ray binaries and at the
centers of our Galaxy, and other galaxies, are indeed black holes. We do not have
yet similarly strong arguments, based on observational data, to prove that these are
the Kerr black holes. Advanced instruments, planned for a near future, will provide
opportunities to probe the spacetime metric around the black hole candidates with a
sufficient space (i.e. angular) and time resolution to obtain constrains.
Acknowledgments My work at the Silesian University in Opava was supported by the Czech
CZ.1.07/2.3.00/20.0071 “Synergy” grant for international collaboration, and at the Institute of
Astronomy in Prague by the Czech ASCR M100031242 grant. I also acknowledge support from
the Polish NCN grant UMO-2011/01/B/ST9/05439.
References
1. Schwarzschild, K.: Über das Gravitationsfeld eines Massenpunktes nach der Einsteinschen
Theorie. Sitzungsberichte der Königlich Preussischen Akademie der Wissenschaften 1, 189
(1916)
2. Chandrasekhar, S.: The maximum mass of ideal white dwarfs. Astrophys. J. 74, 81 (1931).
doi:10.1086/143324
3. Oppenheimer, J.R., Snyder, H.: On continued gravitational contraction. Phys. Rev. 56, 455
(1939). doi:10.1103/PhysRev.56.455
4. Abramowicz, M.A., Jaroszyński, M., Kato, S., et al.: Leaving the innermost stable circular
orbit: the inner edge of a black-hole accretion disk at various luminosities. A&A 521, A15
(2010). doi:10.1051/0004-6361/201014467
5. Kerr, R.P.: Gravitational field of a spinning mass as an example of algebraically special metrics.
6. Dovciak, M.: Radiation of accretion discs in strong gravity. ArXiv e-prints arXiv:astro-ph/0
411605 (2004)
7. Garcia, M.R., McClintock, J.E., Narayan, R., et al.: New evidence for black hole event horizons
from Chandra. Astrophys. J. Letters 553, L47 (2001). doi:10.1086/320494
8. McClintock, J.E., Narayan, R., Rybicki, G.B.: On the lack of thermal emission from the qui-
escent black hole XTE J1118+480: Evidence for the event horizon. Astrophys. J. 615, 402
(2004). doi:10.1086/424474
9. Narayan, R., McClintock, J.E.: Advection-dominated accretion and the black hole event hori-
zon. New Astr. Rev. 51, 733 (2008). doi:10.1016/j.newar.2008.03.002
10. Menou, K., Esin, A.A., Narayan, R., et al.: Black hole and neutron star transients in quiescence.
Astrophys. J. 520, 276 (1999). doi:10.1086/307443
11. Narayan, R.: George Darwin lecture: Evidence for the black hole event horizon. Astron. Geo-
phys. 44(6), 060000 (2003). doi:10.1046/j.1468-4004.2003.44622.x
12. Abramowicz, M.A., Kluźniak, W., Lasota, J.P.: No observational proof of the black-hole event-
horizon. A&A 396, L31 (2002). doi:10.1051/0004-6361:20021645
13. Lemos, J.P.S., Zaslavskii, O.B.: Black hole mimickers: regular versus singular behavior. Phys.
14. Yuan, F., Quataert, E., Narayan, R.: Nonthermal electrons in radiatively inefficient accretion
flow models of sagittarius a*. Astrophys. J. 598, 301 (2003). doi:10.1086/378716
15. Broderick, A.E., Narayan, R.: On the nature of the compact dark mass at the galactic center.
Astrophys. J. Letters 638, L21 (2006). doi:10.1086/500930
16. Penrose, R.: Gravitational collapse: the role of general relativity. Nuovo Cimento Rivista Serie
1, 252 (1969)
17. Penrose, R., Floyd, R.M.: Extraction of rotational energy from a black hole. Nat. Phys. Sci.
229, 177 (1971). doi:10.1038/physci229177a0
18. Bardeen, J.M., Press, W.H., Teukolsky, S.A.: Rotating black holes: locally nonrotating frames,
energy extraction, and scalar synchrotron radiation. Astrophys. J. 178, 347 (1972). doi:10.
1086/151796
19. Wald, R.M.: Energy limits on the Penrose process. Astrophys. J. 191, 231 (1974). doi:10.1086/
152959
20. Kovetz, A., Piran, T.: The efficiency of the Penrose process. Nuovo Cimento Lettere 12(2),
39–42 (1975)
21. Piran, T., Shaham, J.: Upper bounds on collisional Penrose processes near rotating black-hole
horizons. Phys. Rev. D 16, 1615 (1977). doi:10.1103/PhysRevD.16.1615
22. Wald, R.M.: General Relativity. University of Chicago Press, Chicago (1984)
23. Bejger, M., Piran, T., Abramowicz, M., Håkanson, F.: Collisional Penrose process near the
horizon of extreme Kerr black holes. Phys. Rev. Lett. 109(12), 121101 (2012). doi:10.1103/
24. Bañados, M., Silk, J., West, S.M.: Kerr black holes as particle accelerators to arbitrarily high
energy. Phys. Rev. Lett. 103(11), 111102 (2009). doi:10.1103/PhysRevLett.103.111102
25. McWilliams, S.T.: Black holes are neither particle accelerators nor dark matter probes. Phys.
Rev. Lett. 110(1), 011102 (2013). doi:10.1103/PhysRevLett.110.011102
26. Tchekhovskoy, A., Narayan, R., McKinney, J.C.: Black hole spin and the radio loud/quiet
dichotomy of active galactic nuclei. Astrophys. J. 711, 50 (2010). doi:10.1088/0004-637X/
711/1/50
27. Tchekhovskoy, A., Narayan, R., McKinney, J.C.: Efficient generation of jets from magnetically
arrested accretion on a rapidly spinning black hole. Mon. Not. Roy. astr. Soc. 418, L79 (2011).
doi:10.1111/j.1745-3933.2011.01147.x
28. McKinney, J.C., Tchekhovskoy, A., Blandford, R.D.: General relativistic magnetohydrody-
namic simulations of magnetically choked accretion flows around black holes. Mon. Not. Roy.
astr. Soc. 423, 3083 (2012). doi:10.1111/j.1365-2966.2012.21074.x
Mon. Not. Roy. astr. Soc. 179, 433 (1977)
30. Narayan, R., Igumenshchev, I.V., Abramowicz, M.A.: Magnetically arrested disk: an energet-
ically efficient accretion flow. Pub. Astr. Soc. Jap. 55, L69 (2003)
31. Paczyński, B.: The inner boundary condition for a thin disk accreting into a black hole. ArXiv
e-prints arXiv:astro-ph/0004129 (2000)
32. Afshordi, N., Paczyński, B.: Geometrically thin disk accreting into a black hole. Astrophys. J.
592, 354 (2003). doi:10.1086/375559
33. Shafee, R., McClintock, J.E., Narayan, R., et al.: Estimating the spin of stellar-mass black
holes by spectral fitting of the X-ray continuum. Astrophys. J. 636, L113 (2006). doi:10.1086/
498938
34. Straub, O., Bursa, M., Sadowski, A., et al.: Testing slim-disk models on the thermal spectra of
LMC X-3. A&A 533, A67 (2011). doi:10.1051/0004-6361/201117385
35. Krolik, J.H.: Magnetized accretion inside the marginally stable orbit around a black hole.
Astrophys. J. 515, L73 (1999). doi:10.1086/311979
36. Balbus, S.A.: On the behaviour of the magnetorotational instability when the
rayleigh criterion is violated. Mon. Not. Roy. astr. Soc. 423, L50 (2012). doi:10.1111/j.1745-
3933.2012.01255.x
37. Shakura, N.I., Sunyaev, R.A.: Black holes in binary systems. Observational appearance. A&A
24, 337 (1973)
38. Novikov, I.D., Thorne, K.S.: Astrophysics of black holes. In: Dewitt, C., Dewitt, B.S. (eds.)
Black Holes (Les Astres Occlus), pp. 343–450 (1973)
39. Balbus, S.A., Hawley, J.F.: A powerful local shear instability in weakly magnetized disks. I—
Linear analysis. II—Nonlinear evolution. Astrophys. J. 376, 214 (1991). doi:10.1086/170270
40. Umurhan, O.M., Menou, K., Regev, O.: Weakly nonlinear analysis of the magnetorotational
instability in a model channel flow. Phys. Rev. Lett. 98(3), 034501 (2007). doi:10.1103/
41. Regev, O., Umurhan, O.M.: On the viability of the shearing box approximation for numerical
studies of mhd turbulence in accretion disks. A&A 481, 21 (2008). doi:10.1051/0004-6361:
20078413
43. Balbus, S.A.: Enhanced angular momentum transport in accretion disks. Ann. Rev. Astr. Ap.
41, 555 (2003). doi:10.1146/annurev.astro.41.081401.155207
44. Paczynski, B., Bisnovatyi-Kogan, G.: A model of a thin accretion disk around a black hole.
Acta Astr. 31, 283 (1981)
45. Muchotrzeb, B., Paczynski, B.: Transonic accretion flow in a thin disk around a black hole.
Acta Astr. 32, 1 (1982)
46. Abramowicz, M.A., Czerny, B., Lasota, J.P., Szuszkiewicz, E.: Slim accretion disks. Astrophys.
J. 332, 646 (1988). doi:10.1086/166683
47. Sadowski, A., Abramowicz, M., Bursa, M., et al.: Relativistic slim disks with vertical structure.
A&A 527, A17 (2011). doi:10.1051/0004-6361/201015256
48. Abramowicz, M.A., Zurek, W.H.: Rotation-induced bistability of transonic accretion onto a
black hole. Astrophys. J. 246, 314 (1981). doi:10.1086/158924
49. Hirose, S., Blaes, O., Krolik, J.H.: Turbulent stresses in local simulations of radiation-dominated
accretion disks, and the possibility of the lightman-eardley instability. Astrophys. J. 704, 781
(2009). doi:10.1088/0004-637X/704/1/781
50. Hirose, S., Krolik, J.H., Blaes, O.: Radiation-dominated disks are thermally stable. Astrophys.
J. 691, 16 (2009). doi:10.1088/0004-637X/691/1/16
51. Shafee, R., McKinney, J.C., Narayan, R., et al.: Three-dimensional simulations of magnetized
thin accretion disks around black holes: stress in the plunging region. Astrophys. J. 687, L25
(2008). doi:10.1086/593148
52. Reynolds, C.S., Fabian, A.C.: Broad iron-Kα emission lines as a diagnostic of black hole spin.
Astrophys. J. 675, 1048 (2008). doi:10.1086/527344
53. Penna, R.F., McKinney, J.C., Narayan, R., et al.: Simulations of magnetized discs around black
holes: effects of black hole spin, disc thickness and magnetic field geometry. Mon. Not. Roy.
astr. Soc. 408, 752 (2010). doi:10.1111/j.1365-2966.2010.17170.x
54. Abramowicz, M., Brandenburg, A., Lasota, J.P.: The dependence of the viscosity in accretion
discs on the shear/vorticity ratio. Mon. Not. Roy. astr. Soc. 281, L21 (1996)
55. R.F. Penna, A. Sadowski, A.K. Kulkarni, R. Narayan, The Shakura-Sunyaev viscosity pre-
scription with variable α(r ), ArXiv e-prints arXiv:1211.0526 [astro-ph.HE] (2012)
56. Krolik, J.H., Hawley, J.F.: Where is the inner edge of an accretion disk around a black hole?
Astrophys. J. 573, 754 (2002). doi:10.1086/340760
57. Abramowicz, M.A., Kluźniak, W.: A precise determination of black hole spin in GRO J1655–
40. A&A 374, L19 (2001). doi:10.1051/0004-6361:20010791
58. Strohmayer, T.E.: Discovery of a 450 Hz quasi-periodic oscillation from the microquasar GRO
J1655-40 with the Rossi X-ray timing explorer. Aastrophys. J. Letters 552, L49 (2001). doi:10.
1086/320258
59. Barret, D., Kluźniak, W., Olive, J.F., Paltani, S., Skinner, G.K.: On the high coherence of kHz
quasi-periodic oscillations. Mon. Not. Roy. astr. Soc. 357, 1288 (2005). doi:10.1111/j.1365-
2966.2005.08734.x
60. Barret, D., Olive, J.F., Miller, M.C.: An abrupt drop in the coherence of the lower kHz quasi-
periodic oscillations in 4U 1636–536. Mon. Not. Roy. astr. Soc. 361, 855 (2005). doi:10.1111/
j.1365-2966.2005.09214.x
61. Bursa, M., Abramowicz, M.A., Karas, V., Kluźniak, W., Schwarzenberg-Czerny, A.: The
timescale of encircling light. In: Hledík, S., Stuchlík, Z. (eds.) Proceedings of RAGtime 8/9:
Workshops on Black Holes and Neutron Stars, pp. 21–25 (2007)
62. Straub, O., Vincent, F.H., Abramowicz, M.A., Gourgoulhon, E., Paumard, T.: Modelling the
black hole silhouette in Sagittarius A* with ion tori. A&A 543, A83 (2012). doi:10.1051/0004-
6361/201219209
Energy Extraction from Spinning Black Holes
Via Relativistic Jets
Ramesh Narayan, Jeffrey E. McClintock and Alexander Tchekhovskoy
Abstract It has for long been an article of faith among astrophysicists that black
hole spin energy is responsible for powering the relativistic jets seen in accreting
black holes. Two recent advances have strengthened the case. First, numerical gen-
eral relativistic magnetohydrodynamic simulations of accreting spinning black holes
show that relativistic jets form spontaneously. In at least some cases, there is unam-
biguous evidence that much of the jet energy comes from the black hole, not the disk.
Second, spin parameters of a number of accreting stellar-mass black holes have been
measured. For ballistic jets from these systems, it is found that the radio luminosity of
the jet correlates with the spin of the black hole. This suggests a causal relationship
between black hole spin and jet power, presumably due to a generalized Penrose
process.
1 Introduction
Relativistic jets are a common feature of accreting black holes (BHs). They are
found in both stellar-mass BHs and supermassive BHs, and are often very powerful.
Understanding how jets form and where they obtain their enormous power is an
active area of research in astrophysics.
R. Narayan (B) · J. E. McClintock

Harvard-Smithsonian Center for Astrophysics, Harvard University,
60 Garden St, Cambridge, MA 02138, USA
e-mail: rnarayan@cfa.harvard.edu
J. E. McClintock
e-mail: jmcclintock@cfa.harvard.edu
A. Tchekhovskoy
Department of Astrophysical Sciences, Peyton Hall, Princeton University,
Princeton, NJ 08544, USA
e-mail: atchekho@princeton.edu
524 R. Narayan et al.
In seminal work, Penrose [1] showed that a spinning BH has free energy that is, in
principle, available to be tapped. This led to the popular idea that the energy source
behind relativistic jets might be the rotational energy of the accreting BH. A number
of astrophysical scenarios have been described in which magnetic fields enable this
process [2–11]. Field lines are kept confined around the BH by an accretion disk,
and the rotation of space-time near the BH twists these lines into helical magnetic
springs which expand under their own pressure, accelerating any attached plasma.
Energy is thereby extracted from the spinning BH and is transported out along the
magnetic field, making a relativistic jet. Although this mechanism requires accretion
of magnetized fluid and is thus not the same as Penrose’s original proposal,1 we
will still refer to it as the “generalized Penrose process” since ultimately the energy
comes from the spin of the BH.
It is not easy to prove that the generalized Penrose process is necessarily in oper-
ation in a given jet. The reason is that jets are always associated with accretion disks,
and the accretion process itself releases gravitational energy, some of which might
flow into the jet. Let us define a jet efficiency factor ηjet ,
≈Pjet ⊗
ηjet = , (1)
≈ Ṁ(rH )⊗c2
where ≈Pjet ⊗ is the time-average power flowing out through the jet and ≈ Ṁ(rH )⊗c2
is the time-average rate at which rest-mass energy flows in through the BH horizon.
Many jets, both those observed and those seen in computer simulations, have values
of ηjet quite a bit less than unity. With such a modest efficiency, the jet power could
easily come from the accretion disk [13–15].
The situation has improved considerably in the last couple of years. As we show
in Sect. 2, numerical simulations have now been carried out where it can be demon-
strated beyond reasonable doubt that the simulated jet obtains power directly from
the BH spin energy. Furthermore, as we discuss in Sect. 3, the first observational
evidence for a correlation between jet power and BH spin has finally been obtained.
The correlation appears to favor a Penrose-like process being the energy source of
jets.
2 Computer Simulations of Black Hole Accretion and Jets
For the last decade or so, it has been possible to simulate numerically the dynamics
of MHD accretion flows in the fixed Kerr metric of a spinning BH. The dynamics
1 Penrose considered a simple model in which particles on negative energy orbits fall into a spinning
BH. Wagh and Dadhich [12] extended the analysis to discrete particle accretion in the presence of
a magnetic field, which introduces additional interesting effects. We do not discuss these particle-
based mechanisms, but focus purely on fluid dynamical processes within the magnetohydrodynamic
(MHD) approximation. We also do not discuss an ongoing controversy on whether or not different
mechanisms based on magnetized fluids differ from one another [6].
Energy Extraction from Spinning Black Holes Via Relativistic Jets 525
of the magnetized fluid are described using the general relativistic MHD (GRMHD)
equations in a fixed space-time, and the simulations are carried out in 3D in order
to capture the magnetorotational instability (MRI), the agency that drives accretion
[16]. Radiation is usually ignored, but this is not considered a problem since jets
are usually found in systems with geometrically thick accretion disks, which are
radiatively inefficient. We describe here one set of numerical experiments [10, 17]
which have been run using the GRMHD code HARM [18] and which are particularly
relevant for understanding the connection between the generalized Penrose process
and jets.
As is standard, the numerical simulations are initialized with an equilibrium gas
torus orbiting in the equatorial plane of a spinning BH. The torus is initially embedded
with a weak seed magnetic field, as shown in panel (a) of Fig. 1. Once the simulation
begins, the magnetic field grows as a result of the MRI [16]. This leads to MHD
turbulence, which in turn drives accretion of mass and magnetic field into the BH.
We define the mass accretion rate,

Ṁ(r ) = − ρu r d Aθ,ϕ , (2)
θ,ϕ
where the integration is over all angles on a sphere of radius r (Boyer-Lindquist

◦
or Kerr-Schild coordinates), d Aθ,ϕ = −gdθ dϕ is the surface area element, ρ
is the density, u r is the contravariant radial component of 4-velocity, and g is the
determinant of the metric. The sign in Eq. (2) is chosen such that Ṁ > 0 corresponds
to mass inflow. One is usually interested in the mass accretion rate at the horizon,
Ṁ(r = rH ). In computing Ṁ(rH ) from the simulations, one waits until the system
has reached approximate steady state. One then computes Ṁ(rH ) over a sequence of
many snapshots in time and then averages to eliminate turbulent fluctuations. This
gives the time-average mass accretion rate ≈ Ṁ(rH )⊗.
Panels (b)–(d) in Fig. 1 show the time evolution of the accretion flow and jet in
a simulation with BH spin a√ ∗ a/M = 0.99, where M is the BH mass [10]. The
steady accretion of magnetized fluid causes magnetic field to accumulate in the inner
regions near the BH. After a while, the field becomes so strong that it compresses the
inner part of the otherwise geometrically thick accretion flow into a thin sheet (panel
b). The effect is to obstruct the accretion flow (panels c and d), leading to what is
known as a magnetically-arrested disk [19, 20] or a magnetically choked accretion
flow [21]. The strong field extracts BH spin energy and forms a powerful outflow.
To understand the energetics, consider the rate of flow of energy,

Ė(r ) = Ttr d Aθ,ϕ , (3)
θ,ϕ
where the stress-energy tensor of the magnetized fluid is

b2 b2 bμ bν
Tνμ = ρ + u g + pg + μ
u uν + p + δνμ − , (4)
4π 8π 4π
(a) (b) (c) (d)
(e)
η
Fig. 1 Formation of a magnetically-arrested disk and ejection of powerful jets in a GRMHD

simulation of magnetized accretion on to a rapidly spinning BH with a√ = 0.99 [10]. The top and
bottom rows in panels (a-d) show a time sequence of equatorial and meridional slices through the
accretion flow. Solid lines show magnetic field lines in the image plane, and color shows log ρ (red
high, blue low). The simulation starts with an equilibrium torus embedded with a weak magnetic
field (panel a). The weakly magnetized orbiting gas is unstable to the MRI, which causes gas and
field to accrete. As large-scale magnetic flux accumulates at the center, a coherent bundle of field
lines forms at the center, which threads the BH and has the configuration of bipolar funnels along
the (vertical) BH rotation axis. These funnels contain strong field and low mass density (lower
panels b, c, d). Helical twisting of the field lines as a result of dragging of frames causes a powerful
outflow of energy through the funnels in the form of twin jets. The outflow efficiency η (panel e),
calculated as in Eq. (5), becomes greater than unity once the flow achieves quasi-steady state at
time t 5000r g /c. This is the key result of the simulation. Having a time-average η > 1 means
that there is a net energy flow out of the BH, i.e., spin energy is extracted from the BH by the
magnetized accretion flow. This constitutes a demonstration of the generalized Penrose process in
the astrophysically relevant context of a magnetized accretion flow
u g and pg are the internal energy and pressure of the gas, bμ is the fluid-frame
magnetic field 4-vector (see Gammie and McKinney [18] for the definition), and
b2 = bμ bμ is the square of the fluid-frame magnetic field strength. The sign of
Eq. (3) is chosen such that Ė(r ) > 0 corresponds to energy inflow. Note that Ttr
includes the inflow of rest mass energy via the term ρu r u t .
Let us define the efficiency with which the accreting BH produces outflowing
energy as
Ṁ(rH )c2 − Ė(rH )
η= , (5)
≈ Ṁ(rH )⊗c2
where we have made η dimensionless by normalizing the right-hand side by the

time-average mass energy accretion rate. To understand the meaning of Eq. (5), con-
sider the simple example of gas falling in radially from infinity, with no radiative
or other energy losses along the way. In this case, we have Ė(rH ) = Ṁ(rH )c2 , i.e.,
the gas carries an energy equal to its rest mass energy into the BH. Hence η = 0,
as appropriate for this example. For a more realistic accretion flow, some energy is
lost by the gas via radiation, winds and jets, and one generally expects the energy
flowing into the BH to be less than the rest mass energy: Ė(rH ) < Ṁ(rH )c2 . This
will result in an efficiency η > 0, where η measures the ratio of the energy returned
to infinity, Ṁ(rH )c2 − Ė(rH ), to the energetic price paid in the form of rest mass
energy flowing into the BH, Ṁ(rH )c2 .
Usually, Ė(rH ) is positive, i.e., there is a net flow of energy into the BH through
the horizon, and η < 1. However, there is no theorem that requires this. Penrose’s [1]
great insight was to realize that it is possible to have Ė(rH ) < 0 (net outward energy
flow as measured at the horizon), and thus η > 1. In the context of an accretion
flow, Ė(rH ) < 0 means that, even though rest mass flows steadily into the BH,
there is a net energy flow out of the BH. As a result, the gravitational mass of the
BH decreases with time. It is the energy associated with this decreasing mass that
enables η to exceed unity. Of course, as the BH loses gravitational mass, it also loses
angular momentum and spins down. This can be verified by considering the angular
momentum flux at the horizon, J˙(rH ), which may be computed as in Eq. (3) but with
Ttr replaced by Tφr (e.g., Tchekhovskoy and McKinney [22]).
Returning to the simulation under consideration, Fig. 1e shows the outflow effi-
ciency η as a function of time. It is seen that the average efficiency exceeds unity
once the flow achieves steady state at time t 5000r g /c, where r g = G M/c2 . The
outflow thus carries away more energy than the entire rest mass energy brought in by
the accretion flow. This is an unambiguous demonstration of the generalized Penrose
process in the astrophysically plausible setting of a magnetized accretion flow on to
a spinning BH. Of course, it is not obvious that the energy necessarily flows out in
a collimated relativistic jet. The quantity η is defined via global integrals (Eqs. 2, 5)
and it does not specify exactly where the outflowing energy ends up. A more detailed
analysis reveals that the bulk of the energy does indeed go into a relativistic jet, while
about 10 % goes into a quasi-relativistic wind [17].
Figure 1 corresponds to an extreme example, viz., a very rapidly spinning BH with
a√ = 0.99. Figure 2 shows results from a parameter study that investigated the effect
of varying a√ . It is seen that the time-average η increases steeply with increasing a√ .
For an accretion flow that corotates with the BH, the power going into the jet can be
well-fit with a power-law dependence,
ηjet ⇔ 0.65a√2 (1 + 0.85a√2 ). (6)
This approximation remains accurate to within 15 % for 0.3 ≤ a√ ≤ 1. For low

spins, the net efficiency derived from the simulations is greater than that predicted
by Eq. (6). For example, as Fig. 2 shows, the simulation gives a non-zero value of η
for a√ = 0, which is inconsistent with Eq. (6). This is because, for a√ = 0, all the
_ _
Fig. 2 Time-average outflow efficiency η versus BH spin parameter a√ for a sequence of GRMHD
simulations of non-radiative BH accretion flows [17]. The efficiency exceeds unity for a√ 0.9.
Negative values of a√ correspond to the accretion flow counter-rotating with respect to the BH
outflow energy comes directly from the accretion flow, most of which goes into a
wind. Nothing comes from the BH, whereas Eq. (6) refers specifically to the efficiency
ηjet associated with jet power from the BH. With increasing BH spin, both the disk
and the hole contribute to energy outflow, with the latter becoming more and more
dominant. For spin values a√ > 0.9, the BH’s contribution is so large that the net
efficiency exceeds unity.
Before leaving this topic, we note that other numerical simulations have used
geometrically thicker accretion configurations than the one shown in Fig. 1 and find
even larger values of η [21, 23].
3 Empirical Evidence for the Generalized Penrose Process
As discussed in Sect. 2, there is definite evidence from computer simulations that

the generalized Penrose process is feasible, and even quite plausible, with magne-
tized accretion flows. We discuss here recent progress on the observational front. In
Sect. 3.1 we briefly summarize efforts to measure spin parameters of astrophysical
BHs. Then in Sect. 3.2 we discuss a correlation that has been found between jet power
and BH spin. Finally in Sect. 3.3 we explain why we think the observational evidence
favors a Penrose-like process rather than disk power.
Fig. 3 Radius of the ISCO RISCO and of the horizon RH in units of G M/c2 plotted as a function of
the black hole spin parameter a√ . Negative values of a√ correspond to retrograde orbits. Note that
RISCO decreases monotonically from 9G M/c2 for a retrograde orbit around a maximally spinning
black hole, to 6G M/c2 for a non-spinning black hole, to G M/c2 for a prograde orbit around a
maximally spinning black hole
3.1 Spin Parameters of Stellar-Mass Black Holes
In 1989, the first practical approach to measuring black hole spin was suggested
[24], viz., modeling the relativistically-broadened Fe K emission line emitted from
the inner regions of an accretion disk. The first compelling observation of such a line
was reported 6 years later [25]. Presently, the spins of more than a dozen black holes
have been estimated by modeling the Fe K line (see Reynolds et al. [26] for a recent
review).
In 1997, a second approach to measuring black hole spin, the “continuum-fitting
method,” was proposed [27]. In this method, one fits the thermal continuum spectrum
of a black hole’s accretion disk to the relativistic model of Novikov and Thorne [28].
One then identifies the inner edge of the modeled disk with the radius RISCO of the
innermost stable circular orbit (ISCO) in the space-time metric. Since RISCO varies
monotonically with respect to the dimensionless BH spin parameter a√ (see Fig. 3),
a measurement of the former immediately provides an estimate of the latter.
In 2006, the continuum-fitting method was employed to estimate the spins of three
stellar-mass BHs [30, 31]. Seven additional spins have since been measured. Table 1
lists the masses and spins of these ten BHs. Readers are referred to a recent review
by the authors [29] for details of the continuum-fitting method and uncertainties in
the derived spin estimates.
The continuum-fitting method is simple and demonstrably robust. It does not make
many assumptions; those few it makes have nearly all been tested and shown to be
valid (see Steiner et al. [29, 32] for details). A significant limitation of the method
is that it is only readily applicable to stellar-mass BHs. For such BHs, however, we
would argue that it is the method of choice. The Fe K method can be applied to both
Table 1 The spins and

System a√ M/M≡
masses of ten stellar-mass
black holes [29] Persistent
Cygnus X-1 >0.95 15.8 ± 1.0
+0.05
LMC X-1 0.92−0.07 10.9 ± 1.4
M33 X-7 0.84 ± 0.05 15.65 ± 1.45
Transient
GRS 1915+105 >0.95 10.1 ± 0.6
4U 1543–47 0.80 ± 0.10 9.4 ± 1.0
GRO J1655–40 0.70 ± 0.10 6.3 ± 0.5
XTE J1550-564 0.34 ± 0.24 9.1 ± 0.6
H1743-322 0.2 ± 0.3 ∇8
LMC X-3 <0.3 7.6 ± 1.6
A0620-00 0.12 ± 0.19 6.6 ± 0.25
stellar-mass and supermassive BHs. For the latter, it is the only method currently
available.
3.2 Correlation Between Black Hole Spin and Jet Radio Power
The 10 stellar-mass BHs in Table 1 are divided into two classes: “persistent” sources,
which are perennially bright in X-rays at a relatively constant level, and “transient”
sources, which have extremely large amplitude outbursts. During outburst, the tran-
sient sources generally reach close to the Eddington luminosity limit (see [33] for a
quantitative discussion of this point). Close to the peak, these systems eject blobs of
plasma that move ballistically outward at relativistic speeds (Lorentz factor Γ > 2).
These ballistic jets are often visible in radio and sometimes in X-rays out to distances
of order a parsec from the BH, i.e., to distances > 1010 G M/c2 . Because ballistic
jets resemble the kiloparsec-scale jets seen in quasars, stellar-mass BHs that produce
them are called microquasars [34].
On general principles, one expects jet power to depend on the BH mass M, its spin
a√ , and the mass accretion rate Ṁ (plus perhaps other qualitative factors such as the
topology of the magnetic field [35, 36]). If one wishes to investigate the dependence
of jet power on a√ , one needs first to eliminate the other two variables. Ballistic
jets from transient stellar-mass BHs are very well-suited for this purpose. First, the
BH masses are similar to better than a factor of two (see Table 1). Second, all these
sources have similar accretion rates, close to the Eddington limit, at the time they
eject their ballistic jets [33]. This leaves a√ as the only remaining variable.
(a) (b) a* : -1 0 0.5 0.7 0.9 0.99

1000
100
Jet Power
10
1
Γ =2
0.1 Γ =5
0.01
9 7 6 5 4 3 2
RISCO / M
Fig. 4 a Plot of jet power, estimated from 5 GHz radio flux at light curve maximum, versus black
hole spin, measured via the continuum-fitting method, for five transient stellar-mass BHs [33,
37]. The dashed line has slope fixed to 2 (see Eq. 6) and is not a fit. b Plot of jet power versus
RISCO /(G M/c2 ). Here jet power has been corrected for beaming assuming jet Lorentz factor Γ = 2
(filled circles) or Γ = 5 (open circles). The two solid lines correspond to fits of a relation of the
form, “Jet Power” ⇒ ΩH2 , where ΩH is the angular frequency of the black hole horizon [33]. Note
that jet power varies by a factor 1000 among the five objects shown
Narayan and McClintock [37] considered the peak radio luminosities of ballistic
jet blobs in four transient BHs, A0620-00, XTE J1550-564, GRO J1655-40, GRS
1915+105, and showed that they correlate well with the corresponding black hole
spins measured via the continuum-fitting method.2 Later, Steiner et al. [33] included
a fifth BH, H1743-322, whose spin had been just measured. Figure 4a shows the
inferred ballistic jet powers of these five objects plotted versus black hole spin. The
quantity “Jet Power” along the vertical axis refers to (ν Sν )D 2 /M, where ν (= 5 GHz)
is the radio frequency at which the measurements are made, Sν is the flux density
in janskys at this frequency at the peak of the ballistic jet radio light curve, D is
the distance in kiloparsecs, and M is the black hole mass in solar units. There is
unmistakeable evidence for a correlation between jet power and a√ . Although there
are only five data points, note that jet power varies by nearly three orders of magnitude
as the spin parameter varies from ⇔ 0.1 to 1.
The very unequal horizontal errorbars in Fig. 4a are a feature of the continuum-
fitting method of measuring a√ . Recall that the method in effect measures RISCO and
then deduces the value of a√ using the mapping shown in Fig. 3. Since the mapping
is highly non-linear, especially as a√ ∈ 1, comparable errors in RISCO correspond to
vastly different uncertainties in a√ . In addition, the use of log a√ along the horizontal
axis tends to stretch errorbars excessively for low spin values. Figure 4b, based on
2 In the case of a fifth transient BH, 4U1543-47, radio observations did not include the peak of the
light curve, so one could only deduce a lower limit to the jet power, which is shown as an open
circle in Fig. 4a.
[33], illustrates these effects. Here the horizontal axis tracks log RISCO rather than
log a√ , and the horizontal errorbars are therefore more nearly equal. The key point
is, regardless of how one plots the data, the correlation between jet power and black
hole spin appears to be strong.
3.3 Why Generalized Penrose Process?
Assuming the correlation shown in Fig. 4 is real, there are two immediate implica-
tions: (i) Ballistic jets in stellar-mass BHs are highly sensitive to the spins of their
underlying BHs. (ii) Spin estimates of stellar-mass BHs obtained via the continuum-
fitting method are sufficiently reliable to reveal this long-sought connection between
relativistic jets and BH spin.
With respect to (i), the mere existence of a correlation does not necessarily imply
that the generalized Penrose process is at work. We know that the accretion disk itself
is capable of producing a jet-like outflow [13–15]. Furthermore, the gravitational
potential well into which an accretion disk falls becomes deeper with increasing BH
spin, since the inner radius of the disk RISCO becomes smaller (Fig. 3). Therefore, a
disk-driven jet is likely to become more powerful with increasing spin. Could this be
the reason for the correlation between jet power and spin seen in Fig. 4? We consider
it unlikely. The radiative efficiency ηdisk of a Novikov-Thorne thin accretion disk
increases only modestly with spin; for the spins of the five objects shown in Fig. 4,
ηdisk = 0.061, 0.069, 0.072, 0.10 and 0.19, respectively, varying by only a factor
of three. Of course, there is no reason why the power of a disk-driven jet should
necessarily scale like ηdisk . Nevertheless, the fact that ηdisk shows only a factor of
three variation makes it implausible that a disk-powered jet could vary in power by
three orders of magnitude.
In contrast, any mechanism that taps directly into the BH spin energy via some kind
of generalized Penrose process can easily account for the observed variation in jet
power. Analytical models of magnetized accretion predict that the jet efficiency factor
should vary as ηjet ⇒ a√2 [2, 3] or ηjet ⇒ ΩH2 [38], where ΩH is the angular frequency
of the BH horizon.3 The dashed line in Fig. 4a corresponds to the former scaling,
and the solid lines in Fig. 4b to the latter scaling; Equation (6), which is obtained
by fitting simulation results, is intermediate between the two. The observational
data agree remarkably well with the predicted scalings, strongly suggesting that the
generalized Penrose process is in operation.
We cannot tell whether the energy extraction in the observed systems is mediated
specifically by magnetic fields (as in the simulations), since there is no way to observe
what is going on near the BH (where the jet is initially launched). Where the ballistic
jet blobs are finally observed they are clearly magnetized—it is what enables the
charged particles to produce radiation via the synchrotron mechanism—but this is at
distances ∇ 1010 G M/c2 .
3 The two scalings agree for small values of a√ , but differ as a√ ∈ 1.

4 Summary
In summary, the case for a generalized version of the Penrose process being the
power source behind astrophysical jets has become significantly stronger in the last
few years. Computer simulations have been very helpful in this regard since they
enable one to study semi-realistic configurations of magnetized accretion flows and
to explore quantitatively how mass, energy and angular momentum flow through
the system. Recent computer experiments find unambiguous indications for energy
extraction from spinning BHs via magnetic fields. Whether these simulated models
describe real BHs in nature is not yet certain. However, completely independent
observational data suggest a link between the spins of transient stellar-mass BHs and
the energy output in ballistic jets ejected from these systems. The jet power increases
steeply with BH spin (Fig. 4), and the dependence is quite similar to that found both
in simple analytical models [2, 3] and in simulations (Fig. 2). Taking all the evidence
into account, the authors believe that Penrose’s seminal ideas on energy extraction
from spinning BHs are relevant for the production of at least some categories of
relativistic astrophysical jets.
RN’s work was supported in part by NASA grant NNX11AE16G. AT was sup-
ported by a Princeton Center for Theoretical Science fellowship and an XSEDE
allocation TG-AST100040 on NICS Kraken and Nautilus and TACC Ranch.
References
1. Penrose, R.: Gravitational collapse: the role of general relativity. Riv. Nuovo Cimento 1, 252
(1969)
2. Ruffini, R., Wilson, J.R.: Relativistic magnetohydrodynamical effects of plasma accreting into
a black hole. Phys. Rev. D 12, 2959 (1975). doi:10.1103/PhysRevD.12.2959
Mon. Not. R. Astron. Soc. 179, 433 (1977)
4. Damour, T., Ruffini, R., Hanni, R.S., Wilson, J.R.: Regions of magnetic support of a plasma
around a black hole. Phys. Rev. D 17, 1518 (1978). doi:10.1103/PhysRevD.17.1518
5. Koide, S., Shibata, K., Kudoh, T., Meier, D.L.: Extraction of black hole rotational energy by
a magnetic field and the formation of relativistic jets. Science 295, 1688 (2002). doi:10.1126/
science.1068240
6. Komissarov, S.S.: Blandford-Znajek mechanism versus Penrose process. J. Korean Phys. Soc.
54, 2503 (2009). doi:10.3938/jkps.54.2503
7. McKinney, J.C., Gammie, C.F.: A measurement of the electromagnetic luminosity of a Kerr
black hole. Astrophys. J. 611, 977 (2004). doi:10.1086/422244
8. McKinney, J.C.: General relativistic magnetohydrodynamic simulations of the jet formation
and large-scale propagation from black hole accretion systems. Mon. Not. R. Astron. Soc. 368,
1561 (2006). doi:10.1111/j.1365-2966.2006.10256.x
9. Beskin, V.S.: MHD Flows in Compact Astrophysical Objects. Astronomy and astrophysics
library. Springer, (2010). doi:10.1007/978-3-642-01290-7
10. Tchekhovskoy, A., Narayan, R., McKinney, J.C.: Efficient generation of jets from magnetically
arrested accretion on a rapidly spinning black hole. Mon. Not. R. Astron. Soc. 418, L79 (2011).
doi:10.1111/j.1745-3933.2011.01147.x
11. Meier, D.L.: Black Hole Astrophysics: The Engine Paradigm. Praxis Books in Astronomy and
Planetary Sciences. Springer, Heidelberg; New York (2012)
12. Wagh, S.M., Dadhich, N.: The energetics of black holes in electromagnetic fields by the Penrose
process. Phys. Rep. 183, 137 (1989). doi:10.1016/0370-1573(89)90156-7
13. Blandford, R.D., Payne, D.G.: Hydromagnetic flows from accretion discs and the production
of radio jets. Mon. Not. R. Astron. Soc. 199, 883 (1982)
14. Ghosh, P., Abramowicz, M.A.: Electromagnetic extraction of rotational energy from disc-fed
black holes: the strength of the Blandford-Znajek process. Mon. Not. R. Astron. Soc. 292, 887
(1997)
15. Livio, M., Ogilvie, G.I., Pringle, J.E.: Extracting energy from black holes: the relative impor-
tance of the Blandford-Znajek mechanism. Astrophys. J. 512, 100 (1999). doi:10.1086/306777
17. Tchekhovskoy, A., McKinney, J.C., Narayan, R.: General relativistic modeling of magnetized
jets from accreting black holes. J. Phys. Conf. Ser. 372, 012040 (2012). doi:10.1088/1742-
6596/372/1/012040
18. Gammie, C.F., McKinney, J.C., Tóth, G.: HARM: a numerical scheme for general relativistic
magnetohydrodynamics. Astrophys. J. 589, 444 (2003). doi:10.1086/374594
19. Narayan, R., Igumenshchev, I.V., Abramowicz, M.A.: Magnetically arrested disk: an energet-
ically efficient accretion flow. Publ. Astron. Soc. Jpn 55, L69 (2003)
20. Igumenshchev, I.V.: Magnetically arrested disks and the origin of Poynting jets: a numerical
study. Astrophys. J. 677, 317 (2008). doi:10.1086/529025
21. McKinney, J.C., Tchekhovskoy, A., Blandford, R.D.: General relativistic magnetohydrody-
namic simulations of magnetically choked accretion flows around black holes. Mon. Not. R.
Astron. Soc. 423, 3083 (2012). doi:10.1111/j.1365-2966.2012.21074.x
22. Penna, R.F., McKinney, J.C., Narayan, R., et al.: Simulations of magnetized discs around black
holes: effects of black hole spin, disc thickness and magnetic field geometry. Mon. Not. R.
Astron. Soc. 408, 752 (2010). doi:10.1111/j.1365-2966.2010.17170.x
23. Tchekhovskoy, A., McKinney, J.C.: Prograde and retrograde black holes: whose jet
is more powerful? Mon. Not. R. Astron. Soc. 423, L55 (2012). doi:10.1111/j.1745-3933.2012.
01256.x
24. Fabian, A.C., Rees, M.J., Stella, L., White, N.E.: X-ray fluorescence from the inner disc in
Cygnus X-1. Mon. Not. R. Astron. Soc. 238, 729 (1989)
25. Tanaka, Y., Nandra, K., Fabian, A.C., et al.: Gravitationally redshifted emission implying an
accretion disk and massive black hole in the active galaxy MCG-6-30-15. Nature 375, 659
(1995). doi:10.1038/375659a0
26. Reynolds, C.S., Brenneman, L.W., Lohfink, A.M., et al.: Probing relativistic astrophysics
around SMBHs: The Suzaku AGN spin survey. In: Petre, R., Mitsuda, K., Angelini, L.
(eds.) SUZAKU 2011: Exploring the X-ray Universe: Suzaku and Beyond, AIP Conference
Proceedings, vol. 1427, pp. 157–164. American Institute of Physics, Melville, NY (2012).
doi:10.1063/1.3696170
27. Zhang, S.N., Cui, W., Chen, W.: Black hole spin in X-ray binaries: observational consequences.
Astrophys. J. Lett. 482, L155 (1997). doi:10.1086/310705
28. Novikov, I.D., Thorne, K.S.: Astrophysics of Black Holes. In: DeWitt, C., DeWitt, B.S. (eds.)
Black Holes. Les Astres Occlus, pp. 343–450. Gordon and Breach, New York; London (1973)
29. McClintock, J.E., Narayan, R.: Black Hole Spin Via Continuum Fitting and the Role of Spin
in Powering Transient Jets. Space Sci Rev. (2013). doi:10.1007/s11214-013-0003-9
30. Shafee, R., McClintock, J.E., Narayan, R., et al.: Estimating the spin of stellar-mass black holes
by spectral fitting of the X-ray continuum. Astrophys. J. Lett. 636, L113 (2006). doi:10.1086/
498938
31. McClintock, J.E., Shafee, R., Narayan, R., et al.: The spin of the near-extreme Kerr black hole
GRS 1915+105. Astrophys. J. 652, 518 (2006). doi:10.1086/508457
32. McClintock, J.E., Narayan, R., Davis, S.W., et al.: Measuring the spins of accreting black holes.
33. Steiner, J.F., McClintock, J.E., Narayan, R.: Jet power and black hole spin: testing an empirical
relationship and using it to predict the spins of six black holes. Astrophys. J. 762, 104 (2013).
doi:10.1088/0004-637X/762/2/104
34. Mirabel, I.F., Rodríguez, L.F.: Sources of relativistic jets in the Galaxy. Annu. Rev. Astron.
Astrophys. 37, 409 (1999). doi:10.1146/annurev.astro.37.1.409
35. Beckwith, K., Hawley, J.F., Krolik, J.H.: The influence of magnetic field geometry on the
evolution of black hole accretion flows: similar disks, drastically different jets. Astrophys. J.
678, 1180 (2008). doi:10.1086/533492
36. McKinney, J.C., Blandford, R.D.: Stability of relativistic jets from rotating, accreting black
holes via fully three-dimensional magnetohydrodynamic simulations. Mon. Not. R. Astron.
Soc. 394, L126 (2009). doi:10.1111/j.1745-3933.2009.00625.x
37. Narayan, R., McClintock, J.E.: Observational evidence for a correlation between jet power and
black hole spin. Mon. Not. R. Astron. Soc. 419, L69 (2012). doi:10.1111/j.1745-3933.2011.
01181.x
38. Tchekhovskoy, A., Narayan, R., McKinney, J.C.: Black hole spin and the radio loud/quiet
dichotomy of active galactic nuclei. Astrophys. J. 711, 50 (2010). doi:10.1088/0004-637X/
711/1/50

General Relativity, Cosmology and Astrophysics Perspectives 100 Years After Einstein's Stay in Prague

Uploaded by

Copyright:

Available Formats

General Relativity, Cosmology and Astrophysics Perspectives 100 Years After Einstein's Stay in Prague

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

General Relativity, Cosmology and Astrophysics Perspectives 100 Years After Einstein's Stay in Prague

Uploaded by

Copyright:

Available Formats

Fundamental Theories of Physics 177

For further volumes:

ISBN 978-3-319-06348-5 ISBN 978-3-319-06349-2 (eBook)

Library of Congress Control Number:2014938217

Springer International Publishing Switzerland 2014

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Gravity and Prague

Classical General Relativity

The problem of measurement, of precise definition of observers and observables

hyperbolic metamaterials, gravitational kinks, Bloch walls, liquid crystal droplets,

It has been a long-standing problem whether the spin–spin interaction between

Cosmology and Quantum Gravity

Numerical Relativity and Relativistic Astrophysics

Relativistic astrophysics celebrated its 50th birthday in December 2013. Seven

eLISA, which should be launched if the LISA Pathfinder is successful. The

Prague, December 2013 Jiří Bičák

Part I Gravity and Prague

Kepler and Mach’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Einstein in Prague: Relativity Then and Now . . . . . . . . . . . . . . . . . . . 33

Part II Classical General Relativity

Observers, Observables and Measurements in General Relativity . . . . 67

Some Links Between General Relativity and Other

The General Relativistic Two Body Problem and the Effective

Gravitational Self-Force: Orbital Mechanics Beyond

Hamiltonian Formalism for Spinning Black Holes

Stability of Marginally Outer Trapped Surfaces and Geometric

Stationary Black-Hole Binaries: A Non-existence Proof . . . . . . . . . . . 209

Dynamic and Thermodynamic Stability of Black Holes

Instability of Anti-de Sitter Spacetime . . . . . . . . . . . . . . . . . . . . . . . . 239

Higher-Dimensional Black Holes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245

Black Holes, Hidden Symmetry and Complete Integrability:

Part III Cosmology and Quantum Gravity

Cosmological Models and Stability. . . . . . . . . . . . . . . . . . . . . . . . . . . 277

Inflation and Birth of Cosmological Perturbations . . . . . . . . . . . . . . . 305

Loop Quantum Gravity and the Planck Regime of Cosmology . . . . . . 323

The Inflationary Origin of the Seeds of Cosmic Structure:

Quantum Gravity: The View From Particle Physics . . . . . . . . . . . . . . 369

Part IV Numerical Relativity and Relativistic Astrophysics

Three Little Pieces for Computer and Relativity. . . . . . . . . . . . . . . . . 391

Instabilities of Relativistic Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

Gravity Talks: Observing the Universe with Gravitational Waves. . . . 459

Einstein’s Gravity as Seen by a Cosmic Lighthouse Keeper . . . . . . . . 483

The Astrophysical Signatures of Black Holes: The Horizon,

J. Bičák and T. Ledvinka (eds.), General Relativity, Cosmology and Astrophysics, 3

Fig. 1 The eccentricity e is E

I begin with some background to Kepler’s discovery of the laws of planetary

2 Some Important Facts of Planetary Motion

1 See [1] for a detailed discussion of the history.

Fig. 3 Illustration of the

3 Greek Planetary Astronomy

5 Acronychal and non-acronychal observations were still vital in Kepler’s work.

Final comments before we move on to Copernicus and Kepler. First, all of

It is ironic that Copernicus stumbled on his revolutionary idea by trying to undo

Fig. 7 Johannes Kepler. The

Fig. 8 This diagram

as those 2◦ corresponding to twice the Earth’s eccentricity. Brahe had continued

tendencies—rectilinear inertial motion and gravitationally induced deflection from

Proposition 1 The areas which revolving bodies describe by radii drawn to an

7 Intermezzo: Christian Doppler

However, he soonrealized that one gets contradictions with conservation of energy