f (R) theories of gravity

Thomas P. Sotiriou
Center for Fundamental Physics, University of Maryland, College Park, MD 20742-4111, USA

Valerio Faraoni

arXiv:0805.1726v4 [gr-qc] 4 Jun 2010

Physics Department, Bishops University, 2600 College St., Sherbrooke, Qu`ebec, Canada J1M 1Z7
Modified gravity theories have received increased attention lately due to combined motivation
coming from high-energy physics, cosmology and astrophysics. Among numerous alternatives
to Einsteins theory of gravity, theories which include higher order curvature invariants, and
specifically the particular class of f (R) theories, have a long history. In the last five years there has
been a new stimulus for their study, leading to a number of interesting results. We review here f (R)
theories of gravity in an attempt to comprehensively present their most important aspects and
cover the largest possible portion of the relevant literature. All known formalisms are presented
metric, Palatini and metric-affine and the following topics are discussed: motivation; actions,
field equations and theoretical aspects; equivalence with other theories; cosmological aspects and
constraints; viability criteria; astrophysical applications.

Contents
I. Introduction
A. Historical
B. Contemporary Motivation
C. f (R) theories as toy theories
II. Actions and field equations
A. Metric formalism
B. Palatini formalism
C. Metric-affine formalism
1. Preliminaries
2. Field Equations

4. Ghost fields
C. The Cauchy problem
1
1
2
3
4
5
6
8
9
10

III. Equivalence with BransDicke theory and

classification of theories
A. Metric formalism
B. Palatini formalism
C. Classification
D. Why f (R) gravity then?

11
12
13
13
14

IV. Cosmological evolution and constraints

A. Background evolution
1. Metric f (R) gravity
2. Palatini f (R) gravity
B. Cosmological eras
C. Dynamics of cosmological perturbations

15
15
16
18
18
19

V. Other standard viability criteria

20
A. Weak-field limit
20
1. The scalar degree of freedom
20
2. Weak-field limit in the metric formalism
22
3. Weak-field limit in the Palatini formalism
25
B. Stability issues
28
1. Ricci stability in the metric formalism
28
2. Gauge-invariant stability of de Sitter space in the
metric formalism
29
3. Ricci stability in the Palatini formalism
31

Present address: Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences, University of
Cambridge, Wilberforce Road, Cambridge, CB3 0WA, UK.
T.Sotiriou@damtp.cam.ac.uk
vfaraoni@ubishops.ca

VI. Confrontation with particle physics and

astrophysics
A. Metric f (R) gravity as dark matter
B. Palatini f (R) gravity and the conflict with the
Standard Model
C. Exact solutions and relevant constraints
1. Vacuum and non-vacuum exact solutions
2. Surface singularities and the incompleteness of
Palatini f (R) gravity
D. Gravitational waves in f (R) gravity
VII. Summary and Conclusions
A. Summary
B. Extensions and new perspectives on f (R) gravity
C. Concluding remarks

31
31

34
34
35
36
36
37
40
40
40
41
42

Acknowledgments

43

References

43

I. INTRODUCTION
A. Historical

As we are approaching the closing of a century after the

introduction of General Relativity (GR) in 1915, questions related to its limitations are becoming more and
more pertinent. However, before coming to the contemporary reasons for challenging a theory as successful as
Einsteins theory, it is worth mentioning that it took only
four years from its introduction for people to start questioning its unique status among gravitation theories. Indeed, it was just 1919 when Weyl, and 1923 when Eddington (the very man that three years earlier had provided
the first experimental verification of GR by measuring
light bending during a solar eclipse) started considering
modifications of the theory by including higher order invariants in its action (Eddington, 1923; Weyl, 1919).

2
These early attempts were triggered mainly by scientific curiosity and a will to question and, therefore, understand the then newly proposed theory. It is quite
straightforward to realize that complicating the action
and, consequently, the field equations with no apparent
theoretical or experimental motivation is not very appealing. However, the motivation was soon to come.
Beginning in the 1960s, there appeared indications
that complicating the gravitational action might indeed
have its merits. GR is not renormalizable and, therefore,
can not be conventionally quantized. In 1962, Utiyama
and De Witt showed that renormalization at one-loop demands that the EinsteinHilbert action be supplemented
by higher order curvature terms (Utiyama and DeWitt,
1962). Later on, Stelle showed that higher order actions are indeed renormalizable (but not unitary) (Stelle,
1977). More recent results show that when quantum
corrections or string theory are taken into account,
the effective low energy gravitational action admits
higher order curvature invariants (Birrell and Davies,
1982; Buchbinder et al., 1992; Vilkovisky, 1992).
Such considerations stimulated the interest of the
scientific community in higher-order theories of gravity,
i.e., modifications of the EinsteinHilbert action in order
to include higher-order curvature invariants with respect
to the Ricci scalar [see (Schmidt, 2007) for a historical
review and a list of references to early work]. However,
the relevance of such terms in the action was considered
to be restricted to very strong gravity regimes and
they were expected to be strongly suppressed by small
couplings, as one would expect when simple effective
field theory considerations are taken into account.
Therefore, corrections to GR were considered to be
important only at scales close to the Planck scale and,
consequently, in the early universe or near black hole singularities and indeed there are relevant studies, such
as the well-known curvature-driven inflation scenario
(Starobinsky, 1980) and attempts to avoid cosmological and black hole singularities (Brandenberger,
1992,
1993,
1995;
Brandenberger et al.,
1993;
Mukhanov and Brandenberger, 1992; Shahid-Saless,
1990; Trodden et al., 1993).
However, it was not
expected that such corrections could affect the gravitational phenomenology at low energies, and consequently
large scales such as, for instance, the late universe.

matter refers to an unkown form of matter, which has

the clustering properties of ordinary matter but has not
yet been detected in the laboratory. The term dark energy is reserved for an unknown form of energy which
not only has not been detected directly, but also does
not cluster as ordinary matter does. More rigorously,
one could use the various energy conditions (Wald, 1984)
to distinguish dark matter and dark energy: Ordinary
matter and dark matter satisfy the Strong Energy Condition, whereas Dark Energy does not. Additionally, dark
energy seems to resemble in high detail a cosmological
constant. Due to its dominance over matter (ordinary
and dark) at present times, the expansion of the universe
seems to be an accelerated one, contrary to past expectations.1
Note that this late time speed-up comes to be added to
an early time accelerated epoch as predicted by the inflationary paradigm (Guth, 1981; Kolb and Turner, 1992;
Linde, 1990). The inflationary epoch is needed to address the so-called horizon, flatness and monopole problems (Kolb and Turner, 1992; Linde, 1990; Misner, 1968;
Weinberg, 1972), as well as to provide the mechanism
that generates primordial inhomogeneities acting as seeds
for the formation of large scale structures (Mukhanov,
2003). Recall also that, in between these two periods
of acceleration, there should be a period of decelerated
expansion, so that the more conventional cosmological
eras of radiation domination and matter domination can
take place. Indeed, there are stringent observational
bounds on the abundances of light elements, such as deuterium, helium and lithium, which require that Big Bang
Nucleosynthesis (BBN), the production of nuclei other
than hydrogen, takes place during radiation domination
(Burles et al., 2001; Carroll and Kaplinghat, 2002). On
the other hand, a matter-dominated era is required for
structure formation to take place.
Puzzling observations do not stop here. Dark matter does not only make its appearance in cosmological
data but also in astrophysical observations. The missing
mass question had already been posed in 1933 for galaxy
clusters (Zwicky, 1933) and in 1959 for individual galaxies (Kahn and Woltjer, 1959) and a satisfactory final answer has been pending ever since (Bosma, 1978; Ellis,
2002; Moore, 2001; Persic et al., 1996; Rubin and Ford,
1970; Rubin et al., 1980) .
One, therefore, has to admit that our current picture

B. Contemporary Motivation

More recently, new evidence coming from astrophysics

and cosmology has revealed a quite unexpected picture of
the universe. Our latest datasets coming from different
sources, such as the Cosmic Microwave Background Radiation (CMBR) and supernovae surveys, seem to indicate
that the energy budget of the universe is the following:
4% ordinary baryonic matter, 20% dark matter and 76%
dark energy (Astier et al., 2006; Eisenstein et al., 2005;
Riess et al., 2004; Spergel et al., 2007). The term dark

Recall that, from GR in the absence of the cosmological constant and under the standard cosmological assumptions (spatial
homogeneity and isotropy etc.), one obtains the second Friedmann equation
a

4 G
=
( + 3P ) ,
(1)
a
3
where a is the scale factor, G is the gravitational constant and
and P are the energy density and the pressure of the cosmological
fluid, respectively. Therefore, if the Strong Energy Condition
+ 3P 0 is satisfied, there can be no acceleration (gravity is
attractive).

3
of the evolution and the matter/energy content of the
universe is at least surprising and definitely calls for an
explanation. The simplest model which adequately fits
the data creating this picture is the so called concordance
model or CDM (-Cold Dark Matter), supplemented by
some inflationary scenario, usually based on some scalar
field called inflaton. Besides not explaining the origin
of the inflaton or the nature of dark matter by itself, the
CDM model is burdened with the well known cosmological constant problems (Carroll, 2001a; Weinberg, 1989):
the magnitude problem, according to which the observed
value of the cosmological constant is extravagantly small
to be attributed to the vacuum energy of matter fields,
and the coincidence problem, which can be summed up
in the question: since there is just an extremely short
period of time in the evolution of the universe in which
the energy density of the cosmological constant is comparable with that of matter, why is this happening today
that we are present to observe it?
These problems make the CDM model more of an
empirical fit to the data whose theoretical motivation
can be regarded as quite poor. Consequently, there
have been several attempts to either directly motivate
the presence of a cosmological constant or to propose
dynamical alternatives to dark energy. Unfortunately,
none of these attempts are problem-free. For instance,
the so-called anthropic reasoning for the magnitude of
(Barrow and Tipler, 1986; Carter, 1974), even when
placed into the firmer grounds through the idea of the
anthropic or string landscape (Susskind, 2003), still
makes many physicists feel uncomfortable due to its
probabilistic nature. On the other hand, simple scenarios for dynamical dark energy, such as quintessence
(Bahcall et al., 1999; Caldwell et al., 1998; Carroll, 1998;
Ostriker and Steinhardt, 1995; Peebles and Ratra, 1988;
Ratra and Peebles, 1988; Wang et al., 2000; Wetterich,
1988) do not seem to be as well motivated theoretically
as one would desire.2
Another perspective towards resolving the issues described above, which might appear as more radical to
some, is the following: gravity is by far the dominant interaction at cosmological scales and, therefore, it is the
force governing the evolution of the universe. Could it be
that our description of the gravitational interaction at the
relevant scales is not sufficiently adequate and stands at
the root of all or some of these problems? Should we consider modifying our theory of gravitation and if so, would
this help in avoiding dark components and answering the
cosmological and astrophysical riddles?
It is rather pointless to argue whether such a perspec-

We are referring here to the fact that, not only the mass of the
scalar turns out to be many orders of magnitude smaller than any
of the masses of the scalar fields usually encountered in particle
physics, but also to the inability to motivate the absence of any
coupling of the scalar field to matter (there is no mechanism or
symmetry preventing this) (Carroll, 2001b).

tive would be better or worse than any of the other solutions already proposed. It is definitely a different way
to address the same problems and, as long as these problems do not find a plausible, well accepted and simple,
solution, it is worth pursuing all alternatives. Additionally, questioning the gravitational theory itself definitely
has its merits: it helps us to obtain a deeper understanding of the relevant issues and of the gravitational interaction, it has high chances to lead to new physics and it has
worked in the past. Recall that the precession of Mercurys orbit was at first attributed to some unobserved
(dark) planet orbiting inside Mercurys orbit, but it
actually took the passage from Newtonian gravity to GR
to be explained.

C. f (R) theories as toy theories

Even if one decides that modifying gravity is the way to

go, this is not an easy task. To begin with, there are numerous ways to deviate from GR. Setting aside the early
attempts to generalize Einsteins theory, most of which
have been shown to be non-viable (Will, 1981), and the
most well known alternative to GR, scalar-tensor theory
(Bergmann, 1968; Brans and Dicke, 1961; Dicke, 1962;
Faraoni, 2004a; Nordtvedt, 1970; Wagoner, 1970), there
are still numerous proposals for modified gravity in contemporary literature. Typical examples are DGP (DvaliGabadadze-Porrati) gravity (Dvali et al., 2000), braneworld gravity (Maartens, 2004), TeVeS (Tensor-VectorScalar) (Bekenstein, 2004) and Einstein-Aether theory
(Jacobson and Mattingly, 2001). The subject of this review is a different class of theories, f (R) theories of gravity. These theories come about by a straightforward generalization of the Lagrangian in the EinsteinHilbert action,
Z

1
SEH =
d4 x g R,
(2)
2
where 8G, G is the gravitational constant, g is
the determinant of the metric and R is the Ricci scalar
(c = ~ = 1), to become a general function of R, i.e.,
Z

1
(3)
S=
d4 x g f (R).
2
Before going further into the discussion of the details
and the history of such actions this will happen in
the forthcoming section some remarks are in order.
We have already mentioned the motivation coming from
high-energy physics for adding higher order invariants
to the gravitational action, as well as a general motivation coming from cosmology and astrophysics for seeking
generalizations of GR. There are, however, still two questions that might be troubling the reader. The first one
is: Why specifically f (R) actions and not more general
ones, which include other higher order invariants, such
as R R ?

4
The answer to this question is twofold. First of all,
there is simplicity: f (R) actions are sufficiently general to encapsulate some of the basic characteristics of
higher-order gravity, but at the same time they are simple enough to be easy to handle. For instance, viewing f
as a series expansion, i.e.,
f (R) = . . . +

1
R2 R3
2
+
+
+ . . . , (4)

2
+
R
+
R2
R
2
3

where the i and i coefficients have the appropriate

dimensions, we see that the action includes a number
of phenomenologically interesting terms. In brief, f (R)
theories make excellent candidates for toy-theoriestools
from which one gains some insight in such gravity modifications. Second, there are serious reasons to believe
that f (R) theories are unique among higher-order gravity theories, in the sense that they seem to be the only
ones which can avoid the long known and fatal Ostrogradski instability (Woodard, 2007).
The second question calling for an answer is related to
a possible loophole that one may have already spotted in
the motivation presented: How can high-energy modifications of the gravitational action have anything to do
with late-time cosmological phenomenology? Wouldnt
effective field theory considerations require that the coefficients in eq. (4) be such, as to make any corrections to
the standard EinsteinHilbert term important only near
the Planck scale?
Conservatively thinking, the answer would be positive. However, one also has to stress two other serious
factors: first, there is a large ambiguity on how gravity really works at small scales or high energies. Indeed
there are certain results already in the literature claiming
that terms responsible for late time gravitational phenomenology might be predicted by some more fundamental theory, such as string theory [see, for instance,
(Nojiri and Odintsov, 2003b)]. On the other hand, one
should not forget that the observationally measured value
of the cosmological constant corresponds to some energy
scale. Effective field theory or any other high-energy theory consideration has thus far failed to predict or explain
it. Yet, it stands as an experimental fact and putting
the number in the right context can be crucial in explaining its value. Therefore, in any phenomenological
approach, its seems inevitable that some parameter will
appear to be unnaturally small at first (the mass of a
scalar, a coefficient of some expansions, etc. according to
the approach). The real question is whether this initial
unnaturalness still has room to be explained.
In other words, in all sincerity, the motivation for infrared modifications of gravity in general and f (R) gravity in particular is, to some extent, hand-waving. However, the importance of the issues leading to this motivation and our inability to find other, more straightforward
and maybe better motivated, successful ways to address
them combined with the significant room for speculation
which our quantum gravity candidates leave, have triggered an increase of interest in modified gravity that is

probably reasonable.
To conclude, when all of the above is taken into account, f (R) gravity should neither be over- nor underestimated. It is an interesting and relatively simple alternative to GR, from the study of which some useful conclusions have been derived already. However, it is still
a toy-theory, as already mentioned; an easy-to-handle
deviation from Einsteins theory mostly to be used in order to understand the principles and limitations of modified gravity. Similar considerations apply to modifying
gravity in general: we are probably far from concluding
whether it is the answer to our problems at the moment.
However, in some sense, such an approach is bound to be
fruitful since, even if it only leads to the conclusion that
GR is the only correct theory of gravitation, it will still
have helped us to both understand GR better and secure
our faith in it.

II. ACTIONS AND FIELD EQUATIONS

As can be found in many textbooks see, for example (Misner et al., 1973; Wald, 1984) there are
actually two variational principles that one can apply
to the EinsteinHilbert action in order to derive Einsteins equations: the standard metric variation and a
less standard variation dubbed Palatini variation [even
though it was Einstein and not Palatini who introduced
it (Ferraris et al., 1982)]. In the latter the metric and the
connection are assumed to be independent variables and
one varies the action with respect to both of them (we
will see how this variation leads to Einsteins equations
shortly), under the important assumption that the matter action does not depend on the connection. The choice
of the variational principle is usually referred to as a formalism, so one can use the terms metric (or second order)
formalism and Palatini (or first order) formalism. However, even though both variational principles lead to the
same field equation for an action whose Lagrangian is linear in R, this is no longer true for a more general action.
Therefore, it is intuitive that there will be two version of
f (R) gravity, according to which variational principle or
formalism is used. Indeed this is the case: f (R) gravity
in the metric formalism is called metric f (R) gravity and
f (R) gravity in the Palatini formalism is called Palatini
f (R) gravity (Buchdahl, 1970).
Finally, there is actually even a third version of f (R) gravity:
metric-affine f (R) gravity
(Sotiriou and Liberati, 2007a,b).
This comes about
if one uses the Palatini variation but abandons the
assumption that the matter action is independent of the
connection. Clearly, metric affine f (R) gravity is the
most general of these theories and reduces to metric or
Palatini f (R) gravity if further assumptions are made.
In this section we will present the actions and field
equations of all three versions of f (R) gravity and point
out their difference. We will also clarify the physical
meaning behind the assumptions that discriminate

5
them.
For an introduction to metric f (R) gravity see
also (Nojiri and Odintsov, 2007a), for a shorter
review of metric and Palatini f (R) gravity see
(Capozziello and Francaviglia, 2008) and for an extensive analysis of all versions of f (R) gravity and other
alternative theories of gravity see (Sotiriou, 2007b).
A. Metric formalism

Beginning from the action (3) and adding a matter

term SM , the total action for f (R) gravity takes the form
Z

1
Smet =
(5)
d4 x g f (R) + SM (g , ),
2
where collectively denotes the matter fields. Variation
with respect to the metric gives, after some manipulations and modulo surface terms
1
f (R)R f (R)g [ g ] f (R) = T ,
2
(6)
where, as usual,
2 SM
T =
,
g g

(7)

a prime denotes differentiation with respect to the argument, is the covariant derivative associated with
the Levi-Civita connection of the metric, and 
. Metric f (R) gravity was first rigorously studied
in (Buchdahl, 1970).3
It has to be stressed that there is a mathematical jump
in deriving eq. (6) from the action (5) having to do with
the surface terms that appear in the variation: as in the
case of the EinsteinHilbert action, the surface terms do
not vanish just by fixing the metric on the boundary.
For the EinsteinHilbert action, however, these terms
gather into a total variation of a quantity. Therefore,
it is possible to add a total divergence to the action in
order to heal it and arrive to a well-defined variational
principle (this is the well known GibbonsHawkingYork
surface term (Gibbons and Hawking, 1977; York, 1972)).
Unfortunately, the surface terms in the variation of the
action (3) do not consist of a total variation of some
quantity (the reader is urged to calculate the variation in
order to verify this fact) and it is not possible to heal
the action by just subtracting some surface term before
performing the variation.
The way out comes from the fact that the action includes higher order derivatives of the metric and, therefore, it should be possible to fix more degrees of freedom
on the boundary than those of the metric itself. There is

Specific attention to higher-dimensional f (R) gravity was paid in

(Gunther et al., 2002, 2003, 2005; Saidov and Zhuk, 2006, 2007).

no unique prescription for such a fixing in the literature so

far. Note also that the choice of fixing is not void of physical meaning, since it will be relevant for the Hamiltonian
formulation of the theory. However, the field equations
(6) would be unaffected by the fixing chosen and from a
purely classical perspective, such as the one followed here,
the field equations are all that one needs [see (Sotiriou,
2007b) for a more detailed discussion on these issues].
Setting aside the complications of the variation we can
now focus on the field equations (6). These are obviously
fourth order partial differential equations in the metric,
since R already includes second derivatives of the latter.
For an action which is linear in R, the fourth order terms
the last two on the left hand side vanish and the
theory reduces to GR.
Notice also that the trace of eq. (6)
f (R)R 2f (R) + 3f = T,

(8)

where T = g T , relates R with T differentially and

not algebraically as in GR, where R = T . This is already an indication that the field equations of f (R) theories will admit a larger variety of solutions than Einsteins
theory. As an example, we mention here that the JebsenBirkhoffs theorem, stating that the Schwarzschild solution is the unique spherically symmetric vacuum solution,
no longer holds in metric f (R) gravity. Without going
into details, let us stress that T = 0 no longer implies
that R = 0, or is even constant.
Eq. (8) will turn out to be very useful in studying various aspects of f (R) gravity, notably its stability and
weak-field limit. For the moment, let us use it to make
some remarks about maximally symmetric solutions. Recall that maximally symmetric solutions lead to a constant Ricci scalar. For R = constant and T = 0, eq. (8)
reduces to
f (R)R 2f (R) = 0,

(9)

which, for a given f , is an algebraic equation in R. If

R = 0 is a root of this equation and one takes this root,
then eq. (6) reduces to R = 0 and the maximally symmetric solution is Minkowski spacetime. On the other
hand, if the root of eq. (9) is R = C, where C is a constant, then eq. (6) reduces to R = g C/4 and the
maximally symmetric solution is de Sitter or anti-de Sitter space depending on the sign of C, just as in GR with
a cosmological constant.
Another issue that should be stressed is that of energy conservation. In metric f (R) gravity the matter
is minimally coupled to the metric. One can, therefore,
use the usual arguments based on the invariance of the
action under diffeomorphisms of the spacetime manifold
[coordinate transformations x x = x + followed by a pullback, with the field vanishing on the
boundary of the spacetime region considered, leave the
physics unchanged, see (Wald, 1984)] to show that T is
divergence-free. The same can be done at the level of the
field equations: a brute force calculation reveals that

6
the left hand side of eq. (6) is divergence-free (generalized
Bianchi identity) implying that T = 0 (Koivisto,
2006a).4
Finally, let us note that it is possible to write the field
equations in the form of Einstein equations with an effective stress-energy tensor composed of curvature terms
moved to the right hand side. This approach is questionable in principle (the theory is not Einsteins theory and
it is artificial to force upon it an interpretation in terms
of Einstein equations) but, in practice, it has been proved
to be useful in scalar-tensor gravity. Specifically, eq. (6)
can be written as
1
g R
2
[f (R) Rf (R)]
T
+ g
=
f (R)
2f (R)
[ f (R) g f (R)]
+
f (R)

G R

(10)

or


(ef f )
T
+
T
,

f (R)

G =

(11)

where the quantity Gef f G/f (R) can be regarded as

the effective gravitational coupling strength in analogy
to what is done in scalar-tensor gravity positivity of
Gef f (equivalent to the requirement that the graviton is
not a ghost) imposes that f (R) > 0. Moreover,
(ef f )
T

1 h f (R) Rf (R)
g + f (R)

2
i
g f (R)

(12)

is an effective stress-energy tensor which does not have

the canonical form quadratic in the first derivatives of
the field f (R), but contains terms linear in the second
derivatives. The effective energy density derived from
it is not positive-definite and none of the energy conditions holds. Again, this situation is analogous to that
occurring in scalar-tensor gravity. The effective stressenergy tensor (12) can be put in the form of a perfect
fluid energy-momentum tensor, which will turn out to be
useful in Sec. IV.

B. Palatini formalism

We have already mentioned that the Einstein equations can be derived using, instead of the standard metric variation of the EinsteinHilbert action, the Palatini
formalism, i.e., an independent variation with respect to

the metric and an independent connection (Palatini variation). The action is formally the same but now the Riemann tensor and the Ricci tensor are constructed with
the independent connection. Note that the metric is not
needed to obtain the latter from the former. For clarity
of notation, we denote the Ricci tensor constructed with
this independent connection as R and the corresponding Ricci scalar5 is R = g R . The action now takes
the form
Z

1
Spal =
(13)
d4 x g f (R) + SM (g , ).
2
GR will come about, as we will see shortly, when f (R) =
R. Note that the matter action SM is assumed to depend only on the metric and the matter fields and not on
the independent connection. This assumption is crucial
for the derivation of Einsteins equations from the linear
version of the action (13) and is the main feature of the
Palatini formalism.
It has already been mentioned that this assumption has
consequences for the physical meaning of the independent
connection (Sotiriou, 2006b,d; Sotiriou and Liberati,
2007b). Let us elaborate on this: recall that an affine
connection usually defines parallel transport and the covariant derivative. On the other hand, the matter action
SM is supposed to be a generally covariant scalar which
includes derivatives of the matter fields. Therefore, these
derivatives ought to be covariant derivatives for a general
matter field. Exceptions exist, such as a scalar field, for
which a covariant and a partial derivative coincide, and
the electromagnetic field, for which one can write a covariant action without the use of the covariant derivative
[it is the exterior derivative that is actually needed, see
next section and (Sotiriou and Liberati, 2007b)]. However, SM should include all possible fields. Therefore,
assuming that SM is independent of the connection can
imply one of two things (Sotiriou, 2006d): either we are
restricting ourselves to specific fields, or we are implicitly assuming that it is the Levi-Civita connection of the
metric that actually defines parallel transport. Since the
first option is implausibly limiting for a gravitational theory, we are left with the conclusion that the independent
connection does not define parallel transport or the
covariant derivative and the geometry is actually pseudoRiemannian. The covariant derivative is actually defined
by the Levi-Civita connection of the metric { }.
This also implies that Palatini f (R) gravity is a metric theory in the sense that it satisfies the metric postulates (Will, 1981). Let us clarify this: matter is minimally coupled to the metric and not coupled to any other
fields. Once again, as in GR or metric f (R) gravity,
one could use diffeomorphism invariance to show that

5
4

Energy-momentum complexes in the spherically symmetric case

have been computed in (Multamaki et al., 2008).

The term f (R) gravity is used generically for a theory in which

the action is some function of some Ricci scalar, not necessarily
R.

7
the stress energy tensor is conserved by the covariant
derivative defined with the Levi-Civita connection of the
T 6= 0). This can
metric, i.e., T = 0 (but
also be shown by using the field equations, which we will
present shortly, in order to calculate the divergence of T
with respect to the Levi-Civita connection of the metric
and show that it vanishes (Barraco et al., 1999; Koivisto,
2006a).6 Clearly then, Palatini f (R) gravity is a metric
theory according to the definition of (Will, 1981) (not
to be confused with the term metric in metric f (R)
gravity, which simply refers to the fact that one only
varies the action with respect to the metric). Conventionally thinking, as a consequence of the covariant conservation of the matter energy-momentum tensor, test
particles should follow geodesics of the metric in Palatini
f (R) gravity. This can be seen by considering a dust
fluid with T = u u and projecting the conservation equation T = 0 onto the fluid four-velocity u .
Similarly, theories that satisfy the metric postulates are
supposed to satisfy the Einstein Equivalence Principle as
well (Will, 1981). Unfortunately, things are more complicated here and, therefore, we set this issue aside for the
moment. We will return to it and attempt to fully clarify
it in Secs. VI.B and VI.C.2. For now, let us proceed with
our discussion of the field equations.
Varying the action (13) independently with respect to
the metric and the connection, respectively, and using
the formula

.
R =

(14)

It is now easy to see how the Palatini formalism leads

to GR when f (R) = R; in this case f (R) = 1 and
eq. (19) becomes the definition of the Levi-Civita connection for the initially independent connection . Then,
R = R , R = R and eq. (18) yields Einsteins equations. This reproduces the result that can be found in
textbooks (Misner et al., 1973; Wald, 1984). Note that
in the Palatini formalism for GR, the fact that the connection turns out to be the Levi-Civita one is a dynamical
feature instead of an a priori assumption.
It is now evident that generalizing the action to be a
general function of R in the Palatini formalism is just
as natural as it is to generalize the EinsteinHilbert
action in the metric formalism.7 Remarkably, even
though the two formalisms give the same results for
linear actions, they lead to different results for more
general actions (Buchdahl, 1970; Burton and Mann,
1998a,b; Exirifard and Sheikh-Jabbari, 2008; Querella,
1998; Shahid-Saless, 1987).
Finally, let us present some useful manipulations of the
field equations. Taking the trace of eq. (18) yields
f (R)R 2f (R) = T.

(20)

As in the metric case, this equation will prove very useful

later on. For a given f , it is an algebraic equation in
R. For all cases in which T = 0, including vacuum and
electrovacuum, R will therefore be a constant and a root
of the equation

yields
1
f (R)R() f (R)g = T ,
2


gf (R)g



)

+
gf (R)g ( = 0,

(16)

where T is defined in the usual way as in eq. (7),

denotes the covariant derivative defined with the independent connection , and () and [] denote symmetrization or anti-symmetrization over the indices and
, respectively. Taking the trace of eq. (16), it can be
easily shown that


gf (R)g = 0,

(17)
which implies that we can bring the field equations into
the more economical form
1
f (R)R() f (R)g = G T ,
2


gf (R)g = 0,

f (R)R 2f (R) = 0.

(15)

(18)

We will not consider cases for which this equation has

no roots since it can be shown that the field equations
are then inconsistent (Ferraris et al., 1992). Therefore,
choices of f that lead to this behaviour should simply
be avoided. Eq. (21) can also be identically satisfied if
f (R) R2 . This very particular choice for f leads to
a conformally invariant theory (Ferraris et al., 1992). As
is apparent from eq. (20), if f (R) R2 then only conformally invariant matter, for which T = 0 identically,
can be coupled to gravity. Matter is not generically conformally invariant though, and so this particular choice
of f is not suitable for a low energy theory of gravity.
We will, therefore, not consider it further [see (Sotiriou,
2006b) for a discussion].
Next, we consider eq. (19). Let us define a metric
conformal to g as

(19)

Energy supertensors and pseudotensors in Palatini f (R) gravity were studied in (Barraco et al., 1999; Borowiec et al., 1994,
1998; Ferraris et al., 1992) and alternative energy definitions
were given in (Deser and Tekin, 2002, 2003a,b, 2007).

(21)

h f (R)g .

(22)

See, however, (Sotiriou, 2007b) for further analysis of the f (R)

action and how it can be derived from first principles in the two
formalisms.

8
It can easily be shown that8

h h = g f (R)g .

(23)

Then, eq. (19) becomes the definition of the LeviCivita connection of h and can be solved algebraically
to give
=

1
h ( h + h h ) ,
2

(24)

Notice that, assuming that we know the root of eq. (20),

R = R(T ), we have completely eliminated the independent connection from this equation. Therefore, we have
successfully reduced the number of field equations to one
and at the same time both sides of eq. (28) depend only
on the metric and the matter fields. In a sense, the theory has been brought to the form of GR with a modified
source.
We can now straightforwardly deduce the following:
When f (R) = R, the theory reduces to GR, as
discussed previously.

or, equivalently, in terms of g ,

h
1 1
=
g (f (R)g ) + (f (R)g )

2 f (R)
i
(f (R)g ) ,
(25)

For matter fields with T = 0, due to eq. (21), R and

consequently f (R) and f (R) are constants and the
theory reduces to GR with a cosmological constant
and a modified coupling constant G/f . If we denote the value of R when T = 0 as R0 , then the
value of the cosmological constant is


f (R0 )
R0
1
R0
=
,
(29)
2
f (R0 )
4

Given that eq. (20) relates R algebraically with T , and

since we have an explicit expression for in terms
of R and g , we can in principle eliminate the independent connection from the field equations and express
them only in terms of the metric and the matter fields.
Actually, the fact that we can algebraically express
in terms of the latter two already indicates that this connections act as some sort of auxiliary field. We will explore this further in Sec. III. For the moment, let us
take into account how the Ricci tensor transforms under
conformal transformations and write
3
1
( f (R)) ( f (R))

2 (f (R))2


1
1

g  f (R). (26)
f (R)
2

where we have used eq. (21). Besides vacuum, T =

0 also for electromagnetic fields, radiation, and any
other conformally invariant type of matter.
In the general case T 6= 0, the modified source on
the right hand side of eq. (28) includes derivatives
of the stress-energy tensor, unlike in GR. These are
implicit in the last two terms of eq. (28), since f is
in practice a function of T , given that9 f = f (R)
and R = R(T ).

R = R +

Contraction with g yields

3
( f (R)) ( f (R))
2(f (R))2
3
+
f (R).
f (R)

R=R +

The serious implications of this last observation will

become clear in Sec. VI.C.1.
C. Metric-affine formalism

(27)

Note the difference between R and the Ricci scalar of h

due to the fact that g is used here for the contraction
of R .
Replacing eqs. (26) and (27) in eq. (18), and after some
easy manipulations, one obtains



f
1
G = T g R
(28)
f
2
f
1
+ ( g ) f
f


3 1
1
2 ( f )( f ) g (f )2 .
2f
2

As we already pointed out, the Palatini formalism

of f (R) gravity relies on the crucial assumption that
the matter action does not depend on the independent
connection. We also argued that this assumption relegates this connection to the role of some sort of auxiliary field and the connection carrying the usual geometrical meaning parallel transport and definition
of the covariant derivative remains the Levi-Civita
connection of the metric. All of these statements will
be supported further in the forthcoming sections, but
for the moment let us consider what would be the outcome if we decided to be faithful to the geometrical interpretation of the independent connection : this

9
8

This calculation holds in four dimensions. When the number of dimensions D is different from 4 then, instead of using eq. (22), the conformal metric h should be introduced as
h [f (R)]2/(D2) g in order for eq. (23) to still hold.

Note that, apart from special cases such as a perfect fluid, T

and consequently T already include first derivatives of the matter
fields, given that the matter action has such a dependence. This
implies that the right hand side of eq. (28) will include at least
second derivatives of the matter fields, and possibly up to third
derivatives.

9
would imply that we would define the covariant derivatives of the matter fields with this connection and, therefore, we would have SM = SM (g , , ). The action of this theory, dubbed metric-affine f (R) gravity
(Sotiriou and Liberati, 2007b), would then be [note the
difference with respect to the action (13)]
Z

1
Sma =
d4 x gf (R) + SM (g , , ). (30)
2
1. Preliminaries

Before going further and deriving field equations from

this action certain issues need to be clarified. First, since
now the matter action depends on the connection, we
should define a quantity representing the variation of SM
with respect to the connection, which mimics the definition of the stress-energy tensor. We call this quantity
the hypermomentum and is defined as (Hehl and Kerling,
1978)
2 SM

.
g

(31)

Additionally, since the connection is now promoted to

the role of a completely independent field, it is interesting
to consider not placing any restrictions to it. Therefore,
besides dropping the assumption that the connection is
related to the metric, we will also drop the assumption
that the connection is symmetric. It is useful to define
the following quantities: the non-metricity tensor
g ,
Q

(32)

which measures the failure of the connection to covariantly conserve the metric, the trace of the non-metricity
tensor with respect to its last two (symmetric) indices,
which is called the Weyl vector,
Q

1
Q ,
4

(33)

and the Cartan torsion tensor

S [] ,

(34)

which is the antisymmetric part of the connection.

By allowing a non-vanishing Cartan torsion tensor we
are allowing the theory to naturally include torsion. Even
though this brings complications, it has been considered by some to be an advantage for a gravity theory since some matter fields, such as Dirac fields, can
be coupled to gravity in a way which might be considered more natural (Hehl et al., 1995): one might expect that at some intermediate or high energy regime,
the spin of particles might interact with the geometry
(in the same sense that macroscopic angular momentum interacts with geometry) and torsion can naturally
arise. Theories with torsion have a long history, probably

starting with the EinsteinCartan(SciamaKibble) theory (Cartan, 1922, 1923, 1924; Hehl et al., 1976; Kibble,
1961; Sciama, 1964). In this theory, as well as in other
theories with an independent connection, some part of
the connection is still related to the metric (e.g., the
non-metricity is set to zero). In our case, the connection is left completely unconstrained and is to be determined by the field equations. Metric-affine gravity with
the linear version of the action (30) was initially proposed in (Hehl and Kerling, 1978) and the generalization
to f (R) actions was considered in (Sotiriou and Liberati,
2007a,b).
Unfortunately, leaving the connection completely unconstrained comes with a complication. Let us consider
the projective transformation
+ ,

(35)

where is an arbitrary covariant vector field. One can

easily show that the Ricci tensor will correspondingly
transform like
R R 2[ ] .

(36)

However, given that the metric is symmetric, this implies

that the curvature scalar does not change
R R,

(37)

i.e., R is invariant under projective transformations.

Hence the EinsteinHilbert action or any other action
built from a function of R, such as the one used here,
is projective invariant in metric-affine gravity. However,
the matter action is not generically projective invariant
and this would be the cause of an inconsistency in the
field equations.
One could try to avoid this problem by generalizing
the gravitational action in order to break projective invariance. This can be done in several ways, such as
allowing for the metric to be non-symmetric as well,
adding higher order curvature invariants or terms including the Cartan torsion tensor [see (Sotiriou, 2007b;
Sotiriou and Liberati, 2007b) for a more detailed discussion]. However, if one wants to stay within the framework
of f (R) gravity, which is our subject here, then there is
only one way to cure this problem: to somehow constrain
the connection. In fact, it is evident from eq. (35) that,
if the connection were symmetric, projective invariance
would be broken. However, one does not have to take
such a drastic measure.
To understand this issue further, we should re-examine
the meaning of projective invariance. This is very similar
to gauge invariance in electromagnetism (EM). It tells us
that the corresponding field, in this case the connections
, can be determined from the field equations up to a
projective transformation [eq. (35)]. Breaking this invariance can therefore come by fixing some degrees of freedom of the field, similarly to gauge fixing. The number
of degrees of freedom which we need to fix is obviously
the number of the components of the four-vector used

10
for the transformation, i.e., simply four. In practice, this
means that we should start by assuming that the connection is not the most general which one can construct,
but satisfies some constraints.
Since the degrees of freedom that we need to fix
are four and seem to be related to the non-symmetric
part of the connection, the most obvious prescription
is to demand that S = S be equal to zero, which
was first suggested in (Sandberg, 1975) for a linear action and shown to work also for an f (R) action in
(Sotiriou and Liberati, 2007b).10 Note that this does
not mean that should vanish, but merely that
= . Imposing this constraint can easily be done
by adding a Lagrange multiplier B . The additional term
in the action will be
Z

SLM = d4 x g B S .
(38)

The action (30) with the addition of the term in eq. (38)
is, therefore, the action of the most general metric-affine
f (R) theory of gravity.
2. Field Equations

We are now ready to vary the action and obtain field

equations. Due to space limitations, we will not present
the various steps of the variation here. Instead we merely
give the formula

+ 2 , (39)
R =
[]

which is useful to those wanting to repeat the variation

as an exercise, and we also stress our definitions for the
covariant derivative
A = A + A A .

(40)
and for the Ricci tensor of an independent connection
R = R
=

(41)

The outcome of varying independently with respect to

the metric, the connection and the Lagrange multiplier
is, respectively,
1
f (R)R() f (R)g = T ,
(42)
2



+ gf (R)g

gf (R)g
g


+2f (R) g S g S + g S
S

10

= 0.

= ( B [ ] ),

(43)

(44)

The proposal of (Hehl and Kerling, 1978) to fix part of the nonmetricity, namely the Weyl vector Q , in order to break projective invariance works only when f (R) = R (Sotiriou, 2007b;
Sotiriou and Liberati, 2007b).

Taking the trace of eq. (43) over the indices and and
using eq. (44) yields
B =

2
.
3

(45)

Therefore, the final form of the field equations is

1
f (R)R() f (R)g = T ,
(46)
2



gf (R)g
+ gf (R)g

g
2
(47)
+2f (R)g S = ( [ ] ),
3
S = 0.
(48)
Next, we examine the role of . By splitting eq. (47)
into a symmetric and an antisymmetric part and performing contractions and manipulations it can be shown
that (Sotiriou and Liberati, 2007b)
[]

= 0 S = 0.

(49)

This straightforwardly implies two things: a) Any torsion

[]
is introduced by matter fields for which
is nonvanishing; b) torsion is not propagating, since it is given
[]
algebraically in terms of the matter fields through .
It can, therefore, only be detected in the presence of such
matter fields. In the absence of the latter, spacetime will
have no torsion.
In a similar fashion, one can use the symmetrized version of eq. (47) to show that the symmetric part of the
()
hypermomentum
is algebraically related to the
non-metricity Q . Therefore, matter fields with non()
vanishing
will introduce non-metricity. However,
in this case things are slightly more complicated because
part of the non-metricity is also due to the functional
form of the Lagrangian itself [see (Sotiriou and Liberati,
2007b)].
We will not perform a detailed study of different matter fields and their role in metric-affine gravity. We
refer the reader to the more exhaustive analysis of
(Sotiriou and Liberati, 2007b) for details and we restrict
ourselves to the following remarks: Obviously, there are
certain types of matter fields for which = 0. Characteristic examples are
A scalar field, since in this case the covariant
derivative can be replaced with a partial derivative. Therefore, the connection does not enter the
matter action.
The electromagnetic field (and gauge fields in general), since the electromagnetic field tensor F is
defined in a covariant manner using the exterior
derivative. This definition remains unaffected when
torsion is included [this can be related to gauge invariance, see (Sotiriou and Liberati, 2007b) for a
discussion].

11
On the contrary, particles with spin, such as Dirac fields,
generically have a non-vanishing hypermomentum and
will, therefore, introduce torsion. A more complicated
case is that of a perfect fluid with vanishing vorticity. If
we set torsion aside, or if we consider a fluid describing
particles that would initially not introduce any torsion
then, as for a usual perfect fluid in GR, the matter action can be written in terms of three scalars: the energy
density, the pressure, and the velocity potential (Schakel,
1996; Stone, 2000). Therefore such a fluid will lead to a
vanishing . However, complications arise when torsion is taken into account: Even though it can be argued
that the spins of the individual particles composing the
fluids will be randomly oriented, and therefore the expectation value for the spin should add up to zero, fluctuations around this value will affect spacetime (Hehl et al.,
1976; Sotiriou and Liberati, 2007b). Of course, such effects will be largely suppressed, especially in situations
in which the energy density is small, such as late time
cosmology.
It should be evident by now that, due to eq. (49), the
field equations of metric f (R) gravity reduce to eqs. (15)
and (16) and, ultimately, to the field equations of Palatini
f (R) gravity (18) and (19), for all cases in which =
0. Consequently, in vacuo, where also T = 0, they
will reduce to the Einstein equations with an effective
cosmological constant given by eq. (29), as discussed at
the end of Sec. II.B for Palatini f (R) gravity.
In conclusion, metric-affine f (R) gravity appears to
be the most general case of f (R) gravity. It includes
enriched phenomenology, such as matter-induced nonmetricity and torsion. It is worth stressing that torsion
comes quite naturally, since it is actually introduced by
particles with spin (excluding gauge fields). Remarkably,
the theory reduces to GR in vacuo or for conformally invariant types of matter, such as the electromagnetic field,
and departs from GR in the same way that Palatini f (R)
gravity does for most matter fields that are usually studied as sources of gravity. However, at the same time, it
exhibits new phenomenology in less studied cases, such
as in the presence of Dirac fields, which include torsion
and non-metricity. Finally, let us repeat once more that
Palatini f (R) gravity, despite appearances, is really a
metric theory according to the definition of (Will, 1981)
(and the geometry is a priori pseudo-Riemannian).11 On
the contrary, metric-affine f (R) gravity is not a metric
theory (hence the name). Consequently, it should also
be clear that T is not divergence-free with respect to
the covariant derivative defined with the Levi-Civita con actually). However, the physical
nection (nor with
meaning of this last statement is questionable and deserves further analysis, since in metric-affine gravity T

11

As mentioned in Sec. II.B, although the metric postulates are

manifestly satisfied, there are ambiguities regarding the physical
interpretation of this property and its relation with the Einstein
Equivalence Principle (see Sec. VI.C.1).

does not really carry the usual meaning of a stress-energy

tensor (for instance, it does not reduce to the special relativistic tensor at an appropriate limit and at the same
time there is also another quantity, the hypermomentum,
which describes matter characteristics).

III. EQUIVALENCE WITH BRANSDICKE THEORY

AND CLASSIFICATION OF THEORIES

In the same way that one can make variable redefinitions in classical mechanics in order to bring an equation
describing a system to a more attractive, or easy to handle, form (and in a very similar way to changing coordinate systems), one can also perform field redefinitions in
a field theory, in order to rewrite the action or the field
equations.
There is no unique prescription for redefining the fields
of a theory. One can introduce auxiliary fields, perform
renormalizations or conformal transformations, or even
simply redefine fields to ones convenience.
It is important to mention that, at least within a classical perspective such as the one followed here, two theories are considered to be dynamically equivalent if, under a suitable redefinition of the gravitational and matter
fields, one can make their field equations coincide. The
same statement can be made at the level of the action.
Dynamically equivalent theories give exactly the same
results when describing a dynamical system which falls
within the purview of these theories. There are clear advantages in exploring the dynamical equivalence between
theories: we can use results already derived for one theory in the study of another, equivalent, theory.
The term dynamical equivalence can be considered
misleading in classical gravity. Within a classical perspective, a theory is fully described by a set of field
equations. When we are referring to gravitation theories,
these equations describe the dynamics of gravitating systems. Therefore, two dynamically equivalent theories are
actually just different representations of the same theory
(which also makes it clear that all allowed representations
can be used on an equal footing).
The issue of distinguishing between truly different theories and different representations of the same theory (or
dynamically equivalent theories) is an intricate one. It
has serious implications and has been the cause of many
misconceptions in the past, especially when conformal
transformations are used in order to redefine the fields
(e.g., the Jordan and Einstein frames in scalar-tensor theory). It goes beyond the scope of this review to present
a detailed analysis of this issue. We refer the reader to
the literature, and specifically to (Sotiriou et al., 2007)
and references therein for a detailed discussion. Here, we
simply mention that, given that they are handled carefully, field redefinitions and different representations of
the same theory are perfectly legitimate and constitute
very useful tools for understanding gravitational theories.
In what follows, we review the equivalence between

12
metric and Palatini f (R) gravity with specific theories
within the BransDicke class with a potential. It is shown
that these versions of f (R) gravity are nothing but different representations of BransDicke theory with Brans
Dicke parameter 0 = 0 and 0 = 3/2, respectively. We
comment on this equivalence and on whether preference
to a specific representation should be an issue. Finally,
we use this equivalence to perform a classification of f (R)
gravity.
A. Metric formalism

It has been noticed quite early that metric quadratic

gravity can be cast into the form of a BransDicke theory
and it did not take long for these results to be extended
to more general actions which are functions of the Ricci
scalar of the metric (Barrow, 1988; Barrow and Cotsakis,
1988; Teyssandier and Tourrenc, 1983; Wands, 1994) [see
also (Flanagan, 2003) and (Cecotti, 1987; Wands, 1994)
for the extension to theories of the type f (R, k R) with
k 1 of interest in supergravity]. This equivalence has
been re-examined recently due to the increased interest
in metric f (R) gravity (Chiba, 2003; Flanagan, 2004a;
Sotiriou, 2006b). Let us present this equivalence in some
detail.
We will work at the level of the action but the same
approach can be used to work directly at the level of the
field equations. We begin with metric f (R) gravity. For
the convenience of the reader, we rewrite here the action
(5):
Z

1
d4 x g f (R) + SM (g , ).
(50)
Smet =
2
One can introduce a new field and write the dynamically equivalent action
Z

1
d4 x g [f () + f ()(R )] +
Smet =
2
+SM (g , ).
(51)
Variation with respect to leads to the equation
f ()(R ) = 0.

(52)

Therefore, = R if f () 6= 0, which reproduces the

action (5).12 Redefining the field by = f () and
setting
V () = () f (()),

(53)

the action takes the form

Z

1
Smet =
d4 x g [R V ()] + SM (g , ). (54)
2

12

The action is sometimes called R-regular by mathematical

physicists if f (R) 6= 0 [e.g., (Magnano and Sokolowski, 1994)].

This is the Jordan frame representation of the action

of a BransDicke theory with BransDicke parameter
0 = 0. An 0 = 0 BransDicke theory [sometimes called massive dilaton gravity (Wands, 1994)]
was originally proposed by (OHanlon, 1972b) in order to generate a Yukawa term in the Newtonian limit
and has been occasionally considered in the literature
(Anderson, 1971; Barber, 2003; Dabrowski et al., 2007;
Davidson, 2005; Deser, 1970; Fujii, 1982; OHanlon,
1972a; OHanlon and Tupper, 1972).
It should be
stressed that the scalar degree of freedom = f () is
quite different from a matter field; for example, like all
nonminimally coupled scalars, it can violate all of the
energy conditions (Faraoni, 2004a).
The field equations corresponding to the action (54)
are

1
T
g V ()

2
1
+ ( g ) ,

R = V ().

G =

(55)
(56)

These field equations could have been derived directly

from eq. (6) using the same field redefinitions that were
mentioned above for the action. By taking the trace of
eq. (55) in order to replace R in eq. (56), one gets
3 + 2V ()

dV
= T.
d

(57)

This last equation determines the dynamics of for given

matter sources.
The condition f 6= 0 for the scalar-tensor theory to be
equivalent to the original f (R) gravity theory can be seen
as the condition that the change of variable = f (R)
needed to express the theory as a BransDicke one (54)
be invertible, i.e., d/dR = f 6= 0. This is a sufficient but not necessary condition for invertibility: it is
only necessary that f (R) be continuous and one-to-one
(Olmo, 2007). By looking at eq. (52), it is seen that
f 6= 0 implies = f (R) and the equivalence of the actions (3) and (51). When f is not defined, or it vanishes,
the equality = f (R) and the equivalence between the
two theories can not be guaranteed (although this it is
not a priori excluded by f = 0).
Finally, let us mention that, as usual in BransDicke
theory and more general scalar-tensor theories, one can
perform a conformal transformation and rewrite the action (54) in what is called the Einstein frame (as opposed
to the Jordan frame). Specifically, by performing the conformal transformation
g g = f (R) g g

(58)

and the scalar field redefinition = f (R) with

r
20 + 3 d

d =
,
(59)
2

13
a scalar-tensor theory is mapped into the Einstein frame
in which the new scalar field couples minimally to
the Ricci curvature and has canonical kinetic energy, as
described by the gravitational action
"
#
Z

p
R
1
(g)
4
. (60)
g
S = d x
U ()
2 2
For the 0 = 0 equivalent of metric f (R) gravity we have
f (R) = e

2
3

= Rf (R) f (R) ,
U ()
2
2 (f (R))

(61)

(62)

and the complete action is

where R = R(),
#
"
Z

p
1
R

+
U ()

Smet
=
d4 x
g
2 2

+SM (e 2/3 g , ). (63)

A direct transformation to the Einstein frame, without
the intermediate passage from the Jordan frame, has
been discovered in (Barrow and Cotsakis, 1988; Whitt,
1984).
We stress once more that the actions (5), (54), and (63)
are nothing but different representations of the same
theory.13 Additionally, there is nothing exceptional
about the Jordan or the Einstein frame of the Brans
Dicke representation, and one can actually find infinitely
many conformal frames (Flanagan, 2004a; Sotiriou et al.,
2007).

B. Palatini formalism

Palatini f (R) gravity can also be cast in the form of a

BransDicke theory with a potential (Flanagan, 2004b;
Olmo, 2005b; Sotiriou, 2006b). As a matter of fact, beginning from the Palatini f (R) action, which we repeat
here for the readers convenience
Z

1
Spal =
d4 x g f (R) + SM (g , ),
(64)
2
and following exactly the same steps as before, i.e., introducing a scalar field which we later redefine in terms
of , yields
Z

1
d4 x g [R V ()] + SM (g , ). (65)
Spal =
2

13

This has been an issue of debate and confusion, see for example
the references in (Faraoni and Nadeau, 2007).

Even though the gravitational part of this action is formally the same as that of the action (54), this action is
not a BransDicke one with 0 = 0: R is not the Ricci
scalar of the metric g . However, we have already seen
that the field equation (18) can be solved algebraically
for the independent connection yielding eq. (25). This
implies that we can replace the connection in the action without affecting the dynamics of the theory (the
independent connection is practically an auxiliary field).
Alternatively, we can directly use eq. (27), which relates
R and R. Therefore, the action (65) can be rewritten,
modulo surface terms, as


Z

1
3
Spal =
V ()
d4 x g R +
2
2
+SM (g , ).
(66)
This is the action of a BransDicke theory with Brans
Dicke parameter 0 = 3/2. The corresponding field
equations obtained from the action (66) through variation with respect to the metric and the scalar are


1
3 
T 2 g +

2
2
V
1
g ,
(67)
+ ( g )

1
 = (R V ) +
.
(68)
3
2

G =

Once again, we can use the trace of eq. (67) in order to

eliminate R in eq. (68) and relate directly to the matter
sources. The outcome is
2V V = T.

(69)

Finally, one can also perform the conformal transformation (58) in order to rewrite the action (66) in the
Einstein frame. The result is
#
"
Z

p
R

4
U () + SM (1 g , ),
Spal = d x
g
2
(70)
where U () = V ()/(2 2 ). Note that here we have not
used any redefinition for the scalar.
To conclude, we have established that Palatini f (R)
gravity can be cast into the form of an 0 = 3/2 Brans
Dicke theory with a potential.

C. Classification

The scope of this section is to present a classification

of the different versions of f (R) gravity. However, before
doing so, some remarks are in order.
Let us, first of all, use the BransDicke representation
of both metric and Palatini f (R) gravity to comment
on the dynamics of these theories. This representation
makes it transparent that metric f (R) gravity has just

14
one extra scalar degree of freedom with respect to GR.
The absence of a kinetic term for the scalar in the action (54) or in eq. (56) should not mislead us to think
that this degree of freedom does not carry dynamics.
As can be seen by eq. (57), is dynamically related to
the matter fields and, therefore, it is a dynamical degree of freedom. Of course, one should also not fail to
mention that eq. (56) does constrain the dynamics of .
In this sense metric f (R) gravity and 0 = 0 Brans
Dicke theory differs from the general BransDicke theories and constitutes a special case. On the other hand,
in the 0 = 3/2 case which corresponds to Palatini
f (R) gravity, the scalar appears to have dynamics in
the action (66) or in eq. (68). However, once again this is
misleading since, as is clear from eq. (69), is in this case
algebraically related to the matter and, therefore, carries
no dynamics of its own [indeed the field eqs. (67) and (69)
could be combined to give eq. (28), eliminating completely]. As a remark, let us state that the equivalence
between Palatini f (R) gravity and 0 = 3/2 Brans
Dicke theory and the clarifications just made highlight
two issues already mentioned: the fact that Palatini f (R)
gravity is a metric theory according to the definition of
(Will, 1981), and the fact that the independent connection is actually some sort of auxiliary field.
The fact that the dynamics of are not transparent at
the level of the action in both cases should not come as
a big surprise: is coupled to the derivatives of the metric (through the coupling with R) and, therefore, partial
integrations to free or g during the variation are
bound to generate dynamical terms even if they are not

and
i g

tiiii
METRIC-AFFINE f (R)

initially present in the action. The 0 = 3/2 case is

even more intricate because the dynamical terms generated through this procedure exactly cancel the existing
one in the action.
We already saw an example of how different representations of the theory can highlight some of its characteristics and be very useful for our understanding of it.
The equivalence between f (R) gravity and BransDicke
theory will turn out to be very useful in the forthcoming
sections.
Until now we have not discussed any possible equivalence between BransDicke theory and metric-affine f (R)
gravity. However, it is quite straightforward to see that
there cannot be any. Metric-affine f (R) gravity is not a
metric theory and, consequently, it can not be cast into
the form of one, such as BransDicke theory. For the
sake of clarity, let us state that one could still start from
the action (30) and follow the steps of the previous section to bring its gravitational part into the form of the
action (66). However, the matter action would have an
explicit dependence from the connection. Additionally,
one would not be able to use eq. (27) to eliminate R in
favour of R since this only holds in Palatini f (R) gravity.
In conclusion, metric-affine f (R) gravity is the most
general case of f (R) gravity. Imposing further assumptions can lead to both metric or Palatini f (R) gravity,
which can be cast into the form of 0 = 0 and 0 = 3/2
BransDicke theories with a potential. In both cases, restricting the functional form of the action leads to GR.
These results are summarized in the schematic diagram
of Fig. 1.

f (R) GRAVITY
JJ
i
JJ
iiii
JJ
independent
JJ
JJ


= J
J

JJ
JJ
JJ
JJ
$
METRIC f (R)
KS
i
iiii

SM = SM (g , )


PALATINI f (R)
UUUU
KS
U

f (R) = R
UU

f (R) 6= 0

UUUU
UU*


BRANSDICKE, 0 = 23

GR

f (R) = R
iiii
i
i
i
ti

f (R) 6= 0


BRANSDICKE, 0 = 0

FIG. 1 Classification of f (R) theories of gravity and equivalent BransDicke theories. The flowchart shows the list of assumptions that are needed to arrive to the various versions of f (R) gravity and GR beginning from the the general f (R) action. It
also includes the equivalent BransDicke classes. Taken from (Sotiriou, 2006b).

D. Why f (R) gravity then?

Since f (R) gravity in both the metric and the Palatini

formalisms can acquire a BransDicke theory represen-

tation, one might be led to ask two questions: first, why

should we consider the f (R) representation and not just

15
work with the BransDicke one, and second, why, since
we know a lot about BransDicke theory, should we regard f (R) gravity as unexplored or interesting?
The answer to the first question is quite straightforward. There is actually no reason to prefer either of the
two representations at least as far as classical gravity is concerned. There can be applications where the
f (R) representation can be more convenient and applications where the BransDicke representation is more convenient. One should probably mention that habit affects
our taste and, therefore, an f (R) representation seems
more appealing to relativists due to its more apparent
geometrical nature, whereas the BransDicke representation seems more appealing to particle physicists. This
issue can have theoretical implications. To give an example: if f (R) gravity is considered as a step towards a more
complicated theory, which generalisation would be more
straightforward will depend on the chosen representation
[see also (Sotiriou et al., 2007) for a discussion].
Whether f (R) theories of gravity are unexplored and
interesting or just an already-studied subcase of Brans
Dicke theory, is a more practical question that certainly
deserves a direct answer. It is indeed true that scalartensor theories and, more precisely, BransDicke theory are well-studied theories which have been extensively
used in many applications, including cosmology. However, the specific choices 0 = 0, 3/2 for the Brans
Dicke parameter are quite exceptional, as already mentioned in the previous section. It is also worthwhile pointing out the following: a) As far as the 0 = 0 case is
concerned, one can probably speculate that it is the apparent absence of the kinetic term for the scalar in the
action which did not seem appealing and prevented the
study of this theory. b) The 0 = 3/2 case leads to a
conformally invariant theory in the absence of the potential [see (Sotiriou, 2006b) and references therein], which
constituted the initial form of BransDicke theory, and
hence it was considered non-viable (a coupling with nonconformally invariant matter is not feasible). However, in
the presence of a potential, the theory no longer has this
feature. Additionally, most calculations which are done
for a general value of 0 in the literature actually exclude
0 = 3/2, mainly because, merely for simplicity purposes, they are done in such a way that the combination
20 + 3 appears in a denominator (see also Sec. V.A).
In any case, the conclusion is that the theories in the
BransDicke class that correspond to metric and Palatini
f (R) gravity had not yet been explored before the recent
re-introduction of f (R) gravity and, as will also become
clear later, several of their special characteristics when
compared with more standard BransDicke theories were
revealed through studies of f (R) gravity.

IV. COSMOLOGICAL EVOLUTION AND CONSTRAINTS

We now turn our attention to cosmology, which motivated the recent surge of interest in f (R) gravity in order

to explain the current cosmic acceleration without the

need for dark energy. Before reviewing how f (R) gravity
might provide a solution to the more recent cosmological
riddles, let us stress that the following criteria must be
satisfied in order for an f (R) model to be theoretically
consistent and compatible with cosmological observations
and experiments. The model must:
have the correct cosmological dynamics;
exhibit the correct behaviour of gravitational perturbations;
generate cosmological perturbations compatible
with the cosmological constraints from the cosmic
microwave background, large scale structure, Big
Bang Nucleosynthesis, and gravity waves.
These are independent requirements to be studied separately, and they must all be satisfied.
A. Background evolution

In cosmology, the identification of our universe with a

FriedmannLemaitreRobertsonWalker (FLRW) spacetime is largely based on the high degree of isotropy
measured in the cosmic microwave background; this
identification relies on a formal result known as the
EhlersGerenSachs (EGS) theorem (Ehlers et al., 1968)
which is a kinematical characterization of FLRW spaces
stating that, if a congruence of timelike freely falling
observers see an isotropic radiation field, then (assuming that isotropy holds about every spatial point)
the spacetime is spatially homogeneous and isotropic
and, therefore, a FLRW one. This applies to a universe filled with any perfect fluid that is geodesic
and barotropic (Clarkson and Barrett, 1999; Ellis et al.,
1983a,b). Moreover, an almost-EGS theorem holds:
spacetimes that are close to satisfying the EGS conditions are close to FLRW universes in an appropriate sense
(Stoeger et al., 1995). One would expect that the EGS
theorem be extended to f (R) gravity; indeed, its validity
for the (metric) theory
Z


1
S=
d4 x g R + R2 + R R + SM (71)
2

was
proved
in
(Maartens and Taylor,
1994;
Taylor and Maartens, 1995) and the generalization to
arbitrary metric f (R) gravity was given by (Rippl et al.,
1996). The validity of the EGS theorem can also be seen
through the equivalence between f (R) and BransDicke
theory: the theorem was extended to scalar-tensor
theories in (Clarkson et al., 2001, 2003). Since metric
and Palatini f (R) gravities are equivalent to = 0
and 0 = 3/2 BransDicke theories respectively, it
seems that the results of (Clarkson et al., 2001, 2003)
can be considered as straightforward generalizations
of the EGS theorem in both versions of f (R) gravity

16
as well. However, in the case of Palatini f (R) gravity
there is still some doubt regarding this issue due to
complications in averaging (Flanagan, 2004b).
1. Metric f (R) gravity

Considering the discussion above, it is valid to use the

FLRW line element




dr2
2
2
2
2
2
2
2
+ r d + sin d
ds = dt + a (t)
1 kr2
(72)
as a local description of spacetime at cosmological scales,
where (t, r, , ) are comoving coordinates. We remind
the reader that k = 1, 0, 1 according to whether the
universe is hyperspherical, spatially flat, or hyperbolic
and that a(t) is called the scale factor. Part of the standard approach, which we follow here as well, is to use
a perfect fluid description for matter with stress-energy
tensor
T = ( + P ) u u + P g ,

(73)

where u denotes the four-velocity of an observer comoving with the fluid and and P are the energy density and
pressure of the fluid, respectively.
Note that the value of k is an external parameter. As
in many other works in the literature, in what follows we
choose k = 0, i.e., we focus on a spatially flat universe.
This choice in made in order to simplify the equations and
should be viewed sceptically. It is sometimes claimed in
the literature that such a choice is favoured by the data.
However, this is not entirely correct. Even though the
data [e.g. (Spergel et al., 2007)] indicate that the current
value of k is very close to zero, it should be stressed that
this does not really reveal the value of k itself. Since
k =

k
,
a2 H 2

under the assumption that k = 0 should be considered

preliminary until the influence of the spatial curvature
is precisely determined, since there are indications that
even a very small value of k may have an effect on them
[see, for instance (Clarkson et al., 2007)].
Returning to our discussion, inserting the flat FLRW
metric in the field equations (6) and assuming that the
stress-energy tensor is that of eq. (73) yields



Rf f
, (75)
H2 =

3H
Rf
3f
2
h

2
2 f + 2H Rf

2H + 3H = P + (R)
f
i
+ 1 (f Rf ) .
+Rf
(76)
2

With some hindsight, we assume that f > 0 in order to have a positive effective gravitational coupling
and f > 0 to avoid the Dolgov-Kawasaki instability
(Dolgov and Kawasaki, 2003a; Faraoni, 2006a) discussed
in Sec. V.B.
A significant part of the motivation for f (R) gravity
is that it can lead to accelerated expansion without the
need for dark energy (or an inflaton field). An easy way
to see this is to define an effective energy density and
pressure of the geometry as
ef f =
Pef f =

(77)
1
2

(f Rf )

, (78)

where ef f has to be non-negative in a spatially flat

FLRW spacetime, as follows from the inspection of
eq. (75) in the limit 0. Then, in vacuo eqs. (75)
and (76) can take the form of the standard Friedmann
equation

(74)

the current value of k is sensitive to the current value

of a(t), i.e. to the amount of expansion the universe has
undergone after the Big Bang. A significant amount of
expansion can easily drive k very close to zero. The
success of the inflationary paradigm is exactly that it
explains the flatness problem how did the universe
become so flat in a dynamical way, allowing us to
avoid fine-tuning the parameter k (the value k = 0 is
statistically exceptional).
The above having been said, choosing k = 0 for simplicity is not a dramatic departure from generality when
it comes to late time cosmology. If it is viewed as an
approximation and not as a choice of an initial condition, then one can say that, since k as inferred from
observations is very close to zero at current times, the
terms related to k will be subdominant in the Friedmann
or generalised Friedmann equations and, therefore, one
could choose to discard them by setting k = 0, without great loss of accuracy. In any case, results derived

Rf f
3H Rf

2f
f
+ Rf
+
R 2 f + 2H Rf

ef f ,
3
a

= [ef f + 3Pef f ] .
a
6

H2 =

(79)
(80)

Hence, in vacuo the curvature correction can be viewed

as an effective fluid.14
The effective equation of state parameter wef f of modified gravity can be expressed as
wef f

14

+ Rf
+ 1 (f Rf )
R 2 f + 2H Rf
Pef f
2
=
.

Rf f

ef f

3H
Rf
2
(81)

Note the following subtlety though: should we have included

matter it would enter the Friedmann equations with a modified
coupling /f . In general this effective fluid representation is
used only for demonstrative purposes and should not be overestimated or misinterpreted.

17
Since the denominator on the right hand side of eq. (81)
is strictly positive, the sign of wef f is determined by its
numerator. In general, for a metric f (R) model to mimic
the de Sitter equation of state wef f = 1, it must be

not acceptable because it leads to f (R)R = Ca0 eH0 t ,

which is absurd because the left hand side is timeindependent (for a de Sitter solution), while the right
hand side depends on time.

RH
f
=
.

f
(R)

One could impose energy conditions for the effective

stress-energy tensor (12) of f (R) gravity. However,
this is not very meaningful from the physical point of
view since it is well known that effective stress-energy
terms originating from the geometry by rewriting the
field equations of alternative gravities as effective Einstein equations do, in general, violate all the energy
conditions [e.g., (Faraoni, 2004a)].15 Also, the concept
of gravitational energy density is, anyway, ill-defined
in GR and in all metric theories of gravity as a consequence of the Equivalence Principle. Moreover, the
violation of the energy conditions makes it possible to
have H > 0 and bouncing universes (Carloni et al., 2006;
Novello and Bergliaffa, 2008).

(82)

Let us also give two simple examples that can be found

in the literature for demonstrative purposes and without
considering their viability: First, one can consider the
function f to be of the form f (R) Rn . It is quite
straightforward to calculate wef f as a function of n if
the scale factor is assumed to be a generic power law
a(t) = a0 (t/t0 ) (a general a(t) would lead to a time
dependent wef f ) (Capozziello et al., 2003). The result is
wef f =

6n2 7n 1
6n2 9n + 3

(83)

for n 6= 1, and is given is terms of n as

=

2n2 + 3n 1
.
n2

(84)

A suitable choice of n can lead to a desired value for

wef f . For instance, n = 2 yields wef f = 1 and = ,
as expected, considering that quadratic corrections to the
Einstein-Hilbert Lagrangian were used in the well known
Starobinsky inflation (Starobinsky, 1980).
The second example which we will refer to is a model
of the form f (R) = R2(n+1) /Rn , where is a suitably
chosen parameter (Carroll et al., 2004). In this case, and
once again if the scale factor is assumed to be a generic
power law, wef f can again be written as a function of n
(Carroll et al., 2004):
wef f = 1 +

2(n + 2)
.
3(2n + 1)(n + 1)

(85)

The most typical model within this class is that with

n = 1 (Carroll et al., 2004), in which case wef f = 2/3.
Note that in this class of models, a positive n implies
the presence of a term inversely proportional to R in the
action, contrary to the situation for the Rn models.
In terms of the quantity (R) f (R) one can rewrite
eq. (81) as




H
H
(86)
= 1+
wef f = 1+2
3H 2
R f 6H
and

The field equations are clearly of fourth order in a(t).

When matter is absent (a situation of interest in early
time inflation or in a very late universe completely dominated by f (R) corrections), a(t) only appears in the combination H a/a.

Since the Hubble parameter H is a

cosmological observable, it is convenient to adopt it as
the (only) dynamical variable; then the field equations
(75) and (76) are of third order in H. This elimination
of a is not possible when k 6= 0, or when a fluid with
density = (a) is included in the picture.
Regarding the dynamical field content of the theory, the fact that quadratic corrections to the
EinsteinHilbert action introduce a massive scalar field
was noted in (Buchbinder et al., 1992; Stelle, 1978,
1977; Strominger, 1984; Utiyama and DeWitt, 1962;
Vilkovisky, 1992); this applies to any f (R) gravity theory
in the metric formalism [see, e.g., (Ferraris et al., 1988;
Hindawi et al., 1996; Olmo, 2007)]. The metric tensor
contains, in principle, various degrees of freedom: spin 2
modes, and vector and scalar modes, which can all be
massless or massive. In GR we find only the massless
graviton but, when the action is allowed to include terms
that depend on R, R R , R R , other modes
show up. In f (R) gravity, a massive scalar mode appears,
which is evident in the equivalence with scalar-tensor theory (see Sec. III). As discussed in Sec. III.C, the scalar
field = R is dynamical in the metric formalism and
non-dynamical in the Palatini formalism.

15

ef f + Pef f

"
H
d
ln
=
=

dt

!#

(87)

An exact de Sitter solution corresponds to = f (R)R =

0, or to = Ca(t) = Ca0 eH0 t , where C 6= 0 is an integration constant. However, the second solution for (t) is

In (Santos et al., 2007), the Null Energy Condition and the

Strong Energy Condition for metric f (R) gravity have been derived by using the Raychaudhuri equation and imposing that
gravity be attractive, whereas for the Weak Energy Condition
and the Dominant Energy Condition an effective stress-energy
tensor which includes the matter was used. In (Perez Bergliaffa,
2006), a different approach was followed, in which the standard
energy conditions on matter where used in an attempt to constrain f (R) gravity.

18
2. Palatini f (R) gravity

As already mentioned, some concerns have been expressed on whether the homogeneity approximation can
justify the use of the FLRW metric as a cosmological solution in Palatini f (R) gravity (Flanagan, 2004b)
[see also (Li et al., 2008)]. Therefore, even though it is
standard practice in the literature to assume a FLRW
background and a perfect fluid description for matter when studying cosmology in Palatini f (R) gravity
[e.g. (Allemandi et al., 2004, 2005a; Amarzguioui et al.,
2006; Meng and Wang, 2003, 2004b, 2005; Sotiriou,
2006a,e; Vollick, 2003)], and we are going to review this
approach here, the reader should approach it with some
reasonable skepticism until this issue is clarified further.
Under the assumptions that the spacetime is indeed
described at cosmological scales by the FLRW metric,
eq. (72), that the stress-energy tensor of matter is that of
eq. (73), and that k = 0, easy manipulations reveal that
the field eqs. (18) and (19) yield the following modified
Friedmann equation [see for instance (Meng and Wang,
2003; Sotiriou, 2006e)]:
1 f
H+
2 f

!2

1 ( + 3P ) 1 f
+
,
6
f
6 f

(88)

where the overdot denotes differentiation with respect to

coordinate time. Note that when f is linear, f = 1
and, therefore, f = 0. Taking into account eq. (20), one
can easily show that in this case eq. (88) reduces to the
standard Friedmann equation.
We will avoid representing the extra terms in eq. (88)
with respect to the standard Friedmann equation as a
an effective stress energy density and pressure since, as
it is not that difficult to see, the former equation does
not really carry more dynamics than the latter. Indeed,
assume as usual that the cosmological fluid is composed
by pressureless dust (Pm = 0) and radiation (Pr = r /3)
and = m + p and P = Pm + Pr where m , p and Pm ,
Pr denote the energy density and the pressure of dust and
radiation, respectively. Due to eq. (20) and the fact that
for radiation T = 0, it is quite straightforward to derive
an algebraic relation between R and the energy density
of the dust. Combining this with energy conservation,
one obtains (Sotiriou, 2006e)
3H(Rf 2f )
R =
.
Rf f

(89)

This equation can be used to replace R in eq. (88), yielding

H2 =

1
2 + Rf f



(Rf 2f ) 2
6f
1 23 ff (Rf
f )

(90)

Considering now that, due to eq. (20), R is just an algebraic function of m , it is easy to realize that eq. (90) is

actually just the usual Friedmann equation with a modified source. The functional form of f will determine how
the dynamics will be affected by this modification.
It seems, therefore, quite intuitive that by tampering
with the function f one can affect the cosmological dynamics in a prescribed way. Indeed it has been shown
that for f (R) = R 2 /(3R) one approaches a de Sitter expansion as the density goes to zero (Vollick, 2003).
In order to match observations of the expansion history,
one needs to choose 1067 (eV)2 1053 m2 . Additionally, in regimes for which , eq. (90) reduces
to high precision to the standard Friedmann equation.
The above can very easily be verified by replacing this
particular choice of f in eq. (90). We refer the reader to
the literature for more details.
One could, of course, consider more general functions
of R. Of particular interest would be having positive powers of R higher than the first power added in
the action (since one could think of the Lagrangian as
a series expansion). Indeed this has been considered
(Meng and Wang, 2003, 2004b, 2005; Sotiriou, 2006a,e).
However, it can be shown that such terms do not really lead to interesting phenomenology as in metric f (R)
gravity: for instance they cannot drive inflation as, unlike in the scenario proposed by (Starobinsky, 1980) in
the metric formalism, here there are no extra dynamics and inflation cannot end gracefully (Meng and Wang,
2004b; Sotiriou, 2006a). As a matter of fact, it is more
likely that positive powers of R will lead to no interesting
cosmological phenomenology unless their coefficients are
large enough to make the models non-viable (Sotiriou,
2006a).

B. Cosmological eras

As stated in the Introduction, the recent flurry of

theoretical activity on f (R) models derives from the
need to explain the present acceleration of the universe discovered with supernovae of type Ia (Astier et al.,
2006; Barris et al., 2004; Filippenko and Riess, 1998;
Knop et al., 2003; Perlmutter et al., 1998; Riess et al.,
1998, 1999, 2004; Schmidt et al., 1998; Tonry et al.,
2003). We have seen in the previous section how f (R)
gravity can achieve cosmic acceleration and an effective
equation of state parameter wef f 1; on the other
hand, it was already known from R2 -inflationary scenarios of the early universe that this is possible, so we are
actually witnessing a resurrection of this theoretical possibility in models of the late universe this parallels
the use of scalar fields to drive early inflation or latetime acceleration in quintessence models. There are also
attempts to unify early inflation and late time acceleration in modified gravity (Bamba and Odintsov, 2008;
Nojiri and Odintsov, 2007d, 2008a,b,c,d). However, any
model attempting to explain the cosmic speed-up at late
times should not spoil the successes of the standard cosmological model which requires a definite sequence of eras

19
to follow each other, including:
1. early inflation
2. a radiation era during which Big Bang Nucleosynthesis occurs;
3. a matter era;
4. the present accelerated epoch, and
5. a future era.
Big Bang Nucleosynthesis is well constrained
see
(Brookfield et al.,
2006;
Clifton and Barrow,
2005a; Evans et al., 2007; Kneller and Steigman,
2004; Lambiase and Scarpetta, 2006; Nakamura et al.,
2006) for such constraints on f (R) models. The matter
era must last long enough to allow the primordial
density perturbations generated during inflation to grow
and become the structures observed in thd universe
today. The future era is usually found to be a de Sitter
attractor solution, or to be truncated at a finite time by
a Big Rip singularity.
Furthermore, there must be smooth transitions between consecutive eras, which may not happen in all
f (R) models. In particular, the exit from the radiation era has been studied and claimed to originate problems for many forms of f (R) in the metric formalism, including f = R 2(n+1) /Rn , n >
0 (Amendola et al., 2007a,b; Brookfield et al., 2006;
Capozziello et al., 2006c; Nojiri and Odintsov, 2006)
[but not in the Palatini formalism (Carvalho et al., 2008;
Fay et al., 2007b)]. However, the usual model f (R) =
R 4 /R with bad behaviour was studied using singular perturbation methods (Evans et al., 2007), definitely
finding a matter era which is also sufficiently long.
Moreover, one can always find choices of the function f (R) with the correct cosmological dynamics in
the following way: one can prescribe the desired
form of the scale factor a(t) and integrate a differential equation for f (R) that produces the desired scale factor (Capozziello et al., 2005b, 2006c;
de la Cruz-Dombriz and Dobado, 2006; Faulkner et al.,
2007; Fay et al., 2007a,b; Hu and Sawicki, 2007a,b;
Multamaki and Vilja, 2006a; Nojiri and Odintsov, 2006,
2007b,c; Song et al., 2007). In general, this designer
f (R) gravity produces forms of the function f (R) that
are rather contrived. Moreover, the prescribed evolution
of the scale factor a(t) does not determine uniquely the
form of f (R) but, at best, only a class of f (R) models (Multamaki and Vilja, 2006a; Sokolowski, 2007b,c;
Starobinsky, 2007). Therefore, the observational data
providing information on the history of a(t) are not sufficient to reconstruct f (R): one needs additional information, which may come from cosmological density perturbations. There remains a caveat on being careful to
terminate the radiation era and allowing a matter era
that is sufficiently long for scalar perturbations to grow.

While sometimes it is possible to find exact solutions

to the cosmological equations, the general behaviour of
the solutions can only be assessed with a phase space
analysis, which constitutes a powerful tool in cosmology (Coley, 2003; Wainwright and Ellis, 1997). In a
spatially flat FLRW universe the dynamical variable is
the Hubble parameter H, and a convenient choice of
phase space variables in this case is (H, R). Then, for
any form of the function f (R), the phase space is a
two-dimensional curved
embedded in the three manifold

dimensional space H, R, R with de Sitter spaces as

fixed points (de Souza and Faraoni, 2007); the structure

of the phase space is simplified with respect to that of
general scalar-tensor cosmology (Faraoni, 2005b).
Studies of the phase space of f (R) cosmology
(not limited to the spatially flat FLRW case) were
common in the pre-1998 literature on R2 -inflation
(Amendola et al.,
1992;
Capozziello et al.,
1993;
Muller et al., 1990; Starobinsky, 1980). The presence or
absence of chaos in metric f (R) gravity was studied in
(Barrow and Cotsakis, 1989, 1991). Such studies with
dynamical system methods have become widespread
with the recent surge of interest in f (R) gravity to
explain the present cosmic acceleration. Of course, detailed phase space analyses are only possible for specific
choices of the function f (R) (Abdelwahab et al., 2008;
Amendola et al., 2007a,b,c; Amendola and Tsujikawa,
2008; Carloni et al., 2008a; Carloni and Dunsby,
2007; Carloni et al., 2005, 2007; Carroll et al., 2005;
Clifton, 2006a, 2007; Clifton and Barrow, 2005b;
Easson, 2004; Fay et al., 2007a,b; Goheer et al., 2007,
2008; Leach et al., 2006, 2007; Li and Barrow, 2007;
Nojiri and Odintsov, 2004b; Sami et al., 2005).

C. Dynamics of cosmological perturbations

Obtaining the correct dynamics of the background

cosmological model is not sufficient for the theory to
be viable: in fact, the FLRW metric can be obtained
as a solution of the field equations of most gravitation
theories, and it is practically impossible to discriminate between f (R) gravity and dark energy theories
(or between different f (R) models) by using only
the unperturbed FLRW cosmological model, i.e., by
using only probes that are sensitive to the expansion
history of the universe. By contrast, the growth of
cosmological perturbations is sensitive to the theory of
gravity adopted and constitutes a possible avenue to
discriminate between dark energy and modified gravity.
Changing the theory of gravity affects the dynamics of
cosmological perturbations and, among other things,
the imprints that these leave in the cosmic microwave
background (which currently provide the most sensitive
cosmological probe) and in galaxy surveys (Knox et al.,
2006; Koivisto, 2006b; Koyama and Maartens, 2006;
Li and Barrow, 2007; Li and Chu, 2006; Sealfon et al.,
2005; Shirata et al., 2005, 2007; Skordis et al.,

20
2006; Song et al., 2007; Stabenau and Jain, 2006;
Tsujikawa,
2007;
White and Kochanek,
2001;
Zhang et al., 2007).
This originated various efforts
to constrain f (R) gravity with cosmic microwave
background data (Amendola and Tsujikawa, 2008;
Appleby and Battye, 2007; Carloni et al., 2008b;
Hu and Sawicki, 2007b; Li and Barrow, 2007; Li et al.,
2007a; Pogosian and Silvestri, 2008; Starobinsky, 2007;
Tsujikawa, 2008; Tsujikawa et al., 2008; Wei and Zhang,
2008).
Most of these works are restricted to specific choices
of the function f (R), but a few general results have
also been obtained. The growth and evolution of local scalar perturbations, which depends on the theory of
gravity employed, was studied in metric f (R) gravity theories which reproduce GR at high curvatures in various
papers (Carroll et al., 2006; de la Cruz-Dombriz et al.,
2008; Song et al., 2007) by assuming a scale factor evolution typical of a CDM model. Vector and tensor
modes are unaffected by f (R) corrections. It is found
that f (R) > 0 is required for the stability of scalar perturbations (Song et al., 2007), which matches the analysis of Sec. (V.B.2) in a locally de Sitter background.
The corrections to the EinsteinHilbert action produce
qualitative differences with respect to Einstein gravity:
they lower the large angle anisotropy of the cosmic microwave background and may help explaining the observed low quadrupole; and they produce different correlations between the cosmic microwave background and
galaxy surveys (Song et al., 2007). Further studies challenge the viability of f (R) gravity in comparison with the
CDM model: in (Bean et al., 2007) it is found that large
scale density fluctuations are suppressed in comparison
to small scales by an amount incompatible with the observational data. This makes it impossible to fit simultaneously large scale data from the cosmic microwave background and small scale data from galaxy surveys. Also,
a quasi-static approximation used in a previous analysis
(Zhang, 2007) is found to be invalid.
In (de la Cruz-Dombriz et al., 2008), the growth of
matter density perturbations is studied in the longitudinal gauge using a fourth order equation for the density contrast /, which reduces to a second order one
for sub-horizon modes. The quasi-static approximation,
which does not hold for general forms of the function
f (R), is however found to be valid for those forms of this
function that describe successfully the present cosmic acceleration and pass the Solar System tests in the weakfield limit. It is interesting that the relation between the
gravitational potentials in the metric which are responsible for gravitational lensing, and the matter overdensities
depends on the theory of gravity; a study of this relation
in f (R) gravity (as well as in other gravitational theories)
is contained in (Zhang et al., 2007).
Cosmological density perturbations in the Palatini
formalism have been studied in (Amarzguioui et al.,
2006; Carroll et al., 2006; Koivisto, 2006b, 2007;
Koivisto and Kurki-Suonio, 2006; Lee, 2007, 2008;

Li et al., 2007b; Uddin et al., 2007).

Two different formalisms developed in (Hwang and Noh, 2002;
Koivisto and Kurki-Suonio, 2006) and (Lue et al., 2004)
were compared for the model f (R) = R 2(n+1) /Rn
and it was found that the two models agree for scenarios
that are close (in parameter space) to the standard concordance model, but give different results for models that
differ significantly from the CDM model. Although this
is not something to worry about in practice (all models
aiming at explaining the observational data are close
to the standard concordance model), it signals the need
to test the validity of perturbation analyses for theories
that do differ significantly from GR in some aspects.
V. OTHER STANDARD VIABILITY CRITERIA

In addition to having the correct cosmological dynamics and the correct evolution of cosmological perturbations, the following criteria must be satisfied in order for
an f (R) model to be theoretically consistent and compatible with experiment. The model must:
have the correct weak-field limit at both the Newtonian and post-Newtonian levels, i.e., one that is
compatible with the available Solar System experiments;
be stable at the classical and semiclassical level (the
checks performed include the study of a matter instability, of gravitational instabilities for de Sitter
space, and of a semiclassical instability with respect
to black hole nucleation);
not contain ghost fields;
admit a well-posed Cauchy problem;
These independent requirements are discussed separately in the following.
A. Weak-field limit

It is obvious that a viable theory of gravity must have

the correct Newtonian and post-Newtonian limits. Indeed, since the modified gravity theories of current interest are explicitly designed to fit the cosmological observations, Solar System tests are more stringent than the
cosmological ones and constitute a real testbed for these
theories.
1. The scalar degree of freedom

It is clear from the equivalence between f (R) and

BransDicke gravities discussed in Sec. III that the former contains a massive scalar field [see eqs. (54) and
(66)]. While in the metric formalism this scalar is dynamical and represents a genuine degree of freedom, it

21
is non-dynamical in the Palatini case. Let us, therefore,
consider the role of the scalar field in the metric formalism as it will turn out to be crucial for the weak-field
limit. Using the notations of Sec. III.A, the action is
given by eq. (54) and the corresponding field equations
by eq. (55).
Equation (52) for has no dynamical content because
it only enforces the equality = R. However, = R is
indeed a dynamical field that satisfies the wave equation,
3f () + 3f ()
+ f () 2f () = T.

(91)

When f 6= 0 a new effective potential W () 6= V () can

be introduced, such that
dW
T f () + 2f ()
=
.
d
3f ()

(92)

The action can be seen as a BransDicke action with

0 = 0 if the field f () = f (R) is used instead of
as the independent BransDicke field:
Z

1
(93)
S=
d4 x g [R V ()] + S (m) ,
2
where V () is given by eq. (53).
Now one may think of studying the dynamics and stability of the model by looking at the shape and extrema
of the effective potential V () but this would be misleading because the dynamics of are not regulated by
V () (indeed, the wave equation (91) does not contain
V ), but are subject the strong constraint = R, and R
(or f (R)) is ruled by the trace equation (8).
The following example shows how the use of the potential V () can be misleading. As is well known, the effective mass of a scalar field (corresponding to the second
derivative of the potential evaluated at the minimum)
controls the range of the force mediated by this field.
Thus, when studying the weak-field limit of the theory it
is important to know the range of the dynamical scalar
field = R present in the metric formalism in addition
to the metric field g , as this field can potentially violate the post-Newtonian constraints obtained from Solar
System experiments if the scalar field gives observable effects at the relevant scales. One way to avoid Solar System constraints, however, is to have have a sufficiently
short range (see Sec. V.A.2 for more details). Consider
the example f (R) = R + aR2 , with a a positive constant.
By naively taking the potential, one obtains
V () = a 2

m21 2

true effective mass is obtained from the trace equation (8)

ruling the evolution of R which, for f (R) = R + aR2 , reduces to
R

R
T
=
,
6a
6a

(95)

and the identification of the mass squared of = R as

m2 =

1
6a

(96)

is straightforward.16 A small enough value of a now leads

to a large value of m and a short range for 17 . The situation is, however, more complicated; the chameleon effect
due to the dependence of the effective mass on the curvature may change the range of the scalar (Faulkner et al.,
2007; Starobinsky, 2007).
For a general f (R) model, the effective mass squared of
= R is obtained in the weak-field limit by considering
a small, spherically symmetric, perturbation of de Sitter
space with constant curvature R0 . One finds


1 f0
m2 =
.
(97)

R
0
3 f0
This equation coincides with eq. (6) of (Muller et al.,
1990), with eq. (26) of (Olmo, 2007), and with eq. (17) of
(Navarro and Van Acoleyen, 2007). It also appears in a
calculation of the propagator for f (R) gravity in a locally
flat background (eq. (8) of (Nunez and Solganik, 2004)).
The same expression is recovered in a gauge-invariant
stability analysis of de Sitter space (Faraoni and Nadeau,
2005) reported in Sec. V.B.2 below.
Another possibility is to consider the field f (R)
instead of = R, and to define the effective mass of
by using the Einstein frame scalar-tensor analog of
f (R) gravity instead of its Jordan frame cousin already
discussed (Chiba, 2003). By performing the conformal
transformation
g g = f (R) g g

(98)

and the scalar field redefinition = f (R) with

r
20 + 3 d
d =
,
(99)
2

a scalar-tensor theory is mapped to the Einstein frame

in which the new scalar field couples minimally to

(94)

with effective mass squared m21 = 2a. Then, the small

values of a generated by quantum corrections to GR imply a small mass m1 and a long range field might
be detectable at Solar System scales (Chiba et al., 2007;
Jin et al., 2006; Olmo, 2007). However, this conclusion is
incorrect because m1 is not the physical mass of . The

16

17

It was already noted by (Stelle, 1978) that an R2 correction to

the EinsteinHilbert Lagrangian generates a Yukawa correction
to the Newtonian potential this has to be kept small at macroscopic scales by giving it a short range.
The deflection of light by the Sun in GR plus quadratic corrections was studied by calculating the Feynman amplitudes for
photon scattering, and it was found that, to linearized order, this
deflection is the same as in GR (Accioly et al., 1999).

22
the Ricci curvature and has canonical kinetic energy, as
described by the action


Z
p
1
(g)
4

g R U () +
S
=
d x
2

(100)
+SM (e 2/3 g , ),
(note once more the non-minimal coupling of the matter
in the Einstein frame). For the 0 = 0 equivalent of
metric f (R) gravity we have
2
f (R) = e 3 ,
(101)

Rf
(R)

f
(R)
=
U ()
(102)
2 ,
2 (f (R))
q
3 f
By using d/d

where R = R().
= 2
f , the effective mass of is defined by
m
2ef f

#
"
d2 U

4f
1 1

+
=
2
3 f
f
d2
(f )

(103)

[this equation appears in the footnote on p. 2 of (Chiba,

2003)]. By assuming a de Sitter background with constant curvature R0 = 12H02 = f0 /(6f0 ), this turns into


m2ef f
1
f0
m
2ef f =
=

R
.
(104)
0
3f0 f0
f0
In the Einstein frame, it is not the mass m
of a particle or
a field that is measurable, but rather the ratio m/
m
u between m
and the Einstein frame unit of mass m
u , which
1/2
is varying, scaling as m
u = [f (R)]
mu = 1/2 mu ,
where mu is the constant unit of mass in the Jordan frame
(Dicke, 1962; Faraoni et al., 1999; Faraoni and Nadeau,
2007). Therefore,
m
2ef f
m2ef f
=
.
m
2u
m2u

(105)

In practice, f (R) is dimensionless and its value must

be of order unity in order to obtain the gravitational coupling strength measured in the Solar System; as a result,
the Einstein frame metric g and the Jordan frame metric g are almost equal, and the same applies to m
u , mu
and to m
ef f , mef f , respectively. Then, the only relevant difference between Einstein and Jordan frames is

the scalar field redefinition .

2. Weak-field limit in the metric formalism

Having discussed the field content of the theory, we

are now ready to discuss the weak-field limit. Having
the correct weak-field limit at the Newtonian and postNewtonian levels is essential for theoretical viability.
From the beginning, works on the weak-field
(Newtonian and post-Newtonian) limit of f (R)

gravity led to opposite results appearing in the

literature
(Accioly et al.,
1999;
Baghram et al.,
2007; Barrow and Clifton, 2006; Capozziello et al.,
2006d,
2007d,e;
Capozziello and Troisi,
2005;
Capozziello and Tsujikawa,
2007;
Cembranos,
2006;
Clifton and Barrow,
2005a,
2006;
Dick,
2004;
Easson,
2004;
Hu and Sawicki,
2007b;
Iorio, 2007; Multamaki and Vilja, 2006b, 2007a;
Navarro and Van Acoleyen, 2005, 2006; Olmo, 2005a,b;
Rajaraman, 2003; Ruggiero and Iorio, 2007; Shao et al.,
2006; Soussa and Woodard, 2004; Zhang, 2007). Moreover, a certain lack of rigour in checking the convergence
of series used in the expansion around a de Sitter
background often left doubts even on results that, a
posteriori, turned out to be correct (Sotiriou, 2006c).

By using the equivalence between f (R) and scalartensor gravity, Chiba originally suggested that all f (R)
theories are ruled out (Chiba, 2003). This claim was
based on the fact that metric f (R) gravity is equivalent
to an 0 = 0 BransDicke theory, while the observational
constraint is |0 | > 40000 (Bertotti et al., 2003). This
is not quite the case and the weak-field limit is more
subtle than it appears, as the discussion of the previous section might have already revealed: The value of
the parametrized post-Newtonian (PPN) parameter ,
on which the observational bounds are directly applicable, is practically independent of the mass of the scalar
only when the latter is small (Wagoner, 1970). In this
case, the constraints on can indeed be turned into constraints on 0 . However, if the mass of this scalar is
large, it dominates over 0 in the expression of and
drives its value to unity. The physical explanation of
this fact, as mentioned previously, is that the scalar becomes short-ranged and, therefore, has no effect at Solar
System scales. Additionally, there is even the possibility
that the effective mass of the scalar field itself is actually
scale-dependent. In this case, the scalar may acquire a
large effective mass at terrestrial and Solar System scales,
shielding it from experiments performed there while being effectively light at cosmological scales. This is the
chameleon mechanism, well-known in quintessence models (Khoury and Weltman, 2004a,b).

Given the above, it is worth examining these issues in

more detail. Even though early doubts about the validity
of the dynamical equivalence with scalar-tensor theory in
the Newtonian limit (Faraoni, 2006b; Kainulainen et al.,
2007a) have now been dissipated (Faraoni, 2007b), a direct approach which does not resort to the scalar-tensor
equivalence is preferable as the former could in principle
hide things (Olmo, 2005b). This was given in the metric formalism, first in the special case (Erickcek et al.,
2006) f (R) = R 4 /R [which is already ruled out
by the Ricci scalar instability (Dolgov and Kawasaki,
2003a; Faraoni, 2006a)] and in the case f (R) = Rn us-

23
ing light deflection and other Solar System experiments18
(Barrow and Clifton, 2006; Clifton and Barrow, 2005a,
2006; Zakharov et al., 2006). Only later was the case
of a general function f (R) discussed (Chiba et al., 2007;
Jin et al., 2006; Olmo, 2007). Chibas result based on the
scalar-tensor equivalence eventually turns out to be valid
subject to certain assumptions which are not always satisfied (Chiba et al., 2007; Jin et al., 2006; Olmo, 2007)
see below. This method, however, does not apply to
the Palatini version of f (R) gravity.
In what follows we adhere to, but streamline, the discussion of (Chiba et al., 2007) with minor modifications,
in order to compute the PPN parameter for metric f (R)
gravity [see also (Olmo, 2007)]. We consider a spherically symmetric, static, non-compact body embedded in
a background de Sitter universe; the latter can exist in
an adiabatic approximation in which the evolution of the
universe is very slow in comparison with local dynamics.
The condition for the existence of a de Sitter space with
R = R0 g /4 and constant curvature R0 = 12H02 is
s
f0

.
(106)
f0 R0 2f0 = 0,
H0 =
6f0
The line element is


ds2 = 1 + 2(r) H02 r2 dt2


+ 1 + 2(r) + H02 r2 dr2 + r2 d2 (107)

in Schwarzschild coordinates, where the post-Newtonian

potentials (r) and (r) are treated as small perturbations.19 The goal is to compute the PPN parameter
= / by solving the equations satisfied by these
potentials. A linearized analysis is performed assuming
|(r)| , |(r)| 1,

r << H01 ,

(109)

where the deviation R1 (r) of the Ricci curvature from

the constant R0 is also a small perturbation.20
Three assumptions are made:

18

19

20

3f0 R1 + (f0 R0 f0 ) R1 = T1 ,

The perihelion precession in modified gravity is studied in

(Baghram et al., 2007; Iorio, 2007; Iorio and Ruggiero, 2007a,b;
Schmidt, 2008).
Isotropic coordinates are usually employed in the study of the
weak-field limit of spherically symmetric metrics; however, the
difference is irrelevant to first order in and (Olmo, 2007).
The solution derived for the spherically symmetric metric is only
valid when mr << 1, where m is the effective mass of the scalar.
If this assumption is not made, then (for example, according
to (Jin et al., 2006)), it would seem that quantum corrections
in f (R) = R + aR2 with a 1024 GeV2 are ruled out by
Solar System constraints, which is not the case because these
corrections are equivalent to a massive scalar field with short
range that is not constrained by the available data.

(110)

where T = T1 , since T is zero in the background. For

a static, spherically symmetric

body, R1 = R1 (r) and
d
1
r2 dR
R1 = 2 R1 = r12 dr
dr . The reduced trace equation (110) then becomes
2 R1 m2 R1 =

,
3f0

(111)

where
m2 =

f0 f0 R0
.
3f0

(112)

By using R0 = 12H02 = 2f0 /f0 , this reduces to

2

(108)

and
R(r) = R0 + R1 (r),

Assumption 1: f (R) is analytical at R0 .

Assumption 2: mr << 1, where m is the effective mass
of the scalar degree of freedom of the theory. In other
words, this scalar field (the Ricci curvature, which is an
extra dynamical quantity in the metric formalism) must
have a range longer than the size of the Solar System
if it is much shorter than, say, 0.2 mm (Hoyle et al.,
2001), the presence of this scalar is effectively hidden
from Solar System and terrestrial experiments. In this
case, this field could not have cosmological effects at late
times, but could only be important in the very early universe at high curvatures, e.g., in Starobinsky-like inflation (Starobinsky, 1980).
Assumption 3: the pressure P 0 for the energymomentum of the local star-like object. The trace of
the corresponding energy-momentum tensor reduces to
T1 .
By expanding f (R) and f (R) around R0 , the trace
equation (153) reduces to

m2 =

(f0 ) 2f0 f0
.
3f0 f0

(113)

This equation is found in various other treatments of perturbations of de Sitter space (Faraoni and Nadeau, 2005;
Navarro and Van Acoleyen, 2007; Nunez and Solganik,
2004; Olmo, 2007).
Assumption 2 that the scalar R1 is light, which enables
the f (R) theory to produce significant cosmological effects at late times, also allows one to neglect21 the term
m2 R1 in eq. (111). The Green function of the equation

1
2 R1 = 3f
is then G(r) = 4r and the solution is
0
R

)
R1 d3 ~x 3f(r
G(r r ), which yields

R1

21

M
12f0 r

(mr << 1) .

(114)

Although (Chiba et al., 2007) provide Green functions in both

cases m2 > 0 and m2 < 0, the latter corresponds to a spacetime
instability and is unphysical. This is is irrelevant in the end
because only the case m2 0 is necessary and used in the
calculation (Faraoni and Lanahan-Tremblay, 2007).

24
Now, the condition m2 r2 << 1 yields



1  f0
R0  r2 << 1


3 f0
and, using H0 r << 1,
 
 f0  2
  r << 1.
 f 
0

(115)

f
f0 R11 3H02 0 R1 f0 1 1 R1
2
+ f0 R1 R11 + f0 R1 = T11 (125)
(116)

Let us use now the full field equations (6); by expanding f (R) and f (R) and using f0 = 6H02 f0 we get

f
f0 R1 + f0 R 3H02 0 R1
2
f0 R1 + f0 R1 R = T . (117)

By using again the assumption H0 r << 1, the

dAlembertian  becomes 2 and, for (, ) = (0, 0),

f
f0 R00 3H02 0 R1 + f0 R1 R00 + f0 2 R1 = .
2
(118)
By computing R00 = 3H02 2 (r) and dropping terms
f0 H02 R1 << f0 2 , etc., we obtain
f0 2 (r) +

f0
R1 f0 2 R1 = .
2

(119)

Recalling that 2 R1 3f
for mr << 1, one obtains
0

f0 2 (r) =

2 f0

R1 .
3
2

(120)

Eq. (120) can be integrated from r = 0 to r > r0 (where

r0 is the radius of the star-like object) to obtain, using
Gauss law,
d
C1

M
2,
=

2
dr
6f0 48f0 r
r

(121)

Rr
where M (r) = 4 0 0 dr (r )2 (r ). The integration constant C1 must be set to zero to guarantee regularity of
the Newtonian potential at r = 0. The potential (r)
then becomes
(r) =

M
M
r.

6f0 r
48f0

(122)

The second term on the right hand side is negligible; in

fact,
 Mr  



 48f0   f0  2
(123)
 M  =   r << 1,
 6f r 
8f0
0

and

(r)

M
.
6f0 r

Let us now find the second potential (r) appearing in

the line element (107). By using the field equations (6)
with (a, b) = (1, 1),

(124)

with T11 0 outside the star, and

d2 2 d
+
,
dr2
r dr
d2 R1
,
g 11 1 1 R1
dr2

R11 3H02

(126)
(127)

and neglecting higher order terms, one obtains (eq. (22)

of (Chiba et al., 2007))
 2

d 2 d
f R1
2f dR1

f0 2 +
0
+ 0
0. (128)
dr
r dr
2
r dr
Now, using eq. (114) for R1 , one concludes that the third
term in eq. (128) is negligible in comparison with the
fourth term. In fact,


 f0 R1   f 


(129)
 2f 2dR   0  r2 << 1.
 0 1
f0
r

dr

Then, using again the expression (114) for dR1 /dr and
eq. (124) for (r), one obtains
d
M
,
=
dr
12f0 r

(130)

which is immediately integrated to

(r) =

M
.
12f0 r

(131)

The post-Newtonian metric (107) therefore gives the

PPN parameter as
=

(r)
1
= .
(r)
2

(132)

This is a gross violation of the experimental bound

| 1| < 2.3105 (Bertotti et al., 2003) and agrees with
0 +1
found
the calculation of the PPN parameter =
0 +2
by using the equivalence of metric f (R) gravity with an
0 = 0 BransDicke theory (Chiba, 2003).
The results of (Chiba et al., 2007) have been reproduced by (Olmo, 2007), who works in isotropic
coordinates with a slightly different approach.
(Kainulainen et al., 2007a) have obtained spherically
symmetric interior solutions matched to the exterior
solutions of metric f (R) gravity and have confirmed the
result = 1/2.
Limits of validity of the previous analysis: One can
contemplate various circumstances in which the assumptions above are not satisfied and the previous analysis

25
breaks down. It is important to ascertain whether these
are physically relevant situations. There are three main
cases to consider.
The case of non-analytic f (R): While (Chiba et al.,
2007) consider functions f (R) that are analytic at the
background value R0 of the Ricci curvature, the situation in which this function is not analytical has been
contemplated briefly in (Jin et al., 2006). Assuming that
f (R) has an isolated singularity at R = Rs , it can be
expressed as the sum of a Laurent series,
f (R) =

+
X

n=0

an (R Rs ) .

(133)

(Jin et al., 2006) note that it must be R 6= Rs in the

dynamics of the universe because a constant curvature
space with R = Rs can not be a solution of the field
equations. Therefore, one can approximate the solution
adiabatically with a de Sitter space with constant curvature R0 6= Rs . The function f (R) is analytical here
and the previous discussion applies. This is not possible if f (R) has anessential
singularity, for example, if

2

(Jin et al., 2006). There is, of

f (R) = R 2 sin R
course, no reason other than Occams razor to exclude
this possibility.
Short range scalar field: If the assumption mr << 1
is not satisfied, the scalar is massive. If its range is sufficiently short, it is effectively hidden from experiments
probing deviations from Newtons law and from other
Newtonian and post-Newtonian experiments in the solar
neighbourhood. This is the case of quadratic quantum
corrections to Einsteins gravity, e.g., f (R) = R + R2 .
If the effective mass is m 103 eV (corresponding to a
fifth force range less than 0.2 mm, the shortest scale
currently accessible to weak-field experiments), this correction is undetectable and yet it can still have large
effects in the early inflationary universe (Starobinsky,
1980). However, it can not work as a model for late
time acceleration.
Chameleon behaviour:
The chameleon effect
(Khoury and Weltman, 2004a,b), originally discovered in scalar field models of dark energy, consists of the
effective mass m of the scalar degree of freedom being
a function of the curvature (or, better, of the energy
density of the local environment), so that m can be large
at Solar System and terrestrial curvatures and densities,
and small at cosmological curvatures and densities
effectively, it is short-ranged in the Solar System and
it becomes long-ranged at cosmological densities thus
causing the acceleration of the universe. The chameleon
effect can be applied to metric f (R) gravity (Cembranos,
2006; Faulkner et al., 2007; Navarro and Van Acoleyen,
2007; Starobinsky, 2007), with the result that theories
of the kind (Amendola et al., 2007a,b,c; Carroll et al.,
2004)
 n
R
2
f (R) = R (1 n)
(134)
2

are compatible with the observations in the region of

the parameter space 0 < n 0.25 with sufficiently
small (Faulkner et al., 2007). Precisely, using the Cassini
bound on the PPN parameter (Bertotti et al., 2003),
the constraint
1
 2(1n)


65n

2
10 2(1n)
3
H0
n(1 n)

(135)

is obtained (Faulkner et al., 2007). Fifth force experiments give the bounds
1

 2(1n)

212n
2

10 1n .
1n
H0
n(1 n)

(136)

Preferred values seem to be m 1050 eV 1017 H0

(Faulkner et al., 2007). Note that n > 0, which guarantees f > 0, is required for Ricci scalar stability (n = 0
reduces the model to GR with a cosmological constant,
but avoiding the latter was exactly the reason why dark
energy and modified gravity were introduced in the first
place).
These models work to explain the current cosmic acceleration because, for small curvatures R, the correction in Rn with n < 1 is larger than the Einstein
Hilbert term R and comes to dominate the dynamics. On the negative side, these theories are observationally indistinguishable from a cosmological constant and they have been dubbed vanilla f (R) gravity
(Amendola et al., 2007a,b,c; Amendola and Tsujikawa,
2008; Faulkner et al., 2007). However, they still have the
advantage of avoiding a fine-tuning problem in at the
price of a much smaller fine-tuning of the parameter .
As for all modified gravity and dark energy models, they
do not address the cosmological constant problem.
The weak-field limit of metric f (R) theories which admit a global Minkowski solution around which to linearize, was studied by (Clifton, 2008). These theories (inP+
cluding, e.g., analytic functions f (R) = n=1 an Rn ) are
not motivated by late time cosmology and the Minkowski
global solution, although present, may not be stable
(Clifton and Barrow, 2005a), which in practice detracts
from the usefulness of this analysis. Several new postNewtonian potentials are found to appear in addition to
the two usual ones (Clifton, 2008).

3. Weak-field limit in the Palatini formalism

Early works on the weak-field limit of Palatini f (R) gravity often led to contradictory results and to several technical problems as well
(Allemandi et al.,
2005b;
Allemandi and Ruggiero,
2007; Barraco and Hamity, 2000; Bustelo and Barraco,
2007; Dominguez and Barraco, 2004; Kainulainen et al.,
2007b; Meng and Wang, 2004a; Olmo, 2005a,b, 2007;
Ruggiero, 2007; Ruggiero and Iorio, 2007; Sotiriou,
2006c) which seem to have been clarified by now.

26
First of all, there seems to have been some confusion
in the literature about the fact that Palatini f (R) gravity reduces to GR with a cosmological constant in vacuum and the consequences that this can have on the
weak-field limit and Solar System tests. It is, of course,
true (see Sec. II.B) that in vacuo Palatini f (R) gravity will have the same solutions of GR plus a cosmological constant and, therefore, the Schwarzschild-(anti)de Sitter solution will be the unique vacuum spherically
symmetric solution (see also Sec. VI.C.1 for a discussion
of the Jebsen-Birkhoff theorem). This was interpreted
in (Allemandi and Ruggiero, 2007; Ruggiero and Iorio,
2007) as an indication that the only parameter that can
be constrained is the effective cosmological constant and,
therefore, models that are cosmologically interesting (for
which this parameter is very small) trivially satisfy Solar
System tests. However, even if one sets aside the fact that
a weak gravity regime is possible inside matter as well,
such claims cannot be correct: they would completely
defeat the purpose of performing a parametrized postNewtonian expansion for any theory for which one can
establish uniqueness of a spherically symmetric solution,
as in this case we would be able to judge Solar System
viability just by considering this vacuum solution (which
would be much simpler).
Indeed, the existence of a spherically symmetric vacuum solution, irrespective of its uniqueness, does not suffice to guarantee a good Newtonian limit. For instance,
the Schwarzschild-de Sitter solution has two free parameters; one of them can be associated with the effective
cosmological constant in a straightforward manner (using
the asymptotics). However, it is not clear how the second parameter, which in GR is identified with the mass
of the object in the Newtonian regime, is related to the
internal structure of the object in Palatini f (R) gravity.
The assumption that it represents the mass defined in
the usual way is not, of course, sufficient. One would
have to actually match the exterior solution to a solution
describing the interior of the Sun within the realm of
the theory in order to express the undetermined parameter in the exterior solution in terms of known physical
quantities, such as Newtons constant and the Newtonian
mass. The essence of the derivation of the Newtonian
limit of the theory consists also in deriving such an explicit relation for this quantity and showing that it agrees
with the Newtonian expression. The parametrized postNewtonian expansion is nothing but an alternative way
to do that without having to solve the full field equations. Therefore, it is clear that more information than
the form of the vacuum solution is needed in order to
check whether the theory can satisfy the Solar System
constraints.
However, some early attempts towards a Newtonian and post-Newtonian expansion were also flawed.
In (Meng and Wang, 2004a) and (Barraco and Hamity,
2000) for instance, a series expansion around a de Sitter
background was performed in order to derive the Newto-

nian limit. Writing

R = R0 + R1 ,

(137)

where R0 is the Ricci curvature of the background and

R1 is a correction, one is tempted to expand in powers
of R1 /R0 regarding the latter as a small quantity. Since
one needs the quantities f (R0 + R1 ) and f (R0 + R1 ),
the usual approach is to Taylor-expand around R = R0
and keep only the leading order terms in R1 . However, it
has been shown in (Sotiriou, 2006c) that this can not be
done for most cosmologically interesting models because
R1 /R0 is not small.
Take as an example the model (Vollick, 2003)
f (R) = R

22
,
R

(138)

and 2 1067 (eV)2 1053 m2 . Expanding as

1
f (R) = f (R0 ) + f (R0 )R1 + f (R0 )R21 + . . . (139)
2
and using eq. (138) yields


22
1 222 2
f (R) = f (R0 ) + 1 + 2 R1
R + . . . , (140)
R0
2 R30 1
where now R0 = 2 . It is then easy to see that the second
term on the right hand side is of the order of R1 , whereas
the third term is of the order of R21 /2 . Therefore, in
order to truncate before the third term, one needs R1
R21 /2 or
2 R1 .

(141)

This is not a stringent constraint: R0 2 and so this is

the usual condition for linearization.
Let us return now to the trace equation (20). For the
model under consideration,


q
1
T 2 T 2 + 1222 .
(142)
R=
2
According to eq. (142), the value of R, and consequently, R1 , is algebraically related to T and, whether
or not the condition (141) is satisfied or not critically
depends on the value of the energy density. To demonstrate this, pick the mean density of the Solar System,
3
1011 gr/cm , which satisfies the weak-field limit criteria. For this value, |2 / T | 1021 , where T .
The physical branch of the solution (142) is the one
with positive sign because, given that T < 0, it ensures
that matter leads to a standard positive curvature in
strong gravity. Then,
R T

322
T

(143)

and R1 T . Thus, 2 /R1 1021 and it

is evident that the required condition does not hold for
some typical densities related to the Newtonian limit.

27
The situation does not improve even with the unphysical branch of eq. (142) with anegative sign. In fact, in
this case, R1 2 [32 /( T ) + 3] and the correction to
the background curvature is of the order 2 and not much
smaller than that, as it would be required in order to
truncate the expansion (140). In (Barraco and Hamity,
2000), this fact was overlooked and only linear terms in
R1 were kept in the expansion of f (R) and f (R) around
R0 . In (Meng and Wang, 2004a), even though this fact
is noticed in the final stages of the analysis and is actually used, the authors do not take it into account properly
from the outset, keeping again only linear terms [see, e.g.,
eq. (11) of (Meng and Wang, 2004a)].
However, the algebraic dependence of R on the density
does not only signal a problem for the approaches just
mentioned. It actually implies that the outcome of the
post-Newtonian expansion itself depends on the density,
as shown in (Olmo, 2005a,b; Sotiriou, 2006c). Consider
for instance, along the lines of (Sotiriou, 2006a,c), the
conformal metric
h = f (R)g

(144)

that was introduced in Sec. II.B [cf. eq. (22)]. In terms

of this metric, the field equations can be written in the
form


1
R
R Rh + (f 1) R h = T ,
2
2f
(145)
and R is the Ricci tensor of the metric h . It is evident that, if f = 1, then h and g coincide and
eq. (145) yields Einsteins equation. However, since R
and consequently f (R) are functions of the energy density, due to eq. (20), the deviation of f from unity will
always depend on the energy density and the functional
form of f . Therefore, one can definitely find some function f which, for some range of energy densities, will give
f = 1 to high precision. However, for the same function
f , there will be large deviations from f = 1 at a different density range. This dependence of the weak-field
limit on the energy density is a novel characteristic of
Palatini f (R) gravity.
This dependence can be made explicit if the problem is approached via the equivalent BransDicke theory (Olmo, 2005a,b). Note that the usual bounds coming from Solar System experiments do not apply in the
0 = 3/2 case, which is equivalent to Palatini f (R)
gravity. This is because the standard treatment of the
post-Newtonian expansion of BransDicke theory, which
one uses to arrive to such bounds, is critically based on
the assumption that 0 6= 3/2 and the term (20 + 3)
frequently appears as a denominator. Making this assumption is not necessary, of course, in order to derive
a post-Newtonian expansion, but is a convenient choice,
which allows for this otherwise general treatment. Therefore, a different approach, such as the one followed in
(Olmo, 2005b), was indeed required for the 0 = 3/2
case. Following the standard assumptions of a postNewtonian expansion around a background specified by a

cosmological solution (Will, 1981), the following relations

were derived for the post-Newtonian limit:


1
V ()
2 h100 (T ) =
,
(146)
2
2




1
+ V ()
2 h1ij + ij (T ) =
, (147)
2
2

where V is the potential of the scalar field and (T )

log[/0 ]. The subscript 0 in 0 , and in any other quantity in the rest of this subsection, denotes that it is evaluated at T = 0.
The solutions of eqs. (146) and (147) are

2Geff M
V0 2
r + (T ),
(148)
+
r
60


V0 2
2Geff M
(1)
r (T ) ij , (149)

hij (t, ~x) =

r
60
R
where M 0 d3 ~x (t, ~x ) /. The effective Newton
constant Geff and the post-Newtonian parameter are
defined as


MV
G
1+
,
(150)
Geff
0
M
(1)

h00 (t, ~x) =

M MV
,
M + MV

(151)

R
where MV 1 0 d3 ~x [V0 /0 V ()/].
As stated in different words in (Olmo,
R 2005b), if the
Newtonian mass is defined as MN d3 ~x (t, ~x ), the
requirement that a theory has a good Newtonian limit
is that Geff M equals GMN , where N denotes Newtonian, and 1 to very high precision. Additionally, the
second term on the right hand side of both eqs. (148)
and (149) should be negligible, since it plays the role of
a cosmological constant term. (T ) should also be small
and have a negligible dependence on T .
Even though it is not impossible, as mentioned before,
to prescribe f such that all of the above are satisfied for
some range of densities within matter (Sotiriou, 2006c),
this does not seem possible over the wide range of densities relevant for the Solar System tests. As a matter
of fact, is nothing but an algebraic function of T and,
therefore, of the density (since is an algebraic function of R). The presence of the (T ) term in eqs. (148)
and (149) signals an algebraic dependence of the postNewtonian metric on the density. This direct dependence
of the metric on the matter field is not only surprising but
also seriously problematic. Besides the fact that it is evident that the theory cannot have the proper Newtonian
limit for all densities (the range of densities for which it
will fail depends on the functional form of f ), consider
the following: What happens to the post-Newtonian metric if a very weak point source (approximated by a delta
function) is taken into account as a perturbation? And
will the post-Newtonian metric be continuous when going
from the interior of a source to the exterior, as it should?

28
We will refrain from further analysis of these issues
here, since evidence coming from considerations different than the post-Newtonian limit, which we will review
shortly, will be of significant help. We will, therefore,
return to this discussion in Sec. VI.C.2.

B. Stability issues

In principle, several kinds of instabilities need to be

considered to make sure that f (R) gravity is a viable
alternative to GR (Calcagni et al., 2006; Chiba, 2005;
De Felice et al., 2006; Sokolowski, 2007a,b,c; Wang,
2005).
The Dolgov-Kawasaki (Dolgov and Kawasaki, 2003a)
instability in the matter sector, specific to metric f (R)
gravity, imposes restrictions on the functional form of
f , and is discussed below. More generally, it is believed that a stable ground state, the existence of which
is necessary in a gravitational theory, should be highly
symmetric, such as the de Sitter, or Minkowski, or
perhaps the Einstein static space. Instabilities of de
Sitter space in the gravity sector have been found in
(Barrow and Hervik, 2006a; Dolgov and Pelliccia, 2006;
Faraoni, 2004b,c, 2005a; Faraoni and Nadeau, 2005) [see
also (Barrow and Ottewill, 1983; Muller et al., 1990) for
pre-1998 discussions], while stability in first loop quantization of f (R) gravity and with respect to black
hole nucleation was studied in (Cognola et al., 2005,
2008; Cognola and Zerbini, 2006; Paul and Paul, 2006,
2005). The linear stability of de Sitter space with respect to homogeneous perturbations in generalized
theo

ries of the form f R, R R , R R was studied in (Cognola and Zerbini, 2008). The stability of
the Einstein static space in metric f (R) gravity with
respect to homogeneous perturbations was studied in
(Boehmer et al., 2007b), while stability of this space with
respect to inhomogeneous isotropic perturbations was established, with a gauge-invariant formalism and under
certain conditions, in (Goswami et al., 2008).

1. Ricci stability in the metric formalism

In the metric formalism, Dolgov and Kawasaki discovered an instability in the prototype model f (R) =
R 4 /R (now called Dolgov-Kawasaki, or Ricci
scalar or matter instability), which manifests itself on
an extremely short time scale and is sufficient to rule out
this model (Dolgov and Kawasaki, 2003a). Their result
was confirmed in (Nojiri and Odintsov, 2003a, 2004b), in
which it was also shown that adding to this specific f (R)
an R2 term removes this instability. The instability was
rediscovered in (Baghram et al., 2007) for a specific form
of the function f (R). The analysis of this instability is
generalized to arbitrary f (R) theories in the metric formalism in the following way (Faraoni, 2006a).
We parametrize the deviations from Einstein gravity

as
f (R) = R + (R),

(152)

where is a small parameter with the dimensions of a

mass squared and is arranged to be dimensionless (in
the example f = R 4 /R, one has = 2 , = 2 /R,
and H0 1033 eV).
By using the trace equation (8)
3f (R) + f (R)R 2f (R) = T,

(153)

and evaluating f ,

( 1)
T
2

R
+
R=
+
.

3
3
3
(154)
We assume that 6= 0: if = 0 on an interval then the
theory reduces to GR. Isolated zeros of , at which the
theory is instantaneously GR, are in principle possible
but will not be considered here.
Consider a small region of spacetime in the weak-field
regime and approximate locally the metric and the curvature by
R +

g = + h ,

R = T + R1 ,

(155)

where is the Minkowski metric and |R1 / T | 1.

This inequality excludes the case of conformally invariant
matter with T = 0, a situation considered later. Equation (155) yields, to first order in R1 ,

1 2 R1 2 T R 1 + 2 T
~ R
~ 1
R



( T + 2)
1
1

,
R1 = T 2 T
+
3

3
(156)
~ and 2 are the gradient and Laplacian in
where
Euclidean three-dimensional space, respectively, and an
overdot denotes differentiation with respect to time. The
function and its derivatives are now evaluated at R =
T . The coefficient of R1 in the fifth term on the left
hand side is the square of an effective mass and is domi1
nated by the term (3 ) due to the extremely small
value of needed for these theories to reproduce the correct cosmological dynamics. Then, the scalar mode R1 of
the f (R) theory is stable if = f > 0, and unstable if
this effective mass is negative, i.e., if = f < 0. The
time scale for this instability to manifest is estimated to
be of the order of the inverse effective mass 1026 s
in the example (R) = 4 /R (Dolgov and Kawasaki,
2003a). The small value of gives a large effective mass
and is responsible for the small time scale over which the
instability develops.
Let us consider now matter with vanishing trace T of
the stress-energy tensor. In this case eq. (156) becomes
2


1 + R1 2 2 R1
~ 1
R
R



2
1
1
+
.
(157)
R1 =
3

29
1

Again, the effective mass term is (3 ) , which has

the sign of f and the previous stability criterion is recovered. The stability condition f (R) 0 is useful to
veto f (R) gravity models.22
When f < 0, the instability of these theories can be
interpreted, following eq. (156), as an instability in the
gravity sector. Equivalently, since it appears inside matter when R starts deviating from T [see eq. (155)], it can
be seen as a matter instability [this is the interpretation
taken in (Dolgov and Kawasaki, 2003a)]. Whether the
instability arises in the gravity or matter sector seems to
be a matter of interpretation.
The instability of stars made of any type of matter
in theories with f < 0 and sufficiently small is confirmed, with a different approach (a generalized variational principle) in (Seifert, 2007), in which the time
scale for instability found by Dolgov and Kawasaki in
the 1/R model is also recovered. The stability condition
f 0 is recovered in studies of cosmological perturbations (Sawicki and Hu, 2007).
The stability condition f (R) 0, expressing the
fact that the scalar degree of freedom is not a ghost,
can be given a simple physical interpretation (Faraoni,
2007b). Assume that the effective gravitational coupling
Gef f (R) G/f (R) is positive; then, if Gef f increases
with the curvature, i.e.,
dGef f
f (R)G
=
2 > 0,
dR
(f (R))

(158)

at large curvature the effect of gravity becomes stronger,

and since R itself generates larger and larger curvature
via eq. (153), the effect of which becomes stronger and
stronger because of an increased Gef f (R), a positive feedback mechanism acts to destabilize the theory. There is
no stable ground state if a small curvature grows and
grows without limit and the system runs away. If instead the effective gravitational coupling decreases when
R increases, which is achieved when f (R) > 0, a negative feedback mechanism operates which compensates
for the increase in R and there is no running away of
the solutions. These considerations have to be inverted
if f < 0, which can only happen if the effective energy
density ef f also becomes negative. This is not a physically meaningful situation because the effective gravitational coupling becomes negative and the tensor field
and the scalar field of metric f (R) gravity become ghosts
(Nunez and Solganik, 2004).
GR, with f (R) = 0 and Gef f = constant, is the borderline case between the two behaviours corresponding to
stability (f > 0) and instability (f < 0), respectively.
Remarkably, besides the Dolgov-Kawasaki instability
which manifest itself in the linearized version of equa-

22

Refs. (Multamaki and Vilja, 2006a; Nojiri, 2004) hinted towards

the stability criterion, but did not fully derive it because a decomposition in orders of 1 was not performed.

tion (153), there are also recent claims that R can be

driven to infinity due to strong non-linear effects related
to the same equation (Appleby and Battye, 2008; Frolov,
2008; Tsujikawa, 2008). More specifically, in (Tsujikawa,
2008) an oscillating mode is found as a solution to the
perturbed version of eq. (153) . This mode appears to
dominate over the matter-induced mode as one goes back
into the past and, therefore, it can violate the stability
conditions. In (Frolov, 2008), eq. (153) was studied, with
the use of a convenient variable redefinition but without
resorting to any perturbative approach. It was found that
there exists a singularity at a finite field value and energy
level. The strongly non-linear character of the equation
allows R to easily reach the singularity in the presence of
matter. As noticed in (Appleby and Battye, 2008), since
when it comes to cosmology the singularity lies in the
past, it can in principle be avoided by choosing appropriate initial conditions and evolving forward in time. This,
of course, might result in a hidden fine-tuning issue.
All three studies mentioned consider models in which
f (R) includes, besides the linear term, only terms which
become important at low curvatures. It is the form of
the effective potential governing the motion of R, which
depends on the functional form of f (R), that determines
how easy it is to drive R to infinity (Frolov, 2008). Therefore, it seems interesting to study how the presence of
terms which become important at large curvatures, such
as positive powers of R, could affect these results. Finally, it would be interesting to see in detail how these
findings manifest themselves in the case of compact objects, and whether there is any relation between this issue
and the Dolgov-Kawasaki instability.

2. Gauge-invariant stability of de Sitter space in the metric

formalism

One can consider the generalized gravity action



Z
()
f (, R)
4
S = d x g

V () ,
2
2
(159)
incorporating both scalar-tensor gravity (if f (, R) =
()R) and modified gravity (if the scalar field is absent and fRR 6= 0). In a spatially flat FLRW universe
the vacuum field equations assume the form


1
2 Rf
f
2

H =
+
+ V 3H f ,(160)
3f 2
2
2


1
(161)
H = 2 + F H f ,
2f


1 d 2 f
dV
+ 3H +
= 0, (162)

+2
2 d

d
where f f /, F f /R, and an overdot denotes
differentiation with respect to t. We choose (H, ) as
dynamical variables; then, the stationary points of the
dynamical system (160)-(162) are de Sitter spaces with

30
constant scalar field (H0 , 0 ). The conditions for these
de Sitter solutions to exist are
6H02 f0 f0 + 2V0 = 0,


dV 
df 

2
= 0,
d 0
d 0

(163)
(164)

where f0 f (0 , R0 ), f0 f (0 , R0 ), V0 V (0 ),
and R0 = 12H02 . The phase space is a curved twodimensional surface embedded in a three-dimensional
space (de Souza and Faraoni, 2007).
Inhomogeneous perturbations of de Sitter space have
been studied using the covariant and gauge-invariant
formalism of (Bardeen, 1980; Ellis and Bruni, 1989;
Ellis et al., 1990, 1989) in a version provided by (Hwang,
1990a,b, 1997, 1998; Hwang and Noh, 1996) for generalized gravity. The metric perturbations are defined by
g00 = a2 (1 + 2AY ) , g0i = a2 B Yi , (165)
gij = a2 [hij (1 + 2HL Y ) + 2HT Yij ] .
(166)
Here the Y are scalar spherical harmonics, hij is the
i is
three-dimensional metric of the FLRW background,
the covariant derivative of hij , and k is the eigenvalue of
i Y = k 2 Y . The Yi and Yij are vector and tensor
i

harmonics satisfying
1
Yi =
i Y,
k

Yij =

1
1
i j Y + Y hij ,
2
k
3

(167)

respectively. The Bardeen gauge-invariant potentials


HT
a
a 
H = HL +
(168)
B H T ,
+
3
k
k

a
a 
B H T
A = A +
k
k

a
1 
(169)
+ B
a HT ,
k
k
the Ellis-Bruni variable
= +


a 
a
B H T ,
k
k

(170)

and analogous gauge-invariant variables f , f , and

R satisfy first order equations given in (Hwang,
1990a,b, 1997, 1998; Hwang and Noh, 1996), which simplify significantly in the de Sitter background (H0 , 0 )
(Faraoni, 2004b, 2005a; Faraoni and Nadeau, 2005).
To first order and in the absence of ordinary matter, vector perturbations do not appear (Hwang, 1990a,b,
1997, 1998; Hwang and Noh, 1996), and de Sitter space is
always stable with respect to first order tensor perturbations. Focusing on scalar perturbations, modified gravity corresponds to 1 and f = f (R) with f (R) 6= 0
in (159). The gauge-invariant perturbations H (from
which one easily obtains A and R) satisfy
 2

k
f0
2

H + 3H0 H +
4H0 + H = 0, (171)
a2
3f0

(Faraoni, 2004b, 2005a; Faraoni and Nadeau, 2005),

where the term k 2 /a2 can be dropped at late times and
for long wavelength modes. Linear stability ensues if the
coefficient of H is non-negative, i.e. (using eq. (163)),
if23
(f0 )2 2f0 f0
0.
f0 f0

(173)

The only term containing the comoving wave vector k

in eq. (171) becomes negligible at late times and/or for
zero-momentum modes and thus the spatial dependence
effectively disappears. In fact, eq. (173) coincides with
the stability condition that can be obtained by a straightforward homogeneous perturbation analysis of eqs. (160)
and (161). As a result, in the stability analysis of de
Sitter space in modified gravity, inhomogeneous perturbations can be ignored and the study can be restricted to
the simpler homogeneous perturbations, which are free of
the notorious gauge-dependence problems. This result,
which could not be reached a priori but relies on the
inhomogeneous perturbation analysis, holds only for de
Sitter spaces and not for different attractor (e.g., powerlaw) solutions that may be present in the phase space.
The stability condition (173) is equivalent to the condition that the scalar field potential in the Einstein frame
of the equivalent BransDicke theory has a minimum at
the configuration identified by the de Sitter space of curvature R0 (Sokolowski, 2007b).
As an example, let us consider the prototype model
f (R) = R

4
.
R

(174)

The background de Sitter space has R0 = 12H02 = 3 2

and the stability condition (173) is never satisfied: this
de Sitter solution is always unstable. An improvement is
obtained by adding a quadratic correction to this model:
4
+ aR2 .
(175)
R
Then, the condition for the existence of a de Sitter solution is again R0 = 3 2 , while the stability condition (173) is satisfied if a > 313 2 , in agreement with
(Nojiri and Odintsov, 2003a, 2004b) who use independent methods.
Different definitions of stability lead to different, albeit close, stability criteria for de Sitter space [see
(Cognola et al., 2005, 2008) for the semiclassical stability
of modified gravity, (Bertolami, 1987) for scalar-tensor
gravity, and (Seifert, 2007) for a variational approach applicable to various alternative gravities].
f (R) = R

23

The generalization of the condition (173) to D spacetime dimensions, derived in (Rador, 2007) for homogeneous perturbations,
is

2
(D 2) f0 Df0 f0
0.
(172)
f0 f0

31
3. Ricci stability in the Palatini formalism

C. The Cauchy problem

For Palatini f (R) gravity the field equations (18)

and (19) are of second order and the trace equation (20)
is

A physical theory must have predictive power and, to

this extent, a well-posed initial value problem is a required feature. GR satisfies this requirement for most
reasonable forms of matter (Wald, 1984). The wellposedness of the Cauchy problem for f (R) gravity is an
open issue. Using harmonic coordinates, Noakes showed
that theories with action
Z

1
S=
d4 x g R + R R + R2 + SM (177)
2

f (R)R 2f (R) = T,

(176)

where R is the Ricci scalar of the non-metric connection

(and not that of the metric connection { } of g ).
Contrary to the metric case, eq. (176) is not an evolution
equation for R; it is not even a differential equation, but
rather an algebraic equation in R once the function f (R)
is specified. This is also the case in GR, in which the
Einstein field equations are of second order and taking
their trace yields R = T . Accordingly, the scalar field
of the equivalent 0 = 3/2 BransDicke theory is not
dynamical. Therefore, the Dolgov-Kawasaki instability
can not occur in Palatini f (R) gravity (Sotiriou, 2007a).

4. Ghost fields

Ghosts (massive states of negative norm that cause

apparent lack of unitarity) appear easily in higher order
gravities. A viable theory should be ghost-free: the
presence of ghosts in f (R, R R , R R )
gravity has been studied in (Buchbinder et al.,
1992; Codello and Percacci, 2006; De Felice, 2007;
De Felice and Hindmarsh, 2007; Stelle, 1978, 1977;
Strominger,
1984;
Utiyama and DeWitt,
1962;
Vilkovisky, 1992). Due to the GaussBonnet identity, if the initial action is linear in R R ,
one can reduce the theory under consideration to24
f (R, R R ) which, in general, contains a massive
spin 2 ghost field in addition to the usual massless graviton and the massive scalar. f (R) theories have no ghosts
(Buchbinder et al., 1992; Ferraris et al., 1988; Stelle,
1978, 1977; Strominger, 1984; Utiyama and DeWitt,
1962; Vilkovisky, 1992), and the stability condition f (R) 0 of (Dolgov and Kawasaki, 2003a;
Faraoni, 2006a) essentially amounts to guarantee that
the scalaron is not a ghost. Theories of the kind
f (R, R R , R R ) in general are plagued by
ghosts [this is the case, for example, of conformal
gravity, as noticed long before the 1998 discovery of
the cosmic acceleration (Riegert, 1984)], but

models
with only f R, R2 4R R + R R terms in
the action have been claimed to be ghost-free (Comelli,
2005; Navarro and Van Acoleyen, 2006).

24

Furthermore, R R can be expressed in terms of R2 in a

FLRW background (Wands, 1994).

in the metric formalism have a well posed initial value

problem in vacuo (Noakes, 1983). By using the dynamical equivalence with the scalar-tensor theory (54)
when f (R) 6= 0, the well-posedness of the Cauchy
problem can be reduced to the analogous problem for
BransDicke gravity with 0 = 0 (metric formalism) or
0 = 3/2 (Palatini formalism). The fact that the initial
value problem is well-posed was demonstrated for particular scalar-tensor theories in (Cocke and Cohen, 1968;
Noakes, 1983) and a general analysis has recently been
presented in (Salgado, 2006; Salgado et al., 2008). This
work, however, does not cover the 0 = 0, 3/2 cases.
A system of 3 + 1 equations of motion is said to be
well-formulated if it can be rewritten as a system of equations which are of only first order in both time and space
derivatives. When this set can be put in the full first
order form
~ (~u) ,
t ~u + M i i ~u = S

(178)

where ~u collectively denotes the fundamental variables

hij , Kij , etc. introduced below, M i is called the charac~ (~u) describes source
teristic matrix of the system, and S
terms and contains only the fundamental variables but
not their derivatives. The initial value formulation is
well-posed if the system of partial differential equations
is symmetric hyperbolic (i.e., the matrices M i are symmetric) and strongly hyperbolic if si M i has a real set of
eigenvalues and a complete set of eigenvectors for any
1-form si , and obeys some boundedness conditions [see
(Solin, 2006)].
The Cauchy problem for metric f (R) gravity is wellformulated and is well-posed in vacuo and with matter,
as shown below. For Palatini f (R) gravity, instead, the
Cauchy problem is unlikely to be well-formulated or wellposed unless the trace of the matter energy-momentum
tensor is constant, due to the presence of higher derivatives of the matter fields in the field equations and to the
impossibility of eliminating them (see below).
A systematic covariant approach to scalar-tensor theories of the form


Z
()R 1
4
S = d x g
W () + SM
2
2
(179)
is due to (Salgado, 2006), who showed that the Cauchy
problem of these theories is well-posed in the absence of

32
matter and well-formulated otherwise. With the exception of 0 = 3/2, as we will see later, most of Salgados
results can be extended to the more general action


Z
()R ()
4

W () +SM ,
S = d x g
2
2
(180)
which contains the additional coupling function ()
(which is different from the BransDicke parameter 0 )
(Lanahan-Tremblay and Faraoni, 2007).
The field equations, after setting = 1 for this section,
are
1 h
G =
( g )

i
+ ( g )


1h
1
+ g

2
i
(m)
(181)
W ()g + T ,

R W () + = 0, (182)
2
2
where a prime denotes differentiation with respect to .
Eq. (181) can be cast in the form of the effective Einstein
(ef f )
equation G = T , with the effective stress-energy
tensor (Salgado, 2006)

1  ()
(ef f )
()
(m)
T
=
T + T
+ T
,
(183)
()
 +

The metric is then

ds2 = N 2 N i Ni dt2 2Ni dtdxi + hij dxi dxj (188)

(i, j = 1, 2, 3), where N > 0, n = N t, N =

h t is the shift vector, while t obeys t t = 1 and
t = N + N n so that N = n t and N n = 0.
The extrinsic curvature of t is
K = h h n

and the 3D covariant derivative of h on t is defined

by
(3)

1 ...
Di T 1 ... 1 ... = h1 1 . . . h1 1 . . . h i (3)
1 ...
T
(190)
for any 3-tensor (3) T 1 ... 1 ... , with Di h = 0. The spatial gradient of the scalar is Q D (where D
denotes the covariant derivative of h ), while its momentum is = Ln = n and


1 hij
Kij = i nj =
+ Di Nj + Dj Ni , (191)
2N
t
1
=
(t + N Q ) ,
(192)
N
t Qi + N l l Qi + Ql i N l = Di (N ) . (193)
(ef f )

The effective stress-energy tensor T

as
(ef f )

and

+ () ( g ) ,
(184)


1
= () g
2
W ()g .
(185)

The trace of the effective Einstein equations yields

( "
#)1 
2
3 ( )
T (m)
 = +
2 W ()
2
2




( + 3 ) c .
+W () +
2
2
(186)
The 3 + 1 ArnowittDeserMisner (ADM) formulation of
the theory proceeds by introducing lapse, shift, extrinsic curvature, and gradients of (Reula, 1998; Salgado,
2006; Wald, 1984). Assume that a time function t exists such that the spacetime (M ,g ) admits a foliation
with hypersurfaces t of constant t with unit timelike
normal na . The 3-metric and projection operator on t
are h = g + n n and h , respectively. Moreover,

n n = 1,

n = 0,

h n = h ,

h h = h .

is decomposed

1
(S + J n + J n + En n ) ,

(194)

where

()
T
= () g
()
T

(189)

(187)


1  ()
()
(m)
S + S + S ,(195)


1  ()
(ef f )
J h T n =
J + J() + J(m) ,(196)



1
(ef
f
)
E n n T =
E () + E () + E (m) . (197)

(ef f )

S h h T

Its trace is T (ef f ) = S E, where S S . The Gauss

Codazzi equations then yield the Hamiltonian constraint
(Salgado, 2006; Wald, 1984)
(3)

R + K 2 Kij K ij = 2E,

(198)

the vector constraint

D l K l i D i K = Ji ,

(199)

and the dynamical equations

t K i j + N l l K i j + K i l j N l K l j l N i + Di Dj N

N
(3) Ri j N N KK i j =
(S E) ji 2Sji , (200)
2

where K K i i . The trace of eq. (200) yields

t K + N l l K +(3) N N Kij K ij =

N
(S + E) , (201)
2

33
where (3) Di Di . Our purpose is to eventually
eliminate all second derivatives.  which is present
in eqs. (195)-(197) can actually be eliminated using
eq. (186), provided that 6= 3( )2 /(2).
To be more precise, a direct calculation yields the and -quantities of eqs. (195)-(197)
E () = (D Q + K) + Q2 ,

(202)

J()
()
S

+ D ) Q ,

(203)

= (D Q + K h )




h Q2 2 Q Q ,

(204)

(K Q

where Q2 Q Q , while

S () = (D Q + K 3) + 32 2Q2 ,
(205)
and

E () =
(206)
2 + Q2 + W (),
2
J() = Q ,
(207)
i
h


()
Q2 2 + W () ,(208)
S = Q Q h
2

S () =
(209)
32 Q2 3W ().
2

The Hamiltonian and the vector constraints become

2h
(3)
(D Q + K) + 2
R + K 2 Kij K ij

2



2
Q2
E (m) + W () , (210)
( + 2 ) =
+
2

1
(Ki c Qc + Di )
Dl K l i Di K +

+ ( + ) Qi =


(m)
Ji

(211)

respectively, and the dynamical equation (200) is

t K

+ N l K
(3)

j
i

+ K l j N Kj l N

+D Dj N R j N N KK i j



N  2
Q 2 + 2W () +  ji
+
2

N
N
+
Dji + K i j
( + ) Qi Qj

i

N h (m)
=
(212)
S
E (m) ji 2S (m) i j ,
2

with trace

t K + N l l K +(3) N N Kij K ij
N

(D Q + K)


N  2
Q (2 + 3 ) 2
+
2

N 
=
2W () 3  + S (m) + E (m) (213)
2

where (Salgado, 2006)

Ln K Q D (ln N ) D Q = 
(
1
T (m)
i
= h
2 W () + W ()

2
3( )
2
+ 2
)



+
( + 3 )

(214)
2
2
In vacuo, the initial data (hij , Kij , , Qi , ) on an initial hypersurface 0 obey (210), (211), and Qi = Di ,
Di Qj = Dj Qi . In the presence of matter, the variables
(m)
(m)
E (m) , Ja , Sab must also be assigned on the initial hypersurface 0 . Fixing a gauge corresponds to specifying
the lapse and the shift vector. The system (210)-(213)
contains only first-order derivatives in both space and
time once the dAlembertian  is written in terms of
, , , and its derivatives by means of eqs. (186)
or (214). As mentioned earlier, this can be done whenever
6= 3( )2 /(2). As already pointed out in (Salgado,
2006) for the specific case = 1, and can be now generalized for any 6= 3( )2 /(2), the reduction to a
first-order system shows that the Cauchy problem is wellposed in vacuo and well-formulated in the presence of
reasonable matter.
Let us now consider the results specific to f (R) gravity.
Recall that BransDicke theory, which is of interest for
us due to its equivalence with f (R) gravity, corresponds
to () = 0 /, with () = , and W 2V . This
yields the constraints
(3)

2
R + K 2 Kij K ij [D Q + K


i


2 h (m)
0
E
+ V () , (215)
2 + Q2 =
+
2

Dl K l i Di K


(m)
0
1
J
+
K i l Ql + Di +
Qi = i , (216)

and the dynamical equations

t K i j + N l l K i j + K i l j N l Kj l l N i + Di Dj N
N i
(3) Ri j N N KK i j +
(2V () + )
2 j

N 0
N
Di Qj + K i j + 2 Qi Qj
+

i

N h (m)
(m)
(217)
ji 2S (m) i j ,
S
E
=
2

t K + N l l K +(3) N N Kij K ij
0 N
N
(D Q + K) 2 2

i
N h
2V () 3 + S (m) + E (m) ,
=
2

(218)

34
with




T (m)
0
3
 =
2 Q2 .
2V ()+V ()+
0 +
2
2

(219)
The condition 6= 3( )2 /(2), which needs to be satisfied in order to for one to be able to use eq. (186) in
order to eliminate  can be written in the BransDicke
theory notation as 0 6= 3/2. One could of course had
guessed that by looking at eq. (219). Therefore, metric f (R) gravity, which is equivalent to 0 = 0 Brans
Dicke gravity, has a well-formulated Cauchy problem in
general and is well-posed in vacuo. Further work by
(Salgado et al., 2008) established the well-posedness of
the Cauchy problem for scalar-tensor gravity with = 1
in the presence of matter; this can be translated into the
well-posedness of metric f (R) gravity with matter along
the lines established above.
How about Palatini f (R) gravity, which, corresponding to 0 = 3/2, is exactly the case that the constraint
0 6= 3/2 excludes? Actually, for this value of the
BransDicke parameter, eq. (69), and consequently at
eq. (219), include no derivatives of . Therefore, one can
actually solve algebraically for . [The same could be
done using eq. (186) in the more general case where is
a function of , when = 3( )2 /(2).] We will not
consider cases for which eq. (69) has no roots or when it
is identically satisfied in vacuo. These cases lead to theories for which, in the Palatini f (R) formulation, eq. (21)
has no roots or when it is identically satisfied in vacuo
respectively. As already mentioned in Sec. II.B, the first
case leads to inconsistent field equations, and the second
to a conformally invariant theory (Ferraris et al., 1992),
see also (Sotiriou, 2006b) for a discussion.
Now, in vacuo one can easily show that the solutions
of eq. (69) or (219) will be of the form = constant.
Therefore, all derivatives of vanish and one conclude
that 0 = 3/2 BransDicke theory or Palatini f (R)
gravity have a well-formulated and well-posed Cauchy
problem.25 This could have been expected, as noticed in
(Olmo and Singh, 2009), considering that Palatini f (R)
gravity reduces to GR with a cosmological constant in
vacuo.
In the presence of matter, things are more complicated.
The solutions of eq. (69) or (219) will give as a function
of T , the trace of the stress-energy tensor. This can still
be used to replace in all equations but it will lead to
terms such as T . Therefore, for the Cauchy problem
to be well-formulated in the presence of matter, one does

25

This has been missed in (Lanahan-Tremblay and Faraoni, 2007),

where it is claimed that the Cauchy problem is not well-posed
because the constraint 0 6= 3/2 does not allow for the use
of eq. (219) in order to eliminate . Note also that in the
absence of a potential (there is no corresponding Palatini f (R)
gravity) 0 = 3/2 BransDicke theory does not have a wellposed Cauchy problem, as noticed in (Noakes, 1983), because
this theory is conformally invariant and is indeterminate.

not only have to assume that the matter is reasonable,

in the sense that the matter fields satisfy a quasilinear,
diagonal, second order hyperbolic system of equations
[see (Wald, 1984)], but also to require that the matter
field equations are such that allow to express all derivatives of T present in eq. (215) to (218) for 0 = 3/2
in terms of only first derivatives of the matter fields. It
seems highly implausible that this requirement can be
fulfilled for generic matter fields. This seems to imply
that 0 = 3/2 BransDicke theory and Palatini f (R)
gravity are unlikely to have a well-formulated Cauchy
problem in the presence of matter fields. However, more
precise conclusions can only be drawn if specific matter
fields are considered on a case by case basis. The complication arising from the appearance of derivatives of T ,
and consequently higher derivatives of the matter fields
in the equations, and seem to be critical on whether the
Cauchy problem can be well-formulated in the presence
of matter will be better understood in Sec. VI.C.2.
VI. CONFRONTATION WITH PARTICLE PHYSICS AND
ASTROPHYSICS
A. Metric f (R) gravity as dark matter

Although most recent motivation for f (R) gravity

originates from the need to find alternatives to the mysterious dark energy at cosmological scales, several authors
adopt the same perspective at galactic and cluster scales,
using metric f (R) gravity as a substitute for dark matter
(Capozziello et al., 2004, 2005a, 2006a, 2007a, 2006b,
2007f; Iorio and Ruggiero, 2007a,b; Jhingan et al., 2008;
Martins and Salucci, 2007; Nojiri and Odintsov, 2008a;
Saffari and Sobouti, 2007; Zhao and Li, 2008). Given
the equivalence between f (R) and scalar-tensor gravity,
these efforts resemble previous attempts to model dark
matter using scalar fields (Alcubierre et al., 2002a,b;
Bernal and Guzman, 2006a,b,c; Bernal and Matos, 2005;
Cervantes-Cota et al., 2007a,b; Matos and Guzman,
2000; Matos et al., 2001; Matos and Urena-Lopez,
2001, 2007; Rodriguez-Meza and Cervantes-Cota, 2004;
Rodriguez-Meza et al., 2005, 2007).
Most works concentrate on models of the form
f (R) = Rn . A theory of this form with n = 1
/2 was studied in (Mendoza and Rosas-Guevara, 2007;
Saffari and Sobouti, 2007) by using spherically symmetric solutions to approximate galaxies. The fit to galaxy
samples yields

0.494
M
7
= (3.07 0.18) 10
,
(220)
1010 M
where M is the mass appearing in the spherically symmetric metric (the mass of the galaxy). Notice that having depending on the mass of each individual galaxy
straightforwardly implies that one cannot fit the data for
all galactic masses with the same choice of f (R). This
make the whole approach highly implausible.

35
(Capozziello et al., 2004, 2005a, 2006a, 2007a,f) computed weak-field limit corrections to the Newtonian
galactic potential and the resulting rotation curves;
when matched to galaxy samples, a best fit yields
n 1.7. (Martins and Salucci, 2007) performed a 2
fit using two broader samples, finding n 2.2 [see
also (Boehmer et al., 2007a) for a variation of this approach focusing on the constant velocity tails of the
rotation curves]. All these values of the parameter n
are in violent contrast with the bounds obtained by
(Barrow and Clifton, 2006; Clifton and Barrow, 2005a,
2006; Zakharov et al., 2006) and have been shown to violate also the current constraints on the precession of perihelia of several Solar System planets (Iorio and Ruggiero,
2007a,b). In addition, the consideration of vacuum metrics used in these works in order to model the gravitational field of galaxies is highly questionable.
The potential obtained in the weak-field limit of f (R)
gravity can affect other aspects of galactic dynamics as
well: the scattering probability of an intruder star and
the relaxation time of a stellar system were studied by
(Hadjimichef and Kokubun, 1997), originally motivated
by quadratic corrections to the EinsteinHilbert action.
B. Palatini f (R) gravity and the conflict with the Standard
Model

One very important and unexpected shortcoming of

Palatini f (R) gravity is that it appears to be in conflict with the Standard Model of particle physics, in the
sense that it introduces non-perturbative corrections to
the matter action (or the field equations) and strong couplings between gravity and matter in the local frame and
at low energies. The reason why we call this shortcoming unexpected is that, judging by the form of the action (13), Palatini f (R) gravity is, as we mentioned, a
metric theory of gravity in the sense that matter is only
coupled minimally to the metric. Therefore, the stress
energy tensor is divergence-free with respect to the metric
covariant derivative, the metric postulates (Will, 1981)
are fulfilled, the theory apparently satisfies the Einstein
Equivalence principle, and the matter action should trivially reduce to that of Special Relativity locally.
Let us see how this conflict comes about. This issue
was first pointed out in (Flanagan, 2004b) using Dirac
particles for the matter action as an example, and later
on studied again in (Iglesias et al., 2007) by assuming
that the matter action is that of the Higgs field [see
also (Olmo, 2008)]. Both calculations use the equivalent BransDicke theory and are performed in the Einstein frame. Although the use of the Einstein frame has
been criticized (Vollick, 2004),26 , this frame is equivalent
to the Jordan frame and both are perfectly suitable for

26

Note that in the case of (Flanagan, 2004b) in which fermions are

used as the matter fields, one could decide to couple the indepen-

performing calculations (Flanagan, 2004a) [see also the

relevant discussion in Sec. III and (Faraoni and Nadeau,
2007; Sotiriou et al., 2007)].
Nevertheless, since test particles are supposed to follow geodesics of the Jordan frame metric, it is this metric which becomes approximately flat in the laboratory
reference frame. Therefore, when the calculations are
performed in the Einstein frame they are less transparent since the actual effects could be confused with frame
effects and vice-versa. Consequently, for simplicity and
clarity, we present the calculation in the Jordan frame,
as it appears in (Barausse et al., 2008b). We begin from
the action (66), which is the Jordan frame equivalent of
Palatini f (R) gravity, and we take matter to be represented by a scalar field H (e.g., the Higgs boson), the
action of which reads


Z
1
m2H 2
4

SM =
(221)
d x g g H H 2 H
2~
~
(in units in which G = c = 1). As an example, we choose
f (R) = R 4 /R (Carroll et al., 2004; Vollick, 2003).
For this choice of f , the potential is V () = 22 (
1)1/2 . To go to the local frame, we want to expand the
action to second order around vacuum. The vacuum of
the action (66) with (221) as a matter action is H = 0,
= 4/3 [using eq. (69)] and g (2 acts as
an effective cosmological constant, so its contribution in
the local frame can be safely neglected).
However, when one tries to use a perturbative expansion for , things stop being straightforward:
is algebraically related to the matter fields as is obvious from eq. (69). Therefore, one gets T /2
m2H H 2 /(~3 2 ) at energies lower than the Higgs mass
(mH 100 1000 GeV). Replacing this expression in
the action (66) perturbed to second order, one immediately obtains that the effective action for the Higgs scalar
is


Z
m2H 2
1
effective
4

SM

d x g
g H H 2 H
2~
~


m2H H 2
m2H (H)2
1+
(222)
+
2 ~ 3
4 ~ 3
at energies E mH . Taking into account the fact that
2 H02 where H01 = 4000 Mpc is the Hubble
radius and H mH because E mH , it is not difficult
to estimate the order of magnitude of the corrections:
at an energy E = 103 eV (corresponding to the length

dent connection to them by allowing it to enter the matter action

and define the covariant derivative (which would be equivalent
to assuming that the spin connection is as independent variable
in a tetrad formalism), as noted in (Vollick, 2005). Although the
results of (Flanagan, 2004b) would cease to hold in this case, this
can not be considered a problem: clearly in this case we would
be talking about a different theory, namely metric-affine f (R)
gravity (Sotiriou and Liberati, 2007b).

36
scale L = ~/E = 2 104 m), the first correction is
of the order m2H H 2 /(2 ~3 ) (H01 /H )2 (mH /MP )2
1, where H = ~/mH 2 1019 2 1018 m
is the Compton wavelength of the Higgs and MP =
(~c5 /G)1/2 = 1.2 1019 GeV is the Planck mass. The
second correction is of the order m2H (H)2 /(4 ~3 )
(H01 /XH )2 (mH /MP )2 (H01 /L)2 1. Clearly, having
such non-perturbative corrections to the local frame matter action is unacceptable.
An alternative way to see the same problem would be
to replace m2H H 2 /(~3 2 ) directly in (66). Then
the coupling of matter to gravity is described by the interaction Lagrangian


2 g
m2H H 2
g
+
Lint
~3
2
"
 1 2 #
H0
m2H H 2
.
(223)
g 1 +

~3
L
This clearly exhibits the fact that gravity becomes nonperturbative at microscopic scales.
It is obvious that the algebraic dependence of on
the matter fields stands at the root of this problem. We
have still not given any explanation for the paradox
of seeing such a behavior in a theory which apparently
satisfies the metric postulates both in the f (R) and the
BransDicke representation. However, this will become
clear in Sec. VI.C.2.
C. Exact solutions and relevant constraints
1. Vacuum and non-vacuum exact solutions

Let us now turn our attention to exact solutions starting from metric f (R) gravity. We have already mentioned
in Sec. II.A that, as it can be seen easily from the form of
the field equations (6), the maximally symmetric vacuum
solution will be either Minkowski spacetime, if R = 0 is a
root of eq. (9), or de Sitter and anti-de Sitter spacetime,
depending on the sign of the root of the same equation.
Things are slightly more complicated for vacuum solutions with less symmetry: by using eq. (6) it is easy to
verify that any vacuum solution (R = g , T = 0)
of Einsteins theory with a (possibly vanishing) cosmological constant, including black hole solutions, is a solution
of metric f (R) gravity (except for pathological cases for
which eq. (9) has no roots). However, the converse is not
true.
For example, when spherical symmetry is imposed, the
Schwarzschild metric is a solution of metric f (R) gravity if R = 0 in vacuum. If R is constant in vacuo,
then Schwarzschild-(anti-)de Sitter spacetime is a solution. As we have already mentioned though, the JebsenBirkhoff theorem (Wald, 1984; Weinberg, 1972) does not
hold in metric f (R) gravity [unless, of course one wishes
to impose further conditions, such as that R is constant
(Capozziello et al., 2007d)]. Therefore, other solutions

can exists as well. An interesting finding is that the cosmic no-hair theorem valid in GR and in pure f (R) gravity
is not valid, in general, in theories of the form
Z

1
S=
d4 x g R + R2 + R R 2 ,
2
(224)
for
which
exact
anisotropic
solutions
that
continue to inflate anisotropically have been
found
(Barrow and Hervik,
2006a,b)
[see
also
(Kluske and Schmidt, 1996; Maeda, 1988)].
However, isotropization during inflation occurs in mixed
f (, R) models (Maeda et al., 1989)
In addition to the exact cosmological solutions
explored for the purpose of explaining the current cosmic acceleration [see, e.g., (Abdalla et al.,
2005; Barrow and Clifton, 2006; Clifton, 2006a,
2007;
Clifton and Barrow,
2005b,
2006)
and
(Capozziello and De Felice, 2008; Capozziello et al.,
2007c; Modak et al., 2005; Vakili, 2008) for an approach
based on Noether symmetries; see (Carloni et al., 2006)
for bouncing solutions and the conditions that they
satisfy], exact spherically symmetric solutions of metric
f (R) gravity have been explored in the literature, with
most recent studies being motivated by the need to understand the weak-field limit of cosmologically-motivated
theories.
Regarding non-vacuum solutions, the most common
matter source is a perfect fluid. Fluid dynamics in metric
f (R) gravity was studied in (Maartens and Taylor,
1994; Mohseni Sadjadi, 2007; Rippl et al., 1996;
Taylor and Maartens, 1995).
Spherically symmetric
solutions were found in (Bronnikov and Chernakova,
2005a,b,c; Bustelo and Barraco, 2007; Capozziello et al.,
2007d; Clifton, 2006a,b; Mignemi and Wiltshire, 1992;
Multamaki and Vilja, 2006b, 2007a,b; Whitt, 1985).
We regret not being able to present these solutions
extensively here due to space limitations and refer the
reader to the literature for more details.
Stability issues for spherically symmetric solutions
were discussed in (Seifert, 2007). In the theory

Z

g 
S = d4 x
R R2 R R + , (225)

where , , and are constants and is the Gauss

Bonnet invariant, the Schwarzschild metric is a solution
and the stability of Schwarzschild black holes was studied in (Whitt, 1985). Surprisingly, it was found that the
massive ghost graviton (poltergeist) present in this theory stabilizes small mass black holes against quantum instabilities [see also (Myers and Simon, 1988, 1989)]. In
the case = = 0, which reduces the theory (225) to
a quadratic f (R) gravity, the stability criterion found
in (Whitt, 1985) reduces to < 0, which corresponds
again to f (R) > 0. For = 0 we recover GR, in
which black holes are stable classically (but not quantummechanically, due to Hawking radiation and their negative specific heat, a feature that persists in f (R) gravity),

37
so the classical stability condition for Schwarzschild black
holes can be enunciated as f (R) 0.
Let us now turn our attention to Palatini f (R) gravity. In this case things are simpler in vacuo: as we saw
in Sec. II.B, the theory reduces in this case (or more
precisely even for matter fields with T =const., where
T is the trace of the stress energy tensor) to GR with
a cosmological constant, which might as well be zero
for some models (Barraco et al., 1999; Borowiec et al.,
1998; Ferraris et al., 1992, 1994). Therefore, it is quite
straightforward that Palatini f (R) gravity will have
the same vacuum solutions as GR with a cosmological constant. Also, the Jebsen-Birkhoff theorem (Wald,
1984; Weinberg, 1972) is valid in the Palatini formalism
(Barausse et al., 2008a,b,c; Kainulainen et al., 2007b).
Cosmological solutions in quadratic gravity were obtained in (Shahid-Saless, 1990, 1991). Spherically symmetric interior solutions in the Palatini formalism can
be found by using the generalization of the TolmanOppenheimer-Volkoff equation valid for these theories, which was found in (Barraco and Hamity, 2000;
Bustelo and Barraco, 2007; Kainulainen et al., 2007b).
Indeed, such solutions have been found and matched
with the unique exterior (anti-)de Sitter solution
(Barraco and Hamity, 1998, 2000; Bustelo and Barraco,
2007; Kainulainen et al., 2007a,b).
Nevertheless, a
matching between exterior and interior which can lead
to a sensible solution throughout spacetime is not always
feasible and this seems to have serious consequences for
the viability of f (R) gravity (Barausse et al., 2008a,b,c).
This is discussed extensively in the next section.
Let us close this section with some remarks on black
hole solutions. As is clear from the above discussion, all
black hole solutions of GR (with a cosmological constant)
will also be solutions of f (R) in both the metric and
the Palatini formalism [see also (Barausse and Sotiriou,
2008; Psaltis et al., 2008)]. However, in the Palatini formalism they will constitute the complete set of black
hole solutions of the theory, whereas in the metric formalism other black hole solutions can exist in principle, as the Jebsen-Birkhoff theorem does not hold. For
a discussion on black hole entropy in f (R) gravity see
(Jacobson et al., 1994, 1995; Vollick, 2007).27
2. Surface singularities and the incompleteness of Palatini f (R)
gravity

In Secs. V.A.3, V.C and VI.B, we already spotted three

serious shortcomings of Palatini f (R) gravity, namely
the algebraic dependence of the post-Newtonian metric
on the density, the complications with the initial value
problem in the presence of matter, and a conflict with

particle physics. In this section we will study static

spherically symmetric interior solutions and their matching to the unique exterior with the same symmetries,
the Schwarzschild-de Sitter solution, along the lines of
(Barausse et al., 2008a,b,c). As we will see, the three
problems mentioned earlier are actually very much related and stem from a very specific characteristic of Palatini f (R) gravity, which the discussion of this section will
help us pin down.
A common way of arriving to a full description of a
spacetime which includes matter is to solve separately
the field equations inside and outside the sources, and
then match the interior and exterior solutions using appropriate junction conditions [called Israel junction condition in GR (Israel, 1966)]. This is what we are going to
attempt here. We already know the exterior solution so,
for the moment, let us focus on the interior. Since we assume that the metric is static and spherically symmetric,
we can write it in the form
ds2 eA(r) dt2 + eB(r) dr2 + r2 d2 .

We can then replace this metric in the field equations

of Palatini f (R) gravity, preferably in eq. (28) which
is the simplest of all the possible reformulations. Assuming also a perfect fluid description for matter with
T = ( + P ) u u + P g , where is the energy density, P is the pressure, and u is the fluid 4-velocity,
and representing d/dr with a prime,28 one arrives at the
equations


1
1 eB
eB

A =
,
(227)

8GrP +
1+
r
F
r


1 eB
eB
+
1
,(228)
+
8Gr +
B =
1+
r
F
r
"  

#
2
F
2F
eB
f
2 3
r
,(229)
+
+
R
4 F
rF
2
F
"
 2 #

rF
3 F
2 F
,

, (230)
r
F
2 F
2F
where F f /R.
To determine an interior
solution we need a generalization of the TolmanOppenheimer-Volkoff (TOV) hydrostatic equilibrium
equation. This has been derived for Palatini f (R) gravity in (Barraco and Hamity, 2000; Bustelo and Barraco,
2007; Kainulainen et al., 2007b): Defining mtot (r)
r(1 eB )/2 and using Eulers equation
P =

28
27

See also (Eling et al., 2006) for a derivation of the field equations
of metric f (R) gravity based on thermodynamical arguments applied to local Rindler horizons.

(226)

A
(P + ) ,
2

(231)

In this section we modify our standard notation for economy

and a prime denotes differentiation with respect to the radial
coordinate instead of differentiation with respect to the argument
of the function.

38
one can use eqs. (227) and (228) to arrive to the generalised TOV equations:
1
( + P )

(232)
1 + r(r 2mtot )



4r3 P
(r 2mtot ) ,
mtot +
F
2


2
1
4r +
mtot
=
+

(+ ) .(233)
1+
F
2
r

P =

mtot

We now have three differential equations, namely

(231), (232) and (233), and four unknown functions,
namely A, mtot (or B), P , and . The missing piece
is the information about the microphysics of the matter configuration under investigation. In the case of a
perfect fluid, this is effectively given by an equation of
state (EOS). A one-parameter EOS relates the pressure
directly to the energy density, i.e., P = P (). A simple
form of such an EOS is a polytropic equation of state
P = k0 , where 0 is the rest-mass density and k and
are constants. This is the case that we will consider
here. Note that the rest-mass density can be expressed
in terms of the energy density and the internal energy
U as 0 = U . Assuming an adiabatic transformation
and using the first law of thermodynamics, one can express the internal energy in terms of the pressure, i.e.,
U = P/( 1). Therefore, the polytropic EOS can be
rewritten as
 1/
P
P
+
,
(234)
=
k
1
giving a direct link between P and .
Without specifying the interior solution, we can already examine the appropriate matching conditions
needed. One needs continuity of the metric and of A
on the surface of the matter configuration (A is given by
a second order differential equation). Since we know that
the exterior solution is unique and it is the Schwarzschildde Sitter solution with a cosmological constant equal to
R0 /4, where R0 is the vacuum value of R (see Sec. II.B),
we can directly write for the exterior

Assuming that, approaching the surface from the interior, A and mtot indeed take the correct values required
for the matching, it can be shown that continuity of A

across the surface requires F (rout ) = 0 for r rout

(Barausse et al., 2008a). Additionally, if this is the case
then (Barausse et al., 2008a)
mtot (rout )
where
C=

r3
R0 .
24

(239)

The sufficient condition for the singularity to occur is that a polytropic EOS with 3/2 < < 2
should adequately describe just the outer layer of
the matter configuration (and not necessarily the
whole configuration).
In practice, there is no dependence of the result
on the functional form of f (R) [a few unrealistic
exceptions can be found in (Barausse et al., 2008a)]
so what is revealed is a generic aspect of Palatini
f (R) gravity as a class of theories.
The singularities discussed are not coordinate, but
true singularities, as can be easily verified by checking that curvature invariants diverge.

(236)

On the other hand, based on the exterior solution, one

gets on the surface


3
2 rout
R0 12m

A (rout ) =
.
(237)
3 12r
rout (R0 rout
out + 24m)

dF d
dF
(P + ) =
(P + ) .
dP
d dP

Let us now examine the behavior of mtot at the surface for different values of the polytropic index . For
1 < < 3/2, C = dC/dP P dC/dP (P + ) 0 at
the surface so that the expression (238) is finite and it
gives continuity of mtot across the surface [cf. eq. (236)].
However, for 3/2 < < 2, C as the surface is approached, provided that dF/dR(R0 ) 6= 0 and
dR/dT (T0 ) 6= 0 (note that these conditions are satisfied
by generic forms of f (R), i.e., whenever an R2 term or
a term inversely proportional to R is present). Therefore, even though mtot remains finite (as can be shown
using the fact that P = 0 at the surface), the divergence of mtot drives to infinity the Riemann tensor of
the metric, R , and curvature invariants, such as R
or R R , as can easily be checked.29 Clearly, such
a singular behaviour is bound to give rise to unphysical
phenomena, such as infinite tidal forces at the surface
(cf. the geodesic deviation equation) which would destroy anything present there. We are, therefore, forced
to conclude that no physically relevant solution exists for
any polytropic EOS with 3/2 < < 2.
The following points about the result just presented
should be stressed:

exp(B(r)) = exp(A(r)) = 1 2m/r R0 r2 /12,

(235)
where and m are integration constants to be fixed by
requiring continuity of the metric coefficients across the
surface, which is implicitly defined by r = rout where
P = = 0. Using the definition of mtot (r) this gives, in
the exterior,
mtot (r) = m +

2
3
2F0 R0 rout
+ rout
R0 8mtot C
, (238)
=
16F0

29

This fact seems to have been missed in (Barraco and Hamity,

2000).

39
The only assumptions made regard the EOS and
the symmetries. Thus, the result applies to all
regimes ranging from Newtonian to strong gravity.
Let us now interpret these results. Obviously, one
could object to the use of the polytropic EOS. Even
though it is extensively used for simple stellar models,
it is not a very realistic description for stellar configurations. However, one does not necessarily have to refer
to stars in order to check whether the issue discussed
here leaves an observable signature. Consider two very
well known matter configurations which are exactly describes by a polytropic EOS: a monoatomic isentropic gas
and a degenerate non-relativistic electron gas. For both
of those cases = 5/3, which is well within the range
for which the singularities appear. Additionally, both of
these configuration can be very well described even with
Newtonian gravity. Yet, Palatini f (R) gravity fails to
provide a reasonable description. Therefore, one could
think of such matter configurations as gedanken experiments which reveal that Palatini f (R) gravity is at best
incomplete (Barausse et al., 2008a,b,c).
On the other hand, the use of the polytropic EOS requires a perfect fluid approximation for the description
of matter. One may, therefore, wish to question whether
the length scale on which the tidal forces become important is larger than the length scale for which the fluid
approximation breaks down (Kainulainen et al., 2007a).
However, quantitative results for tidal forces have been
given in (Barausse et al., 2008b), and it has been shown
that the length scales at which the tidal forces become
relevant are indeed larger than it would be required for
the fluid approximation to break down. The observable
consequences on stellar configurations have also been discussed there. To this, one could also add that a theory
which requires a full description of the microscopic structure of the system in order to provide a macroscopic description of the dynamics is not very appealing anyway.
In any case, it should be stated that the problem discussed is not specific to the polytropic EOS. The use of
the latter only simplifies the calculation and allows an
analytic approach. The root of the problem actually lies
with the differential structure of Palatini f (R) gravity.
Consider the field equations in the form (28): it is
not difficult to notice that these are second order partial
differential equations in the metric. However, since f is a
function of R, and R is an algebraic function of T due to
eq. (20), the right hand side of eq. (28) includes second
derivatives of T . Now, T , being the trace of the stress
energy tensor, will include up to first order derivatives of
the matter fields (assuming that the matter action has
to lead to second order field equations when varied with
respect to the matter fields). Consequently, eq. (28) can
be up to third order in the matter fields!
In GR and most of its alternatives, the field equations
are only of first order in the matter fields. This guarantees that gravity is a cumulative effect: the metric is
generated by an integral over the matter sources and,
therefore, any discontinuities (or even singularities) in

the latter and their derivatives, which are allowed, will

not become discontinuities or singularities of the metric,
which are not allowed [see (Barausse et al., 2008b) for
a detailed discussion]. This characteristic is not present
in Palatini f (R) gravity and creates an algebraic dependence of the metric on the matter fields.
The polytropic description not only does not cause this
problem but, as a matter of fact, it makes it less acute
than it is, simply because in the fluid approximation the
stress-energy tensor does not include derivatives of the
matter fields and effectively smoothes out the matter
distribution. Actually, the fact that the metric is extremely sensitive to rapid changes of the matter field has
been exhibited also in the interior of stars described by
realistic tabulated EOSs in (Barausse et al., 2008a).
One should not be puzzled by the fact that this awkward differential structure of Palatini f (R) gravity is not
manifest in the f (R) formulation of the theory [and the
field eqs. (18) and (19)]. We have already mentioned that
the independent connection is actually an auxiliary field
and the presence of auxiliary fields can always be misleading when it comes to the dynamics. In fact, it just
takes a closer look to realize that the Palatini f (R) action does not contain any derivatives of the metric and
is of only first order in the derivatives of the connection.
Now, given that the connection turns out to be an auxiliary field and can be algebraically related to derivatives
of the matter and of the metric, it no longer comes as a
surprise that the outcome is a theory with higher differential order in the matter than the metric.
By now, the fact that the post-Newtonian metric turns
out to be algebraically dependent on the density, as discussed in Sec. V.A.3, should no longer sound surprising:
it is merely a manifestation of the problem discussed here
in the weak field regime. The fact that it is unlikely that
the Cauchy problem will be well-formulated in the presence of matter also originates from the same feature of
Palatini f (R) gravity, as already mentioned. Similarly,
the fact that a theory which manifestly satisfies the metric postulates and, therefore, is expected to satisfy the
Equivalence Principle, actually exhibits unexpected phenomenology in local non-gravitational experiments and
conflicts with the Standard Model, as shown in Sec. VI.B,
also ceases to be a puzzle: the algebraic dependence of
the connection on the derivatives of matter fields (as the
former is an auxiliary field) makes the matter enter the
gravitational action through the back door. This introduces strong couplings between matter and gravity
and self-interactions of the matter fields which manifest
themselves in the local frame. Alternatively, if one completely eliminates the connection (or the scalar field in
the equivalent BransDicke representation) at the level
of the action, or attempts to write down an action which
leads to the field eqs. (28) directly through metric variation, then this action would have to include higher order
derivatives of the matter field and self-interactions in the
matter sector. In this sense, the f (R) representation
is simply misleading [see also (Sotiriou et al., 2007) for

40
a general discussion of representation issues in gravitational theories].

D. Gravitational waves in f (R) gravity

By now it is clear that the metric tensor of f (R) gravity contains, in addition to the usual massless spin 2
graviton, a massive scalar that shows up in gravitational
waves in the metric version of these theories (in the Palatini version, this scalar is not dynamical and does not
propagate). A scalar gravitational wave mode is familiar
from scalar-tensor gravity (Will, 1981), to which f (R)
gravity is equivalent. Because this scalar is massive,
it propagates at a speed lower than the speed of light
and of massless tensor modes and is, in principle, detectable in the arrival times of signals from an exploding
supernova when gravitational wave detectors are sufficiently sensitive [this possibility has been pointed out as a
discriminator between Tensor-Vector-Scalar theories and
GR (Kahya and Woodard, 2007)]. This massive scalar
mode is longitudinal and is of dipole nature to lowest order (Corda, 2007c; Will, 1981). The study of its generation, propagation, and detection falls within the purview
of scalar-tensor gravity (Will, 1981). The propagation of
gravitational waves in the specific model f (R) = Rn was
studied in (Mendoza and Rosas-Guevara, 2007) where
the massive scalar mode is, however, missed.
The generation of gravitational waves specifically in
f (R) gravity has not received much attention in the
literature. Even though the fact that the black hole
solutions of GR will also be solutions of metric f (R)
gravity (without the converse being true) implies that
determining the geometry around a black hole is unlikely to provide evidence for such modifications of gravity (Psaltis et al., 2008), solutions describing perturbed
black holes do behave differently and could, therefore,
leave a detectable imprint on gravitational wave radiation (Barausse and Sotiriou, 2008). Note the analogy to
the fact that cosmological FLRW solutions are shared by
most gravitational theories, but cosmological perturbations reveal more about the underlying theory of gravity
than the exact solutions themselves. Additionally, gravitational radiation from binary systems would probably
be more revealing than that coming from perturbed black
holes when it comes to modified gravity.
The detection of gravitational waves generated in the
theories f (R) = 1/R [already ruled out by Solar System
data (Barrow and Clifton, 2006; Clifton and Barrow,
2005a, 2006)] and f (R) = R + aR2 were studied in
(Corda, 2007a,c) and (Corda, 2007b), respectively.
The study of cosmological gravitational waves in f (R)
gravity is perhaps more promising than that of astrophysically generated waves. The stochastic gravitational
wave background produced in the early universe was analyzed in (Capozziello et al., 2007b, 2008). The authors of
this last reference consider the model f (R) = R1+ and
derive an evolution equation for the metric perturbations

hij = h(t) eikl k eij in a background

FLRW universe with
n
t
scale factor a(t) = a0 t0 :
+ (3n 2) h + k 2 a0
h
t

t
t0

2

n h = 0.

(240)

This can be solved in terms of Bessel functions; plots of

these wave amplitudes are reported in (Capozziello et al.,
2008) for various values of the parameter , but the limit
0 < 7.2 1019 obtained by (Barrow and Clifton,
2006; Clifton and Barrow, 2005a, 2006) leaves little hope
of detecting f (R) effects in the gravitational wave background.
(Ananda et al., 2008) give a covariant and gaugeinvariant description of gravitational waves in a perturbed FLRW universe filled with a barotropic perfect
fluid in the toy model f (R) = Rn . The perturbation
equations are solved (again, in terms of Bessel functions
of the first and second kind) in the approximation of
scales much larger or much smaller than the Hubble radius H 1 , finding a high sensitivity of the tensor modes
evolution to the value of the parameter n. In particular,
a tensor mode is found that grows during the radiationdominated era, with potential implications for detectability in advanced space interferometers. This study, and
others of this kind expected to appear in future literature,
are in the spirit of discriminating between dark energy
and dark gravity, or even between different f (R) theories
(if this class is taken seriously), when gravitational wave
observations will be available: as already remarked, this
is not possible by considering only unperturbed FLRW
solutions.
VII. SUMMARY AND CONCLUSIONS
A. Summary

While we have presented f (R) gravity as a class of toy

theories, various authors elevate modified gravity, in one
or the other of its incarnations corresponding to specific
choices of the function f (R), to the role of a fully realistic
model to be compared in detail with cosmological observations, and to be distinguished from other models. A
large fraction of the works in the literature is actually devoted to specific models corresponding to definite choices
of the function f (R), and to specific parametrizations.
Besides the power law and power series of R models
which we have already mentioned extensively, some other
typical examples are functions which contain terms such
as ln(R) (Nojiri and Odintsov, 2004b; Perez Bergliaffa,
2006) or eR (Abdelwahab et al., 2008; Bean et al., 2007;
Carloni et al., 2006; Song et al., 2007) or are more inm
volved functions of R, such as f (R) = Ra (R 1 ) +
n
b (R 2 ) with n, m, a, b > 0 (Nojiri and Odintsov,
2003a) . Some models have actually been tailored to
pass all or most of the known constraints, such as the
one proposed in (Starobinsky, 2007) where f (R) = R +

41
R0 [(1 + R2 /R02 )n 1] with n, > 0 and R0 being of
the order of H02 . Here we have tried to avoid considering
specific models and we have attempted to collect general, model-independent results, with the viewpoint that
these theories are to be seen more as toy theories than
definitive and realistic models.
We are now ready to summarize the main results on
f (R) gravity. On the theoretical side, we have explored
all three versions of f (R) gravity: metric, Palatini and
metric-affine. Several issues concerning dynamics, degrees of freedom, matter couplings, etc. have been extensively discussed. The dynamical equivalence between
both metric/Palatini f (R) gravity and BransDicke theory has been, and continues to be, a very useful tool to
study these theories given some knowledge of the aspects
of interest in scalar-tensor gravity. At the same time, the
study of f (R) gravity itself has provided new insight in
the two previously unexplored cases of BransDicke theory with 0 = 0 and 0 = 3/2. We have also considered
most of the applications of f (R) gravity to both cosmology and astrophysics. Finally, we have explored a large
number of possible ways to constrain f (R) theories and
check their viability. In fact, many avatars of f (R) have
been shown to be subject to potentially fatal troubles,
such as a grossly incorrect post-Newtonian limit, short
time scale instabilities, the absence of a matter era, conflict with particle physics or astrophysics, etc.
To avoid repetition, we will not attempt to summarize here all of the theoretical issues, the applications or
the constraints discussed. This, besides being redundant,
would not be very helpful to the reader, as, in most cases,
the insight gained cannot be summarized in a sentence or
two. Specifically, some of the constraints that have been
derived in the literature are not model or parametrization
independent (and the usefulness of some parametrizations is questionable). This does not allow for them to
be expressed in a straightforward manner through simple
mathematical equations applicable directly to a general
function f (R). Particular examples of such constraints
are those coming from cosmology (background evolution,
perturbations, etc.).
However, we have encountered cases in which clear-cut
viability criteria are indeed easy to derive. We would,
therefore, like to make a specific mention of those. A
brief list of quick-and-easy-to-use results is:
In metric f (R) gravity, the Dolgov-Kawasaki instability is avoided if and only if f (R) 0. The
stability condition of de Sitter space is expressed
by eq. (173).
Metric f (R) gravity might pass the weak-field limit
test and at the same time constitute an alternative
to dark energy only if the chameleon mechanism
is effectivethis restricts the possible forms of the
function f (R) in a way that can not be specified by
a simple formula.

tions. These conclusions are essentially modelindependent. (However, this theory could potentially be fixed by adding extra terms quadratic in
the Ricci and/or Riemann tensors, which would
raise the order of the equations.)
Metric-affine gravity as an extension of the Palatini
formalism is not sufficiently developed yet. At the
moment of writing, it is not clear whether it suffers
or not of the same problems that afflict the Palatini
formalism.
Of course, as already mentioned, the situation is often
more involved and cannot be summarized with a quick
recipe. We invite the reader to consult the previous sections and especially the references that they contain.
B. Extensions and new perspectives on f (R) gravity

We have treated f (R) gravity here as a toy theory

and, as stated in the Introduction, one of its merits is
its relative simplicity. However, we have seen a number
of viability issues related to such theories. One obvious
way to address this issue is to generalize the action even
further in order to avoid these problems, at the cost of
increased complexity. Several extensions of f (R) gravity
exist. Analyzing them in detail goes beyond the scope of
this review, but let us make a brief mention of the most
straightforward of them.
We have already discussed the possibility of having
higher order curvature invariants, such as R R , in
the action. In fact, from a dimensional analysis perspective, the terms R2 and R R should appear at the
same order. However, theories of this sort seem to be
burdened with what is called the Ostrogradski instability (Woodard, 2007). Ostrogradskis theorem states that
there is a linear instability in the Hamiltonians associated
with Lagrangians which depend upon higher than first
order derivatives in such a way that the dependence cannot be eliminated by partial integration (Ostrogradski,
1850). f (R) gravity seems to be the only case that manages to avoid this theorem (Woodard, 2007) and it obviously does not seem very appealing to extend it in a way
that will spoil this.30
The alert reader has probably noticed that the above
holds true only for metric f (R) gravity. In Palatini f (R)
gravity (and metric-affine f (R) gravity), as it was mentioned earlier, one could add more dynamics to the action
without having to worry about making it second order
in the fields. Recall that, in practice, the independent
connection is an auxiliary field. For instance, the term
R R still contains only first derivatives of the connection. In fact, since we have traced the root of some of

30

Palatini f (R) gravity suffered multiple deaths,

due to the differential structure of its field equa-

However, one could consider adding a function of the

GaussBonnet invariant G = R2 4R R R R
(Cognola et al., 2006; Nojiri and Odintsov, 2005).

42
the most crucial viability issues of Palatini f (R) gravity
to the lack of dynamics in the gravity sector, such generalizations could actually help by promoting the connection from the role of an auxiliary field to that of a truly
dynamical field (Barausse et al., 2008b). Such generalizations have been considered in (Li et al., 2007a).
Another extension of metric f (R) gravity that appeared recently is that in which the action includes also
an explicit coupling between R and the matter fields.
In (Bertolami et al., 2007; Bertolami and Paramos, 2007;
Boehmer et al., 2008) the following action was considered:


Z
f1 (R)
4
S = d x g
+ [1 + f2 (R)] Lm , (241)
2
where Lm is the matter Lagrangian and f1,2 are (a priori arbitrary) functions of the Ricci curvature R. Since
the matter is not minimally coupled to R, such theories
will not lead to energy conservation and will generically
exhibit a violation of the Equivalence Principle (which
could potentially be controlled by the parameter ).
The motivation for considering such an action spelled
out in (Bertolami et al., 2007) was that the nonconservation of energy could lead to extra forces, which in
turn might give rise to phenomenology similar to MOdified Newtonian Dynamics (MOND) (Milgrom, 1983)
on galactic scales. Other variants of this action have
also been considered elsewhere: in (Nojiri and Odintsov,
2004a), as an alternative to dark energy by setting
f1 (R) = R and keeping only the nonminimal coupling of
matter to the Ricci curvature; in (Dolgov and Kawasaki,
2003b; Mukohyama and Randall, 2004), where the idea
of making the kinetic term of a (minimally coupled)
scalar field dependent on the curvature, while keeping
f1 (R) = R was exploited in attempts to cure the cosmological constant problem; in (Bertolami and Paramos,
2007) the consequences of such a theory for stellar equilibrium were studied; finally, generalized constraints in
order to avoid the instability discussed in Sec. V.B.1 were
derived in (Faraoni, 2007a). The viability of theories with
such couplings between R and matter is still under investigation. However, the case in which both f1 and f2 are
linear has been shown to be non-viable (Sotiriou, 2008)
and, for the more general case of the action (241), serious doubts have been expressed (Sotiriou and Faraoni,
2008) on whether extra forces are indeed present in galactic environments and, therefore, whether this theory can
really account for the MOND-like phenomenology that
initially motivated its use in (Bertolami et al., 2007) as
a substitute for dark matter.
One could also consider extensions of f (R) gravity in
which extra fields appearing in the action couple to different curvature invariants. A simple example with a scalar
field is the action (159), which is sometimes dubbed extended quintessence (Perrotta et al., 2000), similarly to
the extended inflation realized with BransDicke theory.
However, such generalizations lie beyond the scope of this
review.

Finally, it is worth mentioning a different perspective

on f (R) gravity. It is common in the literature that we reviewed here to treat f (R) gravity as an exact theory: the
generalized action is used to derive field equations, the solutions of which describe the exact dynamics of the gravitational field (in spite of the fact that the action might
be only an approximation and the theory merely a toy
theory). A different approach (Bel and Zia, 1985; Simon,
1990) which was recently revived in (DeDeo and Psaltis,
2007) is that of treating metric f (R) as an effective field
theory. That is, to assume that the extra terms are an
artifact of some expansion of which we are considering
only the leading order terms. Now, when we consider a
correction to the usual EinsteinHilbert term, this correction has to be suppressed by some coefficient. This
approach assumes that this coefficient controls the order
of the expansion and, therefore, the field equations and
their solutions are only to be trusted to the order with
which that coefficient appears in the action (higher orders are to be discarded). Such an approach is based on
two assumptions: first, some power (or function) of the
coefficient of the correction considered should be present
in all terms of the expansion; second, the extra degrees
of freedom (which manifest themselves as higher order
derivatives in metric f (R) gravity) are actually an artifact of the expansion (and there would be a cancellation
if all orders where considered). This way, one can do
away with these extra degrees of freedom just by proper
power counting. Since many of the viability issues troubling higher order actions are related to the presence of
such degrees of freedom (e.g., classical instabilities), removing these degrees of freedom could certainly alleviate many problems (DeDeo and Psaltis, 2007). However,
the assumptions on which this approach is based should
not be underestimated either. For instance, early results
that showed renormalization of higher order actions were
based on an exact treatment, i.e., it is fourth order gravity that is renormalizable (Stelle, 1977). Even though,
from one hand, the effective field theory approach seems
very reasonable (these actions are regarded as low energy
limits of a more fundamental theory anyway), there is no
guarantee that extra degrees of freedom should indeed
not be present in a non-perturbative regime.

C. Concluding remarks

Our goal was to present a comprehensive but still thorough review of f (R) gravity in order to provide a starting
point for the reader less experienced in this field and a
reference guide for the expert. However, even though we
have attempted to cover all angles, no review can replace
an actual study of the literature itself. It seems inevitable
that certain aspects of f (R) might have been omitted, or
analyzed less than rigorously and, therefore, the reader
is urged to resort to the original sources.
Although many shortcomings of f (R) gravity have
been presented which may reduce the initial enthusiasm

43
with which one might have approached this field, the fact
that such theories are mostly considered as toy theories
should not be missed. The fast progress in this field,
especially in the last five years, is probably obvious by
now. And very useful lessons, which have helped significantly in the understanding of (classical) gravity, have
been learned in the study of f (R) gravity. In this sense,
the statement made in the Introduction that f (R) gravity is a very useful toy theory seems to be fully justified.
Remarkably, there are still unexplored aspects of f (R)
theories or their extensions, such as those mentioned in
the previous section, which can turn out to be fruitful.

Acknowledgments

It is a pleasure to acknowledge discussions with, and/or

comments by, L. Amendola, E. Barausse, J. Barrow, S.
Capozziello, P. Dunsby, F. Finelli, A. Frolov, W. Hu, T.
Jacobson, S. Joras, T. Koivisto, P. Labelle, N. LanahanTremblay, S. Liberati, F. Lobo, J. Miller, I. Navarro,
S. Odintsov, G. Olmo, A. Starobinsky, H. Stefancic, N.
Straumann, K. Van Acoleyen, M. Visser, and D. Vollick. The work of T. P. S. was supported by the National Science Foundation under grant PHYS-0601800.
V. F. acknowledges support by the Natural Sciences and
Engineering Research Council of Canada (NSERC) and
by a Bishops University Research Grant.

References
Abdalla, M. C. B., S. Nojiri, and S. D. Odintsov, 2005, Class.
Quant. Grav. 22, L35.
Abdelwahab, M., S. Carloni, and P. K. S. Dunsby, 2008, Class.
Quant. Grav. 25, 135002.
Accioly, A., S. Ragusa, E. C. de Rey Neto, and H. Mukai,
1999, N. Cimento B114, 595.
Alcubierre, M., et al., 2002a, Class. Quant. Grav. 19, 5017.
Alcubierre, M., et al., 2002b, eprint astro-ph/0204307.
Allemandi, G., A. Borowiec, and M. Francaviglia, 2004, Phys.
Rev. D70, 043524.
Allemandi, G., A. Borowiec, M. Francaviglia, and S. D.
Odintsov, 2005a, Phys. Rev. D72, 063505.
Allemandi, G., M. Francaviglia, M. L. Ruggiero, and
A. Tartaglia, 2005b, Gen. Rel. Grav. 37, 1891.
Allemandi, G., and M. L. Ruggiero, 2007, Gen. Rel. Grav.
39, 1381.
Amarzguioui, M., O. Elgaroy, D. F. Mota, and T. Multamaki,
2006, Astron. Astrophys. 454, 707.
Amendola, L., S. Capozziello, M. Litterio, and F. Occhionero,
1992, Phys. Rev. D45, 417.
Amendola, L., R. Gannouji, D. Polarski, and S. Tsujikawa,
2007a, Phys. Rev. D75, 083504.
Amendola, L., D. Polarski, and S. Tsujikawa, 2007b, Phys.
Rev. Lett. 98, 131302.
Amendola, L., D. Polarski, and S. Tsujikawa, 2007c, Int. J.
Mod. Phys. D16, 1555.
Amendola, L., and S. Tsujikawa, 2008, Phys. Lett. B660,
125.

Ananda, K. N., S. Carloni, and P. K. S. Dunsby, 2008, Phys.

Rev. D77, 024033.
Anderson, J. L., 1971, Phys. Rev. D3, 1689.
Appleby, S. A., and R. A. Battye, 2007, Phys. Lett. B654, 7.
Appleby, S. A., and R. A. Battye, 2008, eprint 0803.1081.
Astier, P., et al. (The SNLS), 2006, Astron. Astrophys. 447,
31.
Baghram, S., M. Farhang, and S. Rahvar, 2007, Phys. Rev.
D75, 044024.
Bahcall, N. A., J. P. Ostriker, S. Perlmutter, and P. J. Steinhardt, 1999, Science 284, 1481.
Bamba, K., and S. D. Odintsov, 2008, JCAP 0804, 024.
Barausse, E., and T. P. Sotiriou, 2008, eprint 0803.3433.
Barausse, E., T. P. Sotiriou, and J. C. Miller, 2008a, Class.
Quant. Grav. 25, 062001.
Barausse, E., T. P. Sotiriou, and J. C. Miller, 2008b, Class.
Quant. Grav. 25, 105008.
Barausse, E., T. P. Sotiriou, and J. C. Miller, 2008c, eprint
0801.4852.
Barber, G. A., 2003, eprint gr-qc/0302088.
Bardeen, J. M., 1980, Phys. Rev. D22, 1882.
Barraco, D. E., E. Dominguez, and R. Guibert, 1999, Phys.
Rev. D60, 044012.
Barraco, D. E., and V. H. Hamity, 1998, Phys. Rev. D57,
954.
Barraco, D. E., and V. H. Hamity, 2000, Phys. Rev. D62,
044027.
Barris, B. J., et al., 2004, Astrophys. J. 602, 571.
Barrow, J. D., 1988, Nucl. Phys. B296, 697.
Barrow, J. D., and T. Clifton, 2006, Class. Quant. Grav. 23,
L1.
Barrow, J. D., and S. Cotsakis, 1988, Phys. Lett. B214, 515.
Barrow, J. D., and S. Cotsakis, 1989, Phys. Lett. B232, 172.
Barrow, J. D., and S. Cotsakis, 1991, Phys. Lett. B258, 299.
Barrow, J. D., and S. Hervik, 2006a, Phys. Rev. D73, 023007.
Barrow, J. D., and S. Hervik, 2006b, Phys. Rev. D74, 124017.
Barrow, J. D., and A. C. Ottewill, 1983, J. Phys. A16, 2757.
Barrow, J. D., and F. J. Tipler, 1986, The Anthropic Cosmological Principle (Clarendon Press, Oxford).
Bean, R., D. Bernat, L. Pogosian, A. Silvestri, and M. Trodden, 2007, Phys. Rev. D75, 064020.
Bekenstein, J. D., 2004, Phys. Rev. D70, 083509.
Bel, L., and H. S. Zia, 1985, Phys. Rev. D32, 3128.
Bergmann, P. G., 1968, Int. J. Theor. Phys. 1, 25.
Bernal, A., and F. S. Guzman, 2006a, AIP Conf. Proc. 841,
441.
Bernal, A., and F. S. Guzman, 2006b, Phys. Rev. D74,
103002.
Bernal, A., and F. S. Guzman, 2006c, Phys. Rev. D74,
063504.
Bernal, A., and T. Matos, 2005, AIP Conf. Proc. 758, 161.
Bertolami, O., 1987, Phys. Lett. B186, 161.
Bertolami, O., C. G. Boehmer, T. Harko, and F. S. N. Lobo,
2007, Phys. Rev. D75, 104016.
Bertolami, O., and J. Paramos, 2007, eprint arXiv:0709.3988
[astro-ph].
Bertotti, B., L. Iess, and P. Tortora, 2003, Nature 425, 374.
Birrell, N. D., and P. C. W. Davies, 1982, Quantum Fields
in Curved Spacetime (Cambridge University Press, Cambridge).
Boehmer, C. G., T. Harko, and F. S. N. Lobo, 2007a, eprint
0709.0046.
Boehmer, C. G., T. Harko, and F. S. N. Lobo, 2008, JCAP
0803, 024.

44
Boehmer, C. G., L. Hollenstein, and F. S. N. Lobo, 2007b,
Phys. Rev. D76, 084005.
Borowiec, A., M. Ferraris, M. Francaviglia, and I. Volovich,
1994, Gen. Rel. Grav. 26, 637.
Borowiec, A., M. Ferraris, M. Francaviglia, and I. Volovich,
1998, Class. Quant. Grav. 15, 43.
Bosma, A., 1978, The distribution and kinematics of neutral
hydrogen in spiral galaxies of various morphological types,
Ph.D. thesis, Rejksuniversiteit Groningen.
Brandenberger, R. H., 1992, eprint gr-qc/9210014.
Brandenberger, R. H., 1993, eprint gr-qc/9302014.
Brandenberger, R. H., 1995, eprint gr-qc/9509059.
Brandenberger, R. H., V. F. Mukhanov, and A. Sornborger,
1993, Phys. Rev. D48, 1629.
Brans, C., and R. H. Dicke, 1961, Phys. Rev. 124, 925.
Bronnikov, K. A., and M. S. Chernakova, 2005a, Russ. Phys.
J. 48, 940.
Bronnikov, K. A., and M. S. Chernakova, 2005b, eprint grqc/0503025.
Bronnikov, K. A., and M. S. Chernakova, 2005c, Grav. Cosmol. 11, 305.
Brookfield, A. W., C. van de Bruck, and L. M. H. Hall, 2006,
Phys. Rev. D74, 064028.
Buchbinder, I. L., S. D. Odintsov, and I. L. Shapiro, 1992, Effective Actions in Quantum Gravity (IOP Publishing, Bristol).
Buchdahl, H. A., 1970, Mon. Not. Roy. Astron. Soc. 150, 1.
Burles, S., K. M. Nollett, and M. S. Turner, 2001, Astrophys.
J. 552, L1.
Burton, H., and R. B. Mann, 1998a, Phys. Rev. D57, 4754.
Burton, H., and R. B. Mann, 1998b, Class. Quant. Grav. 15,
1375.
Bustelo, A. J., and D. E. Barraco, 2007, Class. Quant. Grav.
24, 2333.
Calcagni, G., B. de Carlos, and A. De Felice, 2006, Nucl.
Phys. B752, 404.
Caldwell, R. R., R. Dave, and P. J. Steinhardt, 1998, Phys.
Rev. Lett. 80, 1582.
Capozziello, S., V. F. Cardone, S. Carloni, and A. Troisi,
2004, Phys. Lett. A326, 292.
Capozziello, S., V. F. Cardone, S. Carloni, and A. Troisi,
2005a, AIP Conf. Proc. 751, 54.
Capozziello, S., V. F. Cardone, and A. Troisi, 2005b, Phys.
Rev. D71, 043503.
Capozziello, S., V. F. Cardone, and A. Troisi, 2006a, JCAP
0608, 001.
Capozziello, S., V. F. Cardone, and A. Troisi, 2007a, Mon.
Not. Roy. Astron. Soc. 375, 1423.
Capozziello, S., S. Carloni, and A. Troisi, 2003, Recent Res.
Dev. Astron. Astrophys. 1, 625.
Capozziello, S., C. Corda, and M. De Laurentis, 2007b, Mod.
Phys. Lett. A22, 1097.
Capozziello, S., and A. De Felice, 2008, eprint 0804.2163.
Capozziello, S., M. De Laurentis, and M. Francaviglia, 2008,
Astropart. Phys. 29, 125.
Capozziello, S., and M. Francaviglia, 2008, Gen. Rel. Grav.
40, 357.
Capozziello, S., S. Nojiri, and S. D. Odintsov, 2006b, Phys.
Lett. B632, 597.
Capozziello, S., S. Nojiri, S. D. Odintsov, and A. Troisi, 2006c,
Phys. Lett. B639, 135.
Capozziello, S., F. Occhionero, and L. Amendola, 1993, Int.
J. Mod. Phys. D1, 615.
Capozziello, S., A. Stabile, and A. Troisi, 2006d, Mod. Phys.

Lett. A21, 2291.

Capozziello, S., A. Stabile, and A. Troisi, 2007c, Class. Quant.
Grav. 24, 2153.
Capozziello, S., A. Stabile, and A. Troisi, 2007d, eprint
arXiv:0709.0891 [gr-qc].
Capozziello, S., A. Stabile, and A. Troisi, 2007e, Phys. Rev.
D76, 104019.
Capozziello, S., and A. Troisi, 2005, Phys. Rev. D72, 044022.
Capozziello, S., A. Troisi, and V. F. Cardone, 2007f, New
Astron. Rev. 51, 341.
Capozziello,
S., and S. Tsujikawa,
2007,
eprint
arXiv:0712.2268 [gr-qc].
Carloni, S., S. Capozziello, J. A. Leach, and P. K. S. Dunsby,
2008a, Class. Quant. Grav. 25, 035008.
Carloni, S., and P. K. S. Dunsby, 2007, J. Phys. A40, 6919.
Carloni, S., P. K. S. Dunsby, S. Capozziello, and A. Troisi,
2005, Class. Quant. Grav. 22, 4839.
Carloni, S., P. K. S. Dunsby, and D. M. Solomons, 2006, Class.
Quant. Grav. 23, 1913.
Carloni, S., P. K. S. Dunsby, and A. Troisi, 2008b, Phys. Rev.
D77, 024024.
Carloni, S., A. Troisi, and P. K. S. Dunsby, 2007, eprint
0706.0452.
Carroll, S. M., 1998, Phys. Rev. Lett. 81, 3067.
Carroll, S. M., 2001a, Living Rev. Rel. 4, 1.
Carroll, S. M., 2001b, eprint astro-ph/0107571.
Carroll, S. M., V. Duvvuri, M. Trodden, and M. S. Turner,
2004, Phys. Rev. D70, 043528.
Carroll, S. M., and M. Kaplinghat, 2002, Phys. Rev. D65,
063507.
Carroll, S. M., I. Sawicki, A. Silvestri, and M. Trodden, 2006,
New J. Phys. 8, 323.
Carroll, S. M., et al., 2005, Phys. Rev. D71, 063513.
Cartan, E., 1922, C. R. Acad. Sci. 174, 593.
Cartan, E., 1923, Ann. Ec. Norm. Sup. 40, 325.
Cartan, E., 1924, Ann. Ec. Norm. Sup. 41, 1.
Carter, B., 1974, in International Astronomical Union Symposium 63: Confrontation of Cosmological Theories with
Observational Data, edited by M. S. Longair (Dordrecht,
Reidel).
Carvalho, F. C., E. M. Santos, J. S. Alcaniz, and J. Santos,
2008, eprint 0804.2878.
Cecotti, S., 1987, Phys. Lett. B190, 86.
Cembranos, J. A. R., 2006, Phys. Rev. D73, 064029.
Cervantes-Cota, J. L., M. A. Rodriguez-Meza, R. Gabbasov,
and J. Klapp, 2007a, eprint 0704.2579.
Cervantes-Cota, J. L., M. A. Rodriguez-Meza, and D. Nunez,
2007b, J. Phys. Conf. Ser. 91, 012007.
Chiba, T., 2003, Phys. Lett. B575, 1.
Chiba, T., 2005, JCAP 0503, 008.
Chiba, T., T. L. Smith, and A. L. Erickcek, 2007, Phys. Rev.
D75, 124014.
Clarkson, C., M. Cortes, and B. A. Bassett, 2007, JCAP
0708, 011.
Clarkson, C. A., and R. Barrett, 1999, Class. Quant. Grav.
16, 3781.
Clarkson, C. A., A. A. Coley, and E. S. D. ONeill, 2001,
Phys. Rev. D64, 063510.
Clarkson, C. A., A. A. Coley, E. S. D. ONeill, R. A. Sussman,
and R. K. Barrett, 2003, Gen. Rel. Grav. 35, 969.
Clifton, T., 2006a, eprint gr-qc/0610071.
Clifton, T., 2006b, Class. Quant. Grav. 23, 7445.
Clifton, T., 2007, Class. Quant. Grav. 24, 5073.
Clifton, T., 2008, Phys. Rev. D77, 024041.

45
Clifton, T., and J. D. Barrow, 2005a, Phys. Rev. D72, 103005.
Clifton, T., and J. D. Barrow, 2005b, Phys. Rev. D72,
123003.
Clifton, T., and J. D. Barrow, 2006, Class. Quant. Grav. 23,
2951.
Cocke, W. J., and J. M. Cohen, 1968, J. Math. Phys 9, 971.
Codello, A., and R. Percacci, 2006, Phys. Rev. Lett. 97,
221301.
Cognola, G., E. Elizalde, S. Nojiri, S. D. Odintsov, and
S. Zerbini, 2005, JCAP 0502, 010.
Cognola, G., E. Elizalde, S. Nojiri, S. D. Odintsov, and
S. Zerbini, 2006, Phys. Rev. D73, 084007.
Cognola, G., M. Gastaldi, and S. Zerbini, 2008, Int. J. Theor.
Phys. 47, 898.
Cognola, G., and S. Zerbini, 2006, J. Phys. A39, 6245.
Cognola, G., and S. Zerbini, 2008, eprint arXiv:0802.3967
[hep-th].
Coley, A. A., 2003, Dynamical systems and cosmology
(Kluwer, Dordrecht).
Comelli, D., 2005, Phys. Rev. D72, 064018.
Corda, C., 2007a, eprint 0710.2605.
Corda, C., 2007b, eprint 0711.4917.
Corda, C., 2007c, JCAP 0704, 009.
de la Cruz-Dombriz, A., and A. Dobado, 2006, Phys. Rev.
D74, 087501.
de la Cruz-Dombriz, A., A. Dobado, and A. L. Maroto, 2008,
eprint arXiv:0802.2999 [astro-ph].
Dabrowski, M. P., T. Denkiewicz, and D. Blaschke, 2007, Annalen Phys. 16, 237.
Davidson, A., 2005, Class. Quant. Grav. 22, 1119.
De Felice, A., 2007, J. Phys. A40, 7061.
De Felice, A., and M. Hindmarsh, 2007, JCAP 0706, 028.
De Felice, A., M. Hindmarsh, and M. Trodden, 2006, JCAP
0608, 005.
DeDeo, S., and D. Psaltis, 2007, eprint 0712.3939.
Deser, S., 1970, Ann. Phys. (NY) 59, 248.
Deser, S., and B. Tekin, 2002, Phys. Rev. Lett. 89, 101101.
Deser, S., and B. Tekin, 2003a, Phys. Rev. D67, 084009.
Deser, S., and B. Tekin, 2003b, Class. Quant. Grav. 20, 4877.
Deser, S., and B. Tekin, 2007, Phys. Rev. D75, 084032.
Dick, R., 2004, Gen. Rel. Grav. 36, 217.
Dicke, R. H., 1962, Phys. Rev. 125, 2163.
Dolgov, A., and D. N. Pelliccia, 2006, Nucl. Phys. B734, 208.
Dolgov, A. D., and M. Kawasaki, 2003a, Phys. Lett. B573,
1.
Dolgov, A. D., and M. Kawasaki, 2003b, eprint astroph/0307442.
Dominguez, A. E., and D. E. Barraco, 2004, Phys. Rev. D70,
043505.
Dvali, G. R., G. Gabadadze, and M. Porrati, 2000, Phys. Lett.
B485, 208.
Easson, D. A., 2004, Int. J. Mod. Phys. A19, 5343.
Eddington, A. S., 1923, The Mathematical Theory of Relativity (Cambridge University Press, Cambridge).
Ehlers, J., P. Geren, and R. K. Sachs, 1968, J. Math. Phys.
9, 1344.
Eisenstein, D. J., et al. (SDSS), 2005, Astrophys. J. 633, 560.
Eling, C., R. Guedens, and T. Jacobson, 2006, Phys. Rev.
Lett. 96, 121301.
Ellis, G. F. R., and M. Bruni, 1989, Phys. Rev. D40, 1804.
Ellis, G. F. R., M. Bruni, and J. Hwang, 1990, Phys. Rev.
D42, 1035.
Ellis, G. F. R., J. Hwang, and M. Bruni, 1989, Phys. Rev.
D40, 1819.

Ellis, G. F. R., D. R. Matravers, and R. Treciokas, 1983a,

Ann. Phys. (NY) 150, 455.
Ellis, G. F. R., R. Treciokas, and D. R. Matravers, 1983b,
Ann. Phys. (NY) 150, 487.
Ellis, J. R., 2002, eprint astro-ph/0204059.
Erickcek, A. L., T. L. Smith, and M. Kamionkowski, 2006,
Phys. Rev. D74, 121501.
Evans, J. D., L. M. H. Hall, and P. Caillol, 2007, eprint
0711.3695.
Exirifard, Q., and M. M. Sheikh-Jabbari, 2008, Phys. Lett.
B661, 158.
Faraoni, V., 2004a, Cosmology in Scalar-Tensor Gravity
(Kluwer Academic, Dordrecht).
Faraoni, V., 2004b, Phys. Rev. D69, 123520.
Faraoni, V., 2004c, Phys. Rev. D70, 044037.
Faraoni, V., 2005a, Phys. Rev. D72, 061501.
Faraoni, V., 2005b, Ann. Phys. 317, 366.
Faraoni, V., 2006a, Phys. Rev. D74, 104017.
Faraoni, V., 2006b, Phys. Rev. D74, 023529.
Faraoni, V., 2007a, Phys. Rev. D76, 127501.
Faraoni, V., 2007b, Phys. Rev. D75, 067302.
Faraoni, V., E. Gunzig, and P. Nardone, 1999, Fund. Cosmic
Phys. 20, 121.
Faraoni, V., and N. Lanahan-Tremblay, 2007, eprint
arXiv:0712.3252 [gr-qc].
Faraoni, V., and S. Nadeau, 2005, Phys. Rev. D72, 124005.
Faraoni, V., and S. Nadeau, 2007, Phys. Rev. D75, 023501.
Faulkner, T., M. Tegmark, E. F. Bunn, and Y. Mao, 2007,
Phys. Rev. D76, 063505.
Fay, S., S. Nesseris, and L. Perivolaropoulos, 2007a, Phys.
Rev. D76, 063504.
Fay, S., R. Tavakol, and S. Tsujikawa, 2007b, Phys. Rev. D75,
063509.
Ferraris, M., M. Francaviglia, and G. Magnano, 1988, Class.
Quant. Grav. 5, L95.
Ferraris, M., M. Francaviglia, and C. Reina, 1982, Gen. Rel.
Grav. 14, 243.
Ferraris, M., M. Francaviglia, and I. Volovich, 1992, eprint
gr-qc/9303007.
Ferraris, M., M. Francaviglia, and I. Volovich, 1994, Class.
Quant. Grav. 11, 1505.
Filippenko, A. V., and A. G. Riess, 1998, Phys. Rept. 307,
31.
Flanagan, E. E., 2003, Class. Quant. Grav. 21, 417.
Flanagan, E. E., 2004a, Class. Quant. Grav. 21, 3817.
Flanagan, E. E., 2004b, Phys. Rev. Lett. 92, 071101.
Frolov, A. V., 2008, eprint 0803.2500.
Fujii, Y., 1982, Phys. Rev. D26, 2580.
Gibbons, G. W., and S. W. Hawking, 1977, Phys. Rev. D
15(10), 2752.
Goheer, N., J. A. Leach, and P. K. S. Dunsby, 2007, Class.
Quant. Grav. 24, 5689.
Goheer, N., J. A. Leach, and P. K. S. Dunsby, 2008, Class.
Quant. Grav. 25, 035013.
Goswami, R., N. Goheer, and P. K. S. Dunsby, 2008, eprint
0804.3528.
Gunther, U., P. Moniz, and A. Zhuk, 2002, Phys. Rev. D66,
044014.
Gunther, U., P. Moniz, and A. Zhuk, 2003, Phys. Rev. D68,
044010.
Gunther, U., A. Zhuk, V. B. Bezerra, and C. Romero, 2005,
Class. Quant. Grav. 22, 3135.
Guth, A. H., 1981, Phys. Rev. D23, 347.
Hadjimichef, D., and F. Kokubun, 1997, Phys. Rev. D55,

46
733.
Hehl, F. W., and G. D. Kerling, 1978, Gen. Rel. Grav. 9, 691.
Hehl, F. W., J. D. McCrea, E. W. Mielke, and Y. Neeman,
1995, Phys. Rept. 258, 1.
Hehl, F. W., P. Von Der Heyde, G. D. Kerlick, and J. M.
Nester, 1976, Rev. Mod. Phys. 48, 393.
Hindawi, A., B. A. Ovrut, and D. Waldram, 1996, Phys. Rev.
D53, 5597.
Hoyle, C. D., et al., 2001, Phys. Rev. Lett. 86, 1418.
Hu, W., and I. Sawicki, 2007a, Phys. Rev. D76, 104043.
Hu, W., and I. Sawicki, 2007b, Phys. Rev. D76, 064004.
Hwang, J. C., 1990a, Class. Quant. Grav. 7, 1613.
Hwang, J. C., 1990b, Phys. Rev. D42, 2601.
Hwang, J.-C., 1997, Class. Quant. Grav. 14, 3327.
Hwang, J.-C., 1998, Class. Quant. Grav. 15, 1401.
Hwang, J.-C., and H. Noh, 1996, Phys. Rev. D54, 1460.
Hwang, J.-C., and H. Noh, 2002, Phys. Rev. D65, 023512.
Iglesias, A., N. Kaloper, A. Padilla, and M. Park, 2007, Phys.
Rev. D76, 104001.
Iorio, L., 2007, eprint arXiv:0710.0022 [gr-qc].
Iorio, L., and M. L. Ruggiero, 2007a, Int. J. Mod. Phys. A22,
5379.
Iorio, L., and M. L. Ruggiero, 2007b, eprint arXiv:0711.0256
[gr-qc].
Israel, W., 1966, Nuovo Cimento 44B, 1, see erratum (Israel,
1967).
Israel, W., 1967, Nuovo Cimento 49B, 463.
Jacobson, T., G. Kang, and R. C. Myers, 1994, Phys. Rev.
D49, 6587.
Jacobson, T., G. Kang, and R. C. Myers, 1995, Phys. Rev.
D52, 3518.
Jacobson, T., and D. Mattingly, 2001, Phys. Rev. D64,
024028.
Jhingan, S., et al., 2008, eprint 0803.2613.
Jin, X.-H., D.-J. Liu, and X.-Z. Li, 2006, eprint astroph/0610854.
Kahn, F. D., and L. Woltjer, 1959, Astrophys. J. 130, 705.
Kahya, E. O., and R. P. Woodard, 2007, Phys. Lett. B652,
213.
Kainulainen, K., J. Piilonen, V. Reijonen, and D. Sunhede,
2007a, Phys. Rev. D76, 024020.
Kainulainen, K., V. Reijonen, and D. Sunhede, 2007b, Phys.
Rev. D76, 043503.
Khoury, J., and A. Weltman, 2004a, Phys. Rev. D69, 044026.
Khoury, J., and A. Weltman, 2004b, Phys. Rev. Lett. 93,
171104.
Kibble, T. W. B., 1961, J. Math. Phys. 2, 212.
Kluske, S., and H.-J. Schmidt, 1996, Astron. Nachr. 317, 337.
Kneller, J. P., and G. Steigman, 2004, New J. Phys. 6, 117.
Knop, R. A., et al. (Supernova Cosmology Project), 2003,
Astrophys. J. 598, 102.
Knox, L., Y.-S. Song, and J. A. Tyson, 2006, Phys. Rev. D74,
023512.
Koivisto, T., 2006a, Class. Quant. Grav. 23, 4289.
Koivisto, T., 2006b, Phys. Rev. D73, 083517.
Koivisto, T., 2007, Phys. Rev. D76, 043527.
Koivisto, T., and H. Kurki-Suonio, 2006, Class. Quant. Grav.
23, 2355.
Kolb, E. W., and M. S. Turner, 1992, The Early Universe
(Addison-Wesley, California).
Koyama, K., and R. Maartens, 2006, JCAP 0601, 016.
Lambiase, G., and G. Scarpetta, 2006, Phys. Rev. D74,
087504.
Lanahan-Tremblay, N., and V. Faraoni, 2007, Class. Quant.

Grav. 24, 5667.

Leach, J. A., S. Carloni, and P. K. S. Dunsby, 2006, Class.
Quant. Grav. 23, 4915.
Leach, J. A., P. K. S. Dunsby, and S. Carloni, 2007, eprint
gr-qc/0702122.
Lee, S., 2007, eprint 0710.2395.
Lee, S., 2008, eprint 0801.4606.
Li, B., and J. D. Barrow, 2007, Phys. Rev. D75, 084010.
Li, B., J. D. Barrow, and D. F. Mota, 2007a, Phys. Rev. D76,
104047.
Li, B., K. C. Chan, and M. C. Chu, 2007b, Phys. Rev. D76,
024002.
Li, B., and M. C. Chu, 2006, Phys. Rev. D74, 104010.
Li, B., D. F. Mota, and D. J. Shaw, 2008, eprint 0801.0603.
Linde, A., 1990, Particle Physics and Inflationary Cosmology
(Harwood Academic Publishers, Switzerland).
Lue, A., R. Scoccimarro, and G. Starkman, 2004, Phys. Rev.
D69, 044005.
Maartens, R., 2004, Living Rev. Rel. 7, 7.
Maartens, R., and D. R. Taylor, 1994, Gen. Rel. Grav. 26,
599.
Maeda, K.-i., 1988, Phys. Rev. D37, 858.
Maeda, K.-i., J. A. Stein-Schabes, and T. Futamase, 1989,
Phys. Rev. D39, 2848.
Magnano, G., and L. M. Sokolowski, 1994, Phys. Rev. D50,
5039.
Martins, C. F., and P. Salucci, 2007, Mon. Not. Roy. Astron.
Soc. 381, 1103.
Matos, T., and F. S. Guzman, 2000, Annalen Phys. 9, S1.
Matos, T., F. S. Guzman, L. A. Urena-Lopez, and D. Nunez,
2001, eprint astro-ph/0102419.
Matos, T., and L. A. Urena-Lopez, 2001, Phys. Rev. D63,
063506.
Matos, T., and L. A. Urena-Lopez, 2007, Gen. Rel. Grav. 39,
1279.
Mendoza, S., and Y. M. Rosas-Guevara, 2007, Astron. Astrophys. 472, 367.
Meng, X.-H., and P. Wang, 2003, eprint astro-ph/0308284.
Meng, X.-H., and P. Wang, 2004a, Gen. Rel. Grav. 36, 1947.
Meng, X.-H., and P. Wang, 2004b, Class. Quant. Grav. 21,
2029.
Meng, X. H., and P. Wang, 2005, Class. Quant. Grav. 22, 23.
Mignemi, S., and D. L. Wiltshire, 1992, Phys. Rev. D46,
1475.
Milgrom, M., 1983, Astrophys. J. 270, 365.
Misner, C. W., 1968, Astrophys. J. 151, 431.
Misner, C. W., K. S. Thorne, and J. A. Wheeler, 1973, Gravitation (Freeman Press, San Francisco).
Modak, B., A. Ghose, and R. N. Bose, 2005, Gen. Rel. Grav.
37, 985.
Mohseni Sadjadi, H., 2007, eprint 0710.3308.
Moore, B., 2001, eprint astro-ph/0103100.
Mukhanov, V. F., 2003, eprint astro-ph/0303077.
Mukhanov, V. F., and R. H. Brandenberger, 1992, Phys. Rev.
Lett. 68, 1969.
Mukohyama, S., and L. Randall, 2004, Phys. Rev. Lett. 92,
211302.
Muller, V., H. J. Schmidt, and A. A. Starobinsky, 1990, Class.
Quant. Grav. 7, 1163.
Multamaki, T., A. Putaja, I. Vilja, and E. C. Vagenas, 2008,
Class. Quant. Grav. 25, 075017.
Multamaki, T., and I. Vilja, 2006a, Phys. Rev. D73, 024018.
Multamaki, T., and I. Vilja, 2006b, Phys. Rev. D74, 064022.
Multamaki, T., and I. Vilja, 2007a, eprint arXiv:0709.3422

47
[astro-ph].
Multamaki, T., and I. Vilja, 2007b, Phys. Rev. D76, 064021.
Myers, R. C., and J. Z. Simon, 1988, Phys. Rev. D38, 2434.
Myers, R. C., and J. Z. Simon, 1989, Gen. Rel. Grav. 21, 761.
Nakamura, R., M. Hashimoto, S. Gamow, and K. Arai, 2006,
Astron. Astrophys. 448, 23.
Navarro, I., and K. Van Acoleyen, 2005, Phys. Lett. B622, 1.
Navarro, I., and K. Van Acoleyen, 2006, JCAP 0603, 008.
Navarro, I., and K. Van Acoleyen, 2007, JCAP 0702, 022.
Noakes, D. R., 1983, J. Math. Phys. 24, 1846.
Nojiri, S., 2004, TSPU Vestnik 44N7, 49.
Nojiri, S., and S. D. Odintsov, 2003a, Phys. Rev. D68,
123512.
Nojiri, S., and S. D. Odintsov, 2003b, Phys. Lett. B576, 5.
Nojiri, S., and S. D. Odintsov, 2004a, Phys. Lett. B599, 137.
Nojiri, S., and S. D. Odintsov, 2004b, Gen. Rel. Grav. 36,
1765.
Nojiri, S., and S. D. Odintsov, 2005, Phys. Lett. B631, 1.
Nojiri, S., and S. D. Odintsov, 2006, Phys. Rev. D74, 086005.
Nojiri, S., and S. D. Odintsov, 2007a, Int. J. Geom. Meth.
Mod. Phys. 4, 115.
Nojiri, S., and S. D. Odintsov, 2007b, J. Phys. Conf. Ser. 66,
012005.
Nojiri, S., and S. D. Odintsov, 2007c, J. Phys. A40, 6725.
Nojiri, S., and S. D. Odintsov, 2007d, Phys. Lett. B657, 238.
Nojiri, S., and S. D. Odintsov, 2008a, eprint 0801.4843.
Nojiri, S., and S. D. Odintsov, 2008b, Phys. Rev. D77,
026007.
Nojiri, S., and S. D. Odintsov, 2008c, Phys. Lett. B659, 821.
Nojiri, S., and S. D. Odintsov, 2008d, eprint 0804.3519.
Nordtvedt, J., Kenneth, 1970, Astrophys. J. 161, 1059.
Novello, M., and S. E. P. Bergliaffa, 2008, eprint 0802.1634.
Nunez, A., and S. Solganik, 2004, eprint hep-th/0403159.
OHanlon, J., 1972a, J. Phys. A 5, 803.
OHanlon, J., 1972b, Phys. Rev. Lett. 29, 137.
OHanlon, J., and B. Tupper, 1972, Nuovo Cimento B 7, 305.
Olmo, G. J., 2005a, Phys. Rev. Lett. 95, 261102.
Olmo, G. J., 2005b, Phys. Rev. D72, 083505.
Olmo, G. J., 2007, Phys. Rev. D75, 023511.
Olmo, G. J., 2008, Phys. Rev. D77, 084021.
Olmo, G. J., and P. Singh, 2009, JCAP 0901, 030.
Ostriker, J. P., and P. J. Steinhardt, 1995, Nature 377, 600.
Ostrogradski, M., 1850, Mem. Ac. St. Petersbourg VI 4, 385.
Paul, B. C., and D. Paul, 2006, Phys. Rev. D74, 084015.
Paul, D., and B. C. Paul, 2005, Phys. Rev. D72, 064012.
Peebles, P. J. E., and B. Ratra, 1988, Astrophys. J. 325, L17.
Perez Bergliaffa, S. E., 2006, Phys. Lett. B642, 311.
Perlmutter, S., et al. (Supernova Cosmology Project), 1998,
Nature 391, 51.
Perrotta, F., C. Baccigalupi, and S. Matarrese, 2000, Phys.
Rev. D61, 023507.
Persic, M., P. Salucci, and F. Stel, 1996, Mon. Not. Roy.
Astron. Soc. 281, 27.
Pogosian, L., and A. Silvestri, 2008, Phys. Rev. D77, 023503.
Psaltis, D., D. Perrodin, K. R. Dienes, and I. Mocioiu, 2008,
Phys. Rev. Lett. 100, 091101.
Querella, L., 1998, eprint gr-qc/9902044.
Rador, T., 2007, Phys. Rev. D75, 064033.
Rajaraman, A., 2003, eprint astro-ph/0311160.
Ratra, B., and P. J. E. Peebles, 1988, Phys. Rev. D37, 3406.
Reula, O. A., 1998, Living Rev. Rel. 1, 3.
Riegert, R. J., 1984, Phys. Lett. A105, 110.
Riess, A. G., et al. (Supernova Search Team), 1998, Astron.
J. 116, 1009.

Riess, A. G., et al., 1999, Astron. J. 118, 2675.

Riess, A. G., et al. (Supernova Search Team), 2004, Astrophys. J. 607, 665.
Rippl, S., H. van Elst, R. K. Tavakol, and D. Taylor, 1996,
Gen. Rel. Grav. 28, 193.
Rodriguez-Meza, M. A., and J. L. Cervantes-Cota, 2004, Mon.
Not. Roy. Astron. Soc. 350, 671.
Rodriguez-Meza, M. A., J. L. Cervantes-Cota, M. I. Pedraza,
J. F. Tlapanco, and D. la Calleja, E. M., 2005, Gen. Rel.
Grav. 37, 823.
Rodriguez-Meza, M. A., A. X. Gonzalez-Morales, R. F. Gabbasov, and J. L. Cervantes-Cota, 2007, J. Phys. Conf. Ser.
91, 012012.
Rubin, V., and W. K. J. Ford, 1970, Astrophys. J. 159, 379.
Rubin, V. C., N. Thonnard, and W. K. J. Ford, 1980, Astrophys. J. 238, 471.
Ruggiero, M. L. (n), 2007, eprint 0712.3218.
Ruggiero, M. L., and L. Iorio, 2007, JCAP 0701, 010.
Saffari, R., and Y. Sobouti, 2007, Astron. Astroph. 472, 833.
Saidov, T., and A. Zhuk, 2006, Grav. Cosmol. 12, 253.
Saidov, T., and A. Zhuk, 2007, Phys. Rev. D75, 084037.
Salgado, M., 2006, Class. Quant. Grav. 23, 4719.
Salgado, M., D. M.-d. Rio, M. Alcubierre, and D. Nunez,
2008, eprint 0801.2372.
Sami, M., A. Toporensky, P. V. Tretjakov, and S. Tsujikawa,
2005, Phys. Lett. B619, 193.
Sandberg, V. D., 1975, Phys. Rev. D12, 3013.
Santos, J., J. S. Alcaniz, M. J. Reboucas, and F. C. Carvalho,
2007, Phys. Rev. D76, 083513.
Sawicki, I., and W. Hu, 2007, Phys. Rev. D75, 127502.
Schakel, A. M. J., 1996, Mod. Phys. Lett. 10, 999.
Schmidt, B. P., et al. (Supernova Search Team), 1998, Astrophys. J. 507, 46.
Schmidt, H.-J., 2007, Int. J. Geom. Meth. Phys. 4, 209.
Schmidt, H.-J., 2008, eprint 0803.0920.
Sciama, D. W., 1964, Rev. Mod. Phys. 36, 463.
Sealfon, C., L. Verde, and R. Jimenez, 2005, Phys. Rev. D71,
083004.
Seifert, M. D., 2007, Phys. Rev. D76, 064002.
Shahid-Saless, B., 1987, Phys. Rev. D35, 467.
Shahid-Saless, B., 1990, J. Math. Phys. 31, 2429.
Shahid-Saless, B., 1991, J. Math. Phys. 32, 694.
Shao, C.-G., R.-G. Cai, B. Wang, and R.-K. Su, 2006, Phys.
Lett. B633, 164.
Shirata, A., T. Shiromizu, N. Yoshida, and Y. Suto, 2005,
Phys. Rev. D71, 064030.
Shirata, A., Y. Suto, C. Hikage, T. Shiromizu, and
N. Yoshida, 2007, Phys. Rev. D76, 044026.
Simon, J. Z., 1990, Phys. Rev. D41, 3720.
Skordis, C., D. F. Mota, P. G. Ferreira, and C. Boehm, 2006,
Phys. Rev. Lett. 96, 011301.
Sokolowski, L. M., 2007a, Class. Quant. Grav. 24, 3391.
Sokolowski, L. M., 2007b, Class. Quant. Grav. 24, 3713.
Sokolowski, L. M., 2007c, eprint gr-qc/0702097.
Solin, P., 2006, Partial Differential Equations And The Finite
Element Method (Wiley, New York).
Song, Y.-S., W. Hu, and I. Sawicki, 2007, Phys. Rev. D75,
044004.
Sotiriou, T. P., 2006a, Class. Quant. Grav. 23, 1253.
Sotiriou, T. P., 2006b, Class. Quant. Grav. 23, 5117.
Sotiriou, T. P., 2006c, Gen. Rel. Grav. 38, 1407.
Sotiriou, T. P., 2006d, eprint gr-qc/0611158.
Sotiriou, T. P., 2006e, Phys. Rev. D73, 063515.
Sotiriou, T. P., 2007a, Phys. Lett. B645, 389.

48
Sotiriou, T. P., 2007b, Modified Actions for Gravity: Theory
and Phenomenology, Ph.D. thesis, SISSA, eprint 0710.4438.
Sotiriou, T. P., 2008, eprint 0805.1160.
Sotiriou, T. P., and V. Faraoni, 2008, eprint 0805.1249.
Sotiriou, T. P., V. Faraoni, and S. Liberati, 2007, eprint
0707.2748.
Sotiriou, T. P., and S. Liberati, 2007a, J. Phys. Conf. Ser. 68,
012022.
Sotiriou, T. P., and S. Liberati, 2007b, Annals Phys. 322,
935.
Soussa, M. E., and R. P. Woodard, 2004, Gen. Rel. Grav. 36,
855.
de Souza, J. C. C., and V. Faraoni, 2007, Class. Quant. Grav.
24, 3637.
Spergel, D. N., et al. (WMAP), 2007, Astrophys. J. Suppl.
170, 377.
Stabenau, H. F., and B. Jain, 2006, Phys. Rev. D74, 084007.
Starobinsky, A. A., 1980, Phys. Lett. B91, 99.
Starobinsky, A. A., 2007, JETP Lett. 86, 157.
Stelle, K., 1978, Gen. Rel. Grav. 9, 353.
Stelle, K. S., 1977, Phys. Rev. D16, 953.
Stoeger, S. J., William R., R. Maartens, and G. F. R. Ellis,
1995, Astrophys. J. 443, 1.
Stone, M., 2000, Phys. Rev. E62, 1341.
Strominger, A., 1984, Phys. Rev. D30, 2257.
Susskind, L., 2003, eprint hep-th/0302219.
Taylor, D. R., and R. Maartens, 1995, Gen. Rel. Grav. 27,
1309.
Teyssandier, P., and P. Tourrenc, 1983, J. Math. Phys. 24,
2793.
Tonry, J. L., et al. (Supernova Search Team), 2003, Astrophys. J. 594, 1.
Trodden, M., V. F. Mukhanov, and R. H. Brandenberger,
1993, Phys. Lett. B316, 483.
Tsujikawa, S., 2007, Phys. Rev. D76, 023514.
Tsujikawa, S., 2008, Phys. Rev. D77, 023507.
Tsujikawa, S., K. Uddin, and R. Tavakol, 2008, Phys. Rev.
D77, 043007.
Uddin, K., J. E. Lidsey, and R. Tavakol, 2007, Class. Quant.

Grav. 24, 3951.

Utiyama, R., and B. S. DeWitt, 1962, J. Math. Phys. 3, 608.
Vakili, B., 2008, eprint 0804.3449.
Vilkovisky, G. A., 1992, Class. Quant. Grav. 9, 895.
Vollick, D. N., 2003, Phys. Rev. D68, 063510.
Vollick, D. N., 2004, Class. Quant. Grav. 21, 3813.
Vollick, D. N., 2005, Phys. Rev. D71, 044020.
Vollick, D. N., 2007, Phys. Rev. D76, 124001.
Wagoner, R. V., 1970, Phys. Rev. D1, 3209.
Wainwright, J., and G. F. R. Ellis, 1997, Dynamical Systems
in Cosmology (CUP, Cambridge).
Wald, R. M., 1984, General Relativity (University of Chicago
Press, United States of America).
Wands, D., 1994, Class. Quant. Grav. 11, 269.
Wang, L.-M., R. R. Caldwell, J. P. Ostriker, and P. J. Steinhardt, 2000, Astrophys. J. 530, 17.
Wang, P., 2005, Phys. Rev. D72, 024030.
Wei, H., and S. N. Zhang, 2008, eprint 0803.3292.
Weinberg, S., 1972, Gravitation and Cosmology (John Wiley
and Sons, United States of America).
Weinberg, S., 1989, Rev. Mod. Phys. 61, 1.
Wetterich, C., 1988, Nucl. Phys. B302, 668.
Weyl, H., 1919, Ann. Phys. 59, 101.
White, M. J., and C. S. Kochanek, 2001, Astrophys. J. 560,
539.
Whitt, B., 1984, Phys. Lett. B145, 176.
Whitt, B., 1985, Phys. Rev. D32, 379.
Will, C. M., 1981, Theory and Experiment in Gravitational
Physics (Cambridge University Press, New York).
Woodard, R. P., 2007, Lect. Notes Phys. 720, 403.
York, J., James W., 1972, Phys. Rev. Lett. 28, 1082.
Zakharov, A. F., A. A. Nucita, F. De Paolis, and G. Ingrosso,
2006, Phys. Rev. D74, 107101.
Zhang, P., M. Liguori, R. Bean, and S. Dodelson, 2007, Phys.
Rev. Lett. 99, 141302.
Zhang, P.-J., 2007, Phys. Rev. D76, 024007.
Zhao, H., and B. Li, 2008, eprint 0804.1588.
Zwicky, F., 1933, Helv. Phys. Acta 6, 110.