Modified Gravity and Cosmology

Timothy Cliftona , Pedro G. Ferreiraa , Antonio Padillab , Constantinos Skordisb

a Department of Astrophysics, University of Oxford, UK.
b School of Physics and Astronomy, University of Nottingham, UK.
arXiv:1106.2476v3 [astro-ph.CO] 12 Mar 2012

Abstract
In this review we present a thoroughly comprehensive survey of recent work on modi-
fied theories of gravity and their cosmological consequences. Amongst other things, we
cover General Relativity, Scalar-Tensor, Einstein-Aether, and Bimetric theories, as well
as TeVeS, f (R), general higher-order theories, Hořava-Lifschitz gravity, Galileons, Ghost
Condensates, and models of extra dimensions including Kaluza-Klein, Randall-Sundrum,
DGP, and higher co-dimension braneworlds. We also review attempts to construct a Pa-
rameterised Post-Friedmannian formalism, that can be used to constrain deviations from
General Relativity in cosmology, and that is suitable for comparison with data on the
largest scales. These subjects have been intensively studied over the past decade, largely
motivated by rapid progress in the field of observational cosmology that now allows,
for the first time, precision tests of fundamental physics on the scale of the observable
Universe. The purpose of this review is to provide a reference tool for researchers and stu-
dents in cosmology and gravitational physics, as well as a self-contained, comprehensive
and up-to-date introduction to the subject as a whole.
Keywords: General Relativity, Gravitational Physics, Cosmology, Modified Gravity

Preprint submitted to Physics Reports March 13, 2012

Contents

1 Introduction 5

2 General Relativity, and its Foundations 11

2.1 Requirements for Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.1.1 The foundations of relativistic theories . . . . . . . . . . . . . . . . 11
2.1.2 Observational tests of metric theories of gravity . . . . . . . . . . . 14
2.1.3 Theoretical considerations . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Einstein’s Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2.1 The field equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.2 The action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Alternative Formulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.1 The Palatini procedure . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 Metric-affine gravity and matter . . . . . . . . . . . . . . . . . . . 27
2.3.3 Other approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4.1 Lovelock’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4.2 Birkhoff’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.3 The no-hair theorems . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5 The Parameterised Post-Newtonian Approach . . . . . . . . . . . . . . . . 32
2.5.1 Parameterised post-Newtonian formalism . . . . . . . . . . . . . . 32
2.5.2 Parameterised post-Newtonian constraints . . . . . . . . . . . . . . 34
2.6 Cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.6.1 The Friedmann-Lemaı̂tre-Robertson-Walker solutions . . . . . . . 37
2.6.2 Cosmological distances . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.6.3 Perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.6.4 Gravitational potentials and observations . . . . . . . . . . . . . . 42
2.6.5 The evidence for the ΛCDM model . . . . . . . . . . . . . . . . . . 44
2.6.6 Shortcomings of the ΛCDM model . . . . . . . . . . . . . . . . . . 46

3 Alternative Theories of Gravity with Extra Fields 49

3.1 Scalar-Tensor Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.1 Action, field equations, and conformal transformations . . . . . . . 49
3.1.2 Brans-Dicke theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.1.3 General scalar-tensor theories . . . . . . . . . . . . . . . . . . . . . 59
3.1.4 The chameleon mechanism . . . . . . . . . . . . . . . . . . . . . . 66
3.2 Einstein-Æther Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.2.1 Modified Newtonian dynamics . . . . . . . . . . . . . . . . . . . . 68
3.2.2 Action and field equations . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.3 FLRW solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2.4 Cosmological perturbations . . . . . . . . . . . . . . . . . . . . . . 71
3.2.5 Observations and constraints . . . . . . . . . . . . . . . . . . . . . 73
3.3 Bimetric Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.1 Rosen’s theory, and non-dynamical metrics . . . . . . . . . . . . . 76
3.3.2 Drummond’s theory . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3.3 Massive gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2
 3.3.4 Bigravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.5 Bimetric MOND . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4 Tensor-Vector-Scalar Theories . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4.1 Actions and field equations . . . . . . . . . . . . . . . . . . . . . . 82
3.4.2 Newtonian and MOND limits . . . . . . . . . . . . . . . . . . . . . 84
3.4.3 Homogeneous and isotropic cosmology . . . . . . . . . . . . . . . . 86
3.4.4 Cosmological perturbation theory . . . . . . . . . . . . . . . . . . . 91
3.4.5 Cosmological observations and constraints . . . . . . . . . . . . . . 93
3.5 Other Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.5.1 The Einstein-Cartan-Sciama-Kibble Theory . . . . . . . . . . . . . 96
3.5.2 Scalar-Tensor-Vector Theory . . . . . . . . . . . . . . . . . . . . . 99

4 Higher Derivative and Non-Local Theories of Gravity 101

4.1 f (R) Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.1.1 Action, field equations and transformations . . . . . . . . . . . . . 102
4.1.2 Weak-field limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.1.3 Exact solutions, and general behaviour . . . . . . . . . . . . . . . . 111
4.1.4 Cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.1.5 Stability issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.2 General combinations of Ricci and Riemann curvature. . . . . . . . . . . . 123
4.2.1 Action and field equations . . . . . . . . . . . . . . . . . . . . . . . 123
4.2.2 Weak-field limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.2.3 Exact solutions, and general behaviour . . . . . . . . . . . . . . . . 126
4.2.4 Physical cosmology and dark energy . . . . . . . . . . . . . . . . . 128
4.2.5 Other topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
4.3 Hořava-Lifschitz Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.3.1 The projectable theory . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.3.2 The non-projectable theory . . . . . . . . . . . . . . . . . . . . . . 144
4.3.3 Aspects of Hořava-Lifschitz cosmology . . . . . . . . . . . . . . . . 146
4.3.4 The ΘCDM model . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.3.5 HMT-da Silva theory . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.4 Galileons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.4.1 Galileon modification of gravity . . . . . . . . . . . . . . . . . . . . 151
4.4.2 Covariant galileon . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
4.4.3 DBI galileon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
4.4.4 Galileon cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.4.5 Multi-galileons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
4.5 Other Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.5.1 Ghost condensates . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.5.2 Non-metric gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
4.5.3 Dark energy from curvature corrections . . . . . . . . . . . . . . . 169

5 Higher Dimensional Theories of Gravity 172

5.1 Kaluza-Klein Theories of Gravity . . . . . . . . . . . . . . . . . . . . . . . 172
5.1.1 Kaluza-Klein compactifications . . . . . . . . . . . . . . . . . . . . 173
5.1.2 Kaluza-Klein cosmology . . . . . . . . . . . . . . . . . . . . . . . . 174
5.2 The Braneworld Paradigm . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
3
 5.2.1 The ADD model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
5.3 Randall-Sundrum Gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.3.1 The RS1 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
5.3.2 The RS2 model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
5.3.3 Other RS-like models . . . . . . . . . . . . . . . . . . . . . . . . . 185
5.3.4 Action and equations of motion . . . . . . . . . . . . . . . . . . . . 188
5.3.5 Linear perturbations in RS1 and RS2 . . . . . . . . . . . . . . . . 189
5.4 Brane Cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
5.4.1 Brane based formalism – covariant formulation . . . . . . . . . . . 197
5.4.2 Bulk based formalism – moving branes in a static bulk . . . . . . . 199
5.4.3 Cosmological perturbations . . . . . . . . . . . . . . . . . . . . . . 200
5.5 Dvali-Gabadadze-Porrati Gravity . . . . . . . . . . . . . . . . . . . . . . . 207
5.5.1 Action, equations of motion, and vacua . . . . . . . . . . . . . . . 207
5.5.2 Linear perturbations on the normal branch . . . . . . . . . . . . . 209
5.5.3 Linear perturbations (and ghosts) on the self-accelerating branch . 211
5.5.4 From strong coupling to the Vainshtein mechanism . . . . . . . . . 214
5.5.5 DGP cosmology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
5.6 Higher Co-Dimension Braneworlds . . . . . . . . . . . . . . . . . . . . . . 228
5.6.1 Cascading gravity . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
5.6.2 Degravitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
5.7 Einstein Gauss-Bonnet Gravity . . . . . . . . . . . . . . . . . . . . . . . . 236
5.7.1 Action, equations of motion, and vacua . . . . . . . . . . . . . . . 237
5.7.2 Kaluza-Klein reduction of EGB gravity . . . . . . . . . . . . . . . 240
5.7.3 Co-dimension one branes in EGB gravity . . . . . . . . . . . . . . 240
5.7.4 Co-dimension two branes in EGB gravity . . . . . . . . . . . . . . 244

6 Parameterised Post-Friedmannian Approaches and Observational Con-

straints 246
6.1 The Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.1.1 Evolution of perturbations on super-horizon scales . . . . . . . . . 248
6.1.2 The simplified PPF approach, and its extensions . . . . . . . . . . 249
6.1.3 The Hu-Sawicki frame-work . . . . . . . . . . . . . . . . . . . . . . 253
6.2 Models for µ and ζ on Sub-Horizon Scales . . . . . . . . . . . . . . . . . . 254
6.2.1 The importance of shear . . . . . . . . . . . . . . . . . . . . . . . . 254
6.2.2 The growth function . . . . . . . . . . . . . . . . . . . . . . . . . . 255
6.2.3 Current constraints on the PPF parameters . . . . . . . . . . . . . 257
6.2.4 Constraining the growth rate . . . . . . . . . . . . . . . . . . . . . 258
6.2.5 The EG diagnostic . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
6.3 Forecasting Constraints from Future Surveys . . . . . . . . . . . . . . . . 260

7 Discussion 262

4
1. Introduction
The General Theory of Relativity is an astounding accomplishment: Together with
quantum field theory, it is now widely considered to be one of the two pillars of modern
physics. The theory itself is couched in the language of differential geometry, and was
a pioneer for the use of modern mathematics in physical theories, leading the way for
the gauge theories and string theories that have followed. It is no exaggeration to say
that General Relativity set a new tone for what a physical theory can be, and has truly
revolutionised our understanding of the Universe.
One of the most striking facts about General Relativity is that, after almost an entire
century, it remains completely unchanged: The field equations that Einstein communica-
tion to the Prussian Academy of Sciences in November 1915 are still our best description
of how space-time behaves on macroscopic scales. These are
8πG
Gµν = Tµν (1)
c4
where Gµν is the Einstein tensor, Tµν is the energy momentum tensor, G is Newton’s
constant, and c is the speed of light. It is these equations that are thought to govern the
expansion of the Universe, the behaviour of black holes, the propagation of gravitational
waves, and the formation of all structures in the Universe from planets and stars all the
way up to the clusters and super-clusters of galaxies that we are discovering today. It
is only in the microscopic world of particles and high energies that General Relativity is
thought to be inadequate. On all other scales it remains the gold standard.
The great success of General Relativity, however, has not stopped alternatives being
proposed. Even during the very early days after Einstein’s publication of his theory there
were proposals being made on how to extend it, and incorporate it in a larger, more
unified theory. Notable examples of this are Eddington’s theory of connections, Weyl’s
scale independent theory, and the higher dimensional theories of Kaluza and Klein. To
some extent, these early papers were known to have been influential on Einstein himself.
They certainly influenced the physicists who came after him.
The ideas developed by Eddington during this period were later picked up by Dirac,
who pointed out the apparent coincidence between the magnitude of Newton’s constant
and the ratio of the mass and scale of the Universe. This relationship between a funda-
mental constant and the dynamical state of a particular solution led Dirac to conjecture
that Newton’s constant may, in fact, be varying with time. The possibility of a varying
Newton’s constant was picked up again in the 1960s by Brans and Dicke who developed
the prototypical version of what are now known as scalar-tensor theories of gravity. These
theories are still the subject of research today, and make up Section 3.1 of our report.
Building on the work of Hermann Weyl, the Soviet physicist Andrei Sakharov pro-
posed in 1967 what would prove to be one of the most enduring theories of modified
gravity. In Sakharov’s approach, the Einstein-Hilbert action, from which the Einstein
field equations can be derived, is simply a first approximation to a much more complicated
action: Fluctuations in space-time itself lead to higher powers corrections to Einstein’s
theory. In 1977 Kellogg Stelle showed formally that these theories are renormalizable in
the presence of matter fields at the one loop level. This discovery was followed by a surge
of interest, that was boosted again later on by the discovery of the potential cosmologi-
cal consequences of these theories, as found by Starobinsky and others. In Section 4 we
review this work.
5
 The idea of constructing a quantum field theory of gravity started to take a front seat
in physics research during the 1970s and 80s, with the rise of super-gravity and super-
string theories. Both of these proposals rely on the introduction of super-symmetry, and
signalled a resurgence in the ideas of Kaluza and Klein involving higher dimensional
spaces. Boosted further by the discovery of D-branes as fundamental objects in string
theories, this avenue of research led to a vastly richer set of structures that one could
consider, and a plethora of proposals were made for how to modify the effective field
equations in four dimensions. In Section 5 we review the literature on this subject.
By the early 1970s, and following the ‘golden age’ of general relativity that took place
in the 1960s, there was a wide array of candidate theories of gravity in existence that
could rival Einstein’s. A formalism was needed to deal with this great abundance of pos-
sibilities, and this was provided in the form of the Parameterised Post-Newtonian (PPN)
formalism by Kenneth Nordtvedt, Kip Thorne and Clifford Will. The PPN formalism
was built on the earlier work of Eddington and Dicke, and allowed for the numerous
theories available at the time to be compared to cutting edge astrophysical observations
such as lunar laser ranging, radio echo, and, in 1974, the Hulse-Taylor binary pulsar.
The PPN formalism provided a clear structure within which one could compare and as-
sess various theories, and has been the benchmark for how theories of gravity should be
evaluated ever since. We will give an outline of the PPN formalism, and the constraints
available within it today, in Section 2.
The limits of General Relativity have again come into focus with the emergence of
the ‘dark universe’ scenario. For almost thirty years there has existed evidence that, if
gravity is governed by Einstein’s field equations, there should be a substantial amount of
‘dark matter’ in galaxies and clusters. More recently, ‘dark energy’ has also been found
to be required in order to explain the apparent accelerating expansion of the Universe.
Indeed, if General Relativity is correct, it now seems that around 96% of the Universe
should be in the form of energy densities that do not interact electromagnetically. Such
an odd composition, favoured at such high confidence, has led some to speculate on the
possibility that General Relativity may not, in fact, be the correct theory of gravity to
describe the Universe on the largest scales. The dark universe may be just another signal
that we need to go beyond Einstein’s theory.
The idea of modifying gravity on cosmological scales has really taken off over the past
decade. This has been triggered, in part, by theoretical developments involving higher
dimensional theories, as well as new developments in constructing renormalizable theories
of gravity. More phenomenologically, Bekenstein’s relativistic formulation of Milgrom’s
Modified Newtonian Dynamics (MoND) has provided a fresh impetus for new study:
What was previously a rule of thumb for how weak gravitational fields might behave in
regions of low acceleration, was suddenly elevated to a theory that could be used to study
cosmology. Insights such as Bertschinger’s realisation that large-scale perturbations in
the Universe can be directly related to the overall expansion rate have also made it
possible to characterise large classes of theories simply in terms of how they make the
Universe evolve. Finally, and just as importantly, there has been tremendous progress
observationally. A key step here has been the measurement of the growth of structure
at redshifts of z ' 0.8, by Guzzo and his collaborators. With these measurements
one can test, and reject, a large number of proposals for modified gravity. This work
is complemented by many others that carefully consider the impact of modifications
to gravity on the cosmic microwave background, weak lensing and a variety of other
6
cosmological probes. As a result, testing gravity has become one of the core tasks of
many current, and future, cosmological missions and surveys.
In this report we aim to provide a comprehensive exposition of the many developments
that have occurred in the field of modified gravity over the past few decades. We will focus
on how these theories differ from General Relativity, and how they can be distinguished
from it, as well as from each other. A vast range of modified theories now exist in the
literature. Some of these have extra scalar, vector or tensor fields in their gravitational
sector; some take Sakharov’s idea in an altogether new direction, modifying gravity in
regions of low, rather than high, curvature; others expand on the ideas first put forward
by Kaluza and Klein, and take them into new realms by invoking new structures. Indeed,
as the reader will see from our table of contents, there are now a great many possible
ways of modifying gravity that can, in principle, be tested against the real Universe. We
will attempt to be as comprehensive in this report as we consider it reasonably possible
to be. That is, we will attempt to cover as many aspects of as many different theories as
we can.
To be able to efficiently assess the different candidate theories of gravity we have opted
to first lay down the foundations of modern gravitational physics and General Relativity
in Section 2. We have aimed to make this a self-contained section that focuses, to some
extent, on why general relativity should be considered ‘special’ among the larger class of
possibilities that we might consider. In this section we also survey the current evidence
for the ‘dark universe’, and explain why it has become the standard paradigm. From here
we move on to discuss and compare alternative theories of gravity and their observational
consequences. While the primary focus of this report is to elucidate particular theories,
we will also briefly delve into the recent attempts that have been made to construct a
formalism, analogous to the PPN formalism, for the cosmological arena. We dub these
approaches ‘Parameterised Post Friedmannian’.
Let us now spell out the conventions and definitions that we will use throughout
this review. We will employ the ‘space-like convention’ for the metric, such that when
it is diagonalised it has the signature (− + ++). We will choose to write space-time
indices using the Greek alphabet, and space indices using the Latin alphabet. Where
convenient, we will also choose to use units such that speed of light is equal to 1. Under
these conventions the line-element for Minkowski space, for example, can then be written

ds2 = ηµν dxµ dxν = −dt2 + dx2 + dy 2 + dz 2 . (2)

For the Riemann and Einstein curvature tensors we will adopt the conventions of Misner,
Thorne and Wheeler [902]:

Rµναβ = ∂α Γµνβ − ∂β Γµνα + Γµσα Γσνβ − Γµσβ Γσνα (3)

1
Gµν = Rµν − gµν R, (4)
2
where Rµν = Rαµαν and R = Rαα . The energy-momentum tensor will be defined with
respect to the Lagrangian density for the matter fields as
2 δLm
T µν = √ , (5)
−g δgµν

7
where the derivative here is a functional one. Throughout this review we will refer to
the energy density of a fluid as ρ, and its isotropic pressure as P . The equation of state,
w, is then defined by
P = wρ. (6)
When writing the Friedmann-Lemaı̂tre-Robertson-Walker (FLRW) line-element we will
use t to denote
R the ‘physical time’ (proper time of observers comoving with the fluid),
and τ = dt/a(t) to denote the ‘conformal time’ coordinate. Unless otherwise stated,
when working with linear perturbations about an FLRW background we will work in the
conformal Newtonian gauge in which
 
ds2 = a2 (τ ) −(1 + 2Ψ)dτ 2 + (1 − 2Φ)qij dxi dxj , (7)

where qij is the metric of a maximally symmetric 3-space with Gaussian curvature κ:

dr2
ds2(3) = qij dxi dxj = + r2 dθ2 + r2 sin2 θdφ2 . (8)
1 − κr2
When dealing with time derivatives in cosmology we will use the dot and prime operators
to refer to derivatives with respect to physical and conformal time, respectively, such that
d
˙ ≡ (9)
dt
0 d
≡ . (10)
dτ
In four dimensional space-time we will denote covariant derivatives with either a semi-
colon or a ∇µ . The four dimensional d’Alembertian will then be defined as

 ≡ g µν ∇µ ∇ν . (11)

On the conformally static three-dimensional space-like hyper-surfaces the grad operator

~ i , while the Laplacian will be given by
will be denoted with an arrow, as ∇
~ i∇
∆ ≡ q ij ∇ ~ j. (12)

As is usual, we will often make use of the definition of the Hubble parameter defined
with respect to both physical and conformal time as
ȧ
H ≡ (13)
a
a0
H ≡ . (14)
a
The definitions we have made here will be restated at various points in the review, so
that each section remains self-contained to a reasonable degree. The exception to this
will be Section 5, on higher dimensional theories, which will require the introduction of
new notation in order to describe quantities in the bulk.
Let us now move onto the definitions of particular terms. We choose to define the
equivalence principles in the following way:
8
 • Weak Equivalence Principle (WEP): All uncharged, freely falling test particles fol-
low the same trajectories, once an initial position and velocity have been prescribed.
• Einstein Equivalence Principle (EEP): The WEP is valid, and furthermore in all
freely falling frames one recovers (locally, and up to tidal gravitational forces) the
same laws of special relativistic physics, independent of position or velocity.

• Strong Equivalence Principle (SEP): The WEP is valid for massive gravitating ob-
jects as well as test particles, and in all freely falling frames one recovers (locally,
and up to tidal gravitational forces) the same special relativistic physics, indepen-
dent of position or velocity.
Of these, the EEP in particular is known to have been very influential in the conception
of General Relativity. One may note that some authors refer to what we have called the
EEP as the ‘strong equivalence principle’.
Let us now define what we mean by ‘General Relativity’. This term is often used
by cosmologists to refer simply to Einstein’s equations. Particle physicists, on the other
hand, refer to any dynamical theory of spin-2 fields that incorporates general covariance
as ‘general relativity’, even if it has field equations that are different from Einstein’s1 . In
this report when we write about ‘General Relativity’ we refer to a theory that simulta-
neously exhibits general covariance, and universal couplings to all matter fields, as well
as satisfying Einstein’s field equations. When we then discuss ‘modified gravity’ this
will refer to any modification of any of these properties. However, it will be clear from
reading through this report that almost all the proposals we report on preserve general
covariance, and the universality of free fall. Let us now clarify further what exactly we
mean by ‘modified’ theories of gravity.
As we will discuss in the next section, the effect of gravity on matter is tightly
constrained to be mediated by interactions of the matter fields with a single rank-2 tensor
field. This does not mean that this field is the only degree of freedom in the theory, but
that whatever other interactions may occur, the effect of gravity on the matter fields can
only be through interactions with the rank-2 tensor (up to additional weak interactions
that are consistent with the available constraints). The term ‘gravitational theory’ can
then be functionally defined by the set of field equations obeyed by the rank-2 tensor,
and any other non-matter fields it interacts with. If these equations are anything other
than Einstein’s equations, then we consider it to be a ‘modified theory of gravity’. We
will not appeal to the action or Lagrangian of the theory itself here; our definition is an
entirely functional one, in terms of the field equations alone.
While we have constructed the definition above to be as simple as possible, there are
of course a number of ambiguities involved. Firstly, exactly what one should consider as
a ‘matter field’ can be somewhat subjective. This is especially true in terms of the exotic
fields that are sometimes introduced into cosmology in order to try and understand
the apparent late-time accelerating expansion of the Universe. Secondly, we have not
defined exactly what we mean by ‘Einstein’s equations’. In four dimensions it is usually
clear what this term refers to, but if we allow for the possibility of extra dimensions

1 Note that under this definition the Einstein-Hilbert and Brans-Dicke Lagrangians, for example,

represent different models of the same theory, which is called General Relativity.

9
then we may choose for it to refer either to the equations derived from an Einstein-
Hilbert action in the higher dimensional space-time, or to the effective set of equations
in four dimensional space-time. Clearly these two possible definitions are not necessarily
consistent with each other. Even in four dimensions it is not always clear if ‘Einstein’s
equations’ include the existence of a non-zero cosmological constant, or not.
To a large extent, the ambiguities just mentioned are a matter of taste, and have no
baring on the physics of the situation. For example, whether one chooses to refer to the
cosmological constant as a modification of gravity, as an additional matter field, or as
part of Einstein’s equations themselves makes no difference to its effect on the expansion
of the Universe. In this case it is only convention that states that the Einstein equations
with Λ is not a modified theory of gravity. Although less established than the case
of the cosmological constant, similar conventions have started to develop around other
modifications to the standard theory. For example, quintessence fields that are minimally
coupled to the metric are usually thought of as additional matter fields, whereas scalar
fields that non-minimally couple to the Einstein-Hilbert term in the action are usually
thought of as being ‘gravitational’ fields (this distinction existing despite what numerous
studies call non-minimally coupled quintessence fields). Although not always clear, we
try to follow what we perceive to be the conventions that exist in the literature in this
regard. We therefore include in this review a section on non-minimally coupled scalar-
tensor theories, but not a section on minimally coupled quintessence fields.

10
2. General Relativity, and its Foundations

General Relativity is the standard theory of gravity. Here we will briefly recap some of
its essential features, and foundations. We will outline the observational tests of gravity
that have been performed on Earth, in the solar system, and in other astrophysical
systems, and we will then explain how and why it is that General Relativity satisfies
them. We will outline why General Relativity should be considered a special theory in
the more general class of theories that one could consider, and will present some of the
theorems it obeys as well as the apparatus that is most frequently used to parameterise
deviations away from it. This will be followed by a discussion of the cosmological solutions
and predictions of the concordance general relativistic ΛCDM model of the Universe.

2.1. Requirements for Validity

In order to construct a relativistic theory of gravity it is of primary importance to
establish the properties it must satisfy in order for it to be considered viable. These
include foundational requirements, such as the universality of free fall and the isotropy
of space, as well as compatibility with a variety of different observations involving the
propagation of light and the orbits of massive bodies. Today, radio and laser signals
can be sent back and forth from the Earth to spacecraft, planets and the moon, and
detailed observations of the orbits of a variety of different astrophysical bodies allow
us to look for ever smaller deviations from Newtonian gravity, as well as entirely new
gravitational effects. It is in this section that we will discuss the gravitational experiments
and observations that have so far been performed in these environments. We will discuss
what they can tell us about relativity theory, and the principles that a theory must obey
in order for it to stand a chance of being considered observationally viable.

2.1.1. The foundations of relativistic theories

First of all let us consider the equivalence principles. We will not insist immediately
that any or all of these principles are valid, but will rather reflect on what can be said
about them experimentally. This will allow us to separate out observations that test
equivalence principles, from observations that test the different gravitational theories
that obey these principles – an approach pioneered by Dicke [423].
The least stringent of the equivalence principles is the WEP. The best evidence in
support of the WEP still comes from Eötvös type experiments that use a torsion balance
to determine the relative acceleration of two different materials towards distant astro-
physical bodies. In reality these materials are self-gravitating, but their mass is usually
small enough that they can effectively be considered to be non-gravitating test particles
in the gravitational field of the astrophysical body. Using beryllium and titanium the
tightest constraint on the relative difference in accelerations of the two bodies, a1 and
a2 , is currently [1110]

|a1 − a2 |
η=2 = (0.3 ± 1.8) × 10−13 . (15)
|a1 + a2 |

This is an improvement of around 4 orders of magnitude on the original results of Eötvös

from 1922 [472]. It is expected that this can be improved upon by up to a further 5
orders of magnitude when space based tests of the equivalence principle are performed
11
[1282]. These null results are generally considered to be a very tight constraint on the
foundations of any relativistic gravitational theory if it is to be thought of as viable: The
WEP must be satisfied, at least up to the accuracy specified in Eq. (15).
Let us now briefly consider the gravitational redshifting of light. This is one of the
three “classic tests” of General Relativity, suggested by Einstein himself in 1916 [465].
It is not, however, a particularly stringent test of relativity theory. If we accept energy-
momentum conservation in a closed system then it is only really a test of the WEP,
and is superseded in its accuracy by the Eötvös experiment we have just discussed. The
argument for this is the following [423, 594]: Consider an atom that initially has an
inertial mass Mi , and a gravitational mass Mg . The atom starts near the ceiling of a
lab of height h, in a static gravitational field of strength g, and with an energy reservoir
on the lab floor beneath it. The atom emits a photon of energy E that then travels
down to the lab floor, such that its energy has been blue-shifted by the gravitational
field to E 0 when it is collected in the reservoir. This process changes the inertial and
gravitational masses of the atom to Mi0 and Mg0 , respectively. The atom is then lowered
to the floor, a process which lowers its total energy by Mg0 gh. At this point, the atom
re-absorbs a photon from the reservoir with energy E 0 = (Mi00 − Mi0 )c2 and is then
raised to its initial position at the ceiling. This last process raises its energy by Mg00 gh,
where here Mi00 and Mg00 are the inertial and gravitational masses of the atom after re-
absorbing the photon. The work done in lowering and raising the atom in this way is
then w = (Mg00 − Mg0 )gh. Recalling that the energy gained by the photon in travelling
from the lab ceiling to the lab floor is E 0 − E, the principle of energy conservation then
tells us that (E 0 − E) = w = (Mg00 − Mg0 )gh. Now, if the WEP is obeyed then Mi = Mg ,
and this equation simply becomes (E 0 − E) = E 0 gh. This, however, is just the usual
expression for gravitational redshift. Crucial here is the assumption that local position
invariance is valid, so that both Mi and Mg are independent of where they are in the lab.
If the laws of physics are position independent, and energy is conserved, gravitational
redshift then simply tests the equivalence of gravitational and inertial masses, which
is what the Eötvös experiment does to higher accuracy. Alternatively, if we take the
WEP to be tightly constrained by the Eötvös experiment, then gravitational redshift
experiments can be used to gain high precision constraints on the position dependence
of the laws of physics [118]. The gravitational redshift effect by itself, however, does not
appear to be able to distinguish between different theories that obey the WEP and local
position invariance. In Dicke’s approach it should therefore be considered as a test of
the foundations of relativistic gravitational theories, rather than a test of the theories
themselves.
The next most stringent equivalence principle is the EEP. Testing this is a consid-
erably more demanding task than was the case for the WEP, as one now not only has
to show that different test particles follow the same trajectories, but also that a whole
set of special relativistic laws are valid in the rest frames of these particles. Despite
the difficulties involved with this, there is still very compelling evidence that the EEP
should also be considered valid to high accuracy. The most accurate and direct of this
evidence is due to the Hughes-Drever experiments [633, 433], which test for local spatial
anisotropies by carefully observing the shape and spacing of atomic spectral lines. The
basic idea here is to determine if any gravitational fields beyond a single rank-2 tensor
are allowed to couple directly to matter fields. To see why this is of importance, let us
first consider a number of point-like particles coupled to a single rank-2 tensor, gµν . The
12
Lagrangian density for such a set of particles is given by
XZ p
L= mI −gµν uµ uν dλ, (16)
I

where mI are the masses of the particles, and uµ = dxµ /dλ is their 4-velocity measured
with respect to some parameter λ. The Euler-Lagrange equations derived from δL =
0 then tell us that the particles in Eq. (16) follow geodesics of the metric gµν , and
Riemannian geometry tells us that at any point we can choose coordinates such that
gµν = ηµν locally. We therefore recover special relativity at every point, and the EEP
is valid. Now, if the matter fields couple to two rank-2 tensors then the argument used
above falls apart. In this case the Lagrangian density of our particles reads
XZ h p p i
L= mI −gµν uµ uν + nI −hµν uµ uν dλ, (17)
I

where hµν is the new tensor, and nI is the coupling of each particle to that field. The
particles above can now no longer be thought of as following the geodesics of any one
metric, as the Euler-Lagrange Equations (17) are not in the form of geodesic equations.
We therefore have no Riemannian geometry with which we can locally transform to
Minkowski space, and the EEP is violated. The relevance of this discussion for the
Hughes-Drever experiments is that EEP violating couplings, such as those in Eq. (17),
cause just the type of spatial anisotropies that these experiments constrain. In this case
the 4-momentum of the test particle in these experiments becomes
mgµν uν nhµν uν
pµ = p +p , (18)
−gαβ uα uβ −hαβ uα uβ
and as gµν and hµν cannot in general be made to be simultaneously spatially isotropic,
we then have that pµ is spatially anisotropic, and should cause the type of shifts and
broadening of spectral lines that Hughes-Drever-type experiments are designed to de-
tect. The current tightest constraints are around 5 orders of magnitude tighter than the
original experiments of Hughes and Drever [765, 301], and yield constraints of the order

n . 10−27 m, (19)

so that couplings to the second metric must be very weak in order to be observationally
viable. This result strongly supports the conclusion that matter fields must be coupled to
a single rank-2 tensor only. It then follows that particles follow geodesics of this metric,
that we can recover special relativity at any point, and hence that the EEP is valid.
It should be noted that these constraints do not apply to gravitational theories with
multiple rank-2 tensor fields that couple to matter in a linear combination, so that they
P (I)
can be written as in Eq. (16) with gµν = I cI hµν , where cI are a set of I constants.
Local spatially isotropy, and the EEP, is always recovered in this case.
Beyond direct experimental tests, such as Hughes-Drever-type experiments, there are
also theoretical reasons to think that the EEP is valid to high accuracy. This is a con-
jecture attributed to Schiff, that states ‘any complete and self-consistent gravitational
theory that obeys the WEP must also satisfy the EEP’. It has been shown using conser-
vation of energy that preferred frame and preferred location effects can cause violations
13
of the WEP [594]. This goes some way towards demonstrating Schiff’s conjecture, but
there is as yet still no incontrovertible proof of its veracity. We will not consider the issue
further here.
The experiments we have just described provide very tight constraints on the WEP,
the EEP, and local position invariance. It is, of course, possible to test various other
aspects of relativistic gravitational theory that one may consider as ‘foundational’ (for
example, the constancy of a constant of nature [1240]). For our present purposes, how-
ever, we are mostly interested in the EEP. Theories that obey the EEP are often described
as being ‘metric’ theories of gravity, as any theory of gravity based on a differentiable
manifold and a metric tensor that couples to matter, as in Eq. (16), can be shown to
have test particles that follow geodesics of the resulting metric space. The basics of
Riemannian geometry then tells us that at every point in the manifold there exists a
tangent plane, which in cases with Lorentzian signature is taken to be Minkowski space.
This allows us to recover special relativity at every point, up to the effects of second
derivatives in the metric (i.e. tidal forces), so that the EEP is satisfied. Validity of the
EEP can then be thought of as implying that the underlying gravitational theory should
be a metric one [1273].

2.1.2. Observational tests of metric theories of gravity

In what follows we will consider gravitational experiments and observations that can
potentially be used to distinguish between different metric theories of gravity.

Solar system tests

As well as the gravitational redshifting of light that we have already mentioned, the
other two ‘classic tests’ of General Relativity are the bending of light rays by the Sun,
and the anomalous perihelion precession of Mercury. These can both be considered tests
of gravitational theories beyond the foundational issues discussed in the previous section.
That is, each of these tests is (potentially) able to distinguish between different metric
theories of gravity. As well as these two tests, there are also a variety of other gravitational
observations that can be performed in the solar system in order to investigate relativistic
gravitational phenomena. A viable theory of gravity must be compatible with all of
them. For convenience we will split these into tests involving null trajectories (such as
light bending) and tests involving time-like trajectories (such as the perihelion precession
of planets).
First of all let us consider tests involving null geodesics. As already mentioned,
the most famous of these is the spatial deflection of star light by the Sun. In General
Relativity the deflection angle, θ, of a photon’s trajectory due to a mass, M , with impact
parameter d, is given by
2M
θ= (1 + cos ϕ) ' 1.7500 , (20)
d
where ϕ is the angle made at the observer between the direction of the incoming photon
and the direction of the mass. The 1.7500 is for a null trajectory that grazes the limb of the
Sun. This result is famously twice the size of the effect that one might naively estimate
using the equivalence principle alone [464]. The tightest observational constraint to date
on θ is due to Shapiro, David, Lebach and Gregory who use around 2500 days worth of
observations taken over a period of 20 years. The data in this study was taken using 87
14
VLBI sites and 541 radio sources, yielding more than 1.7 × 106 measurements that use
standard correction and delay rate estimation procedures. The result of this is [1131]

θ = (0.99992 ± 0.00023) × 1.7500 , (21)

which is around 3 orders of magnitude better than the observations of Eddington in 1919.
A further, and currently more constraining, test of metric theories of gravity using
null trajectories involves the Shapiro time-delay effect [1130]. Here the deflection in time
is taken into account when a photon passes through the gravitational field of a massive
object, as well as the deflection in space that is familiar from the lensing effects discussed
above. The effect of this in General Relativity is to cause a time delay, ∆t, for a light-like
signal reflected off a distant test object given by
  " 2  #!
4r1 r2 d au
∆t = 4M ln 2
' 20 12 − ln µs, (22)
d R r2

where r1 and r2 (both assumed  d) are the distances of the observer and test object
from an object of mass M , respectively. The second equality here is the approximate
magnitude of this effect when the photons pass close by the Sun, and the observer is on
Earth. Here we have written R as the radius of the Sun, and au as the astronomical
unit. The best constraint on gravity using this effect is currently due to Bertotti, Iess
and Tortora using radio links with the Cassini spacecraft between the 6th of June and
the 7th of July 2002 [147]. These observations result in the constraint

∆t = (1.00001 ± 0.00001)∆tGR , (23)

where ∆tGR is the expected time-delay due to general relativity. The Shapiro time-delay
effect in fact constrains the same aspect of relativistic gravity as the spatial deflection
of light (this will become clear when we introduce the parameterised post-Newtonian
formalism later on). This aspect is sometimes called the ‘unit curvature’ of space.
Let us now consider tests involving time-like trajectories. The ‘classical’ test of Gen-
eral Relativity that falls into this category is the anomalous perihelion precession of
Mercury (this is called a test, despite the fact that it was discovered long before General
Relativity [777]). In Newtonian physics the perihelion of a test particle orbiting an iso-
lated point-like mass stays in a fixed position, relative to the fixed stars. Adding other
massive objects into the system perturbs this orbit, as does allowing the central mass to
have a non-zero quadrupole moment, so that the perihelion of the test particle’s orbit
slowly starts to precess. In the solar system the precession of the equinoxes of the co-
ordinate system contribute about 502500 per century to Mercury’s perihelion precession,
while the other planets contribute about 53100 per century. The Sun also has a non-zero
quadrupole moment, which contributes a further 0.02500 per century. Taking all of these
effects into account, it still appears that the orbit of Mercury in the solar system has an
anomalous perihelion precession that cannot be explained by the available visible matter,
and Newtonian gravity alone. Calculating this anomalous shift exactly is a complicated
matter, and depends on the exact values of the quantities described above. In Table 1
we display the observed anomalous perihelion precession of Mercury, ∆ω, as calculated
by various different groups. For a more detailed overview of the issues involved, and a
number of other results, the reader is referred to [1039]. In relativistic theories of gravity
15
 Source ∆ω /( 00 per century)
Anderson et al. [48] 42.94 ± 0.20
Anderson et al. [49] 43.13 ± 0.14
Krasinsky et al. [746]: EPM1988 42.984 ± 0.061
DE200 42.977 ± 0.061
Pitjeva [1040]: EPM1988 42.963 ± 0.052
DE200 42.969 ± 0.052

Table 1: The value of the perihelion precession of Mercury obtained from observations by various authors.
The acronyms EPM1988 and DE200 refer to different numerical ephemerides, which are reviewed in
[1041].

the additional post-Newtonian gravitational potentials mean that the perihelion of a test
particle orbiting an isolated mass is no longer fixed, as these potentials do not drop off
as ∼ 1/r2 . There is therefore an additional contribution to the perihelion precession,
which is sensitive to the relative magnitude and form of the gravitational potentials, and
hence the underlying relativistic theory. For General Relativity, the predicted anomalous
precession of a two body system is given by
6πM
∆ω = ' 42.9800 , (24)
p
where m is the total mass of the two bodies, and p is the semi-latus rectum of the orbit.
The last equality is for the Sun-Mercury system, and is compatible with the observations
shown in Table 1. Each relativistic theory predicts its own value of ∆ω, and by comparing
to observations such as those in Table 1 we can therefore constrain them. This test is an
additional one beyond those based on null geodesics alone as it tests not only the ‘unit
curvature’ of space, but also the non-linear terms in the space-time geometry, as well as
preferred frame effects.
Another very useful test involving time-like geodesics involves looking for the ‘Nordtvedt
effect’ [986]. This effect is the name given to violations of the SEP. In the previous section
we only considered tests of the WEP and EEP, which provide strong evidence that viable
gravitational theories should be ‘metric’ ones. Now, it is entirely possible to satisfy the
WEP and EEP, with a metric theory of gravity, while violating the SEP. Such violations
do not occur in General Relativity, but do in most other theories. Every test of the
Nordtvedt effect is therefore a potential killing test of general relativity, if it delivers a
non-null result. To date, the most successful approach in searching for SEP violations
is to use the Earth-Moon system in the gravitational field of the Sun as a giant Eötvös
experiment. The difference between this and the laboratory experiments described in
the previous section is that while the gravitational fields of the masses in WEP Eötvös
experiments are entirely negligible, this is no longer the case with the Earth and Moon.
By tracking the separation of the Earth and Moon to high precision, using lasers reflected
off reflectors left on the Moon by the Apollo 11 mission in 1969, it is then possible to
gain the constraint [1277]
η = (−1.0 ± 1.4) × 10−13 , (25)
where η is defined as in Eq. (15). This is indeed a null result, consistent with General
Relativity, and is tighter even than the current best laboratory constraint on the WEP. It
16
can therefore be used to constrain possible deviations from General Relativity, and in fact
constrains a similar (but not identical) set of gravitational potentials to the perihelion
precession described previously.
A third solar system test involving time-like geodesics is the observation of spinning
objects in orbit. While currently less constraining than the other tests discussed so
far, these observations allow insight into an entirely relativistic type of gravitational be-
haviour: gravitomagnetism. This is the generation of gravitational fields by the rotation
of massive objects, and was discovered in the very early days of General Relativity by
Lense and Thirring [1204, 785]. The basic idea here is that massive objects should ‘drag’
space around with them as they rotate, a concept that is in good keeping with Mach’s
principle. Although one can convincingly argue that the same aspects of the gravita-
tional field that cause frame-dragging are also being tested by perihelion precession and
the Nordtvedt effect, it is not true that in these cases the gravitational fields in question
are being communicated through the rotation of matter. Now, in the case of General
Relativity it can be shown that the precession of a spin vector S along the trajectory
of a freely-falling gyroscope in orbit around an isolated rotating massive body at rest is
given by
dS
= Ω × S, (26)
dτ
where
3 1
Ω = v × ∇U − ∇ × g. (27)
2 2
Here we have written the vector g = g0i , and have taken v and U to be the velocity of
the gyroscope and the Newtonian potential at the gyroscope, respectively. The first term
in (27) is called ‘geodetic precession’, and is caused by the ‘unit curvature’ of the space.
This effect exists independent of the massive bodies rotation. The second term in (27)
is the Lense-Thirring term, and causes the frame-dragging discussed above. The most
accurate measurement of this effect claimed so far is at the level of 5 − 10% accuracy,
and has been made using the LAser GEOdynamics Satellites (LAGEOS) [302] (there
has, however, been some dispute of this result [641, 642]). The Gravity Probe B mission
is a more tailor made experiment which was put in orbit around the Earth between April
2004 and September 2005. The current accuracy of results from this mission are at the
level of ∼ 15% [476], although this could improve further after additional analysis is
performed.
All of the tests discussed so far in this section have been for long-ranged modifications
to Newtonian gravity. As well as these, however, there are a host of alternatives to Gen-
eral Relativity that also predict short-ranged deviations from 1/r2 gravity. These range
from extra-dimensional theories [673, 702], to fourth-order theories [307] and bimetric
theories [308], all of which predict ‘Yukawa’ potentials of the form
Z 0
ρ(x0 )e−|x−x |/λ 3 03
U =α d x , (28)
|x − x0 |

where α parameterises the ‘strength’ of the interaction, and λ parameterises its range.
The genericity of these potentials, often referred to as ‘fifth-forces’, provides strong mo-
tivation for experimental attempts to detect them. Unfortunately, due to the their scale
dependence, one can no longer simply look for the extra force on one particular scale, and
17
then extrapolate the result to all scales. Instead, observations must be made on a whole
range of different scales, so that we end up with constraints on α at various different val-
ues of λ. These observations are taken from a variety of different sources, with the scale of
the phenomenon being observed typically constraining λ of similar size. So, for example,
on the larger end of the observationally probed scale we have planetary orbits [1197] and
lunar laser ranging [1277] constraining α . 10−8 between 108 m . λ . 1012 m. On inter-
mediate scales the LAGEOS satellite, and observations of gravitational accelerations at
the top of towers and under the oceans provide constraints of α that range from α . 10−8
at λ ∼ 107 m [1053] to α . 10−3 at 10−1 m . λ . 104 m [459, 1312]. At smaller scales
laboratory searches must be performed, and current constraints in this regime range
from α . 10−2 at λ ∼ 10−2 m, to α . 106 at λ ∼ 10−5 m [624, 819, 291]. Weaker con-
straints at still smaller scales are available using the Casimir effect. For a fuller discussion
of these searches, and the experiments and observations involved, the reader is referred
to the reviews by Fischbach and Talmadge [507], and Adelberger, Heckel and Nelson [11].

Gravitational waves, and binary pulsars

A generic prediction of all known relativistic theories of gravity is the existence of

gravitational waves: Propagating gravitational disturbances in the metric itself. How-
ever, while all known relativistic gravitational theories predict gravitational radiation,
they do not all predict the same type of radiation as the quadrupolar, null radiation
that we are familiar with from General Relativity. It is therefore the case that while
the mere existence of gravitational radiation is not itself enough to effectively discrimi-
nate between different gravitational theories, the type of gravitational radiation that is
observed is. The potential differences between different types of gravitational radiation
can take a number of different forms, which we will now discuss.
Firstly, one could attempt to determine the propagation speed of gravitational waves.
In General Relativity it is the case that gravitational waves have a velocity that is strictly
equal to that of the speed of light in vacuum. Generically, however, this is not true: Some
theories predict null gravitational radiation, and others do not. So, for example, if one
were able to detect gravitational waves from nearby supernovae, then comparing the
arrival time of this radiation with the arrival time of the electromagnetic radiation would
provide a potentially killing test of General Relativity. There are, however, a number
of different theories that predict null gravitational radiation. Tests of the velocity of
gravitational waves therefore have the potential to rule out a number of theories, but by
themselves are not sufficient to distinguish any one in particular.
A second, more discriminating test, is of the polarity of gravitational radiation. Gen-
eral Relativity predicts radiation with helicity modes ±2 only, and so far is the only
proposed theory of gravity that does so. In general, there are six different polarisation
states – one for each of the six ‘electric’ components of the Riemann tensor, R0i0j . These
correspond to the two modes familiar from General Relativity, as well as two modes
with helicity ±1, and two further modes with helicity 0. One of these helicity-0 modes
corresponds to an additional oscillation in the plane orthogonal to the wave vector k µ ,
while the remaining 3 modes all correspond to oscillations in a plane containing k µ . The
extent to which observations of these modes can constrain gravitational theory depends
on whether or not the source of the radiation can be reliably identified. If the source
can be identified, then the vector k µ is known, and one should then be able to uniquely
18
identify the individual polarisation modes discussed above. We then have 6 different
tests of relativistic gravitational theory – one for each of the modes. In the absence of
any knowledge of k µ , however, one cannot necessarily uniquely identify all of the modes
that are present in a gravitational wave, although it may still be possible to constrain
the modes being observed to a limited number of possibilities.
Direct observations of gravitational waves, of the kind discussed above, provide an
excellent opportunity to further constrain gravity. Indeed, some theories can be shown
to be indistinguishable from General Relativity using post-Newtonian gravitational phe-
nomena in the solar system alone, while being easily distinguishable when one also con-
siders gravitational radiation. This is the case with Rosen’s bimetric theory of gravity
[1068, 779, 1275]. To date, however, the direct detection of gravitational radiation has
yet to be performed. At present the highest accuracy null-observations of gravitational
radiation are those of the Laser Interferometer Gravitational-wave Observatory (LIGO).
This experiment consists of two sites in the USA (one in Livingston, Louisiana and one in
Richland, Washington). Each site is an independent interferometer constructed from two
4 km arms, along which laser beams are shone. The experiment has an accuracy capable
of detecting oscillations in space at the level of ∼ 1 part in 1021 , but has yet to make a
positive detection. Further experiments are planned for the future, including Advanced
LIGO, which is scheduled to start in 2014, and the Laser Interferometer Space Antenna
(LISA). Both Advanced LIGO and LISA are expected to make positive detections of
gravitational waves.
Another way to search for gravitational waves is to look for their influence on the
systems that emitted them. In this regard binary pulsar systems are of particular interest.
Pulsars are rapidly rotating neutron stars that emit a beam of electromagnetic radiation,
and were first observed in 1967 [604]. When these beams pass over the Earth, as the
star rotates, we observe regular pulses of radiation. The first pulsar observed in a binary
system was PSR B1913+16 in 1974, by Russell Hulse and Joseph Taylor [634]. This
is a particularly ‘clean’ binary system of a pulsar with rotational period ∼ 59ms in
orbit around another neutron star. Binary pulsars are of particular significance for
gravitational physics for a number of reasons. Firstly, they can be highly relativistic.
The Hulse-Taylor binary system, for example, exhibits a relativistic periastron advance
that is more than 30 000 times that of the Mercury-Sun system. In this regard they
provide an important compliment to the observations of post-Newtonian gravity that we
observe in the solar system. Secondly, they are a source of gravitational waves. Given
the high degree of accuracy to which the orbits of these systems are known, the change
in angular momentum due to gravitational radiation can be determined and observed.
In the Hulse-Taylor system the observed decrease in orbital period over the past 30 years
is 0.997 ± 0.002 of the rate predicted by General Relativity [1267]. Finally, neutron stars
are composed of a type of compact matter that is of particular interest for the study of
self-gravitational effects. For a review of pulsars in this context the reader is referred to
[1177].
There are large number of relativistic parameters that can be probed by observations
of binary pulsar systems [359]. To date, however, the most constrained are the 5 ‘post-
Keplerian’ effects, which are the rate of periastron advance, the rate of change of orbital
period, the gravitational redshift, and two Shapiro time-delay effects. These effects are
familiar from the solar system tests discussed above, apart from the change in orbital
period that is negligible in the solar system. One further effect that has been measured
19
only relatively recently is the ‘geodetic’ precession of the pulsar spin vector about its
angular momentum vector [744]. This is a purely relativistic effect that is observed via
changes in the observed pulse profile over a period of time that can be attributed to
our line of sight to the pulsar crossing the emitting region at varying positions due to
the precession. The determination of the precession rate using these observations is,
however, complicated somewhat by a degeneracy between the a priori unknown shape of
the emitting region and the geometry of the system as a whole [1268, 314].
Not all of the post-Keplerian effects are always apparent in any given binary system,
and not all provide independent tests of gravity. For example, in the Hulse-Taylor binary
only three of these effects can be observed (the inclination angle of the system on the
sky is too large to observe any significant Shapiro delay), and there are two unknown
quantities in the system (the masses of the pulsar, and that of its companion). The Hulse-
Taylor binary therefore provides only 3 − 2 = 1 test of relativistic gravity. The recently
discovered ‘Double Pulsar’ PSR J0737-3039A/B [840], however, does significantly better
[745]. All five post-Keplerian effects are visible in this system, and because both neutron
stars are observable as pulsars the ratio of their masses can be directly inferred from their
orbits. This leaves only one unknown quantity, and hence gives 5 − 1 = 4 independent
tests of relativistic gravity. So far, all binary pulsar tests of gravity, including those of
the double pulsar, are consistent with General Relativity.
Finally, let us return to constraining gravitational theory through the emission of
gravitational waves. The effect of emitting gravitational radiation from a binary system
is to change its orbital period. In General Relativity we know that only quadrupole
radiation with positive energy should be emitted from a system. For most relativistic
theories, however, dipole gravitational radiation is also expected, and sometimes this can
carry away negative energy. The existence of dipole radiation is sometimes attributed
to violations of the SEP, whereby the centre of the mass responsible for gravitational
radiation is no longer the same as the centre of inertial mass. If the centre of inertial mass
is what stays fixed, then the centre of mass responsible for the gravitational radiation can
move and generate dipole radiation. Dipole radiation is expected to be most dominant in
binary systems with high eccentricity, and where the companion mass is a white dwarf.
No evidence for dipolar radiation yet exists [154, 776]. Null observations that attest
to this result therefore allow for experimental limits to be set on theories that predict
positive energy dipolar radiation. The lack of any observation of dipolar radiation can
also be used to rule out with high confidence theories that allow negative energy dipolar
radiation, such as Rosen’s theory [1275].

2.1.3. Theoretical considerations

As he developed the Special Theory of Relativity, it is often assumed that Einstein’s
inspiration came from experiments pointing towards the constancy of the speed of light.
It is true that he was certainly aware of these experiments, but he was also inspired by
theory, specifically his faith in the principle of relativity and the validity of Maxwell’s
equations in any inertial frame. So too, in developing models of modified gravity, we
should not only take our lead from observation but also from theory. Indeed, theoretical
considerations are a very powerful tool in testing new models. Typically these involve
the study of classical and quantum fluctuations about classical solutions. Do the clas-
sical fluctuations propagate super-luminally? Can we excite a ghost? Do the quantum
fluctuations become strongly coupled at some unacceptably low energy scale?
20
Ghosts

Ghosts are a common feature of many modified gravity models that hope to explain
dark energy. Intuitively it is easy to see why this might be the case. To get cosmic
acceleration we need an additional repulsive force to act between massive objects at
large distances. If this force is to be mediated by a particle of even spin, such as a scalar
(spin 0) or a tensor (spin 2), then the kinetic term describing this must have the “wrong”
sign2 , that is, it must be a ghost.
We should be clear about the distinction between the kind of ghost that arises in
certain modified gravity models and the Faddeev-Popov ghost used in the quantisation
of non-abelian gauge theories. The latter is introduced in the path integral to absorb
unphysical gauge degrees of freedom. It does not describe a physical particle and can
only appear as an internal line in Feynman diagrams. In contrast, the ghosts that haunt
modified gravity describe physical excitations and can appear as external lines in Feynman
diagrams.
When a physical ghost is present one has a choice: Accept the existence of negative
norm states and abandon unitarity, or else accept that the energy eigenvalues of the ghost
are negative [317]. Since the former renders the entire quantum description completely
non-sensical, one usually accepts the latter. However, it now follows that the ghost
will generate instabilities if it couples to other, more conventional, fields. When these
fields are already excited, the ghost can and will continually dump its energy into the
“conventional” sector through classical processes, since its energy is unbounded from
below. Even in vacuum, one will get the spontaneous (quantum) production of ghost-
non-ghost pairs, and in a Lorentz invariant theory, the production rate is divergent [317].
There are a few ways to try to exorcise the ghost. One is to isolate it somehow,
such that it completely decouples from other fields. Another option is to make it heavy,
so much so that its mass exceeds the cut-off for the effective theory describing the rel-
evant fluctuations, and one can happily integrate it out. A third option is to break
Lorentz invariance, perhaps spontaneously, so that one can introduce an explicit Lorentz
non-invariant cut-off to regulate the production rate of ghost-non-ghost pairs (see, for
example, [653]). However, perhaps the safest way to deal with a ghost is to dismiss as
unphysical those solutions of a theory upon which the ghost can fluctuate. This school
of thought is exploited to good effect in the ghost condensate model [590].

Strong coupling

Some modified gravity models are said to suffer from “strong coupling” problems.
Given a classical solution to the field equations, this refers to quantum fluctuations on
that solution becoming strongly coupled at an unacceptably low scale. For example, in
DGP gravity, quantum fluctuations on the Minkowski vacuum becomes strongly coupled
at around Λ ∼ 10−13 eV ∼ 1/(1000km). In other words, for scattering processes above
Λ, perturbative quantum field theory on the vacuum is no longer well defined, and one

2 In our conventions, the Lagrangian for a canonical scalar is L = − 12 (∂ψ)2 , whereas a ghost has
L= + 12 (∂ψ)2 .
21
must sum up the contribution from all the multi-loop diagrams. One then has complete
loss of predictivity. Furthermore, the classical solution itself is meaningless at distances
below Λ−1 since it would require a scattering process involving energies above the cut-off
to probe its structure.
The strong coupling scale is, of course, dependent on the background classical solution,
and may even depend on position in space-time. Whether the inferred strong coupling
scale is acceptable, or not, again depends on the background. For example, strong
coupling at 1000 km on the Minkowski vacuum of DGP gravity is not really an issue as
Minkowski space does not represent a good approximation to the classical solution in the
vicinity of the Earth. Indeed, for the classical solutions sourced by the Earth to leading
order, quantum fluctuations will become strongly coupled at some scale that depends on
the radial distance from the Earth’s centre. Computed at the Earth’s surface one should
require that this lies below an meV since quantum gravity effects have yet to show up in
any lab based experiments up to this scale.
It has actually been argued that strong coupling on the vacuum can be a virtue
in modified gravity models [447]. This is because it can be linked to a breakdown of
classical perturbation theory, which is necessary for the successful implementation of the
Vainshtein mechanism [1241, 399]. We discuss the Vainshtein mechanism and strong
coupling in some detail in the context of DGP gravity in Section 5.5.4. Here we will
make some generic statements. Consider a model of gravity that deviates from GR at
large distances. To be significant in terms of understanding dark energy, this deviation
must be at least O(1) on cosmological scales, but be suppressed down to . O(10−5 ) on
Solar System scales. Therefore, the field or fields that are responsible for the modification
must be screened within the Solar System. How can this screening occur? One way is for
the fields to interact so strongly that they are frozen together, so much so that they are
unable to propagate freely. This is the idea behind the Vainshtein mechanism – higher
order derivative interactions help to suppress the extra modes near the source (the Sun).
Alternative ways to screen the extra fields have been suggested in the form of the
chameleon [689, 688], and the symmetron [608] mechanisms. Both methods exploit the
dependence of the effective potential on the environment. For the chameleon, the mass of
the field is environmentally dependent, getting heavy in the Solar System. For the sym-
metron, the strength of the matter coupling is (indirectly) environmentally dependent,
tending to zero near a heavy source.

2.2. Einstein’s Theory

Having considered the requirements that must be satisfied by a viable relativistic
theory of gravity, let us now consider Einstein’s theory of General Relativity in particular.
General Relativity satisfies all of the requirements described in the previous section, either
by construction (for the foundational requirements) or by trial (in the case of tests of
metric theories of gravity).
General Relativity is a gravitational theory that treats space-time as a 4-dimensional
manifold. The connection associated with covariant differentiation, Γµαβ , should be
viewed as an additional structure on this manifold, which, in general, can be decomposed
into parts that are symmetric or antisymmetric in its last two indices:

Γµαβ = Γµ(αβ) + Γµ[αβ] . (29)

22
In General Relativity we take Γµ[αβ] = 0, or, in the language of differential geometry,
we assume that torsion vanishes. We are then left with only the symmetric part of the
connection, which describes the curvature of the manifold.
Now, to define distances on the manifold one also requires a metric tensor, gµν . Along
the curve γ this gives the measure of distance
Z
p
s= dλ gµν ẋµ ẋν , (30)
γ

where λ is a parameter along the curve, xµ = xµ (λ), and over-dots here mean dif-
ferentiation with respect to λ. The metric should also be considered as an additional
structure on the manifold, which is in general independent from the connection. The
relationship between the connection and the metric is defined via the non-metricity ten-
sor, Qµαβ ≡ ∇µ gαβ . In General Relativity it is assumed that the non-metricity tensor
vanishes. We can now use the metric to define the Levi-Civita connection, which has
components given by the Christoffel symbols:
 
µ µ 1
Γ αβ = ≡ g µν (gαν,β + gβν,α − gαβ,ν ) . (31)
αβ 2

To summarise, as a consequence of the two assumptions Qµαβ = 0 and Γµ[αβ] = 0 ,

the components of the connection are uniquely given by the Christoffel symbols via (31),
and so the connection, and all geometric quantities derived from it, are defined entirely
in terms of the metric. In General Relativity, therefore, the metric tells us everything
there is to know about both distances and parallel transport in the space-time manifold.
The resulting set of structures is known as a Riemannian manifold (or, more accu-
rately, pseudo-Riemannian in the case where the metric is not positive definite, as is
required to recover special relativity in the tangent space to a point in space-time). Rie-
mannian manifolds have a number of useful properties including tangent vectors being
parallel to themselves along geodesics, the geodesic completeness of space-time implying
the metric completeness of space-time, and a particularly simple form for the contracted
Bianchi identities:  
1
∇µ Rµν − g µν R = 0, (32)
2
where Rµν and R = g µν Rµν are the Ricci tensor and scalar curvature, respectively . This
last equation is of great significance for Einstein’s equations.

2.2.1. The field equations

Having briefly discussed the geometric assumptions implicit in General Relativity, let
us now display the field equations of this theory:
1
Rµν − gµν R = 8πGTµν − gµν Λ. (33)
2
Here Tµν is the energy-momentum tensor of matter fields in the space-time, and Λ is the
cosmological constant. These equations are formulated such that energy-momentum is
a conserved quantity (due to the contracted Bianchi identity and metric-compatibility
of the connection), so that special relativity can be recovered in the neighbourhood of
23
every point in space-time (up to tidal forces), and so that the usual Newtonian Poisson
equation for weak gravitational fields is recovered in non-inertial frames kept at a fixed
space-like distances from massive objects (up to small corrections).
The Field Equations (33) are a set of 10 generally covariant, quasi-linear second-order
PDEs in 4 variables, for the 10 independent components of the metric tensor. They
constitute 4 constraint equations and 6 evolution equations, with the contracted Bianchi
identities ensuring that the constraint equations are always satisfied. Furthermore, the
conserved nature of Tµν and the Riemannian nature of the manifold ensure that the
WEP and EEP are always satisfied: Massless test particles follow geodesics, and in any
freely falling frame one can always choose ‘normal coordinates’ so that local space-time
is well described as Minkowski space.

2.2.2. The action

As with most field theories, the Field Equations (33) can be derived from the variation
of an action. In the case of General Relativity this is the Einstein-Hilbert action:
Z Z
1 √
S= −g(R − 2Λ)d4 x + Lm (gµν , ψ)d4 x, (34)
16πG
where Lm is the Lagrangian density of the matter fields, ψ, and the gravitational La-
√
grangian density has been taken to be Lg = −g(R − 2Λ)/16πG. Let us now assume
the Ricci scalar to be a function of the metric only, so that R = R(g). Variation of Eq.
(34) with respect to the metric tensor then gives the Field Equations (33), where

2 δLm
T µν ≡ √ . (35)
−g δgµν
√
The factors of −g are included in Eq. (34) to ensure that the Ls transform as scalar
densities under coordinate transformations, i.e. as
 µ
∂x
L̄ = det L, (36)
∂ x̄ν

under coordinate transformations x̄µ = x̄µ (xν ). This property ensures S is invariant
under general coordinate transformation, and that the resulting tensor field equations are
divergence free (i.e. the contracted Bianchi identities and energy-momentum conservation
equations are automatically satisfied).
We have outlined here how Einstein’s equations can be obtained from the variation of
an invariant action with respect to the metric, once it has been assumed that the space-
time manifold is Riemannian. The vanishing of torsion and non-metricity then tell us that
the metric is the only independent structure on the manifold, and the invariant action
principle ensures that we end up with a set of tensor field equations in which energy-
momentum is conserved. Because of this formulation the WEP and EEP are satisfied
identically. Now, when considering alternative theories of gravity one often wants to
modify the field equations while conserving these basic properties. Modified theories of
gravity are therefore often formulated in a similar way; from the metric variation of an
invariant action principle under the assumption of Riemannian geometry, with a universal
coupling of all matter fields to the same metric.
24
2.3. Alternative Formulations
The discussion in the previous section involved deriving Einstein’s equations under
the a priori assumption of Riemannian geometry (i.e. assuming to begin with that the
torsion vanishes and that the connection is metric compatible). In this case the metric is
the only remaining geometric structure, and a simple metric variation of the action is the
only option. We can, however, be less restrictive in specifying the type of geometry we
wish to consider. For the case of the Einstein-Hilbert action, Eq. (34), this usually still
leads to the Einstein Equations (33). For alternative theories of gravity, however, this is
often not the case: Different variational procedures, and different assumptions about the
geometric structures on the manifold, can lead to different field equations. It is for this
reason that we now outline some alternative formulations of General Relativity. A large
collection of many such formulations can be found in [1032].

2.3.1. The Palatini procedure

The most well known deviation from the metric variation approach is the ‘Palatini
procedure’ [1020]. Here the connection is no longer immediately assumed to be metric
compatible, but is still assumed to be symmetric and thus torsionless. In addition, all
matter fields are still taken to couple universally to the metric only3 . The action to be
varied is then
Z Z
1 √  µν Γ  4
S= −g g Rµν − 2Λ d x + Lm (gµν , ψ)d4 x, (37)
16πG

where Γ Rµν indicates that the Ricci tensor here is defined with respect to the connec-
tion and not the metric (at this stage the metric and connection are still independent
variables), and is given by
Γ
Rµν = ∂α Γαµν − ∂µ Γααν + Γβ βα Γαµν − Γαµβ Γβ αν . (38)

The Ricci tensor defined above, as well as the Einstein tensor derived from it, are in gen-
eral asymmetric. However, they become symmetric as soon as we assume the connection
is symmetric. Variation of Eq. (37) with respect to the connection gives the condition
that the connection is in fact the Levi-Civita connection. Variation with respect to the
metric then recovers the Einstein equations.
If the torsionless condition on the connection is dropped then complications arise, as
has been shown by Hehl and Kerlick [600]. The general form of the connection can be
shown to be given by
   
µ µ µ µ
Γ αβ = + J αβ = + K µαβ + Lµαβ . (39)
αβ αβ

The tensor field K µαβ is the contorsion tensor, that can be defined in terms of the
antisymmetric components of the connection, known as the torsion, as

K µαβ ≡ S µαβ − Sαβ µ − Sβα µ , (40)

3 This assumption has limited validity, however, as it cannot be applied to tensor fields without using

a covariant derivative.
25
where S µαβ = Γµ[αβ] is the torsion tensor. The tensor field Lµαβ is defined in terms of
the non-metricity tensor as
1 µ 
Lµαβ ≡ Q αβ − Qαβ µ − Qβαµ . (41)
2
To avoid confusion, we continue to denote the covariant derivative associated with the
Levi-Civita connection as ∇µ , while we use Γ ∇µ to denote the covariant derivative
associated with Γµαβ .
Varying the action with respect to the metric gµν we find the analogue of the Einstein
equations:
G(µν) + Λgµν = 8πGT̃µν . (42)
One should note that only the symmetric part of the Einstein tensor appears here, and
that we have used T̃µν rather than Tµν to emphasise the  fact that T̃µν is defined at
δLm 
constant Γµαβ in the variation, i.e. T̃µν = − √2−g δg µν
Γ
. On the other hand Tµν =

−√ 2 δLm
µν
 . This is not an important distinction at this stage, as we have assumed
−g δg J
that matter field do not couple to the connection, and hence T̃µν = Tµν . It will, however,
be important in the following subsection and in Section 3.5.1. √
8πG δ( −gR)
Varying with respect to the connection defines the Palatini tensor as Pµ αβ = √ µ
−g δΓ αβ
,
that can be written as
Pαµβ = Sµαβ + 2gµ[α Sβ] + gµ[α Qβ] − gµ[α Q̄ν]βν , (43)
where Sµ = S αµα , and where we have split the non-metricity tensor into trace and
traceless parts as Qµαβ = Qµ gαβ + Q̄µαβ , with g αβ Q̄µαβ = 0. The Palatini tensor has
only 60 independent components because it is identically traceless: Pα µα = 0. Now, the
second field equation is the vanishing of the Palatini tensor,
Pµ αβ = 0, (44)
but this provides only 60 constraints among the 64 independent components of the con-
nection. In fact it may be shown that the equation Pµ αβ = 0 is equivalent to the
connection taking the following form [600]:
   
µ 1 µ 2
Γµαβ = − Qα δ µβ = + Sα δ µβ . (45)
αβ 2 αβ 3
Clearly then, there are 4 degrees of freedom left undetermined by the field equations.
Thus the Palatini approach in its most general form does not lead to a unique set of field
equations4 .
The constraint Qµ = 0 is sufficient to produce a consistent theory.
R This, however, has
√
to be imposed as a Lagrange multiplier in the action via a term d4 x −gλα Qα . Once
this is done, one recovers General Relativity uniquely. For theories of gravity other than
General Relativity the difference between the metric variation and the Palatini procedure
is even more significant: The resulting field equations are, in general, different. This will
be spelt out explicitly for some specific theories in the sections that follow.
4 It is often said that the Palatini procedure uniquely recovers GR. As we have seen, however, this

is a myth. It does so only after further assumptions, for instance that the torsion vanishes, or that the
connection is metric compatible, or that Qα = 0. To make the Palatini variation well defined one has
to impose such conditions in the action by means of Lagrange multipliers.
26
2.3.2. Metric-affine gravity and matter
A further generalisation of the metric variation approach is to keep the metric and
connection completely independent, as discussed above, and further allow matter to
couple not only to the metric, but also the connection [600]. In this case the action takes
the form
Z Z
1 √
S= −g(g µν Γ Rµν − 2Λ)d4 x + Lm (gµν , Γµαβ , ψ)d4 x, (46)
16πG
where Γµαβ and gµν are once again independent. Performing the variations we recover
Eq. (42) as before, and
Pµ αβ = 8πG∆µαβ , (47)
δLm
where ∆µαβ = − √1−g δΓµ is called the hypermomentum tensor [600, 601].
αβ

In this case Tµν 6= T̃µν , but it is straightforward to find that

h i
ρ ρ
Tµν = T̃µν + ∇ρ ∆ρ(µν) − ∆(µ ν) − ∆(µν) , (48)

where ∇µ is the covariant derivative associated with the Levi-Civita connection.

Equation (47) can be shown to be self-inconsistent for reasonable forms of matter,
as the Palatini tensor is invariant under projective transformations of the form Γµαβ →
Γµαβ + λα δ µβ , while there is no reason to suspect this invariance is exhibited by the
matter fields and hence the hypermomentum. Equivalently, the Palatini tensor obeys
the identity Pαµ α = 0, while there is no reason that this should identically hold for the
hypermomentum5 . One way to impose self-consistency is to demand that both torsion
and non-metricity must vanish (by using Lagrange multipliers in the action), leading
again to General Relativity. This type of self-consistency is, however, very strong, and
weaker constraints have been found in [600]. One such weaker constraint leads to the
Einstein-Cartan-Sciama-Kibble theory [601], that we shall briefly describe in Section
3.5.1.

2.3.3. Other approaches

There are a variety of other formalisms that one can use to derive Einstein’s equations.
We will not go into the full details of all of these here, but merely mention some of the
approaches that exist in the literature. For brevity we will only consider vacuum general
relativity here.
In the ‘vierbein’ formalism the Einstein-Hilbert action can be written
Z
S = d4 x e eµα̂ eνβ̂ Rµν α̂β̂ , (49)

5 Consider for example a simple Einstein-Æther model for which the matter action is S
M =
√
d4 x −g [α∇µ Aν ∇ν Aµ + λ(Aµ Aµ + 1)]. The hypermomentum is ∆µαβ = −2Aβ ∇µ Aα which clearly
R
does not obey ∆αµα = 0. The variation done this way is inconsistent. On the other hand using the
√
Lagrange constraint d4 x −gβµ αβ J µαβ in the action imposes J µαβ = 0, and hence the vanishing of
R

the Palatini tensor. This leads to a modified Eq. (47), as Pµ αβ = 0 = 8πG(∆µαβ − βµ αβ ), and to a
modified Eq. (42), which now includes derivatives of βµ αβ . After using Eq. (48), however, the resulting
equations are completely equivalent to the metric variation.
27
where indices with hats correspond to a basis in the tangent space defined by the set
of contravariant vectors, eµ̂µ , with determinant e = det[eµ̂µ ]. The inverse of eµ̂µ is eµµ̂ ,
such that eρ µ̂ eν̂ ρ = δ µ̂ν̂ , and eρ̂ µ eν ρ̂ = δ µν . The metric tensor is constructed as gµν =
ηµ̂ν̂ eµµ̂ eν ν̂ . The spin connection ωµ α̂β̂ then defines a space-time and Lorentz covariant
derivative, Dµ , as Dµ vνρ̂ = ∇µ vνρ̂ + ωµ ρ̂λ̂ vνλ̂ , where ∇µ is the Levi-Civita connection6 . The
curvature tensor Rµν α̂β̂ is defined in terms of the spin connection as
ν̂ ν̂
Rµν µ̂ν̂ ≡ ∂µ ων µ̂ν̂ − ∂ν ωµ µ̂ν̂ + ωµ µ̂ρ̂ ων ρ̂ − ων µ̂ρ̂ ωµρ̂ . (50)
Variation now proceeds as in the Palatini formalism by assuming that the spin connection
and vierbein are independent fields, from which one obtains the two field equations
D[µ eα̂ν] = 0, (51)
and
α µ ν 1
Gαρ̂ ≡ eα̂ eρ̂ eβ̂ Rµν
α̂β̂
− (eµα̂ eνβ̂ Rµν α̂β̂ )eα
ρ̂ = 0, (52)
2
where Gµν̂ is the Einstein tensor. Equation (51) can be used to obtain the spin connection
in terms of the partial derivatives of the vierbein, and the resulting relation implies that
ωµ α̂β̂ is torsion-less, i.e. one recovers Cartan’s first structure equation, deµ̂ +ω µ̂ν̂ ∧eν̂ = 0.
The second equation says that the vacuum Einstein equations are recovered.
Another interesting alternative formulation of General Relativity is given by the Ple-
banski formalism [1042]. It is derived from the action
Z
1
S = ΣAB ∧ RAB − ΨABCD ΣAB ∧ ΣCD , (53)
2
where upper case indices denote two component spinor indices to be raised and lowered
with AB and its inverse, and where the exterior product ∧ acts on space-time indices,
which have been suppressed. The curvature 2-form RAB ≡ dωAB + ωA C ∧ ωCB is defined
with respect to a spin connection 1-form ωA B . Variation of this action with respect to
ΨABCD and ωAB then tells us that the 2-form ΣAB is the exterior product of some set
0
of 1-forms that we can identify with the tetrad θAA , and that the connection ωAB is
torsion-free with respect to ΣAB . Using this together with the variation of the action
with respect to ΣAB then gives the vacuum Einstein equations, where the metric is given
0
by g = θAA ⊗ θAA0 .
One further alternative formulation of General Relativity is the purely affine ‘Edding-
ton formalism’ [460]. In previous subsections we outlined how one can either treat the
metric as the only independent structure on the manifold, or treat the metric and con-
nection as being two independent structures. Another approach is to take the connection
as the only structure on the manifold. In this case, the simplest way of constructing a
Lagrangian density with the correct weight (and without a metric) is to simply take the
square root of the determinant of the Ricci tensor itself:
Z q
S= −det[Rµν (Γ)]d4 x. (54)

6 Given a metric, g
µν , the Levi-Civita connection can always be defined. The question is whether
that is the connection that is used to define parallel transport.
28
Varying this action with respect to the connection then gives the field equations
q 
µν
∇ρ −det[Rαβ (Γ)]R = 0, (55)

which can be shown to be equivalent to Einstein’s equations in vacuum with a cosmolog-

ical constant, if we take the connection to be the Levi-Civita connection. Due to the lack
of a metric in the action for this theory, however, it is not a trivial matter to introduce
matter fields into the theory [69].
Finally, let us mention that approaches exist that treat gravity as simply a spin-2
field on flat space [114, 115]. It has been conjectured that one could reconstruct the
Einstein-Hilbert action in such an approach by considering consistency conditions order
by order in perturbation theory. This will, of course, be an invalid treatment when
gravity is strong, and in cosmology.

2.4. Theorems
There a number of theorems in General Relativity that are of great importance for
the structure of the theory itself, as well as for the solutions to the field equations. These
theorems underpin a lot of the acquired intuition on how gravity should function in
different environments, and what the resulting phenomenology should be. In alternative
theories of gravity, however, the theorems of General Relativity often fail, allowing new
behaviours that would otherwise be impossible.
Here we briefly recap what we consider to be some of the most important theorems
of General Relativity. In later sections we will show how these theorems are violated in
alternative theories, and discuss the consequences of this.

2.4.1. Lovelock’s theorem

Lovelock’s theorem [831, 832] limits the theories that one can construct from the
metric tensor alone. To enunciate this theorem, let us begin by assuming that the metric
tensor is the only field involved in the gravitational action. If the action can be written
in terms of the metric tensor gµν alone, then we can write
Z
S = d4 xL(gµν ). (56)

If this action contains up to second derivatives of gµν , then extremising it with respect
to the metric gives the Euler-Lagrange expression
  
µν d ∂L d ∂L ∂L
E [L] = ρ − λ − , (57)
dx ∂gµν,ρ dx ∂gµν,ρλ ∂gµν
and the Euler-Lagrange equation is E µν (L) = 0. Lovelock’s theorem can then be stated
as the following:
Theorem 2.1. (Lovelock’s Theorem)
The only possible second-order Euler-Lagrange expression obtainable in a four dimen-
sional space from a scalar density of the form L = L(gµν ) is
 
√ 1 µν √
E = α −g R − g R + λ −gg µν ,
µν µν
(58)
2
29
where α and λ are constants, and Rµν and R are the Ricci tensor and scalar curvature,
respectively.
This powerful theorem means that if we try to create any gravitational theory in a
four-dimensional Riemannian space from an action principle involving the metric tensor
and its derivatives only, then the only field equations that are second order or less are
Einstein’s equations and/or a cosmological constant. This does not, however, imply that
the Einstein-Hilbert action is the only action constructed from gµν that results in the
Einstein equations. In fact, in four dimensions or less one finds that the most general
such action is
√ √ √  
L = α −gR − 2λ −g + βµνρλ Rαβµν Rαβρλ + γ −g R2 − 4Rµν Rν µ + Rµν ρλ Rρλµν ,

where β and γ are also constants. The third and fourth terms in this expression do not,
however, contribute to the Euler-Lagrange equations as
h i
E µν αβρλ Rγδαβ Rγδρλ = 0 (59)
h√  i
E µν −g R2 − 4Rαβ Rβα + Rαβρλ Rρλαβ = 0, (60)

where the action of E µν on any function X is defined as in Eq. (57). The first of these
equations is valid in any number of dimensions, and the second is valid in four dimensions
only.
Lovelock’s theorem means that to construct metric theories of gravity with field equa-
tions that differ from those of General Relativity we must do one (or more) of the fol-
lowing:
• Consider other fields, beyond (or rather than) the metric tensor.

• Accept higher than second derivatives of the metric in the field equations.
• Work in a space with dimensionality different from four.
• Give up on either rank (2,0) tensor field equations, symmetry of the field equations
under exchange of indices, or divergence-free field equations.

• Give up locality.
The first three of these will be the subject of the next three sections of this report. The
fourth option requires giving up on deriving field equations from the metric variation of
an action principle, and will not be considered further here.

2.4.2. Birkhoff ’s theorem

Birkhoff’s theorem7 is of great significance for the weak-field limit of General Rela-
tivity. The theorem states [160]

7 This theorem is commonly attributed to Birkhoff, although it was already published two years earlier

by Jebsen [659]. It is not to be confused with Birkhoff ’s pointwise ergodic theorem.

30
Theorem 2.2. (Birkhoff’s Theorem)
All spherically symmetric solutions of Einstein’s equations in vacuum must be static and
asymptotically flat (in the absence of Λ).
Strictly speaking, there are very few situations in the real Universe in which Birkhoff’s
theorem is of direct applicability: Exact spherical symmetry and true vacuums are rarely,
if ever, observed. Nevertheless, Birkhoff’s theorem is very influential in how we under-
stand the gravitational field around (approximately) isolated masses. It provides strong
support for the relativistic extension of our Newtonian intuition that far from such ob-
jects their gravitational influence should become negligible, or, equivalently, space-time
should be asymptotically flat8 . We can therefore proceed with some confidence in treat-
ing the weak-field limit of General Relativity as a perturbation about Minkowski space.
Birkhoff’s theorem also tells us that certain types of gravitational radiation (from a star
that pulsates in a spherically symmetric fashion, for example) are not possible.
As we will show below, Birkhoff’s theorem does not hold in many alternative theories
of gravity. We therefore have less justification, aside from our own intuition, in treating
the weak field limit of these theories as perturbations about Minkowski space. We must
instead be more careful, as the space-time we perform our expansion around can have
asymptotic curvature, leading to either time or space-dependence of the background (or
some combination of the two). What is more, the perturbations themselves may be time-
dependent, and their form can be sensitive to the type of asymptotic curvature that the
background exhibits. Behaviours such as these are not expected in General Relativity
[837].

2.4.3. The no-hair theorems

These theorems are named after the phrase coined by Wheeler that “black holes have
no hair”. The first of these theorems was given by Israel and showed that the only
static uncharged asymptotically flat black hole solution to Einstein’s equations is the
Schwarzschild solution [650]. He later extended this theorem to include charged objects
[651], and Carter extended it to black holes with angular momentum [262]. The theorem
is therefore often stated today as “the generic final state of gravitational collapse is a
Kerr-Newman black hole, fully specified by its mass, angular momentum, and charge ”
[1254].
Complementary to the black hole no-hair theorems is the no-hair ‘theorem’ of de Sitter
space. The claim here is that in the context of General Relativity with a cosmological
constant all expanding universe solutions should evolve towards de Sitter space. This
has been shown explicitly by Wald for all Bianchi type models9 [1255].
These theorems play an important role in General Relativity and cosmology. Some
progress has been made in extending them to alternative theories of gravity, but there
have also been explicit examples of them being violated in particular theories. This will
be discussed further in subsequent sections.

8 Of course, in a cosmological setting asymptotic regions are never realised as we will eventually come
across the other masses in the Universe.
9 Except type-IX universes with large amounts of spatial curvature.

31
2.5. The Parameterised Post-Newtonian Approach
This section is a recap of the Parameterised Post-Newtonian (PPN) formalism that
is widely used by both theoretical and observational gravitational physicists. The idea
here is to create a construction that encompasses a wide array of different gravitational
theories, and that contains parameters that can be constrained by observations in a rea-
sonably straightforward fashion. In this way labour can be saved on both the theoretical
and observational ends of the spectrum: Observers can apply their results to constrain
a wide array of theories without having to trawl through the details of the individual
theories themselves, and theorists can straightforwardly constrain their new theories by
comparing to the already established bounds on the PPN parameters without having to
re-calculate individual gravitational phenomena. To date, this approach has been highly
successful, and in the following sections of this report we will often refer to it. We will
therefore outline here how the PPN formalism proceeds. For a more detailed explanation
of the principles and consequences of this formalism the reader is referred to [1274].

2.5.1. Parameterised post-Newtonian formalism

The PPN formalism is a perturbative treatment of weak-field gravity, and therefore
requires a small parameter to expand in. For this purpose an “order of smallness” is
defined by
P
U ∼ v2 ∼ ∼ Π ∼ O(2),
ρ
where U is the Newtonian potential, v is the 3-velocity of a fluid element, P is the pressure
of the fluid, ρ is its rest-mass density and Π is the ratio of energy density to rest-mass
density. Time derivatives are also taken to have an order of smallness associated with
them, relative to spatial derivatives:
|∂/∂t|
∼ O(1).
|∂/∂x|
Here we have chosen to set c = 1. The PPN formalism now proceeds as an expansion in
this order of smallness.
For time-like particles coupled to the metric only the equations of motion show that
the level of approximation required to recover the Newtonian limit is g00 to O(2), with
no other knowledge of other metric components beyond the background level being nec-
essary. The post-Newtonian limit for time-like particles, however, requires a knowledge
of

g00 to O(4)
g0i to O(3)
gij to O(2).

Latin letters here are used to denote spatial indices. To obtain the Newtonian limit of null
particles we only need to know the metric to background order: Light follows straight
lines, to Newtonian accuracy. The post-Newtonian limit of null particles requires a
knowledge of g00 and gij both to O(2).
The way in which the PPN formalism then proceeds is as follows. First one identifies
the different fields in the theory. All dynamical fields should then be perturbed from
32
their expected background values, and the perturbations assigned an appropriate order
of smallness each. For theories containing a metric the appropriate expansion is usually
(2) (4)
g00 = −1 + h00 + h00 + O(6) (61)
(3)
g0i = h0i + O(5) (62)
(2)
gij = δij + hij + O(4), (63)

where superscripts in brackets denote the order of smallness of the term. If, for example,
the theory contains an additional scalar field, then the usual expansion for this quantity
is
φ = φ0 + ϕ(2) + ϕ(4) + O(6), (64)
where φ0 is the constant background value of φ. Additional vector and tensor gravita-
tional fields can be specified in a corresponding way.
The energy-momentum tensor in the PPN formalism is then taken to be that of a
perfect fluid. To the relevant order, the components of this tensor are given by

T00 = ρ(1 + Π + v 2 − h00 ) + O(6) (65)

T0i = −ρvi + O(5) (66)
Tij = ρvi vj + P δij + O(6). (67)

Taking these expressions, the field equations for the theory in question, and substituting
in the perturbed expressions for the dynamical fields in the theory, as prescribed above,
the field equations can then be solved for order by order in the smallness parameter.
(2)
The first step in such calculations is usually to solve for h00 . With this solution in
(2) (3) (4)
hand, one then proceeds to solve for hij and h0i simultaneously, and finally h00 can be
solved for. If additional fields exist, beyond the metric, then these quantities must also
(2)
be solved for to increasing order of smallness as the calculation proceeds. In finding hij ,
(3) (4)
h0i and h00 one needs to specify a gauge. After such a specification one still, of course,
has the freedom to make additional gauge transformations of the form xµ → xµ + ξ µ ,
where ξ µ is O(2) or smaller. This freedom should be used at the end of the process to
transform the metric that has been obtained into the “standard post-Newtonian gauge”.
This is a gauge in which the spatial part of the metric is diagonal, and terms containing
time derivatives are removed. Once this has been done then one is in possession of the
PPN limit of the theory in question.
We have so far outlined the procedure that one needs to follow in order to gain the
appropriate form of the metric that couples to matter fields in the weak-field limit. Once
done, the result can then be compared to the ‘PPN test metric’ below:

g00 = −1 + 2GU − 2βG2 U 2 − 2ξG2 ΦW + (2γ + 2 + α3 + β1 − 2ξ)GΦ1

+2(1 + 3γ − 2β + β2 + ξ)G2 Φ2 + 2(1 + β3 )GΦ3 − (β1 − 2ξ)GA
+2(3γ + 3β4 − 2ξ)GΦ4
1 1
g0i = − (3 + 4γ + α1 − α2 + β1 − 2ξ)GVi − (1 + α2 − β1 + 2ξ)GWi
2 2
gij = (1 + 2γGU )δij .
33
Here β, γ, ξ, β1 , β2 , β3 , β4 , α1 , α2 and α3 are the ‘post-Newtonian parameters’, U is
the Newtonian gravitational potential that solves the Newtonian Poisson equation, and
ΦW , Φ1 , Φ2 , Φ3 , Φ4 , A, Vi and Wi are the ‘post-Newtonian gravitational potentials’
(the precise form of these potentials is given in [1274]). The particular combination of
parameters before each of these potentials is chosen here so that they have particular
physical significance, once gravitational phenomena have been computed.

2.5.2. Parameterised post-Newtonian constraints

Comparison of the weak field metric of a particular theory with the PPN test metric
above allows one to read off values for the PPN parameters β, γ, ξ, β1 , β2 , β3 , β4 , α1 , α2
and α3 for the theory in question. The test metric has been constructed to include the
type of potentials that often appear when one modifies gravity10 . The great utility of
the PPN formalism is that observers can take the PPN test metric above and constrain
the parameters without having a particular theory in mind. These constraints can then
be applied directly to a large number of gravitational theories, without having to work
out how complicated gravitational phenomena work in each theory individually.
In General Relativity we have that β = γ = 1 and ξ = β1 = β2 = β3 = β4 = α1 =
α2 = α3 = 0. Other theories will predict other values for these parameters, and we will
discuss these on a case by case basis in the sections that follow. Observationally, one can
use the gravitational phenomena discussed in Section 1 to impose the constraints that
follow.
As already discussed, observations that involve only null geodesics are sensitive to
(2) (2)
the Newtonian part of the metric, g00 , and the term gij only. These two terms involve
the PPN parameter γ only. We can now use constraints on the bending of light by the
Sun to get a constraint on γ. Using the PPN test metric the predicted bending of light
that one should observe is [1274]

m (1 + γ)
θ = 2(1 + γ) = θGR , (68)
r 2
where m is the mass of the Sun, r is its radius, and θGR is the general relativistic
prediction. Using the observed value of θ given in Section 1 then gives [1131]

γ − 1 = (−1.7 ± 4.5) × 10−4 , (69)

which is consistent with the general relativistic value of γ = 1. Similarly, we can use the
PPN test metric to find that the Shapiro time delay effect is given by [1274]

(1 + γ)
∆t = ∆tGR , (70)
2
where subscript GR again means the value of this quantity as predicted by General
Relativity. Taking the observed value of ∆t given in Section 1 then gives the even tighter
constraint [147]
γ − 1 = (2.1 ± 2.3) × 10−5 , (71)

10 It is not, however, an exhaustive collection of all possible potentials, and in some theories it is

occasionally necessary to include additional terms.

34
again consistent with γ = 1. It can now be clearly seen that the bending of light by the
Sun, and the Shapiro time delay effect do, in fact, constrain the same aspect of space-time
geometry. They can therefore be considered as complimentary to each other.
If we now consider observations of gravitational phenomena that involve time-like
geodesics then we are able to observe, potentially, all of the post-Newtonian potentials in
the PPN test metric. This becomes clear from the expression for perihelion precession,
which now becomes
  2 
6πM 1 1 µ r
∆ω = (2 + 2γ − β) + (2α1 − α2 + α3 + 2β2 ) + J2 ,
p 3 6 M 2M p
where M is the total mass of the two bodies involved, µ is their reduced mass, and p
is the semi-latus rectum of the orbit. The affect of modifying the geometry can be seen
here to be degenerate with the effect due to the solar quadrupole moment, J2 . Once
the value of this quantity is known, however, then one is able to gain constraints on
the above combination of β, γ, α1 , α2 , α3 and β2 . This can be done for any or all of
the observations of the perihelion precession of Mercury given in Section 1, and if we
take the value of γ to be that given by Eq. (71), as well as11 α1 ∼ α2 ∼ α3 ∼ β2 ∼ 0
and a reasonable value of J2 ∼ 10−7 , then this gives constraints on β of the order
β − 1 ∼ O(10−3 ) or O(10−4 ). However, as already noted, these constraints are somewhat
sensitive to a number of assumptions about the orbits of the other planets, as well as the
solar quadrupole moment.
The Nordtvedt effect is similarly an observation of time-like geodesics. In this case it
is convenient to define the ‘Nordtvedt parameter’
10 2 2 1
ηN ≡ 4β − γ − 3 − ξ − α1 + α2 − β1 − β2 , (72)
3 3 3 3
which is not to be confused with the equivalence principle violation parameter η defined
in Equation (15). The observations of Williams, Turyshev and Boggs [1277] then give the
constraint ηN = (4.4 ± 4.5) × 10−4 , which, if we again take γ to be given by observations
of the Shapiro time delay effect with all other PPN parameters being zero, gives us
β − 1 = (1.2 ± 1.1) × 10−4 , (73)
which is a much cleaner constraint on β than those which can be derived from observations
of the perihelion precession of Mercury.
In ‘conservative’ theories of gravity it is usually only the PPN parameters β and
γ (and sometimes ξ) that vary from their general relativistic values. These quantities
are often interpreted as the degree of non-linearity in the gravitational theory, and the
amount of spatial curvature per unit mass that is produced, respectively. The other
parameters ξ, αi and βi are usually interpreted as corresponding to preferred location
effects, preferred frame effect and the violation of conserved quantities. When considering
theories in which such effects are expected to be absent it is therefore usual to assume
that these parameters are all zero, and to search instead for constraints on β and γ.
Of course, one can subject the ξ, αi and βi parameters to observational scrutiny in a
number of ways. The table below gives a selection of the tightest constraints currently
available:

11 These values will be given some justification shortly.

35
 Parameter Limit Source
ξ 10−3 Ocean tides [1274]
α1 10−3 Lunar laser ranging [928]
α2 4 × 10−7 Alignment of Sun’s spin axis with ecliptic [987]
α3 4 × 10−20 Pulsar acceleration [1176]

Further constraints and discussion on the βi parameters can be found in [1274]. For more
details of the observations leading to these constraints on ξ and αi the reader is referred
to the source material cited above and [1274].
The constraints on the PPN parameters that we have discussed above are all, to
date, in reasonably good agreement with General Relativity, and it is likely that future
observations of, for example, the ‘double pulsar’ [840, 745] will tighten these constraints
even further in coming years. This excellent concordance of numerous different physical
phenomena means that one must reconcile any alterations to General Relativity with
observations in weak field systems that appear to be narrowing down on a general rel-
ativistic description. As we will describe in the sections that follow, this places tight
constraints on a variety of different modified theories of gravity: It must be the case
that any alternative theories that we consider should reproduce General Relativity in
the appropriate weak field limit, or at least something very close to it.
There are a number of mechanisms that have been considered in the literature that
allow for a general relativistic weak field limit even in theories that are, in general, very
different from General Relativity. These include the Vainshtein mechanism [1241] which
occurs when large derivative interactions are present, the Chameleon mechanism of theo-
ries with non-minimal coupling to scalar fields [689], as well as the attractor mechanism of
Damour and Nordtvedt [356]. These different approaches allow, potentially, for theories
that deviate considerably from General Relativity to exist without disturbing gravita-
tional physics in the solar system to a large extent. They are thought to be successful in
a number of different environments, and have sometimes been applied to situations that
are quite different to the ones in which they were originally conceived.
As well as successful reproductions of general relativistic behaviour, however, there
have also been a number cases found in the literature of theories that produce weak field
gravity that is surprisingly inconsistent with the predictions of General Relativity. Per-
haps the most famous of these is the van Dam-Veltman-Zakharov (vDVZ) discontinuity
that was originally found in the context of Pauli-Fierz gravity [1243, 1296] (a theory with
one dynamical metric, and one non-dynamical a priori specified metric). Here the gravi-
ton acquires a mass through the introduction of terms into the gravitational Lagrangian
that, in the weak field limit, look like mass terms for the perturbations hµν around
Minkowski space, i.e. like m2 hµν hµν . Naively one might then expect in the limit m → 0,
when the graviton becomes massless, that the zero mass theory of General Relativity
should be recovered. This is, however, not the case. Instead one finds from the study of
linear perturbations around Minkowski space that γ → 1/2, which can be seen from the
constraints above to be in strong disagreement with a number of different observations,
including light bending and time delay effects. The general relativistic limit in this case
is therefore a singular one, and any finite but non-zero graviton mass, no matter how
small, appears to give results that are incompatible with observations. Similar results
36
have also been found in some theories of gravity constructed from more general functions
of the Ricci curvature than the Einstein-Hilbert action [294], and are expected in other
theories as well. In these cases one must either abandon the theory as being incompatible
with observations, or show that the treatment being applied is unsatisfactory because,
for example, one of the mechanisms discussed previously should be applied.
Issues such as those just discussed can make the study of weak field gravity in modified
theories a more complicated subject than it is in General Relativity. One must be careful
to make sure that the treatments being applied are justifiable, that the limits of the theory
take the expected form (rather than being singular), and that non-linear mechanisms and
non-perturbative effects are being fully taken into account. How this should be done for
specific modified theories of gravity will be the subject of subsequent sections. In some
cases it is still an active area of research.

2.6. Cosmology
We now turn to cosmology, which forms a major part of this review. In this sec-
tion we first describe cosmology from the point of view of General Relativity, including
Friedmann-Lemaı̂tre-Robertson-Walker (FLRW) solutions, cosmic distance measures and
cosmological perturbation theory. We then consider the observational evidence that has
led to the rise of the “Dark Sector”, thus arriving at the so-called ΛCDM ‘concordance
model’. We end this section with a short discussion of the successes of ΛCDM, its pre-
dictions and potential shortcomings.

2.6.1. The Friedmann-Lemaı̂tre-Robertson-Walker solutions

The Robertson-Walker metric in the synchronous coordinate system is

ds2 = −dt2 + a2 (t)qij dxi dxj , (74)

where qij is a maximally symmetric 3-metric of Gaussian curvature κ. In a spherically

symmetric coordinate system this can be written
 
2 2 2 dr2 2
ds = −dt + a + r dΩ
1 − κr2

where κ is a real constant. If κ = 0, the hyper-surfaces of constant t are flat, if κ > 0

they are positively curved, and if κ < 0 they are negatively curved. The function a(t) is
called the scale factor, and we assume it to be normalised to unity today.
The Hubble parameter is defined as
1 da
H= , (75)
a dt
with H0 = H(a = 1) being the Hubble constant, i.e. the value of the Hubble parameter
at the present time.
The dynamics of the scale factor is given by the Friedmann equation
X
3H 2 = 8πG ρi (76)
i

37
where ρi are the energy densities of all possible fluids, including photons ργ , neutrinos
ρν (possibly with mass mν ), pressureless matter ρm , and spatial curvature ρκ . We may
also define the relative densities
ρi
Ωi = , (77)
ρT
3H 2
P
where ρT is the total energy
P density ρT = 8πG = i ρi . The Friedmann equation then
becomes the constraint i Ωi = 1.
If each fluid is uncoupled then energy-momentum conservation gives

ρ̇i + 3H(1 + wi )ρi = 0, (78)

where w is the equation of state, defined by P = wρ. For the known forms of matter
wγ = 13 , wm = 0, wκ = − 31 , and wν is in the range [0, 13 ]. We may solve Eq. (78) for a few
cases of interest, and then determine the dynamics of the scale factor. For radiation we
3κ −2
obtain ρr = ρ0r a−4 , for pressureless matter ρm = ρ0m a−3 , for curvature ρκ = − 8πG a ,
1
and for a cosmological constant ρΛ = 8πG Λ.
A general analytic solution in the case where all the above fluids are present is im-
possible. However, analytic solutions can be found in certain special cases. If a single
fluid is present and w is constant then a3(1+w)/2 = 3(1+w) 2 H0 t, provided w 6= −1. For
√
2/3
the case of radiation we get a = 2H0 t, and for pressureless matter a = 3H20 t . The
case of a cosmological constant is special: One obtains a = eH0 t , the de Sitter solution,
in which space (in this coordinate system) is exponentially expanding.
In many cases of interest it is convenient to use a different time coordinate, the
conformal time, τ , defined by dt = adτ . In a radiation dominated universe we then have
a(τ ) = H0 τ , in a matter dominated universe a = 41 (H0 τ )2 , and for the de Sitter universe
a = H0 (τ1∞ −τ ) , where τ∞ is the value of the conformal time at a → ∞. In a universe
√
filled with both radiation and matter we get a = Ω0r H0 τ + Ω0m 2
4 (H0 τ ) . A summary
of these solutions is shown in Table 2.

Matter type a(t) a(τ )

radiation a = (2H0 t)1/2 a = H0 τ
dust a = ( 32 H0 t)2/3 a=√ 1
4 (H0 τ )
2
Ω0m 2
radiation & dust complicated a = Ω0r H0 τ + 4 (H0 τ )
Λ a = e H0 t a = H0 (τ1∞ −τ )

Table 2: A summary of particular solutions to the Friedmann equation.

2.6.2. Cosmological distances

Given a Friedmann universe obeying Einstein’s field equations, it is useful to define
observables that characterise the background evolution. Distances play an important role
if we are to map out its behaviour (see [612] for a more detailed explanation). Hubble’s
law v = H0 d allows us to define a Hubble time, tH = H10 = 9.78 × 109 h−1 yr and the
Hubble distance, DH = Hc0 = 3000 h−1 Mpc. We can also integrate along a light ray to

38
get the comoving distance:
Z t0
cdt0
DC = c .
t a(t0 )
2
From −κ = Ωκ /DH , and performing the radial integral (assuming the observer is at
r = 0), we have
 DH √
Z DM 

√
Ωκ
sinh−1 [ Ωκ DM /DH ] for Ωκ > 0
dr D for Ωκ = 0
DC = √ = M p
0 1 − κr2 
 √DH sin−1 [ |Ωκ |DM /DH ] for Ωκ < 0,
|Ωκ |

where the proper motion distance (also known as the transverse comoving distance) is
DM . This can be rewritten as
 DH √
 Ωκ sinh[ Ωκ DC /DH ]
 for Ωκ > 0
√

DM = D C p for Ωκ = 0

 √DH sin[ |Ωκ |DC /DH ] for Ωκ < 0.
|Ωκ |

It is then possible to find an expression for the angular diameter distance:

DM
DA = .
1+z
Hence, if we know the proper size of an object and its redshift we can work out, for
a given universe, the angular diameter distance, DA . If we measure the brightness or
luminosity of an object, we know that the flux of that object at a distance DL is given by
L
F = 4πD 2 , where DL is aptly known as the luminosity distance and is related to other
L
distances through:
DL = (1 + z)DM = (1 + z)2 DA .
This relation is a consequence of Etherington’s theorem [474], and holds in any metric
theory of gravity, irrespective of the field equations. It is however violated if the photon
number is not conserved (e.g. due to photon-axion mixing), or if photons are extinguished
due to the presence of dust. It turns out that in astronomy one often works with a
logarithmic scale, i.e. with magnitudes. One can then define the distance modulus:
 
DL
DM ≡ 5 log ,
10 pc

which can be measured from the apparent magnitude, m, (related to the flux at the
observer), and the absolute magnitude, M , (what the magnitude would be if the observer
was at 10 pc from the source) through m = M + DM .
Finally, let us consider Hubble’s law. Take two objects that are a distance d apart,
and Taylor expand the scale factor today to find
1
a(t) = a(t0 ) + ȧ(t0 )[t − t0 ] + ä(t0 )[t − t0 ]2 + · · · .
2
39
On small scales the distance to the emitter is roughly related to the time of emission, t,
by d ' c(t0 − t). We can then rewrite the above expression as
 2
−1 d q0 H02 d
(1 + z) = 1 − H0 − + ··· ,
c 2 c

where q0 = −äa/ȧ2 |t=t0 is the deceleration parameter. On small scales and at small
redshifts we then have Hubble’s law, cz = H0 d.
To constrain the background evolution it is necessary to have good distance mea-
surements. So, for example, with measurement of supernovae light curves at different
redshifts it is, in principle, possible to measure DL (z). Alternatively one might try to
measure DA (z) by observing known length scales in the universe. This has been done
spectacularly well with the sound horizon of the cosmic microwave background at redshift
z ' 1100. More tentatively, there is a constraint on a combination of DA (z) and H(z)
using the imprint of acoustic oscillations of baryons on galaxy clustering at moderate to
low redshifts, z ' 0.1 − 0.3.

2.6.3. Perturbation theory

We now turn to perturbation theory, which is an indispensable tool for making predic-
tions for a variety of cosmological observations. For extensive treatments of cosmological
perturbation theory the reader is referred to [90, 717, 469, 917]. Here we shall only
consider scalar fluctuations, for which the perturbed FLRW metric can be written
    
ds2 = a2 −(1 − 2Ξ)dτ 2 − 2(∇ ~ i β)dτ dxi + 1 + 1 χ qij + Dij ν dxi dxj , (79)
3

where Dij ≡ ∇~ i∇ ~ i is
~ j − 1 qij ∆ is a trace-less spatial derivative operator. We note that ∇
3
the covariant derivative compatible with the 3-metric qij . Perfect fluids with shear have
energy-momentum tensors that can be written as

Tµν = (ρ + P )uµ uν + P gµν + Σµν , (80)

where ρ is the energy density, P is the pressure, uµ the 4-velocity of the fluid (normalised
to uµ uµ = −1), and Σµν is the anisotropic stress tensor which obeys uµ Σµν = Σµµ = 0.
In a homogeneous and isotropic space Σµν = 0, and uµ is aligned with the time direction
such that in the coordinate system used above it has components uµ = (a, ~0). For first
order scalar perturbations we can parameterise T µν as

T 00 = −ρδ (81)
T 0i = −(ρ + P )∇~ iθ (82)
T i0 ~ i (θ − β)
= (ρ + P )∇ (83)
T ij = δP δ ij + (ρ + P )Dij Σ, (84)

~ i θ). Here δP is the pressure perturbation, and

while the fluid velocity is uµ = a(1 − Ξ, ∇
Σ the scalar anisotropic stress.
For any variable X, its perturbation δX is not necessarily an observable quantity,
and may depend on a gauge. In particular, one can always define a new perturbation
40
δX0 = δX+Lξ X̄ through the Lie derivative acting on the background tensor X̄ through a
vector field ξ µ . The perturbations δX are thus in general gauge-dependent12 . For scalar
perturbations we can write ξµ = a(−ξ, ∇ ~ i ψ), and then find how our variables transform
under gauge transformations using the Lie derivative. All of them, apart from Σ, are
gauge-dependent, with transformations given by
0
Ξ → Ξ − ξa β → β + a1 [ξ + Hψ − ψ 0 ]
χ → χ + a1 [6Hξ + 2∆ψ] ν → ν + a2 ψ
δ → δ − a3 (1 + w)Hξ θ → θ + a1 ξ
δP δP 1 0
ρ → ρ + a [w − 3w(1 + w)H] ξ Σ → Σ.

0
where H = aa .
Given our set of perturbation variables, two linear combinations of them can be
removed13 (set to zero). Popular gauges are
• Newtonian gauge: ν = β = 0. The remaining metric perturbations give rise to the
Newtonian potentials Φ = − 16 χ and Ψ = −Ξ.
• Synchronous gauge: Ξ = β = 0 (this does not completely fix the gauge). The
remaining metric perturbations are related to the Ma-Bertschinger [841] variables
as χ = h and −k 2 ν = h + 6η.
• Comoving gauge: θ = ν = 0. Strictly speaking there is a multitude of comoving
gauges depending on which velocity θ is set to zero. Thus we may speak of a
”baryon comoving gauge” if θb =P0, a ”photon comoving gauge” if θγ = 0, the total
X (ρX +PX )θX
matter comoving gauge if θT = P (ρX +PX ) = 0, etc.
X

• Uniform density gauge: δ = ν = 0. Once again there is a multitude of uniform

density gauges depending on which density fluctuation is set to zero, as in the
comoving gauges above.
• Spatially flat gauge: χ = ν = 0.
It is possible to find combinations of perturbation variables that are gauge invariant,
but note that there are an infinite number of them as any linear combination of gauge-
invariant variables is also gauge-invariant. Two popular gauge-invariant metric variables
are the Bardeen potentials Φ̂ and Ψ̂:
1 1
Φ̂ ≡ − (χ − ∆ν) + H(ν 0 + 2β), (85)
6 2
and
1 1
Ψ̂ ≡ −Ξ − (ν 00 + 2β 0 ) − H(ν 0 + 2β). (86)
2 2

12 The Stewart-Walker lemma [1188] states that the only gauge-invariant perturbed tensors are those

that have background values that are either zero or a constant multiple of the identity matrix.
13 One has to be careful and not over constrain the gauge by removing two combinations that transform

with the same gauge variable, e.g. δ and θ both transform with ξ and therefore cannot be set to zero
simultaneously.
41
The Newtonian gauge is special in this case as Φ̂ = Φ and Ψ̂ = Ψ. From now on we
will refer to Φ and Ψ without a “hat” as the Newtonian gauge potentials. The Einstein
equations in the Newtonian gauge give
X
2(∆ + 3κ)Φ − 6H(Φ0 + HΨ) = 8πGa2 ρi δi (87)
i
X
0 2
2(Φ + HΨ) = 8πGa (ρi + Pi )θi (88)
i

  X
1 1
Φ + HΨ + 2HΦ + 2H + H + ∆ Ψ − ( ∆ + κ)Φ = 4πGa2
00 0 0 0 2
δPi (89)
3 3 i

and
X
Φ − Ψ = 8πGa2 (ρi + Pi )Σi (90)
i

Combining Eqs. (87) and (88) we can find Φ in terms of the matter variables as
X
2(∆ + 3κ)Φ = 3H2 Ωi [δi + 3H(1 + wi )θi ] (91)
i

while Ψ is then obtained using Eq. (90).

Finally, all scalar modes can be decomposed in terms of a complete set of eigen-modes
of the Laplace-Beltrami operator. For example, a variable A can be decomposed as
Z
A(xi , t) = d3 k Y (xj , kk ) Ã(ki , t),


where the eigen-modes, Y (xj , kk ), obey ∆ + k 2 Y = 0 . In the special case of topolog-
j
ically trivial and spatially flat hyper-surfaces of constant t, we simply have Y = eikj x .
The integral transform above is then a Fourier transform. The value of k depends on the
geometry and topology
p of the spatial hyper-surfaces: In the case of trivial topology k
takes values k = k∗2 − κ, where√ k∗ is continuous and obeys k∗ ≥ 0 for zero or negative
spatial curvature, while k∗ = N κ for positive spatial curvature, where N ≥ 3 is an
integer.

2.6.4. Gravitational potentials and observations

One of the main sources of information in cosmology is through the observation of
perturbations about a Friedmann background. Such perturbations can be probed through
their effects on the dynamics of particles and light, which we will now describe (see [1028]
for further details):

• Density fluctuations: Fluctuations in the matter density field, δ(x), will reflect
various properties of the cosmological model. The simplest approach is to assume
that δ(x) is a multivariate Gaussian random field that is entirely described by the
power spectrum, P (k), defined by

h|δk |2 i ≡ P (k), (92)

42
 where δk is the Fourier transform of δ(x). The shape of the power spectrum
contains a wealth of information: The amplitude of clustering as a function of scale,
its redshift dependence, how its shape on small scales is distorted by small scale
velocities (known as redshift space distortions), and acoustic features imprinted by
the baryons from pre-recombination (known as baryon acoustic oscillations) can
all be used as distance indicators. The power spectrum can be estimated from
surveys of galaxies or clusters of galaxies, the clustering properties of which can be
directly related to the amplitude of fluctuations in the density field (under certain
assumptions of how galaxies (or clusters) trace the density field (known as bias)).
• Peculiar velocities: The motion of galaxies relative to the Hubble flow, v i , is de-
scribed by the non-relativistic geodesic equation given above. In the linear regime,
the peculiar velocity can be related directly to the density field via the gravitational
potential:

k i δk
vki = −iaf H0 ,
k2
where f ≡ d ln δ/d ln a and we have assumed the general relativistic result Φ =
Ψ. Peculiar velocities will be observable through their effects on the redshift of
objects, either in redshift galaxy surveys (through their distortion of P (k)), or
when supplemented with independent distance measurements of each object (using
the Tully-Fisher relation or supernova light curves) in peculiar velocity surveys.
• Anisotropies in the Cosmic Microwave Background (CMB): The CMB will be sen-
sitive to density fluctuations, peculiar velocities, and the gravitational potentials.
It is usual to characterise anisotropies in the CMB in terms of δTT(n̂) , the dimen-
sionless deviations of the black-body temperature of the Universe in a direction
given by the unit vector, n̂. We can expand

δT (n̂) X
= a`m Y`m (n̂),
T
`m

where we haveP spherical harmonics, Y`m (n̂), and define the angular power spectrum
1 2
C` = 2`+1 m h|a`m | i. Like P (k), the C` s contain a wealth of information about
the cosmological model. It is now instructive to delve slightly further into the form
of δTT(n̂) . We can schematically split CMB anisotropies into three cosmological
contributions,

δT (n̂) δT (n̂)  δT (n̂)  δT (n̂) 

= LS
+ ISW
+ SEC
,
T T T T
where the first term encompasses all effects from the surface of last scattering, the
second term (the Integrated Sachs Wolfe effect) is due to integrated effects along
the line of sight, and the last term encompasses secondary effects such as weak
lensing of the CMB, the Sunyaev-Zel’dovich effect and other such contributions.
Let us focus on the effect of the gravitational potentials, the consequence of which
we can see through the geodesic equation for light rays given above [150]. The

43
 accumulated redshift of a beam of light along the line of sight is given by
Z z
Eobs
(1 + z) = = 1 + Ψem − Ψobs − (Φ̇ + Ψ̇)(dτ /dz) dz + higher order terms.
Eem 0

where factors such as the integrated visibility function have been ignored for sim-
plicity. The first term is the Sachs-Wolfe effect and, in the case of the CMB, will
give a redshift to the photons as they climb out of potential wells at the surface
of last scattering. The second term is the ISW effect, and depends on the time
dependence of the gravitational potentials along the line of sight, as advertised
above.
• Weak lensing: Lensing arise when photon light rays are deflected due to the
gravitational potentials along the line of sight. The deflection angle is given by
δ θ~ = −∇
~ ⊥ (Φ + Ψ)dτ , and allows us to relate the true position, θ~true , to the de-
flected position, θ~def , via θ~true = θ~def − rrLS
L
~ where rL (rLS ) is the distance to
δ θ,
the lens (between the lens and the source). In practise we probe the gradient of
the deflection through the inverse magnification matrix:
Z
∂ θ~true zS
rL rLS ~ ~
M−1 = =I+ ∇⊥ ∇⊥ (Φ + Ψ)dτ.
∂ θ~def 0 rS

This two by two matrix is parameterised by the convergence, K, and shear param-
eters γ1 and γ2 . In the case of small deflections this gives

(93)
 
1 + K + γ1 γ2
M= . (94)
γ2 1 + K − γ1
This information can be extracted from imaging surveys of distant galaxies. The
galaxy shapes (or ellipticities) will be distorted by the intervening gravitational
potentials. These distortions will induce correlations between the galaxy shapes
that will reflect the underlying cosmology. Lensing will, of course, also affect the
CMB photons as they pass through potential wells.

2.6.5. The evidence for the ΛCDM model

There is currently a consensus that in an FLRW Universe that is governed by Ein-
stein’s field equations, roughly 95% of the overall energy density must be ‘dark’ in order
to be compatible with observations. The current best fit model claims that about 25%
of this dark material is in the form of a non-relativistic, non-interacting form of matter
called dark matter, and that the remaining 70% is in the form of a non-clustering form
of energy density with a negative equation of state known as dark energy.
The broad case for Cold Dark Matter (CDM) is as follows [1028]:
• The rotation curves of galaxies tend to flatten out at large radii. This flattening
can be explained if the baryonic part of the galaxy resides in a halo of dark matter
with a density profile that falls of as 1/r2 .

44
 • Clusters of galaxies appear to have deeper potential wells than would be inferred
from baryonic matter. This is manifest in the motions of galaxies, as well as
the X-ray temperature of gas, and weak lensing measurements of the integrated
gravitational potentials. Dark matter halos surrounding clusters explain all these
observations.

• Diffusion damping during recombination is expected to wipe out all small-scale

structure in baryons, preventing the formation of galaxies at late times. Dark
matter, however, can sustain structure during the damping regime, and will seed
the formation of galaxies.
The case for dark energy has been around since the early 1980s. After the proposal of
the original models of inflation, the idea that the Universe should have Euclidean spatial
geometry became ever more entrenched in the standard lore. Given that baryons made
up a small fraction of the total energy budget, and that dark matter makes up about
25%, there was clearly a shortfall of pressureless matter at late times. Furthermore,
estimates of the ages of globular clusters of around 12-14 billion years were incompatible
with a flat, matter dominated universe [1228].
There was also tentative evidence from large-scale clustering that a flat, cold dark
matter dominated Universe could not explain some of the observations. Most notably,
an analysis of the APM galaxy catalogue in [463] seemed to show that a Universe with
a cosmological constant might explain the amount of galaxy clustering on a wide range
of scales. Now, with the advent of what has been dubbed “precision cosmology” in the
late 1990s, the evidence for dark energy has become even more compelling. In particular,
the following results make a strong case for presence of an energy density with negative
equation of state:
• Measurements of the luminosity distance of type Ia supernovae are consistent with
a universe with a cosmological constant, and inconsistent with a flat, matter domi-
nated universe or an open universe [1034, 1063]. The latest results seem to constrain
the equation of state, w = P/ρ ' −1.068+0.080
−0.082 [330, 1190].

• Measurements of the CMB anisotropies from large to small scales [732, 442], com-
bined with measurements of galaxy clustering from the Sloan Digital Sky Sur-
vey (SDSS) [1058], greatly favour a model with ΩΛ = 0.725 ± 0.016 and w =
−1.10 ± 0.14.

• The cross correlation between the ISW effect from the CMB and a variety of surveys
of large-scale structure favour w = −1.01+0.30
−0.40 , at around 4σ [534, 611].

• The number density of clusters of galaxies as a function of redshift disfavour a flat,

matter dominated universe. The presence of massive clusters at high redshift point
accelerating expansion out to redshift z ' 2 [32].

Although each individual observation may be subject to a variety of interpretations, and

different systematic effects, the overall concordance is remarkable. Indeed, the model
that best fits these observations is now known as the concordance model, or ΛCDM.

45
2.6.6. Shortcomings of the ΛCDM model
Perhaps the most serious problem with ΛCDM is the cosmological constant problem:
That the observed value of Λ is around 120 orders of magnitude smaller than the naive
4
expectation that it should be of the Planck Mass, MPl . Super-Symmetric (SUSY) theories
can lower this expectation to that of the SUSY breaking scale, but this still required a
bare Λ0 to cancel the vacuum energy coming from the SUSY symmetry breaking scale
to about 60 decimal places. One could consider arguing that some unknown physics at
high energies may provide a mechanism for achieving this level of fine-tuning, but this
seems unlikely as the problem already manifests itself at low energies.
Now, suppose that we want to describe all physics up to scales just above the electron
mass. Then the contribution to the vacuum energy Λ will include a bare term Λ1 , a term
coming from the electron and a term coming from the neutrino. This is schematically
given by
Λ = Λ1 + ce m4e + cν m4ν . . . ,
where ce and cν are coefficients. If we now lower the energy below the electron mass,
and integrate out the electron, we instead have

Λ = Λ0 + cν m4ν . . . ,

for a new bare term Λ0 . To get the same observable vacuum energy, Λ, we must now
have that Λ1 and Λ0 cancel to 32 decimal places.
It may be thought that there could exist some mechanism that relaxes the effective
cosmological constant14 to zero dynamically, but Weinberg [1265] has shown that this is
impossible. Suppose that there is a set of N scalars, φA , that are responsible for driving
the effective Λ to zero. These scalars will contribute an effective potential, V (φA ), to
the cosmological constant. If we are to approach a global Minkowski metric at these
energy levels, then V (φA ) must cancel the other contributions to Λ to high accuracy as
the fields settle to the minimum. However, this is hardly a readjustment mechanism:
If the cosmological constant changes slightly, then the mechanism fails. This proof as-
sumes Poincaré invariance in the scalar sector, which could, however, be considered an
unnecessary assumption (see Horndeski’s theory in Section 3.1.3).
The present value of Λ, as implied by cosmological observations, has another potential
problem associated with it: It has an energy density of the same order of magnitude as
the average matter density in the Universe today,

ρΛ |a=1 ∼ ρm |a=1 .

These two quantities scale with the size of the Universe in very different ways, and so
their similarity at the present time appears naively to be somewhat of a coincidence.
Hence, this problem is sometimes referred to as the coincidence problem.
Aside from the problems of the cosmological constant, there are some problems that
plague dark matter as well. The first is another coincidence problem: Why is the dark
matter energy density so close to the baryon energy density? This is actually worse
than it might seem. Baryons are produced non-thermally, out of equilibrium. CDM is

14 By effective cosmological constant we mean the Ricci curvature of the vacuum.

46
usually thought to be produced thermally, as weak interaction cross-sections naturally
give rise to the right dark matter abundance via thermal production. But how can two
components that have very different production mechanisms have very similar energy
densities15 ? Solutions to this puzzle have been proposed [91, 679, 1205, 699, 680, 365,
210, 587, 586, 876, 27] but they typically require additional particles to those that form
the dark matter, and there is as yet no well accepted mechanism.
Other problems with dark matter are observational, and we will discuss them only
briefly. The density profile of CDM, as determined from N-body simulations, is inferred
to be cuspy. For example the Navarro-Frenk-White (NFW) profile [946] gives ρCDM ∝ 1r
close to the centre of a halo. Other simulations give similar results: ρCDM ∝ r−α with
α ∼ 0.7 − 1.5. Galaxies, however, are observed to have cores such that ρ flattens out
at the centre. This is the cusp problem [366] and proposed solutions within the CDM
paradigm include self-interacting dark matter [1174], fuzzy dark matter [628], or various
feedback processes that expel dark matter. Note that simulations do not have enough
resolution to probe the small scales where the problem manifests itself, but rely instead
on extrapolations. However, simulations with increasingly smaller resolutions (although
still above the probed scales) have not indicated any kind of alleviation to the cusp
problem.
Another problem is that of missing satellites [867, 211]. The CDM paradigm predicts
a rich sub-structure within the main galactic halo that should lead to numerous dwarf
galaxies orbiting the main galaxy. Indeed, simulations indicate that about 500 satellite
galaxies should be orbiting the Milky way [909]. On the contrary, however, only about
30 such dwarfs have been observed. A possible resolution within the CDM paradigm is
that most of these galaxies are dark galaxies, i.e. have very little or no stars in them,
and are instead completely dominated by dark matter [1146].
A third problem is the tight correlation between dark matter and baryons in galaxies
that manifests itself in a universal acceleration scale, a0 ∼ 1.2 × 10−10 m s−2 [895, 1094],
the Tully-Fisher relation [879, 878], and the Faber-Jackson relation [1092]. Within the
CDM paradigm, such correlations are not expected to be present, as baryons should not
know how the dark matter behaves. For further apparent discrepancies between ΛCDM
and small scale observations the reader is referred to [756].
On cluster scales and larger, the ΛCDM model can boast of success coming from a
host of observations: Strong and weak lensing of clusters, X-ray observations of clusters,
the CMB angular power spectrum, the matter power spectrum, P (k), and supernova
data. Yet there are a few cases of interesting discrepancies. The collision velocity of
the bullet cluster [318] may be so large that the probability of it occurring in a ΛCDM
scenario is at best ∼ 10−9 [780]. In [510], however, the opposite conclusion is reached, so
this appears far from settled. Cosmological voids seem to be more empty of galaxies than
expected, as has been championed by Peebles [1030]. The CMB angular power spectrum
has a lack of large-scale power above 60o [1172] (although the statistical significance of
this is debatable, due to cosmic variance). Certain violations of statistical isotropy or
other anomalies on large scales in the CMB have also been reported [591, 337, 1294]. It
remains to be seen whether these are really problems with ΛCDM, if they are due to

15 There are also non-thermal candidates for dark matter, e.g. axions, but this does not change the

argument.

47
systematic effects, or if they are statistical flukes. These difficulties do, however, provide
some motivation for looking at alternatives to ΛCDM.

48
3. Alternative Theories of Gravity with Extra Fields

In General Relativity the gravitational force is mediated by a single rank-2 tensor

field, or a massless spin-2 particle in the quantum field theory picture. While there
is good reason to couple matter fields to gravity in this way, there is less reason to
think that the field equations of gravity should not contain other fields, and one is in
general free to speculate on the existence of such additional fields in the gravitational
sector. The simplest scenario that one could consider in this context is the addition of
an extra scalar field, but one might also choose to consider extra vectors, tensors, or
even higher rank fields [511, 1247]. Of course, the effect of such additional fields needs
to be suppressed at scales where General Relativity has been well tested, such as in
the lab or solar system. This is usually achieved making couplings very weak, although
novel screening mechanisms such as the chameleon mechanism [689, 688] and Vainshtein
mechanism [1241] have also been explored.
This section represents an overview of four-dimensional gravity theories with extra
fields, focusing on additional scalars, vectors and tensors. We note that some theories
in other sections of this review can also be considered as theories with extra fields (e.g.
f (R) gravity, galileons, and ghost condensates). The reader is referred to later sections
for details of this.

3.1. Scalar-Tensor Theories

The scalar-tensor theories of gravity are some of the most established and well studied
alternative theories of gravity that exist in the literature. They are often used as the
prototypical way in which deviations from General Relativity are modelled, and are of
particular interest as the relatively simple structure of their field equations allow exact
analytic solutions to be found in a number of physically interesting situations. Scalar-
tensor theories arise naturally as the dimensionally reduced effective theories of higher
dimensional theories, such as Kaluza-Klein and string models. They are also often used
as simple ways to self-consistently model possible variations in Newton’s constant, G.

3.1.1. Action, field equations, and conformal transformations

A general form of the scalar-tensor theory can be derived from the Lagrangian density
[142, 988, 1253]
1 √
L= −g [f (φ)R − g(φ)∇µ φ∇µ φ − 2Λ(φ)] + Lm (Ψ, h(φ)gµν ), (95)
16π
where f, g, h and Λ are arbitrary functions of the scalar field φ and Lm is the Lagrangian
density of the matter fields Ψ. The function h(φ) can be absorbed into the metric by a
conformal transformation of the form [422]

h(φ)gµν → gµν . (96)

The conformal frame picked out by this choice is one in which there is no direct interaction
between the scalar field and matter fields, and is usually referred to as the Jordan frame.
As discussed in previous sections, test-particles in this conformal frame follow geodesics
of the metric to which they are coupled, and the weak equivalence principle is satisfied
for massless test particles. The effect of this transformation on the remainder of the
49
Lagrangian can then be absorbed into redefinitions of the as yet unspecified functions f ,
g and Λ.
By a redefinition of the scalar field φ we can now set f (φ) → φ, without loss of
generality. The Lagrangian density (95) can then be written as
 
1 √ ω(φ)
L= −g φR − ∇µ φ∇µ φ − 2Λ(φ) + Lm (Ψ, gµν ), (97)
16π φ

where ω(φ) is an arbitrary function, often referred to as the ‘coupling parameter’, and
Λ is a φ-dependent generalisation of the cosmological constant. This theory reduces
to the well known Brans-Dicke theory [184] in the limit ω → constant and Λ → 0,
and approaches General Relativity with a cosmological constant in the limit ω → ∞,
ω 0 /ω 2 → 0 and Λ → constant.
The variation of the action derived from integrating (97) over all space, with respect
to g µν , gives the field equations
 
1ω 2 ω
φGµν + φ + (∇φ) + Λ gµν − ∇µ ∇ν φ − ∇µ φ∇ν φ = 8πTµν . (98)
2φ φ

Now, as well as the metric tensor gµν , these theories also contain the dynamical scalar
field φ, and so we must vary the action derived from Eq. (97) with respect to this
additional degree of freedom. After eliminating R with the trace of (98), this yields

(2ω + 3)φ + ω 0 (∇φ)2 + 4Λ − 2φΛ0 = 8πT. (99)

where primes here denote differentiation with respect to φ. These are the field equations
of the scalar-tensor theories of gravity.
It is well known that these theories admit the very useful property of being ‘confor-
mally equivalent’ to General Relativity. By this we mean that under a transformation
of the metric that alters scales, but not angles, one can find a new metric that obeys
the Einstein equation, with the scalar contributing as an ordinary matter field. This
does not, however, mean that scalar-tensor theories are the same as General Relativity,
as the metric that couples to matter fields must also transform. The theory that is re-
covered after conformally transforming is one in which the metric obeys a set of fields
equations similar to Einstein’s, but with an unusual matter content that does not follow
geodesics of the new metric (with the exception of radiation fields, or null geodesics,
which are themselves conformally invariant). This property of scalar-tensor theories can
sometimes allow their field equations to be manipulated into more familiar forms, that
allow solutions to be found more readily.
To be explicit, a conformal transformation of the metric gµν into ḡµν can be written

gµν = e2Γ(x) ḡµν , (100)

where Γ(x) is an arbitrary function of the space-time coordinates xµ . The line-element

is then correspondingly transformed as ds2 = e2Γ(x) ds̄2 , and the square root of the
√ √
determinant of the metric as −g = e4Γ −ḡ (in four dimensions). After performing
such a transformation we can use the term ‘conformal frame’ to distinguish the new,
rescaled metric from the original.
50
 Among the infinite possible conformal frames we can then identify two which have
particular significance: The Jordan frame and the Einstein frame. The Jordan frame is
the one in which the energy-momentum tensor is covariantly conserved and in which test-
particles follow geodesics of the metric. This is the frame picked out by the transformation
(96), and is the one in which scalar-tensor theories are most usually formulated. The
Einstein frame is the conformal frame in which the field equations of the theory take the
form of the Einstein equations with the scalar contributing as an ordinary scalar field,
as discussed above.
Under the transformation (100) it can be shown that the Ricci tensor and Ricci scalar
transform as


Rµν = R̄µν − 2∇ ¯ µ∇¯ ν Γ + 2∇¯ µ Γ∇
¯ ν Γ − 2∇
¯ α Γ∇
¯ α Γ + Γ
¯ ḡµν (101)
2Γ
e R = R̄ − 6∇ ¯ µ Γ∇¯ Γ − 6Γ,
µ ¯ (102)

while the d’Alembertian transforms as e2Γ φ = φ¯ + 2∇ ¯ µ Γ∇

¯ µ φ. Here, over-bars on op-
erators or indices denote that they are defined using the metric ḡµν . Under these trans-
formations we will now show how the scalar-tensor theories defined by the Lagrangian
(97), in the Jordan frame, can all be transformed into the Einstein frame.
First, consider the term in Eq. (97) containing the Ricci scalar, which under the
conformal transformation (100) becomes
1 √ ¯
L1 = −ḡφe2Γ (R̄ − 6ḡ µν Γ,µ Γ,ν − 6Γ). (103)
16π
The non-minimal coupling to the Ricci scalar can now be removed by making the choice
of conformal factor e2Γ = φ−1 such that gµν = ḡµν /φ. This defines the conformal trans-
formation between Jordan and Einstein frames in the scalar-tensor theories. Applying
the transformation (100) to the rest of Eq. (97) then gives
1 √

L= −ḡ R̄ − 2(3 + 2ω)∇ ¯ µ Γ∇
¯ µ Γ − 2e4Γ Λ + Lm (Ψ, e2Γ ḡµν ). (104)
16π
p
Now, by making the definitions 4π/(3 + 2ω) ≡ ∂Γ/∂ψ and 8πV (ψ) ≡ e4Γ Λ, for the
scalar ψ and the function V (ψ), we can write the transformed Lagrangian (104) as
 
1 √ √ 1¯ ¯ µ ψ + V (ψ) + Lm (Ψ, e2Γ ḡµν ).
L= −ḡ R̄ − −ḡ ∇µ ψ ∇ (105)
16π 2
In the absence of any matter fields the scalar-tensor theories can now be clearly seen to
be conformally related to Einstein’s theory in the presence of a scalar field in a potential.
This potential disappears when Λ = 0
In the Brans-Dicke theory [184] the coupling parameter ω is constant, and the scalar
fields φ and ψ are therefore related by
s
16π
ln φ = ψ.
(3 + 2ω)

For more general theories with ω = ω(φ) the definition of ψ must be integrated to obtain
a relation between φ and ψ. By extremising the action (105) with respect to ḡµν and ψ
51
we get the Einstein frame field equations
   
¯ ¯ 1¯ ¯ α
Ḡµν = 8π T̄µν + ∇µ ψ ∇ν ψ − ∇α ψ ∇ ψ + V ḡµν (106)
2
and
√
˜ − dV = − 4παT̃
ψ (107)
dψ
where α−2 = 3 + 2ω and where we have defined the energy-momentum tensor T̄µν with
respect to ḡµν so that T̄ µν = e6Γ T µν . It can now be explicitly seen that while the Jordan
frame energy-momentum tensor is covariantly
√ conserved, ∇µ T µν = 0, its counterpart in
¯ µν
the Einstein frame is not, ∇µ T̄ = 4παT̄ ∇ ψ. ¯ ν

3.1.2. Brans-Dicke theory

The Brans-Dicke theory is given by the Lagrangian density (97) with ω =constant,
and Λ = 0 [184]. The behaviour of this theory in the vicinity of isolated masses is well
understood, and in the case of static and spherical symmetry can be solved exactly by
the line-element [185]
ds2 = −e2α dt2 + e2β (dr2 + r2 dΩ2 )
where α = α(r) and β = β(r) are given by one of the following four solutions:
" # λ1
B
α α0 1− r
I e =e B
1+ r
 2 " B
# (λ−C−1)
λ
B 1 −
eβ = eβ0 1 + r
r 1 + Br
" # Cλ
1 − Br
φ = φ0
1 + Br
2 r
II α = α0 + tan−1
Λ B  
2(C + 1) r r2
−1
β = β0 − tan − ln
Λ B (r2 + B 2 )
−1 r
φ = φ e Λ tan ( B )
2C
0
r
III α = α0 −
B
r (C + 1)r
β = β0 − 2 ln +
B B
Cr
φ = φ0 e − B
1
IV α = α0 −
Br
(C + 1)
β = β0 +
Br
C
− Br
φ = φ0 e .
52
Here we have defined λ2 ≡ (C + 1)2 − C (1 − ωC/2) > 0 in solution I, and Λ2 ≡
C (1 − ωC/2) − (C + 1)2√> 0 in solution II. The constant C is arbitrary in I and II,
and given by C = (−1 ± −2ω − 3)/(ω + 2) in III and IV. The constants B, α0 , β0 and
φ0 are arbitrary throughout.
Now, while solution I is valid for all values of ω, solutions II, III and IV are only
valid for ω < −3/2. Solution I is also known to be conformally related to the minimally
coupled massless scalar field solution of Buchdahl [207]. It can be seen that these solutions
are not all independent of each other. By a transformation of the form r → 1/r and
some redefinition of constants, solution II can be transformed into the ω < −3/2 range
of solution I [152] and solution III can be transformed into solution IV [153]. It was
also shown in [152] that the independent solutions I and IV are both conformally related
to the general solution of the static, spherically symmetric case in the Einstein frame, as
found by Wyman [1289].
These solutions are very useful for understanding the gravitational fields around an
isolated body in Brans-Dicke theory, but are not the only spherically symmetric vacuum
solutions of the Brans-Dicke field equations. A non-static spherically symmetric vacuum
exact solution is also known [313]:

α(1− √13β )
ds̄2 = −A(r) dt2
√
−α(1+ √13β )
2(β− 3)
√  2 
+ A(r) t 3β− 3 dr + A(r)r2 (dθ2 + sin2 θdφ2 ) , (108)

with  ± 2β
1
2C √
φ(r, t) = 1 − t2/( 3β−1) , (109)
r
√
3
√
where we have A(r) = 1 − 2C r , α = ± 2 , β = 2ω + 3, and C =constant. This
solution reduces to a flat vacuum FLRW metric in the limit C → 0 (an inhomogeneous
solution requires C 6= 0). The metric (108) is spatially homogeneous at large r and has
singularities at t = 0 and r = 2C; the coordinates r and t therefore cover the ranges
0 ≤ t < ∞ and 2C ≤ r < ∞. This solution is known to be conformally related to [635],
and shows explicitly the lack of validity of Birkhoff’s theorem in Brans-Dicke theory.
It also reduces to the Schwarzschild solution when ω → ∞. Black hole solutions in
Brans-Dicke theory with a power-law potential have been investigated in [850].
Let us now consider the weak field limit of this theory. Following the PPN prescription
outlined in previous sections one can straightforwardly find that the relevant values for
the PPN parameters are:
1+ω
βP P N = 1 and γP P N = , (110)
2+ω
with all other parameters equalling zero. The value of Newton’s constant can also be
shown to be given by  
4 + 2ω 1
G= , (111)
3 + 2ω φ0
where φ0 is the background (unperturbed) value of the scalar field. It can be seen
that in the general relativistic limit ω → ∞ we then recover the usual values of the
53
PPN parameters, and that for finite ω the only parameter that deviates from its general
relativistic value is γ.
This value of γ is valid for both the static and non-static exact solutions shown above.
It is interesting to note, however, that it is not the value of γ that one should expect
to measure outside of a black hole that has formed from gravitational collapse in this
theory. Such an object can be shown to have an external gravitational field with γ = 1
[1107], as predicted by Hawking [595]. This does not, however, mean that gravitational
collapse to a black hole proceeds in the same way in Brans-Dicke theory as it does in
General Relativity. In the Brans-Dicke case apparent horizons are allowed to pass outside
of the event horizon, scalar gravitational waves are emitted during the collapse, and the
surface area of the event horizon can decrease with time. Such behaviour does not occur
in General Relativity, and is allowed here because Brans-Dicke theory can violate the
condition Rµν k a k ν ≥ 0, where k a ka = 0. The problem of understanding black hole
thermodynamics in Brans-Dicke theory has been addressed in [674]. Here it was found
that the expression for the entropy of a black hole with an horizon Σ of area A is given
by Z
1 p φA
SBH = d2 x g (2) φ = , (112)
4 Σ 4
such that SBH is always non-decreasing, even if the area decreases. This shows that
the second law of black hole thermodynamics can indeed be extended to Brans-Dicke
theories, with the effective gravitational constant being replaced by 1/φ. For an intuitive
interpretation of this result in the Einstein frame, and for further discussion on this topic,
the reader is referred to [489].
Having discussed the gravitational fields of point-like objects in Brans-Dicke theory,
let us now proceed to use observations of weak field phenomena to constrain the theory.
This can be done most effectively using the constraint on γ given in Equation (71),
derived from observations of the time delay of radio signals from the Cassini spacecraft
as it passed behind the Sun. Together with the expression (110), shown above, this gives
the 2σ constraint on the coupling parameter

ω & 40 000. (113)

This is a very restricting constraint on the theory, and shows that deviations of this
kind from General Relativity must be very small indeed (see the following subsection,
however, for a discussion of scalar-tensor theories that can evade this bound while still
exhibiting significantly different behaviour to General Relativity in the early universe).
Let us now proceed to discuss the cosmology of Brans-Dicke theory. Using the usual
FLRW line-element, and assuming a perfect fluid matter content, the field equations
reduce to:

8πρ κ φ̇ ω φ̇2
H2 = − 2 −H + (114)
3φ a φ 6 φ2
φ̈ 8π (ρ − 3P ) φ̇
= − 3H , (115)
φ φ (2ω + 3) φ

where over-dots denote differentiation with respect to the proper time of a comoving
observer, H = ȧ/a, and ρ̇ + 3H(ρ + P ) = 0. The general solutions to Eq. (114) and (115)
54
are now fully understood [576, 95]. At early times the vacuum solutions of O’Hanlon and
Tupper [993] are recovered, while at late-times one approaches the power-law solutions
of Nariai [938] (when κ = 0):
2
a(t) = t2[1+ω(1−W )]/[4+3ω(1−W )]
(116)
2
φ(t) = φ0 t[2(1−3W )]/[4+3ω(1−W )]
, (117)
where16 p = W ρ. These solutions can be considered “Machian” in the sense that the
matter fields are driving the expansion of the Universe, rather than φ.
Let us now consider the general FLRW solutions in terms of a transformed time
coordinate η = η(t). Such solutions can be found any equations of state W [576], but
here let us consider only the radiation dominated solutions with W = 1/3. In this case
the new time coordinate η is simply the conformal time τ given by adτ ≡ dt, and the
general solution for ω > −3/2 is
1
2+
√1 1
2−
√1
2 1+ 2 ω 2 1+ 2 ω
a(τ ) = a1 (τ + τ+ ) 3 (τ + τ− ) 3 (118)
− √1 + √1
2 1+ 2 ω 2 1+ 2 ω
φ(τ ) = φ1 (τ + τ+ ) 3 (τ + τ− ) 3 (119)

where τ± , a1 and φ1 are integration constants, and where 8πρr0 /3φ1 a21 = 1. For ω <
−3/2, however, we instead find
 
q
2 exp  q −1 τ + τ− ,
a(τ ) = a1 (τ + τ− )2 + τ+ tan−1 (120)
2 τ+
3 |ω| − 1
 
2 τ + τ−
φ(τ ) = φ1  q tan−1 . (121)
2
|ω| − 1 τ +
3

For ω > −3/2 (ω < −3/2) we see that the scale factor here undergoes an initial period
of rapid (slow) expansion and at late times is attracted towards the solution a(τ ) ∝ τ ,
1
or, equivalently, a(t) ∝ t 2 . Similarly, φ can be seen to be changing rapidly at early
times and slowly at late times. These two different behaviours, at early and late times,
can be attributed to periods of free scalar–field domination and radiation domination,
respectively. If ρr0 = 0 is chosen then these solutions become vacuum ones that are
driven by the φ field alone, and for ω < −3/2 the initial singularity can be seen to be
avoided. Corresponding behaviour can also be shown to exist for other equations of state,
W . For a more detailed discussion of this phenomenon we refer the reader to [105].
Unlike in General Relativity, in the Brans-Dicke theory it is also possible to have
spatially flat and positively curved exact vacuum solutions. Spatially flat solutions can

16 Note that in this section we use an upper case W to denote the equation of state of the fluid, rather

than the usual lower case w used in the rest of the review. This is to avoid confusion with the coupling
parameter ω.

55
be found by assuming φ ∝ tx and a ∝ ty , and by setting a(0) = 0. When κ = 0 the
vacuum Brans-Dicke equations are then solved by [993]
1
√ −1
a(t) ∝ t 3 (1+2(1− 3(3+2ω)) ) , (122)
 −2(1−√3(3+2ω))−1
t
φ(t) ∝ . (123)
t0
For spatially closed solutions one can follow the method prescribed in [95]. Here one
defines a new quantity y ≡ φa2 , and uses the conformal time coordinate τ , to write the
field equations as
√
(ln φ),τ = 3Ay −1 (2ω + 3)−1/2 and y,2τ = −4κy 2 + A2 ,
where√ A is a √
constant. For κ > 0 these equations can be integrated to find y =
(A/2 κ) sin(2 κ(τ − τ0 )), which then gives the solutions
√
q
3
φ(τ ) ∝ tan ( κ(τ − τ0 )),
(2ω+3)
(124)
√
sin1/2 (2 κ(τ − τ0 ))
a(τ ) ∝ . (125)
√
q
3
tan 4(2ω+3) ( κ(τ − τ0 ))
Spatially flat and closed vacuum FLRW solutions such as those shown here do not exist
in General Relativity, and show the potential for interesting new behaviour at early times
in scalar-tensor theories of gravity. Phase plane analyses of perfect fluid FLRW solutions
to the Brans-Dicke field equations have been performed in [731, 1096, 613].
A number of anisotropic cosmological solutions of the Brans-Dicke field equations are
also known. Bianchi type-I solutions have been found in [1075, 133], type-II in [822,
580, 825], type-III solutions in [826], type-V solutions in [821, 823, 827, 828, 581], type-
V I0 and V Ih solutions in [829, 830, 131], type-V IIh solutions in [824, 579], type-V III
solutions in [825, 820], type-IX solutions in [825, 820], and Kantowski-Sachs solutions in
[826]. Inhomogeneous cosmological solutions have also been found [313], and braneworld
cosmologies have been considered in [75, 1054]. We will not reproduce any of these
solutions here, but rather refer the reader to the citations above, and references therein.
For a discussion of the cosmic no-hair theorems in Brans-Dicke theory the reader is
referred to [582], where it is shown that these theorems are valid without imposing any
strong constraints on the coupling constant, ω, so that initially anisotropic universes can
evolve towards an isotropic final state.
Now let us consider perturbed FLRW space-times, within which cosmological obser-
vations are usually interpreted. For the Brans-Dicke theory these equations have been
studied many times before, starting with [939]. Here we will present these equations in
the synchronous gauge and with κ = 0, as found in [290]. In this case the equations take
on a simpler form. For the more general case the reader is referred to [934], or to the
ω =constant limit of Eqs. (145)-(149) in Section 3.1.3 for the corresponding equations
in the conformal Newtonian gauge. Now, the perturbed metric can be written as
gµν = ḡµν + a2 (τ )hµν , (126)
where a(τ ) is the FLRW scale factor, ḡµν is the unperturbed FLRW metric with κ = 0,
and hµν is the perturbation that satisfies h00 = h0ν = 0 in the synchronous gauge. We
56
can then proceed as normal, and decompose the remaining non-zero hij perturbations
into harmonic modes, and decouple the scalar, vector and tensor components. The scalar
part of the perturbations can be written as in Section 2
1
hij = hqij + Dij ν, (127)
3
where the synchronous gauge has been adopted, while δφ is the perturbation to the Brans-
~ i∇
Dicke scalar. The Dij operator, as in Section 2, is defined by Dij = ∇ ~ j − 1 qij ∆. As
3
usual we define η = −(h + k ν)/6 (see section 2). The perturbed equations are17 :
2

   0  0
2 1 φ0 0 8πa2 X φ δφ
− 2k η + H + h = ρf δf + ω − 3H
2φ φ φ φ
f
" #
2
ω φ0 δφ
− k 2 + 3H2 + (128)
2 φ2 φ
 
0 8πa2 X 1 0 1 φ0
2η = (ρf + Pf )θf + δφ − H−ω δφ(129)
φ φ φ φ
f
 
1 00 φ0 8πa2 δφ
ν + H+ ν0 + η = (ρ + P )Σf + (130)
2 2φ φ φ

and
1 8πa2 X
δφ00 + 2Hδφ0 + k 2 δφ + φ0 h0 = (δρf − 3δPf ). (131)
2 2ω + 3
f

Here the perturbations to the energy density and pressure of the non-interacting fluids
f are written as δρf and δPf , with the peculiar velocity potentials and anisotropic
stress written as θf and Σf , respectively. Primes denote differentiation with respect to
conformal time, τ .
For the tensor modes we can write the metric perturbations as hij = h̃T Qij , where
Qij is a harmonic function, and with no tensor component involved in δφ. The evolution
equation for h̃T is then given by

8πa2 X
h̃00T + 2Hh̃0T + k 2 h̃T = (ρf + Pf )Σ̃f , (132)
φ
f

where Σ̃f is the tensor contribution to the anisotropic stress of the fluid f . We will not
write the vector perturbation equations here, which are not expected to be significant
for most cosmological applications. For perturbation equations written in terms of gauge
invariant variables the reader is referred to [1285] for the covariant approach, or [934] for
the Bardeen variable approach (for the Brans-Dicke theory one should take ω =constant
in this last reference).
The background cosmological evolution and perturbations can be used to place con-
straints on Brans-Dicke theory from a number of different sources. The CMB is one

17 Various typos in the corresponding equations in [290] have been corrected.

57
such source, and can be used to place constraints on the coupling parameter, ω. This
has been done a number of times in the literature [935, 7, 1284], with the latest results
based on constraints given by the WMAP 5 year data, the ACBAR 2007 data, the CBI
polarisation data, and the BOOMERanG 2003 flight, together with large-scale structure
data from the SDSS data release 4, giving ω > 97.8 or ω < −120.0 to 2σ [1284]. This
is in keeping with the results of [7], but significantly weaker than those claimed by [935]
of ω > 1000 to 2σ based on the WMAP first year data. Among the detailed processes
that lead to these constraints one can see that the change in the horizon size at matter-
radiation equality is altered in Brans-Dicke theory due to the different expansion rates
[799]. This length scale is imprinted on the spectrum of perturbations as during the
radiation era perturbations inside the horizon are effectively frozen, while during matter
domination perturbations grow on all length scales. Different expansion rates also affect
the horizon size at recombination, which affects the level of ‘Silk damping’ that occurs on
small scales due to viscosity and heat conduction. What is more, the thickness of the last
scattering surface is also changed, which affects anisotropy on small scales through the
exponential damping which has its cutoff determined by this quantity. The upcoming
data from the Planck satellite is, of course, expected to tighten the constraints given
above still further.
Another cosmological probe that has been extensively applied to Brans-Dicke theory
is that of the primordial nucleosynthesis of light elements [1292, 59, 354, 265, 266, 1126,
312]. In the Brans-Dicke theory the scalar field φ is approximately constant during the
epoch of radiation domination. Nucleosynthesis therefore proceeds largely as in a general
relativistic cosmology (up to the effect of ‘kicks’ on the scalar field due to the annihilation
of electron-positron pairs [358]), but with a different value of G during this process, and
hence a different expansion rate. Of course, the time at which weak interactions freeze
out in the early universe is determined by equality between the rate of the relevant
weak interactions and the Hubble rate. When the weak interaction rate is the greater
then the ratio of neutrons to protons it tracks its equilibrium value, while if the Hubble
rate is greater than the weak–interaction rate then the ratio of neutrons to protons is
effectively ‘frozen–in’, and β–decay is the only weak process that still operates with any
efficiency. This is the case until the onset of deuterium formation, at which time the
neutrons become bound and β–decay ceases. Now, the onset of deuterium formation is
primarily determined by the photon to baryon ratio, ηγ , which inhibits the formation
of deuterium nuclei until the critical temperature for photodissociation is past. As the
vast majority of neutrons finally end up in 4 He the primordial abundance of this element
is influenced most significantly by the number of neutrons at the onset of deuterium
formation, which is sensitive to the temperature of weak–interaction freeze–out, and
hence the Hubble rate, and so G, at this time. Conversely, the primordial abundances of
the other light elements are mostly sensitive to the temperature at deuterium formation,
and hence ηγ , when nuclear reactions occur and the light elements form. The reader is
referred to [83] for further discussion of these points. The typical bounds that can be
achieved on the coupling parameter from observations of element abundances are then
given by ω & 300 or ω . −30, assuming the power-law solutions (116) and (117). By
using the general solutions (118)-(121), however, these bounds can be somewhat relaxed
or tightened, depending on the behaviour of φ in the early universe [312].
While the cosmological bounds discussed above are weaker than those derived in the
solar system, and in binary pulsars, they probe a very different physical environment and
58
scale. They are therefore usually considered complimentary to the constraints imposed
from observations of weak field gravity, and a useful consistency check. After all, one may
wish to consider theories in which the coupling parameter ω varies throughout cosmic
history. Theories in which such behaviour can occur explicitly are the subject of Section
3.1.3.

3.1.3. General scalar-tensor theories

The Brans-Dicke theory that has so far been considered is a very special scalar-tensor
theory, with only a single constant parameter. The more general class of scalar-tensor
theories contains two free functions, given by ω(φ) and Λ(φ) in Eq. (97). Let us now
consider these more general theories.
First of all let us consider the case in which Λ(φ) = 0. Such theories have been
well studied in the literature, and are often used to model the possibility of having
a coupling parameter ω in the early universe that is small enough to have interesting
effects, while being large enough in the late universe to be compatible with the stringent
bounds imposed upon such couplings by observations of gravitational phenomena in the
solar system, and other nearby astrophysical systems. This interest is bolstered by the
presence of an attractor mechanism that ensures General Relativity is recovered as a
stable asymptote at late times in FLRW cosmology [356]. We will explain this attractor
in more detail below.
Of course, in generalising the Brans-Dicke theory we want to know what the con-
sequences are for constraints imposed in the weak field limit. The extra complication
caused by allowing ω to be a function of φ means that exact solutions are hard to find.
Perturbative analyses can still be readily performed, however, leading to the PPN pa-
rameters
dω/dφ 1+ω
βP P N = 1 + and γP P N = , (133)
(4 + 2ω)(3 + 2ω)2 2+ω
with all other parameters equalling zero. The value of γ here can be seen to be the same as
in the Brans-Dicke theory, while the value of β reduces to the Brans-Dicke (and General
Relativity) value of unity when ω =constant. Observations from the Cassini satellite
therefore place upon ω the same tight constraint as in Brans-Dicke theory (ω & 40 000
to 2σ). This constraint, however, now only applies to the local value of ω (i.e. with
the present day value of φ in the solar system). The variation of ω with φ can then be
constrained by observations of post-Newtonian phenomena that constrain β, such as the
lunar laser ranging experiments described in previous sections. To constrain ω for other
values of φ, however, requires making observations in other physical environments, such
as in the early universe, or near black holes.
Let us now consider the cosmological solutions of these theories. It has been shown
by Clarkson, Coley and O’Neill in [303] that the Ehlers-Geren-Sachs theorem can be
extended to cover scalar-tensor theories of gravity. Taking the FLRW line-element, and
assuming a perfect fluid matter content, the field equations in this case reduce to

8πρ κ φ̇ ω φ̇2
H2 = − 2 −H + (134)
3φ a φ 6 φ2
φ̈ 8π (ρ − 3P ) φ̇ (dω/dφ)φ̇2
= − 3H − , (135)
φ φ (2ω + 3) φ (2ω + 3)φ
59
where over-dots again denote differentiation with respect to the proper time of comoving
observers. These equations are similar to those of the Brans-Dicke theory, Eqs. (114)
and (115), except for the extra term on the RHS of Eq. (135). Exact solutions with
κ = 0 have been found to Eqs. (134) and (135) in [94, 1004, 108, 942], and vacuum
and radiation dominated solutions for arbitrary spatial curvature have been found in
[95, 899, 111]. Some of the methods used in these papers are extended to anisotropic
cosmologies in [898], and the asymptotics of FLRW cosmologies in scalar-tensor theories
have been studied in [112, 1125]. Exact homogeneous and anisotropic solutions are
found in [327, 156] that act as past and future attractors for the general solution. Exact
homogeneous self-similar solutions are found in [132], and inhomogeneous self-similar
solutions are found in [157]. We will not reproduce these solutions here, some of which
can be quite complicated, but will instead return to the attractor mechanism expounded
in [356].
This mechanism is most easily seen in the Einstein conformal frame, given by the
Lagrangian (105), such that for a spatially flat FLRW geometry the evolution equation
for the scalar field can be written as
8π √
ψ 00 + 4π(1 − w)ψ 0 + 4π(1 − 3w)α = 0, (136)
(3 − 4πψ 02 )
where here primes denote differentiation with respect to the natural log of the Einstein
framepscale factor, ā, and w is the equation of state P = wρ. The reader will recall that
ψ = (3 + 2ω)/16π ln φ is the scalar field in the Einstein frame, and α−2 = 3+2ω denotes
the strength of coupling between the scalar and tensor degrees of freedom. Equation
(136) is clearly the equation for a simple harmonic oscillator with a dynamical mass,
a damping force given by −4π(1 − w), and a driving√force R given by the gradient of a
potential (1 − 3w)Γ, where the reader will recall Γ = 4π αdψ. This interpretation of
Γ as an effective potential is often used to justify an expansion of the form
β0
Γ = α0 (ψ − ψ0 ) + (ψ − ψ0 )2 + O((ψ − ψ0 )3 ), (137)
2
where ψ0 is an assumed local minimum of Γ(ψ), and α0 and β0 are constants. In terms
of this parameterisation the PPN parameters βP P N and γP P N then become
α02 β0
1 − βP P N = − (138)
2(1 + α02 )2
2α02
1 − γP P N = . (139)
1 + α02
The requirement of positive mass in (136) can also be seen to be equivalent to the
requirement of positive energy density, ρ̄, in the Einstein frame.
The cosmological dynamics that result from Eq. (136) are that ψ, and hence φ,
approach a constant value during the radiation dominated epoch. This is due to the
vanishing of the ‘potential’ in (136) when w = 1/3, and the negativity of the effective
‘damping force’. Once radiation domination ends, however, and matter domination be-
gins, then the scalar field rolls down to the minimum, ψ0 , of the now non-zero potential
Γ(ψ) (assuming such a minimum exists). Once this minimum is reached, after some
possible oscillations in the case of an under-damped system, then we are left with α = 0,
60
which is the general relativistic limit of these theories. This is a very useful general prop-
erty of any scalar-tensor theory which has a local minimum in its parameter Γ(ψ), and
means that interesting new behaviour is possible at early times, while still being (poten-
tially) compatible with observations that appear to point towards General Relativity at
late-time.
Using order-of-magnitude approximations, the authors of [356] claim that this attrac-
tor mechanism is powerful enough to drive the value of the PPN parameter 1 − γ down to
values of as low as ∼ 10−7 . This is a couple of order of magnitudes below the level that
is probed by even the observations of the Cassini spacecraft, but is not inconceivably
small. In particular, it may be that upcoming observations of binary pulsar systems
could achieve such levels. Further predictions of this scenario are a possible oscillation
in the effective value of Newton’s constant near the beginning of the matter dominated
epoch of the Universe’s history, as well as a prediction for the locally measured value of
βP P N given by
β0
βP P N − 1 = (1 − γ 2 ), (140)
32π
where β0 is defined in Eq. (137). The validity and limitations of these results are
extended, and are further studied in [357, 1095].
Let us now consider perturbations around a general FLRW background, in these
generalised theories. We will work in the conformal Newtonian gauge, which has the
usual correspondence with Bardeen’s gauge invariant variables. Tensor perturbations
on cosmological backgrounds have been studied in [109], while the scalar part of the
perturbed line-element takes the form
 
ds2 = a2 −(1 + 2Ψ)dτ 2 + (1 − 2Φ)qij dxi dxj , (141)

where we have used conformal time, τ , and qij is now the metric of a static 3-space with
constant curvature. Perturbations to the scalar field and energy momentum tensor are
given by δφ and

δT 00 = −δρ (142)
δT 0i ~ iθ
= −(ρ + P )∇ (143)
δT ij = δP δ ij + (ρ + P )Dij Σ, (144)

where ρ, P and θ are the total energy density, pressure and peculiar velocity of the matter
fields. The first-order perturbation equations are then given by [934]
"   # "  #
2 2
2 a0 a0 0 2 3δφ a0
3 Ψ + 3 Φ + (k − 3κ)Φ + 2 +κ (145)
a2 a a a φ a
  0    0 
8π 1 a 0 0 a 0 2
= − δρ − 2 6 Ψ + 3Φ φ − 3 δφ − k δφ
φ a φ a a
 2 "  2  0  #
δφ φ0 dω ω δφ φ0 φ 0 φ02
− 2 + − δφ + Ψ ,
2a φ dφ a2 φ 2 φ φ φ

61
 " "    0 2 # #
0
2 a0 0 a0 a k2 00 a0 0 k 2
Ψ + 2 + − Ψ + Φ + 2 Φ + Φ − κΦ (146)
a2 a a a 3 a 3
"     #
0 2
8πδP δφ a0 a0
= + 2 2 + +κ
φ a φ a a
   
1 a0 a0 2k 2 δφ
− 2 2φ00 Ψ + φ0 Ψ0 + 2 Ψ + 2Φ − δφ00 − δφ0 −
a φ a a 3
 0 2 "     #
2
δφ φ dω ω δφ φ0 φ0 φ02
+ 2 − 2 − δφ0 + Ψ ,
2a φ dφ a φ 2 φ φ φ
and
   0  
2 a0 0 8π 1 a 0 0 ωφ0 δφ
2
Ψ + Φ = (ρ + P )θ − 2
δφ + φ Ψ − δφ + 2 2 . (147)
a a φ a φ a a φ
We also have the perturbed scalar field equation
 
00 a0 0 2 00 0 0 a0 0 8πa2 (δρ − 3δP )
δφ + 2 δφ + k δφ − 2φ Ψ − φ Ψ + 4 Ψ + 3Φ − (148)
a a (2ω + 3)
"   #
a2 d2 ω φ02 δφ 2 dω 0 0 02

2 dω 00 a0 0
= − + 2 φ δφ − φ Ψ + 2 φ + 2 φ δφ ,
(2ω + 3) dφ2 a2 a dφ a dφ a
as well as the condition
8πa2 δφ
Φ−Ψ= (ρ + P )Σ + . (149)
φ φ
This last equation shows that, unlike in General Relativity, Φ 6= Ψ when anisotropic
stresses vanishes (unless the perturbations to the scalar field also vanish). Primes here
denote differentiation with respect to the conformal time, τ , and k is the wave-number
of the perturbation.
Using the equations given above with κ = 0, an analysis of the first year WMAP data
has been performed and used to constrain the parameters α0 and β0 of the attractor
model in [935]. The authors of this study find that the following constraint can be
imposed on these parameters at the 2σ level of significance:
α0 < 5 × 10−4−7β0 . (150)
One should bear in mind here that as β0 → 0 Brans-Dicke theory is recovered, and
as α0 → 0 General Relativity is recovered. As a corollary of this result these authors
also constrain the value of Newton’s constant at recombination to be no more than 5%
different from the value measured in the solar system today, at the 2σ confidence level.
The effect of allowing a non-zero spatial curvature should be expected to weaken these
bounds.
Big bang nucleosynthesis has also been explored in the context of general scalar-tensor
theories [1210, 358, 319, 773]. In [358] it is found that the inferred upper bound on the
baryon density in the Universe is relatively insensitive to the presence of a gravitational
scalar field, and that the parameters of the attractor model must satisfy the constraint
 −1.5
2 −6.5 −1 Ωm h2
α0 . 10 β0 , (151)
0.15
62
when β0 & 0.5. For β0 . 0.5 these bounds are weakened by a few orders of magni-
tude. These results are extended and refined in [319], who also allow for a non-zero
self-interaction potential for the scalar field. The apparent tension between observed and
theoretically predicted abundances of Lithium-7 is addressed in the context of scalar-
tensor theories in [773]. Here the authors point out that a period of expansion slower
than in General Relativity before primordial nucleosynthesis, together with a period of
more rapid expansion during nucleosynthesis, can resolve this conflict. They find such be-
haviour in numerous scalar-tensor gravity theories, both with and without self-interaction
potentials.
Inflation in scalar-tensor theories of gravity has been extensively studied, often under
the name ‘extended inflation’, as coined by La and Steinhardt for the case of Brans-Dicke
theory [762]. The motivation behind this is the possibility of producing a successful
inflationary phase transition from a false vacuum state, thus avoiding the fine tuning
problems associated with ‘new inflation’. Unfortunately, it was soon found that bubble
collisions at the end of inflation produce unacceptable fluctuations in the CMB [1264, 761,
804]. Suggestions to improve this situation were to include a self-interaction potential
for the scalar field [763], generalise the couplings of the Brans-Dicke scalar to other
fields [617] (see also [614, 1260]), include quantum effects [616], add additional couplings
between the inflaton and the space-time curvature [775], or to consider more general
scalar-tensor theories of gravity [1185, 526]. The latter of these approaches was dubbed
‘hyper-extended inflation’. The inflationary solutions of general scalar-tensor theories
have been studied in detail in [106, 212, 86], and specific models that could be compatible
with observations were proposed in [554]. Density perturbations in inflationary scalar-
tensor scenarios have been investigated extensively in [861, 578, 1127, 937, 908, 798, 410,
412, 345, 411, 760, 1181, 1193, 863, 1262]. Studies of topological defects [333, 1089], black
holes [625], gravitational waves [1229, 145], baryogenesis [99, 100], baryon asymmetry
[1295], dark matter [874, 875], the formation of voids [802, 803], bubble nucleation rates
and dynamics [1283, 1088], reheating [328], stochastic inflation [523, 525, 524, 1191, 1192],
slow roll inflation [96, 527, 1212], non-Gaussianity [1047], isotropisation of the Universe
[583, 584], and quantum cosmology [533] have all also been performed in the context
of inflation in scalar-tensor theories. The initial conditions for inflation in scalar-tensor
theories have been considered in [407, 441, 406]. For further details the reader is referred
to 1993 review of extended inflation by Steinhardt [1184].
Theories of gravity with non-minimally coupled scalar fields and non-zero self-interaction
potentials have been studied by a number of authors under the name ‘extended quintessence’
[1035, 74, 35, 1061, 1109]. Such theories can act as dark energy as well as model possible
deviations from General Relativity at early times. These papers include studies of small
angle CMB temperature and polarisation power spectra, the integrated Sachs-Wolfe ef-
fect, the matter power spectrum, supernovae observations and the affects that should be
expected on weak lensing observations. The FLRW solutions of theories with power-law
self-interaction potentials have been studied in further detail in [252], where the attractor
mechanism to general relativity is investigated, as well the presence of periods of accel-
erating expansion at late and early times. Late-time acceleration in models without a
potential for the scalar field is studied in [483].
Another interesting possibility in scalar-tensor theories of gravity is the idea of ‘grav-
itational memory’, proposed by Barrow in [93]. The idea here is that when a black
hole forms one of two things can happen (or some combination of them). Firstly, the
63
Schwarzschild radius of a black hole, which is given by rS = 2G(t)m, could vary as the
value of the scalar field controlling the value of G varies in the background universe. In
this case there is no such thing as a static black hole solution to the gravitational field
equations, unless the black hole exists in a static universe. Secondly, the Schwarzschild
radius of a black hole could be frozen in at its value when the black hole formed, so
that rS = 2G(tf )m, where tf is the time when the black hole formed. In this case
black holes that formed early on in the Universe’s history would remember, in some
sense, the conditions of the early universe, this being reflected in the value of G(tf ). As
Barrow points out, these two possibilities have consequences for the evaporation, and
explosion, of black holes in the late universe. This idea has motivated a number of
studies on the gravitational field of collapsed objects in scalar-tensor theories of gravity
[97, 1211, 1213, 654, 593, 860, 948, 947]. One particularly interesting approach is that
of matched asymptotic expansions, which suggests that the first option is followed, and
black holes do not have any gravitational memory [1135, 1136, 1134].

Horndeski’s theory

The most general four dimensional scalar-tensor theory with second-order field equa-
tions was worked out by Horndeski in [623]. It has the following Lagrangian

αβγ 4
LH = δµνσ κ1 ∇µ ∇α φRβγνσ − κ1,X ∇µ ∇α φ∇ν ∇β φ∇σ ∇γ φ
3

+κ3 ∇α φ∇µ φRβγνσ − 4κ3,X ∇α φ∇µ φ∇ν ∇β φ∇σ ∇γ φ
h i
αβ
+δµν (F + 2W )Rαβµν − 4F,X ∇µ ∇α φ∇ν ∇β φ + 2κ8 ∇α φ∇µ φ∇ν ∇β φ
−3[2(F + 2W ),φ + Xκ8 ]∇µ ∇µ φ + κ9 (φ, X), (152)
[ν ν ]
where X = ∇µ φ∇µ φ, and δµν11νµ22...ν 1 ν2 n
...µn = n!δµ1 δµ2 ...δµn . The theory depends on four arbi-
n

trary functions of φ and X, κi = κi (φ, X) as well as F = F (φ, X), which is constrained

so that F,X = κ1,φ − κ3 − 2Xκ3,X . Note that W = W (φ), which means that it can
be absorbed into a redefinition of F (φ, X). This paper is not very well known, and as
a result Horndeski’s theory has not been well explored. It has, however, been recently
resurrected in [276], where aspects of the theory on FLRW backgrounds were studied.
The effective Lagrangian describing the cosmology in the minisuperspace approximation
is given by
3 
X κ
Leff
H (a, ȧ, φ, φ̇) = a
3
An − Bn 2 H n , (153)
n=0
a
where H = ȧ/a is the Hubble parameter, and where we have
A0 = −Q̃7,φ φ̇ + κ9 (154)
2
B0 = Q̃1,φ φ̇ + 12κ3 φ̇ − 12F (155)
3
A1 = −12F,φ φ̇ + 3(Q7 φ̇ − Q̃7 ) + 6κ8 φ̇ (156)
B1 = −Q1 φ̇ + Q̃1 (157)
2
A2 = −12F − 12F,A φ̇ (158)
3
A3 = 8κ1,A φ̇ , (159)
64
where
∂ Q̃1 ∂ Q̃7
Q1 = = −12κ1 , and Q7 = = 6F,φ − 3φ̇2 κ8 . (160)
∂ φ̇ ∂ φ̇
It is assumed that matter is minimally coupled to the metric gµν , and not to the scalar.
Indeed, it is argued that if the equivalence principle is to hold then this can be as-
sumed without further loss of generality. The cosmological field equations are presented
implicitly as a generalised Friedmann equation:
 
1 ∂Leff
H ∂Leff
H eff
ȧ + φ̇ − LH = −ρ, (161)
a3 ∂ ȧ ∂ φ̇
and the scalar equation of motion
 
∂Leff
H d ∂Leff
H
− = 0, (162)
∂φ dt ∂ φ̇
where ρ is the energy density of the cosmological fluid.
In [276] the authors look for those corners of Horndeski’s theory that admit a self-
tuning mechanism. They demand that the vacuum space-time is Minkowski, irrespective
of the value of the cosmological constant, and that this should remain true even after
a phase transition in which the cosmological constant changes by some amount. This
is not in violation of Weinberg’s theorem since Poincaré invariance is explicitly broken
by the scalar. These considerations reduce Horndeski’s theory to four base Lagrangians
known as the Fab Four:
√
Ljohn = −gVjohn (φ)Gµν ∇µ φ∇ν φ (163)
√
Lpaul = −gVpaul (φ)P µναβ ∇µ φ∇α φ∇ν ∇β φ (164)
√
Lgeorge = −gVgeorge (φ)R (165)
√
Lringo = −gVringo (φ)Ĝ, (166)

where Ĝ = Rµναβ Rµναβ − 4Rµν Rµν + R2 is the Gauss-Bonnet combination, and

P µναβ = −Rµν αβ + 2Rµ[α δ νβ] − 2Rν [α δ µβ] − Rδ µ[α δ νβ]

is the double dual of the Riemann tensor. These terms give rise to self-tuning cosmologies
for κ < 0. The relevant cosmological field equations are given by
Hjohn + Hpaul + Hgeorge + Hringo = − [ρΛ + ρmatter ] , (167)
where we have separated the net cosmological constant contribution, ρΛ , and the matter
contribution, ρmatter , and where
 κ
Hjohn = 3Vjohn (φ)φ̇2 3H 2 + 2
 a
κ
Hpaul = −3Vpaul (φ)φ̇3 H 5H 2 + 3 2
a
  0 
2 κ Vgeorge
Hgeorge = −6Vgeorge (φ) H + 2 + H φ̇
a Vgeorge
 κ
0 2
Hringo = −24Vringo (φ)φ̇H H + 2 .
a
65
The scalar equations of motion are Ejohn + Epaul + Egeorge + Eringo = 0 where
d h 3 i
Ejohn = 6 a Vjohn (φ)φ̇∆2 − 3a3 Vjohn 0
(φ)φ̇2 ∆2
dt
d h 3 i
Epaul = −9 a Vpaul (φ)φ̇2 H∆2 + 3a3 Vpaul 0
(φ)φ̇3 H∆2
dt
d  3 0 
Egeorge = −6 a Vgeorge (φ)∆1 + 6a3 Vgeorge
00
(φ)φ̇∆1
dt
+6a3 Vgeorge
0
(φ)∆21
  
0 d 3 κ 2
Eringo = −24Vringo (φ) a ∆ 1 + ∆3 ,
dt a2 3
 √ n
−κ
and we define ∆n = H n − a . We see that the self-tuning is achieved at the level
of the scalar equation of motion, since on a Minkowski solution one has H 2 = − aκ2 =⇒
∆n = 0 for n ≥ 1. In vacuum, the cosmological constant controls the value of the scalar
via the generalised Friedmann equation. A detailed study of the phenomenology of the
fab four has yet to be carried out, but the authors of [276] argue that the ‘john’ and
‘paul’ terms are expected to play a crucial role, as their derivative interactions could give
rise to Vainshtein effects that could help pass solar system constraints. The Vainshtein
mechanism is discussed in detail in Section 5.5.4.
Note that it has been shown that Horndeski’s general theory is equivalent to [402] in
four dimensions [710]. Aspects of cosmological perturbations are studied in [710] that
may be applied to the Fab Four in the appropriate special case.

3.1.4. The chameleon mechanism

The ‘chameleon mechanism’ was introduced as a concept in gravitational physics by
Khoury and Weltman in [689, 688]. The basic concept here is that if we consider theories
with a non-minimally coupled scalar field, then in the presence of other matter fields
these scalars can acquire an effective mass parameter that is environmentally dependent.
One can then potentially satisfy the tight constraints on non-minimally coupled scalar
degrees of freedom that are imposed in relatively dense environments, such as exist in
the solar system, while still having interesting new behaviour in less dense environments,
such as those that can exist in cosmology.
This mechanism is usually formulated in the Einstein conformal frame, where the
coupling between the scalar curvature and scalar field is minimal, but where the scalar
field couples non-minimally to matter fields. The relevant action is then
 
√ 1 1 √
L = −ḡ R̄ − ḡ µν ψ,µ ψ,ν − V (ψ) + Lm (Ψi , e2 8πβi ψ ḡµν ), (168)
16π 2
√
where, in the notation used in Eq. (105), we have taken Γ = 8πβi ψ, and where the βi
are a set of constants denoting the coupling of ψ to each of the i matter fields Ψi . In
the scalar-tensor theories so far discussed the scalar field should be considered coupled
to each of the matter fields with the same universal coupling, which in the Brans-Dicke
theory is given by β −2 = 2(3 + 2ω). Assuming such a coupling, the non-relativistic limit
of the scalar field equation can then be written as
dVeff
∇2 ψ = , (169)
dψ
66
 √
where Veff (ψ) ≡ V (ψ) + ρe 8πβψ . This new ‘effective potential’ can be seen to be depen-
dent on the ambient energy density, and if ψ and β are both positive then any runaway
potential with dV /dψ < 0 will result in an effective potential with a local minimum whose
position depends on ρ. What is more, for couplings of the type specified in Eq. (168)
the local effective mass of the scalar field ψ, given by mψ = d2 Veff /dψ 2 , can be seen to
generically increase with increasing ρ. Hence, the name ‘chameleon’.
The behaviour of scalar fields outside of massive objects, when the chameleon mech-
anism is present, can be shown to be crucially dependent on the ratio of ∆ψ to Φc .
Here ∆ψ denotes the difference in the value of the scalar field inside the object, ψc , and
asymptotically, ψ∞ , while Φc is the value of the Newtonian potential at the surface of
the object, where r = Rc . More precisely, when one satisfies the condition
√
8π(ψ∞ − φc )
1 (170)
6βΦc
then the resulting configuration of gravitational fields is found to be one in which ψ
occupies the minimal of the effective potential inside the bulk of the massive object,
except for a thin region of depth ∆Rc just below its surface where the value of ψ rises.
Outside of the object ψ increases further, and approaches its asymptotic value ψ∞ as
 
2β 3∆Rc Mc e−m∞ (r−Rc )
ψ ' ψ∞ − √ , (171)
8π Rc r

where m∞ is the effective mass of the field at asymptotically large distances from the
object of mass Mc . Now, the ratio of the thickness of the shell just below the object’s
surface to the object’s overall radius, ∆Rc /Rc , can be shown to be well approximated by
the LHS of Eq. (170). The condition given in (170) is then equivalent to the condition
that a ‘thin shell’ should be present, with ∆Rc /Rc  1.
If the ‘thin shell’ condition is not met then one instead has ψ ∼ ψ∞ everywhere, and
the exterior solution is given by

2β Mc e−m∞ (r−Rc )
ψ ' ψ∞ − √ . (172)
8π r

A comparison of Eq. (172) with Eq. (171) immediately shows that without a thin shell
variations in ψ are no longer suppressed by the small factor of 3∆Rc /Rc , and that we
should therefore expect in this case more obvious consequences to the existence of ψ
within the vicinity of massive objects. Khoury and Weltman proceed to argue that in
order to avoid violations of the weak equivalence principle, and unacceptable deviations
from the predictions of General Relativity in the solar system, we should require that
the Earth, and other astrophysical bodies, should satisfy the thin shell condition [688].
This idea of a scalar field with an environmentally dependent mass has sparked
widespread interest since it was proposed. In particular, it allows for the possibility of
measuring fifth forces, or violations of the weak equivalence principle, that are different in
space than they are on Earth, [689, 688, 499, 912, 1236, 190, 913, 1199, 1091, 1225, 198].
It can act as dark energy [191, 197], and has been studied in the context of structure
formation [196, 188], as well as a number of other cosmological scenarios [912, 190, 913,
191, 484, 239, 362]. The effect of ‘chameleon particles’ on searches for axion-like particles
67
and experiments involving magnetic fields have been studied in [192, 14, 537], and their
effect on the propagation of light in astrophysics in [219, 220]. Experimental searches for
chameleons have now been performed by GammeV [298, 1182, 1238, 1183], and ADMX
[1079], which have started to constrain the viable parameter space of these theories.
Other tests of this scenario are also proposed in [187, 789, 788, 1108, 199, 186].

3.2. Einstein-Æther Theories

Vector-tensor theories, in the form of Einstein-æther theories, have had a revival over
the past decade, and are now often used as a counterfoil to test General Relativity. They
have the particular property that they single out a preferred reference frame and have
become somewhat of a theoretical workhorse for studying violations of Lorentz symmetry
in gravitation. In the Einstein-æther theory [655] violations of Lorentz invariance arise
within the framework of a diffeomorphism-invariant theory, and their modern incarna-
tions are a refinement of the gravitationally coupled vector field theories first proposed
by Will and Nordtvedt in the 1970s [1276, 989]. The presence of a Lorentz-violating
vector field, henceforth called the æther, can dramatically affect cosmology: It can lead
to a renormalisation of the Newton constant [257], leave an imprint on perturbations in
the early universe [809, 677], and in more elaborate actions it can even affect the growth
rate of structure in the Universe [1311, 1314, 589].

3.2.1. Modified Newtonian dynamics

Some of the theories that we will discuss in this subsection and the next have been
constructed to give modifications to Newtonian gravity on galactic scales. To be more
specific, they should lead to Milgrom’s Modified Newtonian Dynamics [895], also known
as MOND, in regimes of low acceleration. Given its relevance for Einstein-æther theories,
we will now briefly describe the motivation for MOND, and how it works. We also briefly
mention some of its successes and failures.
MOND was first proposed as a possible explanation of the need for dark matter in
galaxies, based on observations of their rotational velocities. With Newtonian gravity
and the visible baryonic matter in galaxies only one expects that the rotational velocity,
vr , should depend on the distance from the centre of the galaxy, r, as vr ∝ r−1/2 . What
is in fact found in observations of spiral galaxies is that vr is approximately constant
at large radii. The conventional answer to this problem is to posit that galaxies sit
in halos of dark matter, with energy density profiles that vary as ρ ∼ r−2 for large r.
Milgrom’s proposal was that, alternatively, Newton’s inverse square law of gravity could
be modified in the low-acceleration regime of galactic dynamics. Such a modification, it
was ventured, may be able to account for the anomalously high rotational velocities in
spiral galaxies without invoking any new matter fields.
In MOND the spherically symmetric gravitational potential has two regimes: High
acceleration and low acceleration. In regions of high acceleration (where |~a|  a0 , for
constant a0 ), it simply satisfies Newton’s second law: ~a = −∇Φ where Φ is the gravi-
tational potential. On the other hand, in the low acceleration regime (where |~a|  a0 ),
Newton’s second law is modified to (|~a/a0 )~a = −∇Φ. Albeit a simple rule of thumb,
Milgrom’s proposal is remarkably successful at fitting a large range of spiral galaxy ob-
servations. Furthermore, it can be used to explain the Tully-Fisher relation that relates
the velocity of rotation of a spiral galaxy with its intrinsic luminosity. Unfortunately

68
MOND is unable to explain the dynamics of clusters of galaxies without recourse to
additional dark matter (possibly in the form of neutrinos), and the behaviour of dwarf
spheroidals in different environments is also problematic. Nevertheless, it is an interesting
proposal that has had a renewed surge of interest in the past decade.
The non-relativistic Poisson equation in MOND can be written as
" ! #
| ~
∇Φ|
∇~ · µ ~
∇Φ = 4πGρ, (173)
a0

where ρ is the energy density in baryons, and the function µ(x) → 1 as x → ∞ and
µ(x) → x as x → 0. There are a variety of proposals for the precise form of µ(x) that
fit observations of galaxies to a greater or lesser degree. As a theory of modified gravity,
however, MOND’s greatest limitation is that it is restricted to non-relativistic regimes.
It therefore cannot be used to make prediction on cosmological scales, nor can it be used
to calculate fundamentally relativistic observables, such as lensing. Many of the theories
that follow in this section have been constructed to address this deficiency: They are
relativistic gravitational theories that have MOND as a non-relativistic limit.

3.2.2. Action and field equations

As the name suggests, vector-tensor theories involve the introduction of a space-time
4-vector field, Aµ . A general action for such theories is given by
Z  
4 √ 1
S = d x −g R + L(g , A ) + SM (g µν , Ψ),
µν ν
(174)
16πG
where Sm is the matter action. Note that the matter fields Ψ in SM couple only to the
metric gµν , and not to Aν .
Let us now focus on Einstein-æther theories, and hence forth consider Aµ to have a
time-like direction. The simplest (and most thoroughly studied) version of the Einstein-
æther theory is quadratic in derivatives of Aν , and has the form
1
LEA (g µν , Aν ) ≡ [K µναβ ∇µ Aα ∇ν Aβ + λ(Aν Aν + 1)], (175)
16πG
where K µναβ ≡ c1 g µν gαβ + c2 δ µα δ νβ + c3 δ µβ δ να − c4 Aµ Aν gαβ and λ is a Lagrange
multiplier. In what follows we will use the notation c12... ≡ c1 + c2 + . . . . We call the
theory derived from Eqs. (174) and (175) the linear Einstein-æther theory.
A more general, non-linear Lagrangian for the æther field can be written in the form

M2 1
LGEA (g µν , Aµ ) = F (K) + λ(Aµ Aµ + 1), (176)
16πG 16πG
where K = K µναβ ∇µ Aα ∇ν Aβ , and M has the dimension of mass. We shall call this a
generalised Einstein-æther theory.
Such actions arise from Lorentz violating physics in quantum gravity. Indeed, the
linear Einstein-æther theory can be constructed using the rules of effective field theory,
and has been shown to be stable with regard to quantum effects [1279]. Such theories,
however, can suffer from instabilities at the classical level, with the onset of caustics in a
finite time [331]. This raises the question of whether the vector field in such theories are
69
merely an effective (possibly composite) degrees of freedom, or whether they are genuine
fundamental fields.
The gravitational field equations for this theory, obtained by varying the action for
the Generalised Einstein-æther theory with respect to g µν are given by
matter
Gµν = T̃µν + 8πGTµν , (177)
1
T̃µν = ∇α (FK (J(µα Aν) − J α(µ Aν) − J(µν) Aα ))
2
1
−FK Y(µν) + gµν M 2 F + λAµ Aν , (178)
2
where FK ≡ dF
dK and J µα ≡ (K µναβ +K νµβα )∇ν Aβ . Brackets around indices denote sym-
∂(K αβ )
metrisation, and Yµν is defined by the functional derivative Yµν = ∇α Aρ ∇β Aσ ∂gµνρσ .
The equations of motion for the vector field, obtained by varying with respect to Aν , are
∇µ (FK J µν ) + FK yν = 2λAν , (179)
∂(K αβρσ )
where we have defined yν = ∇α Aρ ∇β Aσ ∂Aν . Finally, variations of the action with
respect to λ fix Aν Aν = −1.

3.2.3. FLRW solutions

In a homogeneous and isotropic universe with perfect fluid matter content, the vector
field will only have a non-vanishing ‘t’ component, so that Aµ = (1, 0, 0, 0). The equations
2
of motion then simplify dramatically, so that ∇µ Aµ = 3H and K = 3 αH M 2 , where α ≡
c1 + 3c2 + c3 . Note that the α we have defined here has the same sign as K. The field
equations then reduce to
  
d F 8πG
1 − αK 1/2 1/2
H2 = ρ, (180)
dK K 3
d
(−2H + FK αH) = 8πG(ρ + P ). (181)
dt
If we now take F (x) = γxn , the modified Friedmann equations become
"  2(n−1) #
H 8πG
1+ H2 = ρ, (182)
M 3

where  ≡ (1 − 2n)γ(−3α)n /6. We also get the relationship

 2(n−1)
6(Ωm − 1) M
γ= , (183)
(1 − 2n)(−3α)n H0
where Ωm ≡ 8πGρ0 /3H02 , and H0 is the Hubble constant today. Let us now consider a
few special cases: If n = 1/2, the Friedmann equations are unchanged ( = 0) and there is
no effect on the background cosmology; with n = 1 we have that  = γα/2 and Newton’s
constant is rescaled by a factor of 1/(1 + ) [257]; if n = 0 we recover a cosmological
constant, Λ ' sign(−γ)M 2 . More generally, we will obtain different regimes depending
on the relative size of each term in the modified Friedmann equation. We can summarise
these behaviours in Figure 3.2.3.
70
Figure 1: A schematic representation of the late-time evolution of FLRW solutions as a function of n
and γ, for n < 1.

3.2.4. Cosmological perturbations

The four-vector Aµ can be perturbed as Aµ = (1 − Ψ, a1 ∇ ~ i V ), where V is a small
2
quantity. Perturbing K to linear order then gives K = K0 + K1 , where K0 = 3 αH M 2 and
K1 = −2 αH 2V
M 2 (k a + 3HΨ + 3Φ̇). The gravitational potentials Ψ and Φ come from the
perturbed metric:
 
ds2 = a2 (τ ) −(1 + 2Ψ)dτ 2 + (1 − 2Φ)qij dxi dxj , (184)

were qij is the unperturbed conformal metric of the hyper-surfaces of constant τ .

The evolution equation for the perturbations in the vector field are

0 = c1 [V 00 + k 2 V + 2HV 0 + 2H2 V + Ψ0 + Φ0 + 2HΨ] (185)

a00
+c2 [k 2 V + 6H2 V − 3 V + 3Φ0 + 3HΨ]
a
2 2 a00
+c3 [k V + 2H V − V + Φ0 + HΨ]
a
FKK
+ [−K1 αH − K00 (−c1 (V 0 + Ψ) + 3c2 HV + c3 HV )].
FK
The perturbation in the vector field is sourced by the two gravitational potentials Φ and

71
Ψ. The first-order perturbations to the vector field’s stress-energy tensor are

a2 δ T̃ 00 = FK c1 [−Hk 2 V − k 2 V 0 − k 2 Ψ] (186)
+FK α[Hk 2 V + 3HΦ0 + 3H2 Ψ] − 3FKK αH2 K1
= FK c1 [−Hk 2 V − k 2 V 0 − k 2 Ψ] + FK α(2n − 1)[Hk 2 V + 3HΦ0 + 3H2 Ψ],
 
a00
a2 δ T̃ 0i = iki FK c1 V 00 + 2HV 0 + V + Ψ0 + HΨ (187)
a
 
a00
+iki FK α 2H2 V − V + iki FKK K00 [c1 (HV + V 0 + Ψ) − αHV ],
a
a2 δ T̃ ij = FK c2 k 2 [2HV + V 0 ]δ ij + FK (c1 + c3 )[2HV + V 0 ]k i kj (188)
 00

a
+FK α 2HΦ0 + Φ00 + 2 Ψ − H2 Ψ + HΨ0 δ ij + FKK (c1 + c3 )K00 V k i kj
a
00
a
−FKK [αK1 + (c1 + c2 + c3 )K1 H2 + αHK10
a
−αK00 Φ0 − 2αK00 HΨ + α ln(FKK )0 K1 H − c2 K00 k 2 V ]δ ij ,

where the second expression for a2 δ T̃ 00 assumes the monomial form for F (K). In the
absence of anisotropic stresses in the matter fields, we may obtain an algebraic relation
between the metric potentials Φ and Ψ by computing the transverse, traceless part of
the perturbed Einstein equations. This gives
3 2 1
k 2 (Ψ − Φ) = a (k̂i k̂j − δij )(δ T̃ji ) (189)
2 3
= (c1 + c3 )k [FK (2HV + V 0 ) + FKK K00 V ].
2

We then find the following expression for the perturbed field equations:
1
k2 Φ = − FK c1 k 2 [V 0 + Ψ + (3 + 2c̃3 )HV ] (190)
2
X

−4πGa2 ρ̄a δa + 3(ρ̄a + P̄a )Hθa .
a

Before we look at the cosmological consequences of these theories, and constraints that
can be imposed on them, it is instructive to study the effect of the vector field during
matter domination. This should allow us some insight into how the growth of structure
proceeds in the generalised Einstein-æther case. First let us consider the simplest case
in which the dominant contribution to the energy density is baryonic, so that we can
treat it as a pressureless perfect fluid. Let us also introduce the new variable V 0 ≡ E.
For illustrative ease we will initially consider
 ponly the case where V is described by
a growing monomial, such that V = V0 ττ0 , During the matter dominated era we
then have a2 δT 00 ' −lE ξ(k)k 2 τ 5+p−6n and k 2 (Ψ − Φ) ' −lS ξ(k)k 2 τ 5+p−6n , where
lE ≡ −(c1 (2 + p)n + 2α(1 − 2n)n), lS ≡ −(c1 + c3 )n(6n − p − 10), and
 p  2 !n−1
1 6−6n H0
ξ(k) ∼ γV0 (k) khub 3αΩm , (191)
τ0 M
72
where khub ≡ 1/τtoday . Hence, the vector field affects the evolution equations for the
matter and metric perturbations only through its contribution to the energy density
and anisotropic stress. On large scales, kτ  1, and assuming adiabatic initial con-
6lS ξ(k)
ditions for the fields δ, Φ, θ, this leads to δ = C1 (k) + (10+p−6n) τ 5+p−6n , where C1 is
a constant of integration and we have omitted the decaying mode. Therefore, even
before horizon crossing, the anisotropic stress term due to the vector field can influ-
ence time evolution of the baryon density contrast. On small scales, kτ  1, we find
( 21 lE +lS )
δ(k, τ ) = C2 (k)τ 2 + (5+p−6n)(10+p−6n) ξ(k)(kτ )2 τ 5+p−6n , where C2 (k) is another constant
of integration. Hence, for sub-horizon modes, the influence of the vector field on the evo-
lution of δ is a combination of its affect on the energy density and anisotropic stress
contributions, though both, in this limit, result in the same contributions to the scale
dependence and time evolution of the density contrast.

3.2.5. Observations and constraints

Let us now consider the constraints that can be imposed on these theories. First
of all we will consider the linear Einstein-æther theory, and then we will consider the
generalised Einstein-æther models.
In the case of the linear Einstein-æther theory, a number of non-cosmological con-
straints on the ci have been derived: Most notably, a Parameterised Post-Newtonian
(PPN) analysis of the theory leads to a reduction in the dimensionality of parameter
space. This is occurs due to the requirement that c2 and c4 must be expressed in terms
of the other two parameters in the theory as c2 = (−2c21 −c1 c3 +c23 )/3c1 and c4 = −c23 /c1 .
Additionally, the squared speeds of the gravitational and æther waves with respect to
the preferred frame must be greater than unity, so as to prevent the generation of vac-
uum Čerenkov radiation by cosmic rays. A final constraint arises from considering the
effects of the æther on the damping rate of binary pulsars. The rate of energy loss in
such systems by gravitational radiation agrees with the prediction of General Relativity
to one part in 103 . In the case of the Einstein-æther theory it has been shown that to
agree with General Relativity in these systems we must require that c+ ≡ c1 + c3 and
c− ≡ c1 − c3 are related by an algebraic constraint. A more exotic, but viable, subset
of the parameter space can be considered if we set c1 = c3 = 0. The PPN and pulsar
constraints are then no longer applicable, and a cosmological analysis is potentially the
only way of constraining the values of the coupling constants.
Using a combination of CMB and large-scale structure data [1315] it is possible to
impose constraints on the coefficients of the theory, ci , as well as the overall energy
density in the æther field. The main effect on the evolution of perturbations is through
the change in the background evolution, and not necessarily through the presence of
perturbations in the vector field. Indeed, artificially switching off the perturbations in
the æther field has essentially no effect on the power spectrum of large-scale structure,
and a small effect (of approximately 10%) on the angular power spectrum of the CMB.
In Figure 2 we plot the join constraints on c+ and c− that can be imposed from these
observables.
If we consider the generalised Einstein-æther theory, we find that the effect on the
CMB is much more pronounced. Let us first consider a Universe with no dark matter,
and in which the perturbations in the æther field simultaneously mimic a perturbed pres-
sureless fluid in the formation of large scale structure, whilst behaving entirely differently
73
 1.0


0.8

0.6
c

0.4

0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
c+

Figure 2: Likelihood plot in the parameter space of −c+ and −c− from observations of the CMB and
large-scale structure. The black lines are the 1 and 2σ contours, for which we have marginalised over the
values of the other parameters. The hatched region is excluded by Čerenkov constraints. The dashed
line indicates the constraints available from binary pulsars.

in the cosmological background [1314]. The first requirement for successful perturbation
evolution is that structure can form at all. One necessary condition for this is that the
sound speed of the structure seed is not too large, since this would wash out structure.
It is therefore required that the sound horizon in the models we are considering should
be less than the smallest scales where linear structure can form: CS kmax τ . 1, where
kmax ∼ 0.2h/M pc. For matter power observations at τ ∼ 3 × 104 , which is the present
epoch, this yields CS . 10−4 .
There are two underlying physical processes that can constrain these models. The
first is a change in the rate of growth of the amplitude of perturbations. This can cause
discrepancies between the amplitudes we expect in the matter power spectrum and the
CMB, since the evolution between the two is different. It can also lead to an integrated
Sachs-Wolfe effect during the matter era, as Φ becomes time dependent. The second
process is due to the increased magnitude Φ − Ψ. This also leads directly to a non-
negligible integrated Sachs-Wolfe effect in the matter dominated era. The details of each
of these processes depends on the functional form of F , the time-dependence of the ξ
growing mode, and the choice of the parameters ci . It is extremely challenging to find
combinations of the parameters that allows for a realistic growth of structure, while
simultaneously ensuring the integrated Sachs-Wolfe effect is acceptably small.
A consequence of these two effects is that it is impossible to find models where the
æther field replaces the dark matter that fit the available cosmological data. This is not
due to the matter power spectra, which can be reasonably fitted to the SDSS data, but
from the CMB. In the low-` regime a large ISW effect is clearly present, destroying any
74
chance of fitting the CMB data at large scales. The positions of the peaks are also poorly
fit by the model. Finally, to fit the matter power spectrum to the data requires rescaled
by a factor 0.02, which corresponds to a galaxy bias of 0.14. Such a scaling is considered
to be improbably small, on physical grounds. All these effects cause severe problems
when attempting to simultaneously fit the CMB and large-scale structure.
Finally, let us consider the possibility of late-time accelerating expansion. A detailed
comparison with the data seems to allow a range of values for the index n, and the three
coupling terms of the theory, which can produce this behaviour. In the limit nae → 0,
however, the æther field behaves exactly as a cosmological constant term.

3.3. Bimetric Theories

In this section we will consider theories that involve two rank-2 tensors. These are
often referred to as “bimetric”, or “tensor-tensor”, theories of gravity. The first formu-
lation of a bimetric theory appears to be due to Rosen [1067, 1068], and involves the
addition of an extra non-dynamical rank-2 tensor into the theory. Rosen’s theory, how-
ever, is now known to lead to the existence of states that are unbounded from below in
their energy. As a result Rosen’s theory predicts the spin up of pulsars, as gravitational
waves with negative energy are emitted. This severely violates the constraints on these
systems that have been imposed by observations of millisecond pulsars [779].
Following in Rosen’s footsteps, there were a number of proposals over the years of
how one could formulate a viable bimetric theory of gravity. Here we highlight what we
consider to be some of the most interesting cases. These include Drummond’s bimetric
(or “bi-vierbein”) theory, which is claimed to mimic the dark matter in spiral galaxies
[434], as well as arguments by Magueijo that bimetric theories could exhibit a variable
speed of light, thus providing a way to model time-varying fundamental constants. More
recently, Bañados and collaborators have shown that a general form of bigravity, which
includes specific forms previously proposed in[647, 355] might allow one to account for
some aspects of the dark sector [85, 70]. Finally, Milgrom has recently proposed a
bimetric theory that reduces to MOND in the appropriate limits. In what we follows,
we will briefly outline each of these theories.
The basic idea behind bimetric, or tensor-tensor, theories is the introduction of a
second ‘metric’ tensor into the theory18 : a dynamical metric, gµν , and a second metric,
g̃αβ . The first of these is usually universally coupled to the matter fields, and is used to
construct the energy-momentum tensor of the non-gravitational fields. It is this field is
used to define the geodesic equations of test particles. The equations that govern gµν ,
however, are not the Einstein field equations: They invariably involve g̃αβ as well.
If the g̃αβ is not dynamical, then it is usually taken to be highly symmetric (i.e.
exhibiting the maximal 10 Killing vectors X, such that LX g̃αβ = 0). An obvious choice
for g̃αβ is the Minkowski metric, ηµν , so that all components of the Riemann tensor
constructed from g̃αβ vanish. Rosen’s bimetric theory is a particular example of such
a construction, as are some attempts to construct a massive theory of gravity. If g̃αβ
is to be dynamical, then a kinetic term of the Einstein-Hilbert form is required in the

18 A second rank-2 tensor would probably be a more accurate description of what is actually being

added here, as the term ‘metric’ implies a particular geometric function. Nevertheless, the term ‘metric’
for this additional field is commonly used, and so we follow this convention here.

75
gravitational action. Coupling terms are then also required between g̃αβ and gµν , with
the matter fields usually coupling to either one, or a combination of both, metrics.

3.3.1. Rosen’s theory, and non-dynamical metrics

As advertised, Rosen’s bimetric theory is constructed with an extra flat metric, g̃αβ =
ηαβ , such that19

R̃µνρσ (g̃αβ ) = 0. (192)

We can now define a covariant derivative in terms of g̃αβ , which we will call ∇ ˜ µ , such
that the field equations for the dynamical metric can be written
√
1 αβ ˜ ˜ 1 ˜ β gν = −8πG √−g (Tµν − 1 gµν g αβ Tαβ )
˜ α gδµ ∇
g̃ ∇α ∇β gµν − g̃ αβ g δ ∇ (193)
2 2 −g̃ 2

The energy-momentum tensor satisfies the conventional conservation equation ∇µ Tµν =

0.
Rosen’s theory has been the subject of a number studies over the years. It has been
found to be extremely successful when subjected to a PPN analysis, and compared to
Solar System observations [779]. In fact, almost all the PPN parameters in Rosen’s
theory are indistinguishable from those of General Relativity. The only exception is
α2 = vg2 /c2 − 1, where vg is the speed of gravitational waves in Rosen’s theorem, and c is
the speed of light. One should note here, however, that vg is not uniquely determined by
the theory, but rather by the cosmological solutions to the theory. One can then adjust
the initial conditions of the Universe in order to tune α2 . If this is done then the theory is
observationally indistinguishable from General Relativity in the weak field, low velocity
regime of post-Newtonian gravitational physics.
Rosen’s theory fails, however, when its predictions for the emission of gravitational
waves are compared to observations of binary pulsars. Will and Eardley found that unless
the binary system under consideration obeys very specific properties, in terms of masses
and mass differences, then Rosen’s theory leads to the emission of a large amount of dipole
gravitational radiation [1275]. This in turn results in a sizeable increase in the orbital
period of the system, which is not observed. Binary pulsar observations are therefore
incompatible with this theory. Rosen later proposed replacing flat space metric by an
a priori specified, but time-varying, cosmological background [1069]. Unfortunately this
does not circumvent the pulsar problem.
Other bimetric theories that also have been proposed with an additional a priori
specified, non-dynamical metric field. These include Rastall’s theory [1055, 1056] and
Lightman and Lee’s theory [808], for which the PPN limits of both theories are known. It
has been conjectured by Will, however, that all such theories that incorporate prior spec-
ified geometry could suffer the same deficiency as Rosen’s, when it comes to calculating
the emission of gravitational radiation from binary systems [1274].

19 Note that this equation can be derived from an action principle by including a space-time dependent

rank-2 tensor as a Lagrange multiplier. We will not go into the details of how to do this here.

76
3.3.2. Drummond’s theory
Let us now turn to a more recent formulation of the bimetric theory. In [434] it
was proposed to work in the vierbein formulation with gαβ = ηAB eA B
α eβ and g̃αβ =
ηÃB̃ ẽà B̃ Ã
α ẽβ . In this case both sets of vierbein are dynamical. While ẽα is used to construct
the Einstein-Hilbert action, eB β is used to construct the action from which the energy-
momentum tensor is derived. The missing pieces of the theory are then a transformation
tensor, MBÃ , and a scalar, φ, which relate ẽÃ B
µ and eµ by

ẽÃ φ Ã B
µ = e M B eµ . (194)

Finally, we then need to define a “linking action”. This is given by Drummond as

Z p
1
SL = d4 x −g̃g̃ µν Tr(jµ jν )
16πG1
Z p
1
+ d4 x −g̃g̃ µν (∂µ φ∂ν φ)
16πG2
Z p m2  Ã A 
1
+ d4 x −g̃ MA MÃ + MÃA MAÃ − γ
16πG1 4
Z p
1
− d4 x −g̃m2 φ2 , (195)
16πG2
where the current jµ is defined as

˜ µ M Ã ]M B ,
jµÃB̃ ≡ g̃ B̃ C̃ [∇ (196)
B C̃

and where G1 and G2 are new gravitational constants, m is a mass parameter, and γ is
a free parameter which corresponds to the cosmological constant. Note that the action
for M is similar to that of the non-linear sigma model found in meson physics.
Drummond has shown that his bimetric theory has a well defined Newtonian limit
and so, in principle, can satisfy the time delay measurements from radio signals. He also
claims that the higher order correction is exactly what is needed to satisfy Solar System
constraints from the precession of the orbit of Mercury, and that in the weak field limit
the dynamical metric gαβ ' ηαβ + hαβ gives rise to a potential of the form
 
GM G1 −mr
h00 = − 1+ e . (197)
r G

Hence, for mr  1 the effective Newton’s constant is GN = G + G1 , while for large scales
GN ' G. Such a correction can alleviate the problem of flat galactic rotation curves that
arises in standard Newtonian gravity with no dark matter, but does not completely
resolve it. Albeit an intriguing proposal for a theory of modified gravity, there has been
little progress in studying the various astrophysical and cosmological consequences of
Drummond’s theory.

3.3.3. Massive gravity

The theory of a single massive spin-2 field can also be considered as a bimetric theory,
with a non-dynamical background metric g̃αβ and a dynamical fluctuation given by gαβ =
77
g̃αβ + hαβ . Taking the background to be Minkowski space, for simplicity, we can then
generate a mass for the spin-2 field hαβ by adding the Pauli-Fierz (PF) term to the
Einstein-Hilbert action [504], resulting in
Z
1 √ m2 √  µν αβ 
SP F = d4 x −gR + −g g g − g µα g µβ hµν hαβ , (198)
16πG 4
where m is a constant mass parameter. It is well known that, in four dimensions, a
massive spin-2 field ought to have five propagating degrees of freedom: Two of helicity
2, two of helicity 1, and one of helicity 0. However, a generic mass term with arbitrary
coefficients will result in higher derivative terms for the helicity-0 mode, giving rise to
an additional ghost-like degree of freedom. The form of the PF mass term is specifi-
cally chosen so that this is not the case to linear order. Massive gravity exhibits some
interesting phenomenology, not least the so-called vDVZ discontinuity, and its possible
resolution via the Vainshtein mechanism. These will be discussed in more detail in the
context of DGP gravity in Section 5.5.
Unfortunately, the PF Lagrangian by itself cannot describe a consistent theory be-
cause the ghost-like mode reappears at non-linear order [175]. This mode is often referred
to as the Boulware-Deser ghost, and it was believed that one could not find generalisa-
tions of the theory that succeeded in eliminating it to all orders [342]. There has, however,
been some recent progress on this issue by de Rham and Gabadadze and collaborators
[517, 380, 383, 382, 385] who have proposed the following action [382, 385]:
Z
1 √ √
SGP F = d4 x −gR + m2 −gU (g, h), (199)
16πG
P4 µ1
where U (g, h) = n=2 am δ[ν 1
. . . δνµmm] Kνµ11 · · · Kνµmm and where the am are constants, and
p 1 µα 1
Kνµ = δνµ + g µα hαν − δνµ = g hαν − g µα hαβ g βγ hγν + . . . . (200)
2 8
It is now clear that the leading order part of the potential gives the PF mass term upon
choosing a2 = 2. To study the behaviour of the theory beyond linear order it is convenient
to restore general coordinate invariance by means of the Stuckelberg trick. To this end
one can perform the following field redefinition [382, 64],

ĥµν ∂φα ∂φβ

hµν = + ηµν − ηαβ µ (201)
Mpl ∂x ∂xν
ĥµν ∂µ ∂ν π η αβ ∂α ∂µ π∂β ∂ν π
= +2 − , (202)
Mpl Λ33 Λ63

where in going from Eq. (201) to Eq. (202) we have set φα = xα − η αβ ∂β π/Λ3 , in
order to focus√on the dynamics of the helicity zero mode. Note that Λ3 = (Mpl m2 )1/3
and Mpl = 1/ 8πG. For the original PF action, Eq. (198), the Boulware-Deser ghost
reveals itself by expanding the action in terms of ĥµν . At zeroth order one finds higher
derivative terms for π that contribute to the equations of motion, indicating the presence
of the ghostly extra mode. In contrast, the generalised PF action is chosen so that the
resulting higher derivative terms contribute a total derivative at zeroth order in ĥµν .
78
This is a crucial first step in avoiding the extra mode. One can go further and study
the theory in the so-called decoupling limit, m → 0, Mpl → ∞, and Λ3 =constant. After
some suitable field redefinitions one finds that the theory contains the quintic galileon
Lagrangian [382]. Note that it does not reproduce the galileon theory discussed in Section
4.4 exactly, since generically there is mixing with a graviton of the form ĥµν Xµν3
, where
3
Xµν is cubic in π. This mixing cannot be eliminated by a local field redefinition and may
have important phenomenological consequences. In particular, when this coupling comes
in with a particular sign it can prevent the recovery of GR inside the Vainshtein radius
around a heavy source [738, 739, 297]. In any event, one can confidently say that the
Boulware-Deser ghost does not appear in the decoupling limit. Of course, it is possible
that this limit corresponds to taking its mass to infinity, and that it will reemerge in the
full theory. Whether or not this is the case has yet to be established.
Self-accelerating and self-tuning cosmologies were studied de Rham and Gabadadze’s
theory in [384], whilst spherically symmetric solutions have also been considered recently
[738, 739, 297].

3.3.4. Bigravity
A class of theories that were first proposed in the 1970s by Isham, Salam, Stradthee
[647], and revisited a few years ago by Kogan [355] and collaborators, have recently been
resurrected by Bañados and collaborators [84, 85, 70] (see for [1150] a short overview).
Further studies of bi-gravity include weak-field solutions and gravitational waves [140],
exact spherically symmetric solutions [141] and the energy of black holes [329]. The
starting point [84] is an extension of Eddington’s affine theory (see section 2) so that the
α
dynamical fields are a metric gµν (with curvature scalar R) and a connection Cµν with
Ricci tensor Kµν [C]. The action is
Z q
1 √ 2
S[g, C] = d4 x −g(R − 2Λ) + 2 −det[gµν − `2 Kµν ] (203)
16πG α`
where Λ is a cosmological constant, α is a dimensionless parameter and ` is a length scale.
It may be shown [85] by introducing a 2nd metric g̃µν corresponding to the connection
α
Cµν that the above theory is a special case of bigravity with action given by
Z  
1 √ p p 1
S= d4 x −g(R − 2Λ) + −g̃(R̃ − 2Λ̃) − −g̃ 2 (g̃ −1 )αβ gαβ , (204)
16πG `
where Λ̃ = `α2 is a cosmological constant term. In these theories, both metrics are used
to build Einstein-Hilbert actions even though only one of them couples to the matter
content.
Such bigravity theories lead to interesting behaviour on cosmological scales [85, 70].
The homogeneous and isotropic FLRW metrics can be written as gαβ = diag(−1, a2 , a2 , a2 )
and g̃αβ = diag(−X 2 , Y 2 , Y 2 , Y 2 ), where X and Y are functions of t alone. The corre-
sponding Friedmann equations are then of the form
8πG
H2 = (ρ̃ + ρ), (205)
3
where ρ̃ = Y 3 /(8πG`2 Xa3 ). This fluid satisfies a conventional conservation equation of
the form
dρ̃
= −3(1 + w̃)ρ̃, (206)
dt
79
where w̃, satisfies a somewhat intricate evolution equation, given by
" s #
dw̃ 3/2
(1 + 3w̃)ρ`
= 2w̃ 1 + 3w̃ + 4(−w̃) Ω̃α − 2 , (207)
dt ρc

where ρc = ρ + ρ̃, ρ` = (8πG`2 )−1 and Ω̃ = ρ̃/ρc . The extra metric here can lead to a
range of interesting behaviours and, in particular, can drive the expansion to a de Sitter
phase, or mimic the effects of dark matter. Anisotropic universes in these models were
studied in [1066].
The cosmological evolution of perturbations in these theories has been worked out
in some detail, and it turns out that the perturbations in the auxiliary field can be
rewritten in the form of a generalised dark matter fluid [626] with density, momentum,
pressure and shear that obey evolution equations. As a result, it is possible to work out
cosmological observables such as CMB anisotropies and large-scale structure. In [70] it
was found that distinctive signatures emerge during periods of accelerated expansion. If
the ρ̃ field dominates, and is responsible for cosmic acceleration, there is a clear instability
in the gravitational potentials; they not only grow but diverge leading very rapidly to an
overwhelming integrated Sachs-Wolfe effect on large scales. It is difficult to reconcile the
angular power spectrum of fluctuations and the power spectrum of the galaxy distribution
predicted by a bimetric theory that unifies the dark sector with current data. If we restrict
ourselves to a regime in which ρ̃ simply behaves as dark matter, however, then the best-fit
bimetric model is entirely indistinguishable from the standard CDM scenario.
Bigravity theory can also be extended to consider more complicated actions, such as
Z p 
1 
SL = − d4 x −g̃ κ0 (g̃ −1 )αβ gαβ + κ1 ((g̃ −1 )αβ gαβ )2 + κ2 (g̃ −1 )αβ (g̃ −1 )αβ ,(208)
16πG
and, although a full analysis of its PPN parameters has been undertaken [308], its cosmol-
ogy remains to be explored. Black holes, and their thermodynamics, have been studied
in bimetric gravity in [71].

3.3.5. Bimetric MOND

A bimetric theory of MOND, somewhat akin to bigravity, has recently been proposed
by Milgrom [896]. The action for bimetric MOND, or BIMOND, is of the form
Z h √ p i
1
S = d4 x β −gR + α −g̃ R̃ − 2(g̃g)1/4 a20 M
16πG
−S̃M (g̃µν , ψ̃) − SM (gµν , ψ) (209)

where M is the interaction term that connects the two metrics, and ψ and ψ̃ are the
matter fields that couple to gµν and g̃µν , respectively. The factor M is a non-linear
function of the tensor Υµν , given by

Υµν = C αµβ C βνα − C αµν C ααα , (210)

where

C αµν = Γ̃αµν − Γαµν , (211)

80
and Γαµν and Γ̃αµν are the Christoffel symbols constructed from gµν and g̃µν , respectively.
Note that the even though the Γs are not tensors, C, constructed in this way, is a tensor.
The constants α and β can be kept unrelated, leading to different gravitational couplings
in the two sectors. If we set α = β, however, but leave β arbitrary, then we get the field
equations
βGµν + Sµν = −8πGTµν (212)
and
β G̃µν + S̃µν = −8πGT̃µν , (213)
which look like the conventional Einstein equations, except for the contributions from
Sµν and S̃µν which contain the interaction terms between the two metrics. These tensors
are quadratic in C αµν , and are non-linear functions of g, g̃ and g µν g̃µν .
This theory has been constructed to reproduce MOND phenomenology on small
scales, in the weak field and low acceleration regime. Its cosmological implications have
been studied in [315, 897], where it was shown that in the high acceleration regime BI-
MOND reproduces conventional FLRW behaviour. In low acceleration regime, however,
we have that the scale factor a(τ ) (where τ is conformal time) can take the form the
form a ' τ p , where [315]
1 − 3w̃
p= . (214)
(1 + 2w + 8nw − 3w̃ − 8nw̃ − 6ww̃)
and where w and w̃ are the equations of state for the matter coupled to gµν and g̃µν , and
n is one of the parameters in M. This leads to an interesting range of behaviours. For
example, it is possible to have a dust filled universe that is static, if the matter coupled
to the second metric is radiation. It is shown in [897] that to calculate fluctuations about
an FLRW background in either metric requires a knowledge of the matter coupled to
both metrics. It is also shown that the growth of fluctuations does not proceed in a
purely Newtonian way, but has a MOND contribution as well.

3.4. Tensor-Vector-Scalar Theories

In General Relativity, the space-time metric gµν is the sole dynamical agent of gravity.
We have seen above that scalar-tensor theories extend this by adding a scalar field that
mediates a spin-0 gravitational interaction, while in Einstein-æther theories one makes
use of a vector field. TeVeS has both of these types of fields as extra degrees of freedom: A
scalar field, φ, and a (dual) vector field, Aµ , both of which participate in the gravitational
sector. Like GR, it obeys the Einstein equivalence principle, but unlike GR it violates
the strong equivalence principle.
TeVeS is a product of past antecedent theories, namely the Aquadratic Lagrangian
theory of gravity (AQUAL) and its relativistic version [125], the phase-coupling gravita-
tion [126], the disformal relativistic scalar field theory [130], and the Sanders’ stratified
vector field theory [1093]. Since its inception [128] TeVeS has been intensively researched,
including studies of cosmology [128, 592, 1153, 419, 427, 1147, 177, 1305, 1148, 502],
spherically symmetric solutions [128, 535, 660, 1082, 774, 1154], gravitational collapse
and stability [1124, 331], solar system tests [128, 535, 124, 1198, 1081], gravitational
lensing [128, 295, 288, 1307, 1306, 287, 498, 1291, 1128, 296, 870], issues of superlumi-
nality [205], and the travel time of gravitational waves [664, 663, 414]. A thorough and
up-to-date review of TeVeS can be found in [1152]. Here we will concentrate mostly on
cosmological features of the theory.
81
3.4.1. Actions and field equations
The original and most common way to specify TeVeS is to write the action in a mixed
frame. That is, we write the action in the “Bekenstein frame” for the gravitational fields,
and in the ‘physical frame’, for the matter fields. In this way we ensure that the Einstein
equivalence principle is obeyed. The three gravitational fields are the metric, g̃µν (with
connection ∇˜ a ), that we refer to as the Bekenstein metric, the Sanders vector field, Aµ ,
and the scalar field, φ. To ensure that the Einstein equivalence principle is obeyed, we
write the action for all matter fields using a single ‘physical metric’, gµν (with connection
∇µ ), that we call the ‘universally coupled metric’ 20 . The universally coupled metric is
algebraically defined via a disformal relation [127] as

gµν = e−2φ g̃µν − 2 sinh(2φ)Aµ Aν . (215)

The vector field is further enforced to be unit-time-like with respect to the Bekenstein
metric, i.e.
g̃ µν Aµ Aν = −1. (216)
The unit-time-like constraint is a phenomenological requirement for the theory to give
an acceptable amount of light bending. Using the unit-time-like constraint, Eq. (216),
it can be shown that the disformal transformation for the inverse metric is

g µν = e2φ g̃ µν + 2 sinh(2φ)Aµ Aν , (217)

where Aµ = g̃ µν Aν . The existence of a scalar and a vector field may seem odd at first,
but they are both the product of a series of extensions from older theories, based on
theoretical and phenomenological constraints.

Actions

TeVeS is based on an action, S, which is split as

S = Sg̃ + SA + Sφ + Sm , (218)

where Sg̃ , SA , Sφ and Sm are the actions for g̃µν , the vector field, Aµ , the scalar field,
φ, and matter fields, respectively.
As already discussed, the action for g̃µν , Aµ , and φ is written using only the Bekenstein
metric, g̃µν , and not gµν , and is such that Sg̃ is of Einstein-Hilbert form
Z p
1
Sg̃ = d4 x −g̃ R̃, (219)
16πG

where g̃ and R̃ are the determinant and scalar curvature of g̃µν , respectively, while G
is the bare gravitational constant. The relation between G and the measured value of
Newton’s constant, GN , will be elaborated on below, in Section 3.4.2.

20 Some work on TeVeS, including the original articles by Sanders [1093] and Bekenstein [128], refer

to the Bekenstein frame metric as the “geometric metric”, and denote it as gµν , while the universally
coupled metric is referred to as the “physical metric”, and is denoted by g̃µν . Since it is more common
to denote the metric which universally couples to matter as gµν , in this review we interchange the tilde.
82
 The action for the vector field, Aµ , is given by
Z p
1
SA = − d4 x −g̃ [KF µν Fµν − 2λ(Aµ Aµ + 1)] , (220)
32πG
where Fµν = ∇µ Aν −∇ν Aµ leads to a Maxwellian kinetic term, λ is a Lagrange multiplier
that ensures the unit-timelike constraint on Aµ , and K is a dimensionless constant.
Indices on Fµν are moved using the Bekenstein metric, i.e. F µν = g̃ µα Fαν . This form
of a vector field action has been considered by Dirac as a way of incorporating electrons
into the electromagnetic potential [424, 425, 426]. More recently it has been considered
as a natural generalisation of GR, in the Einstein-æther theories discussed in Section 3.2
[656, 655].
The action for the scalar field, φ, is given by
Z p h i
1 ˜ µ φ∇
˜ ν φ + V (µ) ,
Sφ = − d4 x −g̃ µ ĝ µν ∇ (221)
16πG
where µ is a non-dynamical dimensionless scalar field, ĝ µν is a new metric defined by

ĝ µν = g̃ µν − Aµ Aν , (222)

and V (µ) is an arbitrary function which typically depends on a scale, `B . Not all choices
of V (µ) give the correct Newtonian or MONDian limits in a quasi-static situation. The
allowed choices are presented in Section 3.4.2. The metric ĝ µν is used in the scalar field
action, rather than g̃ µν , to avoid the superluminal propagation of perturbations. Note
that it is possible to write the TeVeS action using ĝ µν , with the consequence of having
more general vector field kinetic terms (see the appendix of [1152]).
It is also easier in some cases to work with an alternative form for the scalar field
action that does not have the non-dynamical field, µ, but rather has the action written
directly in terms of a non-canonical kinetic term for φ given by a free function f (X),
with X defined by
X = `2B ĝ µν ∇µ φ∇ν φ. (223)
df
The field µ is then given in terms of f (X) by µ = dX , while f (X) can be related to V
by f = µX + `2B V .
The matter fields in the action are coupled only to the ‘universally coupled metric’,
gµν , and thus their action is of the form
Z
√
Sm [g, χA , ∇χA ] = d4 x −g L[g, χA , ∇χA ], (224)

for some generic collection of matter fields, χA . The matter stress-energy tensor is then
defined with respect to δSm in the usual way.
It should be stressed that the action for the scalar field has been constructed such
that the theory has a MONDian non-relativistic limit, under the right conditions, for
specific choices of functions V (µ) (or equivalently F (X)). The action for the vector field
has no particular significance other than the fact that it is simple. More general actions
can be considered without destroying the MOND limit, but that in addition provide new
features or improved phenomenology.
83
The field equations

The field equations of TeVeS are found using a variational principle. This gives two
constraint equations, namely the unit-timelike constraint, given in Eq. (216), and the
µ-constraint:
dV
ĝ µν ∇µ φ∇ν φ = − , (225)
dµ
that is used to find µ as a function of ∇µ φ. The field equations for g̃µν are given by
 
G̃µν = 8πG Tµν + 2(1 − e−4φ )Aα Tα(µ Aν)
h i
+µ ∇ ˜ µ φ∇
˜ ν φ − 2Aα ∇ ˜ ν) φ + 1 (µV 0 − V ) g̃µν
˜ α φ A(µ ∇
  2
α 1 αβ
+K F µ Fαν − F Fαβ g̃µν − λAµ Aν , (226)
4

where G̃µν is the Einstein tensor constructed from g̃µν . The field equations for the vector
field, Aµ , are

˜ α F αµ = −λAµ − µAν ∇
K∇ ˜ ν φ∇
˜ µ φ + 8πG(1 − e−4φ )Aν Tνµ , (227)

and the field equation for the scalar field, φ, is

h i  
˜ µ µĝ µν ∇
∇ ˜ ν φ = 8πGe−2φ g µν + 2e−2φ Aµ Aν Tµν . (228)

The Lagrange multiplier can be solved for by contracting Eq. (227) with Aµ .

3.4.2. Newtonian and MOND limits

We now describe the quasi-static limit, as relevant for establishing the existence of the
Newtonian and MONDian limits. The details of the derivation we consider here can be
found in [128], while an alternative shorter derivation in the spirit of the PPN formalism
is given in [1152].
It can be shown that the PPN parameter γ is unity in TeVeS, hence the universally
coupled metric can be written as

ds2 = −(1 + 2Φ)dt2 + (1 − 2Φ)δij dxi dxj , (229)

~
where Φ is related to the acceleration of particles, ~a, by ~a = −∇Φ. The scalar field is
perturbed as φ = φc + ϕ, where φc is the cosmological value of φ. The Bekenstein metric
takes a similar form to gµν :

ds̃2 = −e−2φc (1 + 2Φ̃)dt2 + e2φc (1 − 2Φ̃)δij dxi dxj . (230)

The vector field does not play a role at this order of perturbations, and is simply given
by
Aµ = e−φc (−1 − Φ̃, 0, 0, 0). (231)

84
The field equations to O(v 2 ) are then
8πG
∆Φ̃ = ρ, (232)
2−K
and
h i
~ · µ∇ϕ
∇ ~ = 8πGρ, (233)

while the potential Φ is given via the disformal transformation Φ = Φ̃+ϕ. Eqs. (232) and
(233) can be solved for any quasi-static situation, regardless of the boundary conditions
or the symmetry of the system in question, provided a function µ(|∇ϕ|)~ is supplied.
To find the Newtonian and MONDian limits we can consider, for simplicity, spher-
ically symmetric situations. In this case we can combine Eqs. (232) and (233) into a
single equation for Φ, called the AQUAL equation:
h i
∇~ · µm ∇Φ
~ = 4πGN ρ, (234)

where
GN µ
µm =  . (235)
2G 1 + µ
2−K

The ratio GN /G is not free, but is found by taking the limit µm → 1, i.e. the Newtonian
limit. Consistency requires that µ → µ0 which is then a constant21 contained in the
function f (or V ). This gives the relation
GN 2 2
= + . (236)
G µ0 (2 − K)
~ N|
|∇Φ
The MOND limit is now clearly recovered as µm → a0 , and we get

~
2G |∇ϕ| 2G 1 φc √
µ→ = e X, (237)
GN a0 GN `B a0
df
where X is given in Eq. (223). Since µ = dX , we may integrate the above equation to
find the function f (X), which in the MOND limit should be given by
2 1
f→   eφc X 3/2 , (238)
3`B a0 1
+ 1
µ0 2−K

where the integration constant has been absorbed into the cosmological constant associ-
ated with the metric g̃µν . Since both X and f are dimensionless we may define a new
constant β0 , such that a0 is a derived quantity given by
2 1
a0 =   eφc , (239)
3β0 `B 1
+ 1
µ0 2−K

21 The 8π
constant µ0 is related to the constant k introduced by Bekenstein as µ0 = k
.
85
and the function f has the MONDian limit f → β0 X 3/2 . Since in the Newtonian limit
we have f → µ0 X, there are at least three constants that can appear in f (X), namely
µ0 , β0 and `B .
In terms of the function dV dV 4
dµ the MONDian limiting case implies that dµ → − 9β02 `2B µ
2

as µ → 0, while it diverging as µ → µ0 in the Newtonian limit. This second limit is

imposed if dV dµ → (µ0 − µ)
−m
, for some constant m. Bekenstein chooses this to be m = 1,
although other choices are equally valid, even functions that have essential singularities.
It is clear from Eq. (239) that a0 depends on the cosmological boundary condition,
φc , which can differ for each system, depending on when it was formed. It could thus be
considered as a slowly varying function of time. This possibility has been investigated
by Bekenstein and Sagi [129], and by Limbach et al. [810].
The two limiting cases for f (X) are somewhat strange. In particular we require that
f (X) → X for X  1 to recover the Newtonian limit, and that f (X) → X 3/2 for
X  1 (i.e. a higher power) to recover the MONDian limit. This signifies that in this
kind of formulation of relativistic MOND (i.e. in terms of a scalar field) the function
f (X) should √ be non-analytic. It further signifies that f (X) can be expanded in positive
powers of X for small X, and in positive powers of X1 for large X, but that these
two expansions cannot be connected. In other words, it is impossible to perturbatively
connect the Newtonian regime with the MONDian regime via a perturbation series in
~
|∇ϕ|.
The Bekenstein free function in [128] is given in the notation used in this review by

dV 3 µ2 (µ − 2µ0 )2
=− 2 2 , (240)
dµ 32π`B µ0 (µ0 − µ)
q
4 2πµ0
which means that β0 = 3 3 , and thus
√
3 1
a0 = √   eφ c . (241)
2 2πµ0 `B 1
+ 1
µ0 2−K

This is in agreement with [129] (the authors of [177] erroneously inverted a fraction in
their definition of a0 ).

3.4.3. Homogeneous and isotropic cosmology

Homogeneous and isotropic Friedmann-Lemaitre-Robertson-Walker (FLRW) solu-
tions to the field equations of TeVeS have been extensively studied [128, 592, 1153,
419, 427, 1147, 177, 1305, 1148, 502]. In this case the universally coupled metric can be
written in the conventional synchronous form as

ds2 = −dt2 + a2 (t)qij dxi dxj , (242)

where a(t) is the ‘physical scale factor’. Here we assume for simplicity that the hyper-
surfaces of constant t are spatially flat (see [1147] for the curved case). The Bekenstein
metric then has a similar form, and can be written as

ds̃2 = −dt̃2 + b2 (t̃)qij dxi dxj , (243)

86
for a second scale factor b(t̃). The disformal transformation relates the two scale factors
by a = be−φ , while the two time coordinates t and t̃ are related by dt = eφ dt̃. The
physical Hubble parameter is defined as usual as by H = aȧ , while the Bekenstein frame
Hubble parameter is H̃ = d dlnt̃ b eφ H + dφ
dt̃
. Cosmological evolution is governed by the
analogue of the Friedmann equation:

3H̃ 2 = 8πGe−2φ (ρφ + ρ) , (244)

where ρ is the energy density of physical matter, which obeys the energy conservation
equation with respect to the universally coupled metric, and where the scalar field energy
density, ρφ , is given by
 
e2φ dV
ρφ = µ +V . (245)
16πG dµ
Similarly, one can define a scalar field pressure by
 
e2φ dV
Pφ = µ −V (246)
16πG dµ

The scalar field evolves according to the two differential equations:

dφ 1
= − Γ, (247)
dt̃ 2µ
and
dΓ
+ 3H̃Γ = 8πGe−2φ̄ (ρ + 3P ), (248)
dt̃
2
where µ is found by inverting φ0 = 21 dV dµ .
It is important to note that
√ the vector field must point to the time direction, so that
it can be written as Aµ = ( g̃00 , ~0). In this case it does not contain any independent
dynamical information, and it does not explicitly contribute to the energy density. Its
only effect is on the disformal transformation which relates the Bekenstein-frame Fried-
mann equation, Eq. (244), with the physical Friedmann equation. This is also true in
cases where the vector field action is generalised, and where the only effect is a constant
rescaling of the left-hand-side of the Bekenstein-frame Friedmann equation, as discussed
in [1148].

FLRW solutions with the Bekenstein function

In the original TeVeS paper [128] Bekenstein studied the cosmological evolution of
an FLRW universe by assuming that the free function is given by Eq. (240). He showed
that the scalar field contribution to the Friedmann equation is very small, and that φ
evolves very little from the early universe until today. He noted that with this choice of
function, a cosmological constant term has to be added in order to have an accelerating
expansion today, as appears to be required by cosmological observations.
Many other studies on cosmology in TeVeS have also used the Bekenstein function,
see for example [592, 1153, 427, 1148]. In particular, Hao and Akhoury noted that the
integration constant obtained by integrating Eq. (240) can be used to get a period
87
of accelerating expansion, and that TeVeS therefore has the potential to act as dark
energy [592]. However, such an integration constant cannot be distinguished from a bare
cosmological constant term in the Bekenstein frame, and so it is somewhat dubious as to
whether this can really be interpreted as dark energy arising from TeVeS. Nevertheless, it
would not be a surprising result if some other TeVeS functions could, in fact account for
dark energy, as the scalar field action in TeVeS close resemblances that of k-essence [65,
66]. Zhao has investigated this issue further [1305](see below).
Exact analytical and numerical solutions with the Bekenstein free function, Eq. (240),
have been found by Skordis et al.in [1153], and by Dodelson and Liguori in [427]. It turns
out that not only, is the scalar field is subdominant, as Bekenstein noted, but its energy
density also tracks the matter fluid energy density. The ratio of the energy density in
the scalar field to that of ordinary matter then remains approximately constant, so that
the scalar field tracks the matter dynamics. One then gets that

(1 + 3w)2
Ωφ = , (249)
6(1 − w)2 µ0

where w is the equation of state of the matter fields22 . Since µ0 is required to be very
large, the energy density in the scalar field is always small, with values typically less than
Ωφ ∼ 10−3 in a realistic situation. Tracker solutions are also present for this choice of
function in versions of TeVeS with more general vector field actions [1148].
In realistic situations, tracking in the radiation era is almost never realised, as has
been noted by Dodelson and Liguori [427]. Rather, during the radiation era, the scalar
field energy density is subdominant but slowly growing, such that φ ∝ a4/5 . However,
upon entering the matter era φ settles into the tracker solution. This transient solution
in the radiation era has been generalised by Skordis to arbitrary initial conditions for
φ, more general free functions (see below), and a general vector field action [1148]. It
should be stressed that the solution in the radiation era is important for setting up initial
conditions for the perturbations about FLRW solutions that are relevant for studying
the CMB radiation and Large-Scale Structure (LSS).
From Eq. (239) we see that a0 for a quasi-static system depends on the cosmological
value of the scalar field at the time the system broke off from the expansion, and collapsed
to form a bound structure. It is then possible that different systems could exhibit different
values of a0 depending on when they formed. The impact of evolving a0 on observations
has been investigated in [129, 810].
Finally, note that the sign of φ̇ changes between the matter and cosmological constant
eras. In doing so, the energy density of the scalar field goes momentarily through zero,
since it is purely kinetic and vanishes for zero φ̇ [1153].

FLRW solutions by generalising the Bekenstein function

Bourliot et al. [177] studied more general free functions, that have the Bekenstein
function as a special case. In particular they introduced two new parameters, a constant,

22 Note that this excludes the case of a stiff fluid with w = 1.

88
µa , and a power index, n, such that the free function is generalised to

dV (n) 3 µ2 (µ − µa µ0 )n
=− 2 2 . (250)
dµ 32π`B µ0 µ0 − µ

This function23 reduces to the Bekenstein function when n = 2 and µa = 2. It retains

the property of having a Newtonian limit as µ → µ0 and a MOND limit as µ → 0. The
cosmological evolution depends on the power index, n. More general functions can also
be constructed by considering the sum of the above prototypical function with arbitrary
P dV (n)
coefficients, i.e. by taking dV
dµ = n cn dµ [177].
Clearly dV
dµ (µa µ0 ) = 0, and at this point φ̇ → 0. Now suppose that the integration
constant is chosen such that V (µa µ0 ) = 0 as well. Then, just like the case of the
Bekenstein function, one finds tracker solutions: The function µ is driven to µ = µa µ0 ,
at which point φ̇ = 0. There are no oscillations around that point, but it is approached
slowly so that it is exactly reached only in the infinite future. The scalar field relative
density is now given by
(1 + 3w)2
Ωφ = , (251)
3µa (1 − w)2 µ0
independent of the value of n. It should also be pointed out that the evolution of the
physical Hubble parameter, H, can be different than the case of GR even in the tracking
1+3µa µ0
phase [177]. For example in the case w = 0 we have H ∝ a−nh , where nh = 2(µ a µ0 −1)
.
Furthermore, just like the Bekenstein case, the radiation era tracker is untenable
for realistic cosmological evolutions, for which µ0 must be large so that Ωφ is small
(. 10−3 ). In this case we once again get a transient solution where the scalar field
evolves as φ ∝ a4/(3+n) [1148]. In the case that the integration constant is chosen
such that V (µa µ0 ) 6= 0 one has an effective cosmological constant present. Thus, once
again, we get tracker solutions until the energy density of the Universe drops to values
comparable with this cosmological constant, at which time tracking comes to an end,
and the Universe enters a de Sitter phase.
The cases −2 < n < 0 turn out to be pathological as they lead to singularities in the
cosmological evolution [177]. The case n = −3 is well behaved when the matter fluid is a
cosmological constant, but is also pathological when w = −1 [177]. The cases for which
n ≤ −4 are well behaved in the sense that no singularities occur in the cosmological
evolution. Contrary to the n ≥ 1 cases, the cosmological evolution drives the function
µ to infinity. Moreover, these cases do not display the tracker solutions of n ≥ 1, but
rather the evolution of ρφ is such that it evolves more rapidly than the matter density,
ρ, and so quickly becomes subdominant. The general relativistic Friedmann equation is
thus recovered, such that 3H 2 = 8πGρ. This also results in H̃ = H, which means that
the scalar field is slowly rolling.
The evolution of the scalar field variables Γ, φ and µ then depends on the equation
of state of the matter fields. If the background fluid is a cosmological constant, then we
get de Sitter solutions for both metrics, and it can be shown that Γ = 2H(e−3Ht − 1).

23 Note that [177] uses a different normalisation for V , and their results can be recovered by rescaling
q
3 n−3
the `B used in this report by `B → `B 2
µ 0 .
89
For the case of a stiff fluid, with w = 1, we get that Γ has power-law solutions that are
inverse powers of t, so that Γ = 6t + Γt30 . A similar situation arises when −1 < w < 1, for
which we get Γ = 2(1+3w) 2 1
1−w H, and the Hubble parameter evolves as H = 3(1+w) t . Notice
that the limit w → 1, for the −1 < w < 1 case, does not smoothly approach the w = 1
case.
Mixing different powers of n ≥ 1 leads once again to tracker solutions. One may
have to add an integration constant in order to keep V (µa µ0 ) = 0, although for certain
combinations of powers n and coefficients ci this is not necessary. Mixing n = 0 with some
other n ≥ 1 cannot remove the pathological situation associated with the n = 0 case.
Mixing n = 0 with both positive and negative powers could however lead to acceptable
cosmological evolution since the effect of the negative power is to drive µ away from the
µ = µa µ0 point. In general, if we mix an arbitrary number of positive and negative
powers we get tracker solutions provided we can expand the new function in positive
definite powers of (µ − µ0a µ0 ), where µ0a is some number different from the old µa .
The observational consequences for the CMB and LSS have not been investigated for
this class of function, unlike the case of the Bekenstein function.

Inflationary/accelerating expansion for general functions

Diaz-Rivera, Samushia and Ratra [419] have studied cases where TeVeS leads to
inflationary, or self-accelerating, solutions. They first consider the vacuum case, in which
they find that de Sitter solutions exist with b ∼ eH̃0 t̃ ,q
where the Bekenstein frame Hubble
µ2 V dV
constant H̃0 is given by the free function as H̃0 = 6 , and where dµ = 0 (i.e. the
0

scalar field is constant, φ = φi ). Such a solution will always exist in vacuum provided
that the free function satisfies dV
dµ (µv ) = 0 and V (µv ) 6= 0, for some constant µv . In that
case, the general solution is not de Sitter since both φ and µ will be time-varying, but
will tend to de Sitter as µ → µv . Indeed, the n ≥ 1 case of Bourliot et al. [177] with an
integration constant is precisely this kind of situation.
In the non-vacuum case, for a fluid with equation of state P = wρ, they make
the ansatz b3(1+w) = e(1+3w)φ . This brings the Friedmann equation into the form
3H̃ 2 = 8πGρ0 + 12 (µ dV
dµ + V ), where ρ0 is the matter density today. Once again, they
assume that the free-function-dependent general solution drives µ to a constant µv , but
φ is evolving. Thus, we must have that φ = φ1 t̃ + φ2 , such that φ̇ = φ1 is a constant. In
order for φ1 to be non-zero we must have dV dµ (µv ) 6= 0. However, there is a drawback to
this approach. As they point out, consistency with the scalar field equation requires that
w < −1. Furthermore, although this solution is a de Sitter solution in the Bekenstein-
frame, it corresponds to a power-law solution for the universally coupled metric. In order
for this power-law solution to lead to acceleration, they find that −5/3 < w < −1. This
range of w corresponds to a phantom fluid.

Accelerated expansion in TeVeS

The simplest case of accelerated expansion in TeVeS is provided by a cosmological

constant term. This is equivalent to adding an integration constant to V (µ) [592, 177],
and it corresponds to the accelerated expansion considered by Diaz-Rivera, Samushia
and Ratra [419] in both the vacuum and non-vacuum cases (see above). These solutions
90
therefore suffer from the usual fine-tuning and coincidence problems, and so it is of
interest to look for accelerating solutions without such a constant, simply by employing
the scalar field (these need not be de Sitter solutions).
Zhao used a function dV 2
dµ ∝ µ to obtain solutions which provide acceleration, and
compared his solution with the SN1a data [1305], finding good agreement. However, it
is not clear whether other observables, such as the CMB angular power spectrum, or
observations of LSS are compatible with this function. Furthermore, this function is not
realistic, as it does not have a Newtonian limit (it is always MONDian). Although no
further studies of accelerated expansion in TeVeS have been performed, it is plausible
to think that certain choices of function could lead to acceleration. This is because the
scalar field action has the same form as a k-essence/k-inflation action [65, 66], which
has been considered as a candidate theory for acceleration. More precisely, the system
of cosmological equations corresponds to k-essence coupled to matter. It is not known
in general whether this type of model has similar features as the uncoupled k-essence
models, although Zhao’s study indicates that this a possibility.

Realistic FLRW cosmology

In TeVeS, cold dark matter is absent. Therefore in order to get acceptable values
for the physical Hubble constant today (i.e. around H0 ∼ 70 km s−1 Mpc−1 ) , we have
to supplement the absence of CDM with something else. The reason for this is simply
that if all the energy density in the Universe today was in the form of baryons, then the
Hubble constant would be lower than what is observed by a factor of ∼5. Possibilities
of what this supplementary material could be include the scalar field itself, massive
neutrinos [1153, 502], and a cosmological constant. At the same time, one has to get
the right angular diameter distance to recombination [502]. These two requirements can
place severe constraints on the allowed form of the free functions.

3.4.4. Cosmological perturbation theory

Cosmological perturbation theory in TeVeS has been formulated to linear order in
[1147], and in variants of TeVeS with more general vector field actions in [1148]. The
scalar modes of the linearly perturbed universally coupled metric are given in the con-
formal Newtonian gauge, as usual, by

ds2 = −a2 (1 + 2Ψ)dτ 2 + a2 (1 − 2Φ)qij dxi dxj , (252)

where τ is conformal time, defined by dt = adτ . Here, we will assume, for simplicity,
that the spatial curvature is zero. The reader is referred to [1147, 1148] for the curved
cases, as well as for an enunciation of vector and tensor perturbations. The scalar field
is perturbed as φ = φ̄ + ϕ, where φ̄ is the FLRW background scalar field, and ϕ is the
perturbation. The vector field is perturbed as Aµ = ae−φ̄ (1 + Ψ − ϕ, ∇ ~ i α), such that
the unit-timelike constraint is satisfied. This removes the time component of Aµ as an
independent dynamical degree of freedom. Thus, there are two additional dynamical
degrees of freedom, when comparing to cosmological perturbation theory in GR: The
scalar field perturbation, ϕ, and the vector field scalar mode, α.
The perturbed field equations for the scalar modes can be found in the conformal
Newtonian gauge in [1153], and in more form (including in the synchronous gauge)
91
in [1147]. Perturbation equations for more general TeVeS actions are given in [1148].
Here we only present the Newtonian gauge equations of the original TeVeS formulation.
0
We define the following variables: Φ̃ = Φ − ϕ, Ψ̃ = Ψ − ϕ, ζ̃ = −(1 − e−4φ̄ )α and H̃ = bb .
The scalar field obeys the two first order equations

µ̄ −3φ̄ 2
eφ̄ h i
γ0 = −3H̃γ + e k ϕ + φ̄0 α − µ̄φ̄0 6Φ̃0 + 2k 2 ζ̃
a  a 
X X  
+8πGae−3φ̄  δρf + 3δPf + (ρ̄f + 3P̄f ) Ψ̃ − 2ϕ  (253)
f f

and
1
ϕ0 = − ae−φ̄ γ + φ̄0 Ψ̃. (254)
2U
The vector field equations are given by
  X
K E 0 + H̃E = −µ̄φ̄0 (ϕ − φ̄0 α) + 8πGa2 (1 − e−4φ̄ ) (ρ̄f + P̄f )(θf − α) (255)
f

and

α0 = E + Ψ̃ + φ̄0 − H α (256)
and finally the Einstein equations are given by the Hamiltonian constraint

h i
−2 k 2 − 3κ Φ̃ − 2e4φ̄ H̃ 3Φ̃0 + k 2 ζ̃ + 3H̃Ψ̃ + ae3φ̄ φ̄0 γ
X
−Kk 2 E = 8πGa2 ρ̄f [δf − 2ϕ] (257)
f

the momentum constraint equation

X
Φ̃0 + κζ̃ + H̃Ψ̃ − µ̄φ̄0 ϕ = 4πGa2 e−4φ̄ (ρ̄f + P̄f )θf (258)
f

and the two propagation equations

 
h i
6Φ̃00 + 2k 2 ζ̃ 0 − e−4φ̄ Ψ̃ + 2e−4φ̄ k 2 − 3κ Φ̃ + 2H̃ 6Φ̃0 + 3Ψ̃0 + 2k 2 ζ̃
h i µ h i
+4φ̄0 3Φ̃0 + k 2 ζ̃ + 3 ae−φ̄ φ̄0 γ + 6 2H̃0 + H̃2 + 4φ̄0 H̃ Ψ̃
U
2 −4φ̄
= 24πGa e (δP − 2P ϕ) (259)

and h   i X
Φ̃ − Ψ̃ + e4φ̄ ζ̃ 0 + 2 H̃ + φ̄0 ζ̃ = 8πGa2 (ρ̄f + P̄f )Σf (260)
f

Let us now turn to the problem of specifying initial conditions for the scalar modes,
which in general should depend on the chosen form of the free function. The exact
adiabatic growing mode in TeVeS, and generalised variants, have been found by Skordis
in [1148], but only for the case of the generalised Bekenstein function. If the free function
92
is such so that the scalar field contribution to the background expansion during the
radiation era is very small, however, then the adiabatic modes for other free functions
should be only marginally different from the ones found in [1148]. In particular, the only
effect should be a difference in the initial conditions of ϕ, which is not expected to make
any difference to observations.
The only study that has been performed of the observational signatures of TeVeS in
the CMB radiation and in LSS is due to Skordis, Mota, Ferreira and Bœhm [1153]. Here
the initial conditions were chosen such that both ϕ and α, as well as their derivatives,
are initially zero. While this is not a purely adiabatic initial condition, it turns out that
it is close enough to ensure that no observable difference can be seen from isocurvature
contamination. Detailed studies of isocurvature modes in TeVeS have not yet been
conducted. In the light of the problems that TeVeS has with observations of the CMB
radiation [1153], however, it may be important to investigate what effects isocurvature
modes are likely to have. Preliminary studies by Mota, Ferreira and Skordis have shown
that setting the vector field perturbations to be large initially can have a significant
impact [910].
In addition to the four regular isocurvature modes that exist in GR, there could in
principle exist four further isocurvature modes in TeVeS: Two associated with the scalar
field, and two associated with the vector field. Preliminary studies by Skordis have shown
that none of the scalar field isocurvature modes are regular in either the synchronous
or conformal Newtonian gauges. Conversely, under certain conditions of the vector field
parameters one of the vector field isocurvature modes can be regular, while the other one
is never regular. Thus, it may be possible to have one regular isocurvature mode in the
TeVeS sector. The observational consequences of this mode are still unknown, as is its
generation method from early universe inflation. Studies of the observable spectra based
on vector or tensor modes are also yet to be conducted although the necessary equations
can be found in [1147, 1148].

3.4.5. Cosmological observations and constraints

Let us now consider the observational signatures of the perturbation theory discussed
above, and how they can be used to constrain TeVeS.

Large-scale structure observations

A traditional criticism of MOND-type theories was their lack of a dark matter com-
ponent, and therefore their presumed inability to form large-scale structure compatible
with current observational data. This criticism was based on intuition formed from a
general relativistic universe filled with baryons only. In that case it is well known that,
since baryons are coupled to photons before recombination, they do not have enough
time to grow into structures on their own. Furthermore, their oscillatory behaviour at
recombination is preserved, and is visible as large oscillation in the observed galaxy power
spectrum Pgg (k). Finally, on scales smaller than the diffusion damping scale they are
exponentially suppressed due to Silk damping. Cold dark matter solves all of these prob-
lems because it does not couple to photons, and therefore can start creating potential
wells early on in the Universe’s history, into which the baryons can fall. This is enough
to generate the right amount of structure, erase most of the oscillations, and overcome
the Silk damping.
93
 TeVeS contains two additional fields, not present in GR, that change the structure of
the equations significantly. The first study of large-scale structure observations in TeVeS
was conducted by Skordis, Mota, Ferreira and Bœhm in [1153]. Here the perturbed
TeVeS equations were solved numerically for the case of the Bekenstein function, and
the effects on the matter power spectrum, P (k), were determined. It was found that
TeVeS can indeed form large-scale structure compatible with observations, depending
on the choice of TeVeS parameters in the free function. In fact, the form of the matter
power spectrum, P (k), in TeVeS looks quite similar to the corresponding spectrum in
ΛCDM. Thus, one has to turn to other observables to distinguish the TeVeS from General
Relativity.
Dodelson and Liguori [427] provided an analytical explanation of the growth of struc-
ture that was found numerically in [1153]. They concluded the growth in TeVeS cannot
be due to the scalar field, as the scalar field perturbations are Bessel functions that decay-
ing in an oscillatory fashion. Instead, they reasoned, the growth of large-scale structure
in TeVeS is due to the vector field perturbation.
Let us see how the vector field leads to growth. Using the tracker solutions in the
matter era, from Bourliot et al. [177], we can find the behaviour of the background
functions a, b and φ̄. Using these in the perturbed field equations, after setting the
scalar field perturbations to zero, it can be shown that in the matter era the vector field
perturbation α obeys the equation
b1 0 b2
α00 + α + 2 α = S(Ψ, Ψ0 , θ) (261)
τ τ
in the conformal Newtonian gauge, where
4(µ0 µa − 1)
b1 = , (262)
µ0 µa + 3
   
2 2 2 4
b2 = µ µ − 5+ µ0 µa + 6 , (263)
(µ0 µa + 3)2 0 a K
and where S is a source term which does not explicitly depend on α. If we simultaneously
take the limits µ0 → ∞ and K → 0, for which Ωφ → 0, meaning that the TeVeS
contribution is absent, then we get b1 → 4 and b2 → 2. In this case the two homogeneous
solutions to Eq. (261) are τ −2 and τ −1 , which are decaying.
Dodelson and Liguori show that the source term S(Ψ, Ψ̇, θ) is not sufficient to create
a growing mode in the general solution to Eq. (261), and that in the general relativistic
limit TeVeS does not, therefore, provide enough growth for structure formation. Now let
us consider the general case. Assuming that the homogeneous solutions to (261) can be
written as τ n , it can be shown that for the generalised Bekenstein function of Bourliot
et al. [177] we can get r
3 1 32
n≈− + 1+ . (264)
2 2 Kµ0 µa
Thus, we can have n > 0, provided that for fixed µ0 µa we also have24

K . 0.01. (265)

24 Smaller values of µ0 µa can also raise this threshold.

94
Figure 3: LEFT: The evolution in redshift of baryon density fluctuations in TeVeS (solid line), and in
ΛCDM (dashed line) for a wavenumber k = 10−3 Mpc−1 . In both cases, the baryon density fluctuates
before recombination, and grows afterwards. In the case of ΛCDM, the baryon density eventually follows
the CDM density fluctuation (dotted line), which starts growing before recombination. In the case of
TeVeS, the baryons grow due to the potential wells formed by the growing mode in the vector field, α
(dot-dashed line). RIGHT: The difference of the two gravitational potentials, Φ − Ψ, for a wavenumber
k = 10−3 Mpc−1 as a function of redshift for both TeVeS (solid line), and ΛCDM (dotted line).

If this condition is met then there can exist a growing mode in α, which in turn feeds
back into the perturbed Einstein equation and sources a non-decaying mode in Φ that
can drive structure formation. This is displayed graphically in the left panel of Figure 3.
It is a striking result that even if the contribution of the TeVeS fields to the background
FLRW equations is negligible (∼ 10−3 or less), one can still get a growing mode that
drives structure formation.

CMB observations

A general relativistic universe dominated by baryons cannot fit the most up to date
observations of anisotropies in the CMB [984]. This is true even if a cosmological constant
and/or three massive neutrinos are incorporated into the matter budget, so that the first
peak of the CMB angular power spectrum is at the right position25 . This, however, is not
proof that only a theory with CDM can fit CMB observations (as claimed in [1155, 1173]).
A prime example to the contrary is the Eddington-Born-Infeld theory [70]. However, the
linear Boltzmann equation, and the resulting CMB angular power spectrum, have been
calculated in TeVeS, using initial conditions that are close to adiabatic [1153]. The re-
sulting fits to the data were poor, at least for the Bekenstein free function, showing that
CMB observations are, nevertheless, problematic for TeVeS. It may be that the new
isocurvature modes discussed above can provide a richer phenomenology but it remains
to be seen whether this can save this theory.

25 In this case the third peak is unacceptably lower than the second.
95
The difference in the gravitational potentials: Φ − Ψ

The result of Dodelson and Liguori [427] have a further direct consequence: The
perturbation equations in TeVeS that relate the difference of the two gravitational po-
tentials, Φ − Ψ, to the shear of matter, have additional contributions coming from the
perturbed vector field, α. This is not due to the existence of the vector field per se, but
comes from the disformal transformation in which the vector field plays an important
part. Indeed, in a single metric theory where the vector field action is Maxwellian, as in
TeVeS, there is no contribution from the vector field to Φ − Ψ. Now, as the vector field
is required to grow in order to drive structure formation, it will also inevitably lead to
growth in Φ − Ψ. This is precisely what we see numerically in the right panel of Figure 3.
If Φ − Ψ can be measured observationally, then it will provide an excellent test of TeVeS.
This possibility is discussed in more detail in section 6.

3.5. Other Theories

3.5.1. The Einstein-Cartan-Sciama-Kibble Theory
In this subsection we will briefly describe the Einstein-Cartan-Sciama-Kibble (ECSK)
theory [259, 260, 261, 1104, 1105, 691]. The ECSK theory is in many cases equivalent to
General Relativity and departs from GR only when at least one matter field has intrinsic
spin. The ECSK theory has been reviewed by Hehl et al. [601], and more recently by
Trautman [1215].

The ECSK theory as a theory with torsion

The ECSK theory is basically General Relativity with the addition of torsion. The
connection is assumed to be metric compatible, but has non-zero torsion, and is thus
given by  
µ
Γµαβ = + K µαβ , (266)
αβ
where K µαβ is the contorsion tensor given in terms of the torsion S µαβ by Eq. (40). The
Riemann tensor is antisymmetric in both the first and the last two indices, and hence
the Ricci tensor is its only unique non-vanishing contraction. It is, however, asymmetric,
and is given by Eq. (38).
The addition of torsion to the connection has a direct consequence on the geometry
of curves. In this case, autoparallels (straightest lines) are not necessarily extremals
(shortest or longest lines) as they are in GR. The former are given by
d2 xµ µ dxα dxβ
+ Γ αβ = 0, (267)
ds2 ds ds
R p
while the latter are found by minimising the proper length γ
−gµν dxµ dxν , and are
given by   α β
d2 xµ µ dx dx
2
+ = 0. (268)
ds αβ ds ds
Spin-less test particles and gauge-fields (e.g. photons) do not feel the torsion and follow
the extremals. However, spinning test particles do feel the torsion and obey analogues
96
of the Mathisson-Papapetrou equations [868, 1024]. The notions of autoparallel and ex-
tremal curves coincide if and only if the torsion is totally antisymmetric. The reader is
referred to [599, 1214, 8] for further discussion.

The action and field equations

The action for the ECSK theory is the same as the one for the metric-affine gravity
described in Eq. (46), with the additional assumption that the connection is metric
compatible. To make sure that no inconsistency arises one has to impose Qµαβ = 0 with
a Lagrange multiplier. Variation with respect to Γµαβ then proceeds in the usual way.
We can follow a different approach, however, and assume from the start that Qµαβ =
0, and that the connection is given by Eq. (266). We can then vary the action, Eq. (46),
without the Lagrange constraint, but instead take as independent variables the metric
and the contorsion. Variation with respect to the inverse metric at constant contorsion
then gives h i
Γ α
G(µν) + ∇α P(µν) − P α(µν) = 8πGTµν , (269)
δLm 

where the matter stress-energy tensor is given by Tµν = − 2√1−g δg µν
K
. Variation with
respect to the contorsion gives

Pµ αβ = 8πGτµ αβ , (270)
δLm
where τµ αβ = − √1−g δK µ is the spin angular momentum tensor of matter. Due to
αβ
the anti-symmetry of the contorsion in the first and third indices we also have that
ταµβ = −τβµα . Note, however, that the metricity assumption also means that the Palatini
tensor simplifies to
Pαµβ = Sµαβ + 2gµ[α Sβ] , (271)
which is also antisymmetric in the first and third indices. Hence no inconsistency arises
from Eq. (270), as it does in the general metric-affine case. Now, Eq. (269) only
determines the symmetric part of the Einstein tensor, but we also need the anti-symmetric
part. Using Eqs. (271) and (38) gives the anti-symmetric part of the Einstein tensor as
∗
G[µν] = R[µν] = ∇α Pµ αν , (272)
∗
where ∇µ = Γ ∇µ + 2Sµ . Equations (269), (270) and (272) form a consistent set of field
equations for the ECSK theory.
We can proceed one further step, however, and consider the torsion rather than the
contorsion as the 2nd independent variable. This leads to the definition of the spin-energy
δLm
potential, µν αβ ≡ − √1−g δS µ . Due to the antisymmetry of the torsion in the last two
αβ
indices, the spin-energy potential also obeys µναβ = −µνβα . It is straightforward to
show that the spin-energy potential and the spin angular momentum tensors are related
to each other by
ταµβ = −µ[αβ]µ (273)
and
µµαβ = τµαβ − τβµα + ταβµ . (274)
97
Carrying out the variation this way defines a new stress-energy tensor, σµν , at constant
δLm 
S µαβ by σµν ≡ − 2√1−g δg µν
S
. This is related to Tµν by
h i
σµν = Tµν − 2 Sβα(ν τµ)αβ + Sαβ(ν τµ)αβ − τ βα(µ Sν)αβ − τ α(µβ Sν)βα . (275)

The field equations obtained from varying the action with respect to the metric are
Γ
 
G(µν) + ∇α Pµν α + Pνµ α
h i
+2 P βα(µ Sν)αβ + P α(µβ Sν)βα − Sβα(ν Pµ)αβ − Sαβ(ν Pµ)αβ = 8πGσµν . (276)

After some algebra, and using Eq. (272) to form the full asymmetric Einstein tensor, we
find that the Einstein equations simplify to
Γ
Gµν = 8πGΣµν , (277)

where
∗
Σµν = σµν + ∇α µµαν (278)
is a new stress-energy tensor. This new stress-energy tensor has a very important inter-
pretation [601]: It is none other than the canonical stress-energy tensor. Thus, in the
ECSK theory the usual symmetrisation procedure of the canonical stress-energy tensor
to obtain the stress-energy tensor that enters the Einstein equations is not necessary.
Equations (277) and (270) form a consistent set of field equations that determine the
geometry of the space-time from the matter stress-energy and spin distribution. They
are supplemented by the conservation laws
∗
∇ν Σµν = 2Σαβ S αµβ − τα βλ Rαµβλ (279)

and
∗
∇α τµ αν = Σ[µν] . (280)
Let us now discuss a further property of the ECSK theory. By inspecting of the field
equations (277) and (270) we notice that there are no derivatives of the torsion appearing
anywhere. Thus, the torsion in the ECSK theory is non-dynamical. Its presence is
directly given in terms of the spin-angular momentum of matter by Eq. (270). This
means that it completely vanishes in vacuum, or in cases where matter does not couple
to contorsion (e.g. scalar and gauge fields). Since S αµν is algebraically determined one
can eliminate it from all of the field equations. The final form of the field equations after
eliminating torsion is then found to be

α β
Gµν = 8πGTµν + (8πG)2 2τβα(µ τ ν) − 2τα τ α(µν) − ταµβ τ αν β

1 
+ 2τ µαβ ταµβ + τ αµβ ταµβ − 2τα τ α gµν , (281)
2

where Gµν is the Einstein tensor of gµν , obtained in the usual way from the Levi-Civita
connection, and τµ = τ ββµ .
98
Consequences of the ECSK theory

We introduced the ECSK theory as a minimal modification of GR through the in-

troduction of torsion and the application of the Palatini procedure. As was shown by
Kibble [691], and then later by Hehl et al. [601], the ECSK theory emerges as the mini-
mal description of space-time when one constructs a local gauge theory of the Poincaré
group. There are, however, a number of unsatisfactory features of this interpretation.
Quantisation of the theory fares no better than in GR. It is still non-renormalizable (the
spin contact interactions are reminiscent of 4-fermion interaction terms in quantum field
theory), and the torsion has vanishing canonical momentum, which makes it hard to
construct a quantum description. The reader is referred to [601] for more details.
One may then ask when is the ECSK theory different from GR? By inspection, Eq.
(281) tells us that the spin-potential modifications enter with an additional power of the
Planck mass. Thus, we only expect the ECSK theory to deviate from GR at very high
spin-densities of matter. For electrons the critical mass density such that spin effects are
important is ∼ 1038 kg m−3 while for neutrons it is ∼ 1045 kg m−3 [602]. Conditions for
such high densities can exist in the early universe.
Cosmology with spin and torsion has been studied in the hope that the cosmo-
logical singularity may be avoided. However, it appears that only under very unre-
alistic circumstances can one avoid the cosmological singularity in the ECSK theory
(see [601, 686, 1195, 757]).

3.5.2. Scalar-Tensor-Vector Theory

As a matter of completeness, we will briefly present the Scalar-Tensor-Vector theory
of gravity (STVG) proposed in [905]. STVG is a theory that contains a vector field, φµ ,
three scalar fields, G, ω and µ, and a metric, gαβ . A key characteristic of this theory is
that the modified acceleration law for weak gravitational fields has a repulsive Yukawa
force, as well as the normal Newtonian force. The action for STVG takes the form

S = SGrav + Sφ + SS + SM , (282)

where
Z  
1 √ 1
SGrav = d4 x −g (R + 2Λ) , (283)
16π G
Z
√ 1
Sφ = d4 x −g[ω( B µν Bµν + V (φ))], (284)
4
Z   
4 √ 1 1 µν ~ ~
SS = = d x −g g ∇µ G∇ν G − V (G)
G3 2
Z   
4 √ 1 1 µν ~ ~
d x −g g ∇µ ω ∇ν ω − V (ω)
G 2
Z   
4 √ 1 1 µν ~ ~
d x −g 2 g ∇µ µ∇ν µ − V (µ) , (285)
µ G 2

and where Bµν = ∂µ φν − ∂ν φµ .

99
 This action has been studied in [905] where the full field equations, the weak field limit,
and cosmological solutions are presented. As would be expected from the complexity of
the action, it is difficult to completely solve this system in the detail required to make
accurate predictions. Hence, the author of [905] has posited a certain number of scaling
relations that translate into spatial dependences for G, ω and µ. It is argued in [905]
that the classical action for this theory could be considered as an effective field theory for
a renormalisation-group-flow-quantum-gravity scenario. The reader is referred to [905]
for further details.

100
4. Higher Derivative and Non-Local Theories of Gravity

Recall from Section 2.4.1 that General Relativity represents the most general the-
ory describing a single metric that in four dimensions has field equations with at most
second-order derivatives [831, 832]. One way to extend GR is therefore to permit the
field equations to be higher than second order. Indeed, such a generalisation might be
considered desirable as it will cause the graviton propagator to fall off more quickly in the
UV, thereby improving the renormalisability properties. Modifying gravity in this way,
however, also has a number of drawbacks. In particular, it can introduce instabilities
into the theory, such as ghost-like degrees of freedom (see Sections 2.1.3, 4.1.5, and 4.2.5,
and [1281] for an overview).
In this section we consider those gravity theories that are higher than second-order in
derivatives. Such theories can have interesting phenomenology, and in many cases can be
shown (or, at least, argued to be) less susceptible to instabilities than one may have ini-
tially suspected. For example, if the higher derivatives act only on what would otherwise
be non-dynamical modes, then they may simply render them dynamical, rather than au-
tomatically generating a ghost. This is what happens in f (R) gravity, where the higher
order derivatives act on the conformal mode that does not propagate in GR. In Hořava-
Lifschitz gravity, as another example, one allows for higher-order spatial derivatives, as
opposed to higher time derivatives, in order to guard against ghost-like instabilities. In
both of these examples the theory can deviate considerably from GR, while still main-
taining some basic stability properties.
This section also includes galileon and ghost condensate theories. Strictly speaking
these are not higher-derivative theories since their field equations are at most second
order in derivatives. In fact, the galileon theory in particular is constructed with this
in mind. Nevertheless we include them in this section as both theories contain non-
trivial derivative interaction terms. We will not discuss theories with infinite derivatives,
as occur in string field theory, or p-adic string theory (see [904] for discussion of such
theories).

4.1. f (R) Theories

Fourth-order theories of gravity have a long history, dating back to as early as 1918
[1269], only a few years after the first papers on General Relativity by Einstein. These
theories generalise the Einstein-Hilbert action by adding additional scalar curvature in-
variants to the action, or by making the action a more general function of the Ricci scalar
then the simple linear one that leads to Einstein’s equations. Here we consider the latter
of these options, a choice that leads by Lovelock’s theorem to fourth-order field equa-
tions for anything except the addition of a constant term to the gravitational Lagrangian.
Such theories, generically referred to as f (R) theories of gravity, have been intensively
studied, and have a number of reviews dedicated to them [969, 1118, 1167, 374, 978].
This interest was stimulated in the 1960s, 70s and 80s by the revelations that the quan-
tisation of matter fields in an unquantised space-time can lead to such theories [1239],
that f (R) theories of gravity can have improved renormalisation properties [1186], and
that they can lead to a period of accelerating expansion early in the Universe’s history
[1179]. More recently they have been of considerable interest as a possible explanation
for the observed late-time accelerating expansion of the Universe.

101
4.1.1. Action, field equations and transformations
The f (R) generalisations of Einstein’s equations are derived from a Lagrangian den-
sity of the form √
L = −gf (R), (286)
√
where the factor of −g is included, as usual, in order to have the proper weight. This
is clearly about as simple a generalisation of the Einstein-Hilbert density as one could
possibly conceive of. The field equations derived from such an action are automatically
generally covariant and Lorentz invariant for exactly the same reasons that Einstein’s
equations are. Unlike the Einstein-Hilbert term, however, the field equations that one
obtains from the least action principle associated with Eq. (286) depend on the varia-
tional principle that one adopts. Different possibilities are the ‘metric variation’ where
the connection is assumed to be the Levi-Civita one, the ‘Palatini approach’ in which
Eq. (286) is varied with respect to the metric and connection independently, and the
‘metric-affine’ approach in which the same process occurs but the matter action is now
taken to be a functional of the connection as well as the metric. In this section we will
mostly be concerned with the metric variational approach, although we will also outline
how the other procedures work below.

Metric variational approach

Let us now derive the field equations in the metric variational approach. Integrating
Eq. (286) over a 4-volume, including a matter term and varying with respect to gµν
yields Z
√ h1 χ i
δS = dΩ −g f g µν δgµν + fR δR + T µν δgµν , (287)
2 2
where fR means the functional derivative of f with respect to R, χ is a constant, and Tµν
is the energy-momentum tensor defined by a variation of the matter action with respect
to gµν in the usual way. Assuming the connection is the Levi-Civita one we can then
write
fR δR ' −[fR Rµν + fR;ρσ (g µν g ρσ − g µρ g νσ )]δgµν , (288)
where ' is used here to mean equal up to surface terms [568]. Looking for a stationary
point of the action, by setting the first variation to zero, then gives
1 χ
fR Rµν − f gµν − fR;µν + gµν fR = Tµν . (289)
2 2
These are the f (R) field equations with which we will be primarily concerned in this
section. It can be seen that for the special case f = R the LHS of Eq. (289) reduces to
the Einstein tensor, and the field equations are second-order in derivatives of the metric.
For all other cases, except an additional constant, the equations in (289) are fourth-order
in derivatives.

Conformal transformation in the metric variational approach

As with scalar-tensor theories, the f (R) theories of gravity derived from the metric
variational approach can be conformally transformed into a frame in which the field
102
equations become those of General Relativity, with a minimally coupled scalar field. This
is sometimes referred to as ‘Bicknell’s theorem’ in the case of f (R) theories, particularly
when the minimally coupled scalar field is in a quadratic potential [155]. In the general
case we consider conformal transformations of the form [101, 853]

ḡµν = fR gµν , (290)

together with the definition r

3
φ≡ ln fR , (291)
χ
which allows the field equations from the metric variational principle, Eqs. (289), to be
transformed into
 
1 χ 1 χ
R̄µν − ḡµν R̄ = φ,µ φ,ν − ḡµν ḡ ρσ φ,ρ φ,σ − ḡµν V + T̄µν . (292)
2 2 2 2

Here T̄µν is a non-conserved energy-momentum tensor, and we have defined

(RfR − f )
V = V (φ) ≡ . (293)
χfR2
Theories derived from an action of the form (286) can therefore always be conformally
transformed into General Relativity with a massless scalar field (as long as fR 6= 0), and
a non-metric coupling to the matter fields.

Legendre transformation in the metric variational approach

As well as conformal transformations, one can also perform Legendre transformations

on f (R) theories in the metric variational approach. Such transformations allow the field
equations of f (R) to take the form of a scalar-tensor theory (albeit it a slightly strange
one). These transformed theories maintain the universal metric coupling of the matter
fields, unlike the case of conformal transformations.
The first step here is to notice that the Eq. (286) can be written in the equivalent
form √
L = −g [f (χ) + f 0 (χ)(R − χ)] , (294)
where χ is a new field, and the prime denotes differentiation. Variation with respect to
χ then gives
f 00 (χ)(R − χ) = 0, (295)
so that χ = R for all f 00 (χ) 6= 0. Substitution of this result back into Eq. (294) then
immediately recovers Eq. (286), showing that the two Lagrangian densities are indeed
equivalent. What is more, the special case f 00 (χ) = 0 can be seen to correspond to the
Einstein-Hilbert action.
Now, if we make the definition
φ ≡ f 0 (χ), (296)
and assume that φ(χ) is an invertible function, then we can define a potential
1
Λ(φ) ≡ [χ(φ)φ − f (χ(φ))] . (297)
2
103
In terms of this new scalar field we can then write the density (294) as
√
L = −g [φR − 2Λ(φ)] , (298)

which is clearly just a scalar-tensor theory, as specified in Eq. (97), with vanishing cou-
pling constant, ω = 0. As we have not transformed the metric, the coupling of this field
to any matter fields that are present remains unchanged.

The Palatini procedure

Starting again from the density (286) we can also proceed in a entirely different way
to the metric variational approach just described. Instead of assuming the connection
from which the Ricci scalar is constructed is the Levi-Civita one, we could instead treat
the metric and connection as independent fields. For the case of General Relativity a
variation with respect to the connection then simply results in the connection being
shown to be the Levi-Civita one, so that the difference between the metric variational
approach and the Palatini approach is a semantic one. For the case of f (R) theories,
however, the Palatini approach leads to an entirely different set of field equations.
Starting with an integral of Eq. (286) over some 4-volume, and extremising with
respect to gµν now gives
1 χ
fR Rµν − gµν f = Tµν , (299)
2 2
where Tµν is once again the usual energy-momentum tensor. In this expression Rµν
is now defined independently from the metric, and R should be taken to be given by
g µν Rµν . The next step is the variation of Eq. (286) with respect to Γµνσ , which results
in √

−gg µν fR ;σ = 0, (300)
where the covariant derivative here should be understood to be taken with respect to
Γµνσ , which is not the Levi-Civita connection unless fR =constant (as is the case in
GR). It is remarkable that the field equations (299) do not involve any derivatives of
the metric, and only first derivatives of the connection. These are a different set of field
equations to Eq. (289), and should be considered a different set of theories to the f (R)
in which R is a priori taken to be constructed from the Levi-Civita connection.
It can be noted from Eq. (300) that even if the connection is not compatible with the
metric gµν , we can still define a new metric, ḡµν = fR gµν , with which it is compatible.
Rewriting the full field equations under this conformal transformation we see that we
recover General Relativity with a minimally coupled scalar field in a potential, but no
kinetic term: √  
L = −ḡ R̄ − 2V (φ) , (301)
where φ ≡ fR and V (φ) = (R(φ)φ − f (φ))/2φ2 . Here R(φ) and f (φ) are given by
inverting the definition of φ, and R̄ should be understood to be constructed from the
metric connection compatible with ḡµν . Transforming back to the original conformal
frame this theory then can be shown to be equivalent to a scalar-tensor theory with
ω = −3/2 and Λ = (RfR − f )/2 [999].
While being an interesting variant on the metric variational incarnation of the f (R)
theories, there are a number of very severe problems in proceeding with the Palatini
104
procedure in this way. Not least of these is the apparent ill-posed nature of the Cauchy
problem in the presence of most matter fields, which is discussed in [766]. Without a
well-posed initial value problem f (R) gravity in the Palatini formalism lacks predictive
power, and hence is not a good candidate for a viable theory of gravity. Furthermore, the
Palatini approach to f (R) gravity also appears to introduce problematic strong couplings
between gravity and matter fields at low energies [509, 640], and singularities in systems
that are usually well described by weak fields [87, 88, 89]. For these reasons we will focus
on f (R) theories in the metric variational approach in the sections that follow.
For further details of the Palatini approach to f (R) gravity, and the results that
follow from it, the reader is referred to [1167]. For studies of weak field gravity in the
Palatini formalism the reader is referred to [92, 429, 881, 999, 30, 1000, 1163, 1001, 1078,
224, 665, 31, 1077], and for cosmology to [1251, 509, 883, 28, 882, 884, 29, 1162, 1164,
724, 34, 727, 258, 793, 1232, 725, 781, 782, 497]. An interesting class of theories that
interpolate between the Palatini approach to f (R) theories and the metric variational
approach to f (R) theories is investigated in [39, 726].

The metric-affine approach

One further approach to the f (R) theories of gravity is the ‘metric-affine’ formula-
tion. Here one again considers the metric and connection to be independent, as in the
Palatini procedure, but now allows the matter action to be a function of both metric and
connection (rather than metric alone, as is the case in Palatini and metric variational
formalisms). The relevant action for the theory then becomes [1168]
Z
√
S = dΩ −gf (R) + Sm (gµν , Γµνσ , Ψ), (302)

where R = g µν Rµν , and Rµν is taken to be a function of the connection only, as in the
Palatini procedure. One can therefore think of the action (302) as a generalisation of
the Palatini action, which is recovered when the dependence of the matter action, Sm ,
on Γµνσ vanishes.
As is the case in General Relativity, the invariance of the Ricci scalar under the
projective transformation Γµνσ → Γµνσ + λν δ µσ can lead to inconsistency of the field
equations, as matter fields do not generically exhibit an invariance of this type. This
invariance can beR cured by adding to the action an additional Lagrange multiplier term
√
of the form S = dΩ −gB µ Λν [νσ] , and results in the field equations
1 χ
fR Rµν − gµν f = Tµν , (303)
2 2
together with Γµ[µν] = 0, and
1 h √
√
i
√ −gfR g µσ ;σ δ νρ − −gfR g µν ;ρ + 2fR g µσ Γν [σρ]
−g
 
χ 2 µ]
= ∆ρµν − ∆σσ[ν δ ρ , (304)
2 3
√ [νρ]
where ∆µνρ ≡ −(2/ −g)δSm /δΓµνρ . It can then be shown that ∆µ = 0 corresponds
(νρ)
to a vanishing of the torsion, and ∆µ 6= 0 introduces non-zero non-metricity [1168].
105
The metric-affine approach has not been studied as intensively as the other approaches to
f (R) gravity that we have already discussed, and will not feature heavily in the sections
that follow.

4.1.2. Weak-field limit

The weak field limit of f (R) theories of gravity has been extensively studied in the
literature. Here we will first consider perturbations about Minkowski space, and then
perturbations around de Sitter space, time dependent backgrounds, and inhomogeneous
backgrounds.

Perturbations about Minkowski space

The first attempt at finding the Newtonian limit of an f (R) theory appears to have
been performed by Pechlaner and Sexl for the case of f = R + αR2 [1029]. The first step
here is to write down the perturbed line-element as

ds2 = −(1 + 2ψ)dt2 + (1 − 2φ)(dx2 + dy 2 + dz 2 ), (305)

neglecting time derivatives and second order terms in φ and ψ one then finds that the
Ricci scalar, R = −2∆ψ + 4∆φ, obeys
χ
6αδR − R = − ρ, (306)
2
and that the potential ψ obeys
χ
(1 + 6α∆)∆ψ = (1 + 8α∆)ρ, (307)
4
where ∆ is the Laplacian on Euclidean 3-space. The derivatives of ρ in this last equation
occur due to replacing terms containing φ with those obtained from taking derivatives
of Eq. (306), and do not occur in the actual field equations themselves.
Inserting a delta function source, ρ = mδ, and integrating Eq. (307), using the
solution to Eq. (306) to find φ, one then gets the solutions [1029]
 
χm e−m0 r
16πψ = − 1+ , (308)
r 3
 
χm e−m0 r
16πφ = − 1− , (309)
r 3

where boundary conditions at infinity have been imposed to eliminate exponentially

increasing modes, and where we have defined the mass
1
m20 ≡ . (310)
6α
Mass terms similar to this continue to exist for more general theories, and more general
situations, as we will outline shortly. One can see straight away that for large masses, with
m0  r−1 , the Yukawa potentials in Eqs. (308) and (309) are exponentially suppressed,
and we recover ψ = φ ∝ −m/r. In the limit of small masses, however, when m0  r−1 ,
106
we instead find that ψ = 2φ ∝ −m/r. For the case of f = R + αR2 gravity we therefore
already expect the PPN parameter γ to be 1 when the mass of the scalar degree of
freedom is large, and 1/2 when it is small.
One is, of course, also interested in other functions of f (R), and can consider the case
of analytic f (R) theories that can be expanded as
∞
X
f (R) = ci Ri , (311)
i=1

where the ci are constants. To full post-Newtonian order the weak field solution in the
presence of a perfect fluid is then given in full generality as [307]

2 2 c2 16c2 7
g00 = − 1 + (U + c2 R) − 2 U 2 + 2 22 R2 − UR − V(U R)
c1 c1 c1 3c21 18πc1
3c2 64 44c2 40c2 ~ ~ )
+ V(R2 ) + 2 V(ρU ) − 2 V(ρR) − V(∇ρ · ∇U
4πc1 9c1 3c1 3c31
40c2 ~ ~ 2 4 6 1
+ 3 2 V(∇ρ · ∇R) + V(ρΠ) + V(ρv 2 ) + V(P ) + X(U R)
c1 c1 c1 c1 6πc1
 
1 c2 c3 4 8c2
− − X(R2 ) − 2 X(ρU ) + 2 X(ρR)
4π c1 2c2 3c1 3c1
2
8c2 ~ ) − 8c2 X(∇ρ
~ · ∇U ~ · ∇R)
~ 2 2
+ 3 X(∇ρ − X(P ) + X(ρΠ) (312)
3c1 c31 c1 3c1
7Vi Wi X(ρvi ) Yi Zi
g0i = − − + − − √ (313)
2c1 2c1 6c1 6c1 6 6c1 c2
 
2
gij = 1 + (U − c2 R) δij (314)
c1

where U is the usual Newtonian potential, and the other potentials are defined as
R ρ(x0 )vi (x0 ) 3 0 R 0 0
)·(x−x0 ))(x−x0 )i 3 0
Vi ≡ |x−x0 | √d x Wi ≡ ρ(x )(v(x |x−x 0 |3 d x
c1
R |x−x0 | R ρ0 v0 ·(x−x0 )(x−x0 )i − 6cc1 |x−x0 | 3 0
q
−
Q0 e 6c2
X(Q) ≡ |x−x0 | d3 x0 Yi ≡ |x−x0 |3 e 2 d x
R Q0 R ρ0 v0 ·(x−x0 )(x−x0 )i − 6cc1 |x−x0 | 3 0
q
03
V(Q) ≡ |x−x 0 | dx Zi ≡ |x−x0 |2 q e 2 d x
R R
q
c c
− 6c1 |x−x0 | 3 0 0
ρ(x ) − 1 |x−x0 |
χ̂ ≡ ρ0 e 2 d x R = 3c12 |x−x 0| e
6c2
d3 x0 ,

where R is the Ricci scalar, P is pressure, ρ is the rest-mass energy density, and Π is
the specific energy density (as defined in [1274]). The terms in Eqs. (313)-(314) that are
functionals of derivatives of ρ, U and R can be recast into a form where such derivatives
do not appear by further manipulations [307].
From the above it can be seen that the results of General Relativity are recovered,
to the appropriate order, when c2 → c3 → 0. For non-vanishing values of the these
constants a large number of extra Yukawa potentials are present, and for large enough
values of c2 we can again see that γ → 1/2, as the scalar degree of freedom becomes
massless.
107
 The study of gravitational waves about a Minkowski space background in f (R) theo-
ries has been undertaken by Berry and Gair in [146]. Here the authors consider analytic
functions of the type prescribed by Eq. (311), and find that an extra mode of oscil-
lation is possible. The gravitational waves emitted by extreme-mass-ratio inspirals are
then calculated, and the authors conclude that current laboratory bounds (that result in
|c2 /c1 | < 10−9 m2 ) mean that the extra oscillatory mode they find cannot be excited by
astrophysical sources.

Perturbations about de Sitter backgrounds

As well as the usual expansions about Minkowski space, in order to determine the
post-Newtonian behaviour of a theory, one is also interested in perturbations about other
backgrounds. Here we will consider de Sitter space as a background. This is not meant to
be an elucidation of cosmological perturbation theory, but rather a consideration of weak
field expansions as applicable to systems such as the solar system and binary pulsars.
Minkowski space is not always a stable solution of f (R) theories of gravity that attempt
to produce self-accelerating cosmologies at late-times, and in these cases time-dependent
backgrounds, and de Sitter space in particular, become of increasing interest for weak
field studies.
Much work has been performed on establishing the weak field limit of f (R) theories
about a de Sitter background, as relevant for theories that try and account for late-
time accelerating expansion without dark energy, see e.g. [1050, 421, 1171, 943, 998,
1129, 1300, 930]. The majority of these studies conclude that, in the absence of extra
mechanisms to mask such behaviour, one should expect to find γ = 1/2. This was shown
in an early paper on the subject in [292], and is the familiar limit of theories in which
a scalar degree of freedom has low mass. Here we will briefly sketch out the method by
which such a result is found for more general f (R), following the approach of [294].
The first step here is to identify a de Sitter solution with constant Ricci curvature
R = R0 = 12H02 . The line-element is then perturbed as

ds2 = −(1 + 2ψ − H02 r2 )dt2 + (1 − 2φ + H02 r2 )dr2 + r2 dΩ2 , (315)

where we have chosen to present the de Sitter background using a static coordinate patch.
We then proceed by perturbing the Ricci scalar as

R = R0 + R1 , (316)

where R1  R0 . The perturbative expansion then linearises all field equations with
respect to φ, ψ, R1 and H02 r2 , and their derivatives, while neglecting all time derivatives.
The case of spherical symmetry is considered for simplicity.
To lowest order the trace of the field equations is now
 
fR − fRR R0 χρ
∆R1 − R1 = − , (317)
3fRR 6fRR

where fR and fRR should be understood to be the value of these quantities at R = R0

(implicit here is an assumption that these quantities are all of the same order of magnitude
as R0 , and that R0 in the weak-field systems under consideration takes the same value as
108
in the cosmological background solution). From cosmological considerations the second
term on the left-hand side of Eq. (317) is then neglected, as the factor in brackets
corresponds to the mass squared of the scalar degree of freedom, which must be small
compared to r−1 in order to have late-time accelerating expansion. The resulting form
of R1 is therefore found to be
χm
R1 ' , (318)
24πfRR r
where m is the mass of the object at the centre of symmetry. Applying the same approx-
imations to the (t, t) component of the field equations results in

χρ R1 fRR 2
∆ψ = − + ∇ R1 , (319)
2fR 2 fR
which on substitution of the expression for R1 gives to lowest order
χm
ψ'− . (320)
12πfR r
The remaining field equations then give φ, to the same order of approximation, as
χm ψ
φ'− ' . (321)
24πfR r 2
This calculation has not been performed in the PPN gauge, which uses an isotropic
spatial coordinate system, but nevertheless one can verify that when interpreted within
the standard PPN framework it does indeed give [1001]
1
γ= . (322)
2
In this section we have discussed de Sitter space as a background to perturb around.
However, establishing whether de Sitter space is, in fact, a stable asymptotic solution of
f (R) theories, and establishing the genericity of initial conditions that lead to de Sitter
space at late times, has not yet been discussed. We will consider this subject in following
sections.

Perturbations about other backgrounds

Having considered the maximally symmetric Minkowski space and de Sitter space
backgrounds, we can now also consider less symmetric spaces to perturb around. This
enterprise is hindered by our ability to find less symmetric solutions to the field equations
(289). We can, however, make progress with some simple cases.
If we consider f (R) theories in which f = R1+δ then one can find exact non-static,
homogeneous and isotropic vacuum solutions [241]. Such solutions can be shown to be,
under certain conditions, stable asymptotic attractors for the general class of spatially
flat, vacuum FLRW solutions [248]. They will be discussed further in the cosmology sec-
tion that follows. For this same class of f (R) theories exact static, spherically symmetric
vacuum solutions are also known [310], which can also be seen to be generic asymp-
totes of the general solution, with the specified symmetries applied [310]. We are now in
109
possession of two exact solutions, for the same theories, which have less than maximal
symmetry, and which can be used as backgrounds to perturb around.
Spherically symmetric, time independent perturbations around the homogeneous,
time-dependent background found in [241] are given to linear order by [304]

ds2 = −(1 + 2ψ)dt2 + a2 (t)(1 − 2φ)(dr2 + r2 dΩ2 ),

δ(1+2δ)
where a(t) = t (1−δ) . The perturbations φ(r) and ψ(r) are then found to be

c1 2c2 (1 − 6δ + 4δ 2 + 4δ 3 ) 2
ψ = − + r
r (5 − 14δ − 12δ 2 )
(1 − 2δ)c1
φ = − − δc2 r2 .
r
The corresponding perturbative analysis about the static, spherically symmetric back-
ground found in [310] gives
(1+2δ)
ds2 = −r2δ (1−δ) (1 + V (r))dt2
(1 − 2δ + 4δ 2 )(1 − 2δ − 2δ 2 )
+ (1 + W (r))dr2 + r2 dΩ2 ,
(1 − δ)2

where V (r) and W (r) are given in full generality by V (r) = c3 V1 (r) + c4 V2 (r) + c5 V3 (r)
and W (r) = −c3 V1 (r) + c4 W2 (r) + c5 W3 (r), where
(1−2δ+4δ 2 )
V1 = −r− (1−δ) (323)

and where V2 , V3 , W2 and W3 are oscillatory modes [310]. It can immediately be seen
that the form of the linearised perturbations around these two backgrounds are quite
different to each other, even though the field equations they obey are identical. One can
verify that an observer in the homogeneous, time-dependent background should measure

β=1 and γ = 1 − 2δ,

which gives δ = −1.1 ± 1.2 × 10−5 when the constraint derived from the Cassini space
probe on γ is applied [147]. On the other hand an observer in the static background
should measure an anomalous extra gravitational force that goes like [310]

δ
F ∼− .
r
When subjected to constraints from observations of the perihelion precession of Mercury,
the presence of this extra force gives δ = 2.7±4.5×10−19 [310]. The different forms of the
gravitational potentials and forces in these examples show that the choice of symmetries
of the background space-time can have important consequences for its weak field limit,
and the constraints on the underlying theory that are derived from it.

Chameleon mechanism

110
 As with a variety of other modified theories of gravity, the ‘Chameleon Mechanism’
has been applied to f (R) theories. This mechanism was outlined in Section 3.1.4, where
a summary of some of the accumulated literature on it was outlined. Here we will simply
reiterate the basic point that this mechanism potentially allows a means by which theories
with a light effective scalar degree of freedom can evade solar system and binary pulsar
tests of the PPN parameter γ by allowing the scalar to acquire a higher mass in the
locale of high mass concentrations, such as the Sun and Earth.
The chameleon mechanism has been applied to specific f (R) theories, and its be-
haviour in this application considered further, in e.g. [267, 945, 1180, 495, 193, 1199,
244, 629]. As with other applications of the Chameleon mechanism, if a ‘thin shell’ is
present then the mass of the scalar degree of freedom in these theories is thought to be
able to be supressed enough to satisfy solar system tests of gravity.

4.1.3. Exact solutions, and general behaviour

Having discussed the weak field solutions which are of interest for inferring constraints
on f (R) theories from observations of gravitational phenomena in the solar system and
binary pulsar systems, let us now consider the behaviour of solutions to the full non-
linear field equations. Here we will be concerned with exact solutions, which can be
obtained in some simple cases, as well as what can be inferred about the general be-
haviour of non-linear solutions by other methods, and what theorems can tell us about
the behaviours that are possible. The relatively simple structure of f (R) theories make
such considerations a feasible proposition. The geodesic deviation equation in general
f (R) theories has been studied in [569].

Isolated masses, and black holes

Progress was made into understanding the static spherically symmetric vacuum so-
lutions of f (R) theories of gravity by Mignemi and Wiltshire in [891]. These authors
consider theories with higher powers of the Ricci scalar added to the Einstein-Hilbert
Lagrangian, and use a dynamical systems analysis to determine the behaviour of the gen-
eral solutions with the specified symmetries. They find the asymptotes of these solutions,
for a variety of different cases, and show that the only static spherically symmetric solu-
tions of the theories they consider that have regular horizons are the Schwarzschild solu-
tions. They further find that by dropping the requirement of regularity the Schwarzschild
solution is also the only solution to these theories that is asymptotically flat.
The black hole ‘no-hair’ theorems have been considered in the context of f (R) =
R + αR2 theories by Whitt [1270]. Collapse to a black hole, however, has not been as
extensively studied in f (R) theories of gravity as it has in Brans-Dicke theory, where
direct numerical calculations have been performed [1107]. Nevertheless, the same logic
that tells us that the vacuum black hole solutions of general relativity are the only vacuum
black hole solution of Brans-Dicke theory that can result from gravitational collapse, also
suggest that this should be true for f (R) gravity. In particular, most of the results of
Hawking on this subject only rely on inequalities of the form
Rµν lµ lν ≥ 0, (324)
where lµ is a null vector, and not on the details of the field equations themselves [595].
This null energy condition is true of the conformally transformed scalar fields equations
111
in Brans-Dicke theory, and is also true in f (R) gravity. It therefore seems reasonable
to expect that the vacuum black hole solutions of f (R) gravity should also be the same
as the vacuum black hole solutions of General Relativity. The subject of black hole
radiation in the context of f (R) gravity has been studied in [200, 322, 16, 543, 378] and
[489] where it was shown that black holes in f (R) gravity have a entropy given by

fR A
S= . (325)
4
The subject of the de Sitter no-hair theorems and isotropisation in f (R) gravity has been
considered by Barrow and Ottewill [110] and Goheer, Leach and Dunsby [540], where
it was shown that flat FLRW isotropic points can exist in the phase plane of Bianchi
solutions, and that de Sitter space can be a stable asymptote of f (R) theories of gravity.
One should, however, be aware that such behaviour depends on the theory in question,
and the initial conditions. For example, for theories with negative powers of R in a
series expansion of their Lagrangian one may generically expect such terms to become
important asymptotically. In this case accelerating power-law expansion is an attractor
solution [306]. This will be discussed further in the cosmology section below.
As well as the black hole solutions of General Relativity, it is known that other
vacuum solutions to f (R) theories of gravity that can describe isolated masses also exist.
Due to the complicated nature of the field equations in these theories, however, only
a few exact solutions that describe these situations have been found. For the case of
f (R) = R1+δ solutions are known that correspond to an isolated mass in a homogeneous
and time dependent background, and an isolated mass in a static, spherically symmetric
background. The former of these solutions is given by the line-element [304]

ds2 = −A1 (r)dt2 + a2 (t)B1 (r)(dr2 + r2 dΩ2 ), (326)

(1+2δ)
where a(t) = tδ (1−δ) , and where A1 (r) and B1 (r) are given by
! q2  4
C1
1− C1
A1 (r) = r
C1
, and B1 (r) = 1+ A(r)q+2δ−1 ,
1+ r
r

where q 2 = 1 − 2δ + 4δ 2 . The latter solution is given by [310]

dr2
ds2 = −A2 (r)dt2 + + r2 dΩ2 (327)
B2 (r)

where
(1+2δ) C2
A2 (r) = r2δ (1−δ) + ,
(1−4δ)
r (1−δ)
!
(1 − δ)2 C2
B2 (r) = 1+ (1−2δ+4δ 2 )
.
(1 − 2δ + 4δ 2 )(1 − 2δ(1 + δ))
r (1−δ)

The constants C1 and C2 appear in these solutions as mass parameters, and it can be
seen that both Eq. (326) and Eq. (327) reduce to the Schwarzschild solution when
112
δ → 0. The problem of static, spherically symmetric solutions in general f (R) has been
considered in [992], where a covariant formalism was developed for studying the problem,
and the non-uniqueness of the Schwarzschild solution was demonstrated. The δ = 1/4
case of Eq. (327) was rediscovered in [243]. Black holes coupled to Yang-Mills
√ fields have
been studied in [872], where an exact solution was found for the case f = R.
These solutions are interesting for a number of reasons. Firstly, they show that the
generalisation of the solutions of General Relativity to other theories of gravity is not
always unique; i.e. there can be multiple solutions in modified theories of gravity that
reduce to the same solution in the limit of general relativity. It may therefore be the
case that one needs to understand the symmetries of the background space-time to a
greater extent than is necessary in General Relativity, in order to fully understand which
solution should be used to model a given situation. Secondly, Eq. (326) shows explicitly
that Birkhoff’s theorem is not valid in general, when one considers generalisations of Ein-
stein’s theory. Spherically symmetric vacuum solutions of modified theories of gravity can
therefore be time-dependent, which can lead to new phenomenology. Birkhoff’s theorem,
in the context of f (R) gravity, has been considered in [490]. Thirdly, Eq. (327) displays
non-trivial asymptotic behaviour as r → ∞. Such behaviour is unexpected in general
relativity, and again opens the window to new phenomenology. The results of Mignemi
and Wiltshire [891] even suggest that such behaviour is generic. A fourth point is that
the solution given in Eq. (326) has been shown in [488] to exhibit a naked singularity.
This has clear implications for the applicability of the cosmic censorship hypothesis to
modified theories of gravity. The Misner-Sharp energy in spherically symmetric space-
times is considered in [225].

Cosmological solutions

A variety of cosmological solutions in f (R) theories of gravity are known, and have
had their stability analysed. Here we will briefly review and provide references to studies
of these solutions.
The conditions for existence and stability of de Sitter solutions in f (R) gravity appears
to have been first studied in [110]. One can show that for any theory for which there is
a value of R which satisfies
fR (RdS )RdS = 2f (RdS ) (328)
there exists a de Sitter solution with RdS = 4Λ. The stability of de Sitter solutions in
f (R) gravity was studied in [110, 1049, 486]. These solutions are of obvious importance
for cosmology at both early and late times. One can note that with f (R) ∝ R2 Eq. (328)
is satisfied with any value of R. Theories with R2 terms in their Lagrangian’s have been
studied extensively, due to the naturalness of adding an R2 term to the Einstein-Hilbert
action, and due to their improved renormalisation properties [1186]. They were also
introduced and studied by Starobinsky for cosmological purposes, and in particular their
ability to give rise to an early non-singular period of accelerating expansion in a natural
way [1179].
Less symmetric cosmological solutions than de Sitter space can also be found for some
f (R) theories. In particular, theories of the type f (R) = R1+δ are again of interest here,
as they admit simple exact solutions. As mentioned in the preceding section, a power-law

113
exact solution for a spatially flat vacuum FLRW solution is known to be given by [241]
1+2δ
a(t) = tδ 1−δ , (329)
and a spatially flat solution in the presence of a perfect barotropic fluid with equation of
state P = wρ is also known [248]
2(1+δ)
a(t) = t 3(1+w) . (330)
The stability of these solutions, and their properties as asymptotes of the general solution,
have been investigated in [248] and [310]. In fact, it has been shown that these are the
only power-law perfect fluid FLRW solutions that exist for any f (R) gravity theory
[539]. Explicit non-power-law general solutions with FLRW symmetries were found in
[305], both with spatial curvature, and in the presence of a perfect fluid. These solutions
show explicitly that in the early universe both non-singular and inflationary behaviour
are possible. The energy conditions in FLRW solutions have been considered in [1033],
and braneworld cosmology in these theories have been considered in [68].
Beyond exact solutions, FLRW cosmological solutions have also been studied in f (R)
theories of gravity using dynamical systems analysis. This has been done for the case of a
number of particular f (R) theories in [248, 310, 790, 5, 538] and also in the general case in
[390, 306, 40, 253]. The dynamical systems approach has even been applied to perturbed
FLRW solutions in [247]. We will discuss perturbed FLRW further in the Cosmology
section that follows. These studies find a variety of interesting cosmological behaviours
at both early and late-times. In particular non-singular and accelerating behaviour in
the early universe is again identified, as well as late-time accelerating expansion, and the
non-sequential domination of higher powers in the Ricci curvature, for analytic f (R), as
the initial singularity is approached. The conditions required for a non-singular ‘bounce’
are given in [249], and oscillating solutions were considered in [208]. There have also
been some concerns expressed as to whether a matter dominated epoch is generically
expected to exist after radiation domination [242, 972, 40, 43, 479]. The inverse problem
of finding particular forms of f (R) that result in pre-specified cosmological evolutions
has been considered in [240, 929, 242, 376, 973, 496, 251, 980]. Such inversions do
not always specify f (R) uniquely [929], and it has been shown that to reproduce exact
ΛCDM evolution with dust only one is forced towards the Einstein-Hilbert action with
a cosmological constant [443].
Exact Bianchi cosmological solutions were discovered for f (R) = Rn theories in [98],
and been studied further in [778, 540], where shear dynamics and isotropisation are
discussed. The special case of n = 2 was studied in [209]. Bianchi type I and V
solutions have been considered in [1132] and [1133], and Bianchi V IIA solutions in [367].
Kantowski-Sachs solutions have been studied in [786]. Other know exact solutions are
the Einstein static universe [309, 790, 549, 1122], and the Gödel universe [309, 1057].
These studies explore the stability of the Einstein static universe, and the existence of
closed time-like curves in the Gödel solution.

4.1.4. Cosmology
Much of the recent motivation for studying f (R) gravity has come from the need to
explain the apparent late-time accelerating expansion of the Universe. Previous motiva-
tion for studying f (R) gravity has also come from cosmological considerations, including
114
the presence of an initial singularity, and early universe inflationary expansion. We will
therefore now present an overview of what we consider to be some of the most relevant
aspects of f (R) gravity for physical cosmology. In terms of the viability of FLRW ge-
ometry in f (R) gravity, the Ehlers-Geren-Sachs theorem of General Relativity has been
extended to cover these theories by Rippl, van Elst, Tavakol, and Taylor in [1065], and
more recently by Faraoni in [487].

Field Equations

To describe the cosmology, up to scalar perturbations, we first define line-element

ds2 = −(1 + 2Ψ)dt2 + a2 (t)(1 − 2Φ)qij dxi dxj , (331)

and the energy-momentum tensor,

T 00 = −ρ − δρ (332)
T 0i ~ iθ
= −(ρ + P )∇ (333)
T ij = P δ ij + δP δ ij + (ρ + P )Dij Σ, (334)

where θ is the peculiar velocity, δP is the pressure perturbation, and Σ is the anisotropic
stress. At zeroth order the Friedmann equations are
 
1 1 κ
H2 = 8πρ − (f − RF ) − 3H Ḟ − 2 (335)
3F 2 a
1 κ
Ḣ = − (8πρ + 8πP + F̈ − H Ḟ ) + 2 (336)
2F a
where the Ricci scalar is given by R = 6(2H 2 + Ḣ + κ/a2 ), and energy-momentum
conservation gives, as usual,
ρ̇ + 3H(ρ + P ) = 0, (337)
where F = fR , over-dots denote derivatives with respect to t, and κ is spatial curvature.
Now let us consider the first-order scalar perturbation equations, which are given in
[637]. Here it is convenient to define a new quantity

χ ≡ 3HΨ + 3Φ̇. (338)

The perturbation equations are then [637]

!  
Ḟ 3 Ḟ 3 k
χ̇ + 2H + χ+ Ψ̇ + 3Ḣ + (2F̈ + H Ḟ ) − 2 Ψ
2F 2F 2F a
  2  
1 k − 6κ
= 8πδρ + 24πδP + 3δ F̈ + 3Hδ Ḟ + − 6H 2 δF (339)
2F a2
and
 
k2 R
δ F̈ + 3Hδ Ḟ + − δF
a2 3
8π   F
= (δρ − 3δP ) + Ḟ (χ + Ψ̇) + 2F̈ + 3H Ḟ Ψ − δR (340)
3 3
115
with fluid evolution equations
 
k2 θ
δ ρ̇ + 3H(δρ + δP ) = (ρ + P ) χ − 3HΨ − (341)
a

and   
(a4 (ρ + P )kθ)˙ k 1 2 2
= Ψ+ δP − (k − 3κ)(ρ + P )Σ . (342)
a4 (ρ + P ) a (ρ + P ) 3
Here the perturbation to the Ricci scalar, δR, is given by
  2  
k (k 2 − 3κ)
δR = −2 χ̇ + 4Hχ − − 3Ḣ Ψ + 2 Φ , (343)
a2 a2

and we have the constraint equations

3 Ḟ 3 h i
χ+ Ψ= 8πa(ρ + P )θ + δ Ḟ − HδF (344)
2F 2F
and
!
Ḟ (k 2 − 3κ) 3H Ḟ
H+ χ+ Φ+ Ψ
2F a2 2F
   
1 2 k2
= − 8πδρ − 3Hδ Ḟ + 3Ḣ + 3H − 2 δF . (345)
2F a

Furthermore, we again have that Ψ 6= Φ, in general. Instead it is the case that

8πa2 (ρ + P )Σ δF
Ψ−Φ=− − . (346)
F F
The equivalent equations to those given above can also be derived in the covariant ap-
proach to cosmological perturbation theory [250]. In the rest of this section we will
consider the consequences of these equations for various cosmological phenomena.

Inflation

The existence of inflation has provided considerable motivation for the study of f (R)
theories of gravity. The pioneering work on this subject was that of Starobinsky in 1980,
who found that theories with R2 corrections to their gravitational Lagrangian can have an
early period of de Sitter expansion [1179]. The spectrum of scalar and tensor fluctuation
generated during this type of inflation have been studied in [916, 1178, 719, 636] where
they were found to compatible with observations of the CMB. Quantum initial conditions
(“tunnelling from nothing”), as well as the process of reheating, were also considered in
[1249].
Inflation in f (R) gravity is particularly transparent in the Einstein conformal frame.
Here, for the Starobinsky model with [1179]

R2
f (R) = R + , (347)
6M 2
116
the conformally transformed theory in vacuum is one in which the minimally coupled
scalar field exists in a potential
3M 2  √χ 
V (φ) = 1 − e− 3 φ . (348)
2χ
This potential is displayed in Figure 4, where it can be seen that slow-roll inflation is
likely to occur in the region φ & mpl , and reheating is feasible during oscillations around
the minimum at φ = 0. This is, of course, exactly the type of behaviour that one wants
for a viable inflaton field.

1.5
VHΦLVH¥L

0.5

0
0 0.5 1 1.5 2
֐mpl

Figure 4: The potential given in Eq. (348), normalised by its asymptotic value as φ → ∞.

In fact, for the theory specified in Eq. (347) one can show that inflation is the
transient attractor of the general solution [852], and that in the region φ  mpl slow-roll
inflation occurs with
Ḣ M2
=− 2 ' , (349)
H 6H 2
and proceeds for N ' (2)−1 e-foldings. We will not proceed with showing the details
of reheating in this model, but only note that around φ ' 0 the potential given in Eq.
(348) is well approximated by V ' 21 M 2 φ2 . For details of how reheating occurs in this
potential the reader is referred to [1249, 892]. Pre-heating in f (R) inflationary models
has been considered in [1222]. Quantum cosmology, instantons, and their implications
for inflation have been studied in [1246, 1027].

Dark Energy

As well as accelerating expansion in the early universe, f (R) theories of gravity are
also capable of producing late-time accelerating expansion. There have been a large num-
ber of papers on this subject. There have also been attempts to construct quintessence-
like f (R) models which produce both early and late acceleration [974, 976, 977, 975, 82].
An easy way to see the potential for late-time accelerating expansion is to consider the
Friedmann equations (335) and (336) in vacuo. One can then identify effective density
117
and pressure parameters by analogy to the Friedmann equations of general relativity.
These are
RF − f − 6H Ḟ
8πρeff = (350)
2F
2F̈ + 4H Ḟ + f − RF
8πPeff = . (351)
2F
The equation of state of this effective fluid is then given by

2F̈ + 4H Ḟ + f − RF
w= , (352)
RF − f − 6H Ḟ
and one can then determine what is required to achieve w < −1/3, and hence accelerating
expansion.
One example of this is the now much considered theory of Carroll, Duvuri, Trodden
and Turner [255]
µ2(n+1)
f (R) = R − , (353)
Rn
where µ is a constant. For power-law evolution the effective equation of state, (352), is
then given at late-times by [255]
2(n + 2)
w = −1 + , (354)
3(2n + 1)(n + 1)
so that if n = 1, and the extra term in the gravitational Lagrangian is inversely propor-
tional to R, then one achieves an equation of state with w = −2/3, and hence accelerating
expansion. In fact, this just corresponds to the power-law solution given in Eq. (329),
for a theory with f (R) ∝ R−n .
The theory specified in Eq. (353) is now known to have a number of deficiencies that
make it non-viable [999, 43, 479, 428, 485, 258, 473, 1102, 122, 1157]. Some of these
have to do with the weak-field limit, which we have already discussed, others come from
cosmology, and yet more are due to stability issues, which we will address in Section
4.1.5. Many of these problems can be traced back to the value of the effective mass in
the scalar degree of freedom of this theory, which is thought to be either too small for
validity of gravitational physics in the solar system, or imaginary, leading to some of the
instabilities just alluded to. Models which have been constructed to try and over-come
these problems, while still leading to accelerating expansion at late-times, are those of
Starobinsky [1180]: "  −n #
R2
f (R) = R − µRc 1 − 1 + 2 , (355)
Rc
Hu and Sawicki [629]:
µRc
f (R) = R − , (356)
1 + (R/Rc )−2n
and Battye and Appleby [56]:
h i
f (R) = R + Rc log e−µ + (1 − e−µ )e−R/Rc (357)
118
where µ, n and Rc are all positive constants. Attempts to construct viable models that
include an early stage of inflationary expansion, as well as late-time accelerating expan-
sion, have been made in [321, 468]. All of these theories rely on the chameleon mechanism
to satisfy solar system constraints on gravity.

Observational Probes

As with many modifications to gravity, cosmological observables can be used to con-

strain f (R) theories of gravity. Here we will briefly survey the literature on this subject.
Primordial nucleosynthesis has been studied in f (R) gravity in [310, 204, 764, 475].
Due to the conformal equivalence between these theories and general relativity, the be-
haviour of cosmological solutions during the radiation dominated epoch are considerably
simplified: They evolve in a similar way to the radiation dominated solutions of GR, but
with a different value of the effective gravitational constant. This situation is familiar
from studies of primordial nucleosynthesis in the scalar-tensor theories of gravity out-
lined in Section 3.1. In the present case the relevant effective gravitational ‘constant’ is
inversely proportional to fR . The value of fR evolves throughout the matter dominated
and accelerating epochs, but is constant during radiation domination. Observations of
the abundances of light elements then provide constraints on the allowed values of fR
during the radiation dominated epoch, and hence constrain the rate of evolution of this
quantity that is allowed during the rest of the Universe’s history.
Other probes of the background expansion of an FLRW universe are the peak posi-
tions of the CMB spectrum of temperature fluctuations, and baryon acoustic oscillations.
Observations of these quantities allow the form of a(t) to be constrained, but due to the
freedom in the choice of f (R) are not able to falsify the most general form of these the-
ories directly [240, 929, 242, 376, 973, 496, 251, 980]. To go further using cosmological
observations we must therefore consider the solutions to the perturbation equations given
above.
The first thing that one may wish to consider is the growth of density perturbations,
δ = δρ/ρ. In a spatially flat universe, manipulation of Eqs. (339)-(346) allows one to
write [1299, 1219, 1226]

4πδρ (4 + 3(a/k)2 M 2 )
δ̈ + 2H δ̇ − = 0, (358)
3fR (1 + (a/k)2 M 2 )

where the mass parameter M is given, just as in the weak field limit discussed in Section
4.1.2, as
fR − RfRR
M2 = . (359)
3fRR
From the third term on the LHS of Eq. (358) it can be seen that the evolution of
δ depends on the magnitude of M , and, in particular, is different in the two regimes
M 2  k 2 /a2 and M 2  k 2 /a2 . When the former is true, the density perturbations
evolve as they do in General Relativity, with an effective Newton’s constant given by
G = 1/fR . For a matter dominated universe this means
2
δ ∝ t3 . (360)

119
In the latter regime, in which M 2  k 2 /a2 , this is no longer true. Here, the third term
in Eq. (358) is modified from its form in GR by a multiplicative factor of 4/3, and the
evolution of δ during the matter dominated era is consequently modified to
√
δ ∝ t( 33−1)/6
. (361)

The transition between these two limits is theory dependent. For studies on this subject
the reader is referred to [1180, 629, 122, 377, 1219, 1226, 1043, 1220, 519, 1221, 914, 940].
Interesting results are that the change in evolution between the two regimes discussed
above is scale dependent. That is, modes with different wave-numbers can evolve in
different ways depending on whether they are larger or smaller than a2 M 2 . This length
scale is therefore imprinted on the density perturbations. Furthermore, oscillating modes
can also become present when M 2  k 2 /a2 , which can lead to undesirable singularities
[515]. The inclusion of an R2 in the gravitational Lagrangian was found to remove these
singularities in [57]. One can also see that fRR > 0 is need for the stability of scalar
modes.
The modified growth of structure just discussed has consequences for large-scale struc-
ture, and the cosmic microwave background, which we will now discuss. The matter
power spectrum in f (R) theories of gravity has been considered in [495, 258, 1102, 122,
724, 1157, 790, 46, 45, 814, 1026], and cluster abundances have been used to constrain
these theories in [1115, 818, 500]. The formation of non-linear structure has also been
considered in [630, 1006, 1007, 1114, 1303]. Cosmic microwave observations are consid-
ered in [790, 1158, 1157, 818] where it is shown that power on large-scales is sensitive
to the modified growth of structure through the integrated Sachs-Wolfe (ISW) effect.
This can lead to damped power for small deviations from GR, or amplified power if the
deviations are large enough. Correlating ISW effects in the CMB with observations of
galaxy number density also leads to tight constraints [1158, 818], due to the sign of the
CMB temperature fluctuation changing if the modification to gravity is large enough.

4.1.5. Stability issues

There are a variety of stability issues that are of concern for f (R) theories of gravity.
These include ghost degrees of freedom, as evidenced in the Ostrogradski instability, as
well as the instabilities found by Frolov, and Dolgov and Kawasaki. Some of these issues
have been mentioned already. In this section we will discuss them further.

Ghosts, and the Ostrogradski instability

In Section 2.1.3, we discussed the problems associated with ghosts – pathological

fields that admit physical states of negative energy, or negative norm when quantised. It
is known that ghosts can occur in general higher derivative theories of gravity, see e.g.
[1187, 236, 606, 607, 293, 944, 392, 113, 990]. They are not, however, as problematic in
f (R) theories as they are for general fourth-order theories, as we will now outline. Let
us first consider the existence of negative energy states in the context of Ostrogradski’s
theorem [1003].
The Ostrogradski instability states that Lagrangians that contain second derivatives,
and are non-linear in those second derivatives, are generically unstable. At first sight
such a result appears to be problematic for f (R) theories of gravity, which are only linear
120
in second-order derivatives of the metric in the case of General Relativity with a possible
cosmological constant. One can show, however, that these instabilities do not occur for
f (R) theories [1281]. This works in the following way: Let us consider a Lagrangian

L = L(g, ġ, g̈), (362)

where dots denote derivatives of g with respect to some parameter λ. Now define a set
of four canonical variables by Q1 ≡ g, Q2 ≡ ġ, and
∂L d ∂L ∂L
P1 ≡ − , and P2 ≡ . (363)
∂ ġ dλ ∂g̈ ∂g̈

If it is now possible to write g̈ = f (Q1 , Q2 , P2 ) then the Hamiltonian of the system can
be written as
H = P1 Q1 + P2 f − L(Q1 , Q2 , f ). (364)
This Hamiltonian, however, is only linear in the momentum P1 , and cannot therefore
be stable. This is Ostrogradski’s instability. Now, f (R) gravity avoids this instability
because one cannot write down the equivalent of g̈ = f (Q1 , Q2 , P2 ) for each component
of the metric. Instead, only a single scalar degree of freedom contains the higher-order
derivatives, and by an appropriate field redefinition one can remove this extra field so
that the redefined metric appears in the Lagrangian only linearly in its second-order
derivatives. This is just the conformal transformation discussed in Section 4.1.1. The
Ostrogradski instability does not, therefore, apply to f (R) theories of gravity [1281].
Let us now consider ghost-like instabilities from the point of view of linear fluctua-
tions. In generic fourth-order theories massive spin-2 degrees of freedom appear along
with a scalar degree of freedom, and the familiar massless spin-2 degree of freedom from
General Relativity. It is the massive spin-2 fields in this situations that present the
generic problem with ghosts. Such fields are absent from f (R) theories, however, which
contain only the massless spin-2 fields of GR, and a single scalar field. Again, this is clear
from the existence of the conformal transformations outlined in Section 4.1.1. The f (R)
theories of gravity therefore do not always suffer from the same problems with ghosts
as more general higher-order theories, which will be discussed in more detail in Section
4.2.5.

Frolov instability

A potential problem with f (R) theories that modify the infra-red limit of General
Relativity has been identified by Frolov in [515]. This instability is caused by the fact
that for the scalar degree of freedom in f (R) theories curvature singularities can occur
at finite field value and energy level, a phenomenon previously investigated in [4, 203].
To illustrate this problem consider the f (R) proposed by Starobinsky, Eq. (355).
The potential for the effective scalar field in this theory is shown in Fig. 5. During
cosmological expansion, the scalar fields associated with FLRW cosmologies roll down
the slope from φ = 0 to the local minimum at φ ' −0.1. The short section of curve
between this minimum and the singularity at φ = 0 is the only part of the potential
the scalar field need experience in the entire history of a perfect FLRW solution. Frolov
121
 1.25

0.75
VHΦL

0.5

0.25

-1 -0.5 0 0.5 1
Φ

Figure 5: The potential for the scalar field in Starobinsky’s theory, Eq. (355), with Rc = 1, n = 1 and
µ = 2.

argues, however, that relatively small perturbation in curvature, caused by collapse of

dust, are enough to push φ back up the potential to the singularity. The existence of
such instability is, of course, undesirable for a physically plausible theory, although it
may be mitigated by the addition of higher power of R to the gravitational Lagrangian
[203, 81, 57].

Dolgov-Kawasaki instability

Finally, let us consider another instability that was initially found by Dolgov and
Kawasaki for the theory given by Eq. (353) with n = 1 [428]. This was later extended
to more general functions of f (R) that modify gravity in the infra-red limit [485] and
formalised better in [1124].
The basic point here is that the trace of the field equations, (289), acts as the prop-
agation equation for scalar degree of freedom. For Eq. (353) with n = 1 this equation
is  4
3µ4 3µ
R− − = 8πρ, (365)
R R2
where we have taken the matter content to be that of dust. Now, de Sitter space is a
solution of this equation, with
1 p
RdS = (8πρ + (8πρ)2 + 12µ4 ) ' 8πρ, (366)
2
for cosmologically relevant µ. If we now consider perturbations around this solution with
122
R = RdS + δR then we get to lowest order that
 
6µ4 3µ4
3 δR + 1 + 2 δR = 0. (367)
RdS RdS

Comparing this with the propagation equation of a massive scalar field gives

RdS R3 (8πρ)3
m2 = − − dS4 ' − ' −106 GeV, (368)
2 6µ 6µ4

where in the last equality the density has been taken to be that of water, ρ ' 103 kg/m3 ,
and µ has been taken to be ∼ 10−33 eV , as required to account for the observed late-
time accelerating expansion of the Universe. This large negative mass corresponds to
a catastrophic instability that should make itself apparent on time scales of ∼ 10−26
seconds.
For more general theories it can be shown that the effective mass squared in the
relevant scalar field equation takes the same sign as fRR [485]. It is therefore the negative
value of this quantity in the theory of Eq. (353) that is responsible for its exhibition
of this instability. Further, one can show that the addition of higher powers of R to
the gravitational Lagrangian again helps defend it from instability [968, 421, 970]. The
Dolgov-Kawasaki instability has been shown not to occur in the Palatini approach [1165].
The existence of this type of instability was rediscovered in [706], for relativistic stars.
The problem in this context has been further studied in [417, 72, 1237, 1206, 707, 901],
where it was shown that the instability can be avoided by changing the equation of state
of the star, adding a divergence to the scalar field potential, or including chameleon
effects. Neutron stars in f (R) gravity have been studied in [332, 1099], and instabilities
in systems with time-dependent mass have been studied in [60].

4.2. General combinations of Ricci and Riemann curvature.

In the previous section we considered theories that generalise the Einstein-Hilbert
action by replacing the Ricci scalar, R, with some non-linear function, f (R). Here we
go further, and allow the action to be a function of not only R, but of any of the three
linear and quadratic contractions of the Riemann curvature tensor26 : R, Rµν Rµν and
Rµνρσ Rµνρσ . A systematic approach to studying theories of this type, based on minimal
sets of curvature invariants, is proposed and studied in [646, 906].

4.2.1. Action and field equations

The most general weight-zero scalar density that one can construct from g, R, Rµν Rµν
and Rµνρσ Rµνρσ alone is given by
√
L = χ−1 −gf (R, Rµν Rµν , Rµνρσ Rµνρσ ) (369)

where f is an arbitrary function of its arguments, and χ is a constant which can be

determined from the Newtonian limit. The action is obtained, as usual, by integrating

26 There is also a fourth possibility, namely µνρσ R τ

τ µν R ρσ [418]. This contraction, however, is of
limited physical interest as it does not affect the field equations, due to its parity.
123
this density, together with that of the matter fields, over all space-time. The addition of
supplementary terms to the density (369), in order to cancel total divergences, and which
can be transformed to integrals on the boundary, can be problematic (see e.g. [849]) and
so, for simplicity, they are usually assumed to vanish.
As with f (R) theories one can proceed by using the metric variation approach (in
which the connection is a priori taken to be given by the Christoffel symbols), using
the Palatini formalism (in which the metric and connection are taken to be independent
fields in the gravitational action), or the metric-affine formalism (in which the metric
and connection are taken to be independent fields throughout the entire action). Here
we will spell out the metric variational approach explicitly, as this is the most commonly
considered form of the theory found in the literature. For studies involving the Palatini
procedure, as applied to the these theories, the reader is referred to [792, 477, 794, 173,
795, 1002, 1070].
Varying the action, derived from integrating Eq. (369) over all space, with respect to
the metric, then gives
Z
−1 √ h1 i
δI = χ dΩ −g f g µν δgµν + fX δX + fY δY + fZ δZ
2
Z h1
√
= χ−1 dΩ −g f g µν δgµν − fX (Rµν δgµν − g µν δRµν ) (370)
2
−2fY (Rρ(µ Rν) ρ δgµν − Rµν δRµν )
i
−2fZ (Rρσ (ν Rµ)σρ δgµν − Rµ νρσ δRµ νρσ ) ,

where we have defined X ≡ R, Y ≡ Rµν Rµν and Z ≡ Rµνρσ Rµνρσ . Using δRmu νρσ =
1 µ
2 g (δgσ;νρ + δgν;σρ − δgσν;ρ − δgρ;νσ − δgν;ρσ + δgρν;σ ) we can then write the grav-
itational part of the action as
Z
√
δI = −χ−1 dΩ −gP µν δgµν , (371)

where
1
P µν ≡ − f g µν + fX Rµν + 2fY Rρ(µ Rν) ρ + 2fZ Rσρ(µ Rν) ρσ
2
+fX;ρσ (g µν g ρσ − g µρ g νσ ) + (fY Rµν ) + g µν (fY Rρσ );ρσ
−2(fY Rρ(µ );ν)ρ − 4(fZ Rσ(µν)ρ );ρσ . (372)

The notation fN here denotes the functional derivative of f with respect to N . Looking
for a stationary point of this action, by setting the first variation to zero, then gives the
field equations
χ
Pµν = Tµν − gµν Λ (373)
2
where matter fields and a cosmological constant have been included. Here, Λ is the cos-
mological constant (defined independent of f (X, Y, Z)) and T µν is the energy-momentum
tensor of the matter fields. These field equations are generically of fourth-order, with the
exception of cases in which the function f is linear in second derivatives of the metric
[418], as occurs in GR.
124
 Unlike the case of f (R) theories of gravity, the theories described by the density given
in Eq. (370) are not, in general, conformally related to General Relativity with a scalar
field.

4.2.2. Weak-field limit

Let us consider the weak field limit of theories with additional terms in their action
that are quadratic in curvature invariants:
√
−g

L= R + αR2 + βRµν Rµν + γRµνρσ Rµνρσ , (374)
χ
where α, β and γ are constants. In this case one can use the well known result that the
Gauss-Bonnet combination of curvature invariants is a total divergence, i.e.

4Rµν Rµν − R2 − Rµνρσ Rµνρσ = total divergence, (375)

which in the action integrates to a boundary term that is usually ignored. By redefining
α and β we can therefore write Eq. (374) in the equivalent form
√
−g

L= R + αR2 + βRµν Rµν , (376)
χ
without any loss of generality in the solutions to the resulting field equations.
If we now substitute Eq. (376) into Eq. (373), to get the field equations, we find that
for the perturbed metric

ds2 = −(1 + 2ψ)dt2 + (1 − 2φ)(dx2 + dy 2 + dz 2 ), (377)

the lowest order equations in the weak-field and slow-motion limit are
χ
2(3α + β)∆R − R = − ρ (378)
2
(4α + β) ∆R − R − 2∆(ψ + β∆ψ) = −χρ, (379)

where the Ricci scalar is given as usual by R = −2∆ψ + 4∆φ. For a delta function
source, ρ = mδ, Eqs. (378) and (379) can then be seen to have the solutions
 
χm e−m1 r 4e−m2 r
16πψ = − 1+ − (380)
r 3 3
 
χm e−m1 r 2e−m2 r
16πφ = − 1− − , (381)
r 3 3
where
1 1
m21 = and m22 = − . (382)
2(3α + β) β
These are the solutions found by Stelle in 1978 [1187]. These solutions can be seen to
reduce to Eqs. (308) and (309) in the limit β → 0− . More generally, however, these
theories can be seen to exhibit massive modes with two different mass parameters. In
order to have non-oscillatory behaviour in the present case we must therefore require
that both 3α + β ≥ 0 and β ≤ 0 be simultaneously satisfied.
125
 If the solutions given in Eqs. (380) and (381) are the correct ones for describing the
space-time geometry around approximately isolated masses, such as the Sun, then one
can immediately see that if m1 and m2 are both large compared to 1/r then one recovers
the general relativistic prediction of γ = 1, just as with f (R) theories. For small masses,
however, the situation is somewhat different from the f (R) case. If both m1 and m2 are
small compared to 1/r then one has that the leading order term in Eqs. (380) and (381)
is a constant (which can be absorbed into coordinate redefinitions), followed by a term
proportional to r. This is a considerable deviation from the behaviour γ → 1/2 that
occurs when m1 r is small and m2 r is large, which is the limit of f (R) gravity with a
low mass parameter. It can also be seen that for m1 ≤ 2m2 gravity is always attractive,
while for m1 > 2m2 it is attractive over large distances, while being repellent over small
distances.

4.2.3. Exact solutions, and general behaviour

Let us now discuss what is known about the solutions to these general fourth-order
theories of gravity in the context of both isolated masses, and cosmological solutions.

Isolated Masses

Motivation for a number of studies in this area has come from Einstein’s particle pro-
gramme, in which one looks for asymptotically flat and singularity free vacuum solutions
which could be used to model particles [466]. While it is known that no such solutions
exist in General Relativity (Lichnerowicz’s theorem [797]), it has been conjectured that
they could exist in fourth-order theories [1252].
By studying the solutions of quadratic theories of the type given in Eq. (376) with
β = −3α it has been shown that the solutions to the linearised vacuum field equations
can be both asymptotically flat as r → ∞, and smooth as r → 0 [503]. These theories
are equivalent to the sum of an Einstein-Hilbert term and a Weyl term. Such results
would initially appear to be encouraging for Einstein’s programme, but it was later
shown that there are, in fact, no solutions with the specified properties that exist within
a neighbourhood of Minkowski space [1116]. This means that if any non-trivial static
spherically symmetric vacuum solutions to these theories exist, that are simultaneously
asymptotically flat and geodesically complete, then they must correspond to very large
energy densities (exceeding the energy density of neutron stars by at least 40 orders of
magnitude [1116]).
The theorems of Lichnerowicz [797] and Israel [650] have more recently been consid-
ered in the context of fourth-order theories of the form given in Eq. (376) by Nelson
[950]. Here it is found that for static space-times with spatial curvature satisfying

m21 − (3)
R ≥ 0 (383)
R̄µν R̄ν µ m21 + R̄µν R̄ν ρ R̄ρµ ≥ 0, (384)

the vacuum field equations imply that all asymptotically constant solutions (or asymp-
totically flat, if the inequalities are saturated) obey Rµν = 0. The expression for m1 is
given in Eq. (382). Over-bars here denote quantities projected into space-like hyper-
surfaces. The spherically symmetric solution to Rµν = 0 is, of course, the Schwarzschild

126
solution, which is geodesically complete only for the case of Minkowski space. Lichnerow-
icz’s theorem can therefore be extended to all theories that obey the inequalities (383)
and (384). It is then shown in [950] that if the spatial curvature satisfies
(3)
R ≤ m2 (385)
R̄µν R̄ν ρ R̄ρµ
≥ −m2 , (386)
R̄µν R̄ν µ

where m2 is given by Eq. (382), and the space-time is asymptotically constant (or
asymptotically flat, if the inequalities are saturated), then the only solutions with m21 ≥ 0
that exist in the region exterior to a closed spherical null surface also obey Rµν = 0. The
only asymptotically constant vacuum solutions with a horizon, that satisfy the bounds
(385) and (386), are therefore the Schwarzschild solutions. This extends the Israel’s no-
hair theorem for black holes to quadratic fourth-order theories of gravity. It is argued in
[950] that the inequalities (383)-(386) should be satisfied everywhere where the spatial
curvature is smaller than the scale of the corrections to the Einstein-Hilbert action. If
these corrections are motivated by quantum considerations, then we should therefore
expect all of the inequalities (383)-(386) to be satisfied for astrophysically interesting
systems. The stability of Schwarzschild black holes in the quadratic theories (376) has
been studied in [1271, 1186].
The initial value problem for quadratic theories, of the type given in Eq. (376), has
also been studied in [967], where it was found to be well-posed.

Cosmological Solutions

There are a number of exact cosmological solutions known to exist for fourth-order
theories containing Rµν Rµν and Rµνρσ Rµνρσ . This simplest of these is, of course, de
Sitter space, which exists for theories with general f (X, Y, Z) in Eq. (369), and a cos-
mological constant, if [309]
1 4
f − Λ = ΛfX + 2Λ2 fY + Λ2 fz , (387)
2 3
where fN denotes differentiation of f with respect to N . The stability of de Sitter space
in quadratic theories, of the type given in Eq. (376), has been studied in [102], and in the
more general case in [325]. Other known exact homogeneous and isotropic cosmological
solutions are the Einstein static universe and the Gödel universe, the existence of which
has been discussed in [309] for arbitrary f (X, Y, Z) (together with the conditions for
the existence of closed time-like curves in the case of latter). The existence of power-
law FLRW solutions, both in vacuum and in the presence of a perfect fluid, has been
discussed by
p Middleton in [887]. Power-law scaling FLRW solutions for theories with
L = R + α R2 − 4Rµν Rµν + Rµνρσ Rµνρσ have been investigated in [1233]. The extent
to which the FLRW solutions of General Relativity can be reproduced in these theories
is discussed in [848, 949].
As well as isotropic cosmological solutions, a number of studies have also been per-
formed of anisotropic cosmological solutions in these theories. The simplest of these
are probably the Bianchi I Kasner-like exact solutions found in [311], which were used
to show that the infinite sequence of anisotropic oscillations that occurs on approach
127
to the initial singularity in General Relativity does not occur in higher-order gravity
theories, except in unphysical situations. This type of solution was further studied in
[887]. Exact Bianchi type II and V Ih solutions were found by Barrow and Hervik in
[102], for quadratic theories of the type (376), and were used to show the lack of valid-
ity of the cosmic no-hair theorems in these theories: Anisotropic inflation with positive
Λ is possible, without de Sitter space as the late-time asymptote. These authors also
considered the general behaviour of Bianchi type I and II solutions in quadratic theo-
ries, where the possibility of a stable isotropic singularity was discovered [103]. Bianchi
type I, IV , V Ih and V IIh universes have been studied in [104], where it was shown
that periods of anisotropic expansion can occur after a near isotropic expansion, and
before re-isotropisation at late-times. Bianchi type V IIA solutions have been studied for
quadratic theories in [367], and Bianchi type IX universes have been studied by Cotsakis
et al. in [340], where the Kasner solution of General Relativity was shown not to be a
stable early asymptote of the quadratic theories given in Eq. (376).
Stability of past isotropic attractors has been the subject of study by Barrow and
Middleton [107, 888]. In the first of these papers the authors demonstrated the stability
of past isotropic solutions to the quadratic theories (376) under scalar, vector and tensor
inhomogeneous perturbations [107]. This supports the hypothesis that small pertur-
bations to the past isotropic attractor form part of the general cosmological solution to
quadratic theories of fourth-order gravity. This study is extended to theories with power-
law curvature terms, (Rµν Rµν )n , in their Lagrangian in [888], where conditions are given
for the stability of early isotropic states. This study also shows the instability of the
exact solution found in [311], as the initial singularity is approached. The asymptotic
behaviour of theories with quadratic corrections to the Einstein-Hilbert were studied,
in the context of string cosmology, in [1090, 236, 1218]. Exponential and power-law
FLRW solutions in higher-dimensional string inspired models are found in [855, 856, 17].
The evolution of FLRW solutions in generalised theories has also been studied using a
dynamical systems analysis in [326].

4.2.4. Physical cosmology and dark energy

Having discussed various cosmological solutions in these general fourth-order theo-
ries, let us now consider their relevance for observational cosmology and dark energy.
We will proceed with this by first discussing studies of more general theories, followed
by theories constructed from the Gauss-Bonnet curvature invariant. The Gauss-Bonnet
invariant has special properties, which we will discuss in the Section 4.2.5.

General Theories

In order to construct cosmological models that can produce late-time accelerating

expansion the authors of [254] considered theories of the type

µ4n+2
L=R+ , (388)
(aR2 + bRµν Rµν+ cRµνρσ Rµνρσ )n

where µ, n, a, b and c are constants. It is found that for these theories there exist

128
power-law attractors for the general spatially flat FLRW solutions, which are given by
√
  8n2 +10n+2−3α± Γ
t 4(n+1)
a(t) = a0 , (389)
t0

where
12a + 4b + 4c
α ≡ (390)
12a + 3b + 2c
Γ ≡ 9n2 α2 − (80n3 + 116n2 + 40n + 4)α
+64n4 + 160n3 + 132n2 + 40n + 4. (391)

The smaller of the exponents in Eq. (389) can be seen to → 0 as n → ∞, while the
larger tends to 4n. For large n it is therefore the case that accelerating expansion can
occur at late times. This is a generalisation of the type of model considered in [255], for
f (R) gravity. In [880] it is shown that while the theory given in Eq. (388) is capable
of explaining the supernova results, to do so and still have an acceptable age for the
Universe it requires the matter content of the Universe to have an equation of state
0.07 ≤ w ≤ 0.21, to 2σ. The FLRW solutions of theories with powers of R, Rµν Rµν and
Rµνρσ Rµνρσ added to the Einstein-Hilbert action were also studied in [455], where the
possibility of late-time accelerating expansion was considered.
Primordial nucleosynthesis in theories with powers of Rµν Rµν added to the Einstein-
Hilbert action have been considered in [957], where constraints from observed element
abundances are imposed. As with scalar-tensor theories, the constraints imposed from big
bang nucleosynthesis are largely due to the different expansion rate during the radiation
dominated period due to a different value of the effective Newton’s constant.
The addition of a conformally invariant term to the Einstein-Hilbert action has been
considered in [953, 954, 951, 952]. In this case the gravitational Lagrangian takes the
form given in Eq. (376) with 3α + β = 0, and the resulting field equations are some-
times known as the ‘Bach-Einstein equations’. The solutions to these equations have
been studied in the context of inflation [953], the evolution of background cosmological
models [954], the observational constraints available from pulsars [951], and weak fields
and gravitational waves [952]. Theories of this type are motivated, in part, from non-
commutative geometry [268]. For a review on short scale modifications of gravity in the
context of non-commutative geometry, see [961].

Theories with L = f (R, R2 − 4Rµν Rµν + Rµνρσ Rµνρσ )

Theories that are functions of the Ricci scalar, R, and the Gauss-Bonnet combination,

Ĝ = R2 − 4Rµν Rµν + Rµνρσ Rµνρσ , (392)

have been particularly well studied, as they are motivated by string theory [886, 530, 529,
981], and have improved stability properties (as will be discussed in Section 4.2.5). The
linear case of f = R+ Ĝ is known to be equivalent to the Einstein-Hilbert Lagrangian in 4
dimensions, up to surface terms, but more general functions, of the type f (R, Ĝ), produce
field equations that differ from those of General Relativity. The mathematical properties
129
of the Gauss-Bonnet tensor, that occurs from varying the action of these theories, as well
as the more general Lovelock tensor, have been studied in [493, 491, 492, 494].
The general behaviour of spatially flat FLRW solution in theories with L = R + f (Ĝ)
has been studied by Zhou, Copeland and Saffin in [1309] using a phase plane analysis.
In this case the Friedmann equations become
χ
3H 2 = ĜfĜ − f − 24H 3 f˙Ĝ + ρ (393)
2
χ
2Ḣ = 8H 3 f˙Ĝ − 16H Ḣ f˙Ĝ − 8H 2 f¨Ĝ + (ρ + P ), (394)
2

where R = 6(Ḣ + 2H 2 ) and Ĝ = 24H 2 (Ḣ + H 2 ). The existence of both stable de Sitter
space, and phantom-like accelerating solutions to the above equations can be demon-
strated, as well as trajectories in the phase space that mimic the evolution of the standard
ΛCDM universe through radiation and matter dominated periods [1309]. The stability of
de Sitter space, as well as radiation and matter dominated epochs has also been studied
by de Felice and Tsujikawa in [371], where the conditions fĜĜ > 0 and fĜĜ → 0+ as
|Ĝ| → ∞ were found to be required for models to be viable. These authors suggest the
following functional forms for f (Ĝ) as examples that satisfy these conditions, and could
produce acceptable expansion histories for the Universe:
! q ! q
Ĝ −1 Ĝ λ Ĝ2
f (Ĝ) = λ p tan − Ĝ∗ ln 1 + − αλ Ĝ∗ (395)
Ĝ∗ Ĝ∗ 2 Ĝ2∗
! q
Ĝ −1 Ĝ
f (Ĝ) = λ p tan − αλ Ĝ∗ , (396)
Ĝ∗ Ĝ∗

where α, λ and Ĝ∗ are constants. It is further claimed that these forms of f (Ĝ) are com-
patible with solar system observations [372], producing corrections to the Schwarzschild
metric that are of the form ∼ H 2 rs2 (r/rs )p , where rs is the Schwarzschild radius of the
Sun, H is the Hubble rate and p is a model dependent quantity. Much larger correction
to General Relativistic predictions are claimed in [364] for theories with polynomial addi-
tions of the Gauss-Bonnet term, Ĝn , to the Einstein-Hilbert action. The cosmologies of
these theories, and theories with inverse powers of αĜ+βR added to the Einstein-Hilbert
action (where α and β are constants), have been considered in [391, 979], while the FLRW
solutions of other R+f (Ĝ) theories have been considered in [971, 323]. The ‘inverse prob-
lem’, of finding FLRW solutions that behave like ΛCDM has been considered for R+f (Ĝ)
theories in [324]. The phase space of FLRW solutions to L = f (R, Ĝ) theories, and the
transition from deceleration to acceleration, has also been studied in [25, 26, 24]. Super-
nova, BAO and CMB observations have been used to constrain L = R + f (Ĝ) theories
in [907].
Linear perturbations around spatially flat FLRW universes have been studied in L =
f (R, Ĝ) theories by de Felice, Gérard and Suyama in [368], and in L = R + f (Ĝ) theories
in particular by Li, Barrow and Mota [791]. The former of these studies uses the velocity
potentials and variational principle approach of Schutz [1119], while the latter uses the
covariant formulation of Ellis and Bruni [469]. In the case of general f (R, Ĝ) it is found
that scalar perturbations in these theories have, in general, six degrees of freedom, two
130
of which propagate on small scales with group velocity [368]

256 Ḣ 2 (fRR fĜĜ − fR2 Ĝ ) k2

vg2 ' − . (397)
3 (8fRĜ H 2 + 16H 4 fĜĜ + fRR )(fR + 4H f˙Ĝ ) a2

There can be seen to be a k-dependence in Eq. (397), which does not occur in General
Relativity. Such a relation had previously been found for the vacuum case in [393],
where it was argued that the space-time is unstable if vg2 < 0, or has super-luminal
modes in short wavelength modes if vg2 > 0. These features are problematic, but can be
avoided in theories that satisfy fRR fĜĜ − fR2 Ĝ = 0. Such theories have scale independent
propagation speeds only, and include the cases of L = f (R), f (Ĝ), Ĝ+f (R) and R+f (Ĝ).
The latter case, being the subject of study in [791], has also been shown to suffer from
matter instabilities. This is due to the evolution equation for perturbations to fĜ , which
we write as , and which obeys [791]
! " ! ! #
˙
27(3 − 4f Ĝθ)
4θ̇ 4θ̇ k 2 θ̇2 2θ2
¨ + θ + ˙ + 1+ 2 − 2 θ̇ + 2 + −  = S, (398)
θ θ a2 θ 9 48θ4 fĜĜ

where θ = 3H is the expansion scalar, and the reader is referred to [791] for the form of
the source term S. For stability it is required that the third term in the square brackets
be positive, and remain dominant over the first, which is expected to be negative during
matter domination. This requires fĜĜ ≥ 0, and for fĜĜ H 6 to remain suitably small, in
order to avoid instabilities [791, 370]. These are strong constraints on the forms of f (Ĝ).
It was further shown in [368] that vector modes in the general case of f (R, Ĝ) decay, and
that the propagation of tensor modes in these theories is model dependent.

4.2.5. Other topics

Let us now consider some remaining topics in fourth-order gravity, that have yet to
be discussed.

Theories with L = f (R, φ, R2 − 4Rµν Rµν + Rµνρσ Rµνρσ )

There has been some study of fourth-order theories that include a scalar field, as
well as the Ricci scalar and the Gauss-Bonnet scalar, in the gravitational action. This
is motivated by the dilaton that arises in string theory compactifications [886], and has
been studied in terms of ‘pre-big bang’ cosmology in [530, 529, 264]. In this scenario
there is an early period of very rapid expansion due to the kinetic term of the scalar
field. Black holes in these theories, and their extensions, have been studied in [890,
889, 678, 22, 1209, 700, 229, 231, 573, 574, 994, 995, 857, 285, 228, 996, 858]. Further
string motivated study of FLRW cosmology in the context of these theories has also been
performed in [53, 1217, 1223, 324, 575].
Late-time acceleration has also been studied in theories where a scalar field has been
included, along with R and G, in the gravitational action [982, 960, 959, 36, 728, 729,
1223, 1097, 324, 37, 983, 80]. These papers have considered the evolution of FLRW
space-times, as well as inflation, structure formation, and the constraints that can be
imposed upon them from supernovae, CMB, BAO, solar system observations, and pri-
mordial nucleosynthesis. These observations place strong constraints on the theories.
131
Greater-than-fourth-order Theories

Another option that has been considered in the literature is that the action itself
could contain derivatives of curvature invariants, so that [206]

L = f (R, R, 2 R, . . . , n R), (399)

where  is the D’Alembertian. Extremising the action associated with this Lagrangian,
by varying the metric, gives the field equations [1117]
1 χ
YRµν − f gµν − Y;µν + gµν Y + Xµν = Tµν , (400)
2 2
where
n 
X 
1 i−1 ;σ i−1


Xµν ≡ gµν (Zi ( R) );σ − Zi;(µ  R ;ν)
,
i=1
2
Xn
∂f
Y ≡ i ,
i=0
∂(i R)
n
X ∂f
Zi ≡ j−i .
j=i
∂(j R)

The field equations (400) can be seen to generically contain derivatives of the metric of
order 2n + 4, so that the familiar fourth-order theories discussed above are recovered
when n = 0. Theories with infinite n have also been considered in the literature, and
have been claimed to be ghost-free [164, 163].
Greater-than-fourth-order theories can be shown to be equivalent, under a conformal
transformation, to General Relativity with two scalar fields [550], and their Newtonian
limit has been considered in [657], where it was found that the familiar form of Newto-
nian potentials and Yukawa potentials can be present. Their consequences for inflation
have been studied in [143, 551, 144, 42, 871, 977, 163], and the attractor nature of de
Sitter space established. Bouncing cosmologies in these theories have been considered
in [164, 162, 163], and the form of the CMB has been investigated in [162]. Primordial
nucleosynthesis has been considered in [957], and the consequences of this type of theory
for dark energy have been considered in [977]. For a more detailed overview of theories
with greater than four derivatives of the metric in their field equations the reader is
referred to [1118].

Conformal Gravity

One more possibility is to completely abandon the Einstein-Hilbert action, even as

a limiting case of the fundamental action. Such a proposal has been advocated by
Mannheim [866] in the case of conformal Gravity. Here the Einstein-Hilbert action is
replaced by
Z
√
SC = −αG d4 x −gCλµνκ C λµνκ , (401)
132
where Cλµνκ is the Weyl tensor, given by
1
Cλµνκ = Rλµνκ + Rαα [gλν gµκ − gλκ gµν ]
6
1
− [gλν Rµκ − gλκ Rµν − gµν Rλκ + gµκ Rλν ],
2
and αG is a dimensionless gravitational self-coupling constant. It has been shown that
such an action can be obtained from the path integral of fermionic degrees of freedom
for the conformal and gauge invariant action of a fermionic field.
The case has been made that such a theory has a number of desirable properties
eluding other higher derivative theories of gravity. Even though the equations are fourth
order, signalling the presence of negative energy states, it has been shown that such states
are completely decoupled from what the authors dub the physical sector [865]. Conformal
gravity can then be held up as a viable theory of quantum gravity. Furthermore, the
peculiar ultraviolet properties of conformal gravity have been argued to lead to a solution
to the cosmological constant problem.
At a classical level, conformal gravity has been shown to have intriguing properties.
For a start, the non-relativistic limit of the field equations leads to a fourth order differen-
tial equation for the gravitational potential, Φ, in which the usual Newtonian potentials
that drop off as 1/r are but one possibility. The general weak gravity potential is of the
form
A
Φ(r) = − + Br.
r
Such a form, it has been argued, can be tuned to fit a range of galaxy rotation curves. As
in some other theories of modified gravity, this is achieved by fixing universal parameters.
This may be contrasted with the usual dark matter prescription in which, for each galaxy,
one can choose the properties of the dark matter halo.
The situation becomes more complicated once one adds couplings to matter fields
[864]. Conformal symmetry is spontaneously broken through a new scalar degree of
freedom, S, such that the general (conformal) matter action is
Z
√ 1 1
IM = −~ d4 x −g[ S µ Sµ − S 2 Rµµ + λS 4 + iψ̄γ µ ∇µ ψ − gS ψ̄ψ],
2 12
where ψ is a fermionic field. One can extend this to more general actions containing
scalar and fermionic fields. The vacuum expectation value of the scalar field is what
then sets the Gravitational constant and the coupling to matter. It can also be used to
renormalise the cosmological constant.
Although some of the quantum properties of conformal gravity have been worked out,
a fully consistent and complete analysis of their cosmology is still lacking. In particular,
and in its current incarnation, in which no dark matter is invoked, it is unclear how the
correct angular diameter distance for the CMB can be obtained.

Theories with L = f (T )

An interesting variant on generalisations of the Einstein-Hilbert action are the L =

f (T ) theories, where T is a contraction of the torsion tensor (defined below). These
133
theories generalise the ‘teleparallel’ approach to General Relativity, which corresponds
to a Lagrangian L = T , from which Einstein’s equations can be derived [1235]. Here T
is defined by
1 1
T = T µνρ Tµνρ + T µνρ Tρνµ − Tµν µ T νρρ , (402)
4 2
where T µνρ is the torsion tensor, defined in terms of the vierbein from gµν = ηαβ hα β
µ hν ,
as

T µνρ = hµα ∂ν hα α
ρ − ∂ρ hν . (403)
This definition is equivalent to setting T µνρ equal to the antisymmetric part of the
Weitzenbock connection. Now, varying the action
Z √
−g
S= f (T )dΩ (404)
16πG
with respect to the vierbein fields gives the field equations
 
1 1
fT Rµν − gµν R + gµν (f − ft T ) (405)
2 2
 
1
+ (Tµνρ + Tρνµ + Tνµρ ) − gµρ T σνσ + gµν T σρσ fT T ∇σ T = 8πGΘµν ,
2

where we have called the energy-momentum tensor Θµν , to distinguish it from the torsion
tensor. It can be seen that in the case f (T ) = T the field equations (405) reduce to
Einstein’s equations, so that the theory L = T is equivalent to the Einstein-Hilbert
action, as stated above. For f 6= T , however, the teleparallel approach outlined here
gives different field equations to the fourth-order theories we have so far considered.
It was within the framework of these generalised equations that Bengochea and Fer-
raro suggested that the late-time accelerating expansion of the Universe could be ac-
counted for without dark energy [135]. These authors considered the particular case
α
f =T − , (406)
(−T )n

where α and n are constants, and constrained the resulting FLRW cosmology they found
with supernovae, BAOs and the CMB. They found the best fitting model has n = −0.10,
Ωm = 0.27, and has the required radiation, matter and accelerating epochs. A large
number of papers have followed [135] in a short space of time, exploring the transition
from deceleration to acceleration, observational constraints, conformal transformations,
and structure formation [815, 1287, 932, 1288, 1227, 289, 134, 1286, 78, 933, 681, 1293,
1308, 79].
It has been shown, however, that these theories do not respect local Lorentz invari-
ance, and have a number of extra degrees of freedom that are not present in General
Relativity as a result [796]. This can seen by noticing that one can write the Ricci scalar
in terms of T µνσ as


R = −T − 2∇µ T νµν . (407)


Now, while R is of course a Lorentz scalar, one can show that ∇µ T νµν is not. It there-
fore follows that T is not a Lorentz scalar either, and so the f (T ) theories do not exhibit
134
local Lorentz symmetry. The exceptional case is f = T , in which case the non-Lorentz
invariant part of the action can be seen from Eq. (407) to be a total divergence, which
does not affect the field equations.

Stability Issues

There are serious concerns with the stability of general theories of the type L =
f (R, Rµν Rµν , Rµνρσ Rµνρσ ). Not least of these is the presence of ‘ghosts’, or perturbative
modes with negative norm, as well as tachyonic instabilities in massive modes [1187, 236,
606, 607, 293, 944, 392, 113, 990].
Let us now outline how ghost terms arise in these theories, following the discussion
of [606, 607, 293]. This starts by considering the quadratic theory
 
√ 1 2 1 2
L = −g R + R − C , (408)
6m20 2m22

where m0 and m2 are constants, and C 2 = Cµνρσ C µνρσ = Rµνρσ Rµνρσ − 2Rµν Rµν + R2
is the square of the Weyl tensor. Any quadratic theory can be written in this form, up
to boundary terms, because of the Gauss-Bonnet identity. We now want to identify the
scalar and spin-two degrees of freedom in Eq. (408), for which it is convenient to introduce
auxiliary fields that play these roles, and to transform so that they have canonical form.
We will do this now, following [606]. By introducing an auxiliary scalar field, ϕ, and
performing a conformal transformation g̃µν = eϕ gµν , Eq. (408) can be rewritten as [606]
" #
p 3  ˜ 2 3m20

−ϕ 2 C̃ 2
L = −g̃ R̃ − ∇ϕ − 1−e − . (409)
2 2 2m22

By extremising this equation with respect to ϕ, and substituting the resulting value of φ
back into the Lagrangian density, one recovers Eq. (408). It can be seen from Eq. (409)
that ϕ now has the kinetic term of a canonical scalar field. By introducing p a second
auxiliary field, πµν , and transforming the metric so that ḡ µν = g̃ µρ Aρν / |A|, where
Aρν = (1 + 12 φ)δρ ν − φρν , we can then rewrite Eq. (409) as [606]
"
√ 3 −1
ν ¯ µ ¯ 3m20
2
L = −ḡ R̄ − A µ
∇ ϕ∇ ν ϕ − p 1 − e−ϕ (410)
2 2 |A|
#
 ρ  m2

−ḡ µν
C µσ C σνρ − C ρµν C σσρ + p 2 φµν φµν − φ2
4 |A|
"
√ 3 ¯µ ¯ 3m20 2 m22

' −ḡ R̄ − ∇ ϕ∇µ ϕ − ϕ + φµν φµν − φ2 (411)
2 2 4
#
1 ¯ ¯ µ ¯ µ φνρ ∇
¯ φ + 2∇
µ νρ ¯ φµν ∇
µ ¯ φ − 2∇
ν ¯ µ φνρ ∇
¯ φ
ρ νµ

− ∇µ φ ∇ φ − ∇ ,
4

where in the second equality we have expanded out to quadratic order in ϕ and φµν
around zero, so that we are considering theories that are close to GR. The field φµν = πµ ν
135
has been introduced here to make clear with respect to which metric the indices are being
raised or lowered: those of πµ ν are raised and lowered with g̃µν , while those of φµν are
raised and lowered with ḡµν . The quantities C µνρ are defined as

1 −1
µσ ¯ ¯ ρ g̃νσ − ∇


¯ σ g̃νρ .
C µνρ = g̃ ∇ν g̃ρσ + ∇ (412)
2
The Lagrangian given in Eq. (411) now has scalar, ϕ, and spin-2 modes, φµν , both in
canonical form. It can be seen from Eq. (411) that for real φµν the spin-2 field does
indeed have the wrong sign before its kinetic term, and is therefore generically a ghost,
while for real ϕ the scalar mode is not [1270]. What is more, if m20 < 0 or m22 < 0 then
the scalar or spin-2 modes exhibit tachyonic instabilities, respectively.
Having outlined the proof for the generic existence of spin-2 ghosts in quadratic
fourth-order theories of gravity, (408), let us now extend this to more general theories of
the form √
L = −gf (X, Y, Z), (413)
where X ≡ R, Y ≡ Rµν Rµν and Z ≡ Rµνρσ Rµνρσ . This demonstration proceeds by
showing that the particle content of theories of the type (413) are the same as the
quadratic theories (408) (at least, when considering fluctuations around de Sitter space),
and then using the result derived above, that these theories generically contain spin-2
ghosts. The first step here is to introduce auxiliary fields φ1 , φ2 and φ3 so that the
Lagrangian density (413) becomes [607, 293]
√
L = −g [f + f1 (X − φ1 ) + f2 (Y − φ2 ) + f3 (Z − φ3 )] , (414)

where fi ≡ ∂f /∂φi , and i = {1, 2, 3}. As long as fi is non-degenerate, extremising with

respect to the these three new fields then gives φ1 = X, φ2 = Y and φ3 = Z, so that
Eq. (413) is recovered. Using the Gauss-Bonnet combination, and discarding boundary
terms, this equation can then be written as [293]
√ h 1
L = −g (f − φ1 f1 − φ2 f2 − φ3 f3 ) + f1 R + (f1 + f3 )R2 (415)
3
1 µνρσ
+ (f2 + 4f3 )Cµνρσ C
2  
1 µνρσ µν 2 2 i
− (f2 + 2f3 ) Cµνρσ C − 2Rµν R + R ,
2 3

where Cµνρσ is the Weyl tensor, and the last quantity in brackets is the Gauss-Bonnet
combination. At this point the theory has been shown to be equal to a scalar-scalar-
scalar-tensor theory, with the three scalars non-minimally coupled to quadratic curvature
invariants. Finally, we can choose to expand the density above up to second order in
fluctuations around a de Sitter background with constant Ricci curvature, R0 , so that
[944]  
√ 1 2 1 2
L = −g −Λ + αR + R − C , (416)
2m20 2m22

136
where boundary terms have been ignored, and we have defined
1 1 1
Λ ≡ hf − XfX + X 2 ( fXX − fY − fZ )
2 4 6
1 1 1 1 1
+X 3 ( fXY + fXZ ) + X 4 ( fY Y + fZZ + fY Z )i0
2 3 8 18 6
2
α ≡ hfX − XfXX − X 2 (fXY + fXZ )
3
1 1 1
−X 3 ( fY Y + fZZ + fY Z )i0
4 9 3
m−2
0 ≡ h(3fXX + 2fY + 2fZ ) + X(3fXY + 2fXZ )
3 1
+X 2 ( fY Y + fZZ + fY Z )i0
4 3
m−2
2 ≡ −hfY + 4fZ i0 ,

where h. . . i0 denotes the value of the quantity inside the brackets on the de Sitter back-
ground. It should be clear that Eq. (416) is identical to Eq. (408), up to the values
of α and Λ. The particle content of general theories of the type (413), on a de Sitter
background, therefore also has a scalar mode with mass m0 , and a ghost-like spin-2 mode
with mass m2 .
As mentioned in Section 4.2.4, it has been suggested that theories that are only
functions of the Ricci scalar, R, and the Gauss-Bonnet combination, G = R2 −4Rµν Rµν +
Rµνρσ Rµνρσ , can evade the ghost problem outlined above [944]. The reason given for this
is that the mass term m−2 2 → 0 as fY → 4fZ , a condition that is satisfied for theories of
the type L = f (X, Z − 4Y ), or, equivalently, L = f (R, G). When m−2 2 vanishes it can
be seen from Eq. (416) that the term responsible for the ghost spin-2 fluctuations will
also disappear. Further requirements for the non-existence of ghosts in f (R, G) theories
are discussed in [392], with particular reference to the model of [254]. Such theories
may still be subjected to constraints on their parameters by the possible existence of
tachyonic instabilities, if m20 < 0. For theories with L = f (R, G) it can be seen that
m−20 = 31 R2 fGG , so that the condition m20 > 0 is equivalent to the stability condition
fGG > 0 found in [371, 370] and [791] in the context of cosmology.

4.3. Hořava-Lifschitz Gravity

Hořava-Lifschitz (HL) gravity was proposed as a toy model of quantum gravity [618,
619, 620]. The model is non-relativistic and relies on anisotropic scaling between space
and time in the UV to help render the theory asymptotically safe. Furthermore, it
was claimed that General Relativity could be recovered in the infra-red by including
additional relevant operators. HL gravity in its various guises has been reviewed in a
number of articles [1016, 1166, 1266, 926, 170].
To understand the idea behind HL gravity we must first understand why perturbative
General Relativity is not UV complete. The non-renormalisability arises because the
coupling constant has negative mass dimension, [G] = −2, and the graviton propagator
scales as 1/p2 . Consider the following scalar field theory,
Z  
1
S = d4 x − (∂ϕ)2 + λϕ6 . (417)
2
137
Again, the propagator scales like 1/p2 , and the coupling constant has mass dimension
[λ] = −2, so schematically at least, one might expect this theory to be non-renormalisable
too. To render the theory asymptotically safe, we need to improve the UV behaviour of
the propagator. One might do this by adding relativistic higher derivative terms to the
action, but this is known to introduce an additional ghost-like degree of freedom. The
reason for the existence of this ghost can be traced back to higher order time deriva-
tives as opposed to space derivatives. This observation suggests an alternative approach.
Let us abandon Lorentz invariance and introduce higher order spatial derivatives with-
out introducing any higher order time derivatives. The former should improve the UV
behaviour of the propagator, whereas the latter guarantees the absence of ghosts. We
therefore modify the kinetic term
1 1 1
− (∂ϕ)2 → ϕ̇2 − ϕ(−∆)z ϕ, (418)
2 2 2
where ∆ is the spatial Laplacian. We now have a non-relativistic dispersion relation
w2 ∝ k 2z , which means that time and space scale differently,

x → lx, and t → lz t, (419)

For large enough z, it follows that the coupling constant has a non-negative scaling
dimension, [λ] = 4z − 6 so we expect the theory to be power counting renormalisable,
and ghost-free. On the flip side, we have broken Lorentz invariance, which is well tested
at low energies. However, we can cope with this by adding a relevant operator of the
form Lrel = 21 c2 ϕ∆ϕ. This leaves the good UV physics unaffected, but allows Lorentz
invariance to be restored as an emergent symmetry in the IR, with an emergent speed of
light c.
In HL gravity, one applies similar logic to the relevant perturbative degrees of freedom,
schematically replacing φ with the graviton, hij . Since we will require time and space
to scale differently in this model, we must first choose a preferred time, t, which in the
language of General Relativity means making an ADM split [902]

ds2 = −N 2 c2 dt2 + qij (dxi + N i dt)(dxj + N j dt), (420)

where qij (x, t) is the spatial metric and N i (x, t) is the shift vector. For the lapse function
we consider two separate scenarios: (i) the projectable case where the lapse N = N (t) is
homogeneous and (ii) the non-projectable case where the lapse N = N (x, t) can depend
on space. Having chosen a preferred time, we no longer have the full diffeomorphism
group, Diff(M), but a subset known as foliation preserving diffeomorphisms, Diff(M, F),
generated by
δt = f (t), and δxi = ξ i (x, t). (421)
Diff(M, F) is defined by the following set of infinitesimal transformations

δN = ∂t (N f ) + ξ i ∂j N (422)
i i i i
δN = ∂t (N f + ξ ) + Lξ N (423)
δqij = f ∂t qij + Lξ qij . (424)

Note that this hard breaking of diffeomorphism invariance is at the root of many of
the problems facing HL gravity as it allows additional degrees of freedom to propagate
138
[230, 278]. To see the extra degree of freedom emerge it is convenient to perform a
Stuckelberg trick [1076], and artificially restore full diffeomorphism invariance at the
expense of introducing a new field – the Stuckelberg field. This field becomes strongly
coupled as the parameters of the low energy theory run towards their diff-invariant values
[278] (see also [167, 735, 1025]).
We can think of the lapse and shift as playing the role of gauge fields in Diff(M, F).
It follows that the projectable case is the more natural since then the gauge fields have
the same space-time dependence as the corresponding generators. Having said that one
might expect it to be easier to match the non-projectable case to General Relativity in
the infra-red.
In any event, the action from these theories is built from objects that are covariant
with respecth to Diff(M, F).i These are the spatial metric, qij , and the extrinsic curvature,
Kij = 1 ~ ~ i is the spatial covariant derivative. In the non-
q̇ij − 2∇(i Nj) , where ∇
2N
projectable case one should also consider terms built from ai = ∇~ i log N [167]. To build
the gravitational analogue of the action (417), we replace the kinetic term such that
1 2 1 √
ϕ̇ → qN (Kij K ij − λK 2 ), (425)
2 κG
where κG is the gravitational coupling with scaling dimension [κG ] = z − 3, and λ is
a dimensionless parameter that also runs with scale. Clearly, for the z = 3 theory
the gravitational coupling constant is dimensionless, which may lead one to suspect the
theory to be power counting renormalisable. For z = 3 the leading order term in the UV
part of the action becomes
1 √
− ϕ(−∆)3 ϕ → −κG qN V6 , (426)
2
where the dimension six contribution to the potential is
~ k Rij ∇
V6 = β ∇ ~ k Rij + . . . . (427)

Here β is a dimensionless parameter, Rij is the spatial Ricci tensor and “. . .” denotes
any of the other possible dimension 6 operators that one might wish to include, e.g.
R3 , R∆R, (ai ai )3 etc.
Now let us consider the type of relevant operators one might add. If we demand our
action to be invariant under spatial parity xi → −xi and time reversal t → −t, then we
only need consider even dimensional operators,
1 √
Lrel = − qN (V2 + κG V4 ). (428)
κG
At dimension four these are [692]

V4 = A1 R2 + A2 Rij Rij + +B1 Rai ai + B2 Rij ai aj + B3 Rai i

+ C1 (ai ai )2 + C2 (ai ai )aj j + C3 (ai i )2 + C4 aij aij , (429)

whereas at dimension two,

V2 = −α(ai ai ) − c2 R. (430)
139
Here we have introduced the notation ai1 ...in = ∇ ~ i1 · · · ∇
~ in log N . Of course, terms of
this form are only relevant in the non-projectable case.
The full z = 3 gravitational theory is now given by
Z
1 √  
S= dtd3 x qN Kij K ij − λK 2 + c2 R + α(ai ai )2 − κG V4 − κ2G V6 +Sm , (431)
κG

where Sm [N, Ni , qij ; Ψ] is the matter part of the action, and V6 is the relevant dimension
6 operator. Note that in the absence of full diffeomorphism invariance we do not require
matter to satisfy energy-momentum conservation [278]. Indeed, in general we expect
to see violation of energy conservation since Diff(M) breaking operators in the gravity
Lagrangian will induce Diff(M) breaking quantum corrections to the matter Lagrangian
[692].
Let us compare this to the Einstein-Hilbert action, written in terms of the ADM
variables as Z  
c4 √ Kij K ij − K 2
SGR = dtd3 x qN + R . (432)
16πG c2
The claim is that λ = 1 and α = 0 are the infra-red fixed points of the renormalisation
group flow. Of course, the parameter α plays no role in the projectable theory, since
any terms containing ai = ∇ ~ i log N will vanish for N = N (t). For both the projectable
and non-projectable theories, the free parameters run to their infra-red fixed points at
low energies, so that the HL action (431) tends towards the Einstein-Hilbert action (432)
with an emergent speed of light, c, and an emergent Newton’s constant, G = κG c2 /16π.
Before delving further into the different manifestations of HL gravity, let us pause
to make a few general comments. The first of these is with regard to the large number
of terms appearing in the potential. We have not bothered to present the contributions
from dimension six operators since they are two numerous27 . To reduce the number of
terms, Hořava originally borrowed the notion of detailed balance from condensed matter
theory [619], but this has since been shown to lead to phenomenological problems [278,
1169, 1170]. Of course, the large number of terms appearing in the potential is really
only an aesthetic concern.
A second, more serious, concern, involves the fine tuning of light cones for each field.
Since Lorentz invariance is not exact there is no symmetry guaranteeing that all fields see
the same emergent light cone. We would like there to be some mechanism suppressing
Lorentz violating operators at low energies but preliminary investigations suggest that
fine tuning is required [1045]. It is possible that supersymmetry may be able help with
this to some extent [565].
Another issue that has yet to be fully explored concerns possible equivalence principle
violations in HL gravity [692]. To see how these might arise it is convenient to go to the
Stuckelberg picture. The Stuckelberg trick was developed in the context of massive gauge
theories [1076], but it has proven very useful in elucidating some of the key physics of
HL gravity as well (see, for example, [278, 167, 692]). Recall that the anisotropic scaling
of space and time requires a hard breaking of Diff(M) down to Diff(M, F). We can

27 Note that the full set of inequivalent terms up to dimension six has been presented for the projectable

case in [1169, 1170].

140
artificially restore full diffeomorphism invariance by redefining the ADM slicing in terms
of the Stuckelberg field φ(x, t). That is, the slicing goes from

t = constant → φ(x, t) = constant. (433)

The unit normal to the spatial

p surfaces is covariant under Diff(M) and given by the space-
time gradient nµ = ∇µ φ/ −(∇φ)2 . We can now express HL gravity as a relativistic
theory involving the space-time metric, gµν , and the Stuckelberg field [167, 692, 532].
We do this by defining the space-time analogue of the spatial metric and the extrinsic
curvature in terms of the projection tensor qµν = gµν +nµ nν and its Lie derivative 21 Ln qµν
[532]. Violations of the EP can occur because the Stuckelberg field can mediate a force
between matter fields carrying Stuckelberg “charge”. As shown in [692], Stuckelberg
“charge” is a measure of violation of energy-momentum conservation, schematically given
by s 
 ∇µ T µν ∇α Tνα 
Γ ∼   6= 0. (434)
Tµν T µν 
This is allowed by foliation preserving diffeomorphisms which simply require [692]

1 δSm nν ∇µ T µν
qαν ∇µ T µν = 0, and √ = −p . (435)
−g δφ −(∇φ)2

Even if the Stuckelberg charges, Γ1 , Γ2 , . . . are always small, violation of the EP can
still be large since the relevant Eötvös parameter η ∼ ΓΓ11 −Γ
+Γ2 really only cares about the
2

charge ratios.

4.3.1. The projectable theory

We now focus on the projectable version of HL gravity, for which the lapse function
is homogeneous, N = N (t). The action is then given by
Z
1 √  
S = dtd3 x qN Kij K ij − λK 2 − V + Sm , (436)
κG

where the potential V = −c2 R+ higher derivative operators. Since the condition of
projectability is imposed at the level of the theory itself, it follows that the Hamiltonian
constraint is non-local:
Z
√   δSm
d3 x q Kij K ij − λK 2 + V = . (437)
δN
In comparison with GR where the Hamiltonian constraint is local, this admits a much
larger class of solutions. Indeed, it has been suggested that the resulting integration
constant can account for dark matter [924], although this may lead to the formation of
caustics and the break down of the theory [167].

Dark matter as an integration constant

141
 To see how this might emerge, we rewrite the action (436) in the following form
Z  
c4 3 √ Kij K ij − K 2
S= dtd x qN + R + Sm
16πG c2
Z
c2 √
+ (1 − λ) dtd3 x qN K 2 + UV corrections. (438)
16πG
Focusing on the low energy theory, the resulting field equations are [924]
Z  
3 √ 8πG
d x −g Gµν − 2 Tµν nµ nν = O(1 − λ)
(4)
(439)
c
 
(4) 8πG
Giµ − 4 Tiµ nµ = O(1 − λ) (440)
c
(4) 8πG
Gij − 4 Tij = O(1 − λ), (441)
c
(4)
where gµν is the full space-time metric, Gµν is the corresponding Einstein tensor, and
nµ = N1 (1, −N i ) is the unit normal to hyper-surfaces of constant t. The stress energy
tensor, Tµν , is not necessarily conserved, as previously stated. Note that the non-local
Hamiltonian constraint, (439), and the local momentum constraint, (440), are preserved
by the dynamical equations (441).
Now, these equations can be rewritten as follows [924]:
8πG
G(4)
µν =
HL
(Tµν + Tµν ) + O(1 − λ), (442)
c4
HL
R √
where Tµν = ρHL nµ nν and d3 x qρHL = 0. Note that this latter condition does not
require ρHL to vanish at all points in space, and one might wish to identify Tµν HL
with a
µ HL
pressureless fluid moving with 4-velocity n . Taking ρ > 0 in our Hubble patch, we
may associate this integration constant with dark matter [924].
This scenario has been criticised in [167], where it is argued that the cosmological
HL
fluid Tµν will inevitably lead to the formation of caustics and the break down of the the-
ory. To see why this might be the case it is convenient to go to the Stuckelberg p picture,
where we identify the unit normal with the space-time gradient nµ = ∇µ φ/ −(∇φ)2 .
HL
Now Tµν behaves like a pressureless fluid, and in General Relativity it is well known
that this will lead to the formation of caustics, due to the attractive nature of grav-
ity. This is not a problem for real dust, as virialisation can occur. However, in the
scenario of [924], the fluid is characterised by the gradient ∇µ φ which is problematic as
the Stuckelberg field is not differentiable at the caustic. These conclusions have been
disputed in [923] where it is argued that as the putative caustic begins to form we enter
a UV regime and the parameter λ runs away from its IR fixed point at λ = 1. At λ 6= 1
it is claimed that an extra repulsive force could ultimately prevent caustics from forming.

Perturbation theory – ghosts, tachyons, and strong coupling

Let us now consider linearised perturbations about a Minkowski background. In what

follows we will work in units where the emergent speed of light is given by c = 1. Given
142
the residual diffeomorphism, in Eqs. (422) to (424), we can choose a gauge defined by
N = 1, Ni = ∂i B + ni , and qij = (1 + 2ζ)δij + hij , (443)
where ni is a divergence-free vector, and hij is a transverse-tracefree tensor. To study
the propagating degrees of freedom we neglect the matter contribution, and integrate out
the constraints. As it is non-local, the Hamiltonian constraint does not affect the local
propagating degrees of freedom. Meanwhile, the momentum constraint yields
1
B=− ζ̇, and ni = 0, (444)
c2s ∆
where
1−λ
c2s = . (445)
3λ − 1
This will shortly be identified with the scalar speed of sound at low energies. Plugging
this into the action and expanding to quadratic order one finds [1170, 926]
Z  
1 1   1
S= dtd3 x − 2 ζ̇ 2 + ζOs ζ + ḣ2ij + hij Ot hij + Sint , (446)
8πG cs 8
where Sint denotes the interactions and
 
∆2 ∆3 ∆2 ∆3
Os = c2s ∆ + 2 + 4 , Ot = ∆ + µ 2 +ν 4 , (447)
kU V kU V kU V kU V
where µ and ν are dimensionless parameters of order one. Here we assume that all
Lorentz
√ symmetry breaking terms in V4 and V6 depend on roughly the same scale, kU V .
1/ 8πG. The dispersion relation for the scalar is given by
 
k4 k6
w2 = c2s k 2 + 2 + 4 . (448)
kU V kU V
Now we see a problem: At low energies, k  kU V , we require c2s > 0 to avoid a tachy-
onic instability, where we identify cs with the speed of propagation of the scalar waves.
However, as we see from the action (446), c2s > 0 yields a ghost, which is far more trou-
bling. We therefore take c2s < 0, but with |cs | being small so as to render the tachyonic
instability mild. But how small does |cs | need to be? The timescale of this instability
is ts ∼ 1/|cs |k > 1/|cs |kU V , and as this would need to exceed the age of the universe,
we infer |cs | < H0 /kU V , where H0 is the current Hubble scale [167]. Furthermore, since
modifications to Newton’s law have been tested down to the meV scale, we may im-
pose kU V > meV. This gives |cs | < 10−30 , or equivalently |1 − λ| < 10−60 , on scales
H0  k  kU V .
What about the interaction terms? At low energies, k  kU V , and working up to
cubic order we find that [735, 1166]
Z 
1 2 ∂i
Sint = dtd x ζ(∂i ζ)2 − 4 ∂t ζ∂i ζ ∂t ζ
3
8πG c ∆
" s 2   #)
3 1 ∂i ∂j 2 1
+ ζ ∂t ζ − + 4 ζ(∂t ζ)2 + . . . . (449)
2 c4s ∆ c2s cs
143
To estimate the strength of these interactions we canonically normalise the quadratic
term (446) in the infra-red by redefining

t̂ p
t= , and ζ= 8πG|cS |ζ̂, (450)
|cs |

so that
s Z 
8πG ∂i
Sint = dt̂d3 x c2s ζ̂(∂i ζ̂)2 − 2∂t̂ ζ̂∂i ζ̂ ∂t̂ ζ̂
|cs |3 ∆
"  2 #)
3 ∂i ∂j 2

2
+ ζ̂ ∂ ζ̂ − 2cs + 1 ζ̂(∂t̂ ζ̂) + . . . . (451)
2 ∆ t̂

p |cs |, we see that the largest cubic interactions become strongly coupled
For small √ at a scale
Λsc ∼ |cs |3 /8πG. Imposing the constraint, |cs | < 10−30 and taking 1/ 8πG ∼ 1018
GeV, we find Λsc . 10−18 eV. This lies well below scale of the UV corrections given
by kU V > meV so we can certainly trust the effective low energy description we have
used to derive this scale. The implications for the theory are profound. The scale
Λsc represents the scale at which perturbative quantum field theory breaks down in
Minkowski space. For scattering processes above Λsc . 10−18 eV we must sum up the
contribution from all multi-loop diagrams. Since the claims of renormalisability are based
on the validity of the perturbative description at all energies, we see that much of the
motivation for studying this theory is lost. We also note that any notion of Minkowski
space is meaningless below distances 1/Λsc & 108 km since one would require a scattering
process at energies above Λsc to probe its structure. In analogy with DGP gravity
(see section 5.5) one might hope to raise the scale of strong coupling by considering
fluctuations on curved backgrounds, for example, on the background gravitational field
generated by the Sun [926]. However, this seems optimistic since Minkowski space is
an excellent approximation28 to thepbackground geometry at distances of order 1/Λsc &
108 km, so the derived scale, Λsc ∼ |cs |3 /8πG should still be reliable.

4.3.2. The non-projectable theory

We now consider the non-projectable theory for which the lapse function can depend
on space, N = N (x, t), just as in General Relativity. This means the Hamiltonian con-
straint is now local and that terms depending on ai = ∇~ i log N could play an interesting
role in the dynamics [169, 168]. Just as in the projectable case, the absence of full dif-
feomorphism invariance allows an extra scalar mode to propagate [278]. The action is
given by
Z   Z
c4 √ Kij K ij − K 2 c2 √
S= dtd3 x qN 2
+ R + α dtd3 x qN ai ai
16πG c 16πG
Z
c2 √
+ (1 − λ) dtd3 x qN K 2 + Sm + UV corrections. (452)
16πG

28 At distances r & 108 km from the Sun, the Newtonian potential V (r) . 10−8 .
144
Again we consider small vacuum perturbations on a Minkowski background, working in
units where c = 1. Given the reduced set of diffeomorphisms, Eqs. (422) to (424), we
cannot gauge away the fluctuations in the lapse function that depend on space. Instead,
we choose a gauge

N = 1 + χ, Ni = ∂i B + ni , and qij = (1 + 2ζ)δij + hij , (453)

where, as before, ni is a divergence free vector, and hij is a transverse-tracefree tensor.

Now integrating out the momentum constraint, one finds
1
B=− ζ̇, ni = 0,
c2s ∆

where c2s is given by equation (445), but, as we will see, we do not identify it with the
speed of sound. The Hamiltonian constraint yields [169, 1025]
2
χ = − ζ.
α
Expanding to quadratic order gives [169, 1025]
Z  
1 3 1  2  1
2 ij
S= dtd x − 2 ζ̇ + ζ Õs ζ + ḣ + h Õt hij + Sint , (454)
8πG cs 8 ij
where Sint denotes the interactions and
 
2 α−2 ∆2 ∆3 ∆2 ∆3
Õs = cs ∆+ 2 + 4 , and Õt = ∆ + µ̃ 2 + ν̃ 4 , (455)
α kU V kU V kU V kU V

where µ̃ and ν̃ are dimensionless parameters of order one. Again we assume that all
Lorentz
√ symmetry breaking terms in V4 and V6 depend on roughly the same scale, kU V .
1/ 8πG. We now write down the dispersion relation for the scalar
 
α−2 2 k4 k6
w2 = c2s k + 2 + 4 . (456)
α kU V kU V
It follows that the speed of propagation of the scalar waves in the non-projectable theory
are given by  
α−2
c̃2s = c2s . (457)
α
The ghost and the tachyon problems can now be avoided by simultaneously taking [169,
1025]
c2s < 0, and 0 < α < 2. (458)
Recall that this was not possible in the projectable case, where we had to accept the
tachyonic instability and use the fact that this should be slow relative to a Hubble time
to place strong bounds on |1 − λ|. Since in the non-projectable case we no longer have
such concerns regarding tachyonic instabilities, the strongest bounds on |1 − λ| and |α|
come from preferred frame effects in the Solar System, requiring [1272]

|1 − λ|, |α| . 10−7 . (459)

145
For λ and α satisfying this bound, the speed of propagation of the scalar is
λ−1
c̃2s ≈ ,
α
which should not be too slow. If the scalar graviton fluctuations propagated significantly
slower than other fields to which they couple, then those fields would decay into scalar
gravitons via the Cerenkov process. This would be particularly worrying for light since
it would result in photon decay and the absence of cosmic rays. To avoid this problem
we take c̃2s ∼ 1, and so |1 − λ| ∼ |α|. Solar system constraints on these theories are
considered further in [643, 644, 645], based on the solutions presented in [684], and black
holes are studied in [226, 227, 232].
What about the interactions? Again, focusing on the low energy theory we find
that thepquantum fluctuations
p on Minkowski space become strongly coupled at the scale
Λ̃sc ∼ |λ − 1|/8πG ∼ |α|/8πG [1025, 692]. Note that this result can be derived using
a direct method similar to the one presented for the projectable case in the previous
section [1025], or using the Stuckelberg method in the decoupling limit [692]. Now given
the bounds in Eq. (459), it follows that the strong coupling scale is Λ̃sc . 1015 GeV.
If kU V > Λsc we can trust our low energy description, and the derivation of this scale.
As explained in the projectable case, strong coupling casts serious doubts on the claims
of renormalisability as these rely on the validity of perturbative quantum field theory.
However, if kU V < Λsc , we cannot trust our derivation of the strong coupling scale, since
the low energy description would not be valid there [168]. In this scenario, new physics
that softens the interactions kicks in at k ∼ kU V . This situation is reminiscent of the
case in string theory where we introduce the string scale just below the Planck scale,
where strong coupling would otherwise occur.
Whilst this seems promising there are some issues facing the non-projectable the-
ory. One of these relates to the formal structure of the theory, and in particular the
constraint algebra, which is dynamically inconsistent. This manifests itself through the
lapse function vanishing asymptotically for generic solutions to the constraint equations
[603]. The asymptotically flat solutions we have just discussed represent a non-generic
subset of measure zero in the space of all solutions.
In the quantum version of the theory, its has been claimed that one must take λ < 1/3
in order to have a stable vacuum [1144]. This is incompatible with phenomenological
requirements for the following reason: We expect there to be 3 fixed points in the renor-
malisation group flow for λ. These are λ = 1 (diff invariance), λ = 1/3 (conformal
invariance) and λ = ∞ [619]. Now, at low energies we require |λ − 1| < 10−7 and
1−λ
c2s = 3λ−1 < 0. This suggests that λ flows from infinity in the UV to λ = 1 in the IR.

4.3.3. Aspects of Hořava-Lifschitz cosmology

HL gravity was first applied to cosmology in [698, 235]. As we have seen, in the UV,
the relevant degrees of freedom have an anisotropic dispersion relation w2 ≈ k 6 /kU 4
V.
This is often at the root of much of the interesting cosmology that has subsequently
arisen, including (i) a scale invariant spectrum of cosmological perturbations, without
early time acceleration [925, 284], (ii) cosmological bounces [179, 234, 522, 854, 351],
(iii) dark matter as an integration constant [924], (iv) chirality of primordial gravity
waves [1196] and (v) enhancement of baryon asymmetry, abundance of gravity waves,
dark matter, and so on [927]. These latter effects occur because the modified dispersion
146
relation results in radiation scaling like 1/a6 as opposed to 1/a4 in the UV regime. We
refer the reader to the following review articles on this subject [926, 1098].
For an FLRW universe, the background cosmology in both the projectable and non-
projectable theories is qualitatively very similar. Choosing units where the emergent
speed of light is c = 1, the Friedmann equation takes the following form:
 
3λ − 1 2 8πG C(t)
H = ρ+ 3 + V (a), (460)
2 3 a

where "    2 #
κ κ κ
V (a) = 2 1 + m 2 a2 +n 2 a2 , (461)
a kU V kU V

and where m and n are dimensionless parameters of order one. Here κ = 0, ±1 is the
spatial curvature. We assume that the matter component with energy density, ρ, and
pressure, P , satisfies the usual energy-conservation law, ρ̇ + 3H(ρ + P ) = 0, although
this is not necessarily required in HL gravity, as we have already discussed.
The contribution from C(t)/a3 depends on the theory in question. For the projectable
theory it corresponds to the “dark matter integration constant” [924], with C(t) →
constant at low energies. For the non-projectable theory there is no such contribution
and C(t) ≡ 0.
We immediately notice that the effective Newton’s constant seen by cosmology differs
from the one derived by comparing the low energy effective action to the Einstein-Hilbert
action:
2
Gcosmo = G. (462)
3λ − 1
Although as λ → 1, at low energies, we see that Gcosmo → G.
To see how HL cosmology can admit a bounce, consider the limiting behaviour of the
right hand side of Eq. (460). Neglecting C(t)/a3 and assuming ρ scales like 1/a3 or 1/a4
−4


2 3
we see that this goes like ∼ nkU V κ/a , which is negative if nκ < 0. By continuity,
this suggests that there exists a∗ for which the right-hand side of Eq. (460) is zero at
a = a∗ . This corresponds to the position of the bounce, since at this point H = 0.
Observational constraints on |λ−1| coming from the background cosmology have been
studied in [445, 446] using BAO+CMB+SN1a, but they are not particularly strong.
At 1σ confidence level they find that |λ − 1| . 0.02, which is far weaker than the
bounds presented in previous sections. Recall that in the projectable theory, stability
considerations require |λ − 1| . 10−60 , whereas in the non-projectable theory preferred
frame effects require |λ − 1| . 10−7 .
Cosmological perturbations in HL gravity have also been considered (see, for example,
[925, 284, 521, 1257, 712, 1258, 713, 1256, 632, 652]). Indeed, for the projectable theory, it
has been claimed that scalar fluctuations on cosmological backgrounds are stable. This is
in contrast to the corresponding fluctuations on Minkowski space [632] which are known
to suffer from either a ghost or tachyonic instability, as we saw in section 4.3.1. Whilst
this may be relevant to long wavelength modes, it is of no consequence on sub-horizon
scales where we can trust the perturbative analysis about Minkowski space, to a good
approximation. Also, gravity waves produced during inflation have been found to be
chiral in HL gravity, thereby representing a robust prediction of the theory [1196].
147
4.3.4. The ΘCDM model
HL gravity represents the UV completion of an interesting cosmological model, dubbed
ΘCDM [171]. In this model, it is assumed that the old cosmological constant problem
is solved in some way, such that the net contribution to the cosmological constant is
vanishing. The model then seeks to explain the tiny, but non-zero, amount of cosmic
acceleration that is currently observed, without any fine tuning. Indeed, it is shown
that the model allows for a technically natural small contribution to cosmic acceleration,
without any corrections from other scales in the theory.
A key assumption corresponds to the fact that Lorentz invariance is broken in the
gravitational sector. Thus the theory contains a unit time-like vector field which may be
generic (as in Einstein-Aether theory) or expressed in terms of the gradient of a scalar
field defining a global time (sometimes called the khronon29 ). The proposed acceleration
mechanism appears generically when we assume the existence of another field, Θ, which
is taken to be invariant under shift transformations. The model is a valid effective field
theory up to a high cut-off just a few orders of magnitude below the Planck scale, with
a UV completion offered by HL gravity in the khronon case.
In the absence of any matter sources (including the cosmological constant) the model
possesses two solutions corresponding to Minkowski and de Sitter space-times. The
former solution is unstable and the presence of an arbitrarily small amount of matter
destroys it. The cosmological evolution of a matter-filled universe is driven to the de
Sitter attractor, with effective equation of state w = −1. The value of the effective
cosmological constant on the de Sitter branch is determined by the lowest dimension
coupling between the Goldstone field and the khronon. Remarkably, it is technically
natural to assume this coupling to be small as it is protected from radiative corrections
by a discrete symmetry. Thus, in the absence of a contribution from the cosmological
constant, the current value of cosmic acceleration would not present any fine-tuning
problem.
Interestingly, the evolution of cosmological perturbations is different in the ΘCDM
and ΛCDM models. In particular, the growth of linear perturbations is enhanced in
ΘCDM as compared to the standard ΛCDM case. The enhancement is most promi-
nent at very large scales of order a few gigaparsecs, but extends also to shorter scales.
Another difference is the appearance of an effective anisotropic stress, resulting in a non-
trivial gravitational slip at very large scales. In principle, these effects may allow one to
discriminate between ΘCDM and ΛCDM in the near future.

4.3.5. HMT-da Silva theory

We have discussed in Sections 4.3.1 and 4.3.2 how the original versions of HL gravity
are plagued with problems at the level of both theory and phenomenology. The root
of this is the breaking of diffeomorphism invariance and the additional scalar degree of
freedom that propagates as a result. With this in mind, Hořava and Melby-Thompson
(HMT) proposed a modified version of the projectable theory possessing an additional
U (1) symmetry [621]. It is claimed that this extra symmetry removes the troublesome
scalar degree of freedom, so that one is left with a spin-2 graviton as the only propagating

29 The khronon field is naturally identified with the Stuckelberg mode in HL gravity at low energies.

148
mode. The HMT action is given by
Z
1 √ n h ~ k Rij )
SHM T = dtd3 x q N Kij K ij − K 2 − V (qij , Rij , ∇
κG
i o
+νΘij (2Kij + ∇ ~ i∇
~ j ν) − A(R − 2Ω) , (463)

where Θij = Rij − 21 Rq ij + Ωq ij , and Ω is a dimensionful coupling constant that can

run with scale. This constant controls the scalar curvature of spatial slices, and can be
thought of as a second cosmological constant. As with the original projectable theory,
we assume N = N (t), with a potential V = −c2 R + . . . containing the usual terms up to
dimension six in order to guarantee the z = 3 scaling in the UV. In addition, however,
HMT theory contains two new fields given by A = A(x, t) and ν = ν(x, t). These are
important in extending the symmetry group to U (1) × Diff(M, F).
The action is invariant under Diff(M, F), with

δN = ∂t (N f ) (464)
i i i i
δN = ∂t (N f + ξ ) + Lξ N (465)
δqij = f ∂t qij + Lξ qij (466)
δA = ∂t (Af ) + ξ i ∂j A (467)
i
δν = f ∂t ν + ξ ∂j ν. (468)

Note that ν transforms as a scalar, whereas A transforms like a spatial scalar and a
temporal vector. Indeed, A transforms exactly as the lapse function would in a non-
projectable theory. This is not a coincidence. One can think of A as being the next to
leading order term in the non-relativistic expansion of the lapse. Of course, one ought
to ask why we have not included the parameter λ in front of the K 2 term in the action,
as in previous versions of HL gravity. According to [621], the parameter λ is fixed to be
equal to one by requiring the action to be invariant under a local U (1) symmetry:
~ iψ
δA = ψ̇ − N i ∇ (469)
δν = −ψ (470)
δN i ~ i ψ.
= N∇ (471)

It is this symmetry that removes the scalar graviton. Furthermore, fixing λ = 1 ensures
no conflict with observational tests of Lorentz violation at low energies. The HMT model
has been applied to cosmology in [1259].
Recently, da Silva has argued that in contrast to the claims of [621], one can ac-
commodate λ 6= 1 and still retain the U (1) × Diff(M, F) invariance [353]. Indeed, he
proposed the following action:
Z
1 √ n h ~ k Rij )
SdaSilva = dtd3 x q N Kij K ij − λK 2 − V (qij , Rij , ∇
κG
i o
+νΘij (2Kij + ∇~ i∇
~ j ν) + (1 − λ)[(∆ν)2 + 2K∆ν] − A(R − 2Ω) , (472)

which is invariant under the same symmetries as (463). Now we must subject λ to the
same constraints as before, in particular those coming from preferred frame effects. There
149
are claims that the extra symmetry will eliminate the scalar graviton even when λ 6= 1
[353], although more detailed study is required to be sure. Preliminary investigations on
this subject have been carried out in [631]. In any event, strong coupling problems have
recently been shown to infect the matter sector of this theory [811], unless one introduces
a low scale of Lorentz violation, in a way that is reminiscent of [169, 168].

4.4. Galileons
Galileon theory [963] was originally developed by Nicolis et al. to facilitate a model
independent analysis of a large class of modified gravity models. In each case, General
Relativity on perturbed Minkowski space is modified by an additional single scalar field,
the galileon, with derivative self-interactions. Although the galileon and the graviton
both couple to matter, any direct coupling between them is neglected to leading order.
The resulting vacuum Lagrangian is invariant under the following shift in the galileon
field
π → π + bµ xµ + c. (473)
This symmetry corresponds to a generalisation of Galilean invariance, hence the name.
The inspiration for the model comes from DGP gravity [454]. In Section 5.5, we will see
how the boundary effective theory on the DGP brane is well described by the following
action, Z
DGP
 
Seff = d4 x LGR + LDGPπ , (474)

where
    
1 1 µν 2 1 1
LGR = h̃ ∂ h̃µν − h̃ηµν + . . . + h̃µν T µν , (475)
16πG 4 2 2
 
1 1  1
LDGP
π = 3π∂ 2 π − rc2 (∂π)2 ∂ 2 π + πT. (476)
16πG 2 2

The Lagrangian (474) has two components: a linearised GR piece, LGR , and a modifica-
tion due to the brane bending mode, LDGP π . It is valid in the so-called decoupling limit
in which all interactions go to zero except the scalar self-interactions. Focusing on the
π-Lagrangian, LDGP
π , Nicolis et al. observed that the vacuum field equations are built
exclusively out of second derivatives, ∂µ ∂ν π. In particular, this means that there are
no terms higher than second order, ensuring a well defined Cauchy problem and avoid-
ing any of the potential problems arising from ghosts in higher derivative theories. In
addition there are no first or zero derivative terms which means that the π Lagrangian
possesses the Galilean symmetry. This is inherited from Poincaré invariance in the bulk
[1017].
One might expect that almost any co-dimension one braneworld model with large
distance deviations from GR will be described, in part, and in some appropriate limit,
by a generalised π Lagrangian possessing the Galilean symmetry. This essentially follows
from the fact that the extrinsic curvature of the brane is Kµν ≈ ∂µ ∂ν π, on scales where
we can neglect background curvature.
We should also note that even if there is no direct coupling to matter and therefore
no modification of gravity, galileons are of interest in their own right as a source of
energy-momentum. In particular, one can potentially obtain violations of the null energy
150
condition without introducing any instability [964, 343]. Generically, however, a single
galileon will result in superluminality, although the situation may be improved by going
to multi-galileon theory (see section 4.4.5).

4.4.1. Galileon modification of gravity

To see how to generalise the decoupling limit of DGP to a larger class of modified
gravity theories, let us consider the amplitude, A, for the exchange of one graviton
0
between two conserved sources, Tµν and Tµν . In General Relativity, this amplitude is
given by  
16πG 0µν 1 0
AGR = Tµν T − TT , (477)
p2 2
where T = Tµµ . We are interested in the case where gravity is modified by an additional
scalar, so that locally we have
1
δA = A − AGR = − T T 0. (478)
αp2
Such a theory can be described by the following action
Z
1 p α 1
S = d4 x −g̃R(g̃) − π∂ 2 π + h̃µν T µν + πT + interactions, (479)
16πG 2 2

where g̃µν = ηµν + h̃µν . The fluctuation h̃µν is identified with the GR graviton, and
as such, for a given source and boundary conditions, it coincides with the linearised
solutions of GR. This statement is true to all orders in the “decoupling” limit
α T µν
Mpl , α, T µν → ∞, 2 = const, and = const, (480)
Mpl Mpl
p
where Mpl = 1/8πG. Note that matter is minimally coupled to the metric gµν =
ηµν + hµν , where the physical graviton is hµν = h̃µν + 2πηµν .

Galileon Action and Equations of Motion

Now suppose we consider the decoupling limit (480) with the additional assumption
that the strength of some of the scalar self-interactions can be held fixed. This amounts
to neglecting the back-reaction of the scalar onto the geometry so that we can consider
it as a field on Minkowski space. We retain some of the scalar self-interactions for the
following reason: we are interested in an O(1) modification of GR on cosmological scales,
but we would like this to be screened down to . O(10−5 ) on solar system scales. As we
will see, the derivative self-interactions can help shut down the scalar at short distances
through Vainshtein screening30 . In the decoupling limit, the action is given by
h i Z
S h̃µν , π = d4 xLGR + Lπ , (481)

30 See section 5.5.4 for a detailed discussion of the Vainshtein mechanism in DGP gravity.
151
where Lπ = Lgal (π, ∂π, ∂∂π) + πT represents the generalisation of the π-Lagrangian in
DGP gravity.
The vacuum part of the generalised π-Lagrangian, Lgal (π, ∂π, ∂∂π) gives second order
field equations, and is assumed to be Galilean invariant in the sense that Lgal → Lgal +
total derivative, when π → π + bµ xµ + c. What is the most general Lagrangian with this
property? The answer is remarkably simple, and in four dimensions is given by [963, 400]
5
X
Lgal (π, ∂π, ∂∂π) = ci Li (π, ∂π, ∂∂π), (482)
i=1

where the ci are constants, and31

L1 = π (483)
1
L2 = − (∂π)2 (484)
2
1
L3 = − ∂ 2 π(∂π)2 (485)
2
1 2 2 
L4 = − (∂ π) − (∂∂π)2 (∂π)2 (486)
2
1 
L5 = − (∂ 2 π)3 − 3(∂ 2 π)(∂∂π)2 + 2(∂∂π)3 (∂π)2 . (487)
2
By construction, the variation of each component is built exclusively out of second deriva-
tives, Z
δ
d4 xLi (π, ∂π, ∂∂π) = Ei (∂∂π),
δπ
where
µ1 µi−1 ν1 νi−1
Ei (∂∂π) = (i − 1)!δ[ν 1
. . . δνi−1 ] (∂µ1 ∂ π) . . . (∂µi−1 ∂ π).
Specifically,

E1 = 1 (488)
E2 = ∂2π (489)
E3 = (∂ 2 π)2 − (∂∂π)2 (490)
2 3 2 2 3
E4 = (∂ π) − 3∂ π(∂∂π) + 2(∂∂π) (491)
E5 = (∂ 2 π)4 − 6(∂ 2 π)2 (∂∂π)2 + 8∂ 2 π(∂∂π)3 + 3[(∂∂π)2 ]2 − 6(∂∂π)4 . (492)

It follows that the field equations for the galileon model are therefore given by the fol-
lowing:
 
1 2 1
− ∂ h̃µν − h̃ηµν + . . . = 8πGTµν , (493)
2 2
5
X
ci Ei (π, ∂π, ∂∂π) = −T. (494)
i=1

31 We define (∂∂π)n = (∂ α α α
α1 ∂ 2 π)(∂α2 ∂ 3 π) . . . (∂αn ∂ 1 π). Note that we have presented the simpler
expressions as suggested by [400].
152
Equation (493) corresponds to the linearised Einstein equations , and so their solution,
h̃µν , corresponds to the standard GR solution for a given source and boundary condi-
tions. The modification of GR is encoded entirely in the solution of the scalar equation
of motion, (494).

Galileon cosmology as a weak field

The galileon theory has been constructed in terms of a tensor and a scalar propa-
gating on a Minkowski background. Whilst it is straightforward to understand the weak
gravitational field in the solar system using this description, it is not clear how one should
describe cosmology. Fortunately, at distances below the curvature scale any metric is well
approximated by a local perturbation about Minkowski space. In what follows, local will
mean local in both space and time, which for cosmological solutions will correspond to
sub-Hubble distances and sub-Hubble times.
Let us consider a spatially flat FLRW space-time. If we take our position to be given
by ~x = ~y = 0 and t = τ = 0, then for |~x|  H −1 and |t|  H we have [963]
 
1 1
ds2 = −dτ 2 + a(τ )2 d~y 2 ≈ 1 − H 2 |~x|2 + (2Ḣ + H 2 )t2 (−dt2 + d~x2 ), (495)
2 2

where the Hubble scale H and its time derivative Ḣ are evaluated now. We recognise
this as a perturbation on Minkowski space in Newtonian gauge,

ds2 ≈ −(1 + 2Ψ)dt2 + (1 − 2Φ)d~x2 , (496)

32
where the Newtonian potentials are
1 1
Ψ = − H 2 |~x|2 + (2Ḣ + H 2 )t2 , and Φ = −Ψ . (497)
4 4
For a given cosmological fluid, the corresponding GR solutions have Hubble parameter
HGR . Since h̃µν agrees with the linearised GR solution, we have h̃tt = −2ΨGR , and
h̃ij = 2ΨGR δij , where

1 2 1
ΨGR = − HGR |~x|2 + (2ḢGR + HGR
2
)t2 . (498)
4 4
Now in our modified theory the physical Hubble parameter (associated with hµν ) differs
from the corresponding GR value, H 6= HGR . Since hµν = h̃µν + 2πηµν , we have a
non-trivial scalar
π = Ψ − ΨGR . (499)
Note that a Galilean transformation π → π +bµ xµ +c merely corresponds to a coordinate
transformation xµ → xµ − cxµ + 12 (xν xν bµ − 2bν xν xµ ) in the physical metric.

32 It may look like that since Φ − Ψ = 2Φ 6= 0 that this construction introduces anisotropic stress.

This is not the case, however, as strictly speaking the condition for the absence of anisotropic stress is
Di j (Φ − Ψ) = 0. It is easy to show that Di j |~
x|2 = 0, hence, no anisotropic stress is present.
153
Self-accelerating solutions

Of particular interest are self-accelerating solutions. A self-accelerating vacuum is

one that accelerates even in the absence of any sources for the fields h̃µν and π. These
are familiar to us from DGP gravity (see Section 5.5), where the self-accelerating solution
is haunted by ghosts. Is the same true in a general galileon scenario? Or can we identify
a scenario that admits a consistent self-accelerating solution?
There is someR ambiguity as to what is actually meant by ‘self-acceleration’ if the
tadpole term d4 xc1 π is present in the galileon Lagrangian. The point is that at the
level of the graviton equations of motion, the source corresponds to the vacuum
R energy, λ.
However, at the level of the scalar equations of motion the tadpole term, d4 x c1 π, has
the effect of renormalising the vacuum energy seen by the π field, λ → λ + c1 . Indeed,
in a braneworld context, one might associate the tadpole with the vacuum energy in
the bulk. To avoid considering a simple cosmological constant, we set the bare vacuum
energy λ = 0, and require the π-tadpole term to vanish, c1 = 0. This guarantees that
Minkowski space is a solution for the physical metric since the field equations are solved
by h̃µν = 0, and π = 0. Note that this Minkowski solution need not be stable, as
in the ghost condensate scenario. On the contrary, our interest is in stable de Sitter
solutions. Given the constraints λ = 0, and c1 = 0, any de Sitter solution is necessarily
self-accelerating.
We now consider maximally symmetric vacua in the absence of vacuum energy, λ = 0,
and the tadpole, c1 = 0. The corresponding GR solution is always Minkowski space,
with h̃µν = 0. Non-trivial solutions for the scalar π = π̄(x), however, could give rise
to self-acceleration. A self-accelerating vacuum with de Sitter curvature H 2 would have
π̄ = − 14 H 2 xµ xµ . Plugging this into the field equations (494), with c1 = 0, and T = 0,
gives [963]
3
− 2c2 H 2 + 3c3 H 4 − 3c4 H 6 + c5 H 8 = 0. (500)
2
Clearly non-trivial solutions exist for suitable choices of the parameters ci , so self-
accelerating solutions also exist.
Are these vacua consistent? To investigate this we need to consider fluctuations h̃µν ,
and δπ = π − π̄ about the self-accelerating vacuum. Because of the Galilean symmetry,
the galileon structure is preserved in the effective theory describing fluctuations,
h i Z
S h̃µν , δπ = d4 xLGR + Lδπ ,

P5
where Lδπ = i=1 di Li (δπ, ∂δπ, ∂∂δπ) + δπT . The coefficientsP can be obtained from the
coefficients in the underlying theory via a linear map di = i Mij cj , where the matrix
Mij depends on the background curvature, H 2 [963].
There are two immediate things to consider: (i) does the spectrum of fluctuations
contain a ghost, and (ii) does the scalar get screened on solar system scales? For a
general galileon theory, to avoid the ghost we must choose parameters such that d2 > 0.
In DGP gravity, where the ghost is known to be present on the self-accelerating branch,
one would find dDGP
2 < 0 (see Section 5.5.3 for further details).
In order to screen the scalar at solar system scales one must appeal to the Vainshtein
mechanism. Again, we discuss the Vainshtein mechanism in detail in the context of DGP
154
gravity in Section 5.5.4. The mechanism works in exactly the same way in a general
galileon theory. For simplicity we assume spherical symmetry for fluctuations on the
self-accelerating background and consider the profile outside of a heavy non-relativistic
source, Tµν = diag(ρ(r), 0, 0, 0). Now, it is well known that the GR solution is given by
the standard Newtonian potential
GM
|h̃µν (r)| ∼ , (501)
r
R
where the mass of the source M = ρ(r)dV . The galileon solution δπ(r) is given by
[963, 223]
 0  0 2  0 3
δπ δπ δπ M
d2 + 2d3 + 2d4 = . (502)
r r r 4πr3
Note that E5 is identically zero when evaluated on a spherically symmetric field. At large
distances one can neglect the higher order terms in Eq. (502) and derive the linearised
solution
M
δπ lin (r) = − . (503)
4πd2 r
Now |h̃µν (r)| ∼ |δπ lin (r)| so we have an O(1) modification of GR. At shorter distances,
the non-linear terms in Equation (502) become important and start to dominate. This
happens at the so-called ‘Vainshtein radius’, given by [223]
( 1/3  1/6 )
d3 M d4 M 2
rV ∼ max , .
d22 d32

Depending on which of the non-linear terms dominates, the profile of the galileon field
changes to
 
1/2 √

 M
nonlin d3 r if the term with d3 dominates
δπ (r) ∼  1/3

 M
d4 r if the term with d4 dominates

For a suitable choice of parameters one can have |h̃µν (r)|  |δπ nonlin (r)| on solar system
scales, and might claim that the modification of gravity does indeed get screened. How-
ever, it is important to note that the Vainshtein mechanism itself has yet to be properly
understood in a well defined and fully covariant theory. We discuss some aspects of this
at the end of Section 5.5.4.
Nonetheless, our galileon analysis indicates that self-accelerating solutions that are
ghost-free and exhibit some form of Vainshtein screening on solar system scales could
exist. However, there are also other concerns. Firstly, we should consider the question
of back-reaction. Our galileon description holds provided we can neglect the effect of the
scalar field back onto the geometry. This turns out not to be problematic provided we
2
take |di | . Mpl /H 2i−4 [963].
More serious concerns appear when we study fluctuations about the spherically sym-
metric solutions we have just described. These can cause problems at both the classical
and the quantum levels. At the quantum level, one must identify the scale at which
155
the quantum fluctuations become strongly coupled, and the radius at which one can no
longer trust the classical background. As the background solution changes with scale, so
does the strong coupling scale. This means that there exists a critical radius at which
the quantum effects start to dominate and one can no longer trust the classical solution.
Aspects of strong coupling in DGP gravity are discussed in Section 5.5.4. Here we note
that for a general galileon model the critical radius at which the theory enters a quantum
fog can sometimes be unacceptably large33 . At the classical level, we find that fluctu-
ations at short distances can sometimes propagate extremely slowly, so much so that a
huge amount of Cerenkov radiation would be emitted as the earth moves through the
solar galileon field. Indeed, to simultaneously avoid problems with Cerenkov emission
and a low scale of strong coupling in a ghost-free theory with self-accelerating solutions,
one must introduce a tadpole. As we have already explained, this could be considered
undesirable as a tadpole will renormalise the vacuum energy seen by the galileon.
Yet another problem concerns radial fluctuations at large distances. These can prop-
agate at superluminal speeds, indicating problems for causality34 . This is known to be
a problem in DGP gravity [9, 609] and can only be avoided in the general case by elimi-
nating all of the interaction terms. This is unacceptable since the interaction terms are
crucial to the successful implementation of the Vainshtein mechanism.
In summary then, while it is possible to obtain self-acceleration in a general galileon
model that avoids some of the problems facing DGP gravity, one cannot find a com-
pletely consistent scenario. However, the situation can be improved by the introduction
of a second galileon [1018], as we will discuss in Section 4.4.5.

Conformal galileon

The conformal galileon is constructed in much the same way as the pure galileon
we have just described, except now we demand that the relevant vacuum Lagrangian,
Lconformal
gal (π, ∂π, ∂∂π), is invariant under the conformal group:

dilations : π(xµ ) → π(bxµ ) + log b (504)

µ µ µ
translations : π(x ) → π(x + a ) (505)
µ µ ν
boosts : π(x ) → π(Λ ν x ) (506)
special conformal
: π(xµ ) → π(xµ + cµ |x|2 − 2(c · x)xµ ) − 2c · x. (507)
transformations

33 Larger than the Schwarzschild radius of the Sun

34 Note, however, that it has been suggested that causal paradoxes associated with superluminality do
not always manifest themselves in theories with non-linear scalar interactions [73].

156
 P
It turns out that Lconformal
gal = conformal
i ci Li , where [963, 389]35

Lconformal
1 = e4π (508)
1
Lconformal
2 = − e2π (∂π)2 (509)
2 
1 1
Lconformal
3 = − ∂ 2 π + (∂π)2 (∂π)2 (510)
2 2
1
Lconformal
4 = e−2π L4 − e−2π (∂π)2 {4(∂π) · (∂∂π) · (∂π)
20
−4(∂π)2 ∂ 2 π + 3[(∂π)2 ]2 (511)
(
1
Lconformal
5 = e −4π
L5 + 3e −4π
(∂π) L4 + [(∂π)2 ]3
2
56
)
5
+ (∂π)2 [(∂π)2 ∂ 2 π − (∂π) · (∂∂π) · (∂π)] . (512)
7

Aspects of the conformal galileon model are studied in [343, 964]. Violations of the null
energy condition here can drive inflationary expansion without introducing instabilities.
There are, however, some issues with superluminality.
A supersymmetric version of the conformal galileon has been obtained in [687] as a
consistent completion of the supersymmetric ghost condensate.

4.4.2. Covariant galileon

The galileon action (481) describes fields propagating on a Minkowski background,
and does not represent a fully covariant theory. Although galileon theory was originally
motivated by co-dimension one braneworld models, it is interesting to consider the four
dimensional covariant completion of the theory in its own right. This has been worked
out in [400, 396], and is given by
Z  
4 √ 1 cov
S[gµν , π] = d x −g R + Lgal + Smatter [ĝµν , ψi ]. (513)
16πG

Naturally
R 4 we recognise the first term in brackets as the standard Einstein-Hilbert action,
1 √
16πG d x −gR. The last term corresponds to the matter action. Note that the matter
fields are minimally coupled to the metric ĝµν = f (π)gµν , where the conformal factor
depends on π. An obvious example would be ĝµν = e2π gµν although this is byP no means
a unique choice. Neglecting the tadpole, the covariant completion of Lgal = i ci Li is

35 Our sign conventions agree with [963], rather than [389].

157
 P
given by Lcov
gal =
cov
i ci Li , where36

1
Lcov
2 = − (∇π)2 (514)
2
1
Lcov
3 = − π(∇π)2 (515)
2 
1 1
Lcov
4 = − (π)2 − (∇∇π)2 − R(∇π)2 (∇π)2 (516)
2 4
"
1
Lcov
5 = − (π)3 − 3(π)(∇∇π)2 + 2(∇∇π)3
2
#
−6Gµν (∇µ π)(∇ν ∇α π)∇α π) (∇π)2 . (517)

Note that for the 4th and 5th order terms one must introduce some non-minimal gravi-
tational coupling to π. This is necessary since the naive covariant completion of L4 and
L5 , with minimal couplings, results in equations of motion containing higher derivatives.
The non-minimal coupling helps to eliminate those higher derivatives. Now although
the field equations in our covariant theory remain at most second order in derivatives,
Galilean invariance is broken. We will not present the field equations here since they are
long and complicated, especially for the higher-order terms. The interested reader can
find them in [400], but should be mindful of the fact that the formulae for Lcov
4 and Lcov
5
presented here differ from those in [400] by an overall factor of 4 and 5, respectively.
In a very recent paper, covariant galileon terms are seen to arise in Kaluza-Klein com-
pactifications of Lovelock actions [1242]. This might have been expected since the under-
lying theory has at most second-order fields, and this is inherited by the dimensionally
reduced theory. We discuss aspects of Lovelock gravity, and in particular, Gauss-Bonnet
gravity, in Section 5.7.

4.4.3. DBI galileon

The galileon Lagrangian Lgal can also be obtained from the non-relativistic limit
of a probe brane in five dimensional Minkowski space [389]. The probe brane action
corresponds to a generalisation of the DBI action, as we will now explain. We take our
bulk coordinates to be (xµ , y), and place the probe brane at y = π(x). The induced
metric on the brane is then given by gµν = ηµν + ∂µ π∂ν π, from which we deduce that
the DBI action is
Z Z p
4 √
SDBI = −λ d x −g = d4 x − λ 1 + (∂π)2 . (518)

For a slowly moving brane (∂π)2  1, the leading order dynamical piece goes like
− λ2 (∂π)2 . To generalise this, we first consider objects that transform covariant on the
brane, and then build a Lagrangian from them that gives rises to field equations that are
at most second-order. The relevant covariant objects are the extrinsic curvature, Kµν ,

36 We define (∇∇π)n = (∇ α α α µν ∇ ∇ π, and (∇π)2 =

α1 ∇ 2 π)(∇α2 ∇ 3 π) . . . (∇aαn ∇ 1 π), π = g µ ν
g µν ∇µ π∇ν π.
158
the induced curvature, Rµναβ , and the covariant derivatives of these quantities. The
generalised DBI action required to guarantee second-order field equations is
X
Sgen-DBI = ci Si , (519)
i

where37
Z
√
S2 = − d4 x −g (520)
Z p
→ − d4 x 1 + (∂π)2 (521)
Z
√
S3 = d4 x −gK (522)
Z
 
→ − d4 xγ ∂ 2 π − γ 2 (∂π) · (∂∂π) · (∂π) (523)

and
Z
√
S4 = − d4 x −gR (524)
Z h
→ − d4 xγ (∂ 2 π)2 − (∂∂π)2

i
+2γ 2 (∂π) · (∂∂π)2 · (∂π) − ∂ 2 π(∂π) · (∂∂π) · (∂π) (525)
Z
3 √
S5 = d4 x −g(J − 2Gµν Kµν ) (526)
2
Z h
→ − d4 xγ 2 (∂ 2 π)3 + 2(∂∂π)3 − 3(∂ 2 π)(∂∂π)2

+6γ 2 ((∂ 2 π)(∂π) · (∂∂π)2 · (∂π) − (∂π) · (∂∂π)3 · (∂π))

i
−3γ 2 ((∂ 2 π)2 − (∂∂π)2 )(∂π) · (∂∂π) · (∂π) , (527)
p
with the Lorentz factor γ = 1/ 1 + (∂π)2 . The expressions for S3 and S5 can be
identified with the boundary terms in General Relativity and in Gauss-Bonnet gravity,
respectively. Of course, the former is the Gibbons-Hawking term, and the latter is the
Myers boundary term [931], discussed in more detail in SectionR 5.7.3. Now, for a slowly
moving probe in Minkowski space, itR can be shown that Si ≈ d4 xLi , which means that,
neglecting the tadpole, Sgen-DBI ≈ d4 xLgal [389]. One can then recover the conformal
galileon by considering a probe brane in AdS, and the covariant galileon by considering
a general bulk geometry [389].
This procedure has recently been extended to probe branes that are curved, giving
rise to a more general class of effective theories on curved space [545, 544] (see, also
[221]). These represent the analogues of galileons and DBI theories living on dS4 and
AdS4 , retaining the same number of symmetries as their flat space counterparts. There is
a rich structure and in some cases the symmetries can even admit non-trivial potentials
beyond the usual tadpoles.

37 We write (∂ µ1 π)(∂µ1 ∂ µ2 π) . . . (∂µn ∂ µn+1 π)(∂µn+1 π) = (∂π) · (∂∂π)n · (∂π).

159
4.4.4. Galileon cosmology
Galileon cosmology encompasses much more than the original model and its covariant
completion. The cosmological behaviour of a number of models that are inspired by the
galileon have also been investigated (see, for example, [1145, 711, 704, 375, 714, 404]).
These include the braiding model [714, 404, 1046], which is described by the following
action Z  
√ 1
S = d4 x −g R + K(φ, X) + G(φ, X)φ , (528)
16πG
where X = −(∇φ)2 . Note that for K = c22 X and G = c22 X we recover the covariant
galileon action up to cubic order. This model still gives rise to second-order field equations
and admits some rich phenomenology. It is claimed that the scalar equation of state can
cross the phantom divide without introducing any instabilities, and results in a blue
tilt for the spectrum tensor perturbations. Constraints on the model coming from large
scale structure and non-Gaussianity have been obtained in [693, 903], respectively. Non-
Gaussianity in DBI galileon inflation has been studied in [1059]. An even more general
class of scalar tensor theories yielding second order field equations has recently been
presented in [402], and is now known to be equivalent to Horndeski’s general theory [623]
in four dimensions [710].
It has been argued that some of these generalised models are perhaps too general
[222]. The point is that there is no symmetry protecting the theory from large radiative
corrections. This can spoil the functional form of the Lagrangian so much so that we
require more input parameters than we can measure, and we lose all predictivity. In
contrast, the pure galileon, conformal galileon and DBI galileon theories are safe against
radiative corrections since they possess additional symmetries that control the form of the
derivative interactions. For this reason, in the remainder of this section we will restrict
attention to those models for which the Galilean invariance is only weakly broken, so
that any radiative corrections that break the galileon symmetry are suppressed.
Let us begin with early universe cosmology and the covariant galileon. This can give
rise to inflation even in the absence of a potential [222]. The theory is radiatively safe
because the terms that break Galilean symmetry are suppressed by powers of Λ/Mpl ,
where Λ is the naive cut-off38 . As with DBI inflation, fluctuations about the quasi de
Sitter background will result in large non-Gaussianities at low sound speeds [903, 222].
It has been argued that what sets this model apart is the fact that the non-Gaussianity is
not constrained to obey fN L ∼ 1/c2s , making it distinguishable from DBI inflation [222],
although this claim has been disputed in [341]. Indeed, it is interesting to note that the
authors of [341] adopt the effective field theory approach to inflation, imposing Galilean
symmetry on the small perturbations around the inflationary background. This permits
additional interactions compared with [222], but maintains stability against radiative
corrections. They find that one can have large (observable) four point functions even
when the three point function is small.
We now turn to the cosmology of the late universe. The late time cosmology of the
covariant galileon has been studied for up to cubic [299, 911], quartic [520, 23], and
quintic scalar interactions [373, 956, 369]. In the latter model, we focus on the role of the
galileon as a dark energy field – it deviates slightly from the original galileon scenario

38 This corresponds to the scale of the galileon self-interactions.

160
[963] because the galileon coupling to matter disappears as Mpl → ∞. In any event, the
model is described by the following action
Z  
√ 1
S = d4 x −g R + Lφ + Lm (gµν , ψ) , (529)
16πG
where
 
c2 c3 c4 1
Lφ = (∇φ)2 + (∇φ)2
φ + (∇φ)2
2(φ) 2
− 2(∇∇φ)2
− R(∇φ)2
2 2Λ3 2Λ6 2
c5  
+ 9
(∇φ) (φ) − 3(φ)(∇∇φ) + 2(∇∇φ) − 6Gµν (∇ φ)(∇ ∇α φ)∇α φ . (530)
2 3 2 3 µ ν
2Λ
As there is no potential, late time acceleration must be driven by the kinetic terms.
There is a late time de Sitter solution characterised by H = HdS =constant and φ̇ =
φ̇dS =constant. The existence of this fixes a relationship between c2 , c3 , c4 and c5 such
that there are only two free parameters, given by
α = c4 x4dS , and β = c5 x5dS ,
√  
where xdS = 8πG φ̇dS /HdS . In [373] conditions are derived that guarantee the
absence of ghosts and imaginary sounds speeds in both the tensor and the scalar sector.
The viable region of parameter space, (α, β), where these conditions are met is presented.
As regards the cosmological evolution, we see that there exists a tracker solution that
approaches the late time de Sitter attractor. In Figure 6, we plot the evolution of the
galileon equation of state for the tracker solution, and for generic initial conditions.
Note that the tracker has a phantom equation of state. Indeed, even for generic initial
conditions, the galileon field is drawn into a phantom phase by the tracker. It turns
out that the tracker solution is disfavoured by a combined data analysis (SNe, BAO,
CMB). The generic case fares rather better, especially if we have non-zero curvature, Ωk ,
although it is still disfavoured with respect to ΛCDM [956].
Matter density perturbations have been studied in detail within the context of this
model in [369], where it is shown that the growth rate of matter perturbations is larger
than in ΛCDM. In the generic case, for suitable choices of α and β, we typically find that
the growth index today is γ0 < 0.4, with large variations at earlier times. This makes the
model easily distinguishable from ΛCDM. Another distinguishing feature is the effective
gravitational potential changing with time, even during matter domination.

4.4.5. Multi-galileons
The extension of the galileon scenario to include multiple scalar fields [1017, 1018,
1019, 396] and even arbitrary p-forms [397] has recently been developed (see [481, 480]
for earlier work). A general multi-galileon theory, in four dimensions, with N real scalar
degrees of freedom is given by the Lagrangian [1019, 1310]
5
X µ ...µ
LN-gal = αi1 ...im δ[ν22...νmm] πi1 ∂µ2∂ ν2 πi2 ...∂µm∂ νm πim , (531)
m=1

µ ...µ
where {αi1 ...im } are free parameters of the theory and δ[ν11...νmm] = m!δ µ1[ν1 ...δ µmνm ] . As
usual, summation over repeated Lorentz (Greek) and galileon indices (Latin) should be
161
 0.0

!0.5
(A) Late!time tracking
wDE

!1.0

!1.5
(B) Tracker
!2.0
0.1 1 10

1"z

Figure 6: Taken from Figure 1 in [369]. The equation of state of the galileon field, wDE , versus redshift,
z, for α = 1.37 and β = 0.44. The evolution is given for generic initial conditions (A) and for the tracker
(B).

understood to be implicit. Note further that we define the m = 1 term of expression

(531) to be αi1 πi1 . The Lagrangian (531) is constructed so that it is invariant under

πi → πi + (bi )µ xµ + ci , i = 1, . . . , N. (532)

One might expect this to appear in the decoupling limit of some co-dimension N braneworld
scenarios, with π1 , . . . , πN corresponding to the position of the brane in the N transverse
directions [1017]. Indeed, one can generalise the formalism discussed in Section 4.4.3
to probe a brane of co-dimension N and recover the multi-galileon theory in the non-
relativistic limit [610].
How many free parameters are there in this theory? We can always choose αi1 ...im to
be symmetric so the total number free parameters is given by

X5   X 5
N +m−1 (N + m − 1)!
= . (533)
m m!(N − 1)!
m=1 m=1

Even for N = 3, this corresponds to 55 free parameters. To reduce the number of pa-
rameters one can consider imposing internal symmetries on the galileon fields [1019],
although this will break the galileon symmetry (532). The phenomenology of spherically
symmetric solutions with an internal SO(N) has been studied and found to suffer from
problems with instabilities and superluminality, at least for standard non-derivative mat-
ter coupling [50]. One can also prove a generalised form a Goldstone’s theorem when
internal symmetries are present [1310].
162
 Of course, the large number of free parameters is less of an issue in the simple case
of N = 2, dubbed bigalileon theory. The phenomenology of this theory was developed
in detail in [1018]. Let us summarise the main results. In direct analogy with the
single galileon case, the bigalileon theory is formulated on sub-horizon scales as fields
propagating on Minkowski space. This time we have our GR graviton, h̃µν , and two
scalar galileons, π and ξ. Only one of the scalars, π, say, couples directly to the trace
of the energy-momentum tensor. The other scalar, ξ, couples indirectly through its
mixing with π, so it still has an important role to play. The governing action is given by
[1017, 1018]
h i Z
S h̃µν , π, ξ = d4 xLGR + Lπ,ξ , (534)

where LGR is given by Eq. (475), and

X
Lπ,ξ = (αm,n π + βm,n ξ)Em,n (∂∂π, ∂∂ξ) + πT, (535)
06m+n64

with
µ1
Em,n = (m + n)!δ[ν1
. . . δνµmm δσρ11 . . . δσρnn ] (∂µ1 ∂ ν1 π) . . .
. . . (∂µm ∂ νm π) (∂ρ1 ∂ σ1 ξ) . . . (∂ρn ∂ σn ξ) . (536)

The physical metric is given by gµν = ηµν + hµν , where hµν = h̃µν + 2πηµν . Given a
source Tµν , h̃µν gives the usual perturbative GR solution, and so 2πηµν gives the modified
gravity correction. The field equations for the scalars are
X X
T+ am,n Em,n = 0, and bm,n Em,n = 0, (537)
06m+n64 06m+n64

where39 am,n = (m + 1)(αm,n + βm+1,n−1 ) and bm,n = (n + 1)(βm,n + αm−1,n+1 ).

In contrast to the single galileon case, self-accelerating solutions can be consistent in
some bigalileon theories. Indeed, one can choose parameters such that we simultaneously
satisfy each of the following: (i) there is no tadpole, (ii) there is a self-accelerating vac-
uum, (iii) fluctuations about the self-accelerating vacuum do not contain a ghost, (iv)
spherically symmetric excitations about the self-accelerating vacuum undergo Vainshtein
screening in the solar system, (v) fluctuations about the spherically symmetric solutions
are never superluminal, (vi) fluctuations about the spherically symmetric solutions never
lead to trouble with excessive emission of Cerenkov radiation, (vii) there is not an unac-
ceptably low momentum scale for strong coupling on the spherically symmetric solution,
and (viii) there are no problems with back-reaction on the spherically symmetric solution
(or the vacuum). This supports the case for considering bigalileon theories as a viable
alternative to dark energy.
One can also develop models of self-tuning in bigalileon theory, where the vacuum en-
ergy does not affect the four dimensional curvature. These models get around Weinberg’s
no-go theorem by breaking Poincaré invariance. Unfortunately, in order for them to re-
main compatible with solar system tests one must limit the amount of vacuum energy
to be . meV.

39 We define α−1,n = βm,−1 = 0.

163
4.5. Other Theories
Let us now consider some further theories that have yet to be discussed. These
are ghost condensate theories, non-metric theories, and the dark energy from curvature
corrections approach of Piazza.

4.5.1. Ghost condensates

Ghost condensate theories involve introducing into the gravitational sector an extra
scalar field, φ, with shift symmetry

φ → φ + constant. (538)

The only terms in the action that can obey this symmetry are derivative ones, and so
the building block for this theory is taken to be

X = ∂µ φ∂ µ φ. (539)

In [590] it was shown that if the leading order term in the action has the wrong sign,
so that φ is a ghost field, it is still possible to construct a theory that is stable to small
fluctuations by including terms that push X to a fixed value, so that

hXi = C. (540)

Theories of this type have a number of interesting properties. For a start, the non-zero
vacuum expectation value of the ghost field signals a spontaneous breaking of Lorentz
invariance. What is more, fluctuations in the ghost field about the vacuum expectation
value appear linearly in the energy-momentum tensor, meaning that anti-gravity is pos-
sible. A further interesting point is that in the weak field limit large ghost condensate
clumps move more slowly than small clumps, with potentially interesting phenomenolog-
ical consequences.
It has been argued in [590] that ghost condensate fields act like the gravitational
counterpart to the Higgs field of the standard model of particle physics. This is because
gravitational fields propagating through the condensate acquire a massive mode, much
like particles acquire mass while propagating through the Higgs field. Ghost condensates
also introduce oscillatory correction to the gravitational potential, with a Jeans instability
that grows with time. For mass parameters of the order 10−3 eV, these corrections occur
on spatial and temporal scales greater than H0−1 . If the massive modes are of order
10MeV, however, then corrections can occur on length scales as small as 1000km, but,
again, only on time scales greater than H0−1 .
The action for ghost condensate theories can be written
Z  
√ R
S = d4 x −g + M 4 P (X) , (541)
16πG
where M is a mass scale (confined to be 1meV ≤ M ≤ 10MeV), and P (X) is a function
that must have a non-zero minimum at X = C in order to be a ghost. For stability we
then require

P 0 (C) ≥ 0 (542)
0 00
P (C) + 2CP (C) ≥ 0. (543)
164
Extremisation of the action (541), with respect to the metric, yields field equations of
the form

Gµν = 8πGTφ,µν , (544)

where

Tφ,µν = M 4 (P (X)gµν + 2P 0 (X)∂µ φ∂ν φ) , (545)

and for simplicity we have not included a term in the action for normal matter fields.
To see why this theory is considered a modified theory of gravity we look at the
perturbed
√ field equations. Writing the metric as gµν = ηµν + hµν , and the ghost field as
φ = Ct + π, the Lagrangian density of the theory becomes
√  
L = −γ F0 (X) + F1 (X)K 2 + F2 (X)K ij Kij + · · · , (546)

where we have chosen a unitary gauge, and where π has been set to zero. The Fn here
are functions of X that are derived from P (X), and K is the extrinsic curvature of the
3 dimensional hyper-surfaces of constant φ. Diffeomorphism invariance in Eq. (546) can
be seen to have been explicitly broken.
Let us now consider cosmology. For a homogeneous and isotropic Universe, the equa-
tion of motion for φ is
d  3 0 
a φ̇P (X) = 0. (547)
dt
If we assume
√ that X → 0 and P 0 (C) → 0, as t → ∞, then Eq. (547) tells us that
φ → ± Ct. The Friedmann equation for this theory is

m2
H2 = (2XP 0 (X) − P (X)) , (548)
3
√
where the new mass parameter is m ≡ 8πGM 2 = M 2 /MP l . The Raychaudhuri equa-
tion is
ä m2
=− (XP 0 (X) + P (X)) . (549)
a 3
Now, it can be shown that the effect of the ghost field on the expansion of the Universe
is such it can mimic radiation domination, matter domination and vacuum domination.
Indeed, the simple choice of P (X) = 21 (X − C)2 leads to a(t) ∝ (Cmt)1/2 at early times,
and a(t) ∝ (Cmt)2/3 at late times. Adding a constant, such that P (X) = 21 (X − C)2 +
Λ/m2 , leads to a period of vacuum domination.
A general class of solutions, with matter sources included, has been studied in [755].
Some of these solutions combine dark matter and dark energy-like behaviour, at the
background level. The behaviour of scalar perturbations in the ghost condensate theory
was worked out in detail in [922]. Modified Newtonian potentials were discovered with

Φ = ΦGR + Φmod (550)

Ψ = ΨGR + Ψmod , (551)
165
where ΦGR and ΨGR take their standard form from General Relativity, while Φmod and
Ψmod are corrections due to the ghost condensate that occur in the limit where the
wavelength of the fluctuation is larger than the symmetry breaking scale. If we consider
the case of de Sitter space, where Φmod = Ψmod , then the evolution equations are
 
α k4 αM 2 k 2 αM 2 k 2
∂t2 Φmod + 3H0 ∂t Φmod + − 2 + 2H 2
0 Φmod = ΦGR , (552)
M a2 4 2MP l a 2 2MP2 l a2

where α is a combination of dimensionless coefficients of O(1) from the action. These

equations shows that the Newtonian part of the potential seeds the modified part.

4.5.2. Non-metric gravity

We will now describe the non-metric gravity theory that deforms GR while keeping
only two dynamical degrees of freedom [136, 137, 138, 747, 748]. In this theory the
fundamental gravitational object is no longer the metric but a triple of 2-forms B i =
i
Bµν dxµ ∧ dxν , where lower-case Latin indices denote internal SU (2) indices and take
values from 1 to 3. The space-time metric is an emergent variable and is given in terms
i
of Bµν as
√ j
−ggµν ∝ ˜αβγδ Bµα
i
Bνβ k
Bγδ ijk (553)
where ˜αβγδ is the completely antisymmetric tensor density having components ±1 in
any coordinate system. The proportionality symbol is used above, rather than equality,
because the metric is defined only up to conformal rescalings. The reason for this is that
i
Bµν is self-dual, i.e. 12 µν ρσ Bρσ
i i
= iBµν is a conformally invariant relation.
The class of theories we will now describe contains only two propagating degrees of
freedom [139, 513], just like GR. Spherically symmetric solutions, as well as black holes,
have been studied [751, 752, 648], and extensions of these ideas to bimetric theories have
also been considered [1175]. Let us now describe the kinematical setup of the theory
before proceeding to discuss its dynamics.

Kinematics

Consider the set of 1-forms {e0 , eI }, the tetrad, where the capital Latin indices denote
internal SO(3) indices (note that this is a different space than SU (2) considered above)
such that
ds2 = gµν dxµ dxν = −e0 e0 + δIJ eI eJ . (554)
From the tetrad we can also define the self-dual 2-forms ΣI(+) , and similarly the anti
self-dual 2-forms ΣI(−) ), by

1
ΣI± = ie0 ∧ eI ∓ I JK eJ ∧ eK . (555)
2

Any other self-dual 2-form can then be decomposed in terms of ΣI(+) . In particular, we
may write
B i = B iI ΣI(+) . (556)

166
From B i we can then define the connection one-forms Ai as40

dB i + ijk Aj ∧ B k = 0. (557)

The above equation can be solved to get

1
Aiµ = B iαβ δjk Bµα
j
∇ν Bνβ
k
, (558)
2 det B
where det B = − 24  Bµ Bν Bρ . Now, since B i is conformally invariant, we know
1 ijk i ν j ρ k µ

that Ai is too. We can now proceed and define the curvature two-forms F i of Ai as

1
F i = dAi + ijk Aj ∧ Ak . (559)
2

Dynamics

The action for this theory takes on the form of the well known BF-theory:
Z  
i 1
S[B, A] = δij B i ∧ F j [A] − V (B i ∧ B j ) + Sm , (560)
8πG 2

where V (M) is a holomorphic function of a complex symmetric 3 × 3 matrix, M, that

is required to be homogeneous of degree one (i.e. V (λM) = λV (M)) so that when it
is applied to a four-form such as B i ∧ B j the result is also a four-form. We introduce
the internal metric hij = B iI B jJ δ IJ and further decompose it into trace and traceless


parts as hij = 13 h δ ij + H ij , where h = δij hij and δij H ij = 0. We can then write V as
V (hij ) = 13 hU (H ij ), and expand U as

1
U (H) = Λ0 − trH2 + O(H3 ), (561)
8`2
where the constant Λ0 plays the role of the cosmological constant, while the constant ` is
a new scale that describes deviations from GR. The minus sign in the 2nd term above is
required to avoid instabilities. In particular, as ` → 0 the theory reduces to the Plebanski
formulation of GR with a cosmological constant (see Section 2.3.3).
As discussed above, the metric here is defined only up to conformal transformations.
In order to couple the theory to matter fields we have to fix this ambiguity, which can
be achieved by the introduction of a new function R(hij ) that is also homogeneous of
degree one. The conformal freedom is then fixed by requiring41 R(h) = 1. In a similar
fashion to V , we can then decompose R as R = 13 hUm (H), and expand Um as
g
Um = 1 − trH2 + O(H3 ), (562)
2

40 Strictlyspeaking this defines a three-form which is then dualised to a one-form.

41 This method of fixing the conformal ambiguity can be shown to arise naturally by considering the
motion of a test body [749].
167
where g is a dimensionless constant that can be of any sign. This new parameter measures
the departure from the Urbantke metric, given by an equality in Eq. (553). Rather than
the two new parameters, g and `, it is sometimes convenient to use the two dimensionless
parameters β and γ, defined by
1 1 4
β=g− , and γ= − . (563)
3 `2 Λ0 3

Cosmological consideration then tell us that 0 < g < 1, and hence − 13 < β < 23 and
γ > 0 [753]. General Relativity is recovered in the limit γ → ∞.
Variation of Eq. (560) with respect to A gives Eq. (557), while variation with respect
to B gives
  
∂U 1 ij ∂Um 1
δij B iI F j = B iI + Λδ − 2πGT + Λm δ ij
B jJ ΣJ(+) − 8πGTIJ ΣJ(−) ,
∂H ij 3 ∂H ij 3
(564)
where Λ and Λm are the Legendre transforms of U and Um respectively, i.e.
∂U ∂Um ij
Λ=U− H ij and Λm = Um − H , (565)
∂H ij ∂H ij
and where T IJ = T̃ µν ΣIµλ ΣJνλ , where T̃ µν is the traceless part of the energy-momentum
tensor, Tµν [750].

Cosmology

The cosmology of this theory has been analysed by Krasnov and Shtanov at the level
of perturbed FLRW solutions [753]. Let us first consider the FLRW solutions of this
theory. For homogeneous and isotropic spaces we have B iI = δ iI . Hence, we can drop
the distinction between i and I and let B i = Σi . We also have H ij = 0, so U = Λ0 and
Um = 1, resulting in R = 13 h and V = 31 hΛ0 . The fixing condition R = 1 then gives
h = 3 and V = Λ0 . Under these conditions the field equations can be written
1 ρ i
Fi = Λ0 Σi − 2πG(P − )Σi − 2πG(ρ + P )Σ̃ . (566)
3 3
For the homogeneous and isotropic space-time we also have e0 = adτ and ei = adxi .
After some algebra this gives Aj = iHdxj and F i = iH0 dτ ∧ dxj − 21 ijk H2 dxj ∧ dxk .
The field equations can then be written

3H2 = 8πGa2 (ρ + ρΛ ), (567)

where PΛ = −ρΛ with Λ0 = 8πGρΛ , and

− 2H0 − H2 = 8πGa2 (P + PΛ ), (568)

where ρ and P do not include the cosmological constant. Thus, for metric backgrounds
the FLRW solutions of this theory are the same as those of General Relativity. The
situation changes, however, when we consider linear fluctuations. In this case one gets
departures from ΛCDM that depend on g and `.
168
 We now consider the perturbed space-time metric in the conformal Newtonian gauge.
The perturbation for B ij is then given (after some convenient gauge-fixing) in terms of
a new scalar mode χ as
1
B ij = δ ij + 2 Dij χ. (569)
2a
The perturbed field equations are then given by [753]
1 2 2 
−k 2 Φ = 4πGa2 ρ [δ + 3H(1 + w)θ] + k k χ − 3Hχ0 , (570)
3a2
1 2 0
Φ0 + HΨ = 4πGa2 (ρ + P )θ − k χ, (571)
3a2

k2 k2
Φ00 + 2HΦ0 + HΨ0 + 2H0 + H2 Ψ + (Φ − Ψ) = 4πGa2 δP + 2 χ, (572)
 3  9a
2
1 k
Φ − Ψ = 8πGa2 (ρ + P )Σ + 2 χ00 + χ , (573)
a 3
and one can show that the Bianchi identities are satisfied independently of the χ terms.
In this sense the χ field is non-dynamical. The remaining equations determine χ in terms
of Φ and Ψ as
 
χ00 − 2Hχ0 − ∆ + 4H2 + Λ0 γ + 8πGβ(ρ − 3P ) χ + a2 (Φ + Ψ) = 0, (574)

where ρ and P does not include the cosmological constant. This equation can be solved
to get χ in terms of Φ + Ψ, hence the χ terms in the field equations can be thought of as
non-local modifications of the Einstein equations.
Krasnov and Shtanov also find the vector and tensor mode equations [753]. Further-
more, they study the evolution of perturbations during inflation, a well as radiation,
matter and Λ dominated epochs, and estimate the effects of the modifications on the
matter power spectrum.

4.5.3. Dark energy from curvature corrections

A proposal for IR modifications of gravity has been put forward by Piazza [1036, 1037].
The starting point for this is the usual semi-classical gravity, where matter fields are
quantised on a curved background manifold. The operators of the matter field theory are
then modified in the IR in a way we will now describe. Schematically, in a cosmological
setup, operators corresponding to Fourier modes of physical momentum k are corrected
by terms of order H 2 /k 2 , where H is the Hubble parameter. These modifications lead
to the apparent existence of Dark Energy, but without introducing a new scale in the
problem.
To illustrate this idea consider the vacuum expectation value of the local energy
density of a massless field42 :
Z " #
fquad (t) flog (t)
0 3
hT 0 (t, ~x)ibare = d k k+ + + ... (575)
k k3
= local terms + non-local terms,

42 see e.g.[1298, 516] for the explicit expression of a massive scalar on flat FLRW background.
169
where spatial homogeneity has been assumed for simplicity. The local terms can be re-
moved by local gravitational counter-terms, while the non-local pieces represent the gen-
uine particle/energy content of the chosen “vacuum” state. The first term contributes to
the cosmological constant, and in flat space-time can be removed by the usual procedure
of normal-ordering (the f ’s vanish in flat space-time). In curved space-time, however, the
presence of the time-dependent f ’s makes the normal-ordering procedure meaningless.
The conjecture of [1036, 1037] is that there exists a theory that resembles semi-classical
GR on small scales, but that has an IR-completion that prohibits the time dependent
pieces in (575). If that is the case then one can still deal with the cosmological constant
term by the usual procedure of normal-ordering, as in flat space-time.
To try to construct such a theory [1036, 1037] propose what they call the Ultra
Strong Equivalence Principle: For each matter field or sector sufficiently decoupled from
all other matter fields, there exists a state (the “vacuum”) for which the expectation
value of the (bare) energy-momentum tensor is the same as in flat space, regardless of
the configuration of the gravitational field.
What this principle aims to achieve is to remove the time-dependent terms in Eq.
(575) by appropriate modifications of semi-classical gravity that manifest themselves
when the Fourier modes have wavelengths comparable to the inverse extrinsic curvature
(i.e. the inverse Hubble radius H −1 ). At the present, a complete theory that imple-
ments this idea is lacking, but a toy-model with massive scalar fields has been considered
in [1036, 1037]. Letting ~n be the comoving momentum that labels operators in Fourier
space (related to physical momentum as ~n/a), the modification to O(H 2 a2 /n2 ) is given
by the modified commutation relation
h i  
(1) (1)† (3) 0 H 2 a2
A~n , A~n0 = δ (~n − ~n ) 1 − + ... , (576)
2n2
(1)
where A~n is the annihilation operator.
h This
i prescription is equivalent to using the
(0) (0)†
standard commutation relation A~n , A~n0 = δ (3) (~n − ~n0 ) for the standard operator
 
(0) 2 2
A~n , but with a modified comoving momentum given by ~k = ~n 1 − H2na2 that locally
defines the infinitesimal translations. In a local neighbourhood (smaller than a Hubble
patch) the above prescription can be shown to cancel the quadratically divergent piece
fquad (t) in Eq. (575). Note that the momentum ~k is not conserved, but ~n is.
To extend the above to the global picture one can use the translation operator
R (1)† (1) R (0)† (0)
e−iλP , where P~ (1) = d3 n ~n A~n A~n = P~ (1) = d3 n ~k A~n A~n is the momentum
(1)

operator constructed with the modified Fourier modes, and λ is the comoving proper
distance to a point far away from the origin. In GR the comoving distance λ = d(t)/a(t)
is a constant given by the ratio of the physical distance, d(t), to the scale factor, a(t).
However, in the present theory one finds instead that
1 3d 2 2
λ̇ = λ (a H ) + higher orders. (577)
4 dt
Comoving distances obeying Eq. (577) are, in fact, already strongly disfavoured by ob-
servations [958]. One may, however, try to explore further whether the dynamical Hubble
scale H(t) itself could provide the scale required by cosmic acceleration by considering
170
the more general expansion

λ̇ = A1 λH + A2 (λH)2 + . . .
d d
+B1 λ2 (aH) + B2 λ3 (a2 H 2 ) + . . . , (578)
dt dt
where Ai and Bi are a set of constants. The authors find that certain regions of the
resulting parameter space can fit the data as well as ΛCDM.

171
5. Higher Dimensional Theories of Gravity

The first systematic studies of higher dimensional geometry date back to the likes
of Riemann, Cayley and Grassmann in the mid nineteenth century. It lies at the heart
of General Relativity, where space and time form part of a curved 3 + 1 dimensional
manifold, as described in Section 2. Of course, Riemannian geometry is not restricted to
3+1 dimensions, so we have the tools to study gravitational theories in higher dimensions.
Indeed, this is more than just a theoretical curiosity. Superstring theory, arguably our
best candidate for a quantum theory of gravity, can only be formulated consistently in
10 dimensions.
The problem now is a phenomenological one: Gravity does not behave like a 10
dimensional force in our experiments and observations. Perhaps the simplest observation
along these lines is the stability of earth’s orbit. In D dimensions of space-time, the
Newtonian potential due to a point source will typically go like 1/rD−3 . For D 6= 4, it
follows that we cannot have stable planetary orbits, and so it is clear that gravity should
not appear 10 dimensional on solar system scales. We use the word appear, because there
exist gravitational models where the extra dimensions are hidden from experiment, but
which open up at shorter and/or larger distances.
In this section we will review various models of higher dimensional gravity that have
been proposed. We will only discuss the case of extra spatial dimensions, although extra
temporal dimensions have been studied (see eg [1141]). One might worry that extra
temporal dimensions lead to problems with causality, as they permit closed time-like
curves in the form of circles in the plane of the two temporal directions.

5.1. Kaluza-Klein Theories of Gravity

Kaluza-Klein (KK) theory grew out of an attempt to unify gravity and electrody-
namics [985, 673, 701, 702]. The basic idea was to consider General Relativity on a 4 + 1
dimensional manifold where one of the spatial dimensions was taken to be small and
compact. One can perform a harmonic expansion of all fields along the extra dimension,
and compute an effective 3 + 1 dimensional theory by integrating out the heavy modes.
This idea has been embraced by string theorists who compactify 10 dimensional string
theories and 11 dimensional supergravity/M-theory on compact manifolds of 6 or 7 di-
mensions respectively, often switching on fluxes and wrapping branes on the compact
space (see [552] for a review). Each different compactification gives a different effective
4-dimensional theory, so much so that we now talk about an entire landscape of effective
theories [1194].
Assuming that the extra dimensions have been stabilised, the late-time dynamics of
KK theories is most easily understood at the level of the 4D effective theory. As we will
show, this will generically correspond to a 4D gravity theory with extra fields, examples
of which are studied in detail in Section 3. At early times, when the 3 dimensional space
is comparable in size to the extra dimensions, the effective description clearly breaks
down. This forms the basis of KK cosmology where one can ask the deeply profound
question of why and how the 3 extended dimensions of space were able to grow large,
while the extra dimensions remained microscopically small. It seems fair to say that a
fully satisfactory answer to this question has yet to emerge.

172
 We will now discuss some aspects of KK theory, starting with an overview of dimen-
sional reduction and effective theory before moving on to a discussion of KK cosmology
at early times. For a more detailed review of KK theory see [76, 1005].

5.1.1. Kaluza-Klein compactifications

To understand the generic features of KK compactifications, it is sufficient to describe
the dimensional reduction of General Relativity on a circle, S 1 . We first define General
Relativity in D = d + 1 dimensions, via the generalised Einstein-Hilbert action
Z
1 √
S[γ] = dD X −γR, (579)
16πGD
where GD is Newton’s constant in D dimensions, γAB is the D dimensional metric
with corresponding Ricci tensor, RAB , and Ricci scalar, R = γ AB RAB . Note that we
are neglecting the matter Lagrangian for brevity. We are assuming that one of the
spatial dimensions is compactified on a circle of radius L/2π. To this end we can define
coordinates X A = (xµ , z), where the coordinate z lies along the compact direction, such
that 0 ≤ z < L.
We can expand the metric as a Fourier series of the form
X (n)
γAB (x, z) = γAB (x)einz/L . (580)
n

We find that this gives an infinite number of extra fields in d dimensions. Modes with
n 6= 0 correspond to massive fields with mass |n|/L, whereas the zero mode corresponds
to a massless field. As we take L to be smaller and smaller we see that the mass of
the first massive field becomes very large. This means that if we compactify on a small
enough circle we can truncate to massless modes in the 4-dimensional theory. Massive
modes will only get excited by scattering processes whose energy lies at or above the
compactification scale 1/L. This also applies to matter fields arising in particle physics.
Indeed, particle physics imposes by far the strongest constraints on the size of the extra
dimension. Standard Model processes have been well tested with great precision down
to distances of the order ∼(TeV)−1 , with no evidence of extra dimensions yet emerging
[941]. Assuming that the extra dimensions are universal, that is the Standard Model
fields can extend all the way into them, we infer that L . 10−19 m. The natural scale of
the compact dimensions is usually taken be Planckian, L ∼ lpl .
(0) (0) (0) (0)
Let us now focus on the zero modes, γAB (x). We could define γµν , γµz and γzz to
be the d-dimensional fields gµν (x), Aµ (x) and φ(x). In effective field theory language,
these will correspond to the metric, gauge field, and dilaton, respectively. In order that
our results are more transparent we will actually define the components of the metric in
the following way:
(0)
γµν = e2αφ gµν + e2βφ Aµ Aν , (0)
γµz = e2βφ Aµ , (0)
γzz = e2βφ , (581)
p
where α = 1/ 2(d − 1)(d − 2), and β = −(d − 2)α. Since we have truncated to the
massless fields, we can integrate out the z part of the action given in Eq. (580). We find
that the d-dimensional effective action is then given by
Z  
L d √ 1 2 1 −2(d−1)αφ 2
Seff [g, A, φ] = d x −g R − (∇φ) − e F , (582)
16πGD 2 4
173
where F 2 = Fµν F µν and Fµν = ∇µ Aν − ∇ν Aµ is the electromagnetic field strength. The
curvature associated with the d dimensional metric, gµν , is described by the Ricci tensor,
Rµν , and Ricci scalar, R = g µν Rµν . What we now have is an Einstein-Maxwell-Dilaton
system in d dimensions. Of course, Kaluza and Klein were particularly interested in the
case of d = 4. They were frustrated by the presence of the dilaton, φ, in the resulting
4-dimensional effective theory. The point is that one cannot simply set the dilaton to
zero and retain a non-trivial Maxwell field, since this would be in conflict with the field
equations arising from Eq. (580),
  
1 1 1
Gµν = ∇µ φ∇ν φ − (∇φ)2 gµν + e−2(d−1)αφ Fµα Fν α − F 2 gµν , (583)
2 2 4
 
∇µ e−2(d−1)αφ Fµν = 0, (584)
1
φ = − (d − 1)αe−2(d−1)αφ F 2 , (585)
2
where Gµν = Rµν − 12 Rgµν is the Einstein tensor in d dimensions. In the usual jargon,
switching off the dilaton does not represent a consistent truncation of the higher dimen-
sional theory [440]. We should also note that the physical size of the compact dimension
is not necessarily given by L, but by Leβφ(x) . If L is to represent an accurate measure
of the compactification scale, we are therefore implicitly assuming that φ is stabilised
close to zero. For this to happen we need to generate a potential for φ that admits a
stable solution– this is known as the problem of moduli stabilisation. In more general
compactifications, moduli potentials can be generated by Casimir effects of fields in the
compact space [54, 55, 238, 335, 336], but the moduli remain unstable [336]. In fact, the
problem of moduli stabilisation has only recently been solved by switching on fluxes to
stabilise the volume of the compact space [662, 661].
There are, of course, many different compactifications that have been studied in the
literature, a detailed analysis of which is clearly beyond the scope of this review (see
[552]). However, aside from details such as the inclusion of fluxes and branes on the
compact space, the general scheme of each compactification is the same as the one we
have just described. Typically, a compactification of, say, 11 dimensional super-gravity
down to four dimensions will give rise to a gravity theory with a plethora of extra fields.
These extra fields include scalars, pseudo-scalars, vectors, pseudo-vectors, and arbitrary
p-forms. Modifications of gravity due to extra fields are studied in detail in Section 3.

5.1.2. Kaluza-Klein cosmology

As we have just discussed, for a phenomenologically viable theory with compact (and
stabilised) extra dimensions, the characteristic size, L, of the compact manifold should
not exceed the scale probed by modern collider experiments, which is currently around
10−19 m [941]. It is amusing to compare this to the characteristic size of the 3 large spatial
dimensions, which is at least a Hubble length c/H0 ∼ 1026 m, or in other words at least
45 orders of magnitude greater. Of course, it was not always like that. In the very early
universe, at times t . L/c, one might expect that all spatial dimensions were of the same
scale, each playing an equally important role in the dynamical evolution. This begs the
question: Why did the Universe evolve into a state where just 3 spatial dimensions grew to
macroscopic scales? Put another way, how does one achieve a dynamical compactification
174
mechanism in the early Universe such that 3 spatial dimensions expand exponentially
to an extremely large size, in contrast to the remaining spatial dimensions? Were those
extra dimensions somehow prevented from growing beyond a certain size, or did they
grow initially and later contract towards their current state?
These questions have led many authors (see, e.g. [514, 33, 730, 1084, 1083, 1085, 1, 3,
2, 997, 555]) to consider the dynamics of anisotropic cosmologies in D = d+1 dimensions,
where d = n + ñ. Indeed, consider the Bianchi-type metric

ds2 = γab dX a dX b = −dt2 + a2 (t)qij (x)dxi dxj + ã2 (t)q̃mn (x̃)dx̃m dx̃n , (586)

where the coordinates xi run over the n spatial dimensions and the coordinates x̃m
run over the ñ spatial dimensions. The n-dimensional metric, qij (x), is taken to have
constant curvature κ, whereas the ñ-dimensional metric, q̃mn (x̃), is taken to have constant
curvature κ̃. The growth of these two spaces is controlled by the relevant scale factors
a(t) and ã(t). Naturally, we will be interested in the case of n = 3, but for the moment
let us keep things general.
We now apply Einstein’s equations in D = d + 1 dimensions,
1
Gab ≡ Rab − Rγab = 8πGD Tab , (587)
2
where the energy-momentum tensor is given by an anisotropic fluid,
 ñ

n
z }| { z }| {
 
Tba = diag −ρ, P, . . . , P , P̃ , . . . , P̃  .

As usual, ρ(t) is the energy density, whereas p(t) is the pressure along the n dimensions
and P̃ (t) is the pressure along the ñ dimensions. Einstein’s Equations (587) then yield
the following [514]

ñ  
n 2 κ 2 κ̃
2 (n − 1) H + a2 + 2 (ñ − 1) H̃ + ã2 + nñH H̃ = 8πGD ρ, (588)

h i
ä 2 κ 8πGD
a + (n − 1) H + a2 + ñH H̃ = d−1 ρ + (ñ − 1)P − ñP̃ , (589)
  h i
¨
ã 2 κ̃ 8πGD
ã + (ñ − 1) H̃ + ã2 + nH H̃ = d−1 ρ − nP + (n − 1)P̃ , (590)

˙ are the Hubble parameters of the two expanding/contracting

where H = ȧ/a and H̃ = ã/ã
spaces. Of course, we also have energy conservation, which gives

ρ̇ + nH(ρ + P ) + ñH̃(ρ + P̃ ) = 0. (591)

Note that we do not necessarily have to assume that the cosmological dynamics is gov-
erned by D-dimensional General Relativity. We can also consider modifications of GR
where additional fields are present. For example, in string gas cosmology [183, 1216], we
consider the action
Z
1 √

S= dD x −γe−2φ R − 4(∇φ)2 − c + Sm [γ, Ψn ], (592)
16πGD
175
where Sm is the matter part of the action, containing the string gas, and the constant
c vanishes in the critical case43 , but not otherwise. The resulting field equations can be
written in the form Gab = 8πGD Tab , where
1 h c i
(m)
Tab = − γab + 8∇a φ∇b φ − 6γab (∇φ)2 − 2(∇a ∇b − γab )φ + e2φ Tab , (593)
8πGD 2
(m) 2 δSm
and Tab = − √−γ δγ ab
. The scalar equation of motion just follows from energy conser-
vation, ∇a Tab = 0.
Let us return to Equations (588)-(590) with a view toward dynamical compactifi-
cation. Many of the earlier works [33, 730, 1084, 1083, 1085, 1, 3] focus on isotropic
perfect fluids, for which P = P̃ = wρ. For simplicity and definiteness, let us follow the
analysis of Abbott, Barr and Ellis [1]. We consider an epoch in which we have radiation
domination, w = 1/(n + ñ), so that the evolution equations read

ñ  
n 2 κ 2 κ̃
2 (n − 1) H + a2 + 2 (ñ − 1) H̃ + ã2 + nñH H̃ = 8πGD ρ, (594)
ä 2 κ

8πGD
a + (n − 1) H + a2 + ñH H̃ = d ρ, (595)
 
¨
ã 2 κ̃ 8πGD
ã + (ñ − 1) H̃ + ã2 + nH H̃ = d ρ. (596)

Now, if the ñ dimensions are taken to be an ñ-sphere (κ̃ > 0), it is clear from Equation
(596) that they will reach a maximum size when H̃ = 0, and will subsequently start to
recollapse. In contrast, we can take the n dimensions to be flat or hyperbolic (κ ≤ 0), so
that these dimensions will never turn around. In fact, one can show that as we start to
approach the singularity of the collapsing sub-space (ã(t) → 0), the n dimensions enter
a phase of accelerated expansion. To see this note that H̃ starts to become large and
negative, and so it is clear from Equation (595) that we will enter a phase with ä > 0.
The typical evolution of the two scale factors is shown in Figure 7. Note that we can
even allow for a sufficiently small κ > 0 and still retain this qualitative behaviour. The
upper bound on κ follows from demanding that the turnaround in a(t) occurs after the
turnaround in ã(t). The bound is not strong enough to be interesting: It merely implies
that today’s Universe is larger than the horizon [1].
Of course, it is clear that the classical equations will start to break down √ in the
neighbourhood of the singular point. The physical radius of the ñ-sphere √ is ã/ κ̃, so
we certainly would not expect to trust our field equations when ã(t) . κ̃LD , where
1/(D−2)
LD ∝ GD is the fundamental Planck length in D dimensions. In [1], it is assumed
that quantum gravity effects will ultimately stabilise the √size of the internal space, ending
the inflationary phase at some time tc , where ã(tc ) = κ̃L, for some compactification
scale L & LD . Albeit without much justification, let us accept this assumption for the
moment, and consider the physically interesting case of n = 3. One might hope that
the inflationary phase is sufficiently long to offer a solution to the flatness, entropy and
horizon problems of the standard cosmology. Consider the entropy problem in particular.
Entropy is indeed released from the extra dimensions into the usual 3 dimensions of

43 For the bosonic string the critical dimension is D = 26, whereas for the superstring the critical

dimension is D = 10.
176
 standard
evolution

∼ t1/2

inflationary
phase

a(t)
t = tc !√
ã(t) ã(t
ã(t c )c )== k̃L
κ̃L

Figure 7: Adapted from Figure 1 in [1]. The typical evolution of the scale factors in the two different
sub-spaces. The scale factor a(t) along the n dimensions grows large, entering an inflationary phase as
the remaining ñ dimensions begin to recollapse. The scale factor ã(t) along the ñ dimension is assumed
to be stabilised by quantum gravity effects at some time tc .

space [33, 1], but only as much as log S ∼ |O(1)| log(LD /L) [730]. Since we demand that
L & LD , this is clearly way short of the total required to solve the entropy problem,
log S ∼ 88. In short, KK inflation does not last long enough to provide an alternative to
scalar field driven inflation.
To get the required expansion of 3-dimensional space, we must therefore include
some additional scalar fields. Adapting [615] slightly, we can mimic a period of slow-roll
inflation by plugging a cosmological constant, Λ, into the Field Equations (588)-(590).
Setting P = P̃ = −ρ, where ρ = Λ/8πGD , we find

ñ  
n 2 κ 2 κ̃
2 (n − 1) H + a2 + 2 (ñ − 1) H̃ + ã2 + nñH H̃ = Λ, (597)
ä 2 κ

8Λ
a + (n − 1) H + a2 + ñH H̃ = d−1 , (598)
 
¨
ã 2 κ̃ 2Λ
ã + (ñ − 1) H̃ + ã2 + nH H̃ = d−1 . (599)

Again, by taking the n dimensions to be flat (κ = 0), and the ñ dimensions to be positively
curved (κ̃ > 0), we find a solution for which the flat directions grow exponentially, and the
spherical dimensions remain fixed with H = H∗ at ã = ã∗ , where H∗2 = ñ−1 κ̃ 2Λ
n ã2∗ = n(d−1) .
The radius of the extra dimensions lies at the maximum of its potential, so this solution
is unstable. Indeed, fluctuations
p reveal that the spherical dimensions collapse to zero size
over a time scale ∆t ∼ (1 + 1 + 8/n)/4H∗ , after which we cannot count on exponential
growth in the flat directions.
We now consider the phenomenologically interesting case of n = 3. To get the required
177
number of ∼ 65 e-folds of inflation along the 3 flat directions we need H∗ ∆t ∼ 65. For
n = 3, H∗ ∆t ∼ 0.729, so once again inflation is cut short far too early. We could imagine
getting around this problem if we could alter the potential for the radius ã, such that it
develops a minimum as well as a maximum, by switching on fluxes [662, 661]. Generically,
it is still very difficult to get enough exponential growth along the familiar 3 dimensions
without causing the extra dimensions to grow alongside them [615]. For further details on
the latest attempts to embed inflation in higher dimensional theories, we refer the reader
to [807, 1048, 812, 666, 218, 873, 166]. Bounds on the variation of fundamental constants
for dynamical compactifications have been studied in [571, 572], while PPN parameters
for KK models in the solar system were computed in [461]. Note that Kaluza-Klein
cosmologies have also recently been applied to the dark energy problem [119, 120].
We end our discussion of Kaluza-Klein cosmology by asking the question: Why are
there 3 large spatial dimensions? We have already alluded to an anthropic explanation
demanding the existence of stable planetary orbits44 . To this we could add the existence
of stable atoms and chemistry, both key to the development of intelligent life, and both
requiring no more than 3 (large) spatial dimensions.
Modern attempts at a dynamical understanding of the dimensionality of space include
String Gas Cosmology [183, 1216] (for reviews see, e.g., [180, 116, 181]). Here the spatial
dimensions are taken to be compact and precisely 3 dimensions are allowed to grow
large due to the annihilation of strings wrapping around those dimensions. The point is
that strings winding around compact dimensions oppose their expansion since the energy
of the string winding modes increases with radius. To allow the compact dimensions to
grow large the winding modes must therefore collide and annihilate with the anti-winding
modes. Generically, we would only expect collisions of 1 + 1 dimensional strings in at
most 3 + 1 dimensions. Thus, the dimensionality of the string controls the dimensionality
of space by allowing at most 3 spatial dimensions to grow to macroscopic scales. Note
that this result is not spoilt by the inclusion of branes wrapping compact directions, as
these happen to fall out of equilibrium before the strings [21].
Whilst this idea has some appeal at first glance, it has not stood up to intense
scrutiny. More detailed quantitative analyses suggest that the desired outcome is not
at all generic, and requires highly fine tuned initial conditions [456, 457, 458]. Whilst
one can engineer an anisotropic set-up allowing 3 dimensions to grow large as desired,
typically the internal dimensions also grow to large sizes, just at a slower rate [456, 457].
In fact, it turns out that either all dimensions grow large since the string gas eventually
annihilates completely, or all dimensions stay small since the string gas gets frozen out
[458]. There are also problems at the level of cosmology. For example, when properly
calculated, the scalar perturbations have a blue power spectrum with n = 5, which is
strongly ruled out by observations [670]. It has been argued that a near scale invariant
spectrum can be obtained if the dilaton gets frozen during the strong coupled Hagedorn
phase in the very early Universe [182]. However, such claims still rely on a semi-classical
treatment of cosmological perturbations that cannot be trusted during the Hagedorn
phase, as the strings are strongly interacting.
In the context of 10 dimensional string theory, other attempts to explain the dimen-

44 On the subject of planetary orbits, it is amusing to note that Kepler himself reasoned that the 3-fold

nature of the Holy Trinity was responsible for the perceived dimensionality of space. Ptolemy is reputed
to have offered some alternative ideas in his work On Dimensionality, but they have since been lost.
178
sionality of our Universe consider that for integer values of n, the inequality 2n < 10 =⇒
n ≤ 4 [444, 683]. This is interesting because it means that the world volume of 3 + 1
dimensional branes (known as 3-branes) are less likely to intersect than those of larger
branes. In particular, Karch and Randall [683] have shown that an FLRW universe
initially filled with equal numbers of branes and anti-branes will ultimately come to be
dominated by 3-branes and 7-branes. This analysis accounts for the fact that larger
branes dilute more slowly, as well as the likelihood of intersections and annihilations
(hence the importance of 7-branes). In a braneworld scenario, this could explain why we
might be more likely to find ourselves living on a 3-brane, as opposed to a larger brane.
The consequences of living on a 3-brane are discussed in detail in the Section 5.2.
Finally, we note that for toroidal compactifications, 3 large spatial dimensions can be
linked to the stability of the small extra dimensions, at least in the presence of solitonic
strings/branes that correspond to point masses in the large dimensions [462].

5.2. The Braneworld Paradigm

The braneworld paradigm [15, 1074, 62, 52, 63] represents a radical alternative to
the standard Kaluza-Klein scenario, discussed in the previous section. In the KK sce-
nario, the extra dimensions must be small and compact, the size of the internal space
constrained by collider experiments to be below the inverse TeV scale. In the braneworld
scenario the extra dimensions can be much larger, perhaps even infinite in extent. This
is made possible by relaxing the assumption of universal extra dimensions.
In the braneworld picture the Standard Model fields are not universal, rather they
are confined to lie on a 3 + 1 dimensional hyper-surface, known as the brane, embedded
in some higher dimensional space-time, known as the bulk. Tests of Standard Model
processes can only constrain how far the brane may extend into the bulk, or, in other
words, the brane thickness. They do not constrain the size of the bulk itself. Such
constraints can only come from gravitational experiments, as gravity is the only force
that extends into the bulk space-time. As is well known, on small scales gravity is
much weaker than the other three fundamental forces, making it difficult to test at short
distances. In fact, the gravitational interaction has only been probed down to ∼ 0.1 mm,
with torsion-balance tests of the inverse square law [10]. It is too simplistic, however, to
suggest that this translates into an upper bound on the radius of the bulk. Gravity is
intimately related to geometry, and, as we shall see, one can warp the bulk geometry such
that an infinitely large extra dimension is still allowed by experiment. For an excellent
introduction to large extra dimensions see [1072].
Before delving into a detailed discussion of the various models, we note in passing
that the braneworld paradigm is well motivated by string theory [622, 838, 52]. As well
as fundamental strings, string theory contains fundamental objects known as D-branes
[1044]. These are extended objects upon which open strings can end. The braneworld
set-up therefore has a natural interpretation in terms of a stack of D-branes embedded in
a higher dimensional target space (see Figure 8). Open strings, with their ends attached
to the D-branes, can be identified with the Standard Model fields bound to the brane.
Only closed strings can propagate through the bulk, and these are identified with the
gravitational interactions.

179
Figure 8: Taken from [51]. Here the brane has 3 macroscopic dimensions, and dk compact dimensions.
The open strings end on the D branes, whereas the closed strings propagate through the bulk.

5.2.1. The ADD model

The braneworld paradigm really began to gather momentum with the seminal work
of Arkani-Hamed, Dimopoulos and Dvali [62], in which the large extra dimension is
exploited in order to explain the vast hierarchy between the electro-weak scale, MEW ∼
TeV, and the Planck scale, Mpl ∼ 1016 TeV. In this scenario the hierarchy does not go
away completely, rather it is reformulated as a hierarchy between the scale of the extra
dimensions, µ ∼ 1/L, and the electro-weak scale. The set-up is as follows: Standard
Model fermions and gauge bosons are localised on a 3 + 1 dimensional domain wall, in
an (effective) D = 4 + n dimensional space-time. We should clarify that D counts the
number of macroscopic dimensions in this scenario. Any microscopic dimensions, with
characteristic size given by the fundamental Planck length, behave as in the standard
KK scenario described previously.
Now, the width of the wall cannot exceed the inverse TeV scale, as explained above.
The bulk space transverse to the wall is compact but much larger than the width of
the wall (L  TeV−1 ), so much so that we can treat the wall as an infinitely thin 3-
brane. In the simplest construction, the bulk action is then described by the generalised
Einstein-Hilbert action Z
M D−2 √
Sbulk = D dD X −γR, (600)
2
where MD is the fundamental Planck scale in D dimensions, γAB is the D dimen-
sional metric with corresponding Ricci tensor, RAB , and Ricci scalar, R = γ AB RAB .
The Planck mass is related to the fundamental Newton’s constant in D dimensions by
180
 2−D
8πGD = MD . At large distances, gravitational interactions along the brane are medi-
ated by the graviton zero mode, which has a homogeneous profile over the extra dimen-
sions. Truncating to the zero mode, we can compute the four-dimensional effective action
describing long distance gravity along the brane by integrating out the macroscopic extra
dimensions. This result in
D−2 Z
MD Vn √
Seff = d4 x −gR, (601)
2
where gµν is the four-dimensional metric on the brane, with Ricci tensor Rµν , and Ricci
scalar R = g µν Rµν , and where the volume of the extra dimensions is given by Vn ∝ Ln .
The effective four-dimensional Planck scale, as seen by an observer on the brane, is then
given by
2 2+n n
Mpl ∼ MD L . (602)
By taking the macroscopic extra dimensions to be sufficiently large, we can eliminate
the standard hierarchy in D dimensions, MD ∼ MEW ∼ 1 TeV, and replace it with a
new hierarchy involving the scale of the extra dimensions, µ ∼ 1/L  MEW . This is not
in violation of short distance gravity tests, at least in D ≥ 6 dimensions. Indeed, in six
dimensions one can eliminate the hierarchy even for millimetre size extra dimensions.
In actual fact, the strongest constraints on the ADD model do not come from short
distance gravity tests, but from astrophysics and cosmology [63]. The problem arises
because the Kaluza-Klein modes can be extremely light, mKK & 1/L & 10−4 eV, and
2 2
extremely numerous, NKK ∼ Mpl /MD . 1032 . This means that even though each mode
is only very weakly coupled, with strength 1/Mpl , scattering processes along the brane,
at energies E & mKK , can produce a copious number of KK gravitons.
The strongest astrophysical constraint comes from the possible emission of KK modes
during the collapse of SN1987a. Requiring this to not be the dominant cooling processes
imposes a lower bound on the fundamental Planck scale. For example, with n = 2 we
have MD ≥ 50 TeV, whereas for n = 3 we have MD ≥ 3 TeV [350].
In cosmology, one has to worry about over-production of KK modes at high temper-
atures, since this may destroy the standard Big Bang picture. In order to be consistent
with Big Bang Nucleosynthesis, and the current composition of the Universe, one must
identify a maximum temperature for the early Universe for a given fundamental scale.
Taking the fundamental scale to be MD ∼ 1 TeV imposes a temperature bound T ≤ 10
MeV for n = 2, rising to T ≤ 10 GeV for n = 6 [588]. A higher fundamental scale will
raise the maximum temperature, but then one loses much of the appeal of the original
model. While a low maximum temperature is not in contradiction with cosmological
data, it does present a challenge to models of baryogenesis and inflation. The tem-
perature bounds can be weakened considerably if one does not require the bulk to be
flat. For example, when the bulk manifold is a compact hyperbolic space, the maximum
temperature can be pushed beyond the GeV scale even for n = 2 [671].

5.3. Randall-Sundrum Gravity

As we have already suggested, in a generic braneworld set-up there is no obvious
reason why one should demand that the bulk space should be flat, as in the ADD model.
In perhaps the most celebrated braneworld model, developed soon after ADD by Randall
and Sundrum [1051, 1052], the bulk is an anti-de Sitter space. There are two versions
181
of the Randall-Sundrum model, generally referred to as Randall-Sundrum I (RS1) [1051]
and Randall-Sundrum II (RS2) [1052]. Somewhat confusingly, the RS1 model contains
two branes, whereas the RS2 model only contains a single brane.
The RS1 model [1051] was also proposed as a resolution to the hierarchy problem. It
improves on the ADD model as the compact extra dimension need not be so large as to
introduce a new hierarchy. This is achieved by exploiting the exponential warp factor to
generate a large bulk volume from a small compactification radius. In contrast, the RS2
model [1052] contains a single brane and a non-compact extra dimension – it is infinite
in extent. This time the bulk warp factor ensures that gravity is localised close to the
brane, so that a brane observer only sees the gravitational effects of the extra dimension
above scales set by the bulk curvature.
Although the phrase Randall-Sundrum gravity really refers to these two original mod-
els, here we extend the definition to include any model with similar features. In this
section we are particularly interested in five-dimensional models containing 3-branes,
with some non-trivial geometry , or “warping”, present in the bulk. We begin with an
overview of some of the models. For further details see, for example, [1008, 845].

5.3.1. The RS1 model

In RS1, we have two 3-branes separated by a region of five-dimensional anti-de Sitter
space [1051]. The branes are located at z = 0, and z = zc , and we impose Z2 symme-
try across each brane. Neglecting Gibbons-Hawking boundary terms [536], the action
describing this model is given by
Z Z zc
M53 √
S= d4 x dz −γ (R − 2Λ)
2 −zc
Z p Z p
− σ+ d4 x −g (+) − σ− d4 x −g (−) , (603)
z=0 z=zc

(+) (−)
where γab is the bulk metric, and gµν and gµν are the metrics on the branes at z = 0
and z = zc , respectively. M5 is the five-dimensional Planck scale and is related to the
five-dimensional Newton’s constant via the standard relation G5 = 1/8πM53 . We also
include a negative bulk cosmological constant, Λ = −6/l2 . If we fine-tune the brane
tensions against Λ, such that
6M53 3
σ+ = −σ− = = ,
l 4πG5 l
then we admit a background solution in which the branes exhibit four-dimensional
Poincaré invariance:
ds2 = e−2|z|/l ηµν dxµ dxν + dz 2 , (604)
for −zc ≤ z ≤ zc . The Z2 symmetry about z = 0 is explicit, whereas the other boundary
condition should be understood to be implicit. The line-element given in Eq. (604)
contains an exponential warp factor that is displayed graphically in Figure 9. In between
the branes we recognise the geometry to be anti-de Sitter space, written in Poincaré
coordinates. Notice the peak in the warp factor at the positive tension brane, and the
trough at the negative tension brane. Although only a toy model, the RS1 set-up is well
motivated by a number of string theory/super-gravity constructions [622, 838, 439].
182
 z=0 z = zc

Figure 9: The behaviour of the warp factor in the RS1 model.

By integrating out the 4D zero-mode we are able to derive the 4D effective Planck
scale on a given brane [1051, 1008]:
 
2
M± = ±M53 l 1 − e∓2zc /l , (605)

where ± labels the sign of the corresponding brane tension. In terms of the effective
Newton’s constants we have
 
G5 ±1
G± = . (606)
l 1 − e∓2zc /l
Now suppose we live on the negative tension brane. If we take the fundamental Planck
scale M5 ∼ TeV, the bulk curvature scale to be just below 1/l ∼ 0.01M5 , and the
distance between the branes to be such that zc /l ∼ 35, we recover the desired effective
Planck scale, M− ∼ mpl ∼ 1016 TeV. Thus, the hierarchy problem has been eliminated
altogether, and not just shifted around, as in the ADD model. In contrast, the hierarchy
is not eliminated if we live on the positive tension brane since then the effective Planck
mass is given by M+ ∼ e−zc /l M− [1051, 1008].
As it stands, the RS1 model is incomplete. The problem is that on either brane the
low energy 4D effective theory is not GR, but Brans-Dicke gravity. The extra scalar
comes from fluctuations in the brane separation, and is sometimes referred to as the
radion [541, 528, 275]. The value of the Brans-Dicke parameter depends on the brane,
and is given by [528]
(±) 3  ±2zc /l 
wBD = e −1 . (607)
2
Observations require this parameter to be large (wBD > 40000, see Section 3.1.2). Note
(+)
that for the positive tension brane wBD can be made arbitrarily large with increasing
(−)
brane separation. The same cannot be said for wBD on the negative tension brane. If
we want to live on the negative tension brane we must generate a mass for the radion to
183
 brane

Figure 10: The behaviour of the warp factor in the RS2 model.

suppress its fluctuations. The Goldberger-Wise mechanism does exactly that, and thus
stabilises the distance between the branes [542].

5.3.2. The RS2 model

The RS2 model is obtained from RS1 by taking the negative tension brane off to
infinity [1052]. The geometry is then described by the metric in Eq. (604) with zc → ∞.
The corresponding warp factor is shown in Figure 10, the single peak at z = 0 indicating
that we have a single brane with positive tension. In this limit M+ → M53 l, so we cannot
eliminate the hierarchy problem as in the RS1 model. Rather, the situation is more akin
to the five-dimensional ADD model, with the curvature scale, 1/l, playing the role of the
compactification scale, µ ∼ 1/L.
What makes the RS2 model interesting is the way in which 4D GR is recovered on the
brane. As we have just seen, in RS1 the observer on the positive tension brane sees a low
(+)
energy gravity
theory corresponding to Brans-Dicke gravity, with a BD parameter wBD =
3 2zc /l
2 e − 1 . In the RS2 limit of infinite brane separation the BD scalar decouples
and one is left with 4D GR. As we will see in detail in Section 5.3.5, even though the
bulk is infinite in extent, gravity is localised on the brane at energies below the bulk
curvature scale, 1/l. As a method to screen the extra dimensions from the low energy
observer, this represents a radical alternative to the standard method of Kaluza-Klein
compactifications.
The key to gravity localisation is that the bulk volume is finite even though it has
infinite extent. This ensures that there is a normalisable graviton zero mode, which, in
the absence of any other massless modes, guarantees 4D GR at low enough energies.
The finite bulk volume arises because the warp factor falls off exponentially as we move
away from the brane. Intuitively, gravity localisation occurs because the warping makes
it difficult for the graviton to propagate too far away from the brane, so much so that
the region of the bulk with z & l has no influence on low energy brane interactions.
Since 1/l sets the scale at which the brane observer starts to become sensitive to the
184
bulk, table top experiments of the inverse square law impose the limit 1/l & 10−4 eV. This
translates into a lower bound on the fundamental Planck scale, M5 & 105 TeV, which
is well above the electro-weak scale. Since TeV scale gravity is not phenomenologically
viable in this case, we abandon any discussion of the hierarchy problem for the single
brane scenario.
Up until now our discussion has centred around weak gravity on the brane. What
about strong gravity in the presence of localised sources? Whilst there have been some
interesting numerical studies (see, e.g., [1278, 758]) exact strong gravity solutions are
rare in RS2. One exception are the solutions for a domain wall localised on the brane
[561, 560]. In contrast, an exact solution describing a braneworld black hole remains
elusive45 . The difficulty arises because the brane is an accelerated surface, so any black
hole residing on the brane must follow an accelerated trajectory. Such an accelerated
black hole requires knowledge of the AdS C-metric, but this solution is unknown in five
dimensions. For a nice review of the search for braneworld black hole solutions see [558],
and more recently [1200].

RS2 and AdS/CFT

Emparan, Fabbri and Kaloper (EFK) have suggested that a static braneworld black
hole does not exist [470]. To understand their argument we must first recall the AdS/CFT
correspondence in which type IIB string theory on AdS5 × S 5 is conjectured to be dual to
N = 4 SU (N ) super-Yang-Mills, in the large N limit [862]. This suggests an alternative
description of RS2 gravity [570]: ‘Gravity on an RS2 brane is dual to a strongly coupled
conformal field theory (CFT) cut off in the UV, minimally coupled to 4D gravity’. There
is plenty of evidence for this holographic description of RS2 (see, for example, [570,
1100, 438, 1009, 556, 1014, 596, 1138, 695], and for a review [1008]). Another way of
describing the correspondence is to say [470]: ‘A classical source on the brane is dual
to a quantum corrected source in four dimensions, with the quantum corrections coming
from the strongly coupled CFT’. The quantum corrections are large because of the large
number of degrees of freedom in the large-N limit. When applied to the problem of
finding a black hole on the brane, this suggests that the solution should not be static
since it should include the back-reaction of the Hawking radiation [470] (see also [1201]).
Fitzpatrick, Randall and Wiseman have disputed this interpretation, pointing out that
the CFT is strongly coupled and may therefore carry fewer degrees of freedom [508].
At this point it is fair to say that as yet there has been no consensus, and the subject
remains an active area of debate (see also [564, 562, 966, 505, 506]).
Note that if we accept the EFK conjecture, we can improve the bound on the bulk
curvature by an order of magnitude, 1/l & 10−3 eV. This is based on the existence of
long-lived black hole X-ray binaries [471]. For smaller values of 1/l these binaries would
have already decayed.

5.3.3. Other RS-like models

One of the characteristic features of the Randall-Sundrum models is the structure of
the bulk geometry, described by a non-factorisable, or warped, metric. One can embrace

45 There have been many attempts to find such a solution, most of which remain unpublished.
185
this structure and consider a whole slew of interesting generalisations. Here we consider
a class of models described by warped geometries of the form

ds2 = a2 (z)ḡµν (x)dxµ dxν + dz 2 . (608)

Many of the most interesting RS-like models exhibit quasi-localisation and give rise to
large-distance modifications of gravity. Perhaps the most celebrated of these is the DGP
model [454], which will be discussed in detail in Section 5.5. Other interesting examples
include the GRS model [563], the asymmetric brane model [1012, 1013] and the CGP
model [274]. We begin with the simplest generalisation, however, proposed by Karch and
Randall [682] (see also [667]).

The Karch-Randall model

Here we take the RS2 model and de-tune the brane tension, σ+ 6= 6M53 /l, so that one
no longer has Poincaré invariance along the brane. For an excess tension, σ+ > 6M53 /l,
the metric ḡµν is de Sitter, whereas for a tension deficit, σ+ < 6M53 /l, ḡµν is anti-de
Sitter. The effective cosmological constant on the brane is given by
" 2 #
σ+ 1
Λ4 = 3 − 2 . (609)
6M53 l

The behaviour of the warp factor also changes. We find that

a(z) = A cosh(c − |z|/l), c = cosh−1 (1/A), for anti-de Sitter, (610)

−1
a(z) = A sinh(c − |z|/l), c = sinh (1/A), for de Sitter, (611)
p
where A = l |Λ4 /3|. The decay of the warp factor away from the brane is greatest
for the dS brane, cutting off the space-time at z = lc. This means that gravity is
more strongly localised than in the standard RS2 scenario. In contrast, for the AdS
brane the warp factor turns around at some finite value of z. This means the bulk
volume is infinite, and gravity is not localised at all on the AdS brane as there is no
normalisable zero mode. Actually, when |Λ4 | is small (compared with M5 ), the AdS
brane exhibits quasi-localisation. This is because there is a normalisable mode that is
ultra-light, with mass m2ultra−light ∼ Λ24 /M5  |Λ4 | [1120]. At intermediate energies,
k  E  multra−light , the light mode behaves as if it were effectively massless and one
recovers 4D gravity. Note that there is no issue with the vDVZ discontinuity in AdS
when m2ultra−light /||Λ4 | . 0.1[721]. We say that gravity is only quasi-localised because
the extra-dimension opens up in the far infra-red, at energies E . multra−light .
Now consider what happens when we introduce a second AdS brane, along the lines
of an AdS generalisation of the RS1 model [720]. The first thing to note is that both
branes can have positive tension since the warp factor turns around. The second thing
to note is that the bulk volume is rendered finite, and so we have a zero mode as well
as the ultra-light mode. At energies above the mass of both modes gravity is mediated
by the exchange of two spin-2 fields, one massless and one massive. This corresponds
to a braneworld realisation of the bigravity scenarios discussed in Section 3.3 (see also
[723, 722, 1011]).
186
The GRS model

The GRS model [563] was developed by Gregory, Rubakov and Sibiryakov. The set-
up contains three Minkowski branes, one with positive tension and two with negative
tension. The positive tension brane is Z2 symmetric and is flanked on either side by a
section of anti-de Sitter space as far as a negative tension brane. Beyond the negative
tension brane lies an infinite region of Minkowski space. The warp factor goes like [563]
(
e−|z|/l |z| < zc
a(z) = −zc /l
(612)
e |z| > zc .

As the bulk volume is infinite, gravity is not localised, although the decaying warp fac-
tor around the positive tension brane gives some degree of quasi-localisation. Again,
although there is no zero mode, there is an ultra-light mode, and the extra dimension
opens up at very large distances. Unfortunately, the GRS model is known to be unstable
due to the presence of a ghost in the spectrum of linearised fluctuations [1038].

The asymmetric brane model

The asymmetric brane model [1012, 1013] (see also [1189]) is a single brane mode,l
like RS2, only without Z2 symmetry imposed across the brane. Indeed, the fundamental
parameters in the bulk are allowed to differ on either side of the brane, including the
bulk cosmological constant and the bulk Planck scales. Allowing the bulk Planck scales
to differ might seem strange, but not if we imagine a string compactification down to
five dimensions in which the dilaton is stabilised at different values on either side of a
domain wall (the brane). If the bulk cosmological constant and Planck scales are given
by ( (
−6/l12 z>0 M1 z>0
Λ= 2
, M= (613)
−6/l2 z<0 M2 z < 0.
The asymmetric model then admits Minkowski branes for a suitably tuned brane tension,
σ = 3(1 M13 /l1 + 2 M23 /l2 ), where 1 = ±1 and 2 = ±1. The corresponding solutions
have a warp factor of the form [1013]
(
e−1 z/l1 z>0
a(z) = (614)
e2 z/l2 z < 0.

The parameters 1 and 2 control whether the warp factor grows ( = −1) or decays
( = −1) away from the brane in a given direction. The model includes RS2 as a special
case.
It is, however, more interesting to consider the case where one of the warp factors
grows away from the brane while the other decays (e.g. 2 = −1 = 1). The bulk volume
is then infinite so that there is no zero mode, but by choosing the scales appropriately
one can engineer a degree of quasi-localisation. The point is that on the growing side the
graviton localises at the AdS boundary, where its sees an effective 4D Planck scale M13 l1 .
By taking this scale to be very large, the effect of localisation close to the AdS boundary
187
is almost decoupled from the gravitational dynamics near the brane. We should note that
the asymmetric model also admits self-accelerating solutions, just as in the DGP model.
In fact, the model shares a number of features with the DGP model. This is no coinci-
dence, since the DGP model can be obtained as a limiting case of the asymmetric model
[736]. It is known that the self-accelerating solutions of DGP contain ghost-like insta-
bilities [273, 1015, 559, 546], so the same is expected to be true for the asymmetric model.

The CGP model

The CGP model [274], developed by Charmousis, Gregory and Padilla, also exhibits
quasi-localisation. It combines the main features of both the asymmetric model and
the DGP model in that there is an induced gravity term, and asymmetry across the
brane. Interestingly, the model contains a new type of cosmological solution that tends
to Minkowski space at very late times, but undergoes an intermediate period of cosmic
acceleration in the presence of ordinary matter. In fact, it corresponds to a braneworld
√
realisation of the Cardassian cosmology, with H 2 ≈ 8πG 3 (ρ + c ρ) [512]. Although a
ghost is present when we introduce a small positive vacuum energy, this decouples in the
Minkowski limit [740].

5.3.4. Action and equations of motion

In each of the models described in our overview, we have a five dimensional bulk
space split into a series of domains separated from each another by 3-branes. Here we
consider the action and field equations for generic models of this type. The 3-branes may
be thought of as the boundaries of the various domains so that the action is given by
Z  3  X Z
5 √ M5 √ 

S= d x −γ R + Lbulk + d4 x −g −∆ M53 K + Lbrane ,
bulk 2
branes brane
(615)
where γab is the bulk metric with corresponding Ricci scalar, R, M5 is the bulk Planck
scale, and Lbulk is the Lagrangian density describing the bulk field content. In principle
both M5 and Lbulk can vary from domain to domain. For each brane gµν is the induced
metric and Lbrane is the Lagrangian density describing the field content on that particular
brane. K = g µν Kµν is the trace of extrinsic curvature, Kµν . This should be evaluated
on either side of the brane as it can differ from side to side. Labelling the two sides of a
given brane, using L and R, we define Kµν |L,R = 12 Ln|L,R gµν , i.e. extrinsic curvature is
given by the Lie derivative of the induced metric, with respect to the unit normal na |L,R .
The unit normal on both sides points from L to R. Note that what appears in the action
is the jump46

∆ M53 K = M53 K|R − M53 K|L .
This corresponds to the Gibbons-Hawking boundary term [536] for the bulk domains
on each side of the brane. The ∆ here is not to be confused with the 3-dimensional
Laplacian.
Now there are two (completely equivalent) ways to treat the brane contributions at
the level of the field equations. One approach is to treat them as delta-function sources in

46 Henceforth we define the jump of any quantity Q across a brane as ∆Q = Q|R − Q|L .
188
the Einstein equations. However, our preferred approach is to explicitly separate the field
equations in the bulk from the boundary conditions at the brane. The bulk equations of
motion are then given by the bulk Einstein equations
1 1 bulk
Gab = Rab − Rγab = 3 Tab , (616)
2 M5
bulk 2 δ
R √
where Tab = − √−γ δγ ab bulk
d5 x −γLbulk is the bulk energy-momentum tensor. The
boundary conditions at Σi are given by the Israel junction conditions [649]
 
∆ M52 (Kµν − Kgµν ) = −Tµν brane
, (617)

(i) R √
where Tµν = − √2−g δgδµν brane d4 x −gLbrane is the brane energy-momentum tensor.
Note that in each of these examples, the bulk geometry is only sourced by a cosmological
bulk
constant, Tab = −M53 Λγab .

5.3.5. Linear perturbations in RS1 and RS2

We now consider the theory of linear perturbations. For brevity, we will restrict at-
tention to RS1 and RS2, although the methods we use are fairly standard, and should
apply to all RS-like models (for further details, see [528, 275, 1038, 1011]).

Weak gravity on a RS1 brane

It is enough to consider RS1, as RS2 can be readily obtained by taking the negative
tension brane to infinity. Recall that the background metric, γ̄ab , is given by Equation
(604), with the positive tension brane (the “+” brane) fixed at z = 0 and the negative
tension brane (the “−” brane) fixed at z = zc . We see that the induced metric on the
(+) (−)
“+” brane is given by ḡµν = ηµν , and on the “−” brane by ḡµν = e−2zc /l ηµν . Note
that we have Z2 symmetry across the branes, so we can restrict attention to 0 ≤ z ≤ zc .
We now consider small perturbations about the background, so that the metric is
given by γab = γ̄ab + δγab . It is convenient to choose Gaussian Normal (GN) gauge,
defined by
δγµz = δγzz = 0. (618)
Actually, this is only a partial gauge fixing. Since we have no additional bulk matter,
we can also take the metric to be transverse and trace-free in the bulk. In other words,
δγµν = χµν (x, z), where
∂ν χνµ = χµµ = 0. (619)
This is known as Randall-Sundrum (RS) gauge [1052]. In RS gauge, the linearised bulk
equations of motion, δGab = 6l2 δγab , yield
 
∂2 4
e2z/l ∂ 2 + 2 − 2 χµν = 0, (620)
∂z l

where ∂ 2 = ∂µ ∂ µ .
Unfortunately, we can no longer assume that the branes are fixed at z = 0 and z = zc .
The presence of matter on the branes will cause them to bend [528] so that they will now
189
be positioned at z = f+ (x) and z = zc + f− (x), for some functions f± that depend only
on the coordinates xµ . This makes it difficult to apply the Israel junction conditions
at the branes. To get round this we can apply a gauge transformation that fixes the
position of the “+” brane, and another that fixes the position of the “−” brane [275],
without spoiling the Gaussian Normal condition (618). This gives rise to two coordinate
patches that are related by a gauge transformation in the region of overlap. We will call
them the “+” patch and the “−” patch accordingly.
We first consider the “+” brane. To fix its position we make the following coordinate
transformation
l
z → z − f+ (x), xµ → xµ + (1 − e−2z/l )∂µ f+ . (621)
2
The “+” brane is now fixed at z = 0, although the other brane is now at z = zc +f− −f+ .
It follows that the metric perturbation in the “+” patch is given by
2
δγµν = χ(+)
µν (x, z) = χµν (x, z) − l(1 − e
−2z/l
)∂µ ∂ν f+ − f+ γ̄µν . (622)
l
Similarly, to fix the position of the “−” brane we let
l 
z → z − f− (x), xµ → xµ + 1 − e−2(z−zc )/l ∂µ f− . (623)
2
Now we have the “−” brane at z = zc , but with the “+” brane at z = f+ − f− . The
metric perturbation in the “−” patch is given by
  2
δγµν = χ(−)
µν (x, z) = χ µν (x, z) − l 1 − e −2(z−zc )/l
∂µ ∂ν f− − f− γ̄µν . (624)
l
(+) (+) (+)
Now the induced metric on the “+” brane is given by gµν = ḡµν + δgµν , where

(+) 2
δgµν = χ(+) (+)
µν (x, 0) = χµν (x, 0) − f+ ḡµν , (625)
l
(−) (−) (−)
whereas on the “−” brane it is given by gµν = ḡµν + δgµν , where

(−) 2
δgµν = χ(−) (−)
µν (x, zc ) = χµν (x, zc ) − f− ḡµν . (626)
l
We are now ready to make use of the linearised Israel junction conditions given by Eq.
(617) at each brane to find
 (±)
∆ M52 δ(Kµν − Kgµν ) (±)
= σ± δgµν (±)
− Tµν , (627)
(±)
where σ± = ±6M53 /l is the tension on the “±’ brane, and Tµν is the energy-momentum
tensor for matter excitations. Now, owing to the Z2 symmetry, the extrinsic curvature
simply changes by a sign when evaluated on either side of a given brane. It follows that
the linearised boundary conditions at each brane are given by
  
∂ 2  (+)
+ χµν  = −Sµν , (628)
∂z l z=0
  
∂ 2  (−)
+ χµν  = −Sµν , (629)
∂z l z=zc
190
where  
± 1 (±) 1 (±) (±)
Sµν (x) = ± T − T ḡ − 2∂µ ∂ν f± , (630)
M53 µν 3 µν

µν (±)
and where T (±) = g(±) Tµν is the trace of the appropriate energy-momentum tensor.
Indeed, taking the trace of these equations, and using the fact that χµν is traceless, we
clearly see that matter on a brane causes it to be bend, such that

∂ 2 f± T (±)
2 =∓ , (631)
a± 6M53

where a± gives the warp factor at the “±” brane (i.e. a+ = 1 and a− = e−zc /l ). Equations
(620), (628) and (629) give the governing differential equations, and a complete set of
boundary conditions for the graviton mode χµν . We now take Fourier transforms along
1
R 4 −ip xµ
the brane directions, Q(x, . . .) → Q̃(p, . . .) = (2π)2 d x e µ Q(x, . . .), to find that
Z
1 µ
χµν (x, z) = d4 p eipµ x χ̃µν (p, z), (632)
(2π)2

where  
∂2 4
−p2 e2z/l + 2 − 2 χ̃µν = 0, (633)
∂z l
and
     
∂ 2  (+) ∂ 2  (−)
+ χ̃µν  = −S̃µν (p), + χ̃µν  = −S̃µν (p). (634)
∂z l z=0 ∂z l z=zc

This system is easily solved to give

χ̃µν (p, z) = C (+) (p, z)S̃µν

(+)
− C (−) (p, z)S̃µν
(−)
, (635)

where
1 h        i
C (+) (p, z) = − I1 plezc /l K2 plez/l + K1 plezc /l I2 plez/l , (636)
p det A(p)

and
1 h    i
C (−) (p, z) = − e−zc /l I1 (pl) K2 plez/l + K1 (pl) I2 plez/l . (637)
p det A(p)

Note that In , and Kn are modified Bessel’s functions of integer order n [6], and
   
det A(p) = I1 (pl) K1 plezc /l − K1 (pl) I1 plezc /l . (638)

From Eq. (631) we also have

Z
1 a2± µ T̃ (±)
f± (x) = ± d4 p eipµ x . (639)
(2π)2 M53 6p2
191
To compute the metric perturbation on each of the branes we simply use Eqs.(622) and
(624), given our knowledge of χµν (x, z) and f± (x). At the positive tension brane we then
have
Z n h
1 1 ipµ xµ α+ (+) (+) i
χ(+)
µν = d4
p e C (+)
(p, 0) T̃ (+)
µν − T̃ ḡµν
(2π)2 M53 2
 
e−zc /l (−) 1
− 2 T̃µν − T̃ (−) ḡµν (−)
+ pure gauge terms, (640)
lp det A(p) 3

whereas at the negative tension brane we have

Z n h
1 1 µ α− (−) (−) i
χ(−)
µν = d4 p eipµ x C (−) (p, zc ) T̃µν
(−)
− T̃ ḡµν
(2π)2 M53 2
 
e−zc /l (+) 1
− 2 T̃µν − T̃ (+) ḡµν
(+)
+ pure gauge terms. (641)
lp det A(p) 3

The parameters α± are crucial as they control the tensor structure of the propagator on
the “±” branes. They are given by
   
2 1 2 e−2zc /l
α+ = 1 + 2 (+) , α− = 1 − 2 (−) . (642)
3 lp C (p, 0) 3 lp C (p, zc )

Using the properties of modified Bessel functions [6], we can show that at low energies,
p  ke−zc /l , we have [528]
Z    +  
1 ipµ xµ 2 wBD + 1
χ(+)
µν ≈ d pe4 (+)
2 p2 T̃µν − + T̃ (+) (+)
ḡµν
(2π)2 M+ 2wBD +3
 
2 (−) 1
+ 2 2 T̃µν − T̃ (−) ḡµν
(−)
+ pure gauge terms, (643)
M− p 3

and
Z    −  
e−2zc /l µ 2 wBD + 1
χ(−)
µν ≈ d4 p eipµ x 2 p2 T̃ (−)
µν − − T̃ (−) (−)
ḡµν
(2π)2 M− 2wBD +3
 
2 (+) 1
+ 2 2 T̃µν − T̃ (+) ḡµν
(+)
+ pure gauge terms, (644)
M+ p 3
±
where M± and wBD are the 4D effective Planck scale and Brans-Dicke parameter on the
“±” brane, respectively. They are given by Equations (605) and (607). We can now see
explicitly how BD gravity emerges as the low energy effective theory in RS1, on both
branes, as claimed in Section 5.3.1.

Graviton spectrum

Let us now pause to comment on the mass spectrum for the graviton. The spectrum
can be obtained by identifying the poles in the propagator. These are given by the
192
solutions to p2 = −m2 , where det A(p) = 0. This gives
   
I1 (iml) K1 imlezc /l − K1 (iml) I1 imlezc /l = 0
   
=⇒ J1 (ml) Y1 mlezc /l − Y1 (ml) J1 mlezc /l = 0, (645)

where J1 and Y1 are Bessel’s functions of order one. By solving this equation we see
that there is a zero mode, and a tower of heavy Kaluza-Klein modes with mass splitting
∆m ∼ 1/l(ezc /l − 1).

Radion effective action

The finite Brans-Dicke parameter indicates the presence of a massless scalar in ad-
dition to the massless graviton. This is due to the radion, or brane bending mode,
δf = f+ − f− . In some RS-like models (e.g. GRS), the radion can exhibit pathological
behaviour that can only be revealed by computing the effective action [1038]. We will
now briefly outline the procedure for doing this.
(±)
Let us consider vacuum fluctuations (Tµν = 0) in the scalar sector. The first thing
to note is that the Field Equations (620), (628) and (629) admit a solution of the form
[275]
l2
χ(rad)
µν = − e2z/l ∂µ ∂ν ψ, (646)
4
where ∂ 2 ψ = 0 and the vacuum boundary conditions require

vac l
f± =− ψ, (647)
2a2±
vac
where f± is the vacuum fluctuation in f± . Note that Equation (631) implies ∂ 2 f± vac
= 0,
2
and so Equation (647) is consistent with ∂ ψ = 0. Equation (647) also imposes a relation
vac vac
between f+ and f− , resulting in a single free scalar degree of freedom, which we take
to be the physical radion mode, ψ. Note that the radion profile in the bulk is localised
close to the “−” brane, in contrast to the graviton zero mode which is localised close
to the “+” brane. We now work in the “+” patch, which has the branes positioned at
z = 0 and z = zc − δf vac . Focusing solely on the scalar sector, the metric perturbation
is given by

l2 2z/l 2 vac
δγµν = − e ∂µ ∂ν ψ − l(1 − e−2z/l )∂µ ∂ν f+
vac
− f+ γ̄µν , δγµz = δγzz = 0. (648)
4 l
In order to integrate out the extra dimension, it is convenient to have both branes fixed.
We can do this with the following coordinate transformation
Z z
vac −2z/l vac
z → z + B(z)δf , xµ → xµ − e ∂µ (δf ) e2y/l B(y)dy, (649)
0

where B(z) is some differentiable function for 0 ≤ z ≤ zc , satisfying B(0) = 0 and

B(zc ) = 1. While this transformation ensures that δγµz is still zero, the price we pay for

193
fixed branes is that we now have non-vanishing δγzz . More precisely,

l2 2z/l 2  vac 
δγµν = hµν = − e ∂µ ∂ν ψ − l(1 − e−2z/l )∂µ ∂ν f+
vac
− f+ − (δf vac )B(z) γ̄µν
4 Z l z
+2e−2z/l ∂µ ∂ν (δf vac ) e2y/l B(y)dy, (650)
0
δγzz = hµz = −2δf vac B 0 (z). (651)

To quadratic order, the effective action is given by

Z zc Z  
M3 √ 6
Seff =− 5 4 ab
dz d x −γ̄ h δ Gab − 2 γab
2 0 l
Z p h i
1
(+)
+ d4 x −ḡ (+) hµν ∆ M52 δ(Kµν − Kgµν ) − σ+ hµν
2 z=0
Z p h i
1
(−)
+ d4 x −ḡ (−) hµν ∆ M52 δ(Kµν − Kgµν ) − σ− hµν . (652)
2 z=zc
 
(±)
It turns out that δ Gµν − 6l2 γµν and ∆ M52 δ(Kµν − Kgµν ) − σ± hµν are identically
zero [1038]. After integrating out a total derivative in z, we arrive at the following
effective action for the radion:
Z
3 2
Sradion = − M− d4 x (∂µ ψ)2 , (653)
2

where M− is given by Equation (605). Thus, as expected, the radion, ψ, behaves as

a massless scalar, and, being localised close to the “−” brane, its coupling strength is
controlled by the scale M− . As we move the branes further and further away from one
another, M− increases and the radion starts to decouple (decoupling completely in the
RS2 limit). Note that in some models, such as GRS [563], the radion effective action
comes in with the wrong overall sign, signalling the presence of a physical ghost [1038].
There is no such pathology in RS1 or RS2.

Weak gravity on a RS2 brane

Let us now focus on RS2 by taking the negative tension brane to infinity. As zc → ∞,
the metric on the remaining positive tension brane is given by
 
Z K1 (pl)
1 µ K2 (pl)  (+) 1 + plK2 (pl) (+) (+) 
χ(+)
µν = d4 p eipµ x T̃ − T̃ ḡµν
(2π)2 M53 pK1 (pl) µν 3

+ pure gauge terms. (654)

At low energies, p  k, we again use the properties of modified Bessel functions [6] to
show that
Z  
(+) 1 4 ipµ xµ 2 (+) 1 (+) (+)
χµν ≈ d pe T̃ − T̃ ḡµν + pure gauge terms. (655)
(2π)2 M42 p2 µν 2
194
It follows that gravity is indeed localised on the brane at low energies, with a 4D effective
Planck scale M42 = M53 l, or, equivalently, with an effective Newton’s constant G4 = G5 /l.
Note that the tensor structure of the propagator matches that of General Relativity, with
a factor of − 12 in front of the trace term. This is where the brane bending mode f+ (x)
plays a crucial role. It cancels part of the graviton zero mode in just the right way
to guarantee good agreement with solar system gravity tests. It is worth noting that
this neat cancellation of terms does not always happen, even in single brane scenarios.
We will see this in DGP gravity, for example, where one has to argue for some sort of
Vainshtein effect to pass observational tests.
Let us finally consider massive modes. In RS2 there is actually a continuum of
massive modes, consistent with the fact that the extra dimension is no longer compact.
However, as we have seen, we still recover 4D GR to leading order. The next to leading
order corrections are obtained by integrating over the continuum, or, equivalently, by
considering the next to leading order expansion in pl above. The result is that the
Newtonian potential reads47
 
1 2l2
V (r) ∝ 1 + 2 + O(l3 /r3 ) .
r 3r

We refer the reader to [528, 1008, 438] for details of the derivation. Note that the 1/r3
correction can be obtained in a dual picture as the one loop CFT correction to the
graviton propagator [438].

5.4. Brane Cosmology

There are two obvious reasons why cosmology offers an interesting arena in which
to develop the braneworld paradigm. The first is that cosmological branes possess a
high degree of symmetry, and this makes it possible to solve the field equations. The
second is that cosmological physics can be tested by a number of observations, ranging
from supernova data to the abundance of light elements. In this section we will study
the cosmology of co-dimension one branes, focusing on the RS2 scenario with a single Z2
symmetric brane. We will review the background dynamics [159, 178] before moving on to
cosmological perturbation theory [716, 1244]. Further details, including generalisations to
multi-brane scenarios with bulk scalar fields, can be found in the following review articles
[1048, 845, 639, 770, 189, 195] (see also [1008, 346, 347, 1139]). Other generalisations
include: the cosmology of branes without Z2 symmetry [1012, 1189, 1143]; anisotropic
braneworlds [843, 966]; and branes for which energy is explicitly transferred between
bulk and brane [694, 695, 1234]. Note that any anisotropy is seen to dissipate on the
brane in RS gravity, a feature that can be identified with CFT particle production in the
holographic picture [966]. Bulk-brane energy transfer has been used to account for dark
energy [694, 695, 1234].
Braneworld cosmology can be studied using two different formalisms: The brane based
formalism, and the bulk based formalism. These two approaches are completely equivalent
and yield a background cosmology governed by the following Friedmann equations [159,

47 Note that this result differs slightly from the original one quoted in [1052].

195
158, 178]
κ Λ4 8πG4  ρ  µ
H2 + = + ρ 1+ + 4, (656)
a2 3 3 2σ a
κ  ρ  2µ
Ḣ − 2 = −4πG4 (ρ + P ) 1 + − 4, (657)
a σ a
where H = ȧ/a is the Hubble parameter along the brane, a(t) is the scale factor, and
κ = 0, ±1 describes the spatial curvature. The brane is sourced by a tension, σ, and
a cosmological fluid with energy density, ρ(t), and pressure, P (t). The parameters Λ4
and G4 denote the effective cosmological constant and Newton’s constant on the brane,
respectively. As in the standard scenario, the Raychaudhuri Equation (657) follows from
the Friedmann Equation (656) and energy conservation,
ρ̇ + 3H(ρ + P ) = 0. (658)
We will now review the derivation of this cosmology using the two equivalent for-
malisms. For the moment, however, let us comment on a few of its important features.
From Equations (656) and (657) we see that the corrections to the standard cosmology
manifest themselves in a term ∝ ρ2 , and a dark radiation term aµ4 = 8πG 3 ρweyl . The
4

latter corresponds to a non-local “Weyl” contribution and can only be fixed by specifying
the bulk geometry. In the holographic description of RS2, the ρ2 corrections contribute
to the conformal anomaly [1138], while the dark radiation is identified with thermal
excitations of the CFT [570, 1100, 1009, 556, 1014, 596].
Both corrections will strongly affect Big Bang nucleosynthesis (BBN), so their mag-
nitude can be constrained by the abundance of light elements. These require that the
dark radiation can be at most 10% of the photon energy density, ρweyl /ργ . 0.01, in the
period following BBN [638]. In addition, the ρ2 corrections to the cosmological evolution
should be negligible after BBN, which imposes a constraint48 on the tension σ & (MeV)4
[159].
The ρ2 corrections play the dominant role in the very early universe and will have a
big impact on the inflationary dynamics. Assuming the inflaton is confined to the brane
along with all the Standard Model fields, one finds that the slow roll parameters are
given by [847]
 0 2 " #
Ḣ 1 V 1 + Vσ
 = − 2 =
, (659)
H 16πG4 V V 2
1 + 2σ
 00  " #
φ̈ 1 V 1
η = − = V
, (660)
H φ̇ 8πG4 V 1 + 2σ

where V (φ) is the inflaton potential. At low energies, V  σ, these match the standard
formulae of 4D General Relativity,
 0 2  00 
1 V 1 V
 ∼ GR = , η ∼ ηGR = .
16πG4 V 8πG4 V

48 Assuming a RS2 scenario with fine tuned tension, σ = 12M 3 /l, the BBN constraint is significantly
5
weaker than the constraint arising from table top experiments quoted in Section 5.3.2, i.e. 1/l & 10−4
eV, giving σ & (100GeV) [189]. Here we have used the fact that M42 = M53 l ∼ 1016 TeV.
4

196
However, at high energies we have
4σ 2σ
 ∼ GR  GR , η ∼ ηGR  ηGR . (661)
V V
This means we can get away with steeper potentials in the braneworld case [847]. This
is essentially because Hubble friction gets enhanced by the ρ2 terms. However, such
potentials are incompatible with observational constraints as they lead to a large tensor-
to-scalar ratio [800].

5.4.1. Brane based formalism – covariant formulation

We shall now derive the background cosmology, Eqs. (656) and (657), using the
brane based formalism. This makes use of the Gauss-Codazzi equations [531, 320], and
the Israel junction conditions to derive the Einstein tensor on the brane [1140]. We
will assume Z2 symmetry, although we refer the reader to [117] for a non-Z2 symmetric
generalisation.
The Gauss-Codazzi equations are the fundamental equations of embedded hyper-
surfaces. The brane can be thought of as an embedding,

xa = X a (ξ µ ), (662)

in the bulk geometry, γab (x). By Z2 symmetry this is the same on both sides of the
brane. We can now define tangent vectors Vµa = ∂X a /∂ξ µ , and the outward pointing
unit normal, na , satisfying

γab na nb = 1, γab na Vµb = 0.

It follows that the induced metric on the brane is given by

gµν = γab (X)Vµa Vνb . (663)

For such an embedding, the Gauss-Codazzi equations [531, 320] give

Rµναβ = Rabcd Vµa Vνb Vαc Vβd + Kµα Kνβ − Kµβ Kνα , (664)
µ a
∇ (Kµν − Kgµν ) = Rab n Vνb , (665)
2 µν a b
R − K + Kµν K = −2Gab n n , (666)

where Rabcd is the Riemann tensor in the bulk, Rµναβ is the Riemann tensor on the
brane, and we recall that the extrinsic curvature of the brane is given by Kµν = 21 Ln gµν ,
the Lie derivative of the induced metric, with respect to the normal.
Now, in the RS2 scenario the dynamics of the brane are governed by the bulk equa-
tions of motion and the Z2 symmetric junction conditions
6
Gab = γab , Kµν − Kγµν = 3σ∗ gµν − 4πG5 Tµν , (667)
l2
with σ∗ = 4πG3
5σ
, where σ is the brane tension and Tµν is the energy-momentum tensor
of additional matter excitations. Of course, in the RS2 scenario we have a fine-tuned
tension such that σ∗ = 1/l, but we will keep things general for the moment.
197
 Consider the Codazzi Equation (665). Because the bulk is only sourced by a cosmo-
logical constant, the right hand side of this equation is identically zero. This is important,
because the junction conditions in Eq. (667) now imply the usual conservation law along
the brane, ∇µ Tµν = 0. To extract information about the Einstein tensor on the brane,
we contract the Gauss Equation (664), and plug in Eq. (667), to give [1140]

Gµν = −Λ4 gµν + 8πG4 Tµν + (4πG5 )2 Πµν − Eµν , (668)

where Λ4 = 3 σ∗2 − 1l2 is the effective cosmological cosmological constant on the brane,
and the effective Newton’s constant is given by

G4 = G5 σ∗ . (669)

The corrections to standard 4D gravity are encoded in a local contribution, Πµν , and a
non-local “Weyl” contribution, Eµν . The local piece is a quadratic combination of the
energy-momentum tensor
1 1 1
Πµν = −Tµα Tνα + T Tµν + T αβ Tαβ gµν − T 2 gµν , (670)
3 2 6
whereas the non-local “Weyl” piece is

Eµν = Cabcd na Vµb nc Vνd , (671)

where Cabcd is the Weyl tensor of the bulk. It is often referred to as the electric part
of the Weyl tensor, and in general one must solve the bulk equations of motion first in
order to evaluate it in full. We should also note that Eµν has vanishing trace, and its
divergence is sourced by the local quadratic contribution

g µν Eµν = 0, ∇µ Eνµ = (4πG5 )2 ∇µ Πµν . (672)

The latter equation follows from the divergence of Equation (668), making use of the
Bianchi identity and the local conservation of energy-momentum, ∇µ Tµν = 0.
Our interest here lies in the cosmology, so let us now assume spatial homogeneity and
isotropy on the brane. The induced metric is given by the usual Friedmann-Lemaı̂tre-
Robertson-Walker (FLRW) metric

ds2 = gµν dξ µ dξ ν = −dt2 + a(t)2 qij dxi dxj , (673)

where qij (x) is the metric of a hyper-surface of constant curvature, κ = 0, ±1. The matter
excitations contribute a cosmological fluid with energy density, ρ(t), and pressure, P (t),
such that its energy-momentum tensor is given by

Tνµ = diag(−ρ(t), P (t), P (t), P (t)). (674)

In the absence of non-trivial sources in the bulk Tνµ is conserved, which means we have
the standard relation given by Equation (658).
It remains to compute the non-local piece, Eµν . Although in general we must solve the
bulk equations of motion to evaluate this, we can exploit the large amount of symmetry

198
to avoid doing so in the current instance. Making use of the Constraints Equations (672),
we can show that  
µ 1 1 1
Eνµ = 4 diag −1, , , ,
a 3 3 3
where µ is an integration constant that should be fixed by the bulk geometry (we will
see in the next section that it can be identified with the mass of a bulk black hole). Note
that Eνµ is conserved on a cosmological background, by virtue of the fact that ∇µ Πµν = 0.
The modified Friedmann Equations (656) and (657) now follow automatically from
Equation (668).

5.4.2. Bulk based formalism – moving branes in a static bulk

The principle limitation of the brane based formalism we have just described is that
it suppresses physics deep inside the bulk. This can be dangerous since it is not at all
obvious if a particular choice of the non-local “Weyl” term, Eµν , at the brane will evolve
into a pathological bulk geometry. Such problems can be avoided by solving for the bulk
geometry first. For example, in the brane based formalism, the parameter µ giving rise
to dark radiation, is just an integration constant and can take any sign. Using the bulk
based formalism, we are able to identify µ with the mass of a bulk black hole, and use
this to constrain it to be positive.
We will now develop the bulk based formalism for cosmological branes. Again, we
will assume Z2 symmetry across the branes, for brevity. The reader can refer to [1008]
for the non-Z2 symmetric generalisation. The bulk based formalism requires us to solve
for the bulk geometry. Since we are interested in cosmological branes (with constant
curvature Euclidean 3-spaces), we study the Einstein equations, Gab = 6l2 γab , with the
following metric ansatz [178]

ds2 = γab dxa dxb = e2ν A−2/3 (−dt2 + dz 2 ) + A2/3 qij dxi dxj , (675)

where A and ν are undetermined functions of t and z, and as before qij (x) is the metric
of a hyper-surface of constant curvature, κ = 0, ±1. Now, in an extremely elegant
calculation, Bowcock et al. [178] were able to prove a generalised form of Birkhoff’s
theorem, showing that the bulk geometry is necessarily given by

dr2
ds2 = −V (r)dτ 2 + + r2 qij dxi dxj , (676)
V (r)

where
r2 µ
V (r) = + κ − 2. (677)
l2 r
For µ > 0, the metric in Eq. (676) takes the form of a (topological) Schwarzschild black
hole in anti-de Sitter space. Here we have written the solution in an explicitly time
independent coordinate system, meaning that we can no longer say that we have a static
brane sitting at a fixed coordinate position. On the contrary, we now have a dynamic
brane, whose trajectory in the these coordinates is more complicated. Braneworld cos-
mology from this perspective was first studied by Ida [639], although moving branes in
a static anti-de Sitter bulk were considered earlier by Kraus [754].

199
 To construct the brane solution, we treat it as an embedding,

τ = τ (t), r = a(t), (678)

of the bulk geometry given in Eq. (676). The induced metric on the brane is then
 
ȧ2
ds2 = gµν dξ µ dξ ν = −V (a)τ̇ 2 + dt2 + a2 (t)qij dxi dxj (679)
V (a)

where over-dots denote ∂/∂t. We are free to choose t to correspond to the proper time
with respect to an observer comoving with the brane. This imposes the condition

ȧ2
− V (a)τ̇ 2 + = −1, (680)
V (a)

ensuring that the brane takes the standard FLRW form, as in Eq. (673). The function
a(t) is then immediately identified with the scale factor along the brane.
The boundary condition at the brane are given in Eq. (667). We must compute the
extrinsic curvature, Kµν = 21 Ln gµν , defined as the Lie derivative of the normal pointing
into the bulk. Assuming we cut away the AdS boundary and retain the region r < a(t),
we find that the inward pointing unit normal is given by

na = (−ȧ, τ̇ , 0, 0, 0), (681)

where we are free to specify that τ̇ > 0. The components of extrinsic curvature are then
given by  
i V τ̇ i t ä + V 0 /2
Kj = δ , Kt = − . (682)
a j V τ̇
In the presence of a cosmological fluid, as given by Equation (674), the junction conditions
in Eq. (667) yield the following:

V τ̇ ρ
= σ∗ (1 + ), (683)
a σ
 
ä + 12 V 0 ρ P
= σ∗ 1 − 2 − 3 . (684)
V τ̇ σ σ
Making use of Equation (680) we then arrive at the modified Friedmann Equations (656)
and (657).

5.4.3. Cosmological perturbations

While the theory describing cosmological perturbations in braneworld gravity has
been well developed in recent years (see, for example, [842, 771, 847, 334, 1087, 991, 801,
202, 201, 548, 772, 547, 918, 919, 921, 920, 596, 716, 768, 1244, 742, 1245, 703, 705, 715,
769, 597, 408, 194, 430, 955, 286, 300, 395, 1062, 787, 245]), even approximate solutions
to the resulting field equations have been notoriously hard to come by. The problem
stems from the fact that one has to solve the fully coupled system of brane and bulk,
which is, in general, a far from trivial task. Indeed, this will generically render the brane
based formalism somewhat incomplete without making ad hoc assumptions about the
200
perturbations of the non-local “Weyl” contribution, Eµν . The bulk based formalism is
better in this respect, but the resulting system is virtually intractable, and only in some
special cases, where the bulk and brane equations become separable, has progress been
made [528, 596, 202, 201, 772]. Note that, unlike in other sections, we will on occasion
include vector and tensor perturbations, as well as scalars. This is because the bulk can
source vector and tensor modes on the brane, giving qualitatively different behaviour to
that seen in standard 4D cosmology.

Cosmological perturbations in the brane based formalism

We shall now review some aspects of cosmological perturbation theory using the brane
based formalism introduced in Section 5.4.1. Further details can be found in [845, 842].
The dynamics on the brane are governed by Equation (668), and so perturbations about
the background cosmology on the brane satisfy

δGµν = 8πG4 δT µν + (4πG5 )2 δΠµν − δEνµ . (685)

To study this we use the standard four-dimensional formalism [90, 717, 469, 917], decom-
posing the system into scalar, vector and tensor perturbations with respect to the spatial
diffeomorphism group in the background cosmology. Working with conformal time as
opposed to proper time, the perturbed metric is given by
h
~ i β)dxi dτ
ds2 = a2 −(1 + 2Ψ)dτ 2 − 2(βi + ∇
h i i
~ (i Aj) + hij dxi dxj , (686)
+ (1 − 2Φ)qij + Dij ν + 2∇

where Ai and βi are transverse vectors on qij , and hij is a transverse and trace-free tensor.
Recall that the operator Dij = ∇ ~ i∇
~ j − 1 qij ∆. The fluctuations energy-momentum on the
3
brane are written in terms of the fluctuations in density, δ, pressure, δP , fluid 3-velocity,
vi , and anisotropic stress, Σij , in the usual way,

δT00 = −ρδ, δTi0 = −(ρ + P )vi , δTji = δP δji + (ρ + P )Σij . (687)

The fluid 3-velocity and anisotropic stress can then be decomposed with respect to their
scalar, vector and tensor components,
(vector) ~ i θ,
vi = θi +∇ (688)
(tensor) ~ (i Σ(vector) + Dij Σ.
Σij = Σ ij +∇ j) (689)

The fluctuation in the quadratic piece is also given in terms of ρδ, δP . . . etc., according
to
2 2
δΠ00 = ρδT00 , δΠ0i = ρδTi0 ,
3 3  
i 1 i 2 i 0 1 k
δΠj = − (ρ + 3P )δTj − (ρ + P )δj δT0 − δTk , (690)
3 3 2

where δTνµ are given by Equations (687).

201
 Now we consider the contribution from the non-local “Weyl” perturbation, and iden-
tify it with the fluctuation in some dark energy energy-momentum tensor

δEµν = −8πG4 δT weyl

µν .

In direct analogy with Equations (687), we can read off corresponding fluctuations in
Weyl energy density, δ weyl , Weyl pressure, δP weyl , Weyl fluid 3-velocity, viweyl , and Weyl
anisotropic stress, Σweyl
ij . We might hope to determine each of these in terms of the local
matter fluctuations, δ, δP . . . etc., by making use of the Constraints Eqs. (672) to linear
order. We find that for scalar perturbations δEµµ = 0 gives49

1 weyl weyl
δP weyl = ρ δ , (691)
3
 
and ∇µ δEνµ − (4πG5 )2 Πµν = 0 gives
 
weyl 0 0 4 1 0 weyl
(δ ) + 4Ψ + ∆ β + ν − θ = 0, (692)
3 2
 
4 weyl weyl 0 2 weyl δ weyl
− ρ (θ ) + Φ − (∆ + 3κ)Σ − =
3 3 4
ρ+P
{ρδ + (ρ + P )[3Hθ − (∆ + 3κ)Σ]} , (693)
σ
where 0 denotes differentiation with respect to conformal time, and H = a0 /a = aH.
Clearly we have too many Weyl unknowns and not enough equations. The bottom line is
weyl
that we need to know the Weyl anisotropic stress δπij explicitly, and for that we need
to abandon the brane based formalism and solve the bulk equations of motion [842].
Some progress can be made at super-horizon scales, since then we can neglect the
spatial gradients in Eqs. (692) and (693), and solve for the Weyl energy density and
Weyl momentum in terms of ρδ, δP . . . etc., thereby closing the system [842, 548, 771].
This simplification has been applied to the study of both density perturbations and
vector perturbations at large scales, revealing qualitatively different behaviour to that
in General Relativity [548, 844]. Regarding density perturbations, it can be shown that
the quantity ρa4 ∆δ will grow during slow-roll inflation on super-horizon scales (it stays
constant in GR) [548]. It can be shown that vector perturbations can be non-vanishing,
even in the absence of vorticity [844].
We can also solve for the (total) curvature perturbation on large scales [771]. Unfor-
tunately, this does not mean we can compute large-scale CMB anisotropies. The problem
is that to evaluate the (non-integrated) Sachs-Wolfe equation we need knowledge of the
metric perturbations. These are sourced by the dark anisotropic stress, according to
 
4 weyl
Φ̂ − Ψ̂ = 8πG4 a2 ρ Σweyl , (694)
3

49 The remaining formulae in this section are taken from [771], where one should identify δρ = ρδ, δq =

−(ρ + P )θ, δπ = (ρ + P )Σ, R = Ψ, A = −Φ, E = 12 ν, B = −β.

202
where Ψ̂ and Φ̂ are the Bardeen gauge invariants for the metric perturbations, and we
have neglected the local anisotropic stress. The braneworld corrections to the Sachs-
Wolfe effect are given by [771]
 
δT δT  8 ργ
=  − Sweyl
T T GR 3 ρcdm
  Z  
4 weyl 16πG4 4 weyl
− 8πG4 a2 ρ Σweyl + 5/2 da a7/2 ρ Σweyl , (695)
3 a 3
where Sweyl is the Weyl entropy perturbation, determined by ρδ weyl .

Cosmological perturbations in the bulk based formalism

As we saw in the previous section, a proper treatment of cosmological perturbation

theory in brane cosmology requires us to solve the coupled system of brane and bulk.
We will now present the details of this in the bulk based formalism, essentially following
[716] (see also [1244, 918, 919, 921, 920]). Note that in this section we deviate from our
usual convention of treating cosmological perturbations with respect to conformal time
on the brane, preferring instead to use proper time of comoving observers, in keeping
with the majority of the relevant literature. For consistency, however, we do define the
brane quantities as in the previous section.
We begin with some notation on the background. The bulk metric is given by
ds2 = γ̄ab dxa dxb = λαβ dxα dxβ + r2 qij dxi dxj , (696)
where λαβ is some two-dimensional metric, and λαβ and r depend only on the first two
coordinates, xα . This corresponds to a section of (topological) AdS-Schwarzschild, and
one can choose a gauge such that
dr2 r2 µ
λαβ dxα dxβ = −V (r)dτ 2 + , V (r) = 2
+ κ − 2. (697)
V (r) l r
However, here we will leave the choice of gauge unspecified so that our analysis can
be applied in other gauges (e.g. Gaussian-Normal gauge). The background embedding
equation is now given by xa = X̄ a (ξ µ ), where ξ µ = (t, ξ i ) are the coordinates along the
brane. We take
X̄ α = X̄ α (t), X̄ i = ξ i ,
and λ (X̄)X̄˙ α X̄˙ β = −1 to ensure that the induced metric on the brane has the standard
αβ
FLRW form, as in Eq. (673), with scale factor a(t) = r(X̄). Over-dots here denote
differentiation with respect to proper time on the brane, t.
We shall now specify the perturbations in the bulk. Decomposing the bulk metric in
terms of scalar, vectors and tensor components (with respect to qij ), we write δγab =
hscalar
ab + hvector
ab + htensor
ab , where,

hscalar = χαβ , hscalar ~ i χα ,

= r∇ hscalar = 2r2 [Aqij + Dij E] , (698)
αβ αi ij

hvector
αβ = 0, hvector
αi = rBαi , hvector
ij
~ (i Hj) ,
= 2r2 ∇ (699)
htensor
αβ = 0, htensor
αi = 0, htensor
ij
2
= r Hij , (700)
203
where Dij = ∇~ i∇
~ j − 1 qij ∆. Here Bαi and Hi are transverse, and Hij is transverse and
3
trace-free. One can identify the following gauge invariants in the bulk [716]:
1
scalars: Yαβ = χαβ − 2D(α Qβ) , Z = A − ∆E − Qα ∂α ln r, (701)
3
vectors: Fαi = Bαi − rDα Hi , (702)
tensors: Hij , (703)
where Dα is the covariant derivative on λαβ , and Qα = r(χα − r∂α E). For simplicity
we will assume that there are no exceptional modes [716] in any sector. Now, the bulk
equations of motion, δGab = 0, are extremely complex in general (see [716]). From a
practical perspective, the only way to proceed is to assume that the background bulk is
maximally symmetric anti de Sitter space, so there is no bulk black hole (i.e. µ = 0)
[918]. The vector and scalar perturbations in the bulk can then be expressed in terms of
a corresponding ‘master variable’. For example, the scalar gauge invariants Yαβ and Z
can be written as
    
1 2 1 1 1 2
Yαβ = Dα Dβ − λαβ D2 − 2 Ω, Z=− Y = D2 − 2 Ω, (704)
r 3 2l 2 6r l
where the scalar master variable Ω satisfies
 
2 α ∆ + 3κ 1
D − 3∂α ln rD + + 2 Ω = 0. (705)
r2 l
For the vectors, we introduce the vector master variable, writing the gauge invariant as
1
Fαi = αβ Dβ Ωi , (706)
r2
where αβ is the total antisymmetric Levi-Civita tensor on λαβ , and Ωi satisfies
  β  
α 5 D Ωi
D r Dβ + (∆ + 2κ)Ωi = 0. (707)
r3
The tensor gauge invariant, Hij , is essentially its own master variable, and satisfies
 
2 α 1
D + 3D ln rDα + 2 (∆ − 2κ) Hij = 0. (708)
r
To solve for these master variables we need to specify boundary conditions at the brane,
as well as the asymptotic boundary conditions. The latter correspond to the condition
of normalisability. Let us now consider the boundary conditions at the brane. The first
thing to note here is that the brane position can fluctuate, so that it now corresponds
to an embedding xa = X̄ a + f a (ξ µ ). We decompose this fluctuation in terms of scalars
~ i ). Secondly, in applying the
f α , , and a transverse vector i , such that f a = (f α , i + ∇
boundary conditions, it is important to express all fields in terms of covariant objects
along the (background) brane. This suggests that all fields of the form W α and W αβ
should be decomposed in terms of their components parallel to the tangent vector X̄˙ α
and along the normal n̄α , as follows:
Wα = −Wk X̄˙ α + W⊥ n̄α , (709)
W αβ = Wkk X̄˙ α X̄˙ β − Wk⊥ X̄˙ α n̄β − W⊥k X̄˙ β n̄α + W⊥⊥ n̄α n̄β . (710)
204
The induced metric on the brane then takes the form

~ i β)dxi dt
ds2 = −(1 + 2Ψ)dt2 − 2a(βi + ∇
h i
~ (i Aj) + hij dxi dxj , (711)
+ a2 (1 − 2Φ)qij + Dij ν + 2∇

with
1 Uk aν̇
Ψ = − Ykk − U̇k + K̄tt U⊥ , β=− − , (712)
2 a 2
1
Φ = −Z − ∆ν + HUk − (D⊥ ln r)U⊥ , ν = 2(E + ), (713)
3
βi = −Fki − aȦi , A i = Hi +  i , (714)
hij = Hij , (715)

where U α = Qα + f α . Note that U α , ν and Ai are invariant under gauge transformations

in the bulk, so these expressions have been written entirely in terms of bulk gauge
invariants.
The gauge invariants on the brane are the ones familiar to us from standard cos-
mological perturbation theory in four dimensions. Given the fluctuation in the brane
energy-momentum tensor

δTtt = −ρδ, δTit = −a(ρ + P )vi , δTji = δP δji + (ρ + P )Σij , (716)

and Equations (688) and (689), we note that the components of the anisotropic stress
(tensor)
Σ, Σ(vector) and Σij are all gauge invariant on the brane, along with the following
scalars,
1
Φ̂ Φ + ∆ν + ȧσg = −Z − (D⊥ ln r)U⊥ ,
= (717)
6
1
Ψ̂ = Ψ − (aσg )˙ = − Ykk + K̄tt U⊥ , (718)
2
θ̂ = θ − σg , (719)
 
P
δ̂ = δ + 3Ha 1 + θ, (720)
ρ
Γ = ρδ − c2s δP, (721)

the vectors θi and β̂i = βi + aȦi = −Fki , and the tensor, hij . Note that

aν̇ Uk
σg = +β =− .
2 a
The goal is now to solve for the master variables, subject to boundary conditions set
by normalisability and the values of the non-dynamical gauge invariants on the brane,
(vector) (tensor)
namely, Γ, Σ, Σi and Σij . We can then use this knowledge to derive the dy-
namical gauge invariants, Φ̂, Ψ̂, β̂i , hij , θ̂, δ̂, and θi . We are already able to express the

205
metric invariants in terms of the master variables (and U⊥ ) as follows:
   
1 1 (∆ + 3κ)
Φ̂ = H Ω̇ − (D⊥ ln r)D⊥ Ω + 2 + Ω (722)
2a l 3a2
−(D⊥ ln r)U⊥ ,
   
1 t

1 2(∆ + 3κ)
Ψ̂ = − Ω̈ − 2H Ω̇ + 2D⊥ ln r − K̄t D⊥ Ω − 2 + Ω (723)
2a l 3a2
+K̄tt U⊥ ,
1
β̂i = − D⊥ Ωi . (724)
a2
To fix U⊥ , and the remaining dynamical gauge invariants, we need to impose boundary
conditions at the brane, given by the linearised Israel junction conditions δ(Kνµ − Kδνµ ) =
−4πG5 δTνµ . For the scalars this gives
  h i
Ω
(∆ + 3κ)D⊥ = 8πG5 a2 ρδ̂ − (∆ + 3κ)[(ρ + P )Σ] , (725)
r
h i
(D⊥ Ω)˙− K̄tt Ω̇ = 8πGa2 (ρ + P )θ̂ − ∂t [a(ρ + P )Σ] , (726)
U⊥ = 4πG5 a2 (ρ + P )Σ, (727)
   2 
−Γ = c2s ρδ̂ + (ρ + P ) Ψ̂ − ∂t aθ̂ + (∆ + 3κ)Σ . (728)
3

We should now view these equations as follows [716]: Equations (725) and (726) fix δ̂
and θ̂ given knowledge of the anisotropic stress, Σ and the master variable, Ω, while
Equation (727) fixes U⊥ . One can then substitute the expression for Ψ̂, δ̂ and θ̂ (in
terms of Ω and Σ) into Equation (728) to derive a Neumann type boundary condition
on Ω. In principle, one should now be able to solve the Master Equation (705), with a
suitable boundary condition derived from Eq. (728), and feed the solution back into the
expressions for Φ̂, Ψ̂, θ̂ and δ̂.
For vectors, the Israel junction conditions yield

(∆ + 2κ)Ωi = 8πG5 a4 (ρ + P )θi , (729)

3 (vector)
Ω̇i = 4πG5 a (ρ + P )Σi . (730)

This time we view Equation (730) as a Dirichlet type boundary condition on the master
variable. We then solve for Ωi using Equation (707), and subsequently fix β̂i and θi using
Equations (724) and (729).
Finally, we consider the tensor. Here we solve Equation (708), subject to the following
junction condition at the brane:
(tensor)
D⊥ Hij = −4πG5 (ρ + P )Σij . (731)

We have now presented the formalism in full, but the task of solving the system
explicitly is another matter altogether. To do so one must choose coordinates for the
background bulk. Whilst the static coordinate system might seem the simplest from
206
the bulk perspective, it is rarely used owing to the fact that a Gaussian Normal (GN)
coordinate system makes it much easier to specify the boundary conditions. For example,
when µ = 0 and κ = 0, the metric in GN coordinates is given by [158]

λαβ dxα dxβ = dz 2 − N 2 (t, z)dt2 , (732)

   
ρ(t)
r(t, z) = a(t) cosh (|z|/l) − 1 + sinh (|z|/l) , (733)
σ

where N = ṙ(t,z)
ȧ(t) . Even in this case analytic solutions can only be obtained in the case
of a background de Sitter brane, when the Master Equation (705) becomes separable
[202, 772]. Numerical solutions, however, have been obtained for scalar perturbations on
a radiation dominated brane [245]. It was found that short wavelength density pertur-
bations are amplified relative to their value GR during horizon reentry, but not so much
that they cause an observable effect in the CMB or in large-scale structure.

5.5. Dvali-Gabadadze-Porrati Gravity

The most celebrated braneworld model exhibiting quasi-localisation and an infra-red
modification of gravity is without doubt the DGP model, developed by Dvali, Gabadadze
and Porrati [454]. The model admits two distinct sectors: The normal branch and the
self-accelerating branch. The latter has generated plenty of interest since it gives rise
to cosmic acceleration without the need for dark energy [394, 398]. However, as we will
see, fluctuations about the self-accelerating vacuum suffer from ghost-like instabilities
[273, 1015, 559, 546]. Although the normal branch is less interesting phenomenologically,
it is fundamentally more healthy and is the closest thing we have to a consistent non-linear
completion of massive gravity. Here the graviton is a resonance of finite width, 1/rc , as
opposed to a massive field. At short distances, r  rc , the brane dynamics do not feel
the width of the resonance, and the theory resembles 4D GR. At large distances, r  rc ,
however, the theory becomes five dimensional as the resonance effectively decays into
continuum Kaluza-Klein modes. It is claimed that the Vainshtein mechanism works well
on the normal branch of DGP, screening the longitudinal graviton without introducing
any new pathological modes, in contrast to massive gravity [405, 399]. The breakdown
of classical perturbation theory at the Vainshtein scale can be linked to quantum fluctu-
ations on the vacuum becoming strongly coupled at around 1000 km [839, 1073, 447].

5.5.1. Action, equations of motion, and vacua

The DGP model contains a single 3-brane embedded in an otherwise empty five
dimensional bulk space-time. Generalisations that include a bulk cosmological constant
and/or bulk branes have been studied [697, 1011]. However, the original DGP action is
given by
Z Z  
3 5 √ 4 √ 3 M42
S = M5 d x −γR + d x −g −2M5 K + R − σ + Lmatter , (734)
M ∂M 2

where, by Z2 symmetry across the brane, we identify the entire bulk space-time with two
identical copies of M, and the brane with the common boundary, ∂M. The bulk metric
is given by γab , with corresponding Ricci scalar, R, and M5 is the Planck scale in the

207
bulk. The induced metric on the brane is given by gµν , and K = g µν Kµν is the trace of
extrinsic curvature, Kµν = 12 Ln gµν . Here we define the unit normal to point into M.
The brane has a bare vacuum energy, or tension, σ, and additional matter contri-
butions contained in Lmatter . However, the key feature in the brane Lagrangian is the
induced curvature term, given by M42 R. Such a term can be generated by matter loop
corrections [161], or finite width effects [263]. The mass scale, M4 , is taken to be Planck-
ian, ∼ 1018 GeV. There is a hierarchy between this scale and the bulk Planck scale,
M4  M5 , which has proven difficult to derive from fundamental theory. Nevertheless,
the hierarchy enables us to identify a crossover scale, rc ∼ M42 /2M53 , below which the
theory looks four dimensional, and above which it looks five dimensional.
The bulk equations of motion are given by the vacuum Einstein equations,
1
Gab = Rab − Rγab = 0, (735)
2
and the boundary conditions at the brane are given by the Israel junction conditions,

2M52 (Kµν − Kgµν ) = M42 Gµν + σgµν − Tµν , (736)

R √
where Tµν = − √2−g δgδµν ∂M d4 x −gLmatter is the energy-momentum tensor for the
additional matter.
Let us now consider the vacua of this theory, corresponding to maximally symmetric
brane solutions with Tµν = 0. For a given tension, there exist two distinct vacua [394,
398]. Assuming σ > 0 for definiteness, one finds that these correspond to de Sitter branes
with intrinsic curvature, " s #
M53 M42 σ
H= + 1+ , (737)
M42 6M53
where  = ±1. In conformal coordinates, the full solution can be written as

ds2 = γ̄ab dxa dxb = e2Hy (dy 2 + ḡµν dxµ dxν ), (738)

where ḡµν dxµ dxν = −dt2 + e2Ht dx2 is the 4D de Sitter line-element written in Poincaré
coordinates. The domain M corresponds to 0 < y < ∞, while the brane is located on
the boundary at y = 0.
The two branches of this solution are often referred to as the normal branch ( = −1),
and the self-accelerating branch ( = 1). The latter is so-called because as we take the
limit of vanishing vacuum energy, σ → 0, the metric on the brane is still asymptotically de
Sitter. The limiting de Sitter curvature is given by the cross-over scale, H = 2M53 /M42 ∼
1/rc . If this were to account for dark energy today, the crossover from four to five
dimensions would have to occur at the horizon scale, placing the fundamental Planck
scale at M5 ∼ 10 MeV or so. However, as we will discuss in Section 5.5.3, this branch
of solution is unstable as it contains physical ghost excitations. The normal branch
is asymptotically Minkowski in the limit σ → 0, and does not suffer from a ghostly
pathology. On this branch it is phenomenologically more interesting to have the cross-
over scale at distances below the horizon scale.
The distinction between branches is best understood by considering their embedding
in the 5D Minkowski bulk. In each case, the brane can be viewed as a 4D hyperboloid
208
Figure 11: Embedding of a de Sitter brane in a flat 5D bulk, taken from [273, 559]. The braneworld
volume is the hyperboloid in the Minkowski bulk. The normal branch corresponds to keeping the interior
of the hyperboloid, while the self-accelerating branch corresponds to the exterior.

of radius 1/H (see Figure 11). The choice of sign,  = ±1, corresponds to the choice
in whether one retains the interior ( = −1) or the exterior ( = 1) of the hyperboloid.
Note that retaining the exterior ensures that the brane behaves like a domain wall with
negative effective tension, even when σ ≥ 0 [559]. This is an early warning sign that the
self-accelerating branch could be pathological.

5.5.2. Linear perturbations on the normal branch

We now consider linear perturbations on the normal branch. For simplicity and
brevity, let us consider the limiting case of vanishing vacuum energy, σ = 0, corresponding
to fluctuations about a Minkowski brane in a 5D Minkowski bulk, γab = ηab + hab . It is
convenient to choose harmonic gauge,
1
∂a hab = ∂b haa , (739)
2
so that the linearised bulk equations of motion take the simple form

[∂y2 + ∂ 2 ]hab = 0, (740)

where ∂ 2 = ∂µ ∂ µ . Taking Fourier transforms with respect to the brane coordinates, so

that Z
1 µ
Q(x, y) → Q̃(p, y) = 2
d4 xeipµ x Q(x, y),
(2π)
it can be seen that the normalisable solution is given by

h̃ab (p, y) = e−py h̃ab (p, 0). (741)

209
We have not yet fully fixed the gauge as there is still a residual symmetry corresponding to
gauge transformations of the form xa → xa + ξ a , where (∂y2 + ∂ 2 )ξ a = 0, or, equivalently,
ξ˜a (p, y) = e−py ξ˜a (p, 0). We now use ξ y to fix the brane position to lie at y = 0, and ξ µ
to set hµy = 0. It follows from the y component of Equation (739) that hyy = h, where
h = η µν hµν . The ν component of Eq. (739) then gives ∂ µ hµν = ∂ν h.
The junction conditions at the brane,

2M53 δ(Kµν − Kgµν ) = M42 δGµν − Tµν ,

now yield the following solution for the metric on the brane:
 
2 1
h̃µν (p, 0) = 2 2 T̃ µν − T̃ η µν + pure gauge pieces. (742)
M4 p + 2M53 p 3

At low energies, p  1/rc = 2M53 /M42 , this becomes

 
1 1
h̃µν (p, 0) ∼ 3 T̃µν − T̃ ηµν , (743)
M5 p 3

which suggests that the potential for a point mass will scale like V (r) ∝ 1/r2 at large
distances, r  rc . This is consistent with 5D gravity, and is to be expected for an
infinitely large 5th dimension. In contrast, at high energies, p  1/rc , we have
 
2 1
h̃µν (p, 0) ∼ 2 2 T̃µν − T̃ ηµν . (744)
M4 p 3

This can be compared with the propagator in four-dimensional General Relativity, h̃GR
µν =
h i
2 1
M 2 p2
T̃µν − 2 T̃ ηµν . Thus we almost have agreement. Indeed, since the scaling with
pl
momenta is the same, so the potential for a point mass will go like V (r) ∝ 1/r for r  rc .
We can understand this as follows: The graviton mediating the interaction between two
point particles along the brane is bound close to the brane by the induced curvature
there. It is only at very large distances, when the induced curvature is insignificant, that
the graviton is able to probe the extra dimension.
However, at this stage it would be premature to say that we have localisation of
gravity at short distances. The problem lies with the tensor structure of the propagator.
As we can see, the GR result differs from DGP in that it has a factor of 21 as opposed
to 13 . This is due to the longitudinal mode of the graviton propagating in DGP but not
in GR, and will inevitably lead to disagreement with observation. To see this, consider
0
the amplitude for the interaction between two sources, Tµν and Tµν , which is given by
0µν
A = hµν T . In GR, we have
 
2 0µν 1 0
ÃGR = 2 2 T̃µν (p)T̃ (−p) − T̃ (p)T̃ (−p) , (745)
Mpl p 2

whereas in DGP, at high energies, we have

 
2 0µν 1 0
ÃDGP ∼ 2 2 T̃µν (p)T̃ (−p) − T̃ (p)T̃ (−p) . (746)
M4 p 3
210
To get the same potential for the interaction of two non-relativistic sources, we require
that50 M42 = 43 Mpl2
. Now consider what happens when one of the sources is relativistic,
say Tµν . We then have T 0 = 0, and so the amplitudes differ by a factor of 34 . So, although
0

we can recover the standard Newtonian potential, the prediction for light bending around
the Sun differs from GR by a factor of 34 .
This is reminiscent of the vDVZ discontinuity in massive gravity [1243, 1296]. In
fact, it is sometimes said that DGP gravity (on the normal branch) is the closest thing
we have to a non-linear completion of massive gravity. Strictly speaking, the behaviour
is more like a massive resonance of width 1/rc , with decay into a continuum of massive
modes occurring at r > rc . However, the similarities with massive gravity do suggest a
possible resolution to this issue of tensor structure, and the resulting phenomenological
problems. In massive gravity, it has been argued that the linearised analysis around a
heavy source breaks down at the so-called Vainshtein radius, which can be much larger
than the Schwarzschild radius [1241]. Below the Vainshtein radius, it is claimed that
non-linear effects help to screen the longitudinal scalar and match the theory to GR
without a vDVZ discontinuity. Similar claims have been made in DGP gravity [399], as
we will discuss in Section 5.5.4.

5.5.3. Linear perturbations (and ghosts) on the self-accelerating branch

Let us consider linearised perturbations and ghosts about the self-accelerating vacuum
described in Section 5.5.1 (see also [273, 559, 1015, 546, 734]). To reveal the ghost in
the cleanest way possible, it is convenient to retain non-zero tension on the background.
It is also enough to consider vacuum fluctuations, so we will set Tµν = 0 for brevity (a
more complete discussion including matter fluctuations can be found in [273]).
Recall that the self-accelerating background solution is given by the geometry in Eq.
(738), with  = +1 and the brane positioned at y = 0. A generic perturbation can be
described by γab = γ̄ab + δγab , with the brane position shifted to y = F (x). We will work
in Gaussian Normal (GN) coordinates, so that
δγyy = δγµy = 0, and δγµν = eHy/2 hµν (x, y). (747)
The tensor hµν can be decomposed in terms of the irreducible representations of the 4D
de Sitter diffeomorphism group as
1 h
hµν = hTµνT + ∇µ Aν + ∇ν Aµ + ∇µ ∇ν φ − ḡµν φ + ḡµν , (748)
4 4
where ∇ is the covariant derivative associated with the 4D de Sitter metric, ḡµν . The
tensor hTµνT is transverse and trace-free, ∇µ hT T µν = ḡ µν hTµνT = 0, and Aµ is transverse,
∇µ Aµ = 0. In addition, we have two scalars, φ and h = ḡ µν hµν .
Following a similar approach to the one outlined for Randall-Sundrum gravity in
Section 5.3.5, we can now fix the position of the brane to be at y = 0 while remaining in
GN gauge by making the following gauge transformation:
e−Hy µ
y → y − F e−Hy , and xµ → xµ − ∇ F. (749)
H
50 To verify this result, simply insert the appropriate energy-momentum tensors for non-relativistic

sources, Tµν = diag(ρ, 0, 0, 0), and Tµν0 = diag(ρ0 , 0, 0, 0), in each expression for the amplitude, and

match the two results.

211
Although the brane position is now fixed at y = 0, the original brane position F (x)
(F )
still enters the dynamics through a book-keeping term, hµν , that modifies the metric
perturbation as
1 h
hµν → hTµνT + ∇µ Aν + ∇ν Aµ + ∇µ ∇ν φ − ḡµν φ + ḡµν + h(F )
µν . (750)
4 4
The book-keeping term is pure gauge in the bulk, and is given by
2 Hy/2

h(F )
µν = e ∇µ ∇ν + H 2 ḡµν F. (751)
H
We can now substitute our modified expression for hµν into the linearised fields equations
in the bulk, δGab = 0, and on the brane, 2M53 δ(Kµν − Kgµν ) = M42 δGµν . It turns out
that the transverse vector, Aµ , is a free field in the linearised theory and can be set to
zero. In addition, the yy and yµ equations in the bulk imply that one can consistently
choose a gauge for which
h = 0, and ( + 4H 2 )φ = 0. (752)
(φ) (F )
Note that we now have hµν = hTµνT + hµν + hµν , where the contribution from φ(x, y) has
been rewritten as

h(φ) 2
µν = ∇µ ∇ν + H ḡµν φ(x, y). (753)
This mode is now entirely transverse and trace-free in its own right. In the absence of
any additional matter on the brane, Tµν = 0, the same is true of the book-keeping mode,
(F )
hµν . This is because the trace of the Israel equation now implies that
( + 4H 2 )F = 0. (754)
The entire perturbation hµν (x, y) is therefore now completely transverse and trace-free.
This greatly simplifies the bulk and brane equations of motion, giving
 
9H 2
 − 2H 2 + ∂y2 − hµν (x, y) = 0 for y > 0, (755)
4
  
3H
M42 ( − 2H 2 ) + 2M53 ∂y − hµν = 0 at y = 0+ . (756)
2
Variables in the tensor and scalar fields can now be separated as follows:
X
hTµνT (x, y) = um (y)χ(m)
µν (x), φ(x, y) = W (y)φ̂(x), (757)
m

(m) (m) (m)

where χµν is 4D tensor field of mass m satisfying ( − 2H 2 )χµν = m2 χµν . Note that
φ̂ is a 4D tachyonic field satisfying ( + 4H 2 )φ̂ = 0. This is a mild instability, related to
the repulsive nature of inflating domain walls.
Assuming that the tensor and scalar equations of motion can be treated indepen-
dently, we find that there is a continuum of normalisable tensor modes with mass
m2 ≥ 9H 2 /4. In addition, there is also a discrete tensor mode with mass
 
2 1 1
md = 3H − , (758)
rc rc
212
 q
9H 2 2
and normalisable wave-function umd (y) = αmd e−y 4 −md . Now, for positive brane
tension, σ > 0, one can easily check that 0 < m2d < 2H 2 . For massive gravitons
propagating in 4D de Sitter, it is well known that masses lying in this range result
in the graviton containing a helicity-0 ghost [605]. The lightest tensor mode in this case
is therefore perturbatively unstable. For negative brane tension we find m2d > 2H 2 , and
there is no helicity-0 ghost in the lightest tensor.
(φ)
Now consider the scalar equations of motion. The first thing to note is that hµν
(φ)
behaves like a transverse trace-free mode with mass m2φ = 2H 2 , because (−2H 2 )hµν =
(φ) (φ)
2H 2 hµν . Since none of the tensor modes have this mass, they are all orthogonal to hµν .
This means it was consistent to assume that the scalar and tensor equations of motion
could indeed be treated independently. It turns out that the scalar has a normalisable
wave-function W (y) = e−Hy/2 , and the 4D scalar φ̂ is sourced by F by the relation
 
2H − r1c
φ̂(x) = αF (x), α = −   . (759)
H H − r1c

(φ)
This is well defined for σ 6= 0, as in this case H 6= 1/rc . hµν (x, y) may now be thought
of as a genuine radion mode, measuring the physical motion of the brane with respect to
infinity. It does not decouple even though we only have a single brane. This property is
related to the fact that the background warp factor, e2Hy , grows as we move deeper into
the bulk.
We have now identified the helicity-0 mode of the lightest tensor as a ghost when
σ > 0. When σ < 0, a calculation of the 4D effective action reveals the ghost to be
the radion (see [273, 1015]). Given that there is always a ghost for non-zero tension, by
continuity one may expect this to remain the case when σ = 0.

The limit of vanishing tension

To study this limit more closely, let us first ask whether we can trust the above
solutions in the limit where σ → 0. In this limit H → 1/rc , and the quantity α becomes
ill-defined. To understand what has gone wrong, note that the mass of the lightest tensor
(φ)
has the limit m2d → 2H 2 . This means that it is no longer orthogonal to the radion, hµν ,
and so we cannot treat the tensor and scalar equations of motion independently. This
behaviour can be traced back to an additional symmetry that appears in the linearised
theory in the limit of vanishing brane tension. It is analogous to the “partially massless
limit” in the theory of a massive graviton propagating in de Sitter space [605]. In that
theory, the equations of motion are invariant under the following redefinition of the
graviton field: √ √
( 2H)
χµν (x) → χ(µν 2H) (x) + (∇µ ∇ν + H 2 ḡµν )ψ(x). (760)
This
√
field redefinition has the effect of extracting out part of the helicity-0 mode from
( 2H)
χµν , and as a result of the symmetry this mode disappears from the spectrum. In
DGP gravity this shift must be accompanied by a shift in the scalar field, φ,
q
9H 2
−m2d
φ(x, y) → φ(x, y) − α√2H e−Hy/2 ψ(x) = φ(x, y) − lim αmd e−y 4 ψ(x), (761)
σ→0
213
in order to render the overall perturbation, hµν (x, y), invariant.
√
These ψ shifts can be
( 2H)
understood as extracting part of the helicity-0 mode from χµν , and absorbing it into a
renormalisation of φ. The symmetry will have the effect of combining the helicity-0 mode
and the radion into a single degree of freedom. It is only after fixing this ψ symmetry that
we can treat the scalar and tensor equations of motion independently of each another.
We could also consider extracting the entire helicity-0 mode and absorbing it into φ, or
vice versa.
Actually, there exists a clever choice of gauge that enables us to take a smooth limit
as σ → 0 [546]. One can then readily calculate the 4D effective action [273], and derive
the corresponding Hamiltonian [546]. It turns out that the Hamiltonian is unbounded
from below, signalling a ghost-like instability [546]. This ghost is a combination of the
radion and helicity-0 mode, and represents the residual scalar degree of freedom left over
after fixing the aforementioned ψ symmetry.

Ghosts

We have shown that for any value of the brane tension, perturbations about the self-
accelerating branch of DGP contain a ghost. As explained in Section 2.1.3, the ghost will
generate catastrophic instabilities as it couples universally with gravitational strength to
the Standard Model fields and the remaining gravity modes. The existence of the ghost
can be trusted as long as we can trust our effective description. It had been argued
that this breaks down at the Hubble energy scale, casting doubt on the ghost’s existence
at sub-horizon distances [401]. However the analysis of [741] suggests that the cut-off
for the effective theory is actually at a much higher energy, being the same as that on
the normal branch. As we will see in the next section, for fluctuations on the vacuum
this corresponds to Λcut-off ∼ 10−13 eV. As this is well above the characteristic scale of
the self-accelerating vacuum, H0 ∼ 10−33 eV, the ghost will cause this vacuum to be
infinitesimally short-lived due to a divergent rate of particle creation51 .
It is also worth pointing out that the ghost can manifest itself beyond perturbation
theory. In the presence of a self-accelerating brane one can accommodate a Schwarzschild
bulk with negative mass, without introducing any naked singularities. This demonstrates
that the five-dimensional energy is unbounded from below, as suggested by the perturba-
tive analysis. Furthermore, standard Euclidean techniques indicate that the spontaneous
nucleation of self-accelerating branes is unsuppressed. This is problematic even on the
normal branch, and suggests that self-accelerating branes should be projected out the
from the theory altogether [559].

5.5.4. From strong coupling to the Vainshtein mechanism

We now return to the normal branch of DGP gravity. In Section 5.5.2, we saw some
similarities between this theory and massive gravity. Indeed, it happens that the brane
to brane graviton propagator is given by
DGP massive
Dµναβ (p) = Dµναβ (p, p/rc ), (762)

51 Note that it has been argued that this choice of vacuum state explicitly breaks de Sitter invariance

[653], and so one is free to impose a Lorentz non-invariant cut-off in the 3-momentum such that the
creation rate is no longer divergent.
214
 massive
where Dµναβ (p, m2 ) is the propagator for 4D massive gravity. Thus, just as massive
gravity suffers from the vDVZ discontinuity [1243, 1296] as m → 0, so does DGP gravity
(normal branch) in the limit p  1/rc . This means the linearised theory is not reduced to
GR at short distances, but to a four-dimensional scalar-tensor theory. If this description
can be trusted at solar system scales, then it leads to wildly inaccurate predictions for
the bending of light around the Sun.
In massive gravity, it turns out that the linearised theory cannot necessarily be
trusted at solar system scales. This is because it breaks down at the Vainshtein ra-
dius, rVmassive ∼ (rs /m4 )1/5 , where rs ∼ 3 km is the Schwarzschild radius of the Sun. For
quantum fluctuations in vacuo, the breakdown of classical perturbation theory at rVmassive
translates into a strong coupling scale Λcut-off ∼ (Mpl m4 )1/5 [64]. Note that the equiv-
alent scales in four-dimensional General Relativity are simply the Schwarzschild radius
and the Planck scale.

Strong coupling in DGP gravity

The situation in DGP gravity (normal branch) is similar to the case of massive gravity,
as we will now show. While aspects of the strong coupling problem were first identified
in [696, 1073], the derivation of the strong coupling scale for quantum fluctuations on
the Minkowski brane is most elegantly presented in [839]. Here we review the results
of this study52 (see also [1017]). Their analysis involves a computation of the boundary
effective action53 , which is found to be
Z " √ ! √ !
4 M42 1 µν 2 −∂ 2 1 2 −∂ 2
Γ[hµν , Nµ , hyy ] = d x h ∂ − hµν − h ∂ − h
4 2 rc 4 rc
p  
1 1 1 p 2 1 1 p
− Nµ −∂ 2 + Nµ + h −∂ hyy − N µ ∂µ hyy − hyy −∂ 2 hyy
rc rc 2rc rc 4rc
Z
1
+ d4 x hµν T µν + Γint [hµν , Nµ , hyy ], (764)
2

where Nµ = hµy and ∂ 2 = ∂µ ∂ µ . The function Γint contains all the higher-order inter-
action terms to be discussed shortly. For now let us focus on the quadratic term. This
can be diagonalised by means of the following field redefinition
p 
2
1
hµν = h̃µν + πηµν , Nµ = ѵ + rc ∂µ π, hyy = −2rc −∂ + π. (765)
rc

52 Correcting a few typos along the way.

53 The boundary effective action, Γ[φ], is obtained by integrating the full action, SM [Φ] + S∂M [φ],
over bulk fields, Φ, satisfying the boundary condition Φ|∂M = φ, i.e.
Z
eiΓ[φ] = d[Φ]ei(SM [Φ]+S∂M [φ]) . (763)
Φ|∂M =φ

215
The effective action can then be written as
Z " √ ! √ !
4 M42 1 µν 2 −∂ 2 1 2 −∂ 2
Γ[h̃µν , ѵ , π] = d x h̃ ∂ − h̃µν − h̃ ∂ − h̃
4 2 rc 4 rc
  √ ! #
1 µ p 2 1 2 −∂ 2
− Ñ −∂ + ѵ + 3π ∂ − π
rc rc rc
Z
1
+ d4 x (h̃µν T µν + πT ) + Γint [h̃µν , ѵ , π]. (766)
2
The next step is to write everything in terms of the canonically normalised fields
r
M4 M4 3
ĥµν = h̃µν , N̂µ = √ ѵ , π̂ = M4 π. (767)
2 2rc 2
This gives
Z √ ! √ !
4 1 µν 2 −∂ 2 1 2 −∂ 2
Γ[ĥµν , N̂µ , π̂] = d x ĥ ∂ − ĥµν − ĥ ∂ − ĥ
2 rc 4 rc
  √ !
1 µ p 2 1 1 2 −∂ 2
− N̂ −∂ + N̂µ + π̂ ∂ − π̂
2 rc 2 rc
Z  
1 1
+ d4 x ĥµν T µν + √ π̂T + Γint [ĥµν , N̂µ , π̂]. (768)
M4 6M4
Turning our attention to the interaction piece, we note that it contains terms that
schematically take the form
Z p Z p 2
d4 xM53 −∂ 2 (hµν )a (Nµ )b (hyy )c , d4 xM42 −∂ 2 (hµν )d , (769)

where a + b + c ≥ 3, and d ≥ 3. By writing these in terms of the canonically normalised

fields one can easily check √
that the largest interaction comes from the term with a = 0
and b + c = 3. Indeed, for −∂ 2  1/rc , the interaction term goes like [839, 1017]
Z
1
Γint = − √ d4 x(∂ π̂)2 ∂ 2 π̂ + sub-leading interactions, (770)
3 6Λ3
where
Λ = (M4 /rc2 )1/3 . (771)
This corresponds to the scale at which quantum fluctuations in vacuum become strongly
coupled. For M4 ∼ Mpl , and rc ∼ 1/H0 , we have Λ ∼ 10−13 eV ∼ 1/(1000 km). In
other words, for scattering processes above 10−13 eV perturbative quantum field theory
in a vacuum is no longer well defined. Of course, it is important to realise that we do
not actually live in a vacuum, and it has been demonstrated that the strong coupling
scale can be raised on curved backgrounds [962]. At the surface of the earth, the new
scale corresponds to about 10−5 eV, indicating the presence of large quantum correc-
tions to the π field at distances below a centimetre. This will not impact much on table
216
top experiments, however, as the π field is expected to be overwhelmed by the classical
graviton at this scale, due to Vainshtein screening.

The π-Lagrangian

To see the emergence of the new strong coupling scale most succinctly, it is convenient
to take the limit in which the troublesome π field decouples from the graviton. This will
also help us in our discussion of the breakdown of classical perturbation theory around
heavy sources and the Vainshtein effect. In any event, we take the so-called decoupling
limit [962]
Tµν
M4 , rc , Tµν → ∞, and Λ, = fixed, (772)
M4
so that all the sub-leading interactions in Eq. (770) go to zero. The limiting theory
should be valid at intermediate scales,
max (L4 , rs )  r  rc ,
where L4 ∼ 1/M4 is the Planck length, and rs is the Schwarzschild radius of the source
(if present). The action goes like
Z
Γ ∼ d4 x [LGR + Lπ ] ,

where
 
M42 1 µν 2 1 2 1
LGR = h̃ ∂ h̃µν − h̃∂ h̃ + h̃µν T µν , (773)
4 2 4 2
M42   1
Lπ = 3π∂ 2 π − rc2 (∂π)2 ∂ 2 π + πT , (774)
4 2
and we have set the free field Nµ = 0. Note that LGR is just the standard Einstein-Hilbert
Lagrangian, expanded to quadratic order about Minkowski space. The modification to
GR is encoded in Lπ . This is often referred to as the π-Lagrangian and much of the
interesting phenomenology of DGP gravity, from strong coupling and Vainshtein effects
on the normal branch to ghosts on the self-accelerating branch, can be studied using this
Lagrangian.
In vacuum (Tµν = 0) the π-Lagrangian possesses a symmetry π → π +aµ xµ +b, where
aµ and b are constants. This is sometimes referred to as Galilean invariance and is the
inspiration for galileon models [963]. The most general Galilean invariant Lagrangians
and will be studied in detail in Section 4.4.
We now return to the question of the strong coupling scale in the theory described
by Eq. (774) in the presence of a classical source. The first thing to note is that our
classical background is no longer the vacuum, but the solution π = πcl (x) of the classical
field equations
  T
3∂ 2 π − rc2 (∂µ ∂ν π)(∂ µ ∂ ν π) − (∂ 2 π)2 = − 2 . (775)
M4
Our interest now lies in the quantum fluctuations, π = πcl (x) + ϕ(x), about this solution.
Plugging this into the Lagrangian given in Eq. (774) we find
M42   1
Lπ = −Z µν (x)∂µ ϕ∂ν ϕ − rc2 (∂ϕ)2 ∂ 2 ϕ + ϕδT , (776)
4 2
217
where

Z µν (x) = 3η µν − 2rc2 ∂ µ ∂ ν − η µν ∂ 2 πcl . (777)
µν
For non-trivial solutions πcl (x) is not constant, and so neither is Z (x). However,
assuming the background varies slowly (relative to the fluctuations) we can treat Z µν as
approximately constant in a neighbourhood of a point in space-time. Further assuming
that the eigen-values of Zνµ (x) are all of similar magnitude, ∼ Z(x), we can identify a
localised strong coupling scale,
p
Λ∗ (x) ∼ Λ Z(x)  Λ. (778)
As we will see shortly, for spherically symmetric solutions inside the Vainshtein radius,
√
rV ∼ (rs rc2 )1/3 , we have πcl ∼ rs r/rc , and so Z(r) ∼ (rV /r)3/2 . On the sphere of
radius r the local strong-coupling scale is
 r 3/4
V
Λ∗ (r) ∼ Λ .
r
It follows that the classical background, πcl , ceases to make sense below a critical length,
rmin , where rmin Λ∗ (rmin ) ∼ 1. This is because scattering processes would need to exceed
the local strong coupling scale to probe the structure of the background at r < rmin .
This short distance cut-off will typically lie well within the Schwarzschild radius of the
source, and therefore outside of the regime of validity of the π-Lagrangian.
We should also consider the implication of the local strong coupling scale for energetic
processes taking place on the surface of the
p Earth. Here the Earth’s gravitational field has
the dominant effect, so we have Λ∗ ∼ ( rc /rH )10−5 eV, where rH ∼ 1/H0 ∼ 1026 m is
the current Hubble radius. For rc ∼ rH this means the fluctuations in the π field becomes
strongly coupled at around 1 cm. However it is important to realise that it is only the
ϕ self-interactions that grow large at this scale. The direct coupling to the graviton and
the coupling to matter are negligible, with the latter going like G4 /Z(r) ∼ 10−15 G4 .
Although the scalar fluctuations enter a quantum fog at the centimetre scale they do not
spoil our classical description as this is dominated by the graviton, which is fifteen orders
of magnitude larger [962].
Finally, let us note another important feature of the scalar dynamics: As Poincaré
invariance is broken by the classical background, πcl , the fluctuations no longer have to
propagate on the light cone. Indeed, one can explicitly show that angular fluctuations
about spherically solutions will be super-luminal, causing problems for causality [9, 609].

The Vainshtein effect in DGP gravity

Let us now consider the Vainshtein effect directly. As we will see, this is a mechanism
in which an additional scalar mode is screened at short distances by non-linear interac-
tions, thereby eliminating the troublesome vDVZ discontinuity. To see how the scalar is
screened in DGP gravity, we consider the classical solution to the field equations around
a heavy non-relativistic source, Tµν = diag(ρ(r), 0, 0, 0), assuming spherical symmetry
for simplicity. If we are to screen the scalar and recover GR at a given scale, the classical
(cl)
graviton solution, h̃µν , should dominate over the corresponding scalar solution, πcl . In
other words, we should schematically have
|h̃(cl)
µν |  |πcl |.
218
 (cl)
Since h̃µν is derived from Eq. (773), it is just the standard linearised GR solution given
by the Newtonian potential,
rs
|h(cl)
µν | ∼ , (779)
r
R
where rs = 2G4 M is the Schwarzschild radius of the source, M = ρ(r)dV is its mass,
and G4 = 1/8πM42 is the four dimensional Newton’s constant.
In general, the dynamics of the scalar are governed by the Field Equations (775). It
follows that the spherically symmetric solution, πcl (r), around our heavy source satisfies

0 2rc2 02 rs
πcl + π = 2, (780)
3r cl 3r
where we have integrated once over a sphere centred at the origin and enclosing the entire
source. Now at large distances we expect the linear term to dominate, giving a solution
(lin) rs
πcl (r) = − . (781)
3r
This scales in the same way as the corresponding graviton solution in Eq. (779), and so
we have an O(1) modification of General Relativity. However, the linearised description
breaks down once the non-linear piece becomes comparable, or, equivalently, when
   
 (lin) 0   rc2 (lin) 02 
(πcl )  ∼  (πcl )  . (782)
r

Substituting our solution from Eq. (781), we see that the linearised theory breaks down
at the Vainshtein radius,
rV ∼ (rs rc2 )1/3 . (783)
For the Sun we have rs ∼ 3 km, and so
 2/3
rc
rV ∼ 1018 m. (784)
rH

As long as the cross-over scale is not too far inside the cosmological horizon, the Vain-
shtein effect extends beyond the edges of the solar system, given by the Oort cloud at
an average of 1016 m or so. For the earth we have rs⊕ ∼ 9 mm, and so
 2/3
rc
rV⊕ ∼ 1017 m. (785)
rH

In each case, the breakdown of classical perturbation theory has the same origin as the
strong coupling of quantum fluctuations in vacuum, namely the self-interaction of the
scalar, π, with scale-dependent coupling. Note that the Vainshtein scale can always be ob-
tained from the strong coupling scale by trading the Planck length for the Schwarzschild
radius of the source.
Now let us ask whether or not the scalar gets screened below the Vainshtein radius.
For r  rV , the linearised solution in Eq. (781) is no longer a good approximation.

219
On the contrary, the non-linear part of Equation (780) will dominate, so that at short
distances we have the solution
√
(nonlin) 2rs r
πcl (r) = . (786)
rc
This implies that at short distances the scalar correction to the Schwarzschild solution
goes like [567]
(nonlin)  3/2
δV |π | r
∼ cl (cl) ∼ . (787)
V |hµν | rV

Given that rV → ∞ as rc → ∞, we see that the scalar does appear to get screened in
this limit, and that one is able to recover GR without any vDVZ discontinuity.
We can now use the correction to the Newtonian force to place bounds on the cross-
over scale. At r ∼ 5 AU fractional corrections to the Sun’s gravitational field should
be . 10−8 [1197], which implies rc & 10−4 rH . Similarly, the corrections to the Earth’s
gravitational field may have an observable effect on the precession of the moon, where
corrections to the potential go like δV /V |moon ∼ 10−13 rH /rc [451].
While we have demonstrated aspects of the Vainshtein mechanism at the level of
the π-Lagrangian, for simplicity, it is worth noting that it was originally discussed using
the full theory [399, 567]. In [399] it was emphasised how one must choose the correct
expansion parameter at a given scale. Standard perturbation theory corresponds to
performing an expansion in rs /r, but this is not a good expansion parameter in the limit
rc → ∞, as then the next-to-leading order terms become singular. At short distances,
the claim is that one should expand in powers of r/rV around the Schwarzschild solution
[567], so the result can be smoothly patched onto the standard perturbative solution in
some neighbourhood of the Vainshtein radius [399].
It is fair to say that the Vainshtein mechanism has not yet been fully explored as a
viable concept. While it is clear that there is a breakdown of linear perturbation theory
at the Vainshtein radius in both massive gravity and DGP, the arguments for a smooth
transition to GR are at best promising, but hardly conclusive. N-body simulations
of large-scale structure in DGP gravity, however, support the case for its successful
implementation [1121, 269]. We will discuss this further in the next section.
One should also worry about the elephant problem54 : An elephant is an extended
object made up of many point particles each with their own Vainshtein radius. Is the
Vainshtein radius of the elephant the same as the Vainshtein radius of a point particle
with the same mass located at the centre of mass of the elephant? While this may not
be a problem when the size of the source particle distribution is much smaller than the
Vainshtein radius, it clearly becomes an issue when the particles are well distributed over
the whole Vainshtein sphere. How does one account for such many particle systems? The
answer is not known and given the role of non-linearities in the Vainshtein mechanism
the problem may well be very complicated, especially in the full covariant theory. See
[672] for a recent discussion.

54 We thank Nemanja Kaloper for this colourful observation.

220
5.5.5. DGP cosmology
In this section we describe DGP cosmology, from the FLRW background, to linear
perturbations and non-linear studies. A thorough review of DGP cosmology up to 2006
can be found in [834].
As discussed in Section 5.4.1, Shiromizu, Maeda and Sasaki [1140] found the 4D Ein-
stein equations of a 3-brane world embedded in a 5D bulk with Z2 symmetry. Applying
this formalism to DGP we get

Gµν = (16πGrc )2 Πµν − Eµν , (788)

where
1 1 1 h αβ i
Πµν = − T̃µλ T̃ λν + T̃ T̃µν + 3T̃ T̃αβ − T̃ 2 gµν , (789)
4 12 24
and
1
T̃µν = Tµν − Gµν . (790)
8πG
The Bianchi identities give

∇ν E νµ = (16πGrc )2 ∇ν Πν µ , (791)

while the matter stress-energy tensor satisfies local energy-momentum conservation:

∇ν T νµ = 0.
The cosmological solutions for an FLRW space-time were found by Deffayet [394] by
considering the embedding of the DGP brane into a Minkowski bulk. This corresponds
to setting Eµν = 0 and plugging an FLRW metric into our field equations (788). Note
that the global structure of the cosmological solutions has been investigated by Lue [833],
while some exact solutions in special cases have been found by Dick [420]. If one were
to generalise the bulk to include 5D Schwarzschild, one can run into an unusual class of
singularity on account of the branch cut in Friedmann equation [1142, 559]. Cosmological
branes in generalised DGP gravity can also mimic a phantom equation of state [1086, 835].
Staying with the simplest case of a Minkowski bulk and a tensionless brane, the 00
component of the 4D field equations (with Eµν = 0) gives the Friedmann equation as
r
κ  κ 8πG
H2 + 2 − H2 + 2 = ρ, (792)
a rc a 3
where  = −1 for the normal branch, and  = 1 for the self-accelerating branch. Clearly, if
ρ = 0 then the normal branch implies Minkowski space, while the self-accelerating branch
gives de Sitter space with Hubble constant H∞ = r1c . This self-accelerating model has
been proposed as an alternative to dark energy [398], but as we have already seen, the
theory contains a ghost in its perturbative spectrum. A phenomenological extension of
Eq. (792) has been considered by Dvali and Turner [453], where for κ = 0 and  = 1 the
2α
term rH2−α is added to Eq. (792).
c
After using Eq. (792) the ij component can be re-worked into a modified Raychaud-
huri equation, as
p
dH 2 κ 3H 2 + 3κ
a2 − 2rc H 2 + aκ2 8πGP
2 + 3H + 2 = − p . (793)
dt a 1 − 2rc H 2 + aκ2
221
It is instructive to cast the background equations into a form resembling an effective
dark-energy fluid with density ρE and pressure PE . This gives
r
3 κ
X = 8πGρE = H2 + 2 , (794)
rc a
and
dH
dt + 3H 2 + 2κ
a2
Y = 8πGPE = − p , (795)
rc H 2 + aκ2
so that we can define the equation of state, wE , as
dH 2κ
PE + 3H 2 + a2
wE = = − dt . (796)
ρE 3H 2 + 3κ
a2

The Bianchi identities, and energy-momentum conservation, ensure that

dρE
+ 3H(ρE + PE ) ≡ 0.
dt
As a consequence, during radiation domination we get wE ≈ − 13 , during matter domina-
tion we get wE ≈ − 12 , during a possible spatial curvature dominated era we get wE ≈ − 32 ,
and eventually during the self-accelerating era we get wE → −1.
Let us now look at the perturbed FLRW universe in DGP. Perturbation theory for
DGP was originally worked out in detail by Deffayet [395], and has since been devel-
oped by a number of authors. Although braneworld cosmological perturbation theory is
discussed in detail in Section 5.4.3, we will now present the formalism for DGP gravity
explicitly, in the interests of self-containment.
Let us consider scalar perturbations in the conformal Newtonian gauge. The compo-
nents of Eµν are then55

E 00 = −µE , (797)
E 0i ~ i ΘE ,
= −∇ (798)
1
Ei j = µE δ i j + Di j ΣE . (799)
3
We find the perturbed Einstein equations are then given by

2 (∆ + 3κ) − 6H (Φ0 + HΨ) = AD 8πGa2 ρδ + BD a2 µE , (800)

0 2 2
(Φ + HΨ) = AD 8πGa (ρ + P ) θ + BD a ΘE , (801)

   
1 1
Φ00 + HΨ0 + 2HΦ0 + 2H0 + H2 + ∆ Ψ − ∆+κ Φ
3 3
 
2 BD 2 2

a2
= AD 4πGa δP − (1 + wE )AD 8πGa ρδ + a µE − µE , (802)
2 2

55 Note that this is consistent with our general analysis for brane world cosmological pertur-

bations, provided one identifies µE = −8πG4 ρweyl δ weyl , ΘE = −8πG4 34 ρweyl θweyl , ΣE =

−8πG4 43 ρweyl Σweyl .

222
 rc2 (X + 3Y )8πG(ρ + P )Σ − 3ΣE
Φ−Ψ = a2 , (803)
3 + rc2 (X + 3Y )

where we recall that primes denote differentiation with respect to conformal time, and
0 2Xrc2
H = aa = aH. Note that we have set AD = 2Xr2 −3 and BD = 2Xr32 −3 = AD − 1,
c c
for simplicity. In the limit rc → ∞ one recovers the familiar perturbation equations of
General Relativity from the above.
We can now use the Bianchi identities to find the field equations for µE and ΘE .
They are
µ0E + 4HµE − ∆ΘE = 0, (804)
and
 
1 H
Θ0E + 4HΘE − µE + (1 + wE ) µE + 3 ΘE
3 a
 
∆ + 3κ 4 1 + wE
+ ΣE + 2 [(2 + 3wE ) Φ − Ψ] = 0. (805)
1 + 3wE 3 a2

It is clear that the above equations are not closed due to the presence of the bulk
anisotropic stress, ΣE . Several authors have applied a variety of approximations to
solve the linear perturbations in the DGP model. Sawicki and Carroll [1101] assumed
that the Weyl perturbations are zero. It is clear, however, from Eq. (805) that this is
not a consistent approximation.
To fully determine the DGP perturbations one must use the five-dimensional equa-
tions. For maximally symmetric 5D space-times, the bulk scalar mode perturbations can
be deduced using a single master variable, Ω, as we discussed in Section 5.4. Assuming
that the bulk cosmological constant is zero, the bulk metric is given by Equation (696),
with
(H 2 + dH )
λαβ = dz 2 − n2 (t, z)dt2 , n(t, z) = 1 + √H 2 +dtκ |z|, (806)
a2
p κ


2
r(t, z) = a 1 +  H + a2 |z| . (807)

Using the Master Equation (705), we find

   
∂ 1 ∂Ω ∂ n ∂Ω ∆ + 3κ n
− − Ω = 0. (808)
∂t nb3 ∂t ∂z b3 ∂z b2 b3

The energy-momentum tensor of the Weyl fluid on the brane can then be related to Ω.
For the case of a spatially flat universe (κ = 0) one finds

1 k4 
µE = − 5 Ω , (809)
3a z=0
 
1 k 2 ∂Ω 
ΘE = − HΩ  , (810)
3 a4 ∂t 
z=0
 2  
1 ∂ Ω ∂Ω k 2 3 dH ∂Ω 
ΣE = − 3 3 2 − 3H + 2Ω− , (811)
6a ∂t ∂t a H dt ∂z z=0

223
where k is the 3-momentum on the homogeneous background. Thus, in general one has
to solve Eq. (808) with appropriate initial and boundary conditions, and then use Eq.
(811) in the perturbed Einstein equations. In practise, one can however apply various
approximations.
Koyama and Maartens [737] assume the small-scale approximation k/a  rc , H.
Under this assumption Eq. (804) implies that ΘE = 0. In quasi-static situations we also
have ∂t Ω ≈ 0 and H1 dH 1 2 2H
dt  H , so that the master equation becomes ∂z Ω − n ∂z Ω −
2
k
a2 n2 Ω = 0. Assuming that the solution of this last equation is regular as z → ∞, it can
k
then be shown that Ω = Ωbr (1 + Hz)− aH when aH/k  1. Therefore, using Eq. (811)
2
one finds that µE = 2k ΣE on the brane, in the quasi-static limit. Inserting this into
Eq. (805) gives ΣE , and therefore µE , in terms of the potentials. This in turn allows us
to eliminate all the Weyl perturbations, to get
 
1
− k 2 Φ = 4πG 1 − ρa2 δM , (812)
3β

and  
2 1
− k Ψ = 4πG 1 + ρa2 δM , (813)
3β
where β = 1 + 2Hrc wE .
Sawicki, Song and Hu [1103] propose a scaling ansatz to close the Bianchi identities,
and solve the master equation on the self-accelerating branch near the cosmological hori-
zon. This allows them to move away from the R a quasi-static regime. The ansatz they use
dã
is Ω = A(p)as G(z/zhor ), where zhor = aH 0 ã2 H(ã) 2 , and s is an exponent that is ap-

proximately constant during times when a particular fluid dominates the expansion. The
master equation then becomes an ordinary differential equation with independent vari-
able z/zhor , and can be solved iteratively by assuming the boundary conditions G(0) = 1
and G(1) = 0. They find that for super-horizon modes s = 6 during radiation domi-
nation, decreasing to s = 4 during matter domination, and finally approaching s = 1
during Λ domination. For all sub-horizon modes s = 3, which reproduces the quasi-
static approximation of Koyama and Maartens [737]. By iteration, one can correct for
the time-dependence of s, as is necessary during transitions between cosmological eras.
Song has performed a similar analysis on the normal branch [1160].
Cardoso et al. [246] solve the master equation numerically by employing null coordi-
nates in the {t, z} plane, thus obtaining the linearised DGP solutions without resorting to
any approximations. They find that the quasi-static approximation is valid to within 5%
for k ≥ 0.01hM pc−1 . Seahra and Hu [1123] develop analytic solutions for both branches
of DGP based on the scaling ansatz, and compare these with numerical solutions in [246].
They find that the analytic/scaling solutions are accurate to within a few percent all the
way to the horizon, and therefore that the use of the scaling ansatz in the observational
constraints imposed in [482, 817] is justifiable.
The FLRW background evolution on the self-accelerating branch has been exten-
sively tested, using a variety of different cosmological observations. Alcaniz [20] used
measurements of the angular size of high redshift compact radio sources [577] to place
constraints on rc . He finds that 4r21H 2 ≥ 0.29 at 1σ, with a best-fit value of rc ∼ 0.94H0−1 .
c 0
Jain, Dev and Alcaniz [658] find that gravitationally lensed QSOs require rc ≥ 1.14H0−1
224
at 1σ. Deffayet et al. [403] use SN-Ia from [1034], and the angular diameter distance
to recombination of pre-WMAP CMB data, while assuming a flat universe. Their SN
analysis gives 4r21H 2 = 0.17+0.03 +0.09 −1
−0.02 at 1σ, which translates to rc = 1.21−0.09 H0 . In-
c 0
cluding CMB data increases the preferred value of ΩM , and leads to a best fit model
with r ∼ 1.4H0−1 . Fairbairn and Goobar [478] use the SNLS SN-Ia data [67] and BAO
data [467] to show that the self-accelerating model is not compatible with a flat universe
at the 99% level. For the generalised model of [453] they find that −0.8 < α < 0.3 at 1σ
(the self-accelerating model corresponds to α = 1). Maartens and Majerotto [846] used
SN-Ia [1064, 67], a CMB shift parameter [1261]), and BAO data [467] to place constraints
on the self-accelerating FLRW background. They find that the self-accelerating model
is consistent with these data sets to within 2σ, but is worse fit than ΛCDM. This puts
tension on the self-accelerating model, but as they point out, it is not necessarily reliable
to use the BAO data for models other than ΛCDM, as ΛCDM is used throughout the
analysis in [467]. Rydbeck, Fairnbairm and Goobar [1080] repeated the analysis of [467]
by including SN-Ia data from ESSENCE [893, 1280], and constraints on the CMB shift
parameter from WMAP-3 [1173] to disfavour the self-accelerating model at the 1 − 2σ
level, depending on whether a peculiar velocity error on the SN-Ia data is included or
not. The inclusion of BAO data from [467] further supports this result.
Moving away from using distance measurements alone, a number of studies have been
performed on the growth of linear structure using the small-scale quasi-static approxi-
mation of Koyama and Maartens [737]. Song, Sawicki and Hu [1161] used the angular
diameter distance to recombination from WMAP-3 [1173], SN-Ia data from [1064, 67],
and constraints on the Hubble constant to exclude the spatially flat self-accelerating
model at 3σ. By allowing non-zero spatial curvature, however, they show that consis-
tency with the data can be improved, but is still marginally worse than spatially flat
ΛCDM. They then use data coming from BAOs, the linear growth of structure, and
ISW and galaxy-ISW correlations to show that any self-accelerating models that shares
the same expansion history as a quintessence-CDM models are strongly disfavoured. As
these authors argue, one must properly take into account the perturbations on the self-
accelerating branch in order to make firm conclusion. Xia considered the generalised
model of Dvali and Turner [453] in conjunction with Union SN-Ia data set [733], CMB
distance measurements, GRB data [1106], and a collection of data on the growth of linear
structure [598, 1248, 1203, 1071, 585, 352, 877]. Using the quasi-static approximation
of Koyama- Maartens [737] to calculate the growth function, Xia finds α < 0.686 at 2σ.
This excludes the self-accelerating value of α = 1.
Fang et al.. [482] perform a thorough test of the self-accelerating model using the
PPF approach of [627, 630]. Using CMB and large-scale structure data these authors
find that the spatially flat model is a 5.3σ poorer fit than ΛCDM, while that open model
is 4.8σ worse. In the latter case non-zero spatial curvature improves the fits for distance
measurements, but worsens those involving the growth of linear structure. One may
speculate that changes to the initial power spectrum may be able to improve the this
situation, but the required reduction in large-scale power also produces unacceptable
reductions in power in the large-scale CMB polarisation spectrum. Changes of this type
cannot therefore save the model. In a follow-up study Lombriser et al. [817] constrained
both branches of the DGP model by employing the PPF description of the CMB and
large-scale structure. These authors find that either brane tension or Λ is required for

225
these models to fit the data well, but that both cases the best fitting models are practically
indistinguishable from ΛCDM. They further find that the cross-over scale is H0 rc > 3 if
spatial curvature is included, and H0 rc > 3.5 in the spatially flat case.
The growth index on small scales, γ, has been calculated by Wei using a Taylor series
expansion in the fractional density parameters ΩDGP and Ωκ [1263]. Ferreira and Skordis
have also calculated this quantity, within a more general frame-work of modified gravity
11 7 93
models, to find γ = 16 − 5632 ΩDGP + 4096 Ω2DGP + O(Ω3DGP ) [501]. This analytic result
is in excellent agreement with numerical studies56 .
Although the DGP model is strongly constrained by the observations we have already
discussed, it is still instructive to consider the substantial work that has been performed
on constructing and studying non-linear structure formation scenarios in these models
[836, 743, 1121, 269, 1112, 1111, 690, 1290, 1113, 1123]. Lue, Scoccimarro and Starkman
[836] have considered spherical perturbations on sub-horizon scales, and derived the grav-
itational force law in a collapsing top-hat model embedded in an expanding background.
They find that the non-linear CDM density contrast evolves as
 
d2 δ M 4 1 dδM dδM
+ 2H − = 4πGeff ρ̄M δM (1 + δM ), (814)
dt2 3 1 + δM dt dt

where  
2 1 √

Geff =G 1+ 1 + D − 1 , (815)
3β D
2
and D = 9β8 2 ΩΩ2 M δM = 89 (1+Ω M)
Ω δ . Their model uncovers a transition point
(1+Ω2M )2 M M
DGP
at D ∼ 1, below which gravity behaves as in GR (in accordance with the Vainshtein
 1/3
16GM rc2
mechanism). This gives a Vainshtein radius of r? = 9β 2 , where M is the mass
of the spherically symmetric object. This result was subsequently re-derived by Koyama
and Silva [741] without the restrictive assumptions of [836]. It can be shown that the
solutions to Eq. (814) are a factor of two larger than the corresponding solutions in
theories that obey Birkhoff’s law, and that have similar expansion histories.
The form of Geff in Eq. (815) is due to an additional degree of freedom in DGP: The
brane-bending mode, as uncovered in [839, 963]. To expand upon this we may rewrite
the RHS of Eq. (814) using 4πGeff δρM = ∆Ψ. The potential Ψ can then be written in
terms of a Poisson potential, ΨP , and a brane-bending mode, ϕ, as Ψ = ΨP + 21 ϕ. These
two new potentials, ΨP and ϕ, satisfy ∆ΨP = 8πGδρM , and ∆ϕ = 8π(Geff − G)δρM .
Ignoring non-local contributions, the brane-bending mode can then be shown to obey
the following equation in the decoupling limit, and in the sub-horizon and quasi-static
regime:
r2 h 2

~ i∇

~ jϕ ∇
i
~ j ϕ = 8πG δρM .
~ i∇
∆ϕ + c (∆ϕ) − ∇ (816)
3β 3β
This equation is closely related to the Field Equation (775), following from the π-
Lagrangian discussed in Section 5.5.4.

56 The terms 3 Ω 1
and 16 Ω2DGP in the Wei [1263] result are incorrect, and appear to have come
16 DGP
from erroneously dropping terms proportional to ΩDGP when going from Eq. (21) to Eq. (22) of that
paper.
226
 Schmidt, Hu and Lima [1113] used the spherical collapse model in DGP to study
the halo mass function, bias and the non-linear matter power spectrum. They find that
top-hat spherical collapse in DGP requires a new, more general method for defining the
virial radius that does not rely on energy conservation. To obtain the comoving number
density of halos per logarithmic interval in the virial mass, and the linear bias, they use
the Sheth-Tormen method [1137], while they use the Navarro-Frenk-White [946] form for
halo profiles. In this way they find that the spherical collapse model agrees well with the
halo mass function and bias obtained from N-body simulations, for both the normal and
the self-accelerating branch. For the non-linear power spectrum, the spherical collapse
model in the self-accelerating branch also matches the simulation results very well. This
is not true, however, for the normal branch, although even in this case the spherical
collapse model predictions are better than those obtained from HALOFIT [1156].
N-body simulations of DGP have been conducted by three independent groups:
Schmidt [1112, 1111], Chan and Scoccimarro [269], and Khoury and Wyman [690]. The
general result, common to all of these studies, is that the brane-bending mode on the
self-accelerating branch provides a repulsive force that greatly suppresses the growth of
structure, while the opposite effect occurs on the normal branch. All three simulations
also display the Vainshtein effect.
The N-body simulations conducted by Schmidt were on both the self-accelerating [1112],
and the normal branch [1111]. Rather than assuming Eq. (815), Schmidt use a relaxation
solver on Eq. (816) to show that the Vainshtein effect is recovered without making any
assumptions about symmetry. Specifically, he finds that the Vainshtein effect is weak-
ened for non-spherically symmetric situations, and in general sets in at smaller scales
than is found in [690]. Like [690], however, he also finds that the HALOFIT model
does not correctly describe the non-linear DGP matter power spectrum. Schmidt then
proceeds to calculate the halo mass function, and shows that the abundance of massive
halos in self-accelerating DGP is much smaller than in CDM models. This last result
puts strong constraints on the self-accelerating model from cluster abundance measure-
ments, independent from of the CMB and large-scale structure constraints discussed
above [482, 817]). For the normal branch, structure is enhanced, and the abundance of
massive halos is larger than a CDM model [1111]. In this case, the halo profiles were
also obtained, and departures from the predictions of GR were seen outside of the halo
virial radius. Finally, Schmidt calculates the bispectrum in both the normal and self-
accelerating branches. The self-accelerating (normal) branch bispectrum is found to be
enhanced (suppressed) for equilateral configurations, but not for squeezed configurations.
This is in agreement with [269], and illustrates the diminishing strength of the Vainshtein
effect for squashed matter configurations.
Scoccimarro [1121] derives the linear and non-linear equations for the growth of struc-
ture in DGP without using the Mukohyama formalism. This results in a set of equations
that includes non-local terms. For example, in the quasi-static limit it is found that
 
a√
∆− −∆ (Φ + Ψ) = 8πGa2 ρδ, (817)
rc
rather than ∆(Φ + Ψ) = 8πGa2 ρδ, as implied by adding the quasi-static expressions
given in Eqs. (812) and (813). Scoccimarro finds that in the linearised quasi-static
limit the bulk behaviour decouples from the brane behaviour, and thus that the non-
local operators can be safely ignored. This ensures the validity of the Koyama-Maartens
227
result [737] on small scales. On larger scales, however, the non-local terms become more
important. On very scales the linear approximation for the brane-bending mode breaks
down, and Scoccimarro finds a non-local and non-linear equation for the potential, Ψ, and
the density perturbation, δ. Chan and √ Scoccimarro √[269] perform N-body simulations by
accounting for the non-local operators −∆ and 1/ −∆. These operators contribute to
the equation for the brane bending mode, ϕ, and the potential, Ψ [1121]. They uncover
the Vainshtein mechanism through a broad transition around k ∼ 2hMpc−1 for z = 1,
and k ∼ 1hMpc−1 for z = 0. They also compute the non-linear matter power spectrum
and bi-spectrum, the CDM mass function, and the halo bias. The results of all this are
in broad agreement with those of Schmidt [1112, 1111].
The simulations of Khoury and Wyman [690] were improved upon in [1290], where
Eq. (816) was solved. It was found that in DGP, and higher dimensional cascading
gravity models, peculiar velocities are enhanced by 24 − 34% compared to CDM [780].
This corresponds to an enhancement by four orders of magnitude in the probability of
the occurrence of high velocity merging system such as the “bullet cluster”.
Scoccimarro [1121] and Koyama, Taruya and Hiramatsu [743] have also developed
two independent techniques based on higher-order perturbation theory in order to find
the non-linear power spectrum. Scoccimarro [1121] shows that the non-linearities coming
from the brane-bending mode can be described by a time and space dependent gravi-
tational constant. He then goes on to develop a re-summation scheme to calculate the
non-linear power spectrum and the bi-spectrum. Koyama, Taruya and Hiramatsu [743]
develop a general perturbation theory method that can be applied in the quasi non-linear
regime of any theory that has an additional scalar degree of freedom (such as Brans-Dicke,
f (R) and DGP). For the case of DGP, they find that their perturbative method recovers
the extreme non-linearity of the Vainshtein mechanism. Their technique has compared
with the HALOFIT mapping [1156], and the non-linear PPF fit of Hu and Sawicki [630].
It is found that HALOFIT over-predicts power on small-scales, in agreement with the
findings of [690, 1112, 1111], while the non-linear PPF fit works well within the regime
of validity of the perturbation theory.

5.6. Higher Co-Dimension Braneworlds

An important characteristic of any braneworld model is the brane’s co-dimension.
This is given by the difference between the dimension of the bulk and the dimension of
the brane. Up to this point we have been concentrating on co-dimension one models.
These are by far the most well developed, mainly because they are much more tractable.
In this section we will discuss braneworld models with co-dimension n ≥ 2. In this case
the gravitational description typically becomes much more complicated.
We have already seen higher co-dimension branes in Section 5.2.1, when discussing
the ADD model [62] in six or more dimensions. This is a particularly simple example
of a higher co-dimension braneworld set-up, corresponding to a tension-less brane in a
Minkowski bulk. In general, of course, one must consider more complicated scenarios,
including the full gravitational back-reaction of brane sources, both in the bulk and on
the brane. Indeed, attempts to solve for a higher co-dimension delta function source
in Einstein gravity will generically result in a singular bulk geometry unless the energy-
momentum of the source is given by pure tension [316]. This is problematic if we insist on
using that infinitely thin defect as a model of our 4-dimensional Universe. However, in a
realistic set-up the defect will not really be a delta function source (it will have some finite
228
thickness, and the would-be singularity will be automatically resolved). Alternatively,
one could consider a genuine delta function source and avoid issues with the singularity
by including higher-order operators in the bulk gravity theory [449, 174, 283, 282].
Despite being extremely difficult to study, higher co-dimension braneworld models
have been discussed in the literature (see, for example, [256, 316, 1250, 965, 13, 448, 449,
450, 518, 436, 669, 668, 381, 386, 388, 379, 387, 900, 12, 338, 339, 174, 280, 281, 277, 1023,
1022, 348, 349]). Much of the interest in this field lies in the fact that these models offer
new ways to think about the cosmological constant problem [256, 316, 1250, 13, 450, 518].
For example, in Section 5.6.2 we will briefly review the importance of higher co-dimension
branes in the degravitation scenario [61].
Co-dimension two branes are particularly interesting as there is reason to believe
that within them one may be able to realise self-tuning of the vacuum energy. This
is because a maximally symmetric co-dimension two brane generically behaves like a
cosmic string in the bulk by forming a conical singularity57 . Changes in tension then
alter the deficit angle, as opposed to the geometry, of the defect [316]. Braneworld
models with two extra dimensions shaped like a rugby ball have been developed with
this in mind [256]. The curvature of the brane is now completely determined by the bulk
cosmological constant and the magnetic flux, independent of the brane tension. One
might then infer that phase transitions on the brane that alter the brane tension do not
alter the curvature. This is not the case, however, as such a transition can affect the
brane curvature via the backdoor, by altering magnetic flux, so there is no self-tuning
[1250, 965]. The situation can sometimes (but not always [214, 215]) be improved by
including super-symmetry and dilaton dynamics in the bulk to protect the relationship
between the bulk cosmological constant and the flux [13, 216, 217, 214, 215]. A low scale
of bulk super-symmetry breaking, set by the size of the rugby ball, can then account for
a small amount of dark energy [216]. Other notable contributions to the literature on
rugby ball compactifications with fluxes and/or scalars include the following: A detailed
discussion of the brane-bulk matching conditions[213, 121], an analysis of the low energy
effective theory [708], and a study of cosmological evolution on the brane [1021].
The rugby ball described above represents a compact extra dimension, with poten-
tially troublesome/interesting light moduli fields [18, 19]. In any event, it is worth pursu-
ing alternative braneworld models with non-compact extra dimensions if they are able to
reproduce 4D gravity at some scale. Co-dimension two branes in Gauss-Bonnet gravity
will be discussed in this context in Section 5.7.4. On the other hand, one might expect
that an infinite bulk should be compatible with some form of quasi-localisation of grav-
ity by considering generalisations of DGP gravity [454] in any number of dimensions.
Such generalisations are sometimes referred to as Brane Induced Gravity (BIG) models
[448, 449, 450, 518, 436, 381, 386, 379, 338, 339, 669, 668]. In the simplest scenario we can
consider BIG models on a single 3-brane of co-dimension n, described by the following
action [448]:
Z 2 Z
M∗2+n 4+n √
Mpl √
S= d x −γR + d4 x −gR. (818)
2 bulk 2 bulk
Using higher derivative operators to resolve the singularity in the bulk propagator58 , we

57 There are exceptions that give rise to curvature singularities as opposed to conical, even for maximal

symmetry on the brane [1208].

58 Note that this theory becomes higher-dimensional at large distances, so the finer details of how the

229
find that for n = 2, 3 the Green’s function in momentum space along the brane takes the
form [449]
2
G̃(p, 0) = , (819)
2 2 M∗2+n
Mpl p + D(p,0)
where Z
dn q 1
D(p, 0) = , (820)
(2π)n p2 + q 2 + (p2 + q 2 )2 /Muv
2

and where Muv . M∗ represents the regularisation scale. Taking Muv ∼ 10−3 eV,
−1
this theory reproduces normal 4D gravity at intermediate scales Muv  r  rc , but
is modified at sub-millimetre scales and at very large distances. It has therefore been
dubbed seesaw gravity in [449]. The large distance cross-over scale can be computed
from Equation (819),  p
 Mpl2 ln(Muv rc ) for n = 2
M∗ q
rc ∼ Mpl M (821)
 2 uv
for n = 3.
M M∗∗

Unfortunately, these constructions suffer from the presence of ghosts in the spectrum of
fluctuations [436].

5.6.1. Cascading gravity

More sophisticated BIG proposals make the transition from 4 + n dimensional gravity
to 4 dimensional gravity via a series of intermediate steps, (4 + n)D → (4 + n − 1)D →
. . . → 5D → 4D [381, 386, 379, 338, 339, 669, 668]. This could potentially help with the
ghost problem. These are sometimes referred to as cascading gravity models as the ef-
fective description tumbles down a cascade of extra dimensions as we come in from large
distances, and increase the resolution of our description. Let us now turn our attention
to this class of gravity models.

The Kaloper-Kiley model

The first braneworld model to explore a cascade from 6D → 5D → 4D gravity

was developed by Kaloper and Kiley [669, 668]. They considered a landscape of models
describing BIG on a 3-brane in six dimensions. The singularities are resolved by modelling
the 3-brane as a cylindrical 4-brane with compact radius r0 . The brane has tension σ5 ,
and some induced curvature weighted by a 5D Planck scale, M5 . The effective 4D
tension and Planck scale at distances r  r0 are then given by σ4eff = 2πr0 σ5 and
M4eff = 2πr0 M53 , respectively. Note that an axion flux is used to cancel the vacuum
pressure in the compact direction q 2 = 2σ5 r02 .
The resulting vacua resemble a cone in the bulk, with the conical singularity cut away
(see Figure 12). The metric is given by

ds26 = ηµν dxµ dxν + dρ2 + f (ρ)2 dφ2 , (822)

theory behaves there will be sensitive to how we resolve the singularity in the bulk [448].

230
Figure 12: Taken from [669]. An illustration of the 2D bulk geometry formed around the resolved brane
in vacuum.

where (
f (ρ) =
ρ   0 ≤ ρ < r0
b
(1 − b) ρ + 1−b r0 ρ > r0 .

The brane tension does not affect the geometry along the brane’s non-compact di-
rections. Instead, it controls the deficit angle measured at infinity, according to b =
2σ5 r0 /M62 , where M6 is the fundamental scale of gravity in the bulk. When the tension
lies below a critical value, σcrit = M62 /2r0 , the deficit angle is less than 2π, and the bulk
space is infinite. For sub-critical branes, σ5 < σcrit , the theory generically resembles
the seesaw gravity theory. We have 4D (Brans-Dicke) gravity at intermediate scales
M4eff p
r0 < r < rc , and a crossover to 6D gravity at a scale rc ∼ (1−b)M 2 ln(2(1 − b)rc /r0 ).
6
Note that the scalar fluctuations naively indicate the possible presence of a ghost, but this
conclusion cannot be trusted since the perturbative theory breaks down due to strong
coupling [669]. This is consistent with the conclusions drawn in [435] for thick brane
regularisations of seesaw gravity.
Things are much more interesting in the near critical limit, for which b = 1 − , where
0 <   1. In this case the brane lies inside a very deep throat, such that the angular
dimension is effectively compactified up to distances of the order r0 / (see Figure 13).
It turns out that perturbation theory is under much better control in this limit, and
that there are no ghosts. The improved behaviour is due to the fact that we make a
series of transitions from 4D → 5D → 6D gravity. Indeed, the theory looks like four
dimensional Brans-Dicke gravity at scales r0 < r < rc , becoming five dimensional beyond
rc ∼ M4eff /M64 r0 . The transition to six dimensions only occurs at very large distances,
r > r0 /, at which point the cylinder decompactifies.
Note that this model does not represent a solution to the cosmological constant prob-
lem. Instead it recasts it in terms of a fine-tuning of the hierarchy between cross-over
scales at each transition.

231
Figure 13: Taken from [669]. An illustration of the 2D bulk geometry formed around the resolved brane
in vacuum, in the near critical limit.

The cascading DGP model

A more recent BIG model exploring these ideas was developed by de Rham et al.,
who coined the phrase cascading gravity [381, 386, 379]. In the simplest model [381], one
has a DGP 3-brane within a DGP 4-brane within a six dimensional bulk. This situation
is described by the following action:
Z Z Z
M4 √ M3 √ M2 √
S= 6 d6 x −g6 R6 + 5 d5 x −g5 R5 + 4 d4 x −g4 R4 , (823)
2 bulk 2 4-brane 2 3-brane
where (g6 )AB , (g5 )ab and (g4 )µν are the metrics in the bulk, on the 4-brane and on the
3-brane, respectively. The corresponding Ricci scalars are given by R6 , R5 and R4 . The
model contains two important mass scales corresponding to the following ratios:
M53 M64
m5 = , and m6 = . (824)
M42 M53
The claim is that the intermediate DGP 4-brane can help to resolve the singularity in the
bulk propagator at the location of the 3-brane [381]. Let us now consider fluctuations
due to a conserved source, Tµν , about a bulk and branes that are all Minkowski. The
field can then be decomposed in terms of a scalar, π, and a transverse and trace-free
tensor, hTT
µν , such that
hµν = hTT
µν + πηµν + gauge terms. (825)
Working in momentum space one finds that the Fourier transformed fields are given by
[381, 386]
  
TT 2 1 1 pµ pν
h̃µν (p) = T̃µν − T̃ ηµν + 2 T̃ , (826)
M42 p2 + g(p2 ) 3 p
 
1 1
π̃(p) = − T̃ , (827)
3M42 p2 − 2g(p2 )
232
where59  √ 2
 πm5 |p −4m26 |

 2 q  p < 2m6
 tanh−1 |p−2m6 |
p+2m6
g(p2 ) = √ 2 (828)

 πm5
|p −4m26 |
p > 2m6 .

 2 −1
q
|p−2m6 |

tan p+2m6

0
The amplitude between two conserved sources on the brane, Tµν and Tµν , now takes the
surprisingly simple form
  
2 1 0µν 1 0 1 p2 + g(p2 ) 0
A= 2 T̃µν T̃ − T̃ T̃ − T̃ T̃ . (829)
M4 p2 + g(p2 ) 3 6 p2 − 2g(p2 )

The last term represents the contribution from the scalar amplitude, whereas the remain-
der is the contribution from the massive spin-2 field. The coefficient of scalar amplitude
changes sign as we flow from the IR to the UV,
(
p2 + g(p2 ) − 12 as p → 0
→ (830)
p2 − 2g(p2 ) 1 as p → ∞,

indicating the presence of a ghost in the UV theory, consistent with the results of [436].
To eliminate this ghost we must add additional operators to the 3-brane action that
modify the scalar propagator in the UV60 . It turns out that this can be achieved simply
by adding tension, λ, to the 3-brane [381, 388]. As the 3-brane is co-dimension 2, its
vacuum solution is unaffected by the change in tension (it is still Minkowski). In contrast,
the 3-brane tension does alter the profile of the fields in the bulk by creating a deficit
angle. This means that kinetic terms describing fluctuations on the new vacuum now
receive corrections from the non-linear bulk interactions. The result is that the scalar
propagator on the 3-brane is modified. One might expect that such effects will necessarily
be suppressed by the five and six dimensional Planck scales, but this is not necessarily so
because of strong coupling of the scalar in the five dimensional boundary effective field
theory [381, 388].
In any event, at higher energies, p  m6 , the modified amplitude goes like
 
  2
2 1 T̃µν T̃ 0µν − 1 T̃ T̃ 0 − 1  p + 2m5 p
Aλ ∼ 2 T̃ T̃ 0  . (831)
M4 p2 + 2m5 p 3 6 p2 1 − 3λ − 4m p
2 22m6 M4 5

2m2 M 2
For large enough tension, λ > 6
3
4
, the scalar amplitude always has the correct sign,
so there is no ghost. However, the tension cannot be arbitrarily large since we require

59 Note that the various formulae for propagators and amplitudes in [381, 386] differ. One must take

m5,6 → 21 m5,6 in going from [381] to [386]. The difference is because only one half of the 6D bulk, and
one half of the 4-brane are considered in [386]. This is a perfectly legitimate truncation by Z2 symmetry
across each of the branes. Here, however, we will adopt the conventions of [381] when expressing our
formulae.
60 It has also been argued that thickening of the brane can help to eliminate the ghost in this model

[386]

233
that the bulk deficit angle is less than 2π. This places an upper bound λ < 2πM64 =
2πm5 m6 M42 . Provided m6 < m5 we therefore have a window,

2m26 M42
< λ < 2πm5 m6 M42 , (832)
3
for which the theory is ghost-free in the UV [381, 388]. The condition m6 < m5 has
added significance in that it permits a transition from 4D to 5D at energies p ∼ m5 . In
particular, we have 4D scalar-tensor gravity in the far UV, for p > m5 > m6 ,
" λ ! #
(4D) 2 0µν 1 1 − m26 M42 0
Aλ ∼ 2 2 T̃µν T̃ − T̃ T̃ , (833)
M4 p 2 1 − 2m3λ
2M 2
6 4

and 5D scalar-tensor gravity for m5 > p > m6 ,

 
(5D) 1 1
Aλ ∼ 3 T̃µν T̃ 0µν − T̃ T̃ 0 . (834)
M5 p 4

In the far infra-red, p < m6 , we recover 6D gravity,

 
(6D) ln(m6 /p) 0µν 1 0
A ∼ T̃µν T̃ − T̃ T̃ . (835)
πM64 4

Thus we have a cascade from 6D → 5D → 4D gravity as we move from large to short

distances. It is interesting to note that the existence of this cascade seems to be closely
related to the absence of ghosts. By taking M4 ∼ Mpl , M5 ∼ 10 MeV and M6 ∼ meV,
the cascade occurs at the current horizon scale, as m6 ∼ m5 ∼ H0 . Note that a cascade
from seven dimensions has also been studied in some detail in [387].
For tensions in the range allowed by Eq. (832) the 4D theory at short distances is a
Brans-Dicke theory that is incompatible with solar system constraints. The authors of
[381] suggest that some sort of Vainshtein mechanism could account for the recovery of
GR at short scales. Indeed, the role of non-linearities in cascading DGP has not yet been
properly investigated, and one may be concerned that there are issues surrounding the
validity of the linearised results (as was found in the sub-critical regime of the Kaloper-
Kiley model [669]).
Another aspect of cascading DGP that requires further study is cosmology (see
[900, 12] for some preliminary work). In [12] a five dimensional proxy theory is used
that is expected to capture many of the salient features of the original model. This
proxy model corresponds to a 5D scalar-tensor theory that, in the limit where the scalar
and tensor degrees of freedom decouple, agrees with the decoupling limit of the 5D
boundary effective field theory of cascading DGP presented in [381]. Of course, it is by
no means unique in this respect.

Induced gravity on intersecting braneworlds

A third BIG model that is very closely related to cascading DGP was developed by
Corradini, Koyama and Tasinato [338, 339]. Here a six dimensional bulk contains two
intersecting DGP 4-branes, each with induced curvature. At the intersection we place a
234
DGP 3-brane, also with some induced curvature. The cosmological evolution is derived
using a formalism based on mirage cosmology [685]. The resulting Friedmann equation
on the 3-brane is then given by [339]
s " s #
2 2 3 H2 4 −1 H2
ρ = 2M4 H + 6M5 k2 1 + 2 + 4M6 tan tan α 1 + 2 , (836)
k2 k2

where α is the angle between the two 4-branes, and k2 is a constant that encodes in-
formation about the warped geometry in the bulk (and can be derived in a non-trivial
way from the parameters of the theory). The model claims to admit self-accelerating
and self-tuning vacua. For fluctuations on a Minkowski 3-brane one may also expect a
cascade from 6D → 5D → 4D gravity as we move from large to short distances, for suit-
ably chosen scales. A thorough perturbative analysis has yet to be done that confirms
this expectation, or the possible existence of ghosts and strong coupling.

5.6.2. Degravitation
Higher co-dimension braneworld models and cascading gravity are expected to play
an important role in realising the degravitation scenario [61]. The idea of degravitation
is best understood by rephrasing the cosmological constant problem. Instead of asking
“why is the vacuum energy so small?”, we ask “why does the vacuum energy hardly
gravitate at all?”. In other words, we accept what our particle physics models are saying
and take the vacuum energy to be up at the (TeV)4 scale or beyond. We then try to
develop a gravity theory that prevents this large vacuum energy from generating a large
amount of curvature.
A phenomenological description of degravitation is given by [61]
G −1 (L2 )Gµν = 8πTµν , (837)
where Newton’s constant, G, has been promoted to a differential operator, G(L2 ),
depending on a length scale, L, and the covariant d’Alembertian operator, . The idea is
that this operator behaves like a high pass filter characterised by the scale L. Sources with
short characteristic wavelengths, l  L, pass through the filter and gravitate normally.
Sources with long characteristic wavelengths, l  L, such as the cosmological constant,
are filtered out and hardly gravitate at all. These considerations amount to imposing
the following limits: G → G for L2 (−) → ∞ and G → 0 for L2 (−) → 0.
Since the covariant derivative does not commute with the d’Alembertian, Equation
(837) suggests that the energy-momentum tensor is not conserved. However, it is impor-
tant to realise that this equation is only expected to describe the linearised dynamics of
the helicity-2 mode of the graviton, given by gµν = ηµν + hµν . The source is conserved
with respect to the full metric, ĝµν = ηµν + hµν + . . ., where the ellipsis denotes the
additional (Stuckelberg) modes that necessarily appear in a fully covariant theory [452].
It is clear that the extra modes must play an important role in the filtering process at
long wavelengths. The transition to normal gravity for short wavelength modes occurs
precisely because these modes get screened by the Vainshtein mechanism [1241] (see our
discussion of the Vainshtein mechanism in the context of DGP gravity in Section 5.5.4).
The filter operator is often parametrised as follows:
 
−1 2 m2 ()
G (L ) = 1 − G−1 , (838)

235
where m2 () ∝ L2(α−1) α , and 0 ≤ α < 1. The upper limit is a necessary, but not
sufficient, condition for degravitation, whereas the lower limit is required by unitarity
[452]. Massive gravity (see Section 3.3.3) corresponds to α = 0, whereas DGP gravity61
(see Section 5.5) corresponds to α = 1/2. It turns out that models with co-dimension
n > 2 are expected to correspond to α = 0 [386].
Using Equation (837) as its starting point, degravitation can be demonstrated in a
cosmological context by the ratio of scalar curvature to Λ scaling as t−1/2 for a broad
class of filters, where t is the proper time of comoving observers [452]. However, the
corresponding cosmological solution arising from a higher co-dimension brane scenario
has yet to be found explicitly.
Working at the level of the phenomenological equations, N-body simulations for de-
gravitation scenarios have been studied by Khoury and Wyman [690]. Their simulation
is based on a modified Poisson equation, given by
"  2α #
k2 1 kL
− 2 + 2 Ψ = 4πGρ. (839)
a L a

Thus, the emergence of the Vainshtein mechanism is encoded in the simulations via Eq.
(815), rather than being recovered from the simulation itself. They find that the matter
power spectrum determined from the simulation agrees well with the HALOFIT [1156]
formula for k <∼ 0.2hMpc−1 . However, for smaller scales the HALOFIT formula over-
predicts power because HALOFIT does not capture the Vainshtein effect. Nevertheless,
they find a way to recalibrate the HALOFIT parameters such that it agrees with their
simulation on all scales.

5.7. Einstein Gauss-Bonnet Gravity

In Section 2.4.1 we outlined how Lovelock’s theorem [831, 832] constrains the class
of theories that can be constructed from the metric tensor alone: In four dimensions,
the most general rank-2 tensor that can be derived from a variational principle and is
(i) symmetric, (ii) divergence free, and (iii) built out of the metric and its first two
derivatives only is given by a linear combination of the metric and the Einstein tensor,
i.e.
(4D) 1
Eµν = − α0 gµν + α1 Gµν . (840)
2
Such a term arises from the variation of the Einstein-Hilbert action in the presence of a
cosmological constant,
Z
(4D) 1 δ √
Eµν =√ µν
d4 x −g (α0 + α1 R) . (841)
−g δg

In four dimensions the Einstein-Hilbert action therefore gives the most general field
equations with the desired properties. In more than four dimensions, however, this result

61 The full DGP theory does not exhibit degravitation, but this is not necessarily the unique theory
with α = 1/2.

236
no longer holds. For D = 5 or 6 dimensions the most general rank-2 tensor satisfying
the same three conditions is given by [831] (see also [767, 1313])

(5D/6D) 1
Eab = − α0 γab + α1 Gab + α2 Hab , (842)
2
where (in D dimensions) γab is the metric, Gab = Rab − 12 Rγab is the Einstein tensor,
and we have introduced the Lovelock tensor [831]
1
Hab = 2RRab − 4Raα Rcb − 4Racbd Rcd + 2Racde Rb cde − Ĝγab , (843)
2

where Ĝ = Rabcd Rabcd − 4Rab Rab + R2 , and Rabcd is the Riemann tensor constructed
from γab . The Lovelock tensor is obtained by variation of the Gauss-Bonnet action,
Z
1 δ √
Hab = √ ab
dD x −g Ĝ. (844)
−γ δγ

These results can generalised still further as the dimensionality of space-time is increased.
√ √
To see how, note that dD x −γR and dD x −γ Ĝ correspond to the Euler classes of order
one and order two [936], and are topological in D ≤ 2 and D ≤ 4 respectively62 . To
generalise Lovelock’s theorem to even higher dimensions we must then add the higher
order Euler classes to the corresponding action, and compute the metric variation. The
Euler class at order k depends on kth powers of curvature. In D dimensions we must
therefore include Euler classes up to order [(D − 1)/2], where the square brackets denote
the integer part. The reader is referred to [270], and references therein, for further details.

5.7.1. Action, equations of motion, and vacua

In five or six dimensions General Relativity is just a special case of a broader class of
theories commonly referred to as Einstein-Gauss-Bonnet (EGB) gravity. These theories
are described by the action
Z   Z
1 D √
S= d x −γ R − 2Λ + αĜ + dD xLm (γab , ψ), (845)
16πGD
where Lm is the Lagrangian density of the higher dimensional matter fields that are
minimally coupled to the metric. The corresponding field equations are given by

Gab + Λγab + αHab = 8πGD Tab , (846)

2 δ
R
where Tab = − √−γ δγ ab
dD xLm (γab , ψ) is the energy-momentum tensor of the matter
fields. As we will see shortly, although Λ acts like a bare cosmological constant, it differs
from the effective cosmological constant, Λeff , seen by the geometry.
The Gauss-Bonnet corrections are weighted by the parameter α. This has dimensions
of [length]2 and is often associated with the slope parameter, α0 , in heterotic string the-
ory. To see why, consider the effective theory describing the dynamics of the heterotic

62 One can check explicitly that the Einstein tensor is identically zero in two dimensions, and that the

Lovelock tensor is identically zero in four dimensions.

237
string on the 10 dimensional target space. Working at tree level in the string coupling
and performing a perturbative expansion in the inverse string tension, µ−1 F ∼ 2πα0 ,
we see that the leading order term gives the standard Einstein-Hilbert action, and the
next to leading order term, after some field redefinitions, gives the Gauss-Bonnet ac-
tion [566, 886]. In fact, one may have expected such a result based on the fact that
string theory is known to be ghost free. As noted by Zwiebach [1316], second-order
corrections to the Einstein-Hilbert action will necessarily give rise to ghosts unless they
appear in the Gauss-Bonnet combination. Note that if we do identify EGB with a stringy
generalisation of General Relativity, we should restrict attention to α ≥ 0 for consistency.

Maximally symmetric vacua

We now consider the maximally symmetric vacuum solutions, γ̄ab , satisfying R̄abcd =
2Λeff
(D−1)(D−2) (γ̄ac γ̄bd − γ̄ad γ̄bc ), where Λeff is the effective cosmological constant seen by the
curvature. There are two possible values for Λeff , given by [279]
r !
± 2Λ
Λeff = ΛCS 1 ± 1 − , (847)
ΛCS

where
1 (D − 1)(D − 2)
ΛCS = − . (848)
4α (D − 3)(D − 4)
For these vacua to be well defined the bare cosmological constant must satisfy the bound
Λ/ΛCS ≤ 1/2. It is easy to check that Λ+ −
eff /ΛCS ≥ 1 ≥ Λeff /ΛCS , with equality when
Λ = ΛCS /2. This is known as the Chern-Simons limit (at least in odd dimensions)
[1297, 344], and corresponds to the case where the two roots coincide. Note that only
the lower root has a smooth limit, Λ− eff → Λ, as α → 0, and as such is often referred
to as the “Einstein” branch. In contrast, the upper root, Λ+ eff , is not smooth as α → 0,
and represents a distinct new feature of EGB gravity that is completely absent in higher
dimensional General Relativity. For this reason, this branch is often referred to as the
“stringy”, or “Gauss-Bonnet”, branch.
We now consider metric perturbations about these vacua, γab = γ̄ab + δγab . The
linearised field equations then take the remarkably simply form

δGab + Λeff δγab = 8πGeff Tab , (849)

where δGab is the linearised Einstein tensor. Thus, we have perturbative Einstein gravity
with an effective Newton’s constant given by
GD
Geff = . (850)
1 − ΛΛCS
eff

Assuming that the bare Newton’s constant is positive, GD > 0, it follows that pertur-
bative gravity on the Einstein branch (Λeff /ΛCS < 1) is essentially well behaved, as
Geff > 0. In contrast, on the Gauss-Bonnet branch (Λeff /ΛCS > 1), we have Geff < 0
indicating the presence of a perturbative ghost [279]. We refer the reader to the closing
paragraphs of Section 5.5.3 for a discussion of the pathologies associated with ghosts.
238
 The above conclusions regarding stability are robust provided we can trust our ef-
fective perturbative description. Here we expect this description to be valid at energies
up to a cut-off, Ecut-off ∼ 1/(Geff )1/(D−2 . As we approach the Chern-Simons limit
(Λ → ΛCS /2), where the two branches coincide, it follows that the cut-off for the effec-
tive description should have the limit Ecut-off → 0. This indicates strong coupling, and a
breakdown of perturbation theory. To analyse the stability of either branch close to this
limit one must study non-perturbative phenomena such as instanton transitions. This
reveals that transitions between branches are unsuppressed in the near Chern-Simons
regime, and that there is very strong mixing between the two (almost degenerate) vacua.
We conclude that neither of them can accurately describe the true quantum vacuum state
in this regime, as both will quickly become littered with bubbles of the other vacuum
[279].

Spherically symmetric solutions

Static spherically symmetric solutions to the vacuum field equations were first dis-
covered by Boulware and Deser [176],

dr2
ds2 = −V (r)dt2 + + r2 dΩD−2 , (851)
V (r)
where dΩD−2 is the metric on a unit (D − 2)-sphere. We have two branches for the
potential, V (r), given by [176, 279]
s !
2ΛCS r2 2Λ M (D − 1)
V± (r) = 1 + 1± 1− − , (852)
(D − 1)(D − 2) ΛCS ΛCS ΩD−2 rD−1

where ΩD−2 is the volume of the unit (D − 2)-sphere. M is an integration constant

that one can identify with the mass of a spherically symmetric source (possibly point-
like). As above, the lower root corresponds to the Einstein branch, and has a smooth
limit as α → 0. The upper root corresponds to the Gauss-Bonnet branch, and does not
have a smooth limit. Of course, the two roots are degenerate in the Chern-Simons case
(Λ = ΛCS /2).
We shall stay away from the Chern-Simons limit in the remainder of this section. Note
that a generalised form of Birkhoff’s theorem now holds that guarantees that Equation
(851) represents the most general spherically symmetric solution, even without the as-
sumption of staticity [272]. Let us consider the properties of this solution for a source
of positive mass, M > 0. On the Einstein branch the singularity at r = 0 is shielded
by an event horizon, and there are no obvious pathologies. This is not the case on the
Gauss-Bonnet branch, where we have a naked time-like singularity.
At asymptotically large radii the metric functions take the form

2Λ±eff r
2
M
V± ≈ − +1± . (853)
(D − 1)(D − 2) ΛCS ΩD−2 rD−3
The asymptotic solution here can be seen to resemble the generalised Schwarzschild-
Tangherlini solution in D dimensional GR [1202], with cosmological constant Λ± eff , and
mass ±M . In particular, on the Gauss-Bonnet branch, it appears as if our solution has
239
negative gravitational energy even for a positive energy source. This conclusion, however,
is incorrect. A proper computation of the gravitational energy taking into account all
of the Gauss-Bonnet corrections reveals the mass of the solution in Eq. (851) is +M on
both branches [415, 416, 1010]. For an in-depth discussion of this, and related stability
issues, the reader is referred to [279]. For an excellent review of Lovelock gravities, and
their black hole solutions, see [270].

5.7.2. Kaluza-Klein reduction of EGB gravity

In Section 5.1.1 we discussed compactification of higher dimensional GR on a circle.
The same operation can now be performed with the EGB action, given in Eq. (845). We
start in five dimensions with coordinates (xµ , z), where the z direction is compact. We
can then dimensionally reduce down to four dimensions using the following ansatz

γµν = gµν + e2φ Aµ Aν , γµz = e2φ Aµ , and γzz = e2φ , (854)

where the metric, gµν , gauge field, Aµ and the dilaton, φ, are all independent of z.
Integrating out the compact dimension we arrive at the following effective theory [915]:
Z
1 √
Seff = d4 x −gLeff , (855)
16πG5
where
 
φ 1 2φ 2 µναβ µν 2
Leff = e R − e F − 2Λ + α(Rµναβ R − 4Rµν R + R )
4
  
3φ 3 2φ µ ν α β 1 2 2
+ αe − e F ν F α F β F µ − (F )
8 2
 
µ ν µν αβ α β α β
− (F α F β + F Fαβ ) R µν − 4R[µ δν] + Rδ[µ δν]
− 2(∇µ Fαβ )(∇α F βµ ) − 2(∇µ Fαµ )(∇ν F αν ) − 12(Fµα ∇α φ)(F µβ ∇β φ) + 6F 2 (∇φ)2
#
+4(Fαβ ∇µ φ)∇µ F αβ − 4(Fαβ ∇µ φ)∇α F βµ − 12(Fαβ ∇β φ)∇µ F αµ
   
+ αeφ 8e−φ ∇µ ∇ν eφ − 2e2φ Fµα Fν α Gµν + 3αe2φ 2F αβ Fβ µ ∇µ ∇α eφ + F 2 eφ ,
(856)

and F 2 = Fµν F µν , with Fµν = ∇µ Aν − ∇ν Aµ . Note that for Λ = 0 and α = 0 we

recover the results of Section 5.1.1, provided we perform a conformal transformation to
the Einstein frame and canonically normalise the dilaton.
One property of the Lagrangian in Eq. (856) is that it gives rise to field equations that
are at most second order in derivatives. This is, of course, inherited from the underlying
theory, and enables us to consider an interesting class of scalar-tensor theories by freezing
the gauge field such that Aµ = 0. We refer the reader to Section 3.1, and [38], for further
discussions on four-dimensional theories of this type.

5.7.3. Co-dimension one branes in EGB gravity

Interest in EGB gravity has really taken off in recent years, largely due to its appli-
cation to braneworlds. We discussed the braneworld paradigm in some detail in Section
240
5.2. Schematically, the physics of 3-branes in five dimensional EGB gravity is more or
less the same as in General Relativity, as we will now discuss.

Generic action and equations of motion

As in the GR case discussed in Section 5.3.4, we split our five dimensional bulk into
a series of domains separated by a series of 3-branes, so that the action is given by
Z  
M53
5 √
S= d x −γ (R + αĜ) + Lbulk
bulk 2
X Z
√   
+ d4 x −g −∆ M53 (K + 2α(J − 2Gµν Kµν )) + Lbrane , (857)
branes brane

where M5 = (1/8πG5 )1/3 is the bulk Planck scale. In comparison with Equation (615) we
have a Gauss-Bonnet correction in the bulk and the corresponding Myers boundary term
[931] on each brane. The latter depends on the extrinsic curvature, Kµν , the induced
Einstein tensor Gµν = Rµν − 21 Rgµν , and the trace J = g µν Jµν , where [363]

1

Jµν = 2KKµα Kνα + Kαβ K αβ Kµν − 2Kµα K αβ Kβν − K 2 Kµν . (858)
3
The bulk field equations are then given by
1 1 bulk
Rab − Rγab + αHab = 3 Tab , (859)
2 M5

with the boundary conditions at the brane given by the “DGW” junction conditions
[363, 553] (see also [409]),
  
∆ M52 Kµν − Kgµν + 2α(3Jµν − Jgµν − 2Pµανβ K αβ ) = −Tµνbrane
. (860)

Here we introduce the double dual of the Riemann tensor, defined as [902]

P µναβ = −Rµν αβ + 2Rµ[α δ νβ] − 2Rν [α δ µβ] − Rδ µ[α δ νβ] . (861)

In what follows we will assume that the bulk geometry is only63 sourced by a cosmological
bulk
constant Tab = M53 Λγab , which we take to be negative, as in the Randall-Sundrum
model, Λ = −6k 2 . We will consider a single brane with tension σ, and some induced
2
curvature, Mind R. Its energy momentum tensor is then given by
brane 2
Tµν = −σgµν − Mind Gµν + Tµν , (862)

where Tµν denotes the contribution from any additional matter excitations on the brane.
For simplicity we will impose Z2 symmetry across the brane (see [1012, 1013] for a dis-
cussion of asymmetric configurations).

63 For generalisations with bulk scalar fields and bulk Maxwell fields see [271] and [805], respectively.
241
Weak gravity on a Minkowski brane

Let us seek Randall-Sundrum-like vacua, corresponding to a Minkowski brane in an

AdS bulk,
ds2 = γ̄ab dxa dxb = dz 2 + e−2|z|/leff ηµν dxµ dxν . (863)
The effective AdS curvature in the bulk is then determined by the bulk equations of
motion (859). We find !
r
1 1 8α
2 = 4α 1± 1− 2 , (864)
leff l
where the lower root corresponds to the Einstein branch and the upper root to the GB
branch. Meanwhile, the vacuum DGW junction conditions at z = 0 impose the following
constraint on the brane tension
 
M53 4α
σ=6 1− 2 . (865)
leff 3leff

Since we are interested in weak gravity on the brane, we consider small fluctuations in
the metric and the brane position. As we saw previously, the linearised equations of
motion are identical to those found in perturbative General Relativity,
6
δGab − 2 δγab = 0. (866)
leff

The linearised junction conditions also take a remarkably simple form,

2(M5eff )3 δ(Kνµ − Kδνµ ) = Mren

2
δGµν − Tνµ , (867)

where we identify the effective five dimensional Planck scale,

 
eff 3 3 4α
(M5 ) = M5 1 − 2 (868)
leff

and the renormalised induced curvature scale,

2 2 8αM53
Mren = Mind + . (869)
leff
The effective Planck scale is consistent with the effective Newton’s constant introduced
in Equation (850), (M5eff ) = (1/8πGeff )1/3 . It follows that our weak gravity description
is identical to the corresponding description in General Relativity, provided we make use
of the effective cosmological constant and Planck scale in the bulk, and renormalise the
induced curvature on the brane. For further details the reader is referred to [270], and
references therein (see also [413]).
The characteristic behaviour of weak gravity on the brane is as follows [270]: At
large distances we recover four dimensional General Relativity, with an effective four
dimensional Planck scale
q q
IR 2 3 2 2 + (M eff )3 l .
Mpl = Mind + M5 leff (1 + 4α/leff ) = Mren 5 eff (870)
242
This holds provided 1/leff 6= 0. For 1/leff = 0 the large distance behaviour is five dimen-
sional owing to the absence of a normalisable zero mode in the Minkowski bulk.
At short distances we recover fourpdimensional Brans-Dicke gravity, with a differ-
ent effective Planck scale64 , MplUV
= Mind 2 + 8αM 3 /l
5 eff = Mren , and a Brans-Dicke
parameter given by
 2  !
IR 2
3 Mind /leff + M53 (1 + 4α/leff
2
) 3 (Mpl )
2w + 3 = = . (871)
4 M53 (1 − 4α/leff
2 ) 4 (M5eff )3 leff

This holds provided Mren 6= 0. If Mren = 0 then the short distance behaviour is five
dimensional since there is no effective induced curvature giving rise to quasi-localisation
near the brane.
These results are consistent with the behaviour that would be expected in linearised
theory on a brane in five dimensional General Relativity, with bulk cosmological con-
2
stant, Λeff = 6/leff , bulk Planck scale, M5eff , and brane induced curvature scale, Mren .
It is interesting to note that the Brans-Dicke parameter gets large close to the Chern-
Simons limit as M5eff → 0. However, as we saw earlier, quantum fluctuations in the bulk
become strongly coupled at ∼ M5eff , so this prediction may be unreliable.

Brane cosmology in EGB gravity

The methods used in Section 5.4 to derive the cosmological evolution of co-dimension
one branes apply equally well to branes in EGB gravity, provided we use the DGW
junction conditions give in Eq. (860). Recall that brane cosmology can be studied using
either the brane-based formalism or the bulk-based formalism. The brane-based formalism
relies on the covariant formulation of the effective Einstein equation on the brane. This
has been worked out for EGB gravity and applied to cosmological branes [859] (see also
[709]). In the bulk based formalism, the generalised form of Birkhoff’s theorem [272]
ensures that the bulk geometry around a FLRW brane is given by

dr2
ds2 = −V (r)dτ 2 + + r2 qij dxi dxj , (872)
V (r)

where V (r) takes the form

r !
r2 8α 8αµ
V± (r) = κ + 1± 1− 2 − 4 , (873)
4α l r

and qij (x) is the metric of a 3-space of constant curvature, κ = 0, ±1. Note that this
generalises the black hole solution given in Eq. (852) to different “horizon” topologies
[233]. Let us once again assume Z2 symmetry across the brane for simplicity. As in
Section 5.4.2, we treat the brane as an embedding τ = τ (t), and r = a(t) in the bulk
geometry, identifying t with the proper time of comoving observers on the brane, and
a(t) with the scale factor. After imposing the DGW boundary conditions given in Eq.

64 There UV in [270].
is a typo in the corresponding expression for Mpl
243
(860), we find that the Hubble parameter, H = ȧ/a, obeys [270] (see also [363, 272, 557,
851, 718])
    r
4α 2 κ  κ − V (a) V (a) ρbrane
1+ 2 H + 2 + H2 + 2 =
3 a a2 a 6M53


2
σ + ρ − 3Mind H 2 + aκ2
= .(874)
6M53

This can be recast as a cubic equation in H 2 + aκ2 , and solved analytically to give a
modified Friedmann equation. The most commonly studied scenario has Mind = 0, with
the bulk taken to lie on the Einstein branch. We then find [363]
κ c+ + c− − 2
H2 + 2
= , (875)
a 8α
where s 2/3
 3/2 r
8α 8αµ α(σ + ρ)2 (σ + ρ) α 
c± =  1− 2 + 4 + ± . (876)
l a 2M56 M53 2

This cosmology can give rise to rapid inflation, just as in Randall-Sundrum cosmology
[806]. However, the Gauss-Bonnet corrections do introduce some new features at the
level of cosmological perturbations. For example, in Randall-Sundrum cosmology the
consistency relation between the tensor-to-scalar ratio, r, and the tensor spectral index,
nT , agrees with the standard GR result, r = −8nT . This relationship is broken by the
GB corrections as the amplitude for the tensor perturbations are no longer monotonically
increasing with scale [437]. We also find that observational constraints of GB brane
inflation are typically softened relative to Randall-Sundrum cosmology, such that certain
“steep” potentials are no longer ruled out [1224].

5.7.4. Co-dimension two branes in EGB gravity

In Section 5.6 we mentioned that higher-order operators could be added to the
bulk gravity to regularise the singularities that appear when we have an infinitely thin
braneworld of co-dimension ≥ 2. One of the ways to do this is to add Gauss-Bonnet
corrections in the bulk [174, 283, 282]. The simplest and most well studied scenario is to
consider co-dimension-2 branes in six dimensional EGB gravity without any additional
sources in the bulk (see, for example, [174, 277, 676, 675, 281, 280, 348, 349]). This gives
Z   Z
√ 1
S= d6 x −γ (R + αĜ) + d4 xLbrane . (877)
bulk 16πG 6 brane

If we assume axial symmetry in the bulk we get

ds2 = γAB dxA dxB = dr2 + L2 (x, r)dθ2 + gµν (x, r)dxµ dxν , (878)

and it turns out that the Einstein tensor on the brane is given by [174, 277]

1 G6
Gµν = − gµν + T brane + f (β)Wµν , (879)
4α α(1 − β) µν
244
where
1
Wµν = Kµλ Kνλ − KKµν + gµν (K 2 − Kλσ K λσ ), (880)
2
and Kµν = 21 ∂r gµν . The parameter β is the deficit angle on the brane, and is often
assumed to be constant, although this need not be case [277]. A field dependent deficit
angle, β = β(x), will lead to two important effects: Transfer of energy between bulk and
brane, and an effective four dimensional Newton’s “constant” that can vary as Geff =
G6
8πα(1−β(x)) [277]. These can clearly be constrained by observation, but it would be very
interesting to study the role varying β could play in attempts to self-tune the vacuum
curvature in co-dimension 2 models.
The function f (β) depends on the mathematical technique used to derive the bound-
ary conditions at the distributional source [277, 676, 675]. The boundary conditions
derived in [174] lead to the condition Wµν |brane = 0, suggesting that Einstein gravity
should be recovered on the brane at all scales, even for an infinitely large bulk. However,
it is now understood that these conclusions rely on the assumption ∂r Kµν = 0 at the
brane, which is too constraining. For constant β the Friedmann equation on the brane
is given by [277]
κ 8πGeff 1 c2
H2 + 2 = ρbrane − + 2 , (881)
a 3 12α ρbrane a8
where c2 is an integration constant. However, as we already emphasised, in general β can
vary, and in this case one cannot find a closed system of equations on the brane [277].

245
6. Parameterised Post-Friedmannian Approaches and Observational Constraints

As expounded in this review, there now exists a vast range of candidate theories of
gravity that modify Einstein’s theory of General Relativity in one way or another. It is
also seems clear that many more such theories are likely to be proposed in the future.
If these theories are to be of any value in understanding and resolving the problems
associated with the Dark Universe then they must be confronted with cosmological data.
This is, in principle, straightforward but time-consuming. It involves working out the
perturbation equations for each and every theory, incorporating them into the Einstein-
Boltzmann solvers and N-body codes, and calculating a list of observables.
This situation is analogous to the experimental study of General Relativity in the
early 1970s, during what has become known as the ‘Golden Age’ of General Relativity.
There, one had a plethora of alternative theories of gravity that needed to be confronted
with constraints from Solar System measurements. The Parameterised Post-Newtonian
(PPN) method was invented in this case as an intermediate step between theory and
experiment. It involves a set of generic parameters that can be easily constrained by
experiments [1207]. Using the PPN method one can then take any given theory, calculate
the PPN parameters it predicts, and compare them with observational constraints. This
process is outlined in Section 2.5.
Over the last few years, the idea of creating such an intermediate step when con-
sidering cosmological constraints has starting taking hold. It has been dubbed by some
the ‘Parameterised Post-Friedmannian approach’, and it attempts to encompass, at the
linear level, the behaviour of a wide array of alternative theories of gravity. We will now
outline the basic idea behind this approach.
The bulk of the cosmological data that can be used to constrain modifications of grav-
ity can be interpreted in terms of perturbations about a Friedmann-Lemaı̂tre-Robertson-
Walker universe. Throughout this review we have presented how the evolution of cosmo-
logical perturbations is modified in these theories, relative to their behaviour in General
Relativity. One can now ask oneself if there is a general way of modifying the equa-
tions of cosmological perturbations such that it will encompass all the theories we have
previously discussed.
The simplest approach, that has been in vogue for the past few years, is to modify
two of the four Einstein field equations as follows:
X
− 2k 2 Φ = 8πµGa2 ρX [δX + 3(1 + wX )HθX ] (882)
X
Φ−Ψ = ζΦ, (883)

where two new functions have been introduced: The effective Newton constant, Geff =
µG, and the gravitational slip, ζ. One can interpret Geff as the inclusion of a form of grav-
itational screening, reducing (or enhancing) the local gravitational force on cosmological
scales. The gravitational slip phenomenologically parametrises the shear that seems to
arise frequently in scenarios of modified gravity. This parametrisation is incredibly useful
for quantifying deviations from General Relativity, and a number of authors have used
it in their analyses of cosmological data [148, 237, 630, 627, 41, 360, 361, 123, 1302].
It is instructive to see in what circumstances such a parametrisation might arise, and
to do this we will develop a consistent formalism in what follows.
246
6.1. The Formalism
In order to generalise the perturbed Einstein equations we follow the approach and
notation used in [1149, 1151, 501, 77]. Here we split the field equations of the theory
in question into a set of evolution equations for the metric, evolution equations for the
additional gravitational fields (if any are present), and a set of constraint equations. The
evolution equations for the metric, and the constraint equations, can then be written
schematically as  
δGmod
µν ≡ δGµν − δUµν = 8πG δTµν + δTµν ,
E
(884)
where δGµν is the perturbed Einstein tensor, δUµν is the contribution of any other terms
that involve perturbations of the metric, δTµν is the perturbed energy-momentum tensor
E
of matter fields in the space-time, and δTµν is the contribution of any terms that involve
perturbations to the additional gravitational fields.
Let us now be more specific. We can define the following new variables: U∆ ≡
~ i UΘ ≡ −a2 δU 0 , UP ≡ δU i and Di UΣ ≡ a2 (δU i − 1 δU k δ i ), as well as the
−a2 δU 00 , ∇ i i j j 3 k j
new gauge invariant
1 0 
Γ̂ ≡ Φ̂ + HΨ̂ , (885)
k
where the Bardeen potentials Φ̂ and Ψ̂ are defined in Eqs. (85) and (86) of Section 2.6.
Dropping the hats (i.e. working in the conformal Newtonian gauge), we can write the
two constraint equations coming from G00 and G0i as
X
− 2k 2 (Φ + 3Hk Γ) = 8πGa2 ρX δX + U∆ (886)
X
X
2
2kΓ = 8πGa (ρX + PX )θX + UΘ , (887)
X

and the two evolution equations coming from the trace and traceless parts of Gij as
  X
0 0 2 k2 k2 1
kΓ + 2kHΓ + H − H − Ψ + Φ = 4πGa2 δPX + UP (888)
3 3 6
X
X
Φ − Ψ = 8πGa2 (ρX + PX )ΣX + UΣ .(889)
X

We can then combine Eqs. (886) and (887) to find a modified Poisson equation:
X
− 2k 2 Φ = 8πGa2 ρX [δX + 3H(1 + wX )θX ] + U∆ + 3HUΘ . (890)
X

Assuming that the theory in question has at most N -time derivatives in its field
equations, and bearing in mind from Eqs. (85), (86) and (885) that Φ̂ and Γ̂ have one
time derivative when expressed in an arbitrary gauge, the components of the tensor Uαβ

247
can be written as
N
X −2 h i
U∆ = k 2−n An Φ̂(n) + En Γ̂(n) , (891)
n=0
N
X −2 h i
UΘ = k 1−n Bn Φ̂(n) + Fn Γ̂(n) , (892)
n=0
N
X −1 h i
UP = k 2−n Cn Φ̂(n) + In Γ̂(n) , (893)
n=0
N
X −1 h i
UΣ = k −n Dn Φ̂(n) + Jn Γ̂(n) , (894)
n=0
n
d
where Φ̂(n) ≡ dτ n Φ̂, and similarly for Γ̂. The coefficients An -Jn depend on time and

scale through the scale factor, a, and wavenumber, k. For the sake of brevity we will
refrain from explicitly stating these dependences.
Now, although we have defined U above in the Newtonian gauge, the individual terms
appearing in the expressions above are all gauge invariant. This fact, however, imposes
further constraints because the gauge-invariant variable Γ̂ contains second derivatives of
the scale factor, when expressed in an arbitrary gauge. Hence, to avoid higher derivatives
of the background appearing in the field equations, we have to set

EN −2 = FN −2 = IN −1 = JN −1 = 0. (895)

We assume that the evolution equations for the matter fields remain unchanged, and that
these equations are supplemented by additional evolution equations for the extra gravi-
tational fields. Finally, the field equations are closed by imposing the Bianchi identities.
This imposes one of two possible options: (i) ∇α Tαβ E
= ∇α Uαβ = 0, which imposes a
series of constraints such that the theory remains consistent, or (ii) ∇α (TαβE
+ Uαβ ) =
0, which is the more general situation. For a detailed discussion of these issues see
[1149, 1151, 77].

6.1.1. Evolution of perturbations on super-horizon scales

In principle it may seem that one should explore the long wavelength behaviour of
cosmological perturbations on a case by case basis. It turns out, however, that the
infinite wavelength mode can be studied simply by considering the evolution of the back-
ground equations. This observation is due to Bertschinger [148], and proceeds as follows.
Consider a background with scale factor a(τ, κ) where κ is the spatial curvature of a
hyper-surface of constant τ :
 
2 2 2 2 1 2 √
ds = a −dτ + dχ + sinhκ ( κχ)dΩ (896)
κ

where sinhκ (x) equals sin(x) for κ > 0, equals x for κ = 0, or equals sinh(x) for κ < 0.
We can now perturb this space-time as κ → κ(1 + δκ ) , and compensate by a change the
coordinates τ → τ + α and χ → χ(1 − 21 δκ ). Here δκ is a constant, while α = α(τ ). Note
that in this case the scale factor a(τ, κ) is perturbed as a(τ + α, κ(1 + δκ )). In words, the
248
scale factor will come out as the solution of some generalised set of Friedmann equations,
and will depend on the spatial curvature, κ.
We can now write this new geometry in the form of a perturbed FLRW metric, with
background curvature κ:
  
2 2 2 2 1 2 √
ds = a −(1 + 2Ψ)dτ + (1 − 2Φ) dχ + sinhκ ( κχ)dΩ , (897)
κ
where
∂ ln a
Ψ(τ ) = δκ + α0 + Hα, (898)
∂
 ln κ 
1 ∂ ln a
Φ(τ ) = − δκ − Hα. (899)
2 ∂ ln κ
One can now eliminate α to find a generic evolution equation that relates δκ with Φ and
Ψ without specifying any particular theory of gravity:
   
1 d a2 Φ 1 d  a  ∂ ln H
= Φ−Ψ+ − δκ , (900)
a2 dτ H 2a dτ H ∂ ln κ
where entropy perturbations have been neglected65 . The constant δκ that remains in the
equation above has a direct physical interpretation: It is twice the comoving curvature
perturbation. Note that we have also assumed that shear perturbations are negligible on
large scales, that local energy-momentum us conserved, and that spatial gradients can
be discarded. By choosing H(a, κ), and a relation between Φ and Ψ, one now has the
evolution equation completely defined. This is a powerful statement, as it means it is
possible to determine the evolution of large-scale perturbations without delving into the
details of the theory.
However, in order to complete the system here one still needs to specify a relation
between Φ and Ψ. As we will describe below, it has become standard practise to assume
the simplest form of the PPF parametrisation, that we described above, on super-horizon
scales. Within this approach comparisons have been made between choices of µ and ζ,
and the outcomes of numerical solutions for specific theories. These comparisons show
reasonable agreement, but, as yet, there is no compelling argument for applying the
simplest PPF parametrisation on large scales. In other words, there is no guarantee that
a simple relation of the form Ψ = ζΦ can encompass all possible theories of gravity. In
fact, it would appear that Eq. (900) does not allow such an interpretation.

6.1.2. The simplified PPF approach, and its extensions

Let us now revisit the simplest, and by far the most popular, version of the Param-
eterised Post Friedmannian approach [148, 237, 630, 627, 41]. Here one considers only
Eqs. (889) and (890). The further assumption that perturbations have no anisotropic
stress (i.e. ΣX = 0) then allows one to reorganise them in the form of Eqs. (882) and
(883). These equations can then be rewritten in terms of the Uαβ tensor by choosing
 
1−µ
U∆ + 3HUθ = 2k 2 Φ,
µ
UΣ = ζΦ.
65 Including entropy perturbations is straightforward, and for this the reader is referred to [148].
249
The coefficients Ai , Bi , Ci and Di must now be chosen to satisfy this condition.
It is understandable why this approach is popular, as it has a number of benefits.
Firstly, it is applicable in the quasi-static regime the arises when k → ∞ (i.e. on small
scales relative to the cosmological horizon). Furthermore, the perturbation equations
now form a closed system. This means that if, for example, we restrict ourselves to a
dust-filled universe then Eqs. (882) and (883), together with
0
δM = −k 2 θM + 3Φ0 (901)
0
θM = −HθM + Ψ, (902)
form a complete system of differential equations that can be straightforwardly solved.
It is instructive at this point to specify the various versions of the simplified PPF
formalism that are currently in use:
• Caldwell, Cooray and Melchiorri [237, 360] introduced the system66
µ = 1
ζ ≡ $ = $0 a3 , (903)
which was later extended to [361]
µ = 1 + µ0 a3 . (904)

• Bertschinger and Zukin [148, 151] proposed a reduced parametrisation that takes
into account the conservation of long wavelength curvature perturbations. Only
one parameter is considered67 :
γBZ = 1 − ζ, (905)
s
that is further refined as γBZ = 1 + βa , where β and s are constants. These
authors make the additional assumptions that γBZ depends only on time, even
on sub-horizon scales, and that the Bertschinger long-wavelength construction (see
Section 6.1.1) can also be extended to sub-horizon scales. The latter of these as-
sumptions allows them to solve for Φ, from which Ψ and δ are then determined. The
assumption of scale-independence is later relaxed further, such that the following
parameterisation can be made
1 + β1 k 2 as
γBZ =
1 + β2 k 2 as
1 + α1 k 2 as
GΦ = µ(1 − ζ) = G ,
1 + α2 k 2 as
68
where αi and βi are constants .

66 The initial convention was that $ = $0 ΩΩ0DE a3 , but the authors changed their convention in
0M
subsequent publications.
67 The symbol used was actually γ, to which we have added a subscript ”BZ” to distinguish it from

other parameters with the same name. A similar approach will be used with the other frame-works
presented here.
68 We kept the original symbol G here, although under our conventions it would be more accurately
Φ
written as GΨ , as it plays the role of an effective gravitational constant for a modified Poisson equation
for Ψ.
250
• Amendola, Kunz and Sapone [41] modify the Poisson equation and the slip by
considering the two functions QA and ηA :

QA = µ,
ηA = −ζ, (906)

while a third variable, ΣA = QA (1 + ηA /2), is also introduced in order to simplify

the relevant expression for gravitational lensing: Φ + Ψ = 2ΣA Φ. A comparison of
this approach is made with DGP and scalar-tensor theories, both on sub-horizon
scales.
• Zhang et al. [1301] make the parameterisation
1
ηZL = ,
1−ζ
G̃ef f = µ(1 − ζ/2). (907)

These authors also introduce the EG statistic, which we describe further below.
• Zhao et al. [1302] modify the Poisson equation using Ψ. They introduce the two
parameters
1
ηZG = ,
1−ζ
µZG = µ(1 − ζ), (908)

as well as a third derived parameter ΣZG = µZG (1 + ηZG )/2 = ΣA . Further

refinements are made by considering two specific models: A case where µZG =
µZG (τ ) and ηZG = ηZG (τ ) transit from their GR values in the early universe to
modified constant values today, and a case where the two functions are pixelised
in the {τ, k} plane by 2 × 2 pixels per function.
• A parametrisation specifically designed for small scales was proposed by Amin,
Blandford and Wagoner [44], where

BA (Hk , a) = (1 − ζ)µ = β0 (a) + β1 (a)Hk + β2 (a)Hk2 + . . .

ΓA (Hk , a) = µ = γ0 (a) + γ1 (a)Hk + γ2 (a)Hk2 + . . . , (909)

while the matter density fluctuation is likewise parametrised by

 
δm (a, k) = δ(k)i δ0 (a) + δ1 (a)Hk + δ2 (a)Hk2 , (910)

where δ(k)i is specified by initial conditions. In this approach different theories

correspond to different sets of the functions {βi , γi , δi }. The authors find the ap-
propriate functions for ΛCDM, scalar-tensor theories, quintessence models, f (R)
theories, and DGP. They also note, however, that not every theory can be ade-
quately matched to this expansion (e.g. k-essence).

251
 The simplified PPF approach, though useful as a phenomenological tool that can
be used to constrain µ and ζ, is not without its problems. For a start, it is not clear
what theories a specific choice of parameters actually encompasses. In [1149, 501, 77]
this problem has been addressed in detail, and it is shown that if one wishes to consider
theories with second order field equations only then the PPF equations become:
X
− 2k 2 Φ = 8πµGa2 ρX [δX + 3(1 + wX )HθX ] (911)
X
g
Φ − Ψ = ζΦ + Φ0 , (912)
k
with the constraint µ = 1 − kg H. In other words, if the field equations of the theory are
second order then the simplified PPF approach is not applicable.
Let us now consider what the simplified PPF approach does correspond to in the
field equations. Taking the expansion for Uαβ , and considering terms up to the lowest
acceptable order, we have:
U∆ = A0 k 2 Φ + A1 kΦ0 + E0 k 2 Γ,
UΘ = B0 kΦ + B1 Φ0 + F0 kΓ,
UP = C0 k 2 Φ + C1 kΦ0 + C2 Φ00 + I0 k 2 Γ + I1 kΓ0 ,
UΣ = D0 Φ.
Applying the Bianchi identities one can then determine the above functions in terms of
ζ = D0 and g̃ = µ1 − 1. This gives F0 = E0 = I0 = I1 = 0 (i.e. no Γ terms appearing),
and
0
 
2 g̃ + H(g̃ + ζ) 1 0
kB0 = 2k , A0 = 2g̃ − 3Hk B0 , C0 = 2ζ + 3 B + 2Hk B0 ,
3H0 − 3H2 − k 2 k 0
g̃
B1 = 2k 2 , A1 = −3Hk B1 , kC1 = 3 (B10 + 2HB1 + kB0 ) .
3H0 − 3H2 − k 2
Since A1 + 3Hk B1 = E0 = F0 = 0, the form of the Poisson Equation (890) is retained.
However, since A1 , B1 and C2 do not vanish, the field equations contain third time
derivatives in arbitrary gauge. This is due to the presence of Φ00 , and corresponds to
a higher-order gravitational theory69 . In other words, one to simplify the equations for
the gravitational slip, but only at the cost of introducing higher-order terms in the other
evolution equations (to enforce consistency).
The above way of reconstructing the U ’s from the simplified approach is not unique.
One can also relax the condition ∇α Tαβ E
= ∇α Uαβ = 0, but this moves us further away
from the simplified approach.
A further problem of the simplified approach is that it is impossible to determine
µ and ζ for a specific theory without solving the field equations for a specific choice of
initial conditions. What we would like to have is a one-to-one correspondence between
the functions that appear in the parametrised frame-work and those that appear in the
theories themselves, without having to solve the field equations every time (just as in the
PPN formalism, discussed in Section 2). This can be achieved by constructing the field
equations as in Section 6.1.
69 Since the form of the Poisson equation is retained, however, both Φ and Ψ remain non-dynamical.

Hence, the higher derivatives do not introduce additional propagating degrees of freedom.
252
6.1.3. The Hu-Sawicki frame-work
Hu and Sawicki introduced a frame-work [630], later generalised by Hu [627], that
depends on a function of time and space, gHS (τ, k), two functions of time, fζ (τ ) and
fG (τ ), and a constant, cΓ . Their frame-work goes beyond the simplified PPF frame-
works described above, and tends to them in either the super-horizon or sub-horizon
limits of the late Universe70 . The field equations in this case are written as

8πGa2 X X
Φ+Ψ = − 2
ρX [δX + 3H(1 + wX )θX ] − 8πGa2 (ρX + PX )ΣX + 2ΓHS ,
cκ k
X X
X
Φ−Ψ = 8πGa2 (ρX + PX )ΣX + gHS (Φ + Ψ) , (913)
X

3κ
where cκ = 1 − k2 .
The variable ΓHS is obtained by solving the differential equation
   
cΓ cΓ k 1
2(1 + gHS ) 1 + 2 Γ0HS + HΓHS + ΓHS + fG (Φ + Ψ)
Hk Hk 2
X 0 
0
= (2HgHS − gHS )(Φ + Ψ) + 8πGa2 gHS ((ρX + PX )ΣX ) + H(ρX + PX )ΣX
X
8πGa2 X 8πGa2
− 2
[gHS (1 + fζ ) + fG ] (ρX + PX )θX + (ρE + PE )θT , (914)
k k2
X
P
(ρ +P )θ
where θT = P X X X X
is calculated in conformal Newtonian gauge71 . In both Eq.
X ρX +PX
(913) and Eq. (914) ρX and PX do not include contributions from the dark energy.
Taking the sub-horizon limit, for which Hk → 0, we get that ΓHS → − 21 fG (Φ + Ψ), and
the system then reduces to the simplified frame-work with
1 + gHS
µ = ,
1 + fG
2gHS
ζ = . (915)
1 + gHS

In the super-horizon limit the system obeys the Bertschinger construction (see Section
6.1.1), with gHS = gHS (τ ) while fζ (τ ) provides the next order correction (beyond the
Bertschinger construction). We see that the sub-horizon limit depends only on gHS and
fG , while the super-horizon limit depends only on gHS and fζ . The constant cΓ controls
the transition scale between these two limits.
As in the simplified approaches, it is not clear from the outset what kind of theo-
ries this frame-work encompasses. One can find fitting functions gHS , fG and fζ that
reproduce the solutions for specific theories, but only after some experimentation.

70 At late times the matter content of the Universe is effectively described by dust and dark energy,

for which the dust has no anisotropic stress.

71 In the original formulation the comoving gauge was used, but we translated the equations here to

make it more consistent with the other approaches we described.

253
6.2. Models for µ and ζ on Sub-Horizon Scales
On sub-horizon scales the situation is greatly simplified. In this case one can make a
quasi-static approximation, and discard all time variations in the perturbed fields. Let
us now consider this regime in a few different models.
To start with we can consider f (R) theories. Following [1043] we then have

Ga2 ρM
− k 2 (Φ + Ψ) = 4π δM ,
fR
2fRR k 2
Ψ−Φ = − fR (Φ + Ψ),
3fRR k 2 + a2 fR
which can be rewritten in the µ-ζ form given above. Note that there is a scale in these
equations, the “Compton wavelength” given by

k 2 fRR
Q≡3 .
a2 fR
For large wavelengths Q → 0, and Ψ − Φ → 0. For small wavelengths Q  1, and
Ψ − Φ ' −(2/3)fR (Φ + Ψ).
Let us now consider DGP models. In this case the expression for gravitational slip
can be found to be [737]
1
µ = 1− ,
3β
3β − 2
ζ = , (916)
3β − 1

where β = 1 + 2Hrc wE (for more details see Section 5.5, and in particular Eqs. (812)
and (813)).
Finally, let us consider scalar-tensor theories. For a generalised coupling parameter,
F (φ), it can be shown that [41]
0
G∗ 2(F + F 2 )
µ =
F Gcav 2F + 3F 0 2
0
F 2
ζ =
F + F 02
The expression above can be very useful in studying the evolution of structure on
observable scales. They can, for example, be used to correctly reproduce the growth rate
of structure, and the effects of lensing, on scales of up to hundreds of Mpc.

6.2.1. The importance of shear

An important issue arises if one wishes to distinguish between the effects of dark
energy and the effects of modified gravity. In [759] the authors constructed an explicit
example of how anisotropic stress could mimic the effect of gravitational slip. To see
this let us consider Eq. (889) with UΣ = D0 Φ. There is clearly a degeneracy between
the two terms on the left hand side. In particular, if we focus on DGP then we have
254
that Σ = 0, but UΣ = 3β−2 72
3β−1 Φ . However, we can of course consider an alternative
model where gravity is unmodified, but the dark energy has anisotropic stress given by
(1 + wE )ΣE = − 2k12 3β−2
3β−1 δM . This would be an odd form of dark energy, but it could
also explain observations that may otherwise have been attributed to gravitational slip.
Distinguishing between such models is, of course, made easier when considering a wide
variety of different scales.

6.2.2. The growth function

The primary effect of modified gravity will be on the growth of structure. The time
evolution of the density field can be a sensitive probe of not only the expansion rate
of the Universe but also its matter content. In a flat, matter dominated universe we
have that δM , the density contrast of matter, evolves as δM ∝ a. We can parametrise
deviations from this behaviour in terms of the growth function, f , given by
0
δM
f≡ , (917)
HδM

or by the introduction of a parameter γ through [1031, 813, 783, 784]

f = ΩγM . (918)

Note that γ and f are not parameters in the usual sense, but are derived quantities (indi-
rect observables, to some extent). It is, however, sometimes convenient to parameterise
some results or processes in terms of these quantities.
For standard growth in the presence of a cosmological constant one has γ ≈ 6/11 to
a very good approximation, although this can change with ΩM . For the case of General
Relativity with a dark energy component with equation of state P = wρ (where w is
constant) we have

γ = γ0 + γ1 ΩE + O(Ω2E ) (919)

where ΩE = 1 − ΩM , and where γ0 and γ1 are given by

3(1 − w)
γ0 = ,
5 − 6w
 
γ0 6w2 − 7w + 2 2 − 3w
γ1 = − γ0 . (920)
2w 5 − 12w 3

In the case of a cosmological constant (i.e. with w = −1) these expressions reduce to
6 15
γ= + ΩE + . . . ,
11 2057
which gives the first order correction to the expression γ ≈ 6/11, given above.

72 Actually, as discussed in Section 5.5, DGP has shear but the quasi-static limit imposes the condition

µE = 2k2 ΣE between energy density associated with the perturbation of the Weyl tensor and the shear.
This in turn generates the term UΣ above.

255
 A natural question to ask is how the growth parameter, γ, depends on modifications
of gravity. There have been a number of attempts at finding analytic expressions that
relate γ to parameters in the underlying theory. This has focused on specific theories, as
well as the extended PPF approach. We now discuss some of the results.
An expression for the growth parameter in the quasi-static limit of f (R) theories was
found in [58]. Here one can define a time dependent mass scale:
1
M 2 (a) = ¯ ,
3fRR

where f¯RR is the second functional derivative of f with respect to the Ricci scalar, R,
evaluated at the General Relativistic value of R. A non-local expression for γ can then
be found of the form
Z
6 ΩΛ k 2 −11/2 a ã3/2 dã
γ= − a 2 2 2
.
11 2ΩM 0 k + ã M (ã)

One can see that for k  aM (a) this expression for γ remains the same as in General
Relativity, but that once k crosses the mass threshold modifications start to kick-in.
In DGP models the correct expression for γ is given by73 [501]
11 7 93 2
γ= + ΩE − Ω + O(Ω3E ). (921)
16 5632 4096 E
This result is in excellent agreement with numerical studies (to within 2% or better for
ΩE < 0.8, and to 5% for ΩE < 0.9).
An attempt at finding approximate analytic expressions for γ was presented in [501].
In this report we will restrict ourselves to limiting cases where ζ ' ζ1 ΩE . If we then
assume that we can Taylor expand γ as in Eq. (919) then we find the coefficients

3(1 − w + ζ1 )
γ0 = , (922)
5 − 6w
3w  
γ1 = −3Y1 − (2 − 3w)Y12 + 4Y2 , (923)
2
where we have defined
1 − w + ζ1
Y1 ≡ ,
w(5 − 6w)
(1 − w)(15w2 − 4w − 1) + ζ1 (9w2 + 2w − 2) ζ12
Y2 ≡ − .
2w2 (12w − 5)(5 − 6w) 2w2 (12w − 5)(5 − 6w)

A comparison between the approximation described above and the exact numerical value
of γ is shown in Figure 14. More general expression for γ that include the impact of µ,
varying w, higher-order terms in ΩE , as well as scale dependent correction in Hk are
presented in [501].

73 Incorrect values for γ in DGP models have been presented in [816, 41, 1263].

256
Figure 14: The growth parameter, γ, for a selection of gravitational slip parameters of the form ζ = ζ1 ΩE ,
as a function of ΩM . The dashed curves are the numerical results for ζ1 = 0, 0.2, 0.4 and 0.6 in ascending
order, and the corresponding analytic approximations are plotted as solid lines.

6.2.3. Current constraints on the PPF parameters

At the time of writing this report, there have only been a few attempts at constraining
the PPF parameters using existing cosmological data sets [123, 361, 1302]. Let us now
consider each of these in turn.
In [123] the authors find constraints on Q = µ and R = 1 − ζ using the Union
set of Supernovae from the Supernovae Cosmology Project (with an additional sample),
the joint 2dF and DR7 SDSS survey estimate of the Baryon Acoustic Oscillations, the
WMAP 7-year CMB Temperature and Polarisation data, the matter power spectrum
from the SDSS DR7 release, auto and cross-correlation functions from 2MASS and the
SDSS LRG catalogue with ISW and the COSMOS weak lensing survey.
A scale and time dependent form of the PPF parameters is proposed:

Q(k, a) = 1 + [Q0 ek/kc + Q∞ (1 − ek/kc ) − 1]as ,

R(k, a) = 1 + [R0 ek/kc + R∞ (1 − ek/kc ) − 1]as .

Setting kc → ∞ and s = 0 the authors then find time independent constraints on the
parameters: 0.97 < Q < 1.01 and 0.99 < R < 1.02 (both at the 95% confidence level).
Their analysis shows that primary constraints come from the WMAP7 data, and that
current weak lensing and cross correlations between galaxies and ISW have a minimal
effect. Allowing time dependence greatly relaxes the constraints on the parameters:
1.04 < Q < 2.66 and −0.22 < R < 1.44 (both at the 95% confidence level). If a
transition scale of kc = 0.01 Mpc−1 is considered, parameters are then constrained to be
0.47 < Q0 < 3.49, −0.80 < R0 < 2.52, 0.97 < Q∞ < 2.65, and −0.43 < R∞ < 1.76.
The analysis of Daniel et al [361] extends the above analysis to include an incarnation
of the CFHTLS lensing survey, which has substantially larger coverage than that of the
COSMOS survey. An earlier release of WMAP data is used, and cross correlations
257
between large scale structure and ISW are not included. The authors focus on scale-
independent parameters, ζ = ω̃ = ω̃0 a3 and µ = µ0 a3 , and find −1.4 < ω̃0 < 2.8 and
−0.67 < µ0 < 2.0. The authors also consider a more elaborate time evolution of both ω̃
and µ by dividing it up into bins in the following redshift ranges: (a) [0, 1] , (b) [1, 2] and
(c) [2, 9]. Setting ω̃ = 0 they then find that µ is constrained to be: −0.074 < µ0a < 0.08,
−0.058 < µ0b < 0.14, −0.023 < µ0c < 0.22. Setting µ = 1 they find that ω̃ is constrained
to be: −0.074 < ω̃0a < 0.08, −0.058 < ω̃0b < 0.14, −0.023 < ω̃0c < 0.22.
In [1302] the authors consider two different parameterisations. In the first they con-
sider a scale independent but time evolving parametrisation for ηZG and µZG , given by
Eq. (908) as
 
1 − µ0 z − zs
µZG (z) = 1 + tanh + µ0 ,
2 ∆z
 
1 − η0 z − zs
ηZG (z) = 1 + tanh + η0 .
2 ∆z

These authors use essentially the same data sets as in [123], albeit with an earlier release
of the WMAP data (5 year) and without BAO constraints. They find that if zs = 1
then the constraints are 0.65 < µ0 < 1.9 and −0.41 < η0 < 2.18, while if zs = 1 the
constraints are 0.68 < µ0 < 1.11 and 0.7 < η0 < 1.9.
In a second parametrisation the authors bin the parameters in both time and scale.
They choose not to bin in η, but rather in ΣZG = µZG (1+η 2
ZG )
. This is the combination
that enters into the calculation of both the ISW effect and weak lensing through their
dependence on Φ + Ψ. They consider the following bins: (a) k ∈ [0, 0.1], z ∈ [1, 2], (b)
k ∈ [0, 0.1], z ∈ [0, 1], (c) k ∈ [0.1, 0.2], z ∈ [1, 1], and (d) k ∈ [0.1, 0.2], z ∈ [1, 2]. The
resulting constraints are then as follows: In bin (a) 0.96 < Σ1 < 1.04, 0.64 < µ1 < 1.42,
in bin (b) 0.93 < Σ2 < 1.07, 0.65 < µ2 < 1.34, in bin (c) 0.58 < Σ3 < 1.02, 0.24 < µ3 <
2.24, in bin (d) 0 < Σ4 < 2.23, 0.05 < µ4 < 2.46.

6.2.4. Constraining the growth rate

Over the past few years there have been attempts to target the growth rate directly
using redshift space distortions. This method involves measuring either the redshift
space correlation function, ξ s (r|| , r⊥ ) (where the r|| and r⊥ correspond to parallel or
perpendicular directions to the line of sight), or the redshift space power spectrum,
Pgs (k). The power spectrum can be related to the real space power spectrum through an
extension of what is known as the ‘Kaiser formula’:
 2 2 2
s 2 2 4 2 k µ σv
P (k) = P (k)[1 + 2µ β + µ β ]G , (924)
H 2 (z)

where β = fb , b is the bias factor, G(x) encodes the non-linear effect due to velocity
dispersion, and µ is the cosine of the k vector with the line of sight. Note that P s (k)
becomes anisotropic (as does the correlation function) and it is through this anisotropy
that one can measure β and hence f .
Until recently, measurements of β were seen as constraints on ΩM . The reason for
6/11
this is that in ΛCDM we have β ' ΩM (z)/b, and hence measurements of β at different
redshifts can be used to reconstruct the history of ΩM . In [585] it is found that the wide
258
part of the VIMOS-VLT Deep Survey (VVDS) can be used to obtain β = 0.70 ± 0.26
at z ' 0.8. This result is then combined with the constraint from the 2dFGRS, β =
0.49 ± 0.09 at z ' 0.15, and the 2dF-SDSS LRG and QSO (2SLAQ) constraint (shown in
Figure 15). The emphasis in [585] was on finding deviations from growth rate in ΛCDM ,
and although not done systematically, their analysis showed that a number of specific
models could be ruled out.

Figure 15: The growth rate, f , as a function of redshift, z, for a number of different models. The symbols
and error bars correspond to constraints from VVDS, 2dFGRS and WiggleZ. Taken from [165].

The analysis of [585] have led to a number of upcoming redshift surveys focusing,
in part, on redshift space distortions. In particular, WiggleZ, FMOS, VIPERS, GAMA
and BOSS, many of whom were primarily targeting large scale structure and the Baryon
Acoustic Oscillations, will now all deliver constraints on β in the near to medium future.
In [1159] it was argued that measurements of the growth of structure through con-
straints on the large scale peculiar velocity dispersion, σv2 (the large-scale redshift space
distortion), can be combined with weak lensing measurements to break the degeneracies,
and target µ more accurately. With measurements at different redshifts it should be
possible to reconstruct the time evolution of µ.

6.2.5. The EG diagnostic

The effect of modified gravity on the gravitational potentials, Φ and Ψ, can in principle
be teased out with a careful combination of measurements. In [1301] it was argued that
this should be possible using a cross-correlation between an estimate of the local velocity
field from redshift space distortions, and the gravitational potentials inferred from a
lensing map from background galaxies. More specifically, from a more general form of

259
Eq. (924),
 
k 2 µ2 σv2
P s (k) = P s (k)[Pg (k) + 2µ2 Pgθ (k) + µ4 Pθ (k)]G , (925)
H 2 (z)

it is possible to estimate the galaxy-velocity power spectrum, Pgθ , in bandpowers Pα ,

and the galaxy-galaxy power spectrum, Pg , in bandpowers Pi . From a weak lensing
measurement of the convergence
Z
1 χs 2
κ= ∇ (Ψ − Φ)W (χ, χs )dχ,
2 0

where W is the lensing kernel, it is possible to construct a cross power spectrum with
P (2) (2)
galaxies, Cκg (`) ' α fα (`)Pα , where Pα is the band power estimate of P∇2 (Ψ−Φ)g .
These band powers can all be collected into one estimator,

Cκg (`, ∆`)

ÊG ≡ P ,
3H02 a−1 α fα (`, ∆` )Pα

where a band averaging over bins of width ∆` has been assumed. It can be shown that
 2 
∇ (Ψ − Φ)
hÊG i = . (926)
3H02 a−1 βδ k=`/χ̄

The diagnostic EG will take different values depending on the theory of gravity:
EG = Ω0 /β in ΛCDM and DGP (with different Ω0 in either case), and EG = Ω0 /(1+fR )β
in F (R). In TeVeS EG is significantly different from the ΛCDM value, and is scale
dependent. Hence, EG is in principle a good diagnostic of the underlying gravitational
theory.
In [1060], the authors attempted to extract an estimate of EG from a sample of 70,205
luminous red galaxies (LRGs) from the SDSS. Their estimate was concentrated at a mean
redshift of z ' 0.32 and they found 8 estimates of EG across a range of between 2h−1 to
50h−1 Mpc, with a mean of EG = 0.40 ± 0.07. This should be compared with the ΛCDM
prediction of EG = 0.408 ± 0.029, the F (R) prediction of EG = 0.328 − 0.365, and an
approximate TeVeS prediction of EG ' 0.22. The authors of [1060] have argued that this
is evidence for the validity of General Relativity on cosmological scales. The results from
[1060] are somewhat tentative and preliminary, but nonetheless promising. A judicious
choice of cosmological parameters may indeed be able to tease out the particularities of
how Φ and Ψ evolve in different theories of gravity. With planned mega surveys such
LAMOST, LSST and SKA it is likely that very tight constraints on EG will be achievable
in the future.

6.3. Forecasting Constraints from Future Surveys

There is still quite a way to go in terms of constraining deviations from General
Relativity. As mentioned in the previous section, with the planned mega surveys it
should be possible to reconstruct the time evolution of µ and ζ with ever increasing
precision [431, 432]. With the projected quality of the data, it also makes sense to
consider model independent parameterisations of µ and ζ using some form of Principal
260
Component Analysis (PCA). The authors in [1304] have done just this, and attempted to
quantify how many eigen-modes in an expansion of µ and ζ can be accurately constrained
with surveys such as Planck, DES, LSST, SNAP and JDEM. They find that while with
DES it should be possible to constrain up to 20 independent eigen-modes, LSST will
improve this by a factor of 5, allowing detailed reconstruction of the time evolution in
modified gravity. They find that in the best case scenario LSST will accurately constrain
the time and scale dependence of µ and ζ in the range 0.5 < z < 2 and 0.04 < k0.16 h−1
Mpc.

261
7. Discussion

In this review we have discussed recent advances in gravitational physics, many of

which have been driven by the discovery of the dark side of the Universe in the late 1990s.
In this final section let us now briefly consider the outlook for future gravity research,
and, in particular, what we consider to be some of the upcoming highlights.
From a theoretical perspective, model building is an important part of understanding
and explaining existing data, as well as making predictions for the future. Tight con-
straints already available on solar system and astrophysical scales, however, mean that
model builders are presented with a choice: They can either study minimal deviations
away from General Relativity, or must otherwise look for mechanisms that hide modifi-
cations to gravity on the scales probed by experiment. The former of these has value for
understanding the special nature of General Relativity, and the consequences of moving
away from it, while the latter provides an exciting opportunity to try and solve some of
the cosmological puzzles that have arisen with the discovery of dark matter and energy.
Modified gravity necessarily involves additional fields, extra dimensions, or broken
symmetries, since we know that GR is the unique diffeomorphism invariant theory of a
single rank-2 tensor that can be constructed from the metric variation of an action in
four dimensions. An important consideration is then whether or not these deviations
manifest themselves at the level of the background cosmology, or merely at the level of
perturbations. Of course, if we wish to account for dark energy, or solve the cosmo-
logical constant problem using modified gravity, these deviations must be manifest in
the solutions of the Friedmann equations. We must also require, however, that they do
not spoil the successful predictions of the standard cosmology, such as the abundance
of light elements, the peak positions of the CMB acoustic spectrum, or the predictions
for baryon acoustic oscillations. This requires the background FLRW cosmology of the
modified theories to closely mimic the standard evolution of ΛCDM from nucleosynthesis
through to matter domination.
Let us now consider the assumptions that go into the standard cosmology. These
include [149]
• Einstein’s field equations.
• The universality of free fall.
• Local Lorentz invariance.
• Three spatial dimensions (below the electro-weak scale, at least).
• Conservation of energy-momentum.
• Homogeneity and isotropy of space on large scales.
• Matter fields being well modelled by fluids of dust and radiation.
These assumptions, when confronted with observations, require the presence of dark
matter and dark energy. From a particle physics perspective this is very discomforting, as
we have discussed in Section 2.6.6, as it requires the vacuum energy to be at least 60 orders
of magnitude smaller than expected from particle theory. This problem is exacerbated
by the fact that contributions from the zero-point energy of massive particles changes as
262
the universe cools and goes through phase transitions. To circumvent the difficulties that
this entails, many model builders therefore assume some unknown symmetry that sets
the vacuum energy, plus any contribution from a bare cosmological constant, to zero.
To account for cosmic acceleration they then require modifications to gravity on large
scales74 .
In some respects addressing the puzzles in cosmology is the easy part of the model
builders job. More difficult is to do this in a way that is consistent with the well estab-
lished observations of relativistic gravitational phenomena. This often involves identify-
ing a screening mechanism that can work on solar system scales. The point here is that
in order to modify the cosmological evolution we have to deviate considerably from GR
in the IR, but IR modifications of gravity are known to cause strong deviations from the
predictions of General Relativity for experiments that involve, for example, the bending
or time delay of light. Deviations on the scales that these observations are made must
therefore be suppressed. Currently the most popular examples of this type of mechanism
are the Vainshtein mechanism and the chameleon mechanism, both of which we have
discussed in this review, and neither of these is without its problems. It is therefore
important for theorists to identify new ways to screen the deviations from GR at short
distances, and we expect this to be an important avenue of future research.
It is also crucial to ensure that our models obey some basic requirements of stability.
There is little motivation for carrying out complicated simulations, and devising expen-
sive experiments, to probe theories that are fundamentally sick. For example, we may
choose to ask if the theory contains ghosts, which can lead to catastrophic instabilities
unless the mass of the ghost lies above the effective theory cut-off, as described in Section
2.1.3. Ghosts are rife in modified theories of gravity, and dealing with them is one of
the principle challenges faced by model builders. Another important thing to establish
is the cut-off for our effective classical theory. Modified theories of gravity can become
strongly coupled at lower than expected energies, and enter a quantum fog. If we want
an effective theory description of gravitational fields at the surface of the earth to be
classical, we need to impose that the cut-off is at least of the order of a few meV , since
classical gravity has been tested in the lab at this scale. Conservatively, one may also
wish to impose that the classical description of gravitational fields around the Sun should
be trustworthy, at least down to its Schwarzschild radius.
Let us now consider the prospect of future input into this field from experiments and
observations. Tests of Lorentz invariance and the weak equivalence principle, that are
already well established on Earth, are now being proposed and planned as space mis-
sions. Space offers a number of benefits over Earth-based experiments, including a lack
of seismic noise, and the fact that cooled atoms stay in interferometers longer. There are
also theoretical reasons for wanting to test these foundational issues of gravity in space,
such as the proposal by Khoury and Weltman that extra gravitational degrees of free-
dom could have an environmental dependence. Space based tests of Lorentz invariance
offer the possibility of improving constraints on violations of this symmetry by orders of
magnitudes (see, e.g. [172, 869]). Space based tests of the equivalence are being planned
by the French space agency CNES, under the name MICROSCOPE (MICRO-Satellite
à trainée Compensée pour l’Observation du Principe d’Équivalence), and by ESA and

74 Some approaches, such as degravitation, (see Section 5.6.2) even use modifications to gravity in order

to screen the effects of the large vacuum energy, thereby removing the need for the unknown symmetry.
263
NASA, under the name STEP (Satellite Test of the Equivalence Principle). These two
missions promise to increase bounds on violations of the weak equivalence principle to
the level of 1 part in 1015 , and 1 part in 1018 , respectively.
With regards to space based tests of metric theories of gravity, there is also hope
for further improvement on current bounds. The Bepi-Columbo Mercury orbiter being
planned by ESA will, after a two year mission, be capable of placing constraints on PPN
parameters of the order γ − 1 ∼ 3 × 10−5 , β − 1 ∼ 3 × 10−4 , and α1 ∼ 10−5 [894]. The
bounds on γ could be improved further by Gaia, a high precision space telescope that
could constrain γ − 1 to around 1 part in 106 . Still greater constraints may be possible
with LATOR (Laser Astronomic Test of Relativity) [1230]. This mission consists of
two satellites that orbit the Sun at 1AU, and will have the potential of being able to
constrain γ − 1 to around 1 part in 108 , and solar frame-dragging to the level of ∼ 1%.
Such constraints are orders of magnitude greater than those that are currently available.
Lunar laser ranging has played an important part in testing gravity over the past
4 decades, since retroreflectors were placed on the moon by Apollo astronauts, and
the Soviet Lunokhod rovers [885]. Improvements in ground based technology during
this time have improved the bounds on PPN parameters that reflected lasers have been
able to impose, but we are now reaching the stage where further gains will be limited
by the retroreflectors themselves. Tightening the current bounds using this technology
will therefore require new space missions, and, in particular, the possibility of planting
retroreflectors on other planets has great potential. Laser ranging of Mars over a 10
year period would allow γ − 1 to be constrained to around 1 part in 106 [1231], and the
Nordtvedt parameter η = 4β − γ − 3 to the level of ∼ 6-2 × 10−6 [47]. Again, these are
order of magnitude improvements on current bounds.
Moving beyond the solar system, binary pulsars are an excellent test of relativistic
gravity, and offer the possibility of becoming more constraining than solar system test in
the near future [1177]. Pulsars also offer the opportunity to test gravity through the emis-
sion of gravitational waves. In particular, pulsar-white dwarf systems have great potential
to constrain the emission of dipolar gravitational, which is a generic prediction of a large
number of modified theories of gravity. Binary systems PSR J1141-6545, J0751+1807
and J1757-5322 are all recently discovered pulsar-white dwarf systems. What is more,
continued observation of existing pulsars also offer the possibility of new tests of gravity,
as, for example, the perihelion of PSR B1913+16 precesses it may soon allow for tests
of the Shapiro time-delay effect. Future prospects for testing gravitational physics using
pulsars are also bright due to the large numbers of these objects that are expected to be
found by the Parkes, Arecibo, and Green Bank telescopes, as well, of course, as the SKA
(Square Kilometre Array). The chance of detecting a pulsar-black hole systems increases
dramatically with large-scale observations of this kind. Such a system would be poten-
tially of great importance for testing strong field gravity. Finally, the double pulsar PSR
J0737-3039A,B also offers a unique test of gravitational physics, with excellent prospects
for improving constraints on gravity in the future as observations of it continue.
In all of these areas it is likely that the bounds on deviations from General Relativity
will continue to be tightened, with lab tests too promising continued improvement. The
Eöt-Wash group at the University of Washington, and others, continue to increase bounds
at ever smaller scales, and even particle experiments using the LHC at CERN are looking
for the signs of the extra-dimensions that are crucial for so many modern theories of
gravity. The future prospects for constraining all of these aspects of gravity means that
264
the extra degrees of freedom in modified theories of gravity will have ever smaller regions
of parameter space in which to hide. This also, of course, means that the possibility
of making a detection of a deviation from General Relativity is improved, if any such
deviations really do exist in nature.
It is, of course, the case that there have been tremendous developments in observa-
tional cosmology over the past couple of decades, and these observations have put gravity
in the spotlight once again. With measurements of the CMB, weak lensing, and galaxy
surveys, as well as probes of the expansion rate with distant supernovae IA s, a strange
Universe has been uncovered in which more than 95% of the energy budget is in some
exotic dark form. The quality of these observations are such that it is difficult to avoid
such a conclusion if the gravitational force arises from Einstein’s theory of gravity. An
alternative point of view is that these observations are pointing at a flaw in our un-
derstanding of the behaviour of the Universe on the largest scales, and of behaviour of
gravity at these distances in particular. Indeed, these observations may be a sign that
we must think beyond Einstein’s General Theory of Relativity, and his field equations.
In the same way that cosmological observations may be hinting at new physics in the
gravitational sector, they can also be used to constrain and even rule out alternatives.
With a vast range of experiments being planned and constructed throughout the world,
it seems that we are a critical juncture in the path to understanding gravity. At ESA, the
satellite mission Euclid is under assessment. This mission could map out vast regions
of space, probing the growth rate and morphology of large-scale structure, both key
observables for constraining theories of gravity through their effects on gravitational
collapse. The SKA is in a planning phase with path finders being constructed on two
continents. SKA will be a vast radio telescope that will generate a survey of up to a
billion radio galaxies, mapping out the evolution of structure back to extremely high
redshifts. These observatories will also be competing against the Large Synoptic Survey
Telescope (LSST) that will image the sky over a period of ten years, building up a survey
of galaxies that will primarily be used for weak lensing.
These larger experiments will be complemented by smaller, more rapid surveys such
FastSound, Weave, Boss, Viper, KIDS, DES and a host of other collaborations that will
produce galaxy and weak lensing surveys on smaller scales, targeting different redshifts.
Interestingly enough, while the original scientific driver for many of these surveys was
to constrain dark energy, they have all taken on board the need to test gravity. Indeed
measuring the growth rate of gravitational collapse, as a test of modified gravity, has
become the core business in all of these surveys.
Einstein developed his General Theory of Relativity almost a century ago, and, al-
though it remains a cornerstone of modern physics, one could argue that of all the
fundamental forces of nature it is gravity that remains the least well understood. This
is almost certainly due to the weakness of the gravitational interaction, which makes
it incredibly difficult to test in the lab experimentally. Inevitably, experiments on the
scale of planets, stars, galaxies, and beyond cannot be performed with the same level of
precision and control as those conducted for the other forces on Earth. Never the less,
technology is now starting to catch up with gravity. The latter half of the twentieth
century may have belonged to the Standard Model of particle physics, but there is every
reason to suspect that the twenty first century will belong to gravity.

265
Acknowledgements

TC and PGF acknowledge the support of the STFC, CERN, the BIPAC, the Oxford
Martin School and Jesus College, Oxford. AP and CS are supported by Royal Society
University Research Fellowships. For comments, discussion and support we wish to thank
Tessa Baker, John Barrow, Cliff Burgess, Christos Charmousis, Ed Copeland, Kenny
Dalglish, Gregory Gabadadze, Nemanja Kaloper, Ian Kimpton, Kazuya Koyama, Ed
Macaulay, João Magueijo, Gustavo Niz, Claudia de Rham, Paul Saffin, Thomas Sotiriou,
Glenn Starkman, Reza Tavakol, Anzhong Wang, Shuang Yong Zhou, Tom Zlosnik, Joe
Zuntz, and Jessica Padilla and her Mum.

266
References

References
[1] R. B. Abbott, S. M. Barr, and S. D. Ellis. Kaluza-Klein Cosmologies and Inflation. Phys. Rev.,
D30:720, 1984.
[2] R. B. Abbott, B. Bednarz, and S. D. Ellis. Cosmological perturbations in Kaluza-Klein models.
Phys. Rev., D33:2147, 1986.
[3] R. B. Abbott, S. D. Ellis, and S. M. Barr. Kaluza-Klein Cosmologies and Inflation. 2. Phys.
Rev., D31:673, 1985.
[4] M. C. B. Abdalla, S. Nojiri, and S. D. Odintsov. Consistent modified gravity: dark energy,
acceleration and the absence of cosmic doomsday. Classical and Quantum Gravity, 22:L35–42,
Mar. 2005, arXiv:hep-th/0409177.
[5] M. Abdelwahab, S. Carloni, and P. K. S. Dunsby. Cosmological dynamics of ‘exponential gravity’.
Classical and Quantum Gravity, 25(13):135002, July 2008, 0706.1375.
[6] M. Abramowitz and I. A. Stegun. Handbook of Mathematical Functions with Formulas, Graphs,
and Mathematical Tables. Dover, New York, 1964.
[7] V. Acquaviva, C. Baccigalupi, S. M. Leach, A. R. Liddle, and F. Perrotta. Structure forma-
tion constraints on the Jordan-Brans-Dicke theory. Phys. Rev. D, 71(10):104025, May 2005,
arXiv:astro-ph/0412052.
[8] W. Adamowicz and A. Trautman. The equivalence principle for spin. Bull. Acad. Pol. Sci., Ser.
Sci. Math. Astron. Phys., 23: 339, 1975.
[9] A. Adams, N. Arkani-Hamed, S. Dubovsky, A. Nicolis, and R. Rattazzi. Causality, analyticity
and an IR obstruction to UV completion. JHEP, 0610:014, 2006, hep-th/0602178.
[10] E. G. Adelberger, J. H. Gundlach, B. R. Heckel, S. Hoedl, and S. Schlamminger. Torsion balance
experiments: A low-energy frontier of particle physics. Prog. Part. Nucl. Phys., 62:102–134, 2009.
[11] E. G. Adelberger, B. R. Heckel, and A. E. Nelson. Tests of the Gravitational Inverse-Square Law.
Annual Review of Nuclear and Particle Science, 53:77–121, Dec. 2003, arXiv:hep-ph/0307284.
[12] N. Agarwal, R. Bean, J. Khoury, and M. Trodden. Cascading Cosmology. Phys. Rev., D81:084020,
2010, 0912.3798.
[13] Y. Aghababaie, C. P. Burgess, S. L. Parameswaran, and F. Quevedo. Towards a naturally small
cosmological constant from branes in 6D supergravity. Nucl. Phys., B680:389–414, 2004, hep-
th/0304256.
[14] M. Ahlers, A. Lindner, A. Ringwald, L. Schrempp, and C. Weniger. Alpenglow: A signature
for chameleons in axionlike particle search experiments. Phys. Rev. D, 77(1):015018, Jan. 2008,
0710.1555.
[15] K. Akama. An early proposal of ’brane world’. Lect. Notes Phys., 176:267–271, 1982, hep-
th/0001113.
[16] M. Akbar and R. Cai. Friedmann equations of FRW universe in scalar tensor gravity, f(R) gravity
and first law of thermodynamics. Physics Letters B, 635:7–10, Mar. 2006, arXiv:hep-th/0602156.
[17] K. Akune, K.-i. Maeda, and N. Ohta. Inflation from superstring and M-theory compactification
with higher order corrections. - II. - Case of quartic Weyl terms. Phys. Rev. D, 73(10):103506,
May 2006, hep-th/0602242.
[18] A. Albrecht, C. Burgess, F. Ravndal, and C. Skordis. Exponentially large extra dimensions.
Phys.Rev., D65:123506, 2002, hep-th/0105261.
[19] A. Albrecht, C. Burgess, F. Ravndal, and C. Skordis. Natural quintessence and large extra
dimensions. Phys.Rev., D65:123507, 2002, astro-ph/0107573.
[20] J. Alcaniz. Some observational consequences of brane world cosmologies. Phys.Rev., D65:123514,
2002, astro-ph/0202492.
[21] S. Alexander, R. H. Brandenberger, and D. A. Easson. Brane gases in the early universe. Phys.
Rev., D62:103509, 2000, hep-th/0005212.
[22] S. O. Alexeyev and M. V. Pomazanov. Black hole solutions with dilatonic hair in higher curvature
gravity. Phys. Rev. D, 55:2110–2118, Feb. 1997, arXiv:hep-th/9605106.
[23] A. Ali, R. Gannouji, and M. Sami. Modified gravity a la Galileon: Late time cosmic acceleration
and observational constraints. Phys.Rev., D82:103015, 2010, 1008.1588.
[24] M. Alimohammadi and N. Agharafiei. Quantum attractors of generalized Gauss-Bonnet dark
energy. ArXiv e-prints, Dec. 2009, 0912.0589.
267
[25] M. Alimohammadi and A. Ghalee. Phase space of generalized Gauss-Bonnet dark energy.
Phys. Rev. D, 80(4):043006, Aug. 2009, 0908.1150.
[26] M. Alimohammadi and A. Ghalee. Remarks on generalized Gauss-Bonnet dark energy.
Phys. Rev. D, 79(6):063006, Mar. 2009, 0811.1286.
[27] R. Allahverdi, B. Dutta, and K. Sinha. Cladogenesis: Baryon-dark matter coincidence from
branchings in moduli decay. Phys. Rev. D, 83(8):083502, Apr 2011.
[28] G. Allemandi, A. Borowiec, and M. Francaviglia. Accelerated cosmological models in first-order
nonlinear gravity. Phys. Rev. D, 70(4):043524, Aug. 2004, arXiv:hep-th/0403264.
[29] G. Allemandi, A. Borowiec, M. Francaviglia, and S. D. Odintsov. Dark energy dominance and
cosmic acceleration in first-order formalism. Phys. Rev. D, 72(6):063505, Sept. 2005, arXiv:gr-
qc/0504057.
[30] G. Allemandi, M. Francaviglia, M. L. Ruggiero, and A. Tartaglia. Post-newtonian parameters
from alternative theories of gravity. General Relativity and Gravitation, 37:1891–1904, Nov. 2005,
arXiv:gr-qc/0506123.
[31] G. Allemandi and M. L. Ruggiero. Constraining extended theories of gravity using Solar System
tests. General Relativity and Gravitation, 39:1381–1388, Sept. 2007, arXiv:astro-ph/0610661.
[32] S. W. Allen, A. E. Evrard, and A. B. Mantz. Cosmological Parameters from Observations of
Galaxy Clusters. ArXiv e-prints, Mar. 2011, 1103.4829.
[33] E. Alvarez and M. Belen Gavela. Entropy from Extra Dimensions. Phys. Rev. Lett., 51:931, 1983.
[34] M. Amarzguioui, Ø. Elgarøy, D. F. Mota, and T. Multamäki. Cosmological constraints on
f(R) gravity theories within the Palatini approach. A&A, 454:707–714, Aug. 2006, arXiv:astro-
ph/0510519.
[35] L. Amendola. Dark energy and the boomerang data. Phys. Rev. Lett., 86(2):196–199, Jan 2001.
[36] L. Amendola, C. Charmousis, and S. C. Davis. Constraints on Gauss Bonnet gravity in dark
energy cosmologies. J. Cosmology Astropart. Phys., 12:20, Dec. 2006, arXiv:hep-th/0506137.
[37] L. Amendola, C. Charmousis, and S. C. Davis. Solar system constraints on Gauss Bonnet medi-
ated dark energy. J. Cosmology Astropart. Phys., 10:4, Oct. 2007, 0704.0175.
[38] L. Amendola, C. Charmousis, and S. C. Davis. Solar System Constraints on Gauss-Bonnet
Mediated Dark Energy. JCAP, 0710:004, 2007, 0704.0175.
[39] L. Amendola, K. Enqvist, and T. Koivisto. Unifying Einstein and Palatini gravities. Phys. Rev. D,
83(4):044016, 2011, 1010.4776.
[40] L. Amendola, R. Gannouji, D. Polarski, and S. Tsujikawa. Conditions for the cosmological
viability of f(R) dark energy models. Phys. Rev. D, 75(8):083504, Apr 2007.
[41] L. Amendola, M. Kunz, and D. Sapone. Measuring the dark side (with weak lensing). J. Cos-
mology Astropart. Phys., 4:13, Apr. 2008, 0704.2421.
[42] L. Amendola, A. B. Mayer, S. Capozziello, S. Gottlober, V. Muller, F. Occhionero, and
H. Schmidt. Generalized sixth-order gravity and inflation. Classical and Quantum Gravity,
10:L43–L47, May 1993.
[43] L. Amendola, D. Polarski, and S. Tsujikawa. Are f(R) Dark Energy Models Cosmologically
Viable? Physical Review Letters, 98(13):131302, Mar. 2007, arXiv:astro-ph/0603703.
[44] M. A. Amin, R. V. Wagoner, and R. D. Blandford. A subhorizon framework for probing the
relationship between the cosmological matter distribution and metric perturbations. MNRAS,
390:131–142, Oct. 2008, 0708.1793.
[45] K. N. Ananda, S. Carloni, and P. K. S. Dunsby. A characteristic signature of fourth order gravity.
ArXiv e-prints, Dec. 2008, 0812.2028.
[46] K. N. Ananda, S. Carloni, and P. K. S. Dunsby. Structure growth in f(R) theories of gravity with
a dust equation of state. Classical and Quantum Gravity, 26(23):235018, Dec. 2009, 0809.3673.
[47] J. D. Anderson, M. Gross, K. L. Nordtvedt, and S. G. Turyshev. The Solar Test of the Equivalence
Principle. ApJ, 459:365, Mar. 1996, arXiv:gr-qc/9510029.
[48] J. D. Anderson, M. A. Slade, R. F. Jurgens, E. L. Lau, X. X. Newhall, and E. Myles. Radar and
spacecraft ranging to Mercury between 1966 and 1988. Proceedings of the Astronomical Society
of Australia, 9:324, 1991.
[49] J. D. Anderson et al. in The Sixth Marcel Grossmann Meeting on general relativity, Kyoto,
Japan. 1992.
[50] M. Andrews, K. Hinterbichler, J. Khoury, and M. Trodden. Instabilities of Spherical Solutions
with Multiple Galileons and SO(N) Symmetry. 2010, 1008.4128.
[51] I. Antoniadis. Gravity modifications from extra dimensions. J. Phys., A40:6657–6667, 2007.
[52] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, and G. R. Dvali. New dimensions at a millimeter
to a Fermi and superstrings at a TeV. Phys. Lett., B436:257–263, 1998, hep-ph/9804398.
268
[53] I. Antoniadis, J. Rizos, and K. Tamvakis. Singularity-free cosmological solutions of the superstring
effective action. Nuclear Physics B, 415:497–514, Mar. 1994, arXiv:hep-th/9305025.
[54] T. Appelquist and A. Chodos. Quantum Effects in Kaluza-Klein Theories. Phys. Rev. Lett.,
50:141, 1983.
[55] T. Appelquist and A. Chodos. The Quantum Dynamics of Kaluza-Klein Theories. Phys. Rev.,
D28:772, 1983.
[56] S. A. Appleby and R. A. Battye. Do consistent f(R) models mimic General Relativity plus
Lambda? Phys.Lett., B654:7–12, 2007, 0705.3199.
[57] S. A. Appleby, R. A. Battye, and A. A. Starobinsky. Curing singularities in cosmological evolution
of f(R) gravity. J. Cosmology Astropart. Phys., 6:5, June 2010, 0909.1737.
[58] S. A. Appleby and J. Weller. Parameterizing scalar-tensor theories for cosmological probes. ArXiv
e-prints, Aug. 2010, 1008.2693.
[59] K. Arai, M. Hashimoto, and T. Fukui. Primordial nucleosynthesis in the Brans-Dicke theory with
a variable cosmological term. A&A, 179:17–22, June 1987.
[60] E. V. Arbuzova, and A. D. Dolgov. Explosive phenomena in modified gravity. Phys. Lett.,
B700:289–293, 2011, 1012.1963.
[61] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, and G. Gabadadze. Nonlocal modification of gravity
and the cosmological constant problem. 2002, hep-th/0209227.
[62] N. Arkani-Hamed, S. Dimopoulos, and G. R. Dvali. The hierarchy problem and new dimensions
at a millimeter. Phys. Lett., B429:263–272, 1998, hep-ph/9803315.
[63] N. Arkani-Hamed, S. Dimopoulos, and G. R. Dvali. Phenomenology, astrophysics and cosmol-
ogy of theories with sub-millimeter dimensions and TeV scale quantum gravity. Phys. Rev.,
D59:086004, 1999, hep-ph/9807344.
[64] N. Arkani-Hamed, H. Georgi, and M. D. Schwartz. Effective field theory for massive gravitons
and gravity in theory space. Annals Phys., 305:96–118, 2003, hep-th/0210184.
[65] C. Armendariz-Picon, T. Damour, and V. F. Mukhanov. k-Inflation. Phys. Lett., B458:209–218,
1999, hep-th/9904075.
[66] C. Armendariz-Picon, V. F. Mukhanov, and P. J. Steinhardt. A dynamical solution to the
problem of a small cosmological constant and late-time cosmic acceleration. Phys. Rev. Lett.,
85:4438–4441, 2000, astro-ph/0004134.
[67] P. Astier et al. The Supernova legacy survey: Measurement of Ωm , ΩΛ and w from the first year
data set. Astron.Astrophys., 447:31–48, 2006, astro-ph/0510447.
[68] K. Atazadeh, M. Farhoudi, and H. R. Sepangi. Accelerating universe in f(R) brane gravity. Phys.
Lett., B660:275–281, 2008, 0801.1398.
[69] M. Bañados and P. G. Ferreira. Eddington’s Theory of Gravity and Its Progeny. Physical Review
Letters, 105(1):011101, July 2010, 1006.1769.
[70] M. Bañados, P. G. Ferreira, and C. Skordis. Eddington-Born-Infeld gravity and the large scale
structure of the Universe. Phys. Rev. D, 79(6):063511, Mar. 2009, 0811.1272.
[71] M. Bañados, A. Gomberoff, and M. Pino. The bigravity black hole and its thermodynamics.
ArXiv e-prints, May 2011, 1105.1172.
[72] E. Babichev and D. Langlois. Relativistic stars in f(R) gravity. Phys. Rev. D, 80(12):121501,
Dec. 2009, 0904.1382.
[73] E. Babichev, V. Mukhanov, and A. Vikman. k-Essence, superluminal propagation, causality and
emergent geometry. JHEP, 0802:101, 2008, 0708.0561.
[74] C. Baccigalupi, S. Matarrese, and F. Perrotta. Tracking extended quintessence. Phys. Rev. D,
62(12):123510, Nov 2000.
[75] A. F. Bahrehbakhsh, M. Farhoudi, and H. Shojaie. FRW cosmology from five dimensional vacuum
Brans-Dicke theory. Gen. Rel. Grav. 43:847–869, 2011, 1005.2501.
[76] D. Bailin and A. Love. Kaluza-Klein theories. Rept. Prog. Phys., 50:1087–1170, 1987.
[77] T. Baker, P. G. Ferreira, C. Skordis, and J. Zuntz. A fully consistent parameterization of modified
gravity. Phys. Rev. D, 84(12):124018, 2011, 1107.0491.
[78] K. Bamba, C. Geng, and C. Lee. Comment on “Einstein’s Other Gravity and the Acceleration
of the Universe”. ArXiv e-prints, Aug. 2010, 1008.4036.
[79] K. Bamba, C. Geng, C. Lee, and L. Luo. Equation of state for dark energy in f(T) gravity. ArXiv
e-prints, Nov. 2010, 1011.0508.
[80] K. Bamba, Z.-K. Guo, and N. Ohta. Accelerating Cosmologies in the Einstein-Gauss-Bonnet
Theory with Dilaton. Prog. Theo. Phys., 118:879-892, 2007, 0707.4334.
[81] K. Bamba, S. Nojiri, and S. D. Odintsov. The future of the universe in modified gravitational
theories: approaching a finite-time future singularity. J. Cosmology Astropart. Phys., 10:45, Oct.
269
 2008, 0807.2575.
[82] K. Bamba and S. D. Odintsov. Inflation and late-time cosmic acceleration in non-minimal
Maxwell-f(R) gravity and the generation of large-scale magnetic fields. J. Cosmology Astropart.
Phys., 4:24, Apr. 2008, 0801.0954.
[83] C. Bambi, M. Giannotti, and F. L. Villante. Response of primordial abundances to a general
modification of gN and/or of the early universe expansion rate. Phys. Rev. D, 71(12):123524,
Jun 2005.
[84] M. Bañados. Eddington-Born-Infeld action for dark matter and dark energy. Phys.Rev.,
D77:123534, 2008, 0801.4103.
[85] M. Banados, A. Gomberoff, D. C. Rodrigues, and C. Skordis. A note on bigravity and dark
matter. Phys. Rev., D79:063515, 2009, 0811.1270.
[86] A. Banerjee, S. B. Dutta Choudhury, N. Banerjee, and A. Sil. Inflationary solutions in general
scalar tensor theory of gravitation. Pramana, 40:31–34, 1993.
[87] E. Barausse, T. P. Sotiriou, and J. C. Miller. A no-go theorem for polytropic spheres in Palatini
f(R) gravity. Class. Quant. Grav., 25:062001, 2008, gr-qc/0703132.
[88] E. Barausse, T. P. Sotiriou, and J. C. Miller. Curvature singularities, tidal forces and the viability
of Palatini f(R) gravity. Class. Quant. Grav., 25:105008, 2008, 0712.1141.
[89] E. Barausse, T. P. Sotiriou, and J. C. Miller. Polytropic spheres in Palatini f(R) gravity. EAS
Publ. Ser., 30:189–192, 2008, 0801.4852.
[90] J. M. Bardeen. Gauge Invariant Cosmological Perturbations. Phys.Rev., D22:1882–1905, 1980.
[91] S. M. Barr, R. S. Chivukula, and E. Farhi. Electroweak fermion number violation and the
production of stable particles in the early universe. Physics Letters B, 241(3):387 – 391, 1990.
[92] D. E. Barraco and V. H. Hamity. Spherically symmetric solutions in f(R) theories of gravity
obtained using the first order formalism. Phys. Rev. D, 62(4):044027, Aug. 2000.
[93] J. D. Barrow. Gravitational memory? Phys. Rev. D, 46(8):R3227–R3230, Oct 1992.
[94] J. D. Barrow. Nonsingular scalar-tensor cosmologies. Phys. Rev. D, 48:3592–3595, Oct. 1993.
[95] J. D. Barrow. Scalar-tensor cosmologies. Phys. Rev. D, 47(12):5329–5335, Jun 1993.
[96] J. D. Barrow. Slow-roll inflation in scalar-tensor theories. Phys. Rev. D, 51:2729–2732, Mar.
1995.
[97] J. D. Barrow and B. J. Carr. Formation and evaporation of primordial black holes in scalar-tensor
gravity theories. Phys. Rev. D, 54(6):3920–3931, Sep 1996.
[98] J. D. Barrow and T. Clifton. Exact cosmological solutions of scale-invariant gravity theories.
Classical and Quantum Gravity, 23:L1–L6, Jan. 2006, arXiv:gr-qc/0509085.
[99] J. D. Barrow, E. J. Copeland, E. W. Kolb, and A. R. Liddle. Baryogenesis in extended inflation.
I. Baryogenesis via production and decay of supermassive bosons. Phys. Rev. D, 43:977–983, Feb.
1991.
[100] J. D. Barrow, E. J. Copeland, E. W. Kolb, and A. R. Liddle. Baryogenesis in extended inflation.
II. Baryogenesis via primordial black holes. Phys. Rev. D, 43:984–994, Feb. 1991.
[101] J. D. Barrow and S. Cotsakis. Inflation and the conformal structure of higher-order gravity
theories. Physics Letters B, 214:515–518, Dec. 1988.
[102] J. D. Barrow and S. Hervik. Anisotropically inflating universes. Phys. Rev., D73:023007, 2006,
gr-qc/0511127.
[103] J. D. Barrow and S. Hervik. Evolution of universes in quadratic theories of gravity. Phys. Rev. D,
74(12):124017, Dec. 2006, arXiv:gr-qc/0610013.
[104] J. D. Barrow and S. Hervik. Simple types of anisotropic inflation. Phys. Rev. D, 81(2):023513,
Jan. 2010, 0911.3805.
[105] J. D. Barrow, D. Kimberly, and J. Magueijo. Bouncing universes with varying constants. Classical
and Quantum Gravity, 21:4289–4296, Sept. 2004, arXiv:astro-ph/0406369.
[106] J. D. Barrow and K. Maeda. Extended inflationary universes. Nuclear Physics B, 341:294–308,
Sept. 1990.
[107] J. D. Barrow and J. Middleton. Stable isotropic cosmological singularities in quadratic gravity.
Phys. Rev. D, 75(12):123515, June 2007, arXiv:gr-qc/0702098.
[108] J. D. Barrow and J. P. Mimoso. Perfect fluid scalar-tensor cosmologies. Phys. Rev. D, 50(6):3746–
3754, Sep 1994.
[109] J. D. Barrow, J. P. Mimoso, and M. R. de Garcia Maia. Amplification of gravitational waves in
scalar-tensor theories of gravity. Phys. Rev. D, 48:3630–3640, Oct. 1993.
[110] J. D. Barrow and A. C. Ottewill. The stability of general relativistic cosmological theory. Journal
of Physics A: Mathematical and General, 16(12):2757, 1983.
[111] J. D. Barrow and P. Parsons. Behavior of cosmological models with varying g. Phys. Rev. D,
270
 55(4):1906–1936, Feb 1997.
[112] J. D. Barrow and D. J. Shaw. Some late-time asymptotics of general scalar-tensor cosmologies.
Classical and Quantum Gravity, 25(8):085012, 2008.
[113] N. H. Barth and S. M. Christensen. Quantizing fourth-order gravity theories: The functional
integral. Phys. Rev. D, 28:1876–1893, Oct. 1983.
[114] Y. V. Baryshev, I. N. Taganov and P. Teerikorpi. Energy-momentum of the gravitational field:
crucial point for gravitation physics and cosmology. Problems of Practical Cosmology, Volume
1, p.276-286, 2008, gr-qc/0809.2323.
[115] Y. V. Baryshev, I. N. Taganov and P. Teerikorpi. New possibilities for observational distinction
between geometrical and field gravity theories. Problems of Practical Cosmology, Volume 1,
p.347-359, 2008, gr-qc/0809.2328.
[116] T. Battefeld and S. Watson. String gas cosmology. Rev. Mod. Phys., 78:435–454, 2006, hep-
th/0510022.
[117] R. A. Battye, B. Carter, A. Mennim, and J.-P. Uzan. Einstein equations for an asymmetric
brane-world. Phys. Rev., D64:124007, 2001, hep-th/0105091.
[118] A. Bauch and S. Weyers. New experimental limit on the validity of local position invariance.
Phys. Rev. D, 65(8):081101, Apr. 2002.
[119] V. Baukh, A. Zhuk, Sp-brane accelerating cosmologies Phys. Rev. D73 (2006) 104016. [hep-
th/0601205].
[120] V. Baukh, A. Zhuk, Dynamical dark energy from extra dimensions Kinematics Phys. Celest.
Bodies 25 (2009) 57-63. [arXiv:0809.1385 [astro-ph]].
[121] A. Bayntun, C. Burgess, and L. van Nierop. Codimension-2 Brane-Bulk Matching: Examples
from Six and Ten Dimensions. New J.Phys., 12:075015, 2010, 0912.3039.
[122] R. Bean, D. Bernat, L. Pogosian, A. Silvestri, and M. Trodden. Dynamics of linear perturbations
in f(R) gravity. Phys. Rev. D, 75(6):064020, Mar. 2007, arXiv:astro-ph/0611321.
[123] R. Bean and M. Tangmatitham. Current constraints on the cosmic growth history. Phys. Rev. D,
81(8):083534, Apr. 2010, 1002.4197.
[124] J. Bekenstein and J. Magueijo. MOND habitats within the solar system. Phys. Rev., D73:103513,
2006, astro-ph/0602266.
[125] J. Bekenstein and M. Milgrom. Does the missing mass problem signal the breakdown of Newtonian
gravity? Astrophys. J., 286:7–14, 1984.
[126] J. D. Bekenstein. Phase coupling gravitation: symmetries and gauge fields. Phys. Lett., B202:497–
500, 1988.
[127] J. D. Bekenstein. The Relation between physical and gravitational geometry. Phys. Rev.,
D48:3641–3647, 1993, gr-qc/9211017.
[128] J. D. Bekenstein. Relativistic gravitation theory for the MOND paradigm. Phys. Rev.,
D70:083509, 2004.
[129] J. D. Bekenstein and E. Sagi. Do Newton’s G and Milgrom’s a0 vary with cosmological epoch?
Phys. Rev., D77:103512, 2008, 0802.1526.
[130] J. D. Bekenstein and R. H. Sanders. Gravitational lenses and unconventional gravity theories.
Astrophys. J., 429:480, 1994, astro-ph/9311062.
[131] J. A. Belinchon. Bianchi V I0 in Scalar and Scalar-Tensor cosmologies. ArXiv e-prints, Sep. 2011,
1109.2880.
[132] J. A. Belinchon. Generalized Self-similar Scalar-Tensor Theories. ArXiv e-prints, Sep. 2011,
1109.2877.
[133] V. A. Belinskij and I. M. Khalatnikov. On the effect of scalar and vector fields on the nature of
the cosmological singularity. Zhurnal Eksperimental noi i Teoreticheskoi Fiziki, 63:1121–1134,
1972.
[134] G. R. Bengochea. Observational information for f(T) theories and Dark Torsion. ArXiv e-prints,
Aug. 2010, 1008.3188.
[135] G. R. Bengochea and R. Ferraro. Dark torsion as the cosmic speed-up. Phys. Rev. D,
79(12):124019, June 2009, 0812.1205.
[136] I. Bengtsson. The Cosmological constants. Phys.Lett., B254:55–60, 1991.
[137] I. Bengtsson. Form connections. 1993, gr-qc/9305004.
[138] I. Bengtsson. Form geometry and the ’t Hooft-Plebanski action. 1995, gr-qc/9502010.
[139] I. Bengtsson. Note on non-metric gravity. Mod.Phys.Lett., A22:1643–1649, 2007, gr-qc/0703114.
[140] Z. Berezhiani, D. Comelli, F. Nesti, and L. Pilo. Spontaneous Lorentz Breaking and Massive
Gravity. Phys. Rev. Lett, 99:131101, 2007, hep-th/0703264.
[141] Z. Berezhiani, D. Comelli, F. Nesti, and L. Pilo. Exact Spherically Symmetric Solutions in
271
 Massive Gravity. JHEP, 0807:130, 2008, 0803.1687.
[142] P. G. Bergmann. Comments on the scalar tensor theory. Int. J. Theor. Phys., 1:25–36, 1968.
[143] A. L. Berkin and K. Maeda. Effects of R3 and RR terms on R2 inflation. Physics Letters B,
245:348–354, Aug. 1990.
[144] A. L. Berkin and K. Maeda. Inflation in generalized Einstein theories. Phys. Rev. D, 44:1691–
1704, Sept. 1991.
[145] M. S. Berman and L. A. Trevisan. Amplification of Gravitational Waves During Inflation in
Brans-Dicke Theory. ArXiv General Relativity and Quantum Cosmology e-prints, Nov. 2001,
arXiv:gr-qc/0111100.
[146] C. P. L. Berry and J. R. Gair. Linearized f(R) Gravity: Gravitational Radiation and Solar System
Tests. ArXiv e-prints, Apr. 2011, 1104.0819.
[147] B. Bertotti, L. Iess, and P. Tortora. A test of general relativity using radio links with the Cassini
spacecraft. Nature, 425:374–376, Sept. 2003.
[148] E. Bertschinger. On the Growth of Perturbations as a Test of Dark Energy and Gravity. ApJ,
648:797–806, Sept. 2006, arXiv:astro-ph/0604485.
[149] E. Bertschinger. Testing General Relativity with Cosmology. The Royal Society, February 2011.
[150] E. Bertschinger. One Gravitational Potential or Two? Forecasts and Tests. Phil. Trans. R. Soc.
A, 369:4947, 2011, 1111.4659.
[151] E. Bertschinger and P. Zukin. Distinguishing modified gravity from dark energy. Phys. Rev. D,
78(2):024015, July 2008, 0801.2431.
[152] A. Bhadra and K. K. Nandi. On the equivalence of the Buchdahl and the Janis-Newman- Win-
nicour solutions. Int. J. Mod. Phys., A16:4543–4545, 2001.
[153] A. Bhadra and K. Sarkar. On static spherically symmetric solutions of the vacuum Brans-Dicke
theory. Gen. Rel. Grav., 37:2189–2199, 2005, gr-qc/0505141.
[154] N. D. R. Bhat, M. Bailes, and J. P. W. Verbiest. Gravitational-radiation losses from the pulsar
white-dwarf binary PSR J1141 6545. Phys. Rev. D, 77(12):124017, June 2008, 0804.0956.
[155] G. V. Bicknell. A geometrical Lagrangian for the neutral scalar meson field. Journal of Physics
A Mathematical General, 7:341–345, Feb. 1974.
[156] A. Billyard, A. Coley, and J. Ibáñez. Asymptotic behavior of cosmological models in scalar-tensor
theories of gravity. Phys. Rev. D, 59(2):023507, Jan. 1999, arXiv:gr-qc/9807055.
[157] A. Billyard, A. Coley, J. Ibáñez, and I. Olasagasti. Inhomogeneous cosmological models in scalar-
tensor theories. Europhysics Letters, 46:832–837, June 1999.
[158] P. Binetruy, C. Deffayet, U. Ellwanger, and D. Langlois. Brane cosmological evolution in a bulk
with cosmological constant. Phys. Lett., B477:285–291, 2000, hep-th/9910219.
[159] P. Binetruy, C. Deffayet, and D. Langlois. Non-conventional cosmology from a brane-universe.
Nucl. Phys., B565:269–287, 2000, hep-th/9905012.
[160] G. D. Birkhoff. Relativity and Modern Physics: Harvard University Press. 1923.
[161] N. Birrell and P. Davies. Quantum fields in curved space. Cambridge Univ Pr, 1984.
[162] T. Biswas, R. Brandenberger, A. Mazumdar, and W. Siegel. Non-perturbative gravity, the Hage-
dorn bounce and the cosmic microwave background. J. Cosmology Astropart. Phys., 12:11, Dec.
2007, arXiv:hep-th/0610274.
[163] T. Biswas, T. Koivisto, and A. Mazumdar. Towards a resolution of the cosmological singularity
in non-local higher derivative theories of gravity. J. Cosmology Astropart. Phys., 11:8, Nov. 2010,
1005.0590.
[164] T. Biswas, A. Mazumdar, and W. Siegel. Bouncing universes in string-inspired gravity. J.
Cosmology Astropart. Phys., 3:9, Mar. 2006, arXiv:hep-th/0508194.
[165] C. Blake et al. The WiggleZ Dark Energy Survey: the growth rate of cosmic structure since
redshift z=0.9. 2011, 1104.2948.
[166] J. J. Blanco-Pillado, D. Buck, E. J. Copeland, M. Gomez-Reino, and N. J. Nunes. Kahler Moduli
Inflation Revisited. JHEP, 01:081, 2010, 0906.3711.
[167] D. Blas, O. Pujolas, and S. Sibiryakov. On the Extra Mode and Inconsistency of Horava Gravity.
JHEP, 0910:029, 2009, 0906.3046.
[168] D. Blas, O. Pujolas, and S. Sibiryakov. Comment on ‘Strong coupling in extended Horava-Lifshitz
gravity’. Phys.Lett., B688:350–355, 2010, 0912.0550.
[169] D. Blas, O. Pujolas, and S. Sibiryakov. Consistent Extension of Horava Gravity. Phys.Rev.Lett.,
104:181302, 2010, 0909.3525.
[170] D. Blas, O. Pujolas, and S. Sibiryakov. Models of non-relativistic quantum gravity: The Good,
the bad and the healthy. 2010, 1007.3503.
[171] D. Blas, S. Sibiryakov, Technically natural dark energy from Lorentz breaking JCAP 1107
272
 (2011) 026. [arXiv:1104.3579 [hep-th]].
[172] R. Bluhm, V. A. Kostelecký, C. D. Lane, and N. Russell. Probing Lorentz and CPT violation
with space-based experiments. Phys. Rev. D, 68(12):125008, Dec. 2003, arXiv:hep-ph/0306190.
[173] M. Borunda, B. Janssen, and M. Bastero-Gil. Palatini versus metric formulation in higher-
curvature gravity. J. Cosmology Astropart. Phys., 11:8, Nov. 2008, 0804.4440.
[174] P. Bostock, R. Gregory, I. Navarro, and J. Santiago. Einstein gravity on the codimension 2-brane?
Phys.Rev.Lett., 92:221601, 2004, hep-th/0311074.
[175] D. Boulware and S. Deser. Can gravitation have a finite range? Phys.Rev., D6:3368–3382, 1972.
[176] D. G. Boulware and S. Deser. String Generated Gravity Models. Phys. Rev. Lett., 55:2656, 1985.
[177] F. Bourliot, P. G. Ferreira, D. F. Mota, and C. Skordis. The cosmological behavior of Bekenstein’s
modified theory of gravity. Phys. Rev., D75:063508, 2007, astro-ph/0611255.
[178] P. Bowcock, C. Charmousis, and R. Gregory. General brane cosmologies and their global space-
time structure. Class. Quant. Grav., 17:4745–4764, 2000, hep-th/0007177.
[179] R. Brandenberger. Matter Bounce in Horava-Lifshitz Cosmology. Phys. Rev., D80:043516, 2009,
0904.2835.
[180] R. H. Brandenberger. Challenges for string gas cosmology. 2005, hep-th/0509099.
[181] R. H. Brandenberger. String Gas Cosmology. 2008, 0808.0746.
[182] R. H. Brandenberger et al. More on the Spectrum of Perturbations in String Gas Cosmology.
JCAP, 0611:009, 2006, hep-th/0608186.
[183] R. H. Brandenberger and C. Vafa. Superstrings in the Early Universe. Nucl. Phys., B316:391,
1989.
[184] C. Brans and R. H. Dicke. Mach’s principle and a relativistic theory of gravitation. Phys. Rev.,
124:925–935, 1961.
[185] C. H. Brans. Mach’s principle and a relativistic theory of gravitation. ii. Phys. Rev., 125(6):2194–
2201, Mar 1962.
[186] P. Brax and C. Burrage. Chameleon Induced Atomic Afterglow. ArXiv e-prints, Sept. 2010,
1009.1065.
[187] P. Brax, C. Burrage, A. Davis, D. Seery, and A. Weltman. Higgs production as a probe of
chameleon dark energy. Phys. Rev. D, 81(10):103524, May 2010, 0911.1267.
[188] P. Brax, R. Rosenfeld, and D. A. Steer. Spherical collapse in chameleon models. J. Cosmology
Astropart. Phys., 8:33, Aug. 2010, 1005.2051.
[189] P. Brax and C. van de Bruck. Cosmology and brane worlds: A review. Class. Quant. Grav.,
20:R201–R232, 2003, hep-th/0303095.
[190] P. Brax, C. van de Bruck, and A. Davis. Compatibility of the Chameleon-Field Model with
Fifth-Force Experiments, Cosmology, and PVLAS and CAST Results. Physical Review Letters,
99(12):121103, Sept. 2007, arXiv:hep-ph/0703243.
[191] P. Brax, C. van de Bruck, A. Davis, J. Khoury, and A. Weltman. Detecting dark energy in orbit:
The cosmological chameleon. Phys. Rev. D, 70(12):123518, Dec. 2004, arXiv:astro-ph/0408415.
[192] P. Brax, C. van de Bruck, A. Davis, D. F. Mota, and D. Shaw. Testing chameleon theories with
light propagating through a magnetic field. Phys. Rev. D, 76(8):085010, Oct. 2007, 0707.2801.
[193] P. Brax, C. van de Bruck, A. Davis, and D. J. Shaw. f(R) gravity and chameleon theories.
Phys. Rev. D, 78(10):104021, Nov. 2008, 0806.3415.
[194] P. Brax, C. van de Bruck, and A. C. Davis. Brane-world cosmology, bulk scalars and perturba-
tions. JHEP, 10:026, 2001, hep-th/0108215.
[195] P. Brax, C. van de Bruck, and A.-C. Davis. Brane world cosmology. Rept. Prog. Phys., 67:2183–
2232, 2004, hep-th/0404011.
[196] P. Brax, C. van de Bruck, A. C. Davis, and A. M. Green. Small scale structure formation in
chameleon cosmology. Physics Letters B, 633:441–446, Feb. 2006, arXiv:astro-ph/0509878.
[197] P. Brax, C. van de Bruck, A. C. Davis, J. Khoury, and A. Weltman. Chameleon Dark Energy.
In C. J. A. P. Martins, P. P. Avelino, M. S. Costa, K. Mack, M. F. Mota, & M. Parry, editor,
Phi in the Sky: The Quest for Cosmological Scalar Fields, volume 736 of American Institute of
Physics Conference Series, pages 105–110, Nov. 2004, arXiv:astro-ph/0410103.
[198] P. Brax, C. van de Bruck, D. F. Mota, N. J. Nunes, and H. A. Winther. Chameleons with Field
Dependent Couplings. ArXiv e-prints, June 2010, 1006.2796.
[199] P. Brax and K. Zioutas. Solar chameleons. Phys. Rev. D, 82(4):043007, Aug. 2010, 1004.1846.
[200] I. Brevik, S. Nojiri, S. D. Odintsov, and L. Vanzo. Entropy and universality of the Cardy-Verlinde
formula in a dark energy universe. Phys. Rev. D, 70(4):043520, Aug. 2004, arXiv:hep-th/0401073.
[201] H. A. Bridgman, K. A. Malik, and D. Wands. Cosmic vorticity on the brane. Phys. Rev.,
D63:084012, 2001, hep-th/0010133.
273
[202] H. A. Bridgman, K. A. Malik, and D. Wands. Cosmological perturbations in the bulk and on the
brane. Phys. Rev., D65:043502, 2002, astro-ph/0107245.
[203] F. Briscese, E. Elizalde, S. Nojiri and S. D. Odintsov. Phantom scalar dark energy as modified
gravity: Understanding the origin of the Big Rip singularity. Phys. Lett. B, 646:105–111, Mar.
2007, arXiv:hep-th/0612220.
[204] A. W. Brookfield, C. van de Bruck, and L. M. H. Hall. Viability of f(R) theories with additional
powers of curvature. Phys. Rev. D, 74(6):064028, Sept. 2006, arXiv:hep-th/0608015.
[205] J.-P. Bruneton. On causality and superluminal behavior in classical field theories. Applications
to k-essence theories and MOND-like theories of gravity. Phys. Rev., D75:085013, 2007, gr-
qc/0607055.
[206] H. Buchdahl. ber die variationsableitung von fundamental-invarianten beliebig hoher ordnung.
Acta Mathematica, 85:63–72, 1964. 10.1007/BF02395741.
[207] H. A. Buchdahl. Reciprocal Static Metrics and Scalar Fields in the General Theory of Relativity.
Phys. Rev., 115:1325–1328, 1959.
[208] H. A. Buchdahl. Non-linear Lagrangians and cosmological theory. MNRAS, 150:1, 1970.
[209] H. A. Buchdahl. The field equations generated by the square of the scalar curvature: solutions
of Kasner type. Journal of Physics A Mathematical General, 11:871–876, May 1978.
[210] M. R. Buckley and L. Randall. Xogenesis. 2010, 1009.0270.
[211] J. S. Bullock. Notes on the Missing Satellites Problem. 2010, 1009.4505.
[212] A. Burd and A. Coley. Extended inflation and generalised scalar-tensor theories. Physics Letters
B, 267:330–336, Sept. 1991.
[213] C. Burgess, D. Hoover, C. de Rham, and G. Tasinato. Effective Field Theories and Matching for
Codimension-2 Branes. JHEP, 0903:124, 2009, 0812.3820.
[214] C. Burgess and L. van Nierop. Bulk Axions, Brane Back-reaction and Fluxes. 2010, 1012.2638.
[215] C. Burgess and L. van Nierop. Large Dimensions and Small Curvatures from Supersymmetric
Brane Back-reaction. 2011, 1101.0152.
[216] C. P. Burgess. Supersymmetric large extra dimensions and the cosmological constant: An update.
Ann. Phys., 313:283–401, 2004, hep-th/0402200.
[217] C. P. Burgess. Towards a natural theory of dark energy: Supersymmetric large extra dimensions.
AIP Conf. Proc., 743:417–449, 2005, hep-th/0411140.
[218] C. P. Burgess. Lectures on Cosmic Inflation and its Potential Stringy Realizations. PoS,
P2GC:008, 2006, 0708.2865.
[219] C. Burrage. Supernova brightening from chameleon-photon mixing. Phys. Rev. D, 77(4):043009,
Feb. 2008, 0711.2966.
[220] C. Burrage, A. Davis, and D. J. Shaw. Detecting chameleons: The astronomical polarization
produced by chameleonlike scalar fields. Phys. Rev. D, 79(4):044028, Feb. 2009, 0809.1763.
[221] C. Burrage, C. de Rham, and L. Heisenberg. de Sitter Galileon. 2011, 1104.0155.
[222] C. Burrage, C. de Rham, D. Seery, and A. J. Tolley. Galileon inflation. JCAP, 1101:014, 2011,
1009.2497.
[223] C. Burrage and D. Seery. Revisiting fifth forces in the Galileon model. JCAP, 1008:011, 2010,
1005.1927.
[224] A. J. Bustelo and D. E. Barraco. Hydrostatic equilibrium equation and Newtonian limit of
the singular f(R) gravity. Classical and Quantum Gravity, 24:2333–2342, May 2007, arXiv:gr-
qc/0611149.
[225] R.-G. Cai, L.-M. Cao, Y.-P. Hu, and N. Ohta. Generalized Misner-Sharp energy in f(R) gravity.
Phys. Rev. D, 80(10):104016, Nov. 2009, 0910.2387.
[226] R.-G. Cai, L.-M. Cao, and N. Ohta. Topological black holes in Horava-Lifshitz gravity.
Phys. Rev. D, 80(2):024003, Jul. 2009, 0904.3670.
[227] R.-G. Cai, L.-M. Cao, and N. Ohta. Thermodynamics of black holes in Horava-Lifshitz gravity.
Phys. Lett. B, 679:504-509, Sep. 2009, 0905.0751.
[228] R.-G. Cai, L.-M. Cao, and N. Ohta. Black holes without mass and entropy in Lovelock gravity.
Phys. Rev. D, 81(2):024018, Jan. 2010, 0911.0245.
[229] R.-G. Cai, C.-M. Chen, K.-I. Maeda, N. Ohta, and D.-W. Pang. Entropy function and universality
of entropy-area relation for small black holes. Phys. Rev. D, 77(6):064030, Mar. 2008, 0712.4212.
[230] R.-G. Cai, B. Hu, and H.-B. Zhang. Dynamical Scalar Degree of Freedom in Horava-Lifshitz
Gravity. Phys.Rev., D80:041501, 2009, 0905.0255.
[231] R.-G. Cai, and N. Ohta. Quantum aspects of black holes, evaporation, thermodynamics, Gravity
in more than four dimensions, Kaluza-Klein theory, unified field theories; alternative theories of
gravity, Classical black holes. Phys. Rev. D, 74(6):064001, Sep. 2006, hep-th/0604088.
274
[232] R.-G. Cai, and N. Ohta. Horizon thermodynamics and gravitational field equations in Horava-
Lifshitz gravity. Phys. Rev. D, 81(8):084061, Apr. 2010, 0910.2307.
[233] R.-G. Cai and K.-S. Soh. Topological black holes in the dimensionally continued gravity.
Phys.Rev., D59:044013, 1999, gr-qc/9808067.
[234] Y.-F. Cai and E. N. Saridakis. Non-singular cosmology in a model of non-relativistic gravity.
JCAP, 0910:020, 2009, 0906.1789.
[235] G. Calcagni. Cosmology of the Lifshitz universe. JHEP, 0909:112, 2009, 0904.0829.
[236] G. Calcagni, S. Tsujikawa, and M. Sami. Dark energy and cosmological solutions in second-order
string gravity. Classical and Quantum Gravity, 22:3977–4006, Oct. 2005, arXiv:hep-th/0505193.
[237] R. Caldwell, A. Cooray, and A. Melchiorri. Constraints on a new post-general relativity cosmo-
logical parameter. Phys. Rev. D, 76(2):023507, July 2007, arXiv:astro-ph/0703375.
[238] P. Candelas and S. Weinberg. Calculation of Gauge Couplings and Compact Circumferences from
Selfconsistent Dimensional Reduction. Nucl. Phys., B237:397, 1984.
[239] F. Cannata and A. Y. Kamenshchik. Chameleon Cosmology Model Describing the Phantom
Divide Line Crossing. ArXiv e-prints, May 2010, 1005.1878.
[240] S. Capozziello, V. F. Cardone, and A. Troisi. Reconciling dark energy models with f(R) theories.
Phys. Rev. D, 71(4):043503, Feb. 2005, arXiv:astro-ph/0501426.
[241] S. Capozziello, S. Carloni, and A. Troisi. Quintessence without scalar fields. ArXiv Astrophysics
e-prints, Mar. 2003, arXiv:astro-ph/0303041.
[242] S. Capozziello, S. Nojiri, S. D. Odintsov, and A. Troisi. Cosmological viability of f(R)-gravity as
an ideal fluid and its compatibility with a matter dominated phase. Physics Letters B, 639:135–
143, Aug. 2006, arXiv:astro-ph/0604431.
[243] S. Capozziello, A. Stabile, and A. Troisi. Spherically symmetric solutions in f(R) gravity via the
Noether symmetry approach. Classical and Quantum Gravity, 24(8):2153, 2007.
[244] S. Capozziello and S. Tsujikawa. Solar system and equivalence principle constraints on f(R)
gravity by the chameleon approach. Phys. Rev. D, 77(10):107501, May 2008, 0712.2268.
[245] A. Cardoso, T. Hiramatsu, K. Koyama, and S. S. Seahra. Scalar perturbations in braneworld
cosmology. JCAP, 0707:008, 2007, 0705.1685.
[246] A. Cardoso, K. Koyama, S. S. Seahra, and F. P. Silva. Cosmological perturbations in the DGP
braneworld: Numeric solution. Phys.Rev., D77:083512, 2008, 0711.2563.
[247] S. Carloni, K. N. Ananda, P. K. S. Dunsby, and M. E. S. Abdelwahab. Unifying the study of
background dynamics and perturbations in f(R) gravity. ArXiv e-prints, Dec. 2008, 0812.2211.
[248] S. Carloni, P. K. S. Dunsby, S. Capozziello, and A. Troisi. Cosmological dynamics of Rn gravity.
Classical and Quantum Gravity, 22:4839–4868, Nov. 2005, arXiv:gr-qc/0410046.
[249] S. Carloni, P. K. S. Dunsby, and D. Solomons. Bounce conditions in f(R) cosmologies. Classical
and Quantum Gravity, 23:1913–1922, Mar. 2006, arXiv:gr-qc/0510130.
[250] S. Carloni, P. K. S. Dunsby, and A. Troisi. Evolution of density perturbations in f(R) gravity.
Phys. Rev. D, 77(2):024024, Jan. 2008, 0707.0106.
[251] S. Carloni, R. Goswami, and P. K. S. Dunsby. A new approach to reconstruction methods in
f(R) gravity. ArXiv e-prints, May 2010, 1005.1840.
[252] S. Carloni, J. A. Leach, S. Capozziello, and P. K. S. Dunsby. Cosmological dynamics of scalar
tensor gravity. Classical and Quantum Gravity, 25(3):035008, Feb. 2008, arXiv:gr-qc/0701009.
[253] S. Carloni, A. Troisi, and P. K. S. Dunsby. Some remarks on the dynamical systems approach to
fourth order gravity. General Relativity and Gravitation, 41:1757–1776, Aug. 2009, 0706.0452.
[254] S. M. Carroll, A. de Felice, V. Duvvuri, D. A. Easson, M. Trodden, and M. S. Turner. Cosmology
of generalized modified gravity models. Phys. Rev. D, 71(6):063513, Mar. 2005, arXiv:astro-
ph/0410031.
[255] S. M. Carroll, V. Duvvuri, M. Trodden, and M. S. Turner. Is cosmic speed-up due to new
gravitational physics? Phys. Rev. D, 70(4):043528, Aug. 2004, arXiv:astro-ph/0306438.
[256] S. M. Carroll and M. M. Guica. Sidestepping the cosmological constant with football shaped
extra dimensions. 2003, hep-th/0302067.
[257] S. M. Carroll and E. A. Lim. Lorentz-violating vector fields slow the universe down. Phys. Rev. D,
70(12):123525, Dec. 2004, arXiv:hep-th/0407149.
[258] S. M. Carroll, I. Sawicki, A. Silvestri, and M. Trodden. Modified-source gravity and cosmological
structure formation. New Journal of Physics, 8:323, Dec. 2006, arXiv:astro-ph/0607458.
[259] É. Cartan. Sur une généralisation de la notion de courbure de Riemann et les espaces à torsion.
C. R. Acad. Sci. , 174, 593, 1922.
[260] É. Cartan. Sur les variétés à connexion affine et la théorie de la relativitée généralisée I. Ann.

275
 Ec. Norm. Sup. , 40, 325, 1923.
[261] É. Cartan. Sur les variétés à connexion affine et la théorie de la relativitée généralisée II. Ann.
Ec. Norm. Sup. , 42, 17, 1925.
[262] B. Carter. Axisymmetric black hole has only two degrees of freedom. Phys. Rev. Lett., 26(6):331–
333, Feb 1971.
[263] B. Carter and R. Gregory. Curvature corrections to dynamics of domain walls. Phys. Rev.,
D51:5839–5846, 1995, hep-th/9410095.
[264] C. Cartier, E. J. Copeland, and R. Madden. The graceful exit in string cosmology. Journal of
High Energy Physics, 1:35, Jan. 2000, arXiv:hep-th/9910169.
[265] J. A. Casas, J. Garcı́a-Bellido, and M. Quirós. Nucleosynthesis Bounds on Jordan-Brans Theories
of Gravity. Modern Physics Letters A, 7:447–456, 1992.
[266] J. A. Casas, J. Garcı́a-Bellido, and M. Quirós. Updating nucleosynthesis bounds on Jordan-
Brans-Dicke theories of gravity. Physics Letters B, 278:94–96, Mar. 1992.
[267] J. A. R. Cembranos. The newtonian limit at intermediate energies. Phys. Rev. D, 73(6):064029,
Mar 2006.
[268] A. H. Chamseddine, A. Connes, and M. Marcolli. Gravity and the standard model with neutrino
mixing. Adv. Theor. Math. Phys., 11:991–1089, 2007, hep-th/0610241.
[269] K. Chan and R. Scoccimarro. Large-Scale Structure in Brane-Induced Gravity II. Numerical
Simulations. Phys.Rev., D80:104005, 2009, 0906.4548.
[270] C. Charmousis. Higher order gravity theories and their black hole solutions. Lect.Notes Phys.,
769:299–346, 2009, 0805.0568.
[271] C. Charmousis, S. C. Davis, and J.-F. Dufaux. Scalar brane backgrounds in higher order curvature
gravity. JHEP, 0312:029, 2003, hep-th/0309083.
[272] C. Charmousis and J.-F. Dufaux. General Gauss-Bonnet brane cosmology. Class.Quant.Grav.,
19:4671–4682, 2002, hep-th/0202107.
[273] C. Charmousis, R. Gregory, N. Kaloper, and A. Padilla. DGP specteroscopy. JHEP, 10:066,
2006, hep-th/0604086.
[274] C. Charmousis, R. Gregory, and A. Padilla. Stealth Acceleration and Modified Gravity. JCAP,
0710:006, 2007, 0706.0857.
[275] C. Charmousis, R. Gregory, and V. A. Rubakov. Wave function of the radion in a brane world.
Phys. Rev., D62:067505, 2000, hep-th/9912160.
[276] C. Charmousis, C. E. J., P. Antonia, and P. M. Saffin. General second order scalar-tensor theory,
self tuning, and the fab four. 2011. submitted.
[277] C. Charmousis, G. Kofinas, and A. Papazoglou. The consistency of codimension-2 braneworlds
and their cosmology. JCAP, 1001:022, 2010, 0907.1640.
[278] C. Charmousis, G. Niz, A. Padilla, and P. M. Saffin. Strong coupling in Horava gravity. JHEP,
0908:070, 2009, 0905.2579.
[279] C. Charmousis and A. Padilla. The Instability of Vacua in Gauss-Bonnet Gravity. JHEP,
0812:038, 2008, 0807.2864.
[280] C. Charmousis and A. Papazoglou. Self-properties of codimension-2 braneworlds. JHEP, 07:062,
2008, 0804.2121.
[281] C. Charmousis and A. Papazoglou. Properties of codimension-2 braneworlds in six-dimensional
Lovelock theory. J. Phys. Conf. Ser., 189:012007, 2009, 0902.2174.
[282] C. Charmousis and R. Zegers. Einstein gravity on an even codimension brane. Phys.Rev.,
D72:064005, 2005, hep-th/0502171.
[283] C. Charmousis and R. Zegers. Matching conditions for a brane of arbitrary codimension. JHEP,
0508:075, 2005, hep-th/0502170.
[284] B. Chen, S. Pi, and J.-Z. Tang. Scale Invariant Power Spectrum in Horava-Lifshitz Cosmology
without Matter. JCAP, 0908:007, 2009, 0905.2300.
[285] C.-M. Chen, D. V. Gal’Tsov, N. Ohta, and D. G. Orlov. Global solutions for higher-dimensional
stretched small black holes. Phys. Rev. D, 81(2):024002, Jan. 2010, 0910.3488.
[286] C.-M. Chen, T. Harko, W. F. Kao, and M. K. Mak. Rotational perturbations of Friedmann-
Robertson-Walker type brane-world cosmological models. Nucl. Phys., B636:159–178, 2002, hep-
th/0201012.
[287] D.-M. Chen. Strong lensing probability in TeVeS theory. JCAP, 0801:006, 2008, 0712.1633.
[288] D.-M. Chen and H. Zhao. Strong gravitational lens probability in TeVeS. Astrophys. J., 650:L9–
L12, 2006, astro-ph/0606506.
[289] S. Chen, J. B. Dent, S. Dutta, and E. N. Saridakis. Cosmological perturbations in f(T) gravity.
ArXiv e-prints, Aug. 2010, 1008.1250.
276
[290] X. Chen and M. Kamionkowski. Cosmic microwave background temperature and polarization
anisotropy in Brans-Dicke cosmology. Phys. Rev. D, 60(10):104036, Nov. 1999, arXiv:astro-
ph/9905368.
[291] J. Chiaverini, S. J. Smullin, A. A. Geraci, D. M. Weld, and A. Kapitulnik. New Experimental
Constraints on Non-Newtonian Forces below 100 µm. Physical Review Letters, 90(15):151101,
Apr. 2003, arXiv:hep-ph/0209325.
[292] T. Chiba. 1/R gravity and scalar-tensor gravity. Physics Letters B, 575:1–3, Nov. 2003,
arXiv:astro-ph/0307338.
[293] T. Chiba. Generalized gravity and a ghost. J. Cosmology Astropart. Phys., 3:8, Mar. 2005,
arXiv:gr-qc/0502070.
[294] T. Chiba, T. L. Smith, and A. L. Erickcek. Solar system constraints to general f (r) gravity.
Phys. Rev. D, 75(12):124014, Jun 2007.
[295] M.-C. Chiu, C.-M. Ko, and Y. Tian. Theoretical aspects of gravitational lensing in TeVeS.
Astrophys. J., 636:565–574, 2006, astro-ph/0507332.
[296] M. C. Chiu, Y. Tian, and C. M. Ko. Necessity of Dark Matter in Modified Newtonian Dynamics
within Galactic Scales? - Testing the Covariant MOND in Elliptical Lenses. 2008, 0812.5011.
[297] G. Chkareuli and D. Pirtskhalava. Vainshtein Mechanism In Λ3 - Theories. 2011, 1105.1783.
[298] A. S. Chou, W. Wester, A. Baumbaugh, H. R. Gustafson, Y. Irizarry-Valle, P. O. Mazur, J. H.
Steffen, R. Tomlin, A. Upadhye, A. Weltman, X. Yang, and J. Yoo. Search for Chameleon
Particles Using a Photon-Regeneration Technique. Physical Review Letters, 102(3):030402, Jan.
2009, 0806.2438.
[299] N. Chow and J. Khoury. Galileon Cosmology. Phys.Rev., D80:024037, 2009, 0905.1325.
[300] D. J. H. Chung and K. Freese. Lensed density perturbations in braneworlds: An alternative to
perturbations from inflation. Phys. Rev., D67:103505, 2003, astro-ph/0202066.
[301] T. E. Chupp, R. J. Hoare, R. A. Loveman, E. R. Oteiza, J. M. Richardson, M. E. Wagshul, and
A. K. Thompson. Results of a new test of local Lorentz invariance: A search for mass anisotropy
in 21 Ne. Physical Review Letters, 63:1541–1545, Oct. 1989.
[302] I. Ciufolini and E. C. Pavlis. A confirmation of the general relativistic prediction of the Lense-
Thirring effect. Nature, 431:958–960, Oct. 2004.
[303] C. A. Clarkson, A. A. Coley, and E. S. D. O’Neill. Cosmic microwave background and scalar-
tensor theories of gravity. Phys. Rev. D, 64(6):063510, Sept. 2001, arXiv:gr-qc/0105026.
[304] T. Clifton. Spherically symmetric solutions to fourth-order theories of gravity. Classical and
Quantum Gravity, 23:7445–7453, Dec. 2006, arXiv:gr-qc/0607096.
[305] T. Clifton. Exact Friedmann solutions in higher-order gravity theories. Classical and Quantum
Gravity, 24:5073–5091, Oct. 2007, arXiv:gr-qc/0703126.
[306] T. Clifton. Higher powers in gravitation. Phys. Rev. D, 78(8):083501, Oct. 2008, 0807.4682.
[307] T. Clifton. Parametrized post-Newtonian limit of fourth-order theories of gravity. Phys. Rev. D,
77(2):024041, Jan. 2008, 0801.0983.
[308] T. Clifton, M. Bañados, and C. Skordis. The Parameterized Post-Newtonian Limit of Bimetric
Theories of Gravity. ArXiv e-prints, June 2010, 1006.5619.
[309] T. Clifton and J. D. Barrow. The existence of Gödel, Einstein, and de Sitter universes.
Phys. Rev. D, 72(12):123003, Dec. 2005, arXiv:gr-qc/0511076.
[310] T. Clifton and J. D. Barrow. The power of general relativity. Phys. Rev. D, 72(10):103005, Nov.
2005, arXiv:gr-qc/0509059.
[311] T. Clifton and J. D. Barrow. Further exact cosmological solutions to higher-order gravity theories.
Classical and Quantum Gravity, 23:2951–2962, May 2006, arXiv:gr-qc/0601118.
[312] T. Clifton, J. D. Barrow, and R. J. Scherrer. Constraints on the Variation of G from Primordial
Nucleosynthesis. Phys. Rev., D71:123526, 2005, astro-ph/0504418.
[313] T. Clifton, D. F. Mota, and J. D. Barrow. Inhomogeneous gravity. Mon. Not. Roy. Astron. Soc.,
358:601, 2005, gr-qc/0406001.
[314] T. Clifton and J. M. Weisberg. A Simple Model for Pulse Profiles from Precessing Pulsars,
with Special Application to Relativistic Binary PSR B1913+16. Astrophys. J., 679:687, 2008,
0708.0993.
[315] T. Clifton and T. G. Zlosnik. FRW cosmology in Milgrom’s bimetric theory of gravity.
Phys. Rev. D, 81(10):103525, May 2010, 1002.1448.
[316] J. M. Cline, J. Descheneau, M. Giovannini, and J. Vinet. Cosmology of codimension-two
braneworlds. JHEP, 06:048, 2003, hep-th/0304147.
[317] J. M. Cline, S. Jeon, and G. D. Moore. The phantom menaced: Constraints on low-energy
effective ghosts. Phys. Rev., D70:043543, 2004, hep-ph/0311312.
277
[318] D. Clowe, M. Bradac, A. H. Gonzalez, M. Markevitch, S. W. Randall, et al. A direct empirical
proof of the existence of dark matter. Astrophys.J., 648:L109–L113, 2006, astro-ph/0608407.
[319] A. Coc, K. A. Olive, J. Uzan, and E. Vangioni. Big bang nucleosynthesis constraints on scalar-
tensor theories of gravity. Phys. Rev. D, 73(8):083525, Apr. 2006, arXiv:astro-ph/0601299.
[320] D. Codazzi. Sulle coordinate curvilinee d’una superficie e dello spazio. Annali di Matematica
Pura ed Applicata (1867-1897), 1(1):293–316, 1867.
[321] G. Cognola, E. Elizalde, S. Nojiri, S. D. Odintsov, L. Sebastiani, and S. Zerbini. Class of viable
modified f(R) gravities describing inflation and the onset of accelerated expansion. Phys. Rev. D,
77(4):046009, Feb. 2008, 0712.4017.
[322] G. Cognola, E. Elizalde, S. Nojiri, S. D. Odintsov, and S. Zerbini. One-loop f(R) gravity in de
Sitter universe. J. Cosmology Astropart. Phys., 2:10, Feb. 2005, arXiv:hep-th/0501096.
[323] G. Cognola, E. Elizalde, S. Nojiri, S. D. Odintsov, and S. Zerbini. Dark energy in modified Gauss-
Bonnet gravity: Late-time acceleration and the hierarchy problem. Phys. Rev. D, 73(8):084007,
Apr. 2006, arXiv:hep-th/0601008.
[324] G. Cognola, E. Elizalde, S. Nojiri, S. D. Odintsov, and S. Zerbini. String-inspired Gauss-Bonnet
gravity reconstructed from the universe expansion history and yielding the transition from matter
dominance to dark energy. Phys. Rev. D, 75(8):086002, Apr. 2007, arXiv:hep-th/0611198.
[325] G. Cognola, M. Gastaldi, and S. Zerbini. On the Stability of a Class of Modified Gravita-
tional Models. International Journal of Theoretical Physics, 47:898–910, Apr. 2008, arXiv:gr-
qc/0701138.
[326] G. Cognola and S. Zerbini. Homogeneous Cosmologies in Generalized Modified Gravity. Inter-
national Journal of Theoretical Physics, 47:3186–3200, Dec. 2008, 0802.3967.
[327] A. A. Coley. Qualitative Properties of Scalar-Tensor Theories of Gravity. General Relativity and
Gravitation, 31:1295, Sept. 1999, arXiv:astro-ph/9910395.
[328] P. D. B. Collins and R. F. Langbein. Thermodynamic aspects of extended inflation. Phys. Rev. D,
47:2302–2312, Mar. 1993.
[329] D. Comelli, M. Crisostomi, F. Nesti, and L. Pilo. Finite energy for a gravitational potential
falling slower than 1/r. Phys. Rev. D, 84(10):104026, Nov. 2011, 1105.3010.
[330] A. Conley, J. Guy, M. Sullivan, N. Regnault, P. Astier, C. Balland, S. Basa, R. G. Carlberg,
D. Fouchez, D. Hardin, I. M. Hook, D. A. Howell, R. Pain, N. Palanque-Delabrouille, K. M. Per-
rett, C. J. Pritchet, J. Rich, V. Ruhlmann-Kleider, D. Balam, S. Baumont, R. S. Ellis, S. Fabbro,
H. K. Fakhouri, N. Fourmanoit, S. González-Gaitán, M. L. Graham, M. J. Hudson, E. Hsiao,
T. Kronborg, C. Lidman, A. M. Mourao, J. D. Neill, S. Perlmutter, P. Ripoche, N. Suzuki, and
E. S. Walker. Supernova Constraints and Systematic Uncertainties from the First Three Years
of the Supernova Legacy Survey. ApJS, 192:1, Jan. 2011.
[331] C. R. Contaldi, T. Wiseman, and B. Withers. TeVeS gets caught on caustics. Phys. Rev.,
D78:044034, 2008, 0802.1215.
[332] A. Cooney, S. Dedeo, and D. Psaltis. Neutron stars in f(R) gravity with perturbative constraints.
Phys. Rev. D, 82(6):064033, Sept. 2010, 0910.5480.
[333] E. J. Copeland, E. W. Kolb, and A. R. Liddle. Topological defects in extended inflation.
Phys. Rev. D, 42:2911–2914, Oct. 1990.
[334] E. J. Copeland, A. R. Liddle, and J. E. Lidsey. Steep inflation: Ending braneworld inflation by
gravitational particle production. Phys. Rev., D64:023509, 2001, astro-ph/0006421.
[335] E. J. Copeland and D. J. Toms. Quantized Antisymmetric Tensor Fields And Selfconsistent
Dimensional Reduction In Higher Dimensional Space-Times. Nucl. Phys., B255:201, 1985.
[336] E. J. Copeland and D. J. Toms. Stability Of Selfconsistent Higher Dimensional Cosmological
Solutions. Phys. Rev., D32:1921, 1985.
[337] C. J. Copi, D. Huterer, D. J. Schwarz, and G. D. Starkman. Large angle anomalies in the CMB.
Adv.Astron., 2010:847541, 2010, 1004.5602.
[338] O. Corradini, K. Koyama, and G. Tasinato. Induced gravity on intersecting brane-worlds Part I:
Maximally symmetric solutions. Phys. Rev., D77:084006, 2008, 0712.0385.
[339] O. Corradini, K. Koyama, and G. Tasinato. Induced gravity on intersecting brane-worlds Part
II: Cosmology. Phys. Rev., D78:124002, 2008, 0803.1850.
[340] S. Cotsakis, J. Demaret, Y. De Rop, and L. Querella. Mixmaster universe in fourth-order gravity
theories. Phys. Rev. D, 48(10):4595–4603, Nov 1993.
[341] P. Creminelli, G. D’Amico, M. Musso, J. Norena, and E. Trincherini. Galilean symmetry in the
effective theory of inflation: new shapes of non-Gaussianity. 2010, 1011.3004.
[342] P. Creminelli, A. Nicolis, M. Papucci, and E. Trincherini. Ghosts in massive gravity. JHEP,
0509:003, 2005, hep-th/0505147.
278
[343] P. Creminelli, A. Nicolis, and E. Trincherini. Galilean Genesis: An Alternative to inflation.
JCAP, 1011:021, 2010, 1007.0027.
[344] J. Crisostomo, R. Troncoso, and J. Zanelli. Black hole scan. Phys.Rev., D62:084013, 2000,
hep-th/0003271.
[345] R. Crittenden and P. J. Steinhardt. Graceful exit in extended inflation and implications for
density perturbations. Physics Letters B, 293:32–36, Oct. 1992, arXiv:astro-ph/9207002.
[346] C. Csaki, M. Graesser, L. Randall, and J. Terning. Cosmology of brane models with radion
stabilization. Phys. Rev., D62:045015, 2000, hep-ph/9911406.
[347] C. Csaki, M. L. Graesser, and G. D. Kribs. Radion dynamics and electroweak physics. Phys.
Rev., D63:065002, 2001, hep-th/0008151.
[348] B. Cuadros-Melgar, E. Papantonopoulos, M. Tsoukalas, and V. Zamarias. Black Holes on Thin
3-branes of Codimension-2 and their Extension into the Bulk. Nucl. Phys., B810:246–265, 2009,
0804.4459.
[349] B. Cuadros-Melgar, E. Papantonopoulos, M. Tsoukalas, and V. Zamarias. Perturbations of
Gauss-Bonnet Black Strings in Codimension-2 Braneworlds. 2010, 1012.4747.
[350] S. Cullen and M. Perelstein. SN1987A constraints on large compact dimensions. Phys. Rev. Lett.,
83:268–271, 1999, hep-ph/9903422.
[351] E. Czuchry. Bounce scenarios in modified Horava-Lifshitz cosmology. 2010, 1008.3410.
[352] J. da Angela, T. Shanks, S. Croom, P. Weilbacher, R. Brunner, et al. The 2dF-SDSS LRG and
QSO Survey: QSO clustering and the L-z degeneracy. Mon.Not.Roy.Astron.Soc., 383:565–580,
2008, astro-ph/0612401.
[353] A. M. da Silva. An Alternative Approach for General Covariant Horava-Lifshitz Gravity and
Matter Coupling. 2010, 1009.4885.
[354] T. Damour and C. Gundlach. Nucleosynthesis constraints on an extended Jordan-Brans-Dicke
theory. Phys. Rev. D, 43(12):3873–3877, Jun 1991.
[355] T. Damour, I. I. Kogan, and A. Papazoglou. Nonlinear bigravity and cosmic acceleration.
Phys. Rev. D, 66(10):104025, Nov. 2002, arXiv:hep-th/0206044.
[356] T. Damour and K. Nordtvedt. General relativity as a cosmological attractor of tensor-scalar
theories. Phys. Rev. Lett., 70(15):2217–2219, Apr 1993.
[357] T. Damour and K. Nordtvedt. Tensor-scalar cosmological models and their relaxation toward
general relativity. Phys. Rev. D, 48(8):3436–3450, Oct 1993.
[358] T. Damour and B. Pichon. Big bang nucleosynthesis and tensor-scalar gravity. Phys. Rev. D,
59(12):123502, May 1999.
[359] T. Damour and J. Taylor. Strong-field tests of relativistic gravity and binary pulsars. Phys. Rev.
D, 45:1840–1868, 1992.
[360] S. F. Daniel, R. R. Caldwell, A. Cooray, and A. Melchiorri. Large scale structure as a probe of
gravitational slip. Phys. Rev. D, 77(10):103513, May 2008, 0802.1068.
[361] S. F. Daniel et al. Testing general relativity with current cosmological data. Phys. Rev.,
D81:123508, 2010. arXiv:1002.1962.
[362] A. Davis, C. A. O. Schelpe, and D. J. Shaw. The Chameleonic Contribution to the SZ Radial
Profile of the Coma Cluster. ArXiv e-prints, Aug. 2010, 1008.1880.
[363] S. C. Davis. Generalized Israel junction conditions for a Gauss-Bonnet brane world. Phys.Rev.,
D67:024030, 2003, hep-th/0208205.
[364] S. C. Davis. Solar System Constraints on f(G) Dark Energy. ArXiv e-prints, Sept. 2007, 0709.4453.
[365] H. Davoudiasl, D. E. Morrissey, K. Sigurdson, and S. Tulin. Unified origin for baryonic visible
matter and antibaryonic dark matter. Phys. Rev. Lett., 105(21):211304, Nov 2010.
[366] W. de Blok. The Core-Cusp Problem. Advances in Astronomy, page 789293, 2010, 0910.3538.
[367] J. A. de Deus and D. Müller. Bianchi VIIA solutions of quadratic gravity. ArXiv e-prints, Mar.
2011, 1103.5450.
[368] A. De Felice, J.-M. Gérard, and T. Suyama. Cosmological perturbation in f(R,G) theories with
a perfect fluid. Phys. Rev. D, 82(6):063526, Sep 2010.
[369] A. De Felice, R. Kase, and S. Tsujikawa. Matter perturbations in Galileon cosmology. 2010,
1011.6132.
[370] A. de Felice, D. F. Mota, and S. Tsujikawa. Matter instabilities in general Gauss-Bonnet gravity.
Phys. Rev. D, 81(2):023532, Jan. 2010, 0911.1811.
[371] A. de Felice and S. Tsujikawa. Construction of cosmologically viable f(G) gravity models. Physics
Letters B, 675:1–8, May 2009, 0810.5712.
[372] A. de Felice and S. Tsujikawa. Solar system constraints on f(G) gravity models. Phys. Rev. D,
80(6):063516, Sept. 2009, 0907.1830.
279
[373] A. De Felice and S. Tsujikawa. Cosmology of a covariant Galileon field. Phys.Rev.Lett.,
105:111301, 2010, 1007.2700.
[374] A. de Felice and S. Tsujikawa. f(R) Theories. Living Reviews in Relativity, 13:3, June 2010,
1002.4928.
[375] A. De Felice and S. Tsujikawa. Generalized Galileon cosmology. 2010, 1008.4236.
[376] Á. de La Cruz-Dombriz and A. Dobado. f(R) gravity without a cosmological constant.
Phys. Rev. D, 74(8):087501, Oct. 2006, arXiv:gr-qc/0607118.
[377] A. de La Cruz-Dombriz, A. Dobado, and A. L. Maroto. Evolution of density perturbations in
f(R) theories of gravity. Phys. Rev. D, 77(12):123515, June 2008, 0802.2999.
[378] A. de La Cruz-Dombriz, A. Dobado, and A. L. Maroto. Black holes in f(R) theories. Phys. Rev. D,
80(12):124011, Dec. 2009, 0907.3872.
[379] C. de Rham. An Introduction to Cascading Gravity and Degravitation. Can. J. Phys., 87:201–
203, 2009, 0810.0269.
[380] C. de Rham. Massive gravity from Dirichlet boundary conditions. Phys.Lett., B688:137–141,
2010, 0910.5474.
[381] C. de Rham et al. Cascading gravity: Extending the Dvali-Gabadadze-Porrati model to higher
dimension. Phys. Rev. Lett., 100:251603, 2008, 0711.2072.
[382] C. de Rham and G. Gabadadze. Generalization of the Fierz-Pauli Action. Phys.Rev., D82:044020,
2010, 1007.0443.
[383] C. de Rham and G. Gabadadze. Selftuned Massive Spin-2. Phys.Lett., B693:334–338, 2010,
1006.4367.
[384] C. de Rham, G. Gabadadze, L. Heisenberg, and D. Pirtskhalava. Cosmic Acceleration and the
Helicity-0 Graviton. 2010, 1010.1780.
[385] C. de Rham, G. Gabadadze, and A. J. Tolley. Resummation of Massive Gravity. 2010, 1011.1232.
[386] C. de Rham, S. Hofmann, J. Khoury, and A. J. Tolley. Cascading Gravity and Degravitation.
JCAP, 0802:011, 2008, 0712.2821.
[387] C. de Rham, J. Khoury, and A. J. Tolley. Flat 3-Brane with Tension in Cascading Gravity. Phys.
Rev. Lett., 103:161601, 2009, 0907.0473.
[388] C. de Rham, J. Khoury, and A. J. Tolley. Cascading Gravity is Ghost Free. Phys.Rev.,
D81:124027, 2010, 1002.1075.
[389] C. de Rham and A. J. Tolley. DBI and the Galileon reunited. JCAP, 1005:015, 2010, 1003.5917.
[390] J. C. C. de Souza and V. Faraoni. The phase-space view of f(R) gravity. Classical and Quantum
Gravity, 24:3637–3648, July 2007, 0706.1223.
[391] A. DeFelice and M. Hindmarsh. Unsuccessful cosmology with modified gravity models. J. Cos-
mology Astropart. Phys., 6:28, June 2007, 0705.3375.
[392] A. DeFelice, M. Hindmarsh, and M. Trodden. Ghosts, instabilities, and superluminal propaga-
tion in modified gravity models. J. Cosmology Astropart. Phys., 8:5, Aug. 2006, arXiv:astro-
ph/0604154.
[393] A. DeFelice and T. Suyama. Vacuum structure for scalar cosmological perturbations in modified
gravity models. J. Cosmology Astropart. Phys., 6:34, June 2009, 0904.2092.
[394] C. Deffayet. Cosmology on a brane in Minkowski bulk. Phys.Lett., B502:199–208, 2001, hep-
th/0010186.
[395] C. Deffayet. On brane world cosmological perturbations. Phys.Rev., D66:103504, 2002, hep-
th/0205084.
[396] C. Deffayet, S. Deser, and G. Esposito-Farese. Generalized Galileons: All scalar models
whose curved background extensions maintain second-order field equations and stress-tensors.
Phys.Rev., D80:064015, 2009, 0906.1967.
[397] C. Deffayet, S. Deser, and G. Esposito-Farese. Arbitrary p-form Galileons. Phys.Rev.,
D82:061501, 2010, 1007.5278.
[398] C. Deffayet, G. R. Dvali, and G. Gabadadze. Accelerated universe from gravity leaking to extra
dimensions. Phys. Rev., D65:044023, 2002, astro-ph/0105068.
[399] C. Deffayet, G. R. Dvali, G. Gabadadze, and A. I. Vainshtein. Nonperturbative continuity in
graviton mass versus perturbative discontinuity. Phys. Rev., D65:044026, 2002, hep-th/0106001.
[400] C. Deffayet, G. Esposito-Farese, and A. Vikman. Covariant Galileon. Phys.Rev., D79:084003,
2009, 0901.1314.
[401] C. Deffayet, G. Gabadadze, and A. Iglesias. Perturbations of Self-Accelerated Universe. JCAP,
0608:012, 2006, hep-th/0607099.
[402] C. Deffayet, X. Gao, D. A. Steer, and G. Zahariade. From k-essence to generalised Galileons.

280
 2011, 1103.3260.
[403] C. Deffayet, S. J. Landau, J. Raux, M. Zaldarriaga, and P. Astier. Supernovae, CMB, and
gravitational leakage into extra dimensions. Phys.Rev., D66:024019, 2002, astro-ph/0201164.
[404] C. Deffayet, O. Pujolas, I. Sawicki, and A. Vikman. Imperfect Dark Energy from Kinetic Gravity
Braiding. JCAP, 1010:026, 2010, 1008.0048.
[405] C. Deffayet and J.-W. Rombouts. Ghosts, strong coupling and accidental symmetries in massive
gravity. Phys. Rev., D72:044003, 2005, gr-qc/0505134.
[406] S. del Campo. Initial conditions for anisotropic extended-type inflationary universes.
Phys. Rev. D, 45:3386–3393, May 1992.
[407] S. del Campo and A. Vilenkin. Initial conditions for extended inflation. Phys. Rev. D, 40:688–690,
July 1989.
[408] N. Deruelle, T. Dolezel, and J. Katz. Perturbations of brane worlds. Phys. Rev., D63:083513,
2001, hep-th/0010215.
[409] N. Deruelle and C. Germani. Smooth branes and junction conditions in Einstein Gauss-Bonnet
gravity. Nuovo Cim., B118:977–988, 2003, gr-qc/0306116.
[410] N. Deruelle, C. Gundlach, and D. Langlois. Perturbations of extended inflation. Phys. Rev. D,
45:3301, May 1992.
[411] N. Deruelle, C. Gundlach, and D. Langlois. Vacuum density fluctuations in extended chaotic
inflation. Phys. Rev. D, 46:5337–5345, Dec. 1992.
[412] N. Deruelle, C. Gundlach, and D. Polarski. The growth of density fluctuations in a simplified
model of extended inflation. Classical and Quantum Gravity, 9:1511–1524, June 1992.
[413] N. Deruelle and M. Sasaki. Newton’s law on an Einstein ‘Gauss-Bonnet’ brane. Prog.Theor.Phys.,
110:441–456, 2003, gr-qc/0306032.
[414] S. Desai, E. O. Kahya, and R. P. Woodard. Reduced time delay for gravitational waves with
dark matter emulators. Phys. Rev., D77:124041, 2008, 0804.3804.
[415] S. Deser and B. Tekin. Gravitational energy in quadratic curvature gravities. Phys.Rev.Lett.,
89:101101, 2002, hep-th/0205318.
[416] S. Deser and B. Tekin. Energy in generic higher curvature gravity theories. Phys.Rev.,
D67:084009, 2003, hep-th/0212292.
[417] A. Dev, D. Jain, S. Jhingan, S. Nojiri, M. Sami, and I. Thongkool. Delicate f(R) gravity models
with a disappearing cosmological constant and observational constraints on the model parameters.
Phys. Rev. D, 78(8):083515, Oct. 2008, 0807.3445.
[418] B. DeWitt. Dynamical theory of groups and fields. Gordon and Breach, 1965.
[419] L. M. Diaz-Rivera, L. Samushia, and B. Ratra. Inflation and accelerated expansion TeVeS cos-
mological solutions. Phys. Rev., D73:083503, 2006, astro-ph/0601153.
[420] R. Dick. Standard cosmology in the DGP brane model. Acta Phys.Polon., B32:3669–3682, 2001,
hep-th/0110162.
[421] R. Dick. Letter: On the Newtonian Limit in Gravity Models with Inverse Powers of R. General
Relativity and Gravitation, 36:217–224, Jan. 2004, arXiv:gr-qc/0307052.
[422] R. H. Dicke. Mach’s principle and invariance under transformation of units. Phys. Rev., 125:2163–
2167, 1962.
[423] R. H. Dicke. The Theoretical Significance of Experimental Relativity. American Journal of
Physics, 34:369–370, Apr. 1966.
[424] P. A. M. Dirac. A new classical theory of electrons I. Proc. Roy. Soc., A209:291, 1951.
[425] P. A. M. Dirac. A new classical theory of electrons II. Proc. Roy. Soc., A212:330, 1952.
[426] P. A. M. Dirac. A new classical theory of electrons III. Proc. Roy. Soc., A223:438, 1954.
[427] S. Dodelson and M. Liguori. Can Cosmic Structure form without Dark Matter? Phys. Rev. Lett.,
97:231301, 2006, astro-ph/0608602.
[428] A. D. Dolgov and M. Kawasaki. Can modified gravity explain accelerated cosmic expansion?
Physics Letters B, 573:1–4, Oct. 2003, arXiv:astro-ph/0307285.
[429] A. E. Domı́nguez and D. E. Barraco. Newtonian limit of the singular f(R) gravity in the Palatini
formalism. Phys. Rev. D, 70(4):043505, Aug. 2004, arXiv:gr-qc/0408069.
[430] M. Dorca and C. van de Bruck. Cosmological perturbations in brane worlds: Brane bending and
anisotropic stresses. Nucl. Phys., B605:215–233, 2001, hep-th/0012116.
[431] J. Dossett, M. Ishak, J. Moldenauer, Y. Gong and A. Wang. Constraints on growth index
parameters from current and future observations. JCAP, 1004:022, 2010, arXiv:1004.3086.
[432] J. Dossett, J. Moldenauer and M. Ishak. Figures of merit and constraints from testing general
relativity using the latest cosmological data set including refined COSMOS 3D weak lensing.
Phys. Rev. D., 84(2):023012, 2011, arXiv:1103.1195.
281
[433] R. W. P. Drever. A search for anisotropy of inertial mass using a free precession technique.
Philosophical Magazine, 6:683–687, May 1961.
[434] I. T. Drummond. Bimetric gravity and “dark matter”. Phys. Rev. D, 63(4):043503, Feb. 2001,
arXiv:astro-ph/0008234.
[435] S. Dubovsky and V. Rubakov. On models of gauge field localization on a brane. Int.J.Mod.Phys.,
A16:4331–4350, 2001, hep-th/0105243.
[436] S. Dubovsky and V. Rubakov. Brane induced gravity in more than one extra dimensions: Vio-
lation of equivalence principle and ghost. Phys.Rev., D67:104014, 2003, hep-th/0212222.
[437] J.-F. Dufaux, J. E. Lidsey, R. Maartens, and M. Sami. Cosmological perturbations from brane
inflation with a Gauss-Bonnet term. Phys.Rev., D70:083525, 2004, hep-th/0404161.
[438] M. J. Duff and J. T. Liu. Complementarity of the Maldacena and Randall-Sundrum pictures.
Class. Quant. Grav., 18:3207–3214, 2001, hep-th/0003237.
[439] M. J. Duff, J. T. Liu, and K. S. Stelle. A supersymmetric type IIB Randall-Sundrum realization.
J. Math. Phys., 42:3027–3047, 2001, hep-th/0007120.
[440] M. J. Duff and C. N. Pope. Consistent truncations in Kaluza-Klein theories. Nucl. Phys.,
B255:355–364, 1985.
[441] M. J. Duncan and L. Gerhard Jensen. Extended inflation in closed cosmologies. Physics Letters
B, 247:246–250, Sept. 1990.
[442] J. Dunkley, R. Hlozek, J. Sievers, V. Acquaviva, P. A. R. Ade, P. Aguirre, M. Amiri, J. W.
Appel, L. F. Barrientos, E. S. Battistelli, J. R. Bond, B. Brown, B. Burger, J. Chervenak,
S. Das, M. J. Devlin, S. R. Dicker, W. Bertrand Doriese, R. Dunner, T. Essinger-Hileman, R. P.
Fisher, J. W. Fowler, A. Hajian, M. Halpern, M. Hasselfield, C. Hernandez-Monteagudo, G. C.
Hilton, M. Hilton, A. D. Hincks, K. M. Huffenberger, D. H. Hughes, J. P. Hughes, L. Infante,
K. D. Irwin, J. B. Juin, M. Kaul, J. Klein, A. Kosowsky, J. M. Lau, M. Limon, Y. Lin, R. H.
Lupton, T. A. Marriage, D. Marsden, P. Mauskopf, F. Menanteau, K. Moodley, H. Moseley,
C. B. Netterfield, M. D. Niemack, M. R. Nolta, L. A. Page, L. Parker, B. Partridge, B. Reid,
N. Sehgal, B. Sherwin, D. N. Spergel, S. T. Staggs, D. S. Swetz, E. R. Switzer, R. Thornton,
H. Trac, C. Tucker, R. Warne, E. Wollack, and Y. Zhao. The Atacama Cosmology Telescope:
Cosmological Parameters from the 2008 Power Spectra. ArXiv e-prints, Sept. 2010, 1009.0866.
[443] P. K. S. Dunsby, E. Elizalde, R. Goswami, S. Odintsov, and D. Saez-Gomez. ΛCDM universe in
f(R) gravity. Phys. Rev. D, 82(2):023519, July 2010, 1005.2205.
[444] R. Durrer, M. Kunz, and M. Sakellariadou. Why do we live in 3+1 dimensions? Phys. Lett.,
B614:125–130, 2005, hep-th/0501163.
[445] S. Dutta and E. N. Saridakis. Observational constraints on Horava-Lifshitz cosmology. JCAP,
1001:013, 2010, 0911.1435.
[446] S. Dutta and E. N. Saridakis. Overall observational constraints on the running parameter λ of
Horava-Lifshitz gravity. JCAP, 1005:013, 2010, 1002.3373.
[447] G. Dvali. Predictive Power of Strong Coupling in Theories with Large Distance Modified Gravity.
New J. Phys., 8:326, 2006, hep-th/0610013.
[448] G. Dvali and G. Gabadadze. Gravity on a brane in infinite volume extra space. Phys.Rev.,
D63:065007, 2001, hep-th/0008054.
[449] G. Dvali, G. Gabadadze, X.-r. Hou, and E. Sefusatti. See-saw modification of gravity. Phys.
Rev., D67:044019, 2003, hep-th/0111266.
[450] G. Dvali, G. Gabadadze, and M. Shifman. Diluting cosmological constant in infinite volume extra
dimensions. Phys.Rev., D67:044020, 2003, hep-th/0202174.
[451] G. Dvali, A. Gruzinov, and M. Zaldarriaga. The Accelerated universe and the moon. Phys.Rev.,
D68:024012, 2003, hep-ph/0212069.
[452] G. Dvali, S. Hofmann, and J. Khoury. Degravitation of the cosmological constant and graviton
width. Phys.Rev., D76:084006, 2007, hep-th/0703027.
[453] G. Dvali and M. S. Turner. Dark energy as a modification of the Friedmann equation. 2003,
astro-ph/0301510.
[454] G. R. Dvali, G. Gabadadze, and M. Porrati. 4D gravity on a brane in 5D Minkowski space. Phys.
Lett., B485:208–214, 2000, hep-th/0005016.
[455] D. A. Easson. Modified Gravitational Theories and Cosmic Acceleration. International Journal
of Modern Physics A, 19:5343–5350, 2004, arXiv:astro-ph/0411209.
[456] R. Easther, B. R. Greene, M. G. Jackson, and D. N. Kabat. Brane gas cosmology in M-theory:
Late time behavior. Phys. Rev., D67:123501, 2003, hep-th/0211124.
[457] R. Easther, B. R. Greene, M. G. Jackson, and D. N. Kabat. Brane gases in the early universe:
Thermodynamics and cosmology. JCAP, 0401:006, 2004, hep-th/0307233.
282
[458] R. Easther, B. R. Greene, M. G. Jackson, and D. N. Kabat. String windings in the early universe.
JCAP, 0502:009, 2005, hep-th/0409121.
[459] D. H. Eckhardt, C. Jekeli, A. R. Lazarewicz, A. J. Romaides, and R. W. Sands. Tower gravity
experiment - Evidence for non-Newtonian gravity. Physical Review Letters, 60:2567–2570, June
1988.
[460] A. S. Eddington. The Mathematical Theory of Relativity. 1923.
[461] M. Eingorn, A. Zhuk, Kaluza-Klein models: can we construct a viable example? Phys. Rev.
D83 (2011) 044005. [arXiv:1010.5740 [gr-qc]].
[462] M. Eingorn, O. R. de Medeiros, L. C. B. Crispino, A. Zhuk, Latent solitons, black strings,
black branes, and equations of state in Kaluza-Klein models Phys. Rev. D84 (2011) 024031.
[arXiv:1101.3910 [gr-qc]].
[463] G. Efstathiou, W. J. Sutherland, and S. J. Maddox. The cosmological constant and cold dark
matter. Nature, 348:705–707, Dec. 1990.
[464] A. Einstein. On The influence of gravitation on the propagation of light. Annalen Phys., 35:898–
908, 1911.
[465] A. Einstein. The foundation of the general theory of relativity. Annalen Phys., 49:769–822, 1916.
[466] A. Einstein and W. Pauli. On the non-existence of regular stationary solutions of relativistic field
equations. The Annals of Mathematics, 44(2):pp. 131–137, 1943.
[467] D. J. Eisenstein et al. Detection of the baryon acoustic peak in the large-scale correlation function
of SDSS luminous red galaxies. Astrophys.J., 633:560–574, 2005, astro-ph/0501171.
[468] E. Elizalde, S. Nojiri, S. D. Odintsov, L. Sebastiani, and S. Zerbini. Nonsingular exponential grav-
ity: A simple theory for early- and late-time accelerated expansion. Phys. Rev. D, 83(8):086006,
2011, 1012.2280.
[469] G. F. R. Ellis and M. Bruni. Covariant and gauge-invariant approach to cosmological density
fluctuations. Phys. Rev. D, 40:1804–1818, Sept. 1989.
[470] R. Emparan, A. Fabbri, and N. Kaloper. Quantum black holes as holograms in AdS braneworlds.
JHEP, 08:043, 2002, hep-th/0206155.
[471] R. Emparan, J. Garcia-Bellido, and N. Kaloper. Black hole astrophysics in AdS braneworlds.
JHEP, 01:079, 2003, hep-th/0212132.
[472] R. V. Eotvos, D. Pekar, and E. Fekete. Contributions to the law of propertionality of inertia and
gravity. Annalen Phys., 68:11–66, 1922.
[473] A. L. Erickcek, T. L. Smith, and M. Kamionkowski. Solar system tests do rule out 1/R gravity.
Phys. Rev. D, 74(12):121501, Dec. 2006, arXiv:astro-ph/0610483.
[474] J. Etherington. Phil. Mag., 15:761, 1933.
[475] J. D. Evans, L. M. H. Hall, and P. Caillol. Standard cosmological evolution in a wide range of
f(R) models. Phys. Rev. D, 77(8):083514, Apr. 2008, 0711.3695.
[476] C. W. F. Everitt et al. Gravity Probe B: Final Results of a Space Experiment to Test General
Relativity. Phys. Rev. Lett., 106:221101, 2011, 1105.3456.
[477] Q. Exirifard and M. M. Sheikh-Jabbari. Lovelock gravity at the crossroads of Palatini and metric
formulations. Physics Letters B, 661:158–161, Mar. 2008, 0705.1879.
[478] M. Fairbairn and A. Goobar. Supernova limits on brane world cosmology. Phys.Lett., B642:432–
435, 2006, astro-ph/0511029.
[479] M. Fairbairn and S. Rydbeck. Expansion history and f(R) modified gravity. J. Cosmology
Astropart. Phys., 12:5, Dec. 2007, arXiv:astro-ph/0701900.
[480] D. Fairlie and J. Govaerts. Euler hierarchies and universal equations. J.Math.Phys., 33:3543–
3566, 1992, hep-th/9204074. Revised version.
[481] D. Fairlie, J. Govaerts, and A. Morozov. Universal field equations with covariant solutions.
Nucl.Phys., B373:214–232, 1992, hep-th/9110022.
[482] W. Fang et al. Challenges to the DGP Model from Horizon-Scale Growth and Geometry. Phys.
Rev., D78:103509, 2008, 0808.2208.
[483] H. Farajollahi, M. Farhoudi, A. Salehi, and H. Shojaie. Chameleonic Generalized Brans–Dicke
model and late-time acceleration. ArXiv e-prints, Sep. 2010, 1009.5059.
[484] H. Farajollahi and A. Salehi. Cosmic Dynamics in Chameleon Cosmology. International Journal
of Modern Physics D, 19:621–633, 2010, 1004.3508.
[485] V. Faraoni. Matter instability in modified gravity. Phys. Rev. D, 74(10):104017, Nov. 2006,
arXiv:astro-ph/0610734.
[486] V. Faraoni. de Sitter space and the equivalence between f(R) and scalar-tensor gravity.
Phys. Rev. D, 75(6):067302, Mar. 2007, arXiv:gr-qc/0703044.
[487] V. Faraoni. Extension of the EGS theorem to metric and Palatini f(R) gravity. ArXiv e-prints,
283
 Nov. 2008, 0811.1870.
[488] V. Faraoni. Clifton’s spherical solution in f(R) vacuum harbours a naked singularity. Classical
and Quantum Gravity, 26(19):195013, Oct. 2009, 0909.0514.
[489] V. Faraoni. Black hole entropy in scalar-tensor and f(R) gravity: an overview. ArXiv e-prints,
May 2010, 1005.2327.
[490] V. Faraoni. Jebsen-Birkhoff theorem in alternative gravity. Phys. Rev. D, 81(4):044002, Feb.
2010, 1001.2287.
[491] M. Farhoudi. Classical Trace Anomaly. Int. J. Mod. Phys. D, 14:1233–1250, 2005.
[492] M. Farhoudi. On higher order gravities, their analogy to GR, and dimensional dependent version
of Duff’s trace anomaly relation. Gen. Rel. Grav., 38:1261–1284, Aug 2006.
[493] M. Farhoudi. Lovelock tensor as generalized Einstein tensor. Gen. Rel. Grav., 41:117–129, Jan
2009.
[494] M. Farhoudi. On Third Order Lagrangians, Weyl Invariants & Classical Trace Anomaly in Six
Dimensions. ArXiv e-prints, Jul. 2011, 1107.1034.
[495] T. Faulkner, M. Tegmark, E. F. Bunn, and Y. Mao. Constraining f(R) gravity as a scalar-tensor
theory. Phys. Rev. D, 76(6):063505, Sep 2007.
[496] S. Fay, S. Nesseris, and L. Perivolaropoulos. Can f(R) modified gravity theories mimic a ΛCDM
cosmology? Phys. Rev. D, 76(6):063504, Sept. 2007, arXiv:gr-qc/0703006.
[497] S. Fay, R. Tavakol, and S. Tsujikawa. f(R) gravity theories in Palatini formalism: Cosmological
dynamics and observational constraints. Phys. Rev. D, 75(6):063509, Mar. 2007, arXiv:astro-
ph/0701479.
[498] M. Feix, C. Fedeli, and M. Bartelmann. Asymmetric Gravitational Lenses in TeVeS and Appli-
cation to the Bullet Cluster. Astron. Astrophys., 480:313, 2008, 0707.0790.
[499] B. Feldman and A. E. Nelson. New regions for a chameleon to hide. Journal of High Energy
Physics, 8:2, Aug. 2006, arXiv:hep-ph/0603057.
[500] S. Ferraro, F. Schmidt, and W. Hu. Cluster Abundance in f(R) Gravity Models. ArXiv e-prints,
Nov. 2010, 1011.0992.
[501] P. G. Ferreira and C. Skordis. The linear growth rate of structure in Parametrized Post Fried-
mannian Universes. Phys.Rev., D81:104020, 2010, arXiv:1003.4231.
[502] P. G. Ferreira, C. Skordis, and C. Zunckel. Dark Matter, Modified Gravity and the Mass of the
Neutrino. Phys. Rev., D78:044043, 2008, 0806.0116.
[503] B. Fiedler and R. Schimming. Singularity-free static centrally symmetric solutions of some fourth
order gravitational field equations. Astronomische Nachrichten, 304:221–229, 1983.
[504] M. Fierz and W. Pauli. On relativistic wave equations for particles of arbitrary spin in an
electromagnetic field. Proc.Roy.Soc.Lond., A173:211–232, 1939.
[505] P. Figueras, J. Lucietti, and T. Wiseman. Ricci solitons, Ricci flow, and strongly coupled CFT
in the Schwarzschild Unruh or Boulware vacua. 2011, 1104.4489.
[506] P. Figueras and T. Wiseman. Gravity and large black holes in Randall-Sundrum II braneworlds.
2011, 1105.2558.
[507] E. Fischbach and C. Talmadge. Six years of the fifth force. Nature, 356:207–215, Mar. 1992.
[508] A. L. Fitzpatrick, L. Randall, and T. Wiseman. On the existence and dynamics of braneworld
black holes. JHEP, 11:033, 2006, hep-th/0608208.
[509] E. E. Flanagan. Palatini Form of 1/R Gravity. Phys. Rev. Lett., 92(7):071101, Feb 2004.
[510] J. Forero-Romero, S. Gottloeber, and G. Yepes. Bullet Clusters in the MareNostrum Universe.
Astrophys.J., 725:598–604, 2010, 1007.3902.
[511] E. Fradkin and M. A. Vasiliev. On the Gravitational Interaction of Massless Higher Spin Fields.
Phys.Lett., B189:89–95, 1987.
[512] K. Freese and M. Lewis. Cardassian Expansion: a Model in which the Universe is Flat, Matter
Dominated, and Accelerating. Phys. Lett., B540:1–8, 2002, astro-ph/0201229.
[513] L. Freidel. Modified gravity without new degrees of freedom. 2008, 0812.3200.
[514] P. G. O. Freund. Kaluza-Klein Cosmologies. Nucl. Phys., B209:146, 1982.
[515] A. V. Frolov. Singularity Problem with f(R) Models for Dark Energy. Physical Review Letters,
101(6):061103, Aug. 2008, 0803.2500.
[516] S. Fulling and L. Parker. Renormalization in the theory of a quantized scalar field interacting
with a robertson-walker spacetime. Annals Phys., 87:176–204, 1974.
[517] G. Gabadadze. General Relativity With An Auxiliary Dimension. Phys.Lett., B681:89–95, 2009,
0908.1112.
[518] G. Gabadadze and M. Shifman. Softly massive gravity. Phys.Rev., D69:124032, 2004, hep-
th/0312289.
284
[519] R. Gannouji, B. Moraes, and D. Polarski. The growth of matter perturbations in f(R) models.
J. Cosmology Astropart. Phys., 2:34, Feb. 2009, 0809.3374.
[520] R. Gannouji and M. Sami. Galileon gravity and its relevance to late time cosmic acceleration.
Phys.Rev., D82:024011, 2010, 1004.2808.
[521] X. Gao, Y. Wang, R. Brandenberger, and A. Riotto. Cosmological Perturbations in Hořava-
Lifshitz Gravity. Phys. Rev., D81:083508, 2010, 0905.3821.
[522] X. Gao, Y. Wang, W. Xue, and R. Brandenberger. Fluctuations in a Hořava-Lifshitz Bouncing
Cosmology. JCAP, 1002:020, 2010, 0911.3196.
[523] J. Garcı́a-Bellido. Jordan-Brans-Dicke stochastic inflation. Nuclear Physics B, 423:221–242, July
1994, arXiv:astro-ph/9401042.
[524] J. Garcı́a-Bellido and A. Linde. Stationary solutions in Brans-Dicke stochastic inflationary cos-
mology. Phys. Rev. D, 52:6730–6738, Dec. 1995, arXiv:gr-qc/9504022.
[525] J. Garcı́a-Bellido, A. Linde, and D. Linde. Fluctuations of the gravitational constant in the infla-
tionary Brans-Dicke cosmology. Phys. Rev. D, 50:730–750, July 1994, arXiv:astro-ph/9312039.
[526] J. Garcı́a-Bellido and M. Quirós. Extended inflation in scalar-tensor theories of gravity. Physics
Letters B, 243:45–51, June 1990.
[527] J. Garcı́a-Bellido and D. Wands. Constraints from inflation on scalar-tensor gravity theories.
Phys. Rev. D, 52:6739–6749, Dec. 1995, arXiv:gr-qc/9506050.
[528] J. Garriga and T. Tanaka. Gravity in the brane-world. Phys. Rev. Lett., 84:2778–2781, 2000,
hep-th/9911055.
[529] M. Gasperini, M. Maggiore, and G. Veneziano. Towards a non-singular pre-big-bang cosmology.
Nuclear Physics B, 494:315–328, Feb. 1997, arXiv:hep-th/9611039.
[530] M. Gasperini and G. Veneziano. Pre-big-bang in string cosmology. Astroparticle Physics, 1:317–
339, July 1993, arXiv:hep-th/9211021.
[531] C. Gauss. Disquisitiones generales circa superficies curvas. Typis Dieterichianis, 1828.
[532] C. Germani, A. Kehagias, and K. Sfetsos. Relativistic Quantum Gravity at a Lifshitz Point.
JHEP, 0909:060, 2009, 0906.1201.
[533] B. Geyer, S. D. Odintsov, and S. Zerbini. Inflationary Brans-Dicke quantum universe. Physics
Letters B, 460:58–62, Aug. 1999, arXiv:gr-qc/9905073.
[534] T. Giannantonio, R. Scranton, R. G. Crittenden, R. C. Nichol, S. P. Boughn, A. D. Myers,
and G. T. Richards. Combined analysis of the integrated Sachs-Wolfe effect and cosmological
implications. Phys. Rev. D, 77(12):123520, June 2008, 0801.4380.
[535] D. Giannios. Spherically symmetric, static spacetimes in TeVeS. Phys. Rev., D71:103511, 2005,
gr-qc/0502122.
[536] G. W. Gibbons and S. W. Hawking. Action Integrals and Partition Functions in Quantum
Gravity. Phys. Rev., D15:2752–2756, 1977.
[537] H. Gies, D. F. Mota, and D. J. Shaw. Hidden in the light: Magnetically induced afterglow from
trapped chameleon fields. Phys. Rev. D, 77(2):025016, Jan. 2008, 0710.1556.
[538] N. Goheer, R. Goswami, and P. K. S. Dunsby. Dynamics of f(R)-cosmologies containing Einstein
static models. Classical and Quantum Gravity, 26(10):105003, May 2009, 0809.5247.
[539] N. Goheer, J. Larena, and P. K. S. Dunsby. Power-law cosmic expansion in f(R) gravity models.
Phys. Rev. D, 80(6):061301, Sept. 2009, 0906.3860.
[540] N. Goheer, J. A. Leach, and P. K. S. Dunsby. Dynamical systems analysis of anisotropic cos-
mologies in Rn -gravity. Classical and Quantum Gravity, 24:5689–5708, Nov. 2007, 0710.0814.
[541] W. D. Goldberger and M. B. Wise. Bulk fields in the Randall-Sundrum compactification scenario.
Phys. Rev., D60:107505, 1999, hep-ph/9907218.
[542] W. D. Goldberger and M. B. Wise. Modulus stabilization with bulk fields. Phys. Rev. Lett.,
83:4922–4925, 1999, hep-ph/9907447.
[543] Y. Gong and A. Wang. Friedmann Equations and Thermodynamics of Apparent Horizons. Phys-
ical Review Letters, 99(21):211301, Nov. 2007, 0704.0793.
[544] G. Goon, K. Hinterbichler, and M. Trodden. General Embedded Brane Effective Field Theories.
2011, 1103.6029.
[545] G. Goon, K. Hinterbichler, and M. Trodden. Symmetries for Galileons and DBI scalars on curved
space. 2011, 1103.5745.
[546] D. Gorbunov, K. Koyama, and S. Sibiryakov. More on ghosts in DGP model. Phys. Rev.,
D73:044016, 2006, hep-th/0512097.
[547] D. S. Gorbunov, V. A. Rubakov, and S. M. Sibiryakov. Gravity waves from inflating brane or
mirrors moving in adS(5). JHEP, 10:015, 2001, hep-th/0108017.
[548] C. Gordon and R. Maartens. Density perturbations in the brane world. Phys. Rev., D63:044022,
285
 2001, hep-th/0009010.
[549] R. Goswami, N. Goheer, and P. K. S. Dunsby. Existence of Einstein static universes and their
stability in fourth-order theories of gravity. Phys. Rev. D, 78(4):044011, Aug. 2008, 0804.3528.
[550] S. Gottlober, H. Schmidt, and A. A. Starobinsky. Sixth-order gravity and conformal transforma-
tions. Classical and Quantum Gravity, 7:893–900, May 1990.
[551] S. Gottloeber, V. Mueller, and H. Schmidt. Generalized inflation from R-cubed and R square R
terms. Astronomische Nachrichten, 312:291–297, July 1991.
[552] M. Grana. Flux compactifications in string theory: A comprehensive review. Phys. Rept., 423:91–
158, 2006, hep-th/0509003.
[553] E. Gravanis and S. Willison. Israel conditions for the Gauss-Bonnet theory and the Friedmann
equation on the brane universe. Phys.Lett., B562:118–126, 2003, hep-th/0209076.
[554] A. M. Green and A. R. Liddle. Conditions for successful extended inflation. Phys. Rev. D,
54:2557–2563, Aug. 1996, arXiv:astro-ph/9604001.
[555] B. R. Greene and J. Levin. Dark Energy and Stabilization of Extra Dimensions. JHEP, 11:096,
2007, 0707.1062.
[556] J. P. Gregory and A. Padilla. Exact braneworld cosmology induced from bulk black holes. Class.
Quant. Grav., 19:4071–4083, 2002, hep-th/0204218.
[557] J. P. Gregory and A. Padilla. Brane world holography in Gauss-Bonnet gravity.
Class.Quant.Grav., 20:4221–4238, 2003, hep-th/0304250.
[558] R. Gregory. Braneworld black holes. Lect. Notes Phys., 769:259–298, 2009, 0804.2595.
[559] R. Gregory, N. Kaloper, R. C. Myers, and A. Padilla. A New Perspective on DGP Gravity.
JHEP, 10:069, 2007, 0707.2666.
[560] R. Gregory and A. Padilla. Braneworld instantons. Class. Quant. Grav., 19:279–302, 2002,
hep-th/0107108.
[561] R. Gregory and A. Padilla. Nested braneworlds and strong brane gravity. Phys. Rev., D65:084013,
2002, hep-th/0104262.
[562] R. Gregory, S. F. Ross, and R. Zegers. Classical and quantum gravity of brane black holes. JHEP,
09:029, 2008, 0802.2037.
[563] R. Gregory, V. A. Rubakov, and S. M. Sibiryakov. Opening up extra dimensions at ultra-large
scales. Phys. Rev. Lett., 84:5928–5931, 2000, hep-th/0002072.
[564] L. Grisa and O. Pujolas. Dressed Domain Walls and Holography. JHEP, 06:059, 2008, 0712.2786.
[565] S. Groot Nibbelink and M. Pospelov. Lorentz violation in supersymmetric field theories.
Phys.Rev.Lett., 94:081601, 2005, hep-ph/0404271.
[566] D. J. Gross and J. H. Sloan. The Quartic Effective Action for the Heterotic String. Nucl. Phys.,
B291:41, 1987.
[567] A. Gruzinov. On the graviton mass. New Astron., 10:311–314, 2005, astro-ph/0112246.
[568] A. Guarnizo, L. Castaeda, and J. M. Tejeiro. Boundary term in metric f(R) gravity: field
equations in the metric formalism. Gen. Rel. Grav. 42:2713, 2010, 1002.0617.
[569] A. Guarnizo, L. Castaeda, and J. M. Tejeiro. Geodesic deviation equation in f(R) gravity. Gen.
Rel. Grav. 43:2713, 2011, 1010.5279.
[570] S. S. Gubser. AdS/CFT and gravity. Phys. Rev., D63:084017, 2001, hep-th/9912001.
[571] U. Gunther, and A. Zhuk, A Note on dynamical stabilization of internal spaces in multidimen-
sional cosmology. Class. Quant. Grav., 18:1441-1460, 2001, hep-ph/0006283.
[572] U. Gunther, A. Starobinsky, and A. Zhuk. Multidimensional cosmological models: Cosmological
and astrophysical implications. Phys. Rev. D, 69:044003, 2004, hep-ph/0306191.
[573] Z.-K. Guo, N. Ohta, and T. Torii. Black Holes in the Dilatonic Einstein-Gauss-Bonnet Theory in
Various Dimensions. I - Asymptotically Flat Black Holes. Prog. Theo. Phys., 120:581-607, Sep.
2008, 0806.2481.
[574] Z.-K. Guo, N. Ohta, and T. Torii. Black Holes in the Dilatonic Einstein-Gauss-Bonnet Theory
in Various Dimensions. II - Asymptotically AdS Topological Black Holes. Prog. Theo. Phys.,
121:253-273, Feb. 2009, 0811.3068.
[575] Z.-K. Guo, N. Ohta, and S. Tsujikawa. Realizing scale-invariant density perturbations in low-
energy effective string theory. Phys. Rev. D, 75(2):023520, Jan. 2007, hep-th/0610336.
[576] L. E. Gurevich, A. M. Finkelstein, and V. A. Ruban. On the Problem of the Initial State in the
Isotropic Scalar-Tensor Cosmology of Brans-Dicke. Ap&SS, 22:231–242, June 1973.
[577] L. Gurvits, K. Kellermann, and S. Frey. The “angular size - redshift” relation for compact radio
structures in quasars and radio galaxies. 1998, astro-ph/9812018.
[578] A. H. Guth and B. Jain. Density fluctuations in extended inflation. Phys. Rev. D, 45:426–432,
Jan. 1992.
286
[579] E. Guzman. Exact Brans-Dicke-Bianchi type-V II(h) perfect fluid solutions. Ap&SS, 152:171–173,
Feb. 1989.
[580] E. Guzmán. General Vacuum Solution for Brans-Dicke-Bianchi Type-II. Ap&SS, 179:331–334,
May 1991.
[581] E. Guzmán. General vacuum solution for Brans-Dicke-Bianchi type V . General Relativity and
Gravitation, 23:1007–1010, Sept. 1991.
[582] E. Guzman. Cosmic no-hair theorem in Brans-Dicke cosmology. Physics Letters B, 391:267–270,
Feb. 1997.
[583] E. Guzmán. Homogeneous Universes in Extended Inflation I. Ap&SS, 249:179–188, Apr. 1997.
[584] E. Guzmán. Homogeneous Universes in Extended Inflation II. Ap&SS, 249:211–221, Apr. 1997.
[585] L. Guzzo et al. A test of the nature of cosmic acceleration using galaxy redshift distortions.
Nature, 451:541, 2008.
[586] N. Haba and S. Matsumoto. Baryogenesis from Dark Sector. 2010, 1008.2487.
[587] L. J. Hall, J. March-Russell, and S. M. West. A Unified Theory of Matter Genesis: Asymmetric
Freeze- In. 2010, 1010.0245.
[588] L. J. Hall and D. Tucker-Smith. Cosmological constraints on theories with large extra dimensions.
Phys. Rev., D60:085008, 1999, hep-ph/9904267.
[589] A. Halle, H. Zhao, and B. Li. Perturbations In A Non-Uniform Dark Energy Fluid: Equations
Reveal Effects of Modified Gravity and Dark Matter. ArXiv e-prints, Nov. 2007, 0711.0958.
[590] N. A. Hamed, H. S. Cheng, M. A. Luty, and S. Mukohyama. Ghost Condensation and a Con-
sistent IR Modification of Gravity. Journal of High Energy Physics, 5:74, May 2004, arXiv:hep-
th/0312099.
[591] F. Hansen, A. Banday, K. Gorski, H. Eriksen, and P. Lilje. Power Asymmetry in Cosmic Mi-
crowave Background Fluctuations from Full Sky to Sub-degree Scales: Is the Universe Isotropic?
Astrophys.J., 704:1448–1458, 2009, 0812.3795.
[592] J. G. Hao and R. Akhoury. Can relativistic MOND theory resolve both the dark matter and dark
energy paradigms? 2005, astro-ph/0504130.
[593] T. Harada, C. Goymer, and B. J. Carr. Tolman-bondi collapse in scalar-tensor theories as a probe
of gravitational memory. Phys. Rev. D, 66(10):104023, Nov 2002.
[594] M. Haugan. Energy conservation and the principle of equivalence. Annals Phys., 118:156–186,
1979.
[595] S. W. Hawking. Black holes in the Brans-Dicke theory of gravitation. Comm. Math. Phys.,
25:167–171, 1972.
[596] S. W. Hawking, T. Hertog, and H. S. Reall. Brane new world. Phys. Rev., D62:043501, 2000,
hep-th/0003052.
[597] S. W. Hawking, T. Hertog, and H. S. Reall. Trace anomaly driven inflation. Phys. Rev.,
D63:083504, 2001, hep-th/0010232.
[598] E. Hawkins, S. Maddox, S. Cole, D. Madgwick, P. Norberg, et al. The 2dF Galaxy Red-
shift Survey: Correlation functions, peculiar velocities and the matter density of the universe.
Mon.Not.Roy.Astron.Soc., 346:78, 2003, astro-ph/0212375.
[599] F. W. Hehl. How does one measure torsion of spacetime? Phys. Lett., A36: 225, 1971.
[600] F. W. Hehl and G. D. Kerlick. Metric-affine variational principles in general relativity. I - Rie-
mannian space-time. General Relativity and Gravitation, 9:691–710, Aug. 1978.
[601] F. W. Hehl, P. Von Der Heyde, G. D. Kerlick and J. M. Nester. General Relativity with Spin
and Torsion: Foundations and Prospects. Rev. Mod. Phys., 48, 393, 1976.
[602] F. W. Hehl and P. Von Der Heyde. Spin and the structure of space-time Annales Poincare
Phys.Theor., 19: 179, 1973.
[603] M. Henneaux, A. Kleinschmidt, and G. L. Gomez. A dynamical inconsistency of Horava gravity.
Phys.Rev., D81:064002, 2010, 0912.0399.
[604] A. Hewish, S. J. Bell, J. D. H. Pilkington, P. F. Scott, and R. A. Collins. Observation of a rapidly
pulsating radio source. Nature, 217:709–713, 1968.
[605] A. Higuchi. Forbidden mass range for spin-2 field theory in de Sitter space-time. Nucl. Phys.,
B282:397, 1987.
[606] A. Hindawi, B. A. Ovrut, and D. Waldram. Consistent spin-two coupling and quadratic gravita-
tion. Phys. Rev. D, 53:5583–5596, May 1996, arXiv:hep-th/9509142.
[607] A. Hindawi, B. A. Ovrut, and D. Waldram. Nontrivial vacua in higher-derivative gravitation.
Phys. Rev. D, 53:5597–5608, May 1996, arXiv:hep-th/9509147.
[608] K. Hinterbichler and J. Khoury. Symmetron Fields: Screening Long-Range Forces Through Local
Symmetry Restoration. Phys. Rev. Lett., 104:231301, 2010, 1001.4525.
287
[609] K. Hinterbichler, A. Nicolis, and M. Porrati. Superluminality in DGP. JHEP, 0909:089, 2009,
0905.2359.
[610] K. Hinterbichler, M. Trodden, and D. Wesley. Multi-field galileons and higher co-dimension
branes. Phys.Rev., D82:124018, 2010, 1008.1305.
[611] S. Ho, C. Hirata, N. Padmanabhan, U. Seljak, and N. Bahcall. Correlation of CMB with
large-scale structure. I. Integrated Sachs-Wolfe tomography and cosmological implications.
Phys. Rev. D, 78(4):043519, Aug. 2008, 0801.0642.
[612] D. W. Hogg. Distance measures in cosmology. ArXiv Astrophysics e-prints, May 1999,
arXiv:astro-ph/9905116.
[613] D. J. Holden and D. Wands. Phase-plane analysis of friedmann-robertson-walker cosmologies in
brans-dicke gravity. Classical and Quantum Gravity, 15(10):3271, 1998.
[614] R. Holman, E. W. Kolb, S. L. Vadas, and Y. Wang. False-vacuum decay in generalized extended
inflation. Physics Letters B, 250:24–28, Nov. 1990.
[615] R. Holman, E. W. Kolb, S. L. Vadas, and Y. Wang. Extended inflation from higher dimensional
theories. Phys. Rev., D43:995–1004, 1991.
[616] R. Holman, E. W. Kolb, S. L. Vadas, and Y. Wang. Scale-invariant extended inflation.
Phys. Rev. D, 43:3833–3845, June 1991.
[617] R. Holman, Y. Wang, and E. W. Kolb. Gravitational couplings of the inflaton in extended
inflation. Physical Review Letters, 65:17–20, July 1990.
[618] P. Horava. Membranes at Quantum Criticality. JHEP, 0903:020, 2009, 0812.4287.
[619] P. Horava. Quantum Gravity at a Lifshitz Point. Phys.Rev., D79:084008, 2009, 0901.3775.
[620] P. Horava. Spectral Dimension of the Universe in Quantum Gravity at a Lifshitz Point.
Phys.Rev.Lett., 102:161301, 2009, 0902.3657.
[621] P. Horava and C. M. Melby-Thompson. General Covariance in Quantum Gravity at a Lifshitz
Point. Phys.Rev., D82:064027, 2010, 1007.2410.
[622] P. Horava and E. Witten. Eleven-Dimensional Supergravity on a Manifold with Boundary. Nucl.
Phys., B475:94–114, 1996, hep-th/9603142.
[623] G. Horndeski. Second-order scalar-tensor field equations in a four-dimensional space. Interna-
tional Journal of Theoretical Physics, 10(6):363–384, 1974.
[624] C. D. Hoyle, U. Schmidt, B. R. Heckel, E. G. Adelberger, J. H. Gundlach, D. J. Kapner, and
H. E. Swanson. Submillimeter Test of the Gravitational Inverse-Square Law: A Search for “Large”
Extra Dimensions. Physical Review Letters, 86:1418–1421, Feb. 2001, arXiv:hep-ph/0011014.
[625] S. D. H. Hsu. Black holes from extended inflation. Physics Letters B, 251:343–348, Nov. 1990.
[626] W. Hu. Structure Formation with Generalized Dark Matter. ApJ, 506:485–494, Oct. 1998,
arXiv:astro-ph/9801234.
[627] W. Hu. Parametrized Post-Friedmann Signatures of Acceleration in the CMB. Phys. Rev.,
D77:103524, 2008, 0801.2433.
[628] W. Hu, R. Barkana, and A. Gruzinov. Cold and fuzzy dark matter. Phys. Rev. Lett., 85:1158–
1161, 2000, astro-ph/0003365.
[629] W. Hu and I. Sawicki. Models of f(R) cosmic acceleration that evade solar system tests. Phys.
Rev. D, 76(6):064004, Sep 2007.
[630] W. Hu and I. Sawicki. Parametrized post-Friedmann framework for modified gravity.
Phys. Rev. D, 76(10):104043, Nov. 2007, 0708.1190.
[631] Y. Huang and A. Wang. Nonrelativistic general covariant theory of gravity with a running
constant λ. 2010, 1011.0739.
[632] Y. Huang, A. Wang, and Q. Wu. Stability of the de Sitter spacetime in Horava-Lifshitz theory.
Mod.Phys.Lett., A25:2267–2279, 2010, 1003.2003.
[633] V. W. Hughes, H. G. Robinson, and V. Beltran-Lopez. Upper Limit for the Anisotropy of Inertial
Mass from Nuclear Resonance Experiments. Physical Review Letters, 4:342–344, Apr. 1960.
[634] R. A. Hulse and J. H. Taylor. Discovery of a pulsar in a binary system. ApJ, 195:L51–L53, Jan.
1975.
[635] V. Husain, E. A. Martinez, and D. Nunez. Exact solution for scalar field collapse. Phys. Rev.,
D50:3783–3786, 1994, gr-qc/9402021.
[636] J. Hwang and H. Noh. f(R) gravity theory and CMBR constraints. Physics Letters B, 506:13–19,
May 2001, arXiv:astro-ph/0102423.
[637] J.-c. Hwang and H. Noh. Gauge-ready formulation of the cosmological kinetic theory in general-
ized gravity theories. Phys. Rev. D, 65(2):023512, Dec 2001.
[638] K. Ichiki, M. Yahiro, T. Kajino, M. Orito, and G. J. Mathews. Observational constraints on dark
radiation in brane cosmology. Phys. Rev., D66:043521, 2002, astro-ph/0203272.
288
[639] D. Ida. Brane-world cosmology. JHEP, 09:014, 2000, gr-qc/9912002.
[640] A. Iglesias, N. Kaloper, A. Padilla, and M. Park. How (not) to use the palatini formulation of
scalar-tensor gravity. Phys. Rev. D, 76(10):104001, Nov 2007.
[641] L. Iorio. On the reliability of the so-far performed tests for measuring the Lense Thirring effect
with the LAGEOS satellites. New Astronomy, 10:603–615, Aug. 2005, arXiv:gr-qc/0411024.
[642] L. Iorio, H. I. M. Lichtenegger, M. L. Ruggiero, and C. Corda. Phenomenology of the Lense-
Thirring effect in the solar system. Astrophys. Sp. Sci., 331:351-395, Feb. 2011.
[643] L. Iorio, and M. L. Ruggiero. Phenomenological Constraints on the Kehagias-Sfetsos Solution
in the HORAVA-LIFSHITZ Gravity from Solar System Orbital Motions. Int. J. Mod. Phys. A,
25:5399-5408, 2010, 0909.2562.
[644] L. Iorio, and M. L. Ruggiero. Constraining the Kehagias-Sfetsos Solution of the Horava-Lifshitz
Modified Gravity with Extrasolar Planets. Open. Ast. J., 3:167-171, Dec. 2010, 0909.5355.
[645] L. Iorio, and M. L. Ruggiero. HORAVA-LIFSHITZ Gravity:. Tighter Constraints for the
Kehagias-Sfetsos Solution from New Solar System Data. Int J. Mod. Phys. D, 20:1079-1093,
2011, 1012.2822.
[646] M. Ishak and J. Moldenhauer. A minimal set of invariants as a systematic approach to higher
order gravity models. JCAP, 1:24, Jan. 2009, 0808.0951.
[647] C. J. Isham, A. Salam, and J. Strathdee. Relativity-f gravity and gravitational collapse. Nature,
244:82, Aug. 1973.
[648] A. Ishibashi and S. Speziale. Spherically symmetric black holes in minimally modified self-dual
gravity. Class.Quant.Grav., 26:175005, 2009, 0904.3914.
[649] W. Israel. Singular hypersurfaces and thin shells in general relativity. Nuovo Cim., B44S10:1,
1966.
[650] W. Israel. Event horizons in static vacuum space-times. Phys. Rev., 164(5):1776–1779, Dec 1967.
[651] W. Israel. Event horizons in static electrovac space-times. Communications in Mathematical
Physics, 8:245–260, Sept. 1968.
[652] K. Izumi and S. Mukohyama. Nonlinear superhorizon perturbations in Horava-Lifshitz gravity.
2011, 1105.0246.
[653] K. Izumi and T. Tanaka. No de Sitter invariant vacuum in massive gravity theory with ghost.
Prog.Theor.Phys., 121:419–426, 2009, 0709.0199.
[654] T. Jacobson. Primordial black hole evolution in tensor-scalar cosmology. Phys. Rev. Lett.,
83(14):2699–2702, Oct 1999.
[655] T. Jacobson. Einstein-aether gravity: a status report. PoS, QG-PH:020, 2007, 0801.1547.
[656] T. Jacobson and D. Mattingly. Gravity with a dynamical preferred frame. Phys. Rev., D64:024028,
2001, gr-qc/0007031.
[657] H. Jaichan. Cosmological perturbations in a sixth-order gravity. Classical and Quantum Gravity,
8:L133–L136, June 1991.
[658] D. Jain, A. Dev, and J. Alcaniz. Brane world cosmologies and statistical properties of gravitational
lenses. Phys.Rev., D66:083511, 2002, astro-ph/0206224.
[659] J. T. Jebsen. Über die allgemeinen kugelsymmetrischen Lösungen der Einsteinschen Gravitation-
sgleichungen im Vakuum. Ark. Mat. Astr. Fys., 15:1, 1921.
[660] X.-h. Jin and X.-z. Li. Oppenheimer-Volkoff Equation in Relativistic MOND Theory. 2006,
gr-qc/0605046.
[661] S. Kachru, R. Kallosh, A. D. Linde, and S. P. Trivedi. De Sitter vacua in string theory. Phys.
Rev., D68:046005, 2003, hep-th/0301240.
[662] S. Kachru, M. B. Schulz, and S. Trivedi. Moduli stabilization from fluxes in a simple IIB orien-
tifold. JHEP, 10:007, 2003, hep-th/0201028.
[663] E. O. Kahya. A Decisive test to confirm or rule out existence of dark matter using gravitational
wave observations. Class. Quant. Grav., 25:184008, 2008, 0801.1984.
[664] E. O. Kahya and R. P. Woodard. A generic test of modified gravity models which emulate dark
matter. Phys. Lett., B652:213–216, 2007, 0705.0153.
[665] K. Kainulainen, V. Reijonen, and D. Sunhede. Interior spacetimes of stars in Palatini f(R) gravity.
Phys. Rev. D, 76(4):043503, Aug. 2007, arXiv:gr-qc/0611132.
[666] R. Kallosh. On Inflation in String Theory. Lect. Notes Phys., 738:119–156, 2008, hep-th/0702059.
[667] N. Kaloper. Bent domain walls as braneworlds. Phys. Rev., D60:123506, 1999, hep-th/9905210.
[668] N. Kaloper. Brane Induced Gravity: Codimension-2. Mod. Phys. Lett., A23:781–796, 2008,
0711.3210.
[669] N. Kaloper and D. Kiley. Charting the Landscape of Modified Gravity. JHEP, 05:045, 2007,
hep-th/0703190.
289
[670] N. Kaloper, L. Kofman, A. D. Linde, and V. Mukhanov. On the new string theory inspired
mechanism of generation of cosmological perturbations. JCAP, 0610:006, 2006, hep-th/0608200.
[671] N. Kaloper, J. March-Russell, G. D. Starkman, and M. Trodden. Compact hyperbolic extra
dimensions: Branes, Kaluza-Klein modes and cosmology. Phys. Rev. Lett., 85:928–931, 2000,
hep-ph/0002001.
[672] N. Kaloper, A. Padilla, N. Tanahashi, Galileon Hairs of Dyson Spheres, Vainshtein’s Coiffure
and Hirsute Bubbles, [arXiv:1106.4827 [hep-th]].
[673] T. Kaluza. On the Problem of Unity in Physics. Sitzungsber. Preuss. Akad. Wiss. Berlin (Math.
Phys. ), 1921:966–972, 1921.
[674] G. Kang. Black hole area in brans-dicke theory. Phys. Rev. D, 54(12):7483–7489, Dec 1996.
[675] S. Kanno and J. Soda. On the higher codimension braneworld. 2004, gr-qc/0410067.
[676] S. Kanno and J. Soda. Quasi-thick codimension 2 braneworld. JCAP, 0407:002, 2004, hep-
th/0404207.
[677] S. Kanno and J. Soda. Lorentz violating inflation. Phys. Rev. D, 74(6):063505, Sept. 2006,
arXiv:hep-th/0604192.
[678] P. Kanti, N. E. Mavromatos, J. Rizos, K. Tamvakis, and E. Winstanley. Dilatonic black holes in
higher curvature string gravity. Phys. Rev. D, 54:5049–5058, Oct. 1996, arXiv:hep-th/9511071.
[679] D. B. Kaplan. Single explanation for both baryon and dark matter densities. Phys. Rev. Lett.,
68(6):741–743, Feb 1992.
[680] D. E. Kaplan, M. A. Luty, and K. M. Zurek. Asymmetric dark matter. Phys. Rev. D,
79(11):115016, Jun 2009.
[681] K. Karami and A. Abdolmaleki. Original and entropy-corrected versions of the holographic and
new agegraphic f(T)-gravity models. ArXiv e-prints, Sept. 2010, 1009.2459.
[682] A. Karch and L. Randall. Locally localized gravity. JHEP, 05:008, 2001, hep-th/0011156.
[683] A. Karch and L. Randall. Relaxing to three dimensions. Phys. Rev. Lett., 95:161601, 2005,
hep-th/0506053.
[684] A. Karch and K. Sfetsos. The black hole and FRW geometries of non-relativistic gravity. Phys.
Lett. B, 678:123-126, Jul. 2009, 0905.0477.
[685] A. Kehagias and E. Kiritsis. Mirage cosmology. JHEP, 9911:022, 1999, hep-th/9910174.
[686] G. D. Kerlick. How does one measure torsion of spacetime? Phys. Lett., A36: 225, 1971.
[687] J. Khoury, J.-L. Lehners, and B. A. Ovrut. Supersymmetric Galileons. 2011, 1103.0003.
[688] J. Khoury and A. Weltman. Chameleon cosmology. Phys. Rev. D, 69(4):044026, Feb. 2004,
arXiv:astro-ph/0309411.
[689] J. Khoury and A. Weltman. Chameleon Fields: Awaiting Surprises for Tests of Gravity in Space.
Physical Review Letters, 93(17):171104, Oct. 2004, arXiv:astro-ph/0309300.
[690] J. Khoury and M. Wyman. N-Body Simulations of DGP and Degravitation Theories. Phys.Rev.,
D80:064023, 2009, 0903.1292.
[691] T. W. B. Kibble. Lorentz invariance and the gravitational field. J. Math. Phys., 2, 212, 1961.
[692] I. Kimpton and A. Padilla. Lessons from the decoupling limit of Horava gravity. JHEP, 1007:014,
2010, 1003.5666.
[693] R. Kimura and K. Yamamoto. Large Scale Structures in Kinetic Gravity Braiding Model. 2010,
1011.2006.
[694] E. Kiritsis, G. Kofinas, N. Tetradis, T. N. Tomaras, V. Zarikas, Cosmological evolution with
brane bulk energy exchange. JHEP 0302 (2003) 035. [hep-th/0207060].
[695] E. Kiritsis, Holography and brane-bulk energy exchange. JCAP 0510 (2005) 014. [hep-
th/0504219].
[696] E. Kiritsis, N. Tetradis, T. N. Tomaras, Induced brane gravity: Realizations and limitations.
JHEP 0108 (2001) 012. [hep-th/0106050].
[697] E. Kiritsis, N. Tetradis, T. N. Tomaras, Induced gravity on RS branes. JHEP 0203 (2002) 019.
[hep-th/0202037].
[698] E. Kiritsis and G. Kofinas. Horava-Lifshitz Cosmology. Nucl.Phys., B821:467–480, 2009,
0904.1334.
[699] R. Kitano, H. Murayama, and M. Ratz. Unified origin of baryons and dark matter. Physics
Letters B, 669(2):145 – 149, 2008.
[700] B. Kleihaus, J. Kunz, and E. Radu. Rotating Black Holes in Dilatonic Einstein-Gauss-Bonnet
Theory. ArXiv e-prints, Jan. 2011, 1101.2868.
[701] O. Klein. Quantentheorie und fünfdimensionale Relativitätstheorie. Zeitschrift fur Physik,
37:895–906, Dec. 1926.
[702] O. Klein. The Atomicity of Electricity as a Quantum Theory Law. Nature, 118:516, Oct. 1926.
290
[703] S. Kobayashi, K. Koyama, and J. Soda. Quantum fluctuations of bulk inflaton in inflationary
brane world. Phys. Lett., B501:157–164, 2001, hep-th/0009160.
[704] T. Kobayashi. Cosmic expansion and growth histories in Galileon scalar-tensor models of dark
energy. Phys.Rev., D81:103533, 2010, 1003.3281.
[705] T. Kobayashi, H. Kudoh, and T. Tanaka. Primordial gravitational waves in inflationary
braneworld. Phys. Rev., D68:044025, 2003, gr-qc/0305006.
[706] T. Kobayashi and K. Maeda. Relativistic stars in f(R) gravity, and absence thereof. Phys. Rev. D,
78(6):064019, Sept. 2008, 0807.2503.
[707] T. Kobayashi and K. Maeda. Can higher curvature corrections cure the singularity problem in
f(R) gravity? Phys. Rev. D, 79(2):024009, Jan. 2009, 0810.5664.
[708] T. Kobayashi, T. Shiromizu, and C. de Rham. Curvature corrections to the low energy effective
theory in 6D regularized braneworlds. Phys.Rev., D77:124012, 2008, 0802.0103.
[709] T. Kobayashi, T. Shiromizu, and N. Deruelle. Low energy effective gravitational equations on a
Gauss-Bonnet brane. Phys.Rev., D74:104031, 2006, hep-th/0608166.
[710] T. Kobayashi, M. Yamaguchi, J. ’i. Yokoyama, Generalized G-inflation: Inflation with the most
general second-order field equations, [arXiv:1105.5723 [hep-th]].
[711] T. Kobayashi, H. Tashiro, and D. Suzuki. Evolution of linear cosmological perturbations and
its observational implications in Galileon-type modified gravity. Phys.Rev., D81:063513, 2010,
0912.4641.
[712] T. Kobayashi, Y. Urakawa, and M. Yamaguchi. Large scale evolution of the curvature perturba-
tion in Horava-Lifshitz cosmology. JCAP, 0911:015, 2009, 0908.1005.
[713] T. Kobayashi, Y. Urakawa, and M. Yamaguchi. Cosmological perturbations in a healthy extension
of Horava gravity. JCAP, 1004:025, 2010, 1002.3101.
[714] T. Kobayashi, M. Yamaguchi, and J. Yokoyama. G-inflation: Inflation driven by the Galileon
field. Phys.Rev.Lett., 105:231302, 2010, 1008.0603.
[715] H. Kodama. Behavior of cosmological perturbations in the brane-world model. 2000, hep-
th/0012132.
[716] H. Kodama, A. Ishibashi, and O. Seto. Brane world cosmology: Gauge-invariant formalism for
perturbation. Phys. Rev., D62:064022, 2000, hep-th/0004160.
[717] H. Kodama and M. Sasaki. Cosmological Perturbation Theory. Prog.Theor.Phys.Suppl., 78:1–
166, 1984.
[718] G. Kofinas, R. Maartens, and E. Papantonopoulos. Brane cosmology with curvature corrections.
JHEP, 0310:066, 2003, hep-th/0307138.
[719] L. A. Kofman, V. F. Mukhanov, and D. I. Pogosian. The evolution of inhomogeneities in infla-
tionary models in the theory of gravitation with higher derivatives. Zhurnal Eksperimental noi i
Teoreticheskoi Fiziki, 93:769–783, Sept. 1987.
[720] I. I. Kogan, S. Mouslopoulos, and A. Papazoglou. A new bigravity model with exclusively positive
branes. Phys. Lett., B501:140–149, 2001, hep-th/0011141.
[721] I. I. Kogan, S. Mouslopoulos, and A. Papazoglou. The m → 0 limit for massive graviton in dS(4)
and AdS(4): How to circumvent the van Dam-Veltman-Zakharov discontinuity. Phys. Lett.,
B503:173–180, 2001, hep-th/0011138.
[722] I. I. Kogan, S. Mouslopoulos, A. Papazoglou, and G. G. Ross. Multi-brane worlds and modification
of gravity at large scales. Nucl. Phys., B595:225–249, 2001, hep-th/0006030.
[723] I. I. Kogan, S. Mouslopoulos, A. Papazoglou, G. G. Ross, and J. Santiago. A three three-brane
universe: New phenomenology for the new millennium? Nucl. Phys., B584:313–328, 2000, hep-
ph/9912552.
[724] T. Koivisto. Matter power spectrum in f(R) gravity. Phys. Rev. D, 73(8):083517, Apr. 2006,
arXiv:astro-ph/0602031.
[725] T. Koivisto. Viable Palatini-f(R) cosmologies with generalized dark matter. Phys. Rev. D,
76(4):043527, Aug. 2007, 0706.0974.
[726] T. Koivisto. New variational principles as alternatives to the Palatini method. Phys. Rev. D,
83(10):101501, 2011, 1103.2743.
[727] T. Koivisto and H. Kurki-Suonio. Cosmological perturbations in the Palatini formulation of
modified gravity. Class. Quant. Grav., 23:2355–2369, 2006, astro-ph/0509422.
[728] T. Koivisto and D. F. Mota. Cosmology and astrophysical constraints of Gauss Bonnet dark
energy. Physics Letters B, 644:104–108, Jan. 2007, arXiv:astro-ph/0606078.
[729] T. Koivisto and D. F. Mota. Gauss-Bonnet quintessence: Background evolution, large scale struc-
ture, and cosmological constraints. Phys. Rev. D, 75(2):023518, Jan. 2007, arXiv:hep-th/0609155.
[730] E. W. Kolb, D. Lindley, and D. Seckel. More Dimensions - Less Entropy. Phys. Rev., D30:1205,
291
 1984.
[731] S. J. Kolitch and D. M. Eardley. Behavior of Friedmann-Robertson-Walker cosmological models
in scalar-tensor gravity. Annals of Physics, 241:128–151, July 1995, arXiv:gr-qc/9405016.
[732] E. Komatsu, K. M. Smith, J. Dunkley, C. L. Bennett, B. Gold, G. Hinshaw, N. Jarosik, D. Larson,
M. R. Nolta, L. Page, D. N. Spergel, M. Halpern, R. S. Hill, A. Kogut, M. Limon, S. S. Meyer,
N. Odegard, G. S. Tucker, J. L. Weiland, E. Wollack, and E. L. Wright. Seven-year Wilkinson
Microwave Anisotropy Probe (WMAP) Observations: Cosmological Interpretation. ApJS, 192:18,
Feb. 2011, 1001.4538.
[733] M. Kowalski et al. Improved Cosmological Constraints from New, Old and Combined Supernova
Datasets. Astrophys.J., 686:749–778, 2008, 0804.4142.
[734] K. Koyama. Ghosts in the self-accelerating universe. Class. Quant. Grav., 24:R231–R253, 2007,
0709.2399.
[735] K. Koyama and F. Arroja. Pathological behaviour of the scalar graviton in Horava-Lifshitz
gravity. JHEP, 1003:061, 2010, 0910.1998.
[736] K. Koyama and K. Koyama. Brane induced gravity from asymmetric compactification. Phys.
Rev., D72:043511, 2005, hep-th/0501232.
[737] K. Koyama and R. Maartens. Structure formation in the Dvali-Gabadadze-Porrati cosmological
model. JCAP, 0601:016, 2006, astro-ph/0511634.
[738] K. Koyama, G. Niz, and G. Tasinato. Analytic solutions in non-linear massive gravity. 2011,
1103.4708.
[739] K. Koyama, G. Niz, and G. Tasinato. Strong interactions and exact solutions in non-linear
massive gravity. 2011, 1104.2143.
[740] K. Koyama, A. Padilla, and F. P. Silva. Ghosts in asymmetric brane gravity and the decoupled
stealth limit. JHEP, 03:134, 2009, 0901.0713.
[741] K. Koyama and F. P. Silva. Non-linear interactions in a cosmological background in the DGP
braneworld. Phys.Rev., D75:084040, 2007, hep-th/0702169.
[742] K. Koyama and J. Soda. Evolution of cosmological perturbations in the brane world. Phys. Rev.,
D62:123502, 2000, hep-th/0005239.
[743] K. Koyama, A. Taruya, and T. Hiramatsu. Non-linear Evolution of Matter Power Spectrum in
Modified Theory of Gravity. Phys.Rev., D79:123512, 2009, 0902.0618.
[744] M. Kramer. Determination of the geometry of the psr b1913+16 system by geodetic precession.
Astrophys. J., 509:856–860, 1998.
[745] M. Kramer, I. H. Stairs, R. N. Manchester, M. A. McLaughlin, A. G. Lyne, R. D. Ferdman,
M. Burgay, D. R. Lorimer, A. Possenti, N. D’Amico, J. M. Sarkissian, G. B. Hobbs, J. E.
Reynolds, P. C. C. Freire, and F. Camilo. Tests of General Relativity from Timing the Double
Pulsar. Science, 314:97–102, Oct. 2006, arXiv:astro-ph/0609417.
[746] G. A. Krasinskii, E. V. Pit’eva, M. L. Sveshnikov, and L. I. Chuniaeva. The motion of major
planets from observations 1769-1988 and some astronomical constants. Celestial Mechanics and
Dynamical Astronomy, 55:1–23, Jan. 1993.
[747] K. Krasnov. Non-Metric Gravity. I. Field Equations. Class.Quant.Grav., 25:025001, 2008, gr-
qc/0703002.
[748] K. Krasnov. On deformations of Ashtekar’s constraint algebra. Phys.Rev.Lett., 100:081102, 2008,
0711.0090.
[749] K. Krasnov. Motion of a ‘small body’ in non-metric gravity. Phys.Rev., D79:044017, 2009,
0812.3603.
[750] K. Krasnov. Plebanski Formulation of General Relativity: A Practical Introduction.
Gen.Rel.Grav., 43:1–15, 2011, 0904.0423.
[751] K. Krasnov and Y. Shtanov. Non-Metric Gravity. II. Spherically Symmetric Solution, Missing
Mass and Redshifts of Quasars. Class.Quant.Grav., 25:025002, 2008, 0705.2047.
[752] K. Krasnov and Y. Shtanov. Halos of Modified Gravity. Int.J.Mod.Phys., D17:2555–2562, 2009,
0805.2668.
[753] K. Krasnov and Y. Shtanov. Cosmological perturbations in a family of deformations of general
relativity. JCAP, 1006:006, 2010, 1002.1210.
[754] P. Kraus. Dynamics of anti-de Sitter domain walls. JHEP, 12:011, 1999, hep-th/9910149.
[755] A. Krause and S. Ng. Ghost Cosmology:. Exact Solutions, Transitions Between Standard Cos-
mologies and Ghost Evolution. International Journal of Modern Physics A, 21:1091–1122, 2006,
arXiv:hep-th/0409241.
[756] P. Kroupa, B. Famaey, K. de Boer, J. Dabringhausen, M. Pawlowski, et al. Local-Group tests
of dark-matter Concordance Cosmology: Towards a new paradigm for structure formation? As-
292
 tron.Astrophys., 523:A32, 2010, 1006.1647.
[757] J. Kuchowicz. The Einstein-Cartan equations in astrophysically interesting situations. I The case
of spherical symmetry.. Act. Phys. Pol., B5: 555, 1975.
[758] H. Kudoh, T. Tanaka, and T. Nakamura. Small localized black holes in braneworld: Formulation
and numerical method. Phys. Rev., D68:024035, 2003, gr-qc/0301089.
[759] M. Kunz and D. Sapone. Dark energy versus modified gravity. Phys. Rev. Lett., 98:121301, 2007,
astro-ph/0612452.
[760] M. Kuwahara, H. Suzuki, and E. Takasugi. Analytic solutions for cosmological fluctuation equa-
tions in extended inflation models. Phys. Rev. D, 50:661–670, July 1994.
[761] D. La and P. J. Steinhardt. Bubble percolation in extended inflationary models. Physics Letters
B, 220:375–378, Apr. 1989.
[762] D. La and P. J. Steinhardt. Extended inflationary cosmology. Physical Review Letters, 62:376–
378, Jan. 1989.
[763] D. La, P. J. Steinhardt, and E. W. Bertschinger. Prescription for successful extended inflation.
Physics Letters B, 231:231–236, Nov. 1989.
[764] G. Lambiase and G. Scarpetta. Baryogenesis in f(R) theories of gravity. Phys. Rev. D,
74(8):087504, Oct. 2006, arXiv:astro-ph/0610367.
[765] S. K. Lamoreaux, J. P. Jacobs, B. R. Heckel, F. J. Raab, and E. N. Fortson. New limits on
spatial anisotropy from optically-pumped 201 Hg and 199 Hg. Physical Review Letters, 57:3125–
3128, Dec. 1986.
[766] N. Lanahan-Tremblay and V. Faraoni. The Cauchy problem of f(R) gravity. Classical and
Quantum Gravity, 24:5667–5679, Nov. 2007, 0709.4414.
[767] C. Lanczos. A remarkable property of the Riemann-Christoffel tensor in four dimensions. Annals
of Mathematics, 39(4):842–850, 1938.
[768] D. Langlois. Brane cosmological perturbations. Phys. Rev., D62:126012, 2000, hep-th/0005025.
[769] D. Langlois. Evolution of cosmological perturbations in a brane- universe. Phys. Rev. Lett.,
86:2212–2215, 2001, hep-th/0010063.
[770] D. Langlois. Brane cosmology: An introduction. Prog. Theor. Phys. Suppl., 148:181–212, 2003,
hep-th/0209261.
[771] D. Langlois, R. Maartens, M. Sasaki, and D. Wands. Large-scale cosmological perturbations on
the brane. Phys. Rev., D63:084009, 2001, hep-th/0012044.
[772] D. Langlois, R. Maartens, and D. Wands. Gravitational waves from inflation on the brane. Phys.
Lett., B489:259–267, 2000, hep-th/0006007.
[773] J. Larena, J. Alimi, and A. Serna. Big Bang Nucleosynthesis in Scalar Tensor Gravity: The Key
Problem of the Primordial 7 Li Abundance. ApJ, 658:1–10, Mar. 2007, arXiv:astro-ph/0511693.
[774] P. D. Lasky, H. Sotani, and D. Giannios. Structure of neutron stars in tensor-vector-scalar theory.
Phys. Rev., D78:104019, 2008, 0811.2006.
[775] A. M. Laycock and A. R. Liddle. Extended inflation with a curvature-coupled inflaton.
Phys. Rev. D, 49:1827–1839, Feb. 1994, arXiv:astro-ph/9306030.
[776] K. Lazaridis, N. Wex, A. Jessner, M. Kramer, B. W. Stappers, G. H. Janssen, G. Desvignes,
M. B. Purver, I. Cognard, G. Theureau, A. G. Lyne, C. A. Jordan, and J. A. Zensus. Generic
tests of the existence of the gravitational dipole radiation and the variation of the gravitational
constant. MNRAS, 400:805–814, Dec. 2009, 0908.0285.
[777] U. J. Le Verrier. Theorie DU mouvement de Mercure. Annales de l’Observatoire de Paris, 5:1,
1859.
[778] J. A. Leach, S. Carloni, and P. K. S. Dunsby. Shear dynamics in Bianchi I cosmologies with
Rn -gravity. Classical and Quantum Gravity, 23:4915–4937, Aug. 2006, arXiv:gr-qc/0603012.
[779] D. L. Lee, W. Ni, C. M. Caves, and C. M. Will. Theoretical frameworks for testing relativistic
gravity. V - Post-Newtonian limit of Rosen’s theory. ApJ, 206:555–558, June 1976.
[780] J. Lee and E. Komatsu. Bullet Cluster: A Challenge to LCDM Cosmology. Astrophys.J., 718:60–
65, 2010, 1003.0939.
[781] S. Lee. Stability of Palatini-f(R) cosmology. ArXiv e-prints, Oct. 2007, 0710.2395.
[782] S. Lee. Palatini f(R) Cosmology. Modern Physics Letters A, 23:1388–1396, 2008, 0801.4606.
[783] S. Lee and K. Ng. Growth index with the exact analytic solution of sub-horizon scale linear
perturbation for dark energy models with constant equation of state. Physics Letters B, 688:1–3,
Apr. 2010, 0906.1643.
[784] S. Lee and K. Ng. Properties of the exact analytic solution of the growth factor and its applica-
tions. Phys. Rev. D, 82(4):043004, Aug. 2010, 0907.2108.
[785] J. Lense and H. Thirring. Über den Einfluß der Eigenrotation der Zentralkörper auf die Bewegung
293
 der Planeten und Monde nach der Einsteinschen Gravitationstheorie. Physikalische Zeitschrift,
19:156, 1918.
[786] G. Leon and E. N. Saridakis. Dynamics of anisotropic f(R) cosmology. ArXiv e-prints, July 2010,
1007.3956.
[787] B. Leong, P. Dunsby, A. Challinor, and A. Lasenby. 1+3 covariant dynamics of scalar perturba-
tions in braneworlds. Phys. Rev., D65:104012, 2002, gr-qc/0111033.
[788] S. A. Levshakov, A. V. Lapinov, C. Henkel, P. Molaro, D. Reimers, M. G. Kozlov, and I. I.
Agafonova. Searching for chameleon-like scalar fields with the ammonia method. II. Mapping of
cold molecular cores in NH3 and HC3N lines. ArXiv e-prints, Aug. 2010, 1008.1160.
[789] S. A. Levshakov, P. Molaro, A. V. Lapinov, D. Reimers, C. Henkel, and T. Sakai. Searching for
chameleon-like scalar fields with the ammonia method. A&A, 512:A44+, Mar. 2010, 0911.3732.
[790] B. Li and J. D. Barrow. Cosmology of f(R) gravity in the metric variational approach.
Phys. Rev. D, 75(8):084010, Apr. 2007, arXiv:gr-qc/0701111.
[791] B. Li, J. D. Barrow, and D. F. Mota. Cosmology of modified Gauss-Bonnet gravity. Phys. Rev. D,
76(4):044027, Aug. 2007, 0705.3795.
[792] B. Li, J. D. Barrow, and D. F. Mota. Cosmology of Ricci-tensor-squared gravity in the Palatini
variational approach. Phys. Rev. D, 76(10):104047, Nov. 2007, 0707.2664.
[793] B. Li, K. C. Chan, and M. Chu. Constraints on f(R) cosmology in the Palatini formalism.
Phys. Rev. D, 76(2):024002, July 2007, arXiv:astro-ph/0610794.
[794] B. Li, D. F. Mota, and D. J. Shaw. Microscopic and macroscopic behaviors of Palatini modified
gravity theories. Phys. Rev. D, 78(6):064018, Sept. 2008, 0805.3428.
[795] B. Li, D. F. Mota, and D. J. Shaw. Indistinguishable Macroscopic Behaviour of Palatini Gravities
and General Relativity. Class. Quant. Grav., 26:055018, 2009, 0801.0603.
[796] B. Li, T. P. Sotiriou, and J. D. Barrow. f(T) gravity and local Lorentz invariance. ArXiv e-prints,
Oct. 2010, 1010.1041.
[797] A. Lichnerowicz. Theories Relativistes de la Gravitation et de l’Electromagnétisme. Masson,
Paris, 1955.
[798] A. R. Liddle, D. H. Lyth, and W. Sutherland. Structure formation from power law (and extended)
inflation. Physics Letters B, 279:244–249, Apr. 1992.
[799] A. R. Liddle, A. Mazumdar, and J. D. Barrow. Radiation-matter transition in jordan-brans-dicke
theory. Phys. Rev. D, 58(2):027302, Jun 1998.
[800] A. R. Liddle and A. J. Smith. Observational constraints on braneworld chaotic inflation. Phys.
Rev., D68:061301, 2003, astro-ph/0307017.
[801] A. R. Liddle and L. A. Urena-Lopez. Curvaton reheating: An application to braneworld inflation.
Phys. Rev., D68:043517, 2003, astro-ph/0302054.
[802] A. R. Liddle and D. Wands. Microwave background constraints on extended inflation voids.
MNRAS, 253:637–648, Dec. 1991.
[803] A. R. Liddle and D. Wands. Hyperextended inflation: Dynamics and constraints. Phys. Rev. D,
45:2665–2673, Apr. 1992.
[804] A. R. Liddle and D. Wands. There is a big-bubble problem in extended inflation. Phys. Rev. D,
46:3655–3658, Oct. 1992.
[805] J. E. Lidsey, S. Nojiri, and S. D. Odintsov. Brane world cosmology in (anti)-de Sitter Einstein-
Gauss-Bonnet-Maxwell gravity. JHEP, 0206:026, 2002, hep-th/0202198.
[806] J. E. Lidsey and N. J. Nunes. Inflation in Gauss-Bonnet brane cosmology. Phys.Rev., D67:103510,
2003, astro-ph/0303168.
[807] J. E. Lidsey, D. Wands, and E. J. Copeland. Superstring cosmology. Phys. Rept., 337:343–492,
2000, hep-th/9909061.
[808] A. P. Lightman and D. L. Lee. New Two-Metric Theory of Gravity with Prior Geometry.
Phys. Rev. D, 8:3293–3302, Nov. 1973.
[809] E. A. Lim. Can we see Lorentz-violating vector fields in the CMB? Phys. Rev. D, 71(6):063504,
Mar. 2005, arXiv:astro-ph/0407437.
[810] C. Limbach, D. Psaltis, and F. Ozel. The Redshift Evolution of the Tully-Fisher Relation as a
Test of Modified Gravity. 2008, 0809.2790.
[811] K. Lin, A. Wang, Q. Wu, and T. Zhu. On strong coupling in nonrelativistic general covariant
theory of gravity. 2011, 1106.1486.
[812] A. D. Linde. Inflation and string cosmology. ECONF, C040802:L024, 2004, hep-th/0503195.
[813] E. V. Linder. Cosmic growth history and expansion history. Phys. Rev. D, 72(4):043529, Aug.
2005, arXiv:astro-ph/0507263.
[814] E. V. Linder. Exponential gravity. Phys. Rev. D, 80(12):123528, Dec. 2009, 0905.2962.
294
[815] E. V. Linder. Einstein’s other gravity and the acceleration of the Universe. Phys. Rev. D,
81(12):127301, June 2010, 1005.3039.
[816] E. V. Linder and R. N. Cahn. Parameterized Beyond-Einstein Growth. Astropart.Phys., 28:481–
488, 2007, astro-ph/0701317.
[817] L. Lombriser, W. Hu, W. Fang, and U. Seljak. Cosmological Constraints on DGP Braneworld
Gravity with Brane Tension. Phys.Rev., D80:063536, 2009, 0905.1112.
[818] L. Lombriser, A. Slosar, U. Seljak, and W. Hu. Constraints on f(R) gravity from probing the
large-scale structure. ArXiv e-prints, Mar. 2010, 1003.3009.
[819] J. C. Long, H. W. Chan, A. B. Churnside, E. A. Gulbis, M. C. M. Varney, and J. C. Price. Upper
limits to submillimetre-range forces from extra space-time dimensions. Nature, 421:922–925, Feb.
2003, arXiv:hep-ph/0210004.
[820] D. Lorenz. Exact Bianchi type-V III and type-IX cosmological models with matter and electro-
magnetic fields. Phys. Rev. D, 22:1848–1852, Oct. 1980.
[821] D. Lorenz-Petzold. The general vacuum Bianchi type-V solution in the Brans-Dicke theory.
Mathematical Proceedings of the Cambridge Philosophical Society, 95:175, 1983.
[822] D. Lorenz-Petzold. Exact Brans-Dicke-Bianchi Type-II Solutions. Progress of Theoretical
Physics, 71:406–408, Feb. 1984.
[823] D. Lorenz-Petzold. Exact Brans-Dicke-Bianchi type-V solutions. Mathematical Proceedings of
the Cambridge Philosophical Society, 96:183, 1984.
[824] D. Lorenz-Petzold. Exact Brans-Dicke-Bianchi type-V II(h) solutions. Ap&SS, 106:409–413, Nov.
1984.
[825] D. Lorenz-Petzold. Exact Brans-Dicke-Bianchi types-II, V III, and IX stiff matter solutions.
Ap&SS, 98:281–286, Jan. 1984.
[826] D. Lorenz-Petzold. Exact Brans-Dicke Friedmann-Robertson-Walker-Bianchi-Kantowski-Sachs
vacuum solutions. Ap&SS, 98:101–113, Jan. 1984.
[827] D. Lorenz-Petzold. The general non-LRS-BDT-Bianchi type-V radiation solution. Ap&SS,
103:397–399, Aug. 1984.
[828] D. Lorenz-Petzold. Electromagnetic Brans-Dicke-Bianchi type-V solutions. Ap&SS, 114:277–284,
Sept. 1985.
[829] D. Lorenz-Petzold. Exact Brans-Dicke-Bianchi type-V Ih solutions. Acta Physica Polonica B,
16:171–178, 1985.
[830] D. Lorenz-Petzold. Bianchi type-V I0 and type-V Ih perfect fluid solutions. Ap&SS, 134:415–416,
June 1987.
[831] D. Lovelock. The Einstein Tensor and Its Generalizations. Journal of Mathematical Physics,
12:498–501, Mar. 1971.
[832] D. Lovelock. The Four-Dimensionality of Space and the Einstein Tensor. Journal of Mathematical
Physics, 13:874–876, June 1972.
[833] A. Lue. Global structure of Deffayet (Dvali-Gabadadze-Porrati) cosmologies. Phys.Rev.,
D67:064004, 2003, hep-th/0208169.
[834] A. Lue. The phenomenology of dvali-gabadadze-porrati cosmologies. Phys.Rept., 423:1–48, 2006,
astro-ph/0510068.
[835] A. Lue, G. D. Starkman, How a brane cosmological constant can trick us into thinking that W ¡
-1 Phys. Rev. D70 (2004) 101501. [arXiv:astro-ph/0408246 [astro-ph]].
[836] A. Lue, R. Scoccimarro, and G. D. Starkman. Probing Newton’s constant on vast scales: DGP
gravity, cosmic acceleration and large scale structure. Phys. Rev., D69:124015, 2004, astro-
ph/0401515.
[837] A. Lue and G. D. Starkman. Squeezing MOND into a Cosmological Scenario. Physical Review
Letters, 92(13):131102, Apr. 2004, arXiv:astro-ph/0310005.
[838] A. Lukas, B. A. Ovrut, K. S. Stelle, and D. Waldram. The universe as a domain wall. Phys.
Rev., D59:086001, 1999, hep-th/9803235.
[839] M. A. Luty, M. Porrati, and R. Rattazzi. Strong interactions and stability in the DGP model.
JHEP, 0309:029, 2003, hep-th/0303116.
[840] A. G. Lyne, M. Burgay, M. Kramer, A. Possenti, R. N. Manchester, F. Camilo, M. A. McLaughlin,
D. R. Lorimer, N. D’Amico, B. C. Joshi, J. Reynolds, and P. C. C. Freire. A Double-Pulsar
System: A Rare Laboratory for Relativistic Gravity and Plasma Physics. Science, 303:1153–
1157, Feb. 2004, arXiv:astro-ph/0401086.
[841] C.-P. Ma and E. Bertschinger. Cosmological perturbation theory in the synchronous and confor-
mal newtonian gauges. Astrophys. J., 455:7, 1995.
[842] R. Maartens. Cosmological dynamics on the brane. Phys. Rev., D62:084023, 2000, hep-
295
 th/0004166.
[843] R. Maartens, V. Sahni, T. D. Saini, Anisotropy dissipation in brane world inflation Phys. Rev.
D63 (2001) 063509. [gr-qc/0011105].
[844] R. Maartens. Geometry and dynamics of the brane-world. 2001, gr-qc/0101059.
[845] R. Maartens and K. Koyama. Brane-World Gravity. Living Rev. Rel., 13:5, 2010, 1004.3962.
[846] R. Maartens and E. Majerotto. Observational constraints on self-accelerating cosmology.
Phys.Rev., D74:023004, 2006, astro-ph/0603353.
[847] R. Maartens, D. Wands, B. A. Bassett, and I. Heard. Chaotic inflation on the brane. Phys. Rev.,
D62:041301, 2000, hep-ph/9912464.
[848] K. I. Macrae and R. J. Riegert. Effect of curvature-squared terms on cosmology. Phys. Rev. D,
24:2555–2560, Nov. 1981.
[849] M. S. Madsen and J. D. Barrow. De Sitter ground states and boundary terms in generalised
gravity. Nuclear Physics B, 323:242–252, Aug. 1989.
[850] H. Maeda and G. Giribet. Lifshitz black holes in Brans-Dicke theory. ArXiv e-prints, May 2011,
1105.1331.
[851] H. Maeda, V. Sahni, and Y. Shtanov. Braneworld dynamics in Einstein-Gauss-Bonnet gravity.
Phys.Rev., D76:104028, 2007, 0708.3237.
[852] K. Maeda. Inflation as a transient attractor in R2 cosmology. Phys. Rev. D, 37:858–862, Feb.
1988.
[853] K. Maeda. Towards the Einstein-Hilbert action via conformal transformation. Phys. Rev. D,
39:3159–3162, May 1989.
[854] K.-i. Maeda, Y. Misonoh, and T. Kobayashi. Oscillating Universe in Horava-Lifshitz Gravity.
Phys. Rev., D82:064024, 2010, 1006.2739.
[855] K.-i. Maeda, and N. Ohta. Inflation from M-theory with fourth-order corrections and large extra
dimensions. Phys. Lett. B, 597:400-407, Sep. 2004, hep-th/0405205.
[856] K.-i. Maeda, and N. Ohta. Inflation from superstring and M-theory compactification with higher
order corrections. Phys. Rev. D, 71(6):063520, Mar. 2005, hep-th/0411093.
[857] K.-i. Maeda, N. Ohta, and Y. Sasagawa. Black hole solutions in string theory with Gauss-Bonnet
curvature correction. Phys. Rev. D, 80(10):104032, Nov. 2009, 0908.4151.
[858] K.-i. Maeda, N. Ohta, and Y. Sasagawa. AdS black hole solutions in dilatonic Einstein-Gauss-
Bonnet gravity. Phys. Rev. D, 83(4):044051, Feb. 2011, 1012.0568.
[859] K.-i. Maeda and T. Torii. Covariant gravitational equations on brane world with Gauss-Bonnet
term. Phys.Rev., D69:024002, 2004, hep-th/0309152.
[860] A. S. Majumdar, D. Gangopadhyay, and L. P. Singh. Evolution of primordial black holes in
Jordan-Brans-Dicke cosmology. MNRAS, 385:1467–1470, Apr. 2008, 0709.3193.
[861] N. Makino and M. Sasaki. The Density Perturbation in the Chaotic Inflation with Non-Minimal
Coupling. Progress of Theoretical Physics, 86:103–118, July 1991.
[862] J. M. Maldacena. The large N limit of superconformal field theories and supergravity. Adv.
Theor. Math. Phys., 2:231–252, 1998, hep-th/9711200.
[863] S. Mallik and D. Rai Chaudhuri. Density perturbation in extended inflation. Phys. Rev. D,
56:625–630, July 1997, arXiv:hep-ph/9612427.
[864] P. D. Mannheim. Cosmic Acceleration as the Solution to the Cosmological Constant Problem.
ApJ, 561:1–12, Nov. 2001, arXiv:astro-ph/9910093.
[865] P. D. Mannheim. Comprehensive solution to the cosmological constant, zero-point energy,
and quantum gravity problems. General Relativity and Gravitation, 43:703–750, Mar. 2011,
0909.0212.
[866] P. D. Mannheim. Making the Case for Conformal Gravity. ArXiv e-prints, Jan. 2011, 1101.2186.
[867] M. Mateo. Dwarf Galaxies of the Local Group. Ann. Rev. Astron. Astrophys., 36:435–506, 1998,
astro-ph/9810070.
[868] M. Mathisson. Neue mechanik materieller systemes. Acta Phys.Polon., 6: 163, 1937.
[869] D. Mattingly. Modern Tests of Lorentz Invariance. Living Reviews in Relativity, 8:5, Sept. 2005,
arXiv:gr-qc/0502097.
[870] N. E. Mavromatos, M. Sakellariadou, and M. F. Yusaf. Can TeVeS avoid Dark Matter on galactic
scales? 2009, 0901.3932.
[871] A. B. Mayer and H. Schmidt. The de Sitter spacetime as attractor solution in eighth-order
gravity. Classical and Quantum Gravity, 10:2441–2446, Nov. 1993.
[872] S. H. Mazharimousavi and M. Halilsoy. Black hole solutions in f(R) gravity coupled with non-
linear Yang-Mills field. ArXiv e-prints, May 2011, 1105.3659.
[873] L. McAllister and E. Silverstein. String Cosmology: A Review. Gen. Rel. Grav., 40:565–605,
296
 2008, 0710.2951.
[874] J. McDonald. Dark matter from an oscillating Brans-Dicke scalar in the extended inflation model.
Phys. Rev. D, 44:2325–2334, Oct. 1991.
[875] J. McDonald. Can a Brans-Dicke scalar account for dark matter in extended inflation models?
Phys. Rev. D, 48:2462–2476, Sept. 1993.
[876] J. McDonald. Baryomorphosis: Relating the Baryon Asymmetry to the ‘WIMP Miracle’. 2010,
1009.3227.
[877] P. McDonald et al. The Linear theory power spectrum from the Lyman-alpha forest in the Sloan
Digital Sky Survey. Astrophys.J., 635:761–783, 2005, astro-ph/0407377.
[878] S. S. McGaugh. A Novel Test of the Modified Newtonian Dynamics with Gas Rich Galaxies.
Phys.Rev.Lett., 106:121303, 2011, 1102.3913.
[879] S. S. McGaugh, J. M. Schombert, G. D. Bothun, and W. de Blok. The Baryonic Tully-Fisher
relation. Astrophys.J., 533:L99–L102, 2000, astro-ph/0003001.
[880] O. Mena, J. Santiago, and J. Weller. Constraining Inverse-Curvature Gravity with Supernovae.
Physical Review Letters, 96(4):041103, Feb. 2006, arXiv:astro-ph/0510453.
[881] X. Meng and P. Wang. Letter: Gravitational Potential in the Palatini Formulation of Modified
Gravity. General Relativity and Gravitation, 36:1947–1954, Aug. 2004, arXiv:gr-qc/0311019.
[882] X. Meng and P. Wang. Palatini Formulation of Modified Gravity with Squared Scalar Curvature.
General Relativity and Gravitation, 36:2673–2680, Dec. 2004, arXiv:astro-ph/0308284.
[883] X. Meng and P. Wang. R2 corrections to the cosmological dynamics of inflation in the Palatini
formulation. Classical and Quantum Gravity, 21:2029–2036, Apr. 2004, arXiv:gr-qc/0402011.
[884] X. Meng and P. Wang. Palatini formulation of the R−1 -modified gravity with an additional
squared scalar curvature term. Classical and Quantum Gravity, 22:23–32, Jan. 2005, arXiv:gr-
qc/0411007.
[885] S. M. Merkowitz. Tests of Gravity Using Lunar Laser Ranging. Living Reviews in Relativity,
13:7, Nov. 2010.
[886] R. R. Metsaev and A. A. Tseytlin. Order α’ (two-loop) equivalence of the string equations of
motion and the σ-model Weyl invariance conditions Dependence on the dilaton and the antisym-
metric tensor. Nuclear Physics B, 293:385–419, 1987.
[887] J. Middleton. On the existence of anisotropic cosmological models in higher order theories of
gravity. Classical and Quantum Gravity, 27(22):225013, Nov. 2010, 1007.4669.
[888] J. Middleton and J. D. Barrow. Stability of an isotropic cosmological singularity in higher-order
gravity. Phys. Rev. D, 77(10):103523, May 2008, 0801.4090.
[889] S. Mignemi. Dyonic black holes in effective string theory. Phys. Rev. D, 51:934–937, Jan. 1995,
arXiv:hep-th/9303102.
[890] S. Mignemi and N. R. Stewart. Charged black holes in effective string theory. Phys. Rev. D,
47:5259–5269, June 1993, arXiv:hep-th/9212146.
[891] S. Mignemi and D. L. Wiltshire. Black holes in higher-derivative gravity theories. Phys. Rev. D,
46(4):1475–1506, Aug 1992.
[892] M. B. Mijić, M. S. Morris, and W. Suen. The R2 cosmology: Inflation without a phase transition.
Phys. Rev. D, 34:2934–2946, Nov. 1986.
[893] G. Miknaitis, G. Pignata, A. Rest, W. Wood-Vasey, S. Blondin, et al. The ESSENCE Supernova
Survey: Survey Optimization, Observations, and Supernova Photometry. Astrophys.J., 666:674–
693, 2007, astro-ph/0701043.
[894] A. Milani, D. Vokrouhlický, D. Villani, C. Bonanno, and A. Rossi. Testing general relativity with
the BepiColombo radio science experiment. Phys. Rev. D, 66(8):082001, Oct. 2002.
[895] M. Milgrom. A Modification of the Newtonian dynamics as a possible alternative to the hidden
mass hypothesis. Astrophys.J., 270:365–370, 1983.
[896] M. Milgrom. Bimetric MOND gravity. Phys. Rev., D80:123536, 2009, 0912.0790.
[897] M. Milgrom. Cosmological fluctuation growth in bimetric MOND. Phys. Rev. D, 82(4):043523,
Aug. 2010, 1006.3809.
[898] J. P. Mimoso and D. Wands. Anisotropic scalar-tensor cosmologies. Phys. Rev. D, 52:5612–5627,
Nov. 1995, arXiv:gr-qc/9501039.
[899] J. P. Mimoso and D. Wands. Massless fields in scalar-tensor cosmologies. Phys. Rev. D, 51(2):477–
489, Jan 1995.
[900] M. Minamitsuji. Self-accelerating solutions in cascading DGP braneworld. Phys. Lett., B684:92–
95, 2010, 0806.2390.
[901] V. Miranda, S. E. Jorás, I. Waga, and M. Quartin. Viable Singularity-Free f(R) Gravity without
a Cosmological Constant. Physical Review Letters, 102(22):221101, June 2009, 0905.1941.
297
[902] C. Misner, K. Thorne, and J. Wheeler. Gravitation. WH Freeman & co, 1973.
[903] S. Mizuno and K. Koyama. Primordial non-Gaussianity from the DBI Galileons. Phys.Rev.,
D82:103518, 2010, 1009.0677.
[904] N. Moeller and B. Zwiebach. Dynamics with infinitely many time derivatives and rolling tachyons.
JHEP, 10:034, 2002, hep-th/0207107.
[905] J. W. Moffat. Scalar tensor vector gravity theory. J. Cosmology Astropart. Phys., 3:4, Mar. 2006,
arXiv:gr-qc/0506021.
[906] J. Moldenhauer and M. Ishak. A minimal set of invariants as a systematic approach to higher
order gravity models: physical and cosmological constraints. J. Cosmology Astropart. Phys.,
12:20, Dec. 2009, 0912.5332.
[907] J. Moldenhauer, M. Ishak, J. Thompson and D. A. Easson. Supernova, baryon acoustic oscilla-
tions, and CMB surface distance constraints on f(G) higher order gravity models. Phys. Rev. D,
81:063514, Mar. 2010, 1004.2459.
[908] S. Mollerach and S. Matarrese. Non-scale-invariant density perturbations from chaotic extended
inflation. Phys. Rev. D, 45:1961–1970, Mar. 1992.
[909] B. Moore, S. Ghigna, F. Governato, G. Lake, T. R. Quinn, et al. Dark matter substructure in
galactic halos. Astrophys.J., 524:L19–L22, 1999, astro-ph/9907411.
[910] D. F. Mota, P. G. Ferreira, and C. Skordis. Cosmic Microwave Background in TeVeS theory.
Work presented at the Alternative gravities workshop in Edinburgh, 2006.
[911] D. F. Mota, M. Sandstad, and T. Zlosnik. Cosmology of the selfaccelerating third order Galileon.
JHEP, 1012:051, 2010, 1009.6151.
[912] D. F. Mota and D. J. Shaw. Strongly Coupled Chameleon Fields: New Horizons in Scalar Field
Theory. Physical Review Letters, 97(15):151102, Oct. 2006, arXiv:hep-ph/0606204.
[913] D. F. Mota and D. J. Shaw. Evading equivalence principle violations, cosmological, and other
experimental constraints in scalar field theories with a strong coupling to matter. Phys. Rev. D,
75(6):063501, Mar 2007.
[914] H. Motohashi, A. A. Starobinsky, and J. Yokoyama. Analytic Solution for Matter Density Pertur-
bations in a Class of Viable Cosmological f(R) Models. International Journal of Modern Physics
D, 18:1731–1740, 2009, 0905.0730.
[915] F. Mueller-Hoissen. Dimensionally continued Euler forms, Kaluza-Klein cosmology and dimen-
sional reduction. Class. Quant. Grav., 3:665, 1986.
[916] V. F. Mukhanov and G. V. Chibisov. Quantum fluctuations and a nonsingular universe. Soviet
Journal of Experimental and Theoretical Physics Letters, 33:532, May 1981.
[917] V. F. Mukhanov, H. Feldman, and R. H. Brandenberger. Theory of cosmological perturbations.
Part 1. Classical perturbations. Part 2. Quantum theory of perturbations. Part 3. Extensions.
Phys.Rept., 215:203–333, 1992.
[918] S. Mukohyama. Gauge-invariant gravitational perturbations of maximally symmetric spacetimes.
Phys. Rev., D62:084015, 2000, hep-th/0004067.
[919] S. Mukohyama. Perturbation of junction condition and doubly gauge- invariant variables. Class.
Quant. Grav., 17:4777–4798, 2000, hep-th/0006146.
[920] S. Mukohyama. Integro-differential equation for brane-world cosmological perturbations. Phys.
Rev., D64:064006, 2001, hep-th/0104185.
[921] S. Mukohyama. Doubly-gauge-invariant formalism of brane-world cosmological perturbations.
2002, hep-th/0202100.
[922] S. Mukohyama. An accelerating universe and cosmological perturbation in the ghost condensate.
J. Cosmology Astropart. Phys., 10:11, Oct. 2006, arXiv:hep-th/0607181.
[923] S. Mukohyama. Caustic avoidance in Horava-Lifshitz gravity. JCAP, 0909:005, 2009, 0906.5069.
[924] S. Mukohyama. Dark matter as integration constant in Horava-Lifshitz gravity. Phys.Rev.,
D80:064005, 2009, 0905.3563.
[925] S. Mukohyama. Scale-invariant cosmological perturbations from Horava-Lifshitz gravity without
inflation. JCAP, 0906:001, 2009, 0904.2190.
[926] S. Mukohyama. Horava-Lifshitz Cosmology: A Review. Class.Quant.Grav., 27:223101, 2010,
1007.5199.
[927] S. Mukohyama, K. Nakayama, F. Takahashi, and S. Yokoyama. Phenomenological Aspects of
Horava-Lifshitz Cosmology. Phys.Lett., B679:6–9, 2009, 0905.0055.
[928] J. Müller, K. Nordtvedt, and D. Vokrouhlický. Improved constraint on the α1 ppn parameter
from lunar motion. Phys. Rev. D, 54:R5927–R5930, 1996.
[929] T. Multamäki and I. Vilja. Cosmological expansion and the uniqueness of the gravitational action.
Phys. Rev. D, 73(2):024018, Jan. 2006, arXiv:astro-ph/0506692.
298
[930] T. Multamäki and I. Vilja. Constraining Newtonian stellar configurations in f(R) theories of
gravity. Physics Letters B, 659:843–846, Feb. 2008, 0709.3422.
[931] R. C. Myers. Higher derivative gravity, surface terms and string theory. Phys. Rev., D36:392,
1987.
[932] R. Myrzakulov. Accelerating universe from F(T) gravities. ArXiv e-prints, June 2010, 1006.1120.
[933] R. Myrzakulov. F(T) gravity and k-essence. ArXiv e-prints, Aug. 2010, 1008.4486.
[934] R. Nagata, T. Chiba, and N. Sugiyama. Observational consequences of the evolution of primordial
fluctuations in scalar-tensor cosmology. Phys. Rev. D, 66(10):103510, Nov 2002.
[935] R. Nagata, T. Chiba, and N. Sugiyama. WMAP constraints on scalar-tensor cosmology and
the variation of the gravitational constant. Phys. Rev. D, 69(8):083512, Apr. 2004, arXiv:astro-
ph/0311274.
[936] M. Nakahara. Geometry, topology, and physics. Taylor & Francis, 2003.
[937] S. Nakamura and A. Hosoya. Density Fluctuation in Extended Inflationary Universe. Progress
of Theoretical Physics, 87:401–407, Feb. 1992.
[938] H. Nariai. On the Green’s Function in an Expanding Universe and Its Role in the Problem of
Mach’s Principle. Progress of Theoretical Physics, 40:49–59, July 1968.
[939] H. Nariai. Gravitational Instability in the Brans-Dicke Cosmology. Progress of Theoretical
Physics, 42:544–554, Sept. 1969.
[940] T. Narikawa and K. Yamamoto. Characterizing the linear growth rate of cosmological density
perturbations in an f(R) model. Phys. Rev. D, 81(4):043528, Feb. 2010, 0912.1445.
[941] P. Nath and M. Yamaguchi. Effects of Extra Space-time Dimensions on the Fermi Constant.
Phys. Rev., D60:116004, 1999, hep-ph/9902323.
[942] A. Navarro, A. Serna, and J. M. Alimi. Asymptotic and exact solutions of perfect-fluid scalar-
tensor cosmologies. Phys. Rev. D, 59(12):124015, May 1999.
[943] I. Navarro and K. van Acoleyen. On the Newtonian limit of generalized modified gravity models.
Physics Letters B, 622:1–5, Aug. 2005, arXiv:gr-qc/0506096.
[944] I. Navarro and K. Van Acoleyen. Consistent long distance modification of gravity from inverse
powers of the curvature. J. Cosmology Astropart. Phys., 3:8, Mar. 2006, arXiv:gr-qc/0511045.
[945] I. Navarro and K. Van Acoleyen. f(R) actions, cosmic acceleration and local tests of gravity. J.
Cosmology Astropart. Phys., 2:22, Feb. 2007, arXiv:gr-qc/0611127.
[946] J. F. Navarro, C. S. Frenk, and S. D. White. The Structure of cold dark matter halos. Astrophys.J.,
462:563–575, 1996, astro-ph/9508025.
[947] B. Nayak, A. Majumdar, and L. Singh. Astrophysical constraints on primordial black holes in
brans-dicke theory. Journal of Cosmology and Astroparticle Physics, 2010(08):039, 2010.
[948] B. Nayak, L. P. Singh, and A. S. Majumdar. Effect of accretion on primordial black holes in
brans-dicke theory. Phys. Rev. D, 80(2):023529, Jul 2009.
[949] W. Nelson. Restricting fourth-order gravity via cosmology. Phys. Rev. D, 82(12):124044, Dec.
2010, 1012.3353.
[950] W. Nelson. Static solutions for fourth order gravity. Phys. Rev. D, 82(10):104026, Nov. 2010,
1010.3986.
[951] W. Nelson, J. Ochoa, and M. Sakellariadou. Constraining the Noncommutative Spectral Action
via Astrophysical Observations. Physical Review Letters, 105(10):101602, Sept. 2010, 1005.4279.
[952] W. Nelson, J. Ochoa, and M. Sakellariadou. Gravitational waves in the spectral action of non-
commutative geometry. Phys. Rev. D, 82(8):085021, Oct. 2010, 1005.4276.
[953] W. Nelson and M. Sakellariadou. Inflation mechanism in asymptotic noncommutative geometry.
Physics Letters B, 680:263–266, Sept. 2009, 0903.1520.
[954] W. Nelson and M. Sakellariadou. Cosmology and the noncommutative approach to the standard
model. Phys. Rev. D, 81(8):085038, Apr. 2010, 0812.1657.
[955] A. Neronov and I. Sachs. On metric perturbations in brane-world scenarios. Phys. Lett.,
B513:173–178, 2001, hep-th/0011254.
[956] S. Nesseris, A. De Felice, and S. Tsujikawa. Observational constraints on Galileon cosmology.
Phys.Rev., D82:124054, 2010, 1010.0407.
[957] S. Nesseris and A. Mazumdar. Newton’s constant in f (R, Rµν Rµν , R) theories of gravity and
constraints from BBN. Phys. Rev. D, 79(10):104006, May 2009, 0902.1185.
[958] S. Nesseris, F. Piazza, and S. Tsujikawa. The universe is accelerating. Do we need a new mass
scale? Phys.Lett., B689:122–128, 2010, 0910.3949.
[959] I. P. Neupane and B. M. N. Carter. Dynamical relaxation of dark energy: A solution to early infla-
tion, late-time acceleration and the cosmological constant problem. Physics Letters B, 638:94–99,
July 2006, arXiv:hep-th/0510109.
299
[960] I. P. Neupane and B. M. N. Carter. Towards inflation and dark energy cosmologies from modified
Gauss Bonnet theory. J. Cosmology Astropart. Phys., 6:4, June 2006, arXiv:hep-th/0512262.
[961] P. Nicolini, Noncommutative Black Holes, The Final Appeal To Quantum Gravity: A Review
Int. J. Mod. Phys. A24 (2009) 1229-1308. [arXiv:0807.1939 [hep-th]].
[962] A. Nicolis and R. Rattazzi. Classical and quantum consistency of the DGP model. JHEP,
0406:059, 2004, hep-th/0404159.
[963] A. Nicolis, R. Rattazzi, and E. Trincherini. The Galileon as a local modification of gravity.
Phys.Rev., D79:064036, 2009, 0811.2197.
[964] A. Nicolis, R. Rattazzi, and E. Trincherini. Energy’s and amplitudes’ positivity. JHEP, 1005:095,
2010, 0912.4258.
[965] H.-P. Nilles, A. Papazoglou, and G. Tasinato. Selftuning and its footprints. Nucl.Phys., B677:405–
429, 2004, hep-th/0309042.
[966] G. Niz, A. Padilla, and H. K. Kunduri. Braneworld Isotropization and Magnetic Fields. JCAP,
0804:012, 2008, 0801.3462.
[967] D. R. Noakes. The initial value formulation of higher derivative gravity. Journal of Mathematical
Physics, 24:1846–1850, July 1983.
[968] S. Nojiri and S. D. Odintsov. Modified gravity with negative and positive powers of curva-
ture: Unification of inflation and cosmic acceleration. Phys. Rev. D, 68(12):123512, Dec. 2003,
arXiv:hep-th/0307288.
[969] S. Nojiri and S. D. Odintsov. Introduction to Modified Gravity and Gravitational Alternative for
Dark Energy. Int. J. Geom. Meth. Mod. Phys., 4:115, 2007, arXiv:hep-th/0601213.
[970] S. Nojiri and S. D. Odintsov. Modified Gravity with ln R Terms and Cosmic Acceleration. General
Relativity and Gravitation, 36:1765–1780, Aug. 2004, arXiv:hep-th/0308176.
[971] S. Nojiri and S. D. Odintsov. Modified Gauss Bonnet theory as gravitational alternative for dark
energy. Physics Letters B, 631:1–6, Dec. 2005, arXiv:hep-th/0508049.
[972] S. Nojiri and S. D. Odintsov. Modified f(R) gravity consistent with realistic cosmology: From
a matter dominated epoch to a dark energy universe. Phys. Rev. D, 74(8):086005, Oct. 2006,
arXiv:hep-th/0608008.
[973] S. Nojiri and S. D. Odintsov. Modified gravity as an alternative for ΛCDM cosmology. Journal
of Physics A Mathematical General, 40:6725–6732, June 2007, arXiv:hep-th/0610164.
[974] S. Nojiri and S. D. Odintsov. Unifying inflation with ΛCDM epoch in modified f(R) gravity
consistent with Solar System tests. Physics Letters B, 657:238–245, Dec. 2007, 0707.1941.
[975] S. Nojiri and S. D. Odintsov. Future evolution and finite-time singularities in f(R) gravity unifying
inflation and cosmic acceleration. Phys. Rev. D, 78(4):046006, Aug. 2008, 0804.3519.
[976] S. Nojiri and S. D. Odintsov. Modified f(R) gravity unifying Rm inflation with the ΛCDM epoch.
Phys. Rev. D, 77(2):026007, Jan. 2008, 0710.1738.
[977] S. Nojiri and S. D. Odintsov. Modified non-local-f(R) gravity as the key for the inflation and
dark energy. Physics Letters B, 659:821–826, Jan. 2008, 0708.0924.
[978] S. Nojiri and S. D. Odintsov. Unified cosmic history in modified gravity: from F(R) theory to
Lorentz non-invariant models. Phys. Rep., 505:59–144, Aug. 2011, 1011.0544.
[979] S. Nojiri, S. D. Odintsov, and O. G. Gorbunova. Dark energy problem: from phantom theory
to modified Gauss Bonnet gravity. Journal of Physics A Mathematical General, 39:6627–6633,
May 2006, arXiv:hep-th/0510183.
[980] S. Nojiri, S. D. Odintsov, and D. Sáez-Gómez. Cosmological reconstruction of realistic modified
f(R) gravities. Physics Letters B, 681:74–80, Oct. 2009, 0908.1269.
[981] S. Nojiri, S. D. Odintsov, and M. Sami. Dark energy cosmology from higher-order, string-inspired
gravity, and its reconstruction. Phys. Rev. D, 74(4):046004, Aug. 2006, arXiv:hep-th/0605039.
[982] S. Nojiri, S. D. Odintsov, and M. Sasaki. Gauss-Bonnet dark energy. Phys. Rev. D, 71(12):123509,
June 2005, arXiv:hep-th/0504052.
[983] S. Nojiri, S. D. Odintsov, and P. V. Tretyakov. From Inflation to Dark Energy in the Non-Minimal
Modified Gravity. Progress of Theoretical Physics Supplement, 172:81–89, 2008, 0710.5232.
[984] M. R. Nolta et al. Five-Year Wilkinson Microwave Anisotropy Probe (WMAP) Observations:
Angular Power Spectra. Astrophys. J. Supp., 180:296–305, 2009, 0803.0593.
[985] G. Nordstrom. On the possibility of unifying the electromagnetic and the gravitational fields.
Phys. Z., 15:504–506, 1914, physics/0702221.
[986] K. Nordtvedt. Equivalence Principle for Massive Bodies. I. Phenomenology. Physical Review,
169:1014–1016, May 1968.
[987] K. Nordtvedt. Probing gravity to the second post-Newtonian order and to one part in 10 to the
7th using the spin axis of the sun. ApJ, 320:871–874, Sept. 1987.
300
 [988] K. Nordtvedt, Jr. PostNewtonian metric for a general class of scalar tensor gravitational theories
and observational consequences. Astrophys. J., 161:1059–1067, 1970.
[989] K. Nordtvedt, Jr. and C. M. Will. Conservation Laws and Preferred Frames in Relativistic
Gravity. II. Experimental Evidence to Rule Out Preferred-Frame Theories of Gravity. ApJ,
177:775, Nov. 1972.
[990] A. Núñez and S. Solganik. Ghost constraints on modified gravity. Physics Letters B, 608:189–193,
Feb. 2005, arXiv:hep-th/0411102.
[991] N. J. Nunes and E. J. Copeland. Tracking quintessential inflation from brane worlds. Phys. Rev.,
D66:043524, 2002, astro-ph/0204115.
[992] A. M. Nzioki, S. Carloni, R. Goswami, and P. K. S. Dunsby. New framework for studying
spherically symmetric static solutions in f(R) gravity. Phys. Rev. D, 81(8):084028, Apr. 2010,
0908.3333.
[993] J. O’ Hanlon and B. O. J. Tupper. Vacuum-field solutions in the Brans-Dicke theory. Nuovo
Cim., B7:305–312, 1972.
[994] N. Ohta, and T. Torii. Black Holes in the Dilatonic Einstein-Gauss-Bonnet Theory in Various
Dimensions. III - Asymptotically AdS Black Holes with k = +/-1. Prog. Theo. Phys., 121:959-981,
May 2009, 0902.4072.
[995] N. Ohta, and T. Torii. Black Holes in the Dilatonic Einstein-Gauss-Bonnet Theory in Various
Dimensions. IV - Topological Black Holes with and without Cosmological Term. Prog. Theo.
Phys., 122:1477-1500, Dec. 2009, 0908.3918.
[996] N. Ohta, and T. Torii. Global Structure of Black Holes in String Theory with Gauss-Bonnet
Correction in Various Dimensions. Prog. Theo. Phys., 124:207-225, Aug. 2010, 1004.2779.
[997] Y. Okada. Inflation in Kaluza-Klein cosmology. Phys. Lett., B150:103, 1985.
[998] G. J. Olmo. The gravity lagrangian according to solar system experiments. Phys. Rev. Lett.,
95(26):261102, Dec 2005.
[999] G. J. Olmo. Post-Newtonian constraints on f(R) cosmologies in metric and Palatini formalism.
Phys. Rev. D, 72(8):083505, Oct. 2005, arXiv:gr-qc/0505135.
[1000] G. J. Olmo. The Gravity Lagrangian According to Solar System Experiments. Physical Review
Letters, 95(26):261102, Dec. 2005, arXiv:gr-qc/0505101.
[1001] G. J. Olmo. Limit to general relativity in f(R) theories of gravity. Phys. Rev. D, 75(2):023511,
Jan. 2007, arXiv:gr-qc/0612047.
[1002] G. J. Olmo, H. Sanchis-Alepuz, and S. Tripathi. Dynamical aspects of generalized Palatini
theories of gravity. Phys. Rev. D, 80(2):024013, July 2009, 0907.2787.
[1003] M. Ostrogradsky. Memoires de l’Academie Imperiale des Science de Saint-Petersbourg, 4:385,
1850.
[1004] A. Oukouiss. Exact and asymptotic solutions to tensor-scalar cosmologies. Nuclear Physics B,
486:413–422, Feb. 1997.
[1005] J. M. Overduin and P. S. Wesson. Kaluza-Klein gravity. Phys. Rept., 283:303–380, 1997, gr-
qc/9805018.
[1006] H. Oyaizu. Nonlinear evolution of f(R) cosmologies. I. Methodology. Phys. Rev. D, 78(12):123523,
Dec. 2008, 0807.2449.
[1007] H. Oyaizu, M. Lima, and W. Hu. Nonlinear evolution of f(R) cosmologies. II. Power spectrum.
Phys. Rev. D, 78(12):123524, Dec. 2008, 0807.2462.
[1008] A. Padilla. Brane world cosmology and holography. 2002, hep-th/0210217.
[1009] A. Padilla. CFTs on non-critical braneworlds. Phys. Lett., B528:274–282, 2002, hep-th/0111247.
[1010] A. Padilla. Surface terms and the Gauss-Bonnet Hamiltonian. Class.Quant.Grav., 20:3129–3150,
2003, gr-qc/0303082.
[1011] A. Padilla. Ghost-free braneworld bigravity. Class. Quant. Grav., 21:2899–2918, 2004, hep-
th/0402079.
[1012] A. Padilla. Cosmic acceleration from asymmetric branes. Class. Quant. Grav., 22:681–694, 2005,
hep-th/0406157.
[1013] A. Padilla. Infra-red modification of gravity from asymmetric branes. Class. Quant. Grav.,
22:1087–1104, 2005, hep-th/0410033.
[1014] A. Padilla. A covariant approach to braneworld holography. Class. Quant. Grav., 23:3983–3992,
2006, hep-th/0512083.
[1015] A. Padilla. A short review of ‘DGP Specteroscopy’. J. Phys., A40:6827–6834, 2007, hep-
th/0610093.
[1016] A. Padilla. The good, the bad and the ugly .... of Horava gravity. J.Phys.Conf.Ser., 259:012033,
2010, 1009.4074.
301
[1017] A. Padilla, P. M. Saffin, and S.-Y. Zhou. Bi-galileon theory I: motivation and formulation. 2010,
1007.5424.
[1018] A. Padilla, P. M. Saffin, and S.-Y. Zhou. Bi-galileon theory II: Phenomenology. 2010, 1008.3312.
[1019] A. Padilla, P. M. Saffin, and S.-Y. Zhou. Multi-galileons, solitons and Derrick’s theorem. 2010,
1008.0745.
[1020] A. Palatini. Deduzione invariantiva deUe equazioni gravitazionali dal principio di Hamilton.
Rend. Circ. Mat. Palermo, 43:203, 1919.
[1021] E. Papantonopoulos and A. Papazoglou. Cosmological evolution of a purely conical codimension-2
brane world. JHEP, 0509:012, 2005, hep-th/0507278.
[1022] E. Papantonopoulos, A. Papazoglou, and V. Zamarias. Regularization of conical singularities in
warped six- dimensional compactifications. JHEP, 03:002, 2007, hep-th/0611311.
[1023] E. Papantonopoulos, A. Papazoglou, and V. Zamarias. Induced cosmology on a regularized brane
in six- dimensional flux compactification. Nucl. Phys., B797:520–536, 2008, 0707.1396.
[1024] A. Papapetrou. Spinning test particles in general relativity. 1 Proc.Roy.Soc.Lond., A209: 248,
1951.
[1025] A. Papazoglou and T. P. Sotiriou. Strong coupling in extended Horava-Lifshitz gravity. Phys.Lett.,
B685:197–200, 2010, 0911.1299.
[1026] C. Park, J. Hwang, and H. Noh. Constraints on a f(R) gravity dark energy model with early
scaling evolution. ArXiv e-prints, Dec. 2010, 1012.1662.
[1027] B. C. Paul, S. Mukherjee, and R. Tavakol. Singular instantons in higher derivative theories.
Phys. Rev. D, 65(6):064020, Mar. 2002, arXiv:hep-th/0201009.
[1028] J. A. Peacock. Cosmological Physics. Jan. 1999.
[1029] E. Pechlaner and R. Sexl. On quadratic lagrangians in General Relativity. Communications in
Mathematical Physics, 2:165–175, Dec. 1966.
[1030] P. Peebles. The void phenomenon. Astrophys.J., 557:495–504, 2001, astro-ph/0101127.
[1031] P. J. E. Peebles. The large-scale structure of the universe. 1980.
[1032] P. Peldan. Actions for gravity, with generalizations: A Review. Class. Quant. Grav., 11:1087–
1132, 1994, gr-qc/9305011.
[1033] S. E. Perez Bergliaffa Constraining f(R) theories with the energy conditions. Phys. Lett. B,
642:311–314, 2006, gr-qc/0608072.
[1034] S. Perlmutter et al. Measurements of Omega and Lambda from 42 high redshift supernovae.
Astrophys.J., 517:565–586, 1999, astro-ph/9812133. The Supernova Cosmology Project.
[1035] F. Perrotta, C. Baccigalupi, and S. Matarrese. Extended quintessence. Phys. Rev. D,
61(2):023507, Dec 1999.
[1036] F. Piazza. Modifying Gravity in the Infra-Red by imposing an ‘Ultra-Strong’ Equivalence Prin-
ciple. 2009, 0904.4299.
[1037] F. Piazza. The IR-Completion of Gravity: What happens at Hubble Scales? New J.Phys.,
11:113050, 2009, 0907.0765.
[1038] L. Pilo, R. Rattazzi, and A. Zaffaroni. The fate of the radion in models with metastable graviton.
JHEP, 07:056, 2000, hep-th/0004028.
[1039] S. Pireaux and J. Rozelot. Solar quadrupole moment and purely relativistic gravitation contri-
butions to Mercury’s perihelion advance. Ap&SS, 284:1159–1194, 2003, arXiv:astro-ph/0109032.
[1040] E. V. Pitjeva. Experimental testing of relativistic effects, variability of the gravitational constant
and topography of Mercury surface from radar observations 1964-1989. Celestial Mechanics and
Dynamical Astronomy, 55:313–321, Apr. 1993.
[1041] E. V. Pitjeva. Modern Numerical Ephemerides of Planets and the Importance of Ranging Ob-
servations for Their Creation. Celestial Mechanics and Dynamical Astronomy, 80:249–271, July
2001.
[1042] J. F. Plebański. On the separation of Einsteinian substructures. Journal of Mathematical Physics,
18:2511–2520, Dec. 1977.
[1043] L. Pogosian and A. Silvestri. Pattern of growth in viable f(R) cosmologies. Phys. Rev. D,
77(2):023503, Jan. 2008, 0709.0296.
[1044] J. Polchinski. Dirichlet-Branes and Ramond-Ramond Charges. Phys. Rev. Lett., 75:4724–4727,
1995, hep-th/9510017.
[1045] M. Pospelov and Y. Shang. On Lorentz violation in Horava-Lifshitz type theories. 2010, 1010.5249.
[1046] O. Pujolas, I. Sawicki, A. Vikman, The Imperfect Fluid behind Kinetic Gravity Braiding
[arXiv:1103.5360 [hep-th]].
[1047] T. Qiu and K. C. Yang. Non-Gaussianities of single field inflation with nonminimal coupling.
Phys. Rev. D, 83(8):084022, 2011, 1012.1697.
302
[1048] F. Quevedo. Lectures on string/brane cosmology. Class. Quant. Grav., 19:5721–5779, 2002,
hep-th/0210292.
[1049] I. Quiros, Y. Leyva, and Y. Napoles. Note on stability of de Sitter solutions of f(R) theories.
Phys. Rev. D, 80(2):024022, July 2009, 0906.1190.
[1050] A. Rajaraman. Newtonian Gravity in Theories with Inverse Powers of R. ArXiv Astrophysics
e-prints, Nov. 2003, arXiv:astro-ph/0311160.
[1051] L. Randall and R. Sundrum. A large mass hierarchy from a small extra dimension. Phys. Rev.
Lett., 83:3370–3373, 1999, hep-ph/9905221.
[1052] L. Randall and R. Sundrum. An alternative to compactification. Phys. Rev. Lett., 83:4690–4693,
1999, hep-th/9906064.
[1053] R. H. Rapp. An estimate of equatorial gravity from terrestrial and satellite data. Geo-
phys. Res. Lett., 14:730–732, July 1987.
[1054] S. M. M. Rasouli, M. Farhoudi, and H. R. Sepangi. An anisotropic cosmological model in a
modified Brans-Dicke theory. Classical and Quantum Gravity, 28(15):155004, 2011, 1105.5086.
[1055] P. Rastall. A theory of gravity. Canadian Journal of Physics, 54:66–75, Jan. 1976.
[1056] P. Rastall. A note on a theory of gravity. Canadian Journal of Physics, 55:38–42, Jan. 1977.
[1057] M. J. Rebouças and J. Santos. Gödel-type universes in f(R) gravity. Phys. Rev. D, 80(6):063009,
Sept. 2009, 0906.5354.
[1058] B. A. Reid, W. J. Percival, D. J. Eisenstein, L. Verde, D. N. Spergel, R. A. Skibba, N. A.
Bahcall, T. Budavari, J. A. Frieman, M. Fukugita, J. R. Gott, J. E. Gunn, Ž. Ivezić, G. R.
Knapp, R. G. Kron, R. H. Lupton, T. A. McKay, A. Meiksin, R. C. Nichol, A. C. Pope, D. J.
Schlegel, D. P. Schneider, C. Stoughton, M. A. Strauss, A. S. Szalay, M. Tegmark, M. S. Vogeley,
D. H. Weinberg, D. G. York, and I. Zehavi. Cosmological constraints from the clustering of the
Sloan Digital Sky Survey DR7 luminous red galaxies. MNRAS, 404:60–85, May 2010, 0907.1659.
[1059] S. Renaux-Petel, Orthogonal non-Gaussianities from Dirac-Born-Infeld Galileon inflation Class.
Quant. Grav. 28 (2011) 182001. [arXiv:1105.6366 [astro-ph.CO]].
[1060] R. Reyes, R. Mandelbaum, U. Seljak, T. Baldauf, J. E. Gunn, L. Lombriser, and R. E. Smith.
Confirmation of general relativity on large scales from weak lensing and galaxy velocities. Nature,
464:256–258, Mar. 2010, 1003.2185.
[1061] A. Riazuelo and J.-P. Uzan. Cosmological observations in scalar-tensor quintessence. Phys. Rev.
D, 66(2):023525, Jul 2002.
[1062] A. Riazuelo, F. Vernizzi, D. A. Steer, and R. Durrer. Gauge invariant cosmological perturbation
theory for braneworlds. 2002, hep-th/0205220.
[1063] A. G. Riess et al. Observational Evidence from Supernovae for an Accelerating Universe and a
Cosmological Constant. Astron. J., 116:1009–1038, 1998, astro-ph/9805201.
[1064] A. G. Riess et al. Type Ia supernova discoveries at z &gt; 1 from the Hubble Space Telescope:
Evidence for past deceleration and constraints on dark energy evolution. Astrophys.J., 607:665–
687, 2004, astro-ph/0402512.
[1065] S. Rippl, H. van Elst, R. Tavakol, and D. Taylor. Kinematics and dynamics of f(R) theories of
gravity. General Relativity and Gravitation, 28:193–205, Feb. 1996, arXiv:gr-qc/9511010.
[1066] D. C. Rodrigues. Evolution of Anisotropies in Eddington-Born-Infeld Cosmology. Phys.Rev.,
D78:063013, 2008, 0806.3613.
[1067] N. Rosen. General Relativity and Flat Space. II. Physical Review, 57:150–153, Jan. 1940.
[1068] N. Rosen. A bi-metric theory of gravitation. General Relativity and Gravitation, 4:435–447, Dec.
1973.
[1069] N. Rosen. Bimetric gravitation theory on a cosmological basis. General Relativity and Gravita-
tion, 9:339–351, Apr. 1978.
[1070] E. Rosenthal. Extended Palatini action for general relativity. Phys. Rev. D, 80(8):084029, Oct.
2009, 0809.2053.
[1071] N. P. Ross, J. da Angela, T. Shanks, D. A. Wake, R. D. Cannon, et al. The 2dF-SDSS
LRG and QSO Survey: The 2-Point Correlation Function and Redshift-Space Distortions.
Mon.Not.Roy.Astron.Soc., 2006, astro-ph/0612400.
[1072] V. A. Rubakov. Large and infinite extra dimensions: An introduction. Phys. Usp., 44:871–893,
2001, hep-ph/0104152.
[1073] V. A. Rubakov. Strong coupling in brane-induced gravity in five dimensions. 2003, hep-
th/0303125.
[1074] V. A. Rubakov and M. E. Shaposhnikov. Do We Live Inside a Domain Wall? Phys. Lett.,
B125:136–138, 1983.
[1075] V. A. Ruban and A. M. Finkel’Shtejn. Generalization of the Taub-Kazner cosmological metric
303
 in the scalar-tensor gravitation theory. Nuovo Cimento Lettere, 5:289–293, 1972.
[1076] H. Ruegg and M. Ruiz-Altaba. The Stuckelberg field. Int.J.Mod.Phys., A19:3265–3348, 2004,
hep-th/0304245.
[1077] M. L. Ruggiero. Gravitational lensing and f(R) theories in the Palatini approach. General
Relativity and Gravitation, 41:1497–1509, July 2009, 0712.3218.
[1078] M. L. Ruggiero and L. Iorio. Solar System planetary orbital motions and f(R) theories of gravity.
J. Cosmology Astropart. Phys., 1:10, Jan. 2007, arXiv:gr-qc/0607093.
[1079] G. Rybka, M. Hotz, L. J. Rosenberg, S. J. Asztalos, G. Carosi, C. Hagmann, D. Kinion, K. van
Bibber, J. Hoskins, C. Martin, P. Sikivie, D. B. Tanner, R. Bradley, and J. Clarke. A Search for
Scalar Chameleons with ADMX. ArXiv e-prints, Apr. 2010, 1004.5160.
[1080] S. Rydbeck, M. Fairbairn, and A. Goobar. Testing the DGP model with ESSENCE. JCAP,
0705:003, 2007, astro-ph/0701495.
[1081] E. Sagi. Preferred frame parameters in the tensor-vector-scalar theory of gravity and its gener-
alization. Phys. Rev., D80:044032, 2009, 0905.4001.
[1082] E. Sagi and J. D. Bekenstein. Black holes in the TeVeS theory of gravity and their thermody-
namics. Phys. Rev., D77:024010, 2008, 0708.2639.
[1083] D. Sahdev. Perfect fluid higher dimensional cosmologies. Phys. Rev., D30:2495–2507, 1984.
[1084] D. Sahdev. Towards a Realistic Kaluza-Klein Cosmology. Phys. Lett., B137:155–159, 1984.
[1085] D. Sahdev. Perfect fluid higher dimensional cosmologies. 2. Phys. Rev., D39:3155, 1989.
[1086] V. Sahni, Y. Shtanov, Brane world models of dark energy JCAP 0311 (2003) 014. [astro-
ph/0202346].
[1087] V. Sahni, M. Sami, and T. Souradeep. Relic gravity waves from brane world inflation. Phys.
Rev., D65:023518, 2002, gr-qc/0105121.
[1088] N. Sakai and K. Maeda. Bubble dynamics and space-time structure in extended inflation.
Phys. Rev. D, 48:5570–5575, Dec. 1993.
[1089] N. Sakai, J. Yokoyama, and K. Maeda. Monopole inflation in Brans-Dicke theory. Phys. Rev. D,
59(10):103504, May 1999, arXiv:gr-qc/9811024.
[1090] M. Sami, A. Toporensky, P. Tretjakov, and S. Tsujikawa. The fate of (phantom) dark energy
universe with string curvature corrections. Physics Letters B, 619:193–200, July 2005, arXiv:hep-
th/0504154.
[1091] H. Sanctuary and R. Sturani. Effective field theory analysis of the self-interacting chameleon.
General Relativity and Gravitation, 42:1953–1967, Aug. 2010, 0809.3156.
[1092] R. Sanders. The Universal Faber-Jackson Relation. 2010, 1002.2765.
[1093] R. H. Sanders. A stratified framework for scalar-tensor theories of Modified Dynamics. Astrophys.
J., 480:492–502, 1997, astro-ph/9612099.
[1094] R. H. Sanders and S. S. McGaugh. Modified Newtonian dynamics as an alternative to dark
matter. Ann.Rev.Astron.Astrophys., 40:263–317, 2002, astro-ph/0204521.
[1095] D. I. Santiago, D. Kalligas, and R. V. Wagoner. Scalar-tensor cosmologies and their late time
evolution. Phys. Rev. D, 58(12):124005, Nov 1998.
[1096] C. Santos and R. Gregory. Cosmology in Brans-Dicke Theory with a Scalar Potential. Annals of
Physics, 258:111–134, July 1997, arXiv:gr-qc/9611065.
[1097] A. K. Sanyal. If Gauss Bonnet interaction plays the role of dark energy. Physics Letters B,
645:1–5, Feb. 2007, arXiv:astro-ph/0608104.
[1098] E. N. Saridakis. Aspects of Horava-Lifshitz cosmology. 2011, 1101.0300.
[1099] A. Savas Arapoglu, C. Deliduman, and K. Yavuz Eksi. Constraints on Perturbative f(R) Gravity
via Neutron Stars. ArXiv e-prints, Mar. 2010, 1003.3179.
[1100] I. Savonije and E. P. Verlinde. CFT and entropy on the brane. Phys. Lett., B507:305–311, 2001,
hep-th/0102042.
[1101] I. Sawicki and S. M. Carroll. Cosmological structure evolution and cmb anisotropies in DGP
braneworlds. 2005, astro-ph/0510364.
[1102] I. Sawicki and W. Hu. Stability of cosmological solutions in f(R) models of gravity. Phys. Rev. D,
75(12):127502, June 2007, arXiv:astro-ph/0702278.
[1103] I. Sawicki, Y.-S. Song, and W. Hu. Near-horizon solution for DGP perturbations. Phys. Rev.,
D75:064002, 2007, astro-ph/0606285.
[1104] D. W. Sciama. On the analogy between charge and spin in General Relativity. (in Recent
developments in General Relativity), Pergamon + PWD (Oxford), 415, 1962.
[1105] D. W. Sciama. The physical structure of General Relativity. Rev. Mod. Phys., 36, 463, 1962.
[1106] B. E. Schaefer. The Hubble Diagram to Redshift greater than 6 from 69 Gamma-Ray Bursts.
Astrophys.J., 660:16–46, 2007, astro-ph/0612285.
304
[1107] M. A. Scheel, S. L. Shapiro, and S. A. Teukolsky. Collapse to black holes in brans-dicke theory.
ii. comparison with general relativity. Phys. Rev. D, 51(8):4236–4249, Apr 1995.
[1108] C. A. O. Schelpe. Chameleon-photon mixing in a primordial magnetic field. Phys. Rev. D,
82(4):044033, Aug. 2010, 1003.0232.
[1109] C. Schimd, J. Uzan, and A. Riazuelo. Weak lensing in scalar-tensor theories of gravity.
Phys. Rev. D, 71(8):083512, Apr. 2005, arXiv:astro-ph/0412120.
[1110] S. Schlamminger, K. Choi, T. A. Wagner, J. H. Gundlach, and E. G. Adelberger. Test of the
Equivalence Principle Using a Rotating Torsion Balance. Physical Review Letters, 100(4):041101,
Feb. 2008, 0712.0607.
[1111] F. Schmidt. Cosmological Simulations of Normal-Branch Braneworld Gravity. Phys.Rev.,
D80:123003, 2009, 0910.0235.
[1112] F. Schmidt. Self-consistent cosmological simulations of dgp braneworld gravity. Phys. Rev.,
D80:043001, 2009.
[1113] F. Schmidt, W. Hu, and M. Lima. Spherical Collapse and the Halo Model in Braneworld Gravity.
Phys.Rev., D81:063005, 2010, 0911.5178.
[1114] F. Schmidt, M. Lima, H. Oyaizu, and W. Hu. Nonlinear evolution of f(R) cosmologies. III. Halo
statistics. Phys. Rev. D, 79(8):083518, Apr. 2009, 0812.0545.
[1115] F. Schmidt, A. Vikhlinin, and W. Hu. Cluster constraints on f(R) gravity. Phys. Rev. D,
80(8):083505, Oct. 2009, 0908.2457.
[1116] H. Schmidt. On static spherically symmetric solutions of the Bach-Einstein gravitational field
equations. Astronomische Nachrichten, 306:67–70, 1985.
[1117] H. Schmidt. Variational derivatives of arbitrarily high order and multi-inflation cosmological
models. Classical and Quantum Gravity, 7:1023–1031, June 1990.
[1118] H. Schmidt. Fourth Order Gravity: Equations, History, and Applications to Cosmology. Inter-
national Journal of Geometric Methods in Modern Physics, 4:209, 2007, arXiv:gr-qc/0602017.
[1119] B. F. Schutz. Perfect Fluids in General Relativity: Velocity Potentials and a Variational Principle.
Phys. Rev. D, 2:2762–2773, Dec. 1970.
[1120] M. D. Schwartz. The emergence of localized gravity. Phys. Lett., B502:223–228, 2001, hep-
th/0011177.
[1121] R. Scoccimarro. Large-Scale Structure in Brane-Induced Gravity I. Perturbation Theory.
Phys.Rev., D80:104006, 2009, 0906.4545.
[1122] S. S. Seahra and C. G. Böhmer. Einstein static universes are unstable in generic f(R) models.
Phys. Rev. D, 79(6):064009, Mar. 2009, 0901.0892.
[1123] S. S. Seahra and W. Hu. Analytic Description of DGP Perturbations on All Scales. Phys.Rev.,
D82:124015, 2010, 1007.4242.
[1124] M. D. Seifert. Stability of spherically symmetric solutions in modified theories of gravity. Phys.
Rev., D76:064002, 2007, gr-qc/0703060.
[1125] A. Serna and J. M. Alimi. Scalar-tensor cosmological models. Phys. Rev. D, 53(6):3074–3086,
Mar 1996.
[1126] A. Serna, R. Dominguez-Tenreiro, and G. Yepes. Primordial nucleosynthesis bounds on the
Brans-Dicke theory. ApJ, 391:433–442, June 1992.
[1127] T. R. Seshadri. Generation of density perturbations by inflation in scalar-tensor gravity theories.
Physics Letters B, 277:67–72, Feb. 1992.
[1128] H. Shan, M. Feix, B. Famaey, and H. Zhao. An analytic model for non-spherical lenses in covariant
MOdified Newtonian Dynamics. Mon. Not. Roy. Astron. Soc., 387:1303–1312, 2008, 0804.2668.
[1129] C. Shao, R. Cai, B. Wang, and R. Su. An alternative explanation of the conflict between 1/R
gravity and solar system tests. Physics Letters B, 633:164–166, Feb. 2006, arXiv:gr-qc/0511034.
[1130] I. I. Shapiro. Fourth Test of General Relativity. Physical Review Letters, 13:789–791, Dec. 1964.
[1131] S. S. Shapiro, J. L. Davis, D. E. Lebach, and J. S. Gregory. Measurement of the Solar Gravita-
tional Deflection of Radio Waves using Geodetic Very-Long-Baseline Interferometry Data, 1979
1999. Physical Review Letters, 92(12):121101, Mar. 2004.
[1132] M. Sharif and M. Farasat Shamir. Exact solutions of Bianchi-type I and V spacetimes in the f(R)
theory of gravity. Classical and Quantum Gravity, 26(23):235020, Dec. 2009, 0910.5787.
[1133] M. Sharif and M. F. Shamir. Non-vacuum Bianchi types I and V in f(R) gravity. General
Relativity and Gravitation, 42:2643–2655, Nov. 2010, 1005.2798.
[1134] D. J. Shaw and J. D. Barrow. Local experiments see cosmologically varying constants. Phys.
Lett., B639:596–599, 2006, gr-qc/0512117.
[1135] D. J. Shaw and J. D. Barrow. The local effects of cosmological variations in physical ‘con-
stants’ and scalar fields. I: Spherically symmetric spacetimes. Phys. Rev., D73:123505, 2006,
305
 gr-qc/0512022.
[1136] D. J. Shaw and J. D. Barrow. The local effects of cosmological variations in physical ‘constants’
and scalar fields. II: Quasi-spherical spacetimes. Phys. Rev., D73:123506, 2006, gr-qc/0601056.
[1137] R. K. Sheth and G. Tormen. Large scale bias and the peak background split.
Mon.Not.Roy.Astron.Soc., 308:119, 1999, astro-ph/9901122.
[1138] T. Shiromizu and D. Ida. Anti-de Sitter no hair, AdS/CFT and the brane-world. Phys. Rev.,
D64:044015, 2001, hep-th/0102035.
[1139] T. Shiromizu and K. Koyama. Low energy effective theory for two brane systems: Covariant
curvature formulation. Phys. Rev., D67:084022, 2003, hep-th/0210066.
[1140] T. Shiromizu, K.-i. Maeda, and M. Sasaki. The Einstein equation on the 3-brane world. Phys.Rev.,
D62:024012, 2000, gr-qc/9910076.
[1141] Y. Shtanov, V. Sahni, Bouncing brane worlds. Phys. Lett. B557 (2003) 1-6. [gr-qc/0208047].
[1142] Y. Shtanov, V. Sahni, Unusual cosmological singularities in brane world models. Class. Quant.
Grav. 19 (2002) L101-L107. [gr-qc/0204040].
[1143] Y. Shtanov, V. Sahni, A. Shafieloo, A. Toporensky, Induced cosmological constant and other
features of asymmetric brane embedding. JCAP 0904 (2009) 023. [arXiv:0901.3074 [gr-qc]].
[1144] F.-W. Shu. Extended Hořava Gravity with Physical Ground-State Wavefunction. 2010, 1009.3677.
[1145] F. P. Silva and K. Koyama. Self-Accelerating Universe in Galileon Cosmology. Phys.Rev.,
D80:121301, 2009, 0909.4538.
[1146] J. D. Simon and M. Geha. The Kinematics of the Ultra-Faint Milky Way Satellites: Solving the
Missing Satellite Problem. Astrophys.J., 670:313–331, 2007, 0706.0516.
[1147] C. Skordis. TeVeS cosmology: Covariant formalism for the background evolution and linear
perturbation theory. Phys. Rev., D74:103513, 2006, astro-ph/0511591.
[1148] C. Skordis. Generalizing TeVeS Cosmology. Phys. Rev., D77:123502, 2008, 0801.1985.
[1149] C. Skordis. Consistent cosmological modifications to the Einstein equations. Phys. Rev. D,
79(12):123527, June 2009, 0806.1238.
[1150] C. Skordis. Eddington-Born-Infeld theory and the dark sector. Nucl.Phys.Proc.Suppl., 194:338–
343, 2009.
[1151] C. Skordis. Modifications of gravity. Nucl.Phys.Proc.Suppl., 194:332–337, 2009.
[1152] C. Skordis. The Tensor-Vector-Scalar theory and its cosmology. Class. Quant. Grav., 26:143001,
2009, 0903.3602.
[1153] C. Skordis, D. F. Mota, P. G. Ferreira, and C. Boehm. Large scale structure in Bekenstein’s
theory of relativistic MOND. Phys. Rev. Lett., 96:011301, 2006, astro-ph/0505519.
[1154] C. Skordis and T. Zlosnik. The Geometry Of Modified Newtonian Dynamics. 2011, 1101.6019.
[1155] A. Slosar, A. Melchiorri, and J. Silk. Did Boomerang hit MOND? Phys. Rev., D72:101301, 2005,
astro-ph/0508048.
[1156] R. Smith et al. Stable clustering, the halo model and nonlinear cosmological power spectra.
Mon.Not.Roy.Astron.Soc., 341:1311, 2003, astro-ph/0207664.
[1157] Y. Song, W. Hu, and I. Sawicki. Large scale structure of f(R) gravity. Phys. Rev. D, 75(4):044004,
Feb. 2007, arXiv:astro-ph/0610532.
[1158] Y. Song, H. Peiris, and W. Hu. Cosmological constraints on f(R) acceleration models.
Phys. Rev. D, 76(6):063517, Sept. 2007, 0706.2399.
[1159] Y. Song, G. Zhao, D. Bacon, K. Koyama, R. C. Nichol, and L. Pogosian. Complementarity of
Weak Lensing and Peculiar Velocity Measurements in Testing General Relativity. ArXiv e-prints,
Nov. 2010, 1011.2106.
[1160] Y.-S. Song. Large Scale Structure Formation of normal branch in DGP brane world model.
Phys.Rev., D77:124031, 2008, 0711.2513.
[1161] Y.-S. Song, I. Sawicki, and W. Hu. Large-scale tests of the DGP model. Phys. Rev., D75:064003,
2007, astro-ph/0606286.
[1162] T. P. Sotiriou. Constraining f(R) gravity in the Palatini formalism. Classical and Quantum
Gravity, 23:1253–1267, Feb. 2006, arXiv:gr-qc/0512017.
[1163] T. P. Sotiriou. The nearly Newtonian regime in non-linear theories of gravity. General Relativity
and Gravitation, 38:1407–1417, Sept. 2006, arXiv:gr-qc/0507027.
[1164] T. P. Sotiriou. Unification of inflation and cosmic acceleration in the Palatini formalism.
Phys. Rev. D, 73(6):063515, Mar. 2006, arXiv:gr-qc/0509029.
[1165] T. P. Sotiriou. Curvature scalar instability in f(R) gravity. Physics Letters B, 645:389–392, Feb.
2007, arXiv:gr-qc/0611107.
[1166] T. P. Sotiriou. Horava-Lifshitz gravity: a status report. 2010, 1010.3218.
[1167] T. P. Sotiriou and V. Faraoni. f(R) theories of gravity. Reviews of Modern Physics, 82:451–497,
306
 Jan. 2010, 0805.1726.
[1168] T. P. Sotiriou and S. Liberati. Metric-affine f(R) theories of gravity. Annals of Physics, 322:935–
966, Apr. 2007, arXiv:gr-qc/0604006.
[1169] T. P. Sotiriou, M. Visser, and S. Weinfurtner. Phenomenologically viable Lorentz-violating quan-
tum gravity. Phys.Rev.Lett., 102:251601, 2009, 0904.4464.
[1170] T. P. Sotiriou, M. Visser, and S. Weinfurtner. Quantum gravity without Lorentz invariance.
JHEP, 0910:033, 2009, 0905.2798.
[1171] M. E. Soussa and R. P. Woodard. Letter: The Force of Gravity from a Lagrangian Containing
Inverse Powers of the Ricci Scalar. General Relativity and Gravitation, 36:855–862, Apr. 2004,
arXiv:astro-ph/0308114.
[1172] D. Spergel et al. First year Wilkinson Microwave Anisotropy Probe (WMAP) observations: Deter-
mination of cosmological parameters. Astrophys.J.Suppl., 148:175–194, 2003, astro-ph/0302209.
[1173] D. N. Spergel et al. Wilkinson Microwave Anisotropy Probe (WMAP) three year results: Impli-
cations for cosmology. Astrophys. J. Suppl., 170:377, 2007, astro-ph/0603449.
[1174] D. N. Spergel and P. J. Steinhardt. Observational evidence for selfinteracting cold dark matter.
Phys.Rev.Lett., 84:3760–3763, 2000, astro-ph/9909386.
[1175] S. Speziale. Bi-metric theory of gravity from the non-chiral Plebanski action. Phys.Rev.,
D82:064003, 2010, 1003.4701.
[1176] I. Stairs, A. Faulkner, A. Lyne, M. Kramer, D. Lorimer, M. McLaughlin, R. Manchester,
G. Hobbs, F. Camilo, A. Possenti, M. Burgay, N. D’Amico, P. Freire, and P. Gregory. Discovery
of three wide-orbit binary pulsars: Implications for binary evolution and equivalence principles.
Astrophys. J., 632:1060–1068, 2005.
[1177] I. H. Stairs. Testing General Relativity with Pulsar Timing. Living Reviews in Relativity, 6:5,
Sept. 2003, arXiv:astro-ph/0307536.
[1178] A. A. Starobinskii. The Perturbation Spectrum Evolving from a Nonsingular Initially De-Sitter
Cosmology and the Microwave Background Anisotropy. Soviet Astronomy Letters, 9:302, June
1983.
[1179] A. A. Starobinsky. A new type of isotropic cosmological models without singularity. Physics
Letters B, 91(1):99 – 102, 1980.
[1180] A. A. Starobinsky. Disappearing cosmological constant in f(R) gravity. JETP Lett., 86:157–163,
2007, 0706.2041.
[1181] A. A. Starobinsky and J. Yokoyama. Density fluctuations in Brans-Dicke inflation. Feb. 1995,
arXiv:gr-qc/9502002.
[1182] J. H. Steffen and Gammev Collaboration. Constraints on chameleons and axions-like particles
from the GammeV experiment. In Identification of Dark Matter 2008, 2008, 0810.5070.
[1183] J. H. Steffen, A. Upadhye, A. Baumbaugh, A. S. Chou, P. O. Mazur, R. Tomlin, A. Weltman,
and W. Wester. Laboratory constraints on chameleon dark energy and power-law fields. ArXiv
e-prints, Oct. 2010, 1010.0988.
[1184] P. J. Steinhardt. Recent advances in extended inflationary cosmology. Classical and Quantum
Gravity, 10:33, Dec. 1993, arXiv:astro-ph/9212004.
[1185] P. J. Steinhardt and F. S. Accetta. Hyperextended inflation. Physical Review Letters, 64:2740–
2743, June 1990.
[1186] K. S. Stelle. Renormalization of higher-derivative quantum gravity. Phys. Rev. D, 16(4):953–969,
Aug 1977.
[1187] K. S. Stelle. Classical gravity with higher derivatives. General Relativity and Gravitation, 9:353–
371, Apr. 1978.
[1188] J. M. Stewart and M. Walker. Perturbations of spacetimes in general relativity. Proc. Roy. Soc.
Lond., A341:49–74, 1974.
[1189] H. Stoica, S. H. H. Tye, and I. Wasserman. Cosmology in the Randall-Sundrum brane world
scenario. Phys. Lett., B482:205–212, 2000, hep-th/0004126.
[1190] E. e. a. Sullivan. SNLS3: Constraints on Dark Energy Combining the Supernova Legacy Survey
Third Year Data with Other Probes. Astrophys. J., 737:102, 2011, 1104.1444.
[1191] M. Susperregi. How typical is General Relativity in Brans-Dicke chaotic inflation? Dec. 1997,
arXiv:gr-qc/9712031.
[1192] M. Susperregi and A. Mazumdar. Extended inflation with an exponential potential. Phys. Rev. D,
58(8):083512, Oct. 1998, arXiv:gr-qc/9804081.
[1193] M. P. K. Susperregi. Spectrum of density fluctuations in Brans-Dicke chaotic inflation.
Phys. Rev. D, 55:560–572, Jan. 1997, arXiv:astro-ph/9606018.
[1194] L. Susskind. The anthropic landscape of string theory. 2003, hep-th/0302219.
307
[1195] J. Tafel. A non-singular homogeneous universe with torsion. Phys. Lett., A45: 341, 1975.
[1196] T. Takahashi, J. Soda, Chiral Primordial Gravitational Waves from a Lifshitz Point Phys. Rev.
Lett. 102 (2009) 231301. [arXiv:0904.0554 [hep-th]].
[1197] C. Talmadge, J. Berthias, R. W. Hellings, and E. M. Standish. Model-independent constraints on
possible modifications of Newtonian gravity. Physical Review Letters, 61:1159–1162, Sept. 1988.
[1198] T. Tamaki. Post-Newtonian parameters in the tensor-vector-scalar theory. Phys. Rev.,
D77:124020, 2008, 0803.4309.
[1199] T. Tamaki and S. Tsujikawa. Revisiting chameleon gravity: Thin-shell and no-shell fields with
appropriate boundary conditions. Phys. Rev. D, 78(8):084028, Oct. 2008, 0808.2284.
[1200] N. Tanahashi and T. Tanaka. Black holes in braneworld models. 2011, 1105.2997.
[1201] T. Tanaka. Classical black hole evaporation in Randall-Sundrum infinite braneworld. Prog.
Theor. Phys. Suppl., 148:307–316, 2003, gr-qc/0203082.
[1202] F. Tangherlini. Schwarzschild field in n dimensions and the dimensionality of space problem. Il
Nuovo Cimento (1955-1965), 27(3):636–651, 1963.
[1203] M. Tegmark et al. Cosmological Constraints from the SDSS Luminous Red Galaxies. Phys.Rev.,
D74:123507, 2006, astro-ph/0608632.
[1204] H. Thirring. Über die Wirkung rotierender ferner Massen in der Einsteinschen Gravitationsthe-
orie. Physikalische Zeitschrift, 19:33, 1918.
[1205] S. Thomas. Baryons and dark matter from the late decay of a supersymmetric condensate.
Physics Letters B, 356(2-3):256 – 263, 1995.
[1206] I. Thongkool, M. Sami, and S. R. Choudhury. How delicate are the f(R) gravity models with a
disappearing cosmological constant? Phys. Rev. D, 80(12):127501, Dec. 2009, 0908.1693.
[1207] K. S. Thorne and C. M. Will. Theoretical Frameworks for Testing Relativistic Gravity. I. Foun-
dations. ApJ, 163:595, Feb. 1971.
[1208] A. J. Tolley, C. Burgess, D. Hoover, and Y. Aghababaie. Bulk singularities and the effective
cosmological constant for higher co-dimension branes. JHEP, 0603:091, 2006, hep-th/0512218.
[1209] T. Torii, H. Yajima, and K. Maeda. Dilatonic black holes with a Gauss-Bonnet term.
Phys. Rev. D, 55:739–753, Jan. 1997, arXiv:gr-qc/9606034.
[1210] D. F. Torres. Nucleosynthesis bounds on scalar-tensor gravity: power-law couplings. Physics
Letters B, 359:249–253, Feb. 1995.
[1211] D. F. Torres. Boson stars in general scalar-tensor gravitation: Equilibrium configurations. Phys.
Rev. D, 56(6):3478–3484, Sep 1997.
[1212] D. F. Torres. Slow roll inflation in nonminimally coupled theories: hyperextended gravity ap-
proach. Physics Letters A, 225:13–17, Feb. 1997, arXiv:gr-qc/9610021.
[1213] D. F. Torres, A. R. Liddle, and F. E. Schunck. Gravitational memory of boson stars. Phys. Rev.
D, 57(8):4821–4825, Apr 1998.
[1214] A. Trautman. On the Einstein-Cartan equations III Bull. Acad. Pol. Sci., Ser. Sci. Math. Astron.
Phys., 20: 895, 1972.
[1215] A. Trautman. Einstein-Cartan theory. gr-qc/0606062, 2006.
[1216] A. A. Tseytlin and C. Vafa. Elements of string cosmology. Nucl. Phys., B372:443–466, 1992,
hep-th/9109048.
[1217] S. Tsujikawa. Density perturbations in the ekpyrotic Universe and string-inspired generalizations.
Physics Letters B, 526:179–185, Feb. 2002, arXiv:gr-qc/0110124.
[1218] S. Tsujikawa. Cosmologies from higher-order string corrections. Annalen der Physik, 518:302–315,
Apr. 2006, arXiv:hep-th/0606040.
[1219] S. Tsujikawa. Matter density perturbations and effective gravitational constant in modified grav-
ity models of dark energy. Phys. Rev. D, 76(2):023514, July 2007, 0705.1032.
[1220] S. Tsujikawa. Observational signatures of f(R) dark energy models that satisfy cosmological and
local gravity constraints. Phys. Rev. D, 77(2):023507, Jan. 2008, 0709.1391.
[1221] S. Tsujikawa, R. Gannouji, B. Moraes, and D. Polarski. Dispersion of growth of matter pertur-
bations in f(R) gravity. Phys. Rev. D, 80(8):084044, Oct. 2009, 0908.2669.
[1222] S. Tsujikawa, K. Maeda, and T. Torii. Preheating with nonminimally coupled scalar fields
in higher-curvature inflation models. Phys. Rev. D, 60(12):123505, Dec. 1999, arXiv:hep-
ph/9906501.
[1223] S. Tsujikawa and M. Sami. String-inspired cosmology: a late time transition from a scaling matter
era to a dark energy universe caused by a Gauss Bonnet coupling. J. Cosmology Astropart. Phys.,
1:6, Jan. 2007, arXiv:hep-th/0608178.
[1224] S. Tsujikawa, M. Sami, and R. Maartens. Observational constraints on braneworld inflation: The
Effect of a Gauss-Bonnet term. Phys.Rev., D70:063525, 2004, astro-ph/0406078.
308
[1225] S. Tsujikawa, T. Tamaki, and R. Tavakol. Chameleon scalar fields in relativistic gravitational
backgrounds. J. Cosmology Astropart. Phys., 5:20, May 2009, 0901.3226.
[1226] S. Tsujikawa, K. Uddin, and R. Tavakol. Density perturbations in f(R) gravity theories in metric
and Palatini formalisms. Phys. Rev. D, 77(4):043007, Feb. 2008, 0712.0082.
[1227] P. Y. Tsyba, I. I. Kulnazarov, K. K. Yerzhanov, and R. Myrzakulov. Pure kinetic k-essence as
the cosmic speed-up and f(T) - gravity. ArXiv e-prints, Aug. 2010, 1008.0779.
[1228] M. S. Turner, G. Steigman, and L. M. Krauss. Flatness of the universe - Reconciling theoretical
prejudices with observational data. Physical Review Letters, 52:2090–2093, June 1984.
[1229] M. S. Turner and F. Wilczek. Relic gravitational waves and extended inflation. Physical Review
Letters, 65:3080–3083, Dec. 1990.
[1230] S. G. Turyshev, M. Shao, and K. Nordtvedt. The laser astrometric test of relativity mission.
Classical and Quantum Gravity, 21:2773–2799, June 2004, arXiv:gr-qc/0311020.
[1231] S. G. Turyshev, J. G. Williams, M. Shao, J. D. Anderson, K. L. Nordtvedt, Jr, and T. W.
Murphy, Jr. Laser Ranging to the Moon, Mars and Beyond. ArXiv General Relativity and
Quantum Cosmology e-prints, Nov. 2004, arXiv:gr-qc/0411082.
[1232] K. Uddin, J. E. Lidsey, and R. Tavakol. Cosmological perturbations in Palatini modified gravity.
Class. Quant. Grav., 24:3951–3962, 2007, 0705.0232.
[1233] K. Uddin, J. E. Lidsey, and R. Tavakol. Cosmological scaling solutions in generalised Gauss-
Bonnet gravity theories. General Relativity and Gravitation, 41:2725–2736, Dec. 2009, 0903.0270.
[1234] K. -i. Umezu, K. Ichiki, T. Kajino, G. J. Mathews, R. Nakamura, M. Yahiro, Observational
constraints on accelerating brane cosmology with exchange between the bulk and brane. Phys.
Rev. D73 (2006) 063527. [astro-ph/0507227].
[1235] A. Unzicker and T. Case. Translation of Einstein’s Attempt of a Unified Field Theory with
Teleparallelism. ArXiv Physics e-prints, Mar. 2005, arXiv:physics/0503046.
[1236] A. Upadhye, S. S. Gubser, and J. Khoury. Unveiling chameleon fields in tests of the gravitational
inverse-square law. Phys. Rev. D, 74(10):104024, Nov. 2006, arXiv:hep-ph/0608186.
[1237] A. Upadhye and W. Hu. Existence of relativistic stars in f(R) gravity. Phys. Rev. D, 80(6):064002,
Sept. 2009, 0905.4055.
[1238] A. Upadhye, J. H. Steffen, and A. Weltman. Constraining chameleon field theories using the
GammeV afterglow experiments. Phys. Rev. D, 81(1):015013, Jan. 2010, 0911.3906.
[1239] R. Utiyama and B. S. Dewitt. Renormalization of a Classical Gravitational Field Interacting
with Quantized Matter Fields. Journal of Mathematical Physics, 3:608–618, July 1962.
[1240] J. Uzan. The fundamental constants and their variation: observational and theoretical status.
Reviews of Modern Physics, 75:403–455, Apr. 2003, arXiv:hep-ph/0205340.
[1241] A. I. Vainshtein. To the problem of nonvanishing gravitation mass. Phys. Lett., B39:393–394,
1972.
[1242] K. Van Acoleyen and J. Van Doorsselaere. Galileons from Lovelock actions. 2011, 1102.0487.
[1243] H. van Dam and M. J. G. Veltman. Massive and massless Yang-Mills and gravitational fields.
Nucl. Phys., B22:397–411, 1970.
[1244] C. van de Bruck, M. Dorca, R. H. Brandenberger, and A. Lukas. Cosmological perturbations in
brane-world theories: Formalism. Phys. Rev., D62:123515, 2000, hep-th/0005032.
[1245] C. van de Bruck, M. Dorca, C. J. A. P. Martins, and M. Parry. Cosmological consequences of
the brane/bulk interaction. Phys. Lett., B495:183–192, 2000, hep-th/0009056.
[1246] H. van Elst, J. E. Lidsey, and R. Tavakol. Quantum cosmology and higher-order Lagrangian
theories. Classical and Quantum Gravity, 11:2483–2497, Oct. 1994, arXiv:gr-qc/9404044.
[1247] M. A. Vasiliev. Consistent equation for interacting gauge fields of all spins in (3+1)-dimensions.
Phys.Lett., B243:378–382, 1990.
[1248] L. Verde, A. F. Heavens, W. J. Percival, S. Matarrese, C. M. Baugh, et al. The 2dF Galaxy
Redshift Survey: The Bias of galaxies and the density of the Universe. Mon.Not.Roy.Astron.Soc.,
335:432, 2002, astro-ph/0112161.
[1249] A. Vilenkin. Classical and quantum cosmology of the Starobinsky inflationary model.
Phys. Rev. D, 32:2511–2521, Nov. 1985.
[1250] J. Vinet and J. M. Cline. Can codimension-two branes solve the cosmological constant problem?
Phys. Rev., D70:083514, 2004, hep-th/0406141.
[1251] D. N. Vollick. 1/R curvature corrections as the source of the cosmological acceleration.
Phys. Rev. D, 68(6):063510, Sept. 2003, arXiv:astro-ph/0306630.
[1252] H. von Borzeszkowski. On singularities of general relativity and gravitational equations of fourth
order. Annalen der Physik, 493:239–248, 1981.
[1253] R. V. Wagoner. Scalar tensor theory and gravitational waves. Phys. Rev., D1:3209–3216, 1970.
309
[1254] R. M. Wald. Final states of gravitational collapse. Phys. Rev. Lett., 26(26):1653–1655, Jun 1971.
[1255] R. M. Wald. Asymptotic behavior of homogeneous cosmological models in the presence of a
positive cosmological constant. Phys. Rev. D, 28(8):2118–2120, Oct 1983.
[1256] A. Wang. Vector and tensor perturbations in Horava-Lifshitz cosmology. Phys.Rev., D82:124063,
2010, 1008.3637.
[1257] A. Wang and R. Maartens. Linear perturbations of cosmological models in the Horava- Lifshitz
theory of gravity without detailed balance. Phys. Rev., D81:024009, 2010, 0907.1748.
[1258] A. Wang, D. Wands, and R. Maartens. Scalar field perturbations in Horava-Lifshitz cosmology.
JCAP, 1003:013, 2010, 0909.5167.
[1259] A. Wang and Y. Wu. Cosmology in nonrelativistic general covariant theory of gravity. 2010,
1009.2089.
[1260] Y. Wang. Constraints on generalized extended inflationary models. Phys. Rev. D, 44:991–998,
Aug. 1991.
[1261] Y. Wang and P. Mukherjee. Robust dark energy constraints from supernovae, galaxy clustering,
and three-year wilkinson microwave anisotropy probe observations. Astrophys.J., 650:1–6, 2006,
astro-ph/0604051.
[1262] J. Weenink and T. Prokopec. Gauge invariant cosmological perturbations for the nonminimally
coupled inflaton field. Phys. Rev. D, 82(12):123510, Dec. 2010, 1007.2133.
[1263] H. Wei. Growth Index of DGP Model and Current Growth Rate Data. Phys.Lett., B664:1–6,
2008, 0802.4122.
[1264] E. J. Weinberg. Some problems with extended inflation. Phys. Rev. D, 40:3950–3959, Dec. 1989.
[1265] S. Weinberg. The Cosmological Constant Problem. Rev.Mod.Phys., 61:1–23, 1989. Morris Loeb
Lectures in Physics, Harvard University, May 2, 3, 5, and 10, 1988.
[1266] S. Weinfurtner, T. P. Sotiriou, and M. Visser. Projectable Horava-Lifshitz gravity in a nutshell.
J.Phys.Conf.Ser., 222:012054, 2010, 1002.0308.
[1267] J. Weisberg, D. J. Nice and J. H. Taylor. Timing Measurements of the Relativistic Binary Pulsar
PSR B1913+16. Astrophys. J., 722:1030–1034, 2010.
[1268] J. Weisberg and J. H. Taylor. General relativistic geodetic spin precession in binary pulsar
b1913+16: Mapping the emission beam in two dimensions. Astrophys. J., 576:942–949, 2002.
[1269] H. Weyl. Gravitation und Elektrizität. Sitzungsber. Preuss. Akad. d. Wiss. Teil, 1(465), 1918.
[1270] B. Whitt. Fourth-order gravity as general relativity plus matter. Physics Letters B, 145:176–178,
Sept. 1984.
[1271] B. Whitt. Stability of Schwarzschild black holes in fourth-order gravity. Phys. Rev. D, 32:379–388,
July 1985.
[1272] C. Will. The confrontation between general relativity and experiment. Living Rev. Rel., 9:3,
2006.
[1273] C. M. Will. in General relativity. An Einstein centenary survey. 1979.
[1274] C. M. Will. Theory and Experiment in Gravitational Physics. Mar. 1993.
[1275] C. M. Will and D. M. Eardley. Dipole gravitational radiation in Rosen’s theory of gravity -
Observable effects in the binary system PSR 1913+16. ApJ, 212:L91–L94, Mar. 1977.
[1276] C. M. Will and K. Nordtvedt, Jr. Conservation Laws and Preferred Frames in Relativistic Gravity.
I. Preferred-Frame Theories and an Extended PPN Formalism. ApJ, 177:757, Nov. 1972.
[1277] J. G. Williams, S. G. Turyshev, and D. H. Boggs. Progress in Lunar Laser Ranging Tests of
Relativistic Gravity. Physical Review Letters, 93(26):261101, Dec. 2004, arXiv:gr-qc/0411113.
[1278] T. Wiseman. Relativistic stars in Randall-Sundrum gravity. Phys. Rev., D65:124007, 2002,
hep-th/0111057.
[1279] B. Withers. Einstein-aether as a quantum effective field theory. Classical and Quantum Gravity,
26(22):225009, Nov. 2009, 0905.2446.
[1280] W. Wood-Vasey et al. Observational Constraints on the Nature of the Dark Energy: First
Cosmological Results from the ESSENCE Supernova Survey. Astrophys.J., 666:694–715, 2007,
astro-ph/0701041.
[1281] R. Woodard. Avoiding Dark Energy with 1/R Modifications of Gravity. In L. Papantonopoulos,
editor, The Invisible Universe: Dark Matter and Dark Energy, volume 720 of Lecture Notes in
Physics, Berlin Springer Verlag, page 403, 2007, arXiv:astro-ph/0601672.
[1282] P. Worden, R. Torii, J. C. Mester, and C. W. F. Everitt. The STEP Payload and Experiment.
Advances in Space Research, 25:1205–1208, 2000.
[1283] A. Wu. Constraints on the nucleation rate in extended inflation. Phys. Rev. D, 45:2653–2664,
Apr. 1992.
[1284] F. Wu and X. Chen. Cosmic microwave background with Brans-Dicke gravity II: constraints with
310
 the WMAP and SDSS data. 2009, 0903.0385.
[1285] F. Wu, L.-e. Qiang, X. Wang, and X. Chen. Cosmic microwave background with Brans-Dicke
Gravity: I. Covariant Formulation. 2009, 0903.0384.
[1286] P. Wu and H. Yu. f(T) models with phantom divide line crossing. ArXiv e-prints, Aug. 2010,
1008.3669.
[1287] P. Wu and H. Yu. Observational constraints on f(T) theory. Physics Letters B, 693:415–420,
Oct. 2010, 1006.0674.
[1288] P. Wu and H. Yu. The dynamical behavior of f(T) theory. Physics Letters B, 692:176–179, Aug.
2010, 1007.2348.
[1289] M. Wyman. Static spherically symmetric scalar fields in general relativity. Phys. Rev., D24:839–
841, 1981.
[1290] M. Wyman and J. Khoury. Enhanced Peculiar Velocities in Brane-Induced Gravity. Phys.Rev.,
D82:044032, 2010, 1004.2046.
[1291] D. Xu et al. Is Gravitational Lensing by Intercluster Filaments Always Negligible? Astrophys.
J., 682:711–720, 2008, 0710.4935.
[1292] J. Yang, D. N. Schramm, G. Steigman, and R. T. Rood. Constraints on cosmology and neutrino
physics from big bang nucleosynthesis. ApJ, 227:697–704, Jan. 1979.
[1293] R. Yang. Conformal transformation in f(T) theories. ArXiv e-prints, Oct. 2010, 1010.1376.
[1294] A. Yoho, F. Ferrer, and G. D. Starkman. Degree-scale anomalies in the CMB: localizing the first
peak dip to a small patch of the north ecliptic sky. 2010, 1005.5389.
[1295] M. Yoshimura. Extended inflation, Brans-Dicke dilaton, and baryon asymmetry. Physical Review
Letters, 66:1559–1562, Mar. 1991.
[1296] V. I. Zakharov. Linearized gravitation theory and the graviton mass. JETP Lett., 12:312, 1970.
[1297] J. Zanelli. Lecture notes on Chern-Simons (super-)gravities. Second edition (February 2008).
2005, hep-th/0502193.
[1298] Y. Zeldovich and A. A. Starobinsky. Particle production and vacuum polarization in an
anisotropic gravitational field. Sov.Phys.JETP, 34:1159–1166, 1972.
[1299] P. Zhang. Testing gravity against the early time integrated Sachs-Wolfe effect. Phys. Rev. D,
73(12):123504, June 2006, arXiv:astro-ph/0511218.
[1300] P. Zhang. Behavior of f(R) gravity in the solar system, galaxies, and clusters. Phys. Rev. D,
76(2):024007, July 2007, arXiv:astro-ph/0701662.
[1301] P. Zhang, M. Liguori, R. Bean, and S. Dodelson. Probing Gravity at Cosmological Scales by Mea-
surements which Test the Relationship between Gravitational Lensing and Matter Overdensity.
Physical Review Letters, 99(14):141302, Oct. 2007, 0704.1932.
[1302] G. Zhao, T. Giannantonio, L. Pogosian, A. Silvestri, D. J. Bacon, K. Koyama, R. C. Nichol,
and Y. Song. Probing modifications of general relativity using current cosmological observations.
Phys. Rev. D, 81(10):103510, May 2010, 1003.0001.
[1303] G. Zhao, B. Li, and K. Koyama. N-body Simulations for f(R) Gravity using a Self-adaptive
Particle-Mesh Code. ArXiv e-prints, Nov. 2010, 1011.1257.
[1304] G. Zhao, L. Pogosian, A. Silvestri, and J. Zylberberg. Cosmological Tests of General Rela-
tivity with Future Tomographic Surveys. Physical Review Letters, 103(24):241301, Dec. 2009,
0905.1326.
[1305] H. Zhao. Constraining TeVeS gravity as effective dark matter and dark energy. Int. J. Mod.
Phys., D16:2055–2063, 2008, astro-ph/0610056.
[1306] H.-S. Zhao. An introduction to gravitational lensing in TeVeS gravity. 2006, astro-ph/0611777.
[1307] H.-S. Zhao, D. J. Bacon, A. N. Taylor, and K. Horne. Testing Bekenstein’s Relativistic MOND
gravity with Gravitational Lensing. Mon. Not. Roy. Astron. Soc., 368:171–186, 2006, astro-
ph/0509590.
[1308] R. Zheng and Q. Huang. Growth factor in f(T) gravity. ArXiv e-prints, Oct. 2010, 1010.3512.
[1309] S. Zhou, E. J. Copeland, and P. M. Saffin. Cosmological constraints on f(G) dark energy models.
J. Cosmology Astropart. Phys., 7:9, July 2009, 0903.4610.
[1310] S.-Y. Zhou. Goldstone’s Theorem and Hamiltonian of Multi-galileon Modified Gravity. 2010,
1011.0863.
[1311] T. G. Zlosnik, P. G. Ferreira, and G. D. Starkman. Growth of structure in theories with a
dynamical preferred frame. Phys. Rev. D, 77(8):084010, Apr. 2008, 0711.0520.
[1312] M. A. Zumberge, J. A. Hildebrand, J. M. Stevenson, R. L. Parker, F. N. Spiess, A. D. Chave,
and M. E. Ander. Submarine measurement of the Newtonian gravitational constant. Physical
Review Letters, 67:3051–3054, Nov. 1991.
[1313] B. Zumino. Gravity theories in more than four dimensions 1. Physics Reports, 137(1):109–114,
311
 1986.
[1314] J. Zuntz, T. G. Zlosnik, F. Bourliot, P. G. Ferreira, and G. D. Starkman. Vector field models of
modified gravity and the dark sector. Phys. Rev. D, 81(10):104015, May 2010, 1002.0849.
[1315] J. A. Zuntz, P. G. Ferreira, and T. G. Zlosnik. Constraining Lorentz Violation with Cosmology.
Physical Review Letters, 101(26):261102, Dec. 2008, 0808.1824.
[1316] B. Zwiebach. Curvature Squared Terms and String Theories. Phys. Lett., B156:315, 1985.

312